The Limits of Syntactic Variation
Linguistik Aktuell/Linguistics Today (LA) Linguistik Aktuell/Linguistics Today (LA)...
85 downloads
1459 Views
4MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
The Limits of Syntactic Variation
Linguistik Aktuell/Linguistics Today (LA) Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.
General Editors Werner Abraham
University of Vienna / Rijksuniversiteit Groningen
Elly van Gelderen
Arizona State University
Advisory Editorial Board Cedric Boeckx
Christer Platzack
Guglielmo Cinque
Ian Roberts
Günther Grewendorf
Lisa deMena Travis
Liliane Haegeman
Sten Vikner
Hubert Haider
C. Jan-Wouter Zwart
Harvard University University of Venice
J.W. Goethe-University, Frankfurt University of Lille, France University of Salzburg
Volume 132 The Limits of Syntactic Variation Edited by Theresa Biberauer
University of Lund
Cambridge University McGill University
University of Aarhus University of Groningen
The Limits of Syntactic Variation
Edited by
Theresa Biberauer University of Cambridge and Stellenbosch University
John Benjamins Publishing Company Amsterdam / Philadelphia
8
TM
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.
Library of Congress Cataloging-in-Publication Data The limits of syntactic variation / edited by Theresa Biberauer. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166-0829 ; v. 132) Includes bibliographical references and index. 1. Grammar, Comparative and general--Syntax. 2. Principles and parameters (Linguistics) 3. Language and languages--Variation. 4. Language acquisition. I. Biberauer, Theresa. P295.L56 2008 415--dc22 2008024028 isbn 978 90 272 5515 0 (Hb; alk. paper)
© 2008 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
Table of contents
Preface and Acknowledgements Introduction
vii 1
Part I. The locus of (parametric) variation Parametric versus functional explanations of syntactic universals Martin Haspelmath
75
Three fundamental issues in parametric linguistics Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
109
On the syntactic flexibility of formal features Hedde Zeijlstra
143
Expletives, datives, and the tension between morphology and syntax Richard S. Kayne
175
Mapping a Parochial Lexicon onto a universal semantics Gillian Ramchand & Peter Svenonius
219
Aspect matters in the middle Marika Lekakou
247
Part II. A classic parameter revisited: The null subject parameter The null subject parameter and correlating properties: The case of Creole languages Marco Nicolis
271
The Case-F valuation parameter in Romance Gerardo Fernández-Salgueiro
295
Silent arguments without pro: The case of Basque Maia Duguine
311
Case morphology and radical pro-drop Ad Neeleman & Kriszta Szendrői
331
Table of contents
Part III. Parametric clustering The macroparameter in a microparametric world Mark C. Baker
351
Topic prominence and null subjects Marcello Modesto
375
Non-configurationality: Free word order and argument drop in Turkish Balkiz Öztürk
411
Diachronic stability and feature interpretability E. Phoevos Panagiotidis
441
Part IV. The acquisition of parameters Can children tell us anything we did not know about parameter clustering? Larisa Avram & Martine Coene
459
Parameter setting and input reduction Arnold Evers & Jacqueline van Kampen
483
Index
517
Preface and Acknowledgements The contributions in this volume represent a selection of the papers that were originally accepted for presentation at the conference on The Structure of Parametric Variation which was held at the University of Newcastle Upon Tyne in September 2005 under the auspices of the Arts and Humanities Research Council (AHRC)-funded project on “Null Subjects and the structure of parametric variation” (AR14458). That conference also included three papers which could unfortunately not be included in the present volume owing to prior publication commitments. These are: • • •
David Adger – Variation without parameters; Sjef Barbiers – Word-order alternations in three-verb clusters as non-parametric variation; and Geoff Horrocks & Melita Stavrou – Parametric variation in the lexicalisation of semantic properties: The role of grammatical aspect in shifting a verb’s basic Aktionsart
My sincerest thanks to all the contributors for the focused, co-operative and extremely patient manner in which they approached the various stages of the publi cation process. Grateful thanks also to the reviewers for their valuable input – in alphabetical order: Karlos Arregi, Anders Holmberg, John A. Hawkins, Kasia Jaszczolt, Chris Johns, Jaklin Kornfilt, Ian Roberts, Michelle Sheehan, Ianthi Tsimpli and David Willis. And to Vicki Hart and Glenda Newton for their careful proofreading at different stages of the process. Additional thanks to Glenda for her help with the Index. Further thanks to the Cambridge students who showed such a lively interest in many of the papers in this volume, particularly – once again in alphabetical order – Bob Davies, Luke Donnan, Emma Game, Nadine Hitchon, Iain Mobbs and Neil Myler. And, last but not least, my thanks to Kees Vaes and Martine van Marsbergen at Benjamins for their patience, support and guidance.
Introduction 1. Introduction Martin Joos’s famous statement that “(l)anguages can differ from each other without limit and in unpredictable ways” (Joos 1957: 96) is often misrepresented as an assertion reflecting the thinking current during the first half of the 20th century.1 In reality, even the American descriptivist Franz Boas, to whom this view is ascribed, did not hold this view. “[I]n all the languages of the world, the number of processes which are utilized to express the relations of terms is limited”, Boas wrote in the introduction to the Handbook of American Indian Languages (1911: 67). More generally, there is also no evidence that other American descriptivists (Sapir, Bloomfield, etc.) held the extreme view expressed by Joos: these linguists appear to have been far less committed to the notion of boundless variation in human languages than to the belief that it was simply too early to speculate about the possible limits on this variation, given how little was known at the time about non-Indo-European languages. Thus Bloomfield (1933: 20), for example, observes that: [t]he fact that some features are … widespread is worthy of notice and calls for explanation; when we have adequate data about many languages, we shall have to return to the problem of general grammar and to explain these similarities and divergences
and he also writes of: [l]ooking forward to a larger synthesis, a General Grammar, which will register similarities between languages (Bloomfield 1934: 46).2
In the 50 years since Joos’s statement, “general grammar” has, in one form or another, been very much at the forefront of linguistic research, with researchers in various frameworks having become interested in language universals and also the associated question of what limits human languages are subject to. Thus the work
. Cf. Thomas (2002) for extensive discussion of this misinterpretation. . Cf. Roberts (2007b) for further discussion of the extent to which linguists of the late 19th and early 20th centuries were interested in comparative issues of this sort.
The limits of syntactic variation
of Greenberg and Chomsky and their followers has always been guided by the fundamental idea that the very striking and seemingly limitless superficial variation between languages in fact conceals “uniformities of universal scope” (Greenberg, Osgood & Jenkins 1961/1963: xv). Furthermore, while the manner in which typologists and generativists have approached the search for these “uniformities”, or “universals” as they are more commonly known, has been very different, there is today agreement between those in the two camps that it is indeed the case that the variation exhibited by the world’s languages is not random. In particular, there is agreement that the syntactic variation across languages is neither “without limit” nor “unpredictable”, and it is this aspect of linguistic variation that is the focus of this volume.
2. The non-randomness of syntactic variation Arguably the most oft-cited illustration of the non-arbitrariness of syntactic variation has its origins in Greenberg’s pioneering work on word-order universals (cf. Greenberg 1963). This work brought to light the surprising fact that seemingly unrelated word-order patternings tend to correlate3 – S(ubject)-O(bject)-V(erb), postverbal auxiliaries and postpositions, for example; VSO, preverbal auxiliaries and prepositions; and so on. Crucially, it also introduced the idea that one might want to think of language types defined in syntactic terms. Thus Greenberg himself distinguished SVO, SOV and VSO languages and proposed that these might be viewed as the “basic word-order types”, with a language’s “type” determining the patterning of various pairs of elements, e.g., V-Auxiliary and N-Adposition order, as noted above, and also Adjective-Noun, Genitive-Noun, and so on. Subsequent work, notably that by Lehmann (1973); Venneman (1974) and Hawkins (1983), refined not only Greenberg’s word-order typology, but also the notion of “harmony” that emerged from it. Thus the well-known generalizations in (1) became established, along with the idea that so-called “harmonic” languages, which conform to these correlations, could be distinguished from so-called “disharmonic” languages, which do not: (1) VO: Aux-V, Prepositions, Noun-Adjective, Noun-Genitive, Noun-Relative OV: V-Aux, Postpositions, Adjective-Noun, Genitive-Noun, Relative-Noun
. In itself, this was not an entirely new observation: as Greenberg (1963: 104, note 4) himself observes, researchers such as Lepsius (1880) and in particular Schmidt (1926) had previously drawn attention to a subset of the word-order correlations identified by Greenberg.
Introduction
Later work, i.a. by Tomlin (1986); Dryer (1992) and by the contributors to the World Atlas of Language Structures (WALS, cf. Haspelmath et al. 2005), has served to reveal the extent to which the Greenbergian language types are in fact instantiated in the world’s languages. Certain correlations are shown to constitute very strong correlations, the OV/VO and adposition one being a case in point, as (2) below shows:
(2) Correlation between OV/VO ordering and adpositional structure in the 892 languages surveyed in WALS (cf. Baker, this volume)
OV & Postposition VO & Preposition OV & Prepostion VO & Postposition
427 (47.87%) 417 (46.75%) 10 (1.12%) 38 (4.26%)
Thus 94.5% of the languages surveyed in the WALS are “harmonic” in respect of their OV/VO and adposition structure. As Dryer (1992, note 17) notes, however: The general conclusion is that completely consistent [i.e., harmonic – TB] types represent a minority, but that the majority of inconsistencies among inconsistent languages can be attributed to a small number of pairs of elements for which there is a skewed distribution, such as the general preferences for NRel (ative) order.
Viewed across the board, then, it is not realistic to think about the Greenbergian word-order-based types in absolute terms – they do not entail absolute universals, but rather statistical universals or universal tendencies (cf. i.a. Greenberg 1963; Hawkins 1983).Worth noting, however, is Dryer’s observation that the limited number of fully harmonic languages is largely due to the existence of a fixed set of harmony-disrupting element-pairs which tend to surface in only one of the two possible orderings relative to each other. What we are left with, then, is the impression that, even though harmonic languages are in the minority, there is a significant amount of non-randomness in the domain of word order, non-randomness which even extends to the domain of exceptions to the observed regularities. And the same also appears to be true in a range of other syntactic and non-syntactic domains (cf. the Universals Archive at http:// typo.uni-konstanz.de/archive/intro/index.php for an inventory of the currently more than 2000 typological universals recorded to date). The question, of course, is why?
The limits of syntactic variation
3. Accounting for limits on syntactic variation 3.1 Two different approaches: Functionalism vs. formalism4 As Haspelmath (this volume) notes, typologists have tended to follow a functionalist line in attempting to answer the question of why languages should display the structural parallels that they do. More specifically, researchers working within the typological framework generally appeal to communicative and/or processing considerations in order to account for the structural skewings that have been observed in the world’s languages (cf. also Hawkins 1988, 2004 and Newmeyer 1998, 2005a for extensive discussion). Thus, for example, Hawkins (2004: 103) proposes that the following parsing efficiency principle plays a central role in accounting for the word-order correlations in (1) and (2):
(3) Minimize Domains (MiD) The human processor prefers to minimize the connected sequences of linguistic forms and their conventionally associated syntactic and semantic properties in which relations of combination and/or dependency are processed. The degree of this preference is proportional to the number of relations whose domains can be minimized in competing sequences, and to the extent of the minimization difference in each domain.
In terms of (3), then, the adjacency of V and P and P and N that is ensured in consistently head-initial and consistently head-final languages (head-initial: [VP V [PP P [NP N …]]] and head-final: [VP [PP [NP … N] P] V]]) minimizes the domain within which the elements making up VP and PP (V and PP and P and NP respectively) will be recognized. The fact that “disharmonic” patterns generally are not as common as “harmonic” ones is thus ascribed to the added processing burden that they entail. In contrast to typologists, generativists in general mostly seek to understand crosslinguistic regularities and gaps by relating these to the influence of an innately specified and formally oriented linguistic capacity, Universal Grammar (UG), on which, more below. It is worth noting, however, that it is not the case that
. As Newmeyer (1998: 7–8) notes, formal is an unhelpfully ambiguous term, meaning both “pertaining to grammatical form (as opposed to meaning and/or use)” and “stated in mathematically precise terms as in formalized”. Clearly, there are formal approaches to language that are only formal in the former sense (cf. Saussurean structuralism), but these will not concern us here. For the purposes of the present discussion, then, formal is to be understood as implying both of the above meanings in the manner generally assumed by generativists. Worth noting in connection with formal in the second sense is that it does not generally in syntactic theory mean anything as strong as “axiomatised” or even “fully formalized”, as it does in, for example, logic, computational work or formal semantics.
Introduction
generativists uniformly reject the possibility that functional and, particularly, processing considerations may determine the distribution of syntactic patterns, although this has undoubtedly been the most common trend to date (see below). Thus, for example, Ackema & Neeleman (2002) argue that the very striking asymmetry between leftward and rightward movement observed in the world’s languages – in terms of which leftward movement is vastly more common than rightward movement and additionally appears to be unbounded in a way that rightward movement is not – can in fact best be understood in parsing terms.5 More specifically, they, like non-generativist psycholinguists (cf. i.a. Just & Carpenter 1992; Gibson 1998), propose that processing of syntactic representations involves a left-to-right parsing procedure which involves both processing and storage: since it is presumably more economical to assign already-parsed structure to short-term memory as soon as this becomes possible rather than to wait until the entire structure has been parsed, it is assumed that already-parsed structure is “closed off ” at appropriate points during the parsing process. This, then, renders its internal structure “invisible” to the parser. If, as is usually thought, the parser employs a filler-driven strategy in dealing with movement dependencies (cf. Clifton & Frazier 1989; Frazier & Flores D’Arcais 1989), it will only postulate a trace after it has encountered a moved element. Clearly, this is always the case when we are dealing with leftward movement (cf. (4a)), but the opposite is true wherever rightward movement has taken place (cf. (4b)): (4) a. wh-element/topicalized dp, etc. ….. twh-element/tDP, etc. b. twh-element/tDP, etc ……… wh-element/topicalized dp, etc.
As (4b) illustrates, rightward movement is problematic within filler-driven, interim storage-oriented approaches to parsing in that it requires post hoc trace insertion in a partially analysed string: left-to-right parsing must be suspended so that a trace can be inserted at an appropriate place in an already-parsed piece of structure located to the left of the point the parser had reached. If it is indeed the case that such structure has been “closed off ” and therefore lacks internal structure, we might expect this kind of trace-insertion to be impossible or at least costly. And the same mode of explanation is available to account for the locality restrictions on rightward movement. As Ross (1967) first observed, this is always clausebounded (cf. the so-called Right Roof Constraint). On the natural assumption that distinct clauses constitute distinct parsing domains, we would expect it to be . Other well-known examples of generativists appealing to parsing to account for formal patterns include: Bever & Langendoen (1971) in their work on the factors determining the availability of relative pronoun omission in English, Fodor (1978, 1984) in her work on nested dependencies, Berwick & Weinberg (1984) in their work on Subjacency, and the work of Janet Dean Fodor & Lyn Frazier more generally.
The limits of syntactic variation
impossible for rightward movement to be unbounded in the manner that leftward movement appears to be. This is because unbounded leftward movement is unproblematic if long-distance movement is indeed, as has generally been assumed since Chomsky (1973), successive-cyclic: if elements move leftward successive-cyclically, each clause will be headed by a trace, facilitating easy tracking of the filler-gap dependency, even where earlier clauses have already been “closed off ” – the trace at the start of every parsing domain alerts the parser to the fact that its search is not yet complete. As far as the parser is concerned, then, filler-gap dependencies can be established in the usual local manner wherever movement is successive-cyclic and leftward. The same is not, however, true for rightward movement beyond the clause to which the moved element belongs. Even if this movement is successive-cyclic, it will still entail the requirement that previously “closed off ” material needs to be “reopened”. This is because traces in clauses preceding that to which the element has moved cannot be postulated until the parser has encountered the gap located at the rightmost edge of the final clause. Long-distance rightward movement therefore entails multiple “reopenings” of the type required in monoclausal rightward-movement structures. Ackema & Neeleman (2002) explicitly indicate why a purely formal account of these phenomena – for example, one appealing to Kayne’s (1994) Antisymmetry proposal, in terms of which all movement is necessarily leftwards because of structural constraints (specifiers all being left-adjoined, with the result that upward movement can only be to the left) – fares less well than a parsing-oriented approach (cf. also Abels & Neeleman 2007). The details of this argument fall beyond the scope of this introduction; the point here is simply that it is not always the case that formalists exclusively seek to understand syntactic distributions and the factors determining them in purely formal terms. And, more generally, there is, as Newmeyer (1998: 21) rightly notes, “nothing in the generative program that demands that all typological facts be attributed to the setting of innately specified parameters” (cf. also Newmeyer 1998, chapter 3). Furthermore, there is also nothing in the generative research programme that excludes the possibility that innately prespecified principles may ultimately be functionally grounded (cf. Müller 2007a, numerous contributions in Good (2008), in particular, Kiparsky 2008, who proposes criteria designed to distinguish UG-based universals from those which have historical – and thus potentially ultimately functional – explanations, and Mobbs 2008, who suggests a “third factor” reinterpretation of Hawkins’s performance principles6). We return to these matters below.
. Kiparsky (ibid.3) explicitly refers to functional principles which “might have become biologized within UG itself ”. Cf. Section 3.4 for discussion of “third factor” considerations within current Chomskyan theory which may also include functional considerations.
Introduction
First, however, let us consider the UG-based approach and, in particular, that which generativists have espoused since the introduction of the Principles and Parameters model in the late 1970s/early 1980s. 3.2 Introducing the Principles and Parameters Framework Universal Grammar (UG) is the innately specified mental endowment which is assumed to be uniquely shared by all humans. In terms of the current general model of UG, the Principles and Parameters (P&P) model first outlined in Chomsky’s 1979 Pisa Lectures (published in revised form as Chomsky 1981), UG consists of invariant principles and unset parameters whose values must be set on the basis of the linguistic input available to language-acquiring children (the primary linguistic data or PLD) – cf. (5): (5) UG (principles & unset parameters) + S0 (the initial state)
PLD →
Adult grammar Ss (the steady state)
Together, the immutable principles and the initially unset parameters (i.e., S0 in (5)) are taken to determine the form of a possible human grammar (i.e., Ss in (5)), where “grammar” is understood as the abstract mental system, including specific parametric settings, – the I-language in Chomsky’s terms (cf. Chomsky 1986 et seq.) – on the basis of which actual utterances are produced. For Chomskyans, then, UG plays a role not only in language acquisition, but also in defining the notion “possible language”, both synchronically and diachronically. And, crucially, the limits it is assumed to impose on “possible languages” and, in particular, the syntactic variation that they may exhibit, are conceived of in formal terms. In P&P’s first incarnation during the Government and Binding (GB) period, a range of principles and parameters were proposed, with the X-bar schema, the Theta Criterion, the Extended Projection Principle (EPP) and Subjacency/Relativised Minimality numbering among the most frequently cited principles and the Head Parameter (HP) and the Null Subject Parameter (NSP) undoubtedly being the most well-known parameters. In terms of the former, there are universal constraints on the way languages construct phrases, the way in which thematic roles like Agent, Patient and Goal are assigned, the positioning of subjects, and the distance over which elements can undergo movement: no languages should violate these principles, with the result that no variation in these domains is expected (but see below). Parameters, on the other hand, do permit variation, although it is variation of a highly restricted kind. Consider, by way of illustration, the HP and the NSP in (6–7).
The limits of syntactic variation
(6) Head Parameter (cf. Travis 1984) Heads X precede/follow their complements a.
X'
b.
X' Comp X → -
X Comp → -
(7) Null Subject Parameter (cf. Rizzi 1982, 1986) Subjects can be pro, where pro is a [+pronominal, -anaphoric] entity which is licensed by X0y where X0y is a subset of the heads in a given language. The grammatical specification of pro is recovered from the features of the licensing head. e.g.
a.
IP Spec pro
b. I'
I [+pronominal] → I licenses pro
IP Spec overt pronoun
...
I'
I ... [–pronominal] → I does not license pro
As shown above, both the HP and the NSP define an innately specified choicepoint, the HP in respect of head-complement directionality and the NSP in respect of the availability of pronominal null subjects. Just like principles, then, parameters impose universal restrictions – they determine the set of structural choices available to languages in various domains, thereby ensuring that all languages make choices about the same set of choice-points. Because languages do not make the same choices in each case, variation results. Thus the P&P approach opens up the possibility of having a model of grammar which entails a finite and potentially relatively simple UG, but which nevertheless leads us to expect what might on the surface seem like considerable crosslinguistic variation – consider the striking differences between languages that opt for different settings of the HP, for example. That is, the P&P approach, unlike those that went before it, facilitates a resolution of the tension between the search for both descriptive and explanatory adequacy, being able to account not only for the (apparent) diversity of syntactic structures, but also for how these could have been successfully acquired given the constraints under which language acquisition takes place (cf. Chomsky 2004b for overview discussion, and also Section 7 below). In accordance with the schema in (5), the values assigned to the various parametric choice-points are assumed to be determined on the basis of specific
Introduction
PLD, signalling the appropriate setting. With the HP, at least in the case of the consistently head-initial/head-final languages (i.e., the ones exhibiting harmonic order in respect of V and its complement, Aux and V(P), Adpositions and their complements, Nouns and their complements, & so on7), it is not hard to see what would be an appropriate cue (cf. Lightfoot 1999 and Dresher 1999, 2003 for discussion of the notion “cue”): any head + complement-containing XP will provide the necessary cue. Thus, for example, any one of the following Japanese structures would enable the learner to set the HP to head-final: (8) a. Sensei-wa [Taro-o] sikata teacher-top Taro-acc scolded ‘The teacher scolded Taro.’ → Objects precedes V b.
[Nihon] kara Japan from ‘from Japan’ → Complement precedes adposition
c. John-ga [Mary- to renaisite] iru John-nom Mary-with in.love is ‘John is in love with Mary.’ → V precedes Auxiliary d.
[John-ga Mary-o nikundeiro] to John-nom Mary-acc be-hating that ‘that John hated Mary’ → Clause precedes Complementizer8
As noted above, however, fully harmonic languages represent the minority of the world’s languages and many languages, including very well studied ones like German, exhibit mixed headedness, with certain phrases seeming to be . It is worth noting that the HP does not, as is often thought, determine the relative ordering of Noun-Adjective as adjectives are not nominal complements. Similarly, the HP has nothing to say about those Noun-Genitive and Noun-Relative pairings which do not involve a complementation relation. The fact that, for example, English exhibits Adjective-Noun rather than the Noun-Adjective order expected in a harmonic head-initial language (cf. (1) above) does not, therefore, reflect a shortcoming of the HP. . If, as Lightfoot (1989, 1991) has suggested, learners do indeed only pay attention to matrix clauses at the stage at which parameter setting is underway, (8d) may play no role in the setting process. As it is only necessary for the child to establish the directionality of a single XP (say, the VP) in order to set the HP, this does not, however, have to pose particular problems. The same is obviously not true for “mixed” languages, though (see main text).
The limits of syntactic variation
consistently head-initial (e.g., CP and DP in German), while others appear to be consistently head-final (VP and TP in German). This is illustrated in (9): (9) a.
… daß er das Buch von Chomsky unter dem Tisch gefunden hat that he the book from Chomsky under the table found has ‘… that he found the book by Chomsky under the table’
b. [CP daß [TP eri [VP ti [DP das Buch von Chomsky] [PP unter dem Tisch] gefunden] hat]] i.e., C, D, N and P precede their complements; V and T follow them.
There also appear to be quite a few languages which exhibit head-directionality of one kind in the clausal domain and the opposite in nominals (cf. Thai, Khmer and Burmese as discussed in Simpson 2005, Bangla as discussed in Bhattacharya 1999, Marathi as discussed in Nayudu 2008, and Gungbe as discussed in Aboh 2004, etc.). This is shown for Gungbe in (10): (10) dàwé l� x� kὲkέ man det bought.perf bicycle ‘The man bought a bicycle.’
(from Aboh 2004: 32)
Worse still, there are further cases where individual lexical items appear to determine the headedness of the phrases in which they surface. Thus German adpositions are generally head-initial (cf. von and unter in (9)), but there are a few exceptions like zufolge (“according to”) and halber (“for the sake of, because of ”) and the “shadow prepositions” (cf. Noonan 2003) herauf (“up”) and drin (“in”) which only surface finally. The “mixed” complementiser class in languages like Bengali, Marathi and Assamese, in which certain complementisers are head-initial while others are head-final, represents another problematic case (cf. Bayer 1999, 2001). Furthermore, there are lexical items like German entlang (“along(side)/beside”) and Afrikaans in (“in”/“into”) which occur both initially and finally, but where positioning determines other morphosyntactic and/or semantic properties – in the case of entlang, which case its complement takes (genitive where entlang is initial and accusative where it is final) and in the case of in, precisely how it is interpreted (as locative “in” when it is initial and as directional “into” when it is final).9 The examples in (11–12) illustrate: (11) a. Sie ist entlang des Flusses gefahren she is along the-gen river driven ‘She drove along the river.’
[German]
. We leave aside here the potential further complication represented by so-called circumpositions (cf. Dutch naar … toe – “to … to”) which, superficially at least, appear to consist of nested adpositional phrases, with a head-initial phrase embedded under a (superficially) head-final one (cf. i.a. van Riemsdijk 1978, 1990; Koopman 1997; Zeller 2001; and den Dikken forthcoming).
Introduction
b. Sie ist den Fluss entlang gefahren she is the-acc river along driven ‘She drove along the river.’ (12) a. Hy loop in die bos he walk in the forest ‘He is walking in the forest.’
[Afrikaans]
b. Hy loop die bos in he walk the forest in ‘He is walking into the forest.’
In at least some cases, then, head-directionality appears to play a lexical disambiguation role.10 Finally, there are elements like the prepositional complement-taking adjective stolz (“proud”) in German and its counterpart in Dutch (trots) and adpositions like ilman (“without”) and kohti (“towards”) in Finnish which can surface either before or after their complements without either morphological or semantic consequences. Consider (13–14) in this connection: (13) a. Die Frau ist stolz auf sich the woman is proud of self ‘The woman is proud of herself.’
[German]
b. Die Frau ist auf sich stolz the woman is of self proud ‘The woman is proud of herself.’ (14) a. kohti Stalinin kuvaa towards Stalin-gen picture ‘towards a picture of Stalin’
[Finnish]
b. Stalinin kuvaa kohti Stalin-gen picture towards ‘towards a picture of Stalin’
. If it is indeed correct to think of lexically-specified head-directionality, one might want to appeal to Elsewhere considerations (cf. Kiparsky 1973) in accounting for the word-order properties of structures containing these elements. More specifically, it could conceivably be the case that the HP is appropriately set for a specific set of syntactic categories, but that lexical items associated with independent (i.e., irregular) linearisation instructions are linearised in accordance with their more specific specification. If this approach is extended to cases permitting both linearisation options (e.g., the German and Dutch adjectives and Finnish adpositions mentioned in the main text), we would need to postulate two distinct lexical entries which are phonologically and semantically identical, but which differ in terms of their linearisation instructions (cf. Adger 2006 for an optionality proposal along these lines).
The limits of syntactic variation
Taken together, these facts raise serious questions about how children successfully identify the appropriate cues to set the HP (cf. Dresher 1999, 2003; Fodor 1998a, 1998b, 2001; Fodor & Sakas 2005; Sakas & Fodor 2001, and Lightfoot 1999 for recent discussion of the general problem of cue identification, and also the contributions by Gianollo, Guardiano & Longobardi (henceforth: GGL), Zeijlstra, Avram & Coene and Evers & van Kampen in this volume). And they also suggest that the HP may be a rather different type of parameter to the NSP, which is usually (though see Sections 3.4 and 5 below) thought to involve a binary choice which is made once and for all for the language as a whole. Specifically, one might view the HP as a parameter which can operate in a more or less “myopic” fashion as outlined below: a. “blindly”, being set once for all categories, thereby delivering harmonic languages; b. category-sensitively, being consistently set in one way for a fixed set of syntactic categories and another way for another fixed set; or c. lexical item-sensitively, being set individually for specific lexical items (although see note 10). Despite this fact, the general assumption about the HP during the GB era was that it was, universally, a parameter specifically associated with the phrase-structure (X-bar) module of the grammar. Turning to the NSP: this parameter clearly involves two components, namely (a) that a null-subject language (henceforth: NSL) include in its pronoun inventory the phonologically null, lexically unspecified pronoun pro and (b) that it include in its inventory of functional categories subject-related ones (e.g., I in the clausal domain and/or D in the nominal domain) which are capable of assigning pro the appropriate φ-features (here: person and number). Non-NSLs, by contrast, lack one or both of these properties. Traditionally, the assumption was that the NSP was set to its positive value – in terms of which children postulate the availability of pro and the appropriate licensing head(s) – on the basis of PLD such as (15) featuring unrealised referential pronominal subjects as in (15a) and unrealised expletive pronominal subjects as in (15b) and (15c): (15) a. Habl- a español speak-3sg Spanish ‘(S)he speaks Spanish.’ b. Piove rain-3sg ‘It is raining.’
[Spanish]
[Italian]
c.
Introduction
Sono venute delle ragazze be-3pl come-f.pl some girls ‘There came some girls.’
[Italian]
Just as was the case with the HP, the NSP was assumed to determine the (un)availability of a range of structures not limited to those which, like those in (15), were generally assumed to constitute the actual cue required to trigger the appropriate setting of the parameter. Thus the NSP sanctioned not only null-subject structures such as those in (15), but also structures like the following: Ha telefonato Gianni have-3sg telephoned Gianni ‘Gianni has called.’
[Italian]
b. Pjos ipes oti egrapse afto to vivlio? who said-2sg that wrote this the book ‘Who did you say wrote this book?’
[Greek]
(16) a.
(16a) represents a so-called Free Inversion structure, in which the overt subject follows the lexical verb rather than occupying the canonical subject position, Spec-IP. (16b), in turn, illustrates the absence of so-called that-trace effects in NSLs: unlike in non-NSLs like English, wh-subjects can be extracted across overtly realised complementisers without resulting in an ungrammatical structure (contrast the English counterpart of (16b), *Who did you say that wrote this book?). Neither of the structures in (16) at first sight seems to relate to the availability of null subjects, but Rizzi’s (1982, 1986) characterization of the NSP links all of these phenomena and additionally also the fact that Romance NSLs exhibit “rich” agreement morphology. The reasoning is as follows: the availability of pro facilitates the existence of null-subject structures such as those in (15), with the “rich” agreement morphology here and elsewhere reflecting the fact that the language has the kind of I required to supply pro with the φ-features it intrinsically lacks. The availability of pro also enables a subject to occur in inverted position as in (16a) since pro is available to satisfy the Extended Projection Principle (EPP), i.e., the requirement that every clause contain a subject located in Spec-IP (cf. Chomsky 1981, 1982). Finally, pro also facilitates the circumvention of that-trace violations, as in (16b), by allowing wh-subjects to be extracted from a lower position, while pro itself once again satisfies the EPP. Clearly, then, the NSP as formulated by Rizzi leads us to expect superficially unrelated phenomena to cluster together.11 And if it is indeed correctly
. Cf. Roberts (2007c) for discussion of three further phenomena that have been proposed to be connected to the NSP: clitic climbing, infinitive-clitic ordering and realisation of an arbitrary subject as a clitic.
The limits of syntactic variation
formulated, we would expect it to impose a universal limit on the types of structures available both in NSLs and in non-NSLs: the former would all be expected to exhibit all of the structures facilitated by pro and a suitable I, while the latter would all be expected to systematically lack these structures. Looking beyond the NSP, this was also the expectation for GB parameters more generally since parameters with a rich deductive structure simultaneously facilitate an account of systematic syntactic correspondences across languages and an account of how it might be possible for children to acquire these systematic correspondences with relative ease. If we consider the NSP, for example, it is arguably the case that unrealised pronominal subjects like those in (15) represent a salient and easily detectable component of the PLD. By contrast, that-trace circumvention structures such as those in (16b) are likely to feature far less frequently in the input and, superficially at least, these structures seem a great deal more complex than unrealised subjectcontaining ones like those in (15). Even more striking in this connection is the scenario faced by children acquiring non-NSLs: in this case, the input does not contain any positive evidence that that-trace structures are unavailable, with the result that one would expect that children would have to be explicitly told that subject wh-extraction is not possible in the presence of an overt complementiser, unlike extraction of other wh-elements. Such direct negative evidence is, however, not usually thought to play a significant role in first language acquisition (cf. i.a. Bowerman 1988; Marcus 1993) and, in the case at hand, it is furthermore not at all clear that adult nativespeakers are actually consciously aware of the existence of this structural gap – it is certainly not one that is noted in prescriptive grammars or explicitly taught in English grammar classes. In both the NSL and the non-NSL case, then, a question arises as to how native speakers are able to acquire structures that are infrequent or even unattested in the input. If we can, however, assume in the NSL case that unrealised pronominal subjects alert children to the other NSP-related properties, the salience and frequency discrepancies noted above need not pose an insurmountable obstacle. Similarly, if we can assume that the obligatoriness of overtly realised subjects (and particularly, overt expletives) serves the same function for children acquiring a non-NSL, we can understand the successful acquisition of that-trace phenomena as the consequence of the fact that the unavailability of wh-subject extraction is simply part of the cluster of NSP-related properties associated with the negative setting of this parameter. More generally, if innately specified parameters do indeed determine the way in which certain types of syntactic phenomena cluster, we would expect, firstly, that the world’s languages (past, present and future) will exhibit certain clusters of properties and not others and, secondly, that language acquisition will proceed in a way which reflects the (incremental) acquisition of relevant parametric settings
Introduction
and that will therefore never feature stages at which children correctly use some of the structures associated with a given parameter, but also others that are associated with the opposite setting of that same parameter. That is, parameters as they are conceived of in the P&P framework aim to contribute to an explanation not only of the limits of syntactic variation, but also of the limits on language acquisition and, by extension, language change. 3.3 Problems with GB principles and parameters As is now well known, many of the principles and parameters proposed during the first phase of the P&P era did not ultimately prove either descriptively or explanatorily adequate. Consider, for example, the fact, already noted above, that, while the HP facilitates an elegant account of languages exhibiting “harmonic” word orders, it is evidently too simplistic to assume that head-directionality is universally fixed once-and-for-all on the basis of the simple binary choice in (6): the empirical evidence clearly points to a much more complex situation. Similarly, the NSP as formulated above leads to various expectations: that NSLs will be morphologically rich,12 that morphologically poor languages will not be NSLs, that the properties governed by the NSP (cf. (15–16) above) will always cluster together, and so on. As work as early as Gilligan (1987) has shown, however, these predictions are not borne out: numerous morphologically rich languages are not NSLs (e.g., Icelandic and Russian), there are various morphologically poor languages that are NSLs (e.g., Chinese and Cheke Holo), and there are also other respects in which the proposed correlations do not always hold (cf. Bantu languages exhibiting referential null subjects, but not Free Inversion and non-NSLs like Icelandic exhibiting the no that-trace effect property of NSLs). Some of these failed predictions have been ascribed to parameter interaction, with the idea being that the non-occurrence of an expected parametric reflex follows from the obscuring effects of the setting of another parameter which interacts with the first. Thus, for example, the non-occurrence of referential null subjects in inflectionally rich Icelandic is commonly ascribed to the positive setting of the V2 Parameter in this language – that is, since the V2 Parameter always requires an XP to occupy the preverbal position in main clauses, preverbal null subjects are
. It is worth being precise here: what is specifically expected is that NSLs will have rich agreement (i.e., person/number) morphology; rich tense, mood or other verbal morphology should not be relevant to the setting of the NSP, as the crucial consideration is that the richness of the verbal agreement morphology should facilitate recovery of pro’s content (cf. Biberauer & Roberts, forthcoming).
The limits of syntactic variation
ruled out in these contexts and consequently the effects of the NSP are masked. Clearly, though, this proposal only accounts for the absence of preverbal null subjects; it tells us nothing about why referential null subjects are also unavailable elsewhere and why expletive null subjects, but not their referential counterparts, may surface in postverbal position. More recently, researchers have adopted a different tack, opting instead to re-examine the morphological paradigms of languages which at first sight appear to qualify as “morphologically rich” and proposing that languages like Icelandic do, after all, systematically differ from NSLs (cf. Müller 2005, 2007b & Tamburelli 2006 for two recent proposals of this type). More generally, however, the notion of “morphological richness” has proven very hard to quantify, not only in the NSP context, but also more generally (cf. Jaeggli & Safir 1989; Vikner 1997; Rohrbacher 1999 for different, but ultimately unsuccessful proposals). Clearly, this is a highly problematic state of affairs for a parameter whose setting was initially thought to be at least partially cued by inflectional wealth. Looking beyond the HP and the NSP, we see that other GB parameters also encountered difficulties of various kinds. Thus, for example, in the V2 context, the discovery that V2 languages are not all “well behaved” and that languages like Icelandic and Yiddish systematically permit V2 in the presence of overtly realised complementisers first raised the question whether one can in fact think of a unitary V2 parameter (cf. Schwartz & Vikner 1996 and Vikner 1995 for relevant discussion). The rise since the late 1980s of so-called “cartographic” approaches to clause structure, in terms of which functional structure is assumed to be a great deal more articulated than initially thought, has only underscored this question (cf. Poletto 2000; Mohr 2005 and Roberts 2005 for illustrations of articulated CP analyses of V2 phenomena). Furthermore, independently of their precise formulation, this and other head-movement-related parameters (e.g., the V-to-I Parameter, the N-to-D Parameter, etc.) would also seem to face a further conceptual question, namely whether they really meet the requirement that parameters should govern clusters of empirical phenomena: head-movement parameters are not generally thought to give rise to the sorts of superficially unrelated phenomena proposed for the NSP.13
. Viewed from a Germanic perspective, the V-to-I parameter could be argued to govern a cluster of unrelated phenomena – consider, for example, Bobaljik & Thráinsson’s (1998) SplitINFL parameter, in terms of which languages exhibiting V-to-I also feature a split IP, full DP Object Shift (OS), and Transitive Expletive Constructions (TECs). This (full) cluster does not, however, seem to be more generally associated with languages which have been analysed as featuring V-to-I movement, e.g., the Celtic languages, the Semitic languages, which at most permit pronominal OS (cf. Shlonsky 1997, chapter 10), and the Romance languages, which also lack non-pronominal OS and TECs of the Germanic type (cf. Cornilescu 2000 for a proposal that Romanian features TECs). Strikingly, Avram and Coene, this volume, propose a very different
Introduction
Baker (this volume) dubs this the extent of variation question, and we return to it in Sections 3.4 and 3.5 below. Related to it is the question that we have already raised in connection with the HP, namely whether binary parameters can in fact hope to account for the attested range of variation within a given domain. Since this question in fact raises a number of important issues, let us consider it in a little more detail on the basis of a previously undiscussed parameter. Both during the GB era and beyond, it has frequently been proposed that the presence vs. absence of wh-movement in a language is determined by the positive vs. negative setting of a Wh-movement Parameter (cf. Fukui 1995; Cheng 1997 and Simpson 2000 for discussion). If this parameter simply offers a binary choice, we would expect to find only two types of languages: wh-movement and wh-in-situ languages, illustrated in (17): (17) a.
Whati did
Steve eat ti?
b. Hufei chi- le shenme Hufei eat-asp thing ‘What did Hufei eat?’
ne Qwh
[Chinese]
In actual fact, however, some languages have been shown to exhibit a mixture of these properties, permitting both movement and in situ structures (cf. Cheng 1997). Crucially, closer investigation has revealed that some of the “mixed” languages do not in fact pose an empirical challenge to the postulation of a binary Wh-movement Parameter. Some, like Egyptian Arabic, Irish, Malagasy and Zulu, can, for example, be shown to involve very different underlying structures: a (pseudo-)cleft (i.e., a non-movement structure) in the (apparent) wh“movement” case and an in situ structure in the non-movement case. This is illustrated in (18): (18) a.
U- bona ubani?14 2sg- see who.1a ‘Who did you see?’
b. Ng- ubani o- m- bona-yo? cop-who.1a rc.2sg-oc.1a- see -rs ‘Who is it that you see?’
[Zulu]
(data from Sabel & Zeller 2006)
cluster of Split-INFL-related effects to that proposed by Bobaljik & Thráinsson. At the very least, then, it does not seem that V-to-I can be viewed as correlating with a single cluster of effects. . Sabel & Zeller’s abbreviations are as follows: 1a = noun class 1a, i.e., that mostly reserved for singular human entities; cop = copula; rc = relative concord/agreement; oc = object clitic; rs = relative suffix.
The limits of syntactic variation
As (18) clearly shows, the usual SVO order is maintained in monoclausal whinterrogatives, with the wh-element surfacing in situ; it is only in the presence of a cleft-marking copula that wh-elements may appear in a moved position, and then it is clearly a relativised one, rather than the clause-initial Spec-CP position associated with wh-movement languages. Similarly, wh-“movement” in Malay and Iraqi Arabic seems to be possible only in focus structures and not in unmarked interrogatives, i.e., wh-movement, where it occurs, takes place as a consequence of focus- rather than specifically whdriven movement (cf. Ouhalla 1996). From the perspective of the Wh-movement Parameter, then, these languages are negatively specified for wh-movement and the superficial “mix” of positioning possibilities can be explained despite the availability of only two mutually exclusive parametric options. Importantly, however, there do appear to be systems within which a putative Wh-movement Parameter could not be set once and for all. Duala, a Bantu language spoken in Cameroon, is a case in point: it permits both wh-movement and wh-in situ in matrix clauses, but necessarily requires wh-movement in embedded contexts (cf. Sabel 1998 for detailed discussion). Languages of this sort clearly cannot be specified as [+movement] as wh-elements are evidently not required to move (there is no (necessary) interpretive difference between movement and non-movement structures). Similarly, they cannot be specified as [–movement] since there is a context (embedded clauses) where movement is systematically required. Here, then, the Wh-Parameter does appear to face a genuine empirical challenge from optionality phenomena. Furthermore, a simple binary parameter also will not be able to account for the optionality exhibited by languages like Bangla, Basque, Imbabura Quechua and others which permit so-called clausal pied-piping to alternate with wh-DP raising, once again without necessary interpretive consequences (cf. Richards 2001 for overview discussion); and the same is true of that found in Slavic and other languages which permit Left Branch Condition/LBC-violating wh-extractions to alternate – often without interpretive consequences – with their non-LBC-violating counterparts (contrast English *Whose did he read book? [LBC violation] which is not available alongside Whose book did he read?). Additionally, a simple binary Wh-movement Parameter cannot explain why some languages (e.g., those belonging to the Slavic family) either obligatorily or optionally front all the wh-words present in multiple wh-questions, whereas others may only front one of them (cf. Bošković 2002 for detailed discussion). Like the “size”-related optionality mentioned above, these differences have the flavour of “sub-parameters” associated with the Wh-movement Parameter initially proposed: more precisely, they would seem to be choices which only come into play following positive setting of the Wh-movement Parameter (cf. also Uriagereka 2007 for recent discussion of parametric structure in terms of parameters which differ in their “deepness”, delivering, in his words (p.1), “progressively deeper forms of variation”).
Introduction
From a parametric perspective, the notion that a parameter that was initially viewed as a unitary Yes/No choice-point in fact turns out to initiate or be part of a sequenced network or hierarchy of related parameters need, of course, not be problematic. We return to this matter in Section 3.3 below. For the moment, it worth noting, however, that the more articulated view of parameters suggested by closer examination of the wh-data also brings with it various complications which should not be overlooked. There is, for example, the matter of how many (sub-)parameters we will ultimately need to postulate, i.e., how many innately specified choice-points UG will need to contain. Related to this is the question of whether each choice-point must be associated with its own cue (see Section 3.2 above and Section 7 below) and, if so, how one can ensure that the cues are “noticed” in the correct sequence by the language acquirer. Specific acquisitional concerns aside, there is also more generally a clear sense in which an articulated Wh-movement parameter is at odds with the original GB notion of parameters as innate entities whose setting “gives” the acquirer a range of language-specific information – some of it “deep” in the sense of not necessarily being superficially evident from the input – “for free”. This is once again a matter to which we will return in Section 7. Here we simply wish to highlight the fact that closer consideration of GB parameters like the Wh-Parameter has led to the discovery that initially proposed parameters do in some cases hold up under closer scrutiny, but that this is not always the case and, furthermore, that it sometimes appears to be necessary to adopt a more articulated view of the structure of a parameter. To conclude, let us briefly consider the relation between GB principles and parameters, and also the fate of principles proposed during the GB era. Whereas principles and parameters are presently often thought of as distinct and separate components of UG – cf. the Wh- or Head-movement parameters discussed above, which are not typically viewed as choice-points (directly) connected to any particular UG principle – the earliest conception of “parameter” entailed a close connection between the two entities: parameters were necessarily “parameterised principles” (cf. Newmeyer 2005a). Consider (19–20) in this connection: (19) A parameterised principle in the phrase structure module: a. Principle: X’ → X; Complement (where; signifies an unordered pair) b. Parameter: the Head Parameter (HP; cf. (6) above) Heads X precede/follow their complements
(20) A parameterised principle in the bounding module:
a. Principle: Subjacency In a structure α … [β … [γ … δ…] …] …, movement cannot apply if β and γ both belong to the class of bounding nodes.
The limits of syntactic variation
b. Parameter: the class of bounding nodes = DP and/or TP and/or CP (where bounding nodes in English = DP and TP, in Italian = DP and CP and in Russian = DP, TP and CP)
As shown above, phrase structure and movement are both subject to inviolable principles (a), but the precise manner in which these principles manifest in different languages is subject to specific points of parametric variation (b). As GB parameters do not all appear to be so directly connected to principles, it seems necessary to acknowledge that the parameters assumed during the GB era were not of a single general type: at the very least, one needs to distinguish parameters associated with principles (“parameterised principles”) from parameters whose choice-points are less directly connected to specific principles, a state of affairs which immediately raises questions about the relation between these two, the “locus” of the latter, and so on. We return to these questions in Section 3.4. The preceding discussion has shown that principles may be directly associated with parametric choice-points. This need, of course, not be the case – consider, for example, the Projection Principle and the Theta Criterion (cf. Section 3.2 above), neither of which entails a (b)-component of the type illustrated in (19–20). Subjacency, too, was initially thought to be an invariant principle (cf. Chomsky 1973), but Rizzi’s (1982) work revealed this assumption to be incorrect. Similarly, work on the wide range of languages that Chomskyans have investigated since the early P&P days has shown that certain further properties initially ascribed universally invariant status also cannot be regarded as such. Let us consider this matter on the basis of the Extended Projection Principle (EPP) in terms of which every clause must have a subject located in Spec-IP (i.e., the canonical subject position; cf. Chomsky 1981, 1982). As is well known, this principle facilitates a unified account of subject and expletive distribution in modern English, but necessitates sometimes quite complex assumptions in the context of the many languages that do not exhibit the same patterns of subject behaviour. Thus, for example, in languages which set the NSP to its positive setting, two assumptions need to be made, namely (a) that these languages feature a particular type of I and (b) that they include in their lexical stock an element, pro, which is able to satisfy the EPP-imposed requirement that Spec-IP be projected and filled (henceforth: the EPP-requirement). In the absence of an EPPrequirement, just the (a)-assumption would be sufficient, and numerous researchers have in fact proposed that it may not be necessary to project and fill Spec-IP in all NSLs (cf. Borer 1986; Barbosa 1995, to appear, Alexiadou & Anagnostopoulou 1998 on so-called “I-subject” approaches to NSLs, and see also the discussion in Section 5 below).15 In other NSLs, postulating pro and therefore also the projection . The “I-subjects” approach necessarily has various consequences for Theta Theory, which
Introduction
of Spec-IP may nevertheless be justified (cf. Sheehan 2006 and Barbosa, to appear for recent discussion and see Section 5 below for further discussion). In the Germanic context too, there are a range of subject-related phenomena which cast doubt on the universal operation of an EPP-requirement. Consider, for example, the fact that subjects in Icelandic, German and Dutch may surface in a position which is clearly lower than the V2-adjacent canonical subject position (Spec-IP). The Icelandic examples in (21) illustrate (subjects in bold; Spec-IP underlined; % indicates that the structure in question is ill-formed in the context in question):16 (21) a.
Við settum allar bækurnar fyrir málstofuna í lestrarsalinn … We put all books-the for seminar-the in reading-room-the
i. … en það hafa margar af þessum bókum verið lesnar áður … but there have many of these books been read before ‘… but many of these books have been read before’ ii. %… en það hafa ___ verið lesnar margar af þessum bókum áður … but there have been read many of these books before b. Við settum færeyskar bækur fyrir málstofuna í lestrarsalinn … We put Faroese books for seminar-the in reading-room-the i. %… en það höfðu íslenskar bækur verið lesnar í fyrra … but there had Icelandic books been read last year ii. … en það höfðu ___ verið lesnar íslenskar bækur í fyrra … but there had been read Icelandic books last year ‘… but Icelandic books had been read last year’ (from Jonas 1996: 63)
As shown above, it is not always necessary for Spec-IP to be filled: where a subject constitutes new information, as in (21bii), it does not undergo raising; it is only where the subject encodes old information, as in (21ai), that raising to Spec-TP is required. Similar effects emerge in German and Dutch, where “low” and raised (i.e., Spec-TP-occupying) subjects are also said to differ in respect of informationstructure-related properties (e.g., specificity; cf. Diesing 1992 and in particular Rosengren 2002). If one is to maintain that these languages have an EPP-requirement, it is necessary to postulate the presence of expletive pro in the case of the structures
we will not go into here, but see Baker (1996, 1997) and Alexiadou & Anagnostopoulou (1998) for discussion. . Note that Icelandic is a symmetric V2 language, exhibiting V2 orders in matrix and embedded contexts. Following Schwartz & Vikner (1996) and many others, I assume that all V2 clauses are CPs in which the finite verb is located in C.
The limits of syntactic variation
featuring “low” subjects. Given the presence of an overt expletive in cases like (21bii), it is, however, not obvious that this is a reasonable postulation, particularly if LF considerations are borne in mind. Further, it is striking that raising vs. nonraising of the subject is necessarily associated with specific interpretive effects, a state of affairs which suggests an optional rather than an obligatory operation (cf. Chomsky 2001: 34 on the so-called Fox-Reinhart intuition on optionality, and also Biberauer & Richards 2006; Biberauer 2004, 2006; Richards & Biberauer 2005 for detailed discussion of the status of the EPP in Germanic). In at least some Germanic varieties, then, it is not so clear that the EPP as originally formulated holds. Looking beyond languages that have traditionally been labelled subjectprominent (cf. Li & Thompson 1976), it is not so clear that the subject-centric EPP accurately captures the grammatical focus of so-called topic-prominent languages (cf. Modesto, this volume for overview discussion). More recently, it has also been proposed that it may be productive to think of languages as being either D-prominent or V-prominent (cf. Davies & Dubinsky 2002), with the latter privileging a verbal rather than a nominal category. On this view, it could then be the case that I – or, following Pollock (1989) and others, the relevant I-related head in a more articulated structure17 – always requires a specifier, but it need not be the case that this specifier is always reserved for the subject or expletive. As with (the relevant subset of) topic-prominent languages, the absence of expletives in V-prominent languages could therefore be viewed as a consequence of the non-pivotal nature of subject/nominal categories in these systems, rather than as an indication of their semi-null-subject nature (cf. Section 5 below). Finally, researchers working on a wide spectrum of languages ranging from Germanic to Austronesian have also considered the possibility that what one might think of as D- and V-prominent languages may in fact differ in respect of the “size” of the category that undergoes raising to Spec-TP (cf. various contributions in Carnie & Guilfoyle 2000 and Carnie, Dooley & Harley 2005, and also Richards &
. It is worth noting that the genuinely universal status of the EPP becomes extremely difficult to maintain in the context of so-called “cartographic” approaches to functional structure within which I may be decomposed into several projections. What analyses couched in these terms have tended to highlight is that languages in fact appear to have available to them a range of IP-internal subject positions (cf. Bobaljik & Jonas 1996; Cardinaletti 1997, 2004; in some cases, subject positions that have been proposed may in fact be located in the lower CP-domain, but we leave this matter aside here). If this is correct, the implication is that one cannot think of a single, EPP-imposed, obligatorily filled subject position in these languages. To “save” the EPP, one could potentially appeal to a “spreading” device of the kind mooted in Chomsky (2008) and in terms of which the phase-head C “spreads” its EPP-feature to one and only one of the I-heads in its domain. In pre-2008 Oh Phases terms, though, it is not clear how this would be achieved.
Introduction
Biberauer 2005 and Travis 2006). In terms of this type of approach, then, the idea is that Spec-TP must be projected and filled, but that is not specifically filled exclusively by a subject; even where a language is D-prominent, it may permit or even require a larger category containing the subject to raise to Spec-TP. Biberauer (2004) argues this to be the case in Afrikaans, where the subject standardly piedpipes its containing vP to Spec-TP, thereby delivering Afrikaans’s characteristic SOVAux ordering, and Biberauer & Roberts (2005, 2008) make a similar case for earlier English. Among the advantages of proposals of this sort is that they can account not just for unexpected subject-related phenomena – e.g., the absence in certain structures of an overtly realised expletive where a language does not generally license expletive pro (expletive pro is not expected to be present if Spec-TP is occupied by a larger category) – but also for unrelated phenomena such as word order and word-order variation. Given the above, then, it would seem that there are a range of empirical and theoretical reasons why it is not feasible to maintain a universally instantiated privileged subject position. One could, of course, maintain the projection-of-Spec-IP component of the EPP independently of its subject connection, but this represents a departure from the principle as originally formulated. What we have seen in this section, therefore, is that a number of prominently cited GB principles and parameters do not appear to be empirically supported in the manner originally predicted. As we will see below, this has led some researchers to reject the usefulness of the parametric enterprise as a whole and to question the extent to which it is in fact feasible to identify universally instantiated limits on variation. In the following section, we will, however, consider an alternative perspective on this matter, one which was, in part, necessarily brought about by the shift to the current phase of the P&P enterprise, namely the Minimalist Program. 3.4 Principles and Parameters in the Minimalist Program The Minimalist Program (MP) has its origins in the theoretical “stock take” that Chomsky & Lasnik initiated at the start of the 1990s, after roughly 10 years of work within the P&P model. Rather than rejecting GB and its predecessors and “starting afresh”, as critics sometimes suggest, it aims to critically evaluate the significant body of theoretical work produced previously and to subject it to the kind of Occam’s Razor assessment that characterises the normal course of scientific research more generally. Additionally, however, the MP, at least as pursued by researchers with a “biolinguistic” commitment (cf. Chomsky 2005, 2007, 2008 for discussion and references), also has at its core a new hypothesis, namely the Strong Minimalist Thesis (SMT) given in (22): (22) The Strong Minimalist Thesis (SMT) Language is an optimal solution to legibility conditions (Chomsky 2000: 97)
The limits of syntactic variation
In terms of the SMT, the starting point for any MP model of the language faculty is that its design is such that it satisfies the requirements that it clearly must meet18 in respect of feeding linguistic information to and from the systems with which it interfaces – the conceptual-intentional (“meaning”) system and the articulatory-perceptual (“sound”) systems – in the most efficient manner possible. The assumption is not that this hypothesis, which in effect postulates a maximally “well-designed” or “perfect” language faculty, is likely to be correct; instead, Chomsky’s rationale is that research within the P&P programme had, by the early 1990s, advanced to the point where enough was understood for a hypothesis like the SMT to serve the kind of heuristic function that is likely to drive the UG enterprise forward (cf. Chomsky 2005, 2007 for discussion). Instead of simply continuing to add to the machinery already proposed in the haphazard way that often becomes inevitable in the context of a widely adopted research enterprise, the suggestion, then, was that the SMT should serve as a yardstick to determine whether particular theoretical notions can and should be maintained or not: the SMT requires that each theoretical assumption be subjected to the question of whether a principled explanation can be found for it, with principled being understood as meaning “imposed by one or both interfaces” and/or “following from general considerations of computational efficiency” (cf. Chomsky 2000, 2001, 2004a, 2004b, 2005, 2007, 2008). A direct consequence of the adoption of the SMT is, of course, the need to minimise the number of purely language-internal entities one postulates. On the architectural front, this led to the elimination of syntax-internal levels (D- and Sstructure) and also, crucially for our purposes, of the modules associated with the GB model of the grammar (the Phrase-Structure, Theta, Case, Bounding, Binding and Control modules). As noted in Section 3.3, at least some GB parameters were assumed to be associated with specific modules of the grammar. Thus the HP, for example, was clearly a phrase-structure module-related parameter, while the Whmovement parameter was associated with the Bounding module. In the absence of the modules with which they were associated, it is clear that these parameters must be “located” elsewhere if they are to be retained within an MP grammar. In principle, the simplified MP architecture, comprising just a Lexicon, the Computational System (which “does” Narrow Syntax) and (possibly) two interface components, PF and LF, allows for two general possibilities: either parameters are located in the Lexicon or they are located at one of the interfaces. The notion that parameters might in fact be associated with lexical items is a long-standing one, first proposed during the GB era as the Lexical Parametrisation
. Chomsky (1995 et seq.) refers to these as “virtually conceptually necessary” requirements.
Introduction
Hypothesis (cf. Borer 1984; cf. also Manzini & Wexler 1987 for further seminal discussion). As noted by Baker (this volume: 253), the idea that much, if not all, parametric variation is located in the Lexicon constitutes a core assumption in minimalist theorising. Baker (ibid) labels it the Borer-Chomsky Conjecture (BCC) and states it as follows: (23) The Borer-Chomsky Conjecture (BCC) All parameters of variation are attributable to differences in the features of particular items (e.g., the functional heads) in the lexicon.
The appeal of the BCC is clear: if lexical items constitute the sole locus of parametric variation, acquiring these will, in effect, amount to acquiring the language as a whole. In other words, on this view, “learning vocabulary” does not just entail learning the idiosyncratic component of language (i.e., that traditionally associated with Saussurean arbitrariness) and, presumably at the same time, also what the phonotactic and morphological properties of the language in question are (i.e., two types of system-defining knowledge); it also entails acquiring the parametric profile of the language, fixing the unvalued parameters of S0 (cf. (5) above). Reconsidering the parameters discussed above, it is clear that a number of them are plausible candidates for reformulation in terms of the BCC. The NSP in (7), for example, in effect specifies two lexical requirements, namely that a language must have both a particular type of functional category (“richly” specified I) and a particular type of pronominal (pro). The HP in (6), by contrast, initially looks like a far less likely candidate, but this parameter is also amenable to reformulation in lexical terms and numerous researchers have in recent years done so, either explicitly or tacitly. Thus one might, for example, assume Kayne’s (1994) Antisymmetry proposal in terms of which phrase structure is exceptionlessly head-initial and propose that surface head-final orderings are the consequence of leftward movement of complements. In practice, what this entails is that the heads which surface head-finally bear a movementtriggering feature (an EPP-feature in current minimalist parlance), whereas those which surface initially do not. On this model, then, at least all the functional heads in harmonic head-final languages, and possibly heads generally, are lexically specified as bearing movement-triggering EPP-features (precisely what determines the “size” of the moved constituent – i.e., whether we see DP-movement or “massive pied-piping” of the sort proposed in so-called roll-up analyses – seems to require independent specification; cf. discussion below).19 By contrast, the (functional) . Note that, in the context of a theory that assumes the distribution of EPP-features to constitute a central point of crosslinguistic variation (cf. also GGL this volume), with functional categories (and possibly lexical items) either bearing or not bearing this movement diacritic, the
The limits of syntactic variation
heads in harmonic head-initial languages systematically lack EPP-features (except where these are optionally present in order to guarantee a non-neutral, movement-derived structure20). In disharmonic languages like German and Gungbe, in which phrases do not all exhibit the same head-directionality, the same system would obtain, with the result that the Lexicon in the case of languages of this type would contain both EPP-bearing (head-final) and EPP-lacking (head-initial) heads. Finally, in languages in which linearisation appears to play a lexical disambiguation role (cf. the examples mentioned in Section 3.2), EPP-features would only be associated with the specific lexical items exhibiting final ordering.21 This last case suggests that not just functional categories, but individual lexical items may bear independent linearisation properties. In the context of the BCC, this does not sound problematic, but it is worth considering the implications of this possibility. One very obvious implication is that it would seem to entail massive redundancy in the context of all languages in which the HP is not operating “myopically” (i.e., lexical item-sensitively; cf. Section 3.3 above) – that is, in all languages where head-directionality is more systematically determined on the basis of either syntactic category membership (the German/Gungbe case) or in an across-theboard fashion (the harmonic cases). In other words, ascribing head-directionality specifications (via diacritics or otherwise) to each and every lexical item would seem, in the vast majority of cases, to be complicating a system which could be significantly simpler if this information were located elsewhere (cf. also Baker, this volume, for the same argument). As such, this kind of proposal would not seem to be in keeping with the SMT in (22).22
GB Extended Projection Principle (discussed in Section 3.3) emerges as a stipulation that would require principled explanation: why would it be necessary for T (=I) in every language to bear an EPP-feature, whereas all other functional categories either may or may not do so? . Cf. Biberauer & Richards (2006) on the difference between obligatory and optional EPPfeatures and the interpretive effects they bring about, and Biberauer, Holmberg & Roberts (2007) for discussion of the necessary distinction between linearisation-ensuring features and those responsible for A’- and possibly also A-movement. . Precisely how this system would account for heads that may appear both initially and finally is not clear. Possibly, the relevant lexical entries would be underspecified for linearization information or perhaps one could, as suggested in note 10, follow Adger (2006) in proposing that certain types of lexically-related optionality may in fact involve two distinct lexical items with the same phonological and semantic characteristics, but different linearisation specifications. Given how strongly languages in general seem to resist this latter “homonymy” scenario, however (cf. i.a. Kroch 1994), the latter may not be a plausible hypothesis. . Worth noting, however, is an alternative perspective on this matter, in terms of which the “redundancy” entailed by assigning an EPP-feature to all the categories in a given system in fact
Introduction
Another serious problem, also discussed in Baker (this volume), with the notion that linearisation information is stored on a lexical-item-by-lexical-item basis is that it leads us to expect “a relatively smooth continuum of languages” (Baker, this volume: 360), in which there are many mixed (i.e., disharmonic) languages of different kinds and relatively few harmonic languages. In reality, however, the facts seem to be rather different. Consider in this connection Greenberg’s original observations, the obvious skewing in favour of harmonic languages registered in (2) above, and Dryer’s (1992, note 17) observation, which points to “disharmonies” having predictable origins not only within languages, but also crosslinguistically. Looking beyond the domain of the HP, Dryer (1998: 123–124) has observed that empirical work strongly suggests that languages “tend to cluster together around a certain type and that they become increasingly infrequent as we move away from the core type”. As Baker (this volume) notes, facts of this kind raise serious questions about the currently quite generally accepted and largely unexamined idea that parametric variation is necessarily lexically specified. Significantly, linearisation was precisely one of the domains that Fukui (1988) argued should fall outside the domain of lexical parameterisation (cf. his Ordering Restriction Hypothesis). Additionally, he also argued that parametric variation should be restricted to functional categories, with lexical categories like N and V, for example, not playing any role in parametric variation (cf. his Functional Parameterisation Hypothesis). This latter idea is one that is currently widely held, with minimalists frequently interpreting the BCC as applying specifically to functional categories (FCs).23 As the contributions by GGL and Kayne in particular highlight, however, it is not necessary to adopt this approach and the presence versus absence of specific lexical items may in fact also play a central role in determining the more general syntactic options available to a given system. We return to this matter in Section 4 below. emerges as highly economical. If we adopt the perspective of a simplicity-driven acquirer who is trying to establish the directionality of the syntactic categories in their language (i.e., whether they are associated with an EPP-feature or not), one could argue that the simplest systems are those in which the acquirer simply postulates C[+EPP] or C[–EPP], i.e., the same EPP-specification for all categories (C) in the system (see i.a. Clark & Roberts 1993; Roberts & Roussou 2003 and Roberts 2007a for discussion of simplicity-driven acquisition). This does, of course, raise the complex question of “markedness” – specifically, whether “harmonic” systems should not actually be easier to acquire than “disharmonic” ones – but we leave this matter aside here. Thanks to Ian Roberts for discussion of this point . Under the view that lexical categories like N and V do not constitute primitive entities, but that “nominalisation” and “verbalisation” are in fact achieved via merger of acategorial roots with functional categories like n and v (cf. Marantz 1997; Chomsky 2001: 9), it is arguably impossible for lexical categories to play a role in parametric variation independently of functional categories (for a very different view on the noun/verb distinction, see Kayne 2008).
The limits of syntactic variation
For the moment, we conclude, firstly, that the BCC is a hypothesis which requires more critical examination than that to which it has been subjected to date and, secondly, that linearisation in particular may well not be the kind of parametric property that is best located in the Lexicon. Leaving aside the objections raised above, there is also the in recent years oft-discussed question of whether linear ordering should play any role in Narrow Syntax (NS), as it would presumably have to if it is already lexically specified. As has been pointed out many times, the computational system does not appear to be sensitive to linear ordering; only to hierarchical structure. As such, one might expect that linear ordering – even if this is ultimately determined via Kayne’s (1994) Linear Correspondence Axiom (LCA) – is only imposed at the stage at which it is self-evidently required, namely at PF (cf. Chomsky 1995: 335, 340). We pursue this matter below. One aspect of linearisation that is, however, usually thought to lend itself to lexical parameterisation is that which was previously encoded by the various movement parameters – V-to-I movement, N-to-D movement, V2 and Whmovement, to name but a few. In each case, the relevant movement requirement can be lexically encoded as a movement diacritic (EPP-feature, etc.), with languages lacking movement in the relevant domains correspondingly lacking this diacritic.24 There is little doubt that the easy “translateability” of the movement parameters into BCC terms has played a significant role within minimalist research in foregrounding the role of the Lexicon in determining parametric variation. It is, however, worth noting that the matter may be significantly more complicated than “recastings” of this kind might suggest. Consider, firstly, the fact that movement parameters, if they are lexically encoded and implemented in the syntax, once again introduce linear ordering into Narrow Syntax. Furthermore, if Chomsky (2001: 37–38) and others are in fact correct in consigning (at least some) headbut not phrasal movement to the PF component (i.e., to an interface), it is not clear how the computational system can effectively “ignore” the movement diacritics that will ultimately result in head-movement effects, while satisfying those which will ultimately deliver phrasal movement. It is, of course, conceivable that head-movement triggers are invisible (i.e., uninterpretable) to the computational system, while XP-triggers are not and that we are thus effectively dealing with two different types of movement diacritic. Given that certain head-movements do appear to take place in the syntax, however (cf. Lechner 2006 and Roberts 2006 on LF effects associated with head-movement), it would seem that there also needs to
. Alternatively, one might, like Brody (1995), Bobaljik (2002), Landau (2006) and others, think in terms of P-features which determine which copy of a chain is ultimately pronounced.
Introduction
be a Narrow Syntax-internally interpretable head-movement trigger – and, additionally, since syntactic relations always involve head-head relations (cf. Chomsky 2004a et seq.), some means of signalling when head- rather than XP-movement is to take place (cf. Watanabe 2006 on “pied-piper” features). If some head-movement is PF-movement, then, quite an extensive inventory of movement-triggers would seem to be required. If it is not, there are still many currently unanswered questions as to how the computational system is able to determine the “size” of the category it is to move (cf. Biberauer & Richards 2006; Biberauer & Roberts 2006 and Watanabe 2006 for some recent discussion). Let us now turn to the alternative potential loci for Minimalist parameters: the interfaces. Since the possibility of LF parameters has not received much attention to date, we will focus on the PF interface here (cf., however, Ramchand & Svenonius, this volume, for discussion of the role that LF may potentially play in determining parametric variation). As noted above, the PF interface would seem the most plausible locus for linearisation generally and therefore also for parametrically variant aspects of linearization.25 Chomsky himself has always held that “[t]here is no linear order in the N[umeration] → LF computation” (Chomsky 1995: 335; cf. also Chomsky 2001: 7) and that linear ordering be imposed at PF, where the LCA may, but need not necessarily, determine linear precedence. Actually formulating a plausible PF linearisation system, let alone a credible PF version of the HP has, however, proven quite difficult. As regards linearisation more generally, the problem is ensuring that the structures sent to PF are in fact unambiguously linearisable. If one assumes that the very restrictive (Spec-Head-Comp) phrase structure assumed by Kayne (1994) is the only kind permitted in NS (i.e., that a very strict version of the X-bar template is in fact encoded as a UG principle), the problem of course becomes trivial: such structure is always linearisable via the LCA. On this view, phrase structure is, however, strictly constrained precisely to ensure a linearisable output, with a “simple PF” coming at the expense of a more complicated computation (including, for example, unrealised heads and specifiers, and “extra” leftward movement wherever head-final structures result – see discussion above and note 22).26 As such, a Kaynian approach to linearisation must be regarded as a departure from
. For a minimalist analysis which proposes relocating the HP to Narrow Syntax, however, see Saito & Fukui’s (1998) Parameterised Merge proposal. . As a result of the latter, particularly if leftward movements may be of the remnant variety often assumed in Kaynian analyses, there is, of course, also one aspect of the PF “computation” that is not simplified, namely the copy spellout process, i.e., determining which copy/copies to spell out and which to suppress (cf. Corver & Nunes 2007 for discussion).
The limits of syntactic variation
the SMT and alternatives should at least be investigated to determine whether a strict LCA-based system in fact constitutes a genuine case where there is a “gap between SMT and the true nature of FL [the Language Faculty – TB]” (Chomsky 2007: 3). An obvious way to relax the Kaynian constraints on phrase structure is to propose, as many researchers following Moro (2000) have done, that syntactic structure need not be linearisable at every stage of the derivation, but that structures must be linearisable at the stage at which they are sent to PF. Within proposals of this kind, linearisation is therefore only partially a PF matter, with initially unlinearisable structures being “rescuable” via subsequent movement operations which may – as in Moro’s Dynamic Antisymmetry approach – or may not be directly aimed at neutralising troublesome “points of symmetry” (i.e., symmetrical and thus unlinearisable structures). Languages may then differ as to the points of symmetry that are generated during the course of the computation27 and possibly also as to when and how they are neutralised (cf. Hsieh & Sybesma 2007 for a phasebased implementation of DA and Takita 2007 for a proposed symmetry-breaking parameter). Approaches of this type can, however, clearly be criticised for “lookahead”, i.e., NS operating in such a way that subsequent requirements – here, PF legibility – can be met.28 We leave this and other matters aside here since our main concern is to consider how feasible it in fact is to completely banish linearisation information from NS. A very different approach is that of Richards (2004). He proposes parameterising the LCA such that PF systematically ignores one set of symmetric c-command relations (cf. also Epstein et al. 1998), thereby capturing the effects of the HP in the manner outlined for VP below: 29 (24) The Parameterised LCA (a PF HP) a. VO = Ignore all O > V b. OV = Ignore all V > O
. Points of symmetry may in principle be either head-head or XP-XP adjunction pairs. . Additionally, it is worth noting that DA-based approaches once again necessitate the postulation of a range of movement-triggering factors since movement in natural languages is clearly not limited to the symmetry-breaking variety. . Richards’s proposal takes as its point of departure the observation that the first two elements Merged in any structure will be heads, i.e., elements which necessarily symmetrically c-command each other and which will therefore also necessarily be unlinearisable in LCA terms without some additional instruction. The Parameterised LCA is intended to supply this instruction and, simultaneously, to state a point of parametric variation. For a very different solution to the head-head merger problem, see Kayne (2008).
Introduction
In terms of (24), then, head-initial languages consistently ignore PF linearisation instructions (i.e., symmetric c-command structures) ordering complements before heads (a), whereas head-final languages consistently ignore the opposite set of linearisation instructions. This system does therefore remove HP-related linearisation information from NS, allowing the HP to be stated as an interface parameter. As with the original HP, however, it is not so clear how an interface HP operates in the context of disharmonic languages, i.e., those which do not set (24) so that it applies across-the-board. One might propose that (24) is sensitive to syntacticcategory information and that lexically specified linearisation instructions can override its across-the-board application (cf. note 10), or possibly that the application of (24) is determined on a phase-by-phase basis, with some languages opting for a subset of phases which are linearised in accordance with (a), while the remainder are linearised in accordance with (b). The implications of these possibilities remain to be worked out in further detail, with linearisation patterns that systematically fail to surface in the world’s languages (i.e., impossible linearisations) potentially representing a crucial empirical testbed (cf. Biberauer, Holmberg & Roberts 2007 for discussion, and also Kural 2005 and Cinque 2005, 2008). For the moment, however, what is worth noting about the HP and its potential MP reformulations is that an exclusively lexically based approach would seem to entail considerable redundancy for harmonic languages in particular, effectively making the most idiosyncratic of the disharmonic exceptions seem the most “natural”, whereas a purely PF-oriented approach seems best suited to the harmonic languages while raising questions as to how the disharmonic orders are to be accounted for. If UG is to contain a linearisation parameter, it clearly needs to be one which could in principle be set so as to generate both the harmonic and the disharmonic word orders and it is not presently clear either how this can be achieved or whether (a) suitable linearisation parameter(s) will require a departure from the SMT. We have already noted above that movement parameters would, at first sight, seem trivially recodable as lexical parameters involving the presence versus absence of movement diacritics. Another possibility is that these parameters may in fact be located at PF. Bobaljik (2002) offers a proposal along these lines, suggesting that the difference between overt and covert movement may in fact not lie in whether an Agree operation between two elements is or is not associated with a movement diacritic (presence of EPP-feature = overt movement; absence of EPPfeature = covert movement), but in which copy of a moved element PF chooses to spell out. In Bobaljik’s terms, then, overt movement entails spelling out the highest copy of a moved element whereas covert movement falls out from spelling out the lowest copy; movement therefore always takes place in NS, but this plays no role in linearisation (cf. Brody 2000 for an in many ways similar approach focused on a different empirical domain). This proposal is particularly appealing in the context
The limits of syntactic variation
of covert movements that were during the GB era assumed to take place in the derivation from S-structure to LF (QR, wh-movement in wh-in situ languages, etc.). If Bobaljik’s proposals are on the right track, it may therefore be the case that GB parameters like the Wh-movement Parameter have both a lexical and an interface component (cf. the oft-discussed correlations between (non-)movement and, inter alia, the availability of what Cheng 1997 designates clause-typing particles and the morphosyntactic structure of wh-elements). To conclude on the question of how GB parameters may be reformulated in the current framework: we have seen that, while the available loci would seem to be clear (the Lexicon and one or more of the interfaces), GB parameters are not equally easily captured in minimalist terms. This may, of course, indicate various things. One possibility – and, given the course of normal scientific progress, it is likely to be a very strong possibility – is that some or all of the GB parameters were misfocused and that they should therefore be viewed, in the context of the P&P enterprise, as “first pass” attempts at pinpointing recurring points of syntactic variation (see Section 3.5 below). Another possibility, which arose in the course of our discussion of the Wh-movement Parameter, is that parameters may in fact be the reflex of potentially quite complex interactions between various aspects of language structure (lexical, morphological and syntactic), with the result that formulating and “localising” them might involve consideration of a greater number of factors than has traditionally been done (but see Haspelmath, this volume, for a different perspective on the likelihood of cross-domain relations). At the most general level, though, what the minimalist heuristic strongly suggests about some of the GB parameters is the fact that they may not ultimately be formulable in a manner which respects the SMT. Let us consider this point in a little more detail. As Chomsky (2007: 1) notes, considerations of language acquisition had, until the advent of the P&P programme, always led Chomskyans to assume that UG must be “rich, highly structured and substantially unique”: if the acquirer was to be able to abduce (i.e., correctly infer on the basis of innately specified knowledge, cf. Peirce 1955) the grammar underlying the PLD to which they are exposed (cf. (5) above), a suitably specific UG seemed a prerequisite. The P&P programme led to the revision of this notion since it opened up the possibility of an underlying system in which parameter settings might differ only slightly, while the surface effects of those slight differences might be extensive (consider, for example, the considerable surface difference between consistently head-initial and consistently head-final languages, where the relevant underlying difference, in GB terms, is simply the way in which the HP is set). Within the MP, consideration of the role of a “third factor”, whose potential relevance had previously only been mentioned in passing, but which was never explicitly included in the acquisitional scenario alongside UG (innate endowment) and PLD (experience), has become central.
Introduction
This “third factor” entails principles of mental/neural organisation that are not specific to the faculty of language, but are common to all cognitive faculties (cf. Chomsky 2005, 2007, 2008, Hauser, Chomsky & Fitch 2002 and the ensuing debate for further discussion). Furthermore, in terms of the hypothesis driving minimalist research – the SMT in (22) – the null hypothesis is that UG, or the specifically linguistic information attributed to the human genome, is minimal. In other words, minimalist research, unlike that which went before, explicitly proceeds on the assumption that, in the absence of evidence to the contrary, as little as possible should be attributed to UG – specifically, only those properties which do not appear to be reducible to more general, possibly partially functionally motivated principles. The question is, as Chomsky (2005: 11) puts it: “How far can we progress in showing that all … language-specific technology is reducible to principled explanation, thus isolating core processes that are essential to the language faculty?”. In the context of SMT-oriented theorising, then, a rich parametric inventory of the type assumed during the GB era would constitute a considerable departure from the SMT and, as such, should only be assumed if the evidence points to the need for it. In the first instance, however, the goal should, as in minimalist theorising more generally, be to consider how much of what was previously proposed can be accounted for by appealing to “non-domain-specific” principles (e.g., those governing efficient computation). The greater the number of language-specific parameters that ultimately needs to be proposed, the more extensive the departure from the SMT will be. As we have seen above, first attempts suggest that it is possible to “locate” some GB parameters in the Lexicon and also that it might be possible to view others as the consequence of particular options which arise at the interfaces. As such, these parameters would represent content that needs to be ascribed specifically to UG, but at least one aspect of their nature – specifically, the fact that they can be formulated in lexical and/or interface terms – is such that one could offer a principled answer to the question: “Why those parameters and not others that one could possibly formulate?”. Answer: the parameters in question are directly associated with the “virtually conceptually necessary” components of the language faculty and do not entail postulating additional modules or otherwise complicating the overall architecture of the language faculty. In the context of the current phase of the P&P programme, then, the question is not just what sorts of innately specified knowledge might facilitate language acquisition in the manner that it occurs, but additionally also whether knowledge that seems to require innate prespecification can in fact be seen to take the form it does by virtue of this having been “imposed” by considerations of “good design”. Insofar as the content of UG and the design of the language faculty more generally can be understood
The limits of syntactic variation
as having the form that they do in virtue of “third factor” considerations, it is, in Chomsky’s terms, possible to go “beyond explanatory adequacy”; and the MP aims to probe the extent to which it is possible to do so, approaching the question both “from above” (i.e., continuing the search for explanatory adequacy) and “from below” (i.e., attempting to discover the extent to which “third factor” considerations have shaped UG). So far, we have speculated that GB parameters which can be viewed as being “located” in one of the “virtually conceptually necessary” components of the language faculty might be thought of as meeting a “good design” requirement, namely that they do not require the postulation of additional, purely language facultyinternal architecture. A further “good design” question that the SMT forces us to ask is whether it is in fact necessary to view each of the specific parameters that were previously identified as independent entities within UG. Consider, for example, the individual head-raising parameters (e.g., V-to-T, N-to-D, V-to-C, etc.): these clearly share the same format – X raises to Y – and the question the SMT raises is whether UG actually requires each of these parameters to be specified individually. One might, for example, alternatively think of a general template which serves, in much the same way as the X-bar template during the GB era, to determine aspects of syntactic structure (cf. GGL, this volume); or, more radically, that even the template may arise as a result of the interplay between PLD and “third factor” considerations (cf. Zeijlstra and Evers & van Kampen, this volume and also Roberts 2007d for different versions of this idea). Evidently, then, there is a very general question within the MP as to the forms that minimalist parameters might take. A further question concerns their status within UG: are they present in “unset” form at S0 or can they arise during the process of acquisition? These are questions which have not featured prominently on the MP research agenda to date, but which clearly need to be addressed if we are to be able to offer a coherent and up-to-date minimalist answer to the question of how principles and parameters constrain syntactic variation. In the absence of a coherent minimalist theory of this type, it is not in fact possible to meaningfully evaluate the success of the P&P enterprise to date. The remainder of Section 3 focuses on a recent evaluation by Newmeyer (2005a, 2005b and 2006) which starkly underlines the need for more systematic work on notions like “minimalist parameter”. 3.5 Recent criticism of the parametric enterprise Newmeyer (2005b: 185) states seven desiderata that a parametric approach to syntactic variation should meet if it is to be superior to a potential rule-based alternative. These include descriptive simplicity, binarity, smallness of number, implicational organisation and a rich deductive structure. He then considers the
Introduction
extent to which GB parameters, notably the NSP, satisfy these desiderata, ultimately concluding that they fall short in various ways, and that the limits of syntactic variation should be sought in “the interplay between unparameterized UG principles, language-particular rules, and processing principles sensitive to structural relations holding among grammatical elements” (2005b: 225). In other words, Newmeyer proposes that parameters should play no role in our model of UG and that the parametric enterprise can be viewed as having failed. It is, however, important to note that his conclusions are based exclusively on consideration of GB parameters.30 As such, they can at most be interpreted as potential conclusions as to the feasibility of the “first pass” parameters of Stage I of the P&P enterprise, and not as conclusions on the enterprise as a whole; as was discussed in Section 3.4, minimalist parameters will necessarily differ from their GB predecessors in various ways and work on precisely what the inventory of minimalist parameters might ultimately look like has scarcely begun. Many of the contributions in this volume are centrally concerned with this question, which was also a major focus of the conference at which they were first presented. What we will briefly consider here is the extent to which some of Newmeyer’s criticisms can also be levelled against the parametric enterprise as it can now be and is now being approached. Our discussion here will focus on the descriptive simplicity, richness of deductive structure and smallness of number desiderata (cf. discussion of implicational organisation in Section 4 below, and Roberts & Holmberg 2005 for discussion of the other desiderata). Firstly, let us briefly consider the matter of descriptive simplicity and the related matter of smallness of number. As discussed in Section 3.4 above, the hypothesis driving minimalist research (the SMT in (22)) takes as its point of departure that UG does not contain domain-specific information. As we have seen, it may well be possible that it need not contain specific parameters of the type assumed during the GB era and that any parametric options that are innately specified may take the form of extremely general and thus also descriptively extremely simple templates (cf. GGL, Zeijlstra and Evers & van Kampen for suggestions as to the format that such templates might take). If this thinking proves to be correct, it may then well be the case that a very small number of (or possibly even no) innately specified templates gives rise to a diverse range of steady-state (Ss) parameters. This line of thinking has an important consequence in the context of the point Newmeyer (2005b: 185) raises regarding the complexity of the variation observed in the world’s languages. According to Newmeyer, this is such that it simply is not
. Cf. Roberts & Holmberg (2005) for a critical evaluation of some of the conclusions Newmeyer draws.
The limits of syntactic variation
feasible to postulate only a small inventory of innately specified parametric options, with the result that one might as well assume that children do not draw on innately specified parametric choices, but instead learn language-specific rules which capture the relevant dimensions of variation. If it can, however, be shown that it is possible for the Ss parameters to be “constructed” during the course of acquisition, that these do not have to be individually prespecified and that their effects might nevertheless be wide-ranging, Newmeyer’s conclusion no longer follows. Let us turn now to the matter of the “rich deductive structure” traditionally ascribed to parameters. As we have seen in Section 3.4, lexically specified parameters associated with specific lexical items and/or functional categories have played a prominent role in minimalist theorising to date. At first sight, these might seem to be parameters which may not always meet the “rich deductive structure” desideratum since their effects are naturally limited to the domains in which the relevant lexical item is present, with the result that lexically specified parameters are often (erroneously – see the discussion of possible encodings of the HP above, and below) equated with microparameters, i.e., parameters that define the kinds of small-scale differences in terms of which closely related dialects may differ (cf. Kayne 2005b for discussion). Newmeyer (2005b, 2006), like i.a. Safir (1987: 78) and Bouchard (2003) before him, is of the view that parameters of this type entirely undermine the rationale for parameters, essentially reducing the term parameter to a synonym for rule. In the words of Newmeyer (2006: 9), “if every difference or even most differences are attributed to a difference in “parameter setting”, then saying that two languages have different parameter settings merely becomes an obfuscatory way of saying that they have different rules”. For Newmeyer, then, only parameters of the type Baker (this volume) designates macro- and medioparameters have a sufficiently rich deductive structure to justify speaking of “parameters” as opposed to “rules”. Like Pica (2001), however, he believes that the past quarter century of research within the parametric framework has shown that parameters of this kind are not empirically motivated, leaving us, at best, with lexically specified microparameters, which, for him, are in fact most realistically viewed as language-specific rules. For Newmeyer, then, one of the central failings of the parametric enterprise is its failure to bring to light plausible, empirically substantiated parameters which have wide-ranging and at first sight seemingly unrelated effects. Worth noting about Newmeyer’s conclusion are the assumptions on which it is based. Firstly, it takes for granted that lexically specified (micro) parameters necessarily have very limited surface effects of the kind that can be very straightforwardly captured in generalised rule format. That microparameters could also entail “deep” effects, with the presence of a lexical item or particular feature specification, for example, determining the (un)availability and/or the form of further, often seemingly quite unrelated structures in the language is not entertained as a
Introduction
possibility. As Kayne’s contribution to this volume and his and other researchers’ microparametric work more generally has strikingly shown, however, points of microparametric variation do not have to differ from medio- or macroparametric ones in this regard (cf. also Rizzi 2004: 332 on this point). Furthermore, as Baker (this volume: 355) points out, “[i]t is perfectly possible that a lexical parameter … could have a substantial impact on the language generated, particularly if it concerned some very prominent item (such as the finite Tense node)”. And, conversely, a macroparameter like the Polysynthesis Parameter might only have a very limited impact on a language lacking extensive agreement (cf. also Baker 1996: note 35). Kayne (2005b: 5–6) likewise registers the opacity surrounding the notion of parametric “size”, pointing out that “what seems “dramatic” [in terms of number of empirical reflexes or range of effects – TB] depends on expectations that may themselves be somewhat arbitrary”. Given all of this, it seems fair to say that it is not at all clear that the discovery and existence of “large” parameters, assuming the meaningfulness of a notion of this type, is crucial to the success of the parametric enterprise. Furthermore, given the potentially extensive effects of “smaller” parameters, it is also cannot be said to follow that the possible non-existence of “large” parameters entails that one might as well conceptualise crosslinguistic variation in terms of language-specific rules. Another point worth noting in this connection is that Newmeyer’s conclusions rest on the assumption that macroparametric research is no longer being seriously entertained by researchers engaged in the parametric enterprise. As Baker himself indicates in his contribution to this volume, it is certainly true that genuinely macroparametric research is not receiving nearly the attention that Kayne-style microparametric research is. This is, however, different from saying that researchers are no longer engaged in this type of research at all, and various contributions to this volume show that this is not in fact the case – cf. Modesto on a potential Topic Parameter, Zeijlstra on a Negative Concord Parameter, Öztürk on a Non-configurationality Parameter and Avram & Coene on a Split INFL Parameter which, unlike Bobaljik & Thráinsson’s (1998) original, is hypothesized to have large-scale effects both in the clausal and the nominal domain. And beyond the articles collected here, Huang’s (2006) work on the synchronic and diachronic consequences of a so-called Analyticity Parameter, in some senses the diametric opposite of Baker’s (1996) Polysynthesis Parameter, constitutes further evidence that this line of research is also receiving attention. Further, this volume also contains papers which consider parameters that might, in Baker’s terms, be thought of a medioparameters – cf. Lekakou’s Aspect Parameter, Fernandez-Salgueiro’s Case-F Valuation Parameter and Neeleman & Szendröi’s Radical Pro-drop Parameter. In sum, then, the contributions in this volume do not support the view that present-day parametric research is exclusively microparametric or that “larger”
The limits of syntactic variation
parameters are no longer on the research agenda. Furthermore, as already noted above and as is also specifically highlighted in Baker’s paper, there is, quite independently of the amount of research being done on non-microparametric syntax, undeniable empirical evidence – uncovered by both typologists and generativists – that the world’s languages robustly exhibit skewings of various kinds; and, moreover, skewings that often cannot be economically stated in rule form (cf. the discussion in Section 3.4 of lexical reformulations of the HP). These empirical facts clearly require explanation, and it is not obvious how a purely (language-specific) rule-based account of the kind Newmeyer espouses gets us any closer to understanding them. To conclude, then: consideration of just three of Newmeyer’s desiderata in relation to the current stage of the parametric enterprise would seem to lead to very different conclusions as to its potential value to those drawn by Newmeyer himself. There is no denying the fact that core questions relating to the number, format, size, interrelationship and, in fact, given Chomsky’s “three factors” speculations, the very existence of innately prespecified parameters do not have clearcut answers at this stage of the P&P enterprise. These are matters which need to be at the forefront of the MP research agenda to a much greater extent than they currently are if a meaningful evaluation of the parametric aspects of the P&P enterprise as it stands now is to become a realistic prospect. That the GB parameters which are the focus of Newmeyer’s evaluation do not constitute plausible innately specified constraints on syntactic variation is beyond dispute; what remains to be seen now is whether or not the same will be true of the current phase of the P&P programme. 3.6 Conclusion This section has shown that the factors determining the limits of syntactic variation are in many respects still poorly understood. Accordingly, questions as fundamental as whether innately specified knowledge plays a role or not and if so, what form such knowledge should take are still under debate. Against this background, what the present volume aims to do, then, is to give a flavour of current thinking on the nature and value of parameters and parametric theory in accounting for syntactic variation. The remainder of this introduction will now focus more specifically on the contents of its sixteen papers and the issues that they address. The volume is organised into four sections which respectively focus on the following topics: • the locus of (parametric) variation (cf. Section 4) • the NSP (cf. Section 5)
Introduction
• parametric clustering (cf. Section 6); and • the acquisition of parameters (cf. Section 7)
4. The locus of parametric variation This section considers the under-discussed question of “where” in terms of the overall architecture of the grammar we might expect points of syntactic variation to be specified. It opens with a paper by Haspelmath which aims to fill a gap in the currently existing literature on (the limits of) syntactic variation by providing (i) a systematic comparison of the different ways in which parametrically and functionally oriented linguists traditionally approach the limits question and (ii) a critical evaluation of the success the two approaches have achieved. Haspelmath’s main contention is that the recurrent patterns in the world’s languages – so-called phenomenological universals in his (2004) terminology – require an observationally adequate account, i.e., one that distinguishes possible from impossible expressions in a language (p. 2) and that does not make any claims about speakers’ knowledge; attempts at achieving any “higher” level of adequacy and, particularly, at trying to gain insight into the workings of the human mind via the study of syntactic limits are not, in Haspelmath’s view, feasible nor, given the availability of functional and processing explanations, even necessary. The proposal is that languages should (contra Meillet 1903: 407) not be viewed as “systems in which everything hangs together”. Consequently, so-called crossdomain implications where, for example, semantic organisation correlates with aspects of phonological organisation, should be expected to be rather rare, whereas so-called intra-domain implications where, for example, one aspect of phonology correlates with one or more other aspects of phonological organisation, need not be so. In keeping with these expectations, typological research has revealed a great many intra-domain implications, many of which are, according to Haspelmath, amenable to a functional explanation (cf. pp. 96–100). Cross-domain implications, on the other hand, have proved rare, producing only weak correlations which are in many cases quite difficult to distinguish from areal effects (cf. p. 19). The conclusion to be drawn from this, then, is argued to be that a research programme which seeks to understand apparently unrelated surface phenomena, particularly where these do not belong to the same domain, as the reflexes of a far smaller number of parameters is misfocused: the data do not warrant such a programme and it is doomed to fail in the same way in which previous attempts to propose “holistic” language types have failed (cf. the fate of the 19th century morphological typologies). Instead, the question that should be being asked about recurrent patterns, given
Introduction
restrictions of computational efficiency and conditions of the interfaces” (p. 118). On this basis, they identify five “core” parameter schemata, leaving open the possibility of further schemata relating to phonetic nullness and lexico-semantic parameterisation. In the context of GGL’s so-called Parameters and Schemata (P&S) model, then, the contents of UG is restricted to a limited number of functional features, and lexical categories, together with “a tiny class of parameter schemata” (p. 122); the parameters with which we are familiar are not therefore linguistic primitives; instead, they are constructed on the basis of innately specified features, lexical categories and schemata, which are the actual primitives. Aside from the new possibilities that this type of approach opens up in the domain of language acquisition (cf. Section 3.4 and 3.5 above), it also makes a very interesting new prediction about the extent to which grammatical variation is expected to emerge over time. As GGL note, current minimalist-inspired theories might lead one to conclude that the amount of variation should decrease over time, whereas the linguistic reality suggests that the opposite is in fact true – consider, for example, the immense grammatical diversity found in Italo-Romance, a language family which has a 1500–2000 year history, compared to the restricted grammatical diversity found in the various varieties of Afrikaans, a language which has only existed as a variety speakers think of as an independent language, distinct from Dutch since the late 19th century (cf. Deumert 2004).32 GGL argue that their P&S model predicts that grammatical diversity will increase over time. In their words: “once the introduction of a parameter schema into the language faculty is justified for a single case, it will … bring about every more potential variation for all possible combinations of relevant entities in the lexicon (features and categories)” (p. 123). Finally, GGL also consider the role that parameters – viewed in the derivative, schemata-determined manner outlined above – might potentially be able to play in facilitating a better understanding of the phylogenetic relations between languages. They argue that the correctness of the phylogenetic predictions made on the basis of MGP, firstly, corroborates the validity of postulating parameters and, secondly and more importantly, opens up the possibility of using parametric values as comparanda in seeking to establish linguistic relationships.
. As Ian Roberts (p.c.) observes, it should, however, be noted that it is too simplistic to view the emergence of grammatical variation as a process which proceeds uniformly: change can sometimes occur in sudden spurts, as seems to have been the case in Middle English (cf. Fischer et al. 2000) and 9th century Icelandic (cf. Hróarsdóttir 2000); conversely, languages may remain fairly static for relatively long periods of time, at least as far as certain properties are concerned – consider, for example, the lack of change in most of the Romance languages with respect to their NSL status.
The limits of syntactic variation
For GGL, then, parametric theory, rigorously pursued, may facilitate insights into a wide range of issues, going well beyond the points at which languages may differ. Zeijlstra’s paper, like GGL’s, pursues the question of what a truly minimalist, plausibly specified UG would contain and the proposal he makes resonates with theirs in various interesting ways. Zeijlstra’s specific focus is the question of the identity of the set of formal features available to UG and his central proposal is that this set is in fact empty. Instead, the inventory of formal features that can head a functional projection in a given language is argued to be fixed on the basis of PLD, during first-language acquisition (cf. also Evers & van Kampen for a similar proposal in another domain). Strikingly, Zeijlstra and GGL’s proposals converge on one of the choice-points requiring fixing during language acquisition – Is feature F formalized as a formal (grammatical) feature? (cf. the first of GGL’s schemata). Where they differ is in respect of the ontological status that they assign to this choice-point: for GGL, it is one of the innately specified schemata in UG, whereas for Zeijlstra, it simply constitutes a linguists’ description of one of the major types of PLD-driven choice that children make during acquisition. In terms of Zeijlstra’s Flexible Formal Feature Hypothesis (FFFH), children must determine for the features encoded in their language whether a given feature is a purely semantic feature which has C-I-relevant content, but none which is relevant to the computational system (i.e., NS) or whether it has both. In the latter case, a formal (i.e., grammatical) feature will be postulated. The cue for deciding the status of a feature is the presence vs. absence in the PLD of semantically vacuous doubling: where it is present, the child “notices” and a formal feature is postulated; where it is absent, there is nothing to “notice” and therefore nothing in effect happens as the FFFH does not assume the prior existence of an entity requiring “negative setting” (cf. also Roberts 2007d: 17–18). Zeijlstra presents the crosslinguistic distribution of Negative Concord phenomena as a case study to illustrate the workings of the proposal, demonstrating that [negation] must be analysed as a formal feature in Negative Concord languages, where negations do not cancel each other out. In socalled Double Negation languages, by contrast, each negative element does in fact bear negative meaning, with the result that children acquiring these languages will conclude that [negation] is always and only a semantic feature. In general, then, the proposal is that instances of syntactic agreement (“concord”) will signal the presence of a formal feature in the language being acquired. As Zeijlstra notes, the proposal has clear implications not only for the description of synchronically attested varieties, but also for the manner in which diachronic changes will take place. In the latter context, it is clear that the loss and/ or introduction of sound-meaning forms (words, clitics, inflections, etc.) bearing features which consequently are or are not doubled would be expected to result
Introduction
in syntactic changes. In the former context, it leads us firstly to expect that languages will differ in respect of their clause structure: Zeijlstra assumes this to be determined by the features which are formalised for a given language, with only formal features being able to project functional categories. It also leads us to expect that it might be possible to identify language types on the basis of the presence vs. absence of certain formal features/functional categories. And finally, it has important consequences for the Borer-Chomsky Conjecture mentioned in Section 3.4 above. If the FFFH is correct, then all parametric variation cannot be reduced to lexical considerations, such as the presence versus absence of functional heads in a given language, as this is simply a reflex of the true locus of parametric variation which is, at it were, “one level up”: whether or not a given semantic feature has been formalised or not. The last three papers in this section, like Zeijlstra’s, focus on specific empirical phenomena in order to make a more general point about the locus of parametric variation. Kayne’s paper offers an excellent illustration of how closely one needs to look at linguistic data in order to avoid the trap – already highlighted by the American Descriptivists (cf. Section 1 above) – of assuming that superficially very similar elements are in fact genuinely exponents of the same category/phenomenon. His, from a syntactician’s perspective, seemingly modest starting point is an apparent syncretism in the clitic system of the Veneto dialects. Employing the methodology of microparametric syntax, i.e., systematic comparison with the behaviour of the apparently corresponding elements in closely related varieties (here: standard Italian and French; cf. Kayne 2005a for more detailed discussion), he shows how misleading both the initial impression of syncretism and the labels traditionally associated with some of the relevant forms can be. Drawing on his own extensive recent work on silence as a point of crosslinguistic and sometimes also parametric variation (cf. Kayne 2005b), he proposes a unified analysis of the Romance facts in terms of which Veneto varieties superficially exhibiting syncretism in fact feature a silent clitic alternating with an overtly realised one, whereas the alternating forms in other Romance varieties are both overt. Focusing more specifically on the overtly realised form, he then shows how microparametric work can lead to new perspectives on the syntax of elements that have received a great deal of attention in earlier generative work. In the case at hand, it is shown how systematic investigation of the distribution of theretype elements in a range of languages points to there in fact being a typology of there-forms. Here, then, we see another case where what is overtly lexicalised (an invariant there-form in all the languages in question) obscures the nature of the feature-bundle that is initially drawn from the Lexicon and that is therefore present in the syntax. In the English case, there can be shown to lexicalise
The limits of syntactic variation
not only what Kayne designates “deictic there”, but i.a. also locative there which in fact modifies silent PLACE, and demonstrative there which modifies silent THING or PERSON. Kayne argues that deictic there is the element traditionally labelled expletive there (cf. There is a book on the table) and that it is in fact no different to locative and demonstrative there in being a modifier associated with a particular DP. Contrary to what is generally believed about expletives, then, he proposes that they originate as part of a so-called big DP (cf. Szabolsci 1983, 1994; Cecchetto 1999, 2000; Boeckx 2003 and Duguine, this volume for discussion of big DPS), subsequently undergoing raising to their surface position. Unlike the “traditional” EPP-oriented explanation (cf. Chomsky 1981), Kayne’s microparametrically based one facilitates an account of a range of puzzling and seemingly unrelated phenomena (cf. Sections 6–14 of his paper), thereby fulfilling one of the “requirements” on parameters discussed by Newmeyer, namely the “rich deductive structure” discussed in Section 3.5 above. Further, it also enables one to make non-obvious and seemingly validated predictions that do not fall out from the traditional analysis and that highlight the important role played by lexical considerations – in particular, which lexical items languages use in which contexts and what this signifies about the nature of silent elements that might additionally be present – both in determining the patterns that are available to languages and those that are universally excluded. More generally, a striking aspect of Kayne’s microparametric, silence-oriented work is its frequent focus on lexical items that had not previously received attention in generative syntactic research. To the extent that consideration of whether and/or how items of this type are lexicalised uncovers new correlations and makes new predictions as to the permitted range of variation, it corroborates Haspelmath’s (p. 78) point that it is not a priori obvious which aspects of language structure will ultimately deliver a window on the limits on variation that Chomskyans seek to identify. It also undoubtedly highlights one of the non-trivial difficulties associated with doing microparametric syntax. Ramchand & Svenonius (R&S) are also centrally concerned with the role lexical items play in determining systematic differences between languages. In their case, however, the focus is specifically on the meaning that lexical items encode and, in particular, on those aspects of meaning which some languages encode lexically, while others appear not to do so. R&S specifically focus on definiteness/specificity, tense and aspect in their chapter, but number, evidentiality and focus are other semantic categories that have frequently been discussed in this connection. Given that the relevant interpretation can still be successfully communicated by speakers, regardless of whether it is lexicalised or not, the question that arises in languages which fail to lexicalise a given meaning is where the “extra” meaning comes from. R&S reject the possibility that such meaning can be “filled in” by a specially designated semantic module of the type assumed by
Introduction
i.a. Chierchia (1998a, 1998b). Instead, they propose that “general non-linguistic properties of the C-I system allow C-I to interpret certain aspects (but not others) of syntactic representations which may be left underspecified” (p. 7) and that there is therefore no need to complicate the architecture of the language faculty by assuming an additional module. More specifically, they propose that certain meaning-components are universally required to be present in the syntax so as to ensure a structure that will be well-formed at C-I. These include Tense (T), which serves to guarantee that the event encoded in the (first phase) verbal domain can be linked to the speech time, thereby creating a well-formed proposition; Aspect (Asp), which is necessary to relate the event domain and the temporal domain; and D, which is necessary to create elements that can function as the arguments of predicates. If this interpretation of the role of the functional categories in question is correct, it is clear that variation in respect of the presence vs. absence of heads forming the functional sequence cannot be a locus of parametric variation, and this is precisely what R&S argue: the functional sequence must always be present in its entirety to ensure that C-I has the “basic skeleton” it requires in order to fulfil its function of collating the information provided to it by the linguistic representation and that coming from the context. Languages may therefore vary in respect of how much their lexical items explicitly encode about the reference of variables like T, Asp and D which are essential to C-I, but the C-I interface also imposes a limit on “true semantic variation” (p. 242) in that these variables must always be represented, providing the starting point for the “enrichment” that ultimately enables all languages to express tensed propositions, events with internal structure, and so on, regardless of whether they lexicalise the semantic categories in question. R&S also address the question of how parametric differences of this sort can be successfully acquired, proposing that children acquiring languages in which optionally lexicalised/syntacticized meaning components are in fact formally encoded acquire these early owing to the fact that they are heavily and unambiguously attested in the PLD. By contrast, children acquiring languages in which one or more of these meaning components is not formally instantiated receive more ambiguous PLD, with the consequence that they only acquire the relevant expressive strategies at a relatively later stage. This proposal would appear to be substantiated by research on the acquisition of definiteness (cf. Anderssen 2006 and also the chapter by Avram & Coene, this volume). Like R&S’s, Lekakou’s paper is also concerned with the question of how differences in the manner in which certain semantic meaning components are encoded can play a role in determining crosslinguistic variation. Her paper specifically focuses on crosslinguistic variation in the morphosyntax of aspect and the consequences that this has for the manner in which middle constructions are realised in different
The limits of syntactic variation
languages. It is well known that middles in some languages (e.g., Greek and French) exhibit (reflexive) passive syntax, whereas those in other languages (e.g., English and Dutch) pattern with unergatives (cf. Lekakou 2005 for detailed comparative discussion). Lekakou argues that this pattern of variation is not accidental and that the mechanism via which a language expresses middle semantics is in fact parametrically determined on the basis of how a language encodes genericity. More specifically, it is proposed that middles be treated as disposition ascriptions, i.e., as structures which enable an object to be ascribed a particular type of generic property such as “good washability” in the case of These jeans wash well. Given this semantic characterization, the manner in which genericity – and, in particular generic quantification over events or so-called VP-level genericity – is encoded becomes relevant (cf. also Condoravdi 1989). According to Lekakou, languages like French and Greek, which encode a distinction between perfective and imperfective aspect, have at their disposal a morphosyntactic means of expressing VP-level genericity, with the consequence that these languages can employ passive structures as a vehicle for middle semantics: because the required generic operator (gen) is encoded in the aspectual morphosyntax (via imperfective aspect), the silent form of the generic pronoun one, ONE*, can be licensed within VP and middles can be derived in the syntax. By contrast, languages like English and Dutch, which do not encode aspectual distinctions morphosyntactically, cannot do so, with the result that middles must be lexically derived. For Lekakou, then, what one might think of as an Aspect Parameter has consequences which extend beyond the manner in which aspect is expressed and, consequently, “recovered” at C-I (cf. R&S’s discussion of this point); it also determines an at first sight completely unrelated structural property, namely the choice of the structure encoding middle semantics. In sum, then, the papers in this section offer a wide range of perspectives regarding the likely locus of different types of syntactic variation, signifying both the range of factors that bear consideration and also the directions that future research might productively take.
5. A classic parameter revisited: The Null Subject Parameter As noted by Baker (this volume: 352), “[h]istory has not been kind to the Pro-drop parameter as originally stated.” Thus, for example, Gilligan’s (1987) investigation of the extent to which the properties Rizzi (1982) originally predicted to cluster together (cf. Section 3.2. above) actually do so has frequently been cited as having revealed the failure of most of the predicted correlations (cf. i.a. Jaeggli & Safir 1989, Newmeyer 2005a,b and Nicolis, this volume; but see also Roberts 2007b for a critical evaluation of the extent to which Gilligan’s investigations
Introduction
genuinely undermine Rizzi’s original proposals). Empirically, then, there are questions about the NSP as originally formulated. And the same is arguably true on the theoretical front. As Holmberg (2005) notes, Rizzi’s pro is incompatible with minimalist proposals in terms of which verbal inflectional features are uninterpretable or unvalued (cf. Chomsky 1995, 2000, 2001) as these cannot then play a role in recovering pro’s grammatical specification. Additionally, one might, in the context of research that takes the SMT as its point of departure, be suspicious of empty categories lacking both PF and LF properties in the way pro was originally assumed to (cf. Chomsky 1995 on Agr). A reconsideration of null subjects and null-subject phenomena therefore also seems called for on theoretical grounds. Given these facts, the question that arises is what we are to make of the NSP’s fate. Recent discussion of this matter has delivered two diametrically opposite answers, which we will now consider in turn. For Newmeyer (2005a,b), the NSP’s empirical failure incontrovertibly points to the failure of the parametric enterprise as a whole. Worth noting, though, is that Newmeyer takes as his point of departure the NSP as originally formulated (cf. Section 3.2. above), with no mention being made of the possibility that this parameter’s “failure” might be the consequence of its simply having been a “first pass” formulation which might be amenable to more successful reformulation. By contrast, the papers in the second section of this book all revisit this classic parameter, attempting to understand how the empirical and theoretical progress that has been made in the quarter century since the NSP’s initial publication might enable us to gain a better understanding of the true nature of the relevant phenomena and of parametric syntax more generally (cf. also Holmberg & Roberts to appear). Tellingly, none of these papers conclude that the original NSP’s failure to stand up to detailed empirical investigation constitutes the death knell for parametric syntax. As the papers by Neeleman & Szendröi and Duguine and also that by Modesto and Öztürk in section III show, there is, in this context and also more generally, good reason to be wary of an assumption that was initially made during the GB era, namely that superficially identical phenomena – in this case, the appearance of a phonetically unrealised subject – are in fact genuinely (underlyingly) the same – in this case, the lexical item pro. What these contributions once again suggest, then, is the aptness in the null-subject context of the American Descriptivists’ warning against simply assuming that one is necessarily dealing with elements of the same kind as those which are familiar to one from well-studied languages. Let us consider this point in a little more detail. Holmberg (2005)’s consideration of Finnish, one of the NSLs that instantiate a counterexample to Rizzi’s original NSP, leads him to conclude that the null subjects found in the world’s languages cannot be thought of as constituting the unitary phenomenon that was, for example, assumed when Gilligan (1987) set out
The limits of syntactic variation
to test the NSP’s crosslinguistic validity. Instead, he proposes that null subjects and NSLs can be of various kinds (cf. also Holmberg, to appear). Neither proposal is entirely novel since numerous researchers, starting with Huang (1984), had previously suggested drawing a distinction between “Romance-style” null-subject languages, which behave in accordance with the NSP, involving the null subject, pro, and “discourse pro-drop” languages, which permit more widespread argument drop, involving an empty category (Huang’s ec) bound by a discourse topic. Rizzi (1986), in turn, also distinguishes “full” NSLs of the Romance type from “semi” NSLs which do not permit referential null subjects, but do license null expletives of various kinds (sometimes, as in Germanic, subject to positional restrictions). Even during the GB era, then, it was acknowledged that it was likely that unrealised subjects might not all be connected to the NSP. More recently, this idea has been the focus of renewed attention, with Tomioka (2003: 336), for example, suggesting that null subjects in the so-called “discourse pro-drop” languages are in fact deleted nominals rather than empty categories of the GB kind. His “Discourse Pro-Drop Generalisation” is given in (25): (25) All languages which allow discourse pro-drop allow (robust) bare NP arguments. … Null pronouns in Discourse Pro-Drop languages are simply the result of N’-deletion/NP-ellipsis without determiner stranding.
For Tomioka, as for Huang, then, null subjects in discourse pro-drop languages are to be viewed as significantly different from those found in Romance-style NSLs, both in terms of their formal status (pro vs. deleted N’/NP) and in terms of the phenomena with which they are parametrically linked (the traditional NSP cluster vs. a distinct Discourse Pro-Drop cluster relating to the structure of nominals more generally). As we will see below, the same is true of the rather different analysis in Neeleman & Szendröi’s contribution to this volume. Holmberg (2005, to appear) builds on these ideas, highlighting a range of circumstances that may give rise to what superficially appears to be a null subject. Specifically, he proposes three major mechanisms that may do so: (26) a. movement, b. deletion, and c. truncation.
(26a) corresponds to the scenario under which movement of a richly inflected verb (i.e., V associated with an agreement complex) into the T-domain gives rise to a structure lacking an overtly realised DP-subject (cf. i.a. Borer 1986; Barbosa 1995; 2001, to appear, Alexiadou & Anagnostopolou 1998 for analyses of Romance-style NSLs along these lines). As outlined in Section 3.3 above, analyses of this type do not necessarily entail the projection of the canonical subject position, Spec-TP, with the consequence that there is no need to postulate the presence of a null
Introduction
pronominal of any kind: the agreement on the verb serves as the subject in this case, with the result that it is not in fact strictly true that no subject is overtly realised as an “I-subject” is present in all the relevant cases (cf. Borer 1986). Holmberg (2005, to appear), Roberts (to appear) and Sheehan (2006) in turn argue that null subjects can also be the consequence of (26b), i.e., a deletion operation (cf. Perlmutter’s original (1971) proposal; see also Duguine, this volume). For these researchers, null subjects arise where a pronominal is deleted in the presence of an appropriately specified T. For some Romance-style NSLs – specifically, those in which overtly realised preverbal subjects do not exhibit the A'-properties that a V-movement analysis would lead us to expect (cf. Sheehan 2006, to appear and Barbosa, to appear for detailed discussion) – they propose that the relevant species of T is associated with a D-feature, with the consequence that deletion of pronominals (viewed as φPs) can result in a referential null subject (cf. Longobardi 1994 and also R&S, this volume, on the role of D). In languages like Finnish, by contrast, the absence of a D-feature on T entails that pronoun (φP) deletion can only result in a non-referential or generic null subject (cf. Holmberg 2005). Referential null subjects are, however, possible in languages of this kind, albeit in a restricted set of contexts (those in which control is possible from the matrix clause – cf. also Holmberg & Sheehan (to appear) and Holmberg, Nayudu & Sheehan (to appear)). The relevant scenarios are schematically represented in (27) (adapted from Holmberg, to appear): (27) a. Basic DP structure ϕP
ϕ
NP
b. Referential null subjects in (some) Romance-style NSLs (strikethrough signifies a deleted lower copy; signifies a deleted pronominal) TP T' T [D, uϕ ...]
vP... ϕP
i.e., D + φ compose to deliver a referential null subject
The limits of syntactic variation
c. Non-referential/generic null subjects in Finnish-style NSLs TP T' T [uϕ ...]
vP... ϕP
i.e., no D present; hence referential null-subjects impossible
d.
Referential null subjects in Finnish-style NSLs [DP Jarii] mainitsi [että φPi oli ostanut talon] Jari mentioned that had bought house ‘Jari mentioned that he had bought a house.’ i.e., D + φ compose to deliver a referential null subject
Leaving aside the details, the crucial point regarding this type of approach is that it opens up the possibility of postulating different types of null-subject languages, depending on the following interacting factors: a. the (substantive) featural make-up of T, b. the structural properties of a language’s pronouns (cf. i.a. Cardinaletti & Starke 1999 and Déchaine & Wiltschko 2002), and, c. the obligatoriness or otherwise of the requirement that Spec-TP be projected (cf. Sheehan 2006 for discussion).33 Appealing to these considerations, Holmberg (2005, to appear) and Holmberg & Sheehan (to appear) and Holmberg, Nayudu & Sheehan (to appear) illustrate how they might enable us to account for the existence of a previously problematic class of NSLs which exhibits both overt and covert expletives and additionally also a restricted subset of referential null subjects – a class they designate partial NSLs. It remains to be seen whether the full predicted range of NSLs is actually attested, but what is immediately clear is that the interaction of relatively uncontroversial properties such as (a–c) predicts a far wider range of NSLs than that assumed during the GB era, a potentially valuable outcome given the findings of Gilligan (1987).
. On the view that Agree-based movement to T is triggered by a movement diacritic (e.g., the EPP-feature discussed in Section 3.4) which is in fact parasitic on some substantive feature (e.g., V, D, etc.), it is clear that (a) and (c) must in fact be viewed as distinct properties.
Introduction
Turning to (26c), it is also possible that unrealised subjects may be the consequence of clausal truncation. Holmberg (2001) suggests that this is the case in the context of replies to Yes/No questions such as that illustrated in (28): (28) a. A: B:
Puhun-ko Joni ranskaa? speaks-Q John French ‘Does John speak French?’ Puhun speaks ‘Yes.’
[Finnish]
b. [CP Puhun-C [IP Joni tpuhun ranskaa]]
As Holmberg (ibid.) shows, the availability of null subjects in structures such as (28) does not correlate with the more general availability of null subjects in a language – besides Finnish, various dialects of Chinese and European Portuguese permit (28)-type structures, whereas languages like Japanese, Persian and Spanish do not. Rather than postulating complex conditions on the licensing of pro, Holmberg therefore suggests that the unrealised subjects in these languages are in fact elided as part of a larger constituent that undergoes deletion. Null-subject structures like Finnish (28) are therefore argued to be as unrelated to the more general null-subject profile of a language as, for example, sluiced structures which may also lack an overtly realised subject, (cf. I know he knows something, but I don’t know [CP what [TP he knows]), but which are never viewed as NSP-related. Given that structures like that in B’s reply are superficially identical to genuine null-subject structures, though, it is clear that care needs to be taken when inspecting data from less familiar languages. Likewise, it emerges that Newmeyer’s (2005b: 200) observation, following Gilligan (1987), that “a solid majority of the world’s languages are null subject” may follow, not from the “unmarked” nature of one or more null-subject-related parameter(s), but from the fact that a range of syntactic operations may result in what superficially looks like a null subject. To summarise, then: if the above is on the right track, it seems fair to say that a quarter century of research on the properties of languages exhibiting null subjects has led to a better understanding not only of the varied nature of null subjects, but also of how null-subject-related parameters might be formulated. Contrary to what Newmeyer proposes, then, the fate of the original NSP seems to represent, if anything, a vindication of the value of the parametric enterprise. The remainder of this section focuses on four recent investigations of nullsubject phenomena. Nicolis’s paper reconsiders some of the correlations that were originally proposed to hold in NSLs generally (cf. Rizzi 1982, 1986), with creole languages forming the major empirical focus of the chapter. Like many of the languages
The limits of syntactic variation
investigated in Gilligan (1987), the creoles under discussion in this chapter attest to the implausibility of viewing the availability of Free Inversion as a precondition for lack of that-trace effects (cf. (16) above). Following recent work by Rizzi & Shlonsky (2005), it is, however, suggested that that-trace effects may in fact be voided via a range of mechanisms, with Free Inversion being just one of these. The central insight in Rizzi’s original work was that that-trace effects will not arise wherever a wh-subject does not have to raise to Spec-TP in order to satisfy the EPP. Taking into account the discussion of the fate of the EPP as originally stated (cf. Section 3.3), it is then clear that languages which are not required to raise subjects to Spec-TP – for example, by virtue of this raising being optional (cf. Biberauer 2006) or by virtue of another XP (e.g., VP or vP) undergoing raising to this position or by virtue of the fact that Spec-TP need not even be projected (cf. V-to-T analyses of null-subject languages) – should not (under the relevant circumstances) exhibit that-trace effects. Similarly, expletive insertion can facilitate subject extraction (contrast *How many people did he say that __ had arrived __? vs. How many people did he say that there had arrived __). What Nicolis specifically argues in this paper is that the lack of Free Inversion in creole languages lacking that-trace effects should not be viewed as fatal for Rizzi’s original proposal; instead, it appears that the languages in question are able to void that-trace violations because they all license null expletives (pro–ref ), i.e., elements which can fill Spec-TP, thereby facilitating subject extraction from a lower position. Pro–ref → lack of that-trace effects can therefore be viewed as a unidirectional correlation relating two at first sight unrelated properties. In a similar vein, Nicolis also considers Proref → Pro–ref , showing how this correlation, likewise, appears to hold universally. Overall, then, the paper shows that careful re-examination of the correlations initially proposed by Rizzi (1982, 1986) does not, as is sometimes suggested, undermine the validity of the original correlations; it simply highlights the fact that there is a wider range of factors in play than was previously thought, with various mechanisms being able to deliver the same surface result (cf. Roberts 2007b for further consideration of this matter). While Nicolis’s chapter focuses on previously proposed correlations, FernándezSalgueiro’s considers a new property which, he argues, should also be related to the parametric cluster associated with Romance-style null-subject languages. The property in question is Further Raising, illustrated in (29): (29) Estes nenos parece que son moi listos these kids seems that are very smart ‘These kids seem to be very smart.’
[Galician]
Fernández-Salgueiro proposes that the Case-F valuation parameter determines whether structures of this kind are available in a given language. In terms of this parameter, the point at which a nominal’s Case features are valued is subject to
Introduction
parametric variation, with those in which Further Raising constructions are possible, for example, being able to delay (Nominative) Case valuation beyond the point at which the nominal’s φ-features have been valued by finite T. In the case of biclausal structures like (29), Case valuation can subsequently take place following Agree with matrix T, with the result that a well-formed structure results. As Fernández-Salgueiro points out, however, it will not be known at the stage of the derivation at which Case valuation is delayed whether a higher clause is to be merged or not; therefore it is possible that nominals in monoclausal structures in languages of this type may ultimately reach PF with an unvalued Case feature. Observing that all the Romance languages permitting Further Raising are also NSLs, he proposes that at least some null subjects in these languages may in fact be nominals with unvalued Case features which consequently cannot be spelled out. On Fernández-Salgueiro’s proposal, then, null subjects may sometimes be the consequence of a repair strategy. Duguine considers a rather different null-subject, or, more properly, nullargument language, Basque. She proposes that this language, which has often been described as a three-way pro-drop language, permitting pro in subject, object and indirect object positions, does not in fact license pro at all. Instead, she argues that the unrealised nominals in this language are in fact elided DPs, which consequently exhibit a number of properties associated with elision structures. Duguine’s arguments centre on the availability of null arguments in non-finite (nominalised) clauses lacking the inflectional richness required to license and identify pro and, in particular, on previously unnoticed interpretive parallels between silent arguments in Basque and elided structures in English. She shows how Basque structures featuring unrealised arguments exhibit the same range of possible strict and sloppy interpretations as English co-ordination structures featuring VP ellipsis in the second conjunct, and, crucially, that the overt argument-/VP-containing counterparts of these structures permit one reading (strict + sloppy) that is not available in the silent argument/VP ellipsis case (cf. Dahl’s puzzle, Dahl 1974). Drawing on Fiengo & May’s (1994) observation that “the effect of ellipsis is … eliminative”, Duguine proposes that it is ellipsis which restricts the range of interpretive possibilities in both cases. Building on this, she then considers the behaviour of topicalised and focalised pronouns vis-à-vis their non-topicalised/-focalised counterparts, showing that the latter once again systematically fail to exhibit the strict + sloppy meaning that is also unavailable where pronouns are elided. Since unrealised arguments can neither be topicalised nor focalised, she concludes that it is really the focalisation or topicalisation of a pronoun – and not simply its realisation – which determines the availability of the “extra” reading in the case of Dahl’s puzzle. According to her proposal, then, pronouns marked with a topic ([T]) or focus ([F]) feature are
The limits of syntactic variation
barred from ellipsis at PF, whereas those lacking a feature of this kind may optionally undergo ellipsis, resulting in null arguments. As far as LF is concerned, the strict + sloppy reading is reserved for [T]-/[F]-marked DPs, with the result that both realised and unrealised non-topicalised/-focalised nominals are correctly predicted to lack this reading. Given the independent evidence that natural languages make use of ellipsis, the fact that Basque null arguments can surface in environments that lack the properties required to license a Rizzian pro and, finally, the range of null-subject phenomena that have recently been said by others (including those mentioned above) to involve deletion, the conclusion is therefore that an ellipsis analysis of Basque null arguments is superior to a pro-based one. The final paper in this section focuses on another type of NSL that has historically posed problems for pro-based accounts, namely the so-called “discourse” or “radical pro-drop” languages. Departing from earlier suggestions about these languages, Neeleman & Szendröi (N&S) propose a novel analysis in terms of which pronoun rather than verbal morphology is crucial in determining the availability of null subjects and arguments more generally. Specifically, they propose that a language will only allow radical pro-drop if its personal pronouns are agglutinating for Case, Number, or some other nominal feature. Languages with fusional pronoun morphology are therefore predicted not to permit this option, and N&S devote their chapter to illustrating the correctness of these predictions, drawing on data made available in the WALS (cf. Neeleman & Szendröi 2007 for earlier discussion of this analysis). The rationale underlying the proposal that agglutinating morphology should be crucial in determining the availability of radical pro-drop is that morphology of this type opens up the possibility that two implications of the Elsewhere Condition (cf. Kiparsky 1973) – spell out the maximum number of features (i.e., the most specific form) and spell out the topmost category – may conflict. Consider (30) which illustrates a pronoun with agglutinating structure (following the model proposed in Weerman & Evers-Vermeul 2002) and the spellout rules that might be associated with it: (30)
KP target of spell-out � rule for Case
K
� DP
D
target of radical pro-drop rule � NP
N
possible target of spellout rules for pronouns � possible target of spellout rules for pronouns ...
As shown above, the various components of pronouns with an agglutinating structure are each associated with their own spellout rule, with the pronoun as a whole
Introduction
consisting of an appropriately concatenated version of these various forms. If unrealised pronouns (null subjects, etc.) are actually deleted pronouns, we can think of these as falling out from a zero spellout rule associated with the entire structure, i.e., the topmost node (KP in (30)). Crucially, the Elsewhere-imposed requirements mentioned above conflict in the context of structures like (30): spelling out KP (i.e., a null pronoun) satisfies the topmost category requirement, whereas spelling out the individual components to generate an overtly realised agglutinating pronoun satisfies the requirement that the maximum number of features be spelled out. Both realisations should therefore be available. In fusional systems, by contrast, pronouns are necessarily spellouts of KP (cf. English him which could be represented as: [KP +p, –a, 3sg, masc, acc]), with the result that the spellout rule assumed to result in radical pro-drop (i.e., [KP +p, –a] ⇔ ∅) will always be blocked by the featurally more specific spellout rule resulting in the realisation of overt pronouns. The prediction therefore is that languages with fusional pronouns will only permit null forms in contexts specified by context-sensitive rules (e.g., [KP +p, –a, φi] ⇔ ∅ / ___ [φi], which would deliver the type of null-subject behaviour that is observed in Romance-style null-subject languages). As N&S show in their paper, investigation of a wide range of agglutinating and fusional languages, many of which have not previously been investigated in the null-subject context, suggests that this proposal may well be on the right track. In sum, then, we see that there is agreement among the authors in this section that null-subject phenomena in the world’s languages cannot simply be viewed as the reflex of a single parameter or set of parameters; instead, it seems to be the case both that different species of unrealised subjects exist and that these may surface as a consequence of quite different considerations. One of these species of null subjects is the “classical” null subject of agreement-rich languages like Italian, but there additionally appear to be several others. Precisely how many distinct types of null subject underlie Newmeyer’s observation regarding the crosslinguistic predominance of “null-subject languages” remains to be seen.
6. Parametric clustering The issue of parametric clustering has already arisen in various places in previous sections, but it lies at the heart of all the papers in this section. As indicated by its title, Baker’s paper concerns itself with the fate of so-called macro- or metaparameters. Statements like Pica’s (2001: v–vi) that “twenty years of intensive descriptive and theoretical research has shown [that] metaparameters do not exist” suggest that history may not have treated macroparameters generally any more kindly than the NSP, but Baker argues that it is not sufficient only to engage in microparametric syntax or to simply assume that the Borer-Chomsky Conjecture/BCC
The limits of syntactic variation
in (23) exhausts the locus of parametric variation. For Baker, typological considerations are crucial here as “the typological record falsifies the view that all syntactic variation is microparametric variation” (8).34 We have already noted in connection with head-directionality in Section 3.4 that languages do not fall on a “smooth continuum” of the type the BCC would lead us to expect. In this chapter, Baker argues that macroparameters are necessary to account for the observed skewing in favour of “pure” language types (e.g., “harmonic” languages in the word-order domain). He offers an illustration of how the macroparametric approach can bring to light new parameters which faithfully represent the observed “pureness”, focusing on two new agreement parameters which emerged from a detailed comparison of Niger-Congo and Indo-European languages (cf. Baker 2008 for further discussion). These are stated in (31): (31) a. Agreement macroparameter 1 A functional head F agrees with NP only if NP asymmetrically c-commands F. (YES: Niger-Congo languages; NO: Indo-European languages) b. Agreement macroparameter 2 A functional head F agrees with NP only if F values the Case feature of NP or vice versa. (NO: Niger-Congo languages; YES: Indo-European languages)
As Baker shows, the settings of these parameters have wide-ranging consequences as F encompasses at least T, C, D, P, v and Lk (which hosts linker particles; cf. Baker & Collins 2006) in the languages investigated to date. As in the case of headdirectionality, BCC-based statements capturing these facts on an FC-by-FC basis would seem to be missing the generalisation highlighted by Baker’s macroparameters, in addition to raising the question why languages with consistent agreement behaviour are significantly more common than languages with inconsistent agreement behaviour. Like Baker’s, Modesto’s chapter also argues for a parametric system which is structured by both macro- and microparameters. His empirical focus is on the null-subject behaviour of three genetically and areally unrelated languages: Brazilian Portuguese (BP), Finnish and Chinese. All three are said to permit null subjects in the absence of rich agreement as a consequence of the fact that they are positively specified for a Topic Prominence Parameter (TPP) approximating that
. Obviously, the same record also poses a significant challenge for adherents of the view that Newmeyer (2005a,b 2006) espouses that crosslinguistic variation is ultimately determined by language-specific rules.
Introduction
first proposed in Li & Thompson (1976). Once again, then, we see a case of nullsubject phenomena which are argued not to fall out directly from the parametric settings that give rise to Romance-style null subjects. Obviously, this raises the question of how a child acquiring a null-subject system will be able to determine what type of system this is. If null subjects in Topic-prominent (Tp) languages in fact co-occur with a cluster of other properties determined by the TPP, the acquisitional problem, however, arguably disappears, and Modesto does indeed postulate a cluster of TPP-related properties, including special topic marking, lack of overt expletives, double subject structures, the absence of passive constructions and more generally topic- as opposed to subject-controlled deletion. Many of these properties are argued to follow from the fact that the EPP-requirement in Tp languages is said to hold in the C-domain and not in the T-domain as in Subjectprominent (Sp) languages like English. Overt expletives are therefore superfluous, there is no need for subjects to be realised in a single “canonical” position and the backgrounding achieved via passive structures can likewise automatically fall out as the consequence of the fact that subjects do not need to be positionally privileged. As Modesto notes, the three languages under consideration in his paper do not uniformly exhibit all the properties which he ascribes to the TPP. Thus, for example, Finnish does have an overt expletive (sitä), although this alternates with a null form (cf. Holmberg 2005 and also Nicolis, this volume for discussion), and it is also not so clear that this language permits subject doubling of the kind typically found in Tp languages like Chinese (cf. Holmberg & Nikanne 2007 for recent discussion of Finnish subject doubling). Clearly, then, there needs to be an account of the lower-level variation found in these languages, with acknowledgement of the possibility that at least some of the properties associated with the TPP may, like those initially ascribed to the NSP, in fact not follow directly from the setting of a macroparameter, but may instead result from various microparametric settings. Modesto addresses this matter by focusing on BP, which emerges as particularly interesting in the context of the discussion by virtue of the fact that it has undergone diachronic change in respect of its TPP setting: whereas European Portuguese is clearly Sp, BP synchronically exhibits a number of clear Tp properties. Given this state of affairs, it must be possible for the macroparametric setting determining a language’s Tp vs. Sp nature to change over time, presumably via intermediate stages (but see Panagiotidis, this volume, for the proposal that macroparameters are relatively resistant to change). Another macroparameter, namely a version of the Non-Configurationality Parameter first proposed by Hale (1980, 1983) is the focus of Öztürk’s contribution. Her paper, which focuses on Turkish, specifically investigates two of the properties associated with this parameter, namely free word order and argument drop.
The limits of syntactic variation
The proposal is that the domain in which argumenthood is established determines whether the language in question will be configurational or non-configurational. For Öztürk, nominals are only arguments if they have the D-related property of referentiality (cf. also R&S, this volume) and the further property of Case. In the case of configurational languages like English, Case needs to be valued by heads (v and T) which Öztürk assumes to be outside the thematic domain within which nominals which ultimately attain argument status are first-merged. Since these heads, which are part of the functional sequence, crucially have a fixed order relative to each other, subjects and objects will, in the absence of A'-movement triggers, generally surface in a fixed order relative to one another. In Turkish, by contrast, the theta-role-assigning heads are also the ones that check (or value) Case-features, with the result that argumenthood in Turkish, unlike that in English, can be established in situ. This means that Turkish differs from English in lacking Case-driven movement out of the thematic domain; where movement out of this domain takes place – and it may well be that “duality of semantics” considerations of the kind discussed in Chomsky (2000 et seq.) always require at least one element to move out of the thematic domain – it will then be A’-movement, which may target any phrase and which thereby creates the impression of non-configurationality. Having established the connection between Turkish non-configurationality and the manner in which argumenthood is established, Öztürk then proceeds to consider how argument drop can be related to this property. Building on the observation that argument drop is not conditioned by morphological considerations – it is possible in adjunct clauses that do not exhibit any verbal agreement morphology – she proposes that the fact that Turkish arguments are already “complete” in their base position means that they can be deleted there. Because English pronouns are not “complete” in this way, they cannot be similarly dropped.35 If this is correct, the implication is that Turkish null arguments are, once again, deleted nominals rather than traditional-style pros. The final chapter in this section also focuses on the parametric properties of nominals, this time in Greek and with specific emphasis on their diachrony. Panagiotidis observes that the diachronic stability of certain syntactic structures in the nominal domain contrasts sharply with the instability of certain other nominal structures, and his paper aims to shed light on the reasons underlying this stability difference. His proposal is that the structures which are most immune to change are those whose availability is determined by macroparametric choices, whereas
. Note that object-drop in English, unlike that in Turkish, never involves referential null objects; only generic or arbitrary objects (i.e., so-called “logical constants”) may be “dropped” (cf. Rizzi 1986 for discussion).
Introduction
those which prove least resistant to change are the reflex of microparametric settings. This proposal obviously rests on the way in which “macroparametric” and “microparametric” are formally characterised, and Panagiotidis offers the following characterisation: macroparametric choices are assumed to involve choices pertaining to the presence versus absence in a grammar of functional categories, while microparametric choices are assumed to relate to the presence versus absence of uninterpretable features on otherwise identical functional categories (cf. also Zeijlstra and GGL, this volume). In the diachronic context, then, the proposal is that the instability of structures determined by microparametric settings is linked to the generally “defective” status of uninterpretable features in the human language faculty. Since (apparently) macroparametric settings also appear to be able to undergo change (consider the case of BP discussed in Modesto’s chapter and also the very familiar and seemingly oft-attested case of OV-VO shift in the domain of word order), it is, however, clear that the proposal cannot be that macroparametric change is impossible. Worth noting in this connection is Baker’s observation that the empirical facts would seem to suggest that syntactic systems reflect a mix of macro- and microparametric properties, with the latter sometimes contributing “noise” around the properties determined by macroparametric settings. If this is correct, one could imagine a scenario in which microparametric changes add up, thereby increasing the overall amount of “noise”, possibly to the point where the original macroparametric setting is largely obscured and therefore amenable to change. As should be obvious from the speculative nature of these comments, understanding of this matter, as of so many relating to parametric clustering, will ultimately depend on the development of a far more fine-grained understanding than we currently have not only of the form of parameters, but also of the manner in which they interact.
7. The acquisition of parameters The final section of the volume consists of two papers focusing on acquisitional aspects of parametric theory. As noted in Section 3.2, parametric approaches to language acquisition have typically assumed that the process of language acquisition entails the setting of innately prespecified parameters on the basis of Primary Linguistic Data/PLD (cf. (5)). As Dresher (2003: 14) notes, the rationale behind this type of approach to language acquisition is that parameter setting should be a “much more manageable learning problem than open-ended induction, or hypothesis formation and testing”. In other words, parameter setting should in part address the oft-noted poverty of the stimulus problem that children encounter in
The limits of syntactic variation
acquisition: although the input is deficient in various ways, this is not fatal as they can draw on what is innately given. For a model of this type to work, it is clear that we need to establish what the innate specification entails, something which the discussion in the preceding sections has revealed currently still to be very much work-in-progress. Additionally, however, there is also an implementation question: how precisely do acquirers make use of innately specified information in order to converge on the appropriate target grammar? As has often been noted (cf. references cited in Section 3.2 and also i.a. Clark & Roberts 1993; Gibson & Wexler 1994; Yang 2002 and Niyogi 2006), this is a non-trivial question to which the answer is also not obvious at this stage: among other things, there is an epistemological problem relating to how acquirers successfully identify the abstract categories that are innately specified in the (untagged) input to which they are exposed; there is also a question as to the “surface visibility” of individual parameters, given the fact that parameters are assumed to interact, with these interactions in some cases resulting in obscuring effects; and, given these facts, there is also the further question of how acquirers are able to set parameters in the correct sequence so as not to fall foul of what Dresher designates the “credit problem” (see below). The contributions in this section consider both content and implementation questions. Avram & Coene’s paper specifically focuses on closer consideration of the relationship between syntax and morphology in acquisition. The proposal is that there is, at least in some cases, a morphological basis for the making of parametric choices and that this basis is more articulated and therefore also more illuminating than previously assumed. More specifically, Avram & Coene (A&C) propose that morphology directly signals information about certain properties of a language’s functional categories (FCs) – e.g., what their component features are, what values are associated with these features, etc. – and that syntactic acquisition involves choosing from the range of possible feature values made available by UG those that correspond to the target (adult) language. Crucially, A&C assume that children execute this selection process in a step-by-step manner, starting with more general decisions regarding the feature-structure of a given FC and thereafter moving on to more specific decisions (as noted below, Evers & van Kampen, this volume, make a similar proposal and offer a detailed consideration of the factors that ensure that children get the sequence right; the idea that acquisition entails a sequence of ever more specified grammars dates back to at least Chomsky 1975). What makes child language acquisition so interesting, according to A&C, is the fact that each parametric choice results in a cluster of syntactic and morphosyntactic surface effects. Child language acquisition, then, is argued to be a potentially valuable window on the correlation between specific (morpho)syntactic properties and the acquisition of a particular feature-structure or feature-value.
Introduction
To illustrate this idea, Avram & Coene consider the acquisition of agreement morphology by two Romanian children. They argue in favour of a unified account of the acquisition of agreement in the verbal and nominal functional domain, with [Person] constituting a privileged feature in the acquisition process by virtue of its importance in determining reference to entities in the discourse. More specifically, they propose that feature valuation in both domains consists of two basic steps: i. determining whether the inflectional head in the relevant domain is [+/–split] (cf. Bobaljik & Thráinsson 1998 for discussion of the Split INFL parameter in the verbal domain). In the verbal domain, this entails determining whether T and AgrS are independently projected and in the nominal domain, whether NumP is projected separately; and ii. wherever the [+split] setting applies, determining whether the separately projected agreement projection has maximally valued [Person] features or not. In the verbal domain, this entails determining whether the AgrS projection is associated with so-called “pronominal agreement”, i.e., the type of agreement which can license referential null subjects, a capacity which many researchers relate to the presence of a [Person] specification (cf. i.a. Rizzi 1982, 1986; Borer 1984; Platzack 1987; Barbosa 1995, to appear, Alexiadou & Anagnostopoulou 1998) or not, and in the nominal domain, it involves determining whether NumP is dominated by a [Person]-bearing DP layer or not. As noted above, each of these steps is said to correlate directly with the emergence of specific morphosyntactic forms and syntactic structures, with the Romanian case study specifically suggesting three developmental stages corresponding to the setting of these parameters. Particularly worth noting is the fact that it appears to be possible to characterise the parametric choices at issue here in “higher order” terms which do not make specific reference to syntactic category: the same choices have to be made in both domains. This once again suggests that more detailed consideration of what is actually specified in UG and of the format that parameters may take could be rewarding. We have already mentioned proposals along these lines by GGL and Zeijlstra, and conclude this section by considering the acquisition-inspired ideas of Evers & van Kampen. Like GGL and Zeijlstra, Evers & van Kampen (EvK) propose that the parameters most commonly identified in the literature are not in fact innate; instead, they argue that these formal properties of the grammatical system originate as “cultural discoveries” which are systematised by a reflexive mind. Their point of departure is Dresher’s (1999, 2003) identification of two serious complications afflicting the traditionally accepted parameter-setting model of language acquisition: the socalled “epistemological” and “credit” problems. As noted above, the first relates
The limits of syntactic variation
to the discrepancy between the very abstract nature of parameters vis-à-vis what is found in the PLD (cf. the discussion of “cues” in Section 3.2), and the second to the fact that parameters are usually thought to interact in specific ways, with these interactions potentially obscuring certain aspects of each other’s settings – a state of affairs which implies that children must follow a specific parameter setting sequence (cf. Chomsky 1975, and also C&A’s paper). Together, the epistemological and credit problems therefore give rise to the questions in (32): (32) a. How is a data set Di+1 selected by a grammar Gi? b. When does a new parameter in G become relevant?
To answer these questions, EvK propose that the language acquisition procedure initially starts by radically reducing the input that it receives. At the earliest stages, this input reduction entails ignoring all grammatical markings (articles, copulas, other auxiliaries, verbal inflections, connectives, etc.) so that utterances can be reduced to an intake of two directly identifiable elements (and the poverty of the stimulus problem is arguably reduced).36 Subsequently, the reduction procedure can be thought of as schematised in (33): (33)
Reduction of input to intake a. First reduction: Replace each functional category still unknown with 〈F?〉. b. Second reduction: Set aside all input strings with more than one 〈F?〉. c. Residue: An intake of the form 〈F?〉/〈Fi〉, where Fi is already known.
Here, the residue in (33c) represents the minimal pattern from which the functional category 〈F?〉 can be acquired within a certain language. According to EvK, the so-called evidence frames which result from input reduction are employed by the language acquisition procedure in order to direct the course of acquisition. Once 〈F?〉 has been identified as a specific category, that category can feature in subsequent evidence frames, with the result that an ordering of acquisition steps emerges in which later steps select more language-specific and more structured contexts. For EvK, then, the order of specific acquisition steps does not necessarily have to be universal since these are determined by evidence frames established on the basis of language-particular PLD, but there is the expectation of a more general ordering of acquisition steps. Furthermore, their proposal predicts that properties acquired earlier are likely to have an effect on the steps that follow, with the result that they may have typological significance of the “macroparametric” type
. Interestingly, EvK suggest that this earliest stage is not beyond the grasp of non-human primates. What differentiates humans from other primates, then, is (33), i.e., the application of an input reduction mechanism.
Introduction
discussed in Baker’s work (cf. also Wexler 1998 on so-called Very Early Parameter Setting/VEPS). To illustrate the workings of their proposal, EvK present a case study from the acquisition of core aspects of Dutch syntax, showing how V2 and V-final parameters may be “discovered” via the strategy in (33). If this is on the right track, it may prove feasible to postulate that speakers of V2 languages ultimately have a “V2 parameter” as part of their adult linguistic competence, but this does not have to entail that an unset form of this parameter was present from birth. Instead, one might think of this parameter, and parameters more generally, as arising as a result of something like Piagetian “epigenesis”, with “third factor”-driven acquisition strategies like EvK’s (33), and also more general UG principles like inclusiveness and locality interacting with the PLD to “create” parametric settings (cf. also Roberts 2007d) and, moreover, doing so in a sequence which evades the credit problem. Importantly, EvK view the “cultural discoveries” as leading to parameter creation and setting, i.e., to the creation of a parametric system which determines intricate and often not at all superficially obvious interrelations between (morpho) syntactic properties; they do not see these as producing either rules (cf. Newmeyer 2005a,b) or construction templates (cf. Tomasello 2003). Their paper, then, aims to illustrate how a very “bare” UG – one which may in fact contain no parameters and possibly only very few constraining principles – might nevertheless result in the types of articulated parametric systems which have been explored and postulated for the last nearly 30 years.
References Abels, K. & Neeleman, A. 2007. Universal 20 without the LCA. MS, University of Tromsø and UCL. (Available on-line at: auf.net/lingBuzz/000279 ) Aboh, E. 2004. The Morphosyntax of Complement-Head Sequences. Clause Structure and Word Order Patterns in Kwa. Oxford: OUP. Ackema, P. & Neeleman, A. 2002. Effects of short-term storage in processing rightward movement. In Storage and Computation in the Language Faculty, S. Nooteboom, F. Weerman & F. Wijnen (Eds), 219–256. Dordrecht: Kluwer. Adger, D. 2006. Combinatorial variation. Journal of Linguistics 42: 503–530. Alexiadou, A. & Anagnostopoulou, E. 1998. Parametrizing AGR: Word order, V-movement and EPP-checking. Natural Language and Linguistic Theory 16: 491–539. Anderssen, M. 2006. The Acquisition of Compositional Definiteness in Norwegian. Ph.D. dissertation, University of Tromsø. Baker, M. 1996. The Polysynthesis Parameter. Oxford: OUP. Baker, M. 1997. Thematic roles and syntactic structure. In Elements of grammar, L. Haegeman (Ed.), 73–137. Dordrecht: Kluwer. Baker, M. 2008. The syntax of Agreement and Concord. Cambridge: CUP. Baker, M. & Collins, C. 2006. Linkers and the internal structure of vP. Natural Language and Linguistic Theory 24: 307–354.
The limits of syntactic variation Barbosa, P. 1995. Null Subjects. Ph.D. dissertation, MIT. Barbosa, P. 2001. On inversion in wh-questions in Romance. In Romance Inversion, A. Hulk & J-Y. Pollock (Eds), 20–90. Oxford: OUP. Barbosa, P. To appear. Two kinds of subject pro. Studia Linguistica. Bayer, J. 1999. Final complementizers in hybrid languages. Journal of Linguistics 35: 233–271. Bayer, J. 2001. Two grammars in one: Sentential complements and complementizers in Bengali and other South Asian languages. In The Yearbook of South Asian Languages: Tokyo Symposium on South Asian Languages – Contact, Convergence and Typology, P. Bhaskarorao & K. Subbarao (Eds), 11–36. New Delhi: Sage. Berwick, R. & Weinberg, A. 1984. The Grammatical Basis of Linguistic Performance. Cambridge MA: The MIT Press. Bever, T. & Langendoen, T. 1971. A dynamic model of the evolution of language. Linguistic Inquiry 2: 433–461. Bhattaycharya, T. 1999. The Structure of the Bangla DP. Ph.D. dissertation, University College London (UCL). Biberauer, T. 2004. Reconsidering the EPP and Spec-TP in Germanic. Cambridge Occasional Papers in Linguistics (COPiL) 1: 15–40. Biberauer, T. 2006. Changes at the edge: Tracking the rise of a canonical subject position in Germanic. Paper given at the Edges in Syntax Conference (Cyprus). Biberauer, T. & Richards, M. 2006. True optionality: When the grammar doesn’t mind. In Minimalist Essays, C. Boeckx (Ed.), 35–67. Amsterdam: John Benjamins. Biberauer, T. & Roberts, I. 2005. Changing EPP-parameters in the history of English: Accounting for variation and change. English Language and Linguistics 9: 5–46. Biberauer, T. & Roberts, I. 2006. The loss of ‘head-final’ orders and remnant fronting in Late Middle English: Causes and consequences. In Comparative Studies in Germanic Syntax: From A(frikaans) to Z(ürich German), J. Hartmann & L. Molnárfi (Eds), 263–297. Amsterdam: John Benjamins. Biberauer, T. & Roberts, I. 2008. Cascading parameter changes: Internally-driven change in Middle and Early Modern English. In Grammatical Change and Linguistic Theory, T. Eythórsson (Ed.), 70–113. Biberauer, T. & Roberts, I. To appear. Subjects, tense and verb movement. In Null Subjects: The Structure of Parametric Variation, A. Holmberg & I. Roberts (Eds). Cambridge: CUP. Biberauer, T., Holmberg, A. & Roberts, I. 2007. Disharmonic word-order systems and the FinalOver-Final Constraint (FOFC). Proceedings of the 33rd Incontro di grammatica generative [Available on-line: http://amsacta.cib.unibo.it/archive/00002397/01/PROCEEINGS_ IGG33.pdf] Bloomfield, L. 1933. Language. New York NY: Henry Holt & Co. Bloomfield, L. 1934. Review of Handbuch der erklärenden Syntax by Wilhelm Havers (Heidelberg: Winters, 1931). Language 10: 32–39. Boas, F. 1911. Introduction. In Handbook of American Indian Languages, P. Holder (Ed.), 1–70. Lincoln NB: University of Nebraska Press. Bobaljik, J. 2002. A-chains at the PF-interface: Copies and ‘covert’ movement. Natural Language and Linguistic Theory 20: 197–267. Bobaljik, J. & Jonas, D. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27: 195–236. Bobaljik, J. & Thráinsson, H. 1998. Two heads aren’t always better than one. Syntax 1: 37–71.
Introduction
Boeckx, C. 2003. Islands and Chains. Resumption as Derivational Residue. Amsterdam: John Benjamins. Borer, H. 1984. Parametric Syntax. Dordrecht: Foris. Borer, H. 1986. I-subjects. Linguistic Inquiry 17: 375–416. Bošković, Z. 2002. On multiple wh-fronting. Linguistic Inquiry 33: 351–383. Bouchard, D. 2003. The origins of language variation. Linguistic Variation Yearbook 3: 1–41. Bowerman, M. 1988. The ‘no negative evidence’ problem: How do children avoid an overly general grammar? In Explaining Language Universals, J. Hawkins (Ed.), 73–101. Oxford: Blackwell. Brody, M. 1995. Lexico-Logical Form: A Radically Minimalist Theory. Cambridge MA: The MIT Press. Brody, M. 2000. Mirror theory: Syntactic representation in perfect syntax. Linguistic Inquiry 31: 29–56. Cardinaletti, A. 1997. Subjects and clause structure. In The New Comparative Syntax, L. Haegeman (Ed.), 33–63. London: Longman. Cardinaletti, A. 2004. Toward a cartography of subject positions. In The Cartography of Syntactic Structures: The Structure of IP and CP, Rizzi, L. (Ed.), 115–165. Oxford: OUP. Cardinaletti, A. & Starke, M. 1999. The typology of structural deficiency: A case study of three classes of pronouns. In Clitics in the Languages of Europe, H. van Riemsdijk (Ed.), 145–233. Berlin: Mouton de Gruyter. Carnie, A. & Guilfoyle, E. 2000. The Syntax of Verb Initial Languages. Oxford: OUP. Carnie, A., Dooley, S. & Harley, H. 2005. Verb First. On the Syntax of Verb-Initial Languages. Amsterdam: John Benjamins. Cecchetto, C. 1999. A comparative analysis of Left and Right Dislocation in Romance. Studia Linguistica 53: 40–67. Cecchetto, C. 2000. Doubling structures and reconstruction. Probus 12: 93–126. Cheng, L. 1997. On the Typology of Wh-Questions. New York NY: Garland. Chierchia, G. 1998a. Plurality of mass nouns and the notion of ‘semantic parameter’. In Events in Grammar, S. Rothstein (Ed.), 53–105. Dordrecht: Kluwer. Chierchia, G. 1998b. Reference to kinds across languages. Natural Language Semantics 6: 339–405. Chomsky, N. 1973. Conditions on transformations. In A Festschrift for Morris Halle, S. Anderson & P. Kiparsky (Eds), 232–286. New York NY: Holt, Rinehart & Winston. Chomsky, N. 1975. Reflections on Language. New York NY: Pantheon. Chomsky, N. 1981. Lectures on Government and Binding Theory. Dordrecht: Foris. Chomsky, N. 1982. Some Concepts and Consequence of the Theory of Government and Binding. Dordrecht: Foris. Chomsky, N. 1986. Knowledge of Language: Its Nature, Origin, and Use. New York NY: Praeger. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik. R. Martin, D. Michaels & J. Uriagereka (Eds), 89–156. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2004a. Beyond explanatory adequacy. In Structures and Beyond. The Cartography of Syntactic Structures, Vol. 4, A. Belletti (Ed.), 104–131. Oxford: OUP.
The limits of syntactic variation Chomsky, N. 2004b. The Generative Enterprise Revisited. Berlin: Mouton de Gruyter. Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, N. 2007. Approaching UG from Below. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax-Semantics, U. Sauerland & H. Gärtner (Eds), 1–29. Berlin: Mouton. Chomsky, N. 2008. Oh phases. In Foundational Issues in Linguistic Theory, R. Freidin, C. Otero & M. Zubizarreta (Eds), 133–166. Cambridge MA: The MIT Press. Cinque, G. 2005. Deriving Greenberg's Universal 20 and its exceptions. Linguistic Inquiry 36: 315–332. Cinque, G. To appear. The fundamental left-right asymmetry of natural languages. To appear in Universals of Language Today, S. Scalise, E. Magni & A. Bisetto (Eds). Berlin: Springer. Clark, R. & Roberts, I. 1993. A computational model of language learnability and language change. Linguistic Inquiry 24: 299–345. Clifton, C. & Frazier, L. 1989. Comprehending sentences with long-distance dependencies. In Linguistic Structure and Language Processing, G. Carlson & M. Tanenhaus (Eds), 273–317. Dordrecht: Kluwer. Condoravdi, C. 1989. The middle: Where semantics and morphology meet. MIT Working Papers in Linguistics 11: 18–30. Cornilescu, A. 2000. The double subject construction in Romanian. In Comparative Studies in Romanian Syntax, V. Motopanyane (Ed.), 83–134. Amsterdam: Elsevier. Corver, N. & Nunes, J. 2007. The Copy Theory of Movement. Amsterdam: John Benjamins. Dahl, Ö. 1974. How to open a sentence. Abstraction in natural language. Logical Grammar Reports 12. Götenberg: University of Götenberg Press. Davies, W. & Dubinsky, S. 2002. Objects and Other Subjects: Grammatical Functions, Functional Categories, and Configurationality. Dordrecht: Kluwer. Déchaine, R-M. & Wiltschko, M. 2002. Decomposing pronouns. Linguistic Inquiry 33: 409–442. Den Dikken, M. forthcoming. On the functional structure of locative and directional PPs. MS, CUNY.(Available on-line at:http://web.gc.cuny.edu/dept/lingu/dendikken/docs/pp_ cinqueandschweikert.pdf) Deumert, A. 2004. Language Standardization and Language Change. The Dynamics of Cape Dutch. Amsterdam: John Benjamins. Diesing, M. 1992. Indefinites. Cambridge MA: The MIT Press. Dresher, E. 1999. Charting the learning path: Cues to parameter setting. Linguistic Inquiry 30: 27–67. Dresher, E. 2003. Meno’s paradox and the acquisition of grammar. In Living on the Edge: 28 Papers in Honour of Jonathan Kaye, S. Ploch (Ed.), 7–27. Berlin: Mouton de Gruyter. Dryer, M. 1992. On the Greenbergian word-order correlations. Language 68: 81–138. Dryer, M. 1998. Why statistical universals are better than absolute universals. Chicago Linguistic Society 33: 123–145. Epstein, D., Groat, E., Kitahara, H. & Kawashima, R. (Eds). 1998. A Derivational Approach to Syntactic Relations. Oxford: OUP. Fiengo, R. & May, R. 1994. Indices and Identity. Cambridge MA: The MIT Press. Fischer, O., van Kemenade, A., Koopman, W. & van der Wurff, W. 2000. The Syntax of Early English. Cambridge: CUP.
Introduction
Fodor, J.D. 1978. Parsing strategies and constraints on transformations. Linguistic Inquiry 9: 427–493. Fodor, J.D. 1984. Constraints on gaps: Is the parser a significant influence? In Explanations for Language Universals, B. Butterworth, B. Comrie & Ö. Dahl (Eds), 9–34. Berlin: Mouton. Fodor, J.D. 1998a. Unambiguous triggers. Linguistic Inquiry 29: 1–36. Fodor, J.D. 1998b. Parsing to learn. Journal of Psycholinguistic Research 27(3): 339–374. Fodor, J.D. 2001. Setting syntactic parameters. In The Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins (Eds), 730–767. Oxford: Blackwell. Fodor, J.D. & Sakas, W. 2005. The subset principle in syntax: Costs of compliance. Journal of Linguistics 41: 513–569. Frazier, L. & Flores D’Arcais, G. 1989. Filler driven parsing: A study of gap-filling in Dutch. Journal of Memory and Language 28: 331–342. Fukui, N. 1988. Deriving the differences between English and Japanese: A case study in parametric syntax. English Linguistics 5: 249–270. Fukui, N. 1995. Theory of Projection in Syntax. Stanford CA: CSLI. Gibson, E. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68: 1–76. Gibson, E. & Wexler, K. 1994. Triggers. Linguistic Inquiry 25: 407–454. Gilligan, G. 1987. A cross-linguistic approach to the pro-drop parameter. Ph.D. dissertation, USC. Good, J. 2008. Linguistic Universals and Language Change. Oxford: OUP. Greenberg, J. 1963. Some universals of grammar with particular reference to the order of meaningful elements. In Universals of Language, J. Greenberg (Ed.), 73–113. Cambridge MA: The MIT Press. Greenberg, J., Osgood, C. & Jenkins, J. 1961. Memorandum concerning language universals. Presented at the Conference on Language Universals. Dobbs Ferry NY, 13–15 April 1961. Published as: Greenberg, J., Osgood, C. & Jenkins, J. 1963. Memorandum concerning language universals. In Universals of Language, J. Greenberg (Ed.), xv–xxvii. Cambridge MA: The MIT Press. Hale, K. 1980. Remarks on Japanese phrase structure: Comments on the papers on Japanese syntax. MIT Working Papers in Linguistics 2: 185–203. Hale, K. 1983. Warlpiri and the grammar of non-configurational languages. Natural Language and Linguistic Theory 1: 5–47. Haspelmath, M. 2004. Does linguistic explanation presuppose linguistic description? Studies in Language 28(3): 554–579. Haspelmath, M., Dryer, M., Gil, D. & Comrie, B. 2005. World Atlas of Language Structures (WALS). Oxford: OUP. Hauser, M., Chomsky, N. & Fitch, T. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 22: 1569–1579. Hawkins, J. 1983. Word Order Universals. New York NY: Academic Press. Hawkins, J. 1988. Explaining Universals. Oxford: Blackwell. Hawkins, J. 2004. Efficiency and Complexity in Grammars. Oxford: OUP. Holmberg, A. 2001. The syntax of yes and no in Finnish. Studia Linguistica 55: 141–175. Holmberg, A. 2005. Is there a pro? Evidence from Finnish. Linguistic Inquiry 36(4): 533–564. Holmberg, A. To appear. Null subject parameters. To appear in: A. Holmberg, & I. Roberts (Eds), Null Subjects and the Structure and Parametric Theory. Cambridge: CUP. Holmberg, A., Nayudu, A. & Sheehan, M. To appear. Control of null subjects in finite clauses in three partial null-subject languages. Studia Linguistica.
The limits of syntactic variation Holmberg, A. & Sheehan, M. To appear. Control into finite clauses in partial null-subject languages. To appear in Null Subjects: The Structure of Parametric Variation, A. Holmberg & I. Roberts (Eds). Cambridge: CUP. Holmberg, A. & Nikanne, U. 2007. Subject doubling in Finnish: The role of deficient pronouns. To appear in: S, Barbiers, M. van der Ham, M. Lekakou & O. Koeneman (Eds), Microvariations in Syntactic Doubling. Oxford: Elsevier. Holmberg, A. & Roberts, I. To appear. Null Subjects: The Structure of Parametric Variation. Cambridge: CUP. Hsieh, F. & Sybesma, R. 2007. On the linearization of Chinese sentence-final particles: Max Spellout and why CP moves. MS, MIT and Leiden University. Hróarsdóttir, T. 2000. Word Order Change in Icelandic. From OV to VO. Amsterdam: John Benjamins. Huang, C-T.J. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531–574. Huang, C-T.J. 2006. The macro-history of Chinese syntax and the theory of change. Talk given at the Chinese Linguistics Workshop (University of Chicago). Jaeggli, O. & Safir, K. 1989. Introduction. In The Null Subject Parameter, O. Jaeggli & K. Safir (Eds), 1–44. Dordrecht: Springer. Jonas, D. 1996. Clause Structure and Verb Syntax in Scandinavian and English. Ph.D. dissertation, Harvard University. Joos, M. 1957. Readings in Linguistics: The Development of Descriptive Linguistics in America since 1925. Washington DC: American Council of Learned Societies. Just, M. & Carpenter, P. 1992. A capacity theory of comprehension: Individual differences in working memory. Psychological Review 98: 122–149. Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press. Kayne, R. 2005a. Movement and Silence. Oxford: OUP. Kayne, R. 2005b. Some notes on comparative syntax, with special reference to English and French. In The Oxford Handbook of Comparative Syntax, G. Cinque & R. Kayne (Eds), 3–69. Oxford: OUP. Kayne, R. 2008. Antisymmetry and the Lexicon. MS, NYU. (Available on-line at: http://ling.auf. net/lingBuzz/000598) Kiparsky, P. 1973. ‘Elsewhere’ in phonology. In A Festschrift for Morris Halle, S. Anderson & P. Kiparsky (Eds), 93–106. New York NY: Holt, Rinehart and Winston. Kiparsky, P. 2008. Universals constrain change; change results in typological generalisations. In Language Universals and Language Change, P.P.J. Good (Ed.), 23–53. Oxford: OUP. Koopman, H. 1997. Prepositions, postpositions, circumpositions and particles: The structure of Dutch PPs. MS, UCLA. (Available at: http://www.linguistics.ucla.edu/people/koopman/ papers/pp.pdf) Kroch, A. 1994. Morphosyntactic variation. In Proceedings of the 30th regional meeting of the Chicago Linguistic Society, K. Beals, R. Knippen, L. Melnar, H. Suzuki, & E. Zeinfeld (Eds), 180–201. Chicago IL: University of Chicago Press. Kural, M. 2005. Tree traversal and word order. Linguistic Inquiry 36: 367–388. Landau, I. 2006. Chain resolution in Hebrew V(P)-fronting. Syntax 9: 32–65. Lechner, W. 2006. Interpretive Effects of Head Movement. (Available on-line at: http://ling.auf. net/lingBuzz/000178) Lehmann, W. 1973. A structural principle of language and its implications. Language 49: 47–66.
Introduction
Lekakou, M. 2005. In the Middle Somewhat Elevated. The Semantics of Middles and its Crosslinguistic Realization. Ph.D. dissertation, UCL. Lepsius, K. 1880. Nubische Grammatik mit einer Einleitung über die Völker und Sprachen Afrika’s. Berlin: Verlag von Wilhelm Hertz. Li, N. & Thompson, S. 1976. Subject and topic: A new typology of language. In Subject and Topic, C. Li (Ed.), 457–489. New York NY: Academic Press. Lightfoot, D. 1989. The child’s trigger experience: Degree-0 learnability. Behavioral and Brain Sciences 12: 321–375. Lightfoot, D. 1991. How to set parameters: Arguments from language change. Cambridge MA: The MIT Press. Lightfoot, D. 1999. The Development of Language: Acquisition, Change and Evolution. Oxford: Blackwell. Longobardi, G. 1994. Reference and proper names: A theory of N-movement in syntax and logical form. Linguistic Inquiry 25: 609–665. Longobardi, G. 2003. Methods in parametric linguistics and cognitive history. Linguistic Variation Yearbook 3: 101–138. Longobardi, G. 2005. A minimalist program for parametric linguistics? In Organizing Grammar: Linguistic Studies for Henk van Riemsdijk, H. Broekhuis, N. Corver, M. Huybrechts, U. Kleinhenz & J. Koster (Eds), 407–414. Berlin: Mouton. Manzini, R. & Wexler, K. 1987. Parameters, Binding Theory and Learnability. Linguistic Inquiry 18: 413–444. Marantz, A. 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. University of Pennsylvania Working Papers in Linguistics 4: 201–225. Marcus, G. 1993. Negative evidence in language acquisition. Cognition 46: 53–85. Meillet, A. 1903. Introduction a l’étude comparative des langues indo-européennes. Paris: Hachette. Mobbs, I. 2008. ‘Functionalism’, the design of the language faculty and (disharmonic) typology. MS, University of Cambridge. Mohr, S. 2005. Clausal Architecture and Subject Positions. Impersonal Constructions in the Germanic Languages. Amsterdam: John Benjamins. Moro, A. 2000. Dynamic Antisymmetry. Cambridge MA: The MIT Press. Müller, G. 2005. Pro-drop and impoverishment. In Form, Structure and Grammar. A Festschrift presented to Günther Grewendorf on the Occasion of his 60th Birthday, P. Brandt & E. Fuss (Eds), 93–115. Tübingen: Narr. Müller, G. 2007a. Formale and funktionale Aspekte abstrakter Analyse: Ikonizität und Minimalität. In Von der Pragmatik zur Grammatik, S. Döring & J. Geilfuss (Eds), 135–161. Leipzig: Leipziger Universitätsverlag. Müller, G. 2007b. Some Consequences of an Impoverishment-Based Approach to Morphological Richness and Pro-Drop. (Available on-line at: http://www.uni-leipzig.de/~muellerg/ mu228.pdf) Nayudu, A. 2008. Issues in the Syntax of Marathi: A Minimalist Approach. Ph.D. dissertation, University of Newcastle. Neeleman, A. & Szendröi, K. 2007. Radical pro-drop and the morphology of pronouns. Linguistic Inquiry 38: 671–714. Newmeyer, F. 1998. Language and Language Function. Cambridge MA: The MIT Press. Newmeyer, F. 2005a. Possible and Probable Languages. A Generative Perspective on Linguistic Typology. Oxford: OUP.
The limits of syntactic variation Newmeyer, F. 2005b. Against a parameter-setting approach to typological variation. Linguistic Variation Yearbook 4: 181–234. Newmeyer, F. 2006. A rejoinder to “On the role of parameters in Universal Grammar: A reply to Newmeyer” by I. Roberts & A. Holmberg. MS, University of Washington (Seattle). (Available on-line at: http://people.pwf.cam.ac.uk/mtb23/NSP/Parametricdebate.html) Niyogi, P. 2006. The Computational Nature of Language Learning and Evolution. Cambridge MA: The MIT Press. Ouhalla, J. 1996. Remarks on the binding properties of wh-pronouns. Linguistic Inquiry 27: 676–707. Peirce, C. 1955. Abduction and induction. In Philosophical papers of Peirce, J. Buchler (Ed.), 151–156. New York NY: Dover Publications. Perlmutter, D. 1971. Deep Structure and Surface Structure Constraints in Syntax. New York NY: Holt Rinehart and Winston. Pica, P. 2001. Introduction. Linguistic Variation Yearbook: v–xii. Platzack, C. 1987. The Scandinavian languages and the null-subject parameter. Natural Language and Linguistic Theory 5: 377–401. Pollock, J-Y. 1989. Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry 20: 365–424. Poletto, C. 2000. The Higher Functional Field: Evidence from Northern Italian Dialects. Oxford: OUP. Richards, M. 2004. Object Shift and Scrambling in North and West Germanic: A Case Study in Symmetrical Syntax. Ph.D. dissertation, University of Cambridge. Richards, M. & Biberauer, T. 2005. Explaining Expl. In The Function of Function Words and Functional Categories, M. den Dikken & C. Tortora (Eds), 115–154. Amsterdam: John Benjamins. Richards, N. 2001. Movement in Language. Interactions and Architectures. Oxford: OUP. van Riemsdijk, H. 1978. A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases. Dordrecht: Foris. van Riemsdijk, H. 1990. Functional prepositions. In Unity in Diversity, H. Pinkster & I. Gene (Eds), 229–241. Dordrecht: Foris. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–557. Rizzi, L. 2004. On the study of the language faculty: Results, developments and perspectives. Linguistic Review 21: 323–344. Rizzi, L. & Shlonsky, U. 2005. Strategies of subject extraction. MS, Siena and Geneva. (Available on-line at: http://www.ciscl.unisi.it/doc/doc_pub/rizzi_shlonsky2005.pdf) Roberts, I. 2005. Principles and Parameters in a VSO Language: A Case Study in Welsh. Oxford: OUP. Roberts, I. 2006. Clitics, head movement and incorporation. MS, University of Cambridge. Roberts, I. 2007a. Diachronic Syntax. Oxford: OUP. Roberts, I. 2007b. Introduction. Comparative Grammar: Critical Concepts in Linguistics. Vol. 1 (Early Work: Pre-1980). London: Routledge. Roberts. I. 2007c. Introduction. Comparative Grammar: Critical Concepts in Linguistics. Vol. 2 (The Null Subject Parameter). London: Routledge. Roberts, I. 2007d. The mystery of the overlooked discipline: Modern syntactic theory and cognitive science. MS, Cambridge University. Roberts, I. To appear. A deletion analysis of null subjects. To appear in Null Subjects: The Structure of Parametric Variation, A. Holmberg & I. Roberts (Eds). Cambridge: CUP.
Introduction
Roberts, I. & Holmberg, A. 2005. On the role of parameters in Universal Grammar: A reply to Newmeyer. In Organizing Grammar. Linguistic Studies in Honor of Henk van Riemsdijk, H. Broekhuis, N. Corver, R. Huybregts, U. Kleinhenz & J. Koster (Eds), 538–553. Berlin: Mouton de Gruyter. Roberts, I. & Roussou, A. 2003. Syntactic Change. A Minimalist Approach to Grammaticalization. Cambridge: CUP. Rohrbacher, B. 1999. Morphology-driven Syntax: A Theory of V-to-I Raising and Pro-drop. Amsterdam: John Benjamins. Rosengren, I. 2002. EPP: A syntactic device in the service of semantics. Studia Linguistica 562: 145–190. Ross, J. 1967. Constraints on Variables in Syntax. Ph.D. dissertation, MIT. Sabel, J. 1998. Principles and Parameters of Wh-Questions. Habilitation dissertation: University of Frankfurt. Sabel, J. & Zeller, J. 2006. Wh-question formation in Nguni. In African Languages and Linguistics in Broad Perspective, J.J. Mugane, J. Hutchison & D. Worman (Eds), 271–283. Somerville MA: Cascadilla. Safir, J. 1987. Comments on Wexler and Manzini. In Parameter Setting, T. Roeper & E. Williams (Eds), 77–89. Dordrecht: Reidel. Saito, M. & Fukui, N. 1998. Order in phrase structure and movement. Linguistic Inquiry 29: 439–474. Sakas, W. & Fodor, J.D. 2001. The structural triggers learner. In Language Acquisition and Learnability, S. Bertolo (Ed.), 172–230. Cambridge: CUP. Schmidt, W. 1926. Die Sprachfamilien und Sprachenkreise der Erde. Heidelberg: Carl Winter. Schwartz, B. & Vikner, S. 1996. The verb always leaves IP in V2 clauses. In Parameters and Functional Heads, A. Belletti & L. Rizzi (Eds), 11–62. Oxford: OUP. Sheehan, M. 2006. The EPP and Null Subjects in Romance. Ph.D. dissertation, Newcastle University. Sheehan, M. To appear. ‘Free’ Inversion in Romance and the Null Subject Parameter. To appear in Null Subjects: The Structure of Parametric Variation, A. Holmberg & I. Roberts (Eds). Cambridge: CUP. Shlonsky, U. 1997. Clause Structure and Word Order in Hebrew and Arabic. Oxford: OUP. Simpson, A. 2000. Wh-Movement and the Theory of Feature-Checking. Amsterdam: John Benjamins. Simpson, A. 2005. Classifiers and DP structure in Southeast Asia. In The Oxford Handbook of Comparative Syntax, G. Cinque & R. Kayne (Eds), 806–838. Oxford: OUP. Szabolcsi, A. 1983. The possessor that ran away from home. The Linguistic Review 3: 89–102. Szabolcsi, A. 1994. The noun phrase. In Syntax and Semantics 27. The Syntactic Structure of Hungarian, F. Kiefer & E. Kiss (Eds), 179–274. San Diego CA: Academic Press. Takita, K. 2007. Order in narrow syntax and PF: Toward a theory of word-order variations. MS, Nanzan University. Tamburelli, M. 2006. Remarks on richness. UCL Working Papers in Linguistics 18: 439–455. Tomasello, M. 2003. Constructing a Language. A Usage-Based Theory of Language Acquisition. Cambridge MA: Harvard University Press. Tomioka, N. 2003. The semantics of Japanese null pronouns and its cross-linguistic implications. In The Interfaces: Deriving and Interpreting Omitted Structures, K. Schwabe & S. Winkler (Eds), 321–340. Amsterdam: John Benjamins. Tomlin, R. 1986. Basic Word Order: Functional Principles. London: Croom Helm.
The limits of syntactic variation Thomas, M. 2002. Roger Bacon and Martin Joos. Generative linguistics’ reading of the past. Historiographia Linguistica XXIX (3): 339–378. Travis, L. 2006. VP-, D-movement languages. In Negation, Tense and Clausal Architecture, R. Zanuttini, H. Campos, E. Herburger & P. Portner (Eds), 127–150. Georgetown DC: Georgetown University Press. Uriagereka, J. 2007. Clarifying the notion ‘parameter’. Biolinguistics 1: 99–113. Venneman, T. 1974. Topics, subjects, and word order: From SXV to SVX via TVX. In Historical Linguistics: Proceedings of the First International Congress of Historical Linguistics, Edinburgh, September 1973, Vol. 2, J. Anderson & C. Jones (Eds), 339–376. Amsterdam: North-Holland. Vikner, S. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: OUP. Vikner, S. 1997. Vo-to-Io movement and inflection for person in all tenses. In The New Comparative Syntax, L. Haegeman (Ed.), 189–213. London: Longman. Watanabe, A. 2006. The pied-piper feature. In Wh-movement: Moving on, L. Cheng & N. Corver (Eds), 47–70. Cambridge MA: The MIT Press. Weerman. F. & Evers-Vermeul, J. 2002. Pronouns and Case. Lingua 112: 301–338. Yang, C. 2002. Knowledge and Learning in Natural Language. Oxford: OUP. Zeller, J. 2001. Particle Verbs and Local Domains. Amsterdam: John Benjamins.
part i
The locus of (parametric) variation
Three fundamental issues in parametric linguistics Chiara Gianollo*, Cristina Guardiano** & Giuseppe Longobardi* In this paper we discuss three compelling questions, which impinge on parameter theory: a. What are the actual parameters of UG? b. What is the form of a possible parameter? c. How are parameter values distributed in space and time? With respect to the first question, we present the guidelines of a heuristic methodology for discovering the parameters of UG, Modularized Global Parametrization, first proposed by Longobardi (2003). Building on the potential of such an approach, we pursue a minimalist critique of the general form of parameters and sketch the programmatic lines of a Principles & Schemata model of UG. To address the third problem, we begin to investigate the implications of a parametric approach to the study of phylogenetic relationships among languages and, more broadly, of cognitive approaches to the study of history. Keywords: cognitive history; evolutionary adequacy; genealogical taxonomy; historical adequacy; Modularized Global Parametrization; Parameter Schemata; parametric minimalism
* Università di Trieste ** Università di Modena e Reggio Emilia We are indebted to several colleagues who commented on previous partial versions of these ideas, in particular to Michael Arbib, Theresa Biberauer, Cédric Boeckx, Noam Chomsky, Richard Kayne, Ayesha Kidwai, Gabriele Rigon, Calvert Watkins, two anonymous referees, and the participants in the conference on “The Structure of Parametric Variation” (Newcastle upon Tyne, September 2005) and the V Asian GLOW (New Delhi, October 2005).
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
1. Theory and metatheory of parameters 1.1 Parametric linguistics We regard parametric linguistics as a specific subfield of the biolinguistic framework initiated by works such as Chomsky (1955, 1965) and Lenneberg (1967). Parametric approaches presuppose the crucial result acquired through this framework, namely that the human language faculty is a complex system of invariant innate knowledge, and focus on the further problem of a formal and principled theory of grammatical diversity, a privileged testing ground for models of the interaction between biologically shaped mental structures and culturally variable information. 1.2 The rationale for parametric linguistics No fewer than four distinct epistemological and historical reasons suggest themselves as the impetus and justification for pursuing parametric linguistics as a specific research program in formal grammar at this stage of the study of human cognition: a. Of the classical cognitive faculties (memory, perception, reasoning, object recognition …), language, and grammar in particular, is especially interesting in that it represents a salient instance of culturally transmitted variation. By that we mean that grammatical differences are environmentally triggered and not genetically inherited, though not accidentally distributed in time, space and social structures:1 it is in this sense that we may, for example, easily talk of a history or theory of French syntax, not of French memory or Spanish vision; b. Although straight empiricist approaches to the problem of language acquisition and knowledge may be considered largely refuted by the rise of the biolinguistic program (most explicitly after Chomsky 1959 or Lenneberg 1964), nobody has so far indisputably assessed the effectiveness of the nativist alternative as a model of the human acquisition of grammar diversity, by implementing a parameter setting system over a sufficiently large and realistic number of parameters (a point most appropriately stressed in Fodor 2001 and also emerging from Yang 2003, but also one which had not gone unnoticed in Chomsky 1995: 7: “The P&P [Principles & Parameters, GGL] model is in part a bold speculation rather than a specific hypothesis. Nevertheless, its basic assumptions seem reasonable … and they do suggest a natural way to resolve the tension between descriptive and explanatory adequacy”); c. Parametric linguistics can provide a theory of language variation practically reconciling synchronic and diachronic concerns (an outcome foreseen in Obenauer 1977: 5 and perspicuously called “Aufhebung der methodologischen
. Cf. Longobardi (2003) for more extensive remarks.
Three fundamental issues in parametric linguistics
Trennung von Syn- und Diachronie auf der Ebene der allgemeinen Theorie” [“suspension of the methodological separation between synchrony and diachrony at the level of general theory’’]); as such, it may best bridge the gap between the two most fruitful paradigms in the history of the language sciences, the historicalcomparative paradigm, founded in the 19th century, and the biolinguistic one, developed within the wider cognitive science framework in the second half of the 20th century. The combination of these two paradigms may best inaugurate a new and very powerful device for the study of human mind and behavior (cf. section 4 below for a specific elaboration), part of an empirical discipline termed cognitive history in Longobardi (2003);2 d. The previous considerations have a further potentially disruptive corollary, affecting the perspectives of history as a science more generally: a philosophical tenet of Chomsky’s minimalist program (recently most insightfully elaborated in Boeckx & Piattelli Palmarini 2005) contends that linguistics, relying on the so called “Galilean style” of inquiry, seems to unveil properties of simplicity, elegance, and symmetry characterizing the principles of the language faculty. If true, this is surprising enough in a certain sense: in biology, to which linguistics, as a cognitive science and, ultimately, a “temporary subject” (Chomsky, Class Lectures, 1983), should presumably be reducible, living systems are often regarded as featuring a range of redundancies and being less governed by simple primitive principles. As suggested by other cases of reduction in the development of the natural sciences, it is possible that parts of biology itself will eventually evolve to accommodate some results of linguistics/cognitive science. In principle, then, the discoveries of linguistics as a whole might contribute to raising new issues and refining the methods and conclusions of even well established sciences, like natural history. The subfield of parametric linguistics might play an analogous role (e.g., importing the Galilean style and the successful idealizations of the cognitive paradigm and testing their effectiveness) in the study of cultural history, i.e., in a classical and central domain of the human sciences. 1.3 Three issues The subfield of parametric linguistics can be minimally defined by three fundamental questions: (1) a. What are the actual parameters of UG? b. What is the form of a possible parameter? c. How are parameter values distributed in space and time?
. The expression was independently used in Netz (1999) and the enterprise may regard Vico (1744) as a philosophical precursor.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
Notice that the first question is obviously pertinent to the classical level of explanatory adequacy of Chomsky (1964), since it aims at characterizing the present form of the language faculty. Questions (1b) and (1c), on the other hand, can both be properly understood as attempts to go “beyond explanatory adequacy” (in the sense of Chomsky 2001b), e.g., pursuing two further levels, technically introduced and termed “evolutionary” and “historical” adequacy in Longobardi (2003). The pursuit of evolutionary and historical adequacy should try to answer questions of the following type:
(2) a. Why did we evolve to have precisely the type of language faculty that we have today? b. Why, in the course of cultural history, did precisely the attested variety of human languages arise?
It is clear how the fundamental issues of parameter theory (1b) and (1c) may properly fall into these two classes of problems. In the specific domain of language, they instantiate much more general questions of anthropology like: why do human cultures vary so noticeably? Why did different cultural features come to be distributed in space and time the way they are? In the field of material culture some of these questions have been most insightfully addressed by ground-breaking works such as Diamond (1997). In the domain of properly cognitive aspects of cultural variation, work is just beginning. We believe that grammatical parameters offer particularly fertile ground for investigation of this kind, owing to certain peculiar features which distinguish parameters from many other cultural variables, even including “ideological” ones (e.g., religion, poetics, and others of the type often elucidated by linguists themselves through investigation of the vocabulary of specific languages).3 The variation exemplified by grammatical parameters is discrete, constrained by precise universal bounds; indeed, it is built into a species-invariant cognitive capacity, which contributes, among other things, to producing largely homogeneous mental structures within each community. Furthermore, knowledge of grammar, including parametric values, is largely beyond the grasp of the speakers’ consciousness (cf. Chomsky 1980, ch. 6, for discussion), virtually excluding that it may be significantly affected by deliberate individual decisions. Finally, we want to add that parameter values sometimes exhibit “long-term” historical persistence, somewhat analogous to the télé-histoires whose study within different civilizations has been advocated by Braudel (1985), for example.
. Especially cf. Watkins (1995); Lazzeroni (1998) among others.
Three fundamental issues in parametric linguistics
All these properties can, we argue, be capitalized on for the research program sketched in section 4 below.
2. Explanatory adequacy in parametric linguistics 2.1 How can we discover the parameters of UG? In order to address (1b) and (1c), it is necessary to provide at least a preliminary and partial answer to (1a), i.e., to attain some initial degree of explanatory adequacy. Basically two methods have been developed to address the question in (1a): (3) a. one relies on classical poverty-of-stimulus arguments b. the other uses fully extended typological coverage in the style of Greenberg (1966)
Either method, in itself, is somewhat problematic. (3a) tries to identify parameters focusing on two or more languages, grouping together seemingly related properties on the grounds of poverty-of-stimulus arguments which take the following form: certain crucial properties of adult grammars are so underrepresented in everyday utterances that they would be unlearnable from Primary Linguistic Data (PLD) unless “deduced” by the child from other, more easily observable, properties of the same parameter. This, of course, establishes typological correlations among the properties in question. This method provides deep and often correct grammatical insights, but, alone, it may lead little further than the study of single parameters in relatively few languages. (3b) is an essentially extensional method, whose application comes up against two types of shortcomings:
(4) a. it can hardly be used for the relevant purposes, in practice, because of the depth of grammatical investigation required by a sound parametric analysis. Thus, for example, descriptions in terms of Dixon’s (1998) Basic Linguistic Theory are often insufficient for a parametrically oriented linguist to determine the setting of actual parametric values (cf. also Baker, this volume: note 5 on the sorts of parametrically relevant information that cannot be gleaned from the World Atlas of Language Structures/WALS); b. it is in any case likely to be insufficient if investigation is not guided by strong abductive and theory-oriented considerations, even if hypothetically extended to all languages, because the cardinality of the universal parameter set is so large as to generate a number of possible languages of which the actual existing or known ones represent an infinitesimally small sample.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
2.2 Modularized Global Parametrization In our view, the methodological dilemma raised by the inadequacies of the methods in (3) is the crucial issue to be addressed in order to establish parametric linguistics as a scientific program. Longobardi (2003: 116–117) has suggested that it can be resolved by adopting the following combination of strategies:
(5) a. starting from a relatively wide sample of languages (no less than 12–15, at least) and language families, but crucially comprising a substantive core of structurally similar varieties exhibiting clear minimal contrasts;4 b. focusing in depth on a limited subset of the parameter system, provided that it can be sufficiently well isolated within the language, so that its parameters may exhibit only limited perturbation by the values of others outside the subset, but sufficient theoretically interesting interactions with others within the subset.
This method (“Modularized Global Parametrization”, henceforth MGP) should eventually secure both some typological coverage (i.e., an attempt at globality) and correct depth of insight, at the acceptable cost of limiting the analysis of the parameter system to specific modules. Some conceptual remarks on the size of the enterprise are in any case in order. The implementation of a parametric approach has sometimes been compared to Mendeleev’s development of the periodic table of elements (cf. i.a. Baker 2001; Moro 2006, 2008): the full deployment of UG parameters should provide us with the exhaustive list of biologically possible grammars, including those (the majority) still unattested, as the incomplete periodic table allowed chemists to predict the existence of certain elements in advance of their empirical discovery. Though epistemologically suggestive, the analogy seems to us not completely appropriate in practice, at least for reasons of order of magnitude. It is already clear that there must exist hundreds of parameters, generating languages in the order of powers of billions, most of which have never been realized and presumably never will (cf. Kayne 2000, ch. 1). Thus, it will hardly be possible to complete an extensional list of possible languages and it will certainly be impossible to actually observe all of them. In any event, it is definitely the case one in practice runs into theoretical combinations of parameter values which immediately correspond to obviously existing languages, but also into others which seem to remain tenaciously unattested. . A point especially stressed in principle and practice by Richard Kayne since Kayne (1975). Unlike Greenberg’s approach, then, this method renounces wide areal, genetic, and typological coverage at least initially, though it may enlarge the sample of languages with less strict adherence to the principle in later stages, once a theoretically stable and widely applicable grid of parameters has been worked out.
Three fundamental issues in parametric linguistics
In this respect, on the grounds of our own descriptive experience, we may suggest a likelihood conjecture, to be validated by further inquiry, about the relation between the set of UG parameters and the set of actual languages:
(6) Completeness Conjecture: Predicted combinations of parameter values are more likely to actually be observed among the world’s languages the more the relevant parameters are independent of specific morphophonological encoding completeness
According to this conjecture, if the setting of a parameter relies heavily on the phonological representation of a certain morpheme (e.g., the “robustness” of agreement in a null subject language, the enclitic status of the definite article etc.), accidental “gaps” among the possible languages are more likely to be found, as variation will be at the border between parametric diversity and the lexical arbitrariness of the sound-meaning pairing. By contrast, parameter values more independent of the phonological coding of particular morphemes (such as, apparently, those governing some movement and order properties) are more likely to be realized in the full range of their possible combinations. On the whole, a closer analogy, at least by size, between even a “modularized” version of the parametric program and the natural sciences could perhaps be sought in the human genome project, dealing with the description of thousands of polymorphic units, potentially giving rise to an innumerable amount of phenotypes. 2.3 An illustration of the MGP method: Table A The method of MGP has been tentatively applied, for both theoretical and historical purposes, to the internal structure of Determiner Phrases (Guardiano and Longobardi 2005, Gianollo, Guardiano and Longobardi 2004), producing successive empirical approximations to a type of representation, conventionally named Table A, like the one reported in the Appendix (Figure 1). This table lists values for 47 parameters (rows) in 24 languages (columns): Italian (It), Spanish (Sp), French (Fr), Portuguese (Ptg), Rumanian (Rum), Latin (Lat), Classical Greek (ClG), New Testament Greek (NTG), Modern Greek (Grk), Gothic (Got), Old English (OE), Modern English (E), German (D), Norwegian (Nor), Bulgarian (Blg), Serbo-Croatian (SC), Russian (Rus), Irish (Ir), Welsh (Wel), Hebrew (Heb), Arabic (Ar), Wolof (Wo), Hungarian (Hu), Finnish (Fin). Charts of the Table A-type are a representational format, ultimately an empirical hypothesis attempting to select and encode various kinds of potentially relevant information. In addition to specified values marked + and −, 0 marks the irrelevance of a certain parameter in the language as a consequence of the partial interactions among parameters; by partial interactions we refer to the numerous
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
cases in which one but not the other of the two values of a parameter m determines the value of parameter n.5 Sometimes such interactions actually surface in two slightly different forms: notationally, 0 will be used if and only if, as a consequence of interactions with m, no construction which could trigger n is ever manifested in the weakly generated language; instead, in other cases, the interaction between m and n may necessarily produce surface consequences which are indeed identical to those obtained when n is set to + or to −, respectively: in such circumstances, it may be mnemonically helpful to use notations such as 0+ or 0− (the difference has no ontological status, though, and is not always straightforward or descriptively useful). The dependence of a parameter upon one of the values of another parameter is signalled by the (conjunctions, disjunctions, or negation of) numbers in the label, say, of parameters 2, 3, 4, 5, etc. For example, “+30 or (−31, +32)” next to 33 would mean that setting 33 is only relevant in languages already marked + in respect of 30 or both − in respect of 31 and + in respect of 32, otherwise the value of 33 will be 0.6 In a few cases, it has so far been impossible to safely set the parameter for ancient languages; this is indicated by a bare “?”. In doubtful cases, “+?” or “−?” are used. Finally, “e” marks the existence of sporadic lexical exceptions to the generally set parameter values. As is inevitable, Table A encodes the set of parameters in a partial order. The main objective criterion followed in choosing such an order is the direction of dependence for interacting parameters: as can be seen from the numbers in the labels, implying parameters always precede their respective implied (dependent) parameters. The fact that parameter values are tightly interrelated and that languages fall short of exploiting the entire space of possible grammatical variation due to such implications has often been pointed out as a major formal feature of parameter sets: cf. Fodor (2001: 735), Baker (2001), Longobardi (2003), and Guardiano and Longobardi (2005). But the rich amount of information empirically collected, along the guidelines of MGP, in a representation such as Table A reveals how strikingly pervasive the phenomenon of partial interaction among parameters turns out to be in a module of grammar of a minimally realistic size. The huge number of 0s bears witness to the high level of potential redundancy that a complete parametric
. The case of total interactions, i.e., full syncretism of both values of two parameters, is not contemplated as such implications are likely to suggest that we have to do with just a single, more abstract, parameter. . Occasionally the valuing of a parameter is signaled as partially dependent not on another parameter but on the morphological properties of some elements of the lexicon.
Three fundamental issues in parametric linguistics
list should present a language learner with. In section 3, we will address the possibility of reducing some of these interactions and redundancies to a formally simpler conception of innate grammatical knowledge. 3. Parameters and evolutionary adequacy 3.1 A new tension Question (1b) may be usefully decomposed into two distinct issues: (7) a. Why do parameters have the form they have? b. Why do we have parameters at all?
Both concern the ultimately evolutionary problem of understanding why variation exists, a problem which is obviously central to a minimalist theory of language. Much of the research so far in the Minimalist Program has dealt with the motivation for principles of UG, namely with the reduction of their permissible formats (theoretically possible “morphologies” in Boeckx and Piattelli Palmarini’s 2005 sense) to the interaction of deeper overarching principles. On the contrary, hardly any investigation has been devoted in the same spirit to the form taken by variation (parameters of UG). Parameters have been postulated mainly on the basis of considerations of explanatory adequacy, i.e., in order to resolve problems of mathematical and psychological feasibility raised by previous models of grammar acquisition. Thus, the classical generative model (e.g., Chomsky 1965, i.e., the so called Aspects model) viewed the Language Acquisition Device (LAD) as consisting of a set of universal principles and of an evaluation metric valuing grammars constructed by the language learner freely within the bounds posed by the universal principles. Within this model, the combined action of these two sub-modules was supposed to provide the child with the highest-valued grammar consistent with primary data. Under this view, variation is essentially what is left out of the scope of UG (i.e., of the biological principles). As Longobardi (2005) notices, this model offered the possibility of a natural, though perhaps “naïve”, explanation for the existence of variation, namely in terms of minimization of the required genetic load: the breadth of variation could be understood as the consequence of a conceivable “economy” condition, active through evolutionary history, on the possible richness of innate mental structures. This condition would limit the amount of a priori knowledge (the number of principles of UG) available to constrain the space of variation. No essential distinction is drawn between diversity due to lexical arbitrariness and grammatical variation, in this respect.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
However, the Aspects model failed to meet the requirements of feasibility as the urge towards descriptive adequacy caused the enrichment of the format of rules and of possible systems, creating unresolved tension with the restrictiveness required to attain explanatory adequacy, i.e., to account for the logical problem of language acquisition. The reconciliation of the conflicting requirements of descriptive and explanatory adequacy was attempted precisely with the Principles & Parameters approach, in which, in principle, grammatical variation is also innately given (exhaustively given, at the appropriate level of idealization), in the form of a presumably finite amount of discrete possibilities. Variability is already present at the initial state of the mind S0 in the form of open parameters, actual varieties are represented by closed parameters at the steady state SS. In this model, however, the very existence of grammatical variation is more problematic, since it is built into the biolinguistic endowment itself: yet, variation remains very wide and cannot obviously be reduced to conceptual necessity in the Minimalist sense. In other words, a new tension arises, this time between explanatory adequacy and evolutionary adequacy (the latter in the sense defined in Longobardi 2003 and highlighted in section 1.3 above): once parameters are included by a theory of UG, minimization of the genetic endowment (produced by the supposed economy constraint on the stored innate knowledge) should probably amount to minimizing the number of parameters as well, thus resulting, if anything, in a reduction, rather than in the observable extension, of the space of variation. We suggest that a possible hint toward the solution of this puzzle could come from a minimalist critique of the format of parameters in order to reveal whether they obey principled restrictions imposed by general properties of computational efficiency and conditions at the interfaces. This involves, as a first step, an operation similar to the one which induced general and pervasive principles from the wide array of transformational rules: in the same way, we believe that, once a full and coherent subsystem of parameters is extensively investigated, it is possible to single out a few general and pervasive schemata governing the form of parameters. As a second step, a minimalist critique of the format of parameters should address the problem of a principled explanation for these schemata, i.e., of the conditions which are supposed to be optimally satisfied by such a parametric system in order to meet evolutionary adequacy. 3.2 Towards parametric minimalism The only well known restriction proposed on the format of parameters so far is the conjecture, stemming from Borer (1984), that parameters are always properties of functional heads of the lexicon. Fundamentally accepting this insight as
Three fundamental issues in parametric linguistics
a point of departure, we will in what follows suggest the possibility of a more articulated restrictive theory and point out some of its desirable consequences. The MGP method, by means of which the database in Table A was produced, is able to prompt some preliminary generalizations in this sense. As a first approximation, it turns out that up to 39 out of the 47 parameters in Table A can be reduced to a form which falls into one of only 5 abstract parameter schemata, listed below, where F and X, Y are variables over features and categories: a. Is F, F a functional feature, grammaticalized?7 b. Is F, F a grammaticalized feature, checked by X, X a category? c. Is F, F a grammaticalized feature, spread on Y, Y a category? d. Is F, F a grammaticalized feature checked by X, strong (i.e., does it overtly attract X)? e. Is F, F a grammaticalized feature, checked by a category X0?8 (8)
By “grammaticalized” in (8a), we mean that the feature must obligatorily occur and be valued in a certain structure, e.g., definiteness is obligatorily marked in argument DPs in certain languages (say English or Italian), but not in others (say Russian or Latin). This does not mean that even the latter languages cannot have lexical items occasionally used to convey the semantic meaning of definiteness (presumably demonstratives can convey such a meaning in every language), but in this case the feature “definiteness” would be regarded as a lexical, not a grammatical one. (8b) asks whether a certain feature requires establishing a relation with a specific (optionally or obligatorily present) category in the structure, creating a dependency (i.e., whether it acts as a probe searching a certain syntactic space for a goal, in Chomsky’s 2001a terminology). (8c) asks if a feature which is interpreted in a certain structural position has uninterpretable counterparts on other categories which depend on it for valuation. (8d) corresponds to the traditional schema inaugurated by Huang (1982), asking whether the dependency mentioned in (8b) involves overt displacement of X, i.e., remerging of X next to F, or not. Finally, (8e) concerns the size of the goal and, also in the wake of the conclusions of Biberauer and Richards’ studies on EPP satisfaction (Richards & Biberauer 2005; Biberauer & Richards 2006), asks whether the goal may have minimal head status or phrasal status (i.e., whether some piedpiping operation is active or not). In Table A, parameters 1, 2, 8, 9, 10, 11, 12, 13, 24, 25 and 29 govern the “grammaticalization” of a given feature within the system and thus conform to the schema
. Cf. also Zeijlstra, this volume. . Cf. Biberauer and Richards (2006).
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
in (8a); parameters 15, 17, 19, 20, 23, 32, 33 and 47 instantiate the “checking” schema in (8b); parameters 3, 4, 6, 7, 18, 19, 20, 28, 34 and 44 are parameters of “spread” (schema (8c)); parameters 14, 21, 22, 35, 36, 37, 38, 39, 40, 42 and 43 are parameters of “strength” (schema (8d)).9 As concerns (8e), in Table A it is likely to be represented by parameter 26, governing raising of nouns over adjectives either as heads or just as full larger projections and presumably distinguishing most European languages from Semitic ones (cf. Fassi Fehri 1993 and Shlonsky 2004). For the time being, we leave open the question whether, considering other domains, two further schemata need to be added to such formats: one might be a schema asking whether a certain category or feature may be phonetically null in a certain situation (e.g., the well known “null pronoun” schema, as inaugurated by Taraldsen 1978 and refined in Rizzi 1986); it is perhaps the case that this sort of parametrization is often reducible to morphophonological factors and combinations of other parametric schemata above. The other potential candidate for schema status is represented by lexicosyntactic parametrization regarding the encoding of some universally definable features – say, [±pronominal], [±anaphoric], [±variable], [±definite], [±deictic] and so on – in different categories, resulting, for example, from various slightly different aggregations of features like [±head], [±maximal], [±functional], [±N], [±V], etc. into lexical items. This latter schema was most insightfully, to our knowledge, used by Sportiche (1986), to account for the peculiarities of Japanese zibun and kare as opposed to English anaphors and pronouns. However, at least in some cases, this schema could perhaps reduce to instances of (8a) and (8b).10 Let us then limit ourselves, very speculatively, to assuming the existence of only the five “core” parameter schemata in (8). From this approach it already follows that certain conceivable types of variation are excluded. For example, one theorem, a conclusion with far-reaching consequences, along the lines of (9) follows: (9) Topological Mapping Theorem: The locus of interpretation of each grammatical feature is universal, not parametrized
Empirical investigation will hopefully corroborate this restrictive conjecture.11
. In other, more theory-neutral terms, they govern the presence of an overt (rather than covert) displacement relation. . Notice that parameters such as 19 and 20 represent two cases in which the difference between the two schemata (8b) and (8c) is, in fact, neutralized, for complex empirical reasons. . Some apparently correct empirical consequences of the Topological Mapping Theorem have already been explored in Longobardi (in press), dealing with parametric variation in the mapping properties of nominal arguments.
Three fundamental issues in parametric linguistics
To exemplify the functioning of the parameter schemata system, consider the answers (10a–c) provided, on the grounds of the evidence tentatively built into the specific parametric grid of Table A, to the first three questions posed in (8) once the values for the variables over features and categories have been fixed: (10) For F = definiteness, X = nouns, Y = adjectives a. Grammaticalization of definiteness (par. 8, 9): Yes = Italian, No = Russian b. Checking of definiteness by nouns (par. 17): Yes = Rumanian, No = Italian c. Spread of definiteness on adjectives (par. 19): Yes = Arabic, No = Italian
As concerns schemata (8d) and (8e), consider (11), with F the referential feature hosted in the D category: (11) For F = D features, X = noun d. Strength of D (par. 14):12 Yes = Italian, No = English e. Minimal size of the goal category (par. 26): Yes = Italian, No = Arabic
In this way, parameter schemata of the type shown in (8) derive actual parameters, which can be literally constructed out of functional features, lexical categories, and parameter schemata, and set under usual assumptions. Along these lines, parameter systems such as Table A could be divided into a number of sub-systems, each singling out the syntactic impact of a given feature. In principle, according to the schemata, each sub-system should reflect, for the relevant feature, the possibility of its grammaticalization, the inventory of lexical categories which can check it, the inventory of lexical items which may host a copy of it, the visibility of the mechanism fulfilling the checking requirement, and the size of the moved element. In fact, as is apparent, Table A is far from trivially instantiating all the possibilities predicted by this organization, i.e., for hardly any features are all theoretically possible parameters displayed. This is due to three main factors: (12) a. Some schemata do not generate any parameter at all for a given feature because interactions among schemata and between schemata and principles block certain combinations. b. Some predicted parameters have not been included as they do not have any clearly syntactic effect, i.e., their setting appears to be reflected on the morphophonological structure of certain words, but not on syntactic computations or the syntax-semantics interface. c. Finally, many predicted parameters, although probably in existence, do not appear in Table A simply because they have not yet been satisfactorily investigated, to our knowledge, within this framework.
. Cf. Longobardi (1994, in press).
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
While factors (12b) and (12c) are the effect of methodological choices or contingent accidents of the present state of the empirical inquiry, factor (12a) is the theoretically most crucial property of a model of UG based on principles and schemata, such as we are going to propose directly, and deserves further attention on the strength of its deep consequences for the overall theory of grammatical variation. 3.3 The Principles & Schemata model If the approach sketched in section 3.2 is in its essentials correct, this may have consequences for our conception of UG itself: it becomes unnecessary to suppose that the initial state of the mind comprises highly specific parameters. Rather, it may just consist of an incomparably more restricted amount of parameter schemata, which combine with the appropriate elements of the lexicon (features and categories) under the relevant triggers in the primary data to yield both the necessary parameters (formulate the questions) and set their values (provide the answers) for each language: (13) Principles & Schemata model: UG = principles and parameter schemata. Parameter schemata at S0, closed parameters at SS
It is then conceivable that parameters which seem to end up being set to 0 (according to the formalism of Table A) in a particular I-language have actually never been present at any state in the mind attaining that I-language. The enormous number of possible core parameters depends, in principle, on the more limited numbers of functional features F and of lexical categories X, Y, combined with the tiny class of parameter schemata. We have noticed above, however, that it is not necessary for all parameter schemata to be realized for every possible functional feature and all potentially relevant categories. This follows from two factors. First, schemata inherently interact with each other: value + for schema (8a), ± grammaticalized F, is an obvious precondition for the instantiation of all the other schemata with respect to the same feature. Similarly, value + for schema (8b), ± F-checking X, is a precondition for (8d), ± strong F, and value + for schema (8d), is in turn, a precondition for (8e), ± minimal size of X. At least a substantial part of the redundancy caused by partial interaction among specific parameters, as noticed in section 2.3 and expressed by the 0s of Table A, is then simply deducible from the more general and abstract interaction among the relevant schemata to which parameters belong. For example, 6 out of the 9 implications induced by parameter 9 of Table A above on successive parameters follow from the more general dependence of the other schemata on (8a). Second, there might be specific principles of UG altogether forbidding variation of an a priori admitted format for particular combinations of features and
Three fundamental issues in parametric linguistics
categories. Descriptively claiming, for example, as is often done, that the so called EPP feature is “universally strong” amounts to preventing a widespread schema of variation among languages (presumably (8d)) from determining differences as to whether the Spec of T is overtly filled or not. 3.4 The ultimate issues of variation Accepting the Principles & Schemata model immediately determines the possibility of huge arithmetic simplification in the primitive axioms of the theory of grammatical variation: exactly as parameters were adopted as cross-constructional generalizations, significantly reducing the amount of apparently atomic points of variation, parameter schemata, in the intended sense, are more abstract, crossparametric entities, allowing further simplification of the set of primitives. At least in theory, this begins to open the possibility of providing a sensible answer to problems (7a) and (7b), because the form and amount of variation itself to be explained is drastically reduced: it will be sufficient to justify the existence of a certain parameter schema through justification (for example, via reduction to “virtual conceptual necessity”, in Chomsky’s 1995 sense) of a single parameter instantiating that schema; and, once a given parameter schema is established, it opens up the possibility of explaining the availability (ultimately, the evolutionary rise) of the whole family of parameters of the same format. But such an approach already relieves the burden of the explanation for the very existence of language diversity (issue (7b)) as well: for, within the proposed model, variation could largely be explained as in the first, pre-Principles & Parameters model. As we have just noted, once the introduction of a parameter schema into the language faculty is justified (e.g., evolutionarily explained, perhaps reducing it to conditions of efficiency on language transmission and use) for a single case, it will be admitted and will bring about ever more potential variation for all possible combinations of relevant entities of the lexicon (features and categories). This holds, unless a further particular principle of UG prohibits certain types of variation: in other words, once a schema has entered UG, reducing variation essentially requires adding to the size of LAD, exactly as in the Aspects model. The kind of explanation in terms of “economy of UG size” usable in that model can therefore be reproposed in the Principles & Schemata approach. Of course, in order for a full minimalist program to be pursued within this approach, it is necessary to show each of the parameter schemata to be indispensable, i.e., reducible to virtual conceptual necessity, or at least to be significantly related to architectural/computational properties present in other biological systems. It is perhaps premature to systematically address this crucial part of the program now, but some exemplification of the required direction of research can be provided.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
For example, (8a) could be motivated again by “economy” constraints on performance (no language could grammaticalize the full set of conceivable functional features) to be spelt out by specific research; (8c) could perhaps ultimately be related to an acquisitional strategy of formal preservation of morphological content under the pervasive diachronic phenomenon of categorial shift or reanalysis (say, of a lexical item from a class where the occurrence of certain features is interpretable to another one where it is not). This preliminary sketch should at least provide a reasonable idea of the heuristic power of the research project advocated here and of the perspective of extending Chomsky’s (1995, 2005) minimalist concerns to the parametric component of UG.
4. Parameters and historical adequacy 4.1 How are parameter values distributed in space and time? As stressed in section 1.2, parametric linguistics enjoys the possibility of fruitfully unifying the two main scientific paradigms in the history of linguistics, which in turn may result in establishing cognitive studies as a historical science as well. To do so, it needs to address precisely questions like (1c) above, an essentially unprecedented task, but one which, at this stage, seems more feasible and realistic than that of addressing (1b). Answering (1c) means searching for at least some significant correlation between the distribution of parameter values and some other known geographical or historical variable. In principle, interesting variables to correlate with parameter values in this sense may be provided by many different sources: e.g., genetic data, geographic distances, distribution of surnames and so on. In the case of languages, however, the most salient historical variable is provided by the classical phylogenetic taxonomies based on lexical evidence, which group languages into families, regarded as the product of divergence from a common protolanguage. Therefore, it is natural to focus precisely on phylogenetic classification methods and the potential contribution of parametric syntax to such issues. But there are also two additional and more general reasons, one sociological and one epistemological, to aim directly at this goal in order to establish parametric syntax, hence, indirectly, modern cognitive science, as a historical discipline: (14) a. phylogenetic hypotheses usually yield insights for neighboring sciences (cf. the so called “New Synthesis” of archaeology, linguistics, genetics …) and may thus justify a historical paradigm in syntax as a contribution to general knowledge, going beyond its intrinsic interest for linguistic theory;
Three fundamental issues in parametric linguistics
b. within linguistics proper, a certain degree of success in syntax-based phylogenetic taxonomy could be regarded as a guarantee that parametric syntax may be successfully pursued at other conceivable levels of historical investigation (cf. Longobardi 2003: 127).
Meeting these two standards enabled the classical comparative method (based on the lexicon, see below) in the 19th century to establish itself and (historical) linguistics as a full scientific paradigm. The second condition is especially relevant: it would indeed be hard to conceive a theory of etymology and of phonological change if the sound shape of the lexicon evolved in so chaotic a way as not to point out salient cues as to the phylogenetic relations of languages. Analogously, one might argue that it will only become plausible to expect success from restrictive theories of syntactic change, such as e.g., Keenan’s (1994, 2000, 2002) Inertia, as well as from attempts to establish scientific historical comparison of syntactic entities13 if the diachronic persistence of syntactic properties is indeed sufficiently robust to allow for correct phylogenetic reconstructions. 4.2 Grammatical typology and lexical comparison To elucidate the role of syntax in phylogenetic linguistics essentially implies beginning to answer a rather general question, conceptually based on distinctions introduced into the linguistic debate by Humboldt’s work and central to his theorizing about classification and historical evolution of languages.14 Let us, quite conventionally, in fact, refer to it as Humboldt’s problem, stating it as follows: (15) Is there any significant relation between grammatical and lexical comparison of languages?
In fact, almost all genealogical classifications in linguistics are traditionally centered around lexical evidence (most superficially macroscopic data), and rarely on syntactic (or, more generally, grammatical15) typology, often uncritically assumed to orthogonally crosscut language families.
. Although this does not, of course, automatically warrant the possibility of reconstructing the actual syntax of protolanguages. On this controversial issue, see the papers collected in Ferraresi & Goldbach (in press). . Cf. Humboldt (1827, 1836 passim). Also cf. Morpurgo Davies (1996: 163 ff.) for discussion. . Grammatical comparison in this sense must not be confused with comparison of the phonological encoding of functional morphemes in the lexicon, sometimes improperly called “grammatical evidence”, as e.g., in Greenberg (2000). In fact, by lexicon we simply mean a set of words, arbitrarily coding sound/meaning relations. By grammar, and more specifically by syntax, we refer very roughly to a set of rules to arrange such words into well formed sentences.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
The question in (15) may be more concretely stated as follows: (16) Given three or more languages, what is the relation between their phylogenetic tree based on syntactic comparison and the one arrived at on the grounds of lexical comparison?
Oversimplifying, one may regard the possible empirical answer as a binary choice: (17) a. Syntax and lexicon provide comparable taxonomic results b. Syntax provides taxonomic results radically at odds with those of the lexicon
We take (17a) to mean that at least significant subportions of the two trees are overlapping. Now, if this is the case, three further possibilities logically arise: (18) a. Syntax provides weaker insights, i.e., the same taxonomic results, but only up to a shallower level of chronological depth (climbing back into the past, the threshold of chance is reached “earlier” and only recent taxa can be safely recognized) b. Syntax and lexicon provide exactly the same tree c. Syntax provides stronger insights, i.e., more ancient taxa can be identified safely (the threshold of chance is reached “later”, i.e., further back in the past)
Let us now consider the properties of some already existing methods of genealogical classification. 4.3 Language taxonomy: Lexical comparison and beyond From the second half of the 19th century onwards, three different types of enterprises have attempted to classify languages and/or populations separated for centuries or millennia into historically significant families. Two have specifically dealt with linguistic evidence, the classical comparative method and Greenberg’s mass comparison. Both methods have produced hypotheses of phylogenetic relatedness by comparing sets of words and morphemes, i.e., items of the lexicon lato sensu, saliently (and crucially for the success of the methods) characterized by Saussurean arbitrariness, i.e., unpredictable pairing of sound and meaning. The classical comparative method, essentially based on the inspection of the lexicon, has provided sharp taxonomic conclusions, adducing as a proof of phylogenetic kinship the identification of phenomena which are patently beyond chance probability, such as “regular” (recurrent) sound correspondences and agreements in irregular morphology, so solving the problem of a safe choice of comparanda. Such a method, one of the greatest discoveries ever in the human sciences, embodies a sharp demarcation criterion between science and pseudo-science in etymology and historical linguistics, at the cost of narrowing the scope of inquiry
Three fundamental issues in parametric linguistics
to chronological spans in which recurrent correspondences (“sound laws”) are recognizable. It has, in fact, provided a proof of the genetic relatedness of the most obvious language families, but turns out not to work equally well for the hypotheses of long-distance grouping of such families into higher taxa, which are supposed to have existed further back in prehistory: in these cases, it is almost impossible to identify exact sound correspondences. Actually, the classical method has been mainly applied a posteriori to languages whose state of cognation was relatively easy to detect in advance of the systematic application of the method itself. Beyond the classical one, the only linguistic method so far proposed has been Greenberg’s multilateral or mass comparison (for example, Greenberg 1987, 2000). Still based on the lexicon, but allegedly fit to meet the needs of long-range comparisons, its basic idea is that the lack of exact sound correspondences can be compensated for by considering sets of lexical elements, expressing predefined meanings (e.g., Swadesh’s lists), not just in pairs of languages, but across a larger number of languages at the same time, in order to mathematically reduce the risk of chance similarity. The identification of supposed cognates is in fact based on mere phonetic and semantic resemblance, not on regularity or identity, and is therefore much less safe from chance. Epistemologically, Greenberg’s method fails to provide as sharp a demarcation criterion as one founded on recurrent sound laws: the crucial difficulty of such an approach lies in the intrinsic uncertainty of the very choice of the chosen comparanda, since it mostly remains unclear whether the compared words in different languages are in fact etymological cognates or merely accidentally similar. Obvious probabilistic questions, often of hardly manageable complexity, arise and receive controversial, but mostly skeptical, answers (cf. some discussion in Ringe 1992, 1996; cf. also Joseph & Salmons 1998, among others). As anticipated, there exists a third, fairly recent, tradition of study which is relevant for taxonomic purposes in historical anthropology: such comparative practice stems from theories of population genetics (cf. Cavalli Sforza, Menozzi, & Piazza 1994; Cavalli Sforza 1996), which have even directly entered the debate about linguistic taxonomies at least since Cavalli Sforza, Piazza, Menozzi, & Mountain (1988). This approach has been considered useful to complement linguistics in the task of classifying populations and languages, especially by longrange comparativists, facing the uncertainties and boldness of their tasks and methods. It is based on the measuring of the distance among human populations with respect to the frequency of alleles in genetic polymorphisms. Here no question arises about the choice of comparanda since they are drawn from a universal list of finite biological options: for example, a blood group must be compared to the same blood group in another population, not to other genetic polymorphisms. The only issue may concern the statistical and empirical significance of the similarities discovered.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
4.4 Parametric comparison To address the questions in (15)–(16), we need to propose a new, lexically “blind”, historical method based on inspection of syntactic evidence. The traditional skepticism concerning the possibility of comparison in historical syntax has been rooted in the idea that syntax does not provide well-defined entities to compare. The parametric model happens to dissipate such an objection, as it may in principle provide precise comparanda, namely parameter values. Recall, thus, that parametric theories try to reduce the core properties of grammatical variation to an ultimately finite set of discrete (ideally binary) choices, closed by each language learner on the basis of environmental evidence (triggers, in Clark & Roberts’ 1993 sense, or cues in Lightfoot 1991) with an operation of selection (rather than instruction, in the terminology insightfully proposed by Piattelli Palmarini 1989) between options innately predefined by UG. Therefore, a parametric model not only allows one to find precise comparanda (as stressed by Roberts 1998), but even provides an invaluable a priori advantage over one of the existing lexical methods, namely Greenberg’s mass comparison: parametric comparison is in the same position as the study of genetic polymorphisms as it compares values (alleles) of parameters (syntactic polymorphisms) in different realizations of universal grammar, i.e., it draws from a universal list of finite biological options (though culturally set); therefore no question about the legitimacy of the comparanda may ever arise. Moreover, for the same reason, even though a parametric comparison method (henceforth PCM) does not challenge the methodological perfection and reliability of the classical comparative method, it displays a potential practical advantage over the latter as well: as noted, the classical comparative method can only be fully applied to sets of languages already exhibiting sufficiently regular correspondences and thus in most cases fails to be useful as a heuristic to detect still controversial language families. By contrast, PCM can, by its very nature, in principle be applied to any set of languages as their parameter values must anyway be possible realizations of the same options, abstractly provided by UG. As in the case of population genetics, what must be ascertained in order to evaluate the effectiveness of PCM is the statistical significance of the comparisons among the relevant entities. A priori, the probative value of parametric comparison is mathematically very high: since just 30 binary independent parameters generate 230 languages, the probability of two languages coinciding in the values of 30 independent parameters with binary equiprobable values is 1/230; for three languages, it is (1/230)2, i.e., less than one in one billion billions. Of course, in the real world, one must expect languages to only partially agree in parameter values;
Three fundamental issues in parametric linguistics
therefore it becomes necessary to compute the probabilistic significance of (partial or full) agreements: as Guardiano & Longobardi (2005: 155–156) first argued, such significance is often still high enough to draw very significant conclusions. But even with respect to population genetics, parametric comparison displays some methodologically profitable features. First, comparanda are discrete entities as opposed to gene frequencies, which are the continuous entities actually measured in geneticists’ comparisons of populations: for most parameters the degree of agreement among speakers of even large communities is such that at the appropriate level of idealization we can consider entire E-languages (in Chomsky’s 1986 sense), like French, German, Wolof, and so on, as homogeneous entities, simplifying mathematical comparisons. Second, parameter settings are completely immune from natural selection on the part of the material environment, which may occasionally be at work even on some genetic markers. In fact, with the development of parameter theory and of PCM, a closer comparison will become possible between genetic and syntactic evolution factors, which seem to correspond well enough with each other: pending further investigation, we suggest that mutation, admixture, and random genetic drift (cf. Cavalli Sforza 2004) may be compared to parameter resetting, borrowing, and random exposure to accidentally different (and heterogeneous) primary corpora, respectively. Future analytical study of each such factor in close comparison to the corresponding biological one may represent the ideal starting point in the attempt to understand cultural evolution along the lines of natural evolution. We are then only left with the significant exception of natural selection, apparently absent from syntactic history, as noted. 4.5 Biological and linguistic data In order for a method of genealogical taxonomy based on syntax to be successful, it must obviously be the case that parametric values exhibit a significant degree of persistence through time. Theoretically, this is a plausible perspective. The inertial view of diachronic syntax advocated in Longobardi (2001) as a specific minimalist implementation of Keenan’s (1994) Inertia proposes that formal features may be more stable than semantic and phonological ones, as their change can only be a direct or indirect consequence of some interface innovation, or of some even more external influence on the grammar, such as borrowing (in a sense extended to include heterogeneous primary data). According to such premises, parametric syntax could be among the aspects of a language most apt to preserve traces of original identity, being more conservative than phonology or the lexicon; this is true at least with respect to those parameters which are not primarily set on the
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
grounds of phonological evidence, potentially an important class for diachronic syntax to be carefully investigated.16 Furthermore, parametric data are likely to be less “superficial” than lexical data in another sense as well: even the classical comparative method may sometimes fail to distinguish borrowing from inheritance of a lexical property, owing to phonological readjustments; in this respect, heavy syntactic borrowing from other E-languages (e.g., areally convergent parameter values) is at least less evident than lexical borrowing. Thomason and Kaufman (1988, especially ch. 3), on the basis of a large number of observations, have suggested that while lexical borrowing may take place heavily and independently of other types of interference in a variety of cultural contact situations, morphosyntactic properties, which constitute the object of parametric variation, are transmitted into another language under only two circumstances: firstly, in the case of language shift on the part of a population (language replacement, as studied in, for example, Renfrew 1987, whether accompanied by gene replacement in the community or not) or, secondly, in the case of close and centuries-long contact involving salient levels of bilingualism. Both situations, even though not strictly speaking phylogenetic in nature, are extremely significant from a historical (and especially pre-historical) point of view and thus naturally recorded by any effective method of historical linguistic analysis. After this discussion, an obvious but important remark, already stressed in Longobardi (2003), is in order: the concept of parametric data is not to be simplistically equated with that of syntactic or even “structural” data. Parameters are abstract cognitive entities often with a robust deductive structure, which capture the cross-linguistically systematic co-variation of many superficial properties by means of the formulation of a unique parameter or of the statement of complex implicational relations between different parameters. In this respect, syntactic parameters, conceived as real mental structures, are intended to entail a radically different epistemological status with respect to surface generalizations such as, for example, VO/OV or N-Gen/Gen-N, which typically subsume many distinct and unrelated types of constructions, and follow from the combinations of more abstract and independent parameters.17 Data in such a format may happen to be even less reliable and more mathematically misleading than phonetically unanalyzed lexical resemblances when employed as purported comparanda for the task of computing degrees of areal or genetic relatedness, or even of typological similarity among languages.
. On this crucial point, see the original insights in Nespor (2001), Nespor, Guasti and Christophe (1996), and Christophe, Nespor, Dupoux, Guasti, and van Ooyen (2003). Also cf. Panagiotidis (this volume) for consideration of factors that may underlie diachronically stable parameters. . Cf. Crisma (2003), Guardiano (2003), Gianollo (2005) and Crisma and Gianollo (2006), among others.
Three fundamental issues in parametric linguistics
These conclusions allow us to stress an even more suggestive analogy between linguistics and biological anthropology. In the study of human biological variation in time and space, external anthropometric traits have been replaced by genetic markers owing to a leap forward in purely theoretical science, i.e., the development of molecular biology which has made available such a new domain of more abstract and subtle entities. As Cavalli Sforza, Menozzi, and Piazza (1994: 18) put it “… the major breakthrough in the study of human variation has been the introduction of genetic markers, which are strictly inherited and basically immune to the problem of rapid changes induced by the environment”. If, as we believe, one of the major breakthroughs in the study of human cultural variation has been the rise of parametric theories in linguistics, which also have been introduced as a result of progress in theoretical science, it is conceivable that parameters may play a historical role analogous to that played by genetic markers. In this sense, we regard lexical data as resembling the external morphological characters used by more traditional biological taxonomies, while syntactic data provided by modern grammatical theories may in principle resemble precisely the molecular evidence exploited in population genetics and more recent biological classifications. 4.6 Syntactic taxonomies The study of the parameter values listed in Table A begins to provide an interesting proof of the validity of the parametrically based approach advocated here and, ultimately, a tentatively positive answer to (15). According to mathematical methods worked out in some detail in Longobardi & Guardiano (in press), it becomes possible from the values of Table A to precisely measure the syntactic distance of all the 276 language pairs generated from the 24 varieties of the sample. Such pairs are listed in Table D in the Appendix (Figure 2) in increasing order of distance. In other words, the pairs of languages at the top of the table are parametrically most similar, while those at the bottom share the fewest parametric properties. Such results can be evaluated against a number of conceptual and empirical standards. 4.6.1 Conceptual tests If PCM is to have any historical significance, it is necessary for three a priori con ditions to be satisfied: (19) a. enough pairs in Table D must be scattered in distance from extremely similar to rather different; b. the probability of chance resemblance for the most similar languages must be significantly low in absolute terms (it needs to be individual-identifying, in Nichols’s 1992 terms);
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
c. the following empirical tendencies should arise: pairs of languages known to be closely related should clearly exhibit more identities than differences (call them positive coefficients), other pairs should tend toward an equal number of identities and differences, and pairs with more differences than identities (negative coefficients) should tend toward non-existence, since a plausible historical explanation is normally available for similarities, but not for differences.
The last point, implicit in all work on historical linguistics, deserves a more formal statement as along the lines of the Anti-Babelic Principle (20) (with obvious reference to Genesis XI, 7–9) and provides an important a priori device to evaluate the reliability of any explicit method of phylogenetic taxonomy: (20) Anti-Babelic Principle: similarities among languages can be due either to historical cause (relatedness, including convergence) or to chance, while differences can only be due to chance (no one ever made languages diverge on purpose)
Given (20), a system of parametric comparison producing too large a number of negative coefficients is unlikely to correctly represent historical reality. The distribution of the pairs in Table D shows that the three conceptual criteria are all almost totally satisfied (see Gianollo, Guardiano & Longobardi 2004 for a more detailed discussion). In particular, only 5 out of the 276 pairs have a distance greater than 0.5, signifying that the differences in their coefficients exceed the identities, giving negative coefficients which violate the Anti-Babelic Principle. 4.6.2 Empirical tests The empirical way to investigate the validity of PCM is to measure the grammatical distance of three or more languages against the lexical distance in cases whose actual history is already well-known (a methodological procedure also adopted in McMahon & McMahon 2003), so trying to directly answer (15)–(16). Consider, first, that each of the 276 pairs may be tentatively classed as instantiating one of the following types of supposed historical relations: (21) a. Strong relations (non-shaded in Table D): a phylogenetic relation between a pair of languages is strong if and only if one of the languages derives from the other or both derive from a common ancestor within a time span of (presumably) at most ±4000 years.18
. Among the strong relations we tentatively included those between Finnish and Hungarian and between Arabic and Hebrew. The latter decision is based on the fact that the variety of Arabic parametrized is the Standard one, still somewhat based on Classical Arabic from the
Three fundamental issues in parametric linguistics
b. Looser relations (shaded light grey in Table D): a phylogenetic relation between a pair of languages is looser if and only if it is not strong but both languages derive from a safely established common ancestor (e.g., PIE). c. Weak relations: a relation between a pair of languages is weak if and only if the pair does not instantiate an independently safely assessed phylogenetic relation (variously shaded in Table D: relations between Semitic and IE in middle grey, relations between Uralic and Semitic-IE in dark grey, relations involving Wolof in black, e.g., Ar-It, Hu-It, Wo-It, …)
Again, the distribution of the pairs in Table D appears to support, this time empirically, the effectiveness of PCM as a device which is able to probe phylogenetic history: out of the 36 defined as strong relations, 27 are among the topmost 31 in Table D (i.e., in the area containing pairs of languages with the most identities), 30 (83.3%) occur among the top 40; out of the 103 weak relations, 27 are at the bottom of Table D (i.e., the section containing pairs of languages with the fewest shared parametric settings), with 80 (77.7%) occuring among the last 100. 4.6.3 Syntactic phylogenies A further test of the results of PCM is provided by formalized attempts to automatically generate a phylogenetic tree, relying on methods and algorithms devised in computational evolutionary biology and now increasingly applied in linguistics (cf. McMahon & McMahon 2003 and Ringe, Warnow & Taylor 2002, among others). Elaborating on the data in Table D by means of algorithms tested for this purpose in Rigon (2005) and more extensively applied in Longobardi & Guardiano (in press), the phylogenetic tree shown in the Appendix (Figure 3) was obtained. The result is encouraging: the only sharp “mistake” occurring in the tree as measured against traditionally assessed phylogenetic hypotheses is the node (taxon) merging Russian and Serbo-Croatian together and excluding Bulgarian, thus missing the South Slavic unity. Some questionable results are also obtained in the Germanic articulation, where the West Germanic unity is obscured by the system’s possible overstating of the relation between Modern English and Norwegian. This is , however, not inconceivably attributable to actual historical influence. All other taxa are either straightforwardly as expected or at least unproven but acceptable empirical hypotheses, especially in view of the fact that the program itself poses some formal limitations on the output tree: it can only generate binary branching trees and all languages are necessarily treated as “leaves” (no motherdaughter relation is admitted among the entities being classified), whatever their chronology of attestation. first millennium AD, while that representing Modern Hebrew is heavily based on the Biblical language.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
4.7 C onclusions and perspectives: Parametric linguistics as cognitive history The still preliminary evidence provided by the set of syntactic parameters of Table A is sufficient to attempt some bold, but potentially exciting speculations. The evidence collected already suggests that taxonomies based on parametric syntax may have significant historical import, and hence supports the relaxing of the traditional view of the orthogonal relation between grammatical typology and language genealogy based on lexical comparison: the comparison of large sets of parameters and the mathematically accurate investigation of the results seriously question the validity of such an enduring but unwarranted assumption of linguistic science, raising the possibility of a positive answer to (15). The impression of orthogonality between parameter values and language families is likely to be largely due to the fact that most works on parametric variation, even including Baker’s (2001) influential book, consider a relatively limited amount of parameters at one time, missing the general perspective offered by Table A. This conclusion has two consequences. First, the correctness of the phylogenetic predictions made by a parametric analysis of the languages of our sample contributes to corroborating the very belief in the reality of parameters. Second, this result can make the study of parameters of potential relevance for historical issues and of interest to the other disciplines engaged in such inquiries. But notice that moving from lexical to parametric comparison in historical linguistics is not just a shift from vocabulary to grammar, but rather to a wholly different level of abstraction in the study of historical variables, i.e., the level of postulated cognitive structures, such as parameters and parameter values, perhaps more remote from common sense than surface patterns of knowledge and behavior, but regarded by scientists as a more reliable and precise object of investigation. Therefore, through parametric linguistics, historical concerns can be reestablished among the central concerns in the study of language and in the wider paradigm of modern cognitive science and the formal methods and idealizations of the latter can be brought into the study of history.
References Baker, M. 2001. The Atoms of Language. New York NY: Basic Books. Biberauer, T. & Richards, M. 2006. True optionality: When the grammar doesn’t mind. In Minimalist Theorizing, C. Boeckx (Ed.), 35–67. Amsterdam: John Benjamins. Boeckx, C. & Piattelli Palmarini, M. 2005. Language as a natural object; Linguistics as a natural science. The Linguistic Review 22: 447–466. Borer, H. 1984. Parametric Syntax. Dordrecht: Foris.
Three fundamental issues in parametric linguistics
Braudel, F. 1985. La Mediterranée. Paris: Flammarion. Cavalli Sforza, L.L. 1996. Geni, popoli e lingue. Milano: Adelphi. Cavalli Sforza, L.L. 2004. L’evoluzione della cultura. Torino: Ed. Codice. Cavalli Sforza, L.L., Menozzi, P. & Piazza, A. 1994. The History and Geography of Human Genes. Princeton NJ: Princeton University Press. Cavalli Sforza, L.L., Piazza, A., Menozzi, P. & Mountain, J. 1988. Reconstruction of human evolution: Bringing together genetic, archeological and linguistic data. Proceedings of NAS USA 85: 6002–6006. Chomsky, N. 1955. The Logical Structure of Linguistic Theory. MS. (Published in 1975, New York NY: Plenum). Chomsky, N. 1959. A review of B.F. Skinner’s Verbal Behavior. Language 35(1): 26–58. Chomsky, N. 1964. Current Issues in Linguistic Theory. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press. Chomsky, N. 1980. Rules and Representations. New York NY: Columbia University Press. Chomsky, N. 1986. Knowledge of Language. New York NY: Praeger. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2001a. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2001b. Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20: 1–28. Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36(1): 1–22. Christophe, A., Nespor, M., Dupoux, E., Guasti, M.T. & van Ooyen, B. 2003. Reflexions on prosodic bootstrapping: Its role for lexical and syntactic acquisition. Developmental Science 6(2): 213–222. Clark, R. & Roberts, I. 1993. A computational model of language learnability and language change. Linguistic Inquiry 24(2): 299–345. Crisma, P. 2003. Genitive constructions in the history of English. In Typological Changes in the Morphosyntax of the Indo-European Languages, G. Banti, P. Di Giovine & P. Ramat (Eds). Munich: Lincom. Crisma, P. & Gianollo, C. 2006. Where did Romance N-raising come from? A parallel study of parameter resetting in Latin and English. In Romance Languages and Linguistic Theory. Selected Papers from “Going Romance” 2004, J. Doetjes & P. Gonzalez (Eds), 71–93. Amsterdam: John Benjamins. Diamond, J. 1997. Guns, Germs and Steel. New York NY: W.W. Norton & Company. Dixon, R. 1998. The Rise and Fall of Languages. Cambridge: CUP. Fassi Fehri, A. 1993. Issues in the structure of Arabic Clauses and Words. Dordrecht: Kluwer. Ferraresi, G. & Goldbach, M. (Eds). In press. Principles of Syntactic Reconstruction. Amsterdam: John Benjamins. Fodor, J.D. 2001. Setting syntactic parameters. In The Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins (Eds), 730–767. Oxford: Blackwell. Gianollo, C. 2005. Constituent Structure and Parametric Resetting in the Latin DP: A Diachronic Study. Ph.D. dissertation, Università di Pisa. Gianollo, C., Guardiano, C. & Longobardi, G. 2004. Historical implications of a formal theory of syntactic variation. Talk presented at DIGS VIII, Yale University. Greenberg, J. 1966. Some universals of grammar with particular reference to the order of meaningful elements. In Universals of Language, J. Greenberg (Ed.), 73–113. Cambridge MA: The MIT Press.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi Greenberg, J. 1987. Language in the Americas. Stanford CA: Stanford University Press. Greenberg, J. 2000. Indoeuropean and its Closest Relatives: The Eurasiatic Language Family. Stanford CA: Stanford University Press. Guardiano, C. 2003. Struttura e storia del sintagma nominale nel greco antico: Ipotesi parame triche. Ph.D. dissertation, Università di Pisa. Guardiano, C. & Longobardi, G. 2005. Parametric comparison and language taxonomy. In Grammaticalization and Parametric Variation, M. Batllori, M.L. Hernanz, C. Picallo & F. Roca (Eds), 149–174. Oxford: OUP. Huang, C.-T. J. 1982. Logical Relations in Chinese and the Theory of Grammar. Ph.D. dissertation, MIT. Humboldt, W. von. 1827. Über den Dualis. In Wilhelm von Humboldt Gesammelte Schriften, im Auftrag der (Königlichen) Preußischen Akademia der Wissenschaften, A. Leitzmann et al. (Eds). Berlin: Behr, VI.1, 1907. Humboldt, W. von. 1836. Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss auf die geistige Entwicklung des Menschengeschlechts. E. Buschmann (Ed.). Gedruckt in der Druckerei der Königlichen Akademie der Wissenschaften, Berlin. (In Wilhelm von Humboldt Gesammelte Schriften, im Auftrag der (Königlichen) Preußischen Akademie der Wissenschaften, A. Leitzmann et al. (Eds). Berlin: Behr VII.1, 1907). Joseph, B. & Salmons J.C. (Eds). 1998. Nostratic, Sifting the Evidence. Amsterdam: John Benjamins. Kayne, R. 1975. French Syntax: The Transformational Cycle. Cambridge MA: The MIT Press. Kayne, R. 2000. Parameters and Universals. Oxford: OUP. Keenan, E. 1994. Creating anaphors. An historical study of the English reflexive pronouns. MS, UCLA. Keenan, E. 2000. An historical explanation of some binding-theoretic facts in English. MS, UCLA. Keenan, E. 2002. Explaining the creation of reflexive pronouns in English. In Studies in the History of the English Language, D. Minkova & R. Stockwell (Eds), 325–354. Berlin: Mouton de Gruyter. Lazzeroni, R. 1998. La cultura indoeuropea. Bari: Laterza. Lenneberg, E. 1964. The capacity for language acquisition. In The Structure of Language, J. Fodor & J. Katz (Eds), 579–603. Englewood Cliffs NJ: Prentice Hall. Lenneberg, E. 1967. Biological Foundations of Language. New York NY: John Wiley & Sons. Lightfoot, D. 1991. How to Set Parameters. Cambridge MA: The MIT Press. Longobardi, G. 1994. Reference and proper names: A theory of N-movement in syntax and logical form. Linguistic Inquiry 25(4): 609–655. Longobardi, G. 2001. Formal syntax, diachronic minimalism and etymology: The history of French chez. Linguistic Inquiry 32(2): 275–302. Longobardi, G. 2003. Methods in parametric linguistics and cognitive history. Linguistic Variation Yearbook 3: 101–138. Longobardi, G. 2005. A minimalist program for parametric linguistics? In Organizing Grammar: Linguistic Studies for Henk van Riemsdijk, H. Broekhuis, N. Corver, M. Huybregts, U. Kleinhenz & J. Koster (Eds), 407–414. Berlin: Mouton de Gruyter. Longobardi, G. In press. Reference to individuals, Person, and the variety of mapping parameters. In Essays on Nominal Determination, A. Klinge & H. Müller (Eds). Amsterdam: John Benjamins.
Three fundamental issues in parametric linguistics
Longobardi, G. & Guardiano, C. In press. Evidence for syntax as signal of historical relatedness. To appear in Lingua. McMahon, A. & McMahon, R. 2003. Finding families: quantitative methods in language classification. Transactions of the Philological Society 101(1): 7–55. Moro, A. 2006. I confini di Babele. Milano: Longanesi. Moro, A. 2008. The Boundaries of Babel. The Brain and the Enigma of Impossible Languages. Cambridge MA: The MIT Press. Morpurgo Davies, A. 1996. La linguistica dell’Ottocento. Bologna: Il Mulino. Nespor, M. 2001. About parameters, prominence and bootstrapping. In Language, Brain and Cognitive Development. Essays in Honour of Jacques Mehler, E. Dupoux (Ed.), 127–142. Cambridge MA: The MIT Press. Nespor, M., Guasti, M.T. & Christophe, A. 1996. Selecting word order: The Rhythmic Activation Principle. In Interfaces in Phonology, U. Kleinhenz (Ed.), 1–26. Berlin. Akademie Verlag. Netz, R. 1999. The Shaping of Deduction in Greek Mathematics: A Study in Cognitive History. Cambridge: CUP. Nichols, J. 1992. Linguistic Diversity in Space and Time. Chicago IL: University of Chicago Press. Obenauer, H.G. 1977. Sprachwandel in der Syntax als Änderung im Regelapparat der Grammatik: die Beziehungen zwischen Syntax und Semantik im Rahmen der Autonomie-These der Transformations-grammatik. Unpublished research project. Piattelli Palmarini, M. 1989. Evolution, selection and cognition: From ‘learning’ to parameter setting in biology and in the study of language. Cognition 31: 1–44. Renfrew, C. 1987. Archaeology and Language: The Puzzle of Indo-European Origins. London: Jonathan Cape. Richards, M. & Biberauer, T. 2005. Explaining Expl. In The Function of Function Words and Functional Categories, M. den Dikken & C. Tortora (Eds), 115–153. Amsterdam: John Benjamins. Rigon, G. 2005. Parametric Explorations in Computational Phylogenies. Tesi di Laurea, Università di Trieste. Ringe, D. 1992. On calculating the factor of chance in language comparison. Transactions of the American Philosophical Society 82(1): 1–110. Ringe, D. 1996. The mathematics of Amerind. Diachronica 13: 135–154. Ringe, D., Warnow, T. & Taylor, A. 2002. Indo-European and computational cladistics. Transactions of the Philological Society 100(1): 59–129. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17(3): 501–557. Roberts, I. 1998. Review of Alice Harris and Lyle Campbell, Historical Syntax in Cross-linguistic Perspective. Romance Philology 51: 363–370. Shlonsky, U. 2004. The form of Semitic noun phrases. Lingua 114: 1465–1526. Sportiche, D. 1986. Zibun. Linguistic Inquiry 17(2): 369–374. Taraldsen, K.T. 1978. On the NIC, vacuous application, and the that-t filter. MS. MIT, distributed by Indiana University Linguistics Club, Bloomington, IN. Thomason, S. & Kaufman, T. 1988. Language Contact, Creolization, and Genetic Linguistics. Berkeley CA: University of California Press. Vico, G.B. 1744. Principi di una scienza nuova intorno alla comune origine delle nazioni. Napoli, 3rd ed. Watkins, C. 1995. How to Kill a Dragon: Aspects of Indo-European Poetics. Oxford: OUP. Yang, C. 2003. Knowledge and Learning in Natural Language. Oxford: OUP.
−
−
−
−
−
+
−
+
+
(16) ± art&compl +art, +compl, +14
(17) ± def. checking N +9
(18) ± def. spread on N +9
(19) ± def. on attrib. +9, (−14 or −17)
(20) ± def. on rel +9
(21) ± high dem.
(22) ± low dem. +21
(23) ± def. check. dem. +9, (−11 or +21)
(24) ± structured APs
+
+
−
(14) ± strong D
(15) ± variable person on D +14
−
0
−
(13) ± gramm. K-to-O
+
+
+
+
−
−
−
−
+
+
0
+
−
+
(12) ± gramm. count specif. +9, −11
+
+
−
+
± gramm. general def. +8
(9)
+
−
+
−
+
+
−
± gramm. anaph. def.
(8)
+
−
+
+
(11) ± gramm.count (‘null article’) −5, +9
± num/gend. on Mod +1 or +2
(7)
+
+
+
+
−
+
−
−
−
−
−
−
+
−
0
+
−
+
+
−
+
−
−e
−e
+
+
−?
+
−
−
−
−
+
+
+?
−
0
+
−
+
+
−
+
−
+
+
0+
+
+
+
0
0
0
?
?
−
0
0−
0
0−
−
−
+
−
+
+
+
+
+
+
−
+
0−
+
+
0−
0+ 0−
−
+
0
−
+
−
0
+
−
+
+
−
+
−
+
+
+
−
+
+
−
+
−
−
+
?
+
−
?
−
−
+
+
−
+
−
+
+
0+ 0+ 0+
+
+
−
+
+
−
+
−
−
+
?
+
−
?
−
−
+
+
−
+
−
+
+
0+
+
+
+
−
+
+
−
+
−
−
−
+
+
−
0
+
−
+
+
−
+
−
+
+
+
0
+
+
−
−
−
−
0
?
?
−
0
0−
−
−
+
−
+
−
+
+
0+ 0+
+
+ +
+
+
0
− −
0
+ +
− −
+ +
+ +
− −
+ +
− −
− +
+ +
+
+
−
+
−
−
−
−
0
?
−
−
0
+
−
+
+
−
+
−
+
+
0
+ +
+ +
− −
+ +
− −
− −
− −
− −
0
+
+
−
+
−
−
+
+
0
0− 0− 0−
−?e − −
−
+?
−
−
+
+
−
+
−
+
+
+
+
−?
+
−
0+
−
+
0
+
+
−
+
−
−
+
+
−
+
−
+
+
0+ 0+ 0+ 0+ 0+
+
+
+
+
+
+
+
−
−
−
+
−
+
+
0 0
0
0
0
?
?
−
+
+
−
−
−
−
+
+
−
+
−
−e
−
−
−
−
−
0
?
−
−
−
−
−
0
?
−? −?
−
−?
−
−
+
+
−
+
−
+
+
+
−
−
−
+
−
−
−
?
+
−
+
−
−
+
+
−
+
−
0
+
−
+
+
−
−
+
+
0− 0−
+
−
+
−
+
+
+
−
−
−
−
−?
− 0−
+ 0−
+
−
−
+
+
+
+
−
0 − − +?
?
?
0
+
+
−
−
−
−
+
0
+
−
−
+
−
−
−
−
+
0
−
+
0
0
0
0
0
−
?
−
0
− 0+ 0−
+
+
+
+
+
+ −
+ 0− −
+
+
−
−
−
+
+
−
+
−
+ +e −
+
−? −? 0+ 0+ 0+ +? −
+
0− 0−
0− 0−
0
0
0
?
?
−
0
0− 0−
0
0− 0−
−
−
+
−
+
+
0+ 0+ 0+ 0+ 0+ 0+ 0+ +
+
Fr Ptg Rum Lat ClG NTG Grk Got OE E D Nor Blg SC Rus Ir Wel Heb Ar Wo Hu Fin
0+ 0+ 0+
+
It Sp
(10) ± gramm. prox. def. +9
± ambiguous bare singulars +3
± num. on APs (BAs) +1 or +2
(4)
± numeral agreement +1 or +2
± num. on N (BNs) +1 or +2
(3)
(5)
± gramm. number −1
(2)
(6)
± gramm. gender
(1)
Appendix Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
+
−
+
+
0+ 0+ 0+
(28) ± context. infl. As +infl. As
(29) ± c selecting N
(30) ± prepos. Gen (vs. infl.Gen)
(31) ± free Gen (non-agr.)−30
+ +
+
(34) ± poss. marking on N
(35) ± N over rel.
− +
+
+
−
+
+
+
+
+
−
(38) ± N over low As +24, +37
(39) ± N over M2 As +25, +38
(40) ± N over high As +39
(41) ± Consistency Princ. −36 or…−40
(42) ± head initial DP
(43) ± pr. (poss.) to Gen −32 or −33
(44) ± agr. poss. +infl.As/Ds, −34
(45) ± obl. poss. form + double form
(46) ± enclitic possessives
(47) ± def ch. poss. +8 or +9, −def ch. Gen −
+
−
+
+
+
+
+
−
+
+
+
+
+
−
−
−
+
+
−
+
−
−
+
+
+
+
+
−
+
+
+
+
+
−
−e
−
0+
+
+
−
+
−
+
−
−
+
+
?
+
+
−
+
+
+
+
+
−
?
?
−
−
+
−
+
−
+
+
−
+
−
+
−
+
−
+
+
−
0
−
+
+
0
+
?
−
+
−
+
0
+
?
0− 0−
0− 0−
0− 0−
−
+
+
−
0+ 0+
0+ 0+
+
−
+
−
+
−
+
−
+
−
+
+
+
?
0−
0−
−
+
+
+
−
−
+
−
−
+
−
+
−
+
−?
+
+
+
−
+
+
−
−
+
+
+
−
+?
−
+
+
+
−
+
+
−
−
+
+
−
−
+
+
+
+
−
−
+
+
+
−
−
−
+
+
−
+
−
0− − 0−
0− + 0−
− + −?
+ +
− −
+ +
− +
−
+
+
+
−
−
−
0+ 0+ 0+ 0+
+ +
+ +
0− +
− −
− −
+ +
−
+
+
−
−
+
−
?
−
+?
+
0
+
?
0− 0−
+
0
− −
+ −
0
−
+ +
+ +
−
+
+
+
+
+
0+ 0+ 0+ 0+
−
+?
+
0
+
+
−
−
+?
+
+
+
−
0− 0− 0− 0− 0−
0− 0−? 0− 0− 0− 0− 0−
−
+
+
+
−
−
+
−
−
+
−
+
−
+
Figure 1. Table A, indicating values for 47 parameters (rows) in 24 languages (columns).
+
+
+
+
+
−
+
+
+
+
+
(36) ± N over ext.arg. −25 or−26
(37) ± N over GenO +36, (+24 or (−30 or +32))
−
−
(33) ± GenS +30 or (−31, +32)
−
−
−e −e
(32) ± GenO +30 or −31
+
+
−
+
−
+
(27) ± free APs in Mod +24, +25, −26
−
−
+
+
(25) ± ordered APs +24
(26) ± inversely ordered APs +25
−
+
+
+
−
−
+
−
−
+
−
−
−
+
+
+
−
−
−
+ +
+ 0
0
+
+
+
+
+
−
−
+
+
+
+
+
+
−
−
+
0
−
+?
+
+
+
−
0
−
+
+
+
+
−
+
−
+
−
+
+
0
+
−
+
−
+
+
0
0
−
+
+
− +
+ −
+
+
−
−
+
−
?
+
+
+
−
+
+
+
+?
−?
−?
+
−
−
0
−
0+ 0+ 0
0+ 0+ 0
0+ 0+ +
?
−
0
0
−
+
+
0?
+?
+
0
0
−
−
−
+
0+ 0+ 0+ 0+
−
0
0
−
0
− 0− 0− 0−
−
+
0
−? −? +
−
−
−
+
0
0
−
−
0+ 0+ +? +? +?
+
−
−
+? +? ?
−
−
−
− 0− −?
0− 0− 0+ −
+
+
0+ 0+ 0+ 0+ −
+
+
−
−
−
+
0− 0− +e +e 0+ 0+ 0
0− 0−
−
+
+
+
−
−
+
−
−
+
−
−
−
+
Three fundamental issues in parametric linguistics
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi
Pairs
Distance
Pairs
Distance
Pairs
Distance
SC, Rus Ir, Wel It, Ptg Sp, Ptg It, Fr OE, D ClG, NTG It, Sp It, Rum Sp, Fr Fr, Ptg E, Nor E, D Got, OE Blg, SC Blg, Rus Ptg, Blg Sp, Rum Ptg, Rum D, Nor NTG, Grk OE, E It, Blg Heb, Ar Rum, Lat Lat, ClG OE, SC OE, Rus Sp, Lat Got, D OE, Nor It, E Fr, E Ptg, E Sp, Blg Fr, Rum ClG, Grk Rum, NTG Rum, Grk Hu, Fin Rum, Got It, D Ptg, D It, Nor Ptg, Nor Lat, Got
0,03030303 0,035714286 0,045454545 0,045454545 0,068181818 0,078947368 0,0875 0,090909091 0,107142857 0,113636364 0,113636364 0,114285714 0,117647059 0,118421053 0,125 0,125 0,125 0,130952381 0,130952381 0,138888889 0,139534884 0,147058824 0,15 0,152777778 0,155172414 0,155172414 0,15625 0,15625 0,160714286 0,166666667 0,166666667 0,171428571 0,171428571 0,171428571 0,175 0,178571429 0,179487179 0,182926829 0,182926829 0,183333333 0,184210526 0,184210526 0,184210526 0,189189189 0,189189189 0,189655172
Lat, SC Lat, Rus OE, Blg Grk, Ar Nor, Blg It, Lat Ptg, Lat Rum, SC Rum, Rus NTG, SC NTG, Rus SC, Ir Rus, Ir Sp, E Got, SC Got, Rus NTG, Got Grk, Got Sp, Ir Sp, Got Sp, D Grk, SC Grk, Rus E, Wel Rum, Blg Grk, Heb Ir, Heb Rum, ClG Sp, Nor D, Blg D, SC D, Rus It, OE Ptg, OE Rum, OE It, SC It, Rus Ptg, SC Ptg, Rus Blg, Heb D, Ir OE, Ir Lat, NTG Lat, Grk Fr, Blg NTG, Blg
0,189655172 0,189655172 0,192307692 0,194444444 0,194444444 0,196428571 0,196428571 0,196969697 0,196969697 0,196969697 0,196969697 0,2 0,2 0,2 0,203125 0,203125 0,205128205 0,205128205 0,207317073 0,210526316 0,210526316 0,212121212 0,212121212 0,212121212 0,2125 0,214285714 0,214285714 0,216216216 0,216216216 0,216216216 0,216666667 0,216666667 0,217948718 0,217948718 0,217948718 0,21875 0,21875 0,21875 0,21875 0,220588235 0,222222222 0,223684211 0,224137931 0,224137931 0,225 0,231707317
It, Ir Sp, Wel Fr, Wel Ptg, Ir Lat, Blg SC, Wel Rus, Wel ClG, Ar E, Blg It, Got Ptg, Got Sp, ClG Ptg, ClG Fr, D Sp, NTG Ptg, NTG Rum, Heb ClG, Heb E, Ir It, Heb Ir, Ar Wel, Heb NTG, Ar Fr, Nor Rum, D Sp, OE Blg, Ir Sp, SC Sp, Rus Got, E D, Heb ClG, Got NTG, Heb D, Wel OE, Wel Grk, Blg NTG, OE It, Wel Fr, Ir Ptg, Wel ClG, Blg Lat, OE Nor, SC Nor, Rus It, NTG It, Grk
0,231707317 0,231707317 0,231707317 0,231707317 0,232142857 0,233333333 0,233333333 0,235294118 0,235294118 0,236842105 0,236842105 0,236842105 0,236842105 0,236842105 0,238095238 0,238095238 0,242424242 0,242424242 0,242424242 0,242857143 0,242857143 0,242857143 0,243243243 0,243243243 0,243243243 0,243589744 0,243589744 0,25 0,25 0,25 0,25 0,25 0,25 0,25 0,25 0,25 0,256097561 0,256097561 0,256097561 0,256097561 0,256756757 0,25862069 0,25862069 0,25862069 0,261904762 0,261904762
Figure 2. Table D, listing the 276 language pairs in increasing order of parametric distance.
Three fundamental issues in parametric linguistics
Pairs
Distance
Pairs
Distance
Pairs
Distance
Sp, Grk Ptg, Grk It, ClG Got, Ir Rum, Nor Got, Nor Fr, Lat Lat, Nor Lat, Ir Fr, OE Grk, Ir Blg, Wel Got, Blg Sp, Heb Ptg, Heb Wel, Ar Rum, Ir NTG, Ir OE, Fin SC, Fin Rus, Fin NTG, D Rum, E OE, Heb Fr, SC Fr, Rus Nor, Ir Grk, OE Fr, Got E, Heb Lat, E Lat, Ar Got, Wel ClG, SC ClG, Rus Grk, Wel Lat, D Fr, Heb ClG, D Rum, Wel NTG, Wel ClG, OE Lat, Heb ClG, Ir Sp, Ar Lat, Wel
0,261904762 0,261904762 0,263157895 0,263888889 0,263888889 0,264705882 0,267857143 0,269230769 0,269230769 0,269230769 0,269230769 0,269230769 0,27027027 0,271428571 0,271428571 0,271428571 0,275 0,275 0,275862069 0,275862069 0,275862069 0,276315789 0,279411765 0,279411765 0,28125 0,28125 0,285714286 0,2875 0,289473684 0,290322581 0,291666667 0,291666667 0,291666667 0,293103448 0,293103448 0,294871795 0,296296296 0,3 0,3 0,3 0,3 0,302631579 0,304347826 0,305555556 0,305555556 0,307692308
SC, Heb Rus, Heb Rum, Ar OE, Wo E, Hu Nor, Wel Blg, Ar E, SC E, Rus D, Fin Grk, D Got, Heb Lat, Fin SC, Ar Rus, Ar D, Ar ClG, Wel It, Ar Ptg, Ar Fr, NTG Fr, Grk Fr, ClG Grk, E Got, Ar OE, Hu Fr, Wo ClG, E Got, Fin Rum, Fin E, Ar Nor, Fin D, Hu OE, Ar It, Hu Fr, Ar Fr, Hu ClG, Hu Nor, Heb SC, Wo Rus, Wo E, Fin NTG, Fin ClG, Nor E, Wo Wel, Wo ClG, Fin
0,307692308 0,307692308 0,308823529 0,3125 0,3125 0,314285714 0,314285714 0,314814815 0,314814815 0,314814815 0,315789474 0,322580645 0,326923077 0,333333333 0,333333333 0,333333333 0,333333333 0,333333333 0,333333333 0,333333333 0,333333333 0,342105263 0,342857143 0,34375 0,347222222 0,35483871 0,359375 0,362068966 0,362068966 0,375 0,38 0,382352941 0,385714286 0,388888889 0,388888889 0,388888889 0,393939394 0,393939394 0,395833333 0,395833333 0,395833333 0,396551724 0,397058824 0,4 0,403225806 0,403846154
Heb, Fin It, Fin Blg, Fin Nor, Ar Got, Wo Got, Hu NTG, E D, Wo Ptg, Hu NTG, Nor It, Wo Ptg, Wo Ir, Fin Wel, Hu Blg, Hu Rum, Hu Grk, Fin NTG, Hu Grk, Nor Grk, Hu Blg, Wo Ir, Wo Grk, Wo NTG, Wo Sp, Hu Ptg, Fin Heb, Hu Sp, Wo Rum, Wo Nor, Hu Ir, Hu Ar, Fin Wel, Fin SC, Hu Rus, Hu Lat, Hu Sp, Fin Fr, Fin Ar, Hu Nor, Wo ClG, Wo Lat, Wo Wo, Hu Heb, Wo Ar, Wo Wo, Fin
0,409090909 0,410714286 0,410714286 0,411764706 0,413793103 0,414285714 0,414285714 0,416666667 0,416666667 0,418918919 0,419354839 0,419354839 0,423076923 0,426470588 0,428571429 0,430555556 0,431034483 0,432432432 0,432432432 0,432432432 0,435483871 0,435483871 0,4375 0,439393939 0,444444444 0,446428571 0,45 0,451612903 0,451612903 0,454545455 0,455882353 0,456521739 0,461538462 0,466666667 0,466666667 0,480769231 0,482142857 0,482142857 0,483870968 0,5 0,5 0,523809524 0,533333333 0,534482759 0,583333333 0,590909091
Figure 2. Continued.
Chiara Gianollo, Cristina Guardiano & Giuseppe Longobardi Finnish Hungarian Arabic Hebrew Welsh Irish NewTestGrk ClassGreek ModGreek Norwegian English German OldEnglish Gothic Portuguese Spanish Italian French Rumanian Latin Russian SerboCroat Bulgarian Wolof Figure 3. Tree produced by an application of Felsenstein’s Kitsch program (PHYLIP 3.66) to the distances in Table D.
On the syntactic flexibility of formal features Hedde Zeijlstra This paper argues that the set of formal features that can head a functional projection is not predetermined by UG, but is derived through L1 acquisition. I formulate a hypothesis that says that every functional category F is realised as a semantic feature [F] unless there are overt doubling effects in the L1 input with respect to F, in which case this feature is then analysed as a formal feature [i/uF]. In the first part of the paper, I provide a theoretical motivation for this hypothesis, and in the second, I test the proposal with a case study, namely the cross-linguistic distribution of Negative Concord (NC). I demonstrate that negation in NC languages must be analysed as a formal feature [i/uneg], whereas negation in Double Negation languages remains a semantic feature [neg] (always interpreted as a negative operator). This paves the way for an explanation of NC in terms of syntactic agreement. In the third part, I argue that the application of the hypothesis to the phenomenon of negation yields two predictions that can be tested empirically. First, I demonstrate how this hypothesis predicts that negative heads (Neg°) can be available only in NC languages; second, that independent change of the syntactic status of negative markers can trigger a change with respect to the occurrence of NC in a particular language. Both predictions are shown to be correct. Finally, I outline the consequences of the proposal presented here for the syntactic structure of the clause and also for the way in which parameters are associated with lexical items. Keywords: negation; negative concord; doubling; formal features; functional projections; redundancy
1. Introduction A central topic in the study of the syntax-semantics interface concerns the question of what exactly constitutes the set of functional projections, or more precisely, what constitutes the set of formal features that are able to project. Since Pollock’s (1989) work on the split-IP hypothesis, many analyses have assumed a rich functional structure, consisting of a UG-based set of functional heads that are present in each clausal domain (cf. Beghelli & Stowell 1997 for quantifier positions,
Hedde Zeijlstra
Rizzi 1997 for the CP domain, Zanuttini 1997a,b for negation, and Cinque 1999 for the IP domain). This approach has become known as the cartographic approach (cf. Cinque 2002, 2006; Rizzi 2004; Belletti 2004 for an overview of recent papers). Under this approach, the set of functional projections is not taken to result from other grammatical properties, but is rather taken as a starting point for grammatical analyses. An alternative view of grammar, standardly referred to as building block grammars or the what you see is what you get (WYSIWYG)1 approach (cf. Iatridou 1990; Grimshaw 1997; Bobaljik & Thrainsson 1998; Koeneman 2000; Neeleman & van de Koot 2002), takes syntactic trees to be as small as possible. In many cases, there appears to be empirical evidence for the presence of a functional projection in a particular clause, e.g., the presence of an overtly realised functional head. The main difference between the building block grammar approach and the cartographic approach (in its most radical sense) is that the presence of a particular functional projection in a particular sentence in a particular language does not, on the first approach, imply its presence in all clauses or in all languages, whereas this is the basic line of reasoning under the latter approach (cf. Cinque 1999; Starke 2004). However, the question of what exactly determines the number and distribution of functional projections remains open. An intermediate approach between cartography and building block/WYSIWYG grammars is Giorgi & Pianesi’s (1997) feature scattering approach. Giorgi & Pianesi argue that the set of formal features is universal and hierarchically ordered. All formal features may in principle project a head. Cross-linguistic differences are then accounted for by allowing different features to syncretise on one formal head, as long as their hierarchical ordering is kept intact. Under this approach, a universal set of formal features is reconciled with cross-linguistic differences in the clausal architecture. It should, however, be acknowledged that the differences between this approach and the cartographic approach are not always as clear as they might initially seem. Only in the most radical cartographic analyses (Cinque 1999; Starke 2004) does each formal feature correspond to a separate functional projection. Rizzi (1997) also allows for feature scattering, e.g., in the case of English [Force] and [Fin(iteness)] when there is no morphosyntactic argument to assume that they must host different head positions. Ultimately, both the cartographic approach and that of Giorgi & Pianesi take the set of formal features to be determined by UG. In the present paper, I propose a radically different perspective on the set of formal features, namely that
. The term WYSIWYG is adopted from Roberts & Roussou (2003); the term building block grammars from Ad Neeleman (p.c.).
On the syntactic flexibility of formal features
UG does not contain any formal features at all, but that the set of formal features (and consequently, the set of functional projections) is created during language acquisition. This paper thus departs from the idea that the set of formal features represents the set of substantive universals or the set of semantic operators. This idea has been attractive because it links cross-linguistic variation to ways of realising these semantic operators, but, as I propose in the present paper, the connection does not have to depend on the stipulated uniformity of the set of formal features. In this paper, I also argue that these operators form the basis for the set of formal features, but that formal features are derived from them, and I provide empirical and theoretical motivations for this claim. As a result, it remains possible to connect cross-linguistic variation to the different ways of expressing particular semantic operators without stipulating a set of formal features that is uniform across languages. The question of what constitutes functional projections and thus the set of formal features that are able to project is not only important for a better understanding of the syntax-semantics interface, but is also of acute relevance to the study of parameters. This is due to Borer’s (1984) assumption that parametric values are associated with properties of functional heads, a view of parametric variation that has been adopted in the Minimalist Program (cf. Chomsky 1995, 2000 and also Baker, this volume). For instance, the Wh (fronting/in situ) parameter follows from the presence of a [WH] feature on C° that either triggers movement of wh-terms to a sentence-initial position or allows them to remain in situ. However, once it is assumed that the pool of formal features in a language is not cross-linguistically identical, it becomes questionable whether parametric variation can still be tied to properties of functional heads. Under a flexible approach, the fact that wh-phrases in wh-languages do not move to sentence-initial position can be said to result from the absence of a formal [Wh] feature that would trigger such movement. Under such an analysis, a putative wh-parameter can no longer be reduced to properties of C° (note that the availability of C° might also be subject to cross-linguistic variation; languages may lack a CP entirely if there is no proper trigger/cue for it during L1 acquisition). Hence a flexible approach to the question of whether the set of formal features is uniform across languages may have farreaching consequences for theories of parametric variation. In the following section, I provide some theoretical background and present my proposal, the Flexible Formal Feature Hypothesis (FFFH), arguing that a particular feature (a negative feature or a tense feature) is analysed as a formal feature able to create a functional projection FP if and only if there are (substantial) instances of doubling effects with respect to F present in language input during first language acquisition. Thereafter, in section 3, I present a case study that illustrates how the FFFH works: negation and Negative Concord. In this section, I demonstrate that
Hedde Zeijlstra
negation is a syntactically flexible functional category: in Negative Concord languages, negation is realised as a formal feature; in Double Negation languages, it is not. In section 4, I argue that if the FFFH is correct, Negative Concord should be analysed as a form of syntactic agreement. I present the outline of such an analysis and show how it solves problems that other attempts at analysing NC have faced. In section 5, two more consequences of the application of the FFFH to negation are discussed, namely those relating to (i) the syntax of (negative) markers and (ii) patterns of diachronic change. Here I show that the FFFH makes correct predictions, thus providing empirical evidence for it. Section 6 concludes. 2. Formal features result from doubling effects In the Minimalist Program (Chomsky 1995, 2000, 2001), Lexical Items (LIs) are assumed to be bundles of three kinds of features: phonological features, semantic features and formal features. In the present paper, the distinction between formal features and semantic features is of particular interest. First, I focus on the question of what exactly the differences between formal and semantic features are. Second, the question arises of how these differences can be acquired during L1 acquisition. 2.1 Formal features As LIs consist of three different kinds of features, three different sets of features can be distinguished: the set of phonological features, the set of formal features and the set of semantic features. Following standard minimalist assumptions about the architecture of grammar, the set of formal features and the set of semantic features intersect, whereas the set of phonological features does not intersect with the other two sets. The relations between the sets are illustrated in (1).
(1) Phonological features
Formal features Semantic features
•
•
•
•
[P]
[uF]
[iF]
[S]
As the sets of formal and semantic features intersect, formal features come in two kinds: formal features that have semantic content and formal features that do not. Therefore every formal feature (i.e., every formal feature on a lexical element) has a value ±interpretable: interpretable formal features can be interpreted at LF, the
On the syntactic flexibility of formal features
interface between grammar and the (semantic) Conceptual-Intentional system; uninterpretable features do not carry any semantic content and should therefore be deleted in the derivation before reaching LF in order not to violate the Principle of Full Interpretation (Chomsky 1995). Uninterpretable features ([uF]s) can be deleted by means of establishing a checking relation with a corresponding interpretable feature [iF]. A good example of a formal feature is the person feature, a so-called ϕ-feature. It is interpretable on DPs, but uninterpretable on verbs. Therefore finite verbs must enter into a relation with a subject so that the uninterpretable person feature on the verb is checked against the interpretable feature on the subject and deleted. An example of a proper semantic feature is sex (as opposed to gender), which does not trigger any syntactic operation. No feature has to be deleted, as sex can always be interpreted. Now the following question arises: how can one distinguish an interpretable formal feature [iF] from a semantic feature [F]? From a semantic perspective the two are indistinguishable, as they have identical semantic content:
(2) ||X[iF]|| = ||X[F]||
Syntactically, there is however a major distinction between an element carrying [iF] and an element carrying [F]. The former, but not the latter is able to check uninterpretable features ([uF]’s). Checking of uninterpretable features thus forms a diagnostic test to distinguish interpretable formal features from purely semantic features. Suppose that a sentence containing a Lexical Item Y carrying a feature [uF] is grammatical. Due to the Principle of Full Interpretation (Chomsky 1995), this feature must have been checked against a feature carrying [iF]. Now suppose that a lexical Item Z is the only element in the proper licensing domain of Y that has an interpretation such that its feature must be either [iF] or [F]. In this case, it must be that Z carries [iF] and not [F] since otherwise the sentence could not have been grammatical. In other words, if a Lexical Item A is able to check a feature [uF] on B, A must carry [iF]. Hence, the occurrence of elements carrying uninterpretable features in grammatical sentences forms the only diagnostic to distinguish elements carrying interpretable formal features from elements carrying purely semantic features. However, this immediately raises the question of how it can be determined whether a particular Lexical Item (LI) carries an uninterpretable formal feature. 2.2 Uninterpretable features and doubling effects The question of how to determine whether an LI carries a formal feature [iF] or a semantic feature [F] reduces to the question of how to determine whether there is
Hedde Zeijlstra
a visible LI that carries a feature [uF].2 This question is much easier to address: LIs carrying [uF]s exhibit (at least) two properties that can easily be recognised; these have already have been mentioned above and are repeated in (3). (3) a. A feature [uF] is semantically vacuous. b. A feature [uF] triggers syntactic operations Move and Agree in order to be deleted.
At first sight, there are therefore three properties that could be harnessed to test for the presence of a feature [uF]: its semantic uninterpretability, the triggering of an operation Move and the triggering of an operation Agree. Below I argue that all three of these properties reduce to one single property: doubling. First, although a feature [uF] is meaningless, it must establish a syntactic relationship with an element that carries [iF] and that must therefore have semantic content. This is illustrated in the following example with the person feature [i/u2sg]: (4) a.
Du komm-st you come-2sg ‘You come’
German
b. [CP Du[i2SG] kommst[u2SG]]
In (4), it is shown that the information that the subject is a 2nd person singular pronoun is encoded twice in the morphosyntax: first by the choice of the subject du, second by the person marker –st on the verbal stem. Since there is only one 2nd person singular subject in the semantics of the sentence, the subject marker on the verb is meaningless. At this stage the question emerges why a certain morpheme is semantically vacuous. In cases of overt doubling, such as (4), the fact that the presence of one semantic operator is manifested twice in the morpho-syntax constitutes evidence that at least one of the two elements must carry no semantic content. But how to analyse cases in which an inflectional morpheme is the only overt marker of a semantic operator? This is, for instance, the case with pro-drop: (5) Cant-a sing-3sg ‘He/she sings’
Italian
. Visible means that the presence of this LI must be clear. This does not mean that this element must have phonological content. It suffices that its presence must be evident, with evidence potentially being supplied by overtness, and also by triggering of syntactic operations or changing the grammatical status of a sentence.
On the syntactic flexibility of formal features
According to Rizzi (1986) among many others, sentences such as (5) are considered to contain an abstract subject (pro) whose presence is marked by the subject marking on the verb. Following this line of reasoning, the subject marking on the verb is then no longer the carrier of the semantics of the 3rd person singular subject. Now, let us discuss the expression of English past tense:
(6) You walked
In (6) it is unclear whether the morpheme –ed on walk is the phonological realisation of the past tense operator. Although this may look like a natural assumption, nothing a priori excludes an analysis where the inflectional morpheme is semantically vacuous and the past tense operator is an abstract adverbial operator (whose presence is marked by the tense marker on the verb). This latter option is available since nothing requires that an LI that carries an interpretable formal feature must be phonologically realised. In the previous section, it has only been argued that [iF] is carried by a visible Lexical Item. If for instance the dependent (the element carrying a [uF]) is phonologically realised (and the sentence is grammatical), the presence of the element carrying [iF] has already been made visible and it therefore does not necessarily have to contain phonological material. Here I follow Ogihara (1995) and von Stechow (2005), who on the basis of their analyses of sequence-of-tense effects and the interaction between tense and (distributive) quantifiers, have convincingly shown that it is impossible to analyse the past tense morpheme as the past tense operator. The first option, in which -ed is the realisation of a semantic past tense operator, is ruled out. Hence -ed must be considered to carry a [uPAST] rather than an [iPAST] feature. Consequently, (6) contains a covert past tense adverbial. The examples in (4)–(5) are already an example of the syntactic operation Agree. In (4), at some point in the derivation, the verb’s [u2sg] feature is checked against a corresponding [i2sg] feature. Without an Agree relation between du and kommst, the sentence would be ungrammatical; if kommst did not have any uninterpretable person features at all, it could not have triggered an Agree relation in the first place. The same holds mutatis mutandis for the relation between the abstract subject and the finite verb in (5) and for the relation between the past tense operator and the tensed verb in (6). In (4), the semantic operator is overtly manifested more than once, a phenomenon that is known as doubling. In (5) and (6), there is only one overt marker of the subject pronoun or the past tense respectively, but there are again multiple elements visible in the morpho-syntax that correspond to a single semantic operator. Therefore, although only one marker has been realised phonologically in (5) and (6), these sentences also exhibit doubling. Agree is always a result of a doubling of features: it can only take place where both a [uF] and an [iF] are present.
Hedde Zeijlstra
Such an Agree relation is not restricted to two elements (one [iF], one [uF]), however; multiple [uF]s can also establish a relation with a single [iF]. Ura (1996) and Hiraiwa (2001, 2005) refer to this phenomenon as multiple Agree. This is illustrated in (7) below for Spanish, where the gender and number features on the noun are also manifested on the determiner and the adjective:3,4 (7) Las chicas guapas the[uFEM][uPL] girls[iFEM][iPL] pretty[uFEM][uPL] ‘The pretty girls’
Spanish
Now, let us examine the operation Move. Checking requirements of uninterpretable features always trigger movement. It follows immediately that Move should follow from doubling of features since Move is a superfunction of Agree (Move = Agree + Pied-piping + Merge). It has been argued that wh-fronting is triggered by an uninterpretable wh-feature [uWH] on C.5 By moving the wh-word, which carries an [iWH] feature, to Spec,CP, the [uWH] feature in C° can be checked against this [iWH]. This is illustrated in (8).
(8)
CP C'
Spec
Who[iWH]i havej
TP
C [uWh]
you tj seen ti
. In this example, I analyse chicas as carrying [ipl]. However, I want to keep the option open that the plural morphology on the noun is also uninterpretable and that an abstract operator is responsible for the plural semantics. . The example may yield some questions about the semantics of numerals. In an example like las dos chicas guapas (the two girls pretty) the numeral dos requires a semantically plural complement. Informally speaking, I suggest, without committing myself, that its semantics must be something like “taking two of …”. On such an approach, dos itself does not have a plural (semantic or formal) feature. Number mismatches in those cases simply follow from mismatches between the lexical semantics of dos and the plural operator, regardless of whether it is hosted on the noun or left abstract. . The question is of course why [uWH] on C° cannot be checked under Agree. This requires an additional explanation. Either one assumes the presence of an EPP feature on C° that requires C°’s Spec position to be filled, or one assumes that Agree is in fact always an instance of Move (feature movement or other movement). Nothing in this paper relies crucially on what motivates Move in this example.
On the syntactic flexibility of formal features
In (8), the wh-feature is present three times in total in the structure: as [iWH] on the wh-word, as [uWH] on C° and as a deleted [iWH] on the trace. Given that the wh-term had to be fronted, it can be determined that C° must contain an uninterpretable feature [uWH]. In other words, Move makes the presence of an uninterpretable feature [uWH] visible although this feature itself has not been spelled-out. Hence Move, too, follows from a double manifestation of the whfeature in the sentence. It of course remains an open question why the checking relation in (8) cannot be established by Agree as well. There is much on-going debate on this issue. In some recent minimalist work, it is assumed that C° in English has an additional EPP feature that is responsible for the movement. For the moment, I will not engage in this discussion. It suffices to say that Move is a superfunction of Agree and since feature doubling is a triggering force behind Agree, it is behind Move too. Consequently, Move cannot take place without being triggered by uninterpretable features. As we have seen, doubling is the driving force behind the existence of uninterpretable features. Without any doubling in natural language, no uninterpretable formal features would exist, and thus no formal features. This immediately raises the question of why there should be doubling in natural language in the first place? In this paper, I do not address this issue as it would require a paper in itself. Ultimately, however, the existence of doubling should follow from interface conditions, as should, ideally, every grammatical property. Since there does not seem to be any semantic advantage to doubling, the question calls for a phonological explanation. In Zeijlstra (2006), I analyse doubling as a result of a phonological desideratum to spell out as many functional markers as possible on one Lexical Element. For a survey of the arguments and explanation, the reader is referred to that paper. Before continuing the discussion, a few words need to be said about the following problem: the fact that [uF] is semantically vacuous seems to be at odds with the fact that it contains the information that it must be checked with a feature ([iF]) that encodes particular semantic information. This problem (an instance of Look Ahead, since apparently the properties that are visible on LF already play a role during the derivation) has been accounted for in terms of feature valuation (cf. Chomsky 2001). The difference between interpretable and uninterpretable features would be rephrased in terms of presyntactically valued versus unvalued features. Valued features are then interpretable at LF, whereas unvalued features are not. Under Agree, unvalued features are valued (without proper valuation, the derivation would crash at LF). After valuation, they are deleted, deletion being a by-product of Agree. However, the idea that features that are valued during syntax are uninterpretable at LF whereas features that are valued in the lexicon are LF interpretable is pure stipulation, and therefore it does not really solve the problem.
Hedde Zeijlstra
I argue that the problem disappears if we look more closely at what counts as syntax and what counts as semantics. In order to keep [uF] semantically vacuous, [uF] may contain only syntactic information. This information encodes that a lexical item carrying [uF] must stand in an Agree relation to an element that carries a feature [iF]. Although [iF] has semantic content, it also has syntactic content. This follows immediately from the architecture of grammar: if [iF] were to lack syntactic content, it could not have been a formal feature in the first place and it would be a purely semantic feature [F]. It is this syntactic content of [iF] that checks [uF] and causes its deletion. Since [iF] is not only a syntactic feature, but also a semantic feature (it is an element at the intersection of the two sets of features – cf. (1)), its syntactic part does not have to be deleted: [iF] has semantic content as well and therefore it does not violate the principle of Full Interpretation at LF. Finally, a word on purely semantic features. Although I adopt the architecture in (1), I am not strongly committed to the existence of semantic features in the model of grammar. A particular element is assigned a particular semantic feature [F] if it contains a semantic operator OpF in its lexical semantics and it does not get formalised, i.e., it is not analysed as [iF]. A negative quantifier (in a Double Negation language) is said to carry a feature [neg] since its lexical semantics carries a logical negation (λPλQ¬∃x[P(x) & Q(x)] for English no, for instance). Under a strong lexical feature approach, the Lexical Item no is composed of a semantic feature [neg] and many other features, such as quantificational features. A more modeltheoretical semantics approach would take [neg] to count as a label encoding that the Lexical Item is semantically negative. On such an approach, purely semantic features no longer have an ontological status. Interpretable formal features would then be purely syntactic features that also label the presence of a particular semantic operator. In this paper, I will restrict myself to the feature approach, but the analysis can also be implemented perfectly in theories without semantic features. Now we can reformulate the answer to the question asked at the beginning of this subsection: how can an [iF] be distinguished from a [F]? The answer is that whenever there is doubling with respect to F, there are (only) formal features ([iF]/ [uF]). Two features [iF] or two features [F] would result in an iterative reading that contains two semantic operators OpF.6 Following this line of reasoning, if there is . The reason why one semantic operator cannot correspond to multiple features [(i)F] is that this leads to a violation of compositionality. Each feature [(i)F] corresponds to a semantic operator OpF, and two elements carrying [(i)F] must give rise to an iterative reading. In order to avoid uninterpretable features, some scholars (Haegeman & Zanuttini 1995; De Swart & Sag 2002) have introduced absorption rules for multiple wh-words or negative quantifiers. However, these mechanisms introduce non-compositionality into the model, a step which should only be taken if it is absolutely unavoidable. This is, however, not the case here, since one still finds many
On the syntactic flexibility of formal features
no doubling with respect to F, there is no reason to assume that F is a formal feature. In those cases, every instance of F always corresponds to a semantic feature [F]. Therefore I put forward the following hypothesis:
(9) Flexible Formal Feature Hypothesis (FFFH)
a. If and only if there are doubling effects with respect to a semantic operator OPF in the language input, all features of F are formal features [i/uF]. b. If there are no doubling effects with respect to a semantic operator OPF in the language input, all features of F are semantic features ([F]).
This hypothesis, if correct, has consequences for the architecture of grammar. It rejects the idea that the set of formal features is fixed by UG, and states that every semantic operator7 can in principle be part of the syntactic vocabulary (i.e., the set of formal features) or remain exclusively within the realm of semantics. Before elaborating the proposal and its consequences in abstract terms, I first provide a case study, which shows that this hypothesis based on the idea that the set of formal features is not predetermined by UG makes correct predictions when applied to a particular empirical phenomenon: negation and Negative Concord. 3. Case study: Negation and Negative Concord Doubling with respect to negation is clearly detectable since two semantic negations always cancel each other out. If two negative elements do not cancel each other out, but yield one semantic negation, at least one of the two negative elements must be uninterpretable. This phenomenon is well described and known as Negative Concord (NC).8 One can distinguish three different types of languages with respect to multiple negation: (i) Double Negation (DN) languages, in which every morphosyntactically negative element corresponds to a semantic negation; (ii) Strict NC languages, in which multiple (clause-internal) negative elements (both negative markers and n-words9) yield only one semantic negation; and (iii) Non-strict NC languages, instances of uninterpretability in natural languages, such as subject marking on verbs. Hence adopting this mechanism does not lead to abolishing uninterpretable material. . For a discussion of what exactly constitutes the class of semantic operators, the reader is referred to von Fintel (1995); Keenan & Stabler (2003) and Roberts & Roussou (2003: ch. 5). . For a more detailed decription of NC effects, see Zeijlstra (2004). . The term n-word is due to Laka (1990) and defined in Giannakidou (2006) as elements that seem to exhibit semantically negative behaviour in some contexts, but semantically non-negative behaviour in other contexts.
Hedde Zeijlstra
where either a preverbal n-word or a preverbal negative marker establishes an NC relation with a postverbal n-word. However, a negative marker in this type of language may not follow preverbal n-words. An example of a DN language is Dutch, an example of a Strict NC language is Czech and an example of a Non-strict NC language is Italian, as is illustrated in (10)–(12) below. (10) a. Jan ziet niemand Jan sees n-body ‘Jan doesn’t see anybody’
Dutch
b. Niemand zegt niets N-body says n-thing ‘Nobody says nothing’ (11) a. Milan *(ne-)vidi nikoho Milan neg.saw n-body ‘Milan doesn’t see anybody’
Czech
b. Dnes *(ne-)volá nikdo Today neg.calls n-body ‘Today nobody calls’ c. Dnes nikdo *(ne-)volá Today n-body neg.calls ‘Today nobody calls’ (12) a.
Gianni *(non) ha telefonato a nessuno Gianni neg has called to n-body ‘Gianni didn’t call anybody’
Italian
b. Ieri *(non) ha telefonato nessuno Yesterday neg has called n-body ‘Yesterday nobody called’ c.
Ieri nessuno (*non) ha telefonato (a nessuno) Yesterday n-body neg has called to n-body ‘Yesterday nobody called (anybody)’
In Dutch, every morphosyntactically negative element corresponds to a semantic negation. These negative elements are either the negative marker niet or a negative quantifier, as illustrated below in (13)–(15). Note that the locus of the negative operator at LF does not coincide with its relative position at surface structure, but this is due to quantifier raising (which is independent of negation) in (13) or V2 in (15). Hence there are no doubling effects with respect to negation. As a result, it follows from the FFFH that negation in Dutch is not formalised (or grammaticalised): the only negative feature [neg] in Dutch is a semantic feature.
On the syntactic flexibility of formal features
(13) Jan doet niets [neg] Jan does n-thing ‘Jan does nothing’ (14)
Niemand komt [neg] N-body comes ‘Nobody is coming’
¬∃x.[thing’(x) & do’(j, x)]10
¬∃x.[person’(x) & come’(x)]
(15) Jan loopt niet [neg] Jan walks neg ‘Jan isn’t walking’
¬walk’(j)
Things are different, however, in NC languages. Let us start by discussing the Nonstrict NC language Italian. In Italian, postverbal n-words obligatorily need to be accompanied by the negative marker non or a preverbal n-word. This means that a large part of negative sentences in the L1 input consists of sentences such as (16).
(16) Gianni non ha visto nessuno [ineg] [uneg] Gianni neg has seen n-body ‘Gianni has seen nobody’
¬∃x.[person’(x) & see’(g, x)]
Since (16) contains more than one negative element, but only one negation in its semantics, only one of the negative elements can be semantically negative and the other one must be semantically non-negative. The latter element must therefore carry an uninterpretable formal negative feature [uneg], and, negation being formalised in this language, the negative operator carries [ineg] and not [neg]. Negation scopes from the position occupied by non.11 Non thus carries [ineg] and nessuno carries [uneg]. This distribution cannot be reversed, since otherwise a sentence such as (17) is expected to be grammatical, contra fact.
(17) *Gianni ha visto nessuno Gianni has seen n-body ‘Gianni hasn’t seen anybody’
The [ineg] feature on non also enables it to express sentential negation. This is shown in (18) where non functions as the negative operator.
. For reasons of readability, tense is neglected in all these readings. . This follows Aristotle/Horn in assuming that negation is not a propositional but a predicative operator.
Hedde Zeijlstra
(18)
Non ha telefonato Gianni [ineg] neg have.3sg telephoned Gianni ‘Gianni hasn’t called’
¬call’(g)
The fact that non is the carrier of [ineg] and n-words carry [uneg] seems to be problematic in one respect, namely that Italian also allows sentences such as (19). Here non is absent (and may not even be included). Hence, all overt negative elements carry [uneg].
(19)
Nessuno ha telefonato a nessuno [uneg] [uneg] N-body have.3sg telephoned to n-body ‘No-one has called anyone’
¬∃x∃y[person’(x) & person’(y) & call’(x, y)]
However, given the grammaticality and the semantics of the sentence, one element must carry the feature [ineg]. Basically, there are two ways out. One possibility is to analyse n-words as being lexically ambiguous between negative quantifiers and non-negative indefinites (cf. Herburger (2001)), but this would render (17) grammatical. The other way is to assume that negation is induced by a (phonologically) abstract negative operator (Op¬), whose presence is marked by the overt n-words. Then (19) would be analysed as follows: (20) Op¬ nessuno ha telefonato a nessuno [ineg] [uneg] [uneg] N-body have.3sg telephoned to n-body
This analysis is supported by the fact that if the subject n-word is focussed and the negative marker non is included, the sentence receives a DN reading. Hence, apart from the presence of non, a second negative operator must be at work. (21) Op¬ nessuno non ha telefonato a nessuno [ineg] [uneg] [ineg] [uneg] N-body neg have.3sg telephoned to n-body
Consequently, given the fact that not every instance of negation in Italian is semantically negative, negation is formalised and every negative element carries a formal negative feature: n-words carry [uneg] and the negative marker non and Op¬ carry [ineg]. In Czech, the application of the FFFH leads to slightly different results. First, since Czech is an NC language, negation must be formalised and n-words are attributed a feature [uneg]. However the (default) assumption that the negative marker carries [ineg] cannot yet be drawn on this basis. The negative operator could also be left abstract. Hence, the value of the formal feature of the negative marker in (22) cannot be determined on the basis of this example.
(22) Milan ne-vidi nikoho [?neg] [uneg] Milan neg.saw n-body
On the syntactic flexibility of formal features
¬∃x.[person’(x) & see’(m, x)]
In Italian we saw that non must be the negative operator, since negation takes scope from the position that it occupies. Consequently, no n-word is allowed to surface to the left of this marker (with the exception of constructions like (21)). However, in Czech n-words are allowed to occur both to the left and to the right of the negative marker. This means that negation cannot scope from the surface position of ne, since otherwise the negative operator would be (asymmetrically) ccommanded by the semantically non-negative n-word, and, contrary to fact, yield a reading in which the indefinite outscopes negation. The only way to analyse ne then, is as a negative marker that carries [uneg] and establishes a feature checking relation (along with the n-words) with a higher abstract negative operator: (23) Op¬ Nikdo ne-volá [ineg] [uneg] [uneg] N-body neg.calls
¬∃x.[person’(x) & call’(x)]
As a final consequence, single occurrences of ne cannot be taken to be realisations of the negative operator, but must be seen as an indicator of such an operator. In (24), the negative marker indicates the presence of Op¬, which in its turn is responsible for the negative semantics of the sentence. (24) Milan Op¬ ne-volá [ineg] [uneg] Milan neg.calls
¬call’(m)
Hence, in Czech even the negative marker is semantically non-negative. Czech and Italian thus differ with respect to the formalisation of negation to the extent that the negative marker in Italian carries [ineg], whereas the negative marker in Czech carries [uneg]. Note that this corresponds to the phonological status of the two markers: in Czech, the negative marker exhibits prefixal behaviour, thus suggesting that it should be treated on a par with tense/agreement morphology. Italian non is a (phonologically stronger) particle that can be semantically active by itself. 4. NC is syntactic agreement 4.1 Analysis The application of the FFFH calls for an analysis of NC as a form of syntactic agreement. Such an approach has been initiated by Ladusaw (1992) and adopted by Brown (1999) and Zeijlstra (2004). It should be noted, however, that these are not the only accounts of NC. Other accounts treat NC as a form of polyadic
Hedde Zeijlstra
quantification (Zanuttini 1991; Haegeman & Zanuttini 1996; De Swart & Sag 2002) or treat n-words as Negative Polarity Items (Laka 1990; Giannakidou 1997, 2000). In this section, I present the outlines of a theory of NC in terms of syntactic agreement. First, I demonstrate how such a theory predicts the correct readings of NC constructions. Second, I show that this approach solves several problems that other theories of NC have been facing. Although space limitations prevent me from addressing these issues in full, I argue that these arguments present robust evidence in favour of the syntactic agreement approach. Note that such an approach follows from the FFFH. Hence, I take the advantages of the syntactic agreement approach to support the FFFH as well. This support aside, I will, in the next section, discuss two further consequences that follow from the syntactic agreement approach that is driven by the FFFH, thus providing additional evidence for both. In accordance with the observations presented above, n-words are taken to be semantically non-negative indefinites that are marked syntactically by means of a feature [uneg]. Such a feature needs to stand in an Agree relation with a negative operator that carries [ineg]. Since a single negative operator may license multiple n-words, NC is thus nothing more than an instance of multiple Agree. The semantic representation of an n-word is thus as in (25). (25) ||n-Q|| = λP.[Q(x) & P(x)], where Q ∈ {Person’, Thing’, Place’, …}
The semantics of the negative operator is then as in (26), where (∃) indicates that this negative operator is able to bind free variables.12 (26) ||Op¬[ineg]|| = ¬(∃)13
A final assumption to be made here is that I allow checking relations between higher [iF]s against lower [uF]s, an assumption made by Adger (2003) among many others during the past few years. On the basis of the above-mentioned assumptions, most of them directly following from the FFFH, one is already able to analyse NC as an instance of Agree: multiple elements carrying [uneg] check their feature against a single negative . In this analysis, I take n-words to be indefinites in the Heimian sense. However, this is not required for this analysis. If n-words are taken to be existential quantifiers, the readings that come out are identical. . In the previous section I used the notion Op¬ in order to represent the abstract negative operator. Strictly speaking, (26) only represents the semantics of this abstract negative operator. However, as I have discussed before, the semantics of the covert and overt negative operator must be identical.
On the syntactic flexibility of formal features
operator that carries [ineg]. However, such an analysis would be overgeneralising as the use of the abstract negative operator is unconstrained. One cannot simply put as many abstract negative operators in the sentence as one likes. In essence, the abstract negative operator Op¬ is a regular lexical item, only with zero phonology. This zero phonology can easily be explained as a result of phonological economy conditions. Hence, the following phonological economy condition is adopted. (27) Phonological Economy: A phonologically empty negative operator may be assumed iff it prevents a derivation from crashing (Given Multiple Agree, no second Op¬ may be assumed if the first one is able to check all present [uneg] features).14
Now, the analysis facilitates explaining NC as a form of syntactic agreement and making correct predictions, both of the readings of NC constructions and their possible occurrences. Let us first discuss Italian, where non is the phonological realisation of the negative operator. A sentence such as (28) has the syntactic form in (29). Under syntactic Agree, all [uneg] features are deleted and as a result, the correct semantic reading falls out immediately, as shown in (30).15 (28) Gianni non telefona a nessuno Gianni neg calls to n-body ‘Gianni doesn’t call anybody’
Italian
(29) [Negº non[ineg] [vP Gianni a nessuno[uneg] telefona]]
. It is unclear how such a condition can be implemented as a rule in a derivational syntactic system. The presence of a [uneg] feature requires the presence of an element carrying [ineg] in the numeration; otherwise the derivation will crash. However, nothing prevents an abstract operator Op¬ carrying [ineg] from entering a numeration without any features [uneg] present. Such sentences should of course not be expected. This is not a problem for this particular analysis; it is a general problem for the inclusion of covert material. It can be solved in two different ways. Either (27) follows from a general licensing constraint on PF that allows covert operators only if their presence has been made visible somehow. Such an account, however, is problematic since Op¬ does not contain any phonological features, and is thus not legible at LF. A more promising account would be to think of (27) as a parsing rule. A hearer needs a trigger to parse the presence of Op¬. If such a trigger is absent, there is no possibility to include Op¬ in the sentence. Such an account is very much in line with Ackema & Neeleman’s (2002) and Abels & Neeleman’s (2006) parsing accounts for the ban on rightward movement. . Neglecting all tense effects.
Hedde Zeijlstra
(30)
Neg’: ¬ Nego: ¬ non
u,e[Person’(u)
& see’(e, g, u)]
vP: λx.[Person’(u) & call’(e, x, u)](g) = Person’(u) & call’(e, g, u)
DP: g
vP: λx.[Person’(u) & call’(e, x, u)
Gianni
vP: λP.[Person’(u) & P(u)](λy.call’(e, x, y)) = Person’(u) & call’(e, x, u)
λx
DP: λP[Person’(u) & P(u)] λy
nessuno
v’: λy.call’(e, x, y) vo: call’(e, x, y) telefona a
In a similar way, the correct syntax and semantics can be provided for Czech NC constructions. Here negation is introduced by Op¬ and both the n-word and the negative marker ne are checked against this negative operator, as shown in (31) (31) Milan ne-vidi nikoho Milan neg.sees n-body ‘Milan doesn’t see anybody’
Czech
(32) [vP Op¬[ineg] [vP Milan nevidi[uneg]i nikoho[uneg] ti]]16 (33)
vP: ¬
u,e[Person’(u)
& see’(e, g, u)]
vP: λx.[Person’(u) & call’(e, x, u)](m) = [Person’(u) & call’(e, m, u)]
¬ Op¬ DP: m Milan
vP: λx.[Person’(u) & see’(e, x, u) λx
vP: λP.[Person’(u) & P(u)](λy.call’(e, x, y)) = Person’(u) & see’(e, x, u)
DP: λP[Person’(u) & P(u)] nikoho
λy
v’: λy.call’(e, x, y) vo: see’(e, x, y) nevidi
. I assume that Czech ne is base-generated on V° by means of head adjunction (cf. Zeijlstra (2004) for a more detailed analysis). Therefore a functional projection NegP is not required and is thus excluded.
On the syntactic flexibility of formal features
The reader will notice that in principle all NC readings can be analysed likewise. Negation is induced either by a negative marker that carries [ineg] or by the abstract negative operator, whose presence is licensed by the economy conditions in (27). 4.2 Advantages of the syntactic agreement analysis I now demonstrate that the analysis presented above has some significant advantages over the previous analyses that view n-words either as negative quantifiers or as NPIs. Although I realise that I cannot do full justice to these approaches in the space available here, I find it necessary to illustrate how the syntactic agreement approach tackles several of the problems that these approaches have been facing. 4.2.1 N-words as negative quantifers Zanuttini (1991), Haegeman & Zanuttini (1996) and De Swart & Sag (2002) argue that n-words are semantically negative unary quantifiers. NC is obtained through a process of polyadic quantification where k n-words turn into one k-ary quantifier (a mechanism that these scholars also adopt for multiple questions). A strong prediction made by such theories is that isolated n-words always keep their negative reading. The question is then why examples such as (17), repeated as (34), are ruled out. (34) *Gianni ha visto nessuno Gianni has seen n-body ‘Gianni hasn’t seen anybody’
Note that the syntactic agreement analysis does not rule out these constructions either. In principle nessuno could be licensed by an abstract negative operator. However, given the economy conditions in (27), this Op¬ must immediately precede nessuno, i.e., be associated with VP. The reading that is then yielded is one in which sentential negation cannot result, as the event variable (introduced in the highest V position, cf. Chung & Ladusaw 2004 amongst others) is not bound by an operator that falls under the scope of negation. Such readings are pragmatically very odd and therefore hardly available. Herburger (2001) presents some examples of such sentences which can be uttered felicitously. (35)
El bébé está mirando a nada17 The baby is looking at n-thing ‘The baby is staring at nothing’ ∃e[look’(e) & Agent(e, b) & ¬∃x[thing’(x) & Patient(e, x)]]
. Example taken from Herburger (2001).
Hedde Zeijlstra
The prediction made by the syntactic agreement proposal, but not by the negative quantifier approach, namely that sentences such as (34) are most often semantically infelicitous rather than syntactically ungrammatical, is thus borne out. Note that under the negative quantifier approach, the parametric variation between NC and DN languages disappears. De Swart & Sag (2002) argue that this distinction is not a matter of grammar, but of language usage and they base their arguments on examples from English and French, in which both readings can often be available (NC being substandardly available in many English dialects). However, these languages are known to be in the process of change with respect to the DN/NC distinction and therefore arguably do not count as proper examples. In other languages, such ambiguity hardly exists. Finally, De Swart & Sag argue that resumption may only take place if two quantificational terms share a particular feature. In the case of resumption of whterms, this is the feature [WH]; in the case of negative terms, this feature is the feature [Anti-Additive]. This feature system prevents resumption of quantifiers of different types. A problem for the negative quantifier analysis is that many n-words may also occur in non-anti-additive downward entailing contexts. (36) Dudo que vayan a encontrar nada18 Doubt.1sg that will.3pl.subj to find n-thing ‘I doubt that they will find anything’
Spanish
On the present analysis, such examples can, however, be explained: despite the fact that it is not anti-additive, doubt can obtain an [ineg] feature during L1 acquisition, which is predicted by the FFFH.19 4.2.2 The Negative Polarity Item approach Another approach takes n-words to be Negative Polarity Items (NPIs), which are licensed by a (possibly abstract) negation (cf. Laka 1990; Giannakidou 1997, 2000). However, these analyses, although much closer to the syntactic agreement
. Example taken from Herburger (2001). . One could ask why downward entailing elements that are not strictly negative may still be assigned an [ineg] feature. In theory, this would enable the learning mechanism to assign [ineg] to a large number of LIs, many of them being semantically non-negative. However, I argue that this is a relic of previous stages of the language. Spanish n-words developed from regular NPIs, which were allowed in such contexts. This analysis is supported by the fact that languages in which n-words have not developed from NPIs (several Slavic languages) do not allow such constructions.
On the syntactic flexibility of formal features
approach as they both take n-words to be semantically non-negative, also face several problems. The first problem is that the distribution of standard NPIs and n-words differs. N-words may only be licensed in syntactically local domains, whereas the licensing conditions for NPIs depends more on their semantic context. NPIs, as opposed to n-words, can be licensed across the clause boundary, as shown in (37) for Greek. (37) I Ariadne dhen ipe oti idhe {*TIPOTA/tipota}20 The Ariadne neg said that saw.3sg n-thing/anything(npi) ‘Ariadne didn’t say she saw anything’
Greek
Since feature checking obeys syntactic locality constraints, this difference is immediately accounted for. It should be remarked, though, that n-words may be licensed across the clause boundary, as shown in (36). This results from the fact that the verb in subordinate clauses has a subjunctive form and subjunctive clauses are generally much weaker with respect to locality effects.21 A final issue to be discussed in this section22 concerns fragmentary answers, taken as evidence by Zanuttini in favour of the negative quantifier approach. Watanabe (2004) argues against Giannakidou’s (2000) analysis in terms of fragmentary answers. Since Giannakidou (2000) argues that n-words in Greek are semantically non-negative, she has to account for the fact that n-words in fragmentary answers yield a reading that includes a negation. She argues that this negation, expressed by dhen, is deleted under ellipsis. Hence, the assumption that n-words are semantically non-negative can be maintained. Watanabe (2004) argues that this analysis violates the condition that ellipsis may only take place under semantic identity (cf. Merchant’s 2001a notion of e-GIVENness). However, as the question does not contain a negation, it may not license ellipsis of the negative marker dhen. Under the syntactic agreement analysis proposed here, dhen is taken to be semantically non-negative (carrying [uneg]), and thus the identity condition is again met. The abstract negative operator then induces the negation in the answer. Note that the negative marker in Non-strict NC languages never follows
. Example taken from Giannakidou (2000): 470. The emphasised TIPOTA is an n-word, and the unemphasised tipota is a plain NPI. . Giannakidou (2000) takes n-words (in Greek) to be universal quantifiers and derives the locality effects from the locality effects of Quantifier Raising. . Another argument often used against the approach that takes n-words to be indefinites/ existentials is the fact that n-words may not be modified by almost. See, however, Penka (2006) who presents a number of arguments that invalidate this test.
Hedde Zeijlstra
an n-word, and therefore no negative marker can be deleted under ellipsis in the first place. (38) a. Q: Ti ides? A: [Op¬ [TIPOTA [dhen[uneg] ida]]] What saw.2sg? N-thing neg saw.1sg ‘What did you see?’ ‘Nothing!’ b. Q: ¿A quién viste? A: [Op¬ [A nadie [vió]]] To what saw.2sg? To n-thing saw.1sg] ‘What did you see?’ ‘Nothing!’
Greek
Spanish
It follows that the syntactic agreement approach, which is derived directly from the FFFH, accounts correctly for NC, and solves many of the problems that the other approaches struggle with. In the following section, I discuss several more predictions made by the FFFH, which turn out to be correct. 5. Consequences The FFFH and the exact analysis of NC in terms of syntactic agreement make several predictions that I discuss in this section. First, I argue that the status of the negative feature (formal or semantic) has some consequences regarding the appearance and distribution of the negative projection (NegP after Pollock 1989). Second, I show that the FFFH makes correct predictions about the consequences of diachronic change with respect to the obligatoriness or optional occurrence of the negative marker. 5.1 Negative features and projections Let us now examine the relation between the formal status of negative features and the syntactic status of negative markers. Negative markers surface in different forms. In some languages (e.g., Turkish) the negative marker is part of the verbal inflectional morphology; in other languages, the negative marker is a bit “stronger”. The Czech particle ne is an independent, but phonologically weak element, while Italian non is a phonologically strong particle.23 German nicht, on the other hand, is even too strong to be a particle and is standardly analysed as an adverb. The examples in (39)–(41) illustrate.
. I refrain from the discussion whether Czech ne should be analysed as a clitic, a prefix or as a particle. It will become clear from the following discussion that the outcome would not be relevant for the final analysis in terms X°/XP status.
On the syntactic flexibility of formal features
(39) John elmalari ser-me-di24 John apples like.neg.past.3sg ‘John doesn’t like apples’ (40) a. Milan ne-volá Milan neg.calls ‘Milan doesn’t call’ b. Gianni non ha telefonato Gianni neg has called ‘Gianni didn’t call’ (41) Hans kommt nicht Hans comes neg ‘Hans doesn’t come’
Turkish (affixal) Czech (weak particle) Italian (strong particle) German (adverbial)
Note also that it is not mandatory that a language should have only one negative marker. Catalan has a strong negative particle no and an additional optional negative adverbial marker (pas) whereas in West Flemish the weak negative particle en is only optionally present, next to the standard adverbial negative marker nie. Standard French even has two obligatory negative markers (ne … pas). The relevant data are presented in (42). (42) a. No serà (pas) facil neg be.fut.3sg neg easy ‘It won’t be easy’ b. Valère (en) klaapt nie Valère neg talks neg ‘Valère doesn’t talk’ c.
Jean ne mange pas Jean neg eats neg ‘Jean doesn’t eat’
Catalan
West Flemish
French
I adopt the standard analysis that negative affixes and weak and strong negative particles should be assigned syntactic head (X°) status, whereas negative adverbials are specifiers/adjuncts, thus exhibiting XP status (cf. Zanuttini 1997a,b; Rowlett 1998; Zanuttini 2001; Merchant 2001b; Zeijlstra 2004). The difference between X° and XP markers affects the projection of functional structure. X° negative markers must (by definition) be able to project, yielding a clausal position Neg°. By contrast, XP negative markers may occupy the specifier position of a projection that is projected by a (possibly abstract) negative head
. Example from Ouhalla (1991), also cited in Zanuttini (2001).
Hedde Zeijlstra
Neg°, Spec,NegP (as is the standard analysis for most adverbial negative markers), but this is not necessarily the case. An XP-negative could also be an adverbial negative marker that occupies an adjunct/specifier position of another projection, for instance a vP adjunct position. In that case, it is not necessary that there should be a special functional projection NegP present in the clausal structure. Now the question follows: when is a negative feature able to project? Giorgi & Pianesi (1997) addressed this problem in terms of their feature scattering principle, arguing that each feature can project a head. However, given the modular view of grammar in which features are divided into different classes (cf. (1) above), the question emerges of what kinds of features can head a projection. One would clearly not argue that every lexical semantic feature or every phonological feature might have its own projection. Feature projection is a syntactic operation, and should thus only apply to material that is visible to syntax. Hence, the most straightforward hypothesis is that only formal features can project. This means that a feature can only head a projection if it is a formal feature [i/uF]. Consequently, it follows immediately that the availability of a negative projection NegP in a particular language depends on the question whether negation is a formal feature [i/uneg] in this language. This leads to the following prediction: only in languages that exhibit doubling effects with respect to negation (i.e., only in NC languages) may NegP be available. This prediction can easily be tested, as it has been argued above that X° negative markers occupy a Neg° position, whereas adverbial negative markers do not have to occupy a Spec,NegP position. Therefore we expect that one can only find negative markers X° in the class of NC languages – see see (43)). (43) a.
NC:
[u/i]/[X]
[u/i]
X
b.
Non-NC: []
[X] [X]
In Zeijlstra (2004), this prediction is tested in three distinct empirical domains – a sample of 267 Dutch dialectal varieties, a sample of 25 historical texts, and a set of 25 other languages from different families – and it is shown to be correct.25 Empirical evidence for the FFFH has thus been provided. . Two kinds of exceptions have been found. First, Standard English, being a non-NC language, allows for the negative marker n’t, which behaves like a negative head. Possibly this is related to the fact English is on its way to transforming itself into an NC language (cf. Zeijlstra 2004). English negation can be said to exhibit doubling effects, as it may trigger movement (negative inversion). Alternatively, Williams (1994) and Zwicky & Pullum (1983) have suggested that the combination of an auxiliary + n’t is lexicalised. Then n’t does not behave like a syntactic head (thanks to an anonymous reviewer for pointing this out to me). Second, a number of South-
On the syntactic flexibility of formal features
5.2 Negation and diachronic change Since Jespersen (1917), it has been known that a large majority of languages have undergone changes with respect to the expression of negation. These changes concern both the syntax of the negative marker and the occurrence of NC. It follows from the previous subsection that these two phenomena are not unrelated. In this subsection, I first discuss how the FFFH applies to Spanish’s development from a Strict NC into a Non-strict NC language. Second, I exemplify the change from Dutch from an NC language into a DN language. 5.2.1 Spanish: From Strict NC to Non-strict NC Old Spanish was a Strict NC language, where a subject n-word was allowed to precede the negative marker no, as is shown for 11th century Spanish in (44).26 (44) Qye a myo Cid Ruy Diaz, que nadi no diessen posada 11th Cent. That to my Lord Ruy Diaz, that n-body neg gave lodging Spanish ‘that nobody gave lodging to my lord Ruy Diaz’27
Given the fact that the language input during L1 acquisition contained expressions of the form in (44), the negative marker was assigned a formal feature [uneg]. However, at some point, speakers began to omit the negative marker no in constructions such as (44), analysed as (45). This change is not surprising since the negative marker in these constructions did not contribute to the semantics of the sentence (the fact that there is an abstract negative marker located in a higher position than nadi follows from the presence of this subject n-word). Hence, the L1 input had the form of (46) with an increasing relative frequency of instances of (47). At a certain point, the absence of cases of no following nadi was thus sufficiently robust for the cue that forces the language learner to assign no the feature [uneg] to disappear. As a result, if no appeared in an NC construction, it was always the highest element. Therefore no got reanalysed as [ineg], leading to the judgements in (48). Following standard semantics of negation, the scope of negation is c-command. Therefore it is always the highest negative (overt or covert) element that must carry [ineg]. This reinterpretation of no is thus correctly predicted by the FFFH. east Asian languages lack n-words. In those languages however, it can be shown that negative markers trigger Move, thus exhibiting a doubling effect as well. . For an overview of the development of Spanish negation, see Herburger (2001) and references therein. See Martins (2000) on negation in Old Romance in general. . Example taken from Herburger (2001).
Hedde Zeijlstra
(45) Op¬ nadi no [ineg] [uneg] [uneg] N-body neg (46) Op¬ nadi (no) [ineg] [uneg] [uneg] N-body neg (47) Op¬ [ineg]
nadi [uneg] N-body
(48) a. No vino nadie neg came n-body ‘Nobody came’
Modern Spanish
b. Nadie (*no) vino N-body neg came ‘Nobody came’
5.2.2 Dutch: From NC to DN Similar observations can be made for Dutch. Middle Dutch was a language that used two negative markers en/ne … niet to express sentential negation, as shown in (49). However, in most cases which contained an n-word, only the preverbal negative marker en/ne was present, as in (50). (49) Dat si niet en sach dat si sochte28 That she neg neg saw that she looked.for ‘That she didn’t see what she was looking for’
Middle Dutch
(50) Ic en sag niemen I neg saw n-body ‘I didn’t see anybody’
Middle Dutch
As in most languages exhibiting two negative markers, one of them disappears in the course of time. 16th and 17th century Holland Dutch in most cases left out the preverbal negative marker en/ne, and exhibited only niet. As a consequence of this development, the presence of en/ne also lost ground in constructions with n-words, resulting in expressions like (51). (51) Ic sag niemen I saw n-body ‘I didn’t see anybody’
. Lanceloet 20042.
17th Cent. Dutch
On the syntactic flexibility of formal features
Hence, the language input contained fewer and fewer constructions like the ones in (52) and more and more expressions in which an n-word was the only negative element in the sentence. As the cue to assign n-words a [uneg] feature gradually disappeared, n-words were no longer analysed as carrying [uneg], but got reanalysed as carrying [neg], as shown in (53).29 (52) a.
Op¬ en niemen [ineg] [uneg] [uneg] neg n-body
b. Op¬ niemen en [ineg] [uneg] [uneg] n-body neg (53) Ic sag niemen [neg] I saw n-body
To conclude, the two developments described above show exactly how a change in the syntax of negative markers leads to a change in the interpretation of multiple negative expressions. Note that these latter changes follow directly from the FFFH and that no other additional account has to be adopted. 6. Conclusions In this paper, I first argued on theoretical grounds that the set of formal features, i.e., the set of features that can head a functional projection, is not provided by UG, but is a result of L1 acquisition. Only those semantic features that exhibit visible doubling effects are formalised (or grammaticalised). This has been formulated in the FFFH. Consequently, as only formal features can project, the number of functional projections FP that a particular grammar has at its disposal is limited by the FFFH. Each grammar makes a particular selection of semantic operators that can be realised as FPs based on the language input available during L1 acquisition. Thus clausal structure is subject to cross-linguistic variation and does not reflect a uniform UG-based template. In the second part of this paper, I applied the FFFH to the domain of negation. Negation is a semantic operator that differs cross-linguistically in the way it surfaces in morphosyntax. Languages differ with respect to whether they exhibit doubling effects (known as NC) and thus the result of this application is that negation . Similarly, the negative marker niet was also analysed as having a [neg] feature.
Hedde Zeijlstra
is only formalised in NC languages. In DN languages, negation is not realised as a formal feature. The consequences of the flexible formal status of negation are empirically testable. Not only do they call for an analysis of NC in terms of syntactic agreement (cf. Zeijlstra 2004, where I show that such an analysis solves many problems that other analyses have been facing). They also make correct predictions about the syntactic status of negative markers, and about the diachronic relation between the syntax of negative marker(s) and the occurrence of NC. First, it is shown that only NC languages may exhibit a negative marker Neg°. Second, it follows that if the (optional) negative marker for independent reasons ceases to occur in particular contexts, this may influence the visible doubling effects and therefore alter the status of the language as a (Strict) NC language. Furthermore, the proposal presented above permits the formulation of predictions in terms of typological implications, which can be tested empirically. This is an interesting result, as it has been questioned whether typological implications count as linguistic evidence.30 I hope to have shown in this paper that typological implications can be used as a testing mechanism for different proposals concerning the status of formal features. Of course, the FFFH is still programmatic in nature. It seems to make correct predictions for negation, but it should be evaluated for a number of other functional categories in order to determine its full strength. It does, however, seem that the evidence presented in this paper sheds more light on how the syntactic vocabulary is created. Finally, as mentioned in the introduction, the adoption of hypotheses such as the FFFH has serious consequences for the conjecture that parametric variation can be reduced to different properties of (functional) heads. In the sections above, strong evidence has been put forward that the negative feature is only formal in some languages. DN languages lack such a formal feature [i/uneg] and therefore never feature a negative head Neg°. Consequently, the NC parameter (±NC) can never be tied down to a value of the formal feature [neg] associated with Neg°. The parametric variation with respect to multiple negation lies one level higher, namely whether or not the semantic operator negation is formalised. Hence, the NC parameter can be reduced to a semantic feature, but not to a syntactic feature. The NC parameter is thus a result of the fact that negation may, but does not have to be formalised, a result of the FFFH. Note that not all parameters follow directly from the FFFH. The Strict vs. Non-strict NC parameter can still be reduced to the
. See Newmeyer (2004) who recently reopened the debate.
On the syntactic flexibility of formal features
[i/u] value of the formal feature [i/uneg] on Neg°. However, the very existence of such a “subparameter” again follows from the FFFH: without its application, no Neg° is available in the first place. If this line of reasoning turns out to be correct, many parameters may be derivable from the FFFH, removing these from UG, much in the same way as the set of formal features. Obviously such a prediction needs to be evaluated for a large number of parameters, but even if it turns out to be incorrect for a number of parameters, it still holds for the NC parameter that it can be derived from L1 acquisition and thus should not be thought of as a linguistic primitive.
References Abels, K. & Neeleman, A. 2006. Universal 20 and the LCA. Ms UCL and CASTL. (http:// ling.auf.net/lingBuzz/000279) Ackema, P. & Neeleman, A. 2002. Effects of short-term storage in processing rightward movement. In Storage and Computation in the Language Faculty, S. Nooteboom, F. Weerman & F. Wijnen (Eds), 219–256. Dordrecht: Kluwer. Adger, D. 2003. Core Syntax: A Minimalist Approach. Oxford: OUP. Beghelli, F. & Stowell, T. 1997. Distributivity and negation. The syntax of each and every. In Ways of scope taking, A. Szabolcsi (Ed.), 71–107. Dordrecht: Kluwer. Belletti, A. (Ed.). 2004. Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3. Oxford: OUP. Bobaljik, J. & Thrainsson, H. 1998. Two heads aren’t always better than one. Syntax 1: 37–71. Borer, H. 1984. Parametric Syntax: Case Studies in Semitic and Romance Languages. Dordrecht: Foris. Brown, S. 1999. The Syntax of Negation in Russian. Stanford CA: CSLI. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step. Essays in Honor of Howard Lasnik, R. Martin, D. Michael & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Cinque, G. 1999. Adverbs and Functional Heads. Oxford: OUP. Cinque, G. (Ed.). 2002. Functional Structure in DP and IP. The Cartography of Syntactic Structures, Vol. 1. Oxford: OUP. Cinque, G. (Ed.). 2006. Restructuring and Functional Heads. The Cartography of Syntactic Structures, Vol. 4. Oxford: OUP. Chung, S. & Ladusaw, B. 2004. Restriction and Saturation. Cambridge MA: The MIT Press. De Swart, H. & Sag, I. 2002. Negative concord in Romance. Linguistics and Philosophy 25: 373–417. von Fintel, K. 1995. The formal semantics of grammaticalization. Proceedings of NELS 25: 175–89.
Hedde Zeijlstra Giannakidou, A. 1997. The landscape of polarity items. Ph.D. dissertation, University of Groningen. Giannakidou, A. 2000. Negative … concord? Natural Language and Linguistic Theory 18: 457–523. Giannakidou, A. 2006. N-words and negative concord. In The Syntax Companian, Vol. 3, M. Everaert, R. Goedemans & H. van Riemsdijk (Eds), 327–391. London: Blackwell. Giorgi, A. & Pianesi, F. 1997. Tense and Aspect: From Semantics to Morphosyntax. Oxford: OUP. Grimshaw, K. 1997. Projections, heads and optimality. Linguistic Inquiry 28: 373–442. Haegeman, L. & Zanuttini, R. 1996. Negative concord in West Flemish. In Parameters and Functional Heads. Essays in Comparative Syntax, A. Belletti & L. Rizzi (Eds), 117–179. Oxford: OUP. Hiraiwa, K. 2001. Multiple agreement and the defective intervention effect. In The Proceedings of the MIT-Harvard Joint Conference (HUMIT 2000), O. Matsushansky et. al. (Eds), 67–80. Cambridge MA: MITWPL. Hiraiwa, K. 2005. Dimensions in syntax. MS, University of Tokyo. Herburger, E. 2001. The negative concord puzzle revisited. Natural Language Semantics 9: 289–333. Iatridou, S. 1990. About AgrP. Linguistic Inquiry 21: 421–59. Jespersen, O. 1917. Negation in English. Copenhagen: A.F. Høst. Keenan, E. & Stabler, E. 2003. Bare Grammar. Lectures on Linguistic Invariants. Stanford CA: CSLI. Koeneman, O. 2000. The Flexible Nature of Verb Movement. Ph.D. dissertation, Utrecht University. Ladusaw, W.A. 1992. Expressing negation. In Proceedings of SALT II, C. Barker & D. Dowty (Eds), 237–259. Cornell NY: Cornell Linguistic Circle. Laka, I. 1990. Negation in Syntax: On the Nature of Functional Categories and Projections. Ph.D. dissertation, MIT. Martins, A.M. 2000. Polarity Items in Romance, Underspecification and Lexical Change. In Diachronic Syntax: Models and Mechanisms, S. Pintzuk, G. Tsoulas & A. Warner (Eds), 191–219. Oxford: OUP. Merchant, J. 2001a. The Syntax of Silence. Oxford: OUP. Merchant, J. 2001b. Why no(t). MS. University of Chicago. Neeleman, A. & van der Koot, H. 2002. The configurational matrix. Linguistic Inquiry 33: 529–574. Newmeyer, F. 2004. Typological evidence and universal grammar. Studies in Language 28: 527–548. Ogihara, T. 1995. The semantics of tense in embedded clauses. Linguistic Inquiry 26: 663–679. Ouhalla, J. 1991. Functional Categories and Parametric Variation. London: Routledge. Penka, D. 2006. Almost there: The meaning of almost. In Proceedings of Sinn und Bedeutung 10, C. Ebert & C. Endriss (Eds), 275–286. Berlin: ZAS. Pollock, J.-Y. 1989. Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20: 365–424. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–557. Rizzi, L. 1997. The fine structure of the left periphery. In Elements of Grammar. Handbook in Generative Syntax, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rizzi, L. 2004. The Structure of CP and IP. The Cartography of Syntactic Structures, 2 Vols. Oxford: OUP.
On the syntactic flexibility of formal features
Roberts, I. & Roussou, A. 2003. Syntactic Change. A Minimalist Approach to Grammaticalisation. Cambridge: CUP. Rowlett, P. 1998. Sentential Negation in French. Oxford: OUP. Starke, M. 2004. On the inexistence of specifiers and the nature of heads. In Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3, A. Belletti (Ed.), 252–268. Oxford: OUP. Stechow, A. von. 2005. Semantisches und morphologisches Tempus: Zur temporalen Orientierung von Einstellungen und Modalen. Neue Beiträge zur Germanistik 4: 3–6. Ura, H. 1996. Multiple Feature-Checking: A Theory of Grammatical Function Splitting. Ph.D. dissertation, MIT. Watanabe, A. 2004. The genesis of negative concord. Linguistic Inquiry 35: 559–612. Williams, E. 1994. Thematic Structure in Syntax. Cambridge MA: The MIT Press. Zanuttini, R. 1991. Syntactic Properties of Sentential Negation. Ph.D. dissertation, University of Pennsylvania. Zanuttini, R. 1997a. Negation and Clausal Structure. A Comparative Study of Romance languages. Oxford: OUP. Zanuttini, R. 1997b. Negation and verb movement. In Elements of Grammar: Handbook in Generative Syntax, L. Haegeman (Ed.), 214–245. Dordrecht: Kluwer. Zanuttini, R. 2001. Sentential negation. In The Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins (Eds), 511–535. London: Blackwell. Zeijlstra, H. 2004. Sentential Negation and Negative Concord. Ph.D. dissertation, University of Amsterdam. Zeijlstra, H. 2006. Functional structure, formal features and parametric variation: consequences of conflicting interface condition. MS, University of Tuebingen. Zwicky, A. & Pullum, G. 1983. Clitisazition vs. inflection: English n’t. Language 59: 502–13.
Historical source Lanceloet – Jonckbloet, W.J.A. (Ed.). 1846–1849. Roman van Lancelot (XIIIe eeuw). Naar het (eenig-bekende) handschrift der Koninklijke Bibliotheek, op gezag van het gouvernement uitgegeven. ’s-Gravenhage. In F. van Oostrom (Ed.). CD-ROM Middelnederlands. Den Haag: SDU.
Expletives, datives, and the tension between morphology and syntax Richard S. Kayne Expletive there and its closest counterparts in other languages are not expletives in Chomsky’s sense (merged directly into a sentential Spec position). They are instead instances of deictic elements originating within their associate (in a way that has something in common with Moro 1997 and, more so, with Sabel 2000, though the proposal made here ties expletive there more closely to other instances of there than do these earlier works). The present proposal sheds new syntactic light on the definiteness effect, as well as on the comparative syntax of possessives and existentials. In some languages, e.g., Paduan, an expletive can be “imported” from existential (sub)sentences into possessive (sub)sentences and from there into dative sentences, giving the impression in the last case of syncretism between deictic clitic and dative clitic, but the correct, more syntactic and less morphological, analysis is that in such languages the deictic clitic is not itself a dative clitic, though it cooccurs with a dative clitic that is silent. Keywords: syncretism; expletive there; definiteness effect; existentials; clitics
1. Introduction Pronominal clitics of the sort found in Romance languages are in many ways obviously part of the syntax of those languages. Yet certain aspects of their behavior can lead to proposals more morphological than syntactic in character, as seen in Perlmutter’s (1971) templatic approach to the question of clitic ordering and clitic combinations. In this paper, I examine a somewhat different aspect of Romance clitics, one for which a more morphological, less syntactic approach might again come to mind. I will argue, however, in favor of an analysis that, using familiar syntactic notions, ties the phenomenon in question more tightly and more fruitfully, I think, to other aspects of Romance (and universal) syntax.1
. The present paper corresponds in varying degrees to the first part of one presented at the Stony Brook Workshop on Romance Clitics (May 2005), at the Cambridge University Graduate
Richard S. Kayne
In the course of so doing, I will be led to reanalyze the status of expletives such as English there and its Romance counterparts, proposing in effect that they are not true expletives. Rather, they originate within their so-called “associate”, in a way that has something in common with Moro (1997) and, more so, with Sabel (2000), though the proposal I will make ties expletive there more closely to various other instances of there than do these earlier works. 2. North Italian ghe A clitic that will have an important role in what follows is the ghe found in many North Italian dialects, in particular in the Veneto area (Padua, Venice, etc.). The behavior of this ghe is different in certain respects from that of comparable clitics in French and Italian. French has a locative clitic y seen in: (1) Jean y a mis le livre. Jean there has put the book ‘Jean has put the book there.’
This y is a clitic in the familiar sense. It cannot be coordinated, or modified, or contrastively stressed. It occurs to the left of the finite verb or auxiliary, in contrast to non-clitic locatives, which follow the verb. In French, this y is distinct from the third person dative clitic, which is lui in the singular and leur in the plural, for example: (2) Jean lui a donné le livre. Jean him/her.dat has given the book ‘Jean has given him/her the book.’
A parallel contrast between locative clitic and third person dative clitic holds for Italian: (3) Gianni ci ha messo il libro. Gianni here has put this book ‘Gianni has put this book here.’
Linguistics Conference (March 2006) and GLOW, Barcelona (April 2006) and in talks at the University of Padua and the University of Siena (March 2006). The proposal on expletive there goes back to talks at NYU (September 2000) and especially at Sophia University, Tokyo (November 2002), as well as to a series of lectures at Leiden University (May/June 2003). Since Perlmutter (1971), many authors have taken a more syntactic (and more fruitful) approach to clitic ordering and clitic combinations; for particularly striking results concerning Romanian, see Savescu Ciucivara (2006).
Expletives, datives, and the tension between morphology and syntax
(4) Gianni gli ha dato il libro. Gianni him/them.dat has given the book ‘Gianni has given him/them the book.’
with ci the locative clitic and gli the dative (le for feminine gender, in the singular). Many Veneto dialects, on the other hand, have a clitic ghe that seems to cover both locative and third person dative, for example, in Paduan: (5) Ghe meto el libro. there I-put the book ‘I put the book there.’ (6) Ghe dago el libro. there I-give the book ‘I am giving him/her/them the book.’
The use of ghe in (6) is systematic for Paduan – there is no distinct dative clitic form for third person, singular or plural.2 The question is how to think of this kind of syncretism. A highly morphological approach might be tempted to say that it is merely a case of two pronominal clitics that happen to be spelled out in the same way. Yet Paduan third person pronouns typically have, like those of French and Italian, an -l- (sometimes pronounced as a y-glide, sometimes not pronounced), and never otherwise have a g(h)- (the h in ghe is just orthographic). The more syntactic approach that I will now develop will try, instead, to find a principled reason for the appearance of an apparently locative clitic in dative sentences like (6). 3. There and ghe as deictics As a first step, it is necessary, I think, to move away from the terminology “locative clitic”, which is misleading in an important respect. This is true for all three languages mentioned so far. Both French and Italian, which clearly distinguish their locative clitic from their dative clitics, have their locative clitic appearing, as is well-known, in sentences that are not locative at all: (7) Jean y pense. Jean of-it thinks ‘Jean thinks of it.’
(French)
. This is so, whether or not the sentence contains an accusative clitic – differently from the Spanish so-called “spurious se” that Perlmutter (1971) discusses.
Richard S. Kayne
(8) Gianni ci pensa. Gianni of-it thinks ‘Gianni thinks of it.’
(Italian)
In these sentences with the verb think, in the sense of “think of/about something”, both French and Italian readily have, when the complement is an unstressed pronominal (especially inanimate), what looks like the locative clitic. In earlier work – Kayne (2004a) – I argued that sentences like (7) and (8) correspond closely to archaic English sentences of the following sort:
(9) We spoke thereof.
These also contain what seems to be a locative there in sentences that do not involve location. Such archaic English sentences (whose counterparts are very much alive in Dutch and German) differ from (7) and (8) in having an overt preposition (in this example, of ) in addition to there. My proposal was to take French and Italian sentences like (7) and (8) to contain a silent counterpart of that preposition, whichever one is appropriate for the verb in question. (Thus (7) will have a silent à and (8) a silent a – essentially the same one in the two languages.) From this perspective, there is a unified phenomenon in (7)–(9) whereby an element that looks like a locative (y or ci or there) appears in a non-locative sentence. The account I proposed goes essentially as follows (using mostly English examples, but the analysis is the same for French and Italian, apart from the orthogonal non-clitic vs. clitic difference). There is not, strictly speaking, locative in any of its uses. Both in (9) and in banal locative sentences like: (10) We went there yesterday.
we have an element there that is the same element as the one found in non-standard English in:3 . And similarly for here in:
(i) This here car ain’t no good.
Although non-standard in English, comparable combinations of there/here with a demonstrative are standard in French:
(ii)
cette voiture-là dem car- there ‘that car there’
(iii)
cette voiture-ci dem car- here ‘this car here’
Expletives, datives, and the tension between morphology and syntax
(11) That there car ain’t no good.
In addition, the proposal is that in both (10) and (11) there modifies a noun, except that in (10) the noun is silent (to be indicated by capital letters), i.e., (10) is to be thought of as: (12) . . . there place . . .
The same holds of (9), except that the silent noun in (9) is not PLACE, but THING:4 (13) . . . there thing of . . .
Exactly parallel, except for the silent preposition (indicated as P) are (7) and (8): (14) . . . y/ci thing P . . .
The question now is how to understand the elements y, ci and there that occur in this range of environments (see also note 3). In (7)–(9), they are not locative in any simple sense. The link with (11) suggests a link with what we call demonstratives, which in some cases do seem related to location, as in pointing contexts: (15) Bring us that book, please.
Yet, as is well-known, demonstratives are not limited to contexts involving location: (16) That book you mentioned yesterday is of little interest.
A familiar idea concerning demonstratives is that they involve deixis, or reference to or orientation with respect to the speaker. The difference between this and that in English could, for example, be put as “in the sphere of the source of the sentence” vs. “not in the sphere of the source of the sentence”. Taking the term “deixis” over to there (and here), I will henceforth speak of deictic there (and deictic here), for all the cases mentioned. In each of (9), (10)
(In French, the demonstrative itself does not vary in form, in contrast to English this vs. that.) On the fact that -là and -ci follow the noun in French, see Bernstein (1997). Sentences like (11) and (i) but with a definite article in place of the demonstrative are also standard in various Scandinavian languages – cf. Leu (2007). In French, there is also a form ici (“here”) in ordinary locative sentences like: (iv) Jean est ici. Jean is here ‘Jean is here’. with an extra i- whose morphemic status remains to be fully understood. . Possibly THING precedes there, and similarly for French and Italian; cf. Kayne (2006).
Richard S. Kayne
and (11) we have an instance of exactly the same deictic there; the three differ with respect to what deictic there modifies (THING in (9), PLACE in (10), and car in (11)5). For the case of deictic there modifiying PLACE, a noun expressing location, I will use the term “locative there”, to be understood solely as an abbreviation for “deictic there modifying PLACE”. From this perspective, all of (9)–(11) contain deictic there. But of the three sentences, only (10) is a locative sentence (in the sense that it contains locative there, by virtue of containing “there PLACE”). (7) and (8) are like (9): they contain deictic y and ci, the French and Italian counterparts of deictic there, but are not locative sentences, since they have no PLACE. Y and ci in (7) and (8) are deictic clitics that are not part of a locative phrase. In: (17) Jean y va. Jean there goes ‘Jean goes there.’
(French)
(18) Gianni ci va. Gianni there goes ‘Gianni goes there.’
(Italian)
y and ci are deictic clitics that are part of a locative phrase, since (17) and (18) contain PLACE. Returning to Paduan, we see that (5) contains a deictic clitic ghe that is part of a locative phrase, given that the verb “put” calls for PLACE (or some overt counterpart). The specificity of Paduan (6), repeated here: (19) Ghe dago el libro. there I-give the book ‘I am giving him/her/them the book.’
is now seen to be that Paduan, unlike standard French and Italian, has a deictic clitic in its third person dative sentences where one might have expected a dative clitic. The question is why. A second question is whether (19) contains PLACE, i.e., whether ghe there is locative. I will argue that it is not. Ghe in (19) will be seen to be deictic, but not locative, just as the clitics y and ci in (7) and (8) are deictic but not locative.
. In colloquial Norwegian, a prenominal (counterpart of) there can in some cases itself modify PLACE, with “there PLACE” then modifying the lexical noun – see Leu (to appear) – whether this is an option in non-standard English is not clear. English certainly allows this post-nominally, as in:
(i)
That car (over) there looks dangerous.
Expletives, datives, and the tension between morphology and syntax
4. Silent DATCL As part of the analysis of Paduan sentences like (19), let me propose that (19) actually does contain a third person dative clitic, of the sort seen overtly in French and Italian in (2) and (4). The difference is that, in Paduan and various other dialects, that clitic is silent. In other words, (19) should be thought of as:6 (20) DATCL ghe dago el libro.
where DATCL represents the silent dative clitic. The fact that ghe appears to “replace” only third person dative clitics must now be interpreted more precisely as meaning that the silent DATCL that ghe cooccurs with in sentences like (19)/(20) is limited to third person and cannot be first or second person (or reflexive). This person restriction can be understood in part as follows. The person property of silent DATCL7 links up here to the fact that various languages (e.g., Somali – cf. Saeed (1993: 174)) have zero forms for (object) third person pronouns in general (even in the absence of any apparent licenser), but not for first or second person. Closer to Paduan, French silent subject clitics, as in: (21) Lui a téléphoné. him has called ‘HE has called.’
with a silent third person subject clitic (in addition to non-clitic lui), are also limited to third person (non-reflexive): (22) *Moi ai téléphoné. me has called
as discussed in Kayne & Pollock (2001, sect. 5). Additional instances of a restriction to third person, in the case of silent pronominals, are discussed in Kayne (2001). (Whether, despite appearances, this restriction can be attributed directly to UG, in which case nothing special would need to be said about Paduan in this regard, remains to be seen.) . Whether DATCL precedes ghe (as I have chosen to indicate it here) or follows it is a potentially important question that will, however, not be relevant to what follows in this paper. The silent clitic postulated here should be compared to those discussed by Benincà (1989) for Friulian, by Roberts (1993) for Valdostano, and by Longa et al. (1998) for various Iberian languages/dialects. . The presence of DATCL appears to be licensed by deictic ghe, though how exactly remains to be worked out – perhaps the presence of ghe makes available a phasal Spec-position into which the dative clitic can “disappear”, a la Kayne (2006).
Richard S. Kayne
By attributing the person restriction holding of Paduan dative sentences with ghe to the presence of DATCL, i.e., by calling DATCL (rather than ghe) the true locus of the person restriction, we account straightforwardly for the fact that the Paduan counterpart of (7) and (8), namely: (23) Giorgio ghe pensa. Giorgio there thinks ‘Giorgi thinks of/about it.’
does not show the same person restriction. Paduan (23), like French (7) and Italian (8), allows reference to a first or second person (or reflexive) object (of the silent preposition) to some extent, especially with CLLD (clitic left dislocation) and with coordination. An example with CLLD in Italian (from Cinque (1990: 59)) with a reflexive is: (24) A se stessa, Maria non ci pensa. to refl. same Maria neg. there thinks ‘Of herself, Maria doesn’t think.’
(A French example with coordination is given in Kayne (1975: sect. 2.7).8 Although sentences like (23) cannot refer to first or second person or reflexive with complete freedom, the restriction appears to be different in kind from the absolute one holding of (19). The reason is that only in (19) is reference mediated by DATCL (rather than by ghe, which is expletive-like in (19), as we shall see later), and silent DATCL is absolutely limited to third person. In (23), on the other hand, DATCL is not present. Instead, we have, as in (14), with P = preposition: (25) . . . ghe THING P . . .
or, more exactly, in those (limited) cases involving reference to a person (as in (24)): (26) . . . ghe PERSON P . . .
Conversely, a phrase of the type “ghe PERSON” is not present in (19). (This is related to the just-mentioned expletive-like character of the ghe of (19).) The presence of DATCL in (19) also plays a role in clitic doubling of the sort seen in: (27) Ghe dago el libro a Giorgio. there I-give the book to Giorgio ‘I give the book to Giorgio.’
. Cf. also van Riemsdijk (1978: 125) on colloquial Dutch R-pronouns being able to refer to humans, perhaps parallel to y/en, ci/ne; cf. also Bennis (1986: 191).
Expletives, datives, and the tension between morphology and syntax
By having DATCL in such sentences, we can (correctly, I think) assimilate this clitic doubling to the parallel well-known phenomenon found with datives in Spanish (although the dative clitic in Spanish is pronounced more than in Paduan). Of particular interest here is the point made by Cordin (1991) for Trentino (which is similar to Paduan in the relevant respects), namely that despite the appearance of the deictic clitic in both dative and locative sentences, there is a difference between them when it comes to clitic doubling. Clitic doubling is compatible (and generally obligatory when there is a non-clitic dative) with the presence of ghe in dative sentences, yet impossible with ghe in locative sentences. (Clitic doubling must be kept distinct from right-dislocation, which is possible with both datives and locatives.) The reason, from the present perspective, is that clitic doubling is dependent on DATCL, and that ghe by itself is not compatible with clitic doubling (for reasons to be elucidated).9 The proposal that there is a silent dative clitic in (27) leads to the expectation that there could be Romance languages in which that dative clitic would be overt at the same time as (the counterpart of) ghe, even in sentences with no locative (i.e., with no PLACE). This expectation is met by some Sardinian, to judge by an example given by Jones (1993: 220): (28) Narrabílis! tell- bi- to-them ‘Tell them!’
where bi is the Sardinian deictic clitic parallel to ghe, and lis is the overt (plural) dative clitic (with the accusative unpronounced, as in some varieties of French).10
5. Expletive there and expletive ghe as deictics From the perspective of the proposal in (20), ghe there is definitely not a dative clitic. It should rather be taken to be exactly the same kind of deictic element as all the other instances of ghe (and there and y and ci) under discussion. At the same time, it does not seem to be identical to any of them.
. The ghe of (23) seems by and large to be incompatible with (non-dislocated) clitic doubling, too, as expected (since there is no DATCL there), although there is the kind of exception mentioned in Kayne (1975: chapter 2, note 51). . If the order of clitics in (28) were to turn out to be limited to imperatives, one would think of Terzi (1999).
Richard S. Kayne
The reason, I think, is that the ghe in Paduan dative sentences that cooccurs with DATCL is an expletive, in the same sense as the ghe of the Paduan sentence: (29) Ghe ze un libro . . . ghe is a book ‘There is a book . . . ’
which is strongly similar to the Italian: (30) C’ è un libro sul tavolo. there is a book on-the table ‘There is a book on the table.’
Burzio (1986: 148) has already argued that in this kind of Italian existential sentence ci/c’ is significantly similar to English expletive there in: (31) There’s a book on the table.
By extension, the ghe of (29) is essentially the same as English expletive there. An important difference, of course, is that English there is subject-like as far as its final position is concerned, while Italian ci shares properties of object clitics and not of subjects (e.g., ci follows negation and certain other object clitics). In this regard, Paduan ghe is very much like ci. But if, in the spirit of Burzio (1986), we set aside the subject position vs. object clitic position difference, we arrive at the conclusion that all of these (ghe, ci, there) in (29)–(31) are expletives in the same sense. Recall that the first part of the answer to the question why Paduan would have deictic ghe in its third person dative sentences was that ghe in such sentences is actually present alongside silent DATCL. The second part, now, is that the ghe of Paduan datives is an expletive ghe of the sort found in (29). This will mean, in a way to be explored shortly, that Paduan dative sentences have exactly the right number of arguments for the number of theta roles. There are still two further questions. First, what exactly do we mean by “expletive”? Second, why is Paduan allowed to have an expletive in dative sentences in the first place? Let me begin with the first question. One of the guiding principles of the discussion so far has been that the various uses of there and of ghe and of ci and y are all reflections of exactly the same deictic element occurring in somewhat different environments. In all the cases discussed, the deictic modified a noun (or NP). That noun is overt in that there car and in the corresponding French cette voiture-là (“that car there”) – see note 3. In the other cases mentioned, the noun modified by the deictic is silent, either PLACE or THING or PERSON. Pursuing this guiding principle further, we unavoidably (and desirably, I will argue) arrive at the conclusion that all the expletive uses of there
Expletives, datives, and the tension between morphology and syntax
and ghe and ci and y (and of Sardinian bi and Catalan hi) are likewise instances of this same deictic element. If English expletive there is the same deictic element as all the other instances of there, then we would expect it, too, (in (31), for example) to modify a noun (or NP). The next question, then, is, what noun? Taking into account examples like: (32) There’s a car in this garage.
and the fact that there is incompatible with this: (33) There’s a car in this here/*there garage.
there is little plausibility to taking there in these examples to modify garage. Much more natural, I think, is the proposal that expletive there is modifying the other noun, the one informally called the “associate”. Put another way, in (32) and (33) expletive there is a deictic modifier of car. More generally: (34) a. In English existential sentences, expletive there is invariably a deictic modifier of the associate. b. The same holds in existentials for Paduan ghe, for Italian ci, for French y, etc.
(Note that given the discussion above, this definitely does not mean that expletive there is locative; expletive there (and its counterparts in other languages) is a deictic element that is non-locative, i.e., that does not modify PLACE.) Adopting (34) means adopting derivations in which expletive there originates within the associate and splits off from it, ending up in subject position (and similarly for the other languages, apart from the object clitic vs. subject position difference). There are two ways to think of this splitting off. One would be to have there raise directly out of the associate containing it. A second would be to have the rest of the associate raise, stranding there and then to have “there + trace of associate” raise further in the manner of remnant movement. There may well be some questions with respect to which the choice between these two approaches to “splitting” is neutral. For others, though, the choice is likely to be meaningful (e.g., for agreement, as discussed below). Thinking more specifically of the kind of remnant movement derivations discussed in Kayne (2002) and works cited there, and of the possible impossibility of extraction of modifiers from within a containing DP, let me adopt the second, remnant movement, approach, which yields (for sentences like (31)) (partial, sketchily illustrated) derivations such as the following: (35)
. . . [there a book] . . . . . . a booki . . . [there ti] . . . . . . is a booki . . . [there ti] . . . . . . [there ti]k is a booki . . . tk . . .
→ raising of a book → merger of V → remnant movement
Richard S. Kayne
and similarly for Paduan, etc. modulo the difference in final landing site (i.e., ghe will end up in an object clitic position, rather than in a subject position).11 The proposal reflected in (35) has expletive there originating as an instance of deictic (nonlocative) there contained within an indefinite DP (using the term DP loosely). In (non-standard) English, however, there appears with an overt noun only in the presence of a demonstrative: (36) that there book; this here book
as opposed to: (37) *the there book; *the here book (38) *a there book; *some here book, etc. (39) *there a book; *here some book, etc.
How plausible is it, then, in particular in light of (38) and (39), to allow the derivation in (35)? One consideration can be put as follows. Given (36), would we have expected (38) or (39) to be acceptable, or not?12 That depends, I think, on how exactly we see the derivation of (36) itself. In particular, if (36) has there or here originating within a relative clause structure, then the impossibility of (38)/(39) is actually a bit surprising, given that relative clauses are in general compatible with both definite and indefinite “heads”.13
. For the Romance languages under discussion, the most direct transposition from (35) would suggest that the expletive clitic (e.g., ghe) is a remnant, rather than a pure head. The possibility that pronominal clitics are always phrasal needs to be examined carefully in any event. A remnant movement approach to pronominal clitics may be suggested, too, by cases of past participle agreement in Italian with ne, which itself shows no phi-features: (i)
Ne ho visti tre. of-them I-have seen.pl three ‘I have seen three of them’.
and (example suggested by Guglielmo Cinque, p.c.): (ii)
Ne ho letta la metà, di quel libro. of-it I-have read.fem the half.fem of that book.masc ‘Of that book, I have read half ’.
. On (37), see Leu (2007). . Indefinite-“headed” relatives are also compatible with “person”-possessives that have something in common with deictics: (i) We need a place of our own.
Expletives, datives, and the tension between morphology and syntax
Kayne (2004a) argued against a relative clause analysis of (36) on what I now think were inconclusive grounds. The relevant data are in part: (40) *that over there book, *this right here book
which contrast with: (41) the book that’s over there; the book that’s right here
Relatives can contain over there or right here, but those combinations are not possible prenominally. Somewhat similarly, there in a relative can be stressed (indicated here by extra spacing and italics) in a way that it cannot be prenominally: (42) the book that’s t h e r e (43) *that t h e r e book (vs. that g r e e n book)
suggesting, apparently, that prenominal deictic there and here must have a source other than within a (reduced) relative. However, there is a narrower conclusion that can be drawn – one that allows these facts to be interpreted as neutral with respect to the relative clause question. This narrower conclusion is that over and right in (40) and stress in (43) are excluded because they all depend on the presence of place,14 i.e., on the presence of a locative phrase. Deictic there and here are by themselves not compatible with over or right or with contrastive stress falling on them. If this is correct, then prenominal deictic there (or here) could well have a relative clause source, as long as the relative lacked place.15
. For a compatible approach to right, see Johnson (1991). Supporting the importance of PLACE are the following, similar to (40):
(i)
We spoke (*over) thereof.
(ii)
We spoke (*right) thereof.
The contrasts seem clear, even relative to archaic English. More part of (a certain) spoken English is:
(iii)
I hereby give you permission.
Again:
(iv) *I over/right hereby give you permission.
. Like simple adjectives, deictics, even if derived from relative clauses, cannot, I think, remain postnominal (contrary to certain reduced relatives like any linguist interested in physics), i.e., (i) and, very clearly, (ii) in English feel as if they only contain a locative:
(i)
that book there
(ii)
the book that’s there
Richard S. Kayne
If prenominal there and here do have a relative clause source, then it is (38)/ (39) that is surprising, and not the presence of “[there a book]” in the derivation proposed in (35).The question remains, then, as to why neither (38) nor (39) is possible. Let me suggest a link to Szabolcsi’s (1983, 1994) analysis of Hungarian possessive sentences. Szabolcsi argues that possessors (in the Hungarian counterparts of simple possessive sentences like John has a sister) originate within a DP (containing a sister) that is the argument of an existential verb (that looks like be). The possessor then moves out of that DP, doing so obligatorily because the DP is indefinite. Putting it slightly differently, a Hungarian DP containing a (relatively nonembedded) possessor must necessarily “split” if that DP is indefinite (and may do so if it is definite). Let me now suggest the same for deictic there, namely that when deictic there is contained in an indefinite DP (and not embedded too far down in it), that indefinite DP must split obligatorily, in the way shown in (35).16 (Why exactly such (non-specific) indefinites must split in these two kinds of cases, and perhaps others, or perhaps all cases, remains to be elucidated.17) (38) and (39) are impossible as intact DPs because they have not split, despite being indefinite. Yet such indefinite DPs containing deictic there (or here) can be legitimate if they do split, and therefore can appear in the initial stage of a derivation such as (35). In summary, then, what we call expletive there is characterized by (34) and, along with its counterparts in various Romance languages, has a derivation of the sort loosely sketched in (35).
In French, on the other hand, (iii) almost certainly can be a non-locative deictic: (iii)
ce livre-là that/this book there ‘that book’
On the order difference between English and French and its relation to the position of adjectives, see Bernstein (1997). On adjectives being derived from relatives, cf. Kayne (1994: sect. 8.4) and references cited there; also Leu (2006). On nouns themselves being derived from relatives, cf. Koopman (2003; 2005). . Maximal parallelism with Hungarian would lead to thinking that Hungarian indefinites containing a possessor split in remnant movement fashion, too. . One should also consider this from the perspective of Sportiche (2002) on D. The Hungarian possessor asymmetry between definite and indefinite DPs has a partial counterpart within English, in the contrast between (i) and (ii):
(i) a friend of John’s (arguably derived from John’s a friend – cf. Kayne (1993: sect. 1.2) – by movement of a friend to the Spec of of) vs.: (ii)
John’s friend
If the containing DP is definite, as in (ii), the kind of movement seen in (i) need not take place.
Expletives, datives, and the tension between morphology and syntax
6. Deictics, demonstratives and indefinites As illustrated in (36)–(39), English deictics have a privileged relation to demonstratives, which are the only determiners that in (non-standard) English can overtly cooccur with a prenominal deictic. On the other hand, not every instance of a demonstrative is compatible with a deictic, even (to my not entirely native ear) in non-standard English. The following seem appreciably less possible than (36): (44) *Your child has never been that there irritable before. (45) *A thesis shouldn’t really be this here short.
(Both of these would be possible without there or here.) The generalization may be that deictics must modify a noun (or projection thereof).18 To the (partial) extent that deictics do have a privileged relation with demonstratives, we are led to ask, given the proposal that deictics can in fact combine with indefinites, whether demonstratives might not be able to combine with indefinites, too, in a way that would support separating both deictics and demonstratives from any intrinsic link to definiteness. Of interest here is Hebrew, as discussed by Sichel (2001), which has the convenient property that its demonstratives cooccur, when in a definite DP, with an overt definite article: (46) ha-yalda the girl ‘that girl’
ha-zot the dem
Yet Hebrew also allows (Sichel: chapter 1, note 6): (47) yalda zot
with no definite article, yet with the same demonstrative element. In addition, while (46) as a direct object would be preceded by the morpheme et that normally precedes definite direct objects, (47) would not be. Sichel concludes that (47) is an instance of a demonstrative that is not definite.19 It seems, then, that demonstratives are no more universally wedded to definites than are, given my proposal that expletive there is a deictic originating within an indefinite, deictics. The plausibility of (35) is thereby enhanced.
. At least in English. Delsing (1993: 136) notes as acceptable the Swedish counterpart of *a so here big car. . Cf. perhaps English:
(i) There’s this guy on the phone for you.
Richard S. Kayne
7. The definiteness effect The order of elements in the constituent “[there a book]” postulated in (35), in which there precedes a, is indirectly supported by (non-standard): (48) these here four books
to the extent that the indefinite article a is akin to numerals, as argued by Perlmutter (1970). In other words, . . . there a . . . in (35) parallels . . . here four . . . in (48). Assume now that the definite article would, in contrast, precede the deictic, as the demonstrative does in the example just given.20 This means that replacing the indefinite article in (35) by a definite article would yield: (49)
. . . [the there book] . . .
(50)
. . . [the there three books] . . .
This contrast in relative position between definite and indefinite article will have an interesting effect. Whereas in (35) a book could raise out from within there a book, that same raising will be precluded in (49) or (50). Such raising of the book or of the three books out of their containing DP is precluded by the fact that in (49) and (50) the book and the three books are not constituents. Therefore there is no way to have, parallel to (35), a derivation that would yield, with expletive there: (51) *There’s the book on the table.
Thus we have, granted that the must precede there or here, the beginning of an account of the core definiteness effect found in (English) existentials.21
. Contrary to Afrikaans, for which something further needs to be said. On relatives preceding D, cf. Whitman (1981) and Kayne (1994: sect. 8.3). . (51) is possible in a “list” context, for example:
(i) (ii)
What should we read? Well, there’s the book on the table. Who can we invite? Well, there’s John.
Perhaps the definites here are embedded within hidden indefinites. On differing sensitivity to definiteness in two dialects of Catalan, see Rigau (2005: 792); similarly, for two varieties of Spanish, Longa et al. (1998: 13). As a reviewer emphasizes, the account suggested for (51) should generalize to (cf. Milsark (1974)):
(iii) *There’s every book on the/a shelf.
on the assumption that this every must precede there as the does in (49), and similarly for other ‘strong determiners’ in Milsark’s sense. The contrast with:
(iv)
There’s every reason to believe we’re right.
Expletives, datives, and the tension between morphology and syntax
In essence, any determiner that must in general precede deictic there will be incompatible with the kind of derivation shown in (35) that underlies existentials that contain expletive there (a particular subcase of deictic there). Starting from (49) or (50), there is another imaginable derivation that needs to be considered, in which “(a) book” or “three books” would be raised out of the containing DP leaving behind ‘the there’. Such a derivation would yield: (52) *The there is (a) book on the table. (53) *The there are three books on the table.
These can be excluded if the presence of the blocks the raising operation.22 That the might have such a blocking effect is a long-standing idea – cf. Fiengo & Higginbotham (1981). Although there are exceptions and although the reason for the blocking effect needs to be made more precise, there is one very sharp case in Romance that seems to fit well with (52)/(53): (54) Jean en a (*les) trois. Jean of-them has the three ‘Jean has three of them.’ (55) Jean en a un/*l’ autre. Jean of-them has an/the other ‘Jean has another of them.’
In these French examples, the extraction of quantitative en (cf. Pollock (1998)) is blocked by the presence of a definite article.23
implies, then, that the every of (iv) is lower than that of (iii), as argued on independent grounds by Postma and Rooryck (1996). . For a partly similar idea, see den Dikken (1997). . Cf. Kayne (1975: chapter 2, note 55). Note that right-dislocation with de (cf. Vinet (1977)), but without en, is possible: (i)
Jean m’ a montré les tiens, de livres. Jean me has shown the yours, of books ‘As for books, John has shown me yours.’
though not with des: (ii) *Jean m’ a montré les tiens, des livres. Jean me has shown the yours of-the books The text discussion assumes that there is no “escape-hatch” available in these cases, either, in particular SpecDP (unlike in the case of (Hungarian) possessors).
Richard S. Kayne
Thinking of (48) and the corresponding non-standard: (56) them there four books
(with them rather than those), one also needs to exclude a derivation that would yield: (57) *Them there were four books on the table an hour ago.
Again, it is plausible that the presence of them blocks the extraction of four books that would have been necessary to derive (57) in a way parallel to the licit derivation indicated in (35). Arguably like the definiteness effect of (51) is (cf. Heim (1987)) the absence of a wide scope reading for three books in There must be three books on the table. From the present perspective, a wide scope (“specific”) reading of three books must in general require the presence of a “specific” D (a more general D than the one restricted to definites), covert in English but arguably overt in Gungbe – Aboh (2004: chapter 3). That D will have the same effect on extraction of there as the definite D.24 (Not surprisingly, then, Cresti (2003) argues that there is a parallel scope restriction with Italian ne (related to the French en of (54) and (55)).) 8. Agreement In the remnant movement derivation given in (35), what ends up in subject position is “[there ti]”, where ti is the trace/copy of “a book”. In the corresponding derivation of: (58) There are three books on the table.
what ends up in subject position is again “[there ti]”, where ti is now the trace/ copy of three books. This may make it possible to take the plural are in (58) to be determined by the plurality of the phrase in subject position, which is not simply expletive (deictic) there, but a bigger phrase containing (the trace/copy of) plural number – cf. also Koopman (2003, 2005). If this is correct, then there may be no need here for downward agreement of the sort proposed by Chomsky (2001).25 . The ill-formedness of (i) can similarly be taken to reflect the presence of a covert definite or specific D (required by the presence of all):
(i) *There will all be three books on the table.
The ill-formedness of (ii) (v. Chomsky (1995: 275)) might be related to this:
(ii) *There seem to each other to be five people here.
. The lack of agreement possible in colloquial English in some cases has a counterpart in Italian – cf. Burzio (1986: 77). On interesting variation concerning agreement within Catalan, see Rigau (2005).
Expletives, datives, and the tension between morphology and syntax
In a similar vein, in the Italian counterpart of (58): (59) Ci sono tre libri sul tavolo. there are three books on-the table ‘There are three books on the table’.
the plural form sono might be determined through direct agreement with the preceding “[ci ti]”, i.e., with the phrase containing the expletive clitic. Alternatively, there may be a Spec-position between ci and sono through which “[ci ti]” has passed.26 9. Why is there the expletive? English expletive there has no counterpart in then. (60) There are problems with your proposal. (61) *Then are problems with your proposal.
If expletives were uninterpretable elements merged directly into a (relatively) high subject position, it would not be immediately clear why English or some nearby (or distant) language could not have then as its expletive. From the present perspective, which takes expletive there to be a deictic element merged within an indefinite DP, we can do somewhat better. First, we can note that the contrast between (60) and (61) is not limited to existential contexts; it is also found in the archaic English construction mentioned earlier, for example: (62) We spoke thereof/*thenof.
and similarly for thereby, which can still be heard: (63) We thereby/*thenby demonstrated . . .
More pointedly, perhaps, deictic there preceding an overt noun has no then counterpart:27 . Cf. also Belletti (2005: 18). . Conversely the non-colloquial (i) has no counterpart with there:
(i) ?his then wife; ?the then president (ii) *his there wife; *the there president
(i) seems more like: (iii) the president then i.e., more like the temporal counterpart of a locative, rather than like the temporal counterpart of a deictic.
Richard S. Kayne
(64) That there/*then car ain’t no good.
A statement touching on all the facts of this section is: (65) Locatives are closer to deictics than are temporals.
More specifically, locatives can be formed by combining a deictic with silent PLACE, yielding what I have been calling locative there (or here). This in effect gives locatives the possibility of being phonetically indistinguishable from the non-locative deictic there seen in (60), (62), (63) and probably in one (perhaps the only – see note 5) reading of (64). The idea behind (65) is that the proper analysis of then in sentences like: (66) They were happy (back) then.
cannot be as simple as the proper analysis of locative there. Put another way, although locative there is deictic there combined with PLACE, temporal then cannot simply correspond to a deictic element combined with TIME. This property of temporals is probably not to be understood in terms of silent TIME being systematically unavailable, given the double possibility indicated in: (67) We’ll be at your place in two hours. (68) We’ll be at your place in two hours’ time.
which makes TIME seems appropriate for (67) – see Kayne (2007a). It may rather be that silent TIME requires a modifier that is itself specified for “time” (as is (two) hours in (67)), whereas PLACE is not so demanding. Alternatively put, both TIME and PLACE require a modifier at least partially specified for, respectively, “time” or “place”.28 The difference, then, would be that there is some partial overlap between “location” and “deixis” itself, but no comparable partial overlap between “time” and “deixis”.29 In effect, location would then be seen as a more concrete, narrowed down (via PLACE) counterpart of deixis, whereas time would not be.
. The PLACE found in:
(i)
Let’s go over to John’s tonight.
is a silent counterpart of the place found in:
(ii)
Let’s go over to John’s place tonight.
which conveys more than just location – it seems close to “home”, in a way that recalls Longobardi (1996). . On probably related differences between locatives and temporals, see Starke (2001).
Expletives, datives, and the tension between morphology and syntax
From this perspective, (66) can be thought of as: (69)
. . . then TIME.
where temporal then is itself specified for time (and requires TIME) – in a way that distinguishes it sharply from deictic there, which is not specified for location (and does not need to cooccur with plACE, though it can). There is consequently no derivation available for (61) that could track the derivation of (60) (that was in effect sketched in (35), which is repeated here): (70)
. . . [there a book] . . . . . . a booki . . . [there ti] . . . . . . is a booki . . . [there ti] . . . . . . [there ti]k is a booki . . . tk . . .
→ raising of a book → merger of V → remnant movement
In addition to there being no expletive then parallel to expletive there, we can also note that there is no modified expletive there: (71) Are there problems with your proposal? (72) *Are over/right there problems with your proposal?
Since expletive there is a subcase of deictic there and since deictic there in general disallows modification (cf. also (40)):30 (73) *That over/right there car ain’t no good.
the unacceptability of (72) is expected. There is no expletive here parallel to expletive there, either:31 (74) *Are here problems with your proposal?
which means that here cannot successfully appear in a derivation like (70). We can express this restriction as: (75) Here can only appear within a definite DP.
. Possible is:
(i)
that car over there
but here we have locative there, i.e., deictic there combined with plACE, with over keyed to plACE. . Possible is:
(i)
Here are several problems for you.
but this is probably an instance of a preposed locative here, i.e., with PLACE.
Richard S. Kayne
In effect, there is more “neutral” than here,32 with this difference in turn probably to be related to what I think are similar differences between that and this, for example: (76) He’s not all that/*all this smart. (77) The behavior of their son is somewhat different from that/*this of their daughter.
and even with what we call complementizer that: (78) They think that/*this everything is fine.
One (plausible) implication of (75) is that here must be contained within a definite DP in both of the following, despite there being no overt definite article: (79) They live here. (80) We hereby declare . . .
(In addition to some silent indicator of definiteness, (79) contains PLACE and (80) THING.) Both locative there and the there of archaic thereof (and non-archaic thereby) have counterparts with where: (81) Where do they live? (82) Whereof have they spoken? (archaic) (83) the plan whereby we . . .
Expletive there does not:33 (84) Where is there/*where a problem?
In this respect the deictic there that we call expletive behaves as the deictic there preceding overt nouns (in non-standard English): (85) that there car (86) *that where car; *this where car; *a where car; *some where car
If where is not a deictic element at all, then (84) is not surprising.34
. Cf. Dasgupta (1992) and Jayaseelan & Hariprasad (2001). Here, this and these may contain a first-person feature or morpheme that does not have or at least can fail to have a counterpart with there, that and those. . Cf. Chomsky (1995: 392), whose argument against (an earlier version of) Moro (1997) does not carry over to the present proposal. . If there, here and where do form a natural class of deictics – see Kayne (2007b)– then more needs to be said.
Expletives, datives, and the tension between morphology and syntax
In summary to these last sections, what we call expletive there is an instance of deictic there initially occurring within an indefinite DP, and then being split away from it, as indicated in (70). 10. Expletive ghe and ci in possessive sentences As earlier (cf. the discussion of (30)), I follow Burzio (1986) in taking Italian ci in existential sentences to be strongly similar to English there, apart from the fact that ci ends up in object clitic position, whereas there ends up in subject position. An example of Italian existential ci is: (87) C’ è un libro sul tavolo. there is a book on-the table ‘There is a book on the table’.
The derivation of (87) will resemble that of (70), modulo the final landing site, and similarly for Paduan ghe in: (88) Ghe ze un libro . . . there is a book ‘There is a book . . . ’
For a certain subset of Italian speakers, the expletive ci of (87) can also appear in simple possessive sentences such as:35 . (89) becomes (much) more widely accepted if expletive ci cooccurs with an accusative or quantitative clitic: (i)
Una sorella, ce l’ ha anche Gianni. a sister there it has also Gianni ‘A sister, Gianni also has one.’
(ii)
Gianni ce n’ ha due. Gianni there of-them has two ‘Gianni has two of them.’
in a way that recalls the improvement in Italian non-dislocated clitic doubling attributable to a second clitic – cf. Cinque (1990: 178). In possessive sentences, this ci cannot undergo clitic climbing: (iii) *Ci vorrei avere una sorella. there I-would-like to-have a sister unless accompanied (perhaps pied-piped) by another clitic, as in: (iv)
Una bella casa, ce la vorrei avere anch’ io. a beautiful house, there it I-would-like to-have also I ‘A beautiful house, I would also like to have one.’
Richard S. Kayne
(89) Gianni c’ ha una sorella. Gianni ci has a sister ‘Gianni has a sister’
For the speakers in question (89) is possible without any locative interpretation being associated with it, i.e., it can correspond perfectly to English: (90) John has a sister.
In the Italian of the relevant speakers, the presence of ci in (89) is generally not obligatory. In many dialects of Italy such a clitic (i.e., the corresponding deictic clitic, e.g., ghe in Paduan) often is obligatory in the equivalent of (89)/(90) – cf. Moro (1997: 237).36 It seems virtually certain, as Moro suggests, that this ci or ghe in possessive sentences is an expletive clitic and is in fact the same expletive clitic as the one found in existential sentences like (87) and (88). The question is why (some) Italian and many dialects of Italy should allow an expletive at all in possessive sentences like (89). I think the answer is to be found in Szabolcsi’s (1983, 1994) analysis of simple possessive sentences in Hungarian. Her proposal, mentioned earlier, is essentially that possessive sentences are based on existentials. The derivation of simple possessive sentences tracks that of existential sentences (which are taken to have a single verb that takes a single argument (apart from a possible additional locative)), with an important twist. In possessive sentences like (90) in Hungarian, the possessor originates within that single argument of the existential and raises out of it, picking up dative Case on the way. Transposition of Szabolcsi’s analysis to English – cf. Kayne (1993) – has the possessor in English raising out of the same single argument of the existential. There are, however, two (linked) differences between Hungarian and English; the possessor in English ends up with nominative Case (apart from ECM contexts) and the verb in English must be have rather than be, as seen in (90). (In the Hungarian counterpart of (90), John would bear dative Case and the verb would be existential be). Szabolcsi’s idea that possessive sentences like (90) embed an existential structure within them clearly makes the appearance of expletive ci in (89) less surprising (and similarly for the relevant dialects). In effect, (89) has “inherited” its expletive ci from the existential embedded within it.
In addition, there are restrictions preventing ci from appearing (postverbally) in non-finite contexts, in many cases. All of this needs further study. . In many dialects, there is a ghe with auxiliary “have”, too. The source for those instances of ghe may lie with the nominalization-like character of past participles.
Expletives, datives, and the tension between morphology and syntax
To see what the derivations might look like, let us begin with one (modeled on (70)) for: (91) There is a sister of John’s in our class.
Setting aside the locative in our class, the derivation proceeds as follows: (92)
. . . [there a sister of John’s]indef.DP . . . . . . [a sister of John’s]i . . . [there ti] . . . . . . is [a sister of John’s]i . . . [there ti] . . . . . . [there ti]k is [a sister of John’s]i . . . tk . . .
Transposing to Italian (89) gives:
(93)
. . . [ci una sorella di Gianni]indef.DP . . . . . . [una sorella di Gianni]i . . . [ci ti] . . . . . . è [una sorella di Gianni]i . . . [ci ti] . . . . . . [ci ti]k è [una sorella di Gianni]i . . . tk . . .
→ raising of a sister of John’s → merger of V → remnant movement
→ raising of una sorella di Gianni → merger of V → remnant movement
If the derivation stopped here, it would correspond to the existential sentence: (94) C’ è una sorella di Gianni . . . there is a sister of Gianni ‘There is a sister of Gianni . . . ’
Extending the derivation in (93) along Szabolcsi’s lines amounts to saying that in Italian the possessor Gianni can subsequently raise out of the phrase una sorella di Gianni, ending up in subject position. With the necessary appearance of “have” instead of “be”, this yields (89). The last step (abstracting away from the question whether “have” is in the numeration or not and also from the question of the preposition di) is then: (95) . . . [ci ti]k ha [una sorella Gianni]i . . . tk . . . → raising of the possessor . . . Giannim [ci ti]k ha [una sorella tm]i . . . tk . . .
Again, this allows us to make sense of the presence of expletive ci in such possessive sentences - ci (and similarly for ghe in the relevant dialects) has been carried over into such possessive sentences from the existential substructure embedded in them. In contrast, of course, expletive there cannot appear in the corresponding English simple possessive sentences: (96) *John has there a sister. (96) is not acceptable and (97) is not a simple possessive sentence, i.e., its there must be interpreted as a locative, parallel to (98): (97) John has a sister there. (98) John has a sister in Chicago.
Richard S. Kayne
This restriction on English, as compared with some Italian and many Italian dialects, is plausibly to be related to the fact that expletive there, unlike expletive ci or ghe, occupies a subject position.37 Thus the impossibility of (96) could be attributed to “competition” between the expletive and the possessor for the same position. Alternatively, in part, one might say that in the spirit of relativized minimality – cf. Rizzi (1990) – the movement of the possessor to the highest subject position is blocked in (96) by the presence of expletive there in an intervening subject position.
11. Comparative syntax of possessives and existentials The account just suggested for (96), although it may well be on the right track for English, does not extend to the fact that the French expletive y that is a close counterpart of Italian ci never occurs in French possessive sentences, in any variety of French, as far as I know: (99) Jean (*y) a une soeur. Jean there) has a sister ‘Jean has a sister’
The reason that (99) with expletive y is impossible cannot be exactly the reason suggested for English (96); the proposal for English does not carry over to French because French expletive y is in an object clitic position (just like Italian ci) and not in subject position. Let me therefore propose a separate account for French (which will in turn not carry back over to English, as it arguably should not). This account of (99) will, in line with earlier discussion, involve the syntax of existentials. In existentials, French has: (100) Il y a un livre sur la table. it there has a book on the table ‘There is a book on the table’
. If the contrast in:
(i) John has a sister.
(ii) *John has the sister.
is due to the same (definiteness) effect found with existentials illustrated in (51), we might be led to postulate a silent there in (i) parallel to the overt ci of (89). This silent there would not interfere with the raising of the possessor (for reasons to be made precise – cf. Kayne (2006)), but it would call for the kind of splitting seen in (70) that leads to the definiteness effect. For relevant discussion, see Szabolcsi (1986).
Expletives, datives, and the tension between morphology and syntax
with y an object clitic-like expletive parallel to Italian ci not only in position, but also in that y, like ci, occurs elsewhere as a locative (with plACE), as in (17), repeated here: (101) Jean y va. Jean there goes ‘Jean goes there.’
and also with THING (and a silent P), as in (7), repeated here: (102) Jean y pense. Jean there thinks ‘Jean thinks about/of it.’
Unlike Italian, though, French existential sentences show what seems to be a second expletive, the subject clitic il, as seen in (100). (Furthermore, the verb in French is “have” rather than “be” – both Italian and English have “be”). The fact that French “have” cooccurs in (100) with expletive y recalls the cooccurrence of Italian ‘have’ with expletive ci in (89), and suggests that French (100), like Italian (89), contains, despite appearances, a possessor subject. The most natural interpretation of this conclusion is, I think, that the possessor subject in question in French is the subject clitic pronoun il. Thinking of Chomsky (1981: 325), a way to put this is to say that the il of (100) is a “quasi-argument” (rather than a true argument). The status of the il of (100) is then significantly similar to that of the il of French weather sentences like: (103) Il pleut. it rains ‘It is raining.’
with which it shares the property of being zero (rather than the usual accusative le/l’) in accusative contexts: (104) ?Un malentendu a misunderstanding
(*l’) a fait y avoir trop d’ enfants it has made there have too-many of children
à la soirée. at the party
‘A misunderstanding led to there being too many children at the party’.
Although embedding an existential like There were too many children at the party under a causative in French is somewhat marginal (perhaps as in English), the
Richard S. Kayne
result is clearly, as seen in (104), appreciably more acceptable without le/l’ than with it. This is the same property that one finds with weather verbs:38 (105) Les savants sont maintenant capables de (*le) faire pleuvoir. the scientists are now capable of it making to-rain ‘Scientists are now able to make it rain’.
The subject il of French existentials also shares with the il of French weather sentences the ability to control PRO (much as in Chomsky’s discussion of English; cf. also Kayne (1979: 713)): (106) Il pourrait y avoir du pain sans y avoir de l’eau. it could there to-have of-the bread without there to-have of the water’ ‘There could be bread without there being water’. (107) Il peut neiger sans pleuvoir. it can to-snow without to-rain ‘It can snow without it raining’.
Note that in the French existential control example (106) il does not appear in the infinitival part (it is replaced by PRO), while y does. This is indirectly related to the fact that English there cannot be a controller: (108) There can’t possibly be a solution without *(there) being a problem. (109) *There was a problem before being a solution.
The reason that English prohibits control with expletive there, in contrast to (106), is that expletive there (like y) is not a quasi-argument (but rather a deictic element that is not an argument at all), contrary to il. (106) can thus have PRO as the subject of the infinitival phrase beginning with sans, whereas in (108) and (109) the subject of the gerund cannot be PRO; put another way, in both languages the existential argument within the controlled infinitive or gerund needs to be merged with a deictic – in French that deictic y can cooccur with PRO since they occupy different types of positions, contrary to what holds in English. In summary, neither of the two elements that might have appeared to be expletives in French (100) is actually an expletive. The il is a quasi-argument, and the y is a deictic modifier that originates within the associate. As for the derivation of a French existential sentence such as (100), the best way to see it is to begin by going back to (93)–(95), consolidated here as the derivation of:
. Why exactly le is impossible here remains to be understood. Note the contrast with English:
(i)
Scientists are now capable of making *(it) be very cold.
Expletives, datives, and the tension between morphology and syntax
(110) Gianni c’ ha una sorella. Gianni there has a sister ‘Gianni has a sister’
which is the Italian possessive sentence containing expletive ci that we saw earlier in (89). The derivation goes as follows:39 (111)
. . . [ci una sorella Gianni]indef.DP . . . → raising of una sorella Gianni . . . [una sorella Gianni]i . . . [ci ti] . . . → merger of V . . . ha [una sorella Gianni]i . . . [ci ti] . . . → remnant movement . . . [ci ti]k ha [una sorella Gianni]i . . . tk . . . → raising of the possessor . . . Giannim [ci ti]k ha [una sorella tm]i . . . tk . . .
The proposal now is that (100) has a similar derivation, except that quasi-argument possessor il replaces the full possessor argument Gianni; un livre replaces una sorella; and y replaces ci: (112)
. . . [ y un livre il]indef.DP . . . → raising of un livre il . . . [ un livre il]i . . . [ y ti] . . . → merger of V . . . a [ un livre il]i . . . [ y ti] . . . → remnant movement . . . [ y ti]k a [ un livre il]i . . . tk . . . → raising of the possessor . . . ilm [ y ti]k a [ un livre tm]i . . . tk . . .
Consider in turn the result of replacing un livre in (100) by une soeur de Jean (‘a sister of J’): (113) Il y a une soeur de Jean dans la cour. it there has a sister of Jean in the courtyard ‘There is a sister of Jean in the courtyard’
This is a sentence whose derivation (abstracting away from the preposition de) must begin with “[ y [une soeur Jean] il]indef.DP” in place of “[ y [un livre] il]indef.DP”. That is, the derivation of (113) must contain an indefinite DP with two possessors, argument Jean and quasi-argument il. This is not implausible (as an instance of recursion), in particular given the existence in English of the quite acceptable (to me): (114) (?)This painting of yours of mine is now quite valuable.
. Here I’m taking ha = “have” to be in the numeration, and abstracting away from the question of preposition incorporation, on which, see Kayne (1993); Rigau (2005) and references cited in those works. The last step of the derivation in (111) involves the extraction of the possessor from within a previously moved constituent. In this respect it recalls (vs. Wexler & Culicover (1980: 278)):
(i)
That’s the problem that we explained to John only part of. (cf. Kayne (1994: 74))
Richard S. Kayne
in the sense of “this painting that you did that I own”.40 Returning to French (99), repeated here: (115) Jean (*y) a une soeur. Jean there has a sister ‘Jean has a sister.’
I think the answer to why it is impossible with expletive y, as opposed to Italian (110) being possible with expletive ci, lies in the following (unidirectional) comparative syntax correlation: (116) If a Romance language allows a clitic counterpart of expletive there in its possessive sentences, then its existential sentences have the verb be (and not have).
Particularly striking here is Catalan, whose deictic clitic hi is in other ways very much like Italian ci. Yet Catalan seems to be like French, rather than like Italian, with respect to (115). (116) claims that Catalan lacks expletive hi in its possessive sentences because Catalan has “have” in its existentials, like French. The next question is why (116) should hold. What I would like to propose as an answer to that question depends on a strong “uniformity”41 assumption about existential sentences (and hence about possessive sentences, given my adoption and extension of Szabolcsi’s analysis), namely that all (Romance) languages have a quasiargument possessor in their existentials (and hence in their possessive sentences), of the sort seen overtly in French il. (This implies that an existential sentence in which the associate contains a full argument possessor actually has two possessors.) Returning to (116) and to the unacceptability of (115) with expletive y, we can now understand (116) as follows. A language that allows an expletive deictic clitic in its possessive sentences, as do Italian and Paduan, must of necessity be a language in which the presence of that expletive clitic does not force the subject position to be filled by the quasi-argument possessor that by hypothesis is found in all simple possessive sentences.42 But in existentials, French and Catalan do have the quasi-argument possessor raising to subject position. (In French, the quasi-argument possessor must generally be pronounced, as opposed to Catalan.) . The fact that (114) is so much better than:
(i)
*My your painting is valuable.
which presumably has something to do with of, needs to be elucidated. . Cf. Chomsky (2001: 2). . At least in possessive sentences of the type under discussion, with an indefinite possessee. Probably the text approach should be extended to cases like:
(i)
John has it/your pencil in his pocket.
though with additional structure – cf. in part note 21.
Expletives, datives, and the tension between morphology and syntax
Consequently, we can exclude French (115) with expletive y by taking the raising of the quasi-argument possessor in French to be more general: (117) In French (and similarly for Catalan), the quasi-argument possessor must, if expletive y is present, raise to subject position. This will exclude (115), as desired.
A more general formulation of (117) would be: (118) If a language has quasi-argument possessor raising in the presence of an overt deictic expletive clitic in some cases (in particular in existentials), then it has it in all cases (thereby blocking full argument possessor raising in the presence of the overt expletive).
Again, (115) is correctly excluded. If y is not pronounced, then the French (and Catalan) quasi-possessor does not (and cannot, for reasons to be determined) raise to subject position, in which case nothing blocks the raising of the full argument possessor Jean, yielding the acceptable variant of (115). Note that in the derivation (112) what is raised into subject position cannot, even in the absence of overt il, be the associate itself: (119) *Un livre y a sur la table. a book there has on the table
This might be due to the fact that in that derivation un livre is not a full DP argument, having been raised out in the first step from within the argument phrase containing y;43 alternatively, there might be an intervention/relativized minimality effect, with y constituting a block. In Italian and in the dialects of Italy, the verb in existentials is “be” rather than “have”. I interpret this to mean that in those languages the quasi-argument possessor has not raised to nominative subject position. The fact that it does not have to, even in the presence of expletive ci or ghe, will make it possible for the full argument possessor to do so even in their presence, yielding sentences like (110), in a way compatible with (116). (On the fact that English acts differently, despite having be in existentials, see the discussion of (96).) 12. Other languages and no languages Some Scandinavian languages are like English in having existentials with be and with a counterpart (der) of subject there.44 This der can be taken to originate as . Cf. Rigau (2005: note 22). . Cf. Allan et al. (1995: sect. 407).
Richard S. Kayne
a deictic modifier of the associate, as discussed for English. Other Scandinavian languages can have be with a subject (det) that seems more like English it or that.45 This det might be akin to French il, i.e., it might be a quasi-argument possessor, though that would leave open (for the time being) why the verb is be; alternatively, this det might have some other status (perhaps related to the it of clefts) that remains to be elucidated. In Swiss German and other southern varieties of German, existentials can have the verb have, with a subject es that corresponds in other ways to English it (examples from Thomas Leu): (120) Es het es buaech uf em tisch. it has a book on the table ‘There is a book on the table.’
This subject es looks very much like a quasi-argument counterpart of French il. Although in (120) there is no overt element corresponding to French y, Swiss German also allows: (121) Da het s es buaech uf em tisch. there has it a book on the table ‘There is a book on the table.’
with an additional da that may correspond to y. Where a Swiss German existential has es het . . . a standard German existential would have es gibt . . . , with (probably) the same es, but with the verb “give”. This use of “give” in existentials has a (nonproductive) counterpart in English: (122) What gives?
In Spanish and Portuguese, the verb in existentials is a form of have, without there being any visible counterpart of French y (except perhaps in the present tense in Spanish). Since Spanish and Portuguese have no visible y elsewhere, either, they plausibly have a silent one in existentials. In addition, they are like Catalan in having no overt counterpart of French il. As we can see, there is substantial variation in the form of existentials across Romance and Germanic. This should not prevent us, however, from “seeing” what is not present. One gap of interest can be illustrated with an (unacceptable) English example: (123) *There has a book on the table.
. Cf. Holmes & Hinchcliffe (1994: 140).
Expletives, datives, and the tension between morphology and syntax
Subject deictic expletive there is not compatible with have in English. But as far as I know, no Romance or Germanic language has an exact counterpart of (123), with verb “have” and a deictic element (rather than a quasi-argument) in subject position. This follows directly from the assumption made earlier that have requires a possessor subject,46 whether quasi-argument (French il, (Swiss) German es) or full argument. Combined with the radically different status of there, which (like French y, Italian ci, Paduan ghe, Catalan hi) is a deictic modifier and not an argument or quasi-argument, this accounts for the general absence of (123). I have been assuming that the French type of existential seen in (100), with verb “have” and with deictic y in object clitic position and quasi-argument possessor il in subject (clitic) position, is closely matched by Catalan (example from Rigau (2005: 777)): (124) Hi ha una biblioteca nova. there has a library new ‘There is a new library.’
with the single difference that Catalan, in a way related to its being a null subject language, has a silent counterpart of il. Less immediate is the answer to the question where the quasi-argument possessor is in Italian existentials: (125) C’ è un libro sul tavolo. there is a book on-the table ‘There is a book on the table.’
Since Italian ci is not in subject position, Italian could perhaps (though not if (118) is correct) be like Catalan in having a silent il there. Alternatively, the quasi-argument possessor in Italian is silent and oblique (thinking of the fact that argument possessors are oblique in many languages47).
. (123) contrasts with:
(i) There has to be a book somewhere.
(ii)
There have been lots of problems.
indicating that auxiliary-like have (and similarly in other languages) does not require a possessor subject in the same way, despite other similarities with main verb have that led Kayne (1993) to assimilate the two to a significant degree. . For interesting discussion, see Hoekstra (1994). Even if oblique, the quasi-argument in Italian must be able to be a controller, given (from Burzio (1986: 174)):
(i)
Potrebbe esserci del pane senza esserci dell’ acqua. to-be there of-the bread without to-be there of-the water ‘There could be bread without there being water’.
could
Richard S. Kayne
13. Existentials and causers As alluded to just above (122), German has one existential with the verb give: (126) Es gibt keine Lösung. it gives no solution ‘There is no solution.’
Since give is normally causative, it seems odd at first glance that (126) could have an existential interpretation. The various languages that use “have” in their existentials (French, Catalan, Spanish, Portuguese, Swiss German and others) appear not to pose a problem of this sort, insofar as “have” is normally not thought of as causative. Yet English “have” can be causative, in sentences with an agentive subject: (127) We’ll have them call you.
Somewhat similarly, although English get seems straightforwardly like an inchoative of have in: (128) They got a large inheritance. and like an inchoative of be in: (129) They got sick.
get can also be causative: (130) They got him arrested. (131) They got us talking to each other. (132) They got us to lend them money.
A possible interpretation of these facts is that (126) actually does have causative content, and so do existentials with have in the various languages mentioned (and arguably, then, existentials with be in languages like English and Italian). This causative content could be understood by thinking of a sentence like There’s no solution as similar to: (133) Something has caused there to be no solution.
in which case the es of (126) could be taken to be a quasi-argument causer. That in turn would lead to the possibility that French il in existential il y a . . . is also a quasi-argument causer48 rather than a quasi-argument possessor, and similarly for the other quasi-argument possessors postulated earlier.
. Leading to the possibility that other instances of French il are, too, for example, those of:
(i) Il
faut que vous partiez. it needs that you leave ‘It is necessary for you to leave’.
Expletives, datives, and the tension between morphology and syntax
Postulating the presence of a non-agentive causer where none is visible is also a tempting option in the case of what are sometimes called “anti-causative” verbs such as sink, which enter into pairs like: (134) The boat sank. (135) The navy/the storm sank the boat.
In (135) the causer can be agentive (the navy) or non-agentive (the storm). At the same time, there is a well-known argument concerning a possible implicit argument in (134), based on the contrast:49 (136) The boat was sunk in order to collect the insurance. (137) *The boat sank in order to collect the insurance.
The deviance of (137) does seem to indicate that (134)/(137) contains no implicit agent, contrary to (136). But it does not exclude the possibility that (134) contains an unpronounced nonagentive causer. Saying that both (126) and (134) contain a non-obvious causer argument is of some interest, if we return to the question of the variation within Romance concerning “have” and “be” in existentials, specifically to the difference between French, which has “have”, and Italian, which has “be”. The reason for thinking this is of interest lies in the fact that French and Italian also differ with respect to verbs like sink, and do so in a way quite parallel to the verb difference in existentials, insofar as French, with anti-causatives like sink, uses auxiliary “have” with past participles, while Italian uses auxiliary “be”: (138) Le bateau a coulé hier. the boat has sunk yesterday ‘The boat sank yesterday’
(French)
(139) La nave è affondata ieri. the boat is sunk yesterday ‘The boat sank yesterday’
(Italian)
(ii)
Il me semble que vous avez it me seems that you have ‘It seems to me that you are right’.
(iii)
Il est important que . . . it is important that ‘It is important that . . . ’
(iv) Il est arrivé quelqu’un. it is arrived someone ‘Someone has arrived’. On (ii), see Rooryck (1997). . Cf. Williams (1985).
raison. reason
Richard S. Kayne
In other words, there may be a generalization that spans existentials and (pastparticipial) anticausatives (a more natural class from the perspective of this section than usually assumed), to the effect that French uses ‘have’ where Italian uses “be”.50 14. Limitations on deictic there as expletive In (my) colloquial English, expletive there occurs only with be. Although the following are possible in some register(s) of English, they are for me impossible in colloquial English:51 (140) There exist solutions to all these problems. (141) There have arrived several letters for you.
In this respect, colloquial English is like both Italian and French, whose deictic expletive ci and y are limited to existentials with be (in Italian) and have (in French) and which do not occur in the counterparts of (140) and (141). In contrast, as discussed by Burzio (1986: chapter 2), Piedmontese expletive clitic ye is found more widely, occurring as it does with all unaccusatives. Why Piedmontese should be freer in this regard than Italian or French is not clear (perhaps there is a link to the fact that Piedmontese object clitics generally follow past participles in a way that Italian and French object clitics do not). Nor is it clear why be is singled out by Italian, French and colloquial English – perhaps it is that be is associated with less structure than any other verb – or even that be is not really a verb (thinking of Postma (1993); cf. also Baker (2003: sect. 2.4)). Considering the partial derivation given earlier: . French and Italian differ in what seems to be the same fashion, when it comes to the auxiliary used with the past participle of seem, appear, disappear and be itself (i.e., French uses auxiliary have and Italian, auxiliary be) – cf. Burzio (1986: 138). Whether some or all of these should be integrated into the text discussion is left an open question, as is the question why die takes auxiliary be in both languages. The contrast between (138) and (139) also seems to be mimicked by French vs. Italian periphrasic causatives (indirectly supporting the text proposal), in the sense that they can passivize in Italian but not in French – cf. Kayne (1985). . As opposed to:
(i) There they go.
like:
(ii)
Here they come.
which do not involve expletive there, despite having some special properties.
(142)
Expletives, datives, and the tension between morphology and syntax
. . . [there a book] . . . → raising of a book . . . a booki . . . [there ti] . . . → merger of V . . . is a booki . . . [there ti] . . . → remnant movement . . . [there ti]k is a booki . . . tk . . .
it might be that the landing site needed for the first movement step is unavailable in these languages except with be. In Piedmontese, expletive ye does not occur with transitives, or even (Luigi Burzio, p.c.) with unaccusatives embedded under an overt causative. Again, it may be that the first step of (142) cannot proceed in the face of the extra structure associated with transitives. Although French unaccusatives do not show expletive y, they do show a subject clitic il that might be taken to be a (different) kind of expletive: (143) Il est arrivé trois lettres. it is arrived three letters ‘There have arrived three letters.’
Alternatively, this il might turn out to be the quasi-argument il discussed earlier – cf. the discussion beginning at (100) and note 48. (Note that this il is not, strictly speaking, limited to unaccusatives – cf. Pollock (1998: note 11).) 15. Datives The last case of parametric variation having to do with expletives that I will touch on has to do with the topic that this paper opened with, namely the deictic clitic ghe that is widely found in North Italian dialects in dative sentences where one might not have expected it. Put another way, these North Italian dialects differ from standard French and from standard Italian, which have retained an overt specifically third person dative clitic. The question how best to understand this difference in syntactic behavior rests in part on how one analyzes the ghe of, for example, Paduan sentences like: (144) Ghe dago un libro (a Gianni). ghe I-give a book to Gianni ‘I give a book to Gianni.’
One of the proposals made earlier was that such sentences contain a silent dative clitic: (145) DatCL ghe dago un libro (a G)
I have also argued in favor of taking all instances of ghe, like all instances of there in English, to be the same element. In this final section, I would like to take the more specific position that the ghe of (144) is the expletive subtype of deictic ghe, i.e., that it originates, like all the
Richard S. Kayne
other expletive deictics so far discussed, within the associate (which in (144) is un libro).52 This proposal rests on two points, the first being that sentences like (144), with a verb corresponding to give, can, following a long tradition, be thought of as arising through the embedding of a have-like structure within a causative one. The second is that have-sentences in a certain number of Italian dialects and for a certain number of Italian speakers show an overt expletive, as discussed earlier beginning with (89). The idea, then, is that expletive ghe appears in (144) via “inheritance”. Dative sentences embed within them possessive sub-sentences that can contain an expletive, in turn inherited from the existential subpart of the possessive structure itself. Thus the ghe of (144) is in essence the familiar expletive of existentials. Needless to say, things are not quite this simple. But we can simply illustrate the first point using French periphrastic causatives of the sort that dativize the embedded subject: (146) Ils ont fait avoir un prix à cet étudiant. they have made have a prize to that student
This sentence, which has the approximate interpretation of: (147) Ils ont donné un prix à cet étudiant. they have given a prize to that student ‘They have given a prize to that student.’
embeds under the causative verb faire a sentence containing avoir (“to-have”). The subject of the embedded verb avoir ends up as the ‘object’ of the (dative) preposition à, in a way discussed in Kayne (2004b). Assume now that a have-sentence embedded under a causative can itself contain an expletive (e.g., ghe), as we know to be visibly possible for have-sentences in various dialects. Then in the relevant dialects, this expletive (deictic) ghe will be visible in their counterparts of (146) and, assuming a parallel derivation, of (147). In which case we have an answer to the question we started with, namely why it is that ghe, an apparent locative (but really a deictic element, as argued above), appears in dative sentences like (144) in the first place. Rather than reflecting syncretism, the presence of ghe in (144) more precisely reflects a piece of the underlying (in part existential) syntactic structure of such sentences.53 . (144) is also possible with definite el libro (“the book”) in place of indefinite un libro. See notes 21 and 42, as well as the possibility, thinking of Koopman (2003, 2005), that definites are or can be built upon indefinites, in relative clause fashion. . This will have to be extended to cover the whole range of dative sentences. Guglielmo Cinque (p.c.) asks why expletive ghe can cooccur with a silent DatCL in (144)/(145), yet never
Expletives, datives, and the tension between morphology and syntax
Left open is the question why we don’t see this expletive in all Romance languages (or in all languages). It does not appear to be possible to say that ghe can be found in dative sentences only if it is found in possessive sentences, since, as Paola Benincà points out (p.c.), Bellunese has ghe in datives but not in possessives.54 Nor, thinking of Italian itself, does the converse seem to hold, since there are Italian speakers who have ci in possessives without having it in datives. Ultimately, the answer will probably involve the question of whether or not clitics belonging to the embedded sentence of a causative structure such as (146) can successfully surface.55 The question also arises as to why ghe (or ci or y or there) appears in the first place in existentials in some languages. French existentials with il y a . . . indicate clearly that filling a subject position cannot be a general answer, since in French the subject position is filled by quasi-argument il, so that the presence of deictic expletive y must rest on other considerations. Part of the answer may be that these deictic elements are an obligatory part of all DPs (including indefinites) in all languages,56 with various factors combining to determine whether they are pronounced, in one or another language, in one or another context.
16. Conclusion Expletive there and its closest counterparts in other languages are not expletives in Chomsky’s sense (merged directly into a sentential Spec-position). They are instead instances of deictic elements originating within their associate.57 In seems to be able to license a silent acc-cl. In sentences with just an accusative object, the answer should be that there is no source for expletive ghe (i.e., no existential substructure) in the first place. In ditransitive sentences, we had an example with a silent accusative in Sardinian in (28); the contrast with Paduan remains to be elucidated. . Perhaps like Bellunese is the popular French described by Postal (1990: 188, note 19) (unless the y there turns out to be a dative Case morpheme). . For some relevant data, see Kayne (1975); Rouveret & Vergnaud (1980) and Burzio (1986). We will also need to understand why in some dialects/languages, such as the Carmignano dialect studied by Penello (2004), ghe can in dative sentences precede the subject clitic. . Cf. Jayaseelan & Hariprasad (2001). How exactly to characterize the interpretive contribution to indefinites of there remains to be elucidated; cf. the discussion of (75). . This is close to Sabel (2000) (cf. also Moro (2000: 125) and Choe (2006)), with the difference that he takes the expletive to correspond to a D-feature within the associate, whereas I have suggested a reduced relative source (cf. the discussion after (43)) that is compatible with taking there to be bi- or perhaps tri-morphemic. That there is at least bi-morphemic is clearly indicated by its
Richard S. Kayne
some languages, e.g., Paduan, these expletives can be “imported” into possessive sentences and (from there) into dative sentences, giving the impression in the latter case of syncretism between deictic clitic and dative clitic, but the correct, more syntactic and less morphological, analysis is that in such languages the deictic clitic is not a dative clitic, though it cooccurs with a dative clitic that is silent.
References Aboh, E.O. 2004. The Morphosyntax of Complement-Head Sequences. Clause Structure and Word Order Patterns in Kwa. Oxford: OUP. Allan, R., Holmes, P. & Lundskaer-Nielsen, T. 1995. Danish. A Comprehensive Grammar. London: Routledge. Baker, M. 2003. Lexical Categories. Verbs, Nouns, and Adjectives. Cambridge: CUP. Belletti, A. 2005. Extended doubling and the VP periphery. Probus 17: 1–35. Benincà, P. 1989. Friaulisch: Interne Sprachgeschichte. Grammatik. Evoluzione della grammatica. In Lexikon der Romanistischen Linguistik, 563–585. Tübingen: Niemeyer. Bennis, H. 1986. Gaps and Dummies. Ph.D. dissertation, Tilburg University. Bernstein, J. 1997. Demonstratives and reinforcers in Romance and Germanic languages. Lingua 102: 87–113. Brustad, K.E. 2000. The Syntax of Spoken Arabic. A Comparative Study of Moroccan, Egyptian, Syrian, and Kuwaiti Dialects. Washington DC: Georgetown University Press. Burzio, L. 1986. Italian Syntax. A Government-Binding Approach. Dordrecht: Reidel. Choe, H.S. 2006. The syntax of existential there and the structure of nominal expressions. Ms. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale. A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Cinque, G. 1990. Types of A’-Dependencies. Cambridge MA: The MIT Press. Cordin, P. 1991. Il raddoppiamento del dativo in trentino. In Tra Rinascimento e strutture attuali. Saggi di linguistica italiana. Atti del I Convegno della SILFI, L. Giannelli (Ed.), 399–410. Torino: Rosenberg and Sellier. Cresti, D. 2003. Aspects of the syntax and semantics of ne. In The Syntax of Italian Dialects, C. Tortora (Ed.), 67–101. Oxford: OUP. Dasgupta, P. 1992. Pronominality and deixis in Bangla. Linguistic Analysis 22: 61–77. Delsing, L.-O. 1993. The Internal Structure of Noun Phrases in the Scandinavian Languages. A Comparative Study. Ph.D. dissertation, University of Lund.
parallelism to here (and perhaps where – cf. note 34), by the plausibility of an initial morpheme th-related to the th-of the, that, etc., and by the relation between Dutch daar (“there”) and Dutch er (“-ere”). Possibly, adapting a suggestion of Pierre Pica’s, one of the morphemes (other than th-) making up there is prepositional, thinking also of the fact that the closest counterpart to there in Egyptian Arabic is prepositional, as brought to my attention by Maha Aboul-Ela – cf. Brustad (2000: 152).
Expletives, datives, and the tension between morphology and syntax
Den Dikken, M. 1997. The syntax of possession and the verb ‘have’. Lingua 101: 129–150. Fiengo, R. & Higginbotham, J. 1981. Opacity in NP. Linguistic Analysis 7: 395–421. Heim, I. 1987. Where does the definiteness restriction apply? Evidence from the definiteness of variables. In The Representation of (In)definiteness, E.J. Reuland & A.G.B. ter Meulen (Eds), 21–42. Cambridge MA: The MIT Press. Hoekstra, T. 1994. HAVE as BE plus or minus. In Paths toward Universal Grammar. Studies in honor of Richard S. Kayne, G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi & R. Zanuttini (Eds), 199–215. Washington DC: Georgetown University Press. Holmes, P. & Hinchliffe, I. 1994. Swedish. A Comprehensive Grammar. London: Routledge. Jayaseelan, K.A. & Hariprasad, M. 2001. Deixis in pronouns and noun phrases. Linguistic Analysis 31: 132–149. Johnson, K. 1991. Object positions. Natural Language and Linguistic Theory 9: 577–636. Jones, M.A. 1993. Sardinian Syntax. London: Routledge. Kayne, R.S. 1975. French Syntax. The Transformational Cycle. Cambridge MA: The MIT Press. Kayne, R.S. 1979. Rightward NP Movement in French and English. Linguistic Inquiry 10: 710–719. Kayne, R.S. 1985. L’accord du participe passé en français et en italien. Modèles Linguistiques VII:73–90. (English translation in Kayne (2000)). Kayne, R.S. 1993. Toward a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. (Reprinted in Kayne (2000)). Kayne, R.S. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press. Kayne, R.S. 2000. Parameters and Universals. Oxford: OUP. Kayne, R.S. 2001. A note on clitic doubling in French. In Current Studies in Italian Syntax. Essays offered to Lorenzo Renzi, G. Cinque & G. Salvi (Eds), 189–212. Amsterdam: NorthHolland. (Reprinted in Kayne (2005)). Kayne, R.S. 2002. On some prepositions that look DP-internal: English of and French de. Catalan Journal of Linguistics 1: 71–115. (Reprinted in Kayne (2005)). Kayne, R.S. 2004a. Here and there. In Syntax, Lexis and Lexicon-Grammar. Papers in honour of Maurice Gross, C. Leclère, E. Laporte, M. Piot & M. Silberztein (Eds.), 253–273. Amsterdam: John Benjamins. (Reprinted in Kayne (2005)). Kayne, R.S. 2004b. Prepositions as probes. In Structures and Beyond. The Cartography of Syntactic Strutures, Vol. 3, A. Belletti (Ed.), 192–212. Oxford: OUP. (Reprinted in Kayne (2005)). Kayne, R.S. 2005. Movement and Silence. Oxford: OUP. Kayne, R.S. 2006. On parameters and on principles of pronunciation. In Organizing Grammar. Linguistic Studies in Honor of Henk van Riemsdijk, H. Broekhuis, N. Corver, R. Huybregts, U. Kleinhenz & J. Koster (Eds), 289–299. Berlin: Mouton de Gruyter. Kayne, R.S. 2007a. Several, few and many. Lingua 117: 832–858. Kayne, R.S. 2007b. A short note on where vs. place. In Miscellanea di studi Linguistici offerti a Laura Vanelli da amici e allievi padovani, R. Maschi, N. Penello & P. Rizzolatti (Eds), 245–257. Udine: Forum. Kayne, R.S. & Pollock, J-Y. 2001. New thoughts on stylistic inversion. In Inversion in Romance, A. Hulk & J.-Y. Pollock (Eds), 107–162. Oxford: OUP. (Reprinted in Kayne (2005)) Koopman, H. 2003. The locality of agreement and the structure of the DP in Maasai. In The Role of Agreement in Natural Language: TLS 5 Proceedings, Texas Linguistic Forum 53, W.E. Griffin (Ed.), 206–227. Austin TX: Texas Linguistic Society. Koopman, H. 2005. On the parallelism of DPs and clauses. Evidence from Kisongo Maasai. In Verb First, On the Syntax of Verb-initial Languages, A. Carnie, H. Harley & S.A. Dooley (Eds), 281–302. Amsterdam: John Benjamins.
Richard S. Kayne Leu, T. 2006. From Greek to Germanic: The structure of adjectival modification. Paper presented at GLOW, Barcelona. Leu, T. 2007. These HERE demonstratives. UPenn Working Papers in Linguistics 13(1): 141–154. Longa, V.M., Lorenzo, G. & Rigau, G. 1998. Subject clitics and clitic recycling: Locative sentences in some Iberian Romance languages. Journal of Linguistics 34: 125–164. Longobardi, G. 1996. The syntax of N-raising: A minimalist theory. OTS Working Papers: 1–55. Milsark, G. 1974. Existential Sentences in English. Ph.D. dissertation, MIT. Moro, A. 1997. The Raising of Predicates. Predicative Noun Phrases and the Theory of Clause Structure. Cambridge: CUP. Moro, A. 2000. Dynamic Antisymmetry. Cambridge MA: The MIT Press. Penello, N. 2004. Capitolo di morfologia e sintassi del dialetto di Carmignano di Brenta. Ph.D. dissertation, Università di Padova. Perlmutter, D.M. 1970. On the article in English. In Progress in Linguistics, M. Bierwisch & K.E. Heidolph (Eds), 233–248. The Hague: Mouton de Gruyter. Perlmutter, D.M. 1971. Deep and Surface Structure Constraints in Syntax. New York NY: Holt, Rinehart and Winston. Pollock, J.-Y. 1998. On the syntax of subnominal clitics: Cliticization and ellipsis. Syntax 1: 300–330. Postal, P.M. 1990. French indirect object demotion. In Studies in Relational Grammar 3, P.M. Postal & B.D. Joseph (Eds), 103–200. Chicago IL: University of Chicago Press. Postma, G. 1993. The syntax of the morphological defectivity of BE. HIL Manuscripts 3: 31–67. Postma, G. & Rooryck, J. 1996. Modality and possession in NPs. In Proceedings of NELS 26, K. Kusumoto (Ed.), 273–298. Amherst MA: GLSA. Riemsdijk, H. van 1978. A Case Study in Syntactic Markedness. Lisse: Peter de Ridder Press. Rigau, G. 2005. Number agreement variation in Catalan dialects. In Handbook of Comparative Syntax, G. Cinque & R. Kayne (Eds), 775–805. Oxford: OUP. Roberts, I. 1993. The nature of subject cliticisation in Franco-Provençal Valdôtain. In Syntactic Theory and the Dialects of Italy, A. Belletti (Ed.), 219–353. Torino: Rosenberg and Sellier. Rooryck, J. 1997. On the interaction between raising and focus in sentential complementation. Studia Linguistica 51: 1–49. Rouveret, A. & Vergnaud, J.R. 1980. Specifying reference to the subject: French causatives and conditions on representations. Linguistic Inquiry 11: 97–202. Rizzi, L. 1990. Relativized Minimality. Cambridge MA: The MIT Press. Sabel, J. 2000. Expletives as features. In WCCFL 19 Proceedings, R. Billerey et al. (Eds), 411–424. Somerville MA: Cascadilla. Saeed, J.I. 1993. Somali Reference Grammar, 2nd revised Edn. Kensington MD: Dunwoody Press. Savescu Ciucivara, O. 2006. Romanian – a PCC Language? Paper presented at LSRL 36, Rutgers University. Sichel, I. 2001. Studies in the Syntax of Pronouns and Features with Particular Reference to Hebrew. Ph.D. dissertation, CUNY, New York. Sportiche, D. 2002. Movement types and triggers. GLOW Newsletter 48: 116–117. Starke, M. 2001. Move Dissolves into Merge: A Theory of Locality. Ph.D. dissertation, University of Geneva. Szabolcsi, A. 1983. The possessor that ran away from home. The Linguistic Review 3: 89–102.
Expletives, datives, and the tension between morphology and syntax
Szabolcsi, A. 1986. From the definiteness effect to lexical integrity. In Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984, W. Abraham & S. de Meij (Eds), 321–348. Amsterdam: John Benjamins. Szabolcsi, A. 1994. The noun phrase. In Syntax and Semantics 27. The Syntactic Structure of Hungarian, F. Kiefer & É.K. Kiss (Eds), 179–274. San Diego CA: Academic Press. Terzi, A. 1999. Clitic combinations, their hosts, and their ordering. Natural Language and Linguistic Theory 17: 85–121. Vinet, M.-T. 1977. Etudes sur le syntagme nominal: La dislocation, l’adjectif ‘propre’. Ph.D. dissertation, University of Paris VIII. Wexler, K. & Culicover, P.W. 1980. Formal Principles of Language Acquisition. Cambridge MA: The MIT Press. Whitman, J. 1981. The internal structure of NP in verb final languages. In Papers from the Seventeenth Regional Meeting of the Chicago Linguistic Society, R.A. Hendrick, C.S. Masik & M.F. Miller (Eds), 411–481. Chicago IL: Chicago University Press. Williams, E. 1985. PRO and subject of NP. Natural Language and Linguistic Theory 3: 297–315.
Mapping a Parochial Lexicon onto a universal semantics Gillian Ramchand & Peter Svenonius CASTL, University of Tromsø
In this paper, we argue that languages differ in what parts of meaning are specified in the syntax and what parts are negotiated by the ConceptualIntentional systems (C-I). This leads to a kind of parametric variation, which we illustrate with examples from Norwegian definiteness, Russian perfectivity, Salish tense, and other natural language phenomena. Our claim is that syntactic variation is not as great as sometimes suggested (for example, we argue that Chinese has a D head and Salish has a T head), but nor are syntactico-semantic representations identical across languages; LFs vary from one language to the next, and in some cases C-I specifies what syntax/semantics does not, particularly when it comes to reference tracking for the variables introduced by the syntax. The result is a very clean system with no semantic module disinct from syntax and hence no distinction between syntactic and semantic parameters. Keywords: syntax; semantics; conceptual-intentional module; modularity; interface; tense; definiteness; Russian; Norwegian
1. Modularity and the structure of the language faculty In this paper we address the question of whether it makes sense to talk of “semantic parameters” in natural language. Essentially, this is a question about modularity. Is there a “semantic module” in which parametric variation can reside? We argue that there is not, though there is variation that is semantic in nature; we propose a way to treat it which does not invoke semantic parameters as they are usually understood. Languages clearly vary in their inventories of words or lexical entries, with different conceptual content and associations. Such differences are to a large extent arbitrary and probably limitless, not reducing to any obvious discrete parametric system. Supposing that those details which are not syntactically relevant can be identified and set aside (we will refer to them as encyclopedic, in essentially Borer’s 2005a sense), it becomes possible to pose questions about that variation which is systematic, as we do here.
Gillian Ramchand & Peter Svenonius
The question of modularity is a fundamental one in theoretical linguistics: to what extent is there compartmentalization of linguistic function? In our case, does a semantic module mediate between syntax proper and what Chomsky (2004) calls the Conceptual-Intentional systems (C-I)? Minimalist reasoning suggests that we explore the possibility that there is no modularization within syntax-semantics, and that a uniform system presents syntax/semantic repre sentations which are interpretable by C-I. In that case, that variation among languages which is not encyclopedic can only reside in featural specifications of lexical items. Within these strict architectural limits, is it possible to make sense of all the variation among languages where meaning seems to be implicated? We will argue that it is both possible and desirable to do so, but that it requires a division of “semantic” information into one subset that is an inherent part of the syn/sem computation, and another that can be cashed out through general conceptual mechanisms. 1.1 Underspecification of lexical content This raises the question of whether C-I is uniform from one language to the next, or subject to variation. Once again, purely conceptual reasoning leads us to the zero hypothesis, that (since we know there is variation in syntax) there is no variation within C-I (more precisely, systematic cross-linguistic variation is not to be explained in terms of different C-I systems). That leaves two possibilities: either the mapping from syntax to C-I is trivial, in which case all syntaxes have to be the same (and highly abstract) at the interface, or else the mapping from syntax to C-I is nontrivial, in which case two languages might express “the same thought” while feeding off different syntax/semantic representations.1 Thus, while we wish to claim that the system constructing syn/sem representations is the same for all languages (a universal computational system), we do not think that this forces the final representations for “the same thought” to be the same, essentially because of the fact that different lexical ingredients may be available in different languages. On one level, this is an obvious point, and one which can be demonstrated for contentful lexical items whose encyclopedic con tent may vary from language to language. For example, to report an act of smoking a cigarette, a Bengali speaker may say (1a), where the verb used is the same one
. Of course, it must be confessed that we have no way of deciding systematically when two utterances in different languages constitute “the same thought.” Levinson (2003) argues, for example, that under certain experimental conditions speakers of different languages will have different conceptualizations of the same situation. We will continue to assume, however, that truth conditions combined with discourse conditions on felicitous usage provide a workable framework for identifying equivalent C-I states, for present purposes.
Mapping a Parochial Lexicon onto a universal semantics
that is used for both eating and drinking as well, while an English speaker might say (1b), using a smoldering inhalant-specific lexical item “smoke.”2 (1) a.
ami sigret khacchi I cigarette imbibe.prog.pres.1sg ‘I am smoking a cigarette.’
b. I am smoking a cigarette.
Here, we claim that there is no reason to believe that the Bengali speaker and the English speaker are expressing “different thoughts” in the relevant sense. Rather, the Bengali speaker uses a more underspecified lexical item referring to general imbibing, whose specific situational reference must be negotiated through sentence-contextual and real world information. Examples such as this can be multiplied, but they are not restricted to purely lexical content words. In many languages, the pronominal system distinguishes between a first person plural inclusive and a first person plural exclusive form. If a language does not distinguish, then its first person plural form must do double duty, and the specific meaning retrieved from context. The same is true for languages with explicit dual marking, exemplified below for Northern Sámi. English does not express dual marking, but uses one word (underspecified) for both situations. The question is whether two concepts correlate with different features in the syn/sem system itself, with morphological homonymy (as in (2)), or whether the C-I system simply translates “we” in English more vaguely with the extra information filled in (when needed) by some nonlinguistic process, as suggested by (3). (2)
First person plural and dual pronouns Syntax/Semantics Northern Sámi English
(3)
Simplified Northern Sámi
English
mii moai we we
Conceptual-Intentional “I and others” “I and one other” “I and others” “I and one other”
Syntax/Semantics
Conceptual-Intentional
mii moai we
“I and others” “I and one other” “I and one or more others”
. Alternatively, the English speaker could use the even more general verb have, which is more general again than the Bengali verb, illustrating the same contrast again but in the opposite direction.
Gillian Ramchand & Peter Svenonius
In this paper, we will claim that the latter situation exists, not just for contentful lexical items, but systematically for a variety of “meanings” that have traditionally been considered functional. This does not mean, however, that variation is limitless. Rather, we defend the position that certain features are universally present in the syn/sem representations constructed by the computational system, but that two things can systematically vary: their particular encyclopedic content, and the information they carry that is relevant to discourse tracking and referential access systems. Deciding which features are universally necessary to a well-formed syn/ sem representation, and which ones can be contextually filled in (and are therefore, on our hypothesis, subject to cross-linguistic variation) is an empirical issue. Acknowledging this potential locus of variation, however, is important in assessing claims about the universal properties of the syn/sem combinatorial system. We assume that the same C-I interpretational mechanisms are available to all languages, and are fed by the same universal syn/sem system. However, the syn/ sem system inevitably underdetermines full contextually augmented meaning. Hence the points of variation in this domain will concern meaning contributions that are not part of the narrow computation proper but which could be part of the lexical information of a particular item in a particular language. At one level, this is the most trivial kind of cross-linguistic variation: if you don’t have a word for tomorrow you can always say “the day after today.” But at another level, it can have serious repercussions especially when the lexical items in question are more abstract, an area of variation that we think has been underappreciated. In this paper, we give some examples of systematic differences between syntactically and postsyntactically provided meaning components, where it is not necessary to assume either that the LF representations are identical or even that each language has a differently constituted semantic component. 1.2 Against a distinct semantic component To distinguish our idea clearly from one which involves a semantic component, we give an example of a specific recent analysis that does assume a semantic component which can be a locus of parametric variation. Chierchia (1998a,b) analyzes the variation in the expression of noun phrases in Chinese, Italian and Russian. Chierchia suggests that Chinese NPs are semantically arguments (type e), whereas Italian NPs are predicates, and are converted into arguments by a syntactic head, D. Russian, on Chierchia’s analysis, represents a third possibility, in which NPs are predicates but are made into arguments by a type-shifting rule. Chierchia’s analysis is designed to account for differences in the distribution of “bare” (determinerless) noun phrases, for example as illustrated in (4), where Italian
Mapping a Parochial Lexicon onto a universal semantics
places strict restrictions on prenominal bare plurals, correlated in his analysis with the conditions on licensing of null D. Chinese, which has type e NPs, and Russian, which can type-shift NP to type e, do not have the same restrictions. (4) a.
Xuesheng lai le. student come perf ‘The student(s) came.’ (Chinese)
b. *Bambini sono venuti da noi. children are come by us (‘Children came to our house’.) (Italian) c.
Student vo-ˇsël v komnatu. student in-came in room ‘A/The student came into the room.’ (Russian)
In Chierchia’s model we see that the very same syntactic form can give rise to two different semantic types in the different languages. The analysis proposed by Chierchia is schematically represented below in table (5). (5) Chierchia’s model Syntax
Semantics
Conceptual-Intentional
Italian
NP DETQ DETQ(NP) DETD DETD(NP)
P 〈e, t〉 Q 〈e, t〉,〈〈e, t〉,t〉〉 GQ 〈〈e, t〉,t〉 ι 〈〈e, t〉,e〉 De
predicate quantificational determiner generalized quantifier definite article definite description
Chinese
NP Ke DET'Q Q' 〈e,〈〈e, t〉,t〉〉 DET'Q(NP) GQ 〈〈e, t〉,t〉
kind quantifier plus lowerer generalized quantifier
Russian
NP DETQ DETQ(NP) NP
predicate quantificational determiner generalized quantifier definite description (by type-shift)
P 〈e, t〉 Q 〈〈e, t〉,〈〈e, t〉,t〉〉 GQ 〈〈e, t 〉,t〉 De
According to Chierchia’s proposal, the type-shifting process used by Russian is generally available, but its use is blocked in Italian by the availablility of a definite article (type-shift as a last resort). The difference between Chinese and Russian is that in Chinese, NPs translate into type e and do not require any determiner to fill argument positions. This is the kind of analysis that we claim to be unavailable in general. Instead, there are only two simple possibilities for the location of this difference between Chinese or Russian and English: it is in the syntax, that is, some other syntactic operator (another kind of D, for example an unpronounced one) is present in the
Gillian Ramchand & Peter Svenonius
apparently determiner-less languages; or it is in C-I, that is, the C-I can interpret an NP as an argument. In fact, recent work in both Chinese (Cheng & Sybesma 1999; Borer 2005b) and in Russian (Englehardt & Trugman 1998; Babyonyshev 1998) has argued for the presence of a null functional head performing the function of D in these languages. For example, Cheng & Sybesma (1999) argue that classifiers in Chinese function in a way very similar to the way determiners do in Italian. They point out that in contexts where a bare N is ungoverned, certain readings are absent, just as in Longobardi’s (1994) analysis of Italian. Furthermore, Cheng & Sybesma argue that in those contexts where Italian allows a seemingly bare N due to N-to-D movement, Chinese has N-to-Cl[assifier] movement. Thus, according to them, argumental noun phrases in Chinese are not simply NPs, but have functional structure, which we can loosely characterize as D for our purposes (though Cheng & Sybesma identify distinctions between the D of Italian and English and the Cl head of Chinese). For Slavic, too, there are various arguments for functional structure in the noun phrase, despite the absence of articles. For example, Rappaport (to appear; 2001) shows that extraction from argumental noun phrases in Polish is bad, but extraction from predicative noun phrases is acceptable, as illustrated in (6). (6) a. Zlikwidowano klub, kto´rego jestem they.disbanded club of.which I.am
honorowym honorary
członkiem. member
‘They closed the club of which I am an honorary member.’
b. *Zlikwidowano they.disbanded
klub, kto´rego wczoraj poznałem club of.which yesterday I.met
honorowego członka. honorary member
(‘They closed the club of which I met an honorary member yesterday.’)
As Rappaport argues, this can be straightforwardly accounted for in terms of a structural difference between DPs and NPs. Of course, an account which does not assume an additional layer of structure in (6b) can be developed, but what it would do is to mimic the effect of a D layer. In a similar vein, Romanova & Diakonova (2003) have argued that in Russian, extraction is possible from some NPs, when interpreted non-specifically, but not from others that are interpreted as specific. This is true even though neither NP shows a determiner overtly. Under this latter kind of analysis, no semantic component would be necessary, and Chierchia’s schema in (6) comes out as the simpler table in (7).
(7) Our model Italian Chinese Russian
Mapping a Parochial Lexicon onto a universal semantics
Syntax/Semantics NP 〈e, t〉 QP 〈〈e, t〉, t〉 D1Pe D2Pe NP 〈e, t〉 QP 〈〈e, t〉, t〉 DPe NP 〈e, t〉 QP 〈〈e, t〉, t〉 DPe
Conceptual-Intensional predicate generalized quantifier definite description indefinite predicate generalized quantifier definite description or indefinite predicate generalized quantifier definite description or indefinite
We think that on conceptual grounds, the theory that does not have a separate semantic component or module is superior to one which multiplies linguistic levels of representation in this way. Such a view would have the burden of proof, and we should prefer any analysis with the same empirical coverage that does not require such a component. 1.3 Against identical LF representations Taken to their logical conclusion, the conceptual reasons for not invoking a semantic component for the analysis of syntactic phenomena (such as the distribution of determiners just discussed) would give rise to an architecture where LF representations of the “same thought” in different languages are analysed as more and more similar to each other, even in the face of increasing abstractness. Such a view would essentially force us into the position of assuming universal syntactic status for every “feature” that shows morphosyntactic representation in even one human language. In our view, however, the assumption of identical LFs would also miss important generalizations. While certain syn/sem elements (such as D, by hypothesis in the example above) are necessary to the computation, some others are clearly negotiated contextually. We suggest that general, non-linguistic properties of the C-I system allow C-I to interpret certain aspects (but not others) of syntactic representations which may be left underspecified by the syntax, for example, coindexation, identification of referents, etc. We suggest that the power of C-I to fill in such information is very different from what is usually envisioned as part of a linguistic semantic component. For example, syn/sem features must be present throughout the syntactic derivation (cf. the Projection Principle of Chomsky 1981); C-I introduces information afterwards (possibly in a cyclic fashion, cf. Chierchia 2004), so that it cannot be referred to by syntactic rules. Furthermore, C-I interpretations are likely to be subject to coercion.
Gillian Ramchand & Peter Svenonius
In the example given above, the difference between Italian on the one hand and Chinese/Russian on the other survives as a difference in the syn/sem representations of the two language types (or LF, in traditional terminology) in our table (7). Specifically, Italian has two distinct D elements each of which carries different information concerning the discourse familiarity of the referent denoted by the DP. In our view, discourse definiteness is precisely the kind of “feature” that can either be expressed by a language’s lexical items, or filled in post-linguistically by the C-I system itself as a matter of crosslinguistic variation. Chinese and Russian both have an “underspecified” D, whose specific interpretation and organisation within a discourse representation is filled in by the C-I system. Because D heads in the phrase structure are required by UG, their presence can be inferred by the language learner and do not require overt morphophonological instantiation; because only one null item is inferred, its semantics must reduce to creating an entity interpreted as type e (the minimum required by the syn/sem computation) with everything else left underspecified. 1.4 Possible refinements Even in languages that are very similar, such as English and Norwegian, we find differences in the distribution and interpretation of articles. For example, Norwegian has suffixal articles which can encode definiteness when used with a bare noun, but which must be “doubled” with a word-like determiner element when an adjective is present (8).
(8) Norwegian double definiteness
a.
et hus ∼ huset a house house.def ‘a house’ – ‘the house’
b. et stort hus ∼ det store huset a big.n house the big.wk house.def ‘a big house’ – ‘the big house’
The definite suffix is also required with demonstratives and possessives (9). (9) a.
den koppen ∼ denne koppen that cup.def this cup.def ‘that cup’ – ‘this cup’
b. koppen min ∼ koppen til Ingebjørg cup.def my cup.def to Ingebjørg ‘my cup’ – ‘Ingebjørg’s cup’
When it comes to interpretational distribution, the definite suffix in Norwegian is required for abstract mass nouns (10a), and for kind denotations (10b)
Mapping a Parochial Lexicon onto a universal semantics
(Halmøy 2005), whereas English requires bare nouns in the first context and allows them in the second, as suggested in the translations. (10) a.
Livet/ *Liv er ikke en film. life.def/ life is not a film ‘Life is not a film’ (*‘The life …’)
b. Lyspæra/ *Lyspærer ble oppfunnet av Edison. lightbulb.def/ lightbulbs became invented by Edison ‘Lightbulbs were invented by Edison’ (or: ‘The light-bulb …’)
In general, the definite suffix in Norwegian may actually be correlated to “specificity” (Anderssen 2006; Julien 2005), while the word-like determiner element is responsible for familiarity. This can be seen in the following examples, from Anderssen (2006): the nouns intensjoner “intentions” and bit “bit, small piece” do not have the definite suffix. To include the definite suffix here (intensjonene, biten) would be grammatical, but would give an inappropriate meaning: an implicit comparison of intentions, an actual smallest piece not eaten. (11) Definite suffix correlated to ‘specificity’: a.
Han har de beste intensjoner. he has the best intentions ‘He has the best intentions.’ (i.e., very good ones)
b. Jeg spiste ikke den minste bit. I ate not the smallest piece ‘I didn’t eat any piece at all.’
In the absence of strong evidence that these correspond to distinct heads within a nominal extended projection, each with a distinct compositional semantic interpretation (and pending such a theory), we speculate that one way to see these facts is as in the table in (12) below. (12) Parametric variation in the negotiation of definiteness
Norwegian
English
Russian
Argumenthood Familiarity Specificity Tracking
Syntax/Semantics Syntax/Semantics Syntax/Semantics C-I
Syntax/Semantics Syntax/Semantics C-I C-I
Syntax/Semantics C-I C-I C-I
In other words, Norwegian has a suffix explicitly signalling specificity as well as a prenominal definite article signalling familiarity, but leaves reference tracking up to C-I; English signals familiarity with the definite article but leaves what we are calling specificity up to C-I, as well as discourse tracking; and Russian, lacking a definite article, resolves familiarity, specificity, and discourse-tracking at C-I.
Gillian Ramchand & Peter Svenonius
This does not mean that Russian has no D, however, as (we assume) D is necessary to provide an argument in the first place (there being no semantic module in which to type-shift independently of the syntactic structure).
2. Discourse is different We have so far exemplified the logic of our position using analyses of definiteness variation and the D head in various languages. This phenomenon is important because it highlights the differences between two distinct domains of meaning. We think that many of the other cases where a semantic module or parameter could be invoked fall into this category as well. In other words, we think there is something distinctive about the kinds of meaning which can in principle be “filled in” within C-I. There is some evidence that elements of meaning that are relevant to establishing cross-sentential relations and general discourse coherence are handled by a different component of the brain than the narrow syntactic computation. For example, Caplan (1992) claims, citing among others studies by Stachowiak et al. (1977), Brookshire & Nicholas (1984), that aphasic subjects with otherwise impaired language abilities show no significant impairment on tasks involving discourse coherence. Caplan notes it as both a surprising and robust finding in this area of psycholinguistics. On the other hand, lexical items that invoke procedural meanings have always been a point of contention for linguists working at the semantics-pragmatics interface. It is well known that languages carry information concerning cross-sentential relations. Certain quantifiers in English carry presuppositional information about the nature of the reference set (e.g., both, neither) (cf. Heim & Kratzer 1998), and others carry presuppositions about the inferential relationship between the current proposition and the previous one (but, however). In the case of English but, Grice (1975) categorized this type of semantic effect as a “conventional implicature” – conventional because it was part of the conventional linguistic meaning of the item, implicature because it contributed to “procedural” or pragmatic inferences (as opposed to purely truth conditional in his terms). Later work has tended to obscure the difference between sentence-internal semantic organisational facts and those relevant to cross-sentential and referential properties. Discourse Representation Theory for example, collects information from different sources to form complete representations which are only then subject to truth conditions (Kamp 1979; Kamp & Reyle 1993). From the point of view of the discourse representations constructed, the contribution made by the syn/sem system can include both logical information and constraints on real
Mapping a Parochial Lexicon onto a universal semantics
world reference. What we are proposing here is that discourse information is in principle different from the syn/sem system in being negotiable, but that information relevant to it can be provided by individual lexical items. This gives rise to an important potential locus of variation. 2.1 Contextual variables It is relevant here to say some words on the relationship between our proposal and the recent debates concerning “minimal propositions” (Bach 1994; Stanley 2000; Recanati 2002). Subscribers to the view of “minimal propositions” (Bach 1994; Recanati 2002) argue that even when linguistic utterances are free from vagueness, indexicality and figurative use they may still express something that underdetermines the actual intended meaning of the speaker. On the other hand, Stanley (2000) claims that genuinely underexplicit sentences that fall short of having truth conditions must nevertheless give rise to syntactic logical forms that do contain the necessary information. As an illustration of the disagreement, consider the sentence in (13) below which is taken from Bach (2007). (13) Everyone in my family went to the wedding.
For Bach, the italicized portion of the sentence has no correspondent at all in the logical form of the sentence uttered without it, but needs to be supplemented at some non-linguistic stage of processing. For Stanley, the implicit restriction on the universal quantifier actually does correspond to something in the logical form, a placeholder in the syntax which is the “contextual variable”. Note that even for Stanley (2000), the specific filled-in information triggered by a contextual variable is not itself explicitly represented in the sentence, so that he too is dealing with some version of a “minimal proposition” even in these cases. There is now a respectable tradition of contextual variables being employed in the syntax of LF representations to provide place-holders for semantically necessary information whose details must nevertheless be provided by context (Stanley 2000; Martí 2003). They have been proposed for a wide variety of cases including the comparison classes for adjectives and restrictions on quantifiers (von Fintel 1995; Martí 2003), where these variables are analogous to pronominal elements. The situation here closely resembles that of pronominal reference, since it concerns the referential properties of the individual variable provided by the nominal (nominal specificity). In general, we are broadly in agreement with the strategy of augmenting system representations with variables for items that might not be morphosyntactially overt. In this sense, we think the Stanley strategy is correct and the existence of null syntactic elements is often forced by the logic of the syn/ sem computation. But we deny the assumption of identical LF representations for
Gillian Ramchand & Peter Svenonius
the “same” thought that would require them in all cases. We think that the elements requiring a place-holder in the syn/sem computation belong to a restricted class. Stanley (2000) claims that the contextual variable must be invoked whenever anything has an impact on truth conditions. … all truth-conditional context dependence results from fixing the values of contextually sensitive elements in the real structure of natural language sentences (Stanley 2000: 392 (our italics))
We adopt a more conservative position here and make no a priori claims about where the line is drawn: any element that is part of the UG syn/sem computation must be present, by definition, but what that consists of is a matter for empirical investigation. It is important for us that the identical LF hypothesis is not invoked to automatically exclude certain classes of solution. Rather, in the cases that we examine here, there seems to be genuine linguistic variation with respect to how much information about the reference of linguistic variables is explicitly encoded in the logical forms of sentences. 2.2 Connection to acquisition If certain features can be provided either by the linguistic system explicitly, or filled in at C-I as a matter of parametric variation, we must address the question of how these systems would be acquired. Our model is partly motivated by developments in acquisition theory. Different kinds of knowledge mature at different ages (cf. Borer & Wexler 1987; Wexler 1999), probably at least in part reflecting the physical development of brain structure. But at the same time, certain phenomena are acquired earlier in some languages than in others, where the signals are less clear. We hypothesize that early acquisition due to heavy exposure leads to the syntacticization of certain phenomena, while late acquisition of the analogous phenomenon in other languages, due to ambiguous signals in the speech of adults, results in the phenomenon being treated in the mapping at the interface with the Conceptual-Intentional component. For example, we argued above that some notion of specificity was syntacticized in Norwegian, but left up to C-I to mediate in English; the evidence we presented for this was that a certain notion of specificity which is left up to context in English is signalled by the definite suffix in Norwegian. As it happens, there is also evidence from acquisition that the definite suffix has a different status from the definite article in English. For one thing, the definite suffix is acquired much earlier in Norwegian than the definite article is in English or German. Anderssen (2006) shows that around the age of two, a Norwegian child is already using the definite suffix in 82% of the
Mapping a Parochial Lexicon onto a universal semantics
contexts in which it is considered obligatory in the adult grammar, and not overgeneralizing it to inappropriate contexts. English two-year-olds, by comparison, are using the definite article in only 40% of obligatory contexts, and German children do not seem to be much different from the English speakers in this regard (see Anderssen 2006: ch. 7 for references and discussion). Furthermore, Englishspeaking children, at least, are known to overgeneralize use of the definite article, using it in contexts in which it is inappropriate (Schaeffer & Matthewson 2005). Anderssen (2006) argues that the fact that the definite suffix is a suffix affects the way it is acquired. Essentially, children who are learning simple vocabulary are exposed to a high number of systematically alternating forms. A typical noun in both English and Norwegian is either monosyllabic or trochaic. Unstressed function words preceding the stressed syllable can be ignored when suffixes cannot, in the identification of the vocabulary item to be learned. This leads to earlier acquisition of a definite suffix than of a free definite article, all else being equal. This means Norwegian children are specifying certain aspects of discourse reference, and are receiving constant input that allows them to check the correctness of their specifications, at an age where English children are not attending to such matters as closely. We speculate that this leads to systematic differences in the way discourse reference is handled in the grammar. Similarly, when English children are learning the correct use of the definite article, Russian children are not.
3. Case study: Tense So far, we have looked at a case that is plausible to interpret as being negotiated within the non-linguistic C-I component: it is clear that cases of reference assignment and knowledge of familiarity conditions in context are crucially involved in issues of definiteness within the nominal domain. We turn now to an apparently different case, although an equally common point of crosslinguistic variation, that of the presence or absence of tense morphology in a language. We will show that, under most recent analyses of tense interpretation, this too hinges on the definiteness or specificity of variable assignment, but this time of the tense variable. It is now acknowledged within the literature on the semantics of tense, that tense should be explicitly represented as a t variable – the referential theory of tense (Partee 1984; Kamp & Reyle 1993 inter alia). Consider the interpretation of the simple past tense in English in the following sentence. (14) I forgot to turn off the stove.
Although the morphological tense marking on the verb in English tells us that the event is located at some time previous to the utterance time, it does not say
Gillian Ramchand & Peter Svenonius
precisely when. In fact, the truth conditions of the utterance here depend crucially on there being a particular time moment (the moments prior to leaving the house) when I failed to turn the stove off, and the sentence is not true if the time variable is simply existentially bound. Heim (1994) goes even further, and assumes that the past tense itself introduces lexical presuppositions restricting the reference time, although this is vague, constraining the reference time to merely precede the utterance time and no more. Crucially the exact reference time is left up to contextual negotiation (via the variable assignment function g in the formula below). (15) ⟨pasti⟩g,c is only defined if g(i) < tC (the utterance time), in which case (from Heim 1994: 144) ⟨pasti⟩g,c = g(i).
We now turn to a particular case study of a language without explicit tense morphology, as analysed by Matthewson (2004; 2006). This is the case of Lillooet Salish. In Lillooet, a “tenseless” sentence may be interpreted as either present or past (unless otherwise noted, all data from Lillooet are taken from Matthewson 2004, 2006).3 (16) a.
táyt-kan hungry-1sg ‘I was hungry/I am hungry.’
b. k’ác-an’-lhkan dry-dir-1sg ‘I dried it/I am drying it.’ c.
sáy’sez’-lhkan play-1sg ‘I played/I am playing.’
d. t’íq-kan arrive-1sg ‘I arrived/I am arriving.’
Matthewson proposes that such sentences in Lillooet contain a phonologically null tense morpheme, tense. According to her, tense introduces a reference time, a variable which receives its value from a contextually determined assignment function (Matthewson 2006: 680). The assignment function in Lillooet differs from the interpretation of English past only in that it introduces no specification of past-ness with respect to the utterance. Instead, all we know is that the reference
. Sentences in Lillooet do get a default interpretation in “out of the blue” contexts dependent on the aktionsart of the verb, but these are defeasible. The addition of temporal adverbials can also be used to further specify the time in a “tenseless” sentence, although we will only be considering bare sentences here.
Mapping a Parochial Lexicon onto a universal semantics
time variable has been introduced and that its reference has to be fixed anywhere up to and including the present moment.4 (17) ⟨tensei⟩g,c is only defined if no part of g(i) is after tC. If defined, ⟨tensei⟩g,c = g(i).
There are two important points here. Firstly, there is evidence that the tense variable is indeed present in the syntax/semantics (i.e., it is not simply added on by some pragmatic component). The evidence for this comes from the fact that these tense variables exhibit anaphoric relationships across sentences that may not be overridden by context. (18) a.
nilh ts7a ta sku´l-a foc here det school-det ‘Here is the school.’
b. alkst lts7a kw s-Rhonda work here det nom-Rhonda ‘Rhonda works here.’ (*‘Rhonda worked here.’) (19) a.
tsícw-kan tu7 áku7 go-1sg past deic ‘I went to Amsterdam.’
Amsterdam-a Amsterdam-det
b. cw7it i qvl-a sman’c n-s-mán’c-em many det.pl bad-det tobacco 1sg.poss.-nom-smoke-mid ‘I smoked a lot of pot.’ (*‘I smoke a lot of pot.’)
The other possibility is that Lillooet Salish actually has two null tense morphemes, one of which is past and the other of which is present. This possibility, as Matthewson shows, is undermined by evidence that in some contexts a “tenseless” sentence may be used to denote an interval that overlaps with both the past and the present utterance time (i.e., which is vague with respect to the choice of whether the event precedes or overlaps the utterance time). (20) Context: Your friends Theresa, Charlie and Marie got drunk at the bar. You are looking after them because you don’t drink. Theresa threw up at 10 PM; Marie hasn’t thrown up at all. Just as Charlie is in the process of throwing up, another friend calls and asks wat’k’ ha i snek’wnuk’wa7-lhkálh-a vomit ynq det.pl friend(pl)-1pl.poss-det ‘Our friends throw up?’
. “Tenseless” sentences in Lillooet cannot be interpreted with future time reference. For the future, an overt, more modal clitic must be used.
Gillian Ramchand & Peter Svenonius
wat’k’ kw s-Theresa mu´ta7 vomit det nom-Theresa and ‘Theresa and Charlie throw up.’
s-Charlie nom-Charlie
Thus, the evidence from Salish shows that what is universal is a T head in the syntax which must be interpreted as locating a reference time variable with respect to the Utterance time. The difference between Salish and English is the amount of information that is provided by the lexical item, as opposed to the amount of information that is negotiated by the context. Since the Salish functional head T provides little in the way of explicit constraints, and doesn’t paradigmatically contrast with another lexical item in T, it can be null.
4. Case study: Perfectivity marking in Russian Another case study that we will examine in some detail is the situation of overt perfectivity marking in a language like Russian. It is traditionally claimed that while Russian has the morphosyntactic category of Aspect, English lacks it. This, then, looks like another case of parametric variation among languages, and the question about semantic versus syntactic parameters arises again. If we are to understand the nature of this variation, it is first important to establish exactly what the perfectivity distinction in Russian actually means semantically, a much vexed question. We will first show that previous proposals fail to adequately capture the full range of facts, and that this casts doubt on a simple characterisation of the difference between English and Russian in terms of the presence (Russian) versus absence (English) of an aspectual head or feature. Rather, in line with what we have argued for Russian definiteness and Salish tense, the difference has to do with how the values for a null functional head are determined: syntactically, in Russian, or pragmatically, in English. 4.1 What perfectivity is not Perfectivity in Russian can be diagnosed by four main language-specific properties: (i) perfectives cannot get a simple ongoing interpretation in the present tense; (ii) perfectives cannot be used as the complements of phasal verbs with meanings such as “begin,” “finish,” or “continue”; (iii) perfectives cannot form present participles; (iv) perfectives combine in discourse to form non-overlapping events in the narrative (see Kamp & Rohrer 1983; Schoorlemmer 1995; Filip 2000; Borik 2002; Romanova 2004 for examples and discussion). Perfectivity in many cases is correlated with the appearance of a prefix on the verb root. This is not a completely reliable diagnostic since some unprefixed roots are “perfective” in the above sense,
Mapping a Parochial Lexicon onto a universal semantics
and some prefixed roots are “imperfective”. This latter possibility is found quite systematically when the imperfective suffix attaches to a prefixed root that would otherwise be perfective. But although there is no single sufficient morphological indicator, there is good evidence that the distinction among stems is linguistically real, and that real morphological devices can combine productively to create members of the two different categories. In English, on the other hand, while there are some basic aktionsartal categories, there are no pairs of morphologically related stems which vary according to the perfectivity parameter – on the contrary, many verbal forms in the past exhibit an ambiguity with respect to interpretation which forces the Russian speaker to make a choice in translation between perfective and imperfective. To this extent, we may say that perfectivity (whatever it is) is a semantically relevant category in Russian, but is not in English. To zero in on exactly what head in the functional sequence of the clause might be relevant here, we need to ask whether perfectivity correlates with “inner aspect” or accomplishment/achievement readings. The answer to this question in the literature has been a disappointing “no”. Basically, the standard tests for telicity (compatibility with “in an hour” vs. “for an hour”) do not work for all perfectives (cf. Borik 2002). For example, the test seems to work for the perfective verbs pereletet “cross” and izlit “outpour” below (examples (21) and (22)). (21) Samolët pere-letel granicu *(za) cˇas. plane across-flew border in hour ‘The plane flew across the border *for an hour/in an hour.’ (22) Ona iz-lila mne duˇsu *(za) cˇas. she out.of-poured me soul in hour ‘She poured out her soul to me *for an hour/in an hour.’
(Svenonius 2004)
However, the very same test fails for a certain class of perfective verbs prefixed with po and pro. Thus, in the case of examples (23) and (24), “in an hour” is bad while “for an hour” is good. (23) Petja po-iskal knigu polˇcasa/*za polˇcasa. Peter po-looked book half.hour/in half.hour ‘Peter looked for a book for half an hour.’ (24) Petja pro-sidel v tjur’me pjat’ let. Peter pro-sat in prison five years ‘Peter was in prison for five years.’
(Borik 2002)
On the other hand, it is also not possible to correlate perfectivity with a “higher” aspectual head encoding boundedness at the sentential level, because one can find sentences which are temporally bounded, without requiring a perfective stem (cf. Borik 2002).
Gillian Ramchand & Peter Svenonius
(25) Petja uˇze peresekal `etot kanal za polˇcasa/*polˇcasa. Peter already crossed.imp this channel in half.hour/half.hour ‘Peter has already crossed this channel in half an hour.’ (Borik 2002)
One other strategy for analysing the semantic difference between the two types of stems is to characterize the perfectivity contrast as involving different “ways of viewing the event”. Classically, the perfective forms are said to view the event from the “outside” as a “completed whole,” while imperfective forms reflect a more “internal” perspective on the event (Isaˇcenko 1960; Comrie 1976), or as emphasising the endpoint of an event (perfective) vs. its visible or “ongoing” portion (Smith 1991). This classical view has also been formalized in more recent work, building on original insights by Reichenbach (1947), with an aspectual head Asp being responsible for ordering the reference time interval with respect to the event time interval in various ways (Klein 1994; Borik 2002; Demirdache & Uribe-Etxebarria 2000). Basically, for Russian, perfective events will be constrained to be included by the reference time (or at least their endpoint will be), while imperfective events surround the reference time. However, this does not strictly work either (see Romanova 2006). As Romanova points out, there are many cases where the imperfective gets a natural reading for which no internal structure is perceivable or relevant (example (26)). (26) General Factual/Present Perfect (PP) Ja ne pojdu v kafe. Ja uˇze ela I not go.perf.1sg in cafe. I already eat.imp.past.fsg ‘I am not going to a cafe. I already ate.’
Moreover there are certain striking pairs of sentences where the imperfective seems to get the “over and done with” reading (the point-like reading), whereas the perfective implies the continuation of the result state. (27) Annulled Result Kto zalezal na cˇerdak Who climb.imp.past.msg on attic ‘Who climbed to the attic? (assumption is that they are not there any more)’ (28) Result Reading Kto zalez na cˇerdak Who climb.perf.past.msg on attic ‘Who climbed to the attic? (assumption is that they are still there)’
This is mysterious under the traditional view, and all of its modern implementations in terms of interval inclusion. 4.2 An Alternative We adopt instead the analysis developed in Ramchand (2004), which connects closely to our analysis of tense and definiteness in the previous section. In Ramchand (2004),
Mapping a Parochial Lexicon onto a universal semantics
it is proposed that perfectivity is actually the aspectual analogue of definiteness definiteness with respect to a reference time variable. The alternative analysis is like many of the recent formal ones in that it builds on the Reichenbachian intuition of the distinction between event time, reference time and utterance time. We continue to assume that there are tense variables in the semantics and that there are two relations at stake, that between an event and a “reference time” (or assertion time) and that between reference time and an utterance time (Reichenbach 1947; Klein 1994; Giorgi & Pianesi 1997, Demirdache & Uribe-Etxebarria 2000). (Basic schema repeated from Giorgi & Pianesi 1997.) (29)
Relation 1: S_R future R_S past (S,R) present
Relation 2: E_R perfect R_E prospective (E,R) neutral
We further assume that the locus of Relation 2 in the above table is an aspectual head, Asp, while the locus of Relation 1 is the tense head, T. So far we are in broad agreement with Klein (1994) and Demirdache & Uribe-Etxebarria (2000). Now instead of assuming that the Asp head relates a reference time interval to an event time interval, what if we assume that the function of the Asp head is to establish/ introduce a time variable (which now represents a point individual) related to the event in a particular way. In other words, every event must have a run-time, or a temporal trace function t(e) (in the terms of Krifka 1989). The job of the Asp head is to specify a temporal individual within that run time, which will then be related to the utterance time via the T head. Thus the Asp head is crucial for relating the verbal eventive domain to the temporal domain: it provides the temporal trace of the event and establishes a temporal individual variable that can be connected to the utterance time. The general compositional schema is shown for the relevant portion of the tree in (30) below. (30)
TP TP (�TP� = λP t �P(t) & t < t*�)
(�TP� = t ��AspP�(t) & t < t*�) AspP
(�AspP� = λt e ��vP�(e) & t ∈τ (e)�)
Asp vP (�vP� = λe�. . . e . . .�) (�Asp� = λPλt e�P(e) & t ∈τ (e)�)
Ramchand (2004) argues that this system can be used to model the “perfective”/“imperfective” contrast, via a system more analogous to the definite/ indefinite contrast than an independently aspectual one. Specifically, the claim is that perfectivity (more particularly, the perfectivity diagnostics) is sensitive to the
Gillian Ramchand & Peter Svenonius
existence of a definite event time given by AspP, as opposed to an indefinite event time given by AspP.55 Although the details of this analysis are beyond the scope of this article, a number of different aspectual heads in Russian can be specified under this analysis, some definite and some indefinite. The null head in Russian, for example (and also probably the -ivaj- imperfective suffix), simply chooses any arbitrary moment within the time trace of the event as its variable. This essentially reduces to an “indefinite” assertion of some time moment once the t becomes existentially quantified ((31)). (31) ⟨Asp⟩ = λPλt∃e[P(e) & t ∈ τ(e)]
On the other hand, certain quantificational prefixes in Russian such as inceptive za-, or delimitative po-, provide more explicit lexical content to the aspectual head, essentially choosing or introducing a specific time moment within the temporal trace of the event ((32) and (33)). (32) ⟨za⟩ = λPλt[P(t) & t occurs at the onset of the temporal trace]
In the case of po, a temporal bound is asserted to exist after a short run time for the event. (33) ⟨po⟩ = λPλt[P(t) & t is a specific moment a short way in to the temporal trace]
There are other details here relevant to establishing the different ways that different prefixes can affect the choice of assertion time, and ensuring that the system works compositionally once prefixes and suffixes are combined, but this would take us too far afield here (the reader is referred to Ramchand 2004 for details). The point of this discussion is to motivate an alternative analysis of the perfective-imperfective contrast available to us which sees the distinction as a contrast in specificity, once the reference time variable is given an explicit representation in the semantics. The claim is that all the forms that give rise to a definite assertion time in Russian behave as “perfective” from the point of view of the Russian internal diagnostics, while the others pattern as “imperfective”. Notice that this also gives us a handle on two further facts about imperfectivity when it arises in languages that previous accounts do not attempt to explain: firstly that in many languages which make a morphological distinction between perfective and imperfective forms, imperfectivity is the unmarked case (Kiparsky 1998); the second is that perfectives have a distinctive behaviour in discourse narrative structures, where chained perfective events
. There are obvious comparisons to be made here with the system advanced in Stowell (1996): Stowell sees his ZeitP as an analogue of the DP within the nominal domain.
Mapping a Parochial Lexicon onto a universal semantics
are ordered rigidly within narrative structure, while imperfective ones classically produce overlapping events (Kamp & Rohrer 1983). This latter point is understandable if a specific anchoring moment is chosen for perfective events which is then used to construct a narrative time line; it also highlights the fact that representing events as perfective has inter-sentential effects on the discourse structure, a fact that we have seen is a hallmark of definiteness in the nominal domain. With this analysis in hand, we can make a proposal about the difference between English and Russian in the verbal domain which is the mirror image of the difference between the two languages in the nominal domain. Consider the sentence pairs from Russian and English respectively, below. (34) a.
Obez’jana jela jabloko (*za) pjatj minut. monkey ate.imp apple in five minutes ‘The monkey ate an apple for five minutes/*in five minutes.’
b. Obez’jana s’jela jabloko *(za) pjatj minut. monkey ate.perf apple in five minutes. ‘The monkey ate the apple in five minutes/*for five minutes.’ (35) a. John ate the apple in an hour. b. The ant ate the apple for a couple of minutes (before returning to the banana).
In Russian, there is no marking of nominal definiteness, and the discourse status of the object referent in (34) is contextually decided (see §4.3 on the connection between perfectivity and the definiteness of the object). The reference time in (34b) is definite in Russian, because a perfective stem has been chosen, and so the expression is interpreted as temporally bounded. Conversely, English explicitly marks definiteness on the object of the verb, but leaves the definiteness of the reference time up to contextual negotiation. This is seen by the fact that the verb “eat,” unlike the Russian equivalent, is possible with different continuations relating to temporal boundedness. The upshot of this is that English can be analysed as having an Asp head in its logical representation, just as Russian really has a D head. The parametric differences reduce to the lexical items that exist in the language, and how much explicit presuppositional content they contain. 4.3 Slavic vs. Germanic: Identical LF representations? We contrast the above analysis with a competing description of the difference between English and Russian (or more generally, Germanic and Slavic) where a different architectural decision is made as to the locus of variation. Namely, the view under which all LFs are at some abstract level exactly the same, but with differences in morphological implementation.
Gillian Ramchand & Peter Svenonius
Following a general strategy established by crosslinguistic work on quantification (Krifka 1987; Bach et al. 1995), Di Sciullo & Slabakova (2005) claim that Slavic and Germanic construct the same abstract representations at LF, differing only in the positional morphological manifestation of a shared set of features related to quantification (on D in Germanic; on V in Slavic). In this way they propose to capture the telicizing effect of quantized DPs (Verkuyl 1972, 1989; Tenny 1987) in Germanic, and conversely the specific or quantizing effect of Slavic perfectivizing prefixes on the interpretation of the Slavic DPs. As attractive as this proposal might seem, however, there is good evidence that tying these two properties together with a shared feature is empirically wrong. The claim for Russian is that marking perfectivity is just Russian’s way of indicating the quantizedness feature, which then affects the interpretation of direct objects by making them quantized. However, as Romanova & Diakonova (2003) have shown, the effects are not so predictable. Specifically, if we diagnose a specific interpretation of the direct object by an ability to take wide scope and to scramble, we see that for one class of perfective verbs the prediction holds up (see (36a) and (36b), from Romanova & Diakonova 2003). (36) a.
Kaˇzdyj student vy-brosil stat’ju. every student out-threw.perf article Can mean: ‘There is a specific article that every student threw out.’
b. Kaˇzdyj immigrant pere-letel granicu. every immigrant across-flew.perf border Can mean: ‘There is a particular border that every immigrant crossed.’
In addition, extraction from within such objects is degraded, if not completely ungrammatical ((37a) and (37b)) as would be expected of objects with a specific or definite interpretation (see Romanova & Diakonova 2003 for details). (37) a. ??O ˇcem John vy-brosil stat’i? about what.loc John out-threw.perf articles About what did John throw out articles?’ b. *S kakoj stranoj John pere-letel granicu? with what.inst country.inst John across-flew.perf border *‘With what country did John cross a border?’
However, these very same tests fail for other classes of perfective verb. Taking the case of po prefixed verbs once again, the following sentence can give rise to either a definite or an indefinite reading of the object (as above, the following data is taken from Romanova & Diakonova 2003). (38) Kaˇzdyj student po-ˇcital stat’i. every student po-read articles ‘Every student read (the) articles for a while.’
Mapping a Parochial Lexicon onto a universal semantics
In addition, if the scope possibilities of this object are assessed with respect to the universal quantifier “every,” both the (a) and (b) scope interpretations are possible. (39) a.
∃y [y are articles] [∀x [x a student] [x has read y]]
b. ∀x [x a student] [∃y [y are articles] [x has read y]]
Consistent with the indefinite/non-specific reading, extraction is possible from within the object of a superlexical po- prefixed verb (see Svenonius 2004; Romanova 2004 and references there on the distinction between lexical and superlexical prefixes). (40) O ˇcem deti po-ˇcitali skazki? about what children po-read tales ‘What did children read tales about?’
So, it does not seem as if perfectivity marking on a V in Russian uniformly gives rise to a reliable correlate in the nominal domain (whether that be specificity, or quantizedness).6 In fact, in a recent experiment, Slabakova herself reports that even though native speakers systematically judge perfective sentences as not ongoing in the discourse, their choice of specific or non-specific interpretations for mass objects is about 50–50 and does not significantly differ between the perfective and imperfective paradigms. The flip-side of the claim, this time for English, also turns out to be false. In a recent paper, Smollett (2005) shows consistent failure of delimitation in English in the presence of quantized objects, even for creation/consumption verbs. Smollett (2005) argues that whether there is a definite endpoint or not is negotiated contextually. We have seen examples of this type already for English in (35) above, repeated here as (41). (41) a. John ate the apple in an hour. b. The ant ate the apple for a couple of minutes (before returning to the banana).
So once again, no property encoded by English DP objects systematically carries over to an aspectual property of the VP. As discussed in the early part of this paper, assuming identical LFs is one attractive strategy for removing the need for a semantic component or a set of semantic parameters. However, the analysis it leads to here seems to be inadequate empirically, and moreover it remains unclear exactly how the parameter in question is to be stated. Under our view, parametric differences can be located within individual
. We are not denying that some such effects are found, but we think these are entailments based on specific properties of the stems that are involved. Crucially, we do not think that a single feature at LF regulates nominal interpretation and aspectual interpretation at the same time, i.e., that it is the same feature, just sitting on two different heads in the different languages.
Gillian Ramchand & Peter Svenonius
lexical items, with no syntactic featural differences and no difference in the functional sequence given by UG countenanced.
5. Conclusion: The locus of semantic crosslinguistic variation Thus, in all the core cases we have examined where serious crosslinguistic variation has been proposed in the inventory and/or interpretation of functional heads, we have found a recurring pattern. In several important cases we find evidence for syn/sem structure where morphology is lacking (e.g., T in Chinese or Salish, D in Russian or Chinese). One plausible assumption about the functional sequence is that it is necessary to construct representations that are well-formed in terms set by the C-I systems. For example, the D head is necessary for the creation of type e elements which can function as the (unsaturated) arguments of predicates. Tense in turn is necessary if the verbal extended domain is to be linked to the speech time in order to make a coherent claim about the world and create a proposition. If we are right about the aspect head, then it too is necessary to create the appropriate connection between the event domain and the temporal domain. In each of these cases, the languages differ in how much is explicitly encoded about the reference of these individual variables by their lexical items, and how much is filled in through context. Since one of the important tasks of the C-I module is to establish particular reference for individual variables, putting together information provided to it by both context and linguistic representation, we think that this is already an unavoidable locus of lexical and crosslinguistic variation. Given that this amount of variation must exist, a strong minimalist hypothesis is that this is in fact the limit of true semantic variation across languages.
Acknowledgements Thanks to David Adger and Eugenia Romanova for discussion. Thanks also to the participants at the Parameters Workshop in Newcastle in September 2005, and to the anonymous reviewers and the editor for their work.
References Anderssen, M. 2006. The Acquisition of Compositional Definiteness in Norwegian. Ph.D. dissertation, University of Tromsø. Babyonyshev, M. 1998. The possessive construction in Russian: A crosslinguistic perspective. Journal of Slavic Linguistics 5(2): 193–234.
Mapping a Parochial Lexicon onto a universal semantics
Bach, E., Jelinek, E., Kratzer, A. & Partee, B.H. (Eds). 1995. Quantification in Natural Languages [Studies in Linguistics and Philosophy 54]. Dordrecht: Kluwer. Bach, K. 1994. Conversational impliciture. Mind and Language 9: 124–162. Bach, K. 2007. Regressions in pragmatics (and semantics). In Advances in Pragmatics, N. BurtonRoberts, (Ed.), London: Palgrave. Borer, H. 2005a. In Name Only. Oxford: OUP. Borer, H. 2005b. The Normal Course of Events. Oxford: OUP. Borer, H. & Wexler, K. 1987. The maturation of syntax. In Parameter Setting, T. Roeper & E. Williams (Eds), 123–172. Dordrecht: Reidel. Borik, O. 2002. Aspect and Reference Time. Ph.D. dissertation, Universiteit Utrecht. Brookshire, R.H. & Nicholas, L.E. 1984. Comprehension of directly and indirectly stated main ideas and details in discourse by brain-damaged and non-brain-damaged listeners. Brain and Language 21: 21–36. Caplan, D. 1992. Language: Structure, Processing, and Disorders. Cambridge MA: The MIT Press. Cheng, L. & Sybesma, R. 1999. Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry 30: 509–542. Chierchia, G. 1998a. Plurality of mass nouns and the notion of ‘semantic parameter’. In Events in Grammar, S. Rothstein (Ed.), 53–105. Dordrecht: Kluwer. Chierchia, G. 1998b. Reference to kinds across languages. Natural Language Semantics 6: 339–405. Chierchia, G. 2004. Scalar implicatures, polarity phenomena and the syntax/pragmatics interface. In Structures and Beyond, A. Belletti (Ed.), 39–103. Oxford: OUP. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 2004. Beyond explanatory adequacy. In Structures and Beyond, Vol. 3, A. Belletti (Ed.), 104–131. Oxford: OUP. Comrie, B. 1976. Aspect: An introduction to the study of verbal aspect and related problems. Cambridge: CUP. Demirdache, H. & Uribe-Etxebarria, M. 2000. The primitives of temporal relations. In Step by Step. Essays on Minimalist Syntax in Honour of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 157–186. Cambridge MA: The MIT Press. Di Sciullo, A.-M. & Slabakova, R. 2005. Quantification and aspect. In Perspectives on Aspect, A. van Hout, H. de Swart & H. Verkuyl (Eds), 61–80. Dordrecht: Kluwer. Englehardt, M. & Trugman, H. 1998. D as a source of adnominal genitive in Russian. In Proceedings of the Sixth Annual Workshop on Formal Approaches to Slavic Linguistics, Ž. Bosković, S. Franks & W. Snyder (Eds), 114–133. Ann Arbor MI: Michigan Slavic Publications. Filip, H. 2000. The quantization puzzle. In Events as Grammatical Objects, C. Tenny & J. Pustejovsky (Eds), 39–96. Stanford CA: CSLI. von Fintel, K. 1995. Restrictions on Quantifier Domains. Ph.D. dissertation, University of Massachusetts. Giorgi, A. & Pianesi, F. 1997. Tense and Aspect: From Semantics to Morphosyntax . Oxford: OUP. Grice, H.P. 1975. Logic and conversation. In Syntax and Semantics, Vol. 3. P. Cole & J. Morgan (Eds), 41–58. New York NY: Academic Press. Halmøy, M. 2005. Bare nouns, number, and (in)definiteness. Paper presented at SCL 21 in Trondheim, June 2005. Heim, I. 1994. Comments on Abusch’s theory of tense. In Ellipsis, Tense, and Questions, H. Kamp (Ed.), 143–170. Amsterdam: University of Amsterdam. Heim, I. & Kratzer, A. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Isačenko, A. 1960. Grammatičeskij stroj russkogo jazyka. Morfologija. Častj vtoraja. Bratislava:
Gillian Ramchand & Peter Svenonius Vydavatelstvo Slovenskej Akadémie vied. Julien, M. 2005. Nominal Phrases from a Scandinavian Perspective. Amsterdam: John Benjamins. Kamp, H. 1979. Some remarks on the logic of change: Part I. In Time, Tense and Quantifiers, C. Rohrer (Ed.), 135–179. Tübingen: Niemeyer. Kamp, H. & Reyle, U. 1993. From Discourse to Logic. Dordrecht: Reidel. Kamp, H. & Rohrer, C. 1983. Tense in texts. In Meaning, Use and Interpretation of Languages, R. Bauerle, M. Schwarze & A. von Stechow (Eds), 250–269. Berlin: Walter de Gruyter. Kiparsky, P. 1998. Partitive case and aspect. In The Projection of Arguments: Lexical and Compositional Factors, W. Geuder & M. Butt (Eds), 265–307. Stanford CA: CSLI. Klein, W. 1994. Time in Language. London: Routledge. Krifka, M. 1987. Nominal reference and temporal constitution: Towards a semantics of quantity. In Proceedings of the 6th Amsterdam Colloquium, J. Groenendijk, M. Stokhof & F. Veltman (Eds), 153–173. Amsterdam: Institute of Linguistic, Logic and Information. Krifka, M. 1989. Nominal reference, temporal constitution and quantification in event semantics. In Semantics and Contextual Expressions, R. Bartsch, J. van Benthem & P. van Emde-Boas (Eds), 75–115. Dordrecht: Foris. Levinson, S.C. 2003. Space in Language and Cognition: Explorations in Cognitive Diversity. Cambridge: CUP. Longobardi, G. 1994. Reference and proper names: A theory of N-movement in syntax and logical form. Linguistic Inquiry 25(4): 609–65. Martí, L. 2003. Contextual Variables. Ph.D. dissertation, University of Connecticut, Storrs. Matthewson, L. 2004. What tense systems can be like: Evidence from a supposedly tenseless language. MS, University of British Columbia. Matthewson, L. 2006. Temporal semantics in a superficially tenseless language. Linguistics and Philosophy 29: 673–713. Partee, B.H. 1984. Nominal and temporal anaphora. Linguistics and Philosophy 7: 243–286. Ramchand, G. 2004. Time and the event: The semantics of Russian prefixes. In Nordlyd, Tromsø Working Papers on Language and Linguistics 32(2): Special issue on Slavic prefixes, P. Svenonius (Ed.), 323–361. Tromsø: University of Tromsø. (Available at: www. ub.uit.no/munin/nordlyd/) Rappaport, G. 2001. Extraction from nominal phrases in Polish and the theory of determiners. Journal of Slavic Linguistics 8(3): 159–198. Rappaport, G. To appear. The Slavic noun phrase in comparative perspective. In Comparative Slavic Morphosyntax, G. Fowler (Ed.). Bloomington IN: Slavica. Recanati, F. 2002. Unarticulated constituents. Linguistics and Philosophy 25: 299–345. Reichenbach, H. 1947. Elements of Symbolic Logic. New York NY: Macmillan. Romanova, E. 2004. Superlexical vs. lexical prefixes. In Nordlyd, Tromsø Working Papers on Language and Linguistics 32(2): Special issue on Slavic prefixes, P. Svenonius (Ed.), 255–278. Tromsø: University of Tromsø. (Available at: www.ub.uit.no/munin/nordlyd/) Romanova, E. 2006. Constructing Perfectivity in Russian. Ph.D. dissertation, University of Tromsø. Romanova, E. & Diakonova, M. 2003. Prefixes and objecthood. A poster presented at CASTL Kick-Off Conference, University of Tromsø, October 2–4, 2003. Schaeffer, J. & Matthewson, L. 2005. Grammar and pragmatics in the acquisition of article systems. Natural Language and Linguistic Theory 23: 53–101. Schoorlemmer, M. 1995. Participial Passive and Aspect in Russian. Ph.D. dissertation, University
Aspect matters in the middle* Marika Lekakou
Meertens Instituut, Amsterdam This paper addresses the variation that the middle construction attests cross-linguistically. In one class of languges middles behave as passives, whereas in another class they pattern with unergative structures. My proposal is that this variation is not accidental, and that it reduces to variation in the morphosyntax of aspect. I put forward a semantic treatment of middles as disposition ascriptions to a Patient/Theme argument. I then show how the morphosyntax of the dispositional generic operator that is argued to be present in such structures determines the syntactic behaviour of middles. Genericity may be morphosyntactically encoded by means of grammatical aspect, in particular in languages where such aspectual distinctions as perfective/imperfective exist. The proposal is that the level at which the generic operator is present correlates with the level at which middles are derived. For instance, in Greek and French, genericity is morphosyntactically encoded and middles are parasitic on passives. In such languages, middles are derived in the syntax, in virtue of the availability in the syntax of the relevant operator. In Germanic languages, by contrast, aspectual distinctions are not encoded in the morphosyntax, and middles are syntactically unergative. This is implemented in terms of a presyntactic derivation. Keywords: middle construction; grammatical aspect; genericity; unaccusativity; disposition ascriptions; presyntactic component
1. Introduction The topic of this paper is the variation attested in the so-called personal middle construction. I will be concerned with middles in the following languages: English, Dutch, German, Greek and French. The relevant data appear in (1).
* This paper builds on parts of my doctoral dissertation (Lekakou 2005) which was carried
out at University College London. In addition to the people acknowledged in the dissertation, I would like to thank two anonymous reviewers for their comments on the present paper. All errors are mine.
Marika Lekakou
(1) a.
This book reads easily.
b. Dit boek leest makkelijk. this book reads easily c.
Dieses Buch liest sich leicht. this book reads refl easily
d. Afto to vivlio diavazete efkola. this the book-nom read-3sg.nonact easily e.
Ce livre se lit facilement. this book refl reads easily
The aim of the paper is twofold: I wish to argue that the crosslinguistic variation in the syntax of middle constructions is parametric, and that the parameter associated with it relates to the aspectual system of different languages. I propose to treat the middle as the ascription to the understood object of a particular type of generic property, which I deem dispositional. The dispositional semantics of middles imposes certain restrictions on their syntax, but at the same time permits a certain kind of freedom which, as we will see, is exploited across languages: middles in one class of languages behave as unaccusative verbs, whereas in a second class of languages they pattern with unergatives. The factor that determines the choice of structure to be employed as a vehicle for the middle interpretation is the means available in a given language for the expression of genericity: languages employ an unaccusative structure for the middle interpretation only if they encode in their aspectual system the distinction generic/non-generic. The paper is organized as follows. In section 2, I establish the existence of two kinds of middle across languages. In section 3, I provide my proposal for the middle semantics, which capitalizes on the genericity of the construction. In section 4, I relate the variation in middles to the way in which genericity is expressed across languages. Section 5 summarizes. 2. Variation in middles As a starting point, we identify middles as generic statements about the understood object. The list of properties in (2), adapted from Ackema & Schoorlemmer (2005), summarizes the core features of middles:
(2) a. The syntactic subject is the argument that would normally be realized internally (the understood/notional object).
b. An otherwise eventive verb becomes a derived stative, and, more precisely, receives a generic interpretation. c.
The Agent is syntactically suppressed and interpreted as arbitrary.
Aspect matters in the middle
Assuming a rigid lexicon–syntax mapping in the spirit of the UTAH (Baker 1988), or a configurational theta-theory `a la Hale & Keyser (1993), we expect middles across languages to behave like passives, i.e., to involve a derived subject and a syntactically represented Agent, given the semantic affinity between the two.1 However, comparative research of the middle construction has revealed that this expectation is not met universally. Instead there seem to exist two different kinds of middle, syntactically speaking. The first type behaves as expected. In Greek and French, middles are syntactically indistinguishable from (formally reflexive) passives. For arguments in favour of the expected unaccusativity of the verb in Greek and French middles, see Tsimpli (1989); Dobrovie-Sorin (2002, 2005); Lekakou (2003, 2005). The fact that the implicit Agent is syntactically represented is evident from its ability to license by-phrases:2 (3) a.
Afto to pediko vivlio djavazete efxarista (akom i ki) this the childrens’ book read-3sg.pass.imperf with pleasure even and
apo megalus. by grown-ups
‘This childrens’ book can be read with pleasure (even) by grown-ups.’
b. Afti i askisi linete (akomi ki) apo anoitus. this the exercise solve-3sg.pass.imperf even and by idiots ‘This exercise can be solved (even) by idiots.’ (4) a.
La Tour Eiffel se voit de loin par tout le monde (qui veut the Tower Eiffel refl sees from afar by all the world who wants
bien la voir). well her see-infin
‘The Eiffel Tower can be seen from afar by everyone (who really wants to see it).’ b. Ces e´toffes se repassent facilement par tout le monde. these fabrics refl iron easily by all the world ‘These fabrics can be ironed easily by everyone.’ (Eric Mathieu, p.c.)
. Cf. Roberts (1987: 207): “the UTAH forces us to adopt a syntactic theory of middles”. . The middle interpretation (on which more presently) of the Greek examples becomes more salient if “even” modifies the by-phrase. For Continental French, it seems that speakers resist most reflexive passives with a by-phrase, on any interpretation, be it generic (middle) or episodic. Eric Mathieu, who provided the examples, informs me that the occurence of by-phrases with reflexive passives is banned by prescriptive grammars of French, but there are speakers who (like himself) accept such sentences. Canadian French is more permissive (cf. Authier & Reed (1996)).
Marika Lekakou
However, Germanic middles are of a different syntactic nature. In a series of articles, Ackema & Schoorlemmer (Ackema & Schoorlemmer 1994, 1995, 2005) argue extensively that middles in English and Dutch display unergative syntax and lack a syntactically represented implicit Agent (contra Stroik (1992); Hoekstra & Roberts (1993)). The same claim (regarding the unergativity of middles) has been made for German by Cabredo-Hofherr (1997). These facts about Germanic middles cannot be accounted for within the confines of the UTAH, according to which we expect a derived subject in the middle; they suggest the existence of a presyntactic level of representation, at which the argument structure of a predicate is manipulated. I will propose such an account of Germanic middles in section 4 (see also Ackema & Schoorlemmer (1995)). The arguments against the unaccusative analysis of Germanic middles center around the failure of middles to pass the unaccusativity diagnostics. For reasons of space, I provide only a brief mention of the supporting evidence from Dutch (see Ackema & Schoorlemmer (1994) and Cabredo-Hofherr (1997) for extensive discussion which also addresses English and German). Sentences (5–7) display the behaviour of unergatives, unaccusatives and middles with respect to the principal tests for unaccusativity in this language. The first diagnostic is auxiliary selection. In Dutch, unaccusatives select zijn “be”, whereas unergatives employ hebben “have”. Movement analyses predict that middles should select zijn, in virtue of featuring a derived subject. However, middles select hebben, thus patterning with unergatives: (5) a.
De taalkundigen *zijn/hebben gedineerd. the linguists are/ have dined
b. De zwaan is/*heeft gestorven. the swan is/ has died c.
Dit vlees heeft/*is altijd gemakkeljk gesneden. this meat has/ is always easily cut
Furthermore, in Dutch, unaccusatives allow both past and present participles as prenominal modifiers, whereas unergatives only allow the present participle prenominally. Again, Dutch middles pattern with unergatives, and not unaccusatives: (6) a.
de stervende zwaan/ de gestorven zwaan the dying swan the died swan
b. de dinerende taalkundigen/ *de gedineerde taalkundigen the dining linguists/ the dined linguists c.
het makkelijk snijdende/ *gesneden vlees the easily cutting cut meat
Aspect matters in the middle
Finally, Dutch middles, like unergatives and unlike unaccusatives, do not allow adjectival passive formation: (7) a.
De kinderen lijken gegroeid. the children look grown
b. *De kinderen lijken gewerkt. the children look worked c. *Dit vlees lijkt gemakkelijk gesneden. this meat looks easily cut
Concerning the syntactic inactivity of the implicit Agent, the most crucial piece of evidence is that, contrary to the implicit Agent of passives, the middle Agent does not license by-phrases:
(8) a. *This book reads easily by anyone.
b. c.
*Dit boek leest makkelijk door iedereen. this book reads easily by anyone *Dieses Buch liest sich von irgendwem leicht. this book reads refl by anyone easily
There are two additional key features of the Germanic middle that are not shared by its Greek and French counterparts. First, adverbial modification is necessary in Germanic middles (Roberts 1987; Condoravdi 1989b; Hoekstra & Roberts 1993; Lekakou 2005, 2006), as shown in (9). In Greek and French, on the other hand, adverbless middles are perfectly acceptable (Fagan 1992; Lekakou 2005, 2006), cf. (10). Second, Germanic languages impose restrictions on what sort of predicate constitutes an eligible middle-candidate (Roberts 1987; Fagan 1992; Ackema & Schoorlemmer 1994, 2005), cf. (11). Greek and French are again more lenient (Fagan 1992; Tsimpli 2004; Lekakou 2005), as indicated by (12).
(9) a.
This book doesn’t read *(easily).
b. Dit boek leest niet *(makkelijk). this book reads not easily c.
Dieses Buch liest sich *(leicht). this book reads refl easily
(10) a.
Afta ta manitaria trogonde. these the mushrooms eat-3pl.nonact ‘These mushrooms are edible.’
b. Cette racine se mange. this root se eats ‘This root is edible.’
Marika Lekakou
(11) a. *Spies do not recognize easily. b. *Dat herkent gemakkelijk. that recognizes easily c. *Diese Krankheit erkennt sich nicht leicht. this disease recognizes refl not easily (12) a.
Ta elinika den kataktonde efkola. the Greek-nom.pl neg acquire-3pl easily.
b. Le grec ne s’acquiert pas the Greek neg refl acquires neg ‘Greek cannot be acquired easily.’
facilement. easily
To sum up, middles across languages do not behave in a syntactically uniform way: middles are unergative in one type of language and unaccusative in another. Quite independently of what the right syntactic derivation for middles is, it is clear that it is an impossible task to coherently characterize in syntactic terms the“middle construction” across languages. In other words, the middle does not exist – not as a syntactic category at least. I submit, in line with Condoravdi (1989b), that the middle is a particular interpretation that independently existing structures receive: passives in Greek and French, unergatives in English, Dutch and German. The one thing common to middles across languages, as far as the syntax is concerned, is that they are parasitic on independently existing structures.3 The central question addressed in this paper is the locus of the crosslinguistic variation, i.e., the question: what determines which structure is employed in a given language as a vehicle for the middle semantics? Is it an accident that Greek and French pattern in the same way with respect to middles? Why doesn’t Dutch have a Greek-style middle?4
. For stylistic purposes and for ease of exposition, I employ the following abbreviations: the term “middle constuction” refers to the independently available structure which encodes the middle interpretation, and “middle formation” to the mechanisms whereby a given structure comes to convey the middle interpretation. . In this paper I concentrate on the syntactic behaviour of the verb and the projectability of the implicit Agent. The remaining two properties (the adverb facts and the restrictions) are connected to the basic unergative/unaccusative bipartition, such that, for instance, it is impossible for a Greek-like language to have middles that require adverbial modification. To show this here, however, would take us too far afield. See Lekakou (2005, 2006) for an account that relates the crosslinguistic discrepancy with respect to adverbial modification to the central parametric distinction argued for here.
Aspect matters in the middle
In order to approach this question, we need to know more about the middle interpretation itself. In section 3 I summarize my proposal that the middle is a disposition ascription to the understood object, i.e., middles are a special type of generic sentence. The crosslinguistic variation relates to the means that different languages have at their disposal with respect to encoding genericity. I show how this works in section 4, where I offer the derivation of the Greek- and Englishtype middle.
3. Middle semantics The essentially unanimous consensus (see Lekakou (2005) for discussion) is that middles are generic sentences. According to Condoravdi (1989b), middles involve generic quantification over events. The semantic representation of (13a) in (13b) involves a generic operator (henceforth Gen) that binds the event variable contributed by the verb: (13) a. This book reads easily. b. Gen [e: book(x), read(e), Patient (e, x)] [easy(e)]
However, middles arguably involve something over and above generic quantification over events. We want their semantics to capture the intuition that middles ascribe a property to their syntactic subject (Fagan 1992; Ackema & Schoorlemmer 2005); this is different from generic quantification over events, which derives a meaning for (13a) roughly as “events of reading this book are generally easy”. I propose a semantics for middles which, in addition to expressing this intuition, can tie together the core properties we attributed to them in (2): the genericity of the sentence, the arbitrarily interpreted Agent and the ‘promotion’ to subject position of the internal argument. The proposal is to treat middles as dispositional generics, which effectively reduces the properties in (2) to the dispositionality component (Lekakou 2004). The profile of dispositional generics is given in (14). As we will see presently, these properties of dispositionals in turn reduce to one, namely (14a). (14) a. Disposition ascriptions employ a VP-level Gen. b. They express ‘‘in virtue of ’’ generalizations. c. They are subject-oriented.
3.1 The dispositional generic VP-operator The standard analysis of generic sentences (Krifka et al. 1995) has it that the generic operator is a null quantificational adverb roughly paraphrasable as usually or generally. As such, Gen is an operator which partitions the clause into restrictor and
Marika Lekakou
scope and which unselectively binds any free variables in its restriction. Moreover, Gen is a modal operator (i.e., it also binds a world variable), and accordingly its semantics is modeled on the semantics of must. In sum, Gen is a quasi-universal modal operator (with a special tolerance to exceptions, see Krifka et al. (1995)). The standard analysis of modal operators assumed is Kratzer (1977), according to whom modals are propositional operators which quantify over possible worlds. There are three components to the meaning of modal operators: the modal relation (necessity or possibility), the modal base (which restricts the operator by providing the set of possible worlds over which the operator quantifies) and the ordering source (which further restricts the modal by specifying an “ideal” world and by ordering the worlds in the modal base on the basis of their similarity to that ideal). My proposal about the middle semantics is inspired by the elaboration of the Kratzerian system offered by Brennan (1993). In a nutshell, I assimilate the dispositional generic operator not to the epistemic modal must, as is standard practice, but to the dispositional operator will that Brennan discusses. Brennan proposes that modal operators do not come only in the sentence-level-operator flavour. Epistemic modals are S(entence)-level, but dynamic modals, she argues, namely ability can and dispositional will, are VP-modifiers. The intuition Brennan captures with this proposal is that VP-level modals relate properties and individuals. Part of her innovation is the proposal that the modal base that restricts dynamic modals consists, not of propositions, but of properties of the subject.5 Crucially it is only properties of the syntactic subject that restrict a dynamic modal operator. This is the formal implementation of the intuition that “in uttering a root modal sentence, the speaker typically relies on information about the syntactic subject” (Brennan 1993: 66). In this sense, dynamic modals are subject-oriented. Brennan uses a battery of diagnostics that argue in favour of the distinction between S-level and VP-level operators. One such test involves in virtue of adverbials (henceforth IVO adverbials/PPs).6 IVO PPs are obligatorily subject-controlled only when combined with dynamic modals (Brennan 1993: 48–52). The following data illustrate this point. (15) They did not award him the prize in virtue of his reputation.
. In Kratzer’s system, modals are restricted by a modal base that consists of propositions. These propositions are introduced by “in view of ”. In Brennan’s sytem, this only holds of epistemic modals. . These IVO PPs correspond to the property that forms part of the modal base, i.e., the property of the syntactic subject that restricts the dynamic modal operator by providing the worlds relevant for quantification.
Aspect matters in the middle
(16) a. Joan can sing arias in virtue of her natural ability. b. In virtue of her patience, Joan will listen to anything. (17) a. ??In virtue of the rock being lightweight, Mary can lift it. b. ??Mary will agree to anything in virtue of the loose atmosphere in the office. (18) a. *In virtue of being a graduate student, Joan may be intelligent. b. *In virtue of winning a Guggenheim, Joan must be intelligent.
(15) shows that IVO-PPs need not be obligatorily subject-controlled in general. However, when combined with VP-modals, they can only be subject-controlled. (16) and (17) illustrate that. In (16) the pronoun inside the IVO-phrase can only be bound by the syntactic subject. In (17) we see the unsuccessful outcome of a nonsubject oriented IVO PP, i.e., of our attempt to restrict the modal by a property that does not reside with the syntactic subject. Finally, (18) demonstrates that subjectoriented in virtue of -PPs are incompatible with epistemic (S-level) modals. My proposal about the semantics of Gen is to apply Brennan’s proposal for dispositional will to the domain of genericity. (For the formal implementation, see Lekakou (2004, 2005).)7 Recent work on genericity (Laca 1990; Scheiner 2003; Van Geenhoven 2003) has suggested that there are several differences among the categories that the label applies to, thus opening up the possibility that there exist different kinds of generic operators. Middles also provide evidence in this direction, which we will review shortly. The relevant distinction for me is the one between habitual generics and dispositional generics. The former are descriptive generalizations, i.e., they report a pattern of recurring events. The latter express ‘in virtue of ’ generalizations, in other words they express a regularity, whose occurrence is linked to properties of the syntactic subject (see Greenberg (2002) for the distinction applied to NPgenericity). Dispositionals, but not habituals, are subject-oriented. Accordingly, habituals employ the standard S-level generic operator, whereas dispositional generics employ a VP-level Gen in the spirit of Brennan’s will. 8 To illustrate, consider the following examples. (19) John goes to school on foot.
. Cf. Bhatt’s (1999) analysis of be able to as involving a Gen whose properties are very similar to those attributed to dispositional Gen. . The proposal made here may apply to two other categories of sentences, though I will not discuss this here. The one is “canonical” disposition ascriptions, as in (i) (Fara 2001):
(i)
Sugar is disposed to dissolve when put in water.
The other case are -able adjectives (Dahl 1975; Chierchia & McConnell-Ginet 1990).
Marika Lekakou
(20) This machine crushes oranges.
(Krifka et al., 1995)
The most salient interpretation of (19) is the habitual one, i.e., the sentence generalizes over events of John going to school and reports that usually these events involve John walking. (20), on the other hand, on the dispositional interpretation is about the ‘‘ability’’ of the machine to crush oranges. We can know whether the sentence is true if we know the properties of the machine. More generally, because habituals assert the existence of a pattern of regularly recurring events, they make a commitment as to the occurrence of prior events; for (19) to be true John needs to have gone to school on foot a sizeable number of times. The truth of dispositional generics, on the other hand, depends on properties of the syntactic subject, and hence can be computed in the absence of any prior occurrences of the event denoted by the verb. (20) on the dispositional reading can be true even if no-one has ever used the machine for crushing oranges. As I mentioned previously, there is also evidence from middles in favour of the distinction between habituals and dispositionals. Nonepisodic passives and middles are not interpreted identically. The nonepisodic passive in (21b) expresses habituality and the middle in (21a) dispositionality. This non-identity explains why (22) is not a contradition. These examples will be brought up again in the last section of this paper. More on middles as disposition ascriptions follows presently.9 (21) a. Plastic cups do not recycle. b. Plastic cups are not recycled. (22) This book reads easily, but it isn’t easily read.
In virtue of involving a VP-level generic operator, dispositional generics are subject-oriented. This does not hold of habituals, which employ the S-level Gen. We can see this in the contrast between (23) and (24). (According to the relevant myth, Midas was endowed with a property which enabled him to turn whatever he touched into gold.) (23) ??Bread is disposed to turn into gold when touched by Midas. (24) Bread turns into gold when touched by Midas.
. Sentence (21a) is one of the few examples of English middles without (overt) adverbial modification. Omission of adverbial modification is discussed in Lekakou (2005, 2006), where the “structural” approach to modification in middles pursued in Roberts (1987); Pitz (1988); Hoekstra & Roberts (1993) is revived. I argue that middles in languages like English, Dutch and German, where the Agent does not project in the syntax, require adverbial modification in order for the implicit Agent to be recovered via identification with the Benefactor/Experiencer of the adverb. It follows that there cannot be adverbless middles in Germanic languages. I show that middles with no overt adverb are interpreted as involving implicit modification.
Aspect matters in the middle
In (23) we are forcing a disposition ascription on “bread”, but this does not work because there is no property inherent in bread which makes it turn into gold when touched by Midas. When we dispose of the “is disposed to” predicate (which I have added in order to make the dispositional reading more salient), the sentence is fine. This is because (24) is a habitual sentence, whose truth does not rely on properties of the subject. 3.2 Middles as disposition ascriptions Let us now return to middles. The notion of a disposition ascription applies quite neatly to middles.10 As expected, in virtue of adverbials can only be subject-controlled. The following examples are adapted from Van Oosten (1977) and Dowty (2000): (25) These clothes wash with no trouble … a. … in virtue of being machine-washable. b. *… in virtue of the technology of my new washing-machine. (26) This car drives well… a. … in virtue of its well-engineered suspension. b. *… in virtue of the smoothness of the asphalt.
Ascribing to middles the semantics of disposition ascriptions enables us to derive the intuition researchers have had concerning the “property” reading they encode (Fagan 1992; Ackema & Schoorlemmer 2005), or Van Oosten’s Responsibility Condition, given in (27): (27) Responsibility condition (Van Oosten (1977)) The subject of a middle (the logical object) must have properties such that it can be understood to be responsible for the action expressed by the predicate.
There are additional benefits to the dispositional semantics. First of all, the genericity of the sentence follows: disposition ascriptions are generic sentences. Futhermore, since dispositional Gen is a VP-level operator, it follows that indefinite subjects of middles can only receive a generic interpretation, and can never be interpreted existentially. This is a welcome result, which does not follow from any account that assumes the run-of-the mill Gen in middles and an analysis `a la Krifka et al. (1995). (28) Small cars park easily. a. ‘Small cars are such that it is easy to park them.’ b. *‘There exist small cars such that it is easy to park them.’
. Cf. the view expressed in Dowty (2000): “the only factors that determine whether a middle construction sentence [sic] is true are properties inherent in the object acted on” (Dowty 2000: 16).
Marika Lekakou
The most significant effect of the dispositional semantics is the argument structure manipulation: because disposition ascriptions privilege the subject position – this is the essence of the subject-orientedness of dispositionals –, we can derive the (syntactic or presyntactic, depending on the language) promotion of the Patient argument to syntactic subject position. This goes hand in hand with the demotion of the Agent, the otherwise most eligible argument for subject position. More specifically, there is evidence that the implicit argument of middles is interpreted as an inherently generic indefinite, namely as the lexical item one. Consider (29), in the following context: I am a lecturer of Linguistics, in discussion with colleagues. I utter (29): (29) Our MA students have a very poor background in semantics, and are for the most part non-native speakers of English. So, I’m looking for a semantics textbook that reads easily. What would you recommend?
The implicit Agent of (29) cannot be construed as anaphoric to the MA students in the previous discourse, but is instead interpreted as a generic indefinite (see also Condoravdi (1989a)).11 This is summarized in (30): (30) The Agent in middles is (interpreted as) ONE*.
4. The analysis 4.1 Proposal and typology The proposal I make is to correlate the syntactic status of the implicit Agent to the nature of the operator that licenses it. Like the lexical item one, ONE* is only licit in the context of genericity (this is what “inherently generic” refers to). I treat the relation between Gen and ONE* as one of licensing, similar to the licensing that holds between, for instance, negation and negative polarity items. That genericity can enter into such a relation is not unheard of; for instance, Chierchia (1995) treats individual-level predicates as inherently generic. The gist of my proposal is that there exists a correlation between licensor and licensee, which is captured in terms of an immediate licensing requirement imposed by ONE*, as stated in (31): (31) ONE* needs to be licensed at the level at which it is expressed.
. I speculate that this is a fact about implicit arguments of dispositional sentences in general. Note the oddness of (i):
(i)
?? Sugar is disposed to dissolve when put into water by John.
Aspect matters in the middle
The stipulation in (31) does the following for us: ONE* can be syntactic, i.e., projected in the syntax, only if its licensor, Gen, is present in the syntax. This is what happens in Greek and French. In the absence of morphosyntactic Gen, licensing of ONE* takes place presyntactically. This is the case for English, Dutch and German. What it means for morphosyntactic Gen to be available is given in (32): (32) A language encodes Gen in imperfective morphology iff in at least one tense it has two distinct verb forms for generic and nongeneric contexts, i.e., iff genericity → imperfectivity.
The typology that emerges on the basis of this proposal is stated in (33): (33) A language will employ an unaccusative structure to convey the middle interpretation iff Gen is encoded in imperfective morphology. Otherwise, an unergative structure is used.
In the next section I discuss in more detail how the proposal made here derives this typology. Before moving on to that, I will subject each of the languages we are considering to the test of (32), and show that Gen is encoded morphosyntactically only in Greek and French. In Greek, all finite verb forms are inflected for aspect. Genericity is encoded in the imperfective; perfective aspect can only give rise to episodic readings (Giannakidou & Zwarts 1999). The following examples illustrate: (34) a.
O Janis egrafe ena grama kathe mera. the-nom John write-past.imperf.3sg one letter every day ‘John used to write a letter every day.’
b. *O Janis egrapse ena grama kathe mera. the-nom John write-past.perf.3sg one letter every day ‘John wrote a letter every day.’
The same holds of French:12 (35) a.
Jean écrivit une lettre hier/ *chaque jour. Jean write-past.perf.3sg one letter yesterday/ every day ‘Jean wrote a letter yesterday/everyday.’
. Despite giving rise to what may seem to be generic readings, the Passé Composé is irrelevant for two reasons. The apparent generic readings are bounded quantificational readings, which are not genuine generics: they involve not an intensional, but an extensional operator. See Lenci and Bertinetto (2000) for an analysis of Italian perfective “habituals” and Lekakou (2005) for discussion of French. Besides, if the bi-clausal analysis of compound tenses (Kayne 1993; Julien 2001) is right, the Passé Composé actually involves bi-clausal structure and hence is not relevant for (32). See the discussion that follows.
Marika Lekakou
b. Jean écrivait une lettre chaque jour. Jean write-past.imperf.3sg one letter every day ‘John used to write one letter every day.’
On the other hand, none of the Germanic languages in our sample have a morphosyntactically expressed Gen at their disposal. As the sentences below indicate, English does not encode the distinction generic-nongeneric in the morphosyntax.13 (36) a. John drove to school (yesterday). b. John drove to school (as a teenager).
The same holds of Dutch (37) and German (38).14 (37) a.
Jan fietste gisteren naar school. Jan cycled yesterday to school ‘Yesterday, John cycled to school.’
b. Als tiener fietste Jan naar school. as teenager cycled Jan to school ‘As a teenager, John cycled to school.’ (38) a.
John ging gestern nachmittag zu Fuß zur Schule. John went yesterday afternoon on foot to.the-dat school ‘John went to school on foot yesterday afternoon.’
b. Als Jugendlicher ging John zu Fuß zur Schule as youngster went John on foot to-the.dat school ‘In his youth, John went to school on foot.’
Before moving on to the ways in which middles are derived, I would like to address a concern voiced by an anonymous reviewer having to do with which forms count as relevant for (32). My claim is that the syntax of middles is determined by .
An anonymous reviewer points out that this may not be entirely correct, in view of (i): (i)
John would drive to school as a teenager/*yesterday.
However, it is not clear that would+VP only gives rise to a generic, and not to an episodic reading. Some speakers of English allow sentences like (i) on a “was scheduled to” or an “insisted to/demanded” reading. Besides that, there is also the conditional use of would+VP, as in John would cut the grass tomorrow if you fixed the lawn mower. These sentences are not generic. On the assumption that would receives a uniform analysis, it is not a means of encoding genericity in the sense of (32). A related note of caution relates to means of expressing habituality and not dispositionality, such as used to; I have ignored the distinction in this section in order to keep the exposition simple, but clearly, these do not contribute (dispositional) Gen. . There is dialectal variation regarding the use of the Present Perfect instead of the Simple Past, but both are compatible with generic and episodic readings.
Aspect matters in the middle
whether genericity is encoded in the morphosyntax (Greek, French) or not (Germanic). An additional condition involves the point at which whatever contributes Gen enters the derivation: there is a “cut-off ” point, beyond which the insertion of a generic operator is not going to give us a middle by licensing ONE* – for one thing, Gen needs to be in the same clause, so bi-clausal constructions are not relevant. Establishing the location of this cut-off point is a matter of future empirical research. For instance, we need to know more about middles in languages like Chinese, where aspectual morphology is not an affix on the verb but is contributed by particles. Unfortunately, it is not so obvious at this stage which structure, if any, constitutes “the middle” in Chinese (as Lisa Cheng informs me in a personal communication). We may thus also be able to shed light on the question of why certain languages, for instance Norwegian, lack middles altogether. To date, there is no research that addresses this question. 4.2 Derivations: the English type In this section I discuss the derivation of the English-type in more detail. The next section is devoted to the Greek middle. The fact that the Agent in Germanic middles is not syntactically represented, in conjunction with the assumption that middles are derived from a transitive entry (which projects the Agent) compels us to regard the middle Agent as an argument represented at a presyntactic level of representation and left unexpressed during the mapping from this level to syntax proper. I will refer to this presyntactic level of representation as conceptual structure (CS). In recent years, it has been proposed that CS is not monolithic, but comprises different levels or tiers. According to Neeleman & van de Koot (2003), who I follow closely in my conception of CS, CS encodes three kinds of information, structured along three sublevels which map onto each other. There is a level of lexical semantics, a level of argument structure, and a level of theta structure (the theta grid of a predicate). Concomitantly, the notion of “argument” is relativized as to the particular level in which it appears. The level of lexical semantics represents information of lexical semantic nature; this is where the difference between, say, “swim” and “dance” is expressed. This level maps onto argument structure, which encodes little (or no) semantic information. Argument structure is the intermediate step between lexical semantics and theta structure, namely the level that represents only syntactically relevant information about a predicate.15 . Theta structure might be a misnomer, because nothing like theta-assignment takes place at this level; only the distinction between external and internal projection is actually encoded and relevant at theta-structure.
Marika Lekakou
The system is such that semantic information is minimized on the way to projection to syntax, and this is to be expected, as certain aspects of semantics are not relevant for the syntax. For an argument to map onto the syntax, it must be represented in the theta grid. Arguments that do not occupy a theta-slot will not be represented syntactically. But such arguments may be represented at levels prior to theta-role structure. We thus have the possibility of mismatches among the three different levels. Germanic middles are precisely an instance of a mismatch, and in particular they are derived at the mapping between lexical semantics and argument structure. Another case of mismatch, according to Neeleman & van de Koot (2003), one between argument structure and theta structure, is unaccusative formation. I will briefly review their proposal about unaccusatives, before turning to middles. (39) illustrates the system I have been describing with reference to transitive cut. The statement of its lexical semantics is from Hale & Keyser (1986). The notation of the argument structure and the theta grid is along the lines of Neeleman & van de Koot (2003) (although my representations gloss over certain aspects), with the underlining used to indicate the external argument, as is standard practice. (39) Conceptual Structure of transitives: cut (Hale & Keyser 1986) a. Lexical semantics: “x produces linear separation in material integrity of y by sharp edge coming into contact with latter.” b. Argument structure: [e x [s y]] c.
Theta grid: (θ, θ)
Argument structure represents information on what Neeleman & van de Koot call “embeddings” of the predicates they distinguish, namely states, events, and interrelations; (39b) roughly says that x is the Initiator of an event e which is causally linked to y being in a state s. The permitted embeddings have to respect the following two rules (which are viewed as meaning postulates): (40) a. In [e x [s /r … y …]], x affects y with the result that y obtains the property expressed by s/r.
b. In [s1 x [s2 … y …]], x experiences y as having the property expressed by s2.
In addition, it is illicit for a predicate to occur without an argument. This reflects the nonexistence in language of zero-argument predicates.16 . As pointed out by an anonymous reviewer, weather predicates do not challenge this generalization, despite taking exclusively expletive subjects, as the expletives are quasi-argumental in the sense of Bennis (1986).
Aspect matters in the middle
According to these authors, what happens in the case of unaccusatives is essentially that the Initiator of the event is not mapped onto the theta grid, and hence does not project in the syntax. Since unaccusatives semantically involve a one-place predicate, their lexical semantics makes no reference to an Initiator. However, there is a level of representation, namely argument structure, where it occupies a slot. This follows from well-formedness conditions on that level. In particular, in order to reflect the change of state interpretation of unaccusatives, argument structure employs the embedding of a state predicate under an event predicate. Representing the event layer itself makes representing the Initiator argument obligatory. I illustrate below with unaccusative break using the notation employed in Neeleman & van de Koot (2003):17 (41) Conceptual Structure of unaccusatives: break a.
Lexical semantics: “y ends up in a state of being broken”
b. Argument structure: λy ∃x [e x [s y]] c.
Theta grid: (θ)
In the terminology of Neeleman & van de Koot, the otherwise external argument undergoes suppression, which in their system is essentially the replacement of a lambda operator by an existential operator. (Note that the term “suppression” is used in a different sense than in Tanya Reinhart’s work, e.g., Reinhart (2003).) The existentially bound variable does not map onto theta structure, which only sees lambdabound variables. In fact, Neeleman & van de Koot (2003) claim that there exists a one-to-one correspondence between lambda bound variables and theta roles. What ensures that the single projecting argument of monadic unaccusatives starts out as an internal and not an external argument is the following principle: (42) Only the most prominent argument variable is linked to an external theta-role.
Because the most prominent argument variable is bound at argument structure by an existential operator, the single argument of unaccusatives is generated internally. Middle formation has to be different. Contrary to what we saw for unaccusatives, in middles the Initiator (an Agent) is semantically present, therefore it has . The lexical semantics description I give from now on will be very schematic. Since lexical semantics contains information on the lexical meaning of predicates, (41a) should explicate the meaning of “break”, in the fashion of the elaborate paraphrases that Hale & Keyser (1986) coined (see the previous example); I have not done this.
Marika Lekakou
to be represented at the level of lexical semantics. More specifically, its interpretation is that of a generic indefinite, our ONE*. ONE* has to be bound by a generic operator. What is more, it has to be licensed immediately. Since syntactic Gen is not available in English, ONE* can only be licensed presyntactically, as dictated by (30) and (31) above. Therefore, it is introduced at the level of lexical semantics and licensed immediately by dispositional Gen, introduced at that level.18 What derives the unergativity of Germanic middles is that ONE* is not mapped onto argument structure, which only represents a single lambda bound variable within a state predicate. This has the immediate effect that middle-read is treated at argument structure as a stative for the purposes of mapping to the theta grid. Stative verbs feature a basegenerated subject, and are thus unergative (see Neeleman & van de Koot (2003) for discussion of the syntactic properties of different classes of stative predicates). (43) Conceptual Structure of middles: read a.
Lexical semantics: “x is such that one reads x”
b. Argument structure: λy [s y] c.
theta grid: (θ)
That the understood external argument is present at the level of lexical semantics is a welcome feature of this analysis, given that Germanic middles involve a semantically present implicit Agent. That they have the same syntax as statives is also wellmotivated, given their interpretation (generics are derived statives). One question that needs to be addressed is what permits us to omit the event layer in the representation of the argument structure of middle-read. For this, I need to assume that “early” binding by Gen is what is relevant. More concretely, the assumption is that if Gen is introduced at the level of lexical semantics, it becomes unnecessary to represent the event predicate of the original, transitive verb. Since the mapping from one level to the next involves essentially “discharging” of semantic information that will not be relevant for the syntax, the fact that the event layer does not have to be represented in this case means that it will not be represented. Given the
. Note that the existence of a presyntactic generic operator is not unheard of: Chierchia (1995) entertains this possibility for individiual level predicates, and discards it not on the basis of reservations against having a lexical Gen, but for reasons independent of our concerns here. Besides, we have already granted life in CS to operators, by allowing the existential operator to bind variables in argument structure.
Aspect matters in the middle
minimization of semantic information on the way to theta structure, the option offered by early insertion of Gen is enforced. 4.3 Derivations: The Greek type Let us now turn to the Greek/French type of middle. ONE* in these languages projects in the syntax, because Gen is available to license it in the syntax. To derive this, we only need to assume a more or less standard account of passive formation: we can assume any analysis of passive that involves A-movement and syntactically representing the Agent (e.g., Baker et al. (1989)). (For the view that passive se realizes the external theta-role and related discussion see Roberts (1987); Kayne (1988); Pesetsky (1995); Authier & Reed (1996); Sportiche (1998); McGinnis (1999).) What enables these languages to resort to passives for the expression of the middle semantics is their aspectual system. 4.4 Further predictions There are two additional aspects of middle formation across languages that we need to explore. The first concerns the question of whether we expect an unergative-middle in Greek and French. The answer to this is negative. I assume that the existence of a syntactically encoded Gen in these languages precludes the existence of a presyntactic Gen. This amounts to a more general assumption that is made tacitly quite generally: if in a language a piece of morphology expresses a certain meaning, the language cannot express that meaning without making use of the means in question. At the risk of belaboring an obvious point, consider the expression of “pasthood”. All of the languages examined here distinguish past from non-past tenses. The assumption introduced here says that past tense meaning cannot be conveyed unless a past tense form is used (modulo the historical present, which arguably involves somewhat different semantics).19 The second question is whether a passive-middle is available in Germanic. Again the answer is negative (contra Lyons (1995)). We have already seen that generic passives do not encode the same interpretation as middles: (44) a. Plastic cups do not recycle. b. Plastic cups are not recycled.
. The assumption I am making can be thought of as a kind of Elsewhere Principle. If past tense forms are marked as encoding [+past], they are more specific with respect to other tense forms, which are not marked as [+past]. Therefore, they will always be preferred to express “pasthood”.
Marika Lekakou
(45) This book reads easily, but it isn’t easily read.
More generally, we observe that ONE* is not a possible interpretation for passive morphology in English. This is because Gen is not syntactically represented. In the examples in (46) it is easily that licenses ONE*.20 (46)
a. b. c. d.
This book is ?*(easily) read by (any)one. Bureaucrats are ?*(easily) bribed by (any)one. This bread is ?*(easily) cut by (any)one. Cotton shirts are ?*(easily) washed by (any)one.
5. Conclusion On the basis of the cross-linguistic variation in the syntax of middles, I have argued that there is no “middle construction” in a syntactic sense; rather, we should conceive of the middle as a particular interpretation that independently existing structures receive. On such an approach, the cross-linguistic variation is far from unexpected (as long as the syntax-semantics interface is not assumed to involve a rigid mapping). The locus of the variation is proposed to be a more general morphosyntactic property of languages, namely how genericity is encoded in different languages. Thus, the parameter that regulates variation in middles reduces to the parameter associated with the morphosyntax of aspect.
References Ackema, P. & Schoorlemmer, M. 1994. The middle construction and the syntax-semantics interface. Lingua 93: 59–90. Ackema, P. & Schoorlemmer, M. 1995. Middles and nonmovement. Linguistic Inquiry 26: 173–197. Ackema, P. & Schoorlemmer, M. 2005. Middles. In The Blackwell Companion to Syntax, Vol. 3, M. Everaert & H. van Riemsdijk (Eds), 131–203. Oxford: Blackwell. Authier, J.-M. & Reed, L. 1996. On the Canadian French Middle. Linguistic Inquiry 27: 513–523. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago IL: University of Chicago Press. Baker, M., Johnson, K. & Roberts, I. 1989. Passive arguments raised. Linguistic Inquiry 20: 219–251.
. As pointed out by an anonymous reviewer, modals can also license ONE*, which is to be expected. The same reviewer wonders whether the facts discussed here bear on the nonexistence of impersonal passives in English. As far as I can see, since impersonal passives in languages that have them (can) involve existential interpretations of the implicit Agent, the latter cannot be identified with ONE*, which is a generic indefinite. Therefore, the lack of impersonal passives in English needs to be explained on independent grounds.
Aspect matters in the middle
Bennis, H. 1986. Gaps and Dummies. Dordrecht: Foris. Bhatt, R. 1999. Covert Modality in Non-Finite Contexts. Ph.D. dissertation, University of Pennsylvania. Brennan, V. 1993. Root and Epistemic Modal Auxiliary Verbs. Ph.D. dissertation, UMass. Cabredo-Hofherr, P. 1997. The German Middle Construction. DEA Thesis, Université Paris 7. Chierchia, G. 1995. Individual level predicates as Inherent Generics. In The Generic Book, G. Carlson & F.J. Pelletier (Eds), 176–237. Chicago IL: Chicago University Press. Chierchia, G. & McConnell-Ginet, S. 1990. An Introduction to Semantics. Cambridge MA: The MIT Press. Condoravdi, C. 1989a. Indefinite and generic pronouns. In Proceedings of WCCFL 8, E.J. Fee & K. Hunt (Eds), 71–84. Stanford CA: CSLI. Condoravdi, C. 1989b. The middle: Where semantics and morphology meet. MIT Working Papers in Linguistics 11: 18–30. Dahl, Ö. 1975. On generics. In Formal Semantics of Natural Language, E. Keenan (Ed.), 99–111. Cambridge: CUP. Dobrovie-Sorin, C. 2002. Agentivity and passivization. Talk given at the Comparative Romance Linguistics Colloquium, Antwerp. Dobrovie-Sorin, C. 2005. The SE-anaphor and its role in argument realization. In The Blackwell Companion to Syntax, Vol. 4, M. Everaert & H. van Riemsdijk (Eds), 118–179. Oxford: Blackwell. Dowty, D. 2000. The semantic asymmetry of ‘argument alternations’ (and why it matters). Groninger Arbeiten zur germanistischen Linguistik 44. (http://www.ling.ohio-state.edu/_dowty) Fagan, S. 1992. The Syntax and Semantics of Middle Constructions. Cambridge: CUP. Fara, M. 2001. Dispositions and their Ascriptions. Ph.D. dissertation, Princeton University. Giannakidou, A. & Zwarts, F. 1999. Temporal, aspectual structure and (non)veridicality. In Tense and Mood Selection, A. Giorgi & J. Higginbotham & F. Pianesi (Eds). Oxford: OUP. Greenberg, Y. 2002. Two types of quantificational modalized genericity, and the interpretation of bare plural and indefinite singular NPs. In Proceedings of SALT 12, B. Jackson (Ed.), Ithaca NY: CLC Publications, Cornell University. Hale, K. & Keyser, S.J. 1986. Some transitivity alternations in English. MIT Working Papers in Linguistics 7. Hale, K. & Keyser, S.J. 1993. On argument structure and the lexical expression of syntactic relations. In The View from Building 20, K. Hale & S.J. Keyser (Eds), 53–109. Cambridge MA: The MIT Press. Hoekstra, T. & Roberts, I. 1993. Middle constructions in Dutch and English. In Knowledge and Language II. Lexical and Conceptual Structure, E. Reuland & W. Abraham (Eds), 183–220. Dordrecht: Kluwer. Julien, M. 2001. The syntax of complex tenses. Journal of Linguistics 18: 125–167. Kayne, R. 1988. Romance se/si. GLOW Newsletter 20. Kayne, R.S. 1993. Towards a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. Kratzer, A. 1977. What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy 1: 337–355. Krifka, M., Pelletier, F.J., Carlson, G., ter Meulen, A., Link, G. & Chierchia, G. 1995. Genericity: An introduction. In The Generic Book, G. Carlson & F.J. Pelletier (Eds), 1–124. Chicago IL: Chicago University Press. Laca, B. 1990. Generic objects. Lingua 81: 25–46. Lekakou, M. 2003. Greek passives on the middle interpretation. In Proceedings of the 6th International Conference on Greek Linguistics.
Marika Lekakou Lekakou, M. 2004. Middles as disposition ascriptions. In Proceedings of Sinn und Bedeutung 8, C. Meier & M. Weisgerber (Eds), 181–196. Konstanz: University of Konstanz. (http://www. ub.uni-konstanz.de/kops/volltexte/2004/1383/). Lekakou, M. 2005. In the Middle Somewhat Elevated. The Semantics of Middles and Its Crosslinguistic Realization. Ph.D. dissertation, University College London. Lekakou, M. 2006. A comparative view of the requirement for adverbial modification in middles. In Demoting the Agent. Passive, Middle and Other Voice-Related Phenomena, B. Lyngfelt & T. Solstad (Eds), 167–196. Amsterdam: John Benjamins. Lenci, A. & Bertinetto, P.M. 2000. Aspects, adverbs, and events: Habituality vs. perfectivity. In Speaking of Events, J. Higginbotham, F. Pianesi & A.C. Varzi (Eds), 245–287. Oxford: OUP. Lyons, C. 1995. Voice, aspect and arbitrary arguments. In Linguistic Theory and the Romance Languages, J.C. Smith & M. Maiden (Eds), 77–114. Amsterdam: John Benjamins. McGinnis, M. 1999. Reflexive clitics and the specifiers of vP. MIT Working Papers in Linguistics 35: 137–160. Neeleman, A. & van de Koot, H. 2003. Bare resultatives. Journal of Comparative Germanic Linguistics 6: 1–52. Pesetsky, D. 1995. Zero Syntax: Experiencers and Cascades. Cambridge MA: The MIT Press. Pitz, A. 1988. Middle Constructions in German. Trondheim Working Papers in Linguistics 6: 1–30. Reinhart, T. 2003. The Theta system – an overview. Theoretical Linguistics 3: 229–290. Roberts, I. 1987. The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris Publications. Scheiner, M. 2003. Temporal anchoring of habituals. In Proceedings of ConSole XI, M. van Koppen, J. Sio & M. de Vos (Eds). (http://www.sole.leidenuniv.nl). Sportiche, D. 1998. Partitions and Atoms of Clause Structure. London: Routledge. Stroik, T. 1992. Middles and Movement. Linguistic Inquiry 23: 127–137. Tsimpli, I. 2004. I foni stin eliniki. (Voice in Greek). MS. University of Thessaloniki. Tsimpli, I.M. 1989. On the properties of the passive affix in Modern Greek. UCL Working Papers in Linguistics 1: 235–260. Van Geenhoven, V. 2003. The semantic diversity of characterizing sentences. Talk given at the 14th Amsterdam Colloquium, Amsterdam, December 2003. Van Oosten, J. 1977. Subjects and agenthood in English. Chicago Linguistic Society 13: 451–471.
part ii
A classic parameter revisited The null subject parameter
The Case-F valuation parameter in Romance* Gerardo Fernández-Salgueiro In this paper I propose that the operation of Agree (Chomsky 2000, 2001) is parameterized with respect to the valuation of Case features. I build here on the ideas put forward in Fernández-Salgueiro (2005), where I argued that Agree has to be modified in certain respects in order to generate Further-Raising constructions, i.e., A-Movement after Spec-Head agreement with ϕ-complete Tense, which is possible in many Romance languages. If this is not done, Further Raising constructions incur a violation of economy constraints on movement, i.e., the principle of Last Resort. I also suggested that Case-F Valuation on subject DPs can be delayed in most Romance languages, but not in other languages like English. Here I develop the features of such a parametric approach in more detail, and comment on acquisition issues and possible consequences for the general theory of parametric variation as understood in the Minimalist Program. Keywords: agree; A-movement; Case; Further Raising; Last Resort; Parametric variation
1. Assumptions about Agree 1.1 The starting point: Chomsky’s (2000, 2001) Agree Chomsky’s (2000, 2001) formulation of Agree constitutes a departure from early Minimalism (Chomsky 1995) in terms of the relation between agreement, Case, and movement. While agreement and Case checking were, in early Minimalism, understood to occur via a Spec-Head relation after (overt or covert) movement of the DP (the so-called Checking Theory), movement in the Minimalist Inquiries/ Derivation by Phase framework is dissociated from Case checking and agreement, which can take place in situ.
*I would like to thank Sam Epstein, Anders Holmberg, Dina Kapetangianni, Hamid Ouali, Acrisio Pires, Teresa Satterfield, Daniel Seely, and Michelle Sheehan for helpful comments. I would also like to thank the audiences at 34th LSRL, WECOL 04, and the Newcastle Workshop on Parametric Variation, where I presented different parts of this work, for their suggestions. All errors are mine.
Gerardo Fernández-Salgueiro
In order to briefly illustrate the Agree approach, consider the following syntactic configuration:1
(1)
Tense vP
Tense [EPP, uφ] DP [uCase, φ]
(...)
In the tree in (1) we have a ϕ-complete Tense head with a selectional feature (EPP) and unvalued ϕ-features (ϕ-Fs), and a DP with interpretable ϕ-Fs and an unvalued Case-F. For the derivation to converge, all unvalued Fs have to be given a value, and the EPP also has to be satisfied by means of Merge. According to Chomsky, the first step is for the unvalued ϕ-Fs of Tense (also called the Probe) to match the interpretable ϕ-Fs of the DP (also called the Goal). This matching relation identifies the elements that are going to undergo Agree. After Match, the probe gets its feature specification from the interpretable ϕ-Fs of the DP, i.e. it values its ϕ-Fs, and the Case-F of the DP gets valued as well. Chomsky claims that after the DP gets its Case-Fs valued, it is “frozen in place”, and unable to undergo further A-Movement. Now that unvalued features have been valued and thus rendered legible at the interfaces, the EPP-F in Tense also needs to be satisfied, via what Chomsky calls “second-merge.” In order to satisfy this selectional feature, the DP raises to the specifier of Tense, yielding (2), where all unvalued features have been valued and all selectional features have been satisfied:
(2)
Tense DP [Case, φ]
Tense
Tense [EPP, φ] DP [Case, φ]
vP (...)
. Here I am only discussing Agree between a subject DP and ϕ-complete Tense. As for all other instances of Agree, I basically follow Chomsky (2001).
The Case-F valuation parameter in Romance
1.2 Problems with Agree Although I agree with Chomsky in that an Agree-based approach to Case checking/ valuation and agreement is conceptually superior to the Checking approach of early Minimalism, I argued in Fernández-Salgueiro (2005; henceforth: F-S) that there are a number of conceptual problems with this particular formulation of Agree with respect to EPP satisfaction (“second-merge”). Here I comment on two of those issues. First, there is a potential paradox in Chomsky’s proposal. If the EPP-F is satisfied by movement of the DP after Case-F valuation applies, how is it that the DP is not “inert” (“frozen in place”) for the purposes of such movement? Epstein and Seely (2002) also noticed this problem, and argued that Last Resort (Chomsky 1995) is consistently violated in Chomsky’s system, since the DP is moving even though it does not have to move to get its Case-F checked/valued. Chomsky’s answer to this question is that there is no order of operations inside the phase, that is, all operations apply simultaneously. If this is true, then the derivational nature of the system is compromised within each phase, since we cannot explain properties of the grammar by examining the logic of a step-by-step structure building mechanism.2 Second, under Chomsky’s approach, there is no clear way to determine which of the copies is interpreted at the interface with the Sensory-Motor Systems (SMS) after Linearization. Nunes (1999, 2004) develops a theory of Linearization of Chains based on the fact that, under the Checking theory, only the topmost copy of the subject DP undergoes Case checking.3 Thanks to this asymmetry between the topmost copy and the lower copies of the DP with respect to Case checking, Nunes derives the fact that it is the upper copy that is generally pronounced from economy considerations. As we saw in section 1.1, under the Agree approach, the DP only raises to the specifier of Tense after its Case-F has been valued, so none of the copies of the DP displays unvalued Case-Fs. Thus, under Chomsky’s approach to Agree, the explanatory force of Nunes’ approach is lost, and we are forced to stipulate that only the upper copy of the DP is visible at the SMS interface.
. I subscribe here to Epstein & Seely’s (2002) criticism of Chomsky’s notion of phase. I will be assuming Chomsky’s probe-goal system but with cyclic spell-out after each Merge operation, or rather, “inspection” of the steps of the derivation by the interfaces (as in Epstein & Seely 2006). . Nunes (1999, 2004) also does not make the assumption that all members of a Chain have to be identical, so the lower copy of a DP whose upper copy has undergone Case checking still has unchecked Case features.
Gerardo Fernández-Salgueiro
1.3 A possible solution In F-S I argued that the above problems can be solved if we allow the EPP-F to be satisfied by raising of the DP before Agree applies between Tense and the DP. This is illustrated in the tree in (3) below:
(3)
Tense DP [Case, φ]
Tense
Tense [EPP, φ] DP [uCase, φ]
vP (...)
Notice that as a consequence of this different timing of EPP-F checking and Agree, the lower copy of the DP still has an unvalued Case-F, which means that Nunes’ (1999, 2004) approach to Chain Linearization can still be formulated, since we now have again asymmetry between the two copies of the DP with respect to Case. Moreover, we can maintain Chomsky’s own observation that a Case feature is what renders a DP active, in the sense that it can undergo A-Movement only if the Case-F on the DP has not been valued. Notice also that this formulation is consistent with a derivational approach to syntax, with no simultaneous operations. Operations apply step by step, with the output of one operation serving as the input of the next operation. One consequence of this approach is that we have to allow Agree to apply between a head and its specifier; in other words, we have to extend the search domain of the Probe (which in Chomsky’s formulation is its c-command domain) to its m-command domain. This does not seem to be problematic conceptually; if the point of restricting the search domain of the Probe is to keep Agree as local an operation as possible, for a Probe to search for a Goal in its specifier indeed seems like a very local search (see also Müller 2004, who reaches a similar conclusion on different grounds).4 (4) provides a summary of the operations involved in Agree, for both Chomsky’s approach and the one I am arguing for here:
. For other recent approaches that claim that Case checking is a Spec-Head relation see Rezac (2003) and Epstein & Seely (2006).
The Case-F valuation parameter in Romance
(4) In situ
Chomsky Match (φ, uφ) Value uφ on Probe Value uCase on Goal EPP-F checking
Steps 1st 2nd 3rd 4th
F-S EPP-F checking Match (φ, uφ) Value uφ on Probe Value uCase on Goal
SpecHead
2. Further-raising and Last Resort 2.1 Further-raising in Romance In this section, I discuss the properties of further-raising constructions (in which a subject can raise to the specifier of a ϕ-complete head and then undergo further A-movement), which are possible in a range of Romance varieties, including Galician, Spanish, European Portuguese, Catalan, and Italian. Then I am going to argue that these constructions cannot be generated under Agree.5 An example of further-raising is given in (5): (5) Estes nenos parece que son moi listos. these kids seems that are very smart ‘These kids seem to be very smart’.
(Galician)
Although an example like (5) looks like a case of topicalization/left-dislocation, more careful analysis shows that this is an instance of A-Movement, despite the fact that estes nenos does not agree with parece. In F-S, I used several tests to show that estes nenos is not base-generated in a topic position and also that the movement of estes nenos is an instance of A-Movement. Here I reproduce some of those tests.6
. French, Brazilian Portuguese, and Romanian are Romance languages that do not display Further Raising. The case of French is expected since it is not a pro-drop language (see section 3.5). Brazilian Portuguese and Romanian display Hyper Raising, which shares some characteristics with Further Raising (see section 2.3). Further Raising is also possible in Basque. . As one reviewer suggests, the same arguments could be used to show that tough-movement is A-Movement, since it also allows idiom chunk raising (Headway isn’t easy to make on these questions) and scope ambiguity (Some problem is easy for everyone to solve). The question that arises then is whether a null operator analysis could be adopted for Further Raising. This does not seem straightforward, given that there are important differences between further raising and tough-movement constructions. First, Further Raising always involves movement from one subject position to another, which is not available in tough-movement, as argued in F-S. Second, the further-raised subject is not an argument of the raising verb, which means that it cannot be base-generated in the matrix clause, which is the standard analysis for tough-movement. The
Gerardo Fernández-Salgueiro
On the one hand, (6) shows that island effects are observed:7 (6) *Estes nenos parece que o ruido que fan é demasiado. these kids seems that the noise that make is too much ‘It seems that the noise that these kids make is too much’.
On the other hand, it can be shown in a number of ways that we are dealing with A-Movement. Here I illustrate two of the common A-Movement tests with Spanish examples. First, (7) and (8) show that the subject of an idiom can be moved, without altering the idiomatic meaning: (7) Mala hierba nunca muere. bad grass never dies ‘The devil looks after himself ’. (8) Mala hierba parece que nunca muere. bad grass seems that never dies ‘The devil always seems to look after himself.’
Second, movement of a quantifier does not freeze its scope. In the simple sentence in (9) below, the existential subject can take scope over a universal object (contra Alexiadou & Anagnostopoulou 1998).8 The scope possibilities remain the same even when the subject is further-raised, as in (10): (9)
Algún problema afecta siempre a toda teoría compleja. some problem affects always to every theory complex ‘Some problem always affects every complex theory’. [some > every, every > some]
(10) Algún problema parece que afecta siempre a toda teoría compleja. some problem seem that affects always to every theory complex ‘Some problem always seems to affect every complex theory’. [some > every, every > some]
parallelism between both constructions is worth exploring further, but it falls outside the scope of this paper. . Most of the examples I will be using here have an alternative reading in which the DP is base-generated in the topic position, in which case an intonation break is needed between the DP and the rest of the sentence. The grammaticality judgments discussed here correspond to readings with plain intonation. . As can be seen, (9) can be taken as evidence that preverbal subjects in Romance cannot always be analyzed as left-dislocated or topicalized (see Alexiadou & Anagnostopoulou 1998 for this kind of analysis).
The Case-F valuation parameter in Romance
In the next section, I show that further raising constructions cannot be generated under Agree in the light of the evidence for an A-Movement analysis just presented. 2.2 Case-Fs, Inert DPs, and Last Resort Given the empirical evidence that further-raising involves A-Movement, and not topicalization/left-dislocation, an obvious incompatibility with the claim that DPs are “frozen in place” after Case-F valuation arises. Consider the derivation in (11) below: : Match and Value : Movement
Tense
(11) (
)
Tense Tense
(...) (...)
*
Tense Tense
DP [Case, φ] Tense [EPP, φ]
DP [uCase, φ]
vP (...)
The last movement of the DP illustrated in (11) is not allowed under Agree, since the DP is “inert” or “frozen in place” after its Case-F gets a value in the embedded clause. This idea is actually a consequence of the principle of Last Resort (Chomsky 1995), which would not allow the DP to move if it does not have to in order to check its Case-F. Languages like English, for example, certainly behave like this, and so a sentence like (12) is ruled out on the basis that the DP John has already checked/valued its Case-F in the embedded clause. It is therefore not allowed to move to a higher specifier position:9 (12) *John seems is smart.
. In this ungrammatical English sentence, John also agrees with seems, which is not what happens in Further Raising, although it is exactly what happens in Hyper Raising (see section 2.3).
Gerardo Fernández-Salgueiro
However, this is exactly what happens in further-raising cases, where the DP is able to undergo A-Movement after Case checking/valuation has occurred. 2.3 Th e typology of (apparently) non-local raising: Super Raising, Hyper Raising and Further Raising In this section, I consider two other types of raising constructions, which should not be confused with the further raising cases I am considering in this paper. Super Raising is non-local A-Movement, where a DP skips an available A- position, as in (13) (taken from Zwart 1997): (13) *John seems that it was told (John) that he could win the race.
Ura (2000) links the availability of such raising constructions in a given language to the availability of multiple specifiers. In fact, Ura argues that languages that allow super-raising also allow multiple subjects. Hyper-raising is similar to further-raising in that a DP can raise from the specifier of a ϕ-complete Tense head with which it agrees to another specifier of a ϕ-complete Tense head, as in the Brazilian Portuguese example in (14): (14) As crianças parecem que gostam da babá. The children seem-3pl that like-3pl of-the baby-sitter ‘The children seem to like the baby-sitter.’
The only difference between Hyper Raising and Further Raising is that in Hyper Raising the DP also agrees with the matrix Tense, whereas in Further Raising such agreement is not allowed; the DP can only agree with the embedded Tense head. Interestingly, no Romance language allows both Hyper and Further Raising; as I pointed out above, Galician, Spanish, European Portuguese, Catalan, and Italian allow Further Raising, whereas Brazilian Portuguese and Romanian allow Hyper Raising (see Zwart 1997; Ferreira 2004). Exploring the relation between Further and Hyper Raising is an interesting question (especially given the fact that they are mutually exclusive), which I leave for further research. 3. (Part of) Agree is parameterized: The timing of Case-F valuation 3.1 Case-F valuation can be delayed in Romance In F-S, I suggested that a possible solution to the problem with Agree with respect to Further Raising that I discussed in section 2.2 above is that different languages may behave differently in the timing of Case-F valuation. Here I explore the details of these parametric differences and several implications of such an instance of parametric variation for the theory of principles and parameters in Minimalism.
The Case-F valuation parameter in Romance
If we keep on assuming Last Resort and we claim that the system only allows a DP to (A-)move if its Case-F has not been valued, we are forced to say that the Case-F of the DP in further-raising constructions does not obligatorily get its value specification in the specifier of Tense when the structural description for Case-F valuation is met. I would like to propose, then, that in languages that allow Further Raising, the ϕ-Fs on the functional head get valued, but Case-F valuation can be delayed.10 If this assumption is on the right track, it will explain why a subject DP can undergo a second instance of A-movement without violating Last Resort. An example of a further raising sentence is given again in (15), and its derivation is illustrated in (16) below (contrast with (11) above): (15) Algúns nenos parece que están tolos. some kids seems that are crazy ‘Some kids seem to be crazy’. (16)
: Match : Movement
Tense DP [Case, φ]
Tense
Tense
(...) (...)
√
(Galician)
Tense DP [uCase, φ]
Tense
Tense [EPP, φ] DP [uCase, φ]
vP (...)
Under the proposed analysis, the DP raises from its thematic position to the specifier of (the embedded) Tense, in order to satisfy the EPP-F in Tense. After this first application of “second-merge”, the ϕ-Fs of the embedded Tense match the ϕ-Fs of the DP. Then the ϕ-Fs of Tense get valued but the valuation of the Case-F of the DP is delayed. The derivation continues and the matrix Tense containing another . Interestingly, Ura (2000) proposes that hyper raising languages optionally undergo Case-F checking in the embedded clause, which is similar to what I am proposing here for Further Raising.
Gerardo Fernández-Salgueiro
EPP-F is inserted in the derivation. Then the dp can raise to satisfy this EPP-F without violating Last Resort, since its Case-F has not yet been valued. After the dp satisfies this higher EPP-F, its Case-F gets valued by virtue of the first Agree operation, one sub-part of which was delayed.11,12 A question that arises under this analysis is how the ϕ-Fs of the matrix Tense head get valued, since the further raised DP cannot agree with it. Given that in these constructions the matrix Tense always displays 3rd person singular agreement, one possibility is that this value is the result of an Agree relation with the embedded CP, which is presumably what happens in a case with no Further Raising, like parece que algúns nenos están locos. As can be seen, the parameterization that I am proposing concerns only one sub-operation within Agree, namely, optional delay of Case-F valuation. This is summarized in the table in (17): (17)
Non-further-raising lgs.
Further-raising lgs.
First
Satisfy EPP-F: Second-Merge
Second
Match ϕ-Fs of Probe and Goal
Third
Value ϕ-Fs of Probe
Fourth
Value Case-F
Value Case-F
Delay Case-F valuation
3.2 Micro- vs. macrovariation At this point, it is worth commenting on the status of this proposal with respect to certain assumptions about the nature of parameterization in the Minimalist Program. An important question that arises related to the parameter I am proposing
. Sam Epstein (p.c.) points out to me that this delay in Case-F valuation involves either global look-back or a violation of the Inclusiveness Condition. For the DP to get its Case-F valued after it satisfies the EPP-F in the matrix clause, we either have to allow the system to inspect previous steps of the derivation in order to check that Case-F valuation was delayed (global look-back) or we have to add some kind of check mark to the DP signaling that its Case-F valuation was delayed (which would violate Inclusiveness). As one reviewer suggests, however, this problem does not arise if we think of features as an attribute: value pair (e.g., [Case: Nom]). If this is true, a dp with an unvalued Case-F is marked as something like [Case: _] and the system can see the feature is unvalued. . I am abstracting away from the question of whether the construction can be iterated (as in estes nenos parece que parece que …), since the data become extremely unclear when more raising verbs are inserted.
The Case-F valuation parameter in Romance
here is whether it should be understood as an instance of microvariation, or macrovariation (see Baker, this volume). In principle, it seems that my proposal fits in well with the minimalist conjecture that parameterization across grammars may be restricted to lexical properties of functional items. In this case, the parameterization that I am proposing here for further-raising languages involves only one functional item, namely, ϕ-complete Tense. This is explicitly stated in the definition of the parameter I am discussing, which I give in (18): (18) The Case-F Valuation Parameter (operative in further-raising languages) After the unvalued ϕ-Fs of a ϕ-complete Tense head get a value from a matching DP, the valuation of the Case-F of the DP can be delayed.
A question that arises regarding (18) is why Case should be singled out in this way, that is, why this delay does not apply to other uninterpretable features, say wh-Fs on the DP, for example. One possible explanation comes from the A/A’ distinction; agreement with Tense is related to A-Movement/-Position properties, so it makes sense that a parameter related to Tense only affects features that play a role in the A-system, namely, Case-Fs. One complication for this microvariation approach is that the parameter that I am proposing affects Case-F valuation in terms of when such valuation occurs, a property that seems harder to relate just to lexical properties of a single functional item (see section 3.3). This therefore actually seems to confirm Baker’s (this volume) claim that variation among languages cannot only be the result of lexical properties of certain functional items (in this case, ϕ-complete Tense and, by extension, the dp agreeing with it). 3.3 V ariation resulting from the timing (not the nature) of syntactic operations Notice that an important feature of the parameterization proposed here is that it does not claim that syntactic operations are different across grammars, but rather, that their timing is different. A ϕ-complete Tense agrees with a DP, gets its ϕ-Fs valued, and values the Case-F of the agreeing DP. All these operations are universal, although Case-F valuation can be delayed in some languages but not in others. If this approach is on the right track, the timing of syntactic operations can be understood to be one of the sources of macrovariation, in Baker’s (this volume) terms (see section 3.2). This kind of solution to the problem of syntactic variation shows some resemblance to Huang’s (1982) approach to wh-in situ questions in Chinese. Huang claims that even though it seems that wh-phrases in Chinese do not move, what happens is that they move at LF, instead of moving at SStructure. Therefore, the movement does not have consequences for PF (since the
Gerardo Fernández-Salgueiro
PF representation is read off the S-Structure representation), but it does for LF (and so the semantics of questions are similar in Chinese and, say, English). It is interesting to see how exactly the same wh-movement operations are understood to apply at different levels of representation in Huang’s GB-based approach. Similarly, I am proposing that the exact same Case-F valuation operation applies at different points in the derivation. As can be seen, the main idea behind the two approaches is really similar, despite the obvious differences in the architectures assumed. In a GB-based approach, there is no notion of “derivational step”; conversely, in Minimalism, there is no notion of “level of representation”. Therefore, parametric variation due to “delay” is necessarily defined in terms of different levels of representation in GB, but in terms of different derivational steps in Minimalism. 3.4 P ossible consequences of the Case-F valuation parameter for the grammar of null subject languages Given that the Case-F Valuation Parameter is operative in the pro-drop Romance languages (with the exception of Romanian), one obvious question that arises is to what extent this parameter can be related to the null subject character of these languages (Rizzi 1982, 1986). The parameter that I am proposing here certainly opens up the possibility of accounting for referential null subjects in these languages in terms of unvalued Case-Fs on subject DPs. To illustrate this, consider the derivation in (19): (19)
Tense DP [uCase, φ]
Tense
Tense [EPP, φ] DP [uCase, φ]
vP (...)
This derivation is actually a sub-part of the derivation in (16), which illustrated how a further-raising example is derived. As can be observed, there is a point in the derivation at which the DP subject still has an unvalued Case-F, despite being in the canonical subject (A-) position. Suppose that the derivation does not continue after it reaches the point in (19). What would happen when this object is interpreted at the interfaces? If we adopt Chomsky’s (2001) system, Case-Fs never actually reach the interface with the Systems of Thought (SOT), so the derivation in (19) cannot be ruled out on the basis of SOT crash. On the SMS interface side, it is clear that the DP could
The Case-F valuation parameter in Romance
not be interpreted, since it lacks information about its Case-F. This alone would be a way to rule (19) out. Suppose, however, that we adopt a version of Perlmutter’s (1971) original approach to null subjects, under which null subjects are generated by deletion of the DP’s phonological features in the phonological component (see also more recent versions of this kind of approach, like Holmberg 2005, Roberts to appear); and we allow such deletion to take place as a repair strategy, only when the derivation has taken the delay of Case-F valuation option.13 This kind of approach would allow us to make the prediction that if a language exhibits Further Raising, then it must allow null subjects as well, since there is no way for the system to know whether the derivation is going to continue after Case-F valuation of the DP is delayed.14 As I pointed out above, all Romance further-raising languages are also null subject languages, which provides initial support for the validity of this implicational universal.15 This approach, though, does not make any reference, in principle, to other factors (syntactic or pragmatic) regulating the distribution of null vs. overt subjects in these languages.16 Further research is needed to evaluate whether this approach is also instrumental in understanding why these languages allow relatively free verb-subject inversion and lack the *that-trace effect (although cf. Nicolis, this volume). 3.5 Questions about parameter setting in acquisition A final important issue that should be discussed is what syntactic cues are available to a child acquiring the Case-F Valuation Parameter. In principle, given that the different sub-operations composing Agree are universal (and follow from optimal design considerations according to Chomsky 2005),
. Notice that this approach would fit in with the Minimalist idea that operations only apply if they have to, for economy reasons. In this case, deletion of the unvalued Case-Fs would apply only in order to salvage the derivation at the SMS interface. . Again, this argument only makes sense if we assume that there is no look-ahead in the system, and operations are performed based only on the information that is available at the current (and previous) steps in the derivation. . I am restricting here the domain of my proposal to the Romance language group, as null subjects in other languages might have very different properties (i.e., partial null subject languages and discourse null subject languages, see Holmberg 2005, Modesto, this volume and Neeleman & Szendröi, this volume). . See Fernández-Salgueiro (2006) for a more detailed analysis of null subjects along these lines.
Gerardo Fernández-Salgueiro
what a child acquiring a language has to do is determine whether Case-F valuation can be delayed in the target language or not. For further-raising languages, it seems that the main cue for the child can be the position of subjects agreeing with Tense. After all, the main difference between further-raising and non-further-raising languages in the linguistic input is that a subject in the former does not have to appear in the same projection (not even in the same clause) as the Tense head it is agreeing with, while in the latter, the relation between the pronounced copy of the subject DP and the agreeing Tense head has to be very local.17 We have already seen that the subject DP in further-raising structures can appear in a different clause than the Tense head it agrees with, as can be seen in (20) (repeated from (15)): (20) Algúns nenos parece que están tolos. some kids seems that are crazy ‘Some kids seem to be crazy’.
(Galician)
Still, in further-raising examples, the further-raised subject DP always agrees in ϕ-Fs with the embedded Tense head, which is evidence that at some point in the derivation they were in a structural relation local enough for Match and ϕ-F valuation to apply (recall that these sub-parts of Agree are assumed to apply universally in a Spec-Head configuration under this approach). Overt agreement morphology, then, seems to be playing a crucial role in providing cues to the child so that the Case-F Valuation Parameter can be correctly set; all the languages that have been discussed here display “rich” agreement morphology, with a distinct overt morphological realization for almost each person and number combination, this will help the child become aware of the agreement relation that always exists between the further-raised DP and the ϕ-Fs of the embedded Tense in all instances of furtherraising present in the linguistic input. 4. Conclusions In this paper, I have tried to show the need for a Case-F Valuation Parameter in Agree in order to be able to generate further-raising constructions, which are possible in many Romance languages. I have also explored the details of such a
. I am abstracting away now from A’-movement operations, which allow non-local (i.e., long-distance) dependencies, even in non-further-raising languages (cf. wh-movement examples like which man do you think Peter said was here?).
The Case-F valuation parameter in Romance
parameter, as well as its implications for the theory of principles and parameters in Minimalism in several domains. These include its relation to the pro-drop parameter, the universality of syntactic operations and the different timing of their application in different languages, syntactic cues available to help parameter setting in acquisition, and also whether this is an instance of microvariation or macrovariation.
References Alexiadou, A. & Anagnostopoulou, E. 1998. Parametrizing AGR: Word order, V-movement, and EPP-checking. Natural Language and Linguistic Theory 16: 491–540. Chomsky, N. 1995. The Minimalist Program. Cambridge MA:The MIT Press. Chomsky, N. 2000. Minimalist inquiries. The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. Epstein, S. & Seely, D. 2002. Rule applications as cycles in a level-free syntax. In Derivation and Explanation in the Minimalist Program, S. Epstein & D. Seely (Eds), 65–89. Oxford: Blackwell. Epstein, S. & Seely, D. 2006. Derivations in Minimalism. Cambridge: CUP. Fernández-Salgueiro, G. 2005. Agree, the EPP-F and further-raising in Spanish. In Experimental and Theoretical Approaches to Romance Linguistics, R. Gess & E. Rubin (Eds), 97–107. Amsterdam: John Benjamins. Fernández-Salgueiro, G. 2006. Romance null subjects at the sensory-motor interface. MS, University of Michigan. Ferreira, M. 2004. Hyperraising and null subjects in Brazilian Portuguese. MIT Working Papers in Linguistics 47: 57–85. Holmberg, A. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36: 533–564. Huang, C.-T.J. 1982. Logical Relations in Chinese and the Theory of Grammar. Ph.D. dissertation, MIT. Müller, G. 2004. Argument encoding and the order of elementary operations. Paper presented at the GLOW 2004, Thessaloniki, Greece. Nunes, J. 1999. Linearization of chains and phonetic realization of chain links. In Working Minimalism, S.D. Epstein & N. Hornstein (Eds), 217–249. Cambridge MA: The MIT Press. Nunes, J. 2004. Linearization of Chains and Sideward Movement. Cambridge MA: The MIT Press. Perlmutter, D. 1971. Deep Structure and Surface Structure Constraints in Syntax. New York NY: Holt Rinehart and Winston. Rezac, M. 2003. The fine structure of cyclic agree. Syntax 6: 156–182. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–557.
Gerardo Fernández-Salgueiro Roberts, I. To appear. A deletion analysis of null subjects: French as a case study. To appear in Null Subjects: The Structure of Parametric Variation, A. Holmberg & I. Roberts (Eds), Cambridge: CUP. Ura, H. 2000. Checking Theory and Grammatical Functions in Universal Grammar. Oxford: OUP. Zwart, J.-W. 1997. Transitive expletive constructions and the evidence supporting the multiple specifier hypothesis. In German: Syntactic Problems Problematic Syntax, W. Abraham & E. van Gelderen (Eds), 105–134. Tübingen: Niemeyer.
Silent arguments without pro The case of Basque* Maia Duguine Basque is traditionally described as a three-way pro-drop language, a paradigmatic instance of the positive setting of the pro-drop parameter. This article investigates the phenomenon of silent arguments in Basque and proposes a novel account of this phonological emptiness. It is argued that Basque silent arguments are best analyzed as elided DPs. This account avoids the shortcomings of the pro-drop analysis with respect to the licensing and interpretation effects of silent arguments, and its interaction with information structure allows a straightforward explanation of the distribution of overt and silent arguments. Keywords: pro-drop; ellipsis; agreement; information structure; focus; topic; Basque
1. The pro-drop parameter and Basque silent arguments In generative linguistic theory, parameters account for language variation. One of the best studied ones, the pro-drop parameter, states that the parameterized properties of inflection allow some languages to license a phonologically empty element pro in certain morphosyntactic configurations. If we adopt Rizzi’s (1986) influential formulation of the pro-drop parameter, for instance, pro has to be licensed and identified by an inflectional head. Within this approach, pro itself
* This research was supported by the grants BFF2002-04238-C02-01 of MCYT-FEDER, UPV-
EHU 9 UPV 00114.130-160.09-2004 U of EHU-U. Basque Country and a Ph.D. research grant by EHU-U. Basque Country. I want to thank Theresa Biberauer, Cedric Boeckx, Hamida Demirdache, Ricardo Etxepare, Tomohiro Fujii, Norbert Hornstein, Javier Ormazabal, Juan Uriagereka, Milan Rezac, the participants of the Conference on the Structure of Parametric Theory (Newcastle, UK, September 2005), the 3rd WoSS (Nantes, September 2005), and the audience of the Syntax Lunch Meeting at University of Maryland, where part of this material was presented, and especially Aritz Irurtzun, Myriam Uribe-Etxebarria and two anonymous reviewers for comments and helpful discussion. Usual disclaimers apply.
Maia Duguine
is an empty category, present as such in the lexicon, identified as [+pronominal] [−anaphoric], and silent in nature (cf. Chomsky 1982). Basque is a language with very rich inflectional morphology. As illustrated in (1), the finite verb, or its auxiliary, can display agreement with the ergative, absolutive and dative arguments:1 (1) Nik Joni artikuluak eman d-i-zki-o-t. I.erg Jon.dat papers(abs) give prs-root-3pl(abs)-3sg.dat-1sg.erg ‘I gave the papers to Jon.’
Furthermore, traditionally, Basque is also described as a three-way pro-drop language (cf. Goenaga 1980; Salaburu 1987; Laka 1988; Ortiz de Urbina 1989; Oyharçabal 1991): as illustrated in (2), the subject, direct object and indirect object (marked ergative, absolutive and dative) can be phonologically empty.2,3 (2) [eERG] [eDAT] [eABS] eman d-i-zki-o-t. give prs-root-3pl(abs)-3sg.dat-1sg.erg ‘I gave them to him/her/it.’
The meaning of the three empty elements in (2) is equivalent to that of personal pronouns nik (1st singular ergative), hari/berari (3rd singular dative) and horiek (3rd plural absolutive) respectively. The traditional account of this type of paradigm is in terms of the pro-drop parameter: Basque has pro in its lexical inventory and this element can be licensed and identified in subject, direct object and indirect object position because of the morphosyntactic properties of inflection. The structure of this paper is as follows. First, I discuss two pieces of data that the pro-drop approach cannot account for: the licensing of silent arguments in non-finite clauses and the impossibility of silent arguments being interpreted as topic or focal (Section 2). Then, I introduce the proposal that silent arguments in this language are regular elided DPs, and I provide evidence in favor of such an approach (Section 3). In Section 4, I return to the question of the participation in the information structure of sentences, spelling out a generalization on the interaction between ellipsis and information structure. Section 5 is the conclusion.
. These are the relevant abbreviations I will use in the text: [e] = phonologically silent argument, aux = inflected auxiliary, erg = ergative, abs = absolutive, dat = dative, sg = singular, pl = plural, neg = negation, nmlz = nominalizer, det = determiner, prs = present, prt = partitive. . In what follows, I will limit myself to the study of silent arguments, that is, I will not consider directly the cases of null expletives or quasi-argumental contexts. . Silent arguments are represented following the “neutral” word order which is standardly assumed to be S-IO-DO-V in ditransitive constructions.
Silent arguments without pro
2. Reanalyzing the pro-drop parameter in Basque In this section, I argue that, contrary to the tenets of the classical analysis, Basque silent arguments cannot be pros. As I show, the pro-drop hypothesis faces two serious problems, one morphosyntactic and the other semantic in nature. The first is that silent arguments are licensed in clauses with no inflection (Section 2.1), and the second is that this hypothesis cannot explain the interpretive and distributive facts of overt pronouns and silent arguments (Section 2.2). 2.1 Pro without agreement The pro-drop parameter is based on the correlation between the licensing of silent arguments and the presence of a “rich” agreement system in languages such as Italian or Spanish. However, there are also languages, such as Japanese or Chinese, that license silent arguments in the absence of apparent morphological agreement (cf. Huang (1984)). Basque presents some intriguing properties in the distribution of silent arguments with respect to agreement. Thus, whereas the language has been claimed to work like an extended case of Italian in finite contexts (cf. section 1.2), it seems to behave like Japanese in non-finite contexts. In particular, as first noted by Ortiz de Urbina (1983), Basque licenses silent arguments in the absence of agreement (see also Ortiz de Urbina 1989; Oyharçabal 1991; Zabala & Odriozola 1996; Elordieta 2002).4 (3) [e] nahi duzu [nik Joni artikuluak ematea]? want aux I.erg Jon.dat papers(abs) give.nmlz.det ‘Do you want [me to give the papers to Jon]?’ (4) [e] nahi duzu [eerg edat eabs ematea]? want aux give.nmlz.det ‘Do you want [(me/us/her/him/it/them) to give (me/us/you/her/him/it/them) to (me/us/you/her/him/it/them)]?’
As illustrated in (3), where the complement clause is nominalized (by the suffix -te), the three arguments of the verb eman “give” are licensed in the absence of overt
. The following data show that silent direct and indirect objects are syntactically projected in nominalized clauses: in (i) the overt or silent dative object can bind the anaphor and in (ii) the silent absolutive object is modified by an adjunct small clause.
(i)
Ez da harrigarria [Joni/[e] bere burua bakarrik gustatzea]. neg aux surprising Jon.to his/her head only please.nmlz.det Lit. ‘It is not surprising that only himself pleases to Jon/him/her.’
(ii)
Nahi duzu [nik Jon/[e] mozorroturik ikustea]? want aux I Jon dressed.up see.nmlz.det ‘Do you want me to see Jon/you/him/her/them dressed up?’
Maia Duguine
inflection. Example (4) further shows that all three of them can be phonologically silent.5 This is not predicted by the pro-drop approach. Zabala & Odriozola (1996) argue, however, that these clauses are only apparently “non-finite”, in that although agreement is not overtly expressed, they project Agr/T, which allows them to license pro. Evidence against this analysis comes from the so-called Person Case Constraint (PCC), a syntactic constraint on agreement relations. In the languages in which the PCC holds, person agreement (1st or 2nd person) with the absolutive object is blocked by the dative object intervener (cf. Ormazabal 2000). As illustrated in (5a), Basque displays PCC effects in finite clauses. However, Albizu (1997) shows that this effect disappears in nominalized ditransitive clauses: here the direct object can be something other than 3rd person (cf. (5b)). This shows that, contrary to the claim in Zabala & Odriozola (1996), the nominalized clause in (5b) does not involve any syntactic agreement.6 (5) a. *Azpisapoek/[e] etsaiari/[e] ni/[e] saldu n-(a)i-o-te. traitors.erg enemy.dat me(abs) sell 1sg.abs-root-3sg.dat-3pl.erg ‘The traitors have sold me to the enemy.’ b. Gaizki iruditzen zait [azpisapoek/[e] ni/[e] etsaiari/[e] saltzea]. wrong seem aux [traitors.erg me(abs) enemy.dat sell.nom.det] ‘Your selling me to the enemy seems wrong to me.’
Coming back to our discussion, given that there is no agreement in infinitives like (3) and (4), we have to conclude that the silent arguments of those non-finite clauses cannot be pros in the traditional sense.7 Hence, a global account of silent arguments in Basque cannot be provided by means of the pro hypothesis.8
. Note that the silent subject of the nominalized clause cannot be coreferential with any of the arguments in the main clause (San Martin (2000) calls this an apparent switch reference). . Thanks to Milan Rezac for suggesting this point. . Note, however, that the broader question about the licensing of overt DPs in non-inflected clauses remains (cf. Ortiz de Urbina 1989; Zabala & Odriozola 1996). . Another possibility would be that silent arguments in non-finite clauses are instances of socalled “null topic-bound” operators (cf. Ortiz de Urbina 1983; Elordieta 2002). But as illustrated in (i), these silent arguments show no subjacency effect:
(i) A: Miren haserre da. Miren angry aux ‘Miren is angry.’ B: Bai. Atzo atera gara [[e] gonbidatzea ahazturik]. yes yesterday go. out aux invite.nmlz.det forget.prt ‘Yes. Yesterday we went out having forgotten to invite (her).’
Silent arguments without pro
2.2 Overt pronouns and silent arguments: Interpretation The basic word order in Basque is SOV, but Basque is also a discourse-configurational language (in the sense of e.g., Kiss 1995). That is, information structure is marked syntactically: the discourse-semantic functions of topic and focus are associated with a particular structural relation (cf. Etxepare & Ortiz de Urbina 2003, and references therein). In particular, focused phrases must show up in the immediate preverbal position and topical phrases must surface to the left of focus (see (6)). However, silent arguments do not participate in this schema: as shown in (7)–(8), in contrast to overt pronouns, they cannot receive a focus or topic interpretation (note that this constraint seems to apply in other pro-drop languages too, cf. Larson & Luján 1989).9
(6) (TopP) (FocP) V Bera etorri da. s/he come aux ‘She came.’ ‘SHE came.’ ‘As for her, she came.’
[‘Neutral’ interpretation] [Focus interpretation] [Topic interpretation]
(8) [e] etorri da. come aux ‘She came.’ # ‘SHE came.’ # ‘As for her, she came.’
[‘Neutral’ interpretation] [No focus interpretation] [No topic interpretation]
(7)
. Note that what I define as topic does not strictly correspond to old or given information. I assume instead that topicalization is a syntactic process (cf. Rizzi 1997). Contrastive topicalization (CT) seems to be such a process in Basque: the sentence in (i) – where the object is fronted – is an appropriate answer to “Who ate the cookie?”, whereas (ii) – where the object is in situ – is not (the focused subject and the verbal complex are in peripheral positions (cf. Ortiz de Urbina 1989; Irurtzun 2006b)). Hence the DP sagarra, to be interpreted as a CT, has to move to the left of focus (to TopP).
(i)
Ez dakit, baina sagarra JONEK jan du. neg aux but apple Jon eat aux ‘I don’t know, but as for the apple, JON ate it.’
(ii) #Ez dakit, baina JONEK jan du sagarra. neg aux but Jon eat aux apple ‘I don’t know, but JON ate the apple.’ Similarly, (8) couldn’t be a proper answer to ‘‘Did Mary and the children finally come?’’: this shows that [e] cannot be a CT, and consequently it is not a syntactic topic marked with a formal feature [T]. Finally, I assume that the German or Chinese ‘‘null topic’’ described in Huang (1984) is not to be subsumed under the phenomenon of topicalization as defined here, but is also a silent argument of the ‘‘pro’’-type (Sigurðsson & Maling 2006), i.e., an elided DP.
Maia Duguine
Consider (9) for instance; the word order is Subject[F]-Verb-Object (the canonical one for subject focalization): its counterpart with an overt pronoun would be completely grammatical. Furthermore, the overt second person subject agreement on the verb makes it clear that a pro there would be properly identified and licensed. Logically then, there is no principled reason for (9) to be out; however, its silent subject simply cannot be interpreted as focused. (9) *[e][F] hautsi duzu basoa, ez nik! break aux.2sg.erg glass neg I.erg ‘(YOU) broke the glass, not me!’
The pro hypothesis by itself cannot explain the interpretive effects of the sentences in (7)–(8) and the ungrammaticality of (9). From the point of view of the traditional analysis of pro, we could imagine two potential ways of accounting for this asymmetry: (i) hypothesis I, stating that lexically, pros cannot bear a focus or topic feature, or (ii) hypothesis II, stating that there is a condition on derivations that requires for instance focus to bear Nuclear Stress at PF and that because of their lack of phonological content, pros cannot support focalization (i.e., they cannot get assigned stress). Various theoretical considerations make hypothesis I untenable. First, it is not clear what the explanation for such a lexical constraint could be. Then, the fact is that really any element in a derivation, even a part of a word or an expletive, can be focused metalinguistically (cf. Artstein 2004; Potts 2007), as illustrated in (10) (assuming the pleonastic “weather” reading of it). Clearly, pro is the only lexical item that cannot be assigned a focus feature in order to be metalinguistically focused. (10) Don’t misunderstand me, I didn’t say ‘Eve is hot’, I said ‘IT is hot!’
In the case of hypothesis II, it is the implementation of the condition on derivations itself that is problematic. The fact is that the realization of topic and focus is not universally dependent on the presence of phonological features of the target constituent (cf. Irurtzun 2006b; Elordieta 2006): in languages such as Wolof (cf. Rialland & Robert 2001), and some varieties of Basque (cf. Hualde, Elordieta & Elordieta 1994; Elordieta 2006) for instance, focus does not have to be realized with nuclear stress assignment. Consequently, a legibility condition on derivations that requires focused elements to bear a nuclear stress at PF cannot be maintained (cf. Irurtzun 2006b).10,11 Summarizing, the intuition behind the hypotheses I and II . Thanks to Aritz Irurtzun for valuable discussion of the focalization facts. . Two anonymous reviewers point out that hypothesis II can be replaced by the more general principle that [F]- or [T]-marked elements cannot be silent, spelling out the standard assumption that being overt is a necessary condition to be focal. However, this principle predicts that there should not be any silent material such as e.g., PRO, silent complementizers, relative operators,
Silent arguments without pro
seems to be on the right track, but the incompatibility of focalization and topicalization with silence cannot be implemented within a classical pro-drop analysis. Finally, I would also like to highlight that recent approaches to the so-called “empty categories” such as the copy theory of movement (cf. Chomsky 1995; Nunes 2004) or the movement approach to control constructions (cf. Hornstein 1999) derive phonological emptiness without having to postulate the existence of inherently silent empty elements such as traces or PRO. This methodological concern makes us wonder whether the silent argument phenomenon could also receive a principled account that does not rely on the stipulation of the existence of a silent empty category. It is a step in that direction that the discussion on Basque presented in the next sections is intended to be. 3. An alternative analysis to the Pro-Drop Parameter: Ellipsis Having discarded the traditional pro-drop analysis, I will now argue in favor of an ellipsis analysis: I propose that Basque silent arguments are actually elided argument DPs (cf. Perlmutter 1968; and cf. Kim (1999) and Holmberg (2005) for similar proposals for Korean and Finnish-type languages respectively). In this analysis, a silent argument will be a DP argument that is not phonologically realized. The first and main advantage of the ellipsis analysis is that it avoids the shortcomings of the pro-drop analysis. On the one hand, whereas the pro-drop analysis relies on licensing and identification conditions that are based on agreement relations, the ellipsis analysis does not. Not only does this analysis allow the licensing of silent arguments in non-finite clauses, but it also predicts such a possibility whenever overt arguments are licensed, as in nominalizations. That is, the question of the licensing of silent arguments reduces to the question of the licensing of overt DPs. On the other hand, in the literature on ellipsis it is assumed (overtly or implicitly; cf. Rooth 1992; Tancredi 1992; Merchant 2001 a.o.) that ellipsis and focalization are in a sense related, but this relation is one of incompatibility: it is what is not focused that can be elided. Hence, an analysis in terms of ellipsis opens the way to explaining why silent arguments cannot be interpreted as focused (see section 4). Finally and crucially, an account based on the independently motivated phenomenon of ellipsis derives the silent nature of silent arguments in a straightforward
or silent determiners in focal structures, contrary to fact. For instance, the presence of PRO in a focal phrase, as in (i), does not lead to ungrammaticality.
(i)
A: What happens? B: [Mary wants PRO to go home][F].
Maia Duguine
way, avoiding the need to stipulate the existence of phonologically empty elements in the lexicon. In the following sections, I give evidence in favor of the ellipsis analysis of Basque silent arguments. First, in section 3.1, I propose a new perspective on the general parallelism that we find between overt pronouns and silent arguments. Then, I discuss the absence of Overt Pronoun Constraint effects (section 3.2) and the existence of silent anaphors (section 3.3). Finally, data related to the interpretive effects specific to ellipsis will provide evidence in favor of the ellipsis hypothesis (section 3.4). 3.1 The identity between overt pronouns and silent arguments In this section, I review some of the basic data introduced in Ortiz de Urbina (1989: chapter 3) to argue in favor of the existence of pro in Basque. The evidence comes from (i) binding conditions, (ii) semantic binding and (iii) weak crossover data. First, overt pronouns and silent arguments are both subject to Condition B of the Binding Theory (cf. Chomsky 1982). On the one hand, both have to be free in their domain: for instance neither the silent [e] nor bera or ni in (11)–(12) can be bound by a c-commanding DP in its domain. (11) *Maiderreki [e]i /berai ikusten du. Maider s/he see aux ‘Maideri sees (heri).’ (12) *Niki [e]i /nii ikusten dut I me see aux ‘I see me.’
On the other hand, overt pronouns and silent arguments can be bound from outside their domain: the silent [e], bera and ni in (13)–(14) can be bound by a DP located in the main clause. (13) Maiderreki esan du [Peiok [e]i / berai ikusi duela] Maider say aux Peio s/he see aux.that ‘Maideri said Peio saw (heri).’ (14) Niki esan dut [Peiok [e]i / nii ikusi nauela] I say aux Peio I see aux.that ‘Ii said Peio saw (mei).’
It seems, then, that overt pronouns and silent arguments cannot be bound in their domain but can be bound from outside their domain: they are both subject to Condition B of the Binding Theory (BT). As pro is also subject to Condition B of BT (cf. Chomsky 1982), Ortiz de Urbina concludes that [e] is an instance of pro.
Silent arguments without pro
The second type of evidence given by Ortiz de Urbina (1989) in favor of the pro-drop analysis comes from semantic binding facts: both the overt pronoun and the silent argument can be logical variables bound by a c-commanding antecedent. (15) Ikasle bakoitzak dio [e] /bera hoberena dela. student each say s/he best aux.that ‘Each student says (s/he) is the best.’ (16) Irakasle askok esan diote ikasle bati [e] /bera hoberena dela. teacher many say aux student one.to s/he best aux.that ‘Many teachers said to a student that (s/he) is the best.’
In (15)–(16), both the overt pronoun and the silent argument can have a variable interpretation (besides the referential reading). Under that reading, the sentence in (15) expresses the proposition that for any student, this student says that he or she is a person that has the property of being the best. The silent argument or the overt pronoun functions then as an individual variable that is bound by the universal quantifier bakoitz “each”. In the same way, and again besides the referential reading, the interpretation of both the overt pronoun and the silent subject in (16) depends on the relative scope of the quantifiers in the main clause. Under the interpretation in which the subject irakasle askok “many teachers” has wide scope over the indirect object ikasle bati “to a student” in the main clause, the subject of the embedded clause is interpreted as a variable. As shown by the sentences in (17) and (18), 1st and 2nd person pronouns can function as bound variables in the same way as 3rd person pronouns. (17), for instance, can be interpreted as expressing the proposition that the speaker is the only person who has the property λx[x got a question that x understood] (besides the reading in which the pronoun or [e] is referential, i.e., referring to the speaker). (17) Bakarrik niki ukan dut [e]i /niki ulertu dudan galdera bat. only I have aux I understand aux.that question one ‘Only I got a question that (I) understood.’ (18) Bakarrik zueki ukan duzue [e]i /zueki ulertu duzuen galdera bat. only you(pl) have aux you(pl) understand aux.that question one ‘Only you got a question that (you) understood.’
Finally, a third argument given by Ortiz de Urbina (1989) in favor of the pro analysis of silent arguments comes from weak crossover effects. The weak crossover effect occurs in configurations in which a wh-element or a quantificational element undergoes movement across a coreferential pronoun by which it is not c-commanded.
Maia Duguine
(19) [[e]i /beraki maite zuen] neskak Joni utzi zuen. s/he love aux.that girl Jon leave aux ‘The girl that (s/he) loved left Jon.’ (20) *nori utzi zuen [[e]i /beraki maite zuen] neskak? who leave aux s/he love aux.that girl ‘Who did the girl that (s/he) loved leave?’
As illustrated in the minimal pair (19)–(20), both the pronoun and the silent argument display weak crossover effects: again, silent arguments show the same behavior as overt pronouns with regards to the phenomena typical of pronominal elements. The results of the tests presented here show the clear pronominal behavior of silent arguments. In the same way as Ortiz de Urbina (1989) took this as being evidence for classifying these silent elements as pro, I take it as being evidence in favor of an ellipsis analysis. Both approaches thus have the same empirical coverage for the type of data just considered. There are other cases, however, that make different predictions; I show in the following sections that the ellipsis analysis is to be preferred over the pro analysis. 3.2 More on semantic binding: The overt pronoun constraint The examples in (15)–(18) above showed that overt pronouns and silent arguments behave the same with regard to semantic binding. It has to be noted though that this result is not predicted by the Overt Pronoun Constraint (Montalbetti 1984), a generalization on the distribution of overt pronouns and pro in which overt pronouns are said to be unable to be bound. (21) Overt Pronoun Constraint (Montalbetti 1984: 94): Overt pronouns cannot link to formal variables iff the alternation overt/empty obtains.
This generalization is one of the most well-grounded arguments in favor of the claim that there exists a silent pronoun with idiosyncratic properties, since, as Montalbetti suggests, “if the interpretation of an element is sensitive to its being lexical or not, then what has been called the Null Subject Parameter cannot be considered solely as a PF phenomenon” (Montalbetti 1984: 74). Thus the non-application of the Overt Pronoun Constraint constitutes evidence against the claim that Basque silent arguments are pros and suggest instead that it is “solely as a PF phenomenon”. 3.3 Silent anaphors In section 3.1, I presented the type of data given in Ortiz de Urbina (1989) that shows that, like pronouns, silent objects are subject to Condition B of BT. However, the data are more complex than that: as is illustrated by the alternation between
Silent arguments without pro
strict and sloppy interpretation for [e] in the examples in (22) and (23), Basque silent arguments can be anaphoric, i.e., subject to Condition A of BT. (22) Jonek bere burua kritikatzen duelarik, Mirenek [e] goraipatzen du. Jon his/her head criticize aux.when Miren praise aux ‘When Jon criticizes himself, Miren praises (him/herself).’ (23) A: Zuk zure burua goraipatzen duzu beti. you your head praise aux always ‘You always praise yourself.’ B: [e] nahi duzu [[e] [e] kritikatzen hastea]? want aux criticize begin.nmlz ‘Do you want me to begin criticizing (you/myself)?’
This type of data is predicted by the DP ellipsis hypothesis: anaphoric DPs as well as pronominal DPs can be the target of ellipsis. Obviously, however, the anaphoric interpretation cannot be directly explained by the pro hypothesis, and thus this type of analysis cannot give a uniform account of the silent argument phenomenon. In these conditions, as far as I can tell, the only way to account for the silent anaphors in (22)–(23) is to analyze them as part of an elided constituent. In particular, it could be the effect of verb-raising followed by VP-ellipsis (cf. Otani & Whitman 1991). However, this option has to be discarded because, on the one hand, V-raising does not take place overtly in Basque (cf. Laka 1990) and on the other hand, as is illustrated by (24), silent anaphors are licensed even in contexts without any parallel VP antecedent, contrary to English VP ellipsis (the possessive pronoun can refer either to Jon or to Peio). (24) A: Jon askotan [[bere txakurra]dp – rekin]pp ateratzen da. Jon often his dog with go.out aux ‘Jon often goes out with his dog.’ B: Peiok aldiz ez du inoiz [e] ateratzen. Peio however neg aux never go.out ‘Peio, however, never walks (his dog).’
In consequence, if silent anaphors are to be explained in terms of ellipsis, the elided constituent will have to be the object DP itself.12 Now, if the existence of DP ellipsis is assumed, it can in principle be extended to account for any type of silent argument. But then, as all cases can be accounted for by the ellipsis analysis, the pro account of non-anaphoric arguments becomes redundant.
. Cf. Kim (1999) for an extended discussion on the hypotheses of VP ellipsis vs. DP ellipsis in Korean.
Maia Duguine
3.4 Not pros but elided DPs: Dahl’s puzzle Dahl’s puzzle or the two-pronoun puzzle is the name given to the interpretive asymmetry that appears in English in certain contexts of VP ellipsis containing pronouns (cf. Dahl 1974; Fiengo & May 1994; Fox 2000 among others). A sentence such as the one in (25) permits four readings for the second part of the parallel conjunction, which are logically predicted by the anaphoric properties of each of the pronouns (strict or sloppy reading, see (27)). However, when the higher VP in the second conjunct is elided, as in (26), one of the four readings “disappears”: crucially, the strict + sloppy reading in (27d). (25) Maxi said hei saw hisi mother, and Oscar said he saw his mother, too. (26) Maxi said hei saw hisi mother, and Oscar did, too. (27)
a. b. c. d.
Oscar said Max saw Max’s mother. Oscar said Oscar saw Oscar’s mother. Oscar said Oscar saw Max’s mother. Oscar said Max saw Oscar’s mother.
[strict + strict] [sloppy + sloppy] [sloppy + strict] [strict + sloppy]
This interpretive asymmetry between the realization and the ellipsis of pronouns is what constitutes Dahl’s puzzle. Let us take Dahl’s puzzle as a property of ellipsis structures. In fact, as pointed out by Fiengo & May (1994), “the effect of ellipsis is … eliminative. Ellipsis no longer preserves anaphoric possibilities; rather, it restricts their realization” (Fiengo & May 1994: 130). If Basque silent arguments are indeed elided elements, they should show the same effect, the same asymmetry between realization and non-realization. Below I show that this prediction is borne out. In a structure with two pronouns equivalent to (25)–(26), each of the two pronouns can be silent or realized. Given the discourse setting in (28), any of the parallel utterances in (28a–d) is felicitous.13,14
. In (28) I have deliberately adopted a structure with two personal pronouns in order to avoid the problematic interactions that would be caused by the possessive pronoun. Note, however, that the effects in English also remain the same as in (25)–(27) when the sentences are equivalent to Basque (28), i.e., when both pronouns are personal pronouns. . In the discourse context, that is, in the first sentence in (28), the two embedded subjects are silent. The use of overt pronouns instead (in both subject positions or just in one) is also possible and permits the establishment of the same anaphoric relation between the DP Miren and both pronouns. But since overt pronouns tend to be interpreted as emphasized, I have chosen to use silent arguments here to avoid interference from other phenomena such as focalization. It would be interesting, however, to see the effects that we get in (28a–d)-type sentences when there are already focused pronouns in the discourse, a matter I leave for further research.
Silent arguments without pro
(28) Mireneki esan digu [[e]i uste duela [[e]i irabaziko duela]] Miren say aux think aux.that win aux.that ‘Miren told us that she thinks she will win.’ a.
Anak esan digu [berak uste duela [berak irabaziko duela]]. Ana say aux s/he think aux.that s/he win aux.that ‘Ana told us that she thinks she will win.’
b. Anak esan digu [[e] uste duela [[e] irabaziko duela]]. c. Anak esan digu [[e] uste duela [berak irabaziko duela]]. d. Anak esan digu [berak uste duela [[e] irabaziko duela]].
As before, the sentences in (28a–d) should in principle give rise to four interpretations, depending on the reading each pronoun gets (i.e., following the adopted designation system, strict or sloppy): (29)
a. b. c. d.
Ana told us that Miren thinks Miren will win. Ana told us that Ana thinks Ana will win. Ana told us that Ana thinks Miren will win. Ana told us that Miren thinks Ana will win.
[strict + strict] [sloppy + sloppy] [sloppy + strict] [strict + sloppy]
However, as in English (25)–(26), the actual possible interpretations do not reflect this logic: whereas (28a) and (28c) can have all four possible readings, (28b) and (28d) – where at least the most embedded subject is silent – can have the readings in (29a,b,c) but not the one in (29d).15 Hence, in the same way as English shows interpretive asymmetries between realization and non-realization of pronominal dependencies in VP ellipsis, Basque shows interpretive asymmetries in the alternation between silent arguments and overt pronouns. Furthermore, the “lost” reading (29d) is the strict + sloppy reading, crucially, the same as the one that is impossible in English VP ellipsis. The natural conclusion to this fact is that the Basque data in (28)–(29) simply illustrate an instance of Dahl’s puzzle. Thus, if we take Dahl’s puzzle to be a property of ellipsis, the result is straightforward: Basque silent arguments are elided constituents. It is also important to emphasize that an analysis in terms of pro-drop would predict that the reading in (29d) is possible for (28b) and (28d). In principle, a pro referring to Ana, just like a pro referring to Miren should be able to be licensed and identified in any of the embedded subject positions in (28). Hence, there is no principled reason in (28b,d) for blocking the combination of a first subject (overt pronoun or pro) referring to Miren and a second more embedded pro referring to
. The specific contexts in which each reading is possible differ from one another. This is also related, as will be shown in section 4, to the “form” that the sentence has to get in terms of information structure in order to be able to display a given reading.
Maia Duguine
Ana. In fact, in a context where, for instance, we know that Ana is an athlete and that her friend Miren always makes good forecasts, (28b,d) can be uttered felicitously as meaning “Ana told us that Miren thinks that Ana will win”. It is only in the type of contexts in (28) that the strict + sloppy reading gets blocked. Thus, an analysis in terms of pro-drop is unable to account for the puzzle that we observe in Basque. In section 4, I return to this question and link the interpretive asymmetries in Dahl’s puzzle data to information structure phenomena.
4. Explaining the distributional facts of overt and elided pronouns In principle, ellipsis has purely phonological effects: i.e., the phonological realization or non-realization of a constituent will not affect the logical form of the sentence (cf. Tancredi 1992; Rooth 1992). In that sense, ellipsis is optional.16 From the point of view of Basque silent arguments, this means that an overt and a covert argument will be semantically equivalent. However this cannot be the whole explanation, since silent arguments cannot be focused or topicalized (cf. section 2.2) and there are interpretive asymmetries in the distribution of overt pronouns and silent arguments (cf. section 3.4). In this section, I show that the second fact can be reduced to the first, and I discuss the interaction between ellipsis and information structure. Dahl’s puzzle effects are explained by the distribution of overt and silent arguments with regard to [T(opic)] and [F(ocus)] features. I will show that it is crucially when the lower subject is part of the information structure of the sentence that the strict + sloppy reading is available. Let us look again at the sentences that do not display the strict + sloppy reading in Basque. Observe (28b) and (28d), repeated here as (33) and (34) respectively: (33) Anak esan digu [[e] uste duela [[e] irabaziko duela]]. Ana say aux think aux.that win aux.that ‘Ana told us that she thinks she will win.’ (34) Anak esan digu [berak uste duela [[e] irabaziko duela]].
While the intermediate subject is elided only in (33), the lower subject is elided in both (33) and (34). This shows that the impossibility of this reading lies in the “absence” of the lower pronoun, and not in the absence of the first or of both
. Note that the targeted element also has to be a syntactically “elidable” constituent, as for instance VP is in English but not in other languages (cf. Lobeck 1995 a.o.). We could then speculate that maybe this is the locus of crosslinguistic variation for silent arguments: DPs could be elidable in Basque, but not in English.
Silent arguments without pro
pronouns. To understand what is going on here, let us look at the sentences that do allow the strict + sloppy reading, the sentences in (28a) and (28c) (repeated here as (35)): (35) Anak esan digu [berak /[e] uste duela [berak irabaziko duela]]. Ana say aux s/he think aux.that s/he win aux.that ‘Ana told us that she thinks she will win.’
Evidence from word-order facts will show that in fact, for the strict + sloppy reading to be available in (35), the lower pronoun has to be focused or topicalized, i.e., (35) will be represented as in (36). As schematized in (6) above, Basque shows a topicfocus-verb adjacency; hence in (37), where the word-order is OV-berak, berak is neither focused nor topicalized. But in (38), where berak is left adjacent to the verb, it can be interpreted as focal, and in (39), being left adjacent to the focused object, it can be interpreted as topic (in the right contexts, of course17). (36) Anak esan digu [berak/[e] uste duela [berak[F] / berak[T] irabaziko duela]]. (37) Anak esan digu [berak/[e] uste duela Ana say
aux s/he
[LASTERKETA[F] irabaziko
think aux.that race
win
duela berak] aux.that s/he ‘Ana told us that she thinks she will win THE RACE.’ (38) Anak esan digu [berak/[e] uste duela [lasterketa[T] Ana say aux s/he think aux.that race BERAK[F] irabaziko duela]]. s/he win aux.that ‘Ana told us that she thinks SHE will win the race.’ (39) Anak esan digu [berak/[e] uste duela [berak[T] LASTERKETA Ana say aux s/he think aux.that s/he race
irabaziko duela]]. win aux.that
‘Ana told us that she thinks as for her, she will win THE RACE.’
The three sentences above allow for the strict + strict, sloppy + sloppy and sloppy + strict readings; but only (38) and (39) allow for the strict + sloppy reading. The . (36) and (38) would be felicitous in a context in which “Miren says that she thinks she will win the TENNIS MATCH” has just been uttered, and (37) after “Miren says that she thinks SHE will win the race” has been uttered, for example (cf. footnote 16).
Silent arguments without pro
interpreted as strict + sloppy since that would imply a violation of the information structure-based constraint on ellipsis stated in (40).20 5. Conclusion Basque has traditionally been described as a three-way pro-drop language, a typical case of the positive setting of the pro-drop parameter. However, given that silent arguments are not licensed by agreement and that the pro-drop analysis cannot account for the distribution of overt pronouns and silent arguments, I arrived at the conclusion that these elements are not pros. The immediate implication is that if, in the case of Basque, the pro-drop parameter plays no role in the fact that arguments can be silenced, a question as to the nature of the “null argument parameter” arises: specifically, is it possible that there are in fact different forms of silent arguments crosslinguistically? Furthermore, looking beyond Basque, it has to be noted that it is a general assumption that silent arguments in pro-drop languages cannot be interpreted as focal. As argued in sections 2 and 3, this raises problems for the pro-drop analysis since it is unable to explain the impossibility of focusing pro in any language. This is, I believe, a crucial problem for the pro-drop parameter. The alternative analysis I have put forth in this paper is based on ellipsis, an operation that is independently motivated. According to this proposal, the distribution of overt and elided pronouns is explained by means of the information structure properties in terms of which a pronoun marked [F] or [T] is barred from ellipsis. The ellipsis analysis has been shown to have broader empirical coverage than the pro-drop analysis. Accepting that the silent argument phenomenon cannot be accounted for by the pro-drop analysis, and knowing that ellipsis is a priori an option available in all languages, what now remains to be seen is whether the present analysis can be extended to other languages and whether it also has greater explanatory power than the pro-drop analysis. This seems plausible. For instance, Holmberg (2005) analyzes 1st and 2nd person subjects in Finnish as deleted DPs. Likewise, Kim (1999) and Tomioka (2003) propose that null arguments in Japanese or Korean are deleted bare NPs. As for Spanish, a typical pro-drop language, Luján
. Note incidentally that the same explanation can be given for Dahl’s puzzle in English. Fiengo & May (1994: 130 fn. 1) acknowledge the importance of the role of focus (“emphasis”) in getting the different readings for (25); this appears however to be really relevant, because the strict + sloppy reading is more clearly available when the second pronoun is emphasized than when it is not (Brendan Costello, Scott Fults, Gillian Ramchand, Heather Taylor p.c.). That is, because this pronoun is focused, the VP it appears in cannot be elided. In consequence the elided version will have a non-focused second pronoun, hence will not permit the strict + sloppy reading.
Maia Duguine
(1999) shows that Montalbetti’s (1984) Overt Pronoun Constraint can be obviated: overt pronouns can be interpreted as bound variables in contexts that require them to be emphasized (cf. also Larson & Luján 1989; and Ortiz de Urbina 1989). This suggests that overt pronouns and silent arguments are the same and that there is no constraint regulating their distribution, apart from the independent constraints imposed by information structure.
References Albizu, P. 1997. Generalized person-case constraint: A case for a syntax-driven inflectional morphology. In Theoretical Issues at the Morphology-Syntax Interface, A. Mendikoetxea & M. Uribe-Etxebarria (Eds), 1–33. Donostia: ASJU. Artstein, R. 2004. Focus below the word level. Natural Language Semantics 12(1): 1–22. Chomsky, N. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge MA: The MIT Press. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Dahl, Ö. 1974. How to open a sentence. Abstraction in natural language. Logical Grammar Reports 12, University of Götenberg. Duguine, M. 2006. The role of information structure in interpretive asymmetries. In Proceedings of the ESSLLI Workshop Ambiguity in Anaphora, R. Artstein & M. Poesio (Eds), 41–48. Málaga: Universidad de Málaga. Elordieta, A. 2002. The role of verbal agreement in licensing null arguments. In Kasu eta Komunztaduraren Gainean, B. Fernandez & P. Albizu (Eds), 113–132. Donostia: EHU. Elordieta, G. 2006. Constraints on intonational prominence of focalized constituents. In Topic and Focus: Intonation and Meaning, C. Lee, M. Gordon & D. Büring (Eds), 1–22. Dordrecht: Kluwer. Etxepare, R. & Ortiz de Urbina, J. 2003. Focalization. In A Grammar of Basque, J.I. Hualde & J. Ortiz de Urbina (Eds), 459–516. Berlin: Mouton de Gruyter. Fiengo, R. & May, R. 1994. Indices and Identity. Cambridge MA: The MIT Press. Fox, D. 2000. Economy and Semantic Interpretation. Cambridge MA: The MIT Press. Goenaga, P. 1980. Gramatika bideetan, 2nd Edn. Donostia: Erein. Holmberg, A. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36(4): 533–564. Hornstein, N. 1999. On control. Linguistic Inquiry 30(1): 69–96. Hualde, J.I., Elordieta, G. & Elordieta, A. 1994. The Basque Dialect of Lekeitio. Supplements to the International Journal of Basque Linguistics and Philology. Donostia: ASJU. Huang, C.-T.J. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15(4): 531–574. Irurtzun, A. 2006a. A derivational approach to the focus structure. MS, EHU-U, Basque Country. Irurtzun, A. 2006b. Focus & clause structuration in the minimalist program. In Minimalist Essays, C. Boeckx (Ed.), 68–96. Amsterdam: John Benjamins. Kim, S. 1999. Sloppy/strict identity, empty objects, and NP ellipsis. Journal of East Asian Linguistics 8: 255–284. Kiss, E.K. 1995. Introduction. In Discourse Configurational Languages, E.K. Kiss (Ed.), 3–27. Oxford: OUP.
Silent arguments without pro
Laka, I. 1988. Configurational heads in inflectional morphology: The structure of the inflected forms in Basque. International Journal of Basque Linguistics and Philology (ASJU) XXXII(2): 343–365. Laka, I. 1990. Negation in Syntax: On the Nature of Functional Categories and Projections. Ph.D. dissertation, MIT. Larson, R. & Luján, M. 1989. Emphatic pronouns. MS, Stony Brook University, New York and University of Texas, Austin. Lobeck, A. 1995. Ellipsis: Functional Heads, Licensing, and Identification. Oxford: OUP. Luján, M. 1999. Expresión y omisión del pronombre personal. In Gramática Descriptiva de la Lengua Española. Vol. 1, I. Bosque & V. Demonte (Eds), 1275–1315. Madrid: Espasa. Merchant, J. 2001. The Syntax of Silence. Oxford: OUP. Montalbetti, M. 1984. After Binding: On the Interpretation of Pronouns. Ph.D. dissertation, MIT. Nunes, J. 2004. Linearization of Chains and Sideward Movement. Cambridge MA: The MIT Press. Ormazabal, J. 2000. A conspiracy theory of case and agreement. In Step by Step: Papers in Honour of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 235–260. Cambridge MA: The MIT Press. Ortiz de Urbina, J. 1983. Empty categories and focus in Basque. Studies in the Linguistic Sciences 15(1): 133–156. Ortiz de Urbina, J. 1989. Parameters in the Grammar of Basque: A GB approach to Basque Syntax. Dordrecht: Foris. Otani, K. & Whitman, J. 1991. V-raising and VP-ellipsis. Linguistic Inquiry 22(2): 345–358. Oyharçabal, B. 1991. Théorie des pronoms nuls et diversité linguistique: Bilan. In Memoriae Mitxelena Magistri Sacrum, Supplements to the International Journal of Basque Linguistics and Philology, J. Lakarra (Ed.), 867–898. Donostia: ASJU. Perlmutter, D.M. 1968. Deep and Surface Structure Constraints in Syntax. Ph.D. dissertation, MIT. Potts, C. 2007. The dimensions of quotation. In Direct Compositionality, C. Barker & P. Jacobson (Eds), 405–431. Oxford: OUP. Rialland, A. & Robert, S. 2001. The intonational system of Wolof. Linguistics 39(5): 893–939. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17(3): 501–557. Rizzi, L. 1997. On the fine structure of the left-periphery. In Elements of Grammar, L. Haegeman (Ed.), 281–337. Dordrecht: Kluwer. Rooth, M. 1992. Ellipsis redundancy and reduction redundancy. In Proceedings of the Stuttgart Ellipsis Workshop, S. Berman & A. Hestvik (Eds). Stuttgart: SBF 340, University of Stuttgart. Salaburu, P. 1987. Kategoria gauzatuak eta kategoria isilak. In Sintaxi arazoak, P. Salaburu (Ed.), 13–34. Donostia: EHU. San Martin, I. 2000. Control in Basque. In Proceedings of WCCFL 19, R. Billerey & D. Lillehaugen (Eds), 425–438. Somerville MA: Cascadilla. Sigurðsson, H. & Maling J. 2006, Argument drop and the empty left edge condition (ELEC). MS, Lund University, Lund and Brandeis University, Waltham MA. Tancredi, C. 1992. Deletion, Deaccenting, and Presupposition. Ph.D. dissertation, MIT. Tomioka, S. 2003. The semantics of Japanese null pronouns and its cross-linguistic implications. In The Interfaces: Deriving and Interpreting Omitted Structures, K. Schwabe & S. Winkler (Eds), 321–339. Amsterdam: John Benjamins. Zabala, I. & Odriozola, J.C. 1996. On the relation between DP and TP. The structure of Basque infinitivals. Catalan Working Papers in Linguistics 5(2): 231–281.
Case morphology and radical pro-drop Ad Neeleman & Kriszta Szendrői This paper focuses on the so-called discourse pro-drop languages, i.e., those permitting radical pro-drop, which affects not just subjects, but other pronominal arguments too. Whereas previous attempts to understand the properties of these languages have focused on verbal agreement, the proposal here is that it is in fact the properties of the pronominal paradigm that determine the availability or otherwise of radical pro-drop. In particular, we propose that radical pro-drop will only be possible in languages with “transparent” (i.e., non-fusional) pronominal paradigms, since pro-drop falls out as the consequence of the zero spell-out of regular pronouns. On the empirical front, we show that investigation of a wide range of languages would seem to confirm our prediction. Keywords: (discourse) pro-drop; pronominal morphology; zero spellout; Elsewhere Condition; typology
1. The problem The generalization that pro-drop is conditioned by rich agreement allows for a very attractive theory that reduces variation in the syntax to variation in the lexicon. The central idea is that languages allow pro-drop to the extent that their verbal agreement paradigm expresses the φ-features necessary for local recovery of the content of dropped arguments (see Taraldsen 1978; Rizzi 1986 among others). The generalization is mainly based on European languages. Richly inflected languages like Italian, Spanish and Greek allow subject drop, but English, Dutch and Swedish, which are poorly inflected, do not. An agreement-based theory of pro-drop faces difficulties with languages like Japanese and Chinese, which lack agreement and nevertheless allow pro-drop. In fact, pro-drop in these languages seems to be more wide-spread than in Italiantype languages: (almost) any pronominal argument can be omitted. The literature refers to this phenomenon as “radical pro-drop”, “rampant pro-drop”, or, perhaps most frequently, “discourse pro-drop”. It is illustrated below (Chinese data from Huang 1984: 533, 563).
Ad Neeleman & Kriszta Szendrői
(1) a. Ø siken-ni otita. exam-dat failed ‘pro failed the exam.’
Japanese
b. Bill-ga Ø setokusuru. Bill-nom persuade ‘Bill persuaded pro.’ c. [Ø mimi-ga] naga- i. ear- nom long- pres ‘pro’s ears are long.’ (2) a. Ø kanjian ta le. see he perf ‘pro saw him.’
Chinese
b. Ta kanjian Ø le. he see perf ‘He saw pro.’ c.
Zhangsan, [Ø baba] hen youqian. Zhangsan father very rich ‘Zhangsan, pro’s father is very rich.’
One possible reaction to these data is to give up on any connection between prodrop and agreement. But such a move would amount to abandoning what insight we have into the phenomenon in favour of descriptivism. It is more desirable to develop a theory that maintains the agreement-based account where it seems relevant, but allows pro-drop in the absence of agreement under certain welldefined circumstances. We propose such a theory in Sections 2–4. What sets apart our proposal from competing theories (Huang 1984; Speas 2004; Tomioka 2003) is that it focuses on the pronominal paradigm. In Neeleman & Szendrői 2007, we argue that a language will only allow radical pro-drop if its personal pronouns are agglutinating for case, number, or some other nominal feature. In this paper we will defend a subpart of this claim, namely that radical pro-drop is blocked by fusional case, but allowed when the expression of case is agglutinative. Thus, the morphological characteristics of the pronominal paradigm determine whether radical pro-drop is allowed. In languages that do not have an agglutinative pronominal paradigm, omission of pronouns is possible, but only in the presence of rich verbal agreement. The proposal thus extends the original idea that variation in the lexicon may determine variation in the syntax. Our proposal does not address the pragmatic conditions under which prodrop can take place in discourse. Rather we intend to find out what grammatical
Case morphology and radical pro-drop
c haracteristics make radical pro-drop available, and what typological predictions can be derived from these. A full theory of pro-drop requires an additional pragmatic component that governs the use of null pronouns in languages whose grammar allows them. There are many proposals in the literature that deal with this aspect of radical pro-drop, but evaluating these is beyond the scope of this paper.
2. Radical pro-drop as a spell-out phenomenon 2.1 The generalizations Let us now explore the idea that the possibility of radical pro-drop in a given language is dependent on the nature of its pronominal paradigm. We believe that the following generalization provides a good approximation of the cross-linguistic distribution of radical pro-drop. A language may drop pronouns if its pronouns either do not vary for case or, if they do vary, case morphology is agglutinating. The two options are exemplified by Chinese and Japanese. Chinese is a language with invariant pronouns. As the example in (3) illustrates, the nominative and accusative forms of the third person, singular, masculine pronoun are identical (see Huang 1984: 533). (3) Ta kanjian ta le. he see he perf ‘He saw him.’
Chinese
Japanese pronouns do inflect for case, but the inflection is clearly separate from the pronominal stem. In other words, the case morphology on pronouns is agglutinative rather than fusional. This is illustrated in (4). (4) Kare-ga kare-o he-nom he-acc ‘He persuades him.’
setokusuru. persuade
Japanese
In contrast, languages in which case on pronouns is fusional do not allow radical pro-drop. This is clearly the case in English (as he saw him / she saw her shows). Hence English pronouns cannot be omitted. The same is true of Italian. The fact that Italian pronouns have fusional case morphology blocks radical pro-drop, with the consequence that omission of arguments is restricted to contexts where it is licensed by agreement. We argue that the correlation between the morphology of pronouns and the availability of radical pro-drop derives from three independently motivated assumptions. First, as originally argued by Perlmutter (1971), null arguments are
Ad Neeleman & Kriszta Szendrői
regular pronouns that fail to be spelled out at PF, rather than special silent lexical items, pro (see also Holmberg 2005). Second, as argued by Weerman & EversVermeul (2002), the spell-out rules for pronouns often target non-terminal nodes, rather than terminals (see also Pannemann & Weerman (2005). Finally, competition between different spell-out rules is regulated by the Elsewhere Condition (see Kiparsky 1973, and subsequent work). We will discuss these assumptions below. 2.2 Pronominal spell-out We start by discussing the nature of spell-out rules for pronouns. For the sake of concreteness, we assume that the extended nominal projection consists of an NP, dominated by a DP, which is in turn dominated by a KP (or Case Phrase). The hypothesis that there is a DP goes back to Abney (1987), and has been widely adopted. Motivation for KP can be found in Bittner & Hale (1996) and Neeleman & Weerman (1999), amongst others. It is conceivable that there is some crosslinguistic variation in the make-up of the extended nominal projection, but we take KP and NP to be universal. KP
(5) K
DP D
NP N
...
There is general agreement that in an ordinary referential noun phrase like the old man the phonological units the, old and man correspond to terminal nodes. In theories that have en bloc lexical insertion of semantic, syntactic and phonological material, this is the only possibility. Phonological units cannot be associated with non-terminal nodes. This is different in theories based on “late spell-out” (see Sproat 1985; Halle & Marantz 1993; Jackendoff 1997; and many others). In such theories, syntactic terminals do not contain phonological information; rather, syntactic representations are associated with phonological material in a mapping procedure at the PF interface. They therefore allow spell-out of terminal nodes, as well as larger chunks of structure. In a recent paper, Weerman & Evers-Vermeul (2002) argue that pronouns very often correspond to larger chunks of structure than D or N. The evidence they provide is partly based on the distribution of Dutch possessive pronouns. As (6a) shows, regular possessive pronouns realize D: the pronoun mijn “my” is in
Case morphology and radical pro-drop
complementary distribution with the determiner de “the”, but not with any other material that can normally appear in the extended nominal projection. A second type of possessive pronoun, given in (6b) has a distribution that suggests that it spells out NP, while a third colloquial variant spells out DP or KP, see (6c).1 (We do not demonstrate this here, but no noun or adjective can be added in (b), while (c) does not permit insertion of noun, adjective or determiner.) (6) a.
MijnD mooie boek is gestolen. my beautiful book is stolen ‘My beautiful book has been stolen.’
Dutch
b. De mijneNP is gestolen. the mine is stolen ‘Mine is been stolen.’ c.
MijnesDP/KP is gestolen. mine is stolen ‘Mine has been stolen.’
Weerman & Evers-Vermeul suggest that personal pronouns like hem “him” correspond to KPs. They cannot be combined with any other material normally hosted by DP, and they vary in form depending on their case: (7) Hij heeft hemKP gisteren nog gezien. He have him yesterday still seen ‘He saw him only yesterday.’
Dutch
If personal pronouns do indeed spell out non-terminal categories, the pronominal paradigm consists of a set of spell-out rules that distinguish between KPs on
. A reviewer suggests an alternative account for the distribution of Dutch pronouns involving head movement. In this view, pronouns would be N or D-heads and move up to D or K. The problem with this analysis is that it predicts that nominal modifiers could appear alongside the pronouns, or be stranded inside the NP by the moved head (Weerman & Evers-Vermeul 2002: 314). But as is clear from (i) this is not the case.
(i) a. *[DP De [NP mooie mijneN]] is gestolen. the beautiful mine is stolen ‘My beautiful one has been stolen.’
b. *[DP MijnesN [NP mooie tN]]] is gestolen. mine beautiful is stolen ‘My beautiful one has been stolen.’ c. *Hij heeft [KP hemN [DP tN [NP mooie tN]]] gisteren nog gezien. he have him beautiful yesterday still seen ‘He saw him, the beautiful one, only yesterday.’
Dutch
Ad Neeleman & Kriszta Szendrői
the basis of their phi-feature composition. For example, hem is introduced by the spell-out rule in (8).2
(8) [KP +p, –a, 3sg, masc, acc] ⇔ /hem/
Of course, the idea that pronouns stand for complete nominal phrases is quite intuitive, and was part of traditional as well as early generative grammar. However, it is not uncontroversial. Since Postal (1966), personal pronouns have been analysed as occupying the D-position. The main evidence for this is based on expressions like us guys, in which a pronoun seems to take an NP complement. However, various linguists have argued that this conclusion is incorrect, and that the relation between us and guys is more like apposition than like complementation. (See Bhat 2004: 50–52 for an overview.) 2.3 Radical pro-drop as zero spell out of regular pronouns Given our emphasis on spell-out, a natural way of analyzing pro-drop is to assume that null arguments are regular pronouns in syntax that fail to be realized at the PF interface. Thus, radical pro-drop would be the result of application of the spell-out rule in (9), which is formulated in such a way that it cannot affect non-nominal arguments, adjuncts, or reflexives.
(9) [KP +p, –a] ⇔ Ø
Note that the claim that pro-drop is zero spell-out of regular pronouns does not imply that overt and covert pronouns are identical in all respects. There is an obvious phonological difference. Therefore, where a pronoun cannot be destressed, it can also not be dropped. Focussed pronouns consequently resist omission, as do pronouns coordinated with some other category. What circumstances require phonological realisation of a pronominal is a matter of debate, but it is clear that under the present proposal contrasts between overt and covert pronouns must be attributed to pragmatic considerations. 2.4 The Elsewhere Condition The Elsewhere Condition was introduced into generative grammar by Kiparsky (1973), although it has a rich history predating the Chomskyan turn. It can be formulated as follows:
. We use the familiar features [+p(ronominal), –a(naphoric)] to indicate that KP is a pronoun. We are not committed to these particular features. What is important for us is that pronouns can be distinguished from other nominal categories, such as R-expressions and anaphors.
Case morphology and radical pro-drop
(10) Let R1 and R2 be competing rules that have D1 and D2 as their respective domains of application. If D1 is a proper subset of D2, then R1 blocks the application of R2 in D1.
The Elsewhere Condition has three consequences for the phonological realization of syntactic structure: (11) a.
All else being equal, a phonological realization of a category C takes priority over a phonological realization of the categories contained in C.
b. All else being equal, a phonological realization of a category C that spells out more of C’s features takes priority over phonological realization that spells out fewer features. c. Optionality results if the phonological realization of a category C spells out fewer of C’s features than the phonological realization of the categories contained in C.
The validity of (11a) can be demonstrated using English irregular verbs. The irregular form went, for example, blocks the regular from go-ed because it realizes a higher level category, and is consequently inserted by a more specific spell-out rule. The effects of (11b) can be observed in almost any inflectional paradigm. For example, German second person singular agreement could either be spelled out using the default third person singular ending or the first person singular ending. Both arguably are compatible with the feature content of the second person singular (see Harley & Ritter 2002). The fact that the second person singular form is used reveals a preference for spell-out rules that realize the maximum number of features present in syntax. The statement in (11c) must hold because the Elsewhere Condition cannot choose between rules whose structural descriptions overlap only partially.
3. Why radical pro-drop is sensitive to the morphology of pronouns We now return to the question at the heart of this paper: what determines the cross-linguistic distribution of radical pro-drop. As a result of the Elsewhere Condition, the general zero spell-out rule in (9) would be blocked by more specific spell-out rules that realize a KP with particular case and phi-features, such as (8). (We repeat these rules below.)
(8) [KP +p, –a, 3sg, masc, acc] ⇔ /hem/
(9) [KP +p, –a] ⇔ Ø
This means that in languages whose pronominal paradigm consists of spell-out rules for KP, a general pro-drop rule would not have any effect. Its application
Ad Neeleman & Kriszta Szendrői
would be systematically suppressed by the more specific spell-out rules that introduce overt pronouns. This does not mean that such languages necessarily lack pro-drop altogether. A context-sensitive spell-out rule could legitimately give rise to zero arguments. Consider a rule that mentions agreement (indicated by co-indexation with an element in the structural description of the rule):3 (12) [KP +p, –a, φi] ⇔ Ø / ___ [φi]
The rule in (12) is not in an Elsewhere relation with the rules that make up the (overt) pronominal paradigm. In order to see this, compare it with (8). The context-sensitive rule is more specific in one sense: it mentions agreement, while the context-free rule does not. On the other hand, a context-free rule like (8) is more specific in that it mentions particular phi-features, which (12) does not. This means that a spell-out rule for an overt pronoun and the context-sensitive prodrop rule are not in an Elsewhere relation: their domains of application are not in a subset-superset relation. Consequently, neither rule blocks the other: languages with fusional pronominal paradigms cannot have radical pro-drop, but they can have pro-drop in the context of rich agreement. The reason why radical pro-drop is blocked in languages with fusional pronominal paradigms is that the relevant spell-out rules all apply to the same category, KP: (13)
target of spell-out → rules for pronouns (8) and context-sensitive K pro-drop rule (12)
←
KP DP D
target of radical pro-drop rule (9)
NP N
...
In order for a general zero spell-out rule like (9) to have an effect, the language in question cannot have other spell-out rules for pronominal KPs. Consider what
. It is often claimed that the richer the agreement, the greater the likelihood of context-sensitive pro-drop (see Ackema et al., 2006 for an overview). In our proposal, this correlation must either be explained on functional grounds, or in terms of the theory of context-sensitive spell-out rules, or a combination of these. We cannot go into this issue here, and will simply take for granted the relevance of rich agreement for context-sensitive pro-drop.
Case morphology and radical pro-drop
happens if the rules that express the pronominal paradigm target lower-level categories, such as K and DP or NP: KP
(14) target of spell-out → rule for case
K
← target of radical pro-drop rule (9) DP
D
NP N
← possible target of spellout rules for pronouns ← possible target of spell-out rules for pronouns ...
An example of a language with this set-up for overt pronouns is Japanese. Recall that this language has independent pronominal stems and case markers. These are inserted by the rules in (15). Application of these rules generates forms like (16). (15)
[NP +p, –a, 1sg] ⇔ /watasi/ [NP +p, –a, 2sg] ⇔ /anata/ [NP +p, –a, 3sg, masc] ⇔ /kare/ [NP +p, –a, 3sg, fem] ⇔ /kanozyo/ [pl] ⇔ /tati/; /ra/
(16) kare-ra-ga he- pl-nom ‘they’
[K nom] ⇔ /ga/ [K acc] ⇔ /o/ [K dat] ⇔ /ni/ [K gen] ⇔ /no/
Japanese (simplified)
Japanese
Similarly, the nominative form of “I” is watasi-ga, the accusative form is watasi-o, etc. Clearly, the general zero spell-out rule in (9) does not stand in an Elsewhere relation with any of the rules in (15), which generate overt pronouns. The structural description of (9) is more specific in one sense, namely in that it spells out a larger chunk of structure than any of the rules in (15). On the other hand, the structural descriptions of the rules in (15) mention features that the zero spell-out rule is insensitive to, which makes them more specific. Hence, the structural description of the zero spell-out rule does not properly include those of the rules for overt pronouns; similarly, none of the structural descriptions of the rules for overt pronouns properly includes that of the zero spell-out rule. As a consequence, there will be no blocking effects between (9) and (15), and omission should be possible for all pronominal arguments. This is indeed a fair characterisation of the situation in Japanese, as was shown in (1) above. Another type of pronominal system that allows radical pro-drop is found in Chinese. In this language, pronouns have invariant forms in subject and object positions. Possessors are derived by adding the particle de. A possible analysis of this
Ad Neeleman & Kriszta Szendrői
situation would be to assume that case is not overtly realised, and that the spell-out rules for pronominal stems target a category lower than KP (for simplicity’s sake, let us say NP). The proposed rules, given in (17), generate pronouns like (18). (17) [NP +p, –a, 1sg] ⇔ /wˇo/ [NP +p, –a, 2sg] ⇔ /nˇı/ [NP +p, –a, 3sg] ⇔ /t/ (18) a.
[K …] ⇔ Ø [poss] ⇔ /de/ [pl] ⇔ /men/
w oˇ - men- Ø I- pl- case ‘we/us’
Chinese
Chinese
b. w oˇ - de I- poss ‘my/mine’
If analyzed along these lines, the situation in Chinese is formally indistinct from that in Japanese. Chinese should therefore allow radical pro-drop. This is, of course, correct (see (2)). For languages with an invariant pronominal paradigm, there is a second possibility. Their pronouns could in principle target KP. Hence, languages with invariant pronouns may or may not be radical pro-drop languages. This seems to be correct. Jamaican Creole, for example, has invariant pronouns and does not allow radical pro-drop: (19)
[KP +p, –a, 1sg] ⇔ /mi/ [KP +p, –a, 1pl] ⇔ /wi/ [KP +p, –a, 2sg] ⇔ /yu/ [KP +p, –a, 2pl] ⇔ /unu/ [KP +p, –a, 3sg] ⇔ /im/ [KP +p, –a, 3pl] ⇔ /dem/ [KP +p, –a, 3sg, neut] ⇔ /i/
(20) a. *(Mi) a rait. I am write ‘I’m writing.’
Jamaican Creole
Jamaican Creole
b. Nobadi neva sii *(im). nobody never see he ‘Nobody ever saw him.’ c.
Dem so feisty in *(dem) ways. they so feisty in they ways ‘They were so feisty in their ways.’
An important question is what properties of a language with invariant pronouns determine whether it has empty case markers (like Chinese), or “fusional” invariant pronouns (like Jamaican Creole). We cannot discuss this issue here due to lack of space, but a concrete proposal can be found in Neeleman & Szendrői (2005, 2007).
Case morphology and radical pro-drop
4. Typological range of the proposal The typological predictions of our proposal can be summarized as follows. First, in languages with pronouns that are fusional for case, like English, radical pro-drop should be impossible. This is because the spell-out rules corresponding to the overt pronominal forms will block the zero spell-out rule, as they realize more features. Second, in languages where case is expressed in a morpheme that is independent from the stem, as in Japanese, we predict that radical pro-drop is possible. The zero spell-out rule that gives rise to radical pro-drop is not blocked by the spell-out rules for overt pronouns, as the latter realize lower-level categories (say, NP rather than KP). Third, in languages whose pronouns are invariant for case there are two options. They may be analyzed as having zero case affixes (as suggested for Chinese), in which case radical pro-drop is possible. Alternatively, their pronouns can be analyzed as “fusional”, in which case radical pro-drop is blocked. In earlier work on the cross-linguistic distribution of radical pro-drop, we argued that it is easy to find examples of the four types of languages allowed by our theory, while the class of languages with fusional pronouns and radical pro-drop is empty. The languages we discussed are enumerated in (21); explicit paradigms and sources are given in Neeleman & Szendrői (2007), where we also discuss why some languages with invariant pronouns fall under (21c), while others fall under (21d). Here, we will simply concentrate on the main prediction of our proposal: the claim that there are no languages of type (21e); that is, languages that combine fusional pronouns with radical pro-drop. (21) a.
Fusional pronouns, no radical pro-drop Afrikaans, Dutch, English, Greek, Italian, Pashto, Swedish
b. Agglutinative for case, radical pro-drop Assamese, Burmese, Hindi/Urdu, Japanese, Korean, Turkish c.
Invariant for case, radical pro-drop Cheke Holo, Chinese, Kokota
d. Invariant for case, no radical pro-drop Jamaican Creole, Papiamentu, Tok Pisin e.
Fusional pronouns, radical pro-drop 〈empty〉
Note that the languages that have radical pro-drop come from different language families. Assamese and Hindi/Urdu are Indo-Aryan languages; Chinese and Burmese are Sino-Tibetan; Cheke Holo and Kokota are Oceanic (Eastern Malayo-Polynesian) and so on. At first sight, it seems that the languages that do not have radical prodrop belong to two classes: they are either creoles or non-Indic Indo-European.
Ad Neeleman & Kriszta Szendrői
This is an artifact of our sample, however. To begin with, Pashto does not fit very well with this classification (it is Iranian, Indo-European). Moreover, Haida, a native American language spoken in Alaska, has obligatory subject pronouns, like English. The paradigm in (22) shows that Haida pronouns are fusional (The strong pronouns are frequently accompanied by a focus marker, but given that focus markers are external to case, this is not sufficient to warrant pro-drop). (22) Haida pronouns: Agent (strong/weak)
Patient (strong/weak)
1sg
hláa/hl
díi
2sg
dáa/dáng
dáng
3sg
‘láa/hal
‘láa/hal
1pl
t’aláng
Íitl’
2pl
daláng
daláng
3pl
tl’áa/tl’
tl’áa/tl’
The sample of languages we have used so far to test our proposal is limited. However, given the nature of the claim we are defending, it is possible to use typological databases to enlarge the empirical grounding of the generalization central to our claim. Such databases might guide us in our endeavor to find languages that are potential counterexamples. Further analysis would then have to determine whether the languages in question are actual counterexamples. There is one database that seems particularly useful, namely the World Atlas of Language Structures (WALS; Haspelmath et al. 2005). The WALS is a large database of structural properties of languages gathered from a wide range of descriptive sources. It consists of maps with accompanying texts on diverse grammatical features. Each map shows between 120 and 1110 languages. Crucially for our present purposes, the WALS is accompanied by a searchable CD-ROM, allowing the user to combine different maps. The type of language that our proposal excludes has fusional case morphology on pronouns, but allows radical pro-drop. We can find languages that potentially fall into this group by combining two maps. The first identifies languages that allow subject omission in the absence of verb-subject agreement; the second identifies languages with independent subject pronouns whose plural is expressed by unanalysable person-number stems (at least in the 1st and 2nd person plural). For example, in Dogon (a Niger-Congo language spoken in Mali), we get the forms mi “I”, emme “we”, u “you.sg” and e “you.pl”. The combination of these two properties gives us languages that may have radical pro-drop (depending on the generality of argument omission) and whose pronouns may express case fusionally. The rationale behind the second part of this statement is that, as already observed by Greenberg (1963: 95), case morphology is generally located outside number morphology. Hence, if a language has agglutinating number morphology, it is unlikely
Case morphology and radical pro-drop
to have fusional case morphology. Conversely, in languages with fusional number morphology, there is some chance that we will find fusional case morphology as well.4 (Of course, it would be better to search for the way case is expressed on pronouns directly, but the WALS does not offer a map for that.) The combination of the two relevant maps yields the following languages: (23) Epena Pedee; Garo; Guugu Yimidhirr; Kayah Li (Eastern); Khmu’; Lezgian; Maybrat; Thai; Yidi; Yoruba.
We can start by removing from this list of potential counterexamples those languages that cannot be classified as allowing radical pro-drop. In Yoruba, for instance, subjects can only be omitted when a third person singular pronoun occurs before the negation marker kò or the future tense marker yó (Bamgbose 1967: 42). This constraint on subject omission implies that Yoruba does not have radical pro-drop. To the best of our knowledge, Kayah Li has not been studied extensively, but the information available to us does not suggest that it allows free omission of pronouns either (see Solnit 1986, 1997). We did not manage to amass sufficient information about Khmu’ to determine whether it is radical pro-drop or not, and will therefore have to put this language to one side. The remaining languages are given in (24).
. A reviewer brings to our attention an interesting case. Brazilian Portuguese is not a radical pro-drop language, although at least at first sight it seems to exhibit agglutinating number morphology: see the –s endings in the 2nd and 3rd person plural forms in (i). The idea that –s expresses number is an appealing one, since –s is the plural marker in the language in general. However, note that most of the accusative forms are not derivable from the nominative ones. So, the –s ending on the accusative forms, is in fact external to case. Given that number marking is almost always internal to case marking (Greenberg 1963: 95), we speculate that the –s ending is not a regular number marking morpheme, but rather that the pronominal forms are spelt-out the same irrespective of whether they are singular or plural, and an additional –s morpheme is added externally to case at PF in a position that does not correspond to the morphosyntactic position of number. This is similar to multiple exponence in Greek strong pronouns, where an additional support morpheme is added onto the case-marked forms at spell-out (see Neeleman & Szendrői 2007).
(i)
Brazilian Portuguese pronouns: 1sg 2sg 3sg msc 3sg fem 1pl 2pl 3pl msc 3pl fem
Nominative eu tu (NBr, NEBr) você/ cê (SBr) ele ela nós/a gente vocês/cês /(vós) eles elas
Accusative me te o/lo a/la nos/a gente vocês/cês /(vos) os/los as/las
Ad Neeleman & Kriszta Szendrői
(24) Epena Pedee; Garo; Guugu Yimidhirr; Lezgian; Maybrat; Thai; Yidi\.
According to our proposal, radical pro-drop is allowed in languages that have pronouns with agglutinative case morphology, and therefore such languages must be removed from the list as well. We already know that Burmese has agglutinative case markers (see (21b)). The same is true of Epena Pedee (see Harms 1994: 58), Garo (see Burling 2003), Lezgian (see Haspelmath 1993: 184) and Yidi\ (Dixon 1977: 168; Martin Haspelmath p.c.). To give some examples, we find forms like (25a) in Epena Pedee (from Harms 1994: 186) and forms like (25b) in Garo (from Burling 2003). (25) a.
mi-a 1sg-erg
b. ching-ko 2sg-acc
We are not confident of the correct analysis of Guugu Yimidhirr pronouns. Guugu Yimidhirr is an Australian language, and so, as expected, it has a large set of agglutinating case markers. Dixon & Blake (1979: 6) state that the Australian languages “seldom exhibit morphophonemic alternations that obscure the agglutinative character of words; there is little fusion of any kind and little suppletion (as in English I/me or go/went for instance)”. This would suggest that Guugu Yimidhirr is no different from Epena Pedee or Garo. But there are some complications in the Guugu Yimidhirr pronominal paradigm. A good overview of the various pronominal forms is given in Haviland (1979: 65–66). In the table below we give a simplified version of the paradigm, abstracting away from regional variation, amongst several other things. (26) Guugu Yimidhirr pronouns (simplified): dat
purp
abes
ades
1sg
ngayu nganh-i
nom
ngadh-u
ngadh-un-ngu
ngadh-un-ga
ngadh-un-gal
2sg
nyundu nhina-in
nhan-u
nhan-un-ngu
nhan-un-ga
nhan-un-gal
3sg
nyulu
nhinhaan-in nhang-u
nhang-un-ngu
nhang-un-ga
nhang-un-gal
ngali-in
ngali-in-ngu
ngali-in-ga
ngali-in-gal
1du in ngali
acc
1du ex ngaliinh ngali-un
ngali-in
ngalinh-un ngalinh-un-ngu ngalinh-un-ga ngalinh-un-gal
1pl
ngana ngali-an
nganang-an nganang-an-ngu nganang-an-ga nganang-an-gal
2du
yubaal yubal-in
yubal-in
yubal-in-ngu
yubal-in-ga
yubal-in-gal
2pl
yurra
yurra-an
yurra-an
yurra-an-ngu
yurra-an-ga
yurra-an-gal
3du
bula
bulang-an
bulang-an
bulang-an-ngu bulang-an-ga bulang-an-gal
3pl
dhana dhana-an
dhana-an
dhana-an-ngu
dhana-an-ga
dhana-an-gal
Case morphology and radical pro-drop
As (26) shows, all non-singular pronouns exhibit perfectly regular non-cumulative agglutinating morphology for all cases: nominative is the unmarked form, there is an accusative/dative form marked by -in/-an/-un, and three oblique cases derived from the accusative/dative by attachment of -ngu, -ga and -gal, respectively. The singular forms are more complex. To begin with, accusative and dative forms are distinguished: the former are marked by -i/-in; the latter by -u. The oblique cases are derived from an allomorph of the dative forms (ending in -un, rather than -u). Again, from the dative onward, the singular paradigm shows perfectly regular non-cumulative agglutinating morphology. The nominative and accusative singular forms are puzzling, though, and in order to extend the analysis of Guugu Yimidhirr pronouns as agglutinating, one must assume that singular stems have three allomorphs: a nominative, an accusative and a dative one. It seems to us that this would be a reasonable assumption for the child to make, given that the paradigm is overwhelmingly regular. But needless to say, more work would be required to determine whether this idealized description of Guugu Yimidhirr pronouns is correct. If we are allowed to remove Guugu Yimidhirr from the list in (24) on the basis of the above considerations, two languages remain: (27) Maybrat; Thai.
Maybrat is arguably a radical pro drop language. Dol (1999: 160) states that subjects and objects may “be omitted if they have been mentioned in earlier in the discourse.” The Maybrat pronominal paradigm is given in (28). It is clear that the possessive forms are formed agglutinatively, while pronouns are otherwise invariant for case. Radical pro-drop in a language with a paradigm of this type is consistent with our hypothesis. (28) Maybrat pronouns Unmarked
Possessive
1sg
tuo
ro-tuo/a-tuo
2sg
nuo
ro-nuo/a-nuo
3sg
au
r-au
3sg msc
ait
r-ait
1pl
amu
r-amu
2pl
anu
r-anu
3pl
ana
r-ana
Thai is also a radical pro-drop language whose pronouns do not express case (see Cooke 1968 and Campbell 1969). So Thai, like Maybrat, does not constitute a
Ad Neeleman & Kriszta Szendrői
counterexample to the generalization that radical pro-drop is restricted to languages whose pronouns are either agglutinating or invariant for case. 5. Concluding remarks To summarize, we argued that the cross-linguistic distribution of radical pro-drop can be captured if three assumptions are made: (i) pro-drop results from zero spellout of regular pronouns; (ii) spell-out rules may mention non-terminal categories as their input; and (iii) spell-out is regulated by the Elsewhere Condition. From these assumptions, it follows that no language with fusional pronominal morphology may have radical pro-drop. In earlier work, we checked this prediction using a small sample of languages. Here, we extended the empirical domain of the proposal by exploring languages identified as potential counterexamples by the WALS. As it turns out, the typology we established earlier is consistent with what appears to be the case in these additional languages. We can identify four groups of languages, with the fifth group (fusional pronoun morphology and radical pro-drop) empty, as predicted: (29) a. Fusional pronouns, no radical pro-drop Afrikaans, Dutch, English, Greek, Haida, Italian, Kayah Li, Pashto, Swedish, Yoruba b. Agglutinative for case, radical pro-drop Assamese, Burmese, Epena Padee, Garo, Guugu Yimidhirr, Hindi/Urdu, Japanese, Korean, Lezgian, Turkish, Yidi\ c.
Invariant for case, radical pro-drop Cheke Holo, Chinese, Kokota, Maybrat, Thai
d. Invariant for case, no radical pro-drop Jamaican Creole, Papiamentu, Tok Pisin e.
Fusional pronouns, radical pro-drop 〈empty〉
As mentioned, one question that remains is what factor determines whether a language with invariant pronouns belongs to class (29c) or to class (29d). This issue is addressed in Neeleman & Szendrői (2007), which also provides a more in-depth study of Maybrat and Thai.
References Abney, S. 1987. The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT. Ackema, P., Brandt, P., Schoorlemmer, M. & Weerman, F. 2006. Agreement and Argument Structure. Oxford: OUP.
Case morphology and radical pro-drop
Bamgbose, A. 1967. A Short Yoruba Grammar. London: Heinemann. Bhat, D.N.S. 2004. Pronouns. Oxford: OUP. Bittner, M. & Hale, K. 1996. The structural determination of case and agreement. Linguistic Inquiry 27: 1–68. Burling, R. 2003. Garo. In The Sino-Tibetan Languages, G. Thurgood & R. LaPolla (Eds), 387–400. London: Curzon Press. Campbell, R.N. 1969. Noun Substitutes in Modern Thai. A Study in Pronominality. The Hague: Mouton. Cooke, J.R. 1968. Pronominal Reference in Thai, Burmese, and Vietnamese. Berkeley CA: University of California Press. Dixon, R. 1977. A grammar of Yidi\. Cambridge: CUP. Dixon, R.M.W. & Blake, B. (Eds). 1979. Handbook of Australian Languages, Vol. 1. Amsterdam: John Benjamins. Dol, P. 1999. A Grammar of Maybrat: A Language of the Bird’s Head, Irian Jaya. Leiden: Grafisch Bedrijf. Greenberg, J. 1963. Some universals of grammar, with particular reference to the order of meaningful elements. In Universals of Language, J. Greenberg, (Ed.), 73–113. Cambridge MA: The MIT Press. Halle, M. & Marantz, A. 1993. Distributed morphology and the pieces of inflection. In The View from Building 20, K. Hale & S.J. Keyser (Eds), 111–176. Cambridge MA: The MIT Press. Harley, H. & Ritter, E. 2002. Person and number in pronouns: A feature-geometric analysis. Language 78: 482–526. Harms, P. 1994. Epena Pedee Syntax: Studies in the languages of Colombia 4. Dallas TX: Summer Institute of Linguistics and the University of Texas at Arlington. Haspelmath, M. 1993. A Grammar of Lezgian. Berlin: Mouton de Gruyter. Haspelmath, M., Dryer, M., Gil, D. & Comrie, B. (Eds). 2005. The World Atlas of Language Structures. Oxford: OUP. Haviland, J. 1979. Guugu Yimidhirr. In Handbook of Australian Languages, Vol. 1, R.M.W. Dixon & B. Blake (Eds), 27–180. Amsterdam: John Benjamins. Holmberg, A. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36(4): 533–564. Huang, J.C.T. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531– 574. Jackendoff, R. 1997. The Architecture of the Language Faculty. Cambridge MA: The MIT Press. Kiparsky, P. 1973. ‘Elsewhere’ in phonology. In A Festshrift for Morris Halle, S. Anderson & P. Kiparsky (Eds), 93–106. New York NY: Holt, Rinehart and Winston. Neeleman, A. & Szendrői, K. 2005. Pro-drop and pronouns. In Proceedings of the WCCFL 24, J. Alderete (Eds), 299–307. Somerville MA: Cascadilla Proceedings Project. Neeleman, A. & Szendrői, K. 2007. Radical pro-drop and the morphology of pronouns. Linguistic Inquiry 38: 671–714. Neeleman, A. & Weerman, F. 1999. Flexible Syntax; A Theory of Case and Arguments. Dordrecht: Kluwer. Pannemann, M. & Weerman, F. 2005. Variation in acquisition as fine-tuning of correspondence rules. Paper presented at TIN-dag, Utrecht University. Perlmutter, D. 1971. Deep and Surface Constraints in Generative Grammar. New York NY: Holt, Rinehart and Winston.
Ad Neeleman & Kriszta Szendrői Postal, P. 1966. On so-called pronouns in English. In Report of the 17th Annual Round Table Meeting on Languages and Linguistics, F. Dineen (Ed.), 177–206. Washington DC: Georgetown University Press. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–57. Solnit, D. 1986. A Grammatical Sketch of Eastern Kayah (Red Karen). Ph.D. dissertation, University of California. Solnit, D. 1997. Eastern Kayah Li: Grammar, Texts, Glossary. Honolulu HI: University of Hawaii Press. Speas, M. 2004. Economy, Agreement and the Representation of Null Arguments. MS, UMass, Amherst. Sproat, R. 1985. On Deriving the Lexicon. Ph.D. dissertation, MIT. Taraldsen, K.T. 1978. On the NIC, Vacuous application and the that-trace filter. MS, MIT, circulated by IULC. Tomioka, S. 2003. The semantics of Japanese null pronouns and its cross-linguistic implications. In The Interfaces: Deriving and Interpreting Omitted Structures, K. Schwabe & S. Winkler (Eds), 321–339. Amsterdam: John Benjamins. Weerman, F. & Evers-Vermeul, J. 2002. Pronouns and case. Lingua 112: 301–338.
part iii
Parametric clustering
Topic prominence and null subjects1 Marcello Modesto This chapter presents data from Brazilian Portuguese, Finnish and Chinese which show that, at least in these languages, phonologically null subjects may be licensed and identified without the participation of verbal agreement. The analysis proposed treats those cases of null subjects as elided topics (PF-deleted elements) and so it relates the presence of null subjects in these languages to a parameter involving topic prominence. Some implications of the analysis are then discussed, based on Brazilian Portuguese data. Keywords: Null Subject Parameter; Topic Prominence; Finnish; Brazilian Portuguese; Chinese
1. Introduction Several authors (notably Rizzi 1982, 1986) have related a cluster of properties to a parameter, which came to be known as the Null Subject Parameter (NSP). These properties include allowing the subject position of finite clauses to be null whether they have a non-referential or a definite pronominal (referential) interpretation; disallowing overt non-referential (expletive) pronouns; allowing free subject inversion; allowing wh-extraction of an embedded subject across an overt complementizer and from embedded questions; and, most importantly, presenting “rich”2
. I am very thankful to Jim Huang, Audrey Li, Zoe Wu, Angela Bartens, Arja Häkkinen, Auli Hakulinen, Helena Halmari, Paivi Koskinen, Minna Niskanen, Anne Vainikka and Maria Vilkuna, for their native judgments and to Anders Holmberg for all his help while writing this paper. I am especially indebted to Hannu Reime for answering all my questions about Finnish. The comments of Jairo Nunes and two anonymous reviewers also helped me to improve this article. Obviously, all errors remaining are my own. . “Rich” appears in scare quotes because the term is used here without a proper definition. As noted by Speas (1994), the property that makes agreement “rich” is difficult to pin down. Most researchers use the term “rich” to mean “bearing enough morphology to provide non-ambiguous information on the person and number (and maybe gender) of the subject.” However, this raises the question of how rich the inflection must be to license null arguments. See Jaeggli & Safir
Marcello Modesto
subject-verb agreement marking. The presence of “rich” agreement has probably been the most discussed of these properties. The reason is that there is irrefutable evidence from several languages that agreement is indeed related to the presence of null subjects (in fact, of null arguments in general), and that it is responsible for allowing them. However, some languages, e.g., Chinese, present null subjects even though there is no agreement. A great deal of effort has been put into explaining how these languages fit into the NSP picture, but numerous problems remain. Another complicating factor is that there are languages which do not comply with the NSP as described by Rizzi. Brazilian Portuguese (BP) presents an agreement paradigm that should most certainly be regarded as “poor”: 2nd and 3rd persons are identical both in the singular and in the plural in all tenses; therefore it is not surprising that postverbal subjects are ungrammatical, except with unaccusative verbs. Null subjects are, however, allowed in BP. Similarly, Finnish has overt non-referential (expletive) pronouns, and yet it allows null referential subjects (cf. Holmberg 2005). There are three ways to go about the problem: either one gives up the idea that there is a parameter setting that involves more than just the presence or absence of null subjects; or one tries to accommodate the not-so-well-behaved languages into the NSP somehow; or, finally, one explains the presence of null subjects in those languages by other means. The first possibility is not interesting and will consequently not be pursued here. The second option was taken by many authors during the 1980s (see especially Huang 1984), with relative success. The third possibility has not, to my knowledge, been explored and it will be this one that I will defend in this paper. Although I will not try to precisely define here what “rich” agreement is, I will assume the intuitive idea that, for agreement to allow null pronominal subjects, it must furnish (at least some) unambiguous information about number and person (and possibly gender in some languages). This excludes languages like Chinese and BP from the group of Null Subject Languages (NSLs) and their noncompliance with the cluster of properties associated with the NSP is therefore straightforwardly explained. The presence of null subjects in these languages, on the other hand, will clearly have to be related to another parameter. Here I will propose that the relevant parameter is the one distinguishing subject-prominent and topic-prominent languages, as discussed in Tsao (1977), Li & Thompson (1976) (and references therein). It will be argued that topic-prominent languages are those in which the EPP is double-faced: not only do clauses need to have a
(1989), Speas (1994), Rohrbacher (1994) and Vikner (1997) for tentative attempts at making the notion of “richness” more precise.
Topic prominence and null subjects
subject in these languages, but they also need to have a “topic”. Matrix topics may then bind ϕPs, in the sense of Holmberg (2005), in the embedded subject position, accounting for the availability of referential null subjects.3
2. Delimiting the problem Huang (1984) showed that there is an important asymmetry between the interpretation of overt pronouns in English, for instance, and an empty category (EC) in languages like Chinese when the pronoun or EC occupies the object position of complement clauses. Compare (1) with (2) (Huang’s (20) and (19) respectively): (1)
a. b. c. d.
He came. Bill saw him. John said that he knew Bill. John said that Bill knew him.
(2) a. e lai-le. come-perf ‘He came.’ b. Lisi hen xihuan e. Lisi very like ‘Lisi likes him a lot.’ c.
Zhangsan shuo [e bu renshi Lisi]. Zhangsan say not know Lisi ‘Zhangsan said he did not know Lisi.’
d. Zhangsan shuo [Lisi bu renshi e]. Zhangsan say Lisi not know ‘Zhangsan said Lisi did not know him.’
There is a striking similarity in distribution and referential possibilities between overt pronouns in English and ECs in Chinese. However, when we turn to (1d) and (2d), an important difference emerges. In (1d), the pronoun is free to refer to the matrix subject John or to someone else. In (2d), on the other hand, the EC object may only refer to someone whose reference is fixed outside of the entire sentence. Importantly, it may not refer to the matrix subject Zhangsan. In order
. An analysis of the BP facts discussed in the text below was put forth in Modesto (2000a,b). This chapter presents a slight revision, including Finnish data and modifying the analysis so that it becomes more compatible with the Minimalist Program (cf. Chomsky 1995, 2000, 2001, 2004, 2005, 2008).
Marcello Modesto
to construct a sentence in Chinese with the same interpretative possibilities as the English counterpart, the EC would have to be replaced by an overt pronoun. Huang explains this difference by taking the EC in (2d) to be a variable bound by a (null) discourse topic. Since it is not identified by agreement, it cannot corefer (with the matrix subject, for instance); it cannot be used deictically, or in an out-ofthe-blue context. The same is true, according to Huang, of the ECs in (2a) and (2b), which are also bound by (null) discourse topics. Since empty categories are not contextually defined, according to recent generative theory, it can be assumed that the null arguments in (2a,b,d) are ϕPs in the sense of Holmberg (2005): a bundle of valued ϕ-features which is not referential by itself and needs to be bound in order to be interpreted, and which is bound by a discourse topic that may itself be null. Since the chain formed by the null topic and the ϕP can be characterized as an A’-chain, the null argument is interpreted at LF as a variable. What about the EC in (2c)? Since it is not identified by agreement, it should not be able to corefer with the matrix subject. Huang argues that the EC is a pronominal which is controlled by the closest nominal element (or simply in its control domain if it has one, as in Huang 1989). Huang therefore solves the problem Chinese poses to the NSP in the following manner: since there is no verbal agreement, null pronominals are not usually allowed, as expected. Where they do occur, however, null arguments are, in fact, variables bound by (null or overt) topics. Null pronominal subjects may perhaps be licensed by the fact that Chinese uniformly lacks agreement, as suggested by Jaeggli & Safir (1989)), but they cannot be identified (by agreement, as subjects of this type are in languages like Italian); therefore they are possible only when controlled by a higher argument. The correlation between pro-drop and richness of agreement is then maintained. Taking the null subject in (2c) to be controlled is, however, problematic; and Huang’s Generalized Control Rule (GCR) cannot account for the whole range of facts. As seen in (3b), the embedded null subject in Chinese does not behave like a normal pronoun in English, nor like a controlled subject:
(3) a.
Bill1 told Joe2 that he1/2 had won the championship.
b. Zhangsan1 gaosu Lisi2 e1/*2 ying le guanjun. Zhangsan tell Lisi win perf championship ‘Zhangsan told Lisi he had won the championship.’ c.
Bill1 told Joe2 e*1/2 to win the championship.
(3a) may be interpreted as saying that Bill was the one giving the good news to Joe that Joe had won the championship. This same reading is absent from (3b). The fact that the EC in (3b) can only take the matrix subject as its antecedent casts serious doubts that ECs in subject position in Chinese find their referent
Topic prominence and null subjects
according to Huang’s GCR. If Control were really involved, one would expect exactly the opposite to take place, i.e., that the EC would take the matrix object as its antecedent (as in (3c)).4 In that case, null embedded subjects in Chinese still present a problem for theories which try to explain the cluster of properties associated with the NSP. In what follows, it will be argued that the peculiar behavior of null embedded subjects in Chinese follows from the fact that all subjects in that language occupy a “topic” position (to be defined more clearly below); therefore a matrix topic/ subject may bind a ϕP in embedded subject position. In this way, all null arguments in Chinese will be treated in the same manner: as ϕPs bound by a (null or overt) topic. BP and Finnish null embedded subjects will be explained in the same manner. Section 3 below gives an overview of the interpretative characteristics of null subjects in BP, Finnish and Chinese. These characteristics are then accounted for in section 4. Section 5 discusses what properties may be involved with the Topic Prominence Parameter and section 6 offers some concluding remarks.
. The preceding statements deserve some clarification. That null subjects in Chinese are subject-oriented (unless interpreted as bound by a discourse topic) was also claimed by Battistella (1985) and this was confirmed by several informants. James Huang (p.c.) agrees that the EC in (3b) takes the matrix subject as its antecedent. However, due to contrasts of the following kind (already mentioned by Battistella), he maintained his skepticism towards the subjectorientation theory (cf. Huang 1989: 21ff): (i) a. Zhangsan1 gaosu Lisi2 e1/*2 bu neng lai. Zhangsan told Lisi not can come ‘Zhangsan told Lisi that he can’t come.’ b.
Zhangsan1 gaosu Lisi2 e*1/2 bu keyi lai. Zhangsan told Lisi not may come ‘Zhangsan told Lisi he may not come.’
I will follow Battistella in assuming that contrasts of the kind shown by the sentences in (i) are explained by pragmatics. In fact, Huang (1984: ff. 6) himself stressed the point that, in languages like Chinese, “pragmatics appears to “override” grammar”. Trying to explain how the EC in object position in (ii) below could be referentially dependent on the matrix subject, Huang claimed that, in that sentence, “the object EC can refer to the matrix subject “thief ”, since this is pragmatically the most natural way to interpret the sentence.” (ii)
Xiaotou yiwei meiyou ren kanjian e, na le dongxi jiu pao. thief think no man see take perf thing then run ‘The thief thought no one saw him so he took the things and ran.’
Whatever the explanation for why sentences with modals behave differently, subject orientation is a factor in sentences which do not contain modals, exemplified by (3b) in the text, and this fact calls for an explanation.
Marcello Modesto
3. The interpretation of null embedded subjects in BP, Finnish and Chinese Languages not related to Chinese, such as BP and Finnish, present similar facts to those discussed in section 2. Null embedded subjects are allowed when they apparently should not be. Agreement in BP has become “poor”, which has caused the subject position generally to be preferably filled by overt pronouns in spoken language (cf. Galves 1993; Duarte 1995; Figueiredo Silva 1996). Null subjects are only allowed when the discourse provides a topic so they can be analyzed in the same way as in Chinese. Embedded null subjects, however, are possible in out-of-the-blue contexts, but, in that case, they take the subject of the next clause up as their antecedent; they cannot be interpreted deictically nor as corefererent with the matrix object (cf. Modesto 2000a,b): (4) a.
O Feco1 convenceu a Dani2 que e1/*2/*3 ganhou o campeonato.5 the Feco convinced the Dani that won the championship ‘Feco convinced Dani that he won championship.’
b. O Feco1 avisou a Dani2 que e1/*2/*3 vai ganhar. the Feco warned the Dani that will win-inf ‘Feco warned Dani that he will win.’
In Finnish, null subjects are only allowed with 1st and 2nd person interpretation despite agreement in the language being “rich”. 3rd person null subjects are not allowed, at least not in matrix contexts. However, as noted by Holmberg (2005), “the use of 1st and 2nd person null subjects is largely restricted to formal varieties of Finnish, including standard written Finnish” (cf. also Vainikka 1989; Vainikka & Levy 1999; Heinonen 1995). He also notes that “in written Finnish the use of the expletive sitä, which is common in spoken language, is proscribed …” So it seems
. Note that convencer is an object control verb, as shown below in (i). That makes even clearer that control is not involved in the identification of the EC in (4a).
(i) O Feco1 convenceu a Dani2 a PRO*1/2 fazer o teste. the Feco convinced to Dani to take-inf the test ‘Feco convinced Dani to take the test.’ Throughout the text, I use the verbs convencer (to convince) and avisar (to warn) as examples, but all the verbs in BP which take a nominal object followed by a sentential argument (including falar (to say), dizer (to tell), informar (to inform), gritar (to yell), assegurar (to assure), among others) behave in exactly the same way. Rodrigues (2004) tried to attribute the facts in (4a) to a peculiarity of the verb convencer, which cannot be correct, as extensively argued in Modesto (2007).
Topic prominence and null subjects
that, although Standard Finnish is a (partial) pro-drop language, colloquial Finnish has become a non-null subject language, like BP. Unexpectedly, however, even colloquial Finnish allows 3rd person null subjects in embedded clauses (in out-of-the-blue contexts). Once again, these embedded null subjects cannot be interpreted deictically nor as corefererent with the matrix object (cf. Gutman 2004; Halmari 1996): (5) a. Liisa1 vakuutti Jussille2, että e1/*2/*3 voi tulla valituksi. Liisa assured Jussi that can come elected ‘Liisa assured Jussi that she can be elected.’ b. Liisa1 takasi Jussille2, että e1/*2/*3 saa ylennyksen. Liisa guaranteed Jussi that may promotion ‘Liisa guaranteed to Jussi that she will get the promotion.’
Besides subject-orientation, null embedded subjects in BP, Finnish and Chinese exhibit a peculiar behavior with respect to their possible interpretations. As noted by Negrão (1997), null subjects in BP can only take c-commanding phrases as their antecedent. Sentences (6b), taken from Vainikka & Levy (1999), and (6c) indicate that the same is true in Finnish and Chinese:6 . An anonymous reviewer notes that Holmberg (2005) gives examples where the subject orientation and/or the c-command requirement seem(s) to be violated in Finnish. According to Holmberg, the allative argument in (i) is a possible antecedent for the null pronoun despite not being a subject; and according to both Holmberg and the reviewer, Tarjalle in (ii) is a possible antecedent, although it does not c-command the null subject: (i) Anu1 sanoi Jarille2 että hän1/2/3/e1/2/*3 ottaa kitaran mukaan. Anu said Jari-all that he takes guitar along ‘Anu said to Jari that he (= Anu or Jari) will take the guitar along’ (ii)
Se oli Tarjalle1 pettymys ettei hän1/2/e1/*2 it was Tarja-all disappointment that.not she
koulussa. school-ine
saanut lukea latinaa could study Latin
‘It was a disappointment to Tarja that she could not study Latin at school’ Both examples pose a problem for the analysis to be presented and will have to be accounted for by some different mechanism. It is interesting to note that both examples would be translated using a subjunctive in Romance languages. Although this is not a definite solution, it is my belief that (ii) is acceptable because the embedded clause may contain a hidden subjunctive. As shown by Landau (2004), subjunctives in Hebrew and Balkan languages, though finite, admit controlled (PRO) subjects. In that case, Finnish may have an additional source for null subjects besides the one argued for in this work. Although Finnish has no morphological subjunctive, the conditional mood is sometimes used where Romance would use a subjunctive. Although the indicative may not be replaced by the conditional in (ii), it may be in (i) and that change makes it easier for speakers to take the allative argument as the antecedent of the null subject:
Marcello Modesto
(6) a. [O amigo do Feco2]1 disse que e1/*2/*3 ganhou a competição. the friend of Feco said that won the championship ‘Feco’s friend said that he won the championship.’ b. [Veljeni2 vaimo]1 oli niin iloinen, ettei e1/*2/*3 voinut nukkua. brother-gen wife was so happy that.not could sleep ‘My brother’s wife was so happy that she could not sleep.’ c. [Zhangsan2 fangwen de ren]1 xiwang e1/*2/*3 neg ying. Zhangsan visit DE person hope can win ‘The person who Zhangsan visited hoped he could win.’
There is a locality restriction involving the antecedent of a null subject in those languages: the closest subject must be the antecedent. (7) a. O Feco1 disse que a Dani2 acha que e*1/2 ganhou na loto. the Feco said that the Dani thinks that won the lottery ‘Feco said that Dani thinks that she won the lottery.’
(iii)
Anu1 sanoi Jarille2 että e2 ottaisi kitaran mukaan. Anu said Jari-all that take-con guitar along ‘Anu told Jari to bring the guitar along.’
The same contrast can be seen in (iv) below (Anders Holmberg, p.c.). The null subject in the (a) sentence, which has a verb in the indicative, can only be interpreted as taking the higher subject (Pekka) as its antecedent. In the (b) sentence, with a conditional, however, both Pekka and Tarja can be the antecedent of the null subject: (iv) a. Pekka1 sanoi Tarjalle2 ettei e1/*2 puhu liian hiljaa. Pekka said Tarja-all that-not speak-ind too low ‘Pekka told Taria that he does not speak too low.’ b. Pekka1 sanoi Tarjalle2 ettei e1/2 puhuisi liian hiljaa. Pekka said Tarja-all that-not speak-con too low ‘Pekka told Tarja that s/he should not speak too low.’ It is possible, then, to imagine that the speakers who accept Jari as the antecedent in (i) are taking the embedded clause to be a subjunctive, despite the lack of morphological evidence. In fact, even for BP, something additional would have to be said for subjects of subjunctive clauses (probably that they involve a case of (finite) control). They do behave differently than subjects of clauses in the indicative. As seen in (v), for instance, the object is (necessarily) the controller of the null embedded subject: (v)
O rei1 ordenou ao soldado2 que e*1/2 matasse o prisioneiro. the king ordered to.the soldier that kill-subj the prisoner ‘The king ordered the soldier to kill the prisoner.’
Topic prominence and null subjects
b. Jukka1 sanoi että Liisa2 ajattelee että e*1/2 oli voittanut arpajaisissa. Jukka said that Liisa thinks that had won lottery ‘Jukka said that Liisa thinks that she won the lottery.’ c.
Zhangsan1 yiwei Lisi2 chengren yiqian e*1/2 zuocuo shi le. Zhangsan think Lisi admit before do.wrong matter asp ‘Zhangsan thinks Lisi admitted that he did wrong.’
As noted by Figueiredo Silva (1996), a null subject cannot take split antecedents in BP. Note that (8a) is perfectly acceptable (in the intended reading) in European Portuguese, a language which is relatively uncontroversially a NSL. The BP sentence, however, is grammatical only on a reading where the null embedded subject is “arbitrary” in the sense of Egerland (2003). Once again, the data can be reproduced in Finnish7 and Chinese: (8) a. *O Feco1 disse pra Dani2 que e1+2 fizeram besteira. the Feco said to.the Dani that did something-wrong ‘Feco told Dani that they did something wrong.’ b. *Jukka1 kysyi vaimoltaan2 e1+2 voivatko mennä Espanjaan lomalle. Jukka asked his.wife can go to Spain for.vacation ‘Jukka asked his wife if they could go to Spain for vacations.’ c. *Zhangsan gaosu Lisi (shuo) e1+2 zuocuo shi le. Zhangsan told Lisi (that) do.wrong matter perf ‘Zhangsan told Lisi that they did something wrong.’
Negrão (1999) and Negrão & Viotti (2000) show that in BP, under VP ellipsis, only a sloppy reading is available for an embedded null subject. With only-NP antecedents, the null embedded subject receives a covariant interpretation, i.e., the invariant interpretation, which is possible when the subject is overt, is absent (on the covariant versus invariant interpretations see Higginbotham 1992; see also Duguine, this volume). According to Rodrigues (2004), the same is true in Finnish in both cases:8 . Sentences (7b) and (8b) are taken from Rodrigues (2004). . Chinese informants were not consistent with respect to VP-ellipsis and sentences with only-NP antecedents. Although speakers perceived only a sloppy reading in VP-ellipsis structures like (ia) below, (ib) was said to be ambiguous. Sentences with only-NP antecedents were also judged to be ambiguous. As suggested by Jairo Nunes (p.c.), Chinese may employ deep anaphora (cf. Hankamer & Sag 1976) in these cases. (i) a. Zhangsan shuo e mei shijian, Lisi ye zheyang shuo. Zhangsan say have-no time Lisi also this-way say ‘Zhangsan said he has no time, Lisi did too.’
Marcello Modesto
(9) a.
O Pedro1 acha que e1 é inteligente e o Paulo também. the Pedro thinks that is intelligent and the Paulo too ‘Pedro thinks that he is intelligent and Paulo does too.’
b. Só o Maluf acha que e vai ganhar as eleições. only the Maluf thinks that will win the elections ‘Only Maluf thinks that he will win the elections.’ (10) a. Jukka1 sanoi että e1 oli voittanut arpajaisissa, ja niin Pekkakin. Jukka said that had won lottery and so Pekka.also ‘Jukka said that he won the lottery and Pekka did too.’ b. Vain Jukka1 ajatteli että e1 oli voittanut arpajaisissa. only Jukka thought that had won lottery ‘Only Jukka thought that he won the lottery.’
The facts listed so far clearly show that the relation between the null embedded subject and its antecedent in BP, Finnish and Chinese involves some kind of variable-binding relation. More evidence pointing towards that conclusion is given by the fact that movement may alter the possible antecedent relations. As seen in (5) above, a matrix object cannot be taken to be the antecedent of a null embedded subject in BP. However, if the matrix object is wh-moved, relativized, clefted or topicalized, then it becomes a possible antecedent.9 In fact, when there is wh-movement, relativization or clefting of the object, it becomes the only possible antecedent for the null subject:10
b. Zhangsan shuo e yiding yingde xuanju, Lisi ye shi. Zhangsan say definitely win election Lisi also be ‘Zhangsan said he would win the election, Lisi also did.’ . Passivization also changes the interpretative possibilities (imp signifies imperfect past tense): (i) a.
O Feco1 viu a Dani2 quando e1/*2 chegava em casa. The Feco saw the Dani when arrive-imp in home. ‘Feco saw Dani when he was arriving home.’
b.
A Dani2 foi vista pelo Feco1 quando e*1/2 chegava em casa. The Dani was seen by Feco when arrive-imp in home ‘Dani was seen by Feco when she was arriving home.’
. The fact that there is a relation between movement and being the antecedent of a null subject in BP was already observed by Modesto (2000a,b). I have tried to use pragmatically neutral examples, since, as noted in footnote 4 above, knowledge of the world may interfere with the readings. As pointed out by Jairo Nunes (p.c.), for instance, the subject in a sentence like (i) below can still be interpreted as the antecedent of the null embedded pronoun, although the object has wh-moved. That reflects the fact that people generally assume students to miss classes more often than teachers do.
Topic prominence and null subjects
(11) a.
Quem2 que o Feco1 avisou t2 que e*1/2 precisa trabalhar até who that the Feco warned that needs work-inf until
mais tarde? more late
‘Who did Feco warn that s/he has to work till late?’
b. O cara2 que o Feco1 avisou t2 que e*1/2 precisa trabalhar mais the guy that the Feco warned that needs work-inf more
já chegou. already arrived
‘The guy who Feco warned that he has to work harder has already arrived.’
c.
Foi a Dani2 que o Feco1 avisou t2 que e*1/2 precisa trabalhar was the Dani that the Feco warned that needs work-inf
até mais tarde. until more late
‘It was Dani that Feco warned that she has to work till late.’
d. A Dani2, o Feco1 avisou t2 que e1/2 precisa trabalhar até the Dani the Feco warned that needs work-inf until
mais tarde. more late
‘(Speaking of) Dani, Feco warned her that s/he has to work till late.’
(12) a.
Quem2 que o Feco1 convenceu t2 que e*1/2 pode se eleger? who that the Feco convinced that can refl elect-inf ‘Who did Feco convince that s/he can get elected?’
b. O cara2 que o Feco1 convenceu t2 que e*1/2 pode se eleger the guy that the Feco convinced that can refl elect-inf
já chegou. already arrived
‘The guy who Feco convinced that he can get elected is already here.’
c.
Foi a Dani2 que o Feco1 convenceu t2 que e*1/2 pode se eleger. was the Dani that the Feco convinced that can refl elect-inf ‘It was Dani that Feco convinced that she can get elected.’
(i) Que professor2 o aluno1 avisou t2 que e1/2 ia faltar a aula? which professor the student warned that was-going miss-inf the class ‘Which professor did the student warn that he was going to miss the class?’
Marcello Modesto
d. A Dani2, o Feco1 convenceu t2 que e1/2 pode se eleger. the Dani the Feco convinced that can refl elect-inf ‘(Speaking of) Dani, Feco convinced her that s/he can get elected.’
In (11a,b,c) and (12a,b,c), the only possible antecedent for the null embedded subject is the moved matrix object. To get a reading where the matrix subject is the antecedent, an overt pronoun is required in the downstairs subject position (and that makes the sentences ambiguous). (11d) and (12d), on the other hand, are ambiguous with an EC in subject position. These facts seem to correlate with another: only a topic can be base-generated in its dislocated position, as shown by the full acceptability of (13d):11 (13) a. *Quem1 que o Feco conhece o jornalista que entrevistou e1? who that the Feco knows the journalist that interviewed ‘Who does Feco know the journalist who interviewed?’ b. *O cara1 que o Feco conhece o jornalista que entrevistou e1 the guy that the Feco knows the journalist that interviewed
já chegou. already arrived
‘The guy that Feco knows the journalist who interviewed him is here.’
c. *É a Dani1 que o Feco conhece o jornalista que entrevistou e1. is the Dani that the Feco knows the journalist that interviewed ‘It is Dani that Feco knows the journalist who interviewed her.’ d. A Dani1, o Feco conhece o jornalista que entrevistou e1. the Dani the Feco knows the journalist that interviewed ‘(Speaking of) Dani, Feco knows the journalist who interviewed her.’
. D-linked wh-phrases and inanimate heads of relative clauses may be base-generated in their dislocated position in BP, as shown in (ia,b). Note, however, that it is probably not the case that the DP can be base-generated in (ib) because it is inanimate, but because being inanimate makes its relation to the ϕP in object position unambiguous. That fact confirms that languages like BP are highly sensitive to pragmatics and knowledge of the world when accessing antecedent possibilities: (i) a. Que professor1 o Feco conhece o jornalista que entrevistou e1? which professor the Feco knows the journalist that interviewed ‘Which professor does Feco know the journalist who interviewed him?’ b. Este é o livro1 que o Feco conhece o jornalista que escreveu e1. this is the book that the Feco knows the journalist that wrote ‘This is the book that Feco knows the journalist who wrote it.’
Topic prominence and null subjects
The ambiguity of (11d) and (12d), then, seems to be caused by the possibility of base-generating the topic, in the following manner: when the topic is moved to its surface position, it seems appropriate to suppose that the topic is the only possible antecedent for the null pronoun, as in the wh-movement, clefting and relativization cases. When the topic is base-generated, on the other hand, the matrix subject becomes a possible antecedent, perhaps the only possible one, as the remarks immediately below seem to indicate. To confirm the relation between movement and the possibility of being the antecedent for the null subject, observe that (13b) can be rescued by (overt) resumption. In that case, the most plausible analysis is that there was no movement (and therefore the island violation is ameliorated). Importantly, when there is resumption, the head of the relative clause is no longer a possible antecedent for the null subject in (14b); that reading is only possible if an overt pronoun occupies the subject position: (14) a. ?O cara2 que o Feco1 conhece o jornalista que entrevistou ele2 the guy that the Feco knows the journalist that interviewed him
já chegou. already arrived ‘The guy that Feco knows the journalist who interviewed him has arrived.’
b. ?O cara2 que o Feco1 convenceu ele2 que e1/*2 pode se the guy that the Feco convinced him that can refl elect-inf
eleger já chegou. already arrived
‘The guy that Feco convinced him that he can get elected has arrived.’
The same happens with topicalized elements. If the topic has been base-generated, as shown by the presence of a resumptive pronoun, it can no longer be taken as the antecedent of the null subject. (15) A Dani2, o Feco1 convenceu ela2 que e1/*2 pode se eleger. the Dani the Feco convinced her that can relf elect-inf ‘(Speaking of) Dani, Feco convinced her that he can get elected.’
The fact that wh-phrases left in situ also cannot be taken as the antecedent of a null embedded subject confirms the claim made here that there is a relation between movement and the choice of antecedent. (16) O Feco1 convenceu quem2 que e1/*2 pode se eleger? the Feco convinced who that can refl elect-inf ‘Who did Feco convince that he can get elected?’
Marcello Modesto
Further confirmation comes from the fact that in Finnish, movement of a matrix object also tends to improve its chances of being interpreted as the antecedent of a null embedded subject: (17) a.
Kenelle2 Liisa1 vakuutti t2 että e1/?2 voi tulla valituksi? to-whom Liisa assured that can come elected ‘Who did Liisa assure that s/he can get elected?’
b. Henkilö2 jolle Liisa1 vakuutti t2 että e1/?2 voi tulla valituksi the person that Liisa assured that can come elected
saapui jo. arrived already
‘The person to whom Liisa assured that s/he can get elected has arrived.’
c.
Jussille2 vakuutti Liisa1 että e1/?2 voi tulla valituksi. Jussi-to assured Liisa that can come elected ‘Jussi, Liisa assured him that s/he can get elected.’
(18) a.
Kenelle2 Liisa1 takasi t2 että e1/?2 saa ylennyksen? to-whom Liisa guaranteed that will-get promotion ‘To whom did Liisa guarantee that s/he will get the promotion?’
b. Henkilö2 jolle Liisa1 takasi t2 että e1/??2 saa ylennyksen the person that Liisa guaranteed that will-get promotion
saapui jo. arrived already
‘The person to whom Liisa guaranteed that s/he will get the promotion has arrived.’ c.
Jussille2 takasi Liisa1 että e1/?2 saa ylennyksen. Jussi-to guaranteed Liisa that will-get promotion ‘To Jussi, Liisa guaranteed that s/he will get the promotion.’
As seen in (5) above, in Finnish, just as in BP, a matrix object cannot be interpreted as the antecedent of a null embedded subject. However, as (17) and (18) show, a whmoved, relativized or topicalized object becomes a possible antecedent. The question mark in the readings where the moved matrix object is the antecedent of the null pronoun means that informants, although accepting that reading, said that the reading where the matrix subject is the antecedent was still the preferred one. Since none of my informants accepted a matrix object as the antecedent in (5) above, the contrast seen in BP is maintained in Finnish. What needs to be explained, though, is the fact that a moved object is the preferred antecedent in BP but not in Finnish. I will present my analysis of the facts and then return to that difference between BP and Finnish.12
.
Unfortunately, Chinese cannot be used to test the relation between movement and
Topic prominence and null subjects
4. The analysis in detail 4.1 A general overview The proposal I want to put forward here is that non-pro-drop languages like BP and Finnish allow null embedded subjects by virtue of being topic-prominent languages. Chinese, on this perspective, is also a non-null subject language since it has no agreement and all occurrences of null subjects can also be explained by the same strategy. The main assumption is that what distinguishes topic-prominent and subject-prominent languages is the fact that, in the former, a head F,13 above TP, always bears an OCC (= EPP) feature (see Chomsky 2004) which causes one constituent necessarily to be moved to its specifier position. Such a moved constituent is interpreted as an unmarked topic, meaning simply that it has a predication relation with the whole TP (see discussion in section 4.2 below). Chinese is uncontroversially a topic-prominent language; Finnish and BP have also been argued to be topic-prominent or discourse-oriented (see Holmberg & Nikanne 2002; and Negrão & Viotti 2000 respectively). The claim is, then, that null embedded subjects are possible in all of these languages because subjects occupy a higher (A-bar) position and so matrix subjects are able to identify null embedded subjects by binding them. An example sentence in BP is presented below (here and below, traces stand for copies deleted at PF and “e” stands for an unpronounced set of features, irrelevant details omitted throughout): (19) [FP o Feco1 [TP t1 convenceu a Dani2 [CP que [FP e1 [TP t1 pode se the Feco convinced the Dani that can refl eleger]]]]] elect-inf
Following Holmberg (2005), I assume that null subjects are non-referential sets of ϕ-features, ϕPs in his terminology. In “rich” agreement languages, referentiality is supplied by verbal agreement, making ϕPs behave like overt pronouns. In the languages discussed here, since agreement cannot provide a reference (or identify) the ϕP, the only way to interpret it will be taking it to be a variable at LF. In other words, ϕPs will only be possible in weak agreement languages when A’-bound.
antecedenthood. On one hand, Chinese does not present (overt) wh-movement. On the other hand, extraction of an object (be it by relativization or topicalization) of verbs taking two complements gives bad results for unclear and independent reasons (Audrey Li p.c.). . Holmberg & Nikanne (2002) refer to F, a category which stands for “Finite”, but which is related to Uriagereka’s (1995) F head, where F stands for “Functional”. Since the category I am employing here corresponds quite closely to the categories used by those authors, I use the same label. However, F here, as in the latter work cited, stands for “Functional”.
Marcello Modesto
Specifically, in (19), the ϕP gets bound by the higher subject that has been moved to SpecFP, to check its OCC feature. The chain formed by the two topics is nothing more than the application of the regular chain formation operation that applies between copies (cf. Nunes 1995). In this case, it may apply to distinct elements since they have the same set of ϕ-features.14 An overt pronoun could be also merged in the embedded subject position in all of those languages. In that case, coreference with the higher subject would be accidental and non-coreference would be possible. The ϕP, however, does not refer by itself and can only be interpreted when bound. It is, therefore, the fact that those languages are topic-prominent, with subjects being moved to an A-bar position (SpecFP), that allows ϕPs in the embedded subject position. This explains why English-type languages do not show null embedded subjects of the kind described here: subjects in English-type languages remain in SpecTP and so cannot variable-bind an embedded subject. Note that any phrase could check F’s OCC feature but, for locality reasons, it is always the subject, the closest element, which is moved to SpecFP, both in the matrix and the embedded clause (unless some other constituent has an uninterpretable feature to check, see section 4.2 below). The fact that objects are not good antecedents for null subjects in those languages is therefore straightforwardly explained. The analysis presented, then, explains all the characteristics listed in section 3. The variable-like characteristics of the embedded null subject have their origins in the fact that the ϕP must be A’-bound by the higher subject/topic. The c-command requirement on antecedents is obviously explained by the fact that chains can only be formed if there is a c-command relation. Locality requirements are also expected: an A’-chain cannot be formed if there is an intervening topic or any other intervening operator. The ban on split antecedents is due to the fact that a ϕP cannot be bound simultaneously by two distinct elements.15 In VP ellipsis and only-NP sentences, the absence of strict and invariant readings is explained
. In other words, it could be assumed that chain formation is possible here because a ϕP is nothing more than a set of ϕ-features and, since the ϕ-features of the embedded and the higher subject are the same, the ϕP is indistinguishable from its antecedent. . Obviously, full pronouns, both those occupying the same unmarked topic position and those in marked topic positions, will behave differently since they can be inherently referential and so will be able to corefer with more than one phrase, as in (i):
(i) O João convenceu a Maria que eles a polícia não pega. the João convinced the Maria that they the police not catch ‘João convinced Maria that the police will not catch THEM.’
Topic prominence and null subjects
since the ϕP is in fact a variable, bound by the preceding topic and cannot therefore be interpreted as referential.16 Although the analysis as presented so far is empirically adequate, many problems have still not been addressed. One may wonder, for instance, what will happen if the subject of a sentence is not referential, which topics are usually thought of as being. Having laid out the gist of the analysis, I will now clarify in section 4.2 why SpecFP may host non-canonical topics. The effects of topicalization, wh-movement and relativization on the choice of the antecedent of the null subject in BP and Finnish are the focus of sections 4.3 and 4.4 respectively. Finally, section 4.5 provides support for the analysis from intervention effects observed in both languages. 4.2 Two topic positions Taking the alternation in (20) as their point of departure, Holmberg & Nikanne (2002) claim that the EPP in Finnish requires a topic in sentence-initial position (which the authors identify as SpecIP). Here topic is understood as an expression denoting an individual or a group already established in the discourse, about which the predicate says something new. (20) a.
Graham Greene on kirjoittanut tämän kirjan. Graham Greene has written this book ‘Graham Greene wrote this book.’
b. Tämän kirjan on kirjoittanut Graham Greene. this book has written Graham Greene ‘This book, Graham Greene wrote.’
Holmberg (2005) says that “if the sentence contains one or more categories which can check the EPP, then one of them must remerge with IP (i.e., move to SpecIP), or an expletive must be merged with IP.” The reason for the conditional clause is that the EPP seems to be suspended in Finnish if the sentence does not contain a referential category capable of serving as a topic, but the idea that any referential category (so not only the subject) may check the EPP is maintained. As noted by
. Note that all the characteristics listed are the same ones encountered when dealing with obligatorily controlled (PRO) subjects. This fact has misled some researchers (e.g., Rodrigues 2004 and Ferreira 2004) into thinking that BP presents a case of finite control, as defined in Landau (2004). Although that may be true in some cases (e.g., when the complement clause is in the subjunctive, cf. footnote 6), it is certainly not true for subjects of indicative verbs, as shown in Modesto (2007).
Marcello Modesto
Holmberg (forthcoming), however, “it is not true that the EPP can only be checked by referential categories, capable of serving a topic function”: (21) a.
Kuka tahansa voi tulla tänne. who ever can come here ‘Anyone can come here.’
b. Tänne voi tulla kuka tahansa. here can come who ever ‘Anyone can come here.’ c. *Voi tulla kuka tahansa (tänne). can come who ever here ‘Anyone can come here.’
In (21a), kuka tahansa is an indeterminate quantifier and, as such, it is not referential or capable of functioning as a topic. The author then resorts to the descriptive generalization that “the subject may check the EPP even if it is not a referential category, but, for example a quantified NP, but non-subjects have to be referential and interpretable as topics, to check the EPP”, as exemplified in (21b). (21c) shows that if neither the subject nor another (referential) category is moved to the topic position, the sentence is ungrammatical, since the EPP-feature is not checked. The relevance of this discussion in the present scenario is that a nonreferential subject may be the antecedent of a null embedded pronoun in Finnish and BP, as noted by an anonymous reviewer (cf. 22a,b). Furthermore, the weak pronoun cê in BP may also fulfill that role, although it may not occur as a (marked) topic (cf. 22c,d): (22) a.
[Kuka tahansa]1 pettyy jollei e1 saa työtä. who ever gets.disappointed if.not gets work ‘Anyone gets disappointed if he doesn’t get work.’
b. Alguém1 acha que e1 sabe a resposta. someone thinks that knows the answer ‘Someone thinks that he knows the answer.’ c.
Cê1 disse que e1 sabe a resposta? you said that knows the answer ‘Did you say that you know the answer?’
d. *Cê/ *alguém1, o Feco disse que e1 sabe a resposta. you/ someone the Feco said that knows the answer ‘Feco said that you/someone knows the answer.’
The existence of sentences such as (21a) and (22a) does not invalidate the argument made here, nor the argument that the EPP in Finnish requires a topic in SpecFP. As discussed in Martins & Nunes (2005), two different topic positions
Topic prominence and null subjects
are independently motivated in BP. They discuss what they call “double subject constructions” in BP: (23) Os professores1 parecem que eles1 gostam da Maria. the teachers seem-3pl that they like of.the Maria ‘The teachers seem to like Maria.’
Modifying their analysis somewhat, we can assume that the higher subject is allowed to surface without a θ-role because it is a topic. In present terms, we can assume that it checks all features of the higher TP and then moves to the SpecFP position to check F’s OCC feature.17 But what is interesting about double subject constructions is that they allow weak pronouns in the subject position: (24) Cê1 parece que cê1 sabe a resposta. you seem that you know the answer ‘You seem to know the answer.’
According to Martins & Nunes (2005), the weak pronoun in (24) occupies an unmarked topic position above TP which does not exclude weak pronouns. I am arguing that all subjects in BP and Finnish occupy this unmarked topic position, which is different to the marked topic position in the left periphery of the sentence, reserved for constituents marked with a topic feature. Holmberg’s generalization that subjects may check the EPP even if they are non-referential categories, whereas non-subjects have to be referential and interpretable as topics to check the EPP in Finnish, can thus be explained. In both Finnish and BP, subjects are moved to SpecFP to check F’s OCC feature, but only because they are the closest goal found by the probe F. They are, then, a kind of grammatical topic, but do not need to be interpreted as fully-fledged semantic topics (nor do they need to have
. Martins & Nunes’ analysis of (23) is that the matrix topic checks the agreement features, but not the nominative Case feature of the matrix T since it is specified for “Topic Case”. The embedded T is defective in the sense that it does not check the Nominative Case of the embedded subject which is then moved in the covert component and adjoin to the matrix T. They claim that the sentence in (i) below is ungrammatical because, if the embedded object were to check the Case of the matrix T in the covert component, the embedded subject should induce a minimality violation. I do not understand how their analysis accords with the Phase Impenetrability Condition of Chomsky (2001, 2004). The interpretation in the text does not incur that problem, but seems to predict that (i), which is taken to be ungrammatical by the authors, is grammatical. In fact, sentences like (23) and (i) do not exist in my dialect, so I have little to say on the contrast between them. (i) *Os professores parecem que a Maria gosta deles. the teachers seem-3pl that the Maria likes them ‘It seems that Maria likes the teachers’
Marcello Modesto
the semantic characteristics of a topic).18 If a non-subject is moved to (a marked) topic position (possibly checking the OCC feature of F on its way there), it is only because it has been merged with a Topic feature that needs to be checked in the left periphery. That position, unlike SpecFP, requires a referential category and excludes weak pronouns. Phrasal structure in BP and Finnish would then include the partial tree in (25), where SpecTopP is reserved for marked topics: (25)
CP TopP FP TP
The word-order differences between the two languages is accounted for if we assume that the verb in Finnish is moved as high as Top0 when that category is merged; otherwise it raises to F0, deriving the order topic-verb-subject. In BP, on the other hand, the verb does not raise as high, staying at T0 (or even below), thereby delivering the order topic-subject-verb. 4.3 The relation between movement and antecedenthood in BP It is clear from the data given in section 3 that the subject in BP is always the antecedent of the embedded null subject except when another constituent has features to check in (and so is moved to) the CP domain. It is, then, very plausible to assume that whenever a constituent other than the subject is moved in BP (either by wh-movement, relativization or topicalization), it is that constituent which will check F’s OCC feature, occupying SpecFP as an intermediate position on its way to SpecCP or other positions.19 Since that constituent (or a copy thereof)
. The distinction made here is then reminiscent of the distinction between ‘primary’ and ‘secondary’ topicalization made by Hope (1974). Hope discusses how languages like Lisu, where every sentence seems to have “topic + comment” structure, may have sentences like (i) below, where a non-referential phrase appears marked as a topic: (i) .
swu nya atha do - a one TOPIC-marker knife forge DECL.-marker ‘Someone is forging a knife.’
Note that moving a constituent to SpecFP is required to check F’s OCC feature and that
Topic prominence and null subjects
occupies the (unmarked) topic position, the topic-chain formed will result in a null subject which has a non-subject as its antecedent. If, however, as seen before, the constituent is base-generated in its ultimate position, which is possible for marked topics, the subject still has to move to SpecFP (to satisfy F’s OCC feature) and so it will be interpreted as the antecedent of the null embedded pronoun. The option of moving the subject to SpecFP and then moving another constituent (the object, for instance) to a higher position is excluded in BP due to Relativized Minimality (as in Rizzi 1990). Sentence (26) below exemplifies this Minimality effect by showing that wh-phrases may not move over a topic; therefore, the structure of a sentence involving wh-movement of an object (such as (12a), repeated as (27a) below) has to be (27b) and not (27c). In (27b), the object wh-phrase moves from its original position to SpecFP and then to its final position in SpecCP. That structure correctly predicts that, when the wh-object is moved, it becomes the only possible antecedent for the embedded null subject. The structure depicted in (27c) is ungrammatical for the same reason (26) is, i.e., Relativized Minimality: (26) *Quem, esses livros, leu? who these books read ‘Who did read these books?’ (27) a.
Quem2 que o Feco1 convenceu t2 que e*1/2 pode se eleger? who that the Feco convinced that can refl elect-inf ‘Who did Feco convince that s/he can get elected?’
b. [CP quem2 que [FP t2 [TP o Feco1 convenceu t2 [CP que [FP e2 [TP who that the Feco convinced that t2 pode se eleger]]]]]] can refl elect-inf c. *[CP quem2 que [FP o Feco1 [TP t1 convenceu t2 [CP que [FP e1 who that the Feco convinced that [TP t1 pode se eleger]]]]]] can refl elect-inf
The same kind of intervention effect can be seen in relative clauses. Sentence (28) shows that movement over a topic is impossible in relative clauses. Therefore, the analysis of a sentence like (12b), repeated as (29a), must be (29b), where the
this constituent does not need to have a Topic feature or any other special feature in order to land in that position. Therefore, even a wh-phrase, which can never be interpreted as a topic, may stop at SpecFP on its way to CP.
Marcello Modesto
subject stays in SpecTP, and not (29c), where the relative operator moves over the subject in SpecFP: (28) *O cara1 que, esses livros2, leu t2 já chegou. the guy that these books read already arrived ‘The guy who read these books has already arrived’ (29) a.
O cara2 que o Feco1 convenceu t2 que e*1/2 pode se eleger the guy that the Feco convinced that can refl elect-inf
já chegou. already arrived
‘The guy who Feco convinced that he can get elected is already here.’
b. [o cara2 [CP Op2 que [FP t2 [TP o Feco1 convenceu t2 [CP que [FP e2 the guy that the Feco convinced that [TP t2 pode se eleger]]]]]]] can refl elect-inf c. *[o cara2 [CP Op2 que [FP o Feco1 [TP t1 convenceu t2 [CP que [FP e1 the guy that the Feco convinced that [TP t1 pode se eleger]]]]]]] can refl elect-inf
Things are slightly different with topicalization. The discussion about sentences (13) above has shown that topics, unlike wh-phrases and relative operators, can be either moved or base-generated. Movement of a topic over another topic is barred by Minimality, as expected (see (30a)). However, a topic may intervene between a base-generated topic and its related position, as shown in (30b). This shows that the structure assigned to sentence (12d), repeated here as (31a), may be (31b or c) but not (31d). That BP allows structures (31b and c) explains why sentences in which an object is topicalized are ambiguous between an interpretation where the antecedent of the embedded null subject is the object, and another where the antecedent is the matrix subject. (30) a. *O Feco1, esses livros2, t1 já leu t2. the Feco these books already read b. O Feco1, esses livros2, ele1 já leu t2. the Feco these books he already read ‘Feco has already read these books’ (31) a.
A Dani2, o Feco1 convenceu t2 que e1/2 pode se eleger. the Dani the Feco convinced that can refl elect-inf ‘(Speaking of) Dani, Feco convinced her that s/he can get elected.’
Topic prominence and null subjects
b. [TopP a Dani2 [FP t2 [TP o Feco1 convenceu t2 [CP que [FP e2 [TP t2 the Dani the Feco convinced that
pode se eleger]]]]]] can refl elect-inf
c. [TopP a Dani2 [FP o Feco1 [TP t1 convenceu e2 [CP que [FP e1 [TP t1 the Dani the Feco convinced that
pode se eleger]]]]]] can refl elect-inf
d. *[TopP a Dani2 [FP o Feco1 [TP t1 convenceu t2 [CP que [FP e2 [TP t2 the Dani the Feco convinced that
pode se eleger]]]]]] can refl elect-inf
I have shown that, due to Minimality, if the subject moves to SpecFP and another phrase moves over it to SpecCP or another position, the result is ungrammatical ((31c) is grammatical only because the higher topic has been base-generated, not moved). However, since the subject has to move to SpecFP when no other constituent is moved to the CP domain, by virtue of its being the closest DP which can satisfy the OCC feature of F, the system seems to involve some kind of look ahead. When the F head is merged in the tree, the system has to know if the object needs to check some feature at a later time in order to allow attraction of the subject DP or not. This problem is avoided if we assume with Chomsky (2001) that all evaluation is done at the phase level. Movement of any constituent to SpecFP is actually free. At the phase level, however, a derivation where an object has been moved to SpecFP crashes as a violation of the Minimal Link Condition, i.e., the subject is a closer phrase which could have checked the OCC feature of F. If, however, at the phase level, the object sits in SpecCP checking a wh-feature or in some other position, checking some other feature, the fact that it has also checked the OCC feature of F is overlooked since any other derivation would have violated Minimality. (For another way of deriving the same facts, see Richards (1998), particularly his discussion of French participle agreement, in which movement of an element may violate Shortest Move if a following application of Move involving that element respects it.). 4.4 The relation between movement and antecedenthood in Finnish Unlike BP, Finnish does not show Minimality effects, i.e., a wh-phrase is moved over a topic without degrading the sentence, as seen in (32): (32) a.
Kuka tämän kirjan on kirjoittanut? who this book has written ‘Who wrote this book?’
Marcello Modesto
b. henkilö joka tämän kirjan on kirjoittanut saapui jo. the person who this book has written arrived already ‘The person who wrote this book has already arrived.’
Structures in which a relativized or wh-object is moved over a subject sitting in SpecFP are, therefore, grammatical in Finnish. Lack of Minimality effects in this language, then, explains why Finnish speakers clearly prefer to take the matrix subject as the antecedent of the null embedded subject, even when an object has been moved. Sentences (17a,b), repeated below as (33a,b), will, then, be assigned the structures in (33c,d) respectively: (33) a.
Kenelle2 Liisa1 vakuutti t2 että e1/?2 voi tulla valituksi? who Liisa assured that can come elected ‘Who did Liisa assure that s/he can get elected?’
b. Henkilö2 jolle Liisa1 vakuutti t2 että e1/?2 voi tulla the person that Liisa assured that can come
valituksi elected
saapui jo. arrived already
‘The person who Liisa assured that s/he can get elected has arrived.’
c. [CP Kenelle2 [FP Liisa1 vakuutti [TP t1 tv t2 [CP että [FP e1 [TP t1 voi who Liisa assured that can
tulla valituksi]]]]]] come elected
d. [ henkilö2 [CP jolle2 [FP Liisa1 vakuutti [TP t1 tv t2 [CP että [FP e1 [TP the person who Liisa assured that t1 voi tulla valituksi]]]]]]] can come elected
The fact that movement of the matrix object makes it possible to interpret the object as the antecedent of the null embedded subject (though as a marked option) is also explained by lack of intervention effects. In (33c,d), the operator in the CP domain binds its variable even though a topic intervenes. Finnish speakers, therefore, seem to allow binding of the null subject by the moved object, although the matrix subject would be a closer binder, as in (34a,b): (34) a. ?[CP kenelle2 [FP Liisa1 vakuutti [TP t1 tv t2 [CP että [FP e2 [TP t2 voi who Liisa assured that can
tulla valituksi]]]]]] come elected
Topic prominence and null subjects
b. ?[henkilö2 [CP jolle2 [FP Liisa1 vakuutti [TP t1 tv t2 [CP että [FP e2 the person who Liisa assured that [T t2 voi tulla valituksi]]]]]] can come elected
Although Finnish does not allow two topics to the left of the verb (whether they are based-generated or not, cf. Holmberg 1998), as shown in (35), movement of a topic to SpecTopP is allowed to cross the element in SpecFP, perhaps in virtue of the movement of the verb to Top0. Verb movement makes the two topics in (36) equidistant (cf. also Öztürk, this volume), allowing the higher topic to bind its variable in object position. The two possible structures for sentence (17c) where the matrix object is topicalized (cf. (36a)), are given in (36b,c): (35) *Jussi tämän kirjan on kirjoittanut (hän). Jussi this book has written he ‘Jussi, this book, he has written.’ (36) a. Jussille2 vakuutti Liisa1 että e1/?2 voi tulla valituksi. to.Jussi assured Lissa that can come elected ‘To Jussi, Liisa assured that s/he can get elected.’ b. [TopP Jussile2 vakuutti [FP Liisa1 tv [TP t1 tv t2 [CP että [FP e1 [TP t1 to.Jussi assured Liisa that
voi tulla valituksi]]]]]] can come elected
c. ?[TopP Jussile2 vakuutti [FP Liisa1 tv [TP t1 tv t2 [CP että [FP e2 [TP t2 to.Jussi assured Liisa that
voi tulla valituksi]]]]]] can come elected
4.5 Intervention effects If it is true that, as I am claiming here, null embedded subjects in topic-prominent languages are bound, forming a (topic-) chain with a matrix constituent, one would expect to find intervention effects of the kind discussed in Rizzi (1990). This expectation is borne out. In the relevant languages, a topic in the embedded clause prevents a chain being formed between the higher and the embedded subject. Therefore, if the embedded subject is null, it must be generic: (37) a.
Oppilas1 tietää ettei e1 pysty ratkaisemaan tehtävää. student knows that-not can solve assignment ‘The student knows that he can’t solve the assignment.’
Marcello Modesto
b. Oppilas1 tietää ettei tehtävää pysty e*1 ratkaisemaan. student knows that-not assignment can solve ‘The student knows that the assignment can’t be solved.’
In (37a), the embedded subject occupies its usual position, SpecFP, and may form a topic-chain with the higher subject oppilas. In (37b), on the other hand, although the embedded subject could still occupy the same position, the presence of the marked topic tehtävää in SpecTopP would prevent chain formation and, consequently, PF-deletion. The absence of a phonologically full subject in that clause, then, can only be interpreted as signaling a generic reading. Consider now BP: (38a) is ambiguous between a generic reading and a bound (coreferent) reading, whereas (38b) has the generic reading only. Interestingly, (38c) also only has a generic reading when the adverb is taken to be related to the embedded event. (38) a.
O Feco1 me falou que e1 vende cachorro quente na praia. the Feco to-me said that sells dog hot on beach ‘Feco told me that he sells hot dogs/hot dogs are sold at the beach.’
b. O Feco1 me falou que na praia e*1 vende cachorro quente. the Feco to-me said that on beach sells dog hot ‘Feco told me that hot dogs are sold at the beach.’ c.
Na praia, o Feco1 me falou que e*1 vende cachorro quente on beach, the Feco to-me said that sells dog hot ‘Feco told me at the beach that he sells hot dogs.’ ‘Feco told me that hot dogs are sold at the beach.’
Sentence (38a) shows that a null subject in BP can be interpreted as bound by the element in matrix SpecFP or that, in some circumstances, it can be interpreted as generic.20 (38b) shows that if another phrase occupies SpecFP or has been basegenerated in SpecTopP, a chain between the two subjects may not be formed and the embedded subject has to be interpreted as generic. Sentence (38c), however, is the most interesting case. If the locative na praia (“at the beach”) is taken to qualify the matrix event, it can already be merged in the higher SpecTopP and the sentence
. The generic reading is absent when the locative adverb is not present: (i) O Feco1 me falou que e1 vende cachorro quente. the Feco to-me said that sells dog hot
‘Feco told me that he sells hot dogs/*hot dogs are sold at the beach.’
That seems to imply that, even in (38a) under the generic reading, the adverb is checking the OCC feature of F0 and has been extraposed.
Topic prominence and null subjects
can be interpreted as meaning that Feco told me, when we were at the beach, that he sells hot dogs (for a living). In this case, the matrix subject has moved to SpecFP. On the other hand, if the locative qualifies the embedded event, it presumably has to be moved from the embedded clause. Moving the locative over the matrix subject in SpecFP would cause a Minimality effect so it has to be moved to SpecFP before moving to SpecTopP. In that case, the matrix subject has to remain in SpecTP and no topic chain can be formed since the two topics do not share the same features and reference. Consequently, the embedded clause only has the generic reading. It is important to note that, in this case, an adverb in sentence-initial position prevents the null embedded subject from being interpreted as coreferent with the matrix subject. That this very surprising state of affairs is accounted for and explained by the analysis put forth here provides strong support to it.
5. The topic prominence parameter 5.1 Topic-prominent languages So far, I have claimed that languages such as Chinese, BP and Finnish do not show all the properties related to the NSP because they are not, in fact, NSLs. The existence of null subjects in these languages is due to the fact that they are topic-prominent languages (and may be due to other factors as well, cf. footnote 6). So, if I am right, topic prominence determines a parameter which also has a cluster of properties associated with it, and one of those properties is the existence of phonological null subjects. In this section, I would like to investigate what other properties may be associated with such a parameter and how to characterize it properly. In their characterization of a typology based on the grammatical relations subject-predicate and topic-comment, Li & Thompson (1976: 459) define four types of languages: … (i) languages that are subject-prominent (a term introduced by E.L. Keenan); (ii) languages that are topic-prominent; (iii) languages that are both subjectprominent and topic-prominent; (iv) languages that are neither subject-prominent nor topic-prominent. In subject-prominent (Sp) languages, the structure of sentences favors a description in which the grammatical relation subject-predicate plays a major role; in topic-prominent (Tp) languages, the basic structure of sentences favors a description in which the grammatical relation topic-comment plays a major role. In type (iii) languages, there are two equally important distinct sentence constructions, the subject-predicate construction and the topic-comment construction; in type (iv) languages, the subject and the topic have merged and are no longer distinguishable in all sentence types.
Marcello Modesto
The authors exemplify the first group with Indo-European languages; the second with Chinese and Lisu (a Lolo-Burmese language); the third with Japanese and Korean; and the forth group with Tagalog. It is unclear (to me) how to characterize this last group, since the authors do not provide a systematic way of telling when a language is neither Sp nor Tp (they only list positive characteristics of Sp and Tp languages, no negative ones). These are the characteristics of Tp languages that Li & Thompson cite, all of which are found in BP, Chinese and Finnish: 1. There is a “surface coding” for topics in those languages; they are coded by a special topic marker (Japanese, Korean) or appear in sentence-initial position (Chinese, BP and Finnish). 2. Passive constructions, which are common in Sp languages, are absent or marginal, and rarely used in speech, in Tp languages. The authors cite Lisu, Lahu and Chinese as examples. Turning to the languages under consideration here, Finnish does not even have a passive construction strictu senso, using an OVS active construction to express the same meaning a passive sentence would have in Sp languages (cf. (20b) above); BP, in turn, can definitely be added to the list of languages where passives are rarely used in colloquial speech, as discussed below. 3. Dummy subjects are not found in Tp languages. Although this is true for Chinese and BP, it is not entirely true for Finnish, but that is not a problem, considering the approach adopted to topic-prominence in this work, namely that Tp languages have an extra EPP-feature to check. In Finnish, that feature is sometimes checked by a dummy subject. 4. Tp languages use “double subject” constructions. Again, Chinese is used as an example and BP fits the picture well, as seen in section 4.2, whereas Finnish, to my knowledge, differs. 5. The topic typically controls co-referential constituent deletion, which is almost exactly what I am arguing for in this work. 6. Tp languages tend to be V-final. Japanese, Korean, Lahu and Lisu exemplify this property. The authors argue that Chinese is in the process of becoming V-final and BP, as discussed below, is pervasively making transitive verbs into intransitives, which makes them sentence-final. Finnish again does not fit the picture exactly, although verb-final constructions may be found in that language as well (cf. Holmberg 2000). 7. In Tp languages, there are no constraints on what constituent may be the topic. The relevance of this last characteristic is questionable since many Sp languages, like English, also do not constrain what may be topicalized. 8. Topic-comment sentences are the basic sentence type in Tp languages. I will not address this last characteristic since ‘basic sentence type’ seems to be a rather ill-defined concept.
Topic prominence and null subjects
Note that Li & Thompson’s fifth characteristic, that the topic typically controls co-referential constituent deletion, implies that all Tp languages present null arguments, which is the main claim made here. It also implies that the authors had in mind a mechanism of producing null arguments (involving deletion) which is rather similar to the one advocated here. Having argued that Chinese, Finnish and BP are, in fact, Tp languages, I will now discuss some characteristics of BP which can be related to the Topic Prominence Parameter. BP seems particularly relevant, as it is a language that has gone through a parameter shift very recently (possibly even in the last few decades). The effects of the shift are therefore rather apparent. 5.2 Th e relation between the null subject and the topic prominence parameter Holmberg & Nikanne (2002) state that “a language is subject-prominent when the argument which is externalized in an active sentence is always the subject … A language is topic-prominent when the argument which is externalized need not be the subject, but can be any category capable of functioning as topic.” They note that Greek and Spanish presumably qualify as Tp languages in their terms, due to the general possibility of OVS orders. Postverbal subjects, however, are characteristic of any NSL of the Romance kind, as discussed in Rizzi (1982, 1986). It may be the case, then, that the NSP is related to the Topic Prominence Parameter and that NSLs can be either Sp or Tp in Li & Thompson’s (1976) terms. European Portuguese (EP) has several characteristics of a Tp language: it uses OVS order when the object has been previously introduced in the discourse (39a), and the topic controls the interpretation of empty constituents (39b). It, however, has to be considered Sp as well since passive and verb-initial constructions are common (39c,d). Finnish, a non-Sp language, on the other hand, usually does not permit verb-initial sentences due to the fact that an argument has to be externalized (as a topic, in Holmberg & Nikanne’s terms) – cf. (40) in this connection: (39) a.
A sopa comeu o João. the soup ate the João ‘João ate the soup.’
b. Essa árvore1, as folhas dela são grandes demais então this tree the leaves of.her are big too so
eu não vou plantar e1. I not go plant-inf
‘This tree, its leaves are too big, so I won’t plant it.’
(EP)
Marcello Modesto
c.
O prédio já foi construído. the building already was built ‘The building has already been built.’
d. Pode-se nadar neste lago. may-SE swim in.this lake ‘One may swim in this lake.’ (40) a. *Istuu mukavasti tässä. sits comfortably here b. Tässä istuu mukavasti. here sits comfortably ‘One sits comfortably here.’
BP has become a non-NSL, as argued here and elsewhere (cf. the collection of papers in Kato & Negrão 2000) and so, if the present speculations are correct, it has become a non-Sp language. An indication that this is indeed correct is that BP is becoming more like Finnish than EP in requiring a topic at the beginning of the sentence:21 (41) a. ??Pode nadar neste lago. may swim-inf in.this lake b. Neste lago pode nadar. in.this lake may swim-inf ‘One may swim in this lake.’
The shift from an NSL (and so from a Sp language) to a non-NSL (and so to a non-Sp language) has had other consequences in BP. As argued in Holmberg (in press) and witnessed in (41b) above, a generic null pronoun does not satisfy the EPP in Finnish or in BP.22 (EPP here refers to the OCC-feature of F0.) In impersonal structures, the languages in question, then, have to merge some referential phrase in the topic (SpecFP) position. This fact has had a domino effect in BP, creating a construction, now pervasive in the language, which is a mixture of a middle and an ergative construction (cf. Lekakou, this volume) that applies indiscriminately to nearly any transitive verb and that is now much more frequently used than the passive construction: (42) O prédio já construiu. the building already constructed ‘The building has already been constructed.’
. Sentence (41a) requires a special intonation to be acceptable. For instance, it is much more acceptable as a question than as a statement. The loss of the Romance SE clitic in BP is probably related to the shift from NSL to non-NSL. . Probably since it stays in vP, as argued by Holmberg (in press).
Topic prominence and null subjects
Although (42) requires adverbial support (a characteristic of the English middle construction), it is compatible with the progressive tense and is not a generic statement in the sense that it may be about specific entities and describe a particular event in time, both characteristics of ergative constructions: (43) a.
Esse prédio está construindo rápido. this building is constructing quick ‘This building is being built quickly.’
b. Esse prédio construiu em junho. this building constructed in June ‘This building was built in June.’
Another tricky characteristic of this construction is that the verb agrees in number with the topic. That seems to imply that, contra Holmberg (in press), the null generic subject is assigned Case, but does not value the ϕ-features of T (at least in BP). The generic pronoun, being invisible to the EPP, remains vP-internal, where it gets its generic interpretation (cf. once again, Lekakou, this volume). Another phrase, therefore, has to (re-)merge in SpecTP and SpecFP to check the two OCC features. While the comments made here are clearly only of a preliminary and therefore necessarily sketchy nature, it is evident that the pervasive use of this construction in BP, and the consequent infrequence of passives, is related to the fact that this language has become non-Sp. 5.3 Scope facts Tp languages – taken to be the ones in which the subject usually occupies a (grammatical) topic position (when there is no overt topicalized element) – are expected to show slightly different interpretations of subjects than languages in which these elements do not occupy the topic position. That is indeed what one finds with respect to scope relations. Inverse scope readings – where a quantified object phrase (QP) scopes over a quantified subject – are common in English and other languages, but they are absent in BP and Chinese:23 (44) a.
Every student admires some professor.
(ambiguous)
. In Finnish, since word order is “free”, inverse scope readings are usually encoded via OVS structures. In other words, object QPs scope over subject QPs when they are moved to a position higher than SpecFP; and subject QPs scope over object QPs in SVO sentences, as expected. However, some of my informants claim that, although the readings reflecting the surface word order are preferred, the sentences are still ambiguous with respect to scope. I will leave a full discussion of Finnish for another occasion.
Marcello Modesto
b. Meige xuesheng dou mai-le yiben shu. (unambiguous) every student all buy-asp one book ‘Every student bought one/a book (but not necessarily the same one).’ c.
Todo professor entrevistou um aluno. every professor interviewed a student ‘Every professor interviewed a (non-specific) student.’
(unambiguous)
The difference between BP and Chinese, on the one hand, and English, on the other, can be explained along the lines of the analysis presented here. Since subjects in the former group of languages occupy a topic position, a higher position than that occupied by subjects in the latter group, inverse scope readings cannot obtain, or are at least are harder to obtain.24 In this way, lack of inverse scope readings can be seen as another property of topic-prominent languages. 6. Conclusions Data presented in this paper show that there is an undeniable relation between movement and the possibility of being the antecedent of a null subject in both BP and Finnish. There also seems to be a relation between Minimality effects and the choice of the antecedent. Both relations appear to be accounted for if we assume that the antecedent of the null subject must pass through the topic position in these languages. Since BP shows Minimality effects, a matrix subject cannot be interpreted as the topic of the sentence when the object is wh-moved or relativized. In Finnish, on the other hand, the subject is most commonly moved to the topic position, with the consequence that no Minimality effects are attested. If the explanation for the BP and Finnish facts is that the subject occupies the topic position (SpecFP) rather than the “canonical subject position” (SpecTP), the same analysis can be argued to hold for null subjects in Chinese-type languages, which are problematic for the NSP. These languages have long been said to be topic-prominent, a property which we here formalize as entailing subject movement to a specially designated topic position (SpecFP or SpecTopP). On the analysis presented here, null subjects in these traditionally “problematic” NSLs can therefore be explained as a property associated with a different parameter: one relating to topic-prominence.
. For reasons of space, I will not give a full account of scope facts in these languages, nor will I comment on the exact nature of the syntax of scope. The interested reader is referred to chapter 4 of Modesto (2000b), where such an account can be found.
Topic prominence and null subjects
References Battistella, E. 1985. On the distribution of PRO in Chinese. Natural Language and Linguistic Theory 3: 317–340. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 1–52. Cambridge MA: The MIT Press. Chomsky, N. 2004. Beyond explanatory adequacy. In Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3, A. Belletti (Ed.), 104–131. Oxford: OUP. Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. Chomsky, N. 2008. On phases. In Foundational Issues in Linguistic Theory. R. Freidin, C. Otero & M. Zubizarreta (Eds) 133–166. Cambridge MA: The MIT Press. Duarte, M.E. 1995. A Perda do Princípio ‘Evite Pronome’ no Português Brasileiro. Ph.D. dissertation, Unicamp, Campinas. Egerland, V. 2003. Impersonal pronouns in Scandinavian and Romance. Working papers in Scandinavian syntax 71: 75–102. Lund: Department of Scandinavian Languages. Ferreira, M. 2004. Hyper-raising and null subjects in Brazilian Portuguese. In MIT Working Papers in Linguistics 47: Collected Papers on Romance Syntax, 57–85. Cambridge MA: MITWPL. Figueiredo Silva, M.C. 1996. A Posição Sujeito no Português Brasileiro: Frases finitas e infinitivas. Campinas: Editora da Unicamp. Galves, C. 1993. O Enfraquecimento da Concordância no Português do Brasil. In Português Brasiliero: Uma viagem diacrônica, I. Roberts & M. Kato (Eds), 387–408. Campinas: Editora da Unicamp. Gutman, E. 2004. Third person null subjects in Hebrew, Finnish and Rumanian: An accessibility-theoretic account. Journal of Linguistics 40: 463–490. Cambridge: CUP. Halmari, H. 1996. On accessibility and coreference. In Reference and Referent Accessibility, T. Fretheim & J. Gundel (Eds), 155–178. Amsterdam: John Benjamins. Hankamer, J. & Sag, I. 1976. Deep and surface anaphora. Linguistic Inquiry 7: 391–428. Heinonen, T. 1995. Null subjects in Finnish: From either-or to more-or-less. MS, Research Institute for the languages of Finland. Higginbotham, J. 1992. Anaphoric reference and common reference. MS, MIT. Holmberg, A. 1998. Word order variation in some European SVO languages: A parametric approach. In Constituent Order in the Languages of Europe, A. Siewierska (Ed.), 553–598. Berlin: Mouton de Gruyter. Holmberg, A. 2000. Deriving OV order in Finnish. In The derivation of VO and OV, P. Svenonius (Ed.), 123–152. Amsterdam: John Benjamins. Holmberg, A. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36(4): 533–564. Holmberg, A. Forthcoming. The empty generic subject pronoun in Finnish. To appear in Null Subjects: The Structure of Parametric Variation, A. Holmberg & I Roberts (Eds). Cambridge: CUP. Holmberg, A. & Nikanne, U. 2002. Expletives, subjects, and topics in Finnish. In Subjects, Expletives, and the EPP, P. Svenonius (Ed.), 71–105. Oxford: OUP.
Marcello Modesto Hope, E. 1974. The Deep Syntax of Lisu Sentences. Canberra: The Australian National University. Huang, C.-T.J. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531–574. Huang, C.-T.J. 1989. Pro-drop in Chinese: A generalized control theory. In The Null Subject Parameter, O. Jaeggli & K. Safir (Eds), 185–214. Dordrecht: Kluwer. Jaeggli, O. & K. Safir (Eds). 1989. The Null Subject Parameter. Dordrecht: Kluwer. Kato, M. & Negrão, E.V. (Eds). 2000. Brazilian Portuguese and the Null Subject Parameter. Frankfurt: Vervuert-Iberoamericana. Landau, I. 2004. The scale of finiteness and the calculus of control. Natural Language and Linguistic Theory 22: 811–877. Li, C.N. & Thompson, S.A. 1976. Subject and topic: A new typology of language. In Subject and Topic, C. Li (Ed.), 457–489. New York NY: Academic Press. Martins, A.M. & Nunes, J. 2005. Raising issues in Brazilian and European Portuguese. Journal of Portuguese Linguistics 4: 53–77. Modesto, M. 2000a. Null Subjects without ‘Rich’ Agreement. In Brazilian Portuguese and the Null Subject Parameter, M. Kato & E. Negrão (Eds), 147–174. Frankfurt: Vervuert-Iberoamericana. Modesto, M. 2000b. On the Identification of Null Arguments. Ph.D. dissertation, University of Southern California, Los Angeles. Modesto, M. 2007. Null subjects in Brazilian Portuguese and Finnish: They are not derived by movement. In New Horizons in the Analysis of Control and Raising, S. Dubinsky & W. Davies (Eds). Dordrecht: Springer. Negrão, E. 1997. Asymmetries in the distribution of overt and empty categories in Brazilian Portuguese. In Clitics Pronouns and Movement, J. Black & V. Motapanyane (Eds), 217–235. Amsterdam: John Benjamins. Negrão, E.V. 1999. O Português Brasileiro: Uma Língua Voltada para o Discurso. MS, University of São Paulo. Negrão, E.V. & Viotti, E. 2000. Brazilian Portuguese as a discourse-oriented language. In Brazilian Portuguese and the Null Subject Parameter, M. Kato & E. Negrão (Eds), 105–125. Frankfurt: Vervuert-Iberoamericana. Nunes, J. 1995. The Copy Theory of Movement and Linearization of Chains in the Minimalist Program. Ph.D. dissertation, University of Maryland, College Park. Richards, N. 1998. The principle of minimal compliance. Linguistic Inquiry 29: 599–629. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–557. Rizzi, L. 1990. Relativized Minimality. Cambridge MA: The MIT Press. Rohrbacher, B. 1994. The Germanic VO languages and the full paradigm: A theory of V to I raising. Ph.D. dissertation, University of Massachusetts, Amherst. Rodrigues, C.A.N. 2004. Impoverished Morphology and A-Movement out of Case Domains. Ph.D. dissertation, University of Maryland, College Park. Speas, M. 1994. Null arguments in a theory of economy of projection. In Functional Projections [University of Massachusetts Occasional Papers in Linguistics 17], E. Benedicto & J. Runner (Eds), 179–208. Amherst MA: GSLA. Tsao, F.F. 1977. A Functional Study of Topic in Chinese: The First Step Towards Discourse Analysis. PhD dissertation, University of Southern California, Los Angeles. Uriagereka, J. 1995. Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26: 79–123.
Topic prominence and null subjects
Vainikka, A. 1989. Deriving Syntactic Representations in Finnish. Ph.D. dissertation, University of Massachusetts, Amherst. Vainikka, A. & Levy, Y. 1999. Empty subjects in Finnish and Hebrew. Natural Language and Linguistic Theory 17: 613–671. Vikner, S. 1997. V0-to-I0 Movement and inflection for person in all tenses. In The New Comparative Syntax, L. Haegeman (Ed.), 189–213. London: Longman.
Non-configurationality Free word order and argument drop in Turkish Balkiz Öztürk This paper investigates free word order and argument drop in Turkish as the two properties typically associated with non-configurationality. We will argue that these properties follow from the lack of case-driven Agree in Turkish phrase structure. That is, the case feature in Turkish is not located on the functional heads T and v, but is available within the thematic domain. In languages like English, on the other hand, given that case is provided by hierarchically ordered functional projections, i.e., TP > vP, arguments forming case-chains with these functional heads get represented within a hierarchical configuration. Lack of such chain formation in Turkish is what creates the flat structure effect, which in turn facilitates free word order and argument drop. Keywords: non-configurationality; null arguments; case-driven Agree; free word order; referentiality
1. Introduction Since Hale (1980, 1983), free word order, extensive argument drop and discontinuous constituents are taken to be three main characteristics that would define a language as non-configurational. Turkish is also a language which exhibits free word order, extensive argument drop and limited discontinuous constituents as illustrated in (1)–(3) respectively, implying that Turkish might also be a nonconfigurational language: (1) a. b.
Adam ev- i bul-du man house-acc find-past ‘Ali found the house.’ Ev-i adam bul-du
(2) (Ben) (onu) sev-di-m I him/her love-past-1ps ‘I loved him/her.’
SOV OSV
c. d. e. f.
Adam bul-du ev-i Ev-i bul-du adam Bul-du adam ev-i Bul-du ev-i adam
SVO OVS VSO VOS
Balkiz Öztürk
(3) a.
Ben Ali-nin kitab-ın-ı oku- du- m I Ali-gen book-3ps-acc read-past-1ps ‘I read Ali’s book.’
b. Ben kitab- ın-ı I book-3ps-acc ‘I read Ali’s book.’ c.
oku- du-m Ali-nin1 read-past-1ps Ali-gen
Ben üç tane elma ye- di-m I 3 cl apple eat- past -1ps ‘I ate three apples.’
d. Ben elma ye- di- m üç I apple eat- past-1ps 3 ‘I ate three apples.’
tane cl
It has been widely noticed in the literature that the free word order observed in Turkish is derived via movement not via base-generation; therefore, it is an instance of scrambling (Kural 1991, 1992; Kornfilt 2003b). Thus, the type of nonconfigurationality exhibited by the phrase structure of Turkish falls under the Japanese-type non-configurationality, that is, non-configurationality as movement, as defined by Baker (2001). This study will investigate the source of the movementbased non-configurationality observed in Turkish and will then focus on how the general properties of non-configurationality are derived.2 As cases of discontinuous constituents are very limited in Turkish, we will mainly focus on free word order and argument drop, leaving the issue of discontinuous constituents for a future study. Section 2 will introduce a new parameter for (non-)configurationality based on how case and referentiality features interact with each other in a language’s phrase structure. The discussion will be mostly based on a comparison between English – a configurational language and Turkish – a language which exhibits non-configurational properties such as the availability of free word order and
. One of the reviewers suggested that given that the head noun agrees with the possessor overtly, this construction might not be a true discontinuous construction. However, Turkish allows for other NP constructions lacking agreement which are discontinuous in nature, as illustrated by the sentence-final classifier phrase in (3d). In this connection, it is, however, important to highlight the fact that classifier phrases tend to appear sentence-finally and cannot appear anywhere else in the sentence if they are to form a discontinuous NP structure. The reason for this limited distribution requires further investigation. . Note that, as will become clearer later in the text, we are still assuming a hierarchical phrase structure for Turkish, similar to Japanese as discussed by Baker (2001), and not a purely flat structure. But we aim to capture the properties of Turkish typically associated with nonconfigurational languages, such as free word order and argument drop, based on the absence of case-driven Agree in Turkish.
Non-configurationality: Free word order and argument drop in Turkish
null arguments. Based on the (non-)configurationality parameter introduced in section 2, section 3 will discuss why free word order is possible in languages like Turkish as opposed to the case observed in languages like English. Section 4 will aim to explain how null arguments are licensed in non-configurational languages. Section 5 concludes. 2. Non-configurationality In the literature, two conditions have been proposed for a predicative NP to become a syntactic argument: (i) assignment of referentiality by a functional category, e.g., by an overt or covert D (Longobardi 1994), and (ii) case checking to make an NP visible for theta role assignment by a functional category e.g., by T or v (Chomsky 1995). Hence, an NP should be checked for case and referentiality in order to become an argument. These two conditions have been proposed independently in the literature. However, it is a well-known fact that there are many languages without determiners where case morphology imposes certain interpretations for NPs, such as definiteness and specificity. In these languages case plays an important role in determining the referentiality of nouns, as illustrated by Turkish in (4) below (Enç 1991). Furthermore, historically, it has also been claimed that the development of article systems follows the deterioration of case systems (Abraham 1997; Osawa 1998, 2000). This implies a direct correlation between referentiality and case. (4) a.
Ali kitap okudu b. Ali kitab-ı okudu Ali book read Ali book-acc read ‘Ali did book-read.’ (non-referential) ‘Ali read a certain book.’ (definite)
In the following, we will argue that case and referentiality are co-dependent features in syntax as they need to be associated with each other to establish the argument status for NPs. With respect to non-configurationality we will propose that the domain where argumenthood is established via associating case and referentiality features will have a direct effect on defining phrase structure either as configurational or non-configurational. 2.1 Case and referentiality in English The standard view on case and referentiality assignment, which is assumed to hold cross-linguistically, is that case and referentiality are assigned by certain specified functional categories, e.g., v and T for case (Chomsky 1995) and D for referentiality (Longobardi 1994). English presents strong evidence for the standard view. Firstly, it is a language with overt morphological determiners. This argues for the presence of DPs. Secondly, it exhibits syntactic phenomena which involve casechecking relations between DPs and the functional categories T and v, such as
Balkiz Öztürk
raising, passive and unaccusative constructions, and ECM. Thus, in compliance with the standard view, D in English introduces the referentiality feature to NPs and case features are introduced by T and v. In English, NPs are merged into their theta positions already embedded within DPs. That is, NPs are introduced associated with a referentiality feature in their theta positions and therefore fulfill the first requirement on argumenthood. Then, to meet the second requirement on argumenthood, namely case, they undergo a checking relation with higher functional heads such as T and v via Agree. The configuration in (5) implies that in English case and referentiality features are introduced in different domains: referentiality features in the theta domain and case within the domain of a higher functional projection, i.e., T and v. That is, argumenthood for an NP is established by forming an Agree chain between the theta position and the case position. Note that in (5) we assume a Neo-Davidsonian structure for English, as proposed by Lin (2001) and Borer (2005), where verbs do not bear any information regarding their argument structure in the lexicon, but arguments are introduced via functional heads: (5)
TP T' T [+case]
vP v' v [+case]
Agree
AgentP
Merge → DP[+ref] Agent
Agree
Agent' ThemeP
Merge → DP[+ref] Theme
Theme' VP V
What is crucial at this point is that argumenthood establishment in English follows not only the theta hierarchy à la the Neo-Davidsonian model but also the
Non-configurationality: Free word order and argument drop in Turkish
hierarchy between TP and vP. TP and vP are strictly hierarchically ordered with respect to one another, which implies a configurational structure. Thus, in English establishment of argumenthood is accomplished in a configurational manner. 2.2 Case and referentiality in Turkish In this section, we will focus on the functional categories associated with case and referentiality features in Turkish and to what extent Turkish complies with the standard view. Let us start with the question of whether it is possible to have a DP as the source of a referentiality feature in Turkish given that Turkish lacks morphological determiners of the type found in many Romance and Germanic languages. Demonstratives, quantifiers and the numeral bir “one” used as a marker of indefinite 3 ness do not qualify as functional heads in Turkish – a strictly head-final language4 - since . Crisma (1997) and Longobardi (2000) state that there are no languages which lack a definite article but have an indefinite one. Yet it is very common for languages to lack the indefinite but have the definite article. Given that Turkish does not have a definite article, to assume that Turkish has an indefinite in the absence of a definite article would make Turkish highly exceptional in terms of language typology. Furthermore, Öztürk (2004, 2005a) provides further evidence for the non-determiner status of bir ‘one’ in Turkish. bir can either precede or follow adjectives and she shows that the preadjectival bir is a numeral and acts as a VP adverb, whereas the postadjectival bir is adjectival in nature. See Öztürk (2004, 2005a) for details. . Note that although Turkish allows for free word order when arguments bear overt case morphology, it enforces strict SOV order canonically if no case marker is available. This suggests that the other word orders are derived from the canonical SOV word order in Turkish, which points to a verb-final phrase structure. (See Kelepir (1996) for the inapplicability of a Kayneian approach for Turkish): (i) Huzur mutluluk ver- ir peace happiness give-aorist ‘Peace gives happiness.’ (ii) Mutluluki huzur ti ver- ir happiness peace give-aorist a. ‘Happiness gives peace.’ b. *‘Peace gives happiness.’ In addition to the head-final VP domain, NPs, PPs, APs are also head-final in nature: (iii) NP → PP N Terör-e karş ı öfke *Öfke terör-e karş ı terror-dat against rage rage terror-dat against ‘rage against terrorism’ (iv) PP → NP P Ali için *için Ali Ali for for Ali ‘for Ali’ (v) AP → PP Adj Terör-e karşı öfke-li *Öfke-li terör-e karşı terror-dat against rage-with rage-with terror-dat against ‘in rage towards terrorism’
Balkiz Öztürk
they are all prenominal.5 Due to the head-final sturcture, it is also not possible to diagnose any head movement (N-to-D), which is taken to be the evidence for the DP projection in languages like Italian (Longobardi 1994).6,7 The distributional restrictions exhibited by bare nouns in languages like Italian are also used to motivate an empty D head. If it is assumed that there are empty determiners in Turkish, then bare nouns in Turkish should also exhibit distributional restrictions. However, this is not the case in Turkish, since bare nouns are totally acceptable in any syntactic position in Turkish, and are interpreted as arguments as seen in (1) above.8 Thus, as Turkish does not exhibit any morpho-syntactic evidence for the presence of an abstract DP, we conclude that there cannot be any overt or covert DPs in Turkish. This raises the question of what takes care of referentiality features in Turkish.9 Chierchia (1998) argues that bare nouns in determinerless languages can be associated with kind, definite and existential readings via covert type-shifters (cf. Ramchand & Svenonius, this volume for a critical discussion of this proposal). The application of covert type-shifters can only be blocked if there are overtly
Furthermore, as Turkish is an agglutinative language, all functional heads which are affixal or clitic in nature appear strictly head-finally. Hence, I assume Turkish to be a strictly head-final language. Also see Taylan (1984) and Özsoy (1998) for arguments for the head-final nature of Turkish phrase structure. . Demonstratives, numerals and quantifiers in Turkish fail the tests for determinerhood proposed by Fukui (1995), as they cannot close off functional projections by occupying a specific position, but can be ordered freely with respect to other NP-modifiers, such as adjectives and relative clauses. They behave as adjectivals à la Giorgi & Longobardi (1991). See Öztürk (2004, 2005a) for details. . Note that unlike N-to-D, there is ample evidence for V-to-T in Turkish, although it is a strictly head-final language as widely discussed in the literature (Kural 1993; Kelepir 2001; Aygen 2002). Negation as a suffix and the question particle as a clitic appear on the verbal complex and their scopal properties help to diagnose V-to-T. Furthermore, presence of object shift also implies V-to-T in compliance with Holmberg’s generalization (Holmberg 1986). . Note that Kornfilt (1995, 2007) and von Heusinger & Kornfilt (2005) assume N-to-D movement. However, they do not provide any syntactic motivation for such a movement. . Even non-referential bare nouns in Turkish, which are not marked with overt case-markers and which tend to occur immediately preverbally (see note 3), can scramble depending on the information structure of the sentence. Thus they do not exhibit distributional restrictions, unlike Italian bare nouns with an unfilled D head. This clearly argues against having an empty D head in their structure. See Uygun (2006) for details of this piece of data in Turkish: (i) a.
Ben kahve iste- me- di- m I coffee want-neg-past-1ps ‘I did not ask for coffee.’
b. Kahve ben iste-me-di-m c. Ben iste-me-di-m kahve.
. For a more detailed discussion of the lack of DP in Turkish, see Öztürk (2004, 2005a).
Non-configurationality: Free word order and argument drop in Turkish
expressed morphological type-shifters such as overt articles that bear the same interpretation provided by covert type-shifters. Given that there are no morphological determiners in Turkish, it is expected that bare nouns are freely associated with any of the covert type-shifters. However, this is not the case in Turkish, as shown in (6). Case-marked bare nouns only allow definite or kind readings. This implies that there must be overt functional categories which block the application of covert type-shifters: (6) a.
Ali çocuğ-u gördü b. Ali dondurma-yı sev-iyor10 Ali child-acc saw Ali ice-cream-acc love-prog ‘Ali saw the/*a child.’ ‘Ali loves ice-cream.’
Turkish has been recognized in literature as a language where overt case morphology has a direct effect on the referentiality interpretation of nouns (Nilsson 1986; Enç 1991; Kornfilt 1995; Aygen 1999, 2002; Kelepir 2001). The fact that in a determinerless language like Turkish, case exhibits a highly systematic interaction with referentiality immediately raises the question of whether case can function as the counterpart of determiners found in other languages. Accusative case in Turkish denotes specificity as in (4b) and definiteness as in (6). Only in the case of accusative morphology can a noun be referentially used; that is, it can introduce a discourse referent, which can act as an antecedent for a pronoun (Ghomeshi 1999). This is illustrated by the contrast between (7a) and (7b): (7) a. *Ali kitap oku- du. Reng-i kırmızı-ydı. Ali book read-past color-3ps red-past ‘Ali did book reading. It was red.’ b. Ali kitab-ı oku- du. Reng-i kırmızı-ydı. Ali book-acc read-past color-3ps red-past ‘Ali read the book. It was red.’
(non-referential)
(referential)
Oblique cases also force a referential interpretation (Aygen 1999, 2002). As seen in (8a), the ablative case strictly implies a definite singular reading, which is referential. Only in the case of a generic operator, such as the aorist in (8b), is it
. As pointed out by one of the reviewers, it is possible to form the construction in (6b) without the accusative case marker, suggesting that both constructions mean the same. However, I take the construction without accusative as an instance of complex predicate formation in Turkish, where the non-case-marked NP is part of the verbal complex, forming an intransitive construction. Due to space limitations, I cannot elaborate on such an analysis here but see Öztürk (2004, 2005a, in press) for details.
Balkiz Öztürk
possible for oblique case-marked nouns to be interpreted as generics, which are also considered to be referential (Carlson 1977):11 (8) a.
Ali köpek- ten kork- tu b. Ali dog- abl afraid- past ‘Ali got afraid of the dog.’
Ali köpek- ten kork- ar Ali dog- abl afraid-aorist ‘Ali is afraid of dogs.’
The presence of a genitive case marker also effects the interpretation of a noun, as seen in (9). Nouns marked with genitive are interpreted as referential at the NP level as in (9a), whereas the absence of genitive implies a non-referential reading as in (9b): (9) a.
Kalem-in kutu-su b. Kalem kutu-su pencil-gen box-3ps pencil box-3ps ‘The box of the pencil’ ‘pencil box’
The discussion above highlights the fact that case in Turkish systematically interacts with the referentiality interpretation of nouns. Note that there is no morpho-syntactic evidence for a DP in Turkish, yet it is not possible for NPs to be associated with covert type-shifters either. This implies that there must be an overt functional element blocking the application of covert type-shifters in Turkish. Given the two conditions on argumenthood, i.e., case and referentiality, as well as the historical correlation between the two, we propose that in Turkish, in the absence of DPs, case checking is responsible for referentiality assignment and acts as a type-shifter.12 Note that since case morphology is strictly head-final such a proposal also respects head directionality in Turkish. Locating both case and referentiality features on the same head has immediate implications for the phrase structure. The proposal that case-checking amounts to referentiality checking in Turkish implies that the two features necessary for argumenthood are introduced within the same domain, namely, within the theta domain by the theta role-introducing functional heads. This also implies that there will not be any case-driven Agree with higher functional heads, such as T and v in Turkish, unlike
. The following example further highlights how oblique case-markers enforce a referential reading. In (ii) when the object is marked with oblique case, it receives a definite singular reading, unlike (i):
(i)
Aylin dün akşam iki saat çocuk bak- tı Aylin yesterday evening two hour child look-past ‘Aylin baby-sat for two hours last night.’
(ii)
Aylin dün akşam iki saat çocuğ-a bak- tı Aylin yesterday evening two hour child-dat look-past ‘Aylin baby-sat the child for two hours last night.’
. Note that this is parallel to the type-shifting properties of strong case as defined by de Hoop (1996).
Non-configurationality: Free word order and argument drop in Turkish
what we see in English. That is, if case is available in the theta domain, T and v should be disassociated from case features in Turkish.13 In the following, we will very briefly take a look at the role of v and T relative to case-checking in Turkish phrase structure in order to understand their role in the non-configurational nature of the language. 2.2.1 vP in Turkish Chomsky (1995) introduces v as the locus of object case-checking. v is also assumed to be the head facilitating the introduction of an external argument into the structure. The idea behind such a configuration follows from the generalization proposed by Burzio (1986). Thus, in languages whose phrase structure includes a vP projection, Burzio’s generalization is respected. Furthermore, it is possible to diagnose vP as a syntactic constituent, and vP exhibits phasehood (cf. Chomsky 2000 et seq.). Due to the active case feature of v, it is not possible in such languages to leave more than one argument VP-internally, as observed by Alexiadou & Anagnastopoulou (2001). Turkish, however, fails to pattern along with languages of this type with respect to vP. First of all, pseudo-incorporation of agents in Turkish challenges Burzio’s Generalization, representing a construction in which the pseudo-incorporated agent is incompatible with control and agent-oriented adverbs, as seen in the contrast between (10a) and (10b), thereby failing the tests for external argumenthood, despite the presence of accusative case on the object. Note that the counterpart of (10b) with a non-incorporated referential subject is fully grammatical as in (10c):14 (10) a.
Ali-yi polis tutukla-dı Ali-acc police arrest- past ‘Police-arresting happened to Ali.’
(Meaning: Ali got arrested.)
b. *Ali-yi [eci sorgula-mak için] kasıtlı.olarak polisi tutukla-dı Ali-acc interrogate-inf for intentionally police arrest-past ‘Police-arresting happened intentionally to Ali to interrogate him.’
. Note that vP is considered to be the node where both the object case and the external argument are introduced in parallel to Burzio’s generalization (Chomsky 1995). However, given the rich literature against Burzio’s generalization (Harley 1995; Pylkkänen 2002; Markman 2003), we dissociate vP from its theta-role related features. That is, we do not view v as the head which introduces the external argument, but only as a case assigner. For the arguments against Burzio’s generalization in Turkish, see section 2.2.1. . Note that one of the reviewers pointed out the possibility that polis “police” in (10) is a true external argument, which has undergone head-incorporation. However, this would make Turkish highly exceptional, as it would imply incorporation of agent arguments, which is strictly banned under Baker (1988). See Öztürk (2004, 2005b, in press) for a pseudo-incorporation analysis of this data.
Balkiz Öztürk
c.
Policei Ali-yi [eci sorgula-mak için] kasıtlı.olarak tutukla-dı police Ali-acc interrogate-inf for intentionally arrest-past ‘The police intentionally arrested Ali to interrogate him.’
Second, it is not possible to diagnose vP as a constituent either via vP-fronting à la Huang (1993) or via ellipsis. As seen in (11), arguments of the verb can be stranded under ellipsis: (11) Ben kitab-ı Ahmed-e ver-eceğ- im, Mehmet de gazete-yi öyle yap-acak I book-acc Ahmet-dat give-fut-1ps Mehmet too paper-acc so do- fut ‘I will give the book to Ahmet and so will Mehmet the newspaper.’
Third, Turkish fails to obey the constraint proposed by Alexiadou & Anagnostopoulou (2001), which suggests that due to the active case feature on the v head, vP can contain no more than one argument with an unchecked case feature by Spell-Out. Based on the scope facts in (12a), where negation introduced right above the theta domain unambiguously scopes over the subject,15 it is seen that more than one argument can be left in situ in its theta position. 16 The fact that a high adverb cannot follow the subject also supports that the subject has not left its theta position:17 (12) a. [CP[TP[NegP[AgentP bütün çocuk-lar (*/?Allahtan)[ThemePo test-e all child-pl luckily that test-dat [VPgir-me- di] take-neg-past ‘Luckily, all children did not take that test.’ (*all > not , not > all) b. [TP Bütün çocuk-lari Allahtan [NegP[VP ti o test-e gir- me- di- ler]]]18 all child-pl luckily that test-dat take-neg-past-pl ‘Luckily, all the children did not take that test. (all > not, *not > all)
Kornfilt (1990), on the other hand, provides the following examples to argue for an identifiable VP level for Turkish, which might imply an identifiable vP level under the current terminology: . Aygen (1998) and Kelepir (2001) assume that negation in Turkish is introduced by NegP projected directly above VP, which includes all arguments of the verb. . This test has been adopted for Turkish from Miyagawa (2001). . Note that some speakers get the narrow scope reading for ‘all’ even in the presence of the plural agreement. However, all my informants agree that in the absence of the plural morpheme, when the subject follows the high adverb, the only reading is where the subject takes narrow scope. . As the discussion implies, the motivation for movement into SpecTP in Turkish is neither case, nor EPP. As widely noticed in the literature, sentence-initial position is reserved for referential elements, which are to be interpreted as old information in Turkish (Taylan 1984). Özsoy (2005) shows that Topic feature in Turkish percolates down from C-to-T and thus is checked by the phrase, which occupies SpecTP. Öztürk (2004, 2005a) proposes that movement into SpecTP is related to check such a topic feature, but not for case or EPP (cf. also Modesto, this volume).
(13) a.
Non-configurationality: Free word order and argument drop in Turkish
Ben kitab-ı Ahmed-e ver-eceğ-im, Mehmet de öyle I book-acc Ahmet-dat give- fut-1ps Mehmet and thus ‘I will give the book to Ahmet and so will Mehmet.’
b. *Ben kitab-ı Ahmed-e ver-eceğ-im, Mehmet de I book- acc Ahmet-dat give- fut -1ps, Mehmet and
gazete- yi öyle newspaper- acc thus
‘I will give the book to Ahmet and so will Mehmet the newspaper.’
Kornfilt (1990) claims that the contrast in (13) argues for a VP level, which includes the verb and the object to the exclusion of the subject. She argues that, since it is not possible to strand the object in (13b), the verb and the object form a constituent. Her example can be used to argue for a vP in the current framework. (13a) can be analyzed as a case where the vP has been substituted after the external argument has moved into SpecTP. This suggests that the verb and the object form a constituent. In (13b), on the other hand, leaving the object stranded leads to ungrammaticality. However, the grammaticality judgments presented by Kornfilt for (13) do not follow from the verb’s forming a constituent with the object but from the fact that the adverb öyle cannot be used by itself without the verb yap- “do”. Therefore, both of the examples in (13) above are equally bad, as suggested by the grammaticality judgments of native speakers of Turkish. That is, (13a) should also be starred. Providing the verb yap- resolves the ungrammaticality as seen in (14): (14) a.
Ben kitab- ı Ahmed- e ver- eceğ- im, Mehmet de öyle I book- acc Ahmet- dat give- fut- 1ps Mehmet and thus
yap-acak do- fut
‘I will give the book to Ahmet and so will Mehmet.’
b. Ben kitab- ı Ahmed-e ver- eceğ- im, Mehmet de I book- acc Ahmet-dat give- fut - 1ps Mehmet and
gazete- yi öyle yap-acak newspaper- acc thus do- fut
‘I will give the book to Ahmet and so will Mehmet the newspaper.’
Thus, her examples do not necessarily point to the impossibility of object stranding. On the contrary, object stranding is totally acceptable in Turkish unlike in English, as illustrated by (11) above, where it is perfectly possible to strand the object, when only the verb is substituted. Thus, we can very confidently conclude that there is no identifiable vP layer in Turkish. Furthermore, Turkish fails to exhibit movement of accusative subjects into the matrix vP position in ECM constructions. As seen in (15b), the embedded subject marked accusative cannot appear to the left of the adverb phrase, which bears the
Balkiz Öztürk
topic marker ise, marking the left-edge of the embedded clause as the topic phrase. See Öztürk (2005b) for discussion of the lack of movement and case-checking with matrix v in ECM constructions in Turkish: (15) a. Cem [bu gece ise Ayşe- yi gel-ecek] san- ıyor Cem tonight as.for Ayşe- acc come- fut think- prog “Cem thinks that as for tonight Ayşe will come.” b. *Cem [Ayşe- yi bu gece ise gel-ecek] san- ıyor Cem Ayşe- acc tonight as.for come- fut think- prog(Gür Oded 2002)
As shown above, Turkish behaves quite differently to languages with a vP projection, implying that either the constraints and observations discussed above are wrong, or that there is something peculiar about the phrase structure of Turkish.19 Given that the above-mentioned constraints are respected by languages said to have vPs, the only way to resolve the dilemma introduced by Turkish is to suggest that there is no Agree relation between an object and v in Turkish and that this is because accusative case is provided not by v, but in situ by the head of ThemeP. In this way, Turkish will cease to be an exception. If we take a step further, given the discussion above, we can argue for eliminating the vP projection totally from the phrase structure of Turkish, implying that the presence versus absence of vP is subject to parametric variation. This is the view we will adopt. In brief, the discussion above argues against the presence of vP as a caseassigning functional projection in the phrase structure of Turkish, as Turkish does not exhibit any motivation for a vP level. That is why Turkish is immune to several restrictions which lead to ungrammaticality in languages with a vP. Thus, we conclude that Turkish does not have a vP and that accusative case is assigned in situ by the theta role-introducing functional heads, Theme and Agent. 2.2.2 TP in Turkish If it is not possible to argue for the presence of v in Turkish, then what is the status of T as the other potential case-provider under the standard view? The presence of T as the locus of the Tense feature in Turkish is almost uncontroversial (Kural 1993; Aygen 1998, 2002; Özsoy 2002; Kelepir 2001; among others). It has generally been assumed that T in Turkish is associated both with the nominative case feature as well as the feature EPP, which requires XP movement into SpecTP (Kornfilt 1984, 2002; . Furthermore, Turkish is incompatible with the tests for defining phasehood introduced by Legate (2003), namely, (i) reconstruction of wh-phrases to the vP edge, (ii) quantifier raising in ACD constructions targeting the vP edge and (iii) parasitic gaps licensed at the vP edge. Turkish is a wh-in situ language, and does not have ACD constructions of the type observed in English (Özsoy 1996). It does not obviously exhibit parasitic gaps either, being an object drop language as shown by İnce (2004). Thus, these tests are irrelevant for Turkish.
Non-configurationality: Free word order and argument drop in Turkish
Özsoy 2002; Aygen 1998, 2002; Kelepir 2001). However, Öztürk (2005a) shows that movement into SpecTP is not obligatory in Turkish, which argues against checking the EPP feature via XP movement. As seen in the raising construction given in (16a), movement into matrix TP is not obligatory. When movement takes place, this triggers overt subject-verb agreement at the matrix level as seen in (16b): (16) a. [TP _____bana [sen üzül- müş- sün] gibi görün-üyor] me you upset- past- 2ps as seem- prog ‘It seems to me that you got upset’ b. [TP Seni bana [ ti üzül- müş- sün] gibi görün-üyor- sun]20 you me upset- past- 2ps as seem- prog- 2ps ‘You seem to me to be upset’
Example (12a) above also highlights the fact that movement into SpecTP is not obligatory in regular transitive constructions in Turkish. Such movement is allowed to create a change in scope facts as shown in (12b). When the quantified subject moves into TP, negation fails to scope over the subject. This movement also triggers overt agreement on the verb. Passives and unaccusatives exhibit a similar behavior; highlighting the same point that movement into SpecTP is not obligatory in Turkish, as shown by the scope facts in (17) and (18) respectively: (17) a. [TP[NegP[AgentP[ThemeP bütün çocuk-lar] [VPçağr]-ıl]- ma]-dı] all child- pl invite-pass-neg-past ‘Not all children were invited (but some were).’ (*all > not, not > all)
. Note that once raising happens, the agreement on the embedded verb seen in (16b) can be dropped as in (i). However, this option is possible only if raising happens otherwise the structure is ungrammatical as shown in (ii): (i) [tp Seni bana [ ti üzül-müş] gibi görün-üyor-sun] you me upset-past as seem-prog-2ps ‘You seem to me to be upset’
(ii) *[tp ______bana [sen üzül-müş-sün] gibi görün-üyor-sun] me you upset-past-2ps as seem-prog-2ps One of the reviewers points out that “under certain intonation patterns, especially if a break is introduced after the embedded subject, example (ii) improves,” as this might imply that the subject has raised when the indirect object is scrambled to its left. In return, parallel to this case, (16a) might also be considered to be a construction where the embedded subject is in the matrix subject position following a scrambled indirect object. However, given the lack of agreement on the matrix verb in (16a) and the fact that matrix SpecTP is not a theta-position, enforcing such a movement seems unnecessary and highly marked, as it requires certain intonation patterns as pointed out by the anonymous reviewer.
Balkiz Öztürk
b. [TPBütün çocuk-lari [NegP[AgentP[ThemeP ti][VPçağr]- ıl]- ma]-dı- lar] all child- pl invite-pass-neg-past-pl ‘All children were such that they were not invited.’ (all > not, *not > all) (18) a. [TP[NegP[ThemeP Bütün çocuklar] [VPgel]-me]-di] all children come-neg-past ‘All children did not come (but some did).’
(*all > not , not > all)
b. [TPBütün çocuk-lari [NegP[ThemeP ti][VPgel]-me]-di-ler] all child-pl come-neg-past-pl ‘All children were such that they did not come.’ (all > not , *not > all)
The fact that impersonal passives derived from unergatives do not involve expletives as shown in (19) and that there are no subjects in pseudo-incorporation of agents as seen in (10) also provides further evidence against filling up SpecTP by an XP category. (19) Koş-ul- du run-pass-past ‘Running happened.’
As shown above, unlike languages like English, Turkish does not exhibit obligatory movement into SpecTP. Within the GB framework, movement into SpecTP is not solely for EPP reasons but also for case, since the T head is assumed to be the case assigner of nominative. However, if considered within the GB framework, Turkish does not exhibit any motivation for movement into SpecTP for case purposes. That is, if there is no case-driven A-movement into SpecTP in GB terms, then the T head cannot be the locus of the nominative case feature. If we extend this GB notion to our current framework, it implies that there cannot be a case-driven Agree relation between the T head and the subject in the theta position, since the T head does not host the case feature. Such an interpretation will be in accordance with what we have been assuming for the locus of case-checking in Turkish, that is, it is done in theta positions via the theta role-introducing functional heads bearing both case and referentiality features. Therefore, we conclude that just like the v head, T also does not play a role in case feature checking.21,22
. Alexiadou & Anagnostopoulou (1998) argue that in pro-drop languages with rich verbal agreement, head movement of the verb can check the EPP. That is, EPP checking does not require MERGE XP, but MOVE X0. Öztürk (1999, 2002) has also independently proposed that in Turkish verbal agreement satisfies the EPP. . In passives in Turkish, similar to English, the object no longer bears accusative, but nominative case. If there is no case-driven Agree with T in the passive, the question is why there is such a case alternation. However, it has often been noticed that case-morphology is independent of case-checking (Kuroda 1988; Miyagawa 1991; Marantz 1991; Harley 1995). Kuroda (1988)
Non-configurationality: Free word order and argument drop in Turkish
2.2.3 Summary As seen in the discussion above, we propose that there is no case-driven Agree with higher functional projections such as vP and TP in Turkish. Turkish lacks vP in its phrase structure, and T also does not play a role in introducing a syntactic case feature. Case, which encodes referentiality features, is introduced within the thematic domain via theta role-introducing functional heads. Thus, a typical transitive construction in Turkish should be represented as follows: (20) a.
Çocuk kitab- ı oku-du. child book-acc read- past ‘The child read the book’
b. ∃e[read (e) & Theme (book) (e) & Agent (child) (e)] c.
TP T' AgentP
T readi
MERGE NP → subject child-
Ag' Ag [+Case, +Ref] ti
ThemeP MERGE NP → object book-
Theme'
VP
Theme [+Case, +Ref] ti V'
V ti
shows that case in Japanese is not sensitive to phrase structure, but to surface order. That is, in Japanese morphological realization of case takes place left to right, following a nominative > dative > accusative order. The same phenomenon has been observed in Turkish as well (Aissen 1974; Zimmer 1976; Taylan 1984; Göksel 1993, among others). Therefore, the fact that there is such morphological case alternation in passives does not necessarily argue for movement into and case-checking with the TP projection. The object as the sole argument NP does not leave its theta position in passives, but gets marked with the mandatory nominative morphology after syntax at the PF level. See Öztürk (2004, 2005a) for details.
Balkiz Öztürk
Note that theta role-introducing functional heads bear the two crucial features for argumenthood, namely, case and referentiality. That is, NPs acquire full argument status in their theta positions without recourse to case-driven Agree. This is a crucial difference between languages like Turkish and English, which has immediate implications for non-configurationality. 2.3 The non-configurationality parameter In the structure in (20c) above, the association of the lexical verb with syntactic arguments introduced by theta role-introducing functional heads is achieved via verb movement. This establishes event identification within the Neo-Davidsonian model. In this model, all theta roles count as equal in terms of their relation to the event structure as represented in (20b) above. This also implies that all theta roles have the same status with respect to their relation to the verb. In Turkish, given that NPs acquire full argument status in their theta positions without Agree with higher functional projections, they remain in their theta positions and, thus, have equal relations with the verb in compliance with the Neo-Davidsonian model. When the verb moves to a higher functional projection such as T, even though theta role-introducing functional heads observe the theta hierarchy, all arguments count as equidistant from the V/T complex because they all have an equivalent relation to the verb.23 This is what causes the “flat phrase structure” effect observed in Turkish, which is taken to be a case of non-configurationality. TP
(21) Spec
T' Theta Domain T
θ1……..θ2……θ3 ← Flat Structure Effect [+case][+case][+case]
In languages like English, however, it is not possible to talk about such equidistance for arguments. In English, too, theta roles are introduced by independent functional projections within the Neo-Davidsonian model. However, argumenthood for NPs is not established in the theta domain, since only the referentiality
. V-to-T is a well-established phenomenon in Turkish phrase structure (Kural 1993; Aygen 2002).
Non-configurationality: Free word order and argument drop in Turkish
feature is available in this domain. NPs in English must form an Agree chain with T and v to get case and thus satisfy the two requirements on argumenthood. The crucial point here is that T and v as functional categories bearing case features are strictly hierarchical (TP > vP). Forming chains with these heads to establish argumenthood causes arguments such as subject and object to be ordered hierarchically with respect to each other. Even in languages where the verb can raise to T, unlike English, if case-driven Agree obtains, arguments of the verb cannot be equidistant to TP since they would be bound in Agree-chains formed with hierarchically ordered functional heads. Thus, what creates the configurational structure in such languages, as well as in English is the obligatoriness of case-driven Agree chains with higher functional heads. The general idea behind this is that if in a language case and referentiality are checked within the domain of a single projection, that is, if there is no case-driven Agree with higher functional heads, it exhibits a non-configurational phrase structure. However, if there is case-driven Agree with hierarchically ordered functional projections, then the language is strictly configurational. Note that in section 2.2.1 we have proposed that Turkish lacks vP as the functional projection responsible for case assignment. This is also very similar to the proposal made by Hale (1980) that non-configurational languages lack the VP node. In the absence of a VP node all arguments can c-command each other, thus, a flat structure effect is created. Therefore, there is no hierarchical difference among arguments of a verb. The account proposed here, on the other hand, goes a step further and presents an explanation for the intuition behind Hale’s (1980) proposal. Lack of an identifiable VP (or in our terms vP) node follows from the absence of case-checking relations with higher functional heads. This variation in the establishment of argumenthood between English and Turkish implies that case-driven Agree with higher functional projections is not a universal pattern in the establishment of argumenthood. Languages can vary based on the functional categories they have associated with case and referentiality features, as well as based on the domain in which these features are checked. We therefore have a point of parametric variation. More specifically, we would like to propose the following non-configurationality-related parameter (see also Öztürk 2004, 2005a): (22) Languages without case-driven Agree are non-configurational. Languages exhibiting case-driven Agree, on the other hand, have a configurational phrase structure.
In accordance with (22), whether a language will display configurational or nonconfigurational properties can be anticipated based on what functional categories and
Balkiz Öztürk
what XP domains are available in the introduction of case and referentiality features for argumenthood.
3. Free word order The parameter based on the presence or absence of case-driven Agree proposed above has immediate implications for the (un)availability of free word order in a language. Going back to Turkish, free word order is a result of the “flat structure” effect yielded by movement of V to T, in the absence of vP. Given the flat structure depicted in (21) above where all arguments have an equal relation to the verb, once the verb moves to T, any argument can be attracted to SpecTP, since this movement makes arguments equidistant to TP.24 Thus, even though arguments are hierarchically ordered following the theta hierarchy, this hierarchy will be overridden by V-to-T movement. However, if there is case-driven Agree with the hierarchically ordered functional projections (TP > vP), movement into SpecTP will only be possible for the element agreeing with T, as the hierarchy between the case-checking functional categories will override the flat structure effect created by the theta domain as in (23).25 That is, the element which agrees with v will be defined as lower in the hierarchy compared to the one which establishes an Agree relation with T. Note that movement of V-to-T is not crucial here. What matters is the hierarchy established in case-checking relations with respect to T and v. Therefore, whether there is V-to-T as in the case of French, or not, as in English, will not change the movement possibilities.
. As discussed in Öztürk (2004, 2005a), scrambling in Turkish exhibits A-movement properties, targeting SpecTP as the landing site, as illustrated by the unambiguous example (ii) below. We suggest that such movement is possible due to the flat structure effect created by V-to-T movement:
(i)
Her çocuk bir kitab-ı okudu every child one book-acc read
a. ‘Every child read a specific book.’ b. ‘Every child read a different book out of a definite set of books.’
(ii)
Bir kitab-ıi her çocuk ti okudu one book-acc every child read ‘Every child read a specific book.’
(SOV) ∃>∀ ∀>∃ (OSV) ∃ > ∀ (*∀ > ∃)
. Note that Boeckx (2008) proposes that there is no sole Agree. That is, Agree always requires movement of the agreeing category.
Non-configurationality: Free word order and argument drop in Turkish
(23)
TP T'
Spec T [+case]
vP Spec
v' v [+case]
Agree
Agree
Theta Domain θ1........θ2
Thus, whether a language will have free word order or not will again be dependent on how argumenthood is established in a language. If argumenthood is established in situ in the theta domain without recourse to case-checking with higher functional heads such as v or T, then any argument can be scrambled into SpecTP if V-to-T movement takes place. However, if argumenthood is established following the hierarchy between vP and TP via forming case-chains, a strictly configurational phrase structure results. Consequently, scrambling will not be possible. 4. Null arguments The parameter proposed for non-configurationality also has implications for the null argument phenomenon observed in Turkish. Turkish exhibits both subject and object drop extensively, as illustrated in (24). However, there is agreement morphology only for subjects, but not for objects. Example (25) presents one instance of a six-way subject agreement paradigm. (24) a. (Ben) (onu) sev- di- m I him/her love- past-1ps ‘I loved him/her.’ (25) a. (Ben) gid-iyor- um c. (O) gid-iyor-∅ I go- prog-1ps s/he/it go-prog.-3ps ‘I am going.’ ‘He/she/it is going.’ b. (Sen) gid-iyor-sun d. (Biz) gid-iyor- uz you go-prog-2ps we go-prog- 1ppl ‘You are going.’ ‘We are going.’
Balkiz Öztürk
e.
(Siz) gid- iyor- sunuz f. (Onlar) gid-iyor-lar you go- prog- 2ppl they go-prog.-3ppl ‘You are going.’ ‘They are going.’
In the literature, the availability of pro-drop in languages like Italian and Spanish is correlated with the existence of a rich agreement system, which helps to recover the content of the empty argument specified with the features [+pronominal, –anaphor] (Taraldsen 1978; Chomsky 1981; Jaeggli 1982; Borer 1983, 1986; Huang 1984; Rizzi 1986; among many others). Based on the subject agreement paradigm illustrated in (25), which is assumed to recover the content of null subjects, Turkish has been analyzed as a pro-drop language on a par with languages like Italian and Spanish (Kornfilt 1984; Taylan 1986; Enç 1986; Özsoy 1988; Sezer 1991). In the case of subject drop in Turkish, the empty element, which occurs in the subject position, clearly exhibits pronominal features. This is illustrated in (26): (26) Johni [eci/j Mary-i sev- diğ- in] -i biliyor John Mary-acc love-nom-3ps-acc knows ‘Johni knows that hei/j loves Mary.’
As seen in (26), the empty category can, on a par with pronominal elements, either be associated with the matrix subject or with someone else already introduced into the discourse. Therefore, it is assumed that the empty subject is pro in Turkish. Both Kornfilt (1984) and Özsoy (1988) argue that pro is licensed via agreement introduced by AgrP. Thus the licensing conditions for pro in Turkish are assumed to be identical to those in Italian as defined by Rizzi (1986): (a) government by a head category such as Infl, and (b) recovery of the content of pro by the agreement specifications on the governing head. However, as discussed in detail by Öztürk (1999, 2002), subject-drop does not always require the presence of verbal agreement in Turkish. This is observed in adjunct clauses that do not allow any verbal agreement morphology:26 (27) a.
Ben gel- ince, Ahmet gid-ecek I come-when Ahmet go- fut ‘When I come, Ahmet will go.’
. Note that we are not considering the possiblity of analyzing the null subjects in adjunct clauses as PRO (see Kornfilt 2003a for an analysis along these lines). The motivation for considering these null elements as PRO is not convincing, but we cannot go into the details because of space considerations here. Furthermore, given the recent literature in which control is reduced to movement, (Hornstein 1999, Boeckx & Hornstein 2004), constructions which do not fall under obligatory control will also be analyzed as pro.
Non-configurationality: Free word order and argument drop in Turkish
b. pro gül- ünce, karn- ım ağrı- yor laugh-when stomach-1ps ache-prog ‘When I laugh, my stomach aches.’ (28) A: Johni Billj gel- ince mi gid-ecek? John Bill come-when Q go- fut ‘Will John go when Bill comes?’ B: Evet, proj gel- ince, proi gid-ecek Yes come-when go- fut ‘Yes, hei will go when hej comes.’ (29) a. proi/j gel- ince, Alii Ahmetj’le konuş-acak come-when Ali Ahmet-with talk- fut ‘When he comes, Ali will talk to Ahmet.’ b. pro*i/j/k gel- ince, öğrenci-leri Ahmetj’le konuş-acak-lar come-when student-pl Ahmet-with talk- fut- pl ‘When he/*they come(s), the students will talk to Ahmet.t’
As shown (27–28), pro in adjunct clauses is licensed without any overt agreement morphology. Furthermore, the empty element seen in such adjunct clauses may have its own φ-specifications. This is illustrated in (29b), where the null element, imposing a strictly 3rd person singular reading, avoids coindexation with the 3rd person plural matrix subject, but can be coindexed with the 3rd person singular matrix object in the absence of any matching 3rd person singular agreement morphology. This also highlights its pronominal features. Given the data above, pro in Turkish does not have to be licensed via agreement, unlike what has been proposed so far in the literature. However, the question is what syntactic conditions license subject pro in adjunct clauses in Turkish. We will return to this shortly. Similar to null subjects in adjunct clauses, null objects in Turkish are also licensed without any agreement morphology. Then the question is whether null objects can also have pronominal features. Huang (1984) introduces the discourse-dependent argument drop strategy in languages like Chinese, which allows both subjects and objects to be null in the absence of any agreement morphology. This challenges the agreement-based pro-drop analysis for Chinese. As noticed by Huang (1984, 1987) and Xu (1986), empty categories in Chinese exhibit different properties than the null arguments found in languages like Spanish and Italian. Empty objects in Chinese receive a bound variable interpretation, allowing a sloppy identity reading. (30) a.
Meigereni [piping le zijii] everyone criticize perf self ‘Everyonei criticized selfi.’
Balkiz Öztürk
b.
Zhishi Zhangsanj mei piping [ej]. only Zhangsan not criticize = ‘Only Zhangsanj did not criticize selfj’ ≠‘Only Zhangsanj did not criticize everyonei’ ≠‘Only Zhangsanj did not criticize Zhangsanj’
(Huang 1987)
Based on these properties of Chinese empty categories, Huang (1987) proposes that the empty category in Chinese is a variable bound by a null topic operator, which has [+anaphor] properties. Thus, (30a) can be considered to be on a par with (31): (31) Zijij Zhangsanj bu piping [ej] self Zhangsan not criticize ‘Selfj Zhangsanj does not criticize [e]j’
Turkish null objects, on the other hand, do not exhibit the strict anaphoric behaviour that Chinese null objects do. İnce (2004) discusses the pronominal nature of null objects in adjunct clauses in Turkish, which implies that null objects can also be analyzed as a pro category: (32) a.
Ahmeti Hasan eci vur- unca, ağla- ma- ya başladı Ahmet Hasan hit- when cry- inf- dat started ‘Ahmet started crying when Hasan hit him.’
(İnce 2004: 3)
b. Eğer eci orada bırak- ır- sa- n, kedi fare- yii yi-yecek if there leave- aorist- cond- 2ps cat mouse- acc eat-fut ‘If you leave it there the cat will eat the mouse.’ c.
Hasan eci azarla-yınca, herkesi ağla-ma- ya başladı Hasan scold- when everyone cry- inf- dat started ‘When Hasan scolded (them), everyone started to cry.’
As seen in (32a), the null object in the adjunct clause can be coindexed with the matrix subject and is coindexed with the embedded object in (32b). (32c) also highlights the non-variable nature of the empty category since it allows only a collective reading, not a distributive reading. In simplex clauses, too, the null object exhibits pronominal features. As seen in (33a), the empty object cannot be coindexed with the matrix subject since this will be a Condition B violation given the pronominal nature of the object. If the null object occurs in an embedded structure, however, it can be coindexed with the matrix subject, given the right context as in (33b):27
. Meral (2006) argues that null objects in certain syntactic configurations in Turkish can be analyzed as resumptive pronouns which are dependent on an operator and behave as split NPs in accordance with the non-configurational nature of Turkish.
Balkiz Öztürk
pair is not available for null objects in Turkish. Then the question is: Given that null objects have certain pronominal features, but there is no agreement to recover them in the context, what licenses their occurrence in Turkish? This is the same question that we have asked for null subjects in adjunct clauses. If both null subjects and null objects in Turkish are pronominal categories that can occur in the absence of agreement morphology, unlike in Italian and Spanish, then what syntactic mechanism is responsible for their licensing? Italian also allows null-objects with arbitrary interpretation in the absence of object agreement morphology: (36) a.
Questa musica rende [______ allegri] this music renders happy-pl ‘This music makes one happy.’
b. Quale musica riconcilia _____ con se stessi? ‘Which music reconciles with SE oneself ’
(Rizzi 1986: 507)
(Rizzi 1986: 514)
(37) Ne tür müzik ec rahatlat-ır? what kind music ease- down ‘What kind of music soothes one?’
Rizzi (1986) shows that null objects in Italian may co-occur with other overt operators as seen in (36b). This suggests that the null category cannot be a variable licensed by a covert topic operator as in Chinese. If it were, then (37) would be ungrammatical, but that is not the case. Thus, Rizzi (1986) concludes that the null object is a pronominal element, i.e., pro. Note that the same test is applicable to Turkish, and null objects can co-occur with other overt operators as shown in (37). However, the occurrence of object pro does not meet the two conditions that license subject pro in Italian, as defined by Rizzi (1986): (a) government by a head category such as Infl, and (b) recovery of the content of pro by the agreement specifications on the governing head. Neither condition holds for object pro. Note that it cannot be assumed that constructions such as (36) do not project an object since the null object in Italian is syntactically active. As extensively discussed by Rizzi (1986), it can act as a controller, as a binder and as a subject of predication for adjunct and argument small clauses. This again raises the same question as the one we asked for Turkish: What licenses pro in the absence of agreement morphology? Rizzi (1986) proposes that the features defining arb interpretation must be transferred onto the object position. This is done by assigning arb to the direct theta-role in the theta-grid of the verb. If we translate Rizzi’s proposal to the NeoDavidsonian model that we are adopting here – which proposes that theta roles are not part of the lexical representation of verbs, but are introduced via functional heads (Lin 2001; Borer 2005) – the content of pro can only be recovered if argument
Non-configurationality: Free word order and argument drop in Turkish
drop is in situ, that is, in the theta position, as the referential index of pro is assigned to the theta position. This implies that empty elements are always recoverable from the event structure. Based on this, we would like to propose that an argument can only be null if it can remain in its theta position. This immediately captures why it is possible to have null pro in the absence of any agreement morphology as in the case of null objects in Turkish and Italian. Such a proposal, however, has immediate implications for the relationship between agreement and pro-licensing. If being in a theta position is enough to license an empty element, then the presence of agreement morphology ceases to be what licenses pro-drop since event structure will be the licenser.30 This is supported by the possibility of having subject drop in languages like Turkish in the absence of agreement in adjunct clauses. Now let us go back to the correlation between non-configurationality and null arguments. Given that non-configurational languages allow their arguments to remain in theta positions as argued in the sections above, then we would expect to find extensive argument drop in these languages. Thus, argument drop would be one of the inevitable characteristics of non-configurational languages such as Turkish. Alexiadou & Anagnostopoulou (1998) argue that in SVO orders in languages like Italian and Spanish, the subject is not in an A-position but in an A-bar position. It is the pronominal agreement which satisfies the EPP feature on T in such languages, with the result that there is no motivation for subjects to move into this position to check EPP. Furthermore, Alexiadou & Anagnostopoulou (2001) discuss the fact that Italian and Spanish also allow VOS. They show that in this order, the object is in a higher position than the subject. Consider the following example from Italian: (38) Ha visitato [ogni soldato]i suai madre has visited every solder his mother ‘His mother visited every soldier.’
The coindexation possibilities in (38) imply that subjects can be left in their theta positions in Italian. Given the proposal above, it is expected for Italian to allow extensive subject drop. Presence of agreement on the verb could be seen due the EPP feature on T, but not as a trigger of pro-drop. Also note that object drop in Italian is restricted to non-referential objects (Rizzi 1986) and that found in Spanish is only possible for narrow scope indefinites as argued by Campos (1986).
. Note that Romero (1999) also makes a similar proposal for the licensing of null arguments in the absence of agreement morphology in languages like Turkish and Japanese. He assumes that thematic projection is uniform and that this allows recovery of the empty categories. However, he assumes that agreement is still crucial for pro-drop in languages like Italian and Spanish.
Balkiz Öztürk
It can be assumed that these objects have to stay in their theta positions due to their semantic properties. Therefore, it is expected that they should undergo pro-drop, being licensed by the event structure.31 In languages like English, on the other hand, pro-drop is not possible even with expletive constructions with postverbal subjects, which are interpreted as taking narrow scope. Lasnik (1995) argues that such subjects also undergo movement out of the VP domain. This is evidenced by the binding facts as in (39). In order for the subject to bind the VP adjunct, it should raise to a position higher than the VP. This implies that even such subjects cannot remain in their theta positions in English; therefore, pro-drop is not expected: (39) There arrived three meni on each otheri’s horses
If this analysis is on the right track, it can be extended to null topic languages like Chinese too. Given that all arguments can remain in their base positions in Chinese, a purely Neo-Davidsonian language, as shown by Lin (2001), it is expected that all arguments can be dropped. It is true that the feature specifications of the empty categories in Chinese differ extensively from those found in Spanish, Italian and Turkish; however, the underlying reason for the presence of null arguments is the same, namely, their being in situ. Thus, we argue for a unified account of null argument phenomena cross-linguistically which we believe is more advantageous with respect to acquisition, but we accept that the feature specification of empty categories can be language-specific.
5. Conclusion In this study, we mainly focused on two properties of Turkish, i.e., free word order and argument drop, which are typically associated with a non-configurational . As pointed out by one of the reviewers, the immediate question that arises given the discussion of how null arguments are licensed in languages like Italian is: Why do such languages have a configurational phrase structure and fail to allow free word order? As extensively discussed in the literature (Burzio 1986; Alexiadou & Anagnostopoulou 1998, 2001), the case feature in languages like Italian is introduced not within the thematic domain, but in the functional domain. On the analysis proposed here, this therefore implies the presence of case-driven Agree with hierarchically ordered higher functional heads in such languages, and that is why we observe strict word order. The fact that arguments can remain in the thematic domain, though, leads to argument drop, as arguments can be licensed through the event structure. This implies that being able to remain within the thematic domain by itself is a sufficient condition to license argument drop and is independent of the non-configurationality parameter based on case-driven Agree proposed here. This option requires further investigation, which I leave for a future study.
Non-configurationality: Free word order and argument drop in Turkish
phrase structure. We proposed that both properties follow from the lack of casedriven Agree as a mechanism to establish argumenthood in Turkish.32 Specifically, we argued that argumenthood based on the two syntactic features, case and referentiality, is established in the theta domain in languages like Turkish as both features are assigned by the same theta role-assigning functional categories. Dissociating case from functional categories such as T and v and moving it into the theta domain creates a “flat structure” effect in Turkish, from which properties associated with non-configurationality, namely free word order and argument drop, follow. In languages like English, on the other hand, given that T and v, as the loci of case features, are involved with the establishment of argumenthood, case-driven Agree chains have to be established, creating a strictly configurational phrase structure. As argumenthood is established configurationally in English, lack of free word order and argument drop in English syntax is immediately accounted for. Thus we have proposed a new way of defining non-configurationality based on how argumenthood is established in a language with respect to the checking domain of case and referentiality features.
References Abraham, W. 1997. The interdependence of case, aspect and referentiality in the history of German. In Parameters of Morphosyntactic Change, A. van Kemenade & N. Vincent (Eds), 29–61. Cambridge: CUP. Aissen, J. 1974. The Syntax of Causative Constructions. Ph.D. dissertation, Harvard University. Alexiadou, A. & Anagnostopoulou, E. 1998. Parameterizing AGR: Word order, V-movement and EPP checking. Natural Language and Linguistic Theory 16: 491–531. Alexiadou, A. & Anagnostopoulou, E. 2001. The subject in situ generalization and the role of case in driving computations. Linguistic Inquiry 32: 193–231. Aygen, G. 1998. Split Infl Hypothesis in Turkish. MA thesis, Boğaziçi University. Aygen, G. 1999. Specificity and subject-object positions/scope interactions in Turkish. MS, Harvard University. Aygen, G. 2002. Finiteness, Case and Clausal Architecture. Ph.D. dissertation, Harvard University. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago IL: Chicago University Press. Baker, M. 2001. The natures of non-configurationality. In The Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins (Eds), 407–438. Oxford: Blackwell. Boeckx, C. 2008. Bare syntax. Oxford: OUP. Boeckx, C. & Hornstein, N. 2004. Movement under control. Linguistic Inquiry 35: 431–452.
. See Öztürk (2004, 2005a) for the extension of this proposal to other languages such as Japanese, Hungarian and Chinese.
Balkiz Öztürk Borer, H. 1983. The projection principle and rules of morphology. Proceedings of NELS 14. Borer, H. 1986. I-subjects. Linguistic Inquiry 17: 375–416. Borer, H. 2005. Structuring Sense: An Exo-skeletal Trilogy. Oxford: OUP. Burzio, L. 1986. Italian Syntax. Dordrecht: Reidel. Campos, H. 1986. Indefinite object drop. Linguistic Inquiry 17: 354–359. Carlson, G. 1977. A unified analysis of the English bare plural. Linguistics and Philosophy 1: 413–457. Chierchia, G. 1998. Reference to kinds across languages. Natural Language Semantics 6: 339–405. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Crisma, P. 1997. Nominals without the articles in the Germanic languages. Ms, Universita Di Padova. de Hoop, H. 1996. Case Configuration and Noun Phrase Interpretation. New York NY: Garland. Enç, M. 1986. Towards a Referential Analysis of Temporal Expresions. Linguistics and Philosophy 9: 405–426. Enç, M. 1991. Semantics of Specificity. Linguistic Inquiry 22: 1–25. Fukui, N. 1995. Theory of Projection in Syntax. Stanford CA: CLSI. Ghomeshi, J. 1999. The variable loci of referentiality in noun phrases. Paper presented at the Annual meeting of the Canadian Linguistics Association, Université de Sherbrooke. Giorgi, A. & Longobardi, G. 1991. The syntax of noun phrases. Cambridge: CUP. Göksel, A. 1993. Levels of Representation and Argument Structure in Turkish. Ph.D. dissertation, SOAS. Gür Öded, İ. 2006. Control in Turkish. MA Thesis, Boğaziçi University. Hale, K. 1980. Remarks on Japanese phrase structure: Comments on the papers on Japanese syntax. MIT Working Papers in Linguistics 2: 185–203. Hale, K. 1983. Warlpiri and the grammar of non-configurational languages. Natural Language and Linguistic Theory 1: 5–47. Harley, H. 1995. Subjects, Events and Licensing. Ph.D. dissertation, MIT. Hoji, H. 1998. Null object and sloppy identity in Japanese. Linguistic Inquiry 29: 127–152. Holmberg, A. 1986. Word Order and Syntactic Features in the Scandinavian Languages and English. Ph.D. dissertation, University of Stockholm. Hornstein, N. 1999. Movement and control. Linguistic Inquiry 30: 69–96. Huang, C.-T.J. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531–574. Huang, C.-T.J. 1987. Comments on Hasegawa’s paper. In Proceedings of the Japanese Syntax Workshop: Issues on Empty Categories. New London CT: Connecticut College. Huang, C.-T.J. 1993. Reconstruction and the structure of VP: Some theoretical consequences. Linquistic Inquiry 24: 103–138. İnce, A. 2004. Parasitic gaps in Turkish. Talk given at ECO5 Syntax Workshop, University of Maryland, College Park. Jaeggli, O. 1982. Topics in Romance Syntax. Dordrecht: Foris. Kelepir, M. 1996. The Implications of Turkish for the Theory of Antisymmetry of Syntax. MA thesis, Boğaziçi University.
Non-configurationality: Free word order and argument drop in Turkish
Kelepir, M. 2001. Topics in Turkish Syntax: Clausal Structure and Scope. Ph.D. dissertation, MIT. Kornfilt, J. 1984. Case Marking, Agreement and Empty Categories in Turkish. Ph.D. dissertation, Harvard University. Kornfilt, J. 1990. Turkish and configurationality. In Current Issues in Turkish Linguistics. Proceedings of ICTL V, B. Rona (Ed.), 111–125. Ankara: Hitit Yayınevi. Kornfilt, J. 1995. Scrambling and incorporation in Turkish. In FAS Papers in Linguistics 1, A. Alexiadou, N. Fuhrhop, P. Law & S. Löhken (Eds), 56–65. Berlin: FAS. Kornfilt, J. 2002. Functional projections and their subjects in Turkish clauses. In The Verb in Turkish, E. Taylan (Ed.), 183–212. Amsterdam: John Benjamins. Kornfilt, J. 2003a. Subject case in Turkish nominalized clauses. In Syntactic Structures and Morphological Information, U. Junghanns & L. Szucsich (Eds), 129–215 Berlin: Mouton de Gruyter. Kornfilt, J. 2003b. Scrambling, Subscrambling and Case in Turkish. In Word Order and Scrambling, S. Karimi (Ed.), 125–155. Oxford: Blackwell. Kornfilt, J. 2007. Agr in Turkish as an expression of categorial features. In Proceedings of WAFL 2, M. Kelepir & B. Öztürk (Eds). Cambridge MA: MITWPL. Kural, M. 1991. Scrambling and Mixed Positions in Turkish. In Proceedings of NELS 21: 259–272. Kural, M. 1992. Properties of Scrambling in Turkish. MS, UCLA. Kural, M. 1993. V-to-I-to-C in Turkish. UCLA Occasional Papers in Linguistics, F. Beghelli & M. Kural (Eds), 17–54. Los Angeles CA: UCLA. Kuroda, S. 1988. Whether we agree or not: A comparative syntax of English and Japanese, Linguistic Investigations XII: 1–47. Lasnik, H. 1995. Case and expletives revisited: On greed and other human failings. Linguistic Inquiry 26: 615–633. Legate, J. 2003. Some interface properties of the phase. Linguistic Inquiry 34: 505–516. Lin, J. 2001. Light Verb Syntax and the Theory of Phrase Structure. Ph.D. dissertation, University of California at Irvine. Longobardi, G. 1994. Reference and proper names: A theory of N-movement in syntax and logical form. Linguistic Inquiry 25: 609–665. Longobardi, G. 2001. The structure of DPs: Some principles, parameters and problems. The Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins, (Eds), 562–601. Oxford: Blackwell. Marantz, A. 1991. Case and licensing. In Proceedings of the Eighth Eastern States Conference on Linguistics (ESCOL ‘91), G. Westphal, B. Ao & H.-R. Chae (Eds), 234–253. Ithaca NY: CLC Publications. Markman, V. 2003. Causatives without causers and Burzio’s generalization, CASTL Conference Handout, University of Tromsø. Meral, H. M. 2006. Null objects in Turkish. MS, Boğaziçi University. Miyagawa, S. 1991. Case realization and scrambling. MS, Ohio State University. Miyagawa, S. 2001. EPP, scrambling, and wh-in-situ. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.), 293–338. Cambridge MA: The MIT Press. Nilsson, B. 1986. Object incorporation in Turkish. In Proceedings of the Turkish Linguistics Conference, A. Aksu-Koç & E. Erguvanlı-Taylan (Eds), 113–128. Istanbul: Boğaziçi University Publications. Osawa, F. 1998. The emergence of the D system and the demise of morphological case in English. UCL Working Papers in Linguistics 10: 467–489.
Balkiz Öztürk Osawa, F. 2000. The historical emergence of DP in English. English Linguistics 17: 51–79. Özsoy, A.S. 1996. The minimalist program and the antecedent contained deletion in Turkish. In Proceedings of the 8th International Conference on Turkish Linguistics, K. İmer & N. Uzun (Eds), 9–15. Ankara: Ankara University Press. Özsoy, S. 1988. Null subject parameter and Turkish. In Studies on Modern Turkish, Proceedings of the Third Conference on Turkish Linguistics, H. Boeschoten & L. Verhoeven (Eds), 82–91. Tilburg: Tilburg University Press. Özsoy, A. 1998. Locative inversion, VP adjunction and Turkish relativisation. In The Mainz meeting; Proceedings of the Seventh International Conference on Turkish Linguistics, L. Johanson, É. Csató, V. Locke, A. Menz & D. Winterling (Eds), 361–375. Wiesbaden: Harrassowitz. Özsoy, A. 2002. On ‘small’ clauses, other ‘bare’ verbal complements and feature checking in Turkish. In The Verb in Turkish, E. Taylan (Ed.), 213–237. Amsterdam: John Benjamins. Özsoy, A. 2005. Topic, focus, multiple Specifiers, multiple Spell-out. Paper presented at the Mediterranean Syntax Meeting, Rhodes. Öztürk, B. 1999. Turkish as a non-pro-drop language. MA thesis, Boğaziçi University. Öztürk, B. 2002. Turkish as a non-pro-drop language. In The Verb in Turkish, E. Taylan (Ed.), 239–259. Amsterdam: John Benjamins. Öztürk, B. 2004. Case, Referentiality and Phrase Structure. Ph.D. dissertation Harvard University. Öztürk, B. 2005a. Case, Referentiality and Phrase Structure. Amsterdam: John Benjamins. Öztürk, B. 2005b. Accusative subjects in Turkish. Paper presented at the Mediterranean Syntax Meeting, Rhodes. Öztürk, B. In press. Incorporating Agents. Lingua. Pylkkänen, L. 2002. Introducing Arguments. Ph.D. dissertation, MIT. Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–557. Romero, J. 1999. The case of agreement. In Papers on Morphology and Syntax, K. Arregi, B. Bruening, C. Krause & V. Lin (Eds), 343–367. Cambridge MA: MITWPL. Sezer, E. 1991. Topics in Turkish Syntax. Ph.D. dissertation, Harvard University. Taraldsen, T. 1978. On the NIC, vacuous application and that-t filter. MS, MIT. Taylan, E. 1984. The Function of Word Order in Turkish Grammar. Berkeley Ca: University of California Press. Taylan, E. 1986. Pronominal vs. zero representation anaphora in Turkish. In Studies in Turkish Linguistics, D. Slobin & K. Zimmer (Eds), 209–232. Amsterdam: John Benjamins. Uygun, D. 2006. Scrambling bare singular nominal objects in Turkish. Paper presented at the 13th International Conference on Turkish Linguistics, Uppsala. Von Heusinger, K. & Kornfilt, J. 2005. The case of the direct object in Turkish. Semantics, syntax and morphology. Turkic languages 9: 3–44. Xu, L. 1986. Free empty categories. Linguistic Inquiry 17: 75–93. Zimmer, K. 1976. Some constraints on Turkish causativization. In The Grammar of Causative Constructions: Syntax and Semantics, M. Shibatani (Ed.), 399–412. New York NY: Academic Press.
Diachronic stability and feature interpretability* E. Phoevos Panagiotidis
This contribution examines diachronic change in the Greek nominal phrase. It proposes an analysis to capture the diachronic stability of certain syntactic structures in the nominal domain, suggesting that points of stability reflect macroparametric rather than microparametric choices, which are more vulnerable to change and loss. Macroparametric choices are assumed to involve choices pertaining to the presence versus absence in a grammar of functional categories, while microparametric choices relate to the presence versus absence of uninterpretable features on otherwise identical functional categories. The difference in diachronic stability is linked to the generally “defective” status of uninterpretable features in the human language faculty. Keywords: macroparameter; microparameter; feature interpretability; functional category; Greek; determiner
1. Introduction This paper proposes a way to account for the persistence of a number of syntactic configurations in spite of diachronic change. The main argument is that structures arising from the presence of uninterpretable features are more vulnerable to diachronic change than those generated by the presence (or absence) of LFinterpretable features. I will proceed to illustrate and argue for this hypothesis by appealing to diachronic aspects of the nominal syntax of Greek. It will be proposed that, despite the great deal of change the language has undergone during roughly
*I wish to thank Theresa Biberauer, Martin Haspelmath, Anders Holmberg, Chris Johns, Richie
Kayne, Pino Longobardi, Ad Neeleman, Fritz Newmeyer, Ian Roberts and Peter Svenonius for comments, pointers and discussion. I also wish to thank two reviewers for their pertinent commentary and overall significant help in revising this chapter. Special thanks have to go to Io Manolessou for all her help and guidance with Greek syntax in its various incarnations. All errors remain mine.
E. Phoevos Panagiotidis
30 centuries, the syntactic properties of the Greek nominal phrase that have proved resistant to change are the surface patterns reflecting the workings of LFinterpretable, as opposed to uninterpretable, features. 2. Micro- versus macro-parametric choices and change Let us first specify the problem to be dealt with. In the generative literature, language variation has typically been conceived as following from parametric differences; in other words, grammars are taken to differ from each other along the lines of different parametric choices (Chomsky 1981 and onwards). Moreover, in work by Lightfoot (1991, 1999); Roberts (1993, 1999); Roberts & Roussou (2003) and elsewhere, syntactic change has been conceived as diachronic parameter resetting. In other words, the setting of a parameter may change over time, yielding different syntactic configurations – say VO instead of OV orders. While this is a theoretical standpoint that has produced a lot of exciting and fruitful research, there is a nontrivial question about language change, namely:
(1) Why are some syntactic configurations more resistant to diachronic change than others?
For a number of researchers, (a variant of) the question in (1) provides reason to believe that not all parameters are created equal. To clarify: it appears that structures resulting from some parameter settings are more resistant to change than those of other parameters and/or their settings. In what follows, I will pursue the intuition that, indeed, not all parameters are equal and will attempt to attribute this difference in diachronic stability to the nature of the parameter underlying the structures in question. 2.1 Micro- vs. macro-parameters An established viewpoint within generative grammar since Chomsky’s (1981) Lectures on Government and Binding is that structures/configurations/constructions are not primitives of the theory, but rather reflexes of deeper, more abstract factors: typically parameters or grammatical features. Following a central idea in minimalist theory, the so-called Borer-Chomsky Conjecture foregrounded in Baker (this volume), let us here adopt the idea that the syntactic structures found in different languages reflect the properties of their functional categories. With the above in mind, I will proceed to capture the difference in the diachronic persistence of certain grammatical properties along the lines of a parametric theory that distinguishes between
Diachronic stability and feature interpretability
(2) a. the number, identity and relative order of functional heads as the locus of macroparametric variation, and b. the distribution of uninterpretable features on functional heads as the locus of microparametric variation.
Thus, macroparametric will henceforth stand for variation and parametric effects stemming from which functional categories are available in a given grammar and from their interpretable feature makeup. By contrast, microparametric will be used to characterise variation and parametric effects due to the presence or absence and, more generally, the distribution of uninterpretable features. The choice of the above terms is neither gratuitous nor without consequence: it is indeed hoped and thus suggested that what has elsewhere (e.g., Kayne 2000, this volume; Baker, this volume) been termed microparametric variation will indeed boil down to the distribution of uninterpretable features on functional elements. However, I will not pursue this matter extensively, choosing instead to concentrate on the effects of (2) on the diachronic stability and resistance to change of three syntactic configurations in Greek. Before doing so, I would like to adumbrate the properties and underlying mechanisms of micro- and macro-parameters as well as the differences between them. Let us begin with Willim’s (2000) argument that there is no (evidence for a) determiner category in Polish, something that would make this language different from, say, English. Should this and similar claims (Fukui & Speas 1986; Fukui 1988 and elsewhere) on the non-availability of particular functional categories in certain natural language grammars be correct, they would herald a first kind of macro parametric variation: that stemming from the presence versus absence of a functional head, D in this case. Subsequently, generalising Ouhalla’s (1991) hypothesis that some languages display the order T > Agr whereas others Agr > T, we could identify a second locus of macroparametric variation among grammars as stemming from the relative order of their functional heads. Finally, a third and more prominent source of macroparametric variation would be the type and number of interpretable features encoded on a functional head: according to Bobaljik & Thráinsson (1998), for example, English has a unitary Infl head encoding both tense and agreementrelated features, unlike Icelandic, which has separate T and Agr functional heads. To ensure that our explanation in terms of (2) of the diachronic persistence of the structures in question is well grounded, let us now contrast the above with what we would call instances of microparametric variation. Let there be two grammars X and Y, which could be individuals’ mental grammars in a synchronic or a diachronic setting. Let X and Y also have the same number and type of functional categories, but with different uninterpretable features on them, triggering different movement and/or agreement options. Then X and Y can be said to differ from each other in a microparametric fashion. Two plausible candidates to exemplify
E. Phoevos Panagiotidis
the above state of affairs are given below: first, consider Germanic varieties such as Standard Dutch and Standard German, featuring non-agreeing complementisers, and closely related West Flemish and Bavarian, which possess agreeing complementisers. We would plausibly wish to conceive the difference between Dutch and West Flemish or between German and Bavarian as a microparametric one, caused by (the lack of) uninterpretable phi-features on the C category. A similar hypothesis has been offered with respect to the availability of negation with imperatives in Mainland Scandinavian varieties. Hence, compatibility between imperatives and the Swedish negative inte and the Norwegian negative ikke microparametrically contrasts with the impossibility of an imperative with negative ikke in Danish. Platzack & Rosengren (1998), building on Zanuttini (1996), attribute this difference to the T-relatedness of negative ikke in Danish, as opposed to the lack of T-relatedness of inte and ikke in the other two varieties. The relevant contrast can then be understood as a matter of which uninterpretable T-related features are encoded on Neg and imperative C respectively.1 From this point onwards, the guiding hypothesis is very simple. It is given in (3):
(3) Macroparametric choices are more persistent diachronically than microparametric ones.
Elaborating slightly, we would expect microparametric choices to be more vulnerable to diachronic variation and change, especially if change is invariably conceived as the result of variation (cf. Lightfoot 1999). This would by no means imply either that macroparameters are immutable – obviously this would be plainly wrong – or that microparameters are bound to change; only that they are more vulnerable to change. Before applying (3) to the diachrony of the Greek nominal phrase and before observing the differences in the diachronic persistence of different configurations, let us establish in more detail what (3) actually means. 2.2 Features and parameters I have argued in (2) that language-specific choices relating to the number, identity and relative order of functional heads are more resistant to change than the particulars of uninterpretable features on functional heads. Nevertheless, if we take seriously the reasonably well-established idea that functional elements are grammarinternal entities, essentially no more than collections of features – see Harley & Ritter (2002a, 2002b) and Hegarty (2005) for excellent illustrations – what are we to make
. I am grateful to Kristin Eide for alerting me to the particulars of negation with imperatives in Scandinavian.
Diachronic stability and feature interpretability
of the distinction expressed in (2) and (3)? In more detail, the difference between macroparametric and microparametric characteristics – and their relative resistance to language change – has been sketched in (2) and (3) as one between, inter alia, the identity of functional heads as opposed to the distribution of uninterpretable features on them. This distinction, however, becomes blurry and indeterminate if functional heads themselves are not primitives (as, in a sense, they are in Cinque 1999), but just sets of features. This is a matter of some concern and, by addressing it, I hope to shed more light on the details of the macro- vs. micro-parametric properties of languages and their diachronic behaviour. In doing so, I will start with two quite straightforward background assumptions, (4) and (5).
(4) Universal Grammar (UG) “provides” the learner with a fixed repertoire of features to choose from and build functional categories.
Essentially, this is the line of research into the nature of functional categories represented by Thráinsson (1996) and Bobaljik & Thráinsson (1998) (see also Borer (2004) for meticulous criticism of this thesis). For purposes of illustration, let us consider a model according to which UG makes available just three features for the “nominal” domain, say [Foc(us)], [Ref(erential)] and [Ind(ividuation)]. Given (4), it follows that the possible number and type of nominal functional heads in grammars would include the following (considering only interpretable features): – a grammar containing just the one “hybrid” [Foc] [Ref] [Ind] functional category; – a grammar containing an independent [Foc] as well as a “hybrid” [Ref] [Ind] category; – a grammar with, likewise, two nominal functional categories, but “hybrid” [Foc] [Ref] and independent [Ind]; – a grammar with three nominal functional categories, namely [Foc], [Ref] and [Ind]; and so on. As not all UG features need be used by a particular grammar, the following options are also available: – a grammar with a single [Ref] [Ind] nominal functional category; – one with two categories, marking just [Ref] and [Ind] respectively; and, once again, so on. To anticipate, we can now say that all the hypothetical grammars listed above would differ from one another in macroparametric terms. But let us now spell out the second necessary assumption:
(5) Functional categories are made up of both LF-interpretable and LF-uninterpretable features.
E. Phoevos Panagiotidis
Let us put the two assumptions together in order to explain the different degree of diachronic stability displayed by different structures. Again, let there be two varieties X and Y. If the grammars of X and Y have carved up the UG feature repertoire in significantly different ways, i.e., if they have combined UG features into different sets of “functional categories” as in (4) – then X and Y differ in macroparametric terms. By (3), it is more difficult for particular sets of interpretable UG features (“functional categories”) to change over time. The question now is whether it can be the case that X and Y have identical functional categories – i.e., that they have carved up the UG feature repertoire identically – and still have differences. I would like to suggest that the answer is yes, although these will be microparametric differences, resulting from the presence or absence of uninterpretable features. Given (3), it is easier for uninterpretable features to be lost. Crucially, microparametric differences can be readily detectable and quite prominent in the input as uninterpretable features trigger agreement and drive movement (Chomsky 2000). Consider again in this connection the Complementiser (non-)agreement phenomena found in closely related Germanic varieties mentioned above. Hence the diachronic loss of a single uninterpretable feature can have widespread surface effects. Nevertheless, if what is claimed here is on the right track, a very interesting result emerges, a hypothesis worth closer empirical investigation. Suppose a grammar X sanctions a battery of phrasal movement operations. Let us also suppose that these movement operations are easily detectable in the input, for example because they target the specifier of CP. What the above discussion predicts is that the surface salience of the particular battery of movement operations cannot guarantee its diachronic persistence. If this battery of movement operations is triggered by a single uninterpretable feature on the C-head of X, as in our hypothetical example here, this will, given (2) and (3), be a configuration quite vulnerable to diachronic change. Although for the moment we have to make do with a hypothetical example, an actual case of a diachronically vulnerable structure with surface prominence, but an uninterpretable feature as its trigger is discussed in subsection 2.3 below. The proposal that emerges is that syntactic change proceeds not by negotiating surface patterns or under pressure from statistical predominance of frequent word orders and configurations, but by renegotiating feature representations on functional categories. The bigger picture of how movement is replaced by base generation as a function of diachronic (uninterpretable) feature loss is painted in Roberts & Roussou (2003); for purposes of exposition, we can say that a diachronic process may:
Diachronic stability and feature interpretability
1. rearrange interpretable features into new functional categories; e.g., by splitting the hypothetical [Foc] [Ref] [Ind] into two heads, [Foc] and [Ref] [Ind] or by unifying [Foc] and [Ref] [Ind] into a single [Foc] [Ref] [Ind] head; 2. eliminate an (interpretable) feature from the particular grammar; e.g., [Foc]; 3. introduce uninterpretable features; e.g., [uphi];2 4. eliminate uninterpretable features; e.g., [uphi]; 5. … The conclusion is straightforward: (introduction and) elimination of uninterpretable features is easier than any of the processes described in 1 and 2. above.3 Let us accordingly revise (2) and (3) as the following hypothesis:
(6) a. Grammars can differ from each other macroparametrically because their functional heads consist of different interpretable features. The feature makeup of functional heads is resistant to change;
b. Grammars can differ from each other microparametrically because their functional heads bear different uninterpretable features. Uninterpretable features – and their surface effects – are less resistant to change.
The hypothesis in (6) runs parallel to language acquisition theories according to which microparametric choices (as defined here) are harder to acquire in first and second language acquisition; in other words, acquisition of uninterpretable features is difficult in First Language Acquisition and, if Hawkins’s (2003) Representational Deficit Hypothesis and Tsimpli (2003) are on the right track, impossible in Second Language Acquisition. The remainder of this paper examines data from the diachrony of the Greek nominal domain, which will be shown to behave in accordance with the proposal in (6).
. I will not focus in this chapter on diachronic processes introducing uninterpretable features, although I effectively consider them to be microparametric in (2) and (6), on a par with the deletion of uninterpretable features. Still, I believe that eliminating an uninterpretable feature from a category F is likely to be easier than inserting one. In any case, processes inserting uninterpretable features clearly must be there, otherwise, as Richie Kayne, Ad Neeleman and Ian Roberts have pointed out to me, grammars would end up converging over time significantly more than they seem to do (see also the discussion in Newmeyer 2005: 101–103 and Gianollo et al., this volume). Language contact may be an example of a process that introduces uninterpretable features (thanks to Peter Svenonius for discussing this with me). . A reviewer points out that this makes intuitive sense since adding and removing interpretable features actually affects the LF representation in a way that adding and removing uninterpretable features does not.
E. Phoevos Panagiotidis
3. The diachrony of Greek nominal phrases Its long attestation makes Greek a particularly rewarding language to investigate from a diachronic perspective. Moreover, because of the language’s uninterrupted status since the 9th century BC as a Kultursprache, both Greek grammar at a number of synchronic stages (those of Homeric, Archaic, Classical Attic, Koine, Medieval and Renaissance varieties as well as post-19th century) and grammatical change across time have been the object of considerable research. Instead of surveying the whole range of (lack of) change within the Greek nominal phrase (henceforth “DP”), I refer interested readers to Manolessou (2000) for a detailed survey and analysis. I will confine myself to three configurations characteristic of the Greek DP and describe how they have or have not changed over time. In doing so, I will largely be following Manolessou (2000). I will also adopt the Determiner Phrase (DP) hypothesis for nominal constituents without further discussion (cf. Abney 1987). Beginning with some descriptive facts, the typical order of constituents within the Greek nominal phrase has been the following throughout the history of the language:
(7) (Demonstrative) > Article > Adjective > (Demonstrative) > (Article) > Noun > Genitive DP
Moreover, Greek DPs have consistently been left-headed and there has always, except in the genitive DP, been gender, number and case concord between all the DP-internal constituents, 3.1 Demonstrative doubling Greek demonstratives in a DP obligatorily co-occur with a definite article. The demonstrative itself may appear in one of two positions: either preceding the rest of the DP, typically surfacing adjacent to the article, or following the adjective (but not in both positions at the same time). This is a state of affairs that has remained stable since at least the time of Herodotus (5th century BC) and is illustrated in (8) and (9) for Classical Attic and Standard Modern Greek respectively. (8) (hautē) hē megalē (hautē) polis this the big this city
Classical Greek
(9) (afti) i meghali (afti) poli this the big this city ‘this big city’
Modern Greek
Detailed treatments of the topic can be found in Stavrou & Horrocks (1989); Brugè (1996); Giusti (1997); Bernstein (1997); Manolessou (2000); Panagiotidis (2000) and Grohmann & Panagiotidis (2004), the latter dubbing this co-occurrence
Diachronic stability and feature interpretability
of a demonstrative with an article Demonstrative Doubling.4 All the above accounts converge in analysing the post-adjectival low position of the demonstrative as the one in which the demonstrative is merged and the pre-article high one as one derived by movement. Quite interestingly, as Manolessou (2000), Panagiotidis (2000) and Campos & Stavrou (2004) explain, the two positions are linked to distinct interpretive effects, with the low position resisting both focus and purely deictic functions, which are however felicitous with the demonstrative positioned high. Hence, when the demonstrative is high, in the pre-article position, it can have a deictic function and receive focal stress; when the demonstrative is left in its low position, it restricts itself to a discourse-anaphoric function and cannot receive focus stress. Grohmann & Panagiotidis (2004) compare the state of affairs outlined above with instances of Demonstrative Doubling in a number of other languages including Spanish, Beiruti Arabic, Macedonian and Romanian.5 Their conclusion is that Demonstrative Doubling results from the interaction of the demonstrative with a “nominal” Focus functional head, which they take to be responsible for the demonstrative’s deictic force when it appears in the high position. Although the existence of a nominal Focus category has been argued for in a number of languages (Ihsane & Puskás 2001; Aboh 2004), its presence constitutes a necessary but not a sufficient condition for the occurrence of the Demonstrative Doubling configuration. What is important to keep in mind at this point is that the diachronic stability of two demonstrative positions being available in Greek correlates with a clear interpretive effect for each position and with the workings of focus and deixis; thus it does not seem to be the effect of uninterpretable features. In the terminology used here, Demonstrative Doubling is the macroparametric effect of the existence of a functional head, nominal Focus, and the fact that this head attracts demonstratives. The diagram below gives the proposed structure for (8) and (9). (10) Demonstrative Doubling FocP Foc afti
DP D
NP
i
meghali afti poli
. The idea here being that the demonstrative is “doubled” by an (apparent) article. . They also derive the obligatory presence of the article as the morphological effect of the demonstrative’s “too-local” movement within the DP’s left periphery (cf. Grohmann 2003).
E. Phoevos Panagiotidis
3.2 Determiner Spreading In Greek, a second article may appear in front of the adjective within the DP. An established term for this construction is Determiner Spreading (after Androutsopoulou 1994) and, again, it is a construction dating back to at least the 5th century BC and Herodotus’ History, see (11) and (12).6 (11) hē polis hē megalē the city the big
Classical Greek
(12) i poli i meghali the city the big ‘the big city’
Modern Greek
Androutsopoulou (1994) has already claimed the configuration to involve a special functional head Def(initeness). This has been challenged by Alexiadou & Wilder (1998), Manolessou (2000) and Marinis & Panagiotidis (2005), both on theoretical grounds and due to the lack of independent empirical evidence for Def. However, it does indeed seem to be the case that this particular construction owes its existence, if not to a nominal functional category unique to Greek, then at least to particular properties of Determiners in Greek. As Manolessou (2000), Campos & Stavrou (2004), Larson & Yamakido (2005) and Marinis & Panagiotidis (2005) observe, Determiner Spreading crucially has a predicative reading – motivating Campos & Stavrou (2004) to independently argue for a small clause projection within DPs with Determiner Spreading.7 For reasons of parsimony, I will follow Marinis & Panagiotidis (2005) here in claiming that Determiner Spreading is possible in Greek not because of a fully-fledged small clause within the DP, but because the Determiner heading the phrase encodes both referential and predicative properties, a situation reminiscent of Holmberg & Nikanne’s (2002) Fin category in Finnish, with its mixed A and Aʹ properties. What this proposal amounts to is the following: there is a variety of Determiner (let’s call it Dpred for lack of a more suitable label) in Greek bearing (an) interpretable feature(s) that enable(s) it to
. The term “polydefiniteness” has also been used in relation to the construction. It is not adopted here because DPs with Determiner Spreading do not have to be definite, see Roussou & Tsimpli (1994), Giannakidou & Stavrou (1999), and Marinis & Panagiotidis (2005). . Campos & Stavrou (2004) argue for a specialised predicative/small clause head within the Greek DP – a macroparametric characteristic according to (6). Their account is compatible with what follows. See also Giusti (2003) and Kolliakou (2004) for alternative analyses.
Diachronic stability and feature interpretability
mediate in a predicative relation between its specifier (the subject of predication, i.e., i poli “the city” in (13) below) and the NP in its complement (the predicate, i.e., meghali eN “big one” in (13) below). Again, the configuration is a macroparametric characteristic given (6) because it is triggered by a functional category, Dpred, that differs from certain other Ds in bearing (an) interpretable feature(s). (13) Determiner Spreading DPpred DP2 i poli
Dpred' Dpred i
NP meghali eN
Of course, as the presence of the elliptical NP meghali eN (“big one”) in (13) indicates, the particular shape of Determiner Spreading is also contingent on the available noun ellipsis possibilities in the language concerned; I will not expand on the matter here, but the reader is referred to Marinis & Panagiotidis (2005) for discussion. However, there is an element of circumstantiality that nicely contrasts with the diachronic stability of Dpred (and, consequently, Determiner Spreading) in the grammar(s) of Greek. Let us compare (11) and (12), displaying nounadjective order, with the examples of Determiner Spreading below from Manolessou (2000), exhibiting adjective-noun orders: (14) *hē megalē hē the big the
polis city
Classical Greek
(15) i meghali i the city the ‘the big city’
poli big
Modern Greek
The question we would like to answer is why the contrast between (14) and (15) obtains. Although the hybrid referential/predicative Dpred category has persisted in the grammar(s) of Greek, an uninterpretable feature associated with it has perished, a situation aptly illustrating the contrast in (6) between the stability of interpretable feature combinations and the vulnerability of uninterpretable features on functional heads. Let us look into the matter by consulting the phrase marker for
E. Phoevos Panagiotidis
Determiner Spreading in its Classical Greek incarnation. This is the same as the one in (13), repeated here as (16): (16) Determiner Spreading DPpred DP hē polis
Dpred' Dpred hē
NP megalē eN
An uninterpretable feature on Dpred could plausibly have been responsible for the impossibility of adjective-noun orders in Classical Greek and well into the Koine period. Although the identity of this feature is not clear at this point, what (16) illustrates is that it required a full DP as its checker (e.g., hē polis “the city”), not an elliptical one (e.g., hē megalē eN “the big one”). The proposal here is therefore that an uninterpretable feature was to blame for the ungrammaticality of (14) in older forms of Greek, and that this uninterpretable feature was lost during the diachronic process. As the hybrid referential/predicative D is still part of the grammar of Greek, Determiner Spreading has not only persisted, but ordering options associated with it have been relaxed, as witnessed by the availability in Modern Greek of both the orders in (12) and (15). To conclude this section: noun-adjective orders with Determiner Spreading, as in (11) and (12), have been possible since the structure was first attested, whereas the reverse adjective-noun orders became possible only (relatively) recently, as shown in (14) versus (15). This can now be easily captured along the following lines: although the combination of interpretable referential and predicative features making up Dpred has remained stable as a functional category, the uninterpretable feature on Dpred formerly responsible for the banning of adjective-noun orders – being itself more vulnerable – has perished, freeing up the ordering possibilities with Determiner Spreading. 3.3 Post-article genitives The very widespread optional movement of genitive DPs to a position immediately after the DP-initial determiner was lost after the Koine period, as the contrast between (17) and (18) shows. Ever since, genitive DPs have only been able to
Diachronic stability and feature interpretability
appear in what appears to be their canonical position, after the Ns of which they are the argument. This is sketched in (7).8 (17) ho [tou basileos] thronos the the.gen king.gen throne
t
Classical Greek
(18) *o [tu vasilia] thronos the the.gen king.gen throne ‘the king’s throne’
t
Modern Greek
Given what we noted at the start of section 2 about the unmarked order of constituents within the Greek DP, the occurrence of the genitive DP between D and N in (17) must be due to leftward XP movement. This optional movement targets the specifier of a nominal functional category higher than N but lower than D (for details, see Manolessou 2000), let us say here for the sake of exposition that it targets the specifier of Num. The trigger for this movement seems to be an uninterpretable feature on Num, something that is suggested by the lack of any interpretive difference between (17) and its non-movement counterpart (Manolessou 2000; p.c.). By (6), this makes post-article genitives a configuration resulting from a microparametric choice, i.e., one, which, in accordance with the hypothesis in (3), is more susceptible to change and loss. Indeed, this is what seems to have happened during the Koine period: the uninterpretable feature on Num attracting the genitive DP perished and post-article genitives became ungrammatical. Interestingly, post-article genitives illustrate the hypothetical state of affairs discussed in section 1.2: I claimed there that the surface prominence of a syntactic configuration will not guarantee its diachronic stability if it is triggered by the presence of an uninterpretable feature. The eventual demise of post-article genitives from Greek lends support to this thesis.
4. Conclusion and beyond Before concluding, let us consider the bigger picture suggested by the Greek diachronic evidence. I used the proposal in (6) to make sense of the fact that configurations like Demonstrative Doubling and Determiner Spreading have been quite stable throughout the language’s history, effectively reducing their persistence to the diachronic stability that particular bundles of interpretable features, i.e., languagespecific functional categories, display. Recall that, by hypothesis, Demonstrative
. That is, unless focused or topicalised, in which case they can appear at the left edge of the DP, just like any other focused or topicalised DP subconstituent.
E. Phoevos Panagiotidis
Doubling, in (8) and (9), is triggered by the presence of a nominal Focus category, whereas Determiner Spreading, in (11) and (12), is a reflex of a hybrid referential/ predicative Dpred, i.e., a particular combination of “referential” and “predicative” features available from UG which has always been available in the grammar of Greek. By (3) and (6) it should be “difficult” to shed interpretable features associated with these two categories or to split them into more heads, and this has proved to be the case. Determiner Spread and Determiner Doubling therefore persist. The stability of the two structures contrasts with the loss of the option of placing genitives right after the article, as illustrated in (17). Although the surface prominence of such a syntactic operation might be expected to make it easy to detect and acquire, it nevertheless seems that something along the lines of the proposal in (6) may be correct, in that its syntactic trigger, an uninterpretable feature on Num, was vulnerable to language change, leading to the loss of the configuration. Another important aspect of the analysis proposed here is this: once we stop thinking of constructions like Determiner Spreading as consolidated “packages” of grammatical properties (i.e., as primitives) and once we understand them to be surface effects of the workings of grammatical features, we can explain how certain constructions are more likely to change than others. The relevant diachronic pattern is stated in (19): (19) Diachronic processes eliminate uninterpretable features (responsible, inter alia, for agreement and movement) more easily than they eliminate interpretable features on functional elements or than they rearrange them across novel functional categories.9
(19) suggests a potentially fruitful direction for future research, namely investigating how the proposals made here might be explicitly tied in with those in the grammaticalisation model proposed in Roberts & Roussou (2003). Roberts & Roussou propose that movement is diachronically dispreferred because it is triggered by “extra features”. Here I have argued that uninterpretable features in particular are the ones that are diachronically unstable. If the proposals in this paper are on the right track, we might expect that many (all?) of Roberts & Roussou’s “extra” features will turn out to be uninterpretable.
References Abney, S. 1987. The English Noun Phrase in it Sentential Aspect. Ph.D. Dissertation, MIT. Aboh, E. 2004. Topic and focus within D. Linguistics in the Netherlands 21(1): 1–12
. On introducing uninterpretable features, see endnote 2.
Diachronic stability and feature interpretability
Alexiadou, A. & Wilder, C. 1998. Adjectival modification and multiple determiners. In Possessors, Predicates and Movement in the DP, A. Alexiadou & C. Wilder (Eds), 303–332. Amsterdam: John Benjamins. Androutsopoulou, A. 1994. The distribution of the definite determiner and the syntax of Greek DPs. In Proceedings of the 30th Regional Meeting of the Chicago Linguistic Society, K. Beals, R. Knippen, L. Melnar, H. Suzuki, E. Zeinfeld (Eds), 16–29. Chicago IL: University of Chicago Press. Bernstein, J. 1997. Demonstratives and reinforcers in Romance and Germanic languages. Lingua 102: 87–113 Bobaljik, J.D. & Thráinsson, H. 1998. Two heads aren’t always better than one. Syntax 1: 37–71 Borer, H. 2004. In Name Only. Oxford: OUP. Brugè, L. 1996. Demonstrative movement in Spanish: A comparative approach. University of Venice Working Papers in Linguistics 6: 1–53. Campos, H. & Stavrou, M. 2004. Polydefinite constructions in Modern Greek and in Aromanian. In Balkan Syntax and Semantics, O. Tomić (Ed.), 136–173. Amsterdam: John Benjamins. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds), 89–155. Cambridge MA: The MIT Press. Cinque, G. 1999. Adverbs and Functional Heads: A Crosslinguistic Perspective. Oxford: OUP. Fukui, N. 1988. Deriving the differences between English and Japanese: A case study in parametric syntax. English Linguistics 5: 249–270. Fukui, N. & Speas, M. 1986. Specifiers and projections. MIT Working Papers in Linguistics 2: 128–172. Giannakidou, A. & Stavrou, M. 1999. Nominalization and ellipsis in the Greek DP. The Linguistic Review 16: 295–331. Giusti, G. 1997. The categorial Status of Determiners. In The New Comparative Syntax, L. Haegeman (Ed.), 95–123. London: Longman. Giusti, G. 2003. The functional structure of noun phrases: A bare phrase structure approach. In Functional Structure in DP and IP. The Cartography of Syntactic Structure, Vol. 1, G. Cinque (Ed.), 54–90. Oxford: OUP. Grohmann, K. 2003. Prolific Domains. On the Anti-Locality of Movement Dependencies. Amsterdam: John Benjamins. Grohmann, K. & Panagiotidis, P. 2004. Demonstrative doubling in Greek. In University of Maryland Working Papers in Linguistics 13, P. Chandra, T. Fujii, U. Soltan & M. Yoshida (Eds), 109–134. College Park MD: The University of Maryland Press. Harley, H. & Ritter, E. 2002a. Structuring the bundle. A universal morphosyntactic feature geometry. In Pronouns – Grammar and Representation, H. Simon & H. Wiese (Eds), 23–39. Amsterdam: John Benjamins. Harley, H. & Ritter, E. 2002b. Person and number in pronouns: A feature-geometric analysis. Language 78: 482–526. Hawkins, R. 2003. “Representational deficit” theories of (adult) SLA: Evidence, counterevidence and implications. Paper presented at the 13th EUROSLA, Edinburgh. Hegarty, M. 2005. A Feature-based Syntax of Functional Categories. Berlin: De Gruyter. Holmberg, A. & Nikanne, U. 2002. Expletives, subjects, and topics in Finnish. In Subjects, Expletives and the EPP, P. Svenonius (Ed.), 71–105. Oxford: OUP. Ihsane, T. & Puskás, G. 2001. Specific is not definite. Generative Grammar in Geneva 2: 39–54.
E. Phoevos Panagiotidis Kayne, R. 2000. Parameters and Universals. Oxford: OUP. Kolliakou, D. 2004. Monadic definites and polydefinites: Their form, meaning and use. Journal of Linguistics 40: 263–323. Larson, R. & Yamakido, H. 2005. Ezafe and the deep position of nominal modifiers. Paper presented at the Barcelona Workshop on Adjectives and Adverbs. Lightfoot, D. 1991. How to Set Parameters. Arguments from Language Change. Cambridge MA: The MIT Press. Lightfoot, D. 1999. The Development of Language. Oxford: Blackwell. Manolessou, I. 2000. Greek Noun Phrase Structure: A Study in Syntactic Evolution. Ph.D. dissertation, Cambridge University. Marinis, T. & Panagiotidis, P. 2005. Determiner spreading as DP-predication. MS, available from LingBuzz. Newmeyer, F. 2005. Possible and Probable Grammars. Oxford: OUP. Ouhalla, J. 1991. Functional Categories and Parametric Variation. London: Routledge. Panagiotidis, P. 2000. Demonstrative determiners and operators: The case of Greek. Lingua 110: 717–742. Platzack, C. & Rosengren, I. 1998. On the subject of imperatives: A minimalist account of the imperative clause. Journal of Comparative Germanic Linguistics 1: 177–224. Roberts, I. 1993. Verbs and Diachronic Syntax. A Comparaive History of English and French. Dordrecht: Kluwer. Roberts, I. 1999. Verb movement and markedness. In Language Creation and Language Change: Creolization, Diachrony, and Development, M. DeGraff (Ed.), 287–327. Cambridge MA: The MIT Press. Roberts, I. & Roussou, A. 2003. Syntactic Change: A Minimalist Approach to Grammaticalization. Cambridge: CUP. Roussou, A. & Tsimpli, I.-M. 1994. On the interaction of case and definiteness in modern Greek. In Themes in Greek Linguistics, I. Philippaki-Warburton, K. Nicolaidis & M. Sifianou (Eds), 69–76. Amsterdam: John Benjamins. Stavrou, M. & Horrocks, G. 1989. Εγκλιτικές και δεικτικές αντωνυμίες μέσα στην ΟΦ (Enclitic and demonstrative pronouns inside NP). In Proceedings of the 10th Annual Meeting of the Department of Linguistics, Faculty of Philosophy, 225–243. Thessaloniki: Aristotle University of Thessaloniki. Thráinsson, H. 1996. On the (non-) universality of functional categories. In Minimal Ideas. Syntactic Studies in the Minimalist Framework, W. Abraham, S.D. Epstein, H. Thráinsson & J.-W. Zwart (Eds), 253–282. Amsterdam: John Benjamins. Tsimpli, I.-M. 2003. Interpretability of features in L1 and L2 acquisition and language attrition. Paper presented at the 13th EUROSLA, Edinburgh. Willim, E. 2000. On the grammar of Polish nominals. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, R. Martin, D. Michaels & R. Uriagereka (Eds), 319–346. Cambridge MA: The MIT Press. Zanuttini, R. 1996. On the relevance of Tense for sentential negation. In Parameters and Functional Heads: Essays in Comparative Syntax, A. Belletti & L. Rizzi (Eds), 181–207. Oxford: OUP.
part iv
The acquisition of parameters
Parameter setting and input reduction1 Arnold Evers & Jacqueline van Kampen The language acquisition procedure identifies certain properties of the target grammar before others. The evidence from the input is processed in a stepwise order. Section 1 equates that order and its typical effects with an order of parameter setting. The question is how the acquisition procedure derives the order from input evidence. Section 2 proposes a systematic input reduction for functional categories as the key; the reduction residue contains no more than a single non-acquired functional category 〈F?〉 that is first seen as an optional element only. If that functional category has turned into the most preferred option, the input reduction shifts its acquisition focus to the next functional category. Section 3 and 4 demonstrate how quantitative proportions within the child’s input reduction determine the underlying order as SOV in Dutch before the V-second shift for root sentences is derived. The child’s input reductions are claimed to follow from ignorance rather than from any a priori information. It is argued that parameters are formal properties of the grammatical system that originate as cultural discoveries made by a reflexive mind rather than being task-specific neural a prioris. Section 5 suggests that this view can be extended to syntactic islands. Keywords: cue; learnability; acquisition of Verb Second (V2); lexical categories; island constraints
1. Parameter setting The twin notions “parameter” and “parameter setting” (Chomsky 1981) are aimed at explaining first that grammatical properties do not vary independently of each other, and second that this fact is significant for the striking success of
. The critical remarks on an earlier version made by Peter Culicover and Bill Philip, as well as comments by two anonymous reviewers did a lot to sharpen our ideas. Nevertheless, the standard phrase “remaining errors are due to the authors” certainly applies in the present case. The research for this paper was supported by NWO (grant 355-070-009, second author).
Arnold Evers & Jacqueline van Kampen
first language acquisition. The classic example is the property of “subject prodrop” acquired by setting a feature value in some functional category I 〈+agr〉. Chomsky (1981: 240) analyzes pro-drop as possibly implying free inversion of the subject phrase, violation of the that-trace filter, and long wh-movement of a subject. An engaging perspective opened up. Suppose that all parameters are binary valued like 〈± subject pro-drop〉 and suppose, more daringly, that the set of possible parameters is limited and predetermined. Is it then possible that setting a limited amount of predetermined parameters might deliver all possible types of core grammars and thus reveal something about the success of spontaneous language acquisition? Two problems with the notion of parameter setting were diagnosed in Dresher (1999): “the epistemological problem” and “the credit problem’’. The very explanatory and abstract nature of the parameter implies that its presence cannot be noticed as a simple property in the primary data. This is the epistemological problem. The credit problem relates to the interaction of parameters. The necessary and intended interaction of parameters prevents their setting by simple observation. When combining a tentative setting of parameters yields results not confirmed by the primary data, it is unclear which parameter setting caused the uninvited superset effect. Both problems for parameter setting (the epistemological problem and the credit problem) follow from the nature of the parameter itself and both seem to disqualify parameter settings as innate bootstraps for first language acquisition. Yet, when we look at first language acquisition itself, there are reasons not to give up on the explanatory potential of the notion “parameter setting” too easily. There is in language acquisition a stepwise appearance of grammatical properties and it would be strange if that progress were not caused by an independent identification of grammatical properties. Child language picks up the properties of a grammar in a non-arbitrary order. Parameter theory implies a hierarchy of grammatical properties as expected by Jakobson (1942) and developed by him for phonology. Parameter theory allows us to see language acquisition as a series of ever more specified grammars Gi that approach the target grammar Gn (Chomsky 1975: 119f).
(1) Go ---- Gi ⇒
Gi+1 ---- Gn
Properties acquired earlier, the left-side steps in (1), are likely to have a lasting effect upon the steps that follow. They are likely to have typological significance and may develop into what Baker dubs “macro-parameters” (cf. Baker 1996, this volume). The right-side steps in (1) are likely to have fewer such consequences. They will introduce variation that is more language-specific and possibly more flexible across history and dialects (cf. Panagiotidis, this volume). It is for that reason that
Parameter setting and input reduction
the reconstruction of the order of acquisition steps is of the utmost importance for understanding how input data may determine the setting of a parameter. The separate and successive identification of grammatical properties in first language acquisition reappears in computational models of the acquisition procedure (Berwick 1985; Berwick & Weinberg 1984; Clark 1992; Gibson & Wexler 1994) as the Single Value Constraint. The Single Value Constraint implies a parametric view of acquisition and holds that the acquisition steps that set the parameters are taken in a linear order that develops as a causal chain. Each new step Gi ⇒ Gi+1 shifts the focus to a new parameter. Plausibly, this happens because an orientation by Gi opens the mind to perceptions that trigger the step to Gi+1. In terms of data selection, Gi makes selections from the raw input and arrives at a data set Di+1 that will redirect the acquisition procedure towards Gi+1. This implies that the acquisition procedure somehow gets information about the questions in (2). (2) a. How is a data set Di+1 selected by Gi? b. When does a new parameter in Gi+1 become relevant?
A mechanism to assemble that information is still needed and the present paper represents an attempt to formulate one. 1.1 Parameter setting models Models of language acquisition by parameter setting have to deal with the problems diagnosed in Dresher (1999). They tend to do this by adding more a priori (innate) information about possible grammars (cf. Hale & Reiss 2003). A priori grammatical factors enable the acquisition procedure to construct a provisional grammar. The learner changes the provisional grammar only when it is inconsistent with the current input sentences (error-driven learning). For example, the Triggering Learning Algorithm (TLA) model by Gibson & Wexler (1994) considers the consequences of the following speculation. Let the primary data be provisionally represented by a string of category labels as known from the adult grammars and let all parameters be given a priori. The learner applies a provisional parameter setting to the primary data of his language. Predictably, this will in general not work. Now the language acquisition procedure is allowed to switch one parameter value. When this move hits the jackpot and arrives at a satisfactory analysis of (a subset of) primary data, the switched parameter is accepted as the correct value of the parameter for that grammar. This is all preliminary assumption by Gibson & Wexler in order to arrive at their genuine research question. Will this method of blind gambling on parameter values ever reach the target grammar? Gibson & Wexler try it out for an imaginary case of 5 given categories {subject, object, adverb, verb, auxiliary} and three parameters
Arnold Evers & Jacqueline van Kampen
{subject-head order, complement-head order, presence or absence of a V-second rule}. The primary data are given as strings of the functional categories and the acquisition procedure must find out how to discriminate between the 23 = 8 possible types of grammar allowed by the model of three binary parameters. It turns out that the method cannot guarantee an arrival at the correct target grammar. A legitimate but with hindsight premature setting of the V-second parameter will cause havoc. It may deliver distributions that require two simultaneous re-settings (instead of one). More than one parameter change for each acquisition step weakens the model as it greatly expands the search space for each step. Gibson & Wexler call such failures a local maximum, where the target grammar represents the absolute maximum. A subsequent study by Kohl (1999) calculated that the chance of running into local maxima rises considerably when the number of parameters rises. However, an unambiguously faultless result is guaranteed when the acquisition procedure is informed about a linear order in which the parameters must get fixed. The demonstration case in Gibson & Wexler (1994) runs fine if an ad hoc decision is added to delay a decision about the V-second parameter. This is interesting since actual language acquisition does indeed show an order of acquisition steps (Brown 1973).2 The new problem is now how the acquisition procedure can be informed about the sequencing restrictions in parameter setting. Wexler (1999) speculates quite generally that the troublemakers among the parameters will not be considered before the brain has matured sufficiently. Fodor (1998, 2001) proposes that a troublesome parameter will not be reset from an initial default status until the learner has spotted a configuration that is an unambiguous context cue for the resetting. The successful cue has to be crafted carefully. It is supposed to be part of the innate equipment and it should protect its innocent users against all the world’s grammatical distributions. Dresher (1999) even advocates a maximum use of context cues (see also Dresher & Kaye 1990; Lightfoot 1991). He argues that a sufficiently developed system of (innate) cues could do without the whole method of error-driven parameter setting.
. By a surprisingly simple approach, Brown (1973) arrived at a picture of the stepwise hierarchy in morphosyntactic progress. In attempting to explain the hierarchy, he introduced a complexity measure employed at the time, namely Derivational Complexity, i.e., the number of transformational steps needed for deriving the structure. This complexity measure was at the time assumed to predict measurable performance delay. When that speculation failed, however, the field of psycholinguists started without much reason to mistrust the very notion of a complexity measure and acquisitionists even gave up trying to understand the problem of order in acquisition steps. Consequently, Brown (1973) was reduced to a conglomeration of remarkable observations. With hindsight, this may have been a strategic research failure.
Parameter setting and input reduction
Other attempts maintain error-driven parameter setting without cues. They evade the local maximum by adding more computational power to the acquisition procedure. For example, they allow several parameter values to be switched at the same time or they allow parameters to be switched on and off for some time in order to escape the effects of a premature setting. Such measures increase the search space for possible solutions and thereby weaken the model. The most interesting variant of these is probably Yang’s (2002) stochastic Variational Model (cf. also Clark 1992). Yang lets his acquisition procedure try out all kinds of parameter settings in a random way. The moment of irreversibly setting any parameter is delayed. Most such arbitrary settings will of course be widely off the mark and not succeed in deriving much of the primary data. Yet, the chance of a parameter being tried out in a plus or minus setting is enhanced by “bonus points” when that setting happened to be involved in a few successful derivations of primary data. By adding a clever arrangement for assigning bonus points, Yang’s computer simulations of the acquisition procedure sift out the correct parameter settings after an acceptable amount of time (program re-runs). There is a remarkable secondary effect. Yang’s system of a priori parameters does not include stipulations about an order of parameter setting. Yet, when applied to the real primary data, the correct parameters come up in a specific order. Some parameters assemble a decisive amount of bonus points before others. Let us remind ourselves of the original problem. There was, as Dresher (1999) diagnosed, an inevitable and principled gap between abstract parameters and primary data. The attempts to bridge the gap between abstract parameters and plausible acquisition procedures consist so far in stipulating more innate structure, such as an (a priori) order of parameter setting, or (a priori) cues added to the parameters, with the effect that the difficulty of local maxima is nullified. Sometimes, as in the case of Yang (2002), there appears to be an innate talent for probability computations that identify abstract pattern factors. One may wonder why the human animal had to come up with languages that require such elaborate innate and task-specific provisions. There are other human activities that display kinds of rule-governed creativity. The art of playing chess in sequences of moves, the art of improvising music in sequences of striking intervals, or the art of crafting sequences of interacting tool sets are all exclusive to human beings. Do these and all other cultural activities require innate task-specific talents as well? And if not, why is language the exception? 1.2 Innate parameters? Chomsky sometimes seems to support such skepticism about task-specific talents. He repeatedly mentions the case of arithmetic (Chomsky 1987, 2005).
Arnold Evers & Jacqueline van Kampen
It has been found that counting five or six bananas is well beyond the grasp of otherwise intelligent chimpanzees. The difficulty seems to be this: it is not an inherent property of a banana to come in as first or second banana. A banana is second banana only because just before some other banana has been indicated as first banana for otherwise pretty arbitrary reasons. In order to count bananas, you must not only be attentive to bananas but also, and simultaneously, to the sequence of your own naming activities. This is more than any chimp with a healthy interest in bananas can work himself up to. The human being, by contrast, has sufficient attention and memory to spare. It keeps track of its own naming activities, sees the parallel between the naming series and the banana series, and gets the counting trick. Each banana is associated with a provisional number name, say “six/sixth”. The name is determined by the preceding number name, say “five/fifth”. Each number name must have a successor number name if the counting game is to go on. Inevitably, the amount of bananas will at a certain moment exceed the amount of numbers provided by some five or ten finger nursery rhyme. Other tricks must follow to provide new number names to measure up to the number of naming activities. Recursively raising ten spread fingers may do, as one still sees at some markets. Then and there, at the market, the world of mathematical structures opens up as a matter of achievements in cultural history. The parallel between the bananas and the mental acts is neither fruit-stuff nor (subjective) mental stuff; it is mathematical stuff. Eventually, mathematical structures can become ever more complex and surprising. It would be curious to propose that a new invention for number structures comes out of an innate neural surprise box. Yet, the inventions are predetermined. For example, the decimal notation for fractions was invented by the Dutch engineer Simon Stevin in his arithmetic book De Thiende (1596, “The Tenth One”), but it had been invented earlier by Islamic astronomers (Al-Kashi Sallam At-sama, 1407, “The Stairway to Heaven”). It answered a system-internal problem with dividing numbers. Nobody would relate that invention to until then inoperative neural constraints in the human brain. When something like the accumulative invention of system-internal solutions holds for mathematical structures, why should it not hold for grammatical structures? It therefore seems reasonable to be skeptical of all assumptions about parameters as innate task-specific a prioris. This seems close to abandoning the notion parameter as such. Yet, that is only apparently so. The parallel with arithmetic may be helpful. It is perfectly possible that an abstract system is learned and applied due to a set of local options that invite some trivial rules of thumb. The true effectiveness of the overall system is only understood afterwards or even not at all. The parameter is effective in the system rather than being an abstract decision by the user.
Parameter setting and input reduction
2. Input reduction Accumulative learning means that one begins with simple problems before one gets a clear picture of the more complex ones. We propose that the language acquisition procedure begins with a radical reduction of the input. The reduction yields a set of utterances that consist of two directly identifiable elements. The reduction follows from a simple common sense principle in (3). We will give it a subtler form in (6). (3) Reduction of input to intake: Leave out what you cannot fit in.
The reduction procedure in (3) initially ignores all grammatical markings (functional categories): articles, copulas, auxiliaries, verbal inflections, connectives. This is because a functional category 〈Fi 〉 indicates a grammatical relation between two phrases [XP [F YP]FP]. It is a word that carries little meaning beyond the syntactic relation. The word and does not mean “pair”, the word but does not mean “objection”, the word is does not mean “property”. Although each of the functional categories in child-directed speech is 100 to 300 times more frequent than an arbitrary denotational word, functional categories cannot be identified and learned before regular combinations of denotational words are perceived and sensibly combined. Hence, sentences in the child-directed speech like Isn’t the bear nice?; The bear is nice, isn’t he?; I want the bear to be nice are turned into [bear nice]. A set of binary word constructions is the result. They relate pragmatically as topic-comment or as operator-comment.3 The operator-like words have an immediate situational deixis. This at least happens in the child’s actual output. A longitudinal analysis for one language and one child is necessary to illustrate how input reduction leads to an order of acquisition steps. The present article considers the acquisition steps in Dutch taken by the child Sarah as a case study. The files are available in CHILDES (van Kampen corpus, MacWhinney 2006). Typical two-word sentences in Dutch child language are given in (4). (4) a.
modal operator + Y (comment) kwi beer (I-want bear) ga slapen (go sleep) moet op (must on/up)
. The example set of typical binary constructions in (4) is not exhaustive. We just give types that are relevant to the present exposition.
Arnold Evers & Jacqueline van Kampen
b.
deictic operator + Y is drinken (is drink) dis pop (this-is dolly) dats beer (that’s bear) dit op (this gone)
c.
X (topic) + Y (comment) beer lief (bear nice) papa slapen (daddy sleep) schoen aan (shoe on)
Most examples of pragmatic illocution operators in Dutch derive from modal verbs in the input: kwi(l) (wish “I want”), kga (intention “I intend/I am going to”), moet (requirement “it should be/it goes”). The (later) subjects are not yet present as phrase or word. They rather are “mode-implied’, since these early operators do not vary in 1st, 2nd, or 3rd person (van Kampen 2006b). The intention is always I want, it must (be that). We feel that it would be adult over-interpretation to analyze these early pragmatic illocution operators as finite verbs. Finite verbs presuppose far more system (a verbal paradigm 〈±finite〉 and a subject-predicate construction) than is available. There is also a statement and name-giving operator dis/dats. It seems to fuse the demonstrative and the copula. It might once again be adult over-interpretation to analyze these early forms as phonetic fusions. They fit the binary frame more naturally as a single illocution operator unit.4
. In a typological study of intransitive predicates in 410 languages, Stassen (1997: 99) observes that in many languages the morphological form of the copula recalls the morphological form of demonstratives. This seems less strange if both elements are picked up in child language as an illocution operator ‘‘statement’’ introducing the utterance with colons. Such an analysis is possible for (at least) Dutch (i), English (ii), and French (iii) child language.
(i)
Dutch (Sarah: week 120, Van Kampen corpus) a. is: gieter/bordje/water/plakkertje (is: can/plate/water/sticker) b. is: heet/lekker/vies (is: hot/nice/dirty) c. is: mij/jou niet (is: mi(ne)/not you(rs)) d. is: op (is: on) e. is: koud buiten (is: cold outside) f. is: deze koud buiten (is: this one cold outside)
(ii)
English (Nina: week 112; Suppes corpus) a. is: Mommy living room b. is: Mommy’s living room c. is: a girl
Parameter setting and input reduction
The binary word utterances have minimal grammatical structure. They either relate two denotational words to each other or they relate a denotational word with a fixed operator. Relating parts of an utterance in a grammatical frame offers the same problem as banana numbering. The succession of elements reflects a structured situation. Again, perceiving the parallels between situational structures and grammatical markings is beyond the grasp of the nonhuman primates. The input reduction (3) will initially set aside all functional categories as unknown, not directly interpretable 〈F?〉. The learner need not be aware of a grammatical difference between denotational words with a direct meaning and functional elements that mark a more abstract level of meaning and categories. The distinction between denotational words and more abstract elements is simply imposed upon the learner as elements provisionally comprehensible {beer, slapen, lief}, {dis, kwil} versus high frequency elements not yet comprehensible {tense -te, agreement -t, articles de, een}. The words with immediate meaning can be acquired separately as names or characterizations within an immediate situation in front of you. Functional categories are left out not because they have no (semantic) meaning, but because their meaning can only be ascertained within the syntactic and morphological oppositions that are still to be acquired. It is in any case a fact that child language begins with binary utterances that lack functional categories, although these categories (articles, auxiliaries, inflectional endings) are highly frequent in the input language. This period without functional categories is the period that Lyons (1979) may have had in mind when he suggested that early child language might have proto-predication as a forerunner of predication and also that child language might have proto-reference as a forerunner of reference. We propose an interpretation of Lyons (1979) in (5). (5) a. proto-reference ~ naming function b. proto-predication ~ comment/characterizing function c. proto-illocution ~ (pragmatic) operator function
The proto-distinctions are made within a situation-bound system. Each element is supported by the situation it is used in. The relation between the elements is
(iii)
French (Grégoire 1; 10.20 week 98; Champaud corpus) a. est: crocodi(le) (is: crocodile) b. est: casquette d’ Adrien (is: cap of Adrien) c. est: chaussette Victor (is: sock Victor)
Stassen’s parallel between the demonstrative and the copula may find its origin in proto-grammar where the distinction D0/I0 is not yet made.
Arnold Evers & Jacqueline van Kampen
pragmatically guessed. There is a cognitive abstract frame that schematizes that situation-bound understanding. The elements that signal an abstract frame are present in the input, but they have been set aside as 〈F?〉 by the reduction to intake. When the relation between two denotational elements X and Y is nevertheless repeatedly and correctly guessed, the learner may observe that an abstract frame for the relation was given in the neglected elements 〈F?〉.5 The utterance bear nice allows the insertion of a copula when it is recognized as a kind of property attribution. We assume that reference, predication, and illocution exist only as factors within a grammatical system. Before the grammatical structure is in place, and the functional categories are acquired, all words are used in a strictly situationbound manner and there is no grammatically encoded reference, predication, or illocution. Since the adult input is dense with functional categories, all with languagespecific restrictions, the acquisition procedure must apply a fixed strategy. We propose, as a specification of (3), the reduction principle in (6). (6)
Reduction of input to intake a. First reduction: Replace each functional category still unknown with 〈F?〉. b. Second reduction: Set aside all input strings with more than one 〈F?〉. c. Residue: An intake of the form 〈F?〉/〈Fi〉.
The residue (6c) is a set of input sentences that yield an acquisitional focus for the category/parameter value 〈Fi〉. The residue represents the minimal pattern from which the functional category 〈Fi〉 can be acquired within a certain language. It is our contention that the language acquisition procedure scans the input for such patterns. We will call these patterns evidence frames. The input reduction is such that the acquisition procedure concentrates on a single grammatical value 〈Fi 〉 before the attention switches to the next acquisition step. Part of the acquisition procedure indeed confirms such a Single Value Constraint. Language is structured in such a way that the evidence frames must turn up in a linear order signalled
. It has been observed by Locke (1997) that early child language – the form without grammatical markings – is processed in the (then dominant) right brain hemisphere. The left hemisphere is more hesitant and applies no reduction of functional markings. It preserves them as phonological peculiarities. When the pragmatic understanding by the right hemisphere becomes more difficult due to the rapid expansion of the lexicon, the left hemisphere contains hints about the relations due to the functional markings it still attends to. The systematic reliance on these hints, the functional markings, represents the grammaticalization of understanding and a shift from the right to the left hemisphere. Locke points out how the shift in grammar is strikingly supported by brain scans of small children (Mills et al. 2004).
Parameter setting and input reduction
by the order of acquisition steps. Note that the order of acquisition steps is not assumed to be universal. The order of acquisition steps is derived from evidence frames and the evidence frames are derived from a natural reduction of the input language. The strategy in (6) is not language-specific for getting the evidence frame, but its outcome, the evidence frame, is just that: a language-specific frame. The reduction of the input to an intake selected for a single grammatical point 〈Fi〉 confronts the acquisition procedure with a repetitive set of short sentences all containing the same functional category in the same position. In a sense, the evidence frame is a language-specific drill for the category 〈Fi〉. An example could be the copula construction in (7). (7) a. Learning step: identify the pragmatically understood 〈F?〉 as 〈Fi〉 b. In diagram form: (bear is nice) 〈FP?〉 〈FP?〉
XP [beer] 〈F?〉 ↓ is
YP lief
– binary pattern in proto-grammar – selection relation – stress pattern – high frequency
The meaning of the binary frame in (7), “property attribution”, is pragmatically guessed and demonstrates the contribution by 〈F?〉. We assume that all functional categories fit such an evidence frame. When 〈F?〉 is identified as the grammatical function 〈Fi〉, a full utterance of the input has been identified. When this acquisition step has been made, the filter in (6) may reapply and select a new intake from the same type of input (cf. the Single Value Constraint, Berwick 1985: 108; Clark 1992: 90). The next functional category then comes into acquisitional focus. Its context, the context of 〈F?/Fi+1〉 allows the presence of 〈Fi〉. 〈Fi〉 passes the filter in (6), since it has been identified and acquired. This suggests an ordering of acquisition steps. The later steps select more language-specific and more structured contexts. Turning this around, we may see the empirical presence of ordered acquisition steps as an indication that a data reduction filter (6) has been at work. If (6) selects the relevant data-set from the input, the category 〈F?〉 becomes the acquisitional focus. That focus turns the diffuse stimulus of the input into the effective stimulus of a temporary intake. The data selection function of the earlier acquisition steps has been noted in Berwick & Weinberg (1984: 284, note 2). It is proposed here that data selection
Arnold Evers & Jacqueline van Kampen
Di+1 by the provisional grammar Gi is crucial. The unreduced input represents a diffuse stimulus, but the selectively reduced input by the procedure in (6) may prove to be a sufficiently focused stimulus. It may eliminate Chomsky’s argument from the poverty of the stimulus. As a first result, the reduction of input to intake provides the learner with elementary Greenberg-type patterns that facilitate the acquisition of the major parameters of the system (Greenberg 1963). Studying the order of the child’s acquisition steps provides the key to the built-in learnability of grammar. The initial acquisition of the copula as an illocution operator for root statements (is lief – “is nice”) subsequently allows the evidence frame [name 〈F?〉 comment] to be solved (beer is lief – “bear is nice”). The copula now marks the comment within a topic-comment structure. The evidence frame has been represented as the change from a binary construct “XP (topic) YP (comment)” to a triple one “XP [〈F?〉 → 〈Fi〉 copula] YP”. This is not close enough to the actual procedure. In the first place, the copula is but one of the elements that appear in the in-between position. This leads to a further, more important consideration. The various operator elements 〈Fi〉 are not freely combinable with any “comment” element. Rather, the operator elements, – the later modals, copulas, aspectuals and auxiliaries – associate with different subclasses in the complement. These stereotype selections can be modeled by a convention. The denotational element in the comment is countermarked in the lexicon as 〈−Fi〉 when it combines with and is selectable by the operator 〈+Fi〉. The associative counter-marking opens the general possibility that denotational elements split up into categories according to their combinatory restrictions with given functional categories.6 The selection restrictions between operator elements and the comment do not change when the binary stage changes into more complex triple combinations, as in (7). When the binary system contains the type of combinations in (8), the acquisition procedure is near to being able to recognize and produce the structures in (9).7
. The evidence frames seem close to the position of Tomasello (2003) that language acquisition is construction-based. There is an important difference, though. The evidence frames are supposed to add syntactic features 〈+Fi〉/〈−Fi〉 to lexical items. Once they have done that, evidence frames disappear from grammar. . The term ‘‘recognize’’ may be somewhat unfortunate. An experimental tradition of the last twenty years has shown that children recognize far more structure than they are in fact able to produce. This is not unlike perceiving a fugue by Bach or a perspectual effect by Esher. Perceiving something does not yet enable one to reproduce anything like it. For some fortunate reason, we can perceive and recognize more than we can produce.
Parameter setting and input reduction
(8) a.
is lief beer lief
(9) a.
is moet
[beer lief ] [beer slapen]
beer beer
[is lief ] [moet slapen]
b.
b.
(is sweet) (bear sweet)
moet slapen beer slapen
(must sleep) (bear sleep)
((now) is (the) bear sweet) ((then) must (the) bear sleep) ((the) bear is sweet) ((the) bear must sleep)
What happens in the three-word stage (9) is a nesting of two types of binary constructions into each other. The selection restrictions are preserved. The inner units [topic comment] in (9a) and [operator comment] in (9b) had already been constructed earlier. The binary system is preserved with either the operator in the peripheral position as in (9a) or the topic as in (9b). The patterns in (9) are reinforced by the maternal input, as indicated by the glosses, e.g., (nu) is beer lief – “(now) is bear nice”. These maternal structures have a subject topic (9b) or a dummy topic followed by subject-inversion (9a), but that much structure cannot yet be perceived by the child. Distinctions like (i) external argument/subject (ii) finite verb and (ii) subject-inversion belong to a more advanced system that is not yet active. It is important to notice this because an analytical perception of (9a) as free inversion of finite verb and subject is the cue in Fodor (1998) and Lightfoot (1999) for setting a V-second parameter. Even if discussions about the A-/A-bar status of the initial phrase are relevant, they cannot be relevant at this stage of acquisition (see van Kampen 1997: chapter 3). The acquisition procedure must first single out a class of words in predicate-final position, make a distinction between the 〈±finite〉 elements in that class, introduce the category V, construe an initial head position of finite verbs and use that position to identify subjects and topics in its specifier.
3. Evidence frames for underlying directionality The acquisition analysis below is based on the assumption that the target, adult Dutch, is an SOV language (subject precedes the predicate and object precedes the verb), with the complication added that the finite verb in root clauses is not in the SOV predicate-final position, but appears in an initial position right after the first major phrase (the V-second position). The marked and prominent position of the finite verb probably serves as a warning that the sentence has an illocutionary value (cf. i.a. van Kampen 1997; Truckenbrodt 2006). Subordinate clauses lack the illocutionary function and do not apply the V-second rule. The finite verb of the subordinate clause is therefore once again in predicatefinal position.
Arnold Evers & Jacqueline van Kampen
(10)
CP topic
CP
arguments Co Predicate or V〈+C, +fin〉 dummy (er) {arguments} V〈+fin〉
The 〈+fin〉-marked verb is associated with two positions, a predicate-final position that serves to subcategorize the denotational verb for its complements, and the Co-position that serves to mark the illocutionary function of the root clause. As in Weerman (1989) for reasons of method, and in van Kampen (1997, 2006b) for learnability reasons, no separate Io/To-position is assumed here (see also Watanabe 1994). The Germanic V-second parameter can now be presented as setting a 〈+C〉 value for 〈+fin〉. A finite verb marked V 〈+fin, +C〉 will move into the Co-position, not unlike the Do-marked 〈+wh〉 will move into the SpecCP. The 〈+C〉-value is de-activated in subordinates by the complementizer constant 〈+C〉. It is not the point of the present section to argue for the minimal structural analysis in (10). The basic point is that any structural analysis of the V-second clauses requires a learnability account of the kind given below. An alternative analysis, for example Zwart (1993), that Dutch is after all consistently head-initial and/or requires a host of additional verbal positions, meets with somewhat different, but no less serious, learnability problems. In all variants, a plain surface distribution must offer the learner access to underlying regularities, preferably with a minimum of a priori stipulations. As noted in section 1.1, Gibson & Wexler (1994) confront the problem of setting the V-second parameter. They assume the Single Value Constraint, which holds that parameter setting is a stepwise procedure where each step sets a single parameter value. They show that the acquisition procedure for the Dutch type of grammar (SOV/V-second) should somehow succeed in setting the underlying Greenberg directionality parameters (subject before predicate, object before verb) before setting the V-second parameter. Yet, there is an input problem here. The V-second surface effect is present in nearly all input sentences (93% in our data), with the result that the “underlying” SOV structure is not unambiguously present. The predicate-final position of denotational verbs comes out in the subgroup of constructions where the finite verb is an auxiliary/modal and the denotational verb is transitive – see (11): (11) a.
Subject Aux 〈+fin〉
papa gaat [een boek daddy goes a book (‘Daddy will read a book’)
[object
V 〈+inf 〉] lezen] read
Parameter setting and input reduction
b. Adjunct Aux 〈+fin〉
[subject object
V 〈+inf〉]
(narrative inversion)
nu gaat [papa een boek lezen] now goes daddy a book read (‘Now daddy will read a book’)
The construction types in (11) represent only a small subset of all input sentences (15%) since explicitly transitive verbs are not that frequently present. It is a legitimate question asking why and how the acquisition procedure should select this subset to construct a grammar. In order to make this problem more clear, Evers & van Kampen (2001, section 3.1) assembled the input given by the mother in the months just before her daughter Sarah acquired the V-second rule (van Kampen corpus, CHILDES). The presentation below (in sections 3.1–3.2) is slightly schematized, but basically the same as earlier. There were 1017 utterances with a verb, 95% of them contained a finite verb, 93% of root sentences with the finite verb in V-second position and 2% subordinates with the finite verb in final position. Two subsets were constructed from the set defined by the presence of a finite verb, one subset containing the constructions with a finite verb and a subject, the other subset containing the construction with a finite verb and an object/complement. The question is whether the subsets could serve to identify fixed positions for subject and complement. The actual input for subject or object identification is complicated by the fact that the denotational theta-assigning verb may move into initial position and precede rather than follow the complement and often, in case of subject-inversion, precede the subject as well, see (12). (12) a.
papa leest een boek tVfin daddy reads a book
(object follows the denotational verb (VO))
b. nu leest papa een boek tVfin(subject follows the verbal predicate (VS)) now reads daddy a book
These distributional complications, consequences of the V-second rule constitute a hindrance for the early acquisition procedure. As long as the underlying structure has not yet been established, the V-second effects appear as mere distributional surface noise. The sentences in (11) first have to be analyzed as revealing an underlying structure. 3.1 Paradoxical input The adult grammar should eventually construct a relation between the subject and the finite verb, but the subgroup defined by the finite verb and the presence of an explicit subject shows the following puzzling percentages. Regardless of whether the finite verb is a theta-assigner or not, the subject precedes the finite verb in 30%
Arnold Evers & Jacqueline van Kampen
and follows it in 70% of the constructions. The percentages 30/70 are proportions between the relevant subsets.8 (13) a.
Subject-V 〈+fin〉 papa moet een boek lezen papa leest een boek dat boek is leuk
(daddy must a book read) (daddy reads a book) (that book is funny)
b.
V 〈+fin〉 -Subject nu gaat papa een boek lezen nu leest papa een boek leest papa een boek? nu is de sap op op is de sap
(now goes daddy a book read) (now reads daddy a book) (reads daddy a book) (now is the juice gone) (gone is the juice)
(30%)
(70%)
It seems then that the distinction 〈±finite〉 verb does not suffice to identify a basic subject configuration. The object-verb directionality needed for the SOV type has its problems as well. The straight Object-V 〈−fin〉 pattern reaches 45% (14a) and trails other transitive constructions (55%). This is due to transitive constructions where the denotational verb moves by the V-second rule or the object gets topicalized (14b). Again, the percentages are proportions within the subset of transitive verb constructions in the maternal input (cf. note 8). (14) a.
XP-Auxfin-[Object-V] papa moet een boek lezen nu gaat papa een boek lezen
(daddy must a book read) (now goes daddy a book read)
(45%)
b.
movement of the denotational verb papa leest een boek nu leest papa een boek leest papa een boek?
(daddy reads a book) (now reads daddy a book) (reads daddy a book?)
c.
topicalization of the object dat boek leest papa dat boek gaat papa lezen
(55%)
(that book reads daddy) (that book goes daddy read)
. This does not contradict the count in Van Kampen (1997: 59) and in Lightfoot (1999: 153) that the finite verb is preceded by a subject in 2/3 of the cases and by a non-subject topic in about 1/3 of the cases, counting simple statements in the maternal input. The actual input is further complicated by the subject inversions due to questions and narrative inversion. Within that larger set, the percentage of subject-V 〈+fin〉 configurations lowers to 30%.
Parameter setting and input reduction
The quantitative proportions in the input for the acquisition of Dutch seem designed to prove Chomsky’s point that the input offers the acquisition procedure no more than a confusing and poor stimulus, if it offers any stimulus at all. On the other hand, no matter how much a priori guidance one is willing to postulate, there can be no warning in the cradle of the type “be careful, within a short time you will enter the domain of an SOV language marked by a V-second root movement”. In other words, the input itself has to offer access to the system. Paradoxically enough, it seems to do so. The Dutch child begins by ignoring the massive evidence for the V-second position of 〈+fin〉. Initially, she strongly prefers the so-called “root infinitives”. This will be shown by longitudinal graphs in the next section. By that approach, she immediately opts for the object-verb order, in spite of the fact that this order constitutes only 45% of the transitive constructions in the input. Finally, the rise of finite verbs in child Dutch correlates with a systematic appearance of the subject, although the subject shows no obvious distributional preference for a standard position in the specifier of the finite phrase (CP). 3.2 Intake by evidence frames Let us consider now how systematic input reduction and its resulting languagespecific evidence frames control the acquisition procedure. As observed by Lebeaux (1988: 11f), the child’s initial structures for two denotational words in proto-grammar represent minimal theta-assigning structures of sister government. If the acquisition procedure is attentive to sister government structures, content elements are perceived as belonging to linearly fixed frames for theta-assignment. The choice of the frame follows from the most frequent PF predicate form. Let the acquisition procedure focus on adjacency when setting up the thetaframes for lexical items. Distributions of transitive structures for deriving the OV order are given in the bracketed phrase in (15). Keep in mind that the child is as yet unaware of the V-second rule/finiteness marking. Hence, she may also consider the bracketed phrase in (15b) as potential evidence for a theta-frame. The remaining transitive structures (15c) are irrelevant and neglected due to the initial data reduction by the adjacency requirement: the OV/VO relations in (15c) are not string-adjacent as the subject intervenes. (15) Frames for Object-Verb/Verb-Object a.
Evidence: adjacent Object-Verb (78%) papa moet een [boek lezen] (daddy must a book read) nu gaat papa een [boek lezen] (now goes daddy read a book) dat [boek leest] papa
Arnold Evers & Jacqueline van Kampen
b. Counter-evidence: adjacent Verb-Object papa [leest (een) boek] (daddy reads a book) c.
No evidence nu leest papa een boek leest papa een boek? dat boek moet papa lezen
(22%)
(now reads daddy a book) (reads daddy a book?) (that book must daddy read)
A parallel evidence filter can be used to determine prospective subject directionality, where the predicate functions as the theta-assigner. Subject directionality holds for all predicates, verbal and non-verbal (copular constructions). Within the present view, early child grammar cannot distinguish these two kinds of predicates, since the verbal paradigm is not yet available. Therefore early child language automatically addresses the more abstract, less specified subject-predicate directionality of the input rather than subject-finite verb directionality.9 (16) Frames for Subject-Predicate/Predicate-Subject a.
Evidence: adjacent Subject-Predicate nu gaat [papa [(een) boek lezen]] [papa leest] een boek nu (is) de [sap op] ik vind [papa lief]
(68%) (now goes daddy a book read) (daddy reads a book) (now is the juice gone) (I find daddy nice)
b. Counter-evidence: adjacent Predicate-Subject
nu [leest papa] een boek [leest papa] een boek?
c.
No evidence
papa moet een boek lezen dat boek is leuk op is de sap
(32%)
(now reads daddy (a book) (reads daddy (a book)?) (daddy must a book read) (that book is funny) (gone is the juice)
The theta evidence frames for the subject argument will place brackets as in (16a), setting aside all the functional categories (auxiliary, modal verb and copula) that precede the subject. That reduction allows the acquisition procedure to tap the rich supply of subject inversion structures in the input. These support the acquisition of . The subject is string-adjacent to the theta assigner within the evidence frame when the early object verb frames in (15a) are read as incorporations (Van Kampen 1997). All early theta assignments are ambiguous in status between theta-frame and stereotype lexical association as Lebeaux (1988: 13) observes. They certainly fit Tomasello’s construction-based grammar in early child language. Explicit argument marking by Do appears later, after V-second, see the graph in (18) section 4. The lexical associations in proto-grammar do exhibit the dominant SOV order and they initially obviate the V-second pattern as shown by the 〈+C, +fin〉 graph for V-second in (18).
Parameter setting and input reduction
theta frames for denotational verbs. Some counter-evidence is, however, brought in by subject-inversion with a denotational verb, as illustrated in (16b). These constructions in (16b) where the denotational verb is finite must either enter the system as fixed idioms or remain unanalyzed. The prospective subject is now determined by lexical structure. In (16c) the copula or modal verb is between the subject and predicate. For that reason, these sentences are irrelevant and not considered as long as the acquisition procedure is still controlled by Lebeaux’s filters of theta-assignment under strict adjacency (Lebeaux 1988). The bracketed evidence frames in (15) and (16) now yield (17) as input evidence for argument order. (17) Directionality: Input percentages on string-adjacent evidence frames Parameter
Ratio in input [+evidence]
Ratio in input [−evidence]
Subject is adjacent to and precedes theta-assigner
68%
32%
Complement is adjacent to and precedes theta-assigner
78%
22%
The bracketed evidence frames in (15a) and (16a) suggest that the acquisition procedure could set the argument frame as a property of the lexical item that assigns the theta roles by the following practical decision. If there is support from some 2/3 of the distributional evidence within the relevant frame, the directionality parameters are set for the underlying SOV order in Dutch. As will become clear in the next section, the stage for the acquisition of the V-second rule has now been set.
4. Order of acquisition steps When a previously unknown category 〈F?〉 has been identified as 〈Fi〉, the acquisition procedure has two expressions at its disposal [X + Y] (bear nice) and [X + 〈Fi〉 + Y] (bear is nice). Eventually, the more specified adult variant will block the less specified child variant. Blocking in language acquisition takes some time and it can be traced by longitudinal graphs (van Kampen 1997). The acquisition graph identifies a ‘parameter setting’ by its blocking effect on the less-specific predecessor. Graphs for different parameters reveal the order of acquisition steps. See the graphs in (18) for Dutch for predicate marking by V-second and argument marking by Do (cf. van Kampen 2004). The graphs are constructed from the Sarah corpus available in CHILDES (cf. MacWhinney 2006). The V-second graph establishes the marking of a predication by a 〈+C, +finite〉 illocution operator
Arnold Evers & Jacqueline van Kampen
in the Co position.10 The Co 〈+fin〉 factor generalizes over a variety of categories {copula, auxiliary, modal, finite morphology}. Similarly, Do 〈+det〉-marking is the explicit marking of arguments for reference. The reference factor also generalizes over a variety of categories{article, demonstrative, possessor, quantifier}. (18) Dutch Sarah: the acquisition of Co 〈+fin〉-marking and Do-marking 100
〈+fin〉
Percentage
80 60 40
〈+det〉
20
100 110 120 130 140 150 160 170 180 190 200 Age in weeks
Dutch Sarah applies systematic 〈+C, +fin〉-marking (V-second) almost half a year earlier than systematic Do-marking (articles and their like). She acquired Co 〈+fin〉 -marking at week 120 and Do-marking at week 145.11 The same order of appearance, the acquisition of finite verbs preceding the acquisition of determiners, has been found for French (cf. van Kampen 2004), English (cf. Brown 1973) and for Rumanian (cf. Avram & Coene 2004, this volume). The Co 〈+fin〉-marking graph of Sarah in (18) simply lists the number of elements in the first or second position that could be seen as a finite verb in the adult system. What actually happens in the rise of the graph is more intricate. The lower part of the graph can be accounted for by a more simple system. There are designated elements that function only as pragmatic illocution operators in the initial utterance position. These are the later modal verbs. They do not yet
. We prefer Co 〈+finite〉 rather than Co 〈+tense〉. The opposition 〈±finite〉 is present in the child’s early grammatical system, whereas the opposition 〈±tense〉 comes in later, when the Vsecond rule is already acquired. The question is not whether the child is cognitively aware of time. Any conscious being is aware of time. The question is when such awareness is systematically expressed in forms of language. The oppositions 〈±finite〉 and 〈±tense〉 are more problematic in the acquisition context for English, but we leave this aside here (but cf. Ramchand & Svenonius, this volume). . It is true that there are examples with a Do-marking and without a Co 〈+fin〉-marking for Sarah, especially around week 120, as the graphs in (9) suggest. Note that this is not relevant to the question of why systematic Co 〈+fin〉-marking precedes systematic Do-marking.
Parameter setting and input reduction
require the presence of a full predication with a topic and a subject. When the graph rises above 50%, finite forms of denotational verbs come in. The second part of the 〈+C, +fin〉 graph rises due to the operators that are derived from denotational verbs. At the same time, the finite form begins to imply a full predication frame (topic, subject and arguments). The child’s optional infinitive constructions (papa boekje lezen – “daddy booklet read”) disappear either for an explicit modal verb (papa moet boekje lezen – “daddy must booklet read”) or for a finite denotational verb (papa leest boekje – “daddy reads booklet”). The V-second construction is, however, anomalous in two ways, as the diagram in (19) shows: (19)
CP topic
CP C° V〈+C, +fin〉
Predicate
{arguments}
t〈+fin〉
There is no designated illocution operator present in C o and there is no theta-assigner present in the predicate-final position. The theta-assigner is a morphologically marked 〈+fin〉 and it appears in the Co position of the illocutionary operator. The set of theta-assigners that allow this switch form a category “verb’’. Some notation is needed to express that the theta/argument frame known from the 〈−fin〉 variant reappears for the 〈+fin〉 variant. The derived secondary effect of the theta structure is expressed by a movement of the 〈+fin〉 form from the predicate-final position into the Co-position. If movement rules are not an option in a grammatical model (e.g., Categorial Grammar, HPSG, LFG), a derived theta frame for 〈+finite/illocutionary〉 could be expressed by a separate frame for the 〈+C, +fin〉 form. The important point for the acquisition procedure is that the V-second rule identifies the Vo in C as a distributional variant of the Vo in predicate-final position. It is not the task of a grammar or an acquisition procedure to merely list variants; they have to capture the underlying identities. One may see the acquisition of the V-second rule in the following way. There was originally an option in proto-grammar. The topic and/or the illocution marker could be left out. That option gets marginalized in favor of a norm that requires both the topic and the illocution marker to be realized. Constructions where a suitable modal or copula is lacking bring in a finite form of the denotational verb as illocution marker and preserve the verb’s full argument structure. Constructions where a suitable topic is lacking conform to the adult pattern via the insertion of a dummy topic or
Arnold Evers & Jacqueline van Kampen
adverb.12 See the examples in (20) and (21). Proto-forms are given in (20). Forms with the first grammatical devices appear in (21), where (21a) realizes all options from the proto-grammar. (22b) realizes the V-second shift of a denotational verb, (22c) the use of a dummy topic and (22d) all the previous devices together. (20) a. beer slapen b. gaat slapen c. (in) bed slapen (21)
a. b. c. d.
beer beer er/d’r er/d’r
gaat slaapt gaat slaapt
(bear sleep) (goes sleep) ((in) bed sleep) in bed slapen in bed een beer slapen een beer in bed
(bear goes in bed sleep) (bear sleeps in bed) (there goes a bear sleep) (there sleeps a bear in bed)
The systematic Co 〈+fin〉-marking and the Do-marking of arguments lead to further acquisition steps, beginning with a grammatical decision procedure on the category membership V versus N (cf. van Kampen 2005). 4.1 Acquisition steps due to local evidence frame IP/DP The systematic marking of comment parts by a variety of 〈F?〉 elements {copula, modal, auxiliary, inflection} divides the denotational elements of the comment into a subclass “V” and a subclass “non-V’. When a denotational element has access to all the Co 〈+fin〉-markings in (22), it is a verb. Verbs and nouns are considered here as derived from the grammatical system. They are not considered to be innately or cognitively given. (22) Xo = V b.
when it may enter the Co contexts in a, b and c ] [〈+C, +modal〉] ….. [ 〈+ infinitive〉 [〈+C, +fin〉] …..
c.
[〈+C, +aux〉]
a.
….. [
] 〈+past participle〉
. V-second and dummy topics are not simultaneously acquired. The systematic introduction of dummy topics does not take place before the acquisition of 〈± definite〉 articles on arguments and the acquisition of Do-marking follows the acquisition of V-second (see (18)). This ordering in terms of which Do-marking on arguments precedes the introduction of dummy topics is predictable since topic selection is sensitive to the ± definite distinction.
Parameter setting and input reduction
The somewhat elaborate context condition in (22) simply states that a denotational element is V when it fits the verbal paradigm.13 The parts of the paradigm (22) refers to are present in early Dutch child language. Other parts of the verbal paradigm, especially the 〈finite〉 oppositions for tense and number, play no part in the acquisition of the category V in Dutch. They appear later around week 140, as we will show in section 4.2. The intention of (22) is to stress that the paradigm is learnable from the Co evidence frame. The acquisition of the category V by means of the evidence frame in (22) may take up to half a year in Dutch child language. This amounts to some high six-digit number of short parallel sentences to get the verbal paradigm and the category V. It was Briscoe (2000) who pointed out that the speed of acquisition steps is to be measured by estimates of the number of learning experiences. Within this perspective, language acquisition is the most intensive onslaught that our brain ever experiences. Grammar therefore results from an unrelenting round-the-clock course that continues for years, almost a brainwashing. Let the rise of the category V be simultaneous with the rise of 〈+C〉 by the stipulation that 〈+C〉 automatically causes a countermarking in its denotational partner 〈+C\−〉. The countermarking simply indicates “fits the 〈+C〉 paradigm in (22)”. The 〈+C\−〉 is better known as V. This morphological partner marking eventually reinterprets all auxiliaries as verbs. The identification of the category V by means of Co 〈+fin〉-marking refers to various constructions. Sometimes the 〈+C, +fin〉 factor is a morphological part of the Vo-element, sometimes it is a factor outside the verbal phrase at a distance from the Vo-element.
. This view on parts of speech is not uncommon. Nouns and verbs are identified as different kinds of words by their morphological paradigms. In as far as a language does not offer such paradigms, it will offer functional words 〈auxs〉 type Co/Io for verbs and 〈articles〉 type Do for nouns. Our main point has been that the grammatical markers that lead up to V and N in syntax will do so in acquisition as well. Baker (2003) argues that, as far as comparative grammar is concerned, one would be better off to see it differently. Many a language lacks paradigms as well as Co/Io and/or Do projections. Baker (2003) considers verbs as inherently predicative (type 〈e, t〉), and nouns as inherently referential (type 〈e〉). The functional categories have, in Baker’s view, the more circumstantial task of indicating types of verbal predication {± modal, ± tense, ± aspect, ± negation} and types of nominal reference {± countable, ± gender classifiable, ± definite}. More importantly, their presence is an addition that can in principle be left out, and is often left out in certain languages. We must at least weaken our statement and claim that our acquisition procedure may succeed even when explicit markings of the verbal paradigm (Io) or the nominal paradigm (Do) are present as options only. If so, optional particles for mood, negation and illocution should be sufficiently present to identify Vo, and occasional demonstratives, possessives and classifiers frequent enough to identify No.
Arnold Evers & Jacqueline van Kampen
The identification of the category N by means of Do-marking looks, by contrast, like a quite simple affair, that involves no paradigms or phrasal networks when the grammar is case-free, as it is in Dutch.
(23)
Xo = N when it may enter the Do context
[Do -]DP
The difficulty with Do and No is more with the acquisition of Do. Do implies two relations, one within the pragmatic discourse network and one within the network of sentence grammar (cf. van Kampen 2004, cf. also Őztürk, this volume, Ramchand & Svenonius, this volume, and Avram & Coene, this volume). The discourse Do varies over two values 〈+definite〉 and 〈−definite〉, which translates into “previously mentioned” and “not previously mentioned’. Sensibly adding the Do-markers implies a daunting pragmatic task: check your frame for presupposition and check your immediate discourse context. This means reference-tracking, something far more complex than name-giving. The Do-marked expression is not a new name. It rather represents the requirement for Do 〈+definite〉 that the temporarily built discourse network should contain a previous mention of the same argument, and for Do 〈−definite〉 that this should not be so. The Do-relation is non-local and determined by pragmatic felicity conditions. In addition, the Do-marked phrase is to be placed in a local syntactic network, where it is connected with the predicate head by means of a standard theta role. As such, one can also see Do-marking as a grammatical expression of UTAH (Baker 1988). Consequently, a Do-marked phrase includes a pragmatic index for reference tracking and an argument index for theta-role assignment (Baker 2003). There are arguments that the referential as well as the case-marking function, belong to the non-denotational Do rather than to the denotational No, although this position is rejected in Baker (2003). In the first place, when the referential index is expressed by articles, one expects that the case endings are in principle present on the article as well, as in German, and not, or only marginally, on the noun. In the second place, when there are no articles in the language, one expects case-marking on the nouns as if the denotational N’s had incorporated a Do, since the same case-marking will appear on the non-denotational free anaphors in such languages. This suggests that the indices for reference and theta role are carried by a non-denotational factor, in casu Do (cf. van Kampen 2004, 2006a and Őztürk, this volume). The non-denotational free anaphors carry the local theta index, as well as the pragmatic reference index. All of this underlines Postal’s view (1966) that personal pronouns and articles are deeply identical and both Do. There is a third argument that articles and pronouns are of the same type and it derives from language acquisition. The acquisition graph for the insertion of obligatory articles and the acquisition graph for free anaphoric pronouns coincide for Dutch Sarah in (24). See van Kampen (2004) for an elaboration.
Parameter setting and input reduction
(24) Sarah (Dutch): Acquisition of Do 〈−pro〉 (articles) and Do 〈+pro〉 (3rd p. pronouns) 100
Percentage
80 60 40
D〈–pro〉
20 0
90
100
110
D〈+pro〉
120 130 Age in weeks
140
150
160
The coincidence of the two graphs is natural when both represent the acquisition of the double indexing system of the Do. Due to Do-marking, the language of the child becomes discourse-bound rather than situation-bound. As such, the rise of the two Do-graphs (〈±pro〉) has a certain magic: it shows how space is created for the full human flexibility with respect to situations. The argument for the D 〈±pro〉 identity in Dutch can be repeated for the acquisition of articles and free anaphoric clitics in Romance languages. In child French, the acquisition of the clitic system immediately follows the acquisition of articles (cf. van Kampen 2004 and references cited there). 4.2 Local evidence frames outweigh mere input frequency We saw in section 3 that frequency plays a role in deriving the basic word order, but that V-second is delayed in spite of its 93% presence in input utterances. Something like this also holds for subject-verb agreement. The initial evidence frame for finite verbs leaves out the ϕ-features of person/ number. Finiteness is first represented in the 3rd person singular only. This is also the default in comparative grammar (Benveniste 1966: 228ff, 255f). The acquisition of Do-marking leads to a subsequent step that brings in the ϕ-feature content of Do {± person, ± number}. The acquisition of ϕ-features on the finite verb 〈+C, +fin〉 follows the acquisition of ϕ-features on Do. There is a measurement problem with “correct agreement”. Irregular paradigms may take more than a year and there are sometimes difficulties with identifying the phonological forms on recordings. What has been done in (25) for ϕ-agreement in the speech of Dutch Sarah is a measurement of clear mistakes in plural agreement. Sarah went down from 7 cases of “wrong” agreement (〈+singular〉) for 12 plural
Arnold Evers & Jacqueline van Kampen
subjects in weeks 130–135, to 0 cases of “wrong” agreement for 13 plural subjects in week 140 (cf. van Kampen 2006a for an elaboration). Late acquisition of agreement has also been reported by Ferdinand (1996) for French, and Avram & Coene (2003) for Rumanian. The Agr-ϕ steps are a matter of weeks whereas the earlier 〈+C, +fin〉 and 〈+D, +det〉 steps took a matter of months. See (25) for Dutch Sarah. (25) Step Co 〈+fin〉 20 wks
Step Do Step Do-ϕ Step Agr-ϕ 25 wks 5 wks 5 wks
The more effective acquisition plausibly relates to the more effective evidence frame. The ϕ-feature acquisitions are supported by a lexicon with categorial marking 〈+fin〉 and 〈±plural〉. The EPP, the nearly obligatory presence of the subject in 〈+fin〉-marked predicates, operates as a frequent and effective evidence frame for ϕ-agreement. The input of adult child-directed speech has of course not been lacking in ϕ-features on 〈+C, +fin〉 and Do at all; rather the ϕ-features could not become part of the intake before the acquisition of 〈+C, +fin〉 (V-second) and the EPP had been established. This is a matter of clearly perceiving initial structure before it can become a quantitative factor in acquisition. The preceding EPP structure is needed to spot the relevant points. The EPP is acquired before agreement marking. From an acquisitional point of view, the later ϕ-agreement therefore appears as a final touch rather than as a structural underpinning.
5. Movement structures The preceding section introduced a movement rule. The movement rules reorder an underlying array of heads and phrases in order to arrive at the perceived surface structure. At least two learnability problems have to be dealt with. (26) a. The perception of underlying structure How can a phrase position be perceived as an antecedent or a gap due to a movement rule? b. The distance problem How are syntactic islands learned as phrases that do not allow the relation of an internal gap to an outside antecedent?
Both problems look more manageable in unification-based approaches that trade in the movement rule for a lexical feature matching between two sister-constituents. Neeleman & van de Koot (2002) derive such an approach from Minimality principles. The left-hand sister X in (27) below is grammatically marked 〈Fa, Fb〉. These features 〈Fa, Fb〉 need a licensing context. For example, wh-phrases in the SpecCP
Parameter setting and input reduction
position are case-marked and preposition-marked as if they occupy an argument position. The right-hand sister Y in (27) should contain a grammatically definable gap based on the same licensing features 〈−/gap Fa, Fb〉. 〈+C〉 movement structure
(27) X-phrase〈Fa, Fb〉
←→
Y-phrase 〈/gap, Fa, Fb〉
Along the same lines, the Dutch finite verb in the Co position in (10) (cf. section 3) carries properties for licensing complements as if it were in the predicate-final head position on the right. Obviously, the learnability of the antecedent-gap relation is on a promising track when the learner already commands a grammar that: (28) a. spots the markings Fa, Fb orphaned in the C-projection on the left and spots their absence in the complement of C on the right. b. projects the licensing markings 〈Fa, Fb〉 according to existing head-complement conventions.
Fortunately, it is as required in (28a). The grammatical licensing markings 〈Fa, Fb〉 that define the antecedent on the left in (27) have been acquired earlier on the righthand side in non-gapped structures marked for preposition, case, and theta role. As we have seen in section 3 for V-second, the acquisition of argument licensing in the Dutch [complement+V] structures precedes the acquisition of the V-second and the wh-movement distributions. The acquisition order is an empirical point and it fits (28a). The procedure to project grammatical features, required in (28b), was acquired when heads were subcategorized for the grammatical categories of their complements. This fits point (28b). The subcategorization of heads works for complements, but not for subjects and adjuncts. The latter, subjects and adjuncts, are in attached non-subcategorizing, non-complement positions. This seems to be the heart of the matter. Subcategorization is learned by assigning to lexical heads frames that specify the order and the categorial properties of their complements. Non-subcategorizing positions are not represented in such frames for lexical heads. If licensing features are necessary, as in (27) above, it follows that gaps in non-complements will not be related to an antecedent structure. A positive definition of the antecedent-gap relation was already given in Kayne (1981) as the government projection path. Note that arrangements as in Kayne (1981), or more recently Neeleman & van de Koot (2002), evade the notion “island’. If Kayne’s projection path, or some equivalent of it, is learnable, the learnability of islands need no longer be considered.
Arnold Evers & Jacqueline van Kampen
The learnability of island constraints has, however, been a topic of considerable debate (Chomsky 1975, 1980: 319; Crain & Nakayama 1987, Linguistic Review 2002). In (20), the copula from the main clause is fronted. Copula movement out of a subject relative like (20) is not possible (subject island and “complex NP’’). (29) Isi any ape that is brainy ti talkative? IP/C° sister adjacency realized by percolation
CP is 〈+C〉
IP/C° sister adjacency not realized by percolation
←→
IP
subject
IP predicate
any ape
〈+C-gap〉 also talkative
CP that is brainy
The point of discussion has been how children learn that the movement is structuredependent where the raw input consists of strings only. Crain & Nakayama (1987), as well as Legate & Yang (Linguistic Review 2002), and Fodor & Crowther (Linguistic Review 2002) relate this acquisitional fact to innate principles, whereas Pullum & Scholz (Linguistic Review 2002) prefer to see a hint in input percentages as sufficient. In the present view, island effects need neither follow from innate principles, nor from input frequency. It is rather a consequence of a feature projection system that has been acquired earlier. The relative clause [that is brainy] does not subcategorize for the noun ape. It is not a complement and must be attached to the DP phrase [any ape]. By consequence, it will be an island by virtue of not being head-governed. In the same vein, the subject phrase does not subcategorize for the copula. It must be attached, in English at least, to the copula phrase. By consequence, it will not be head-governed and it will automatically be an island. A gap within a subject phrase cannot and will not be noted in the projection line that the subject phrase is attached to. If it has already been established in earlier acquisition steps that the projection-line subcategorizes for the grammatical properties of its complements and that the subject is defined by an attachment to the marked predicate, the rest follows. Suppose some overzealous and slightly unethical psycho-linguist were to prescribe a daily regimen of nursery rhymes for a whole Kindergarten class, and suppose the rhymes were maliciously construed with gaps in
Parameter setting and input reduction
the subject phrase. Then we may now predict in advance that these gaps would not be acquired. They can only be acquired when the pupils first unlearn the idea that subjects are in a non-subcategorizing attached position. Unlearning that property comes close to abandoning the (English) notion of subject. In short, it is a misconstrual of the acquisition problem to worry about the learnability of islands or the learnability of long wh-movement. These phenomena are not learned at all. They rather follow from the combined effect of more elementary properties. Just like other parametric consequences, they are a surprise effect. Such effects are a characteristic of combinatory systems, where early elementary decisions have complex and unforeseen consequences.
6. Conclusion Parameters, not unlike Jakobson’s (1942) feature-oppositions, should clarify how the oppositions and implications in comparative grammar can be analyzed as resulting from choices in language acquisition. As pointed out by Dresher (1999), there are, however, clear difficulties faced by parametric approaches to language acquisition. UG parameters, as linguists generally conceive of them, are far removed from the primary data in language acquisition. An acquisition procedure based on parameter setting must be able to analyze the primary data in quite an abstract way (subject, object, predicate, auxiliary, complementizer, X-bar principles, islands, etc.). This seems impossible without already knowing the very grammar that is still to be acquired. That is Dresher’s epistemological problem. At the same time, the acquisition procedure must be attentive to the fact that the raw data follow from an interaction of several parameters, Dresher’s credit problem. There have been several attempts to bridge the gap. For the epistemological problem, Dresher (1999) proposed that each parameter was extended with a cue – a property more directly recognizable in the primary data (cf. also Lightfoot 1999). The cue would obligatorily trigger parameter setting and reduce the search space. As far as the credit problem is concerned, ordering of parameter setting is a key property. Ordered parameter setting is also postulated in Gibson & Wexler (1994) and it likewise appears spontaneously in the Variational Model of Yang (2002). What is significant about the idea of sequential ordering of parameter setting in the theoretical models of Gibson & Wexler (1994) and Yang (2002) is that language acquisition does indeed show a linear order of steps. The present paper directs the attention to the actual order in parameter setting. It proposes that the order of acquisition steps in child language can be predicted from reductions of the adult input language. The adult input continues to be recognized as a confused and poor stimulus, but reduction according to the acquisitional
Arnold Evers & Jacqueline van Kampen
principle (6) yields a series of intakes that are focused on single values 〈Fi〉. As such, principle (6) is highly effective. A prime example has been the paradoxical acquisition of the underlying word order and the V-second rule in Dutch. In general, when reduced structures can be extended to a real input structure by the addition of one functional category 〈Fi〉, they start to function as an evidence frame for that category 〈Fi〉. Evidence frames remind one of the cues in Dresher’s view on parameter setting. Yet, the evidence frame is derived from the reduced input. It is not an a priori given. The evidence frame is close to Tomasello’s (2003) “linguistic construction’’, but also different from it. The evidence frame is aimed at building abstract combinatorial categories in the lexicon. The reduction method (6) is universal and not language-specific, but neither the evidence frame itself, nor the functional category it identifies are given a priori. The reduction procedure directs the focus of the acquisition procedure to a single functional category. It is a naturally occurring “structure drill’’. The fact that many languages make use of the same parameters need not be derived from a languagespecific genetic endowment. Rather, parameters represent parallel solutions that appear in other cultural constructs as well. Elementary constructions, typological distinctions and steps in language acquisition remain closely related in the following three points. 1. The present approach to parameter setting obviates both the epistemological problem and the credit problem highlighted by Dresher (1999). The reduced structures focus on a specific acquisitional step 〈Fi〉, a “parameter setting’’, solving the epistemological problem. Setting the parameter is a causal effect and irreversible. The interaction of parameter settings will lead to superset effects, but not to credit problems. The unintended superset effects are rather the merit of the whole approach. The superset effects must be either correct or easily correctable by an additional micro-parameter. Otherwise the system will not survive. 2. The longitudinal analysis of child language may show how macro- and microparameters appear as a hierarchy of acquisition steps due to successively weaker input reductions. The Greenberg-type implications for universals follow from that hierarchy. The earlier steps (macro-parameters) set the stage for the later ones (micro-parameters). 3. There may be a type of combinatorial (generative) system that allows natural decoding and that enables a learner to acquire the system from scratch. The question is what such systems actually entail and whether the core grammars of human language are such systems. In the best of all possible worlds, minimal properties of grammar like “locality” and “inclusiveness” are stable, because they guarantee the survival of the system by keeping it learnable due to parameter setting.
Parameter setting and input reduction
The usual contention that parameters are somehow a priori options that require specific postulations about the brain has here been given up. A grammar is rather seen as a highly learnable cultural invention, not unlike number systems. This entails a change of focus. The number of (micro)-parameters and their ranking in the acquisition hierarchy becomes the major point of interest.
References Avram, L. & Coene, M. 2003. Why is it difficult to reach Agreement?. In Romance Linguistics. Theory and Acquisition, A.T. Pérez-Leroux & Y. Roberge (Eds), 247–262. Amsterdam: John Benjamins. Avram, L. & Coene, M. 2004. What early clitics can tell us about early subjects. In Proceedings of GALA 2003, J. van Kampen & S. Baauw (Eds), 93–102. Utrecht: LOT Series. Baker, M.C. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago IL: University of Chicago Press. Baker, M.C. 1996. The Polysynthesis Parameter. Oxford: OUP. Baker, M.C. 2003. Lexical Categories. Verbs, Nouns and Adjectives. Cambridge: CUP. Benveniste, E. 1966. Problèmes de Linguistique Générale. Paris: Gallimard. Berwick, R. 1985. The Acquisition of Syntactic Knowledge. Cambridge MA: The MIT Press. Berwick, R. & Weinberg, A. 1984. The Grammatical Basis of Linguistic Performance. Cambridge MA: The MIT Press. Briscoe, E. 2000. Grammatical acquisition: Inductive bias and co-evolution of language and the Language Acquisition Device. Language 76(2): 245–296. Brown, R. 1973. A First Language: The Early Stages. Cambridge MA: Harvard University Press. Chomsky, N. 1975. Reflections on Language. New York NY: Pantheon. Chomsky, N. 1980. Rules and Representations. Oxford: Blackwell. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1987. Language and Problems of Knowledge. The Managua Lectures. Cambridge MA: The MIT Press. Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36(1): 1–22. Clark, R. 1992. The selection of syntactic knowledge. Language Acquisition 2(2): 83–149. Crain, S. & Nakayama, M. 1987. Structure dependence in children’s language. Language 62: 522–543. Dresher, E. 1999. Charting the learning path: cues to parameter setting. Linguistic Inquiry 30: 27–67. Dresher, E. & Kaye, J. 1990. A computational learning model for metrical phonology. Cognition 34: 137–195. Evers, A. & Kampen, J. van. 2001. E-language, I-language and the order of parameter setting. UiL OTS Working Papers 00105-S-S. Utrecht: UiL. http://www-uilots.let.uu.nl/research/ Publications/Publications.htm Ferdinand, A. 1996. The Development of Functional Categories. The Acquisition of the Subject in French. Ph.D. dissertation, University of Leiden. Fodor, J.D. 1998. Unambiguous triggers. Linguistic Inquiry 29: 1–36. Fodor, J.D. 2001. Setting syntactic parameters. In Handbook of Contemporary Syntactic Theory, M. Baltin & C. Collins (Eds), 730–767. Oxford: Blackwell.
Arnold Evers & Jacqueline van Kampen Fodor, J.D. & Crowther, C. 2002. Understanding stimulus poverty arguments. Linguistic Review 19(1/2): 105–145. Gibson, E. & Wexler, K. 1994. Triggers. Linguistic Inquiry 25(3): 407–454. Greenberg, J. 1963. Some universals of grammar with particular reference to the order of meaningful elements. In Universals of Language, J. Greenberg (Ed.), 73–113, Cambridge MA: The MIT Press. Hale, M. & Reiss, C. 2003. The subset principle in phonological acquisition: Why the tabula can’t be rasa. Journal of Linguistics 39: 219–244. Jakobson, R. 1942. Kindersprache, Aphasie und Allgemeine Lautgesätze. Upsala: Uppsala Universitets Årsskrift. Kampen, J. van. 1997. First Steps in Wh-movement. Delft: Eburon. Kampen, J. van. 2004. Learnability order in the French pronominal system. In Selected Papers from Going Romance 2002, R. Bok-Bennema, B. Hollebrandse, B. Kampers-Manhe & P. Sleeman (Eds), 163–183. Amsterdam: John Benjamins. Kampen, J. van. 2005. Language-specific bootstraps for UG categories. International Journal of Bilingualism 9(2): 253–277. Kampen, J. van. 2006a. The acquisition of the standard EPP in Dutch and French. In Studies on Agreement, J. Costa & M. Figueiredo Silva (Eds), 99–121. Amsterdam: John Benjamins. Kampen, J. van. 2006b. Early operators and late topic-drop/pro-drop. In The Acquisition of Syntax in Romance Languages, V. Torrens & L. Escobar (Eds), 203–223. Amsterdam: John Benjamins. Kayne, R. 1981. ECP extensions. Linguistic Inquiry 12: 93–133. Kohl, K. 1999. An Analysis of Finite Parameter Learning in Linguistic Space. MA thesis, MIT. Lebeaux, D. 1988. Language Acquisition and the Form of Grammar. Ph.D. dissertation, University of Massachusetts. Legate, J. & Yang, C. 2002. Empirical re-assessment of poverty arguments. Linguistic Review 19(1/2): 151–162. Lightfoot, D. 1991. How to Set Parameters. Cambridge MA: The MIT Press. Lightfoot, D. 1999. The Development of Language: Acquisition, Change and Evolution. Oxford: Blackwell. Locke, J. 1997. A theory of neurolinguistic development. Brain and Language 58: 265–326. Lyons, J. 1979. Deixis and anaphora. In The Development of Conversation and Discourse, T. Myers (Ed.), 88–103. Edinburgh: EUP. MacWhinney, B. 2006. The CHILDES Project: Tools for Analyzing Talk, 3rd edn. Mahwah NJ: Lawrence Erlbaum. Mills, D., Prat, C., Zangl, R., Stager, C., Neville, H. & Werker, J. 2004. Language experience and the organization of brain activity to phonetically similar words: ERP evidence from 14- and 20-month-olds. Journal of Cognitive Neuroscience 16: 1452–1464. Neeleman, A. & Koot, H. van de. 2002. The configurational matrix. Linguistic Inquiry 33(4), 529–574. Postal, P. 1966. On so-called pronouns in English. In Monograph Series on Language and Linguistics 19, F. Dinneen (Ed.), 177–206. Georgetown DC: Georgetown University Press. Pullum, G. & Scholz, B. 2002. Empirical assessment of stimulus poverty arguments. Linguistic Review 19(1/2): 9–50. Stassen, L. 1997. Intransitive Predication. Oxford: OUP.
Parameter setting and input reduction
Tomasello, M. 2003. Constructing a Language. A Usage-Based Theory of Language Acquisition. Cambridge MA: Harvard University Press. Truckenbrodt, H. 2006. On the semantic motivation of syntactic verb movement to C in German. Theoretical Linguistics 32(3): 257–306. Watanabe, A, 1994. The role of triggers in the extended split Infl hypothesis: Unlearnable parameter settings. Studia Linguistica 48: 156–178. Weerman, F. 1989. The V2 Conspiracy. Dordrecht: ICG. Wexler, K. 1999. Maturation and growth of grammar. In Handbook of Child Language Acquisition, W. Ritchie & T. Bhatia (Eds), 55–105. San Diego CA: Academic Press. Yang, C. 2002. Knowledge and Learning in Natural Language. Oxford: OUP. Zwart, C. J.-W. 1993. Dutch Syntax. A Minimalist Approach. Ph.D. dissertation, University of Groningen.
Index
A acquisition, 7–8, 14, 15, 27, 32–34, 36, 41, 42, 59–63, 80, 94, 100, 110, 117, 118, 145, 146, 162, 167, 169, 171, 230, 231, 307, 308, 436, 447, 459, 460, 471–79, 483–87, 489, 492–97, 499–513 – cue, 9, 11, 13, 19, 42, 145, 167, 169, 307, 308, 486, 487, 495, 511, 512 – parameter setting, 9, 36, 59, 62, 63, 86, 110, 129, 274, 307, 309, 357, 359, 370, 371, 376, 442, 460, 483–87, 496, 501, 511, 512 – Single Value Constraint, 485, 492, 493, 496 adposition, 3, 9, 10, 11, 359, 362 – postposition, 2, 3 – preposition, 2, 3, 10, 11, 178, 179, 182, 199, 203, 212, 509 Afrikaans, 10, 11, 23, 41, 190, 341, 346, 465 Agree, 31, 50, 53, 148–52, 158, 159, 295–99, 302–08, 364, 414, 415–22, 426–29, 436 agreement (see also concord), 13, 15, 37, 42, 48, 49, 55, 56, 58, 61, 90, 100, 115, 143, 146, 157–59, 161–64, 170, 185, 186, 192, 193, 275, 288, 289, 295, 297, 302, 304, 305, 308, 311–14, 316, 317, 327, 331–33, 337, 338, 342, 351, 353, 359, 361, 363–71, 375–76, 378, 380, 389, 393, 397, 412, 420, 423, 424, 429–31, 434, 435,
443, 446, 454, 459–62, 465–67, 470, 471, 474–75, 491, 507, 508 Alexiadou, 20, 21, 48, 61, 277, 300, 419, 420, 424, 435, 436, 450 Anagnostopoulou, 20, 21, 61, 277, 300, 420, 424, 435, 436 articulatory-perceptual/ A-P interface (see interfaces) aspect, 37, 44–46, 234, 235, 242, 247, 259, 266, 271, 505 Avram, 12, 16, 37, 45, 60, 61, 461, 464, 470, 472–78, 502, 506, 508 B Baker, 3, 17, 21, 25–27, 36–38, 46, 55, 56, 59, 63, 77, 79–81, 83–87, 113, 114, 116, 134, 145, 210, 249, 265, 273, 305, 352, 353, 356, 358, 362, 363, 364, 365, 367–70, 351–373, 412, 419, 442, 443, 484, 505, 506 Basque, 18, 53, 54, 278, 299, 311–18, 320–27, 371 den Besten, 279, 281, 286 Biberauer, 15, 22, 23, 26, 29, 31, 52, 109, 119 biolinguistics, 23, 110, 111, 118 Bobaljik, 16, 17, 22, 28, 31, 32, 37, 61, 144, 443, 445, 459, 461 Borer, 20, 25, 43, 48, 49, 55, 61, 118, 145, 219, 224, 230, 351, 353–56, 358–60, 363, 364, 369, 372, 414, 430, 434, 442, 445 Borer-Chomsky Conjecture (see Lexical
Parameterisation Hypothesis) Brazilian Portuguese (see Portuguese) C Case, 24, 37, 52–54, 56, 58, 91, 198, 213, 277, 295–99, 301–308, 314, 334, 337, 364–369, 393, 405, 411–30, 436–37, 474–75, 506 case, 97, 98, 331–35, 339–46, 448, 475, 506 Celtic (see also Irish, Welsh), 16 Chinese, 15, 17, 51, 56, 57, 219, 222–26, 242, 261, 305, 306, 313, 315, 331–33, 339–41, 346, 375–406, 431–37 Chierchia, 45, 222–25, 255, 258, 264, 416 Chomsky, 2, 6–8, 13, 20, 22–25, 27–30, 32–34, 38, 40, 44, 47, 58, 60, 62, 76, 78–80, 83, 86, 90, 110–12, 117, 119, 123, 124, 129, 145–47, 151, 175, 192, 196, 201, 202, 204, 213, 220, 225, 274, 278, 295–98, 301, 306, 307, 312, 317, 318, 351, 353, 377, 389, 393, 397, 413, 419, 430, 442, 446, 483, 484, 487, 494, 499, 510 Cinque, 31, 77, 82, 85, 144, 182, 186, 197, 212, 445, 460 clitic, 13, 43, 175–88, 193, 197, 198, 200–05, 209–11, 213, 214, 233, 280, 281, 359, 404, 416, 471, 476–79, 507
Index Coene, 12, 16, 37, 45, 60, 61, 460, 464, 470, 472–77, 502, 506, 508 Comparative Method, 125, 126, 128, 130 competence, 63, 99, 101 conceptual-intentional interface (see interfaces) concord, 42, 364, 448 –negative concord, 153–164 copy, 28, 29, 31, 192, 297, 298, 308, 389, 390 creoles, 51, 52, 271–92, 341 cue (see acquisition) Czech, 154, 156, 157, 160, 164, 165 D Danish, 444, 465 deletion (see also ellipsis), 48, 49, 51, 54, 57, 151, 152, 307, 400, 402, 403, 447 demonstrative, 44, 178, 179, 186, 189, 190, 226, 362, 415, 416, 448, 449, 490, 491 determiner, 83, 115, 150, 191, 223, 224, 226, 227, 312, 335, 362, 367, 415, 441, 443, 448, 450–54, 468, 473, 476 doubling, 42, 57, 145, 146–53, 153–57, 166, 167, 169, 170, 182, 183, 197, 284, 359, 448, 449, 453, 454 Dresher, 9, 12, 59–61, 484–87, 511, 512 Duguine, 44, 47, 49, 53, 54 326, 383 Dutch, 10, 11, 21, 41, 46, 63, 154, 155, 166–68, 178, 182, 214, 247, 250–52, 256, 259, 260, 279, 331, 334, 335, 341, 346, 362, 444, 464, 465, 472, 474, 483, 489, 490, 495–96, 499, 501, 502, 505–509, 512
E Economy, 90, 91, 98, 117, 118, 123, 124, 159, 161, 297, 307, 326, 463
ellipsis (see also deletion), 48, 51, 53, 54, 163, 64, 311, 312, 315, 317, 318, 320–24, 326, 327, 375, 383, 390, 420, 451 Elsewhere Condition, 11, 54, 55, 265, 331, 334, 336–37, 346 English, 5, 9, 13, 14, 18, 20, 23, 41, 43, 46, 53, 55, 57, 58, 77, 84, 85, 87, 99, 115, 119–21, 133, 144, 149, 151, 152, 162, 166, 176, 178–80, 184–90, 192, 193, 196–203, 205–08, 210, 211, 221, 223, 224, 226–28, 230–32, 234, 235, 239, 241, 247, 250, 252, 253, 256, 258–61, 264, 266, 272, 276, 279–82, 285, 290, 291, 301, 306, 321–24, 327, 331, 333, 337, 341, 342, 344, 346, 354, 355, 357–59, 362, 365, 367, 377, 378, 390, 402, 405, 406, 411–15, 419, 421, 422, 424, 426–28, 436, 437, 443, 461, 468, 472, 476, 490, 502, 510, 511 explanation, 15, 24, 26, 33, 39, 44, 75, 78, 79, 84–88, 91–93, 96, 98–101, 117, 118, 123, 132, 143, 150, 311, 316, 324, 327, 406, 427, 443 – functional explanation/ account, 4–6, 39, 75, 78, 79, 91, 96, 99–101, 159 – generative explanation/ account, 6, 43, 44, 75–79, 81–87, 90, 92, 94–101, 117, 273, 274, 311, 336, 353, 360, 378, 442, 512 expletive, 12, 14, 16, 20–23, 44, 48, 50, 52, 57, 84, 87, 175, 176, 182–86, 188–93, 195–205, 207, 210–14, 262, 271, 275–78, 281–83, 284–91, 312, 316, 365, 375, 376, 380, 391, 424, 436, 461
Extended Projection Principle (EPP) 7, 13, 20–23, European Portuguese (see Portuguese) Evers, 12, 34, 35, 42, 60, 61, 62, 497
F Faroese, 462, 463 features – Case features, 52, 53, 56, 58, 277, 295, 297, 298, 364, 393, 411, 415, 419–21, 424, 425, 427, 436 – feature checking, 157, 163, 424 – feature valuation, 61, 151, 459, 460, 469, 471, 479 – formal features, 42, 43, 116, 129, 143–53, 156, 166, 167, 169–71, 326, 460 – phi (φ)-features, 12, 13, 53, 147, 186, 296, 331, 336–38, 364, 365, 378, 389, 390, 405, 444, 467, 507, 508 Fernandez-Salgueiro, 37 Finnish, 11, 47, 49–51, 56, 57, 115, 132, 276–78, 291, 317, 327, 371, 375–79, 380–88, 389, 391–94, 397–99, 401–06, 450 Focus, 18, 44, 53, 79, 284, 285, 311, 315–18, 324–27, 336, 342, 365, 366, 449, 453–55 French, 43, 46, 51, 85, 110, 115, 129, 162, 165, 176–85, 188, 191, 192, 200–02, 204–13, 247, 249, 251, 252, 259, 261, 265, 279–81, 288–90, 299, 352, 354, 397, 428, 464–66, 472, 476, 490, 491, 502, 507, 508 Fukui, 17, 27, 29, 80, 82, 83, 85, 353, 416, 443 Full Interpretation, 147, 152 functional category, 13, 25–27, 36, 43, 45, 60, 62, 143, 146, 369, 370, 413, 415,
Index 417, 427, 428, 441–43, 445–47, 450–53, 460, 483, 484, 489, 491–93, 500, 512 Free Inversion, 13, 15, 52, 271–73, 278, 279, 283–85, 291, 484, 495
G genericity, 46, 247, 248, 253, 255, 257–61, 266 German, 9–11, 21, 26, 115, 129, 148, 164, 165, 178, 206–208, 230, 231, 247, 250, 252, 256, 259, 260, 315, 337, 367, 444, 464, 465, 474, 506 Germanic, 16, 21, 22, 48, 85, 133, 206, 207, 239–240, 247, 250, 251, 256, 260–62, 264, 265, 415, 444, 446, 464, 496 Gianollo, 12, 40–42, 115, 130, 132, 447 Greek, 13, 46, 58, 115, 163, 164, 247, 249, 251–53, 259, 261, 265, 331, 341, 343, 346, 367, 403, 441–44, 447–54, 473 Greenberg, 2, 3, 27, 77, 83, 89, 92–96, 113, 114, 125–28, 342, 343, 475, 494, 496, 512 Guardiano, 12, 40–42, 115, 116, 129–33
H Haegeman, 76, 78, 80, 82, 83, 152, 158, 161, 280, 464 Haspelmath, 3, 4, 32, 39, 40, 44, 78, 90–93, 95, 97, 99–101, 342, 344, 351 Hawkins, 2–4, 6, 83, 92, 97, 99, 447 Head Parameter (HP) (see parameter) Holmberg, 26, 31, 35, 47–51, 57, 81, 84–85, 273, 274, 276, 277, 307, 317, 327, 334, 376–78, 380–82, 389, 391–93, 399, 402–05, 416, 450
Huang, 37, 48, 119, 305, 306, 313, 315, 331–33, 376–79, 420, 430–32 Hungarian, 115, 132, 188, 191, 198, 437
I Iatridou, 144 interfaces, 24, 28, 29, 31–33, 45, 121, 129, 143, 145, 147, 151, 219, 220, 228, 230, 266, 297, 306, 307, 334, 336 – articulatory-perceptual (A-P), 24 – conceptual-intentional (C-I), 24, 45, 46, 146, 219–28, 230 – Logical Form (LF), 22, 25, 28, 29, 32, 47, 54, 146, 147, 151, 152, 154, 159, 219–22, 225, 226, 229, 230, 239–42, 305, 306, 324, 326, 378, 389, 441, 442, 445, 447, 462 470, 471, 490, 507 – Phonological Form (PF), 24, 28–31, 47, 53, 54, 159, 305, 306, 316, 320, 326, 334, 336, 343, 375, 389, 400, 407, 425, 499, 507 Irish, 17, 115, 142 Italian, 12, 13, 20, 43, 55, 84, 87, 88, 115, 119, 121, 148, 154–57, 159, 164, 165, 176–82, 184–86, 192, 193, 197–201, 203–205, 207–13, 222–26, 259, 272, 276, 280, 299, 302, 313, 331, 333, 341, 346, 352, 378, 416, 430, 431, 434–36, 465, 473, 476
J Japanese, 9, 51, 77, 80, 85, 120, 313, 327, 331–33, 339–41, 346, 355, 358, 359, 402, 412, 425, 435, 437, 464 van Kampen, 12, 34, 35, 42, 60, 61, 489, 490, 495–98, 500–502, 504, 506–08
K Kayne, 6, 25, 27–30, 36, 37, 43, 44, 82, 85–87, 114, 178, 179, 181–83, 185, 187, 188, 190, 191, 194, 196, 198, 200, 202, 203, 207, 210, 212, 213, 259, 265, 284, 352, 353, 355–59, 362, 363, 367, 371, 443, 509 Keenan, 96, 125, 129, 153, 401 Kiparsky, 6, 11, 54, 78, 238, 334, 336 L Last Resort, 223, 295, 297, 299, 301, 303, 304 Linear Correspondence Axiom (LCA) – (see phrase structure) learnability, 86, 483, 494, 496, 508–11 Lekakou, 37, 45, 46, 247, 249, 251–53, 255, 256, 259, 404, 405 Lexical Parameterisation Hypothesis (Borer-Chomsky Conjecture), 25, 27, 28, 43, 55, 351, 353–72, 442 LF interface (see interfaces) Lightfoot, 9, 12, 78, 82, 83, 88, 128, 442, 444, 486, 495, 498, 511 Logical Form (LF) (see interfaces) Longobardi, 12, 40, 49, 109–12, 114–18, 120, 121, 125, 129–33, 194, 224, 413, 415, 416, 467, 468, 479 M Match, 296, 301, 303, 308 Merge, 29, 30, 58, 150, 193, 296, 297, 303, 355, 390, 404, 405, 424 middle, 45, 46, 247–66, 404, 405 Minimalist Program, 23, 90, 111, 117, 123, 145, 146, 295, 304, 353–55, 377 Modesto, 22, 37, 47, 56, 57, 59, 307, 377, 380, 384, 391, 406, 420
Index morphological richness, 16, 275, 279, 284 movement – A’-movement (A-bar movement), 26, 58, 308 –Wh-movement, 14, 17–19, 32, 82, 84, 306, 308, 375, 384, 387, 389, 391, 394, 395, 509, 511 – A-movement, 296–305, 424, 428 – EPP feature, 22, 25–28, 31, 50, 123, 150, 151, 277, 392, 402, 423, 435, 474 – head movement –V-to-T movement, 16–17, 28, 34, 52, 275, 283, 416, 426, 428, 429, 461, 462, 464, 466 –V-to-C movement (Verb Second/V2) 15, 16, 21, 28, 34, 63, 83, 154, 464, 483, 486, 495–504, 507–509, 512 –N-to-D movement, 16, 28, 34, 224, 416 – locality, 5, 63, 83, 163, 287, 382, 390, 512 – pied-piping, 18, 25, 150 – reconstruction, 422, 485 Move, 148, 150, 167, 397 N Neeleman, 5, 6, 37, 47, 48, 54, 55, 77, 144, 159, 261–64, 276, 307, 332, 334, 340, 341, 343, 346, 464, 508, 509 negation, 42, 95, 116, 143–46, 152–57, 160–63, 166–70, 184, 258, 271, 343, 416, 420, 423, 444, 462, 505 – negative polarity item, 158, 162, 258 Newmeyer, 4, 6, 19, 34–38, 44, 46, 47, 51, 55, 56, 63, 80, 81, 84, 86–88, 91, 92, 100, 170, 271–73, 291, 447 Nicolis, 46, 51, 52, 57, 275, 279, 283, 284, 307 Norwegian, 115, 133, 180, 219, 226, 227, 230, 231, 261, 444, 465, 467 null elements
– null argument, 53, 54, 59, 275, 276, 282, 284, 327, 333, 336, 375–376, 378, 403, 411, 413, 429, 436, 464 – null expletive, 48–50, 52, 84, 271, 272, 275–78, 281–91, 312, 365, 389, 413, 435 – null operator, 299 – pro, 8, 12–15, 20–23, 25, 37, 46–48, 51–55, 83–85, 148, 149, 235, 271, 273, 275, 276, 278–82, 284, 289–91, 299, 306, 309, 311–21, 323, 324, 326, 327, 331–34, 336–46, 351–53, 368, 378, 389, 391, 424, 430–32, 434–36, 463–66, 471, 476, 478, 484, 507 – PRO, 202, 316, 317, 380, 381, 430 null subject languages (NSL) 12–16, 20, 41, 47–54, 115, 272–76, 279, 287–89, 299, 306, 307, 311, 312, 315, 327, 331, 343, 341, 343, 345, 376, 381, 383, 389, 401, 403, 404, 407, 425, 430, 463–65 Null Subject Parameter (NSP) 7, 8, 12–16, 20, 25, 35, 37, 38, 46–48, 51, 55, 57, 83–85, 271–76, 291, 309, 311–13, 317, 320, 327, 351–53, 375, 379, 401, 403, 406, 476, 478 O Öztürk, 37, 47, 57, 58, 399, 415– 17, 419, 420, 422–25, 427, 428, 430, 437 Optimality Theory (OT), 75, 79, 87–91, 97, 101 Optionality, 11, 18, 22, 26, 337 P Panagiotidis, 57–59, 130, 372, 448–51, 484 parameter 6–9, 12–20, 24, 25, 27–41, 44, 46–47, 51,
52, 55–63, 75, 79–89, 94–96, 101, 109–124, 128–31, 134, 143, 145, 170, 171, 219, 228, 234, 235, 241, 248, 271, 273–76, 295, 302, 304–309, 311–13, 320, 327, 351–66, 369–72, 375, 376, 401–406, 413, 426–29, 441–44, 460, 478, 483–87, 492, 494–96, 501, 511–13 – Head (Directionality) Parameter (HP) 7–13, 15–17, 19, 24–27, 29–32, 36, 38, 56, 82–84, 358–63, 418, 495–501 –head-final, 4, 9, 10, 25, 26, 29, 31, 32, 355, 356, 359–62, 415, 416, 418, 464 –head-initial, 4, 9, 10, 25, 26, 31, 32, 355, 356, 359–62, 496 – macroparameter, 37, 55–57, 75, 84, 86, 94–95, 101, 274, 351, 354, 355, 357, 359–61, 363, 364, 371, 441, 444 – microparameter, 36, 56, 86, 101, 351–53, 356, 358–61, 369, 371, 441 – Split INFL parameter, 37, 61 – Wh-parameter, 18–19, 145 parsing, 4–6, 159, 360 passive, 46, 57, 84, 251, 256, 265, 266, 365, 402–404, 414, 424 Phonological Form (PF) (see interfaces) phrase structure – Bare Phrase Structure, 326 – binarity, 34 – (non-) configurational(ity), 57, 58, 83, 84, 356, 411–37 – Linear Correspondence Axiom (LCA), 28–30, 87 performance, 6, 99–101, 124, 486
Index PF interface (see interfaces) phase, 22, 30, 31, 45, 297, 393, 397, 419, 422 Pollock, 22, 143, 164, 181, 191, 211 Portuguese, 115, 142, 206, 208, 279, 284, 299, 302 – Brazilian Portuguese, 56, 284, 299, 302, 343, 375, 376, 466 – European Portuguese, 51, 57, 277, 299, 302, 383, 403, 466 Principles and Parameters (see principle and parameter) principle – parsing or other functionally motivated principle, 4, 6, 33, 34, 99 – UG principle 6, 7, 13, 19, 20, 23, 26, 29, 33, 35, 63, 79, 87, 90, 111, 117, 121–23, 132, 147, 152, 166, 225, 263, 265, 274, 278, 287, 295, 316, 351, 354, 357, 463, 492, 508, 510–12 R reflexive, 46, 61, 97, 99, 181–182, 249, 336, 483 Ramchand, 29, 44, 45, 236–38, 416, 473, 502, 506 Rizzi, 8, 13, 20, 37, 46–48, 51, 52, 58, 61, 84, 120, 144, 149, 200, 271–75, 278, 280, 285–89, 291, 292, 306, 311, 315, 331, 351, 355, 375, 376, 395, 399, 403, 430, 434, 435, 463 Roberts, 1, 13, 15, 16, 23, 26–29, 31, 34, 35, 41, 42, 46, 47, 49, 52, 60, 63, 81, 82, 84, 85, 128, 144, 153, 181, 249–51, 256, 265, 273–75, 281, 307, 442, 446, 454
Romance, 13, 16, 41, 43, 48, 49, 52, 53, 55, 57, 85, 167, 175, 176, 183, 186, 188, 191, 204, 206, 207, 209, 213, 299, 300, 302, 306–08, 352, 359, 368, 381, 403, 404, 415, 464, 507 Romanian, 16, 61, 115, 121, 176, 299, 302, 306, 449, 460, 465, 469, 471–74, 476, 502, 508 Roussou, 27, 144, 153, 442, 446, 450, 454 Russian, 15, 20, 115, 119, 121, 133, 219, 222–28, 231, 234–36, 238–42, 464, 467
312, 314–16, 325, 326, 375–80, 386, 387, 389–406, 420, 422, 432, 434, 436, 489, 490, 494, 495, 498, 503, 504 Turkish, 57–58, 164, 165, 341, 346, 371, 411–13, 415–37 typology, 2, 43, 88, 89, 92, 94, 95, 125, 134, 258, 259, 302, 331, 346, 351, 401, 415 U unaccusative, 97, 210, 211, 248, 250–52, 259, 262, 263, 277, 283, 365, 376, 414, 423 universal(s) 1–3, 6, 39, 75–101, 120, 144, 145, 220, 222, 225, 234, 254, 278, 305, 307, 316, 475 Universal Grammar (UG) 4, 6, 7, 9, 19, 24, 28, 31–35, 40–42, 60, 61, 63, 75, 78–80, 82, 86–88, 90–93, 95, 99, 101, 109, 113–15, 117,119, 122–24, 128, 143–45, 153, 169, 171, 181, 226, 230, 242, 445, 446, 454, 460, 465, 511
S scope, 155, 157, 161, 167, 192, 240, 241, 254, 299, 300, 319, 405, 406, 420, 423, 435, 436 scrambling, 412, 428, 429 semantic operators, 145, 152, 153, 169 Spanish, 12, 51, 115, 150, 162, 164, 167, 168, 177, 183, 190, 206, 208, 277–79, 281, 299, 300, 302, 313, 327, 331, 352, 403, 430, 431, 434–36, 449, 465, 476 Spell-out, 29, 31, 54, 151, 297, 331, 333–41, 343, 346, 420 Svenonius, 29, 44, 45, 235, 241, 416, 473, 502, 506 Swedish, 189, 331, 341, 346, 444, 465, 472, 473 Szendröi 37, 47, 48, 54, 77, 276, 307, 332, 340, 341, 343, 346
Y Yiddish, 16, 365, 370, 462–64
T that-trace effects, 13–15, 52, 84, 307, 485 topic, 22, 37, 48, 53, 56, 57, 79, 88, 299, 300, 311,
Z Zeijlstra, 12, 34, 35, 37, 42, 43, 59, 61, 119, 151, 153, 157, 160, 165, 166, 170 Zulu, 17
V Verb Second (V2) (see head movement: V-to-C movement) W Welsh, 115, 142, 368
Linguistik Aktuell/Linguistics Today A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 136 Kandybowicz, Jason: The Grammar of Repetition. Nupe grammar at the syntax–phonology interface. xiii, 159 pp. + index. Expected November 2008 135 Lewis, William D., Simin Karimi, Heidi Harley and Scott O. Farrar (eds.): Time and Again. Theoretical Perspectives on Formal Linguistics. In honor of D. Terrence Langendoen. xiv, 261 pp. + index. Expected November 2008 134 Armon-Lotem, Sharon, Gabi Danon and Susan Rothstein (eds.): Current Issues in Generative Hebrew Linguistics. vi, 388 pp. + index. Expected October 2008 133 MacDonald, Jonathan E.: The Syntactic Nature of Inner Aspect. A minimalist perspective. xv, 234 pp. + index. Expected November 2008 132 Biberauer, Theresa (ed.): The Limits of Syntactic Variation. 2008. vii, 521 pp. 131 De Cat, Cécile and Katherine Demuth (eds.): The Bantu–Romance Connection. A comparative investigation of verbal agreement, DPs, and information structure. xix, 350 pp. + index. Expected September 2008 130 Kallulli, Dalina and Liliane Tasmowski (eds.): Clitic Doubling in the Balkan Languages. ix, 431 pp. + index. Expected October 2008 129 Sturgeon, Anne: The Left Periphery. The interaction of syntax, pragmatics and prosody in Czech. xi, 139 pp. + index. Expected October 2008 128 Taleghani, Azita H.: Modality, Aspect and Negation in Persian. 2008. ix, 183 pp. 127 Durrleman-Tame, Stephanie: The Syntax of Jamaican Creole. A cartographic perspective. 2008. xii, 190 pp. 126 Schäfer, Florian: The Syntax of (Anti-)Causatives. External arguments in change-of-state contexts. 2008. xi, 324 pp. 125 Rothstein, Björn: The Perfect Time Span. On the present perfect in German, Swedish and English. 2008. xi, 171 pp. 124 Ihsane, Tabea: The Layered DP. Form and meaning of French indefinites. 2008. ix, 260 pp. 123 Stoyanova, Marina: Unique Focus. Languages without multiple wh-questions. 2008. xi, 184 pp. 122 Oosterhof, Albert: The Semantics of Generics in Dutch and Related Languages. 2008. xviii, 286 pp. 121 Tungseth, Mai Ellin: Verbal Prepositions and Argument Structure. Path, place and possession in Norwegian. 2008. ix, 187 pp. 120 Asbury, Anna, Jakub Dotlačil, Berit Gehrke and Rick Nouwen (eds.): Syntax and Semantics of Spatial P. 2008. vi, 416 pp. 119 Fortuny, Jordi: The Emergence of Order in Syntax. 2008. viii, 211 pp. 118 Jäger, Agnes: History of German Negation. 2008. ix, 350 pp. 117 Haugen, Jason D.: Morphology at the Interfaces. Reduplication and Noun Incorporation in Uto-Aztecan. 2008. xv, 257 pp. 116 Endo, Yoshio: Locality and Information Structure. A cartographic approach to Japanese. 2007. x, 235 pp. 115 Putnam, Michael T.: Scrambling and the Survive Principle. 2007. x, 216 pp. 114 Lee-Schoenfeld, Vera: Beyond Coherence. The syntax of opacity in German. 2007. viii, 206 pp. 113 Eythórsson, Thórhallur (ed.): Grammatical Change and Linguistic Theory. The Rosendal papers. 2008. vi, 441 pp. 112 Axel, Katrin: Studies on Old High German Syntax. Left sentence periphery, verb placement and verbsecond. 2007. xii, 364 pp. 111 Eguren, Luis and Olga Fernández Soriano (eds.): Coreference, Modality, and Focus. Studies on the syntax–semantics interface. 2007. xii, 239 pp. 110 Rothstein, Susan (ed.): Theoretical and Crosslinguistic Approaches to the Semantics of Aspect. 2008. viii, 453 pp. 109 Chocano, Gema: Narrow Syntax and Phonological Form. Scrambling in the Germanic languages. 2007. x, 333 pp. 108 Reuland, Eric, Tanmoy Bhattacharya and Giorgos Spathas (eds.): Argument Structure. 2007. xviii, 243 pp.
107 Corver, Norbert and Jairo Nunes (eds.): The Copy Theory of Movement. 2007. vi, 388 pp. 106 Dehé, Nicole and Yordanka Kavalova (eds.): Parentheticals. 2007. xii, 314 pp. 105 Haumann, Dagmar: Adverb Licensing and Clause Structure in English. 2007. ix, 438 pp. 104 Jeong, Youngmi: Applicatives. Structure and interpretation from a minimalist perspective. 2007. vii, 144 pp. 103 Wurff, Wim van der (ed.): Imperative Clauses in Generative Grammar. Studies in honour of Frits Beukema. 2007. viii, 352 pp. 102 Bayer, Josef, Tanmoy Bhattacharya and M.T. Hany Babu (eds.): Linguistic Theory and South Asian Languages. Essays in honour of K. A. Jayaseelan. 2007. x, 282 pp. 101 Karimi, Simin, Vida Samiian and Wendy K. Wilkins (eds.): Phrasal and Clausal Architecture. Syntactic derivation and interpretation. In honor of Joseph E. Emonds. 2007. vi, 424 pp. 100 Schwabe, Kerstin and Susanne Winkler (eds.): On Information Structure, Meaning and Form. Generalizations across languages. 2007. vii, 570 pp. 99 Martínez-Gil, Fernando and Sonia Colina (eds.): Optimality-Theoretic Studies in Spanish Phonology. 2007. viii, 564 pp. 98 Pires, Acrisio: The Minimalist Syntax of Defective Domains. Gerunds and infinitives. 2006. xiv, 188 pp. 97 Hartmann, Jutta M. and László Molnárfi (eds.): Comparative Studies in Germanic Syntax. From Afrikaans to Zurich German. 2006. vi, 332 pp. 96 Lyngfelt, Benjamin and Torgrim Solstad (eds.): Demoting the Agent. Passive, middle and other voice phenomena. 2006. x, 333 pp. 95 Vogeleer, Svetlana and Liliane Tasmowski (eds.): Non-definiteness and Plurality. 2006. vi, 358 pp. 94 Arche, María J.: Individuals in Time. Tense, aspect and the individual/stage distinction. 2006. xiv, 281 pp. 93 Progovac, Ljiljana, Kate Paesani, Eugenia Casielles and Ellen Barton (eds.): The Syntax of Nonsententials. Multidisciplinary perspectives. 2006. x, 372 pp. 92 Boeckx, Cedric (ed.): Agreement Systems. 2006. ix, 346 pp. 91 Boeckx, Cedric (ed.): Minimalist Essays. 2006. xvi, 399 pp. 90 Dalmi, Gréte: The Role of Agreement in Non-Finite Predication. 2005. xvi, 222 pp. 89 Velde, John R. te: Deriving Coordinate Symmetries. A phase-based approach integrating Select, Merge, Copy and Match. 2006. x, 385 pp. 88 Mohr, Sabine: Clausal Architecture and Subject Positions. Impersonal constructions in the Germanic languages. 2005. viii, 207 pp. 87 Julien, Marit: Nominal Phrases from a Scandinavian Perspective. 2005. xvi, 348 pp. 86 Costa, João and Maria Cristina Figueiredo Silva (eds.): Studies on Agreement. 2006. vi, 285 pp. 85 Mikkelsen, Line: Copular Clauses. Specification, predication and equation. 2005. viii, 210 pp. 84 Pafel, Jürgen: Quantifier Scope in German. 2006. xvi, 312 pp. 83 Schweikert, Walter: The Order of Prepositional Phrases in the Structure of the Clause. 2005. xii, 338 pp. 82 Quinn, Heidi: The Distribution of Pronoun Case Forms in English. 2005. xii, 409 pp. 81 FuSS, Eric: The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal inflection. 2005. xii, 336 pp. 80 Burkhardt, Petra: The Syntax–Discourse Interface. Representing and interpreting dependency. 2005. xii, 259 pp. 79 Schmid, Tanja: Infinitival Syntax. Infinitivus Pro Participio as a repair strategy. 2005. xiv, 251 pp. 78 Dikken, Marcel den and Christina Tortora (eds.): The Function of Function Words and Functional Categories. 2005. vii, 292 pp. 77 Öztürk, Balkız: Case, Referentiality and Phrase Structure. 2005. x, 268 pp. 76 Stavrou, Melita and Arhonto Terzi (eds.): Advances in Greek Generative Syntax. In honor of Dimitra Theophanopoulou-Kontou. 2005. viii, 366 pp. 75 Di Sciullo, Anna Maria (ed.): UG and External Systems. Language, brain and computation. 2005. xviii, 398 pp. 74 Heggie, Lorie and Francisco Ordóñez (eds.): Clitic and Affix Combinations. Theoretical perspectives. 2005. viii, 390 pp. 73 Carnie, Andrew, Heidi Harley and Sheila Ann Dooley (eds.): Verb First. On the syntax of verbinitial languages. 2005. xiv, 434 pp.
72 FuSS, Eric and Carola Trips (eds.): Diachronic Clues to Synchronic Grammar. 2004. viii, 228 pp. 71 Gelderen, Elly van: Grammaticalization as Economy. 2004. xvi, 320 pp. 70 Austin, Jennifer R., Stefan Engelberg and Gisa Rauh (eds.): Adverbials. The interplay between meaning, context, and syntactic structure. 2004. x, 346 pp. 69 Kiss, Katalin É. and Henk van Riemsdijk (eds.): Verb Clusters. A study of Hungarian, German and Dutch. 2004. vi, 514 pp. 68 Breul, Carsten: Focus Structure in Generative Grammar. An integrated syntactic, semantic and intonational approach. 2004. x, 432 pp. 67 Mišeska Tomić, Olga (ed.): Balkan Syntax and Semantics. 2004. xvi, 499 pp. 66 Grohmann, Kleanthes K.: Prolific Domains. On the Anti-Locality of movement dependencies. 2003. xvi, 372 pp. 65 Manninen, Satu Helena: Small Phrase Layers. A study of Finnish Manner Adverbials. 2003. xii, 275 pp. 64 Boeckx, Cedric and Kleanthes K. Grohmann (eds.): Multiple Wh-Fronting. 2003. x, 292 pp. 63 Boeckx, Cedric: Islands and Chains. Resumption as stranding. 2003. xii, 224 pp. 62 Carnie, Andrew, Heidi Harley and MaryAnn Willie (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. 2003. xii, 378 pp. 61 Schwabe, Kerstin and Susanne Winkler (eds.): The Interfaces. Deriving and interpreting omitted structures. 2003. vi, 403 pp. 60 Trips, Carola: From OV to VO in Early Middle English. 2002. xiv, 359 pp. 59 Dehé, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. xii, 305 pp. 58 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. 2003. vi, 309 pp. 57 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. 2003. vi, 405 pp. 56 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. 2003. x, 295 pp. 55 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 1: The syntax and semantics of noun phrases. 2003. vi, 362 pp. 54 Baptista, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2003. xxii, 294 pp. (incl. CD-rom). 53 Zwart, C. Jan-Wouter and Werner Abraham (eds.): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26–27, 2000). 2002. xiv, 407 pp. 52 Simon, Horst J. and Heike Wiese (eds.): Pronouns – Grammar and Representation. 2002. xii, 294 pp. 51 Gerlach, Birgit: Clitics between Syntax and Lexicon. 2002. xii, 282 pp. 50 Steinbach, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. xii, 340 pp. 49 Alexiadou, Artemis (ed.): Theoretical Approaches to Universals. 2002. viii, 319 pp. 48 Alexiadou, Artemis, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Gärtner (eds.): Dimensions of Movement. From features to remnants. 2002. vi, 345 pp. 47 Barbiers, Sjef, Frits Beukema and Wim van der Wurff (eds.): Modality and its Interaction with the Verbal System. 2002. x, 290 pp. 46 Panagiotidis, E. Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. x, 214 pp. 45 Abraham, Werner and C. Jan-Wouter Zwart (eds.): Issues in Formal German(ic) Typology. 2002. xviii, 336 pp. 44 Taylan, Eser Erguvanlı (ed.): The Verb in Turkish. 2002. xviii, 267 pp. 43 Featherston, Sam: Empty Categories in Sentence Processing. 2001. xvi, 279 pp. 42 Alexiadou, Artemis: Functional Structure in Nominals. Nominalization and ergativity. 2001. x, 233 pp. 41 Zeller, Jochen: Particle Verbs and Local Domains. 2001. xii, 325 pp. 40 Hoeksema, Jack, Hotze Rullmann, Víctor Sánchez-Valencia and Ton van der Wouden (eds.): Perspectives on Negation and Polarity Items. 2001. xii, 368 pp. 39 Gelderen, Elly van: A History of English Reflexive Pronouns. Person, Self, and Interpretability. 2000. xiv, 279 pp.
38 Meinunger, André: Syntactic Aspects of Topic and Comment. 2000. xii, 247 pp. 37 Lutz, Uli, Gereon Müller and Arnim von Stechow (eds.): Wh-Scope Marking. 2000. vi, 483 pp. 36 Gerlach, Birgit and Janet Grijzenhout (eds.): Clitics in Phonology, Morphology and Syntax. 2001. xii, 441 pp. 35 Hróarsdóttir, Thorbjörg: Word Order Change in Icelandic. From OV to VO. 2001. xiv, 385 pp. 34 Reuland, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. xii, 255 pp. 33 Puskás, Genoveva: Word Order in Hungarian. The syntax of Ā-positions. 2000. xvi, 398 pp. 32 Alexiadou, Artemis, Paul Law, André Meinunger and Chris Wilder (eds.): The Syntax of Relative Clauses. 2000. vi, 397 pp. 31 Svenonius, Peter (ed.): The Derivation of VO and OV. 2000. vi, 372 pp. 30 Beukema, Frits and Marcel den Dikken (eds.): Clitic Phenomena in European Languages. 2000. x, 324 pp. 29 Miyamoto, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 2000. xiv, 232 pp. 28 Hermans, Ben and Marc van Oostendorp (eds.): The Derivational Residue in Phonological Optimality Theory. 2000. viii, 322 pp. 27 Růžička, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. x, 206 pp. 26 Ackema, Peter: Issues in Morphosyntax. 1999. viii, 310 pp. 25 Felser, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999. xiv, 278 pp. 24 Rebuschi, Georges and Laurice Tuller (eds.): The Grammar of Focus. 1999. vi, 366 pp. 23 Giannakidou, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. xvi, 282 pp. 22 Alexiadou, Artemis and Chris Wilder (eds.): Possessors, Predicates and Movement in the Determiner Phrase. 1998. vi, 388 pp. 21 Klein, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. x, 232 pp. 20 Laenzlinger, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. x, 371 pp. 19 Josefsson, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. ix, 199 pp. 18 Alexiadou, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. x, 256 pp. 17 Beermann, Dorothee, David LeBlanc and Henk van Riemsdijk (eds.): Rightward Movement. 1997. vi, 410 pp. 16 Liu, Feng-hsi: Scope and Specificity. 1997. viii, 187 pp. 15 Rohrbacher, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and prodrop. 1999. viii, 296 pp. 14 Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts (eds.): Materials on Left Dislocation. 1997. viii, 349 pp. 13 Alexiadou, Artemis and T. Alan Hall (eds.): Studies on Universal Grammar and Typological Variation. 1997. viii, 252 pp. 12 Abraham, Werner, Samuel David Epstein, Höskuldur Thráinsson and C. Jan-Wouter Zwart (eds.): Minimal Ideas. Syntactic studies in the minimalist framework. 1996. xii, 364 pp. 11 Lutz, Uli and Jürgen Pafel (eds.): On Extraction and Extraposition in German. 1996. xii, 315 pp. 10 Cinque, Guglielmo and Giuliana Giusti (eds.): Advances in Roumanian Linguistics. 1995. xi, 172 pp. 9 Gelderen, Elly van: The Rise of Functional Categories. 1993. x, 224 pp. 8 Fanselow, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. xvii, 232 pp. 7 Åfarlí, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. xii, 177 pp. 6 Bhatt, Christa, Elisabeth Löbel and Claudia Maria Schmidt (eds.): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. ix, 187 pp. 5 Grewendorf, Günther and Wolfgang Sternefeld (eds.): Scrambling and Barriers. 1990. vi, 442 pp. 4 Abraham, Werner and Sjaak De Meij (eds.): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. v, 349 pp. 3 Abraham, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. vi, 242 pp.
2 1
Ehlich, Konrad and Jürgen Rehbein: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. viii, 150 pp. With many photographic ills. Klappenbach, Ruth (1911–1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene MaligeKlappenbach. (Written in German). 1980. xxiii, 313 pp.