Correlatives Cross-Linguistically (Language Faculty and Beyond)

Correlatives Cross-Linguistically

Language Faculty and Beyond Internal and External Variation in Linguistics Language Faculty and Beyond (LFAB) focuses on research that contributes to a deeper understanding of the properties of languages as a result of the Language Faculty and its interface with other domains of the mind/brain. While the series will pay particular attention to the traditional tension between descriptive and explanatory adequacy, the series will also address issues such as the level of linguistic design, through new lines of inquiry often referred to as ‘physiological linguistics’ or ‘biolinguistics’. LFAB aims to publish studies from the point of view of internal and external factors which bear on the nature of micro- and macro-variation as, for example, understood in the minimalist approach to language.

Editors Kleanthes K. Grohmann University of Cyprus

Pierre Pica CNRS, Paris

Advisory Board Paola Benincà

Anders Holmberg

Cedric Boeckx

Lyle Jenkins

Guglielmo Cinque

Richard K. Larson

Noam Chomsky

Andrew Ira Nevins

University of Padova, Italy ICREA/Universitat Autònoma de Barcelona University of Venice, Italy Massachusetts Institute of Technology, Cambridge, USA

Stephen Crain

Macquarie University, Sydney, Australia

Marcel den Dikken

CUNY Graduate Center, New York, USA

Naama Friedmann

Tel Aviv University, Israel

University of Newcastle, UK Biolinguistics Institute, Cambridge, USA SUNY, Stony Brook, USA Harvard University, Cambridge, USA

Alain Rouveret

University of Paris VII, France

Esther Torrego

University of Massachusetts, Boston USA

Anna Papafragou

University of Delaware, Newark, USA

Akira Watanabe

University of Tokyo, Japan

Volume 1 Correlatives Cross-Linguistically Edited by Anikó Lipták

Correlatives Cross-Linguistically

Edited by

Anikó Lipták Leiden 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 Correlatives cross-linguistically / edited by Anikó Lipták.        p. cm. (Language Faculty and Beyond: Internal and External Variation in Linguistics, issn 1877-6531 ; v. 1) Includes bibliographical references and index. 1.  Grammar, Comparative and general--Relative clauses.  I. Lipták, Anikó Klára, 1973P297.C57    2009 415--dc22 2009009890 isbn 978 90 272 0818 7 (Hb; alk. paper) isbn 978 90 272 8942 1 (EB)

© 2009 – 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 The landscape of correlatives: An empirical and analytical survey Anikó Lipták

vii 1

Part 1.  Correlatives and related constructions What don’t wh-questions, free relatives, and correlatives have in common? Barbara Citko

49

Basque correlatives and their kin in the history of Northern Basque Georges Rebuschi

81

{Relative {conditional {correlative clauses}}} Boban Arsenijević

131

Relatively different: Italian Sign Language relative clauses in a typological perspective 157 Chiara Branchini and Caterina Donati Part 2.  The derivation of correlatives The syntax of the Tibetan correlative Seth Cable

195

Adjunction, features and locality in Sanskrit and Hindi/Urdu correlatives Alice Davison

223

Comparative correlatives and successive cyclicity Marcel den Dikken

263

Part 3.  The matching effect On the matching requirement in correlatives Tommi Tsz-Cheung Leung

309



Correlatives Cross-Linguistically

Matching effects in the temporal and locative domains Rajesh Bhatt and Anikó Lipták

343

Index

373

Preface and acknowledgements This book brings together recent work in generative syntax on the construction type called correlative relative constructions. While this kind of construction has been wellstudied in Hindi and closely related tongues, correlatives have not been systematically studied in other languages. Recent years have brought welcome developments in this domain, as some researchers have brought other, unrelated languages into the light of research on correlatives. This book presents these pioneering articles in a single volume. I thank Johan Rooryck for putting the idea of this book in my head back in 2004, and all those who helped me realize it, first and foremost the series editors Kleanthes Grohmann and Pierre Pica, and the authors of the articles, who kindly accepted my invitation to this volume. Special thanks are due to those who served as reviewers: Klaus Abels, Rajesh Bhatt, Valentina Bianchi, Veneeta Dayal, Marcel den Dikken, Alex Grosu, István Kenesei, Carol Neidle, Roland Pfau, Philomen Probert, Luis Vicente, Mark de Vos and Mark de Vries. I am happy to have had the help of Mark de Vos in checking the English of certain papers. Last but not least let me express my gratitude to those authors who had to put up with a lot of waiting to see their articles printed. I am indebted to the Netherlands Organization for Scientific Research (NWO) for financing a research project on correlatives between 2002 and 2006, titled Correlatives: typological and theoretical perspectives (nr. 355–70–013). This project allowed me to work on the topic myself and to come in contact with the researchers of this volume. I am also grateful to the Leiden University Centre of Linguistics (LUCL) for housing the project. Leiden, January 2009. Anikó Lipták

The landscape of correlatives An empirical and analytical survey Anikó Lipták

1. What is a correlative? The word correlative has multiple uses in linguistics. It is sometimes used to refer to pairs of words that show up linked to each other, across phrases or whole clauses. This is the sense in which grammars refer to “correlative (adverbs)” or “correlative subordinators” to describe pairs of words like if…then..., (al)though… yet/nevertheless..., as… so..., either... or... (Quirk et al. 1972, Chung 2004, Johannessen 2005). The term correlative is also used to refer to combinations of a clause and a pronominal linked to it. In a few cases, this means the combination of an argumental clause and its sentential pronominal (e.g. Berman et al 1998), as in I couldn’t believe it that John won the lottery. The pronominal here is called correlative to refer to the fact that it is related to the embedded clause, whose argument slot it occupies next to the verb. Even more frequently, however, the term correlative is used to refer to combinations of a relative clause and a possibly non-adjacent nominal expression linked to it. This is the way in which the typological literature refers to relative clause constructions that instantiate a non-local relativization strategy well-known in the ancient Indo-European languages like Sanskrit, Latin, Greek and Hittite (Haudry 1973) and in modern Indo-Aryan languages like Hindi (Srivastav 1991, Dayal 1996, Bhatt 2003). This book is about this type of relativization construction. In a correlative relativization strategy a left-peripheral relative clause is linked to a (possibly phonetically unrealized) nominal correlate in the clause that follows the relative clause. An illustrative example is given from Hindi – perhaps the most well-known and most cited example of a correlative, from Srivastav (1991: example 3a):1 (1)

[jo laRkii khaRii hai ] vo lambii hai rel girl standing is that tall is lit. Which girl is standing, that is tall. ‘The girl who is standing is tall.’



Anikó Lipták

The defining property of correlative constructions is the left peripheral position of the relative clause. As we can see in example (1), the left peripheral relative clause (also called the protasis) is linked to the main clause (the apodosis) by a correlate, a nominal expression. The latter, vo ‘that’ in (1), picks out the same referent as the relative clause and occupies the same argument slot. The schematic representation of a correlative construction can thus be captured by the structure in (2):

(2) [correlative clause... relative phrase... ] [main clause... correlate...]

2. Why are correlatives interesting? The interesting property of correlative constructions is that while they are used as equivalents of English-type headed relatives, their syntax and semantics differ from these. The syntactic and semantic differences give rise to a set of properties that are not found with English-type headed relative clauses. These properties are summarized in (3): (3) Characteristic properties of correlatives i. a peripheral position of the relative clause ii. the possibility of spelling out the nominal head both in the relative clause and in the correlate iii. demonstrative requirement on the correlate iv. the availability of multiple relative phrases In the following, these special properties will be illustrated for Hindi, following Srivastav (1991)/Dayal (1996).2

2.1

The position of the relative clause

The first characteristic property of correlatives is their placement. Correlatives predominantly occur in the left periphery, in a position that is not necessarily adjacent to the correlate nominal expression (Dayal 1996): (4) [jo vahaaN khaRii hai] raam us laRii-ko jaantaa hai rel there standing is Ram that girl-acc know is ‘Ram knows the girl who is standing there.’ The peripheral position of the relative does not necessarily mean initial position in the sentence. Topics of various types can precede the correlative: (5) kal [jo vahaaN khaRii hai] raam us laRii-se mil-egaa tomorrow rel there standing is Ram that girl-with meet-fut ‘Tomorrow Ram will meet the girl who is standing there.’



The landscape of correlatives

The left peripheral placement of correlatives clearly contrasts with the distribution of headed relatives. Headed relatives either occur next to the nominal head they modify (cf. 6a), or they occur to the right of it at a distance (cf. 6b). Unlike correlatives, a relative clause that originates from a headed construction can never precede the modified nominal in the left periphery, cf. (6c): (6) a. John called somebody [who he knows from school] yesterday. b. John called somebody yesterday [who he knows from school]. c. *[Who he knows from school] John called someone yesterday. The left peripheral position of correlative clauses is thus a distinctive characteristic that sets them apart from headed relatives in English, and from headed relatives in other languages, too. Hindi correlatives contrast with Hindi headed relatives in the same way. Hindi also possesses a postnominal headed relativization strategy, where the relative clause occupies a clause-internal position, necessarily right-adjacent to the head noun: (7) vo laRkii [jo khaRii hai] lambii hai that girl rel standing is tall is ‘The girl who is standing is tall.’ Such relative clauses cannot be found non-adjacent to their head in sentence-internal position: (8) *vo laRkii lambii [jo khaRii hai] hai that girl tall rel standing is is This contrasts with correlatives, where the relative clause can be placed non-adjacent to the nominal it modifies, as was shown in (4).

2.2

The position of the head NP

The second characteristic property of correlatives concerns the distribution of the common noun they modify. This common noun can be spelled out either inside the relative clause, as was shown in (1) or inside the correlate (cf. 9a), or both inside the relative and in the correlate phrase at the same time, as shown in (9b): (9) a. [jo khaRii hai] vo laRkii lambii hai rel standing is that girl tall is b.

[jo laRkii khaRii hai] vo laRkii lambii hai rel girl standing is that girl tall is lit. Which girl is standing, that is tall. ‘The girl who is standing is tall.’





Anikó Lipták

Headed relatives contrast with correlatives in that they do not allow for the option where the nominal appears both in the head position and in the relative clause.3 (10) a. vo laRkii [jo khaRii hai] lambii hai that girl rel standing is tall is b. *vo laRkii [jo laRkii khaRii hai ] lambii hai that girl rel girl standing is tall is lit. *That girl which girl is standing, is tall. ‘The girl who is standing is tall.’

2.3

The nature of the correlate

The third characteristic property of correlatives is related to the correlate in the main clause. This item has to be a definite phrase with a special requirement: it has to contain a demonstrative item. This is the ‘demonstrative requirement’ referred to by Srivastav (1991)/Dayal (1996). If the correlate does not contain a demonstrative, ungrammaticality results, even in cases where the correlate is a definite phrase otherwise, like in the following sentence (bare nouns are definite in Hindi): (11) *[jo laRkii khaRii hai] laRkii lambii hai rel girl standing is girl tall is ‘The girl who is standing is tall.’ Indefinite phrases like do ‘two’ are similarly ruled out as correlates (12a), although they are fine when the relative clause follows them, in the headed relative pattern (12b): (12) a. *[jo laRkiyaaN khaRii haiN] do lambii haiN rel girl standing are two tall are b. do laRkiyaaN [jo khaRii haiN] lambii haiN two girls rel standing are tall are ‘Two girls who are standing are tall.’ (12a) can be saved by turning do into a partitive phrase, by adding a demonstrative un-meN.se ‘of them’ to it: (13) jo laRkiyaaN khaRii haiN un-meN.se do lambii haiN rel girls standing are that-part two tall are ‘Two of the girls who are standing are tall.’ Apart from definite DPs with a demonstrative, universal quantifiers like sab ‘all’ or dono ‘both’ can also appear as correlate phrases. These are however not exceptions from the demonstrative requirement as these quantifiers can also appear with a demonstrative (ve) without a difference in meaning, suggesting that when this demonstrative is not spelled out, it is present covertly (Dayal 1996).



The landscape of correlatives

2.4

Multiple relatives

Another, very remarkable, property of correlatives is that they can contain multiple instances of relative pronouns, to be matched with the same number of correlate phrases in the main clause: (14)

[jis laRkii-ne jis laRke-ke saath khelaa] us-ne us-ko haraayaa rel girl-erg rel boy-gen with played that-erg that-acc defeated lit. Which girl played with which boy, she defeated him. ‘Every girl defeated the boy she played with.’

Relative clauses with multiple relative pronouns are unique to correlative constructions. A relative clause containing multiple relative pronouns cannot follow multiple nominal phrases as heads, as one relative clause cannot be headed by two phrases at the same time: (15) *us-ne us-ko [jis laRkii-ne jis laRke-ke saath khelaa] haraayaa that-erg that-acc rel girl-erg rel boy-gen with played defeated ‘idem’ The requirement that there be the exact same number of correlates as relative phrases is referred to as the matching requirement (see Leung this volume). The four properties reviewed above are typical of correlatives and do not characterize headed relatives either in Hindi or in English. This reinforces the suspicion that correlatives are fundamentally different from headed relatives. The relation between the relative clause and the main clause demonstrative phrase is not that of noun modification as known in the case of headed relatives.

2.5

Comparison with relatives on the right periphery

It must be noted that correlative clauses are distinct from relative clauses that appear on the right periphery of clauses in what can be taken to be a position reached by extraposition. In Hindi, headed relative clauses can be extraposed to the right, just like in English. According to Srivastav (1991)/Dayal (1996) and Bhatt (2003), right extraposed relatives differ from correlatives in that properties (3ii-iv) do not characterize these.4 To start with the second property, right peripheral relatives cannot contain a common noun in the relative phrase, similarly to headed relatives (cf. 10b) above. The judgments reported below come from Srivastav (1991), but note that Mahajan (2000) considers (16b) to be grammatical: (16) a. vo laRkii lambii hai [jo khaRii hai] that girl tall is rel standing is b. *vo laRkii lambii hai [jo laRkii khaRii hai] that girls tall is rel girl standing is





Anikó Lipták

c. *vo lambii hai [jo laRkii khaRii hai] that tall is rel girl standing is ‘The girl who is standing is tall.’ Concerning the demonstrative restriction, the literature (Srivastav (1991)/Dayal (1996), Bhatt (2003), Mahajan (2000)) agrees that it cannot be found among right peripheral relatives. So the following example is good without any demonstrative (compare the ungrammaticality of (12a)): (17) do laRkiyaaN lambii haiN [jo khaRii haiN] two girls tall are rel standing are ‘Two girls who are standing are tall.’ Property (iv), the availability of multiple relative phrases does not characterize relatives on the right periphery, either:5 (18) *us-ne us-ko haraayaa [jis laRkii-ne jis laRke-ke saath khelaa] that-erg that-acc defeated rel girl-erg rel boy-gen with played intended: ‘Every girl defeated the boy she played with.’ The examples above indicate that right-peripheral relatives cannot contain a common noun ‘head’, cannot host multiple relative phrases, but can have indefinite phrases as correlates. As the reader can ascertain, the same set of properties characterize headed relatives, too. This makes it entirely plausible that the right-peripheral relatives originate as headed relatives, and undergo extraposition to the right. Correlatives on the other hand are arguably not derived from headed relatives via a mechanism of extraposition similar to that of extraposition to the right. In the light of the above discussion the conclusion presents itself that left peripheral relatives constitute a relativization strategy on their own, vindicating the use of a special term, correlativization, for this relative clause formation type. The schematic representation of correlatives (cf. 19a), as opposed to headed and extraposed relatives is shown in (19c). Rel-XP stands for relative phrase and Dem-XP for the correlative phrase. (19) Relative clause types a. [correlative clause Rel-XP...][main clause ... Dem-XP...] b. [main clause ... [NP [relative clause Rel-XP... ]]] c. [main clause ... [NP]i...] [relative clause Rel-XP... ]i

correlative headed extraposed

The structural differences sketched in (19) is what Srivastav (1991) and articles in its wake subscribe to, including the articles in this volume. It has to be noted that there have also been proposals that do not treat correlatives as a relativization strategy distinct from the derivation of headed and extraposed relatives. In these proposals headed relatives and correlatives receive a uniform account. Both types of proposals will be reviewed in Sections 5.1. and 5.2.



The landscape of correlatives

3. Correlatives in the typology of relative clauses The typological literature (Downing 1973, Lehmann 1984, Keenan 1985) also recognizes that correlatives instantiate a typologically distinct type of relativization, in which the relative clause is positioned at the periphery of the main clause. Correlatives are one of the four main types of relative clause formation that can be differentiated according to parameters like the presence of a subordinating nominal head and the position of the relative clause with respect to the modified nominal. The four main types of relative clauses are: postnominal relatives, prenominal relatives, internally headed (also called circumnominal) relatives and correlatives. As was shown in the previous section, correlatives differ from pre- and postnominal relatives in that they do not follow or precede the nominal they modify in an adjacent manner. They are not embedded in a relativized noun phrase. This of course does not mean that they are not subordinated clauses, but the subordinator in this case is a clausal constituent: the main clause. Unlike postnominal and prenominal relatives, correlatives can contain their head noun inside the clause as was shown in (9) above. In this respect they are similar to internally headed relatives. Yet the two differ, too, in several other respects. Firstly, in the case of correlatives the modified nominal need not be spelled out inside the relative, it can also be represented outside the correlative clause (cf. example (9a)). Secondly, while the internal head is always fronted in correlatives, it is not always fronted in internally headed relatives. Thirdly, internally headed relative clauses do not contain a relative pronoun, while correlatives do. The position of correlatives is also different from other types of relative clauses. Correlatives are relative clauses that do not occupy a sentence-internal position corresponding to an argument/adjunct slot, rather they occur in a left-adjoined position in the matrix clause. Such a placement sets them apart from other types of relatives, as the following schematic representation, adapted from De Vries (2002), shows (N stands for the nominal the (cor)relative modifies): (20) a. postnominal relatives: [matrix clause ... [N [relative clause...] ... ] b. prenominal relatives: [matrix clause ... [[relative clause...] N]... ] c. internally headed relatives: [matrix clause ... [relative clause... N...]... ] d. correlatives: [matrix clause (...) [relative clause (N)... ] [matrix clause ... Dem (N)...]] The fact that correlatives do not occupy sentence-internal positions, coupled with the fact that they do not exhibit external determiners, nominalizing suffixes and case endings of various sorts (including adpositions) made researchers like Keenan (1985) or Dayal (1996) conclude that correlatives are not nominal in nature, they do not correspond





Anikó Lipták

to a DP externally. Rather, they are bare sentences, i.e. CPs or IPs. In this they differ from internally headed relatives, which are externally nominal (DPs), evidenced by possible nominal morphology on the relative clause (Culy 1990). Correlatives for this reason cannot be considered to be extraposed internally headed relatives.6 Turning now to semantic typology, and the question how semantic and syntactic types of relative clauses correlate, correlatives seem to be more like restrictive relatives than appositive ones. Grosu & Landman (1998), however, define correlatives – together with free relatives, degree relatives (also called amount relatives) and Quechuatype internally headed relatives – to be of a 'third kind’. The special, third-kind nature of correlatives is due to a meaning component that does not characterize either restrictives or appositives: maximalizing semantics.7 Relative clauses with maximalizing meaning are distinct from restrictive and appositive relatives when it comes to the importance of the head noun for the meaning of the whole construction and with respect to the relative clause. Representing these on a semantic scale indicating the importance of external and internal material, as in (21), we can place the three types in the following way: (21) Appositives sortal external

Restrictives

Maximalizing relatives sortal internal

On the left side of the scale we find so-called sortal-external relatives, where the external material is most important. Sortal external are appositives, and to a less extent, restrictives. With appositives, the reference of the construction cannot be derived from material inside the relative clause. Restrictives are less dependent on external material only, since both internal and external material is crucial for interpretation, but they can still be considered sortal-external. Correlatives and other maximalizing relatives are at the opposite end of the scale, being sortal-internal: the content of the relative clause is more important than external material, when the latter is present. Now, what is exactly the import of maximalizing semantics? Maximalizing means that correlatives always refer to a maximal individual that has the property denoted by the relative clause. In other words, they pick out a maximal individual or maximal degree or the maximal set of individuals/degrees as their denotation. Maximalizing semantics is due to a maximalization operation, which, in the realm of relative clause constructions characterizes free relatives and degree relatives as well.8 To illustrate the effect of maximalization, consider the following degree relative: (22) I invited the boys that there were in the classroom.

[maximalizing relative]

(22) implies that I invited all boys in the classroom. The relative clause here, that there were in the classroom, is clearly not a restrictive clause. If it was, there could not occur in it: (23) I invited the boys who (*there) were in the classroom.

[restrictive relative]



The landscape of correlatives

The difference between restrictive and maximalizing relatives is that the restrictive in (23) singles out boys in the classroom, out of a larger group of boys, while the degree relative in (22) does not make reference to such a larger group, rather, it refers to the maximal ‘amount’ of boys. To illustrate, let us imagine that there are five boys in a classroom. If there are five, it is also true that there are four, three or two boys there. These amounts, however, are not available as the reference of the degree relative, instead only the maximum number of boys is taken, i.e. the denotation is maximalized. The same maximalization applies in correlatives. Consider (24): (24) [jo laRke KhaRe haiN], ve lambe haiN. rel boys standing are those tall are lit. Which boys are standing, they are tall. ‘Every boy who is standing is tall.’ In a similar vein as in the case of the degree relative above, the correlative here refers to all boys standing, which corresponds to the universal meaning of correlatives. When the correlative is singular, it picks out an atomic individual, which is necessarily unique. This gives rise to the characteristic definite meaning of the correlative, and makes the correlative analogous to a definite description. As Grosu and Landman (1998) argue, the definite nature of correlatives explains why the correlate DP must be definite or universal: (25)

[jo laRke KhaRe haiN], ve/dono/sab/*do/*kuch/*adhiktam rel boys standing are those/both/all/*two/*few/*most lambe haiN. tall are lit. Which boys are standing, they/both/all/*two/*few/*most are tall. ‘Those/both/all boys who are standing are tall.’

That this effect is due to maximalization can be shown by the fact that degree relatives are similarly selective when it comes to their head. They only allow definite DPs in head position: (26) I invited {the/the ten/the many/the few/*ten*many/ *some} boys in the classroom. According to Grosu and Landman (1998), there is yet another property of correlatives that could fall out from the maximalizing semantics. As (27) shows, correlatives do not stack (but see Davison this volume for an exception), similarly to degree relatives (cf. 28a) and contrary to restrictives (cf. 28b): (27) *[jo laRkii KhaRii hai] [jo ravii-kii dost hai], vo rel girl standing is rel Ravi-gen friend is that bahut lambii hai. very tall is lit. Which girl is standing, [*who is Ravi’s friend ], she is very tall.





Anikó Lipták

(28) a. I invited the girls that there were in the classroom (*that there were there to study). b. I invited the girls who (*there) were in the classroom, who were there to study. We will come back to the explanation for the impossibility of stacking in Section 6.2 below.

4. The cross-linguistic distribution of correlatives Correlatives are typologically rare constructions among the world’s languages. Downing (1973) argues are they are limited to head-final (OV) languages, according to an implicational universal.  As Keenan (1985) and also De Vries (2002) point out, this generalization needs to be qualified, as head finality is not universal among correlative languages. Rigid verb final languages like Japanese or Turkish do not feature correlatives, “loose” head final languages on the other hand do. Loose head final languages are those that allow some noun phrases, especially heavy noun phrases to occur in postverbal positions, without any special effect of foregrounding or backgrounding. Apart from loose head final languages, languages with exceedingly free word order, like early Sanskrit or Medieval Russian (cf. 29) also had correlatives: (29)

I kotoruju zvezdu potrebno bylo nam videt’ and which.acc star necessary was us see.inf tu zvezdu zaslonilo tucheju that star covered cloud.by ‘The star we needed to see was covered by cloud.’

Similarly to Medieval Russian, present-day Slavic languages also feature correlatives. The documented languages here are: Bulgarian, Macedonian, Serbo-Croatian (Izvorski 1996, Arsenijević this volume) and Polish (Citko this volume). Hungarian, another free word order language, also has correlatives (Bhatt and Lipták this volume, Lipták 2008). The following compendium gives an exhaustive list of languages that have correlatives according to our present knowledge. To start with the Indian peninsula, correlatives were used in Sanskrit (Andrews 1985, Davison this volume). Modern Indo-Aryan languages also use correlatives, with the exception of Southern Konkani, Saurashtri, and Sinhalese. Correlatives are documented in the following Indo-Aryan languages: Assamese (Masica 1991), Bengali (Dasgupta 1980, Bagchi 1994), Bhojpuri (Grierson 1883, Shukla 1981), Dakkhini Urdu (Schmidt 1981), Gujarati (Cardona 1965, Lambert 1971), Hindi-Urdu (Kachru 1973, Srivastav 1991, Dayal 1996), Kashmiri (Wali and Koul 1997), Maithili (Grierson 1883, Yadav 1996), Marathi (Junghare 1973, Berntsen and Nimbkar 1975, Pandharipande 1997), Nepali (Masica 1991, Anderson 2007a,b), Oriya (Sahoo and Hellan 1998), Punjabi (Bhatia 1993), Sindhi (Trumpp 1872). Dravidian languages also have correlatives:



The landscape of correlatives

these can be found in Kannada (Sridhar 1990), Malayalam (Asher and Kumari 1997), Tamil (Asher 1982) and Telugu (Krishnamurti and Gwynn 1986). The literature is divided as to whether correlatives in these languages are borrowed from Indo-Aryan (Nadkarni 1970) or indigenous phenomena (Lakshmi Bai 1985). Burushaski, a language isolate in Pakistan and India also has correlatives (Tifou and Patry 1995, Berger 1998). Among Indo-European languages, the following have or had correlatives: Latin (Gildersleeve and Lodge 1974, Lehmann 1984, Bianchi 2000), Old English (Curme 1912), Hittite (Berman 1972, Raman 1973, Bach and Cooper 1978, Garrett 1994, Probert 2006), Lycian (Garrett 1994), Medieval Russian (Keenan 1985), Bulgarian, Macedonian, and Serbo-Croatian (Izvorski 1996, Arsenijević this volume), Polish (Citko this volume). From other language families we find correlatives in Bambara (Zribi-Herz and Hanne 1995), Basque (Rebuschi 2003, this volume, Lipták and Rebuschi to appear), Huallaga Quechua (Weber 1983), Hungarian (Bhatt and Lipták this volume, Lipták 2008), Tibetan (Cable this volume), Warlpiri (Hale 1976, Keenan 1985). Apart from the above list of languages, De Vries (2002: 388) also lists the following languages that have correlatives, based on various typological sources that do not always quote attested data: Avestic, Diegueño, Erzya, Farsi, Gaididj, Hurric, Kala Lagaw Ya, Mandinka/Maninka, Mohave, Vai and Wappo. When it comes to the distribution of correlatives cross-linguistically, a separate mention must be made about comparative correlatives, which are constructions of the type (30) in English. (30) The more you read, the less you understand. As Den Dikken (2005) shows, constructions of this type are best analyzed as correlatives (for details, see Section 5.3 below). This conclusion is based on a detailed study of comparative correlatives in a variety of languages, including German, Dutch, Russian, Polish and Hungarian and various stages of English. As this list also shows, comparative correlatives are very wide-spread cross-linguistically, more wide-spread than ordinary correlatives. Many languages that do not feature correlatives of the Hindi type productively have comparative correlatives: e.g. English, German, Dutch, French, Maltese or Greek (cf. Beck 1997, Culicover and Jackendoff 1999, Borsley 2003, Den Dikken 2005). In these languages comparative correlatives are handed down from earlier stages of the language – if we can believe Haudry's (1973) diachronic study in claiming that headed relatives in present-day languages are descendants of correlative constructions at earlier stages of the language (on the productive nature of correlativization at earlier stages of English, see Geis 1985 as well). Whether or not such a diachronic development can indeed be attested (for arguments to the opposite, see König and van der Auwera 1988, Probert 2006, Rebuschi this volume), comparative correlatives are not the only correlative-looking constructions in non-correlative languages. These languages frequently feature proverbs with a correlative structure, like the English proverb Where there’s a will, there’s a way.





Anikó Lipták

5. Syntactic approaches to correlatives Syntactic approaches to correlatives concern themselves with the following issues: i. What is the position of the correlative clause? ii. How does it come to occupy this position? iii. What kind of relationship does the correlative clause entertain with the correlate in the main clause? In the following sections, we turn to these issues in turn.

5.1

The position of the correlative clause

As far as the position of the correlative is concerned, all researchers agree that the surface position of correlatives is one adjoined to a clausal projection – at least in the overwhelming majority of correlatives.9 For Hindi, this projection is taken to be IP in Dayal (1996), based on the observation that correlatives can be preceded by topics (see also example (5) above): (31) kaun aayegaa [jo laRkii vahaaN rahtii hai] us-ko maalum hai who come rel girl there live is that-dat known is lit. Who will come, which girl lives there, she knows. The IP-adjoined position of correlatives is also taken for granted in Dwivedi (1994), Mahajan (2000) and Bhatt (2003). Special attention to the attachment site of the correlative in comparative correlative constructions is given in Den Dikken (2005, this volume). Den Dikken shows that the correlative clause in Dutch can either adjoin to CP or IP, depending on the category of the main clause, which in turn depends on the context. In root contexts, the main clause is a CP, and adjunction takes place to this CP. In embedded contexts, the main clause is only an IP, and adjunction is at this level. The difference in category is evidenced by word order differences in the main clause: V2 effects to the right of the comparative correlative in root contexts and obligatory verb final order in embedded ones. The left peripheral position of correlatives has been related to that of topics in Lipták’s (2005, 2008) work on Hungarian and Anderson’s (2007a,b) on Nepali.10 In these languages, correlatives participate in a discourse strategy marking certain topic constituents. The type of topics they instantiate can be language specific. In Hungarian correlatives have the interpretation of aboutness topics, compatible with either old or new information. Their role can furthermore be likened to ‘simplifying’ left dislocates, in the definition of Prince (1998). Both correlatives and left dislocates simplify processing and pronunciation, i.e. they “lift the burden off ” the sentence internal material by placing new information into a separate discourse unit in the higher left periphery. In Nepali, as Anderson (2007b) shows, correlatives – similarly to left dislocates – express



The landscape of correlatives

familiar topics, i.e. those that are salient in the discourse. Nepali correlatives cannot denote a brand new referent. Coupled with their topic function, correlatives exhibit syntactic properties of topical constituents as well, as is shown to be the case for Hungarian in Lipták (2005, 2008). Correlative clauses line up in the left periphery among other topic constituents and can undergo long distance movement of the sort ordinary topics can. Topic syntax of the correlative clause has also been detected in Hittite and Lycian (cf. Garrett 1994). In Lycian, one can even find morphological evidence for the topic status of correlatives, as both correlatives and ordinary topics are followed by the same marker me (see Garrett (1994) for specific examples).

5.2

The derivation of correlatives

Turning now to questions (ii) and (iii) about the placement of correlatives and the relationship they entertain with the correlate, there have been several proposals about these in the literature. These cluster in two families of approaches: the so-called uniformity accounts on the one hand and non-uniformity accounts on the other. Uniformity accounts defend a view that correlatives are derived from underlying headed relatives. Non-uniformity accounts posit that the derivations of headed relatives and correlatives are different. 5.2.1 Uniformity accounts Uniformity accounts do not subscribe to the conclusion we presented at the end of Section 2, namely that correlatives are fundamentally distinct both from headed and right extraposed relatives. Approaches to correlatives that argue for a uniform treatment between correlatives and other types of finite relatives can be found in the following works, all proposed for Hindi: Verma (1966), Junghare (1973), Kachru (1973), Wali (1982), Subbarao (1984), Bains (1989) and Mahajan (2000). In these proposals all relative clause types are derived from headed relatives, including correlatives. Correlatives start out as modifiers of a noun phrase and are taken to move to the left by adjunction. The head NP that is left behind after movement undergoes pronominalization and shows up as a demonstrative expression. Among the uniformity accounts we need to dedicate special attention to Mahajan (2000), which is cast in the antisymmetry framework of Kayne (1994). Mahajan argues that all types of Hindi relatives are headed relatives and should receive a head raising analysis. Thus, the relativized NP is generated inside the relative clause and is moved to an IP-initial position by scrambling. This scrambling step is followed by an optional movement of the NP into Sp,CP of the relative clause, leaving the relativizer behind in IP. The derivation of correlatives starts out with the building of a headed relative in these steps, too, and proceeds with the application of two more operations: scrambling of the whole relative clause to the left and some deletion operation in either the fronted relative or its copy. Deletion can apply to different parts of the structure (sometimes





Anikó Lipták

to non-constituents), deriving all the structures that surface as well-formed outputs. These structures are shown in the following representations (Rel stands for the relative marker; the DP containing Dem(onstrative) and a CP corresponds to the relative clause that has been scrambled to the beginning of the main clause IP): (32)

a. b. c. d.

[DP Dem [DP Dem [DP Dem [DP Dem

[CP [RelP Rel NP... ]]] [IP ... [DP Dem [CP [RelP Rel NP... ]]] [IP ... [DP Dem [CP NP [RelP Rel... ]]] [IP ... [DP Dem [CP NP [RelP Rel... ]]] [IP ... [DP Dem

[CP [RelP Rel NP... ]]] ] [CP [RelP Rel NP... ]]] ] [CP NP [RelP Rel... ]]] ] [CP NP [RelP Rel... ]]] ]

The scrambling step of the derivation explains why the correlate needs to be a definite item (although it says nothing about the obligatoriness of a demonstrative in it). Since only definite phrases can undergo scrambling, indefinite correlates are ruled out. Double spellout of the head NP is accounted for by allowing for various pronunciation possibilities for the nominal phrase as well. This NP undergoing movement to Sp,CP inside the relative clause can be spelled out twice, both inside RelP in the moved copy of the relative and in Sp,CP in the original copy of the relative: (33) [DP Dem [CP NP [RelP Rel NP... ]]][IP ... [DP Dem[CP NP [RelP Rel NP... ]]] ] The same instance of double spellout cannot happen in headed relatives, since, as Mahajan argues following Kayne (1994), two copies of the same item can only be spelled out simultaneously if they do not c-command each other. Postposed relatives, which in this account are also derived via the above derivational steps, followed by some remnant movement steps, are predicted to allow for a double spellout of the NP, too, since in this case there is no c-command relation between the two copies, either. This squares with the facts according to the judgments of Mahajan and speakers in Delhi he consulted, who accept double spellout of the head noun in postposed relatives. Thus, these speakers accept example (16b), which is considered ungrammatical in Srivastav (1991)/Dayal (1996) and Bhatt (2003): (34) vo laRkii lambii hai [jo laRkii khaRii hai] that girls tall is rel girl standing is ‘The girl who is standing is tall.’ It is important to note that Mahajan opens up the empirical domain of correlativization not only in reporting dialectal differences in the acceptability of cases like (34) (as well as the acceptability of multiple relatives on the right periphery, see fn. 4), but also in that he considers cases where the left peripheral relative is preceded by a demonstrative expression, cf. the structures in (32a,c). While these often occur in informal speech (Rajesh Bhatt p.c), they are usually not accounted for in theoretical works. A real life example corresponding to structure (32a) is provided in (35):



The landscape of correlatives

(35) vo [jo aadmii sita-ko acchaa lagtaa hai] mujhe vo pasand nahĩ: hai that rel man Sita-dat nice seem is I.dat that pleasing not is ‘I do not like the man who Sita likes.’ 5.2.2 Non-uniformity accounts The underlying idea of non-uniformity accounts of correlativization is that correlatives are fundamentally different from headed relatives, so much so that a uniform treatment of the two types is not feasible. Different incarnations of the non-uniformity approach can be found in Donaldson (1971), Downing (1973), Bach and Cooper (1978), Dasgupta (1980), Lehmann (1984), Keenan (1985), Andrews (1985), Srivastav (1991), Dayal (1996), Izvorski (1996) and Bhatt (2003). Of these, we only deal with the most recent four pieces of work in detail here, because in these the syntactic (and sometimes also semantic) analysis of the correlative construction is placed center-stage. In these proposals we find three basic types of approaches centering around the question of how the correlative clause combines with the main clause and what kind of relationship it entertains with its correlate. These approaches differ along the lines of two ingredients of the analysis, positing (i) base-generation vs. movement of the relative clause; (ii) local modification vs. binding of the correlate by the correlative clause. In what can be termed the high-adjunction & binding account, proposed by Srivastav (1991)/Dayal (1996), correlatives differ from headed relative constructions in that the correlate phrase and the relative clause do not form a constituent at any point of the derivation. Instead, the correlative clause is base-generated adjoined to the main clause IP from the left. From its left adjoined position, the correlative binds the correlate, which is an ordinary phrase in the main clause. This binding relation is quantificational.  The correlative behaves as a generalized quantifier and the correlate as a variable, which is akin to an overtly spelled out A-bar trace. This configuration can be read off the structure in (36): (36) high-adjunction & binding approach [IP [CorrelCP ... RelXP... ]i [IP ... DemXPi...]] Besides Srivastav (1991)/Dayal (1996), Bhatt (2003) also uses high adjunction for the derivation of multiple correlatives (see below).11 Evidence for the quantificational nature of the correlative comes from the observation that correlatives occupy a left-adjoined position that is similar to that of raised quantifiers. The trace-kind of behavior of the correlate on the other hand follows from the presence of locality effects between correlative and correlate. Such locality effects subsume island violations of the usual kind, exemplified by the CNPC violation in (37a), with actual data in (37b): (37) a. [CorrelCP ]k[IP ... [DP DP [RelCP ...DemXPk...]]... ] b. *[ jo vahaaN rahtaa hai] mujhe vo kahaani rel there stay is I.dat that story





Anikó Lipták



jo Arundhati-ne us-ke.baare.me likhii pasand hai rel Arundhati-erg that-about write pleasing is lit. Who lives there, I like the story that Arundhati wrote about that boy.

For the semantic computation of the high-adjunction & binding account, see Section 6.1 below. An entirely different way of cashing out locality effects is found in what can be referred to as the low adjunction & movement account. This proposal, worked out in Bhatt (2003) for single correlatives, has it that the correlative modifies the correlate phrase locally by forming a complex adjunction structure with it in the base. From this low position, the correlative optionally moves out to adjoin to IP via an operation such as A-bar scrambling or QR. If the correlative moves, the correlate phrase can undergo optional scrambling as well: (38) [IP [CorrelCP ... RelXP... ]i[IP ... (DemXPj) [...[[ t i ] DemXPj ] ...]]] Since the low adjunction account operates with movement of the correlative to the left periphery, it predicts island effects of the kind observed in (37) to be the result of this movement operation. Other arguments in favor of the low-adjunction analysis come from reconstruction effects, both in the domain of condition C effects as well as pronominal binding facts. Here I exemplify these with a binding principle C effect (Bhatt 2003): (39) a. [CorrelCP R-expl.... ]k [ pronl DemXPk...] b. *[jo laRkii Sita-kol pyaar kar-tii hai]k us-nel rel girl Sita-acc love do is that-erg us-kok ţhukraa di-yaa that-acc reject give-pfv ‘Hei rejected the girl who loves Sitaj.’ The name (Sita-ko) contained in the correlative cannot be coreferential with the pronoun (us-ne) in the matrix clause, which argues for a reconstruction step that takes the correlative back to a position c-commanded by this matrix pronominal. If correlatives originate from a DemXP-adjoined position and undergo obligatory reconstruction at LF, as shown in (40), the observed coreference relations are ruled out as a binding principle C violation. (40) [CorrelCP R-expl.... ]k[pronl [CorrelCP R-expl.... ] DemXPk...]] The most striking piece of evidence for the low adjunction & movement account comes from data whose relevance is somewhat underrated in other works (with the exception to Wali 1982): the possibility of generating the correlative clause and the correlate as a constituent in overt syntax as well. The existence of such structures have been acknowledged as a possibility in Dayal (1996), who, quoting Wali (1982), cites the following case:



The landscape of correlatives 

(41) [DP [jo ayee] un-kaa kaam] [DP[jo gaye ] un-ke rel came they-gen work rel left they-gen kaam-se ] behtar hai work-than better is ‘The work of those who came is better than the work of those who left.’ As indicated by the bracketing, we find two pairs of correlative and correlate phrase forming a constituent DP in this sentence. A similar configuration is found in (42). This example also contains two correlative-correlate sequences, each sequence involving the correlative clause adjacent to its own demonstrative: (42)

Ram-ne [ jo laRkaa tumhaare piichhe hai ] [DemXP us laRke-ko] Ram-erg rel boy your behind is that boy-dat [jo kitaab Shantiniketan-ne chhaapii thii] [DemXP vo kitaab] dii [rel book Shantiniketan-erg print-pfv was that book give-pfv ‘Ram gave the book that Shantiniketan had published to the boy behind you.’

Facts like this follow without further assumptions if we assume that correlative clause and correlate form a constituent at some level and can be moved as one constituent in the syntax. These remarkable complex DP-structures force us to allow for DP-adjunction for the correlative at least as a possibility, and together with the observed locality effects they clearly vindicate the low adjunction & movement analysis. Multiple correlatives receive a distinct treatment in Bhatt’s analysis, due to the fact that they behave differently from single correlatives both according to the evidence of locality effects and the impossibility of complex formation. Multiple correlatives do not give rise to DP-adjunction structures and they do not reconstruct into the main clause, either. There is no restriction on coreference between a pronoun in the matrix clause and a name contained in a multiple correlative adjoined to the clause, for example: (43) [jis-ne Ram-ko jise di-yaa ] us-ne us-se rel-erg Ram-acc rel-dat give-pfv that-erg that-ins us-kii taariif kii that-gen praise did ‘For x and y, such that x gave Ram to y, Ram praised x to y.’ This shows that multiple correlatives do not undergo movement to the left periphery, rather they are base generated adjoined to IP. The observed locality effects have also gained a third kind of explanation in the literature. In Izvorski (1996), which discusses Hindi and South Slavic correlatives, the correlative clause is base-adjoined to the main clause and the correlate demonstrative phrase is argued to undergo focus movement to the left periphery, predicting locality effects. The movement of the correlate takes place to Spec,CP via A-bar movement, a step that is covert in Hindi and overt in Slavic, as Izvorski claims. (44) [CP [CorrelCP ... RelXP... ] [DemXPi] [CP ... [ DemXPi ]...]]



Anikó Lipták

This high adjunction & correlate-raising account can be viewed as a combination of the two approaches mentioned above. It keeps the high, CP/IP-adjoined position for the correlative and takes care of locality effects via arguing for the raising of the DemXP in an A-bar manner. According to Izvorski, the movement step depicted in (44) is parameterized according to the properties of wh-movement in a given language: it takes place overtly in overt movement languages like Bulgarian and Serbian, and covertly in covert wh-movement languages like Hindi. Bhatt (2003) notes, however, that while this parametric account is theoretically elegant, covert correlate phrase movement is unlikely to take place in Hindi, as in this language finite clauses are islands to covert movement. Izvorski’s account is also slightly improved upon by Lipták (2005) who shows that the overt movement step of the correlate phrase can also be a process of topicalization, which is an option utilized in Hungarian (Lipták 2008) and in Serbian (Arsenijević this volume) as well.

5.3

Comparative correlatives

As briefly mentioned in Section 4 above, comparative correlatives (CCs) are constructions expressing comparison between two clauses, of the type in (45) in English: (45) The more you read, the less you understand. The first of the two clauses expresses a condition under which the second clause is true. Due to this conditional import, these sentences are also sometimes referred to as ‘comparative conditionals’ in some works, like in Beck (1997). The correlative nature of these constructions is quite obvious in languages with correlatives, like Hindi. In these languages comparatives of this sort are expressed via the means of ordinary correlativization. Consider the following example (quoted from Den Dikken (2005), who attributes it to Rajesh Bhatt p.c.): (46) [jiitnaa suuraj chamk-aa] utnii(-hii) ThanD baRh-ii how.much sun shine-pfv that-much(-only) cold increase-pfv ‘The more the sun shone, the colder it got.’ In this example, just like in ordinary correlatives, we find a left peripheral relative clause adjoined to a main clause and linked to a demonstrative pronominal (utnii(hii)). While this example does not pose any problem for a syntactic analysis, the underlying structure of the equivalent construction in other, non-correlative languages, like the English (45) is not evident at first sight, because the construction has some quirky properties that are difficult to explain rightaway. Let us illustrate two of these. First, according to the evidence of locality effects, the ‘the...’ phrases that introduce each clause are fillers similar to ordinary wh-phrases in that they undergo movement to Sp,CP12 and bind a trace:



The landscape of correlatives 

(47) a. *The more Mary knows a man who ti eats, the poorer she gets. b. *The more he eats, the pooreri he knows a woman who gets ti. At the same time, these phrases are not normal wh-phrases and cannot be subsumed under degree expressions of other types, like so or all the more either, argue Culicover and Jackendoff in Culicover and Jackendoff (1999). The second quirky property concerns extraction. Although both clauses are similar to wh-clauses and thus should constitute islands, both clauses are extractable from in English: (48) a. This is the sort of problem whichi the sooner you solve ti, the more easily satisfy the folks up at corporate headquarters. b. The folks up at corporate headquarters are the sort of people whoj the sooner you solve this problem, the more easily you’ll satisfy tj. Extraction should be less problematic for the second clause, as this clause functions as the main clause to which the first is subordinated to, as reflected by the choice of tagquestions, among other things: (49) The more we eat, the angrier you get, don’t {you /*we}? Yet, if anything, extraction from the first clause in English CCs is easier than extraction from the second clause. Concentrating on these quirky properties of the construction, Culicover and Jackendoff conclude that the English CC embodies a ‘syntactic nut’, a construction type that does not conform to principles of UG grammar. Instead, it is sui generis – at least when it comes to the the-phrases and the combination of the clauses. About the latter, it is concluded that both clauses of the comparative correlative have the status of coordinate clauses in the syntax, while in the semantics the first clause is subordinated. To counter Culicover and Jackendoff ’s conclusion about the syntactic lawlessness of CCs, Den Dikken (2005) subjects correlative comparatives to meticulous scrutiny in a handful of languages, involving Dutch, German, Hungarian, Russian and various stages of English (see also Bhatt (2009) on Greek). Taking the lead of the evidence in (49) for the subordinated nature of the first clause, and working his way into the microscopic structure of the the-phrases Den Dikken shows that the internal composition of CCs does obey UG principles and that this construction furthermore should be analyzed as genuine and cross-linguistically consistent correlative constructions. In these, we find the first clause as a relative clause adjoined to the second clause, with a basic structure as in (50):13 (50) [correlative clause [the more]i I read ti][main clause [ the more ]j I understand tj] The correlative nature of the construction manifests itself in various ways, including the subordinate nature of the first clause as well as the fact that this construction is



Anikó Lipták

always bi-clausal, which the author relates to the fact that correlatives do no stack (compare to (27) above): (51) *The more you eat, the fatter you get, the sooner you die. Concerning extraction facts, Den Dikken shows that the above observed quirks are particular to English and do not characterize other languages. In German or Dutch, extraction can never take place from the first clause and can only take place from the second clause if that has the comparative phrases in initial position (in-situ placement of the correlate is also allowed in these languages). Such a state of affairs is entirely expected if CCs have an underlying structure as correlatives, but does not follow if they are coordinated clauses as suggested by Culicover and Jackendoff. The latter scenario would also make wrong predictions about extraction of predicate nominals from both clauses. These can extract out of clauses combined by ordinary coordination, but not out of comparative correlatives:14 (52) a. the kind of doctor Op that [he would very much want to be t] but [does not consider himself capable of becoming t] b. *the kind of doctor Op that [the more he wants to be t], [the less able he will be able to actually become t] Turning now to the nature of the fronted comparative phrase, its morphosyntax also complies with X-bar theory in Den Dikken’s analysis. The-phrases in English and their cross-linguistic equivalents are run-of-the-mill degree expressions, DegPs, whose specifier contains a prepositional measure phrase. Evidence for such a complexity comes from modern Russian (cf. 53) or from 16th century English examples (cf. 54): (53) naskol’ko luchshe mashina nastol’ko ona dorozhe by.how.much better car by.that.much it more.expensive ‘The better the car, the more expensive it is.’ (54) by how much the lesse he looked for his discourse, by so much the more he lyked it As the Russian example clearly evidences, the comparative expression contains a relative operator in the first clause, and a demonstrative expression in the second. A parallel representation can also be assigned to the English example, as illustrated in (55). (55) a. DegP in correlative clause: relative phrase [DegP [PP by [QP how much]] [Deg’ the [AP lesse]]] b. DegP in main clause: demonstrative phrase [DegP [PP by [QP so much]] [Deg’ the [AP more]]] What makes comparative phrases somewhat peculiar in comparative correlatives is that parts of the DegP – the Deg head, the measure phrase or the preposition introducing the measure phrase – need not be overt in some languages. In modern English



The landscape of correlatives 

the-phrases, for example, the measure PP is covert, and we only get to see the Deg head (the) and the comparative AP in the fronted comparative phrase. While this discussion only concentrated on the most difficult puzzles that comparative correlatives present the theorist with, it is clear that these constructions can fruitfully be subjected to an analysis in terms of a correlative structure, concerning both the combination of clauses and the morphosyntactic composition of the relative clause.

6. Semantic approaches to correlatives The semantic composition of correlative constructions requires special attention if one analyzes these along the lines of non-uniformity approaches. If correlative constructions are not assembled in the same way as headed relatives, the basic tenet of non-uniformity accounts, their interpretation must proceed differently from headed relatives, too –assuming a compositional syntax-semantics correspondence like Montague’s approach. The compositional interpretation for headed relatives, following Partee (1975), involves combining (both in the syntax and the semantics) the relative clause with a common noun, and applying the definite article to the result. As Bach and Cooper (1978) noticed, the same interpretation is not available for correlatives, since at the point where the correlative is inserted into the structure, the correlate DP has already been composed and interpreted. To solve this problem, Bach and Cooper assumed that all relative-modified nominals – whether next to their modifying relative clause or at a distance from it – have an implicit property variable (R), which gets filled in (via lambda-abstraction) by the relative clause. This allows the relative clause that is not a constituent of the head DP to be interpreted inside that DP. In this model, the difference between headed relatives and correlatives is that R gets filled in at the DP level in the case of ordinary headed relatives, and at the clausal level in the case of correlatives.

6.1

Dayal’s (1996) approach to correlatives

A criticism of this uniform semantic approach to correlatives and headed relatives was provided in works by Veneeta Dayal in Srivastav (1991) and Dayal (1996). Using syntactic evidence about the distinct nature of correlativization and headed relative formation (see Section 2 above) she has shown that correlatives are not ordinary noun modifiers, and thus a uniform account of headed relatives and correlatives is mistaken. Instead, correlatives instantiate a strategy based on a different interpretive mechanism, that of quantification. This conclusion suggests itself quite naturally if we consider that (i) correlatives occupy a position where quantified phrases are interpreted in languages (i.e. IP adjoined position) (ii) the relationship between the correlative clause and its correlate shows typical properties of operator-variable relationships such as island violations (recall Section 5.2.2.).



Anikó Lipták

In Dayal’s account, correlative constructions are interpreted according to the rules of quantification. The correlative clause is a generalized quantifier that needs to bind an argumental variable in the main clause. Since the correlative does not move out of the main clause, the argumental variable that it binds cannot be a trace, instead it has to be an overt pronominal. The correlate DP is such a pronominal expression (the demonstrative element inside being the variable), and can be considered similar to a phonetically realized trace, a resumptive pronoun. What kind of quantifier is the correlative? As was shown in Section 3, correlatives have a special meaning component: maximalizing semantics. Maximalization, when applied to degrees, restricts the set of degrees to the singleton set containing the maximal degree (if there is one). When applied to individuals, maximalization results in a definite reading, which means that the correlative is interpreted as a singular definite description denoting a unique individual when the relative operator has singular morphology, and a plural definite when the relative operator is plural – the exact same interpretation free relatives receive in Jacobson (1995). Correlatives can thus be considered generalized quantifiers over maximal individuals. Uniqueness can be absent under two conditions: so-called quantificational variability effects (QVE) and relatives with an ever-type suffix bringing in free choice interpretation. Both occur with generic tenses only. QVE shows up with adverbs of quantification like often, illustrated in (56): (56) [jo laRkii mehnat kartii hai ] vo aksar safal ho-tii hai rel girl effort do is that often successful be-hab is lit. Which girl makes an effort, she is often successful. ‘A unique girl who makes an effort is often successful.’/’Most girls who work hard are successful.’ As the translation shows, the sentence has two readings. Under the second, ‘variable’ reading there is no uniqueness: the correlative does not denote a unique girl. Dayal (1995, 1996) argues that we can preserve a uniqueness analysis if we treat quantifier variability via quantifying over situations. In this approach, the variable reading can be paraphrased as: ‘most situations that involve a unique girl making an effort, are situations in which this unique girl is successful’. Uniqueness can be checked for minimal situations, so in cases where there are more girls making an effort, there will also be minimal situations in which a single girl makes an effort. This way, we allow for uniqueness and at the same time we allow for the number of situations for often to quantify over. The other apparent exception to uniqueness are correlatives with the particle -bhii. This particle has an interpretation similar to English -ever. This morpheme can have a free choice reading or it can indicate that the identity of the individual denoted by the relative is not known to the speaker: (57) [jo-bhii laRkii mehnat kartii hai] vo safal ho-tii hai rel-ever girl effort do is that successful be-hab is ‘Whichever girl makes an effort, she is successful.’



The landscape of correlatives 

As mentioned above, the effect here is also dependent on the tense of the clause. The absence of uniqueness only shows up on a generic interpretation of the sentence. If the relative has episodic tense, -bhii receives the ‘unknown identity’ interpretation: (58) [jo-bhii laRkii vahaaN khaRii hai] vo ravi-kii dost hai rel-ever girl there standing is she Ravi-gen friend is ‘Whichever girl is standing there, she is Ravi’s friend.’ This shows that the uniqueness effect is dissipated not by -bhii itself, but by genericity. Before going on it has to be noted that free choice readings of correlatives are also available in other languages, sometimes even without an overt -ever suffix on the relative phrase. In a language like Hungarian, there is actually a tendency to interpret all correlatives with generic tense and what can be called a free choice interpretation: (59) [Aki szorgalmasan dolgozik], azjutalmat kap. rel.who diligently works that reward gets ‘Who works diligently will get a reward.’ Non-generic tense, on the other hand, just like in Hindi, requires a uniqueness interpretation: (60) [Aki először lépett be], azt nem ismerem. who first entered in that.acc not know.1sg ‘I do not know the person who entered first.’

6.2

Semantics for single correlatives

Dayal (1996) follows Jacobson’s (1995) study of free relatives in assigning correlatives the semantics of definite descriptions, and attributes definiteness to the complementizer of the relative clause. This complementizer is thought of as a special kind of definite determiner corresponding to a two-place operator. The role of this operator is to map the intersection of the properties designated by the NP in the relative phrase and by the predicate of the clause to the maximal individual within this intersection. The set of properties of this individual is designated by the correlative that is shifted to a generalized quantifier. This generalized quantifier is then applied to the matrix clause, which is taken to designate a property, with the correlate phrase construed as a variable inside it. This approach can explain why correlatives are sensitive to the type of their correlate – a property we dubbed 'the demonstrative requirement' in Section 2. Note first that headed relatives are not sensitive to whether there is a determiner in their head phrase, due to the fact that they attach to (and take scope over) the level of the noun only. The determiner layer builds in only after the noun has combined with the relative clause. In the case of correlatives, on the other hand, correlative clause and correlate DP have a different relationship: since the correlate DP is a bound variable, it needs to have a variable in its denotation. This is the demonstrative element, which can get



Anikó Lipták

bound from outside when it is not interpreted deictically. Definite DPs of other types, like proper names, are robustly impossible correlates, as these cannot vary with the relative clause. Somewhat different alternatives to Dayal's account have been provided by Grosu and Landman (1998), Grosu (2000, 2002) and Gajewski (2008). These differ in where and how they take maximalization to be active in the relative clause – notably, the very aspect of any analysis that involves some kind of arbitrariness. Gajewski (2008), which is designed to account for multiple correlatives (see below), derives the maximality effect from the relative phrase instead of the relative complementizer. Grosu (2000, 2002) (following the insights of Grosu and Landman (1998)), assumes together with Dayal that the source of maximalization is the complementizer C, but considers this to be the effect of a feature [DEF], which is similar, but not equivalent to a definite article. The difference is that [DEF] applying to the relative CP does not shift its type either higher or lower, rather, it triggers the mapping of a set to a singleton set, thus preserving the type of the input. As a result of this feature, the relative CP is interpreted as the singleton set whose unique member is the output of a maximalizing operation. Further, Grosu & Landman (1998) differ from Dayal also in their treatment of the correlate. They do agree that there is a variable in the place of the correlate (bound by abstraction), but they take the correlate itself to contribute to the building of the quantificational correlative clause. They take this correlate phrase to be interpreted in the position of an external head of the correlative clause. As we have briefly mentioned above in Section 3, the Grosu-Landman proposal neatly explains the general tendency that correlatives do not stack15, with reference to the singleton status of CP. Stacking of relatives would require the intersection of the denotation of two sets. Restrictive relatives for example, which intersect with the meaning of their head, allow stacking. Maximalizing relatives, like correlatives, on the other hand, do not involve set intersection, as maximalization always creates a singleton set. Take for example two correlatives, each denoting a singleton set. If the two contain distinct members, their intersection is empty, and if they contain identical members, their intersection is identical, tautologous. In both cases the outcome of intersection is infelicitous and stacking is thus ruled out. The Grosu-Landman analysis also explains why the correlate needs to be a definite phrase. The correlative is only compatible with deteminers that do not single out subsets of the set denoted by the correlative. Since the relative CP denotes a singleton set, the correlate that is external to it can only be felicitous in case its implications are consistent with the uniqueness of CP’s only member. Definite determiners and universals are consistent, but indefinites are not since they carry non-uniqueness implications. While the definite nature of the correlate is thus explained, it remains unclear in this account why the definite correlate needs to contain a demonstrative expression.



The landscape of correlatives 

6.3

Semantics for multiple correlatives

Multiple correlatives require separate mention in any semantic analysis, as their properties are slightly different from single correlatives. Multiple correlatives contain two (or more) relative phrases, which correspond to two (or more) correlates in the main clause. The number of relative and correlate phrases has to match: (61) [jis laRkii-ne jis laRke-ke saath khelaa] us-ne rel girl-erg rel boy-gen with played that-erg us-ko haraayaa that-acc defeated lit. Which girl played with which boy, she defeated him. ‘The girl who played with a boy, defeated him.’/’Every girl defeated the boy she played with.’ As shown by the translation, the correlative need not refer to a single girl-boy pair, but can also quantify universally over multiple girl-boy pairs. The second option is interesting since the relative phrases are both singular. To derive this universal force of the correlative, Andrews (1985) suggested that the correlative has the same type of quantificational structure as a conditional. Replacing the relative phrases with an indefinite, we can recast the relative clause as a conditional and get the same universal meaning: (62) ∀x,y [girl'(x) ∧ boy'(y) ∧ played-with'(x,y)][defeated'(x,y)] Dayal (1996), however, refutes this conditional-based analysis with two pieces of evidence. First, while correlatives need to match up with correlate phrases in a one-to-one manner, conditionals do not have such a matching requirement. Second, an analysis of correlatives as conditionals would entail that we predict that singular correlatives are always interpreted as universals, contrary to fact. Dayal’s analysis of multiple correlatives is a combination of her generalized quantifier approach to single correlatives and her analysis of multiple questions in Dayal (1996). The parallel with multiple questions is based on the observation that both multiple questions and multiple relatives have a universal meaning, and both denote functional relations.16 The latter means that the first relative phrase needs to be exhausted (all girls had to play with a boy) and there has to be uniqueness in that for each member of the relative-phrase there can be no more than one pair (no girl can have played with more than one boy). Dayal proposes that multiple relatives differ from single ones in that the relative complementizer takes multiple heads and denotes a set of relations, not properties. So a multiple relative with n relative phrases denotes a set of n-place relations, determined by a unique function. The correlative combines with the main clause that in the case of (61) denotes a relation, due to the demonstratives that are abstracted over. This accounts for the matching requirement: quantification is only defined if the main clause also denotes an n-place relation, so it has to have n demonstratives. The uniqueness/maximalizing effect present in correlatives results in the



Anikó Lipták

above specified uniqueness condition on the number of pairs associated with the first relative phrase. Gajewski (2008) puts forward an improvement over Dayal’s analysis, by showing that it can simplified. The way to simplify it is to follow Jacobson’s (1995) analysis of free relatives even more closely in the derivation of the definite meaning: the two operations that derive it, maximalization and iota-shift need to be kept separate, and should be allowed to apply at different points of the derivation. Maximalization in this account is due to the semantics of the relative operator. This operator maps a predicate into a predicate over one individual (this is the same effect that is attributed to the feature [DEF] in Grosu and Landman 1998). Iota-shift is another operation that can apply to the maximalized predicate. It can raise the type of this predicate (denoting a singleton set) to an individual, denoting the unique member of this set. Gajewski shows that a double correlative can be derived by applying maximalization to the clausal node containing the lower relative phrase and the rest of the sentence. When the second relative phrase builds in, giving us pairs of individuals, the presupposition of this maximalization is projected (in the manner of Heim 1983) and this gives rise to the observed uniqueness meaning (no girl has played with more than one boy). Before the correlative combines with the main clause, maximalization happens again, now to the whole correlative, optionally preceded by a pluralization operation. When pluralization does not apply, the result of maximalization is a singleton set containing one pair of individuals. After iota-shift of this set and combination with the main clause, which is taken to be a two-place predicate just like in Dayal’s account, we get a single pair reading. If pluralization takes place before the correlative is maximalized, the outcome is similar except that we get a set of pairs. This corresponds to the universal reading.

6.4

The relation between correlatives and conditionals

Before closing this section, we need to turn to another area of research on correlatives, that which covers the relationship between conditionals and correlatives, discussed in various works on ordinary correlatives (Andrews 1985, Geis 1985, von Fintel 1994, Izvorski 1996, Cheng and Huang 1996, Dayal 1996, Bittner 2001, Bhatt and Pancheva 2006) as well as on comparative correlatives (McCawley 1988, Michaelis 1994, Beck 1997, Culicover and Jackendoff 1999). This area is an important field of study as there exist several interpretive and formal parallels between correlative and conditional constructions. There are morphological, syntactic and semantic similarities. Concerning morphological marking, in languages where correlativization is a productive strategy correlatives and conditionals often use the same marker of subordination. As Cable (this volume) shows, Tibetan correlatives contain a particle na, as seen in (64), which is also productively used in conditional statements, like in (63): (63) [ Kyodrang Lhasa la ’gro na ] nga ’gro gi yin. you Lhasa dat go if I go non.pst aux ‘If you go to Lhasa, I will go there.’



The landscape of correlatives 

(64) [Khyodra-s gyag gare nyos yod na ] nga-s de bsad pa yin. you-erg yak what buy aux if I-erg that kill perf aux ‘I killed whatever yak you bought.’ A similar pattern is observed in Basque (Rebuschi this volume). In present day Basque conditionals use ba- ‘if ’, while correlatives use bait- as a finite complementizer. According to the evidence of some texts, however, there have been quite many interchanges between the two forms at earlier stages of the language. At some stages, baitcould replace ba- and vice versa.17 Apart from complementizers, the relative pronouns and main clause correlates can also be selected from the same pronominal paradigms in languages. The following example from Marathi (originally from Pandharipande 1997, quoted by Bhatt and Pancheva 2006) shows this parallel very clearly. The first example is a conditional, the second a correlative construction:18 (65) (dzar) tyāne abhyās kelā tar to pā hoīl if he studying do.pst.3msg then he pass be.fut.3sg ‘If he studies, he will pass (the exam).’ (66)

dzo mānūs tudzhyā śedzārī rāhto which man your neighborhood.in live.prs.3msg to mānūs lekhak āhe that man writer is ‘The man who lives in your neighborhood is a writer.’

Parallels between conditionals and correlatives extend beyond the use of the same morphological markings or the selection of pronouns. In their syntax, the two kinds of construction also show many similarities. To start with a basic one, they both involve a bi-clausal structure with a subordinate clause adjoined to the main clause. As Bhatt and Pancheva (2006) show, sentence initial conditionals adjoin to CP/IP, just like correlatives. Furthermore, conditionals, just like correlatives, can be coindexed with a proform – this form in English is then – whose placement observes conditions also found with correlate phrases, see the details of a particular parallel concerning stacking possibilities in Bhatt and Pancheva (2006).19 Based on such syntactic parallels, Bhatt and Pancheva conclude that conditional clauses with then are correlative structures, and in line with this they are also interpreted just like correlatives. Semantically, the most obvious parallels between correlatives and conditionals are the maximalization effects that characterize conditional clauses just as much as free relatives: they are both definite descriptions, the difference boiling down to the type of entity they denote. While correlatives are definite descriptions over individuals, conditionals are definite descriptions of possible worlds (Bhatt and Pancheva 2006). A slightly different formulation is found in Bittner (2001), who is concerned with the logical representation of individuals and possibilities, from the point of view of centering theory. She shows that conditional clauses center a possibility,



Anikó Lipták

while correlatives center an individual. In the footsteps of these accounts, yet reversing the direction of assimilation, Arsenijević (this volume) argues that correlatives are a subcase of conditionals.20 Based on the observation that Serbian correlatives and conditionals show parallel syntactic and semantic behavior, this work treats correlatives as a subtype of conditional clauses, a type that involves extreme non-specific wh-expressions in topic positions in the CP-domain. Conditionals in turn are analyzed as yes-no relative clauses, thus coming a full circle and claiming in effect that conditionals are a subcase of relativization, just like in Bhatt and Pancheva’s and Bittner’s accounts. The semantic parallel between conditionals and correlatives is demonstrated very clearly by the following Warlpiri example, where one and the same construction is actually ambiguous between a ‘correlative’, individual reading and a conditional ‘possibility/possible world’ reading (Bittner 2001, quoting Hale 1976): (67)

Maliki-rli kaji-ngki yarlki-rni nyuntu dog-erg ‘same.topic’.3sg.2sg bite-non.pst you ngula-ju kapi-rna luwa-rni ngajulu-rlu. dem-top fut-1sg.3sg shoot-non.pst me-erg (i) ‘As for the dog that bites you, I’ll shoot it.’ (ii) ‘If a dog bites you, then I’ll shoot it.’

(individual-centered) (possibility-centered)

As the interpretations show, the subordinate clause can denote either an individual (reading (i)) or a possibility (reading (ii)). The kind of ambiguity indicated in (67) also arises in correlatives with generic or habitual tense in other languages. Depending on the language, the conditional reading can be predominant over the individual reading. This seems to be a case in Hungarian, or Slavic (Arsenijević, p.c.). Consider the following example and its predominant, free choice/conditional interpretation:21 (68) Amelyik kutya közel jön hozzám, azt elkergetem. rel.which dog close comes to.me that.acc chase.away.1sg ‘Which(ever) dog comes close to me, I’ll chase it away.’ = ‘If a dog comes close to me, I’ll chase it away.’ Note that the second reading resembles examples of ‘donkey’ anaphora: there is a covert universal operator whose restriction is the first clause and the main clause contains an E-type pronoun azt ‘that-acc’. Correlatives clearly differ from regular headed relatives in the availability of this donkey-type reading. While such paraphrase relations between conditionals and correlatives might be suggestive of full equivalence, there are semantic differences between the two types of construction. An obvious one is that not all correlatives can receive a conditional-type free choice interpretation, but rather a definite interpretation denoting a single unique individual. A second difference concerns symmetric versus asymmetric readings in relation to the proportion problem of donkey sentences (Kadmon 1987). The problem



The landscape of correlatives 

concerns the anchoring possibilities of an adverb of quantification – whether it is anchored to one or all of the indefinites in a given sentence: (69) If a farmer owns a donkey, he is usually rich. The symmetric reading of (69) is one in which the adverb usually is anchored to both a farmer and a donkey. In this reading the sentence says that in most cases involving a farmer-donkey pair, the farmer is rich. In an asymmetric reading, the adverb is anchored either to a farmer only or to a donkey only, and not to farmer-donkey pairs. Now, Hindi correlative clauses (Cheng and Huang 1996, referring to Utpal Lahiri p.c.), can only have asymmetric readings. Conditionals on the other hand allow for a symmetric reading. This distinction argues for keeping the two types of constructions separate, as done for example in Cheng and Huang (1996), who demarcate Chinese ‘bare conditionals’, a Chinese construction whose semantics is parallel to (69), from Hindi-type correlative constructions on the basis of the presence vs. absence of the symmetric readings.

7. The contents of this volume The articles in this volume make theoretical and empirical advances in the study of correlatives in the fields of syntax and semantics. All articles use the toolbox of generative theoretical syntax (in particular, the Government and Binding and Minimalist frameworks) in the study of the phenomenon, and many of them build on various existing theories of relativization or structure building. The empirical contribution of the articles is evident when we consider that the articles provide in-depth studies of particular languages, some of which have not been studied extensively in connection with correlatives. The languages covered are: Basque (Rebuschi), Dutch (Den Dikken), Hindi (Bhatt and Lipták, Davison), Hungarian (Bhatt and Lipták), Polish (Citko), Sanskrit (Davison), Serbian (Arsenijevic), Tibetan (Cable), Italian Sign Language (Branchini and Donati). With this array of language-particular studies, the present collection of essays drastically extends the scope of previous research whose focus fell exclusively on Indo-Aryan languages. The articles are grouped in three sections according to the topic they place center-stage: the relation between correlatives and related constructions (Part I); the syntactic derivation of correlatives (Part II) and the explanation behind the matching effect (Part III). The first group of articles (Part I) examines syntactic properties of correlatives in comparison with other constructions that are syntactically or functionally similar to correlatives: wh-questions, free relative clauses, conditionals and leftward extraposed headed relative clauses. The central question in these articles is to what extent correlatives share a common syntax with these constructions, and where the differences, when present, stem from.



Anikó Lipták

The second group of articles (Part II) centers around the question how correlative constructions are derived in the syntax. The topics handled here concern the internal build-up of the correlative clause; its placement in the sentence; its relation with the correlate phrase as well as extraction possibilities out of correlative constructions. The third group of articles (Part III) centers around a very specific property of correlatives: the matching requirement, which obtains between the relative phrase in the correlative clause and the correlate phrase in the main clause. One kind of matching discussed here is matching in number, that is, that there is always the same number of relative phrases as correlates. Another kind of matching that can be observed is a rather particular type of matching: case matching. This obtains between the relative phrase and the main clause correlate in correlative constructions expressing temporal relations and, to a lesser degree, those expressing locative relations. In what follows, the reader finds a summary of the contents of the volume. Part I. Correlatives and related constructions The book’s opening article, Barbara Citko’s What don’t wh-questions, free relatives and correlatives have in common? discusses the internal syntax of Polish correlatives (both standard and comparative correlatives) in comparison to two other wh-constructions: questions and free relatives. Similarities between these three constructions immediately meet the eye, as they all involve wh-fronting and by and large use the same range of wh-phrases. Differences, however, show up when we look at other properties, like the availability of multiple wh-movement, the possibility of left branch extraction and the presence of reconstruction effects. Concerning these areas, correlative clauses pattern with questions and not with free relatives. Both correlatives and questions allow multiple occurrences of wh-phrases and superiority violations among these. They also allow left branch extractions of the type where the fronted wh-phrase extracts from a left branch of a noun phrase, stranding the NP behind. Reconstruction effects are also exhibited by both correlatives and questions. The fronted wh-phrase reconstructs into its original position, according to the evidence of anaphor binding, BP-C effects or variable binding. Free relatives on the other hand do not allow multiple wh-movement and left branch extractions and they show anti-reconstruction effects in the same contexts: they allow both reconstructed and non-reconstructed interpretations. The paper shows that in these properties free relatives are exactly parallel to headed relative clauses, due to the fact that the two share a common underlying structure. In both cases the clause is dominated by a DP layer. This layer is filled with a DP that is generated externally in the case of headed relatives, and with a wh-DP that is moved into this position in the case of free relatives. Correlatives and questions on the other hand are not headed by a DP-layer, they are headless CP structures, containing the fronted wh-phrase in Sp,CP. As for the specifics of the derivation of headed relatives, the author argues for a matching analysis (also proposed in Citko 2001), in which the head DP is generated



The landscape of correlatives

outside the relative clause. In this analysis, the wh-phrase, promoted from inside the clause, shows up without its NP subpart as a result of ellipsis applying under identity with the head NP. Free relatives are claimed to have a similar structure except that their external head is derived from the fronted wh-phrase that raises out of the CP into the dominating DP layer. This last step in the derivation explains why BP-C effects are missing in the case of free relatives: the wh-phrase in the DP cannot reconstruct back into the clause. The headed nature of free relatives also explains why they do not allow multiple wh-phrases. Since the clause cannot be headed by more than one head, this restricts the number of wh-phrases to one. The ban on left branch extraction arguably follows from the fact that the wh-phrase cannot strand its NP as the relative clause needs to have the nominal phrase in the head position. While the comparison in Citko’s article is prompted by syntactic similarities between questions, free relatives and correlatives, Georges Rebuschi’s article, Basque correlatives and their kin in the history of Northern Basque, examines both syntactic and semantic “relatives” of correlatives. This article systematically compares Northern Basque correlatives to various types of free and headed relatives, wh-interrogatives and ordinary as well as so-called “no matter” conditionals. Besides documenting these in 400 years of language history, mostly reflected in a variety of Bible translations, the article sets out to see whether correlative clauses in Basque constitute a special type of sentences, or whether they can be subsumed under other types. The findings of the paper support a view that considers correlatives an independent construction, but one that shares many properties with other types. Correlative clauses in Basque are wh-clauses introduced by a special complementizer bait­-, which always appears enclitic on the sentence-final auxiliary. Correlatives always occur in the left periphery, linked to the main clause by a demonstrative pronominal, which is often preceded by a linker element, homophonous with the conjunction eta ‘and’. Semantically, correlatives are very close to another type of free relative in Basque, the so-called ‘semi-free relative’, which can also occur in the left periphery, but which is never connected to the main clause by means of an eta linker. Semi-free relatives are moreover different from correlatives in their internal syntax as well: semi-free relatives contain no wh-operator and exhibit a different kind of complementizer, -en. Last but not least, while correlatives are externally CPs, semi-free relatives are clearly DPs according to the evidence of case-marking differences. Correlatives also share some, though not all, syntactic properties with headed whrelatives (referred to as ‘appositive’ relatives in the paper) and interrogative wh-clauses. The C-domain of correlatives is similar to headed wh-relatives to the extent that it contains the same complementizer, but differs in the range of wh-phrases allowed: headed relatives, unlike correlatives, do not allow for the whole range of interrogative wh-expressions. Embedded interrogatives in this respect square better with correlatives, allowing for the same set of wh-elements, yet syntactically, the two have a different clause structure: while in interrogatives the wh-phrases are necessarily adjacent to





Anikó Lipták

the finite verb form, in correlatives they are not. So-called indefinite free relatives (clauses of the type C’e [chi dice sempre di si] there-is who says always of yes ‘There is always someone who says yes’ in Italian) do not share their internal or external syntax fully with correlatives, either, in Basque. This can be seen from the fact that unlike correlatives, they are typically postverbal and may contain no finite morphology or complementizers. Apart from the above wh-constructions and relative constructions special attention is dedicated to the similarities between conditionals (both ordinary and “no matter” types) and correlatives. Such a comparison is interesting in the light of the literature alluded to in Section 6.4 above, which argues that conditionals are a subkind of correlatives. In Basque, one finds both morphological and semantic evidence for such a view. The semantic equivalence between the two constructions is shown by the fact that in past centuries Latin conditionals have often been translated as correlatives, and correlatives as conditionals. Morphologically, the link can be established between the marking of the dependent clause. While conditionals use ba- ‘if ’, and correlatives use the finite complementizer ‘bait-’, there are quite many interchanges between the two forms. At different stages of the language bait- could replace ba- and ba- could replace bait-. Looking at morphological parallels of this type, the paper considers the validity of theories that derive the two forms from the same underlying source. It then argues, albeit tentatively, that the similarity between correlatives and conditionals is prompted by an overall semantic parallelism rather than by an underlying syntax. The kind of semantic parallelism referred to in Rebuschi’s article is the driving force behind Boban Arsenijević’ paper, {Relative {conditional {correlative clauses}}}, which is an investigation into the relationship between correlatives and conditionals. The account found in this paper is partly similar to those that analyze conditionals as correlatives (Bhatt and Pancheva 2006), but goes one step further in that it completely assimilates the two clause types into one category. Conditionals are analyzed as yes-no relative clauses, restrictive clauses in which the truth value of a proposition is restricted. The proposition represented by the conditional clause restricts the set of worlds compatible with the proposition represented by the head clause. Syntactically, the locus of modification is a functional projection called WorldP, the projection that specifies the truth value of clauses by containing the feature world with a value [actual] or [possible]. WorldP is often lexicalized in languages as forms of then when the clause hosts a conditional. Once this analysis of conditionals is in place, the paper goes on to show, mostly on the basis of Serbo-Croatian correlatives, that correlatives are a subtype of conditionals as defined above. They are clauses that modify the actual/possible world content of the main clause, just like ordinary conditionals. The syntactic difference of course is that correlatives contain a wh-item, which this analysis treats not as a relativizer, but as an extreme non-specific item (in the sense of Farkas 2002), similar in meaning to free



The landscape of correlatives 

choice ‘any’ phrases. The fact that it can occur together with a conditional marker supports the analysis of correlatives as conditionals: (70) Ko na brdo ak’ i malo stoji, više vidi od onog pod brdo. who on hil if and little stands more see from that under hill ‘The one who stands on a hill even a little bit sees more than the one at the foot of the hill.’/‘Anyone who stands on a hill even a little bit sees more than the one at the foot of the hill.’ The wh-non-specific expression appears at the beginning of the clause due to its topic function. On its way to the CP, it lands in the WorldP projection of the conditional, to establish a dependency between the interpretation of the wh-expression and that of its clause. This results in the fact that the worlds denoted by the clause vary with the referents of the wh-expression. The correlative demonstrative in this analysis is a purely anaphoric item that entertains only a semantic, but no syntactic relation with the correlative clause. The article Relatively different: Italian Sign Language relative clauses in a typological perspective by Chiara Branchini and Caterina Donati is the only article in this collection that does not argue for but against an analysis in terms of correlativization. The object of study in this article is relative clauses found in Italian Sign Language (LIS), the language of the Italian Deaf community. The authors explicitly argue against an analysis of these constructions in terms of correlativization, as proposed in Cecchetto, Geraci and Zucchi (2006). Italian Sign Language is a head-final SOV language. It employs a strategy for relativization involving a bi-clausal construction. Of the two clauses, the sentence initial clause is marked by two means. It is marked by a special sign, referred to as PE, which is coreferential with an NP inside the clause, realized through agreement in space. In addition, the relative clause is also marked by non-manual marking consisting of raised eyebrows, and tension of eyes and upper cheeks extending over all or a part of the relative clause. The main clause to which this relative clause attaches contains a pronominal correlate or a gap, interpreted as coreferential with the NP the PE sign refers to. Branchini and Donati show that the proper analysis of this construction treats the relative clause as extraposed to the left of the clause from the position of the correlate. The relative clause itself has a restrictive meaning and can be considered to be an internally headed relative clause, similar to those in Japanese in the analysis of Shimoyama (1999). A correlative analysis of the facts, such as the one put forward by Cecchetto, Geraci and Zucchi (2006), is thus misguided. This analysis has it that the relative clause is adjoined to the main clause, which contains an E-type pronominal correlate (never a full NP), possibly pro-dropped. PE inside the relative clause is a demonstrative moved to Sp,CP, whose role is to turn the relative clause into a generalized quantifier, similarly to the analysis of Hindi correlatives in Dayal (1996). The relative clause in this account is furthermore claimed to be appositive in meaning.



Anikó Lipták

Branchini and Donati show that this analysis cannot be right for various reasons. To start with the last mentioned claim, the relative clause has restrictive semantics according to a battery of tests. Second, the relative clause is not a CP category, but a nominal one, evidenced by the fact that it can be modified by nominal modifiers like first, which in LIS resist an adverbial analysis. Nominalization is taken to be the result of the PE element, which acts as a nominalizer in other contexts as well. PE is thus analyzed as a D head, and is generated next to the NP inside the relative, where it can surface as well, at least for some signers. For the majority of signers though it moves from inside the relative clause to the head position of the clause, to create a dependency between two nominal positions, one internal and one external to the clause. This movement nominalizes the entire clause. Concerning the nature of the alleged correlate phrase in the main clause, it is argued that this element cannot be a dropped pronoun, as it can be covert in cases where pro-drop is impossible, for example with oblique NPs. For these reasons, the correlate should better be analyzed as a trace, which can be either phonetically empty or spelled out as a resumptive pronoun. This trace is a trace of the relative clause that moved out of the main clause via extraposition to the left, evidenced by island sensitivity: the relative clause cannot escape from out of an island. The extraposed nature of the clause is further evidenced by the fact that the non-manual marking for relative clauses contains at least two components, raised eyebrows and tensed eyes, the latter of which appears to single out extraposed constituents in other contexts, too. Part II. The derivation of correlatives The second part of this book contains three language-specific case studies of correlatives, centering around the question how correlative constructions are derived. This part commences with a paper The syntax of the Tibetan Correlative by Seth Cable, which explores the properties and structure of Tibetan Correlative constructions, as found in the Lhasa dialect of Central Tibet. Tibetan is a rigidly verb final language with a correlativization strategy. This strategy involves a left peripheral finite relative clause that involves a wh- or relative operator, interpretively linked to a demonstrative correlate in sentence-internal position. It is argued that such constructions are correlatives, and yield meanings similar to that of Hindi correlatives and English free relatives: either a universal reading or a definite one (see (59) and (60) above). An interesting property of the correlative clause in Tibetan is that it features an element that also functions as the conditional marker elsewhere, yet correlatives are not conditionals, for lack of full semantic parallels. The bulk of the paper concerns itself with the structural analysis of single Tibetan correlatives. Following Srivastav (1991) and Bhatt (2003), the author considers the availability of three strategies for the derivation of the Tibetan correlative: (i) low attachment of the correlative CP to the correlative DP; (ii) low attachment of the correlative clause to the correlative DP followed by movement to IP-adjoined position;



The landscape of correlatives 

and (iii) high attachment to IP via base-generation. Cable shows that all three strategies can manifest themselves in the formation of Tibetan correlatives. In some cases, the correlative is formed by low attachment, evidenced by the availability of coordination structures of the type where two correlative-correlate pairs are coordinated and form a constituent. IP-adjunction via movement (option ii) is evidenced by binding. Quantifiers that occur lower than the correlative clause can bind a variable inside this clause, suggesting that the correlative has undergone movement to its surface position. IP-adjunction via base-generation (option iii) seems to exist as well, as the relation between the correlative clause and the correlative DP is not island sensitive (the correlative DP can be found inside relative clauses of all types, with the correlative clause outside these). It is also possible to co-index an R-expression inside the correlative clause with a DP that occurs lower than the clause. The paper closes with some speculations at to why Tibetan can employ all three possibilities side by side, while Hindi, as shown by Bhatt (2003), only allows for option (i) and (ii), at least in single relatives (multiple relatives use option (iii)). According to Cable, the answer to this cross-linguistic difference might lie in the distinct agreement properties of the two languages. While in Hindi, the correlative operator in the correlative CP agrees in phi-features with the correlative DP, in Tibetan it does not. According to Cable, only languages where there is no such phi-feature agreement allow for variable attachment sites for the correlative, both a high one and a low one, without any preference for either. While Cable’s work uncovers derivational possibilities within one language, Alice Davison’s article Adjunction, features and locality in Sanskrit and Hindi/Urdu correlatives aims to explain variation among two closely related languages: Sanskrit and Hindi/Urdu. Sanskrit being the predecessor of Hindi/Urdu, correlatives share quite many properties in the two languages, but differences are also attested. The basic tenet of the article is that the differences can be traced back to the general property of Sanskrit that it avoids embedding and uses juxtaposition instead. In Sanskrit, clauses are linked in a very loose paratactic way, without syntactic encoding of subordination. This results in the fact that correlatives are never found adjoined to a nominal phrase, a strategy allowed in Hindi/Urdu. Instead, Sanskrit correlative clauses can only adjoin at the clausal level. Clausal adjunction furthermore is also different in the two languages. In Sanskrit we find symmetric adjunction, while in Hindi/Urdu we find asymmetric adjunction. Asymmetric adjunction means subordination as we know it: the subordinate correlative clause (a CP) does not project in category when it adjoins to the host clause (a TP). Symmetric adjunction on the other hand means that either clauses can project. According to the paper, this happens because the adjunction site is different in this case: the correlative (a CP) adjoins at the CP level. The two types of adjunction bring about a difference in c-command possibilities: in symmetric adjunction, but not in asymmetric adjunction, both CPs c-command the constituents of the other clause. Evidence for the symmetric adjunction analysis of Sanskrit correlatives comes from various considerations. One involves the occurrence of sentence-oriented



Anikó Lipták

particles like indeed, surely, furthermore, which are characteristic of independent clauses. These particles occur in the clause initial string of clauses and fill head positions in the CP domain.22 Interestingly, in the realm of correlative constructions, such particles can be found both in the correlative clause and the main clause. This evidences that the correlative clause is a CP itself, and it adjoins at the level of the CP. If Sanskrit would use asymmetric adjunction just like Hindi and adjoin the correlative CP to the main clause TP, the relative clause would be preceded by such particles of the clause-initial string. But sentences of this kind are impossible in Sanskrit. The paper puts forward the claim that a well-formedness condition requiring that the correlative clause c-command the correlate, assumed to be operative in Hindi/ Urdu but inoperative in Sanskrit, can furthermore explain differences in stacking. Hindi/Urdu does not allow correlatives to stack, Sanskrit on the other hand does. Davison takes stacking to be a configuration where two relatives adjoin to each other before adjoining to the main clause, a configuration in which the first relative does not c-command into the main clause. It is shown that this stacking configuration is only allowed in languages where the correlative does not need to c-command its correlate phrase, as is the case in Sanskrit. The same condition also rules out iterated relative clauses of the sort where one correlative clause finds its correlate in a non-adjacent clause. Such combinations are ungrammatical in Hindi/Urdu but good in Sanskrit, due to the fact that in Sanskrit correlatives need not c-command the correlate. The last paper in Part II, Marcel den Dikken’s article Comparative correlatives and successive cyclicity focuses on a special and more widely occurring type of correlatives: comparative correlatives (see Section 4 and 5.3 above). Within the realm of such constructions, the paper deals with one particular aspect of comparative correlatives in Dutch: locality. The author investigates to what extent comparative conditional phrases can undergo long-distance movement and how their behavior with respect to locality reflects on the nature of the comparative correlative construction. The paper starts by painting the empirical and theoretical lie of the land of Dutch comparative correlatives. Partly building on earlier work, Den Dikken argues that the head clause in Dutch comparative correlatives can be an IP or a CP, the difference depending on its root vs. embedded nature. Corresponding to this categorial difference, the position of the correlative clause and that of the comparative correlate in the main clause varies as well: the correlative clause attaches either to the main clause IP (in embedded contexts) or to the main clause CP (in root ones); the correlate phrase on the other hand either undergoes movement to Sp,CP or adjunction to IP. These ingredients of the analysis, coupled with the requirement that the fronted correlative particle and the sentence-initial relative clause must always be adjacent, give us all observed word order possibilities within the main clause: V2 effects when the correlate adjoins to IP, and lack of the possibility of V2 when the correlate is in Sp,CP. The same ingredients of the analysis are also instrumental to the analysis of locality effects. In the latter domain, the two clauses of the Dutch comparative correlative



The landscape of correlatives 

construction behave differently when it comes to long-distance movement of the comparative degree phrase: the relative clause freely allows for movement of the degree phrase, while extraction from the main clause results in degradation for many speakers. The judgments are sensitive to various factors, including the choice of the correlative particle used ([+wh] or [–wh] phrases), the finite/non-finite distinction in the clause, and word order. The article shows that the observed patterns neatly fall into place once it is assumed that whenever the compara­tive lands in SpecCP (as it does in the relative clause), long-distance depen­dencies are gram­matical, due to successivecyclic movement through intermediate SpecCP positions. When the comparative lands in an IP–adjoined position (as it often does in the headclause), severe locality restrictions emerge. These results are entirely predictable from the Principle of Unambiguous Binding (Müller & Sternefeld (1993)), a principle that regulates movement dependences by ruling out variables that are both bound from an A-bar specifier and an A-bar adjunction position. Part III. The matching effect On the matching requirement in correlatives Tommi Tsz-Cheung Leung sets out to discuss the requirement that regulates the distribution of relative phrases and demonstrative phrases in the correlative and the main clause respectively. In most cases, correlatives exhibit perfect matching between these items in both single and multiple cases. On the one hand, we find exactly as many relative phrases as demonstrative ones, on the other, the two share syntactic features such as category features, number and gender features. Existing accounts of correlatives have difficulties in accounting for matching in the case of multiple correlatives, since in this case the correlative clause cannot entertain a local relation with both demonstrative phrases at the same time, a local relation that would be necessary for the sharing of features. The new proposal put forward in this paper postulates that the required local relationship is possible nevertheless but crucially does not obtain between the correlative clause and the demonstrative, rather, it obtains between the relative phrase itself and the demonstrative expression. In this account, the local relationship is established at the base, where the relative XP and the demonstrative XP form what is called a “doubling constituent” in which the two parts are in a “contextual relation” in the sense of Vergnaud (2003). Contextual relatedness roughly corresponds here to the idea that at some derivational stage the elements should be grouped together. Starting from an underlying doubling constituent [Relative phrase-Demonstrative correlate], the surface word order observed in correlatives is derived via movement. The relative phrase moves out of the doubling constituent leaving the demonstrative behind. This step is similar to movement accounts of floating quantification (as in Sportiche 1988), resumption (as in Boeckx 2003) or antecedent-pronoun relations (as in Kayne 2002). The difference from these accounts is that the movement of the relative phrase needs to proceed via special means, that of sideway movement (in the sense of Nunes



Anikó Lipták

2004) since the target position of movement is inside the correlative clause, subordinated to the main clause. An elegant feature of this account is that it can handle single and multiple correlatives in the same fashion. In the case of multiple relatives, the derivation starts out from multiple occurrences of Relative phrase-Demonstrative correlate doubles and involves the application of multiple sideway movements. Having seen properties and possible explanations of the general matching effect in correlatives, the article by Rajesh Bhatt and Anikó Lipták, Matching effects in the temporal and locative domains turns to a particular – and yet undocumented – kind of matching observed in correlative constructions: case matching. This kind of matching, which is studied in Hindi and in Hungarian, does not apply to ordinary correlatives that denote individuals, but shows up in correlatives that abstract over times and locations. In such correlatives, both single and multiple, the case on the relative pronoun has to be identical to the case on the correlate expression in the main clause. To illustrate, take the Hungarian example in (71): if the temporal connective till is present on the relative pronoun, it has to be present on the correlate phrase, too. (71) Ameddig János alszik, {addig /* akkor} Mari hazajön. when-till János sleeps then-till then Mari comes.home ‘Till the time John sleeps, {till then Mari comes home / *at that time Mari comes home}.’ Due to this matching requirement, the combinations when− then; till when − till then; since when − since then are well-formed. Non-matching combinations are ungrammatical, with the exception of when− till then, when− since then, which are possible in Hindi but not in Hungarian. In the realm of locative correlatives, facts are somewhat similar in Hungarian: case matching is obligatory, although violations give rise to milder ungrammaticality. Hindi, on the other hand, shows no matching effects whatsoever. The paper offers speculations on the explanation behind this peculiar matching requirement. It shows that this phenomenon is not morphologically conditioned like in the case of ordinary free relatives, where matching is a way of solving the morphological conflict that arises because the relative pronoun has to carry both the case assigned to it inside the relative clause (internal case) and the case assigned to the free relative as a whole (external case). Since in the case of correlatives internal and external case are carried by two distinct items, the relative phrase and the correlate phrase respectively, matching is never predicted to occur. The observed pattern in the realm of temporal/locative correlatives is more likely to have a semantic explanation, to be found in the fact that when-clauses denote a different type of temporal entity than till/ since when-clauses. While the latter denotes an interval, the former denotes a point of time. This constrains their possible combinations with time denoting then vs. interval denoting since/till then phrases.



The landscape of correlatives 

References Abeillé, Anne & Robert D. Borsley. 2008. Comparative correlatives and parameters. Lingua 118. 1139–1157. Anderson, Corinna. 2007a. A non-constituent analysis of Nepali correlatives. Paper presented at LSA, Anaheim, California. Anderson, Corinna. 2007b. Discourse specialization in correlatives: a South Asian perspective. Paper presented at Interdisciplinary Approaches to Relative Clauses, Cambridge University. Andrews, Avery. 1985. Studies in the syntax of relative and comparative Clauses. Outstanding dissertations in linguistics. New York: Garland. Arsenijević, Boban. This volume. {Relative {conditional {correlative clauses}}}. Asher, Ronald E. 1982. Tamil. Descriptive Grammars. London: Routledge. Asher, Ronald E. & T. C. Kumari. 1997. Malayalam. Descriptive Grammars. London: Routledge. Bach, Emmon & Cooper, Robin H. 1978. The NP-S analysis of relative clauses and compositional semantics. Linguistics and Philosophy 2. 145–50. Bagchi, Tista. 1994. Bangla correlative pronouns, relative clause order and D-linking. In Miriam Butt, Tracy Holloway King & Gillian Ramchand (eds), Theoretical perspectives on word order in South Asian languages, 13–30. Stanford, CA: CSLI. Bains, Gurpreet. 1989. Complex structures in Hindi-Urdu: Explorations in government and binding theory. New York University, NY: PhD dissertation. Beck, Sigrid. 1997. On the semantics of correlative conditionals. Linguistics and Philosophy 20. 229–271. Berger, Hermann. 1998. Die Burushaski-Sprache von Hunza und Nager, 3 volumes. Wiesbaden: Harrassowitz. Berman, Howard. 1972. Relative clauses in Hittite. In Paul M. Peranteau, Judith N. Levi & Gloria C. Phares (eds), The Chicago which hunt: Papers from the relative clause festival, 1–8. Chicago: Chicago Linguistics Society. Berman, Judith, Stefanie Dipper, Christian Fortmann & Jonas Kuhn. 1998. Argument clauses and correlative es in German – deriving discourse properties in a unification analysis. In Miriam Butt & Tracy Holloway King (eds), Proceedings of the LFG98 conference. University of Queensland, Brisbane/Stanford: CSLI publications. Berntsen, Maxine & Jai Nimbkar. 1975. A Marathi reference grammar. South Asia Regional Studies. University of Pennsylvania. Bhatia, Tej K. 1993. Punjabi: a cognitive-descriptive grammar. Descriptive Grammars. London: Routledge. Bhatt, Rajesh. 2003. Locality in correlatives. Natural Language and Linguistic Theory 21. 485–541. Bhatt, Rajesh. 2009. Comments on “Comparative Correlatives in Greek: The Syntax of oso”. In Claire Halpert, Jeremy Hartman & David Hill (eds), Proceedings of the 2007 Workshop in Greek Syntax and Semantics at MIT, MITWPL 57. 107–122. Boston: MIT Press. Bhatt, Rajesh & Roumyana Pancheva. 2006. Conditionals. In Martin Everaert, Henk van Riemsdijk, Rob Goedemans & Bart Hollebrandse. The Blackwell Companion to Syntax I, 638–687. Oxford: Blackwell. Bhatt, Rajesh & Anikó Lipták. This volume. Matching effects in the temporal and locative domains. Bianchi, Valentina. 1999. Consequences of antisymmetry: Headed relative clauses. Berlin: Mouton de Gruyter.



Anikó Lipták Bianchi, Valentina. 2000. Some issues in the syntax of relative determiners. In Artemis Alexiadou, Paul Law, André Meinunger & Chris Wilder (eds), The Syntax of relative clauses, 53– 81. Amsterdam: John Benjamins. Bittner, Maria. 2001. Topical referents for individuals and possibilities. In Rachel Hastings, Brendan Jackson & Zsófia Zvolenszky (eds.), Proceedings of SALT XI, 36–55. CLC, Ithaca. Boeckx, Cedric. 2003. Islands and Chains. Amsterdam: John Benjamins. Borsley, Robert. 2003. On the Polish periphery: Comparative correlatives in Polish. Manuscript. University of Essex. Cable, Seth. This volume. The syntax of the Tibetan correlative. Cardona, George. 1965. A Gujarati reference grammar. Philadelphia: The University of Pennsylvania Press. Cecchetto, Carlo, Carlo Geraci & Sandro Zucchi. 2006. Strategies of relativization in Italian Sign Language. Natural Language and Linguistic Theory 24. 945–975. Cheng, Lisa L-S. & C-T. Jim Huang. 1996. Two types of donkey sentences. Natural Language Semantics 4. 121–163. Chung, Daeho. 2004. Semantics and syntax of correlative adverbs. Studies in Generative Grammar 14. 307–328. Citko, Barbara. 2001. An argument in favor of the deletion under identity account of relative clauses. In Min-Joo Kim & Uri Strauss (eds), Proceedings of the 31st North East Linguistics Society, 131–145. Amherst, MA: GLSA Publications. Citko, Barbara. This volume. What don’t wh-questions, free relatives and correlatives have in common? Culicover, Peter & Ray Jackendoff. 1999. The view from the periphery: The English comparative correlative. Linguistic Inquiry 30. 543–71. Culy, Christopher. 1990. The syntax and semantics of internally headed relative clauses. Stanford, CA: Stanford University dissertation. Curme, George O. 1912. A history of the English relative construction. The Journal of English and Germanic Philology 11. 10–29, 180–204, 355–380. Dasgupta, Probal. 1980. Questions and relative and complement clauses in a Bangla grammar. New York, NY: New York University dissertation. Davison, Alice. This volume. Adjunction, features and locality in Sanskrit and Hindi/Urdu correlatives. Dayal, Veneeta. 1995. Quantification in correlatives. In Emmon Bach, Eloise Jelinek, Angelika Kratzer & Barbara H. Partee (eds), Quantification in natural languages. Studies in Linguistics and Philosophy 54, 179–205. Dordrecht: Kluwer Academic Publishers. Dayal, Veneeta. 1996. Locality in wh-quantification: Questions and relative clauses in Hindi. Studies in Linguistics and Philosophy 62. Dordrecht: Kluwer Academic Publishers. Dikken, Marcel den. 2005. Comparative correlatives comparatively. Linguistic Inquiry 36. 497–532. Dikken, Marcel den. This volume. Comparative correlatives and successive cyclicity. Donaldson, Susan K. 1971. Movement in restrictive relative clauses in Hindi. Studies in the Linguistic Sciences 1. 1–74. Downing, Bruce. 1973. Correlative relative clauses in universal grammar. Minnesota Working Papers in Linguistics and Philosophy of Language 2, 1–17. Minneapolis, Minnesota: University of Minnesota. Dwivedi, Veena Dhar. 1994. Topicalization in Hindi and the correlative construction. In Miriam Butt, Tracy Holloway H. King & Gilian Ramchand (eds), Theoretical perspectives on word order in South Asian languages, 91–117. CSLI Publications.



The landscape of correlatives  Evans, Gareth. 1980. Pronouns. Linguistic Inquiry 11. 337–362. Farkas, Donka. F. 2002. Extreme non-specificity in Romanian. In Claire Beyssade, Reineke BokBennema, Frank Drijkoningen & Paola Monachesi (eds.), Romance languages and linguistic theory 2000. 127–153. Amsterdam/Philadelphia: John Benjamins. Fintel, Kai von. 1994. Restrictions on quantifier domains. Amherst, MA: University of Amherst, Massachusetts dissertation. Gajewski, Jon. 2008. On the semantics of Hindi-Urdu multiple correlatives. Linguistic Inquiry 39. 327–334. Garrett, Andrew. 1994. Relative clause syntax in Lycian and Hittite. Die Sprache 36. 29–69. Geis, Michael L. 1985. The syntax of conditional sentences. In Michael L. Geis (ed.), Studies in Generalized Phrase Structure Grammar. 130–159. Columbus, OH: Department of Linguistics, OSU. Gildersleeve, Basil L. & Gonzalez Lodge. 1974. Latin grammar. Macmillan, London. Grierson, George A. 1883. Seven grammars of the dialects and subdialects of the Bihari language. Delhi: Bharatiya Publishing House. Groenendijk, Jeroen & Martin Stokhof. 1982. Semantic analysis of wh-complements. Linguistics and Philosophy 5. 175–223. Grosu, Alexander. 2000. Type resolution in relative constructions. Feature marking and dependency encoding. In Artemis Alexiadou, Paul Law, André Meinunger & Chris Wilder (eds), The syntax of relative clauses. 83–120. Amsterdam and Philadelphia, John Benjamins. Grosu, Alexander. 2002. Strange relatives at the interface of two millennia. Glot International 6. 145–167. Grosu, Alexander & Fred Landman. 1998. Strange relatives of the third kind. Natural Language Semantics 6. 125–170. Hale, Kenneth L. 1976. The adjoined relative clause in Australian. In Robert M. W. Dixon (ed.), Grammatical categories in Australian languages. 78–105. Canberra: Australian Institute of Aboriginial Studies. Haudry, Jean. 1973. Parataxe, hypotaxe et corrélation dans la phrase latine. Bulletin de la Société Linguistique de Paris 68. 147–186. Heim, Irene. 1983. On the projection problem for presuppositions. In Michael Barlow, Daniel Flickinger, & Michael Wescoat (eds), West Coast Conference on Formal Linguistics 2. 114– 126. Stanford, CA: Stanford University, Stanford Linguistics Association. Izvorski, Roumyana. 1996. The syntax and semantics of correlative proforms. In Kiyomi Kusumoto (ed.), Proceedings of NELS 26, 133–147. Amherst, MA: GLSA. Jacobson, Pauline. 1995. On the quantificational force of English free relatives. In Emmon Bach, Eloise Jelinek, Angelika Kratzer & Barbara H. Partee (eds.), Quantification in natural language. Studies in Linguistics and Philosophy 54. 451–486. Dordrecht: Kluwer Academic Publishers. Johannessen, Bondi J. 2005. The syntax of correlative adverbs. Lingua 115, 4. 419–433 Junghare, Indira 1973. Restrictive relative clauses in Marathi. Indian Linguistics 34. 251–262. Kachru, Yamuna. 1973. Some aspects of pronominalization and relative clause construction in Hindi-Urdu. In Braj B. Kachru (ed.), Papers on South Asian Linguistics. Studies in the Linguistic Sciences 3.2, 87–103. Urbana, Illinois: University of Illinois. Kadmon, Nirit. 1987. On unique and non-unique reference and asymmetric quantification. New York: Garland. Kayne, Richard S. 2002. Pronouns and their antecedents. In Samuel D. Epstein & Daniel T. Seely (eds), Derivations and explanation in the minimalist program, 133–183. Oxford: Blackwell. Kayne, Richard S. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press.



Anikó Lipták Keenan, Edward. 1985. Relative clauses. In Timothy Shopen (ed.), Language typology and syntactic description II, 141–170. Cambridge, England: Cambridge University Press. König, Ekkehard & Johan van der Auwera. 1988. Clause integration in German and Dutch conditionals, concessive conditionals and concessives. In John Haiman and Sandra A. Thompson (eds), Clause combining in grammar and discourse, 101–129. Amsterdam: John Benjamins. Krishnamurti, Bhadriraju & J.P.L. Gwynn. 1986. A grammar of modern Telugu. Oxford: Oxford University Press. Lafon, René. 1966. La particule bait en basque: ses emplois morphologiques et syntaxiques. Bulletin de la Société de linguistique 61.1. Reprinted in Jean Haritschelhar and Zabaltzagarai Charritton (eds). 1999. Vasconiana. 667–693. Bilbao: Euskaltzaindia. Lakshmi Bai, B. 1985. Some notes on correlative constructions in Dravidian. In Veneeta Z. Acson & Richard L. Leed (eds), For Gordon H. Fairbanks, 181–90. Honolulu: University of Hawaii Press. Lambert, Hester M. 1971. A Gujarati language course. Cambridge, England: Cambridge University Press. Lehmann, Christian. 1984. Der Relativsatz. Tübingen: Gunter Narr Verlag. Leung, Tommi Tsz-Cheung. This volume. On the matching requirement in correlatives. Lipták, Anikó. 2005. Correlative topicalization. Manuscript. Leiden University. Lipták, Anikó. 2008. On the correlative nature of Hungarian left peripheral relatives. In Benjamin Shaer, Philippa Cook, Werner Frey & Claudia Maienborn (eds), Dislocation: Syntactic, semantic and discourse perspectives, 398–430. London: Routledge. Lipták, Anikó & Georges Rebuschi. To appear. Types and distribution of Northern Basque correlatives. In Ch. Borgoyari, N.Ch.-D. Khuong and R.S. Sinha (eds), 5th Asian GLOW Proceedings. Mahajan, Anoop. 2000. Relative asymmetries and Hindi correlatives. In Artemis Alexiadou, André Meinunger, Chris Wilder & Paul Law (eds), The syntax of relative clauses, 201–229. Amsterdam: John Benjamins. Masica, Colin. 1991. The Indo-Aryan languages. Cambridge Language Surveys. Cambridge, England: Cambridge University Press. McCawley, James D. 1988. The comparative conditional construction in English, German, and Chinese. In Shelley Axmaker, Annie Jaisser, & Helen Singmaster (eds), Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society, 176–187. Berkeley: University of California, Berkeley Linguistics Society. McCawley, James D. 1992. Remarks on adsentential, adnominal, and extraposed relative clauses in Hindi. Presented at SALA, Stanford. Michaelis, Laura. 1994. A case of constructional polysemy in Latin. Studies in Language 18. 45–70. Müller, Gereon & Wolfgang Sternefeld. 1993. Improper movement and unambiguous binding. Linguistic Inquiry 24. 461–507. Nadkarni, M. V. 1970. Noun phrase embedding structures in Kannada and Konkani. Los Angeles, CA: University of California dissertation. Nunes, Jairo. 2004. Linearization of chains and sideward movement. Cambridge, MA: MIT Press. Pandharipande, Rajeshwari V. 1997. Marathi: a descriptive grammar. Descriptive Grammars. London: Routledge. Partee, Barbara. 1975. Montague grammar and transformational grammar. Linguistic Inquiry 6. 203–300. Prince, Ellen. 1998. On the limits of syntax, with reference to left dislocation and topicalization. Syntax and Semantics 29. 281–302. New York: Academic Press.



The landscape of correlatives  Probert, Philomen. 2006. Clause boundaries in Old Hittite relative sentences. Transactions of the Philological Society 104. 17–83. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik. 1972. A Grammar of contemporary English. London: Longman. Raman, Carol F. 1973. The Old Hittite relative construction. Austin, TX: University of Texas dissertation. Rebuschi, Georges. 2003. Towards a theoretical treatment of left-dislocated subordinate clauses. In Jesus M. Makazaga & Bernard Oyharçabal (eds), P. Lafitteren sortzearen mendemugako biltzarra, I: Gramatika gaiak, coll. IKER. 14. 1. 395–416. Bilbao: Euskaltzaindia. Rebuschi, Georges. This volume. Basque correlatives and their kin in the history of Northern Basque. Rullmann, Hotze. 1995. Maximality in the semantics of wh-constructions. Amherst, MA: University of Massachusetts dissertation. Sahoo, Kalyanamalini & Lars Hellan. 1998. Multiple correlativization in Oriya. Talk presented at ConSole98, University of Bergen, Norway. Schmidt, Ruth Laila. 1981. Dakkhini Urdu. New Delhi: Bahri Publications. Shimoyama, Junko. 1999. Internally headed relative clauses in Japanese and E-type Anaphora. Journal of East Asian Linguistics 8. 147–182. Shukla, Shaligram. 1981. Bhojpuri Grammar. Washington: Georgetown University Press. Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19. 425–449. Sridhar, Shikaripur N. 1990. Kannada. Descriptive Grammars. London: Routledge. Srivastav, Veneeta. 1991. The syntax and semantics of correlatives. Natural Language and Linguistic Theory 9. 637–686. Stechow, Arnim von. 1984. Comparing semantic theories of comparison, Journal of Semantics 1. 1–77. Subbarao, Karumuri. 1984. Complementation in Hindi syntax. Academic Publications, Delhi. Tiffou, Etienne & Richard R. Patry. 1995. La relative en bourouchaski du Yasin. Bulletin de la Société de Linguistique 90.1. 335–390. Trumpp, Ernest. 1872. Grammar of the Sindhi language compared with the Sankrit-Prakrit and the cognate Indian vernaculars. Osnabrück: Biblio Verlag. Vergnaud, Jean-Roger. 2003. On a certain notion of “occurrence”: the source of metrical structure, and of much more. In Stefan Ploch (ed.), Living on the edge, 599–632. Berlin: Mouton de Gruyter. Verma, Manindra K. 1966. The noun phrase in Hindi and English. New Delhi: Motilal. Vries, Mark de. 2002. The syntax of relativization. LOT dissertation series 53. Wali, Kashi. 1982. Marathi correlatives: a conspectus. In P. J. Mistry (ed.), South Asian Review: studies in South Asian languages and linguistics. 78–88. Jacksonville, Florida: South Asian Literary Association. Wali, Kashi & Omkar N. Koul. 1997. Kashmiri: a cognitive-descriptive grammar. Descriptive Grammars. London: Routledge. Weber, David J. 1983. Relativization and nominalized clauses in Huallaga (Huanuco) Quechua. Berkeley/Los Angeles/London: University of California Press. Yadav, Ramawatar. 1996. A reference grammar of Maithili. Trends in Linguistics: Documentation 11. Berlin: Mouton de Gruyter. Zribi-Hertz, Anne & Jean-François Hanne. 1995. Pronoms, déterminants et relatives en bambara de Bamako. Linguistique africaine 15. 91–135.



Anikó Lipták

Endnotes 1. Here and in the examples below, the kind of Hindi orthography is used that represents retroflexes by capitalization, nasal vowels by following the vowel by the capitalized nasal, and long vowels by the doubling of the vowel. The glosses used in the text are as follows: acc: accusative; aux: auxiliary; com: comitative; comp: complementizer; dat: dative; erg: ergative; gen: genitive; fut: future; hab: habitual; inf: infinitive; ins: instrumental; non.pst: non-past; part: partitive; pfv: perfective; pl: plural; pot: potential; prs: present; prog: progressive; prosp: prospective; pst: part; sg: singular; rel: relative morpheme, 1/2/3: person. 2. For an overview of the semantics of correlatives, which is different from that of headed clauses, too, see Sections 3 and 6 below. In these sections it will be shown that correlatives are neither restrictive nor appositive, rather they represent a third type, that of definite or “maximalizing” relatives. 3. The third option, where the nominal only occurs in the relative clause is possible in correlatives, too. Concerning the same in headed relatives, Srivastav (1991)/Dayal (1996) considers these ungrammatical, but Mahajan (2000) accepts them: (i) %vo [jo laRkii khaRii hai] lambii hai that rel girl standing is tall is ‘The girl who is standing is tall.’ 4. Mahajan (2000) holds a different view here. For him (and some Delhi speakers he consulted), (16b) and (18) are grammatical sentences. 5. According to Srivastav (1991), fn. 15, right-peripheral multiple relatives are fine for some speakers but need special intonation. Mahajan (2000) finds these completely grammatical even without special intonation. 6. In Lehmann’s (1984) functional classification of relative constructions there is a fifth type, extraposed relatives. Extraposed and correlatives form a natural group of “co-relatives”, which, unlike prenominal relatives, postnominal relatives and internally headed relatives are not subordinated to a nominal, but rather are satellites to another clause. 7. Dayal (1996) also recognizes that correlatives are neither restrictives (‘noun modifiers’) nor appositives, but instantiate a distinct type, relatives that are definite, displaying uniqueness/ maximality effects. She argues that correlatives share this property with free relatives and internally headed relatives. As Grosu and Landmann (1998) and Grosu (2002) have shown, internally headed relatives are not uniform in this respect: only the Quechua-type internally headed relative clause (also found in Japanese, Korean or Navajo) can be said to be maximalizing. 8. Maximalization does not only occur in relative clause constructions, but other types of constructions, too, like plural anaphora (Evans 1980, Kadmon 1987), questions (Groenendijk and Stokhof 1982, Rullmann 1995) and comparatives (von Stechow 1984). 9. See Section 5.2.2 for some cases where the correlative adjoins to a nominal category. 10. Anderson (2007b) considers topicality a special function, which is restricted to correlatives in languages where there is also an unrestricted alternative strategy of relativization available. In such languages correlatives can specialize to take up a topical discourse role. 11. A partly similar account was proposed by Rebuschi (2003) for Northern Basque, with the difference that correlatives in this account do not behave as generalized quantifiers.



The landscape of correlatives  12. That the position of the comparative phrase is Sp,CP is evidenced by the possibility of spelling out the complementizer that in both clauses:

(i) The more that you read, the less that you understand.

This “doubly-filled comp” violation is another quirky property of the CC construction in English. 13. Abeillé and Borsley (2008) note that cross-linguistic consistency of CCs is not absolute, thus (50) might not be applicable in all languages. Some languages might use different strategies for the expression of contents similar to that of comparative correlatives. An example might be French, which does not pattern with English CCs in that the first clause in a CC construction is not a relative clause and seems to be coordinated to the second clause. Among some other pieces of evidence, Abeillé and Borsley remark that the construction can even contain a coordinator between the two clauses: (i) Plus je lis (et) plus je comprends. more I read and more I understand ‘The more I read, the more I understand.’ I t is interesting to note in this context that the presence of coordinators would not in and of itself argue against the correlative nature of the construction, as optional coordinators are not unknown in the realm of correlatives. Burushaski (Tiffou and Patry 1995) and Basque (Lipták and Rebuschi to appear) exhibit them: (ii) Amenmoinga barin écam (ka) mo gusmoina γare sail ayét. which.com words do.fut.1sg and the woman.com with walk do.not ‘Do not walk with the woman with whom I talk.’ (iii)

Nork ere huts egiten bait du, (eta) hura Peiok who.erg ever mistake doing comp aux and that Peio.erg zigortuko du. punish.prosp aux



‘Who(ever) makes a mistake, Peio will punish him.’

14. With the help of this example Den Dikken shows that multiple extraction out of CCs is a case of a parasitic gap construction, and not ATB-movement. 15. I refer to this as a tendency, since Sanskrit (Davison (this volume), mentioned also in Lehmann (1984)) is an exception. Davison argues that the availability of stacking is not semantically but syntactically conditioned. She takes stacking to be a configuration where the two relatives adjoin to each other before adjoinging to the main clause. Such a structure is only allowed in a language where the correlative does not have to comply with the requirment that it has to c-command its correlate. If two CPs adjoin to each other, one of them, namely the one that does not project, does not c-command into the main clause. Sanskrit, which has no c-command requirement allows for stacking. Hindi on the other hand has the c-command requirement, consequently, stacking is not allowed. 16. Dayal notes that the functional reading of correlatives is only available when the main clause can also denote a relation. When that is not the case, like in the following example quoted by McCawley (1992), we get a single pair interpretation: (i) [jo laRkii jis laRke-se baat kar rahii hai] ve ek saath rel girl rel boy-ins talk do.prog is they together



Anikó Lipták sinemaa jaayeNge movie go.fut ‘Which (particular) girl is talking to which boy, they will go to the movies together.’ Gajewski (2008, fn. 3) notes without illustration that exhaustivity and uniqueness of the pairing can be missing in other contexts, too, for some speakers. 17. Lafon (1996) suggests that ba- and bait- originate from the same morpheme, a positive assertive particle bai ‘yes’. See Rebuschi (this volume) for more details and objections to this hypothesis. 18. Not all correlative languages express conditionals with a correlative structure. Hindi, for example, does not. 19. Arsenijević (this volume) points out that conditionals and correlatives show the same kind of embedding possibilities in Serbian. For instance, they cannot be embedded in a relative clause and can only be embedded under verbs of saying/believing under special conditions. 20. Andrews (1985) also cashed out the idea that correlatives are conditionals in the analysis of multiple relatives. The parallel between the two is semantic: the meaning of the correlative can be captured by replacing the relative phrases with an indefinite, and recasting the relative clause as a conditional. Such a conditional analysis for multiple relatives, however, is unlikely to be on the right track as this account undergenerates the possible meanings. Next to this universal meaning, multiple correlatives can also have a non-universal meaning where they denote a single pair of individuals. 21. This kind of free choice readings are also available in Hindi -bhii clauses as was noted above. In this language, generic tense is also required for this reading. 22. This string of particles are completely impossible in either main or dependent clauses in Hindi/Urdu.

part 1

Correlatives and related constructions

What don’t wh-questions, free relatives, and correlatives have in common? Barbara Citko This paper examines the behavior of Polish wh-questions, standard correlatives, comparative correlatives, and free relatives with respect to the availability of multiple wh-pronouns, superiority effects, left branch extraction, and reconstruction effects. It shows that with respect to all these diagnostics correlatives pattern with wh-questions. It further shows that free relatives differ from both correlatives and wh-questions with respect to the same diagnostics, and pattern together with headed relatives. Keywords: correlatives, free relatives, head account, reconstruction effects

1. Introduction Various similarities between wh-questions, free relatives, and correlatives have received a lot of attention in generative literature. The differences between them, however, have not been studied to the same extent. My main goal in this paper is to remedy this gap. The three constructions I will be comparing are exemplified in (1–3), respectively.

(1) What is John working on right now?



(2) I like what John is working on right now.



(3) * What John is working on right now, this (topic) I like.

wh-questions free relatives correlatives

What distinguishes correlatives from free relatives (as well as headed ones) is the fact that in a correlative construction the modifying clause is in a left peripheral position and the matrix clause contains an anaphoric correlate. The ungrammatical status of the example given in (3) above shows that correlatives are not possible in English. They are quite common, however, in Indo-Aryan (as shown by Srivastav (1991), Dayal (1996), Bhatt (2003), among many others), Finno-Ugric (as shown by Lipták (2005)), and Slavic languages (as shown by Izvorski (1996) and Rudin (2006)). The following examples provide an illustration of correlatives in these languages.1



Barbara Citko

(4) jo laRkii khaRii hai vo lambii hai. which girl standing be.pr she tall be.pr ‘Which girl is standing, she is tall.’

[Hin] (Dayal 1996: 160)

(5) Aki korán jöt azt ingyen beengedték. rel.who early came that.acc freely pv.admitted.3pl ‘Those who came early were admitted for free.’

[Hun]

(6) a. Kogo ljublju, togo poceluju. whom love.1sg dem kiss.1sg ‘I’ll kiss who I love.’

[Rus]

b. Kogo lubisz, tego pocałuj.2 whom like.2sg dem kiss.imp ‘Kiss the one you like.’

[Pol]

c.

[Bul]

Kolkoto pari iska tolkova misli how.much money wants that.much thinks če šte i dam. that will her give ‘She thinks that I’ll give her as much money as she wants.’

In what follows, I will further distinguish standard correlatives, illustrated in (4–6) above, from the so-called comparative correlatives, illustrated in (7a-d), which differ from standard ones in that the first clause is interpreted as a conditional rather than a restrictive one. Den Dikken (2005) argues, based on cross-linguistic as well as diachronic evidence, that comparative correlatives should be structurally assimilated to standard ones. The parallelism between the two is evident in the examples given in (7b-d); in all of them the adjunct clause contains a relative or an interrogative pronoun and the matrix one a corresponding demonstrative correlate:3, 4

(7) a. The more you smile, the happier you get.

b.

jiitnaa suuraj chamk-aa, utnii(-hii) [Hin] how.much sun shine that.much(-only) ThanD baRh-ii cold.f increase ‘The more the sun shone, the colder it got.’ (Den Dikken 2005: 499)

c.

Minél többet olvasol, annál többet [Hun] what.adess more read.2sg that.addess more.acc {megértesz / értesz meg}. pv.understand.2sg understand.2sg pv ‘The more you read, the more you understand.’ (Den Dikken 2005: 500)



Wh-questions, free relatives and correlatives

d.

Im więcej się uśmiechasz, tym lepiej rel more refl smile dem.instr better się czujesz. refl feel.2sg ‘The more you smile, the better you feel.’

[Pol]

Wh-questions, correlatives, and free relatives are typically analyzed as involving the structures given in (8a-c). All involve a CP with an interrogative (or relative) pronoun in its specifier. Furthermore, in languages with overt wh-movement, which are the languages I focus on in this paper, the wh-pronoun undergoes the same kind of movement in all three constructions. (8) a. wh-questions CP 3 whi C’ 3 C TP 5 … ti … b. correlatives CPmatrix wo CPrelative CPmatrix 3 3 whi C’ dem C’ 2 2 C TP C TP 5 5 … ti … … ti … c. free relatives on the Comp Account (Groos and van Riemsdijk 1981) DP 3 DP CP ∅ 3 whi C’ 3 C TP 5 … ti … Different researchers have different views on the category status of the matrix clause in a correlative construction. Dayal (1996) and Bhatt (2003) take it to be an IP rather than





Barbara Citko

a CP. However, there is convincing evidence, due to Izvorski (1996) (see also Den Dikken (this volume)), that at least in some languages the matrix clause is a CP (rather than IP). I will come back to this difference in Section 2 of this paper. Furthermore, the structure given for correlatives in (8b) does not specify how the relative clause gets into its CP adjoined position. Different researchers have different views on this matter as well. Srivastav (1991) proposes that it is base-generated there and binds the demonstrative pronoun in the matrix clause. Bhatt (2003), on the other hand, argues that the relative clause starts out adjoined to the demonstrative pronoun inside the matrix clause, and moves to its final adjunct position.5 Given the structures in (8a-c), we would expect free relatives and correlatives (standard and comparative alike) to behave similarly to wh-questions with respect to wh-movement diagnostics. My main goal in this paper is to test this prediction. I will use the following diagnostics: the availability of multiple wh-pronouns, the presence or absence of superiority effects, the availability of left branch extraction, and reconstruction effects. The main empirical focus of the paper will be on Slavic languages, which are known since Wachowicz (1974) and Rudin (1988) to allow multiple whfronting and, and since Ross (1967), to allow violations of the left branch condition in wh-questions. Most of the new data will come from Polish, a West Slavic language; however, comparisons with other Slavic (as well as non-Slavic) languages will also be made where relevant. My secondary goal in this paper is to compare the behavior of standard and comparative correlatives with respect to these diagnostics. To preview what is to come, I will show that Polish wh-questions and standard correlatives pattern alike; free relatives, however, show a different behavior with respect to the availability of multiple wh-pronouns, left branch extraction, and Principle C reconstruction effects. I will further show that comparative correlatives behave differently from standard ones with respect to the availability of multiple relative phrases.

2. Similarities 2.1

Use of wh-pronouns

There are a number of well-documented similarities between wh-questions, free relatives, and correlatives. For example, in Polish (as well as many other languages), all three use the same range of interrogative pronouns. Some representative examples are given below. 6 (9) a. Co chcesz? what want.2sg ‘What do you want?’ b. Dostaniesz co(kolwiek) chcesz. get.2sg what(ever) want.2sg ‘You will get whatever you want.’



Wh-questions, free relatives and correlatives 

c. Co chcesz, to dostaniesz. what want.2sg dem get.2sg ‘You will get what you want.’ (10) a. Kto mi pomoże? who me help.3sg ‘Who will help me?’ b.

Kto(kolwiek) mi pomoże, zostanie moim whoever me help.3sg become.3sg my najlepszym przyjacielem. best friend ‘Whoever helps me will become my best friend.’

c.

Kto mi pomoże, ten zostanie moim who me help.3sg dem become.3sg my najlepszym przyjacielem. best friend ‘The one who helps me will become my best friend.’

(11) a. Jak zaśpiewasz? how sing.2sg ‘How will you sing?’ b. Zaśpiewam jak(kolwiek) mi każesz. sing.1sg how(ever) me tell.2sg ‘I will sing however you tell me (to sing).’ c. Jak mi każesz, tak zaśpiewam. how me tell.2sg dem sing.1sg ‘I will sing the way you tell me to.’ There are three exceptions to the generalization that questions, correlatives, and free relatives use the same range of wh-pronouns. First, there are two comparative correlative strategies in Polish. One involves the expected combination of a wh-word in the relative clause and a demonstrative correlate in the matrix clause. The other strategy, however, does not use a wh-pronoun but a frozen relative pronoun im.7 This relative pronoun is only used in comparative correlatives. The two strategies are illustrated in (12a-b), respectively. In both of them, the demonstrative correlative and the wh (or relative) pronoun are marked Instrumental. (12) a. Czym starsze, tym głupsze. what.instr older dem.instr dumber ‘The older, the dumber’ b. Im starsze, tym głupsze. rel older dem.instr dumber ‘The older, the dumber’



Barbara Citko

Second, the parallelism between the three constructions breaks down when it comes to the wh-pronoun why. Polish has two lexical items corresponding to the English word why. I will follow Stepanov and Tsai’s (2005) terminology and refer to one of them as a reason why, glossed below as whyR, and the other one as a purpose why, glossed below as whyP. The difference between the two can best be seen by looking at the appropriate answers to the questions they ask. The reason why questions the reason behind the event, whereas a purpose why questions its purpose. (13) a. Dlaczego Jan wyszedł? whyR Jan leave.p.3sg ‘Why did Jan leave?’ b. Jan wyszedł, bo był zmęczony. Jan leave.p.3sg because was tired ‘Jan left because he was tired.’ (14) a. Po co Jan wyszedł? whyP Jan leave.p.3sg ‘Why did Jan leave?’ b. Jan wyszedł żeby kupić gazetę. Jan leave.p.3sg in-order-to buy.inf newspaper ‘Jan left in order to buy a newspaper.’ Both reason and purpose why are possible in correlatives, as shown below: (15) a.

Dlaczego Maria zrezygnowała, dlatego Jan whyR Maria resign.p.3sg dem Jan też powinien zrezygnować. also should resign.inf ‘Jan should resign for the same reason Maria resigned.’

b. Po co Jan wyszedł, po to samo Maria też wyszła. whyP Jan leave.p.3sg dem same Maria also leave.p.3sg ‘Jan left for the same reason Maria left.’ In free relatives, however, there is a contrast between a reason and a purpose why.8 Free relatives with a purpose why are markedly better than the ones with a reason why, as shown below: (16) a. * Jan wyszedł dlaczegokolwiek Maria wyszła. Jan leave.p.3sg whyeverR Maria leave.p.3sg ‘Jan left whyever Maria left.’



Wh-questions, free relatives and correlatives 

b. ?Jan wyszedł po cokolwiek Maria wyszła. Jan leave.p.3sg whyeverP Maria leave.p.3sg ‘Jan left whyever Maria left.’ And third, there is one wh-pronoun which is possible in questions and free relatives but not in correlatives. The pronoun in question is the wh-pronoun który ‘which’. (17) a. Który student dostał nagrodę? which student get.p.3sg prize ‘Which student got the prize?’ b.

Którykolwiek student napisał najlepszy whichever student write.p.3sg best doktorat powinien dostać nagrodę. dissertation should get.inf prize ‘Whichever student wrote the best dissertation should get the prize.’

c. * Który student napisał najlepszy doktorat, ten which student write.p.3sg best dissertation dem powinien dostać nagrodę. should get.inf prize ‘The student who wrote the best dissertation should get the prize.’ What seems to be responsible for the ungrammatical status of (17c) is a gap in the morphological paradigm. As the paradigm in (18) shows, Polish interrogative and demonstrative pronouns are morphologically related. In the demonstrative series, the dental stop t (or d) replaces the initial consonant of the corresponding interrogative. The only exception is the wh-pronoun który ‘which’, which does not seem to have a corresponding demonstrative. (18)

2.2

co jak kiedy jaki ile dlaczego po co który kto gdzie

what how when what-kind-of how-many whyR whyP which who where

to tak wtedy taki tyle dlatego po to ?? ten tam

this this-way then this-kind-of that-many for that for that this there

Movement of the wh-pronoun

Another well-documented range of similarities between wh-questions, free relatives and the two types of correlatives involves movement diagnostics.



Barbara Citko

In Polish (as well as in all the other Slavic languages), movement of the wh-pronoun is obligatory in all four constructions, as shown by the contrast in grammaticality between the grammatical examples in (19a-d), involving overt movement of the wh-pronoun, and the ungrammatical ones in (20a-d), involving wh-in-situ. (19) a. Co Jan chce ti? what Jan want.3sg ‘What does Jan want?’ b. Jan chce co(kolwieki) Maria chce ti. Jan want.3sg whatever Maria want.3sg ‘Jan wants whaever Maria wants.’ c. Coi Maria chce ti, to Jan chce. what Maria want.3sg dem Jan want.3sg ‘Jan wants the thing Maria wants.’ d. Im więcej czytasz, tym jesteś mądrzejszy. rel more read.2sg dem.instr be.2sg smarter ‘The more you read, the smarter you get.’ (20) a. * Jan chce co? Jan want.3sg what ‘What does Jan want?’ b. * Jan chce Maria chce co(kolwiek). Jan want.3sg Maria want.3sg what(ever) ‘Jan wants whatever Maria wants.’ c. * Jan chce co, to Maria chce. Jan want.3sg what dem Maria want.3sg ‘Jan wants what Maria wants.’ d. * Więcej czytasz im, tym mądrzejszy jesteś. more read.2sg rel dem.instr marter be.2sg ‘The more you read, the smarter you get.’ Furthermore, in both types of correlatives, movement of the demonstrative correlate is also obligatory, as shown by the ungrammatical status of (21a-b). (21) a. * Co Jan chce, dostanie to. what Jan want.3sg get.3sg dem ‘Jan will get what he wants.’ b. * Im więcej czytasz, mądrzejszy jesteś tym. rel more read.2sg smarter be.2sg dem ‘The more you read, the smarter you get.’



Wh-questions, free relatives and correlatives 

The obligatory movement of the correlative pronoun shows that at least in some languages the matrix clause is a CP. Izvorski (1996) makes a similar observation based on Bulgarian facts (see also Den Dikken (this volume) for relevant discussion).

3. Differences 3.1

The availability of multiple wh-pronouns

The first difference between questions, correlatives, and free relatives to consider involves the availability of multiple wh-pronouns. Both questions and correlatives with multiple wh-pronouns are grammatical, as shown in (22a) and (22b), respectively..9 Free relatives with multiple wh-pronouns, however, are ungrammatical, as shown in (22c). (22) a. Kto co chce? who what want.3sg ‘Who wants what?’ b. Kto co chce, ten to dostanie. who what want.3sg dem.nom dem.acc get.3sg ‘Everyone gets what they want.’ c. * Kto(kolwiek) cokolwiek zrobił będzie moim przyjacielem. who(ever) what(ever) did be.3sg my friend. ‘Everyone can take what he/she wants.’ As a side note, let me note that correlatives behave similarly to wh-questions with respect to superiority effects. In Polish, both allow violations of superiority, as shown by the grammatical status of the following examples: (23) a. Co kto chciał? what who want.p.3sg ‘Who wants what?’ b. Co kto chciał, ten to dostał. what who want.p.3sg dem.nom dem.acc get.p.3sg ‘Everyone got what they wanted.’ It seems quite natural to assume then that whatever independent factors are responsible for the lack of superiority effects in wh-questions are also at play in correlatives.10 For further discussion of what these factors might be, see Citko (1998), Bošković (2002), and Richards (2001), among many others.



Barbara Citko

Demonstrative correlates inside the matrix clause behave similarly to wh-pronouns inside the adjunct clause in that they also undergo obligatory fronting: (24) a. * Komu co Jan dał, temu who.dat what.acc Jan give.p.3sg dem.dat Maria zabierze to. Maria take.back.p.3sg dem.acc b. * Komu co Jan dał, who.dat what.acc Jan give.p.3sg Maria zabierze temu to. Maria take.back.p.3sg dem.dat dem.acc ‘Maria took back from everyone what Jan gave to the person.’ Furthermore, their movement also allows violations of superiority: (25) a. Kto co chce, ten to dostanie. who what want.3sg dem.nom dem.acc get.3sg ‘Everyone will get what they want.’ b. Kto co chce, to ten dostanie. who what want.3sg dem.acc dem.nom get.3sg ‘Everyone will get what they want.’ The movement of the wh-pronouns inside the relative clause and the movement of the demonstrative pronouns inside the matrix clause are independent of each other in a sense that the ordering of the fronted wh-pronouns does not have to parallel the ordering of the fronted demonstratives: (26) a.

Komu co Jan dał, who.dat what.acc Jan give.p.3sg, to temu Maria zabierze. dem.acc dem.dat Maria take.back.p.3sg ‘Maria took back from everyone what Jan gave to the person.’

b. Co kto chciał, ten to dostanie. what.acc who.nom want. p.3sg dem.nom dem.acc get.3sg ‘Everyone will get what they wanted.’ Interestingly, comparative correlatives differ from standard correlatives in that they do not allow multiple relative pronouns with corresponding multiple correlates. (27) * Im częściej im dłużej biegasz, tym bedziesz rel more.often rel longer run.2sg dem be.2sg silniejszy tym bardziej wysportowany. stronger dem more athletic ‘The more and the longer you run, the stronger and more athletic you will become.’



Wh-questions, free relatives and correlatives 

This might have something to do with the general semantics of this construction. In English, multiple comparative correlatives, in which one of the compared elements stays in situ, are also impossible. (28) * The more often you run the longer, the stronger you get the sooner. To sum up this section, we have seen that standard correlatives pattern with wh-questions with respect to multiple fronting (both of the relative phrases and the demonstrative pronouns) and with respect to superiority effects. Free relatives and comparative correlatives, on the other hand, differ from both wh-questions and standard correlatives in that they disallow multiple wh-pronouns. Their behavior with respect to superiority effects is thus not testable. In the next section, I turn to another difference between free relatives on the one hand, and correlatives and wh-questions on the other.

3.2

Left branch extraction

As is well-known since Ross (1967), Slavic languages (with the exception of Bulgarian) allow left branch extraction. The Polish examples in (29a-b) show that both left branch extraction and pied-piping are possible. Discourse considerations dictate the choice between the two: 11 (29) a. Którąi Jan przeczytał ti książkę? which Jan read.p.3sg book ‘Which book did Jan read?’ b. Którą książkęi Jan przeczytał ti? which book Jan read.p.3sg ‘Which book did Jan read?’ Both standard and comparative correlatives pattern with wh-questions in that they also allow left branch extraction: (30) a. Jakii chcesz ti samochód, taki sobie kup. what.kind.of want.2sg car dem refl.dat buy.imp ‘Buy yourself whatever car you want.’ b. Im więceji czytam ti książek, tym więcej rozumiem. rel more read.1sg books.gen dem more understand.1sg ‘The more books I read, the more I understand.’ (Borsley 2004: 67) Not surprisingly, pied-piped variants are also grammatical: (31) a. Jaki samochódi chcesz ti, taki sobie kup. what.kind.of car want.2sg dem refl buy.imp ‘lit. What car you want, that car you buy.’

 Barbara Citko

b. Im więcej książeki czytam ti, tym więcej rozumiem. rel more books read.1sg dem more understand.1sg ‘The more books I read, the more I understand.’ Interestingly, there is one difference between correlatives with left branch extraction and the ones without it. In left branch extraction cases, the repetition of the stranded nominal in the matrix clause is somewhat degraded. This is not the case in correlatives without left branch extraction. (32) a. ?? Jakii chcesz ti samochód, takii sobie kup ti samochód. what.kind.of want.2sg car dem refl buy.imp car ‘Buy the kind of car you want.’ b.

Jaki samochódi chcesz ti, what.kind.of car want.2g taki samochódi sobie kup ti. dem car refl buy.imp ‘What car you want, that car you buy.’

This effect can be derived from independent discourse properties of left branch extraction. Typically, the nominal stranded by left branch extraction is interpreted as focused.12 In the ungrammatical examples above, however, the repeated nominal inside the matrix cannot be felicitously interpreted as focused, since it is no longer new information. This effect is not limited to correlatives. In Citko (2006), I show that acrossthe-board wh-questions are subject to a similar restriction. They are ungrammatical if the nominals stranded by left branch extraction are identical:13 (33) a. * Jakii Maria chce ti samochód what.kind.of Maria want.3sg car.acc a Jan sobie kupił ti samochód? and Jan refl buy.p.3sg car.acc ‘What kind of car does Mary want and John bought? b. * Ilei Maria napisała ti książek how.many Maria write.p.3sg books a Jan przeczytał ti książek? and Jan read.p.3sg books ‘How many books did Maria write and how many books did Jan read?’ Furthermore, the status of both correlatives and ATB questions improves if the stranded nominals are distinct, which adds plausibility to the view that discourse considerations are at play here.



Wh-questions, free relatives and correlatives 

(34) a.

Ilei Jan sprzedał ti samochodów, how.many Jan sell.p.3g cars tyle Marek musi sprzedać ti cięzarówek. dem Marek must sell.inf trucks ‘Marek must sell as many trucks as Jan has sold cars.’

b.

Ilei Maria napisała ti książek how.many Maria write p.3g books a Jan przeczytał ti artykułów? and Jan read p.3sg articles ‘How many books did Maria write and how many articles did Jan read?’

So far, we have seen that correlatives pattern with wh-questions with respect to left branch extraction. What is interesting is that free relatives differ from both questions and correlatives in that they disallow left branch extraction:14 (35) a. * Jan podrobi kogokolwieki Maria podrobiła ti podpis. Jan forge.3sg whosever Maria forge.p.3g signature ‘Jan will forge whosever signature Maria forged.’ b. * Jan wyrzuci kogokolwieki zobaczył ti nazwisko Jan fire.3sg whosever see.p.3g name na tej liście. on this list ‘Jan will fire whosever name he saw on this list.’ Interestingly, pied-piping improves the status of both (35a) and (35b): (36) a. Jan podrobi kogokolwiek podpisi Maria podrobiła ti. Jan forge.3sg whosever signature Maria forge.p.3sg ‘Jan will forge whosever signature Maria forged.’ b. Jan wyrzuci kogokolwiek nazwiskoi zobaczył ti na tej liście. Jan fire.3sg whosever name see.p.3sg on this list ‘Jan will fire whosever name he saw on this list.’ However, the example in (36b) is only grammatical if the entire wh-phrase kogokolwiek nazwisko ‘whosever name’ determines the denotation of the entire free relative. It is thus grammatical with the interpretation that Jan is throwing away the names that he saw on the list, not the people whose names he saw on the list. This interpretation is possible due to the fact that the Polish verb wyrzucić can mean either ‘to fire (someone)’ or ‘to throw away (someone or something)’. A similar observation regarding the denotation of English free relatives was made by Grosu (2003), who notes that free relatives ‘allow wh-phrases of arbitrary complexity, so long as the entire phrase contributes to determining the nature of the designatum of



Barbara Citko

the SFR [standard free relative, B.C.]’ (Grosu 2003: 253). In the ungrammatical example given in (37a) below only the left branch determines the reference of the free relative. (37) a * John will fire whose signature appears on the list. b. John will forge whose signature you want me to forge. Grosu’s generalization, however, seems only partially correct. As pointed out by Jacobson (1995), the ungrammatical (37a) becomes grammatical if ever is added to the wh-pronoun:15 (38) John will fire whosever signature appears on the list. Grosu’s observation thus does not seem to apply to free relatives with ever. At present, I do not have an explanation for why the presence or absence of ever should affect the grammaticality of free relatives such as the one in (37a). One thing I would like to note is that the ameliorating effect of ever goes beyond English and beyond free relatives with possessor wh-pronouns. In Polish, all free relatives with complex wh-phrases are ungrammatical without kolwiek ‘ever’, irrespective of the function of the wh-pronoun: (39) Czytam jakie*(kolwiek) książki Maria czyta. read.1sg what(ever) books Maria read.3sg ‘I read what books Maria reads.’ Free relatives headed by simple wh-pronouns are possible without kolwiek ‘ever’: (40) Przeczytam co(kolwiek) Maria przeczytała. read.1sg what(ever) Maria read.p.3g ‘I will read(ever) what Maria read.’ Donati (2006: 32) notes a similar contrast in Italian: (41) a. * Ho mangiano quanti biscotti have.1sg eaten how.many cookies hai preparato. have.2sg prepared ‘I have eaten what cookies you have prepared.’

[Ita]

b. Mangerò qualunque biscotto preparerai. eat.fut.1sg whatever cookie prepare.fut.2sg ‘I will eat whatever cookie you will prepare.’ In English, free relatives headed by singular wh-phrases are subject to a similar restriction. Larson (1987) and Grosu (1996) point out that they are ungrammatical without ever: (42) a. John will execute what*(ever) order Mary gives him. (Grosu 1996: 261) b. I will accept what *(ever) mission you entrust me with.



Wh-questions, free relatives and correlatives 

This is not the case with plural wh-phrases: (43) a. John will execute what orders Mary gives him. b. I will accept what missions you entrust me with. Grosu attributes this effect to the properties of what in relative clauses. He suggests that what carries the implication that the set of individuals designated by the common noun has a (relatively) low and not precisely specified cardinality (Grosu 1996: 261). It is this property of the relative what that prevents it from combining with singular nouns, whose cardinality is fixed to one. Grosu’s explanation seems unsatisfactory for the following three reasons. First, it seems to imply the existence of two lexical items what: an interrogative what, which does not carry the low cardinality implication, and the relative what, which does. Second, it does not really explain why adding ever should affect the cardinality implication of what. And third, it does not extend to languages like Polish or Italian, in which all free relatives headed by complex wh-phrases without ever are ungrammatical. To sum up, we have seen that Polish correlatives pattern with wh-questions with respect to the availability of left branch extraction, multiple wh-fronting, and superiority. Free relatives, on the other hand, differ from both questions and correlatives with respect to these diagnostics. In particular, they disallow left branch extraction, and are incompatible with multiple wh-phrases. In the next section, I turn my focus to reconstruction effects, yet another diagnostic that will help us establish the position of the wh-pronoun in all the three constructions under consideration here.

3.3

Reconstruction effects

The diagnostics I will use to establish the presence (or absence) or reconstruction effects in Polish questions, correlatives, and free relatives involve anaphor binding, variable binding and Principle C effects.16 As expected, wh-phrases in wh-questions undergo reconstruction with respect to these diagnostics, as shown in (44a-c). (44) a. b. c.

anaphor binding: [Ile swoichj portretów]i Janj zamówił ti? how.many relf portraits Jan order.p.3sg ‘How many portraits of himself did Jan order?’ variable binding: [Ile portretów swojejj córki]i każdyj król zamówił ti? how.many portraits refl daughter every king order.p.3sg ‘How many portraits of his daughter did every king order?’ Principle C effects: *[Ile portretów Janaj]i onj zamówił ti? how.many portraits Jan.gen he order.p.3sg ‘How many portraits of Jan did he order?’



Barbara Citko

Standard correlatives behave similarly: (45) a. anaphor binding: [Ile swoichj zdjeć]i Janj zamówił ti, tyle dostał. how.many refl pictures Jan order.p.3sg dem get.p.3sg ‘Jan got however many pictures of himself he ordered.’ b. variable binding: [Ile portretów swojejj żony]i każdyj król how.many portraits refl wife every king zamówił ti, tyle dostał. order.p.3sg dem get.p.3sg ‘Every king got however many portraits of his wife he ordered.’ c. Principle C effects: * [Ile portretów Janaj]i onj zamówił ti, tyle dostanie. how.many portraits Jan.gen he order.p.3sg dem get.3sg ‘He will get as many portraits of Jan as he ordered.’ So do comparative correlatives: (46) a. anaphor binding: [Im więcej swoichj portretów]i Janj zamówi ti, rel more refl portraits Jan order.3sg tym bardziej będzie zadowolony. dem more be.3sg happy ‘The more portraits of himself Jan orders, the happier he will be.’ b. variable binding: [Im więcej portretów swojejj córki]i każdyj ojciec rel more portraits refl daughter every father zamówi ti, tym wiecej będzie go to kosztowało. order.3sg dem more be.3sg him this cost ‘The more portraits of his daughter every father orders, the more it will cost him.’ c. Principle C effects: *[Im więcej portretów Janaj]i onj zamówi ti, rel more portraits Jan.gen he order.3sg tym wiecej będzie go to kosztowało. dem more be.3sg him this cost ‘The more portraits of Jan he orders, the more it will cost him.’



Wh-questions, free relatives and correlatives 

Free relatives allow reconstructed interpretations with respect to anaphor and variable binding. What is interesting, though, is that they also allow non-reconstructed interpretations, as shown below: (47) a. b.

variable binding: Mistrzi namaluje [ilekolwiek portretów swojejj/i master paint.3sg however.many portraits refl żony]k każdyj król zamówi tk. wife every king order.3sg ‘The master will paint however many portraits of his wife every king orders.’ anaphor binding: Mistrzi namaluje [ilekolwiek swoichj/i master paint.3sg however.many refl portretów]k Janj zamówi tk. portraits Jan order.3sg ‘The master will paint however many portraits of himself Jan orders.’

With respect to Principle C effects, the non-reconstructed interpretations are also possible, as shown by the grammatical status of the following examples: (48) a. ? Kupię na aukcji [ilekolwiek autoportretów buy.1sg on auction however.many selfportraits Picassai]k oni namalował tk w okresie niebieskim. Picasso.gen he paint.p.3sg in period blue ‘I will buy at an auction however many selfportraits of Picasso he painted in the Blue period.’ b. ? Przeczytamy [ilekolwiek książek Tołstojai o Rosji read.1pl however.many books Tolstoy.gen about Russia w okresie carskim]k oni napisał tk na wygnaniu. in period tsar he write.p.3sg on exile ‘We will read however many of Tolstoy’s books about Russia in the era of the tsars he wrote in exile.’ There seems to be a considerable amount of cross-linguistic variation with respect to Principle C effects in free relatives. Even Slavic languages do not seem to be uniform in this respect. Gračanin-Yuksek (2008) notes that Croatian free relatives do show Principle C reconstruction effects, thus arguing in favor of the Comp Account.17 (49) *Vid će nagraditi koju god Danovuj djecu onj preporuči.18 Vid will reward which ever Dan’s children he recommends ‘Vid will reward whichever of Dan’s children he recommends.’ (Gračanin-Yuksek 2008: 13)



Barbara Citko

At present, I do not have data showing how Croatian free relatives behave with respect to the other two diagnostics considered here: the availability of multiple wh-pronouns and left branch extraction. However, the fact that there is variation even among such closely related languages suggests to me that both the Comp and Head Accounts might be right, for different languages though. Given this micro-variation, arguing against or in favour of either the Head or the Comp Account based on only one language seems to me to be a misguided effort. The main point that emerges from the discussion is that Polish free relatives differ from both questions and correlatives. The table in (50) summarizes these differences: (50) Summary of Facts

wh-questions free relatives standard correlatives comparative correlatives

multiple wh-fronting

lack of superiority effects

left branch extraction

anti-reconstruction effects

√ * √

√ N/A √

√ * √

* √ *

*

N/A

√

*

In the next section, I will focus on the analysis of the facts summarized in the table. I will abstract away from the differences between standard and comparative correlatives, and will focus on the differences between questions and correlatives on the one hand, and free relatives on the other hand.

4. Analysis The discussion given in the previous section shows that correlatives pattern with whquestions with respect to the availability of multiple wh-pronouns, left branch extraction, and reconstruction effects. The parallelism between questions and correlatives is not surprising, given the widely accepted analysis of correlatives, on which the relative clause is a CP and the wh-pronoun undergoes the same (obligatory in Slavic) movement in correlative clauses as a wh-phrase does in wh-questions. We have also seen evidence that in Slavic languages the matrix clause in correlatives has to be a CP as well. The evidence came from the fact the demonstrative correlates undergo the same kind of movement as wh-phrases. What I would like to focus on in the rest of the paper concerns the lack of parallelism between free relatives on the one hand, and questions and correlatives on the other. This is not what we expect given the structure assigned to free relatives on the Comp Account, which is repeated below:



Wh-questions, free relatives and correlatives 

(51)

DP 3 DP CP ∅ 3 whi C’ 3 C TP 5 … ti …

We have seen that Polish free relatives are impossible with multiple wh-pronouns, disallow left extraction, and show Principle C anti-reconstruction effects. What is interesting is that in all these respects, free relatives pattern with headed relatives.19 Thus, Polish headed relatives also disallow multiple interrogative (or relative) pronouns: (52) * Samochód, który kto kupił jest warty dużo pieniędzy. car which who buy.p.3sg is worth much money ‘The car that somebody bought is worth a lot of money.’ There are two different ways to think about left branch extraction in headed relatives. The relative operator allows both pied-piping and left branch extraction, as shown in (53a-b). (53) a. książka, której okładkęi Maria zaprojektowała ti book whose cover Maria design.p.3sg ‘a book whose cover Maria designed’ b. książka, któreji Maria zaprojektowała ti okładkę book whose Maria design.p.3sg cover ‘a book whose cover Maria designed’ There is another potential instance of left branch extraction in headed relatives. It only arises on the head promotion account of relative clauses, such as the one proposed by Kayne (1994). Such a derivation is schematized in (54). First, the wh-phrase which book moves from the relative clause internal position to [Spec,CP]. Next, the head book moves to the specifier of the which phrase: (54)

DP wo D CP the wo DPj C’ 3 3 bookj D’ C TP 2 D NP Mary recommended ti which tj



Barbara Citko

Thus, whatever mechanism is responsible for left branch extraction in questions should also be allowed in relatives. A relative with left branch extraction from the head would simply involve movement of the wh-determiner instead of the movement of the entire wh-phrase: (55)

DP 3 D CP the 3 DP C’ g 3 D C TP whichi Mary recommended ti book

Such relatives, however, are ungrammatical:20 (56) a. * Kupię teni, który mi poleciłaś ti samochód. buy.1sg this which me recommend.p.3sg car ‘I will buy the car that you recommended to me.’ b. * Opublikujemy teni, który editor zaakceptował ti artykuł. publish.1pl this which editor accept.p.3sg article ‘We will publish the article that the editor accepted.’ These examples violate a very intuitive requirement that headed relatives need overt heads. Since the prediction that headed relatives should allow such left branch extraction arises only if they involve head promotion, we need independent evidence for head promotion in Polish headed relatives. Such evidence comes from standard diagnostics, such as anaphor binding (57a) and bound variable interpretation (57b). (57) a.

Wiersze o sobiei, które Mariai napisała poems about refl which Maria write.p.3sg są bardzo znane. are very famous ‘The poems about herself that Mary wrote are very famous.’

b.

Zdjęcie swojeji dziewczyny, które każdy żołnierzi picture refl girlfriend which every soldier ma w portflefu jest mu bardzo drogie. has in wallet is him very precious. ‘A picture of his girlfriend that every soldier has in his wallet is very precious to him.’

More evidence comes from the ambiguity of relative clauses of the kind given in (58). This example is modelled upon Bhatt’s (2002) English examples, which Bhatt showed



Wh-questions, free relatives and correlatives 

to be ambiguous between the so-called high and a low reading. The two readings are paraphrased in (58b-c). Furthermore, Bhatt showed that the low reading requires reconstruction into the relative clause internal position, which is only possible on the promotion analysis. (58) a. b.

Pierwsza książka, którą Jan powiedział, że Tołstoj napisał first book which Jan say.p.3sg that Tolstoy write.p.3sg ‘the first book that Jan said that Tolstoy wrote’ high reading: In 1990, John said that Tolstoy had written Anna Karenina; in 1991, John said that Tolstoy had written War and Peace. c. low reading: John said that the first book that Tolstoy had written was War and Peace.

So far, the reconstruction behavior of headed relatives parallels that of wh-questions. The two diverge, however, with respect to Principle C effects. The lack of Principle C effects in English relatives was first noted Munn (1994) (see also Sauerland (1998), Safir (1999), Citko (2002)). As shown in (59) and (60), Polish headed relatives also differ from wh-questions in that they do not show Principle C effects. (59) a. * [Którą studentkę Kowalskiegoi]j oni nam polecił tj? which student Kowalskiego.gen he us recommend.p.3sg ‘Which student of Kowalski did he give the best reference? b.

Studentka Kowalskiegoi, którą oni polecił ti, student Kowalski.gen which he recommend.p.3sg została przez nas zatrudniona. was by us employed ‘The student of Kowalski’s that he recommended to us was hired.’

(60) a. * [Ilu studentów Kowalskiegoi]j oni nam polecił tj? how.many students Kowalski.gen he us recommend.p.3sg ‘How many students of Kowalski did he recommend to us?’ b.

Zatrudnimy tylu studentów Kowalskiegoi, hire.1pl that.many students Kowalski.gen ilu oni nam polecił. how.many he us recommend.p.3sg ‘We will hire as many of Kowalski’s students as he recommended to us.’

The lack of Principle C effects is a puzzle on the head promotion analysis sketched above, on which the relative head undergoes wh-movement and ends up in [Spec,CP]. However, it receives a straightforward explanation on the so-called matching account, on which the head is base-generated in the head position, and what moves from the relative clause internal position is an empty operator (or relative pronoun). Such a matching account, however, will fail to account for the presence of any reconstruction effects whatsoever. What we need then is an account that would capture both



Barbara Citko

reconstruction and anti-reconstruction effects. This is precisely what a slightly modified version of the matching account, which I argued for in Citko (2001), is designed to do. On this account, the head is base-generated in the relative clause external position and a matching wh-phrase is raised from the relative clause internal position. The raised wh-phrase undergoes deletion under identity with the external head. The deleted element is indicated with a strikethrough in (61). (61)

DP wo DP CP 5 3 the book DP C’ 6 2 which booki C TP Mary recommended ti

The lack of Principle C effects follows from the fact that the lower copy can undergo deletion under identity with an external head. This version of the matching analysis can also explain the lack of left branch extraction in headed relatives. Irrespective of whether left branch extraction has taken place or not, the nominal (whether it is stranded or pied-piped) is deleted under identity with the external head. The ungrammaticality of the examples in (56a-b) thus follows from a simple assumption that they are missing (a base-generated) external head. A popular alternative explanation for the lack of Principle C effects involves the mechanism of Vehicle Change of Fiengo and May (1994). Vehicle Change replaces a name with its ‘pronominal correlate’, i.e. a pronoun bearing the same index as an offending name. Safir (1999) argues that Vehicle Change is responsible for the lack of Principle C effects in headed relatives, as shown in (62a-c). (62) a. A picture of Johni which hei thought Mary would like to have a picture of John was recently stolen. (raising of the DP a picture of John) b. A picture of Johni which hei thought Mary would like to have a picture of John was recently stolen. (reconstruction) c. A picture of Johni which hei thought Mary would like to have picture of himi was recently stolen. (Vehicle Change) In Citko (2001), I argued against the Vehicle Change mechanism on the following grounds. Vehicle Change was originally meant to explain the lack of Principle C effects in ellipsis (not movement) constructions. Extending it to constructions involving



Wh-questions, free relatives and correlatives 

movement predicts the lack of Principle C effects in many environments in which they do occur. For example, it is not clear to me how the Vehicle Change mechanism would derive the contrast between questions and relative clauses.21 Furthermore, the Vehicle Change account does not explain the correlation between Principle C effects and variable binding, scope, or idiom interpretation. Sauerland (1998) shows that in cases in which reconstruction is forced for variable binding, idiom chunk interpretation or scope, Principle C effects re-emerge. For example, in (63) the adjacency requirement on idiom chunk interpretation forces the interpretation of the head in a relative clause internal position, which yields a Principle C effect. (63) * The headway on Mary’si project that shei made pleased the boss. The question I would like to turn to now is why free relatives pattern with headed relatives, and correlatives with wh-questions. Conceptually the most straightforward way to account for the parallelism between free relatives and headed relatives would be to say that they involve the same structure. What this implies is that the wh-phrase in free relatives occupies the same position as the head in headed relatives. This is exactly the case if we adopt an alternative to the Comp Account, the so-called Head Account, on which the head in a free relative occupies the same position as the head in a headed relative (see Bresnan and Grimshaw (1978), Larson (1987), Larson (1998), Citko (2002), Bury (2003), Donati (2006) for various versions of this account). The original version of the Head Account, due to Bresnan and Grimshaw (1978), involves basegeneration of the wh-pronoun and deletion in the relative clause internal position. I will adopt a more ‘updated’ version of the Head Account, referred to in the literature as a derived Head Account, on which the wh-phrase moves to the head position (rather than being base-generated there). (64)

DP 3 DP CP 5 2 whatever C TP Tom reads twhatever

Now the similarities between free relatives and their headed counterparts with respect to the availability of multiple wh-pronouns, the availability of left branch extraction, and Principle C reconstruction effects can be attributed to the similarities in structure between the two. The lack of multiple wh-pronouns in free relatives follows from the fact that there is not enough room for multiple heads. On the Comp Account, the ungrammaticality of free relatives with multiple wh-pronouns would be hard to explain, since a wh-pronoun in a free relative occupies the same position as a wh-pronoun in a wh-question. The lack of left branch extraction in free relatives can also be attributed to the requirement that they have an overt nominal head. Similarly, the lack of Principle C



Barbara Citko

effects can be attributed to the fact that the wh-pronoun in free relatives is not in a [Spec,CP] position. Therefore, it does not have to undergo obligatory reconstruction. A natural question to ask is whether assigning the same structural position to heads in relative clauses and wh-pronouns in free relatives does not imply that there should be no differences whatsoever between the two. The two are well-known to differ with respect to the compatibility with overt complementizers and matching effects. For example, free relatives differ from headed ones in that they disallow overt complementizers (or relative pronouns). This is true of Polish as well as English free relatives:22 (65) a. * John reads whatever books that/which Tom reads. b. * Jan czyta jakiekolwiek książki, co/które Tomek czyta. Jan read.3sg whatever books comp/which Tom read.3sg ‘John reads whatever books Tom reads.’ Second, free relatives, unlike headed ones, exhibit matching effects. In headed relatives, the case of the external head does not have to match the case assigned to the internal operator trace, as shown in (66). The external head is Accusative, whereas the internal operator is Nominative.23 (66) Widziałem tego pana, który zbił ci szybę. see.p.3sg this.acc man.acc which.nom break.p.3sg you.dat glass ‘I saw the man that broke your glass. (Borsley 1997: 638) However, a corresponding free relative, that is a free relative in which the case of the whpronoun does not match the case of the relative clause internal gap, is ungrammatical. (67) * Widziałem kogokolwiek/ ktokolwiek zbił ci szybę. see.p.3sg whoever.acc/whoever.nom break.p.3sg you.dat glass ‘I saw whoever broke your glass.’ Interestingly, there is a fair amount of variation across languages and speakers with respect to matching in free relatives. Vogel and Zugck (2003) note that in some dialects of German, free relatives allow case mismatches as long as the realized case is higher on the case hierarchy than the suppressed case.24 In (68), the wh-pronoun is Dative even though the matrix verb assigns Accusative case. This kind of matching seems to be sensitive to case hierarchy; Dative, being higher on the case hierarchy than Accusative, is the case that wins. (68) Ich lade ein wem ich begegne. I invite who.dat I meet ‘I invite who I meet.’

[Ger] (Vogel and Zugck 2003: 110)

Further evidence that there is a fair amount of crosslinguistic variation with respect to matching in free relatives comes from languages like Icelandic, whose free relatives differ from the languages considered so far in this paper in two respects. First, its free



Wh-questions, free relatives and correlatives 

relatives allow overt complementizers. Second, the case realized on the wh-pronoun is the case assigned by the matrix verb.25 (69) a ?Eg hjálpa hverjum/ *hvern (sem) ég elska. I help who.dat/ who.acc (that) I like ‘I help whom I like.’ b. ?Eg elska *hverjum/hvern (sem) ég hjálpa. I like who.dat/who.acc (that) I help ‘I like whom I help.’

[Ice]

(Vogel 2000: 7)

The fact that there is so much crosslinguistic variation with respect to matching in free relatives suggests that the structure we assign to them has to be flexible enough to allow matching as well as non-matching. One way to capture this variation is to assume that case can be assigned to the wh-pronoun at different points in the derivation in different languages.

5. Conclusion To conclude briefly, I have examined in this paper the behaviour of Polish wh-questions, standard correlatives, comparative correlatives, and free relatives with respect to the availability of multiple wh-pronouns, superiority effects, left branch extraction, and reconstruction effects. I have shown that with respect to all these diagnostics, correlatives pattern with wh-questions. Free relatives, on the other hand, behave differently with respect to the same diagnostics. I have accounted for these differences by proposing a similar analysis for questions and correlatives, and distinguishing them from both free and headed relatives.

Acknowledgements An earlier version of this paper was presented at the 2006 Annual Meeting of the Linguistic Society of America. I would like to thank the audience at the meeting for useful feedback. I am also indebted to Anikó Lipták and an anonymous reviewer for many useful and inspiring comments, which led to substantial improvements in the paper. I alone am responsible for any remaining errors and omissions.

References Bhatt, Rajesh. 2002. The raising analysis of relative clauses: evidence from adjectival modification. Natural Language Semantics 10. 43–90.



Barbara Citko Bhatt, Rajesh. 2003. The locality in correlativization. Natural Language and Linguistic Theory 21. 485–541. Borsley, Robert D. 1997. Relative clauses and the theory of phrase structure. Linguistic Inquiry 28. 629–647. Borsley, Robert D. 2004. The view from the Polish periphery. In Olga Arnaudova, Wayles Browne, Maria Luisa Rivero & Daniela Stojanović (eds), Annual Workshop on Formal Approaches to Slavic Languages: the Ottawa Meeting 2003, 59–91. Ann Arbor, MI: Michigan Slavic Publications. Borsley, Robert D. & Ewa Jaworska. 1988. A note on prepositions and Case marking in Polish. Linguistic Inquiry 19. 685–691. Bošković, Željko. 2002. On multiple wh-fronting. Linguistic Inquiry 33. 351–383. Bošković, Željko. 2005. On the locality of left branch extraction and the structure of NP. Studia Linguistica 59. 1–45. Bresnan, Joan & Joan Grimshaw. 1978. The syntax of free relatives in English. Linguistic Inquiry 9. 331–391. Bury, Dirk. 2003. Phrase Structure and Derived Heads. Doctoral dissertation, University College London. Citko, Barbara. 1998. On Multiple WH Movement in Slavic. In Željko Bošković, Steven Franks & William Snyder (eds), Formal Approaches to Slavic Linguistics: The Connecticut Meeting, 97–114. Ann Arbor, MI: Michigan Slavic Publications. Citko, Barbara. 2001. Deletion under identity account of relative clauses. In Minjoo Kim & Uri Strauss (eds), Proceedings of the 31st North East Linguistics Society. Amherst, MA: GLSA Publications. Citko, Barbara. 2002. Anti-reconstruction effects in free relatives: a new argument against the Comp Account. Linguistic Inquiry 33. 507–511. Citko, Barbara. 2006. On the interaction of left branch extraction and across-the-board whmovement. Syntax 9. 225–247. Corver, Norbert. 1990. The syntax of left branch extraction. Doctoral dissertation, Tilburg University. Dayal, Veneeta. 1996. Locality in wh-quantification: Questions and relative clauses in Hindi. Studies in Linguistics and Philosophy 62. Dordrecht: Kluwer Academic Publishers. Dikken, Marcel den. 2005. Comparative correlatives comparatively. Linguistic Inquiry 36. 497–533. Dikken, Marcel den. This volume. Comparative correlatives and successive cyclicity. Donati, Caterina. 2006. On Wh-head-movement. In Lisa L. Cheng & Norbert Corver (eds), Whmovement on the Move. 21–46. Cambridge, MA: MIT Press. Fanselow, Gisbert & Damir Ćavar. 2001. Distributed Deletion. In Artemis Alexiadou (ed) Universals of Language: Proceedings of the 1999 GLOW Colloqium. Amsterdam: Benjamins. Fiengo, Robert and Robart May. 1994. Indices and Identity. Cambridge, MA: MIT Press. Franks, Steven & Ljiljana Progovac. 1994. On the placement of Serbo-Croatian clitics. Indiana Linguistic Studies 7. 69–78. Gračanin-Yuksek, Martina. 2008. Free relatives in Croatian. Arguments for the COMP Account. Linguistic Inquiry 39. 275–294. Groos, Aneeke & Henk van Riemsdijk. 1981. Matching effects in free relatives: a parameter of core grammar. In Adriana Belletti, Luciana Brandi & Luigi Rizzi (eds), Theory of Markedness in Generative Grammar: Proceedings of the IVth GLOW Conference, 171–216. Pisa: Scuola Normale Superiore di Pisa.



Wh-questions, free relatives and correlatives  Grosu, Alexander. 1996. The Proper Analysis of “Missing-P” Free Relative Constructions. Linguistic Inquiry 27. 257–293. Grosu, Alexander. 2003. A unified theory of ‘standard’ and ‘transparent’ free relatives. Natural Language and Linguistic Theory 21. 247–331. Izvorski, Roumyana. 1996. The syntax and semantics of correlative proforms. In Kiyomi Kusumoto (ed), Proceedings of NELS 26, 133–147. Amherst, MA: GLSA. Jacobson, Pauline. 1995. On the quantificational force of English free relatives. In Emond Bach, Eloise Jelinek, Angelika Kratzer & Barbara H. Partee (eds), Quantification in Natural Languages [Studies in Linguistics and Philosophy 54], 451–486. Dordrecht/Reidel: Kluwer Academic Publishers. Kayne, Richard. 1994. Antisymmetry of Syntax. Cambridge: MA: MIT Press. Larson, Richard. 1987. Missing prepositions and the analysis of English free relative clauses. Linguistic Inquiry 19. 239–266. Larson, Richard. 1998. Free relative clauses and missing P’s: reply to Grosu. Unpublished manuscript, Stony Brook University. Lipták, Anikó. 2005. Correlative topicalization. Manuscript, Leiden University. Munn, Alan. 1994. A minimalist account of reconstruction asymmetries. In Proceedings of NELS 24, 397–410. Amherst, MA.: GLSA. Richards, Norvin. 2001. Movement in Language: Interactions and Architectures. Oxford: Oxford University Press. Rudin, Catherine. 1988. On multiple questions and multiple wh-fronting. Natural Language and Linguistic Theory 6. 445–501. Rudin, Catherine. 2006. Multiple wh relatives in Slavic. Paper presented at the 15th Annual Conference on Formal Approaches to Slavic Linguistics, University of Toronto. Ross, John R. 1967. Constraints on variables in syntax. Doctoral dissertation, MIT. Safir, Kenneth. 1999. A-bar reconstruction and vehicle change in A-bar chains. Linguistic Inquiry 30. 587–621. Sauerland, Uli. 1998. The meaning of chains. Doctoral dissertation, MIT. Srivastav, Veneeta. 1991.The syntax and semantics of correlatives. Natural Language and Linguistic Theory 9. 637–686. Stepanov, Artur & Dylan Tsai. 2005. The Hows and Whys of How and Why: Merger and licensing of wh-adjuncts. Paper presented at the 2005 Wh-Fest, Department of Linguistics, University of Maryland. Uriagereka, Juan. 1988. On government. Doctoral dissertation, University of Connecticut. Vogel, Ralf. 2000. Free relatives in OT syntax. In Gisbert Fanselow & Caroline Féry (eds), Resolving Conflicts in Grammars: Optimality Theory in Syntax, Morphology, and Phonology [Linguistische Berichte Sonderheft 11], 119–162. Hamburg: Helmut Buske Verlag. Vogel, Ralf & Marco Zugck. 2003. Counting Markedness. A corpus investigation on German free relative constructions. In Susann Fischer, Ruben van de Vijver & Ralf Vogel (eds), Experimental Studies in Linguistics 1 [Linguistics in Potsdam 21], 105–122. Wachowicz, Krystyna. 1974. Against the universality of a single wh-question movement. Foundations of Language 11. 155–166. Zubizarreta, Maria Luisa. 1998. Prosody, Focus, and Word Order. Cambridge: MA: MIT Press.



Barbara Citko

Endnotes 1. I will use the following abbreviations throughout the paper: ACC: accusative; ADESS: adessive, COMP: complementizer, DAT: dative, DEM: demonstrative, ERG: ergative, F: feminine, GEN: genitive, IMP: imperative, INF: infinititive, INSTR: instrumental, P: past; PL: plural; PR: present; PV: preverbal element; REFL: reflexive; REL: relative; SG: singular. In the Polish examples case is only glossed when relevant. 2. The Polish judgments reported in this paper are my own, unless otherwise noted. 3. Another possible distinction, brought to my attention by an anonymous reviewer, is between correlatives containing full DPs, such as the Hindi one in (4) and the ones containing just relative pronouns, such as the ones in (5) and (6a-b). For the purposes of this paper, I will abstract away from this potentially significant distinction. 4. An anonymous reviewer wonders whether the Slavic and Hungarian correlatives cited here are true correlatives, sharing all the characteristics of Hindi correlatives. I follow the consensus in the literature regarding the correlative status of these constructions, leaving open the possibility that they might differ from their Hindi counterparts. 5. Furthermore, Bhatt proposes that multiple correlatives, that is correlatives with multiple relative pronouns, involve a different derivation in which the relative clause is base-generated in a high adjoined position. He derives the difference between the two types of correlatives from a locality on Merge condition, which merges elements as early as possible. 6. The use of wh-pronouns in free relatives and correlatives is not universal, however. Other languages, such as Hindi or Hungarian, use wh-pronouns in wh-questions, but relative pronouns in correlatives. 7.

To my ear, this strategy is more common, and I will use it in the remainder of this paper.

8. The restrictions on free relatives with why seem to be quite common cross-linguistically. Larson (1987) points out a similar restriction in English. (i) * John left whyever Mary left. This gap is quite puzzling, given the fact that why is possible as a relative pronoun in headed relatives (ii), and that the form whyever exists as an emphatic form (iii).

(ii) The reason why John left is unclear.



(iii) Whyever would anyone do it?

9. The availability of ‘multiple heads’ is common from a crosslinguistic perspective, as shown by the following examples from Hindi and Hungarian. (i) jis laRkiine jis laRkeko dekhaa usne usko passand kiyaa. rel girl.erg rel boy.acc see.p.3sg that.erg that.acc like.p.3sg ‘Which girl saw which boy, she liked him.’ (ii) Aki amit kér az azt elveheti. rel.who rel.what.acc want.3sg that that.acc take.pot.3sg

[Hin]

[Hun]

10. This is not true of all Slavic languages that allow violations of superiority in questions. As shown by Bošković (2002), Serbo-Croatian correlatives differ from (simple) questions in that they obey superiority:



Wh-questions, free relatives and correlatives  (i) Ko koga voli, taj o njemu i govori. who whom loves that.one about him even talks ‘Everyone talks about the person they love.’

[SC]

(ii) ?*[Koga ko voli], taj o njemu/o njemu taj i govori.

11. The exact mechanism behind left branch extraction is not crucial for present purposes. It has been linked to the lack of overt determiners or the freedom of word order. Uriagereka (1988), Corver (1990), and Bošković (2005) correlate the availability of left branch extraction with the lack of overt articles in Slavic, which allows for a subjacency style of account of left branch extraction. In English-type languages, there is a DP layer; hence movement of the whdeterminer crosses two bounding nodes. In Slavic-type languages, on the other hand, there is no DP layer, hence movement of the wh-determiner (or wh-adjective to be more accurate) crosses just one bounding node. Alternatively, left branch extraction can been analyzed as involving remnant movement (Borsley and Jaworska (1988), Franks and Progovac (1994)), or discontinuous deletion (Fanselow and Cavar 2001). Even though the choice between the three accounts is not crucial, one thing is worth noting; only the subjacency-style approach involves movement of the actual left branch. Both the remnant movement approach and the discontinuous movement approach ‘create’ the illusion of left branch extraction through other means. 12. A separate issue is whether such discourse considerations are the driving force behind movement, or simply a ‘side-effect’ of movement driven by independent formal reasons. See Zubizarreta (1998) for arguments that movement can be prosodically driven. On Zubizarreta’s view, movement out of a VP can be driven by the prosodic need for the focused constituent to occupy the most deeply embedded structure. 13. This effect is stronger in ATB questions, which has to do with an independent requirement that in ATB questions the overt elements in the second conjunct have to be distinct from their correspondents in the first conjunct. 14. The Polish examples in (35) are modelled on the English examples discussed by Grosu (2003), given in (37a-b). 15. More examples of free relatives in which the wh-possessor determines the reference of the free relative are given in (i-ii). As the paraphrases in (iii-iv) show, both free relatives refer to people, not cars or ideas.

(i) Whosever car this is had better watch out.



(ii) Whosever idea this is is brilliant.



(iii) The person whose car this is had better watch out.



(iv) The person whose idea this is is brilliant.

16. Other diagnostics, which I will not consider in this paper, involve idiom chunk interpretation and scope. 17. Croatian free relatives also behave differently from Polish ones with respect to anaphor binding: (i) Vidi će nagraditi koje god svoje*i/j dijete Danj preporuči. Vid will reward which ever refl child Dan recommend.3sg ‘Vid will reward whichever of his children Dan recommends.’ (ii) Jani nagrodzi którekolwiek Jan reward.3sg whichever

[Cro]

(Gračanin-Yuksek 2008) ze swoichi/j dzieci Tomekj from refl children Tom

[Pol]



Barbara Citko mu poleci. him recommend.3sg 18. The Polish equivalent of (49) is grammatical on the coindexed reading: (i)

Jan nagrodzi którekolwiek z Tomkai dzieci oni Jan reward.3sg whichever from Tom.gen children he mu poleci. him recommend.3sg ‘Jan will reward whichever of Tom’s children he will recommend to him.’

To my ear, this example sounds even better than the examples given in (48) above. 19. This by no means implies that there are no differences between headed and free relatives whatsoever. As pointed out by an anonymous reviewer, the two differ with respect to relative pronoun selection, pied-piping, stacking, and interpretation. Since the focus of this paper is on the differences between questions, correlatives, and free relatives (rather than free relatives and headed relatives), I will leave these differences aside. 20. An anonymous reviewer suggests that the examples in (56a-b) could be excluded on independent grounds, such as Bianchi’s (1999) idea that feature checking requirements force raising of the actual head. However, if LBE involves raising of the entire DP and discontinuous spell-out (as argued for by Fanselow and Cavar 2001), this argument is not going to work. 21. Safir questions the existence of any contrast between wh-questions and relatives with respect to Principle C effects. He gives a number of examples of wh-questions where the Principle C effects do not seem to arise, contra the predictions made by the copy theory of movement (Safir 1999: 609):

(i) Which biography of Picassoi do you think hei wants to read?



(ii) Which witness’s attack on Leei did hei try to get expunged from the trail records?



(iii) Which criticism of Leei did hei choose to ignore?



(iv) Which evaluation of Lee’si physical fitness did hei use when he applied to NASA for space training?

(v) Whose allegation that Leei was less than truthful did hei refute vehemently? The judgments regarding these sentences vary, and I do not at present have an explanation for this variation. Safir takes the lack of Principle C effects to be the general pattern, and attributes the violations of Principle C effects, where they do occur, to independent factors. A more standard view, which I am assuming here, is to assume that the presence of Principle C violations in wh-questions is the norm, and to treat the lack of Principle C effects, where it does occur, as an instance of accidental coreference. Thus, I will proceed on the assumption that the contrast between wh-questions and relatives is real, and needs to be accounted for. 22. The ungrammatical status of the Polish free relative in (65b) is even more striking given the fact that Polish does not allow headed relatives without overt complementizers or relative pronouns: (i) * Jan czyta te same książki Maria czyta. Jan read.3sg these same books Maria read.3sg ‘Jan reads the same books Maria reads.’ 23. At first sight, the lack of matching in headed relatives seems to be a problem for head promotion analyses. The problem disappears, Kayne (1994) argues, if we assume that what undergoes raising from the relative clause internal position is not the entire DP but an NP.



Wh-questions, free relatives and correlatives  24. Not all speakers of German allow non-matching free relatives. Vogel (2000) distinguishes three German dialects which differ with respect to matching effects in free relatives. The differences between them follow from case hierarchy. 25. The fact that the complementizer is possible in free relatives is quite unusual. It does suggest that at least in some languages the distinction between free and headed relatives is not as clear cut as it is normally thought to be.

Basque correlatives and their kin in the history of Northern Basque Georges Rebuschi Relative correlatives in Northern Basque are more restricted than in some other languages. First, they are stylistically marked as highly rhetorical, second, they only impart a generic reading, and thus do not function like ordinary restrictive correlatives as in e.g. Hindi. This paper, which is mainly descriptive, will be concerned with the syntactic differences that help distinguish between correlative relatives and other non-root clauses with and without a wh-item: embedded questions, appositive relatives, and indefinite free relatives, semifree relatives, conditional and unconditional protases. Hitherto unnoticed examples are provided which show how the close semantic connection between the various types of subordinated clauses examined here have triggered some interesting overlaps in the history of the language. Keywords: Basque, embedded questions, free relatives, semi-free relatives, unconditional clauses

1. Introduction 1.1

The Basque language

Basque is a language isolate, spoken on both sides of the French-Spanish border on the Atlantic coast. Three basic groups of dialects are recognized today (Zuazo 1998): Western (or Biscayan, bizkaiera), Central (or Guipuzcoan, gipuzkera), and Eastern, a term which subsumes most of the varieties spoken in Spanish Navarre (Nafarroa), and those spoken in the French Basque Country – the “Northern dialects”. Sub-varieties are countless, but it is customary to distinguish, among the varieties of Northern Basque, between, from the Atlantic coast eastwards: Labourdin Basque (lapurtera), Low Navarrese (behe-nafarrera) and Souletin Basque (zuberera). The first book ever printed in Basque, Etchepare (1545), was written in Low Navarrese. The two other monuments of Basque literature were also written in Northern Basque: Leiçarraga’s (1571) translation of the New Testament aimed at a cross-dialectal language, but its basis was fundamentally



Georges Rebuschi

Labourdin, and the Basque classical text par excellence, Axular’s (1643) Gero, was written in the same dialect. Until the end of the 19th century, Basque literature was mainly religious, and mostly consisted of translations. As we shall see, this can turn into an advantage, since the translations provide us with many paraphrases of what are basically the same source texts (mainly the four Gospels, and Thomas a Kempis’ Imitation of Christ). Independently of the relative wealth and ancientness of Northern Basque in general, this paper will concentrate on this variety of Basque because in this variety complex correlative sentences (as they will be defined presently) have been attested throughout its history, and are still alive today (although they are stylistically marked as literary or rhetorical).

1.2

Why this study?

Pretheoretically, correlative structures are pervasive in natural languages. The word “correlative” can refer to (pairs of) words, as in: “Both, but not each, is also used as a correlative with and in linked predications” (Quirk et al. 1972: 364). The same authors also use “subordinating correlatives” (p. 528) and “correlative subordinators” (p. 728) indiscriminately to describe pairs like ‘if… then, (al)though… yet/nevertheless, as… so […]’. Greenbaum (1969: 120) refers to “correlative constructions” when discussing the impossible ellipsis of the subject in sentences like Although they did not like the music, yet *(they) applauded vigorously, and he uses the expression “correlative structure” to refer to an example like It’s not that he’s young; it’s just that he’s inexperienced (id.: 120).1 Many recent studies of correlative constructions have been devoted to a specific subclass of complex correlative sentences, henceforth CCSs, which are well-kown in the ancient Indo-European languages (Latin, Greek, Sanskrit, Hittite) – see e.g. Haudry (1973) for an overwiew – and are still attested in modern Indo-European languages: the Slavic languages (Izvorski 1996, Boškovič 1997, and Citko this volume) or Hindi (Srivastav 1991, Dayal 1996, Bhatt 2003), but also in languages that belong to quite distinct phyla such as Bambara (Zribi-Hertz & Hanne 1995), Burushaski (Tiffou & Patry 1995), Hungarian (Lipták 2000, 2005), and Tibetan (Cable 2005, this volume). This subclass of CCSs is characterised by a left-peripheral relative clause or protasis linked to a (possibly phonetically unrealised) pronominal correlate in the main clause that follows it, also known as the apodosis. In the examples below, the paired items are in italics. (1) a.

bambara (Zribi-Hertz & Hanne 1995) 2 Musa ye uru min san, n ye o ye Musa prf knife rel buy 1sg prf it see ‘I saw the knife that Musa bought.’



Basque correlatives and their kin 

b.

burushaski (Tiffou & Patry 1995) amenmoiNga barin écam (ka) mo gusmoina which.com words I.will.do and the woman.com are sail ayét. with walk don’t ‘Dont (take a) walk with the woman with whom I’ll speak.’

It has become customary in contemporary formal linguistics to use the phrase “correlative clauses” to refer not to the apodosis which contains the pronominal correlate, as is standard usage in the literature devoted to the ancient languages, but rather to the protasis itself. I will therefore altogether ignore that expression in this paper, and use the expression “correlative protasis” (or CorPs) to refer to the clausal material in the left periphery of the main clause. Basque CCSs are of special interest for at least three reasons.3 A first reason is that it is possible to observe some facts concerning the evolution of CCSs and of related constructions in the period of four and a half centuries, reflected in the texts to be inspected. The second reason is typological.  Whereas in all the languages mentioned above the CorP can function as a restrictor over a domain given by the correlate in the apodosis (e.g. ‘the woman’ in (1b)), in Basque the correlate does not define anything at all. Lexically headed DPs normally do not appear in the apodosis, instead we have a pronominal correlate in the form of a demonstrative. This demonstrative furthermore does not “choose” between human beings or inanimate entities: all the restrictions are provided by the CorP itself. Thus, if the (semantic) predicates of the protasis and apodosis are compatible with both human beings and things, as in (2), it is only the whelement in the protasis that reveals the nature of the entities referred to. (2) Nor / zer ere maite bait-dut, (eta) hura begira-tuko dut. who what ever love bait-‹a3e1› and dem preserve-pros ‹a3e1› Lit., ‘Whoever/whatever I like/love, and them/that I will look after.’ It follows that when the wh-phrase is not discourse-linked, (Northern) Basque CorPs are free relatives (FRs) which are generically or universally understood: they cannot denote maximal individuals, because they are concerned with open, rather than closed, sets. A final reason for the study of Basque CCSs is that, next to correlatives, all the dialects have a very distinct way of building what can generally also translate into free relatives: by using “semi-free relatives” (SFRs) to be described in 2.3. To summarise, Basque CCSs actually appear at the intersection of three families of constructions that resemble each other either in their syntax or in their semantics: (i) semi-free relatives, (ii) non-independent wh-clauses of various types, and (iii) conditional clauses which contain either an existential pronoun like ‘someone’ or ‘something’, or a negative polarity item like ‘anybody’ or ‘anything’. The purpose of this paper, next to providing a general description of Basque CCSs, is to review these three construction types and to show how they differ from CCSs.



Georges Rebuschi

1.3

Organisation of the paper

This paper is basically descriptive and is organized as follows. The general background data will be provided in Section 2, together with a description of common restrictive and appositive relatives, and of the semi-free relatives alluded to above. Section 3 will turn to the essential properties of CorPs and complex correlative sentences and will compare CorPs and semi-free relatives. Section 4 will be devoted to a systematic comparison of CorPs with other non-independent wh-clauses. Section 5 will examine “generic” conditional sentences, and observe some intriguing converging properties that they share with CCSs. Finally, Section 6 will briefly recapitulate some of the results obtained in the preceding sections.

2. Background data This section will introduce the reader to some general properties of Basque. The following topics will be handled: case-marking and basic word- and phrase-order (2.1); the various ways of marking distinct types of embedded clauses (2.2); the most common way of building relative clauses (2.3); syntactically appositive relatives (2.4); and some properties of appositive relatives and SFRs (2.5).

2.1

Some typological properties of (standard) Basque

Basque has an absolutive/ergative case system and the basic structure of the simple sentence is generally considered to be SOV, as shown in (3a-b). Note that case-endings appear on the right edge of the NP/DP, not on every element the NP/DP contains. (3) a. [Gizon gaztea]-k [liburu zaharra] irakurr-i du. man young.sg-e book old.sg read-prf ‹a3e3› ‘The young man has read the/an old book.’ b. [Gizon zaharra] etorr-i da. man old.sg come-prf ‹a3› ‘The old man has come.’ The finite verb form (FVF), which is either an auxiliary or directly inflectable lexical verb, is cross-referenced pluripersonally if the verb has more than one arguments. In the present tense, the absolutive morpheme is a prefix, and the dative and ergative agreement morphemes are suffixes. The choice of the auxiliary depends on the transitivity of the verb (and its possible dative extension), not on the choice of the participial suffix, as shown in (4). Note also that the synthetic conjugation, when it is possible, as in (4c), is semantically progressive, whereas its periphrastic or com­pound counterpart is iterative or habitual.



Basque correlatives and their kin 

(4) a. Gizona-k liburua irakurr-i/tzen/iko du. man.sg-e book.sg read-prf/impf/pros ‹a3e3› ‘The man has read / reads / will read a/the book.’ b. Haurra etorr-i/tzen/iko da child.sg come-prf/impf/pros ‹a3› ‘The child has come / comes / will come.’ c. Haurra (ba-)dator. child.sg prtc-come‹3a› ‘The child is coming.’ Owing to this rich verbal morphology, Basque is a generalised pro-drop language: subjects, but also direct and indirect objects need not be expressed by phonetically realised pronouns. The basic SOV word order in Basque can be affected by various processes, in particular those that raise various items towards the functional projections in the left periphery of the sentence. In fact, SOV as such typically occurs in totally rhematic sentences, contributing the information required by a question like Zer gertatu/gertatzen da? ‘What’s happened/happening?’. Whereas as soon as either a topic or a focussed item must appear, Basque looks much more like a “discourse configurational language”, to borrow É.  Kiss’ (1995) typological characterisation. Further, interrogative whphrases must, at least in today’s Basque4, be left-adjacent to the verb complex (the FVF if the lexical verb is inflected or the sequence participle + auxiliary otherwise), as shown in (5a-c). Focussed phrases and replies to wh-questions must occupy the same linear position as shown in (6a-c), where the small caps indicate prosodic salience. (5) a. Liburu hau, nor-k irakurr-i du? book this who-e read-prf ‹a3e3› ‘This book, who has read (it)?’ b. Nor-k irakurr-i du liburu hau? who-e read-prf ‹a3e3› book this ‘Who has read this book?’ c. *Nor-k liburu hau irakurr-i du? who-e book this read-prf ‹a3e3› (6) a. Patxi-k irakurr-i du liburu hori. Patxi-e read-prf ‹a3e3› book that ‘(It’s) Patxi (who) has read that book.’ b. Liburu hori, Patxi-k irakurr-i du. book that Patxi-e read-prf ‹a3e3› ‘That book, (it is) Patxi (who) has read (it).’ c. *Patxi-k liburu hori irakurr-i du. Patxi-e book that read-prf ‹a3e3›



Georges Rebuschi

(5c) and (6c) are ungrammatical in today’s Basque, and were highly marked in the older language, because the wh-word and the focussed element – the latter either indicating contrast or providing (new) information – do not immediately precede the verb complex.

2.2

Embedded clauses

The marking of embedded clauses is done by four distinct morphemes that function as complementiser heads and can be affixed or cliticised to the FVF. The first two of these morphemes are clearly suffixes. One, ‑(e)la, basically marks completive clauses, as in (7): (7) Patxi-k uste du [haurra-k liburua irakurr-iko du-ela]. Patxi-e belief has child.sg-e book.sg read-pros ‹a3e3›-comp ‘Patxi thinks that the child will read the book.’ The other suffix is ‑(e)n; it occurs in indirect questions (8a), in some exclamative sentences (8b), and, as we shall see in 2.3, in relative clauses which contain an abstract, or phonetically empty, relative operator. (8) a. Ez dakit [nor etorr-i d-en]. neg I-know who come-prf ‹a3›-en ‘I don’t know who has come.’

[den=da+‑en]

b. Zenbat jende etorr-i d-en! how.many people come-prf ‹a3›-en ‘How many people have come!’

(Etxepare 2003: 564)

The other items are ba- ‘if ’, which can be found with indicative mood, as in (9a), or other moods (potential, irrealis…), and bait-, which serves several purposes, among which that of marking the FVF in relative clauses that display a wh-word or phrase, cf. (9b).5 In (9c), we illustrate successively the relative-marking function of bait- on the one hand, and its function of marking that the clause is governed by specific items (hain… non ‘such/so much… that’). (9) a. Adi-tzen ba-nau, uler-tuko du. hear-impf if-‹a1e3› understand-pros ‹a3e3› ‘If s/he hears me, s/he will understand.’ b.

Ba-dire6 … kasu batzu, zeinetan barur egin prtc-are case some which.pl.loc fast make behar bai-ta. must/necessary bait-‹a3› ‘There are some cases in which one must fast.’

(Duhalde 1809: 5)



Basque correlatives and their kin 

c. Hauk dira Eskuara-ren hitzak, […] zein-eta-n these are Basque-gen words.pl which-pl-loc edirei-ten bai-tira hain […] signifikanza misteriotsuak, find-impf bait-‹a6› so meaning mysterious.pl non bai-tirurite Philosophoe-n sentenziak. that bait-they.look.like philosopher.pl-gen maxim.pl (Etcheverry of Sare, [1712] 1907: 45) ‘These are Basque words, in which such mysterious meanings are found that they resemble the maxims of philosophers […].’

2.3

‑(E)n relatives and semi-free relatives

Consider the DPs in (10b,c). The bracketed sequences correspond to sentence (3a), repeated without the adjectives (for better readibility) as (10a), with a gap in object or subject position respectively (marked by e). They are restrictive relatives which modify the nouns liburu and gizon (or, more technically, restrict the domain introduced by the translation of the nouns):7 (10) a. Gizon-a-k liburu-a irakurr-i du. man-sg-e book-sg read-prf ‹a3e3› ‘The man has read the/a book.’ b. [gizon-a-k e irakurr-i du-en] liburu-a man-sg-e read-prf ‹a3e3›-en book-sg ‘the book [that the man has read –]’ c. [[e liburu-a irakurr-i du-en] gizon-a] zuhurr-a da. book-sg read-prf ‹a3e›-en man-sg wise-sg he-is ‘[The man [that – has read the book]] is wise.’ As could be expected, the case of the dp that contains the restrictive relative is linked to the function of that dp in the higher clause. Thus, in the case of (10c), in spite of the fact that the pivot, or relativised position in the embedded clause, corresponds to that of the subject of a transitive verb, the whole DP is in the absolutive or zero case (there is no case suffix on gizon-a). On the other hand, in (11) below, the relativised position is that of the subject of an unaccusative verb (cf. the aux. da = a3, lit. ‘he is’), a DP that would be in the absolutive case if it were phonetically realised. The whole DP on the other hand is in the ergative case, because it is the subject of the transitive verb irakurri du ‘he has read it’. (11) [[ e Etorr.i d-en] gizon-a]-k liburu-a irakurr-i du. come.prf ‹a3›-en man-sg-e book-sg read-prf ‹a3e3› ‘The man that has come has read the book.’



Georges Rebuschi

The head noun in these kinds of relatives can also be elided, as in (12), where the second occurrence of gizon ‘man’ is missing, represented by means of a Ø sign in the second relative clause. (12) [[[e Liburu-a irakur-tzen du-en] gizon]-a] eta [[[e book-sg read-impf ‹a3e3›-en man-sg and astekari-a irakur-tzen du-en]Ø]-a] gazte.ak8 dira. weekly-sg read-impf ‹a3e3›-en-Ø-sg young.pl they.are ‘The man that reads the book and the one that reads the weekly are young.’ Lit., ‘… and [the – [that [– reads the weekly]]]…’ In another context, a DP like the second conjunct in (12) could also be interpreted as a free relative, i.e. a relative that does not modify any “understood” or contextually recoverable NP: (13) [Irakur-tzen du-en-Ø-a] zuhurr-a da. read-impf ‹a3e3›-en-Ø-sg wise-sg is ‘He who reads is wise.’ It must be underlined that if the bracketed sequence in (13) is interpreted as referring to human beings, this is merely an effect of the lexical choice of the verb and our knowledge of the world. With a verb like erori ‘fall (down)’, for instance, the expression is totally ambiguous between a reading referring to human entities and a reading referring to non-human ones, whether it is in the singular, or in the plural: (14) a. eror-i d-en-Ø-a fall-prf ‹a3›-en-Ø-sg ‘which has fallen’ b. eror-i dir-en-Ø-ak fall-prf ‹a6›-en-Ø-pl ‘those [±human] that have fallen’ I label this kind of headless relative clauses “semi-free relatives”, SFRs for short, “semi” because they are dominated by the functional projections characteristic of nominal expressions: a Number Phrase and a DP (see Rebuschi 2000 for arguments). For the syntactic representation of these relatives, in particular for the representation of the functional layers above the CP, I will adopt the analysis of Artiagoitia (2002) for Basque DPs. According to this there are two options concerning the insertion, or “external merge”, of the article ‑a(k): one is directly under D°, if this node carries a [+definite] feature, the other is under Num° if D° is [–definite]. In the latter case, the number suffix will later raise to D° so as to carry a possible case suffix. Transferring this analysis to SFRs, the result is (15a) or (15b). Op here represents the abstract relative operator in the specifier of the relative CP.



Basque correlatives and their kin 

(15) a. DP 2 NumP D° 2 -a NP Num° 2 [+sg] CP NP 2 Ø Op C’ 2 IP C°  -en b. DP 2 NumP D° 2 [–def] NP Num° 2 [+sg] CP NP -a 2 Ø Op C’ 2 IP C°  -enElliptical DPs stricto sensu, like the one in (12) – i.e. those that contain an empty anaphoric NP – must of course have a structure like (15a). As for SFRs, I tentatively argued elsewhere (Rebuschi 2001) that a simpler structure, in which the CP would be the direct complement of the Num(ber) head, should suffice. Yet, this is not obviously true. For instance, an SFR in the locative case has three readings: (16)

irakur-tzen du-en-Ø-e-an read-impf ‹a3e3›-en-(Ø-)sg-loc (i) anaphoric interpretation: ‘in the (one) that s/he reads’ (ii) generic FR interpretation over objects: ‘in what s/he reads’ (iii) generic FR interpretation over time points or intervals: ‘when s/he reads’

One may therefore infer that an empty NP is always present, which minimally possesses either a [+entity] feature, as in (i) and (ii), or a [+time] feature, as in (iii).

 Georges Rebuschi

2.4

Appositive relatives

A different kind of relative clause, which Basque has always displayed9, is introduced by the wh-phrase zein- ‘which’. This kind of relative clause uses the bait-complementiser illustrated in (9b,c) above. Already in the 16th century, the wh-word could be left unpronounced (a stylistically preferred variant today, but a much rarer option in the past). Whether their wh-element is realised or not, bait- relative clauses always follow the “modified” nominal expression: they are either simply adjoined to a case-marked DP, like gure eskuara(k) in (17a), or extraposed to the right of the clause that contains the relevant DP, as in (17b). (17) a.

Lege hau segi-tu du gure eskuara-k, law this follow-prf ‹a3e3› our Basque.language-e zein-a hanbat mende-z mantena-tu eta which‑sg so.many century-inst maintain-prf and mantena-tzen bai-ta. (Etcheverry of Sare [1712] 1907: 23) maintain-impf bait-‹a3› ‘Our Basque language, which has defended itself for so many centuries, has conformed to this law.’

b.

Eneko, eta Semeno Nafarroa-ko Errege iza-n ziren, Eneko and Semeno Navarre-gen king be-prf ‹a1[pst]› zein-e-n izene-k aski klarki aditzera zein-pl-gen name.pl-e enough clearly to.understand emai-ten bai-tarokute Eskualdun izenak dire-la. give‑impf bait-‹a3d3e6› Basque name.pl they.are-comp ‘Eneko and Semeno were kings of Navarre, whose names let sufficiently clearly understand that they are Basque names.’ (id.: 34)

Owing to the “purification” of the language advocated at the end of the 19th century, however, this type of relatives tends to be rejected as a Romance calque today, especially when used with an intended restrictive meaning. In the Central and Western dialects, the complementiser ‑(e)n (or its variant ‑(a)n), mentioned in 2.3 above, was, and still is, used in wh-relative clauses. Appositive relatives can be adjoined to a demonstrative pronoun (with the same interpretation as elliptical ‑en.a(.k) expressions in some contexts). In the Labourdin dialect, from the 16th century to the end of the 19th, a special pronoun, haina, lit. ‘the such’ was used instead of a demonstrative when the intended meaning was that of a “real”, non-anaphoric free relative, cf. (18b). This paraphrases the SFR in (18a) (more on haina in 3.1).



Basque correlatives and their kin 

(18) Mt 11,27 – LV: ‘Et nemo novit Filium nisi Pater, neque Patrem quis novit, nisi Filius et cui voluerit Filius revelare.’ / KJV ‘And no man knoweth the Son, but the Father; neither knoweth any man the Father, save the Son, and he to whomsoever the Son will reveal him.’ a. Eta nihor-k ez du ezagu-tzen Semea and nobody-e neg ‹a3e3› know-impf Son.sg Aita-k baizen, eta nihor-k ez du ezagu-tzen Father-e but and nobody-e neg ‹a3e3› know-impf Aita [[Semea-k baizen] eta [[Semea-k Father-sg Son.sg-e but and Son.sg-e erakutsi iza-n nahi dio-en-a-k] baizen]]. show aux-prf want ‹a3d3e3›-en-sg-e but (Haraneder 1740) b.

eta haina-k zein-a-ri Seme-a-k nahi iza-n and such.sg-e zein-sg-d son-sg-e want aux-prf bai-tio ezagut-araz-i. (Duvoisin 1859–65) bait-‹a3d3e3› know-fact-prf Lit., ‘[…] and the-such to whom the Son has wanted to reveal  (him).’

As (19) shows, the appositive relative could also be extraposed in such cases. (19) Imit. 1,25,3 – Thomas’s original Latin text: Illi maxime præ aliis in virtutibus proficiunt, qui ea quæ sibi magis gravia et contraria sunt vincere nituntur. Benham’s (1905) translation: ‘Nevertheless they advance above all others in virtue who strive manfully to conquer those things which are most grievous and contrary to them.’ Bertzenaz hainak gehienean berthutea-n aitzina-tzen dire, otherwise such.pl mostly virtue.sg-loc advance-impf ‹a6› zein-ak animo gehiago-rekin ensaia-tzen bai-tire who/which-pl courage more-com try-impf bait-‹a6› traba-rik handi-en-e-n […] garaitzerat (Chourio 1720) obstacle-prtc big-most-pl-gen to.surmount

2.5

Further properties of appositive relatives and SFRs

SFRs can be right-adjoined to a DP and can be coordinated with appositive relatives introduced by a wh-phrase (zoin- in (20a) is a dialectal variant of zein-): (20) a.

Mundu hun-en konprenitzeko, behar da izaite world this-gen to.understand necessary is being bat, bethi-tik d-en-a, nihor-k egi-n ez one always-abl is‑en-sg nobody-e make-prf neg



Georges Rebuschi



du-en-a, eta zoin-a-k egin bai-titu ‹a3e3›-en-sg and which-sg-e made bait-‹a6e3› dire-n guzi-ak. (Lapeyre [1891] 1982: § 36) are-en all-pl ‘In order to understand this world, we need a being that has always been, that nobody has made, and that has made everything that is.’

b.

Asko ezagu-tu dituzket ni baino gazte-ago-ak, ni-k many know-prf ‹a6e1› 1sg than young-more-pl I-e bezain luzeki bizitzeko iduri ederr-a as-much long to.live appearance beautiful-sg zu-te-n-ak, eta zein-ak ez-pai-tire orai they-had-en-pl and zein‑pl neg-bait-are now mundu-a-n. (Duhalde 1809: 121) world-sg-loc ‘I have known many people younger than I who looked as if they  were to live as long as myself, and who are no longer in this  world.’

Such examples show that the interpretation of both types of clauses must sometimes be restrictive rather than appositive. However, I shall stick to the label “appositive relatives” to reflect their syntactic structure, i.e. the fact that they are undeniably adjoined to a case-marked DP, rather than c‑commanded by D°. Interestingly, fully case-marked SFRs can be conjoined and still denote a single (possibly maximal) individual, as shown by (21) where, baina(n) ‘but’, just as the source text(s),10 blocks a different reading. (21) Mt 10,28 – LV: ‘Et nolite timere eos, qui occidunt corpus, animam autem non possunt occidere; sed potius eum timete, qui potest et animam et corpus perdere in gehenna.’ KJV: ‘And fear not them which kill the body, but are not able to kill the soul: but rather fear him which is able to destroy both soul and body in hell.’ a. Eta et-zaretela beldur gorputz‑a hil-tzen and neg-be[imp5’] afraid body-sg kill-impf dute‑n‑en, eta arima ezin hil ‹a3e3›‑en-pl-gen and soul.sg neg √kill dezakete-n-e-n: baina aitzitik zareten ‹a3e6[pot]›‑en‑pl‑gen but on.the.contrary be[imp5’] beldur arima eta gorputza gehena-n gal afraid soul.sg and body.sg hell-loc √ruin/lose ahal ditzake-n-a-ren. (Leiçarraga 1571) can ‹a6e3[pot]›‑en‑sg‑gen



Basque correlatives and their kin 

b.

Ez duzuela beldurr-ik gorphutz-a hil‑tzen neg have[imp] fear-part body.sg kill-impf dute-n-e-z eta arima ezin hil ‹a3e6›‑en-pl-inst and soul.sg cannot kill dezakete-n-e-z: bainan zarete-n beldur ‹a3e6[pot]›‑en-pl-inst but be[imp] afraid hobeki bai arima eta bai gorputza ifernua‑n better prtc soul.sg and prtc body.sg hell.sg-loc gal-araz detzake-n-a-z. (Haraneder 1740) √lose-fact ‹a6e3[pot]›‑en-sg-inst

c.

Ez izan beldur-rik [[gorputza hil-tzen dute‑n‑e‑i] neg √have fear-part body.sg kill-impf ‹a3e6›‑en-pl-d baina [arima ezin hil dezakete-n-e-i]]. (“eheg” 1980)11 but soul.sg cannot √kill ‹a3e6[pot]›‑en-pl‑d

(Note that neither qui in Latin, nor them which (=those who) in the King James version are repeated in the second conjunct.) Returning to the discussion of the structural properties in (15), we can now say that SFRs are contained in a DP whose head is [±definite]. If it is [‑definite], the interpretation will yield a property (an option that is compositionally compatible with SFRs that are interpreted as restrictive). If the head is [+definite], it will provide either a maximal individual, if the given set is closed, or a generalised quantifier (with universal or generic force), if the set is open. When they are not interpreted as properties, SFRs can appear in argumental position, cf. (22), be focussed, cf. (23), or left-dislocated, cf. (24). (22) a. Erakaslea-k [huts egi-nen du-en-a] zigor-tuko du. teacher.sg-e mistake make-pros ‹a3e3›-en-sg punish-pros ‹a3e3› ‘The teacher will punish [the one who will make a mistake].’ b. [Huts egi-nen du-en-a] zigortua iza-nen da. mistake make-pros ‹a3e3›-en-sg punished.sg be-pros ‹a3› ‘The one who will make a mistake shall be punished.’ (23) [Huts egin-en du-en-a] izan-en da zigortua. mistake make-pros ‹a3e3›-en-sg be-pros ‹a3› punished.sg ‘(It’s) The one who will make a mistake (that) shall be punished.’ (24) [Huts eginen duena], hura zigortua izanen da. … dem … Lit., ‘The one who will make a mistake, that (one) shall be punished.’ Zein(a) + bait- appositive clauses on the other hand cannot appear in such positions as these. Interestingly, they also very rarely appear under the scope of an existential predicate



Georges Rebuschi

like ba-da ‘there is’ (or ba-dire ‘there are’ as in (9b)). When they do, they are very strongly felt as Latin calques. SFRs, on the other hand, are perfect in such contexts.12

3. Basque correlative protases and complex correlative sentences 3.1

Introduction: “nor/zer… bait-” free relatives

In addition to the relatives described above, there exists yet another type of free relative, which is to be found in the correlative protasis proper (CorP) of complex correlative sentences (CCSs), already illustrated in (2) above. The following excerpts from the NT illustrate CorPs, and CCSs as a whole, across four and a half centuries of Northern Basque texts. These examples clearly indicate, in comparison with the source texts (the Latin Vulgate for Haraneder and Duvoisin), that the CCSs have a universal flavour to them. (25) Mt 12,50 – LV: ‘Quicumque enim fecerit voluntatem Patris mei, qui in caelis est, ipse meus frater et soror et mater est’ / KJV: ‘For whosoever shall do the will of my Father which is in heaven, the same is my brother, and sister, and mother.’ Ezen nor-k ere egi-nen bai-tu ene Aita for who-e ever do-prosp bait-‹a3e3› my father zeru-e-tan d-en-a-ren13 borondatea, hura da ene heaven-pl-loc is-en-sg-gen will.sg dem is my anaye, eta arreba, eta ama. brother and sister and mother (Leiçarraga 1571) (26)

Mt 10,22 – LV: ‘Qui autem perseveraverit in finem, hic salvus erit.’ KJV: ‘… but he that endureth to the end shall be saved.’ Bainan nor-k ere irau-nen bai-du akhabantza-raino, but who-e ever last-pros bait-‹a3e3›14 end-until eta hura iza-nen da salbatua. (Haraneder 1740) and dem be-prosp ‹a3› saved.sg

(27) 1 Jn 2,10–11 – LV ‘Qui diligit fratrem suum, in lumine manet… qui autem odit fratrem suum, in tenebris est…’ / KJV ‘He that loveth his brother abideth in the light… But he that hateth his brother is in darkness.’ Nor-k ere maite baitu bere anaia, haina who-e ever love bait-‹a3e3› his brother such.sg argia-n dago, […] bainan nor-k ere higuin baitu light.sg-loc is but who-e ever hate bait-‹a3-e3› bere anaia, ilhunbe-tan da haina. his brother darkness-loc is such.sg (Duvoisin 1859–65)



Basque correlatives and their kin 

Needless to say, these structures (which do not translate a Latin correlative in (25)), are not only attested in translations, but also in original texts. (28a) is from the first book ever printed in Basque, and (28b) from the late 18th century. (28) a.

Ni-k nor-gatik pena bait-ut, hark ene eztu I-e who-for pain.sg bait‑I.have dem-e of.me neg.has axol-ik. (Etchepare 1545: VII, 20) concern-part ‘He/she/the one because of/for whom I suffer does not worry for  me.’

b.

Norat ere arbola eror-tzen bai-da, han bethi whither ever tree.sg fall-impf bait-‹a1› there always ego-nen da; zer ere estatu-tan hil-en stay-pros ‹a3› what ever state-loc die-pros bai-zare, hala eternitate guzia-n iza-nen zare. bait-‹a5› thus eternity all.sg-loc be-prosp ‹a5› ‘Wherever a tree falls, it will stay there; in whatever state you will  die, you will remain in the same state in all eternity.’ (Baratciart 1787: 156)

The protasis of each of the sentences in (25)-(28) contains an interrogative wh-word, nor who’ and zer ‘what’. As has been pointed out above in 2.4, and as the next section will show, these wh-phrases cannot be used in appositive relatives. The apodosis of CCSs contains a pronominal correlate, either a demonstrative, cf. hura ‘that (one)’ or its manner counterpart hala ‘thus’ or haina, the special item alluded to above. Haina had most of the properties of what could be called a lexicalisation of an E-type pronoun: it could not be used deictically, could not anaphorically refer to an entity, and could not be c‑commanded by (even) a non-referential antecedent in an argumental position (see Rebuschi 1998 for details). Yet another noteworthy feature of CCSs, illustrated in (26), is that the protasis and the apodosis can be “linked” by what is otherwise an ordinary conjunction, eta ‘and’. In fact, eta in this context is very frequent with some authors, and fairly rare with others. Most examples of CCSs given by Oyharçabal (2003: 818–821) contain eta. When eta is present, the correlative demonstrative is often found focussed in a position right-adjacent to eta. Eta may cliticize to the correlative protasis, an independently attested property of eta as a real conjunction, cf. Rotaetxe Amusategi (2004).

3.2

Are there “nor-/zer-…bait-” relatives in argumental position?

Some nor-/zer- free relatives apparently appear in the right periphery of a sentence. For illustration, consider (29a), which roughly repeats the contents of Mt 11,27 in (18). In this example, a wh-relative seems to be coordinated with a DP, Semeak ‘the Son-e’. But such constructions should not be dealt with independently of the paraphrase provided by (29b), which is much preferred stylistically today.



Georges Rebuschi

(29) Lk 10,22 – LV: ‘[…] nemo scit […] qui sit Pater, nisi [[Filius] et [cui voluerit Filius revelare]].’ / KJV: ‘[…] no man knoweth who […] the Fatheri is, [[but the Sonj], and [he to whom the Son will reveal himi]].’ a. Nehor-k ez-taki […] nor d‑en Aita, [[Semea-k baizen], nobody-e neg-knows who is‑en Father Son.sg-e but eta [nor-i ere Semea-k manifestatu nahi uka‑nen and who-d ere Son.sg-e impart want aux‑pros bai‑trauka]]. (Leiçarraga 1571) bait-‹a3d3e3› Lit., ‘… and to-whomever the Son will want to reveal (him).’ b.

Nihor-k ez daki […] nor d-en Aita [[Semea-k nobody neg knows who is-en Father Son.sg-e baizen] eta [nor-i ere Semea-k agertu nahi-ko but and who-d ever Son.sg-e reveal want-pros bai-dio, eta har‑k baizen]] (Haran. 1740) bait-‹a3d3e3› and dem-e but Lit., ‘…[[but to the Son] and [but to whom the Sonj will want to  reveal himi]k and hek]]]’

Constructions like (29b) are also attested in original texts, cf. (30): (30)

(Munduko gauza guztiek kondenatzen dute alferra.) … Bertze guztie-k enplega-tzen [d]ute bere denbora, other all.pl-e employ-impf bait-‹a3e6› their time.sg [zer-tako egin-ak bai-tira, har-tan] what-for made-pl bait-are dem-loc (Ax. [1643] 1968: § 241) ‘(All the things in the world condemn lazy people.) […] apart from hem, everybody else spends their time in that for which they are/have been created.’ (Lit., ‘they spend their time for what they have been made, in that.’)

Importantly, the second conjunct in (29b), [nori ere Semeak agertu nahiko baidio, eta hark baizen] also contains eta ‘and’, just like (26) above. As for its structural position, it it very difficult to imagine that this linker could be inside a nominal expression consisting of a nor-free relative on its left, and a demonstrative on its right, not to mention the pause. Things become clearer once we assume that (30) is derived via movement from a structure like (31a): all the material that follows the CorP and the pronominal correlate hartan, between square brackets, raises to the immediate left of the wh-word zertako ‘what-for’ to yield (30). In (31a), which represents the relevant structure before movement, the landing site, marked by a dash, can be taken to be the specifier of one of the iterable Topic Phrases of Rizzi (1997). (31b) represents the structure after movement: it crucially contains a trace at the end (see Rebuschi 2006 for details).



Basque correlatives and their kin 

(31) a.

Bertze guztie-k [[CorP zer-tako eginak bai-tira], other all.pl-e what-for done.pl bait-are [har-tan [enplegatzen dute bere denbora]]]. dem-loc use-impf ‹a3e6› their time.sg

b. Bertze guztie-k [enplega-tzen dute bere denbora]i other all.pl-e use-impf ‹a3e6› their time.sg [[CorP zer-tako egin-ak bai-tira], [hartan ti]]. what-for done-pl bait-are dem.loc Returning to (29b) now, the predicate ez daki nor den Aita ‘does not know who the Father is’ must simply be assumed to simultaneously bind two empty positions, following the two occurrences of baizen ‘except’. This is a clear case of an ATB phenomenon, understandable if the conjuncts are not two DPs, but two clauses, just as the string zertako eginak baitira, hartan in (30) is in fact a clausal structure with a gap, as shown in (32a). Likewise, the correct representation of the relevant part of (29a) must be assumed to be as in (32b), since neither eta nor the pronominal correlate are obligatory. (32) a.

Nihor-k [ez daki nor d-en Aita]i [[Semea-k baizen ti] nobody neg knows who is-(e)n Father Son.sg-e but eta [nor-i ere Semea-k agertu nahi-ko bai-dio, and who-d ever son.sg-e reveal want-pros bait-‹a3d3e3› eta har-k baizen ti]]. and dem-e but

b.

Nehor-k [ez-taki nor d-en Aita]i, [[Semea-k baizen ti], eta nobody-e neg-knows who is-en Father Son.sg-e but and [nor-i ere Semea-k manifestatu nahi ukan-en bai-trauka ti]].15 who-d ever Son.sg-e impart want aux[tr]-prosp bait-‹a3d3e3›

3.3

Differences between CorPs and SFRs

Taking into account the fact that all the Basque dialects easily topicalise SFRs, and that even in Northern Basque, the latter are less marked stylistically, it is natural to find translations which display either construction for the same original text, a fact illustrated in (33). (33) Mt 10,40 – LV: ‘Qui recipit vos, me recipit.’ / KJV: ‘He that receiveth you receiveth me.’ a. Zu-ek16 rezebi-tzen zaituzte-n-a-k, ni rezebi-tzen nau. you[5’] receive-impf ‹a5’e3›-en-sg-e me receive-impf ‹a1e3› (Leiçarraga 1571)



Georges Rebuschi

b.

Nor-k ere zu-ek har-tzen bait-zaituzte, who ever you[5’] receive-impf bait-‹a5’e3› eta har-k ni har-tzen nau. and dem-e me receive-impf ‹a1e3›

(Haraneder 1740)

c. Zu-ek har-tzen zaituzte-n-e-k, ni naute har-tzen. you[pl] receive-impf ‹a5’e6›-en-pl-e me ‹a1e6› receive-impf (Duvoisin 1859–65) However, the fact that CorPs can be paraphrased by SFRs does not constitue evidence that they have much in common syntactically: their internal structure is obviously quite different, as the reader can also ascertain. First, CorPs have an explicit wh-word or phrase in their left periphery, whilst SFRs lack such phrases. They may only have an abstract operator. Second, the complementiser affix associated with the FVF is distinct in the two cases. While SFRs use ‑(e)n, CorPs feature bait-. The third difference affects syntactic composition. As noted in 2.3. above, SFRs display a Number and D projection (cf. (15)). In the case of CorPs, there is no decisive argument to hypothesize that anything dominates the CP. This difference gives rise to differences when it comes to quantification. As (34a,b) shows, SFR can be modified by the universally quantifiers guzia (which is always marked for number and sits in the head of the DP), or oro (which is not marked for number, and is right-adjoined to the DP). Such items can never follow a CorP. (34) Imit. 3,47,1 – Thomas: Modicum est, et breve omne quod transit cum tempore. Benham: ‘Little and short is all that passeth away with time.’ a. Aphurra eta laburra da denbora-rekin iragai-ten de-n guzia. little.sg and short.sg is time.sg-com pass-impf ‹a1›-en all.sg (Pouvreau 1669) b.

Egun-e-kin bat-ean itzal-tzen dir-en-ak oro, day-pl-com one-loc disappear-impf ‹a1›-en-pl all gauza guti dira eta guti dute irau-ten. thing little are and little ‹a3e6› persist-impf

(Léon 1929)

While CorPs themselves cannot be so quantified, it must be noted that their pronominal correlates in the apodosis can be universally quantified, as shown in (35a,b). (35) a.

Derradan huts-ik egin gabe zer ere say[subj]‹a3e1› fault-part make without what ever eskiriba-tuko baita euskara-z, hura guztia, write-prosp bait.‹a3› Basque-inst dem all.sg



Basque correlatives and their kin 



euskaldun-e-n buruzagi bezala, zu-ri dagotzu-la. Basque-pl-gen leader as you-d pertains‹a3d5›-C° ‘Let me say without making any mistake that whatever will be written in Basque, all that is due to you, as leader of the Basques.’ (Axular [1643] 1964: § 5) b. Mt 7,12 – LV: ‘Omnia ergo, quaecumque vultis ut faciant vobis homines, ita et vos facite eis.’ / KJV: ‘Therefore all things whatsoever ye would that men should do to you, do ye even so to them.’ Zer ere beraz nahi bai-tuzue egin what ever then want bait-‹a3e5’› √do diezazuete‑n zu-e-i gizone-k, eta he‑k ‹a3d5’e6[subj]›-C° you[5’]-d man.pl-e and dem-pl guzi-ak egin diotzatzuete zuek ere h‑e‑i. all-pl √do ‹a3d6e5’pl[imp]› you[5’].e too dem-pl-d (Harriet 1855) The possibility of such quantification suggests that in these cases the CorPs must be interpreted as a property. Were it not a property, the ban on vacuous quantification would not be respected. Turning now to differences between syntactic positions SFRs and CorPs can occupy in the sentence, there are also two salient ones. First, as was remarked in 2.3., SFRs, due to the fact that they contain extended nominal projections, can appear in any position where ordinary DPs occur. In particular, they can be found in argumental positions, as (non-topicalized) subjects or objects. They can also occur in the CP domain, for example in ordinary topic and focus positions, cf. (21)-(24) above. Correlative protases on the other hand cannot be focussed: if their content must be focussed, it is the pronominal correlate which occupies the focus position in the left periphery (see Rebuschi & Lipták (to appear) for examples and discussion). As a second difference it must be noted that the presence of eta lit. ‘and’ as a linking element between relative and main clause is only attested with correlatives (cf. 26). When they are leftperipheral, SFRs cannot be linked to the main clause by the word eta. Before closing this section, it must be noted that although SFRs and CorPs are clearly distinct syntactically, the two types of clauses can be – quite unexpectedly – coordinated with each other (cf. (20)). Here again, the conjoined phrases do not yield two entities, or two generalised quantifiers, but two properties narrowing down the relevant domain. Now consider (36), from the Easternmost Northern dialect, Souletin Basque. The first domain is provided by a CorP, and is next restricted by an SFR, whilst

 Georges Rebuschi

in (37) (in Baztanese, spoken in Spanish Navarre, but very close to Labourdin Basque), it is the opposite situation that obtains: (36) Imit. 1,20,4 – Thomas: O, qui omnem vanam sollicitudinem amputaret, et duntaxat […] divina cogitaret, […] quam magnam pacem et quietem possideret. / Benham: ‘O how great peace and quiet should he possess, who would cast off all vain care, and think only of […] divine things […]!’ O [[nur-k ere khen be-litzake arranküra banuak Oh who-e ever √remove bait-‹a6e3› worry vain.pl oro], eta [zelüko … gaize-tan baizik gogua ezar-tzen all and heavenly thing.pl-loc but mind.sg put-impf e-li-an-a-k]], ala be-iliro har‑ek bake neg-‹a3e3›-en-sg-e prtc bait-‹a3e3[pot]› dem-e peace eta phausü handi bat goza! (Maister 1757) and rest big one √enjoy (37) Mt 5,19 – LV ‘Qui ergo solverit unum de mandatis istis minimis et docuerit sic homines, minimus vocabitur in regno caelorum […]’ kjv: ‘Whosoever therefore shall break one of these least commandments, and shall teach men so, he shall be called the least in the kingdom of heaven […]’ Orrengatik, [[nor-k ere aus-ten bai-tu manamendu for-that, who-e ever break-impf bait-‹a3e3› commandment o-tarik ttipiena], eta [ola gizone-i those-abl smallest.sg and thus man.pl-d erakus-ten du-en-a]], soil ttarr-a dei-tu-a iza-in teach-impf ‹a3e3›-en-sg mere small-sg call-prf-sg be-pros da (pro) zeruetako erreinua-n […]. (Echenique, c. 1857) ‹a3› heavenly kingdom.sg-loc Lit., ‘Who[sg] ever breaks… and the – that teaches[sg], (pro[+sg]) will[sg] be called…’ Such examples of syntactically unbalanced coordination clearly show that CorPs and SFRs must share one important feature, and, as was suggested in 2.5 for coordinated SFRs and appositive relatives, this feature may well be that both types of structures can denote properties.

4. Differences between CorPs and other dependent wh-clauses Up to now, we have identified embedded questions (cf. 8a) and appositive relatives as non-independent wh-clauses distinct from CorPs. Leaving aside exclamatives (cf. 8b), there are three more types of constructions that are semantically/syntactically similar



Basque correlatives and their kin 

to correlatives: indefinite free relatives, unconditional clauses and interrogatives. This section will review these types of constructions in turn. It will be shown that correlatives differ from all these. The section will close with a recapitulation that summarizes all the differences, together with the differences discussed in the previous section, concerning relative clauses.

4.1

Indefinite free relatives

The expression “indefinite free relatives” (henceforth IFR) has been introduced by Caponigro (2001) and Grosu (2004). It refers to wh-clauses found in existential contexts. The wh-clause can be preceded by existential ‘be’ or its two-place equivalent ‘have’. The following examples are borrowed from Caponigro (2001): (38) a. C’è [chi dice sempre di sì]. there’s who says always of yes ‘There’s somebody/people who always says/say yes.’

(Italian)

b. Est’ [s kem pogovorit’]. is with whom to.talk ‘There is somebody with whom one could talk.’ c. Van [kivel beszélni]. is who-com to.talk ‘There is/are someone/people to talk to.’

(Russian)

(Hungarian)

(39) a. Non aveva [dove nascondersi in caso di pericolo]. (Italian) not had where to.hide.himself in case of danger ‘He didn’t have a place/places where he could hide in case of danger.’ b. Toj ima [s kogo da govori]. he has with whom prtc talk.3sg ‘He has somebody to talk to.’ c. eyn li [im mi le-daber]. not.is to.me with who to.talk ‘I don’t have anybody to talk to.’

(Bulgarian)

(Modern Hebrew)

Basque displays the same possibilities, as illustrated by the following excerpts. The respective verb forms are either finite, and can therefore receive a C° affix, cf. (40)-(42), or non-finite, cf. (43). In the latter case the absence of any auxiliary blocks the appearance of aspectual suffixes on the lexical verb’s root (represented by ‘√’ in the glosses), and the presence of comps like bait- or ‑en. (40) Mat 8,20 – LV: ‘Filius autem hominis non habet ubi caput reclinet.’ KJV: ‘but the Son of man hath not where to lay his head.’

 Georges Rebuschi



Baina gizona-ren Semea-k ez-tik but man.sg-gen son.sg-e neg-has[alloc]17 [non bere burua reposa deza-n]. where his head. sg √lay ‹ae3[subj]›-en

(Leiçarraga 1571)

(41) Imit. III,3,4 – Thomas: Qui habet verba mea et spernit ea, habet qui judicet eum in novissimo die. / Benham (III,3,5): ‘He who hath My words and rejecteth them, hath one who shall judge him at the last day.’ Nor-k ere adi-tzen bai-titu ene hitz-ak eta who-e ere hear-impf bait-‹a3e3› my word-pl and ez-pai-tu he-taz kontu-rik egi-ten, ba-du neg-bait-‹a3e3› dem-inst heed-part do-impf prtc-has [nor-k hura juiea-tuko du-en azken egun-ean]. who-e dem judge-prosp ‹a3e3›-en last day-sg.loc (Pouvreau 1669) (Lit., ‘He who hears my words and does not pay attention to them has who will judge him on the last day.’) (42) Jn 12,48 – LV: ‘Qui […] non accipit verba mea, habet qui iudicet eum. / KJV: ‘He that receiveth not my words, hath one that judgeth him.’ ene hitzak onhes-ten ez ditu-en-a-k ba-du my words receive-impf neg ‹a6e3›-en-sg-e prtc-has [nor-k juia-turen bai-tu]. who-e judge-pros bait-‹a3e3› (Duvoisin 1859–65) (43) [same verse as in (40)] a. Gizona-ren Semea-k ez du [non burua pausa]. man.sg-gen Son.sg-e neg has where head.sg √put Lit., ‘But the Son of Man does not have where to put/rest (his) head.’ (Haraneder 1740) b. Gizona-ren Semea-k ez du … [non etzan burua] man.sg-gen Son.sg-e neg has where √lay head.sg (ditto) (Duvoisin 1959–65) In today’s Basque, structures like (43), where the verb is unmarked for aspect and the auxiliary is elided, are preferred, although the subjunctive of (40) and the future (indicative) of (41) and (42) are still acceptable.18 When we compare IFRs to CorPs, we can notice that their internal properties, albeit somewhat similar, are still not completely identical. IFRs allow the mood to be subjunctive rather than indicative as shown above19 and they also differ with respect to the presence of the particle ere, which has been translated into ‘ever’ in this sort of context, but which may also mean ‘even’ and ‘also/too’. While this particle characterizes CorPs (see Section 3), the reader will have noticed that ere is absent in all of the foregoing examples of IFRs. It needs to be noted that ere is not completely ruled out



Basque correlatives and their kin 

here: my informants do not reject the presence of this particle in (44), for example. Rather, they find it slightly degraded: (44) Gai hor-taz, ez dut zer-taz (?ere) mintza. matter dem-inst neg I.have what-inst ere √speak ‘I do not have anything to say on that matter.’ (Lit., ‘On that subject, I don’t have what(ever)-about (to) speak.’) Returning to the first point, note that the ellipsis of the aspectual suffix and the associated auxiliary is always possible in both correlative protases and apodoses – and is even the main, if not sole, possibility of forming (comparative) CCSs in central, Guipuzcoan, Basque. See (45), which is quite natural in Northern Basque too: (45) Nola bizi, hala hil. how √live thus √die ‘One dies the way one has lived.’

(Zavala 1985: II, 70, #2223)

IFRs and CorPs are also quite different from the point of view of external properties. When it comes to syntactic position, IFRs must appear post verbally so as to be in the scope of the existential verbs ‘be’ and ‘have’, while CorPs are by definition located in the left periphery of the main clause. IFRs cannot be paraphrased by SFRs either, although the latter, contrary to CorPs, can also appear post verbally.

4.2

Unconditional clauses

The phrase “unconditional clauses” is, as far as I know, due to Zaefferer (1990, 1991). This nice term refers to “no matter…” clauses, i.e. clauses that explicitly express the idea that their content is, contra what could be expected, not relevant to the truth value of the apodosis (cf. Izvorski 2000, who labels them “free adjunct free relatives”, and Tredinnick 2003). They are relevant to our concerns for at least two reasons. In the earliest centuries of the history of Basque, unconditional sentences always had the same internal form as correlatives, see (46), or (47), which dates back to the late 19th century. These sentences are still judged acceptable today. (46) Rom 2,1 – LV: ‘Propter quod inexcusabilis es o homo omnis qui iudicas […].’ / KJV: ‘Therefore thou art inexcusable, o man, whosoever thou art that judgest […].’ a. Halalotz eskusa gabe aiz o gizona therefore excuse without thou.art o man.sg nor ere bait-aiz bertzeak juieatzen ditua-n-a. who ever bait-thou.art other.pl judge.impf ‹a6e2›-en-sg (Leiçarraga 1571)

 Georges Rebuschi

b. Hargatik zare ezin-esenkusatuzkoa, o gizona, because.of.that you.are inexcusable.sg o man.sg nor ere bait-zare bertze-ak jujea-tzen ditutzu-n-a. who ere bait-you.are other-pl judge-impf ‹a6e5›-en-sg (Haraneder 1740) (47) Nor ere bait-zare, […], barkha zadazu ene bizitasuna. who ever bait-you.are √forgive ‹a3d1e5[imp]› my briskness.sg ‘Who ever you are, forgive my briskness.’ (Hiriart-Urruty 1891) Paraphrases of these clauses appeared in the 18th century (today, they are less marked stylistically as rhetorical); they contain a free choice indefinite, and their FVF is in the subjunctive mood, as in (48). (48) Rom 2,1 [same NT verse as in (46)] a. Hargatik, o gizon juiatzen hari haiz-en-a, because.of.that o man judging busy thou.art-en-sg nor-nahi izan hadien, ez haiz aitzakia-dun. anybody be ‹a2[subj]› neg thou.art pretext-having (Lit., ‘Therefore, o man that judges [others], be you anybody, you have no excuse.’) (Duvoisin 1859–65) b. …zu-k, juje-a-ren-a egi-ten duzu-n-a-k, you-e the-Ø-of-the-judge-sg do-impf ‹a3e5›-en-sg-e nor-nahi izan zaitez-en, estakururik ez duzu. anybody be ‹a5[subj]›-en pretext.part neg you.have (Etchehandy 1999) The free-choice indefinites of subjunctive unconditional sentences are formed with a wh-word or root (belonging to the “interrogative” paradigm, not the relative one), either followed by nahi lit. ‘want’, or preceded by edo (conjunctive) ‘or’. The case suffixes always follow the wh-word in edo-… forms, but appear either on the wh-item, or on nahi, and even sometimes on both, according to the subdialects, e.g., in the ergative: edo-nor.k / nor.k-nahi / nor-nahi.k / nor.k-nahi.k (more complex forms can also be found). Unconditionals can also be found as paraphrases of CorPs and left-peripheral SFRs. This possibility is illustrated in (49a-c), where the similarity of meaning is underlined by the parallelism in form both in the Elizabethan authorised version and in the other Basque translations given.20 (49) Mt, 12,32 – LV: Et quicumque dixerit verbum contra Filium hominis, remittetur ei; qui autem dixerit contra Spiritum Sanctum […]. / KJV: ‘And whosoever speaketh a word against the Son of man, it shall be forgiven him: but whosoever speaketh against the Holy Ghost, it shall not be forgiven him […].’



Basque correlatives and their kin 

a. Eta nor ere mintza-turen bai-ta gizona-ren and who ever speak-pros bait-‹a3› man.sg-gen Semea-ren kontra, barka-turen zaio har-i; Son.sg-gen against forgive-prosp ‹a3d3› dem-d baina nor ere mintza-turen bai-ta Spiritu but who ever speak-pros bait-‹a3› ghost saindua-ren kontra, et-zaio barka-turen har-i. holy-gen against neg-‹a3d3› forgive-prosp dem-d (Leiçarraga 1571) Lit. ‘And who(ever) will speak against…, it shall be forgiven to that one; but who(ever) …, it shall not be forgiven to that one.’ b.

Eta nor-nahi mintza dadien gizona-ren Semea-ren and who-want √speak ‹a3[subj]› man.sg-gen Son.sg-gen kontra, barkhakizun da Ø; bainan Izpiritu saindua-ren against forgivable is pro but ghost holy.sg-gen kontra mintza-tu daite-n-a-ri et-zaio […] against speak-pros ‹a3[fut]›-en-sg-d neg-‹a3d3› barkha-turen. (Duvoisin 1859–65) forgive-pros (lit., ‘And [should] anyboby speak against the Son of man, he/it can be forgiven; but to who(m)ever has spoken against the Holy Spirit, it shall not be forgiven.’)

c.

Gizona-ren Semea-ren kontra mintza-tuko d-en-a-ri man.sg-gen son.sg-gen against speak-pros ‹a3›-en-sg-d barkatua izan-en zaio. Izpiritu Saindua-ren kontra forgiven.sg be-pros ‹a3d3› Ghost Holy.sg-gen against mintza-tuko d-en-a-ri aldiz et-zaio speak-pros ‹a3›‑en-sg-d however neg-‹a3-d3› barka-tu-a izan-en […] (Léon 1946) forgive-prf-sg be-pros Lit., ‘To the (one) that will speak…; to the one that will speak…’

4.3

Appositive relatives, CorPs and interrogatives: The wh-phrase

When comparing correlatives and wh-constructions, interrogatives also need to be mentioned. These differ from CorPs most clearly in their wh-phrase. We have already mentioned that CorPs select their basic wh-words, nor ‘who’ and zer ‘what’, from the interrogative paradigm. We have also mentioned that these words cannot be used as relative pronouns in appositive clauses, whether these clauses are adjoined to a DP or the E-type pronoun haina, or whether they are extraposed. For a full description of the facts, more factors must be taken into account.

 Georges Rebuschi

First, the general relative pronoun zein can also be used as an interrogative, but its use is usually restricted to closed (D-linked) sets, which already belong to the universe of discourse. Accordingly, the normal translation of zein is ‘which one(s)’.21 Starting from the 18th century, however, an important distinction has been made between two of its occurences. Its bare form that functions as an interrogative pronoun is also used in CorPs. A form marked for number on the other hand is used as a relative pronoun, as shown in the following examples. (50)

[Correlative use: invariable/no number indicated] Mt 26,48 – LV: ‘Quemcumque osculatus fuero, ipse est!’ KJV: ‘Whomsoever I shall kiss, that same is he.’ Zein-i ere musu ni-k eman-en bai-diot, which-d ever kiss I-e give-pros bait-‹a3d3e1› eta hura diteke. and dem it.will.be

(51)

[Relative use: singular] Mt 1,21– LV: ‘pariet autem filium, et vocabis nomen eius Iesum.’ KJV: ‘And she shall bring forth a son, and thou shalt call his name Jesus.’ Erdi-ko da seme batez, zein-a-ri ema-nen bear-pros ‹a3› son one.inst which-sg-d give-pros bai-diote izen-a Jesus. (ibid.) bait-‹a3d3e6› name-sg Jesus (Lit. ‘She will bear a son, to whom they will give the name Jesus.’)

(Haraneder 1740)

(52) [Relative use: plural] Lk 26,48 – LV: ‘[…] non omnes capiunt verbum istud sed quibus datum est.’ / KJV: ‘All men cannot receive this saying, save they to whom it is given.’ Ez dire guzi-ak gai hortakotz, hainak baizen, neg they-are all-pl able for.that such.pl but zein-e-i eman izan baitzaie. (ibid.) which-pl-d given aux-perf bait-‹a3d6› (53a,b) illustrate the use of bare zein as a relative pronoun in the 16th and 17th centuries. (Haraneder 1740 and Duvoisin 1859–65 have the plural form zein-ak in their translations of (53a), so that their FVFs naturally agree in plurality.) (53) a. Mt 7,15– LV: ‘Attendite a falsis prophetis, qui veniunt ad vos in vestimentis ovium…’/ KJV: ‘Beware of false prophets, which come to you in sheep’s clothing…’ Begirauzue […] profeta falsue-tarik, zein beware[imp]‹a3-e5’› prophet false.pl-abl which ethor-ten bai-tirade zu-etara ardi abitu-rekin… come-impf bait-‹a6› you[5]-to sheep clothing-com (Leiçarraga 1571)



Basque correlatives and their kin 

b. Nola ageri baita Ejipto-ko Piramidet-an, how apparent bait-is Egypt-gen pyramid.pl-loc zein eragin bait-zit-uen errege Faraon-ek… which build-fact-prf bait‑‹a6e3[past]› king Pharaoh-e (Axular [1643] 1964: § 16) (Lit., ‘ How [clearly] it appears in Egypt’s pyramids, which Pharaoh had made build…’) Next to form, there is also another property that distinguishes interrogative and correlative wh-words from those used in relative clauses: their position in the clause in which they appear. Thus, whereas relative pronouns must be clause initial (cf. Rizzi (1997), and much ensuing work), interrogative and correlative wh-items need not be. An interrogative wh-word can be preceded by various topicalised phrases, as shown in (28a) above, where the subject pronoun is topicalised. In the next two examples, the same is shown for CorPs. In (54) a direct object precedes the wh-phrase, and in (55), a quantifier does. (54) Lk 8,21 – LV: ‘Mater mea et fratres mei hi sunt, qui verbum Dei audiunt et faciunt.’ / KJV: ‘My mother and my brethren are these which hear the word of God, and do it.’ [Jainkoa-ren hitza nor-k ere entzu-ten bait-du God-gen word.sg who-e ever hear-impf bait-‹a3e3› eta obra-tzen], hura da ene ama eta ene aneia and work-impf dem is my mother and my brother (Léon 1946) (55)

Holakoe-tan [bakotxa-k zer ere bait-du bere such.pl-loc each.sg-e what ever bait–‹a3e3› himself baitan], eta har-tarik ari da. (Hiriart-Urruty [1893] 1972: 49) in and that-abl busy is ‘In such cases, whatever anyone has in himself, that is what he uses.’ Lit., ‘each/every one whatever s/he has…’

Finally, we must also mention a difference between the distribution of the wh-phrases in CorPs and in questions. While focussed phrases can intervene between the whword and the finite verb form in CorPs, they cannot do so in an interrogative sentence, regardless of whether the latter is embedded or not. This is shown in (56) and (57), where the focussed adjuncts are in italics and small caps. (56) Gal 6,16 LV ‘et quicumque hanc regulam secuti fuerint pax super illos…’ / KJV ‘And as many as walk according to this rule, peace be on them…’ Eta nor ere regla hun-en araura ebil-ten and who ever rule that-gen according behave-impf bai-tirade, bakea izan-en da hai-e-n gainean…22 bait‑‹a6› peace.sg be-prosp ‹a1› dem-pl-gen on (Leiç. 1571)

 Georges Rebuschi

(57) Mt 19,29 – LV: ‘Et omnis, qui reliquit domos vel fratres […] aut agros propter nomen meum, centuplum accipiet…’ / KJV ‘And every one that hath forsaken houses, or brethren […] or lands, for my name’s sake, shall receive an hundredfold…’ Eta nor-k ere ene izenaren aria-z utz-iko bai-titu and who-e ever my name’s sake-inst leave-pros bait-‹a6e3› etxe-a, edo anai-ak […] edo lurr-ak, haina-k house-sg or brother-pl or land-pl such.sg-e ehunkun iza-nen […] du. (Duvoisin 1859–65) hundredfold have-pros ‹a3e3› Concerning the other constructions handled in this section, a focussed element can also intervene between a wh-word and the FVF in IFRs, indicating some relationship between CorPs and IFRs. In the case of unconditional clauses, the focus cannot intervene between the wh-word and the FVF, since the wh-word is locally focussed and must therefore be left-adjacent to the inflected verb.

4.4

Summary of findings and the structure of CorPs

In this 4th section, we have described and compared five types of (non-independent) wh-clauses in Basque: appositive relatives, wh-questions, correlative clauses, indefinite free relatives and unconditional clauses. Their properties are listed Table (1). Table 1.  Comparative properties of CorPs and other non-independent wh-clauses Appositive Embedded relative clauses questions

Correlative protases

IFRs

wh-elements zein-*(a) /*nor zein(*a) / nor zein(*a) / nor nor obligatory initial yes no no no position of wh-item C° affix on FVF bait-en /*baitbait-/?-en -en/baitpresence of ere after no no preferred no the wh-word or phrase obligatory left-adjacenno yes no no cy of wh-item with respect to FVF position to the right of possible possible impossible obligathe main clause’s FVF tory insertion of eta (DNA) (DNA) possible (DNA) between the protasis and the apodosis possible paraphrase by yes no yes no a SFR

Unconditional clauses nor no baitoptional no

possible impossible

no



Basque correlatives and their kin 

The first four properties concern the internal structure of the compared clauses, the next three their combinatorial properties, and the last one, their ability to be paraphrased by an SFR. As can be seen, CorPs clearly differ from every other type, in that: i. they can be linked to the apodosis by the word eta; ii. they must occur in the left periphery of the main clause – if the argumentation in 3.2 above is correct, this is simply because they are directly merged there. Concerning other properties, correlative clauses are similar to embedded questions as far as the choice of their wh-element is concerned. Yet, there are differences, too. The wh-phrase is preferably followed by ere ‘ever’ in CorPs, which cannot appear either in embedded or in root questions. Correlatives also share some properties with indefinite free relatives, except for the presence of ere, and their respective positions with respect to the main clause. Finally, correlatives and appositive relatives are the only types of clauses that can be paraphrased by SFRs. The other constructions also differ from each other in various ways. For instance, the fact that the wh-phrase must be left-adjacent to the verb complex in its own minimal clause, and the fact that the C° affix cannot be bait‑ in embedded interrogatives, clearly sets these apart from the other types. Note by the way that the C° suffix ‑en in CorPs, which is normal in the varieties spoken across the French-Spanish border, is also, albeit very rarely, attested in Northern Basque. (58) illustrates this point, wich justifies the question mark in Table (1), 3rd line. (58) Lev 24,19 LV: ‘Qui inrogaverit maculam cuilibet civium suorum sicut fecit fiet ei.’ / KJV: ‘And if a man cause a blemish in his neighbour: as he has done, so shall it be done to him.’ [Nor-k ere bere herritarretarik edozein-i gaizki egin who-e ever his neighbours.abl anyone-d evil do.pros bait-dioke], [[nola bera-k egin du-en], bait-‹a3d3e3› how he.himself-e do-prf ‹a3e3›-en hala egin-a izan-en zaio]]. (Duvoisin 1859–65) thus do.prf-sg be-pros ‹a3d3› On the other hand, the differences between IFRs and unconditional clauses seem fairly small, in spite of what seems to be irreducible semantic differences between them. More work is necessary to derive at least some of those bundles of properties from more abstract principles. A word is also in order concerning the internal syntax of correlative clauses. We can infer from the foregoing discussion that the structure of CorPs from the 17th century on can best be captured by the representation in (59). The basic idea is that the wh-item in these clauses is located in the specifier of a special Topic Phrase. The head of this TopP is normally lexically realised by ere. This optionally visible head may take a FocusP or, more directly and more often, a Finiteness Phrase (FinP) as its complement.

 Georges Rebuschi

For cases like (45), I assume a possible zero realisation of the inflectional material within the latter. TopP can itself be dominated by other TopP (see Rizzi 1997). (59) The left periphery of CorPs ForceP 2 TopP* 2 TopP 2 wh-P Top’ 2 Top° FocP ere 2 FinP For a more detailed representation concerning the presence of quantifiers, illustrated in (55), as well as the position of multiple wh-phrases, when present, see Rebuschi and Lipták (to appear).

5. Correlative protases and conditionals It is well-kown that complex correlative sentences whose tense is generic or habitual can be paraphrased by conditional sentences whose protases contain either an indefinite pronoun like somebody, something, or a NPI like anybody, anything. In this section, we turn to a comparison between correlatives and such conditionals. We will show that correlatives are not only paraphrasable by conditionals, but that there are intriguing convergences in the very morphological material used in both types of sentences.

5.1

Basque conditionals

Let us start with a brief description of conditional sentences and their basic semantic similarity with other types of sentences. Basque conditional clauses always have a morpheme ba- ‘if ’ prefixed to their finite verb form. This ba- may be given some emphasis by being replaced by balinba (or preceded by the orthographically independent word balin);23 moreover, an allomorphic variant of balin, baldin (and longer derived forms, such as baldin-eta, baldin.eta.ria(k)) may occupy the initial position in the clause.24 Although it is rare to find both initial baldin and the emphatic balin.ba together, it is not impossible, cf. (60a,b), where the subordinate clause is, however, to the right of the main clause.



Basque correlatives and their kin 

(60) Imit. I,23,4 – Dabit namque magnam fiduciam moriendi perfectus contemptus mundi. […] / Benham: ‘For a perfect contempt of the world, a fervent desire to excel in virtue […] [is something] which shall give great confidence of a happy death.’ a. Heriotze on eta dohatsu bat-en esparanza har death good and happy one-gen hope √receive ahal dezakegu baldin balin-ba-dugu mundua-rentzat can ‹a1e4[pot]› baldin balin-if-we.have world.sg‑for mezprezio oso bat. (Chourio 1720) disdain full one (Lit. ‘We can have the hope of a good and happy death if we have total contempt for the world.’) b. Ba-dakite nehon-dik neholere ez prtc-they.know nowhere-abl in.no.way neg detzazkegu-la guk hemen bazterre-tako berriak ‹a6e4[pot]›-C° we-e here place.pl-gen news.pl xuxen eman eskuaraz, zoin beren egunetan, correctly give Basque-inst each their days-loc baldin ez balin-ba-dugu igortzaile-rik. baldin neg balin-if-we.have sender-part (Hiriart-Urruty [1893] 1995) ‘They know that we here can absolutely not give correct news from every place, from day to day, if we have no correspondent.’

5.2

CCSs as paraphrases of conditional sentences

Resuming our review of various translations of the same texts, the following examples show the various ways in which Latin indefinite conditional sentences could be translated: the translations either provide a similar conditional clause (cf. 61a,b), or a CorP. (cf. 62), or yet a left-peripheral SFR (cf. 63).25 (61) Jn 14,23 – LV: ‘Si quis diligit me, sermonem meum servabit […]’ KJV: ‘If a man love me, he will keep my words […]’ Imit II,1,2– Benham: ‘if any man loveth me he will keep my words […]’ a. Baldin norbeit-ek on ba-darizt ni-ri, ene baldin somebody-e good if-he.finds me-d my hitz-a begira-turen du. (Leiçarraga 1571: conditional) word-sg keep-pros ‹a3e3› b. Nihor-k maite ba-nau, begira-tuko du ene hitza. anybody-e love if-‹a1e3› keep-pros ‹a3e3› my word.sg (Pou. 1669: conditional)

 Georges Rebuschi

(62) Ni maite nau-en-a-k ene hitz-ak beira-tuko ditu. me love ‹a1e3›-en-sg-e my word-pl keep-prosp ‹a6e3› (Lit., ‘The – that loves me…’) (Chourio 1720: SFR) (63)

Nurk ere ni maite bei-nai, ene hitzak who-e ere me love bait-‹a1e3› my words begira-türen dütü. keep-pros ‹a6e3› (Lit., ‘Whoever loves me…’)

(Maister 1757: CorP)

Although the examples above are chronologically ordered, the variation displayed is not diachronical. Other excerpts would show a different distribution of the syntactic variants, as witnessed by the other translations I have found of Jn 14,23. To wit, Haraneder (1740), Harriet (1855) and Duvoisin (1859–65) use a conditional (both with norbait); Inchauspé (1883), “Ezkila” (1974) and Etchehandy (1999) use a SFR; Haristoy (1896) and Léon (1929, Imit., and 1946, NT) use a CorP. The reverse situation holds too, as shown in (64): here the Latin text, which contains first a universally quantified left peripheral FR, and second a CorP, is translated into two conditional sentences. Haraneder’s and Harriet’s translations are added, because their first protasis is, morpho-syntactically, an unconditional clause. (64) Mt, 12,32 – LV: Et quicumque dixerit verbum contra Filium hominis, remittetur ei; qui autem dixerit contra Spiritum Sanctum […]. / KJV: ‘And whosoever speaketh a word against the Son of man, it shall be forgiven him: but whosoever speaketh against the Holy Ghost, it shall not be forgiven him […].’ Norbeit-ek errai-ten ba-du solas zerbeit Gizona-ren somebody-e say-impf if-‹a3e3› word some man.sg-gen Semea-ren kontra, barka-tuko zaio; norbeitek, aldiz, Son.sg-gen against forgive-pros ‹a3d3› somebody-e however errai-ten ba-du Izpiritu Saindua-ren kontra, ez zaio say-impf if-‹a3e3› Ghost Holy.sg-gen against neg ‹a3d3› barka-tuko […] (“Ezkila” 1974) forgive-pros (Lit. ‘If somebody says something …, it shall be forgiven him; but if he says [it/ so] against the Holy Ghost, it shall not be forgiven him.’) (65) a.

Nor nahi mintzatu diteken Gizona-ren Semea-ren kontra anybody speak-prf ‹a3›‑en man.sg-gen Son.sg-gen against barkha-tuko zaio; bainan baldin nihor mintzatu ba-da… forgive-pros ‹a3d3› but baldin anybody speak-prf if-‹a3› (Lit., ‘That anybody should have spoken against the Holy Ghost […]; but if anybody has spoken […]’ (Haraneder 1740)



Basque correlatives and their kin 

b.

Nor nahi gaizki mintza dadin gizonaren […], anybody badly √speak ‹a3[subj]› man-gen Semea-z, bainan baldin nehor Ispiritu Saindua-z son.sg-inst but baldin anybody ghost holy.sg-inst gaizki mintza-tzen ba-da […] badly speak-impf if-‹a3› (Lit., ‘let anybody speak evil of the Holy Ghost […]; but if anybody speaks evil […]’) (Harriet 1855)

Other translations of this verse are: Leiçarraga (1571) with two CorPs; AnBa (1828, Labourdin Basque) and Iribarnegaray (c. 1860, Low Navarrese) with two conditional protases; Salaberry (1856, Low Navarrese) and Léon (1946, Low Navarrese) with two SFRs; Cazenave (c. 1860, Low Navarrese), with first a CorP and second, an SFR; Etchehandy 1999 (“Navarro-Labourdin”), with first an SFR and second a conditional protasis; finally, Inchauspé (1856, Souletin) used agentive nominalisation: mintzazalia, lit., ‘the speaker’ twice. In fact, the only historical change really attested is the lesser and lesser frequency of appearance of the NPI (nihor / nehor ‘anybody’ or ‘nobody’ according to the context) in conditional protases, as against the indefinite/existential norbait ‘somebody’. However, the latter word was already used in the 16th century, as illustrated in (61a).

5.3

Correlative and conditional protases: ba- and bait-

Let us now turn to some properties that correlative and conditional protases share or have shared. Curiously enough, some texts display sentences in which bait- is used instead of conditional ba-, and vice versa. As early as the 16th century, one example of bait- replacing conditional ba- ‘if ’ is attested in the NT’s translation. (66) 1-Jn 4,11 – LV: ‘Carissimi, si sic Deus dilexit nos: et nos debemus alterutrum diligere.’ / KJV: ‘Beloved, if God so loved us, we ought also to love one another.’ Maiteak, baldin hunela Iainkoa-k onhets-i bai-kaitu, beloved-pl baldin so God-e love-prf bait-‹a4e3› gu-k ere behar dugu elkar onhetsi. (Leiçarraga 1571) we-e too must ‹a3e4› each.other love A century later, the same construction is still attested, see (67), but bait- can be substituted for ba- even if baldin is not present, as in (68). (67) Imit. 3,37,4 – Thomas: […] nisi integra resignatione, et quotidiana sui immolatione prius facta […] / Benham: ‘[…] unless they first entirely resign themselves and daily offer themselves up as a sacrifice […]’ baldin lehenago bere buruaren sakrifizio oso eta baldin firstly their own sacrifice full and

 Georges Rebuschi



bethiereko bat-ez ene-ganat diren guziaz eternal one-inst me-towards are-en all-inst abandona-tzen ez-pait-zaizkit… forsake-impf neg-bait-‹a5d1›

(Arambillaga 1684)

(68) Imit. 3,59 – Thomas: … si tu ipse non assistas, juves, confortes, consoleris, instruas et custodias. / Benham: ‘… if Thou Thyself do not assist, help, strengthen, comfort, instruct, keep in safety.’ zuk zerorrek […] lagun-tzen, sustenga-tzen, borthiz-ten, you-e yourself help-impf sustain-impf strengthen-impf argi-tzen eta goarda-tzen ez-pai-nauzu. (ibid.) enlighten-impf and keep-impf neg-bait-‹a1e5› Besides, in his Grammaire basque, Ithurry ([1895] 1920: 360) notes that Axular (the most acclaimed classical writer in Northern Basque) used “bait- instead of ba- when conditional ‘if ’ was followed by bezala ‘as’ and bezanbat ‘as much as’.” But Axular was not the only author to use bait- in morphologically marked conditional contexts (whereby I mean that the subordinators equivalent to ‘if ’ were not enough to licence bait- instead of ba-: the very conjugation of the FVF had to be “conditional” in the sense that it roughly corresponded, and still does, to the English preterite with a modal, counterfactual or “irrealis” value).26 Thus, the same construction is attested in the 18th century too, both in the Labourdin and the Souletin varieties of Northern Basque, as shown in (69) and (70). (69)

Halarikan-ere nahi dut Oihenarto-k dio-en-a, egia however want ‹a3e1› Oyhenarte-e says‑en-sg truth.sg bai-liz bezala besarkatu. bait-it.were as embrace (Etcheverry of Sare [1712] 1907: 76) ‘However, I want to espouse Oyhenarte’s thesis [lit., ‘what O. says’], as if it were the truth.’

(70) Imit. III,30,3 – Thomas: … tamquam omnis spes sit ablata emergendi. / Benham: ‘… as if all hope of escape were taken away.’ … antik jalkiteko esperantxa osoki galdü bei-liz bezala. there-abl to.get.out hope.sg completely lost bait-it.were as (Maister 1757) Intriguingly, the opposite phenomenon, whereby ba- is substituted for bait-, seems to have started only once bait- could no longer replace ba-. The first example I have found only dates from the late 18th century, see (71).27 (71) Jn 20,23 – LV: ‘Quorum remiseritis peccata, remissa sunt eis; quorum retinueritis, retenta sunt.’ / KJV: ‘Whose soever sins ye remit, they are remitted unto them; and whose soever sins ye retain, they are retained.’



Basque correlatives and their kin 

Nor-i ere barka-tzen ba-diozkatzue bekatuak, ta who-d ever forgive-impf ba-‹a6d3e5’› sin.pl and ha-e-i barka-tuko zaizte, ta nor-i ere ez dem-pl-d forgive-pros ‹a6d6› and who-d ever neg ba-diozkatzue barka-tzen ta ha-e-i ez zaizte barka-tuko. ba-‹a6d3e5’› forgive-impf and dem-pl-d neg ‹a6d6› forgive-pros (Ubillos 1785: 151) More examples can be found in the 19th century, see (72), (73), and even in the second half of the 20th century, cf. (74). (72) Mt 5,19 – LV: ‘Qui ergo solverit unum de mandatis istis minimis et docuerit sic homines, minimus vocabitur in regno caelorum; qui autem fecerit et docuerit, hic magnus vocabitur in regno caelorum.’ KJV: ‘Whosoever therefore shall break one of these least commandments, and shall teach men so, he shall be called the least in the kingdom of heaven: but whosoever shall do and teach them, the same shall be called great in the kingdom of heaven.’ Artakotz, nork ere aus-ten ba-du manamendu ttipi therefore who-e ever break-impf ba-‹a3e3› order small otarik bat, … arras ttikia deitua izain da those-abl one very small.sg called.sg be-pros ‹a3› zeruetako erreinuan; bano nork ere egi-ten heaven-gen kingdom-sg-loc but who-e ever do-impf ba-du, eta erakus-ten ba-du, ura deitu-a ba-‹a3e3› and teach-impf ba-‹a3e3 ›, dem called-sg izain da andia zeruetako erreinuan. be-pros ‹a1› great.sg heaven.gen kingdom.sg.loc (Echenique 1857) (73) Rom 5,18 – LV: ‘Igitur sicut per unius delictum in omnes homines in condemnationem sic et per unius iustitiam in omnes homines in iustificationem vitae.’ / KJV: ‘Therefore as by the offence of one judgment came upon all men to condemnation; even so by the righteousness of one the free gift came upon all men unto justification of life.’ Beraz nola baten bekhatua-ren bidez kondenamendua-ren thus how one-gen sin.sg-gen via condemnation-gen azpira erori ba-dire gizon guziak, hala ere bat-en under fallen ba-‹a6› man all.pl thus also one-gen zuzentasun-a-k zuzen eragin ditu guziak. righteousness-sg-e right made-prf ‹a6e3› all.pl (Duvoisin 1859–65)

 Georges Rebuschi

(74) Jn 8,12 – LV: ‘Qui sequitur me, non ambulabit in tenebris, sed habebit lucem vitae.’ / KJV: ‘He that followeth me shall not walk in darkness, but shall have the light of life.’ nor ere jarraiki-tzen ba-zaut, hura ez dabil ilhunbe-tan, who ever follow-impf ba-‹a3d1› dem neg walks darkness-loc baina uka-nen du harek biziko argia. but have-pros ‹a3e3› dem-e life-gen light.sg (Soubelet 1969: 104)

5.4

On the relationship between correlative sentences and conditional sentences

Bittner (2001) argues that complex conditional sentences are a subkind of complex correlative sentences (on a similar view, see Arsenijević, this volume). She bases her conclusion on Marathi examples in which wh-words and the pronominal correlates on the one hand, and the counterparts of ‘if ’ and ‘then’ on the other exhibit the same pairs of morphemes: j- in the protasis, and t- in the apodosis. However, while conditional sentences are organised on a protasis-apodosis base in many, if not all, natural languages, not all languages have correlative sentences in the sense that Burushaski, Hindi, Hungarian, Latin, or Polish do, in particular when the pronominal correlate qualified by the CorP denotes an individual. On the other hand, comparative correlatives are much more frequent (see den Dikken this volume and references cited there), perhaps just as universal as complex conditional sentences are. It would then be more natural to consider CCSs to be a special subkind of conditional sentences, or, perhaps, to show that both structures are fundamentally similar, with something special to add as far as CCSs are concerned. From this point of view, it is worth recalling Lafon’s (1966) hypothesis, revived by Trask (1997), according to which “originally” ba- and bait- were one and the same morpheme, the former a simplification, and the latter a phonological “hardening” (which can be independently justified)28 of the positive assertive particle bai ‘yes’. Lafon did not mention the ba-/bait- interchanges described in 5.3, but remarked, among other things, that bait functions either as a suffix in indefinite pronouns (in nor-bait ‘someone’, zer-bait ‘something’, non-bait ‘somewhere’, etc.), or as a subordinating prefix in appositive relatives and in correlative clauses. His general thesis is thus that the nor/ zer paradigm may have been ambiguous between an interrogative and an indefinite pronoun in the prehistory of the language. If left alone, these pronouns would have later specialized as interrogative pronouns, but the insertion of assertive bai(t)- in the sentence would have triggered either its suffixation to the wh-words, thereby forcing their unambiguous reinterpretation as indefinites, or its prefixation to the finite verb



Basque correlatives and their kin 

form, yielding a “subordinate” clause interpretable either as a correlative or a conditional protasis, as summarized in (75)-(77): (75) Prehistoric period (I) a. ◊Nor dator.29 wh-[+hum] comes ‘Someone is coming.’ b. ◊Nor dator? ‘Who is coming?’ (76) Prehistoric period (II) a. ◊Nor dator? (same as (75b)) b. ◊Nor bai(t) dator wh-[+hum] bait comes ‘Someone is coming’ [cf. modern Basque Norbait ba-dator] ‘Someone is coming’ [cf. the non-emphatic assertion (77a) below] (77) Historically attested Basque a. Norbait dator. someone comes ‘Someone is coming.’ b. Nor (ere) bai(t)-dator… wh-[+hum] ever bait-comes ‘Who(ever) comes…’ c. ??Nor / nehor30 / nor.bait ba-dator… who anybody / somebody if-comes ‘If someone/anyone is coming…’ An independent argument for an autonomous status for bait- at an earlier stage of the history of the language is that, in the oldest printed text, the indefinite/existential norbait, in the ergative case, instead of being followed by the case suffix as it has always been since the 17th century, viz. norbait-ek (with an epenthetic ‑e‑), occurs twice under the form: nor-k bait in Etchepare’s poems, cf. (78) (remark also that such examples do not have a universal interpretation, in spite of the presence of ere). (78) a. Nor-k-bait-ere amoria niri daraut mutha-tu. who-e-bait-ever love(r).sg me-d ‹a3d1e3› change-prf ‘Someone has changed my mistress.’ Lit. ‘Someone has-to-me changed the lover.’ (Etchepare [1545] 1980: VII,7) b. Nor-k-bait-ere egin deraut malizia handia. who-e-bait-ever do.prf ‹a3d1e3› wickedness big.sg ‘Someone has done a lot of harm to me.’

(id.: XIII,31)

 Georges Rebuschi

Attractive though it may be, and even adding to Lafon’s arguments those of 5.2 and 5.3 above, the proposed reconstruction remains highly tentative to say the least, because there is, as far as I know, no means to connect the indefiniteness of the suffix ‑bait in the norbait, zerbait indefinite pronoun series, and the complementiser status of bait- in relative clauses and CorPs. Moreover, if both CCSs and conditional sentences do optionally exhibit a syntactic “linker”, they are quite distinct: eta in the former case, and the grammaticalized time adverbial orduan ‘then’ in the latter case. Finally, whilst it is possible to establish some intuitive connection between bai ‘yes’ and the stressed ba- of the modern Basque Norbait ba‑dator mentioned in (76b), it is much more difficult to suggest anything of the kind concerning unstressed ba- ‘if ’ and bait‑, all the more as the latter two items can be preceded by the negative particle ez, which cannot be concatenated with emphatic ba. The following examples illustrate negative clauses containing conditional ba- and protatic bait- respectively. (79) Imit. 1,2,1 – Thomas: Si scirem omnia quæ in mundo sunt, et non essem in charitate, quid me juavert coram Deo […]? / Benham: ‘If I knew all the things that are in the world, and were not in charity, what should it help me before God […]?’ a. Mundu-ko gauza guziez jakinsun ba-nintz, eta world-gen thing all.pl.inst knowing if-I.were and ez ba-nu karitatea, zer probetxu nuke neg if-I.had charity.sg what benefit I.would.have Jainkoaren aitzinean […]? (Chourio 1720) God-gen before b.

Nakitzan ordian mündüko gaizak oro, I-know[subj] then world-gen thing-pl all ez-palin ba-nü31 karitatia, zer balio neg-balin if-I.had charity.sg what value lüke jakite harek […]? would-have knowledge dem.e

(Maister 1757)

c. … karitate-rik ez-ba-nu, zer probetxu nuke […]? charity-part neg-if-I.had what benefit I.would.have (Léon 1929) As was noted above, according to the texts we have, ba- only began to replace bait- in CorPs after the non-standard use of bait- in conditional protases fell into disuse. Whatever the relevance of that fact is, there are independent facts that tend to show that the similarities between CCSs and (indefinite) conditional sentences may well be due to a late convergence (perhaps prompted by the overall semantic parallelism of the two complex constructions), rather than to a (pre)historic common origin.



Basque correlatives and their kin 

A case in point is provided by the presence of eta that occurs not in between the protasis and the apodosis of CSSs (i.e. as the head of a special Topic Phrase), but following the wh-item of CorPs on the one hand, or immediately to the right of the would be “pleonastic” baldin of conditional protases on the other. Insofar as CorPs are concerned, the most interesting case is provided by the free relatives whose wh-item is noiz ‘when’, because of their semantic similarity with conditional protases. According to the authors, four sub-cases must be distinguised: noiz can either be used alone, or followed by eta, or followed by ere, or by eta ere. Likewise, baldin can be used on its own, or followed by eta, or yet by a longer form, -eta-ria(k). Interestingly, Leiçarraga’s (1571) NT contains many cases of baldin-eta, but none of noiz-eta; the same is true of Etcheverry of Ciboure (1636), and, to a lesser degree, of Etcheverry of Sare (1712), who uses, however, noiz-eta-ere. Likewise, Haraneder (1740, 1750) uses baldin.eta.ria(k) in conditional clauses, but never noiz-eta without ere following. But the reverse situation is also attested: Axular (1643) and Laphitz (1867) never use baldin-eta(‑ria(k)), but they do use noiz-eta (without ere) quite systematically. Axular does not have a single instance of either noiz-ere or of noiz-eta-ere, whereas Laphitz does not use noiz-ere but has one occurrence of noiz-eta-ere.

6. Conclusions The main issue raised in this paper was whether Basque complex correlative sentences constitute a special type of sentences, or whether it is possible to reduce them to other existing constructions. To this effect correlative protases have been compared with various other clauses: semi-free relatives, dependent clauses containing an explicit whword (appositive relatives, indefinite free relatives, embedded questions and unconditionals), and conditional protases. The results of the comparison were summarized in Table 1 in Section 4.4 and the special section on conditionals in 5. Here we recapitulate the most important findings. i. Concerning the link between semi free relatives (SFRs) and CorPs, we have shown that they arguably have different syntactic structures. While SFRs are nominal in nature (DPs), CorPs are clausal (CPs). The two are also different from each other in the way they are connected with the main clause. While both types of relative clauses can appear in the left periphery and may be resumed by a pronominal correlate, the two types differ in the presence of the linker eta, which may only accompany CorPs in complex correlative sentences. Although we have left their semantic analysis completely in the vague so far, it is worth noticing that just like SFRs, CorPs seem to be neutral with respect to the three basic types e, <e,t> and <<e,t>,t> they represent. They sometimes yield entities, as in (50), sometimes properties, as in (36) and (37), and sometimes generalised quantifiers, when the whole CCS predicates something of the elements of a closed set. It is therefore

 Georges Rebuschi

tempting to suggest that CorPs compositionally provide a property (the movement of the wh-word creating a gap, and the wh-word itself transmitting a λ feature to the C° it specifies):32 the type shifts identified by Partee (1987) would then allow a reinterpretation triggered by contextual and pragmatic factors (and the possible “widening effect” of ere, to borrow Dayal’s (1997) expression).33 ii. Concerning the link between correlatives and dependent wh-clauses, we can say that unconditional or “no matter clauses” are the type that most closely approximates CorPs. In particular, ere is much preferred today in CorPs, and unconditionals are the only other type of wh-clauses which also allows its presence. This corroborates the idea that the basic contribution of ere is to express that there is no other property than those specified in the clause that is relevant (as in the case of CCSs) or irrelevant (as in the case of unconditionals) to the rest of the complex sentence, i.e. to the content of the apodosis. iii. Finally, we have compared CSSs to conditional sentences whose apodoses contain an indefinite pronoun or a negative polarity item.34 As an anonymous reviewer remarked after reading the first version of this paper, which was much more sympathetic to Lafon’s theory than the present version, what the diachronic data really show is that there has clearly been a convergence in the means of expressing CCSs and conditional sentences, probably because of their semantic closeness. Given the above results, we hope to have contributed to a better understanding of complex correlative sentences. Next to asking whether CCSs constitute a special type of sentences, the final answer to which must unfortunately await further research, we have also touched upon questions of structural properties and syntactic composition. We have entertained the view that the CorP’s position in the main clause might follow from properties of the linker eta. This element seems to select a wh-free relative clause in its specifier, and a FocusP à la Rizzi as its complement. If this is indeed the case, the fact that a CorP must by definition have a pronominal correlate (hosted by SpecFocusP), whereas unconditional clauses need not have one, would be quite natural.  It remains to be seen if correlatives in every language are amenable to an analysis along this line. Hopefully, a careful examination of CCSs across many languages will sooner or later provide a possible clue to this question.

Acknowledgements First and foremost, I must thank Anikó Lipták, not only for having asked me to write this paper, but also for her keen interest in Basque syntax and for her precious remarks on two preliminary versions of this paper. I am grateful to Aurelia Arkotxa, Esther Mortalena, Beñat Oyharçabal and Marie Pourquié for discussions concerning both the acceptability of older (attested) examples and the grammaticality of those that pertain to the contemporary language. I also wish to thank two anonymous reviewers, who pointed out many shortcomings in the first version of this paper, and forced me to revise some of my analyses. Needless to say, all remaining errors are



Basque correlatives and their kin  mine only. This paper is an outgrowth of the collective work undertaken between 2003 and 2005 in the research group “Structure de la phrase / La périphérie gauche” funded by CNRS (Fédération Typologie et universaux linguistiques).

References (A) Basque sources35

[AnBa 1828] Anonymous. 1828. Jesus-Christo gure Jaunaren laur Ebanyelioac […]. Bayonne: Lamaignière, 1828. Arambillaga, d’ A. 1684. Jesu Christoren Imitacionea [books III and IV]. Bayonne: Fauvet.* Astarloa, Pedro. 1816–18. Urteco Domeca gustijetaraco verbaldi icasibidecuac […]. [Cited in Arejita (1978)]. Axular, Pedro. 1643. Gero. Bordeaux: Milanges. Facsim.: Bilbao: Euskaltzaindia, 1988. Edition with modernized spelling by Luis Villasante, Barcelona: Juan Flors, 1964. Baratciart, André. 1787. Guiristinoqui bicitceco eta hiltceco moldea. Bayonne: Fauvet-Duharte.* Cazenave [Casenave]. A. c. 1860. Jesu Khristoin Ebanjelio Seindia San Mathiuïn aäbera. Manuscript. In Rosa Miren Pagola et al (eds), Bonaparteko Ondareko Eskuizkribuak, Ekialdeko behe-nafarrera, I, 17–76. Bilbao, Deustuko Unibertsitatea, 1999. Chourio, Michel de. 1720. Jesu-Christoren Imitacionea […] Escararat itçulia. Reprint, Bayonne: Trebos, 1788; facsim. of the latter, San Sebastián: Hordago-Lur, 1979.* Duhalde, Martin. 1809. Meditacioneac […].Facsim., San-Sebastián: Hordago-Lur, 1978.* Duvoisin, Jean-Pierre. 1859–65. Bible edo Testament Zahar eta Berria […]. London. Facsim., Bilbao: Gran Enciclopedia Vasca, 1972. “EAB” [anonymous collective work]. 1994. Elizen arteko Biblia. Lazkao: Sociedades Bíblicas Unidas & Bilbao: Bibli Elkarte Batuak & Euskal Eizbarrutiak. Echenique, Bruno. 1857. El Evangelio segun San Mateo […]. London. Facsim. in Louis-Lucien Bonaparte, Opera Omnia Vasconice, II, 91–151. Bilbao: Euskaltzaindia, 1991. “EHEG” [anonymous collective work]. 1980. Itun Berria. San-Sebastián: Idatz (Herriko Elizbarrutietako Gotzaiak). Etchehandy, Marcel. 1999. Testamendu Berria. Belloc: Biblia Elkartea & Madrid: Sociedades Bíblicas Unidas. Etchepare, Bernat. 1545. Linguae vasconum primitiae. Edited by P. Altuna, Bilbao: Mensajero, 1980.* Etcheverry of Ciboure, Joanes [Etxeberri Ziburukoa]. 1636. elizara erabiltzeko liburua. New printing, Bordeaux: Mongiron Milanges, 1665.* Etcheverry of Sare, Joanes [Etxeberri Sarakoa, J.]. 1712. Obras vascongadas. Ed. by Julio de Urquijo é Ibarra, Paris: Geuthner, 1907.* “Ezkila” [anonymous collective work]. 1974. Jesu Kristoren Berri Ona. Bonloc (pyrénées-Atlantiques): Editions Ezkila. Haraneder, Joanes. 1740. Jesu Christoren Evangelio Saindua. Edited by P. Altuna, Bilbao: Euskal­ tzaindia, 1990.* Haraneder, Joanes. 1750. Gudu izpirituala. New edition. Bayonne: Cluzeau, 1827.* Haristoy, Pierre. 1896. Jesu-Kristoren imitazionea. Pau: Sylvain Dufau & Vignancour. Harriet, Maurice. 1855. Iesu-Christo gure Iaunaren Testament Berria […]. Bayonne: Lasserre. Hiriart-Urruty, Jean. 1891. Biba eskuara, in Eskualduna (13/11). Reprinted in Inaki Camino (ed.), Gontzetarik jalgiaraziak. Bilbao & San Sebastián: Euskal Editoreen Elkartea, 1995.*

 Georges Rebuschi Hiriart-Urruty, Jean. 1893. Gure indarrak. In Eskualduna (06/10). Reprinted in Pierre Lafitte (ed.), Mintzaira, Aurpegia, Gizon. Oñate: Editorial Franciscana Aránzazu, 1971.* Inchauspé, Emmanuel. 1856. Jesus-Kristen Ebanjelio Saintia Sen Mathiuren arauéra. Bayonne: Lamaignère. Facsim. In Louis-Lucien Bonaparte (1991) Opera Omnia Vasconice, II. 167– 243. Bilbao: Euskaltzaindia. Inchauspé, Emmanuel. 1883. Jesu-Kristen imitacionia. Bayonne: Lamaignère. Iribarnegaray, Vicaire –. c. 1860. Yesu Chrichtoin Ebanyeliua San Mathioin arabera. Manuscript. Published in Rosa Miren Pagola et al (eds), Bonaparte Ondarreko Eskuizkribuak, Mendebaldeko behe-nafarrera. 19–81. Bilbao: Deustuko Unibertsitatea, 1999. Kerexeta, Jaime. 1976. Euskal-Biblia (bizkaieraz). Bilbao: Bilbo’ko Elizbarrutiko Gotzaintza. Lapeyre, Etienne. 1891. Credo edo sinhesten dut esplikatua. Edited by Luis Villasante. Bilbao: Euskaltzaindia, 1982.* Laphitz, Franzisko. 1867. Bi saindu hescualdunen bizia […]. Facsim., San-Sebastián: Hordago/ Lur, 1978.* Larréguy, Bernard. 1775. Testamen çaharreco eta berrico historioa […]. Bayonne: Fauvet-Duharte, 1775.* Léon, Léon. 1929. Jesu-Kristoren Imitazionea. Turnhout, Belgium: Brepols.* Léon, Léon. 1946. Jesu-Kristo gure Jaunaren Ebanjelio Saindua. Ustaritz: published by the author. Leiçarraga, Johannes. 1571. Iesus Christ Gure Iaunaren Testamentu Berria – Kalendera – ABC edo Christinoen Instructionea. La Rochelle. Reprinted in facsim. as: I. Leiçarragas Baskische Bücher von 1571, by Theodor Linschmann and Hugo Schuchardt, Strasburg, 1900.* Maister, Martin. 1757. Jesu-Kristen Imitacionia. Pau: Dugué & Desbaratz. Pouvreau, Silvain. 1669. Iesusen Imitacionea. Edited with modernised spelling by José María Satrústegui, San-Sebastián: Hordago/Lur, 1979.* Salaberry d’Ibarolle, M., 1856. L’Evangile selon Saint Mathieu […]. Facsim. in Louis-Lucien Bonaparte, Opera Omnia Vasconice, III, 335–420. Bilbao: Euskaltzaindia, 1991. Soubelet, Domingo. 1969. Jesus. Ligugé (Vienne, France): Presses d’Aubin. Ubillos, Juan Antonio. 1785. Christau doctriñ berri-ecarlea […].Facsim. with an Introduction by Patxi Altuna. Bilbao: Euskaltzaindia, 1989. Zavala, Antonio. 1985. Esaera Zaararen bilduma berria I, II. Tolosa: Auspoa, #184 and #185.

Sources of the Latin texts and their translations.

The Bible. Latin Vulgate. Downloadable at: http://www.thelatinlibrary.com/bible.html The Bible. King James’ Version. Thomas à Kempis. ± 1425. De imitatione Christi. Downloadable at: http://www.thelatinlibrary. com/kempis.html English Translation: Benham, W., 1905. Imitation of Christ. London: Routledge & New York: E.P. Dutton.

(B) Linguistics

Agud, Manuel & Antonio Tovar. 1989. Materiales para un diccionario etimológico de la lengua vasca (V). Anuario del Seminario de Filología Vasca “Julio de Urquijo”, 23 (2). 463–532. Arejita, Adolfo. 1978. Euskal Joskera [Basque syntax]. Durango: Zugaza. Arsenijević, Boban. This volume. {Relative {conditional {correlative clauses}}}. Artiagoitia, Xabier. 2002. The functional structure of the Basque noun phrase. In Xabier Artiagoitia et al. (eds), Erramu Boneta: Festschrift for Rudolf P. G. de Rijk, 73–90. San-Sebastián /



Basque correlatives and their kin  Donostia & Vitoria / Gasteiz: Diputación Foral de Gipuzkoa and Universidad del País Vasco, Supplements of ASJU 44. Bhatt, Rajesh. 2003. Locality in Correlatives. Natural Language and Linguistic Theory 21(3). 485–541. Bittner, Maria. 2001. Topical referents for individuals and possibilities. In Rachel Hastings, Brendan Jackson & Zsófia Zvolenszky (eds.), Proceedings of SALT XI, 36–55. CLC, Ithaca. Boškovič, Zeljko. 1997. Superiority effects with multiple wh-fronting in Serbo-Croatian. Lingua 102. 1–20. Cable, Seth. 2005. The syntax and semantics of the Tibetan correlative. Manuscript. Downloadable at: http://web.mit.edu/scable/www/work/papers/Tibetan-Correlative-SynSem.pdf (1 June 2006) Cable, Seth. This volume. The syntax of the Tibetan correlative. Caponigro, Ivano. 2001. On the semantics of indefinite free relatives. In Marjo van Koppen, Joanna Sio & Mark de Vos (eds), Proceedings of ConSOLE X, 49–62. Leiden: SOLE. Caponigro, Ivano. 2004. On the source of maximality in wh-constructions crosslinguistically. In Mary Andronis, Erin Debenport, Anne Pycha & Keiko Yoshimura (eds), Chicago Linguistics Society, 38 (1). 125–144. Chomsky, Noam. 1956. Three models for the description of language. I.R.E. Transactions of Information Theory 2. 133–124. Citko, Barbara. This volume. What don’t wh-questions, free relatives, and correlatives have in common? Dayal, Veneeta. 1996. Locality in wh-quantification: Questions and relative clauses in Hindi. Studies in Linguistics and Philosophy 62. Dordrecht: Kluwer Academic Publishers. Dayal, Veneeta. 1997. Free relatives and ever: identity and free choice readings. In Aaron Lawson (ed.), Proceedings of SALT VII, 99–116. Ithaca, NY: CLC Publications. Dikken, Marcel den. This volume. Comparative correlatives and successive cyclicity. é. Kiss, Katalin. 1995. NP Movement, Operator Movement, and Scrambling in Hungarian. In Katalin é. Kiss (ed.), Discourse Configurational Languages, 207–243. Oxford: Oxford University Press. Etxepare, Ricardo. 2003. Exclamatives. In José I. Hualde & Jon Ortiz de Urbina (eds), A Grammar of Basque, 564–572. Berlin: Mouton de Gruyter. Greenbaum, Sidney. 1969. Studies in English adverbial usage. London: Longman. Grosu, Alexander. 2004. The syntax-semantics of modal existential wh constructions. In Olga Tomic (ed.), Balkan Syntax and Semantics, 405–438. Amster­dam/Philadelphia: John Benjamins. Haudry, Jean. 1973. Parataxe, hypotaxe et corrélation dans la phrase latine. Bulletin de la Société Linguistique de Paris 68 (1). 147–186. Hualde, José Ignacio & Jon Ortiz de Urina. 2003. A grammar of Basque. Berlin/New York: Mouton de Gruyter. Ithurry, Abbé. 1895. Grammaire basque. Dialecte labourdin. Bayonne: New Printing (Facsim. 1979, San Sebastian: Hordago). Izvorski, Roumyana. 1996. The syntax and semantics of correlative proforms. In Kiyomi Kusumoto (ed.), Proceedings of NELS 26, 133–147. Amherst, MA: GLSA. Izvorski, Roumyana. 2000. Free adjunct free relatives. In: Roger Billerey & Brook D. Lillehaugen (eds), WCCFL Proceedings, 232–245. Somerville, MA: Cascadilla Press. Lafon, René. 1944. Le système du verbe basque au XVIe siècle, I, II. Bordeaux: Delmas and Université de Bordeaux. Facsim., San Sebastián: Elkar, 1980.

 Georges Rebuschi Lafon, René. 1966. La particule bait en basque: ses emplois morphologiques et syntaxiques. Bulletin de la Société de linguistique 61.1. Reprinted in Jean Haritschelhar and Pierre Charritton (eds), 1999. Vasconiana. 667–693. Bilbao: Euskaltzaindia. Levin, Beth. 1983. Ergativity. Doctoral dissertation, MIT. Lipták, Anikó. 2000. Multiple relatives as relatives of questions. In Gábor Alberti & István Kenesei (eds), Approaches to Hungarian, VII: Papers from the Pécs Conference, 153–177. Szeged: JATE Press. Lipták, Anikó. 2005. On the correlative nature of Hungarian left peripheral relatives. In Benjamins Shaer, Werner Frey & Claudia Maienborn (eds), Proceedings of the Dislocated Elements Workshop, ZAS Papers in Linguistics, 35:1. 287–313. Berlin, ZAS. Oyharçabal, Bernard. 1987. Etude descriptive de constructions complexes en basque. Paris: Université de Paris VII dissertation [Thèse d’Etat]. Oyharçabal, Bernard. 2003. Relatives. In José.I. Hualde & Jon Ortiz de Urbina (eds), A grammar of Basque, 762–823. Berlin: Mouton de Gruyter. Partee, Barbara. 1987. Noun phrase interpretation and type-shifting principles. In Jeroen Groenendijk, Dick de Jong & Martin Stokhof (eds), Studies in discourse representation theory and the theory of generalized quantifiers, 115–143. Dordrecht: Foris. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik. 1972. A Grammar of contemporary English. London: Longman. Rebuschi, Georges. 1998. Nouvelles remarques sur ‘haina’. Lapurdum 3. 53–75. [Bayonne, France.] Rebuschi, Georges. 2000. A propos d’une construction non-standard du basque. Lapurdum 5. 237–282. [Bayonne, France]. Rebuschi, Georges. 2001. Note sur les phrases complexes à protase corrélative du basque. Lapurdum 6. 261–289. [Bayonne, France]. Rebuschi, Georges. 2005. Generalizing the antisymmetric analysis of coordination to nominal modification. Lingua 115 (4). 445–459. Rebuschi, Georges. 2006. A new look at Northern Basque wh-free relative clauses. In Beatriz Fernández & Itziar Laka (eds), Andolin gogoan; Essays in Honour of Professor Eguzkitza, 779–798. Vitoria, Spain: Servicio Editorial de la Universidad del País Vasco. Rebuschi, Georges and Anikó Lipták. To appear. Northern Basque free relatives and the left periphery. In Hans G. Obenauer (ed.), Studies on the left periphery [provisional title]. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of grammar. Handbook of generative syntax, 281–337. Dordrecht: Kluwer. Rotaetxe Amusategi, Karmele. 2004. (E)ta enklitikoa sailkapen eskalarrean [=A scalar analysis of the enclitic (e)ta, in Basque]. Fontes Lingvæ Vasconum 95. 115–139 [Pamplona, Spain]. Srivastav, Veneeta. 1991. The Syntax and Semantics of Correlatives. Natural Language and Linguistic Theory 9 (4). 637–686. Tiffou, Etienne & Richard R. Patry. 1995. La relative en bourouchaski du Yasin. Bulletin de la Société de Linguistique 90 (1). 335–390. Trask, Robert L. 1997. The history of Basque. London: Routledge. Tredinnick Victoria. 2003. Modal Flavor and Quantificational Force in Free Relatives with ever. NELS 34 poster session, available at: http://www.ic.sunysb.edu/clubs/nels/abstracts/tredinnick.pdf. (17 December 2008) Zaefferer, Dietmar. 1990. Conditionals and Unconditionals in Universal Grammar and Situation Semantics. In Robin Cooper et al. (eds), Situation Theory and its Applications I, 471– 492. Stanford, CA: CSLI.



Basque correlatives and their kin  Zaefferer, Dietmar. 1991. Conditionals and Unconditionals: Cross-linguistic and Logical Aspects. In Dietmar Zaefferer, (ed.), Semantic universals and universal semantics, 210–236. Dordrecht: Foris. Zribi-Hertz, Anne & Jean-François Hanne. 1995. Pronoms, déterminants et relatives en bambara de Bamako. Linguistique africaine 15. 91–135. Zuazo, Koldo. 1998. Euskalkiak, gaur. [The Basque dialects, today]. Fontes Lingvæ Vasconum 78, 191–233. [Pamplona, Spain].

Endnotes 1. Chomsky (1956) mentions similar sentences, namely “If S1, then S2” and “Either S3, or S4” as examples of “dependencies” that Markovian chains cannot account for, but, as far I am aware, he never dealt with them or their like in any of his transformational models. 2. The abbreviations used in the glosses are the following: abl, ablative; abs, absolutive; acc, accusative; adn, adnominalizing suffix; alloc, allocutive; aux, auxiliary; C°, see comp; com, comitative; comp, complementiser; d, dative; dem, demonstrative; e, ergative; fact, factitive; fut, future; gen, genitive; hum, human; imp, imperative; impf, imperfective (also iterative, as far as Basque is concerned); inst, instrumental; loc, locative; part, partitive; prf, perfective; pl, plural; pot, potential; pros, prospective; prtc, particle; pst, past; rel, relative marker; sg, singular; subj, subjunctive; unacc, unaccusative. The symbol ‘√’ indicates that a verb is devoid of any aspectual suffix. The zero or absolutive case suffix is not glossed in the Basque examples. In the glosses of the Basque auxiliaries, the cross reference to the arguments will be represented as follows: ‹a3e1›: ‘absolutive, 3sg–ergative, 1sg’, ‹a5d6›: ‘absolutive, 2pl–dative, 3pl’, etc.; ‘5’ denotes the morphologically plural, but referentially singular, polite 2nd person, and ‘5’’ the referentially plural second person. Frequently used abbreviations in the text are: CCS: complex correlative sentence; CorP: correlative protasis; FinP: Finiteness Phrase; FR: free relative clause; FVF: finite verb form; IFR: indefinite free relative. The abbreviations concerning the references are: Imit: Thomas à Kempis’ De Imitatione Christi and its various translations into Basque; KJV: the King James’ Version of the Bible; LV: Latin Vulgate. Reference to the passages of the Bible will be given according to custom. 3. Not to mention the fact that only a very few pages are devoted to CCSs in Oyharçabal (2003: 818–821). 4. Quite a few counter-examples to that generalisation can be found until well into the 18th century, but not later on, if I am not mistaken. 5. Bait- normally loses its final ‑t before plosive consonants, but in the frequent ‑t+d- sequence (where d‑ encodes a3), the d‑ is generally dropped, except for some authors who drop the ‑t, e.g. baita or baida, bait- + da ‘he/she/it is’ ≈ ‹a3›. 6. Conditional ba- ‘if ’ is unstressed, as in (9a). In (9b), however, the positive assertive particle, whose presence is triggered by the FVF being in sentence initial position, is stressed. 7. This is an oversimplification. Although these relatives are typically restrictive, they can also convey an appositive value. See (i) for instance: (i) Acts 15,8 – LV ‘Et qui novit corda Deus testimonium perhibuit…’ KJV: ‘And God that knows about the hearts, testified about them.’

 Georges Rebuschi

Bihotz-e-tako berri daki-en Jainkoa-k, lekhukotasun heart-pl-gen news knows-en God-e testimony egi-n izan du. do-perf aux.perf ‹a3e3›

(Duvoisin 1859–65)

More recent Northern translations, such as “Ezkila” (1974), and even the now received ecumenical translation of the Bible (EAB 1994), much more typical of the Central and Western dialects, offer the same type of construction for this verse. 8. The ending ‑ak corresponds both to absolutive plural, and ergative singular; here, it is the former reading that is compatible with the grammatical context, but in (10a,b) or (11a), it is the latter. In the word for word glosses, the irregular absolutive plural ending is not segmented: “‑ak”, but the regular ergative singular is: “‑a‑k”. With visible case endings, such as ergative ‑k or instrumental ‑z, plurality is unambiguously marked by the morpheme ‑e‑. 9. Or almost always: none are to be found in Etchepare (1545). 10. Leiçarraga’s sources are complex, cf. Lafon (1944, I: 50), but some Greek version was pro­ bably used along with a French translation (recall that a Greek text was also the basis of the King James Version). 11. I have added this example, written in euskara batua or “unified Basque”, a standardised written language which borrows much more from the Southern (Central and Western) dialects than from the Eastern ones, among which those spoken in France. This is because I intend to show that the construction and its interpretation have remained stable during the whole history of the language, and therefore cannot be regarded as marginal. 12. The cases illustrated (21a-c) are obviously more complex. One possible account is to consider that the first case-marked conjunct refers to a maximal individual, and that eta materializes the connective that is usually, but not always, left phonetically unrealised cross-liguistically when a restrictive relative modifies its antecedent (see Rebuschi 2005 for independent arguments and evidence). 13. Note that the restrictive relative clause within this DP follows, rather than precedes, the head-noun. See also (48a) infra. On the (non-un)grammaticality of this word-order, see Oyharçabal (1987). 14. Iraun ‘to last, persist’ is a so-called “unergative” verb; as a consequence, its unique argument is in the ergative case (see Levin 1983 on this). 15. The last clause in (32b) also contains a zero resumptive or correlative demonstrative (a little pro). Let us add that the case clash in the second conjunct of (29a) between the wh-item nor-i (which is in the dative because it is the indirect object of the verb manifestatu ‘impart’) and the case required for the subject of the main clause predicate ‘know the Father’ is only licit – at least today –, precisely because it is a second conjunct: otherwise, such structures have been only possible since the 18th century if the same morphological case is assigned both by the main clause verb, and by the embedded verb too. A detailed analysis of this evolution must be left for future research. 16. There is no distinction between the absolutive and ergative cases for this plural pronoun, ‘ye-all’. 17. A finite verb is [allocutive] if it incorporates a morpheme indicating familiarity (as here) or, in fewer dialects, respect, with respect to the hearer, without the latter corresponding to an argument of the verb.



Basque correlatives and their kin  18. The periphrastic subjunctive mood is built with a second series of auxiliaries, e.g. with the root ‑eza- rather than ‑u- for ‘have’ with transitive verbs, and the root of (or perhaps a zero aspectual suffix on) the lexical verb. As for the Basque periphrastic “future”, it is built by the combination of the prospective aspectual ending on the lexical verb and the inflected auxiliary in the present tense. 19. Interestingly, in older Biscayan, CCSs existed, and could be in the subjunctive, as shown in (i) (in Southern Basque, what are verbal roots in Northern Basque also carry the ‑perf suffix, but the root of the auxiliary is different from the one used in the indicative mood). (i) Jn 20,23 LV: ‘Quorum remiseritis peccata, remissa sunt eis; quorum retinueritis, retenta sunt.’ / KJV: ‘Whose soever sins ye remit, they are remitted unto them; and whose soever sins ye retain, they are retained.’ Nor-tzu-ben pekatu-bak parkatu daikezubeza-n, who-pl-gen sin-pl forgive ‹a6e5’[subj]›-en parka-tuko jakez, eta nor-tzu-ben-ak geratu forgive-pros ‹a6d6› and who-pl-gen-pl leave daikezubeza-n geratuak izango dira. ‹a6e5’[subj]›-en leave.perf.pl aux.pros ‹a6› (Astarloa 1816–18, cited in Arejita 1978: 135) Moreover, the subjunctive mood can also be used in SFRs which have a generic or universal value, in certain varieties of Bizcayan Basque again. A 20th translation of the Bible into this dialect thus reads for the same passage: (ii) Zuek pekatu-ak parkatu dagiezuez-an-a-i you-pl sin-pl forgive ‹a6d6e5’[subj]›-en-pl-d parka-tu-ta gera-tuko yakez; zu-ek pekatu-ak forgive-perf-ta stay-prosp ‹a6d6› you-pl sin-pl lotu dagiezuez-an-a-i lo-tu-ta retain ‹a6d6e5’[subj]›-en-pl-d retain-perf-ta geratuko yakez (Kerexeta 1976) stay.pros ‹a6d6› Lit., ‘[Those to whom you will forgive [the sins]i]j, theyi will remain-to-themj forgiven; [those to whom you will retain [the sins]i]j, theyi will remain-to-themj retained.’ 20. Likewise, for Mt 5,31–32, the KJV has whosoever three times, Leiçarraga (1571) and Léon (1946) have three correlatives (in nork ere bait‑), and “Ezkila” (1974) and Etchehandy (1999) have three SFRs – but Duvoisin (1859–65) has two unconditionals and one SFR, whereas the Latin Vulgate respectively displays quicumque, omnis qui, and qui in that order. 21. Zein can also function as a wh-determiner or modifier, as in zein gizon? ‘which man/men?’ in interrogatives, or zein gizon ere ‘whichever man’ in CorPs, in which case no number affix can appear; on the contrary, in exclamative clauses, the number suffix is compulsory: Zein gizon*(a) (den)! ‘What a man (he is)!’ 22. Example (56) also shows that the default singular agreement in the FVF of the CorP was not the only option at that time (but it is nowdays), since it displays a plural morpheme for the argument in the absolutive case. Nowadays, a plural correlate is still possible in the apodosis, but within the protasis, the only possible agreement-marking in the FVF is the default value, the singular. 23. There is no known etymology for balin and baldin, and only the latter is mentioned (as a mere variant of ba‑) in Agud and Tovar (1989: 531).

 Georges Rebuschi 24. The situation was more confusing until the end of the 17th century. For instance, Axular (1643) did not distinguish between balin and baldin, but the description in the text is correct as far as later stages of the language are concerned – contra the description provided in Hualde & de Urbina eds. (2003: 724–726), which only reflects Southern Basque grammar. 25. The Basque texts are either translations of the NT itself, or of the same verse as quoted in Thomas’ Imitation, the original version of which contains the very words of the Latin Vulgate. 26. The relevant FVFs are characterized by a prefix l- rather than d- in the present tense, and z- in the past tense, when the argument or all the arguments are 3rd person(s). But if there is a 1st or 2nd person prefix, it is identical with the one used in the past tense, the difference lying in the fact that the irrealis FVF is not followed by the past tense marker ‑(e)n. 27. The author, and the dialect of his book, are Guipuzcoan, i.e. Central, not Eastern, Basque, but this sentence is clearly borrowed from a neighbouring dialect: P. Altuna, in his introduction to the facsimile republication of the book in 1989, explicitly wonders whether the sentence is really Guipuzcoan, i.e. Central, Basque – and it is obviously not. 28. On the one hand, diphthongs devoice a following voiced plosive at morpheme boundaries, as in N.B. zai-ku ou zau-ku for /zai‑gu/‘it is to us’ ‹a3d4›, and transform fricatives into affricates, as in N.B. d(er)au-tzu for /d(er)au-zu/ ‘s/he-has-it-to-you’, cf. standardised and Guipuzcoan Basque di-zu, id. On the other hand, a ‑t- often appears for instance in compounds such as sutondo ‘fireplace’, from su ‘fire’ and ondo ‘side’. Whence for instance, on the one hand, baikara for bai(t)-gara ‘bait-we-are’, and, on the other hand, baitaiz for bai-(h)aiz ‘bait-thou art’ (as in (46) in 4.2). 29. The symbol “◊” indicates an obviously unattested, but reconstructed, (word or sentence) form, to be distinguished from the use of ‘*’ as marking ungrammatical ones. 30. Nehor (and its variant nihor) results from the prefixation of the negative morpheme ez, whose consonant traditionally fell before n- and l-, as is attested by the traditional spelling e-naiz of ez naiz ‘I am not’, or e-lianak in (36). The evolution must have been: ◊e-nor > ◊ne-nor (under the influence of surrounding Romance negative morphemes) > nehor, with a well-attested replacement of intervocalic -n- by -h- (on “◊”, see endnote 29 above). 31. The position of the negative morpheme ez to the left of balin indicates that balin plus ba- are one syntactic word – as opposed to baldin… ba‑, since in this case ez must immediately precede ba‑. A clear illustration is provided by Larréguy (1775, chap. 31): (i)

Othoitz-tu zuen Laban ez gaitzezte-az pray-perf ‹a3e3[past]› Laban neg getting.angry-inst baldin ez balin ba-zen altxa-tzen hari baldin neg balin if-‹a3[past]› get.up-impf dem-d agur egiteko. hail to.do



‘She prayed Laban not to be angry if she did not stand up to greet him.’

32. See Caponigro (2004) for a variant, according to which the translation of the wh-words is: λXλx[P(x)∧X(x)], where the property variable P ranges over animates, inanimates, location, manner, according tho the specific morpheme associated with the wh-prefix (who, what, where, how… respectively). 33. The well-known ban on indefinite (pro)nominal correlates in the apodoses – one of the strongest arguments in favour of a “maximalisation” analysis of the semantic contribution of CorPs across languages – can be analysed differently. If indefinite pronouns and nominal ex-



Basque correlatives and their kin  pressions are interpreted either as properties or as variables ranging over entities, the translation of the CorPs themselves into properties will of course make no sense. But if correlates like hura lit. ‘that one’ or haina lit. ‘the such’ are interpreted as iota-like expressions, their translation will necessarily contain two property variables – one of which will be bound by the contents of the CorP. If haina is rather analysed as a universal quantifier, the CorP will provide the required domain of restriction. 34. Interestingly, the presence of a pronominal correlate is not necessary either, cf. If anybody enters the room, tell him/her/them to go away, but also If anybody enters the room, pretend you’re asleep. However, the relevance of the content of the protasis is indisputable in both cases. 35. The references followed by a star (*) are freely downloadable at: http://klasikoak.armiarma. com/alfa.htm.

{Relative {conditional {correlative clauses}}} Boban Arsenijević After introducing an analysis of conditional clauses as yes-no relatives, this paper presents an analysis of correlative constructions in terms of conditional clauses. As opposed to many accounts of correlative clauses that treat the correlated argument as the relativized one, the present analysis takes the actual relativization site in correlatives to be the functional projection determining the set of worlds compatible with the propositional meaning in its complement, similarly to conditional clauses. The relative phrase is treated as an extreme non-specific expression with a topic feature, which undergoes movement to the left periphery. The correlate demonstrative in the matrix clause is argued to be purely anaphoric, without any syntactic relation with the extreme non-specific expression in the relative clause. Keywords: conditional, interrogative, weak existential, relativization, Serbo-Croatian

1. Introduction This paper proposes a novel analysis of correlative clauses, as conditional clauses involving extreme non-specific wh-expressions with some kind of topicality. I argue, based on a large number of semantic, syntactic and morphological similarities, that correlative clauses are a subtype of conditional clauses. The wh-expression involved in the correlative clause is analyzed not as a relativizer, but as an independent expression, sensitive to polarity environments, and fronted for its topical discourse nature. The demonstrative in the main clause is analyzed as an anaphoric expression, referring back to the discourse referent introduced by the wh-expression in the correlative clause. Section 2 prepares the ground for the analysis of correlatives by presenting an outline of an analysis of conditional clauses as yes-no relative clauses. Conditional clauses are presented as relative clauses with their truth value, i.e. the set of worlds compatible with their meaning, lexicalized by a relativizer: the conditional marker, counterpart of the English if. This relativizer relates the clause with the host head from the head clause. This host, in turn, can be lexicalized by a conditional demonstrative, counterpart of the English then, or it can be left without lexicalization.

 Boban Arsenijević

In Section 3, I discuss wh-expressions involved in correlativization. I argue that all wh-expressions involve the feature of extreme non-specificity, i.e. a weak existential commitment. They can appear in three flavors: question words, characterized by the additional presence of a question-feature, relativizers, characterized by the relation they establish with the head, and independent extreme nonspecific expressions, characterized by polarity sensitivity. Based on data from Serbo-Croatian, I argue that the latter include the correlating wh-expression of the correlative clause. The core of the analysis of correlative clauses in terms of conditional clauses is presented in Section 4. A number of semantic, syntactic and morphological similarities between the two clause types, such as maximization effects, preference for preposing, overt appearance of conditional markers in both the head and the correlative clause of correlative sentences, and constructions which seem to have all properties of both clause types, are taken as a starting argument in favor of providing a unified analysis for the two types of clauses. I propose an analysis, which presents correlative clauses as conditional clauses with several special properties, among which the presence of a topical extreme non-specific wh-expression referred to by an anaphoric demonstrative in the main clause. Additional semantic parallels, supporting the proposed analysis are provided in Section 5. Here, the distinction between two types of conditionals, those generated at the IP level, and those generated at the CP level, is transferred into the domain of correlatives, and shown to hold there as well. Section 6 discusses the correlative demonstrative. I first show that major arguments for the analysis of the correlative clause as a relative clause modifying the demonstrative can be overcome, and do not exclude the present analysis. Then, I discuss a number of special properties of the correlative demonstrative observed in Serbo-Croatian, such as its relation with the conditional demonstrative and the systematic ban on the use of one member of the demonstrative paradigm in correlative sentences. Section 7 concludes.

2. Conditionals: yes-no relative clauses It is an old linguistic observation that questions and relative clauses have something in common, and many attempts have been made to formalize this similarity. The contemporary linguistic theory disposes with deep insights into the nature of both related classes, and many parallels can be identified. Both questions and relative clauses involve an underspecified, variable-like element that requires further specification. In questions, this requirement is marked by the interrogative force, which introduces an expectation of further specification in the discourse. In relatives, the required specification comes from the head that hosts the relative clause. These considerations set the ground for explaining other parallels between the two classes, such as the tendency to displace the wh-element into the higher structural domains.



{Relative {conditional {correlative clauses}}} 

It is normally considered that the parallel as outlined only relates to one kind of questions – those involving wh-elements. This implies that yes-no questions have no relative counterparts. Were this correct, it would be a curious fact, requiring an explanation in terms of either treating yes-no questions as improper questions or the parallel observed as a partial one. I argue, however, that this is not correct, and that there are yes-no relative clauses – those usually referred to as conditional clauses – with the same degree of parallelism with yes-no questions as in the wh-paradigm. Let me outline several empirical facts pointing towards this parallel. First, in many languages conditional clauses, just like yes-no questions, can be derived by subjectverb inversion (1a). Moreover, in a number of languages, markers usually involved in yes-no questions are also used in conditional clauses, as in the examples in (1b-c) from Serbo-Croatian (S-C for short).1

(1) a. Had you given me some trust, we might have had a nice friendship.

b. Imaš li problema? have.2s cl.interr problems.gen ‘Do you have any problems?’

[S-C]

c. Imaš li problema, zovi šefa. have.2s cl.interr problems.gen call.imp boss.acc ‘If there are any problems, call the boss.’ Parallels in the other direction, with conditional markers appearing in questions, are also common. A nice example is the English embedded yes-no question in (2).

(2) I asked him if he had any coins, and he didn’t answer.

On the semantic side, parallels are equally striking. Yes-no questions have the interrogative force operating over their truth-value, i.e. over the set of worlds compatible with the proposition denoted by the sentence. Conditional clauses establish a dependency between their truth-value and the truth-value of the head clause. The proposition represented by the conditional clause restricts the set of worlds compatible with the proposition represented by the head clause. In other words, a proper relative clause emerges. If a declarative main clause normally asserts in the following manner: the relevant worlds are compatible with the predication P, conditional clauses have the form: the relevant worlds that are compatible with the predication Pcond. are compatible with the predication P. Observe the example in (3).

(3) If John can’t swim, he‘ll have to walk around the lake.

The two clauses of this sentence share the same discourse and hence the same contextual restrictions. Their meaning can be paraphrased as reference to the possible worlds in which John has to walk around the lake, restricted by their compatibility with his impossibility to swim.

 Boban Arsenijević

This is, in fact, expected: if subjects, objects, place, time, manner, and basically all standard syntactic functions, can take relative clauses, why would the truth-value of the sentence be an exception? Descriptively speaking, relative clauses are restrictive modifiers of different types of linguistic elements. Moreover, they are modifiers that have the broadest domain of use, in terms of structure and categories. In language use, when the intended modification cannot be realized by an adjective, adverb, or other lexical modifier, relative clauses usually come handy: they are not picky about the category to attach to, and they realize a very wide range of, possibly very rich, meanings. The truth-value of the sentence is clearly modifiable – there are a number of lexical modifiers, such as always, usually, seldom, that are used for this purpose. It would be strange under these circumstances if there were no relative clauses with the same effect. That conditionals modify the same semantic component that is modified by adverbs like usually, seldom, always is not a novel claim. Kratzer (1986) suggests that conditionals restrict epistemic modals, which means that a) head clauses in conditional sentences involve some (silent) modality, and b) this modality, similar to those realized by epistemic modal adverbials, is restricted by the conditional clause. Hence, conditional clauses are restrictive modifiers attaching at approximately the same range of structural positions as the epistemic adverbs. The only ingredient that needs to be added is that they are to be seen as relative clauses. I argue that this not only has the methodological advantage of unifying two classes of subordinated clauses into one, but also is empirically correct, and opens space for a different analysis of other phenomena, such as correlative clauses.2 In order to show that conditional clauses, as epistemic modifiers, are to be seen as relative clauses, it is useful to first identify a relativizing element in the CP of the conditional that relates the relative clause with its head. In the purest case, this means that the relativizer has to represent the truth-value (the epistemic outcome), or the set of worlds compatible with the conditional clause, which restricts the corresponding semantic component of the head clause. I propose that the relativizer in conditional clauses is the if-element, i.e. its counterpart in any other language. Note that in some languages, like German, this element morphologically and etymologically belongs to the class of wh-elements. What distinguishes conditional clauses from other relatives is their usual lack of overt hosting heads. Like any other relative clause, the conditional clause should have a hosting head, and like any linguistic element, this head is expected to have overt phonological marking. A slight problem for this is that, usually, lexicons of natural languages do not have overt elements surfacing in the head of the projection encoding the (possible) world(s) in which the proposition is to be interpreted. But then, the domain involved likes to borrow from the temporal domain – the semantics of epistemic modifiers, for instance, often involves temporal meanings (e.g. seldom, always, frequently). then (i.e. this being the case, in the worlds with this property), why not borrow one to represent the head as well? What kind of element do we need? It should be light, like a pronoun, and it should have a temporal meaning. Languages



{Relative {conditional {correlative clauses}}} 

usually have no temporal personal pronouns, so the next best choice is a temporal demonstrative, i.e. then (meaning something like ‘this being the case’, or ‘in the worlds with this property’), as in (4).

(4) If John can’t swim, then he‘ll have to walk around the lake.

The sentence in (4) may have some different semantic or pragmatic effects from that in (3), coming from the overt lexicalization of what I argue to be the (possible) world(s) head (i.e. from its discourse function that licenses its overt realization), but still, they share the default, neutral, conditional meaning. Then in this sentence really acts as a proform, it makes the hosting head of the relative clause overt, deriving the default pattern of a relative clause.3 However, the lexical semantics of this proform brings in the feature [distance], characteristic of demonstratives and specified for then to bear the value [distal] (as opposed to ‘now’ with the value [proximal], see Jayaseelan & Hariprasad 2001 for a discussion). This aspect becomes relevant in the domain of correlatives, and is discussed in Section 6. The structure we arrived at is presented in Diagram 1. The head of the projection related to the truth-value of the clause, marked here neutrally as WorldP, contains the feature world with any possible value. This means that the sentence does not necessarily pick up the actual world: it can also refer to (sets of) possible worlds, or can have a generic meanning. It specifies that whatever worlds are referred to by the WorldP, these are compatible with the predication generated below WorldP. In addition to this, WorldP is restricted by a relative clause, with the same feature [world] directly under the scope of a weak existential quantifier weak_∃ (discussed in Section 3) and with another predication. The interpretation derived, in which worlds compatible with the meaning of the relative clause restrict over the worlds compatible with the head clause. Just like with other restrictive relative clauses, an asymmetry arises: the dependency involved is oriented. WorldP

weak_∃ /if/

[world] /then/

[world] …

Diagram 1.

…

 Boban Arsenijević

The relative clause restricts the head clause, so that the set of worlds denoted by the head clause before the modification by the relative clause is a superset of the set resulting after the modification. Due to the conservativity of modification, the denotation of the relative clause has to be determined within the domain specified by the denotation of the head clause, which it modifies. Further asymmetries may arise when one of the two clauses is additionally epistemically specified, by overt modifiers or from the discourse. For instance, the proposition in the conditional clause may be old in the discourse, but still without explicitly established quantification (e.g. subject to question, or other uncertain status of the world(s) involved – or simply subject to expressive and not assertive use). I come back to these issues in Section 5, but only to the extent it is relevant for the discussion of correlative sentences. Before moving on to discuss aspects relevant for correlative clauses, let me briefly highlight one interesting consequence of the presented analysis of conditionals. An old debate in the analysis of conditional clauses, stemming, to my knowledge, from Lewis (1975), relates to the capacity of conditional clauses to (seemingly) act like quantifier restrictors. Observe the near equivalence of the sentences in (5) and (6).

(5) Every student will pass the exam if she answers at least 5 questions correctly.



(6) Every student that answers at least 5 questions correctly will pass the exam.

Moreover, conditional clauses show different variants of what Lewis (1975) refers to as unselective binding. This is best exemplified by the so-called donkey anaphora sentences as in (7).

(7) If a farmer owns a donkey, he usually beats it.

The availability of the readings referred to as symmetric, where pairs of farmers and donkeys are predicated over (if they satisfy the own predicate, they usually satisfy the beat predicate) signals that it is not the regular kind of restriction. In conditional clauses, it seems that the restriction takes a broader scope, and then somehow matches the restricted quantifiers with the structural positions of the conditional clause ‘in which’ they are restricted. Higginbotham (2003) shows that the two types of sentences in (5) and (6) have different truth conditions. The difference can be described as the availability of broader, even generic, readings for the conditional sentences. More precisely, even referents outside of the set that the conditional clause determines for the relevant arguments in the actual world, have to satisfy the implication specified in the conditional sentences – even for the worlds which turn out not to be the actual world, but only possible ones. To illustrate, observe again (5) and (6), and imagine that there are 10 relevant students, and there is one among them, who wants a very high grade: she would accept to pass the exam only if she answered 10 questions correctly. This immediately makes at least one reading of (5) false, irrespective of what was really going to happen in the exam,



{Relative {conditional {correlative clauses}}} 

because there is a possibility that this described student answers 9 questions correctly, and still not pass the exam. Now take a simple situation in which 3 students answered more than 5 questions correctly, and 7 did not. If the three students that did it pass the exam, i.e. if the ambitious student is not among them, (6) is true. Hence, while (6) is only sensitive to the actual world, (5), at least in some of its readings, extends its ‘restriction’ into possible worlds. No matter what actually happened, if there was a possibility of an outcome in which the sentence in (5) would be evaluated as false – it receives the general value false. Or, in yet another formulation, both sentences talk about the future, but the sentence in (5) talks about any possible future scenario, while the one in (6) talks only about the one that turns out to be part of the future of the actual world.4 This asymmetry follows straightforwardly from the analysis presented in this section. Donkey anaphora sentences then follow as well. The conditional clause restricts over possible worlds. If it introduces a non-specific expression, it will also introduce a referent that is part of the restriction over the possible worlds introduced in the very conditional clause. Due to restriction over worlds, such a referent will also be present in every world referred to by the main clause. Therefore, and especially with some help from the pragmatics, it will be a prominent referent to be picked up by an adequate proform (personal, clitic, demonstrative) in the main clause. This is how the conditional clause provides restrictions for two arguments of the main clause in (7). This aspect is further elaborated on in the following discussion.

3. Wh-expressions as extreme non-specific items In this section, I argue that the minimal semantic content shared by all wh-expressions is the weak_∃ quantification, i.e. the non-specific interpretation with a weak existential commitment (see Farkas 2002 for an extensive discussion of this property). I present the behavior of wh-expressions in S-C, as well as in other languages, when they are used without the relativizing or question-forming function, as weak Polarity Sensitive Items (PSI), and argue that wh-expressions, which involve weak_∃ quantification, when surfacing high enough in the structure, may ‘take over’ the lexicalization of the weak_∃ quantification over the [world] feature of the conditional clause. Such conditional clauses surface with a fronted wh-element and without an obligatory conditional marker, constructions usually referred to as correlative sentences, the topic of Section 4. Observe the three contexts of the use of bare wh-elements in (8): as question words, relativizers and PSIs.

 Boban Arsenijević

(8) a. b. c.

Who kissed Mary? The one who did it should be ashamed. Jesi li doneo šta? aux.2s cl.interr brought what.acc ‘Did you bring anything?’ (lit. ‘Did you bring what?’)

[S-C]

The same holds for heavier wh-expressions in (9) (9) a. b. c.

Which boy kissed Mary? The box which fell from the table was yellow. Jesi li doneo koju olovku? aux.2s cl.interr brought which.acc pen.acc ‘Did you bring any pen(s)?’ (lit. ‘Did you bring which pen?’)

[S-C]

In the first two uses (8a-b)/(9a-b), wh-elements have interpretations related to the higher clausal domain (the left periphery). In the PSI use, their interpretation is, at first sight, independent of the CP domain. However, under a closer look, even in this use, wh-elements involve some discourse-oriented semantics. More precisely, they involve instructions that the discourse domain in which they refer can, or must, be broadened (cf. Chierchia 2006) and that there is no commitment about the quantity, or even existence of any salient referents for the expressions involved. S-C wh-PSIs, although with morphological singular forms, have denotations unspecified for quantity: they can refer to one or more counting units of the relevant kind, and when pragmatically salient even to parts of individual units. This brings S-C wh-PSIs very close to Farkas (2002)’s definition of extreme non-specifics, but even less specified than their Romanian counterparts, since they are also unspecified for quantity. Indeed, they satisfy all the requirements and fit all the contexts that Farkas introduces as characterizing properties of these elements (most prominently, they are good in ‘environments which neither assert nor presuppose the existence of a verifying value in the relevant context’ – Farkas 2002:  135). I refer to these elements in the rest of the paper as the wh-Polarity Sensitive Items (wh-PSIs). Arguably, the other two uses of wh-items, questions and relative clauses, also involve the weak existential commitment. This question is not directly relevant for the topic of the paper, and hence will not be discussed.

4. Correlatives: Conditionals with topical wh-PSIs If we now combine what we have: the analysis of conditional clauses as world-related relative clauses and the analysis of wh-elements as extreme non-specific items, we can represent conditional clauses as clauses with a weak_∃ quantification in the WorldP. An intriguing empirical question emerges: what happens if a conditional clause involves an additional wh-expression (an argument or an adjunct) with weak_∃



{Relative {conditional {correlative clauses}}} 

quantification? The simple case is presented in (10), where a conditional clause, as a well-behaved environment, takes an extreme non-specific wh-expression (so far, I have ignored the demonstrative in the head clause; it will be discussed in Section 6). (10)

Ako je (ko) već (ko) ustao (*ko), (taj) [S-C] if aux who already who raised who that (onda) neka (taj) i izađe (*taj). then let that and go.out that ‘If anyone already stood up, (then) let him also go out.’/ ‘Whoever stood up, let him also go out.’

Note that even in an overt conditional clause, the extreme non-specific wh-expression must move to the CP domain. It appears after the conditional marker (in the present approach the yes-no relativizer), after the second position clitic-cluster, or after a topical adverbial, but no other element may appear to its left, as illustrated in (10). Another related requirement of correlative clauses applies: the demonstrative has to be fronted in the head clause, unless it is focalized and stressed and the rest of the clause is topical. However, the demonstrative can also be preceded by the topicalized material and the second position clitic cluster. While the wh-expression in the correlative clause cannot drop, the demonstrative can also be left out. A special case of conditional clauses with extreme non-specific wh-expressions is when the wh-expression is topicalized. In such a case, its topic feature drags it to CP. This, I argue, derives correlative sentences. One of the intermediate landing positions of the wh-expression in such a conditional clause is SpecWorldP. This results in establishing a dependency between the interpretation of the wh-expression and that of the clause, i.e. between the weak_∃ quantification components in the two phrases. The dependency is interpreted as a kind of distribution: the worlds denoted in the correlative clause vary with the referents chosen for the wh-expression, which in turn has a free-choice capacity (due to extreme non-specificity). This gives the structures in (11), represented in Diagram 2. (11) a. b.

Head clause [… [WorldP [Rel.Clause] [modal value/Ø]] … ] Relative (i.e. conditional) clause [TopP [WhP weak_∃ [topic]] [topic] [WorldP [WhP weak­_∃ [topic]] weak_∃ [world] …]

 Boban Arsenijević TopP Top'

weak_∃ [topic]

WorldP

[topic] weak_∃ [topic]

World' …

Diagram 2.

I will not go into a deeper discussion of the value of the WorldP head in the head clause, but would like to note that it may contain no value at all, in which case the modal semantic component is simply an implicature. Even if its value is that of the actual world, the actual world gets further restricted, and modalized, by the conditional clause. The presence of a restrictive modifier implicates restriction over a broader set, which could only contain possible worlds. On the other hand, the more costly solution of simply marking by a feature in the head of WorldP that the worlds are possible is available as well, just slightly less attractive. Perhaps there can be empirical ways to decide. In the relative clause, which is conditional because what it restricts over is worlds, there is a topic projection, taking an extreme non-specific wh-expression. This expression is base generated in some lower position. On its way to TopP, and especially because it involves weak_∃ quantification, it had to have an intermediate landing in WorldP. This structure is the one I propose for correlative (and preposed free relative) clauses, illustrated in (12).5 (12) Ko je ustao, (taj) (onda) neka i izađe. [S-C] who aux raised that.one then let and go.out ‘Let those who stood up (also) go out.’/‘Let anyone who stood up (also) go out.’ This example is just a variant of (10), with the non-specific wh-expression being topical. Its intermediate landing in WorldP licenses the absence of lexical realization of the head of this projection, usually realized as ako ‘if ’. However, the drop of ako is optional, and although it is strongly preferred, in certain contexts ako appears overtly.



{Relative {conditional {correlative clauses}}} 

(13) a.

Ko na brdo ak’ i malo stoji, više vidi od onog [S-C] who on hill if and little stands more see from that pod brdo. under hill. ‘The one who stands on a hill even a little bit sees more than the one at the foot of the hill.’/‘Anyone who stands on a hill even a little bit sees more than the one at the foot of the hill.’

b.

Ko tebe ako nečim udari, who you.acc if something.inst hits ti tome vrati istom merom. you that.one.dat return.imp same.inst measure.inst ‘If anyone hits you with something, hit him in the same way as he hit you.’ / ‘Whoever hits you with something, hit him in the same way as he hit you.’

In a number of other languages, or dialects, a counterpart of if can, or even has to, appear in the correlative clause (cf. Tibetan in Cable this volume, Dutch dialects in den Dikken this volume). The drop may be related to the general tendency of a language to elide redundant material. Anyway, the fact that correlative clauses can contain an overt if-element, supports the claim that correlatives are a special kind of conditionals. Similarities between correlatives and conditionals have been observed and discussed for a couple decades by now. Among the most obvious parallels are the maximization effects (over the possible worlds compatible with the condition and over the referents satisfying the description from the extreme non-specific wh-expression), preference for preposing (conditional clauses, and even more so correlative clauses, require special intonation and information structure when they appear to the right of the head clause) and the possibility of deriving generic statements (see (14)) (for a more thorough discussion of the similarities between the two types of sentences, see among others Bittner 2001, Bhatt & Pancheva 2006, Rebuschi this volume). In addition to this, the following aspects need to be mentioned. Cheng & Huang (1996) argue that a distinction is to be made between conditional and correlative clauses. Their argument is primarily based on symmetric and asymmetric readings (briefly commented on in Section 5). They link the availability of symmetric readings in sentences with multiple non-specifics to (donkey) conditionals, and the lack thereof as a property of correlatives. A symmetric reading of (14a) would be that for any pair of a farmer and a donkey, if they enter the owning relation with role distribution as in the conditional clause, then they also enter the beating relation, as specified in the main clause. The construction is asymmetric if it bans this type of readings, and allows only for single elements to be varied across the domain (for any farmer, if there is a donkey…). It appears, however, that S-C correlatives, and Hungarian too, according to some speakers at least, do get symmetric readings. This is the case with the sentence in (14b), which is why one of the English translations of (14b) nearly

 Boban Arsenijević

corresponds to (14a). Availability of symmetric readings hence appears to be a parametric phenomenon, rather than a universal difference between constructions. (14) a. b.

If a farmer owns a donkey, he usually beats it. Ko ima magarca, taj ga i bije. [S-C] who has donkey, that cl.acc and beats ‘Anyone who owns a donkey beats it.’/ ’The one that owns a donkey beats it.’ / ‘For any donkey and any person, if the person owns the donkey, she also beats it.’ (this, symmetric, reading is more available if ima ‘have and magarca ‘donkey’ invert their order)

Bittner (2001) argues, on theoretical grounds, that conditionals are a type of correlatives. She approaches the issue from the aspect of reference to possibilities, and points to the similarities between correlative and conditional clauses in this respect. She argues that correlative clauses systematically have one reading in which they involve possible worlds, and that, with some simplification, conditional clauses are a type of correlatives involved most directly in the domain of possibilities, rather than individuals. Rebuschi (this volume, Section 5) contributes some empirical arguments indicating the close relation between conditionals and correlatives. Most strikingly, he provides spontaneous Basque translations of Latin and English sentences, where conditionals are often translated into correlatives, as well as the other way around. Even the same sentence often receives different translations, some as a conditional, and some as a correlative. Rebuschi reasons that similarities between the two constructions are mostly semantic, and do not entail a shared syntactic structure. This paper takes a stronger position, and argues that correlatives are a subtype of conditional clauses, i.e. that they share part of their structure, but that correlatives have some additional special properties. In addition to the general fact that conditionals are more universally present in languages of the world, more concrete arguments, in particular lexical and syntactic ones, are provided in what follows. Among the empirical facts that support the analysis, most important are those related to the semantic similarities between conditional and correlative clauses. These involve complex issues, some of which are separately discussed in Section 5. Other arguments include the possible, or obligatory, appearance of an if-element in correlatives in some languages. It is strengthened by the even much deeper parallels between conditionals with extreme non-specific elements and correlatives, illustrated in (10) and (12). S-C correlatives show another relevant similarity with conditionals: at least some of them may involve the temporal demonstrative ‘onda’, normally used in conditionals, as illustrated in (12) above. The same holds for the closest Cantonese counterparts of correlatives, as in (15), where zau ‘almost’ is the same word typically used in conditional clauses as the counterpart of the English then (Joanna Sio, p.c.) (see footnote 3).



{Relative {conditional {correlative clauses}}} 

(15) yau muigwai ge neoizai zau hoji japlei have rose adn girl zau can enter ‘The/any girl with a rose can enter.’

[Cantonese]

The possibility of lexical realization of elements that bear conditional semantics or are typically used in conditional sentences (cross-linguistic counterparts of if and then) indicates that the conditional structure is underlyingly present in correlative constructions. For different reasons, from information structure to semantic relations between quantifiers involved, these elements are in most cases left without overt lexicalization. In addition to this, as Rebuschi (this volume) points out, Basque correlatives involve a separate paradigm of wh-elements, distinct from those normally involved in relative clauses in being not specified for number. This may be a consequence of certain agreement that relative wh-elements undergo in the structure they appear in, and independent wh-expressions do not. Finally, when S-C conditionals and correlatives are embedded under saying/believing verbs, they trigger the same effects. Unless they are fronted to the left of the main clause, overtly epistemically modified, or separated by an intonation break, embedding results in strong degradation. This is illustrated in (16). What is more, both the conditional and the correlative clause only receive the possible worlds readings, even if the context suggests an actual world reading with definite referents only (apart from the extreme non-specific wh-expression, which naturally cannot be definite). (16) a.

Jovan kaže da ??(#) ako je [S-C] Jovan says that [pause] if aux Marija ispraznila frižider pre odlaska, Petru Marija emptied fridge before leaving Petar.dat nije ostalo hladnog piva. not.aux stayed cold.gen beer.gen ‘Jovan says that if Marija *(indeed/actually) emptied the fridge before leaving, Petar was left without cold beer.’

b.

Jovan kaže da ??(#) ko je ispraznio frižider pre Jovan says that [pause] who aux emptied fridge before odlaska, taj nije mislio glavom. leaving that not.aux thought head.inst ‘Jovan says that whoever emptied the fridge, he didn’t think with his head.’ / *‘Jovan says that the one who emptied the fridge didn’t think with his head.’

Fronting makes both clauses grammatical.

 Boban Arsenijević

(17) a.

Ako je Marija ispraznila frižider pre odlaska, Jovan kaže if aux Marija emptied fridge before leaving Jovan says da Petru nije ostalo hladnog piva. that Petar.dat not.aux stayed cold.gen beer.gen ‘Jovan says that if Marija emptied the fridge before leaving, Petar was left without cold beer.’

b.

Ko je ispraznio frižider pre odlaska, Jovan kaže da who aux emptied fridge before leaving Jovan says that taj nije mislio glavom. that not.aux thought head.inst ‘Jovan says that whoever emptied the fridge, he didn’t think with his head.’ / *‘Jovan says that the one who emptied the fridge didn’t think with his head.’

The two types of clauses also share other properties related to embedding, such as the impossibility of embedding inside a relative clause. Moreover, just like with conditional clauses, there can be only one correlative clause in a sentence (unless they are conjoined). (18) a. #/*If Mary comes, if John wakes up, they will fight. intended: ‘If Mary comes under the condition that John wakes up, they will fight’. b. #/*Ko voli da kuva, ko voli da jede, taj tog hrani. who likes to cook who likes to eat that.nom that.acc feeds intended:‘The one who likes to cook feeds the one who likes to eat.’/ ‘Whoever likes to cook feeds anyone who likes to eat.’ There are some possible counterarguments to the proposed analysis. The following two have been pointed out to me by Rajesh Bhatt (p.c.). The first one is a universal asymmetry between conditional and correlative constructions. While in correlatives, the correlative wh-expression must have a match in the head clause (see Leung, this volume), no such requirement is observed in proper conditional clauses. Observe the minimal pair in (19): (19) a. */#Ko tebe (ako) nečim udari, ti idi kući. who you.acc if something.inst hits you go home b. Ako tebe ko nečim udari, ti idi kući. if you.acc who something.inst hits you go home ‘If anyone hits you with something, go home.’ Only the sentence in (19a) involves a real correlative construction, because only there is the weak existential expression topical. The sentence in (19b) is simply a conditional clause with a weak existential expression. Without a match for the weak existential whexpression, the correlative in (19a) is bad, while the conditional in (19b) is fine. But the



{Relative {conditional {correlative clauses}}} 

proposed analysis has a possible explanation for this asymmetry. The wh-expression, involving weak existential quantification, has a landing site at WorldP, where it establishes a distributive interpretation with respect to the weak existential quantification in this projection (the relation that licenses the lack of lexical realization of the material in WorldP). The nature of quantification involved does not allow for any interpretive effects of this distribution. However, it seems that the topicality of the wh-expression targets not the referent of the expression, but exactly the distribution between the two weak existentials. In other words, the correlative construction always says something about the distributive relation between a set of worlds (denoted by the underlying conditional clause) and a set of referents (objects, times, places etc. – denoted by the wh-expression). This topicality requires that the head clause, which contains the new information about this distributive relation, also involves reference to both elements involved in distribution: the set of worlds (WorldP of the head clause) and the set of referents (the demonstrative). Facts about the demonstratives discussed in the following section add further support to this view. The second counterargument relates to the fact that in certain languages, the correlative construction shows properties that do not really follow from the presented analysis. In particular, Hindi correlatives do not necessarily involve generic or modal nuances: they may mean exactly the same that regular relative clauses mean in other languages. As discussed in the following section, in relation to example (20), correlative as well as conditional clauses allow for a possible reading involving only the actual world. For this type of readings, correlative clauses indeed have the meaning quite similar to that of regular relatives. They still talk about the distribution between worlds and objects, times, places or other entities, but there is only one world involved in this distribution: the actual one, and only one entity distributed over it. Perhaps due to lower productivity or lack of regular relative clauses, this particular interpretation of correlatives may become more prominent in some languages, such as Hindi. The explanation is then a strategy of language use, rather than a fact about the structure of the observed construction. At the end of Section 5, I briefly discuss one possible scenario which would derive exactly the asymmetry observed between Hindi and languages like Hungarian or Serbo-Croatian.

5. Two kinds of conditionals In this section, I point out some parallels between conditional and correlative clauses based on the available observations and analyses, as well as some differences, which I try to explain in terms of the special properties of correlative clauses. It has been claimed for Hindi correlative that their head clauses are always IPs, and never CPs (Dayal 1996 and Bhatt 2003). However, Izvorski (1996) and Citko (this volume), based on Slavic data, as well as den Dikken (this volume) based on Dutch, show that head clauses of correlative sentences can be IPs or CPs. In parallel, Haegeman

 Boban Arsenijević

(2003) convincingly argues that there are two types of conditionals, event- and premiseconditionals, which differ in the structural domain of the matrix clause that they modify, and which can be IP or CP. The parallel speaks for itself, but its deeper semantic effects make it even stronger. Semantic effects of the base generation positions that Haegeman discusses for conditional clauses are mostly related to the epistemic status of the conditional clause. In the higher position, the conditional clause is interpreted with an ‘echoic’ component, as something that has the status of an interlocutor’s contribution to the discourse, rather than a neutral objective possibility in the world. In the lower position, the conditional clause is more integrated with the eventuality denoted by the VP of the head clause. In terms of the present analysis, the epistemically active base generation position would be WorldP, the default target of epistemic modification. The availability of a lower base generation position requires a modification of the analysis. The immediate question in the accommodation of the present analysis to Haegeman’s observations is what projection in the (extended-)IP domain can be modified by a conditional clause. Going back to the initial stage of the proposal, the idea that conditional clauses are yes-no relative clauses, a natural candidate emerges: NegP. Conditionals are then yes-no relatives at the level of NegP, or at the level of WorldP. In the former case, they restrict the truth conditions of an eventuality, and in the latter, the worlds of the proposition. This is illustrated in (20). (20) a. If John emptied the fridge yesterday before he left, there is no cold beer in the house. (i) In the actual world of the present, the NEG-availability of the beer is restricted by the emptying of the fridge yesterday. (ii) Possible worlds in which there is no cold beer in the house are restricted to those in which John emptied the fridge before leaving. b. If, as you say, John emptied the fridge, there is no cold beer in the house. Possible worlds in which there is no cold beer in the house are restricted to be compatible with the collocutor’s contribution to the discourse. Important asymmetry between the two examples is that (20a) has an actual world reading, with definite and actual speech-, reference- and event-time, while (20b) has no choice but to receive a possible worlds interpretation, because of the epistemic modification by as you say. The reading that distinguishes (20a) from (20b) corresponds to the underlying structure in which the conditional clause is generated in NegP. The type of reading involving possible worlds has the conditional clause generated in WorldP. Bittner (2001) shows that the same ambiguity, which she attributes to the centering of worlds versus the centering of individuals, is systematically present in correlatives. Observe the correlative sentence in (21).



{Relative {conditional {correlative clauses}}} 

(21)

Koji student je prvi ušao, taj je položio ispit. which student aux first entered, that aux passed exam ‘The student who entered first passed the exam.’ (the actual world) ‘Any student who entered first passed the exam.’ (possible worlds, in most contexts reduced to only one FC student, due to the pragmatics of entering first)

It has one reading that refers to the actual world, and states that a definite and topical eventuality of passing a definite and topical exam involves as its theme a particular definite and topical (or possibly focalized) student who entered first. The other reading involves possible worlds, and states that for any student that passed the exam, the possible world in which this happened is characterized by the same student that entered first. The latter reading has a strong epistemic component, and may even be taken as a generic statement: entering first implies passing the exam. The structural explanation presented for conditional clauses applies here as well. When the correlative clause modifies NegP, one subcase is that, in the CP domain, there is WorldP that specifies the sentence as referring to the actual world. When the correlative clause modifies WorldP, it always entails epistemic modification, and hence also possible worlds. It is not excluded, however, that some languages parametrically ban one of the two structural patterns, and hence also that the Hindi correlatives are indeed, as Dayal (1996) and Bhatt (2003) argue, only of the IP level, while in languages like Hungarian or Serbo-Croatian they may also appear within the CP layer. The ban of the attachment in the CP domain would result in the exclusive availability of the actual world reading (unless specified otherwise by an independent adverbial or other element, but this is orthogonal to the issue discussed). This means that there is no distribution between worlds and the referential domain of the wh-expression, and that the resulting meaning of the correlative sentence is very close to the meaning of sentences involving relative clauses as modifiers, licensing the use of correlatives in the fashion in which relative clauses are used in other languages. Also, it may be the case that the (un)availability of the CP level conditionals and correlatives provides an explanation for the (un)availability of symmetric readings in Hindi correlatives (cf. Cheng & Huang 1996). Symmetric readings might be explained in terms of assigning the same discourse role to all the symmetric elements. Unavailability of the CP in the correlative sentence of a language, and the clear asymmetry in the correlative clause, where topicality is the crucial issue, reduces the possibility of deriving symmetric readings.6 This brings us to one important difference between regular conditionals and correlatives. Conditional clauses appear to involve maximization over possible worlds, while correlative clauses maximize primarily over the domain of the extreme nonspecific wh-expression. This property makes correlatives more similar to free and regular restrictive relatives, than to conditionals. However, this property is easily derivable from the fact that the relevant wh-expression is topical, and has an intermediate landing site in WorldP. In this way, it establishes a relation with the quantification

 Boban Arsenijević

in WorldP; the result is that possible worlds are bound by, i.e. distributed over, members of the referential domain of the extreme non-specific wh-expression.

6. The demonstrative The present analysis takes wh-expressions in correlative clauses to be independently generated extreme non-specific expressions, rather than relativizers rigidly related to certain antecedents. These expressions move to the left periphery of the correlative clause because they are topical, and not because they need to establish some relation with the contents of the head clause. The coreference between the demonstrative in the head clause and the wh-expression in the correlative clause is just discourse-reference. This view is similar to that of Bittner (2001), and goes against the analysis put forth in Bhatt (2003), in which the correlative is base generated as a relative clause adjoined to the demonstrative. Bhatt (2003)’s crucial argument for analyzing correlative clauses as adjuncts to the correlated demonstrative expressions is their sensitivity to island effects. S-C correlatives, just like the ones in Hindi, show these effects, as illustrated in (22). (22) */#Koji student je prvi ušao, which student aux first entered, čitam rad koji je taj student pisao. read.1sg paper which aux that student written intended: ‘I’m reading a paper of the student that entered first.’ First, note that the same effect appears in proper conditional clauses with extreme non-specific arguments. (23)

??Ako ko uđe, pročitaću rad koji je taj pisao. if who enters read.fut.1s paper which aux that written ‘If someone/?anyone enters, I’ll read the paper that he wrote.’/ *‘Whoever enters, I’ll read the paper that he wrote.’

Although (23) is not completely out, it is certainly not as good as (10), repeated in (24), and it loses the semantic parallel with correlatives, illustrated by the two readings available for (24). (24)

Ako je (ko) već (ko) ustao (*ko), if aux who already who raised who (taj) (onda) neka (taj) (i) izađe (*taj). that then let that and go.out that ‘If anyone already stood up, (then) let him also go out.’/ ‘Whoever stood up, let him also go out.’



{Relative {conditional {correlative clauses}}} 

This indicates that anaphoric reference per se does not suffice for a proper analysis of correlatives and their conditional counterparts, since anaphoric reference is available in (23), but the relevant semantics does not obtain. But at the same time, it signals that the reason for the degradation of (22) is not that the correlative clause is generated inside an island and then moved in an illegal way. The movement rather involves some other element, most likely the demonstrative, and the correlative clause is generated in the same way as the conditional clause in (23), as argued in this paper. It has been suggested that the generalization about the demonstrative is in fact that it needs to be at the left edge of the head clause, and in some languages also adjacent to the correlative clause (den Dikken this volume, Section 2). Combined with the facts in (23) and (24), including the loss of the correlative semantics triggered by the drop of the demonstrative, this indicates that the island effects may in fact only relate to the extraction of the demonstrative into the relevant position where its discourse role is interpreted, or for the availability of the relevant discourse role in the problematic positions. This implies that discourse functions, and information structure in general, play an important role in correlative sentences. This has already been subject to detailed analyses in the literature (among others, Lipták 2005 and Bittner 2001). Impossibility to realize the adequate information structure leads to impossibility to realize the correlative construction. Indeed, my personal intuition about the S-C example in (22), which was confirmed by several other speakers, is that it is not really ungrammatical, but pragmatically infelicitous. The speaker’s instinctive reaction is to correct it by extracting the demonstrative out of the relative clause, but this results in a clear ungrammaticality, as in (25). (25) *Koji student je prvi ušao, taj student čitam which student aux first entered that student read.1s rad koji je pisao. paper which aux written intended: ‘I’m reading a paper of the student that entered first.’ An argument against analyzing correlatives as relative clauses is that, in some languages, the wh-expression in the correlative clause can be preceded by different topical material (see Rebuschi this volume, and Section 4 of the present paper). This is not a property shown by relativizers in these languages: they have to appear clause-initially. In addition to this, in favor of anaphoric reference rather than relativization, observe that when the demonstrative is not strongly focalized, the structure with the correlative to the right of the head clause is out in S-C. Observe the example in (26). (26) Taj (student) je položio ispit, koji (*student) je ušao. that (student) aux passed exam which student aux entered ‘The student who entered passed the exam.’ / *‘Whichever student entered, he passed the exam.’

 Boban Arsenijević

With a neutral intonation, with no narrow focus, this sentence is fine, or just slightly degraded, but only if read as a regular relative. It does not at all have the correlative structure available, which is clearly indicated semantically by the absence of the free choice and generic readings and syntactically by the ban on the overt realization of the head noun in the subordinate clause. The deletion of the head noun is possible, but it has to be licensed by the context (topicality of a definite group of students or of the generic meaning of the noun). Under a particular information structure, namely contrastive focus on both ušao ‘entered’ and taj ‘that’, the correlative reading emerges again. However, this is expected under the present analysis, if the head clause is pied-piped to the focus position, over the previously fronted correlative. Note that the same effect as above is achieved if the conditional clause containing an extreme non-specific wh-expression appears to the right of the head clause, as in (27). Since there is no regular relative reanalysis for the proper conditional clause, the result is just uncontroversial ungrammaticality. Lack of either topic or focus on the wh-expression in the conditional clause blocks even the information structure with contrastivity that saved the correlative counterpart in (26). (27) *Taj je položio ispit, ako je koji student ušao. that aux passed exam if aux which student entered intended: ‘The student who entered passed the exam.’ Finally, let me present another important property of the demonstrative in the correlative construction. S-C has a tripartite paradigm of demonstratives, with three different values for the distance feature. These can be formalized in the following way. One member of the paradigm is characterized by the morpheme -v-, and has the value [+proximal, –distal], relating to the immediate domain of the speaker, as well as to the most recently mentioned referents in the discourse. The member characterized by the morpheme -t-, has the value [–proximal, –distal], and is related to the most prominent discourse domain and its referents, as well as to the contents introduced by, or related to, the collocutors. It is also the neutral member of the paradigm: in situations where proximity is not specified, or different dimensions that determine it clash with each other, this form is used as an underspecified solution (this is why it is specified with both negative values). Finally, the member marked by the morpheme ‑n- has the value [–proximal, +distal] and relates to the members and domains of the discourse with a low level of discourse prominence, outside the immediate domains of all interlocutors, or related to an interlocutor which is absent from the communication situation (who is an interlocutor e.g. through indirect speech).7 The paradigm extends to temporal, spatial, manner, and other demonstratives. Notice that one of the three demonstrative forms, the proximal -v- person, is systematically banned from the correlative construction (28).



{Relative {conditional {correlative clauses}}} 

(28) Ko je ustao, onaj/taj/*ovaj neka izađe. who aux raised that-n/this/that-v let go.out ‘Let those who stood up go out.’/’If anyone stood up, let them go out.’ This looks like a weird regularity. Why would any demonstrative be excluded from the correlative sentence, especially because the antecedent is very recently mentioned, it is prominent (see Bittner 2001 for a formal representation of its centered nature), and quite possibly introduced into the discourse by the speaker, all exactly properties of the excluded demonstrative? A natural move at this point would be to see if the same holds for conditional clauses. (29) Ako pada kiša, onda/tada/*sada Jovan ne dolazi. if fall rain then-n/then-t/now Jovan not comes ‘If it rains, then Jovan won’t come.’ Indeed, the proximal demonstrative is out in conditionals too. Moreover, conditionals offer an explanation for this fact. The proximal demonstrative in the temporal domain, unlike any other proximal demonstrative, including the spatial one, has a deictic meaning. It does not have the discourse-based, relative, meaning of ‘this time’, but really has the deictic denotation of now. This makes it a bad candidate to refer to worlds, and in fact makes it a non-demonstrative. We can now go further in uncovering the nature of the demonstratives in correlative sentences. At first glance, it still seems unclear why this special property of temporal demonstratives in conditional sentences would extend to other demonstratives, even if correlatives are, as argued here, a subtype of conditionals. Going through the paradigm in (28) again, it appears that there is a slight semantic difference between the two versions of the correlative sentence, with the -n- and with the -t- demonstrative. The difference is that in the latter variant, the possible worlds introduced in the correlative clause are accessible to the speaker, which is not the case in the former. In the one case, with the -t- demonstrative, the speaker has means to determine the actual referent of the demonstrative for any situation that appears to be not only possible, but also part of the actual world (as it is the case in the actual world reading with only one maximized referent for the demonstrative, which is usually present in correlatives). In the other case, with the -n- demonstrative, the speaker (similarly perhaps to the interlocutors) would not be able to identify the referent of the demonstrative even if there is exactly one, specific, candidate. This is proven by the fact that with the -n- demonstrative, (28) cannot have the actual world reading, in which the speaker sees the student who stood up and wants him to leave. As pointed out many times in this paper, this type of reading is normally available in correlatives, and, as expected, (28) with the -t- demonstrative also has it. Since the correlative clause provides the referent with a maximizing description, it is by default specific within its domain, and its maximizing description is directly available to the speaker. This means that in the default case, the referent of the

 Boban Arsenijević

demonstrative should be identifiable by the speaker, and in the default case the -t-demonstrative should be the one that is used. Indeed, this is the case in S-C: correlatives with the -n- demonstrative are clearly marked, and in the vast majority of cases, the -t- demonstrative is the one that is used. The only way to make the referent of a demonstrative, which is provided with a maximizing description, unavailable to the speaker is to embed it in a world, or situation, that is unavailable to the speaker. This means that the proximity feature of the demonstrative is in fact related to the status of the context (world, situation) to which its referent belongs, and not directly to the status of the referent itself. This means that we have to look at the conditional demonstratives again, and see if their behavior can tell us something interesting. While the -t- demonstrative is the default choice for different kinds of referents in the S-C correlative clause, the conditional demonstrative has the -n- form as its default value. The sentence in (30) would normally be used with onda, and in order to have tada, one needs a special information structural status for the condition: it has to be part of the ground information, already talked about, and assigned some properties. (30) Ako pada kiša, onda/tada/*sada Jovan ne dolazi. if fall rain then-n/then-t/now Jovan not comes ‘If it rains, then Jovan won’t come.’ Explanation of this fact requires a brief excursion into the semantics of S-C temporal demonstratives. As a neutral temporal demonstrative, oriented towards the most prominent discourse-referent of the relevant, temporal type, tada always picks up exactly the temporal interval that is most recently reached in the relevant discourse (see ter Meulen 1995 for an elaborated picture of the temporal dynamics of discourse). It is normally translated to English as ‘at that time’. For this reason, in narration, for the very common goal of sequencing temporal intervals, i.e. of switching from one interval to the next, possibly adjacent, relevant one, the demonstrative that is used is onda. (31) a. Jovan je zaplivao. Tada se (i) Marija probudila. Jovan aux swimming then-t refl and Marija woke.up ‘Jovan started swimming. At the same time, Marija woke up’. b. Jovan je zaplivao. Onda se (i) Marija probudila. Jovan aux swimming. then-n refl and Marija woke.up ‘Jovan started swimming. Then, i.e. (right) after that, Marija woke up’. In conditional sentences, the temporal demonstrative refers to the relevant, epistemically modified, worlds. In other words, it points to the discourse representation right after (the possible case of) an update by the contents of the conditional clause. This corresponds to the ‘right after that’, rather than the ‘at the same time’ meaning in (31). Its meaning involves the component of consequence: one predicate being the case, we can move forward to another predicate. This move forward is a move into the domain of worlds of weak existential commitment.



{Relative {conditional {correlative clauses}}} 

This defines two possible cases for the interpretation of the extreme non-specific argument of the correlative clause. This argument is also characterized by weak existential commitment, but it differs from the conditional clause in not being involved in the relative clause structure, in which the conditional demonstrative takes place as the head, and the conditional clause as the relative clause. In this sense, the correlative demonstrative has an independent interpretation, and can be interpreted directly in the discourse domain that the speaker has access to. Still, it can also be interpreted as member of the set of worlds denoted by the conditional clause, and hence have its referent inaccessible to the speaker. The latter case is marked by the -n- form of the demonstrative, marking the distal nature of the worlds where the referent belongs to, i.e. borrowing it from the conditional demonstrative to mark the dependency it establishes with respect to the conditional clause (recall that the conditional demonstrative can be present in the head clause of a correlative sentence, just like a correlative demonstrative). The distinct behavior of the two demonstratives crucially comes from their different syntactic relation to the relevant segment of the correlative/conditional clause. The conditional demonstrative, as discussed in Section 2, is a lexicalization of the head modified by the conditional clause, the relation it establishes with the conditional clause is that between a head and a relative clause. The correlative demonstrative, however, presents an independent nominal expression, anaphorically referring to the referent, or set of referents, introduced by the extreme non-specific wh-expression in the correlative clause. The inaccessibility of the correlated referents marked by -n- demonstratives to the speaker, and its link with the conditional demonstrative, would be hard to explain under the analysis in which the correlative clause is generated on the demonstrative. Outside correlative clauses, -n- demonstratives do not differ from other demonstratives in the accessibility of their referents.

7. Conclusion In this paper I proposed an analysis of conditional clauses in terms of yes-no relativization. I argued that the relation between conditional clauses and yes-no questions is the same as that between wh-relative clauses and wh-questions. The conditional clause, i.e. the yes-no relative, introduces a restriction over the worlds compatible with the relevant proposition in the head clause. I showed that this view makes correct predictions and accounts for facts that have long been subject to debate. Building on this proposal, correlative clauses were analyzed as conditionals with topical extreme non-specific arguments. The aggregate empirical coverage of the analysis is quite broad, including some novel observations made on Serbo-Croatian correlatives. The analysis provides explanation for a number of striking similarities between conditional and correlative clauses, starting from semantics to syntax and to

 Boban Arsenijević

morphology. It also reduces the inventory of clause types by two: correlative clauses are presented as a subtype of conditional clauses, and conditional clauses are analyzed as relative clauses. In the correlative clause, the correlated element, normally lexicalized as a whphrase, is analyzed as a topical extreme non-specific expression. Its topicalization derives several important aspects of the grammar of correlative sentences, such as the distribution of worlds over the denotation of the wh-phrase, or the (optional) drop of the conditional marker. In the main clause, I analyzed the correlative demonstrative as an anaphoric expression, rather than the head of a relative clause. I suggested that the seeming island effects are a matter of the (un)availability of the relevant discourse function in the relevant position, rather than a real piece of evidence that correlative clauses are generated locally to the demonstrative and then moved to the sentential left periphery. I showed that with respect to the position of base generation, correlatives parallel conditional clauses, as predicted by the analysis. Based on novel data from Serbo-Croatian correlatives, I showed that the correlative demonstrative indeed behaves different from the conditional demonstrative. Being the head of a relative, i.e. conditional, clause, the conditional demonstrative is restricted in its meaning and form by the semantics contributed by the conditional clause. As an anaphoric expression, the correlative demonstrative is less dependent in its semantics and the form it takes. However, its meaning may be directly dependent on the meaning of the conditional clause and its demonstrative, in which case a morphological dependency is observed as well. The proposed analysis also accounts for the similarities between correlative clauses and conditionals, by having the former as a subtype of the latter, between correlatives and free relatives, in involving an independent extreme non-specific wh-expression, and between correlatives and headed relative clauses, through the relative nature of the conditional structure, which underlies the structure of correlative clauses.8

Acknowledgements I am indebted to Wolfram Hinzen and to the editor of this volume for fruitful discussions, and for inspiring and encouraging me to write this article. I also thank Rajesh Bhatt, Lisa Cheng, František Kratochvil, Anikó Lipták, Joanna Sio and Leo Wong for helping me find my way through the data. The present work is a result of my engagement in the project “The origins of truth and the origins of the sentence”, funded by NWO (Netherlands Organization for Scientific Research).



{Relative {conditional {correlative clauses}}} 

References Arsenijević, Boban. 2009. Clausal complementation as relativization. Lingua, 119.1. 39–50. Bittner, Maria 2001. Topical referents for individuals and possibilities. In Rachel Hastings, Brendan Jackson and Zsófia Zvolenszky (eds), Proceedings of SALT X, 36–55. Ithaca: CLC. Bhatt, Rajesh. 2003. Locality in Correlatives. Natural Language and Linguistic Theory 21. 485–541. Bhatt, Rajesh & Roumyana Pancheva. 2006. Conditionals. In Martin Everaert, Henk van Riemsdijk, Rob Goedemans & Bart Hollebrandse (eds), The Blackwell Companion to Syntax, v. I, 638–687. Malden, MA: Blackwell. Cable, Seth. This volume. The syntax of the Tibetan correlative. Cheng, Lisa L-S. & James C-T. Huang. 1996. Two types of donkey sentences. Natural Language Semantics 4. 121–163. Chierchia, Genaro. 2006. Broaden your views: Implicatures of domain widening and the ‘logicality’ of language. Linguistic Inquiry 37. 535–590. Citko, Barbara. This volume. What don’t wh-questions, free relatives and correlatives have in common? Dayal, Veneeta. 1996. Locality in Wh-Quantification: Questions and Relative Clauses in Hindi. Dordrecht: Kluwer. den Dikken, Marcel. This volume. Comparative correlatives and successive cyclicity. Farkas, Donka. 2002. Extreme non-specificity in Romanian. In Claire Beyssade, Reineke BokBennema, Frank Drijkoningen & Paola Monachesi (eds), Romance Languages and Linguistic Theory 2000, 127–153. Amsterdam/Philadelphia: John Benjamins. Haegeman, Liliane. 2003. Conditional clauses: external and internal syntax. Mind and Language. Numéro spécial 18–4. 317–339. Higginbotham, James. 2003. Conditionals and compositionality. In John Hawthorne (ed.), Philosophical Perspectives 17. 181–194. Izvorski, Roumyana. 1996. The syntax and semantics of correlative proforms. In Kiyomi Kusumoto (ed.), Proceedings of NELS 26, 133–147. GLSA Publications, Amherst. Jayaseelan, Karattuparambil A. & M. Hariprasad. 2001. Deixis in Pronouns and Noun Phrases. Linguistic Analysis 31. 132–149. Kratzer, Angelika. 1986. Conditionals. In Anne M. Farley, Peter T. Farley, and Karl-Erik McCullough (eds), Chicago Linguistics Society 22(2), 1–15. Leung, Tommi Tsz-Cheung. This volume. On the matching requirement in correlatives. Lewis, David. 1975. Adverbs of Quantification. In Edward L. Keenan (ed.), Formal Semantics of Natural Language, 3–15. Cambridge: Cambridge University Press. Lipták, Anikó. 2005. Correlative Topicalization. Ms. University of Leiden. ter Meulen, Alice G. B. 1995. Representing time in natural language. The dynamic interpretation of tense and aspect. Cambridge: MIT Press, Bradford Books. Rebuschi, Georges. This volume. Basque correlatives and their kin in the history of Northern Basque.

 Boban Arsenijević

Endnotes 1. The paper uses the following abbreviations: and: adnominal marker, aux: auxiliary, cl: clitic, imp: imperative, inf: infinitive, interr: interrogative, fut: future, refl: reflexive, nom/ acc/dat/gen/inst: nominative/accusative/dative/genitive/instrumental case; S/P: singular/plural. Person and number is only indicated on verb forms that are first (1) or second (2) person. 2. See also Arsenijević (2009), where complement clauses are analyzed as relative clauses with the relativization site in the projection of force, which leads to generalizing over relative clauses as the only type of embedded clauses. 3. Not all languages use the temporal demonstrative for this purpose. In Cantonese, for instance, the element ‘zau’ is used (Joanna Sio p.c.), meaning something like almost. It is still an element from the temporal domain. See example (15) below. 4. It is possible, as suggested in Section 5 in relation to examples in (20), that the sentence in (6) is ambiguous between the two described meanings, but still, the one in (5) can have only the one that Higginbotham ascribes to it. 5. On the special information structural nature of the fronted wh-expression in correlatives, see Bittner (2001), who describes it in terms of the centering theory. 6. It is possible that in some languages, like Hindi, correlative clauses involve further structural restrictions, which make them even more different from those in Hungarian or Slavic languages. 7. Morpheme -t- also derives second person pronouns, and morpheme -n- derives third person pronouns. 8. An analysis of free relatives where they present an independent extreme non-specific whexpression which takes a relative clause, and reduces the head and the relativizer by haplology, accounts for the requirement of S-C free relatives as in (i-iii) that the head clause and the relative clause assign the same case to the wh-expression. Observe the data below. (i) Ako ko [ko god je zakasnio] ima pitanje… if who who ever aux been.late has question ‘If anyone, who was late, has a question…’ (ii) Pozvao sam koga [koga god si ti pozvao]. invited aux.1s who.acc who.acc ever aux you invited ‘I invited whoever you invited.’ (iii) Pozvao sam koga [*(??ko) god je tebe pozvao]. invited aux.1s who.acc who.nom ever aux you.acc invited intended: ‘I invited anyone who invited you.’

Relatively different Italian Sign Language relative clauses in a typological perspective Chiara Branchini and Caterina Donati Italian Sign Language displays a dedicated structure expressing relativization: a biclausal construction consisting of an embedded clause containing an antecedent, followed by a main clause containing a gap or a pronoun coreferential with the antecedent. This paper compares two possible analyses for this type of construction: as a correlative structure, as recently proposed by Cecchetto et al. 2006, or as a non-correlative internally headed relative clause. Evidence for the nominal status of the clause, for its extraposed nature, and for the trace-like nature of the gap in the main clause is provided and discussed. This leads to the conclusion that the non-correlative analysis is more suitable. As for their interpretative status, we argue that these clauses are restrictive relative clauses in the light of a battery of diagnostics. Keywords: Italian Sign Language, relative, correlative, internally headed relative, restrictive, head-movement

1. Introduction Italian Sign Language (LIS) is the language of the Italian Deaf community, consisting of about 60.000 speakers spread all over the Italian peninsula, and, in most cases, invisible and marginalized by the Italian (hearing) society.1 Like most other sign languages, LIS is very poorly known to core linguistic research, its original and intricate properties being either ignored or relegated to the field of “applied linguistics”. This paper aims to show that a sign language such as LIS can and must provide important data for our understanding of core syntactic categories and operations. Starting from the seminal work of Cecchetto, Geraci and Zucchi 2006, we provide a new analysis of the peculiar relative construction observable in LIS. In doing so, we reflect on the interpretive differences between internally and externally headed relative clauses and call for a reconsideration of some of the boundaries standardly posited between relativization strategies in typological studies.

 Chiara Branchini and Caterina Donati

After clarifying the methodology adopted for data collection (Section 2), we briefly present some basic properties of LIS syntax (Section 3). Section 4 presents the construction to which the paper is devoted, the structure translating English relative clauses, which we label ‘PE-clauses’. Then two distinctive features of the PE-clause are discussed. These are, namely: (i) that a PE-clause is always located to the left of the main clause and (ii) that a PE-clause always has an antecedent internal to it. These characteristics are important in the discussion of two competing analyses (§ 5): what we call the correlative analysis (as specifically proposed by Cecchetto et al. 2006: § 4.1), and what we shall label the ‘extraposed internally headed analysis’. Articulated evidence pointing toward the latter is then provided (§ 5.2, § 5.3), and a new implementation of the analysis is discussed and tentatively extended to similar constructions in other languages (§ 6). Lastly, Section 7 discusses whether PE-clauses are to be analyzed as restrictive or as appositive in nature. After questioning the validity of some apparent evidence supporting their appositive status, we conclude in the light of a battery of standard diagnostics that PE-clauses are restrictive.

2. Research methodology Our data come from five native deaf signers of LIS, that is, from Deaf people who have been exposed to Italian Sign Language from birth. L1 speakers represent a minority in the community of deaf signers where about 95% are born from hearing parents. The informants come from different regions of Italy (two pairs of siblings from Ancona and a signer from Rome) and they all share a native linguistic competence, a strong Deaf cultural identity and a good knowledge of Italian as a second language. All of them are active members of the local Deaf community. While the siblings from Ancona had never been interviewed before in the framework of a linguistic research, the signer from Rome is a LIS teacher and had participated in other linguistic inquiries in the past (coordinated by Virginia Volterra’s research group in Rome). Despite some lexical differences reducible to dialectal variation and their initial disparity in meta-linguistic awareness, our informants provided us with a homogenous collection of LIS data, all using the same construction when asked to produce the equivalent of an Italian relative clause. The data were elicited through the aid of a LIS interpreter who presented the signers with a relevant context to which she referred with some questions. At a very early stage, when we needed to elicit highly specific constructions, such as relative clauses headed by quantifiers or numerals, the signers were presented with written Italian stimulus sentences and were asked to produce the most natural LIS translation given a particular contextual background for their interpretation. The procedure was videotaped and analyzed through SignStream, a software package developed at Boston University to create and analyze sign language databases.2 The data provided in this paper follow the standard convention of using capitalized words



Italian Sign Language relative clauses 

for signs. Coindexing signals coreference. The line over the glosses stands for the nonmanual component realized simultaneously to the sign(s) it dominates.3

3. Some basic syntactic properties of LIS In this section we will present a brief description of some basic syntactic properties of LIS. This is meant to provide the reader with the necessary background for evaluating the analysis of LIS relativization strategies. We will briefly show how space and directionality are used for linguistic purposes and how the simultaneous realization of the manual and non-manual markings (NMMs, i.e. face expressions, body lean, head tilt and movement of the shoulders) enter into the building of syntactic structures. The basic word order of LIS is SOV, as illustrated in (1). (1) dog cat chase ‘The dog chases the cat’ Together with many derived orders due to extrapositions, SVO is also attested, but seems to be employed mostly when signers interact with non-natives who are influenced by the dominant SVO order of Italian.4 All functional categories within the clausal domain are located postverbally. The aspectual marker DONE, which marks the verbal action as completed, occurs after the verb, as in (2). (2) dog cat chase done ‘The dog has chased the cat’ Modals are also located postverbally, cf. (3). (3) dog cat chase can ‘The dog can chase the cat’ Negation is also found after the verb, as shown in (4) ________neg (4) dog cat chase not ‘The dog does not chase the cat’ The same holds for (manual) manner adverbs5, see (5). (5) anna leave on-time ‘Anna leaves on time’ Finally, (6) shows that the wh-phrase is also at the right periphery of the sentence.

 Chiara Branchini and Caterina Donati

(6) cat chase who ‘Who chases the cat?’6 We can now attempt to combine these various postverbal elements. Manner adverbs, when present, need to precede the aspectual marker done, negation, and modal verbs, cf. see (7). (7)

anna leave on-time done/not/must a. ‘Anna has left on time’ b. ‘Anna did not leave on time’ c. ‘Anna must leave on time’

Negation and done (aspect) are in complementary distribution perhaps due to some semantic incompatibility (Zucchi 2003); the same holds for modals and done. Modals precede negation (8). _______neg (8) anna leave must not ‘Anna must not leave’ The wh-phrase is always the rightmost element of the clause. _______neg (9) cat chase must not who ‘Who must not chase the cat?’ This distribution suggests that the functional projections hosting all these elements are located to the right of the verb phrase. Turning now to the left field, time adverbials are always found in sentence-initial position as in (10). (10) yesterday dog cat chase done ‘Yesterday the dog has chased the cat’ Putting all this information together, we claim, following Cecchetto et al. 2006, that LIS is a head-final language. We assume that the subject originates in Spec,VP, and raises to Spec,IP, while the object remains in its base position within the VP. Manner adverbs are taken to be right adjoined to VP and time adverbials are taken to be left adjoined to IP. Negation occupies a position between CP and IP (see Geraci 2006 for a detailed analysis). As for the wh-phrase, the analysis proposed by Neidle et al. 2000 for American Sign Language (ASL) and extended to LIS by Cecchetto and Zucchi 2004, locates it within a right-branching Spec,CP. This issue is controversial, at least for ASL, see Wilbur and Patschke 1999 and Petronio and Lillo Martin 1997. We will not take any position on this, as it is not relevant for our analysis. The tree diagram in (11) adapted from Cecchetto et al. 2006 illustrates all these basic claims.



Italian Sign Language relative clauses 

(11)

CP 2 C' WH 2 NegP C° 2 (Spec) Neg' 2 IP Neg° AdvP 2 Time IP 2 Subject I' 2 AspP I/Mod 2 (Spec) Asp' 2 VP Asp 2 done VP AdvP 2 Manner object V

It is also important to understand agreement patterns in LIS. LIS allows both subject and object drop, with the verb agreeing in space with the position(s) associated with these non-overt argument(s). (12) proi igivej proj ‘He/she gives him/her’ Agreement in space is a general mechanism in LIS, as in other sign languages. In (12) the movement of the sign for give is modified so as to start in the position in space associated with the subject and to end in the position associated with the object. Agreement in space is thus an agreement marker. For sake of simplicity, we shall not spell out verbal agreement in our glosses. Another important function of agreement in space is illustrated in (13), where the sign for his (POSSi) is articulated in the position associated with its antecedent friend. (13) friendi poss sons possi all leave done ‘My friend’s sons all left’ Agreement here signals coreference.7 While this is always the case with signs produced in the neutral space, this is not necessarily so with body-anchored signs. The glossing

 Chiara Branchini and Caterina Donati

we adopt here does not make any distinction between neutral and body-anchored signs. We simply use coindexing to mark the linguistic phenomenon of coreferentiality, independent of its realization. Finally, LIS, like all sign languages, exploits non-manual means to mark a number of syntactic features, such as negation, question, condition, topic, focus etc. To illustrate, the following LIS examples are minimal pairs where the negative (14a), interrogative (14b), conditional (14c) status of the clause is given by different non-manual markers. ____neg (14) a. dog cat chase ‘The dog does not chase the cat’ _____________y/n b. dog cat chase ‘Does the dog chase the cat?’ ____________cond c. dog cat chase cat scared ‘If the dog chases the cat, the cat gets scared’. Expressed crudely, ‘negation’ corresponds to furrowed eyebrows and a left-to-right shaking of the head; ‘yes/no’ corresponds to raised eyebrows, wide eyes and a forward leaning of the body; ‘conditional’ corresponds to raised eyebrows, a general tension of eyes and cheeks and forward head tilt.8 According to a number of studies (Neidle et al. 2000, Neidle 2002, Petronio and Lillo-Martin 1997, Pfau 2004, for ASL and other sign languages, Cecchetto and Zucchi 2004 for LIS), the extent of the spreading of the NMM is an indicator of syntactic dependency. To illustrate, in the example (6) repeated here as (15), the wh-element might, in principle, be interpreted both in object and in subject position. The spreading extent of the wh-NMM disambiguates it: in (15a), the non-manual spreading signals that the wh-element is interpreted in object position, while in (15b), spreading signals that the wh-element is interpreted in subject position. (15) cat chase who _________wh a. cat chase who ‘Who does the cat chase?’ ______________wh b. cat chase who ‘Who chases the cat?’ As is clear from this overview, much of LIS syntax still awaits a proper description. This is definitely the case for the nominal domain. LIS does not have any (manual) sign for definite or indefinite articles, but other determiners, such as numerals, demonstratives and quantifiers are attested, and seem to occur both pre-nominally and



Italian Sign Language relative clauses 

post-nominally. For example, in a sentence like (16), any ordering of N and D seems to be acceptable. However, there is a preference for the determiner to follow the noun. We are not able to provide an adequate account of these ordering possibilities but, given the systematic complement-head order illustrated in (11), we will provisionally assume N-D to be the basic order. (16) dogi ixi catj ixj chase ‘This dog chases that cat’ This assumption will play a role in the analysis we will propose for LIS relative clauses.

4. LIS PE-clauses Our data attest to the presence of a specialized syntactic structure used by native signers as a strategy for relativization. This is a bi-clausal construction including a sentence-initial clause (which we will call ‘PE-clause’ due to the presence of a specific sign we label ‘PE’) and a sentence-final clause. The PE-clause, which cannot be produced on its own, is overtly identified by two superficial characteristics: PE and rel. Each of these will be discussed in turn. i. The PE9 marker co-occurs with the silent articulation of a labial stop. PE is realized manually with the index finger stretched out and shaken downwards. It occurs systematically at the right edge of the clause (following modals, aspect and negation, cf. Cecchetto et al.  2006). PE is coreferential with an NP within the clause, and this coreferentiality can be realized through agreement in Space. ii. The non-manual marker glossed as ‘rel’ consists of raised eyebrows and tension of eyes and upper cheeks. Its realization and spreading reaches its maximal intensity over the sign PE, although its expression may vary. This will be discussed presently. The second clause, which we might call the ‘main’ clause since it can be uttered in isolation, contains what we might descriptively call the ‘correlate’: a gap or an indexical NP (glossed IX) interpreted as coreferential with the same NP to which PE corefers. Following traditional analyses, we shall call this pivotal NP, occurring within the sentence-initial clause, the ‘antecedent’. To illustrate, the examples in (17) and (18) provide a minimal pair, where the presence or the absence of the properties (i) and (ii) listed above correlates with a sharp difference in interpretation: (17) is a coordination of two independent sentences, while (18) is a relative construction. (17) dogi cat chase (ixi) home come done ‘The dog chased the cat and came home’

 Chiara Branchini and Caterina Donati

_________________rel (18) dogi cat chase pei (ixi) home come done ‘The dog that chased the cat came home’ Cecchetto et al. 2006 show that the antecedent cannot be fully reduplicated in the main clause: no full NP can be realized in the position of the correlate, as the agrammaticality of (19) shows. __________________rel (19) *dogi cat chase pei dogi home come done ‘The dog that chased the cat came home’ As for the position of PE, we have assumed so far that it is the rightmost element within the PE-clause, but we shall demonstrate that it does not actually belong to the main clause. In (20) the time adverbial yesterday refers to the matrix clause and follows PE. Subj-Subj _______________________rel (20) today mani pie bring pei yesterday (ixi) dance ‘The man that brought the pie today danced yesterday’ In (20) the correlate is the subject of the main clause, and the antecedent NP (MAN) is understood as the subject of the PE-clause. In (21) the correlate is the object of the main clause and the antecedent (DOG) is the subject of the PE-clause. Since time adverbials always mark the left edges of their clauses, (20) is clear evidence that PE does not belong to the main clause. The clause boundary following PE can also be signaled through prosodic means: a pause, an eye blink, the abrupt ending of the NMM characterizing the PE-clause. In what follows, we provide a full-fledged paradigm instantiating all possible combinations of the functions carried out by the antecedent in the PE-clause and the correlate (overt or silent) in the main clause. Subj-Obj ___________________rel (21) dogi ixi eat a-lot pei doctor vet (ixi) bring ‘I took to the vet the dog that eats a lot’ In (22), the correlate in the main clause is in subject position and the antecedent (DOG) corresponds to the object of the PE-clause. Obj-Subj ____________________________rel (22) doctor vet ixi dogi bring pei (ixi) eat a-lot ‘The dog that I took to the vet eats a lot’



Italian Sign Language relative clauses 

In (23), the correlate corresponds to an object position in the main clause and the antecedent (DOG) is the object of the PE-clause. Obj-Obj ______________________rel (23) yesterday dogi find pei paolok ixk (ixi) wash ‘Paolo washes the dog that I found yesterday’ Finally in (24) through (26), we find various combinations where the antecedent is an adjunct and the correlate an argument (24); both are adjuncts (25); the antecedent is an argument and the correlate in the main clause an adjunct (26). Adj-Arg rel (24) girli dual-pron study together pei giannij ixj like ixi ‘Gianni likes the girl with whom I study’ Adj-Adj __________________________rel (25) son my play near tablei pei (ixi) ix key forgot ‘I forgot the key near the table where my son plays’ Arg-Adj rel (26) maria tablei burn pei (ixi) near child play ‘The child plays near the table that Maria burnt’ From the paradigm sketched above, we observe that the antecedent is always internal to the PE-clause. This is confirmed by the distribution of time adverbials, which, as we have seen, are always in sentence initial position in LIS, and precede the antecedent, as illustrated in (23). Our data do not show different relativization patterns corresponding to the different roles carried out by the antecedent in relation to the PE-clause and to the main clause. More specifically, the PE-clause always translates the equivalent of the Italian relative clause.10 As (27) shows, this conclusion also holds for abstract antecedents. _____________________________rel (27) paolo maria ideai suggest pei important ‘The idea that Paolo suggested to Maria is important’ As the data show, there is some variation in the spreading of the NMM ‘rel’. While the most frequent pattern is that of ‘rel’ spreading over the entire PE-clause, other options are also produced by our informants. ‘rel’ might also be realized only over PE, as in

 Chiara Branchini and Caterina Donati

(24) repeated here as (28), or over the material intervening between the antecedent and PE, as in (29). rel (28) girli dual-pron study together pei giannij ixj like ixi ‘Gianni likes the girl with whom I study’ _____________rel (29) studenti exam done pei alli pass ‘The students that took the exam all passed’ While all these spreading possibilities are consistent with the analysis of PE clauses we shall provide in the next section, in the interests of conciseness, we ignore the reasons driving the choice between them.

5. Correlatives or internally headed relative clauses? We shall now turn towards an analysis and try to interpret these data in the light of the standard typology of relativization strategies. We can start out by safely claiming what PE-causes are not. PE-clauses are not externally headed relative clauses, since the antecedent is systematically situated within the clause itself (Cecchetto et al.  2006). Moreover, we know that PE-clauses are not free relatives since the defining property of free relatives is that they lack an antecedent in the form of an overt NP. We are left with two relativization strategies that appear to be compatible with our data: correlative clauses, and extraposed, internally headed, relative clauses. These two analyses are sketched in the abstract structures given in (30) and (31), respectively. (30) IP correlative wo CP IP 3 3 ... PEi... ...pronouni... (31) IP extraposed, internally headed relative 3 DP IP 2 2 CP D° ...DP... 2 ... PE While (30) and (31) share the property of having the antecedent located internal to the clause, they display three crucial differences. The first one concerns the categorial status of the PE-clause: it is a simple CP in (30), a (complex) DP in (31). The second



Italian Sign Language relative clauses 

difference has to do with the relation of the PE-clause with the main clause. Given the correlative structure in (30), the two clauses start out as two independent adjoined clauses.11 In (31), on the other hand, the PE-clause is first merged within the main clause, in the position of the correlate (the base position for relatives) and then extraposed to a left-adjoined position. The third difference concerns the nature of the nominal correlate in the main clause: it is a pronoun in (30); a trace in (31). In the remainder of this section we shall compare these two analyses in detail using a number of empirical facts concerning PE and its clause. Rather than referring to the correlative option in its abstract version in (30), we will discuss the actual analysis proposed by Cecchetto et al. 2006, which is essentially an implementation of this very option.12

5.1

A correlative analysis: Cecchetto et al. 2006

Cecchetto et al. 2006 argue for a structure as in (32). (32) [IP [CP…..npi ti ……pei ] [IP …….pronouni …..]] The defining traits of (32) are the following. The construction is formed by two conjoined clauses. The first clause (PE13-clause) involves the movement of the demonstrative-like element PE from its base position to the right of the antecedent NP, to the Spec,CP of the clause (which is arguably on the right, cf. (11) above). The second clause is a simple IP, with an E-type pronominal correlate referring to the antecedent NP. Since LIS is a pro-drop language, such a pronominal can be non-overt (pro). As evidence for the movement of PE, Cecchetto et al. 2006 provide the sentence in (33), produced by one of their informants (and considered marginally acceptable by the others). (33) ?? boyi pei phone leave done ‘The boy that called left’ Such optional (overt) movement is motivated within the semantic analysis provided by Cecchetto et al. 2006. We will not review this proposal in any detail. We can simply say that, under this analysis, PE is the element that has the function of connecting the CP to the main clause. It moves either overtly or covertly to Spec, CP in order to scope over its clause and take the clause as its argument. In this position, PE converts the PE-clause into a generalized quantifier (see Dayal’s 1996 analysis for Hindi correlatives). The analysis of the main clause pronoun as an E-type anaphor is related to another crucial property attributed to the construction by Cecchetto et al., namely, that PE-clauses are nonrestrictive. Since there seems to be no direct link between the syntactic type of a relativization strategy and its restrictive or nonrestrictive interpretation, we shall postpone the discussion of this issue to Section 7 and keep it separate from the assessment of the syntactic analysis attributed to the PE-clause. The analysis in (32) has a number of straightforward advantages: first of all, it derives the systematic occurrence of the PE-clause at the left without any further

 Chiara Branchini and Caterina Donati

assumption. Second, given that full NPs cannot be bound while pronominals can, it correctly predicts that the main clause can never contain a full NP in this position. Finally, the claim that (32) is a correlative construction is consistent with classical typological generalizations according to which correlatives are a prerogative of SOV languages (Downing 1973)14. Let us now go back to the three distinctive properties differentiating the correlative analysis from the alternative analysis: the categorial status of the PE-clause; the nature of the correlate and the relation of the PE-clause to the main clause. Let’s discuss some empirical data that might orient the choice of the most suitable structural analysis.

5.2

Evidence for the nominal status of the PE-clause

As we already mentioned in the discussion of the two abstract structures in (30) and (31), a crucial difference between the two available analyses concerns the categorical status of the PE-clause: a simple clause under the correlative analysis; a nominal clause (i.e. a relative) under the internally headed relative analysis. Any evidence of nominal properties of the PE-clause would then advocate the latter. In what follows we shall present some evidence in favour of this analysis. The first fact concerns the possibility of having the PE-clause modified by ordinals, as illustrated in (34). _____________ ______rel (34) first womani kissi pei bank work ‘The first woman I kissed works in a bank’ As the translation clearly shows, the ordinal FIRST in (34) does not simply modify the antecedent WOMAN (the sentence does not entail that I kissed the first woman ever existed, or the first one in a row, etc.), but the entire PE-clause itself, namely the woman I kissed. FIRST cannot therefore be assigned a position internal to the clause, but rather a position external to it. However, ordinals cannot usually modify clauses (not even extraposed ones), neither in LIS nor in any language. Ordinals can only modify nominals, either simple ones as in (35a) or complex ones as in (35b). (35) a. The first woman/one b. The first woman I kissed There might, in principle, be an alternative analysis for (34) which does not imply that the PE-clause has any nominal feature. In this analysis FIRST is an adverbial modifying the verb, as in the woman I kissed first.15 However, this alternative can be excluded since LIS displays no lexical ambiguity in this respect, having two separate signs for ordinal FIRST and adverbial FIRST. A second piece of evidence for the nominal status of the PE-clause might come from a closer examination of the nature and position of PE itself. First of all, notice that



Italian Sign Language relative clauses 

PE is not exclusively used in the context which we have labeled as the PE-clause. We also find it in nominal contexts, as illustrated in (36) through (39). (36) first pe ‘The first one’ (37) small pe ‘The small one’ (38) red pe ‘The red one’ (39) my pe ‘Mine’ In all these examples, PE appears to carry out the function of a determiner nominalizing the constituent. In other cases, it can co-occur with an NP as a determiner strongly marking the referent, as in (40).16 ________rb (40) housei pei annaj ixj buy want ‘It is a house that Anna wants to buy’ Finally, besides its interpretation and distribution, PE also shares with other LIS determiners a similar morphology: both PE and, say, the equivalent of the demonstrative that can agree in Space with an NP which may either be present in the clause or in the discourse context. If PE is to be analyzed as a determiner, it is predicted that it should occur optionally alongside the antecedent. This is attested in Cecchetto et al. 2006 (see (33) above) and confirmed by some of our informants. (41) illustrates this option again. __________________________rel (41) childi pei competition win teacher prize give ‘The teacher gives a prize to the child who has won the competition’ As for the standard position of PE at the right periphery of the PE-clause, we already know that it is located in a position higher then Negation, Aspect and Modal, presumably in the CP area. We will argue in Section 6 that it occurs in a position high enough to project its categorical D status, nominalizing the entire clause.

5.3

Evidence for the correlate as a trace

Recall that the antecedent in the PE-clause is always coreferential with a nominal in the main clause, which we called the correlate. One crucial difference between the two analyses we are comparing concerns the nature of this correlate: it is an anaphoric pronoun in the correlative analysis; a nominal trace (that of the extraposed relative clause) in the extraposed, internally headed analysis. As a matter of fact, the data show

 Chiara Branchini and Caterina Donati

a systematic optionality of the realization of the correlate, which can be either a gap or an indexical NP. Both analyses easily account for this optionality. Cecchetto et al. 2006 relate the possibility of omitting the pronoun to the pro-drop nature of LIS. As for the extraposition analysis, it treats the optional pronoun as a resumptive strategy spelling out the trace of the relative clause. This is reported to be a widespread mechanism both in sign languages and in spoken ones. Notice that the correlate can also be realized as a quantifier, as shown in (42) and (43). rel (42) childreni cake eat pei today all [e] stomachache ‘All the children that ate the cake today have stomachache’ ... ________________rel (43) boyi exam done pei pass nobody17 ‘No boy that took the exam passed’ The distribution and interpretation of quantifiers in our LIS construction is quite intricate and interesting. We shall discuss it in detail in Section 7.2. For the time being, it is sufficient to show that the presence of ALL or NOBODY in the correlate position is compatible with both analyses. Under the extrapositon analysis, ALL can be interpreted as a floating quantifier (Sportiche 1988), and under the correlative analysis, ALL can be seen as a possible realization of the anaphoric element. The two analyses differ, however, in their predictions about cases where the correlate is in an oblique position.18 The correlative analysis predicts that the pronoun should be obligatorily realized since the verb morphology cannot license pro in that position; the extraposition analysis predicts that the same optionality observed in argument positions should hold. The LIS data appear to confirm the prediction of the extraposed internally headed analysis. As shown in (44) and (45), the correlate can be a gap even in an oblique adjunct position. (46) shows that pro is not licensed in an oblique position in LIS. __________________________rel (44) son my play near tablei PEi (IXi) ix key forgot ‘I forgot the key on the table near which my son plays’ rel (45) maria tablei burn pei (ixi) near child play ‘The child plays near the table that Maria burnt’ (46) *children pizza eat near *‘Children eat pizza near’



Italian Sign Language relative clauses 

5.4

Evidence for extraposition

The third distinctive feature discriminating between the two analyses we are discussing concerns the relation between the two clauses of the construction. In the correlative option, the PE-clause is merged in a position left-adjoined to the main clause, while in the alternative, relative-clause analysis, it is merged in a position internal to the main clause (the correlate position) and extraposed to the left. In what follows we will present some facts, all pointing towards an extraposed analysis for the PE-clause. The first piece of evidence has to do with the NMM obligatorily associated with the PE-clause. Available descriptions of the various NMMs in LIS and their respective functions are far from being exhaustive in the literature.19 However, as far as we know, at least one component of the relevant NMM is not the exclusive prerogative of PEclauses. Rather, it is found in a number of syntactic environments, in different combinations with other markers. Trying to characterize what we have labeled ‘rel’ in some detail, we can say that it is a complex marker which includes at least two components: raised eyebrows (‘rb’) and ‘tensed eyes’ (‘te’) – a peculiar tension in the upper area of the face including cheeks and eyes. While ‘rb’ is indeed present in many syntactic environments such as yes/no questions, conditionals, topics and focus constructions, ‘te’ appears to be unique to extraposed constituents. To illustrate, while (47), (48), and (49) exhibit raised eyebrows, ‘te’ is only found in (48) to mark the topic, and in (49) to mark contrastive focus. _____________y/n (47) dog cat chase ‘Does the dog chase the cat?’ ____________topic (48) yes car silver, ix see done ‘Yes, the silver car, I saw it’ ..___focus (49) car silver, ix buy done ‘The silver car, I bought’ We can take this observation as a provisional piece of evidence that PE-clauses belong to the family of extraposed constituents. Interestingly, similar conclusions have been drawn with respect to relative clauses in other sign languages, where relative clauses share both the NMMs and their position with extraposed constituents, see Aarons 1994, Neidle et al. 2000 among others for (cor)relative clauses in ASL and Pfau and Steinbach 2005 for relative clauses in German Sign Language (DGS). The extraposed status of the PE-clause is reinforced by other evidence. LIS syntax systematically requires embedded clauses to sit in a peripheral position, either at the left or at the right. While (50) shows that the unmarked position of the object is before

 Chiara Branchini and Caterina Donati

the verb want, the contrast between (51) and (52) vs. (53) shows that complement clauses cannot remain in their basic position. (50) paolo apple want ‘Paolo wants an apple’ …_____________________te (51) paolo want mariai house possi sell ________________te (52) maria house sell paolo want (53) *paolo maria house poss sell want ‘Paolo wants Maria to sell her house’ If we assume that PE-clauses are extraposed, then we can account for their position and the position of complement clauses with one hypothesis, namely that any embedded clausal structure needs to be marginalized in LIS.20 The exact reason underlying this hypothesis remains unclear but is perhaps linked to cognitive considerations. Notice that when complement clauses are extraposed, they are marked by the marking associated with the PE-clauses in particular and with extraposed constituents in general, i.e. ‘tensed eyes’. The only residual peculiarity of PE-clauses in this framework is that they tend to dislocate more to the left and somewhat less to the right. The reason for this is unclear and remains to be investigated in future research. A different kind of evidence for the extraposed status of the PE-clause concerns reconstruction. For example, consider (54). ____________________________rel (54) one womani make-up not pei ix meet never ‘I never met any woman who doesn’t wear make-up’ As the translation makes clear, the PE-clause falls under the scope of matrix negation even though, at first glance, the negator itself does not appear to c-command the PEclause. This suggests that, for interpretative reasons, the PE-clause is ultimately reconstructed into a position internal to the main clause, that of the correlate. This is exactly what is predicted under the extraposition analysis, where the PE-clause gets merged in that very position. In a similar way, in (55), the quantifier NOBODY in the matrix clause binds the possessive pronoun occurring to its left, within the relative CP. _________________________________rel (55) professor possi computerk buy pek studenti steal nobody ‘No student steals the computer that his professor bought’ A final piece of evidence pointing toward the extraposition of the PE-clause is illustrated in the following examples. While in (56) the PE-clause can be reconstructed in a position internal to the main clause crossing a simple clause boundary, I know, in (57), the relative clause the teacher who gave a price is an island blocking reconstruction



Italian Sign Language relative clauses 

of the relative clause the child who wins. This sensitivity to islands is clear evidence for movement of the PE-clause. __________________________rel (56) [childi competition win pei] [ix know teacher price child competition win pe give] ‘I know that the teacher gave a price to the child who won the competition’ (57) *[childi competition win pei] [teacherk price child competition win pe give pek] [ix know] ‘I know the teacher that gave a price to the child who won the competition’ We can conclude from this brief survey that we have good reasons to believe that the PE-clause is extraposed. Notice, however, that the extraposed status of the clause is not necessarily an argument against a correlative analysis: Bhatt 2003 argues convincingly that Hindi correlative structures are not base generated in their left adjoined position, but rather moved there from a position internal to the main clause. If this were a general fact about correlatives, then the distance between correlative structures and extraposed internally headed relative clauses would be significantly reduced.21

5.5

Concluding remarks

Before concluding this section and turning to a full-fledged analysis, there is a final piece of data that is worth mentioning as it seems to provide evidence against a correlative analysis: PE-clauses can stack, as illustrated in (58). ______________________________rel (58) vasei see done pei today buy pei expensive22 ‘The vase that I saw that I bought today is expensive’ Stacking is a general property of relative clauses. The English translation in (58) is itself an example of stacking. On the other hand, the impossibility of stacking is reported as a typical property of correlatives (Srivastav, 1991, Bhatt 2003, Mc Cawley 2004, amongst others). In this section, we have discussed a number of data pointing towards the nominal status of the PE- clause (Section 5.2), the trace status of the correlate (Section 5.3), and the extraposed status of the PE-clause (Section 5.4). As we observed at the beginning of the section, these are the crucial properties that force us to discard a correlative analysis, such as the one illustrated in Section 5.1. In the next section, we analyse the PE-clause as an extraposed internally headed relative clause.

 Chiara Branchini and Caterina Donati

6. An extraposed internally headed relative clause analysis Suppose that relatives are defined by two crucial properties: (i) they are clausal structures endowed with D-like features (they are so-called ‘complex nominal phrases’ in GB terms); (ii) they involve a dependency between two (nominal)23 positions: one internal to the clause itself, and one external to it, corresponding to the position where the clause is interpreted. To illustrate this, look at the English example in (59) which is derived in the spirit of the so-called ‘raising analysis’ proposed by Kayne 1994 and developed further in Bianchi 1999 for externally headed relative clauses. (59) The dog that chased the cat (came home). DP 2 D CP the 2 NP C' dog 2 [that dog chased the cat] In (59) the two properties (i) and (ii) above are realized through two different syntactic entities. The D-like feature of the clause is provided through the merging of an external D head. The dependency is achieved by the movement of an NP from a position internal to the clause to a position external to it. Notice, however, that there is nothing intrinsic in the definition of relative clauses that requires the two properties (i) and (ii) to be realized through the activation of two distinct elements. This observation is crucial in order to understand the mechanisms behind LIS PE-clauses. The basic idea is that the two properties may also be carried out by the simple movement of one and only element: PE. To clarify this, consider the structure in (60). (60)

DP/CP 3 IP D/C 2 PE [dog pe cat chase]

In (60), the determiner PE moves to a peripheral position (C°) from an internal position. In doing so, PE realizes property (ii), namely the establishing of a dependency between two positions just as dog does in (59). The only difference is that this dependency is achieved through head movement in (60), and through phrase movement in (59). Since PE is a determiner located in the head position of the clause, PE furthermore endows the clause with the required D-like feature (property (i)). The difference with (59) is simply that the D head nominalizing the clause is moved in (60), while it is merged in (59). The notation given in (60), where a double status is assigned to the very



Italian Sign Language relative clauses 

same head and its projection, is meant to signal precisely this: that the C head of the clause derivationally acquires the status of a D head (projecting DP) by hosting the PE head which has head-moved to that position.24 Remember that PE can also be signed in its base position alongside the NP which it usually strands (at least for some signers). See (41) repeated here as (61). __________________________rel (61) childi pei competition win teacher prize give ‘The teacher gives a prize to the child who has won the competition’ Cases like (61) provide important evidence for the derivative status of the external position of PE, which is a crucial component of our analysis. Presumably, in cases like (61), the PE-clause receives its relative properties through the covert movement of PE. This way of interpreting relativization can be usefully extended to structures apparently not related to sign languages. Consider the following data, corresponding to relative clauses in Japanese, Diegueño, Quechua, Tibetan, respectively. (62) Yoko-wa [[Taro-ga sara-no ue-ni keeki-o iota]-no]-o tabeta Yoko-top Taro-nom plate-ge on-loc cake-acc put-nmk-acc ate Yoko ate a cake which Taro put on the plate’ (Japanese: Shimoyama 1999: 147)25 (63) [Tenay ?ewa:ø ?ewu:w]- pu] -Ly ?ciyawx yesterday house-acc I-saw def-in I.will.sing ‘I will sing in the house that I saw yesterday’ (Diegueño: Keenan 1985: 162) (64) [[Kan kwitsa-man kwintu-ta willa] -shka]-ka llapa sumaj-mi you girl-to story-acc tell -nml-top very pretty-validator ‘The girl to whom you told the story is very pretty’ (Quechua: Comrie 1981: 139)26 (65) [[Peemε thep khii-pa] the ] nee yin Peem-erg book-abs carry-part the-abs I.gen be ‘The book Peem carried is mine’ (Tibetan: Keenan 1985: 161)27 These sentences have a lot in common: they all belong to head-final languages; they all involve internally headed relative clauses (within brackets); they all display a determiner-like element (either in the form of a nominalizing morpheme or in the form of a free determiner) at the right edge of the relative clause. We can straightforwardly account for these three facts if we extend the analysis proposed above for LIS (60) to these structures: they might all realize the defining relativization properties (i) and (ii) by moving a determiner head to the C position of the clause.

 Chiara Branchini and Caterina Donati CP/DP

(66)

C°/D°

IP

[…NP D°…] Unfortunately, we do not have any specific knowledge of the relativization strategies in the languages cited above. Therefore, we cannot provide any rigorous typological evidence at this stage that the nominalizers actually move. Nevertheless, we believe that something can be said, although very briefly, for Japanese. Notice that besides relative clauses, the determiner-like element –no shares with PE the same general distribution. They both occur in a number of contexts in which they carry out the same nominalizing function. The following data in Japanese28 perfectly overlap with the data concerning the distribution of PE presented in Section 5.2. (67) a. b. c. d.

Akai-no red-nml ‘The red one’ saisho-no first-nml ‘The first one’ watashi-no I-nml ‘Mine’ chishai-no little-nml ‘The little one’

These data, which we cannot discuss here in detail for obvious reasons of space, suggest that a similar derivation for relative clauses involving PE and –no is plausible. Before concluding, it is worth pointing out an important advantage of our analysis presented here. As has already been demonstrated in (59) and (60), the same function of connecting two nominal positions (property (ii) of relativization) is carried out by two different elements: a nominal phrase in externally headed relative clauses; a determiner head in internally headed relative clauses. This implies that the two types of relatives do not share the same LF representation. This is very different from more standard analyses reducing internally headed relatives to externally headed relatives. By assuming that the antecedent NP moves covertly at LF, they imply that the two types of relatives share the same LF representation (see Ito 1986 for Japanese, Broadwell



Italian Sign Language relative clauses 

1986 for Choctaw, Cole 1987 for Ancash and Imbabura Quechua, Lefebvre and Muysken 1988 for Quechua, Watanabe 1992 for Japanese amongst others). Shimoyama 1999 shows that the two types of relatives in Japanese display important interpretive differences and, in particular, that the antecedent NP is never externalized in internally headed relatives. This is exactly what is predicted given the two structural hypotheses in (59) and (60). Section 7 will be devoted to discussing some interpretive properties of LIS PE-clauses together with Shimoyama’s observations in more detail. Finally, while (60) provides an analysis for the internal structure of the PE-clause and its nominal status, we have not yet accounted for its extraposed position. Like any other relative clause, the PE-clause originates within the main clause in the position where it is interpreted. As for any embedded clause in LIS, it is extraposed.29 We will not characterize any further the position it occupies in the left periphery, due to the lack of a fine-grained cartography of the left periphery in LIS. The diagram in (69) provides an illustration of the overall structure of a sentence like (68). _________________rel (68) dogi cat chase pei home come ‘The dog that chased the cat came home’ (69) IP wo DP IP 3 6 IP D° dp home come 6 PE dog pe cat chase From the data presented in this section, we can conclude that PE-clauses seem to fit naturally within a typology of relativization which is well attested across a number of genetically unrelated languages. As for sign languages, the structure in (69) seems to present an interesting analogy with ASL internally headed relative clauses as described in a classical paper by Liddell 1978 and illustrated by the sentence in (70).30 ________________________________r (70) recently dog that chase cat come home ‘The dog that recently chased the cat came home’ More specifically, (70) shares with PE-clauses the following properties: i. a specific NMM identifying the relative; ii. a demonstrative-like element glossed THAT in (70) coreferent with the antecedent; iii. a strong preference for a sentence initial position. More investigation is needed to verify whether these superficial similarities are derivable from a common analysis such as the one proposed here. We refer the reader to

 Chiara Branchini and Caterina Donati

Pfau and Steinbach 2005 for a typological survey of relativization strategies in the world’s sign languages.

7. Restrictive or appositive? One issue remains to be discussed, namely the question of whether PE-clauses are restrictive or appositive. This question is only partially related to the conclusions about the structure of the PE-clause reached above. Cecchetto et al. 2006 argue for the nonrestrictive interpretation of PE-clauses on the basis of two facts. Let us briefly illustrate them and assess whether their conclusion can be maintained.

7.1

Apparent reversability

The first fact discussed by Cecchetto et al. concerns the alleged ‘reversability’ of the clauses. They observe that their informants, when asked to translate an Italian sentence corresponding to (71), spontaneously produce two reversed constructions, namely those in (72a-b), which can thus be claimed to be equivalent. (71)

Maria kissed a boy that left

(72) a. maria boyi kiss pei leave done ‘A/the boy that Maria kissed left.’ b. boyi leave done pei maria kiss ‘Maria kissed a/the boy that left’ However, when it comes to English restrictive relative clauses, such equivalence does not hold, since the two ‘reversed’ sentences clearly carry different presuppositions. (73) a. Mary kissed the boy that left b. The boy that Mary kissed left (73a) presupposes that there is only one boy such that he left and (73b) presupposes that there is only one boy such that Mary kissed him. In contrast, the equivalence of the pair in (72) is reminiscent of what holds for English appositive relative clauses, as illustrated in (74), where commas stand for a marked intonational pause characteristic of appositives. (74) a. Mary kissed the boy, who left b. The boy, who Mary kissed, left The similarity between the LIS pair in (72) and the two English appositive relative clauses in (74) is striking at first sight. The similarity is interpreted by Cecchetto et al. as indicating that LIS PE-clauses are appositive as well. However, this conclusion does not withstand deeper investigation. The similarity breaks down as soon as a



Italian Sign Language relative clauses 

context of utterance is provided. When trying to reduplicate the results of Cecchetto et al.  with our informants, we found out that the ‘reversed’ PE-clauses in (72a-b) are elicited by different contexts. More precisely, (72a) is only felicitous as answer to the question: ‘Who left?’, while (72b) answers a question like: ‘Who did Mary kiss?’. This kind of context sensitivity clearly shows that the two sentences can indeed carry two different presuppositions, just as the English restrictive relative clauses in (73) do. On the other hand, nothing similar happens with the pair of English appositive relative clauses in (74), whose equivalence remains constant in any given context. When no context is provided, the equivalence of the two ‘reversed’ sentences in (72) observed by Cecchetto et al. might be due to a different factor, namely the ambiguity of the antecedent in LIS. Recall that LIS has no overt means to distinguish definite and indefinite NPs. It follows that any relative antecedent in LIS is ambiguous between a definite reading (like the readings of (73a-b) which are not equivalent at all) and an indefinite reading (like the ones in in (75a-b) which are indeed reversible). (75) a. Mary kissed a boy that left b. A boy that Mary kissed left Here the equivalence in presupposition of the two sentences is not due to the status of the relative clause (which is restrictive), but to the indefiniteness of its antecedent. The same is likely to hold in LIS in any case (when no context is provided) -- any antecedent being systematically ambiguous with respect to its definiteness. Summarizing the argument so far, we can say that the (limited) equivalence of (72a) and (72b) does not provide any evidence for the nonrestrictive status of PE-clauses.

7.2

An unexpected entailment

The second piece of evidence for the appositive nature of LIS relatives discussed by Cecchetto et al concerns the interaction of PE-clauses with universal quantifiers. They observe that while ALL is compatible with the PE-clause, it yields an unexpected interpretation, in which the domain of the quantifier is not restricted by the PE-clause. More precisely, (76) surprisingly entails that all the boys left and all called. Compare this interpretation with that of the English sentence in (77) where no such entailment holds and the domain of the quantifier is restricted by the relative clause. (76) all boysi leave pei ixi call (77) All the boys that left called Cecchetto et al. qualify the LIS interpretation as typically appositive, since it is typically displayed in English by nonrestrictive relative clauses, as (78). (78) All the boys, who left, called. We think, however, that the interpretation of ALL in (76) does not necessarily force the conclusion that PE-clauses are appositive. Remember that under the analysis we have

 Chiara Branchini and Caterina Donati

adopted (see (69) above), no NP is ever externalized from the PE-clause at any level of representation. This is a crucial feature setting apart internally headed relative clauses from externally headed relative clauses. Suppose, for instance, that quantifiers need to be merged with an NP. This means that when ALL is signed next to the antecedent, it is internal to the relative clause in LIS (76) (the relative being internally headed) but external to the relative clause in English (77) (the relative being externally headed). Under this view, the entailment observed in LIS would be a consequence of the fact that ALL is internal to the relative clause. Indeed, the very same entailment holds in English when all is stranded in the base position of the antecedent such as in (79). (79) ?The boys that all left called Although marginal, the relative clause in (79) clearly entails that all the boys left and all the boys called. Notice that (79) contrasts with (77) only in that all sits within the relative clause in the former but not in the latter. Japanese provides independent evidence that we might be on the right track when we are claiming that what is really relevant for the interpretation of quantifiers is the position of the antecedent: internal or external to the relative clause. Shimoyama (1999: 149–150) discusses a minimal pair reported in (80). (80a) is an internally headed relative clause; (80b) is a minimally different externally headed relative clause. (80) a.

Taro-wa [[ Yoko-ga reezooko-ni kukkii-o Taro-top Yoko-nom fridge-loc cookie-acc hotondo irete-oita]- no]-o paatii-ni motte itta. most put-aux nml-acc party-to brought ‘Yoko put most cookies in the refrigerator and Taro brought them to the party.’

b.

Taro-wa [[ Yoko-ga reezooko-ni ∅ irete-oita ] Taro-top Yoko-nom fridge-loc put-aux kukkii-o hotondo] paatii-ni motte itta cookie-acc most party-to brought ‘Taro brought most cookies that Yoko had put in the refrigerator to the party.’

The difference in interpretation which is informally indicated in the translations provided by Shimoyama himself is strikingly reminiscent of the contrast between LIS and English discussed above in (76–77). Quoting Shimoyama, “the two sentences do not share truth conditions (…) The translation for <(80b)> shows that the relative clause ‘Yoko put x in the refrigerator’ constitutes the restriction for the domain of ‘most’ along with ‘cookies’, whereas the translation for <(80a)> shows that the domain of ‘most’ is restricted only by ‘cookies’, but not by the rest of the relative-clause ‘Yoko put x in the refrigerator’” (Shimoyama 1999: 149–150). If we assume that the quantifier needs to be merged with the antecedent, we can account for all these data in a unified way: in internally headed relative clauses, the quantifier is internal to the clause and therefore unrestricted in its domain (80a,76);



Italian Sign Language relative clauses 

the same interpretation holds in sentence (79), an English externally headed relative in which the quantifier is stranded within the clause. The other reading, where the relative clause restricts the domain of the quantifier, can only be obtained in externally headed relative clauses, with the quantifier merged with the external antecedent, cf. (77, 80b).31 If we are on the right track, then the data discussed by Cecchetto et al. cannot per se force the conclusion that PE-clauses are nonrestrictive.

7.3

Testing the status of PE-clauses

When providing a context for the interpretation of the data we elicited, we had the strong suspicion that PE-clauses are restrictive. We therefore tested the clauses for a set of properties standardly associated with restrictivity, using them as a tentative diagnostics for their status. As we shall see, this set of properties is quite heterogeneous, some clearly highlighting a syntactic definition of restrictivity, others pointing to more interpretive characteristics which simply appear to operatively distinguish restrictives from appositives in many languages. The cross-linguistic validity of these properties as diagnostic for the appositive/restrictive status, used in various works before us (see for example Del Gobbo 2005 for a similar procedure on Chinese relatives) encouraged us to apply them to LIS. Although they are standardly meant to be diagnostics for restrictivity in externally headed relative clauses, we tentatively tried to extend their scope to LIS internally headed relative clauses, and see whether this would yield a consistent picture. For each property discussed, we shall first briefly define it, then illustrate it with an English example and finally provide the corresponding LIS sentence we elicited from our informants. i. Sentential adverbs While sentential adverbs can appear inside nonrestrictive relative clauses (81a), they cannot appear inside restrictive relative clauses (81b) (Ogle 1974). (81) a. The boys, who have, by the way, lost the case, should give up b. *The boys who have by the way lost the case should give up As (82) shows, no sentential adverb can appear in a LIS PE-clause. ________________________________rel (82) *womani man by-the-way kiss pei pasta make can *‘The woman that by the way kissed the man can make pasta’ ii. Pronominal head While nonrestrictive relative clauses can modify pronouns (83a), restrictive relative clauses may not (83b). (83) a. We, who are women, think that you, who are men, should go now b. *We who are women think that you who are men should go now

 Chiara Branchini and Caterina Donati

The ungrammaticality of (84) shows that a pronoun (i.e. an indexical NP: IX) cannot be the antecedent of a PE-clause. _____________________________rel (84) *yesterday ixi fell-off bike pei today new glasses buy want *‘You that yesterday fell off the bike today want to buy new glasses’ iii. Matrix negation While an NP modified by a restrictive relative clause can be within the scope of matrix negation (85b), a nonrestrictive relative clause cannot (85a) (Dermidache 1991). (85) a. *I haven’t met a girl, who doesn’t like to wear make-up b. I haven’t met a girl who doesn’t like to wear make-up A relevant sentence was already discussed above in Section 5.3, as a piece of evidence for the extraposed status of the PE-clause. (86) repeats this sentence. Its interpretation shows that the PE-clause is in the scope of the matrix negation. ____________________________rel (86) one womani make-up not pei ix meet never ‘I never met any woman who doesn’t wear make-up’ iv. Stacking While restrictive relative clauses can stack (87b), nonrestrictive relative clauses cannot (87a) (McCawley 1988). (87) a. #The tiger, which was 5 weeks old, which was fed twice a day, ate only fish32 b. The tiger that I saw that I wanted to buy was expensive Stacking was already discussed above as well, as evidence against a correlative analysis for PE clauses, see (58) above repeated here as (88). _______________________________rel (88) vasei see done pei today i buy pei expensive ‘The vase that I saw that I bought today is expensive’ v. Proper name head While nonrestrictive relative clauses can be used to qualify unmodified proper names (89a), restrictive relative clauses cannot (89b). (89) a. John, who you saw yesterday, is a good friend b. *John who you saw yesterday is a good friend As the contrast in (90) shows, a PE-clause is incompatible with a proper noun antecedent. The only way to have a sentence modifying a proper name is by omitting both PE and the non-manual marking ‘rel’.



Italian Sign Language relative clauses 

(90) a. b.

_________________________rel *mariai cake cook like pei prepare done ‘Maria who likes to cook cakes has prepared one’ maria cake cook like prepare done ‘Maria, who likes to cook cakes, prepared one’ ‘Maria likes to cook cakes and has prepared one’

Whether the sentence in (90b) is some kind of an appositive relative or a simple coordination we do not know. Interestingly enough, it was produced by our informants in a context that was meant to elicit a relative clause. vi. Ordinal head An ordinal preceding the antecedent of a restrictive relative clause is restricted in its domain by the relative clause, as seen in (91b). An ordinal preceding the antecedent of a nonrestrictive relative clause on the other hand is only restricted in its domain by the antecedent (cf. 91a). (91) a. The first woman, whom I kissed, works in a bank b. The first woman that I kissed works in a bank The relevant piece of data from LIS, which was discussed in detail in Section 5.2, is repeated here. ____________________rel (92) first womani kissi pei bank work The first woman I kissed works in a bank’ In the LIS sentence in (92), just like in the restrictive relative in (91b), but unlike in the nonrestrictive relative clause in (91a), the ordinal does not just refer to the noun woman, but rather to the first woman I kissed. (vii) Intentional verbs While restrictive relatives can be in the scope of intentional verbs (93b), nonrestrictive relatives cannot (93a) (Zhang 2001). (93) a. #Gianni thinks that Mary likes the men, who own big cars b. Gianni thinks that Mary likes the men who own big cars The interpretation of (94) corresponds to that of (93b): the PE-clause is in the scope of think. …_________________________________rel (94) gianni think meni car cl.big-car pei maria like33 ‘Gianni thinks that Maria likes men that own big cars’.

 Chiara Branchini and Caterina Donati

viii. Ellipsis While the antecedent of a VP ellipsis may include a restrictive relative clause (95b), it may not include a nonrestrictive one (95a). (95) a. My sister likes pizza, which by the way I cook well, and my brother does not (= like pizza) b. My sister likes the cakes I bake, and my brother does not (= like the cakes I bake) The interpretation of (96) shows that what gets reconstructed in the second conjunct includes the PE-clause. ___________________rel ________________rb (96) cakei ixi ixk cook pei sister possk like brother possk not ‘My sister likes the cake that I bake, my brother does not (= like the cake that I bake)’ Other standard diagnostics were not included in the testing because they are not extendable to LIS for independent reasons. For example, a useful diagnostic for singling out appositive clauses in English is heavy pied-piping, shown in the contrast between (97a) and (97b). (97) a. The windows, the curtains of which I really dislike, are very wide b. *The windows the curtains of which I really dislike are very wide It is clear that a category such as pied-piping plays no role in PE-clauses, where the antecedent does not move at all. Summarizing, Table 1 shows the behavior of restrictive and nonrestrictive relatives in English and of LIS PE-clauses with respect to each property that we have discussed in this section. These phenomena provide a very clear and consistent description: they all point unambiguously to the restrictive nature of LIS PE-clauses. Table 1.  The properties of appositive and restrictive relative clauses in comparison to LIS relatives Property sentential adverbs pronominal head matrix negation stacking proper name head ordinal head intentional verbs ellipsis reconstruction

English restrictives

English appositives

LIS relatives

no no yes yes no yes yes yes

yes yes no no yes no no no

no no yes yes no yes yes yes



Italian Sign Language relative clauses 

8. Conclusions In this paper, we have analyzed in detail the construction functionally dedicated to express relativization in Italian Sign language, a biclausal sentence containing what we have labeled, a PE-clause, followed by a main clause. Starting from the important descriptive results of the first pioneering study on these structures (Cecchetto et al. 2006), we have compared two different analyses, which appeared to be compatible with the data. One of these identifies the structure as a correlative construction, as argued in Cecchetto et al. 2006, while the other one analyzes the PE-clause as an internally headed relative clause extraposed to sentence intial position. At first sight, both analyses appear to be compatible with the LIS data. However, a closer examination has revealed that the extraposed internally headed analysis is more adequate. From the evidence discussed, we rejected the correlative analysis by concluding that PE-clauses are relatives in that they have nominal status, they are extraposed in the left periphery from the main clause position where they are interpreted and they leave a trace which is optionally spelled out by a resumptive pronoun. More precisely, relying on the observation that PE is a determiner-like element and that its position at the end of the relative clause is derived, we argued that the relative status of the clause is the result of the PE head moving to C. Being a head, PE in C projects, endowing the clause with the D-like feature necessary for its nominal distribution. By connecting an internal position (that of the antecedent) to the material in the sentence external to the clause, it triggers the interpretation of the sentence as a relative. Under this analysis the conclusion that PE-clauses are relatives does not imply that they share the same LF representation as externally headed relative clauses. This important corollary is discussed in some detail. The final section is dedicated to verifing whether the distinction of restrictive vs. nonrestrictive is nevertheless relevant for PE-clauses. We reached the conclusion that they behave as restrictive relative clauses with respect to all the properties tested.

Acknowledgements This paper would have never existed without the unvaluable collaboration of our informants, whom we thank dearly: Corrado D’Aversa, who also allowed the recordings of his interviews to be made available online; Remo and Romolo Leonori; Fiorella and Pantelis Yacommidakis and Mauro Mottinelli. Thanks are also due to all the members of the Deaf community of Ancona, and especially to the president of the ENS of Ancona Adalberto Nisi and the interpreter Marzia Branchini. For their kind hospitality in Rome, we also thank the ENS of Rome and the SILIS group. We would also like to express our gratitude to Alessandra Brezzi for valuable discussion on Chinese and Japanese relative clauses and to Karine Arnéodo who provided the Japanese data. Earlier versions of this paper were presented at SIGNA VOLANT in Milan, GLOW Asia in Delhi, IGG in Florence, Urbino graduate seminar, CNR seminar in Rome, AFLico in Lille, ALT7 in Paris, Interdisciplinary Approaches to Relative Clauses in Cambridge, seminar in Rovereto:

 Chiara Branchini and Caterina Donati we thank the audiences for their precious questions and comments. In particular, we benefitted a lot from discussing with Carlo Cecchetto, Carlo Geraci, Sandro Zucchi in Milan, and Michela Ippolito, Robert Lee, Carol Neidle, Michael Schlang in Boston. We finally need to thank two anonymous reviewers as well as Anikó Lipták and Carlo Cecchetto, who read an earlier version and helped improving it with their comments. Technical supplies for video recordings were financed through a research grant to Branchini by Regione Marche, Assessorato ai Servizi Sociali, and by the University of Urbino.

References Aarons, Debra. 1994. Aspects of the syntax of American Sign Language. Doctoral dissertation. Boston University. Bahan, Benjamin. 1996. Non-manual realization of agreement in American Sign Language. Doctoral dissertation. Boston University. Bhatt, Rajesh. 2003. Locality in correlatives. Natural Language and Linguistic Theory 21. 485–541. Bianchi, Valentina. 1999. Consequences of Antisymmetry: Headed relative clauses. Berlin: Mouton de Gruyter. Branchini, Chiara. 2007. On Relativization in Italian Sign Language. Doctoral dissertation. Università di Urbino. Broadwell, George A. 1986. A-bar Anaphora and Relative Clauses. In S. Berman, S., Choe, J-W., McDonough, J. (eds), Proceedings of the 16th Conference of the Northeast Linguistic Society, 47–56. Amherst, MA: GLSA. Cable, Seth. This volume. The syntax of the Tibetan correlative. Caselli, M. Christina, Simonetta Maragna & Virginia Volterra. 2006. Linguaggio e sordità. Bologna: Il Mulino. Cecchetto, Carlo & Sandro Zucchi. 2004. Why is Spec, CP on the Right in Sign Languages? Glow Newsletter. Cecchetto, Carlo & Caterina Donati (in press). On Labeling: Principle C and Head Movement. Syntax. Cecchetto, Carlo, Carlo Geraci & Sandro Zucchi. 2006. Strategies of relativization in Italian Sign Language. Natural Language and Linguistic Theory 24. 945–975. Chomsky, Noam. 2004. On Phases. Unpublished Ms. MIT. [Available on-line at: http://www.phon.ucl.ac.uk/home/hans/mrg/chomsky_onphases_1204.pdf Citko, Barbara. This volume. What don’t wh-questions, free relatives, and correlatives have in common? Cole, Peter. 1987. The structure of Internally Headed Relative Clauses. Natural Language and Linguistic Theory 5. 277–302. Comrie, Bernard. 1981. Language Universals and Linguistic Typology: Syntax and Morphology. Chicago: University of Chicago Press. Dayal, Veneeta. 1996. Locality in WH Quantification: Questions and Relative Clauses in Hindi. Dordrecht: Kluwer. Del Gobbo, Francesca. 2005. Chinese Relative Clauses: Restrictive, Descriptive or Appositive. In Nicola Munaro, Walter Schweikert & Giuseppina Turano (eds), Contributions to the 30th Incontro di Grammatica Generativa, 207–305. Venezia: Libreria Editrice Cafoscarina.



Italian Sign Language relative clauses  Dermidache, Hamida. 1991. Resumptive Chains in Restrictive Relatives, Appositives and Dislocation Structures. Doctoral dissertation. MIT. Donati, Caterina. 2006. On Wh-head-movement. In Lisa L. Cheng & Norbert Corver (eds), Whmovement on the Move, 21–46. Cambridge, MA: MIT Press. Downing, Bruce. 1973. Correlative relative clauses in universal grammar. Minnesota Working Papers in Linguistics and Philosophy of Language 2, 1–17. Emonds, Joseph. 1979. Appositive Relatives have no Properties. Linguistic Inquiry 10. 211–243. Franchi, Maria Luisa. 2004. Componenti Non Manuali. In Virginia Volterra. (ed.), La Lingua Italiana dei Segni, 159–178. Bologna: Il Mulino. Geraci, Carlo. 2004. L’ordine delle parole nella LIS. In Albano F. Leoni et al. (eds), Parlato italiano, Atti del convegno nazionale di Napoli. Napoli: D’Auria Editore. Geraci, Carlo. 2006. Negation in Italian Sign Language (LIS). In Leah Bateman & Cherlon Ussery (eds), Proceedings of the North East Linguistic Society 35, 217–229. Amherst, MA: GLSA. Geraci, Carlo, Marta Gozzi, Constanza Papagno & Carlo Cecchetto. 2008. How grammar can cope with limited short-term memory: simultaneity and seriality in sign languages. Cognition 106 (2). 780–804. Grosu, Alexander. 2002. Strange relatives at the interface of two millennia. Glot International 6. 145–167. Ito, Junko. 1986. Head-Movement at LF and PF. University of Massachusetts Occasional Papers in Linguistics 11. 109–138. Jackendoff, Ray. 1977. X’ Syntax: A study of Phrase Structure. Cambridge, MA: MIT Press. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Keenan, Edward L. 1985. Relative Clauses. In Timothy Shopen (ed.), Language Typology and Syntactic Description II: Complex Constructions, 141–170. Cambridge, MA: Cambridge University Press. Lefebvre, Claire & Pieter Muysken. 1988. Mixed Categories: Nominalizations in Quechua. Dordrecht: Kluwer. Liddell, S. 1978. Nonmanual Signals and Relative Clauses in ASL. In Siple, P. Ed. Understanding Language through Sign Language Research, 59–90. New York: Academic Press. Lipták, Anikó. 2005. Correlative Topicalization. Unpublished Manuscript. LUCL, Leiden University. McCawley, James D. 1988. The syntactic phenomena of English. Chicago: Chicago University Press. McCawley, James D. 2004. Remarks on Adsentential, Adnominal, and Extraposed Relative Clauses in Hindi. In Veneeta Dayal, Mahajan, Anoup (eds), Clause Structure in South Asian Languages. Dordrecht: Kluwer. Neidle, Carol. 2002. Language across modalities: ASL focus and question constructions. In Pierre Pica & Johan Rooryck (eds), Linguistic Variation Yearbook, 71–98. Amsterdam: John Benjamins. Neidle, Carol, Judy Anne Kegl, Dawn MacLaughlin, Benjamin Bahan & Robert G. Lee. 2000. The Syntax of American Sign Language. Cambridge, MA: MIT Press. Neidle, Carol, Stan Sclaroff & Vassilis Athitsos. 2001. Sign StreamTM: A Tool for Linguistic and Computer Vision Research on Visual-Gestural Language Data. Behaviour Research Methods, Instruments and Computers 33. 311–320. Ogle, Richard. 1974. Natural Order and Dislocated Syntax. Doctoral dissertation. UCLA. Petronio, Karen & Lillo Martin, Diane. 1997. Wh-movement and the Position of Spec, CP: Evidence from American Sign Language. Language 73. 18–57. Pancheva, Roumyana. 1997. The syntax and semantics of correlative proforms. In Kiyomi Kusumoto (ed.), Proceedings of NELS 26, 133–147. Amherst, MA: GLSA.

 Chiara Branchini and Caterina Donati Pfau, Roland. 2004. Doing it the right way: Wh-movement in Indo-Pakistani sign language. Talk given at Zentrum fuer Allgemeine Spachwissenschaft, Typologie & Universalien Forschung, Berlin. Pfau, Roland & Marcus Steinbach. 2005. Relative Clauses in German Sign Language: Extraposition and Reconstruction. Leah Bateman & Cherlon Ussery (eds), Proceedings of the North East Linguistic Society 35. 507- 521. Amherst, MA: GLSA. Ross, John, R. 1967. Constraints on Variables in Syntax. Doctoral dissertation, MIT. Sells, Peter. 1985. Restrictive and Non-Restrictive Modification. CSLI Report. Stanford, CA. Shimoyama, Junko. 1999. Internally Headed Relative Clauses in Japanese and E-type Anaphora. Journal of East Asian Linguistics 8, 147–182. Sportiche, Dominique. 1988. A Theory of Floating Quantifiers and its Corollaries for Constituent Structure. Linguistic Inquiry 19. 425–449. Srivastav, Veneeta. 1991. The Syntax and Semantics of Correlatives. Natural Languages and Lingusitic Theory, 9. 637–686. Volterra, Virginia. 2004. La Lingua Italiana dei Segni. Bologna: Il Mulino. Watanabe, Akira. 1992. Wh-in-situ, Subjacency, and Chain Formation. MIT occasional papers in Linguistics 2. Wilbur, Ronnie B. & Cynthia Patschke. 1999. Syntactic Correlates of Brow Raise in ASL. Sign Language and Linguistics 2. 3–40. Williamson, Janis. 1987. An Indefiniteness Restriction for Relative Clauses in Lakhota. In Eric Reuland & Alice ter Meulen (eds), The Representation of (In)definiteness, 168–190. Cambridge, MA: MIT Press. Zhang, Niina. 2001. Sell Non-Restrictive Relative Clauses, Manuscript. National Chung Cheng University. Zucchi, Sandro. 2003. The Semantics of FATTO: a case of grammaticalization. Paper presented at the 14th Amsterdam Colloquium. Zucchi, Sandro. 2004. Monsters in the Visual Mode? Manuscript. Università degli Studi di Milano.

Endnotes 1. See Caselli, Maragna and Volterra 2006 for a survey of the socio-linguistic and socio-cultural situation of the Italian Deaf community. 2. Information on SignStream is available at http://www.bu.edu/asllrp/SignStream. See Neidle, Sclaroff and Athitsos 2001 for details. 3. The clips corresponding to the examples discussed are available at: www.psicologia.unimib. it/relativeclauses.php 4. As recently demonstrated in Geraci 2004. 5. Notice that there are three typologies of manner adverbials in LIS: some of them are expressed manually as a separate sign (e.g. ON TIME in (5)); others are incorporated in the verb by altering its movement; others are realized non-manually simultaneously to the verb. 6. As glossed here, (6) might be ambiguous (Who chases the cat?/Who does the cat chase?). See below for the role of NMMs for disambiguating this kind of sentences.



Italian Sign Language relative clauses  7. Bahan 1996 shows for ASL that coreference and verbal agreement can be signaled by nonmanual markers as well. Although no specific research on this has been carried out yet, it seems to be true in LIS as well. 8. See Volterra 2004, Zucchi 2004 for a more accurate characterization of the non-manual markings in LIS. Notice that NMMs can co-occur with a corresponding manual sign, as shown in (4) repeated here in (i) for negation, this having an impact on its spreading extension and realization: see Geraci 2006 on negation.



_______neg

(i) dog cat chase not ‘The dog does not chase the cat’ 9. ‘PE’ corresponds to ‘PROREL’ in the glosses provided by Cecchetto et al. 2006, where this element was first identified. 10. Cecchetto et al. 2006 report frequent ‘reversals’, where the input relative clause is translated in the LIS counterpart as a main clause and vice versa. The informants would give e.g. (ii) to express the sentence in (i) below.

(i) Maria kissed a boy that left



(ii) mariai boyj ikissj pej leave done

“The boy that Maria kissed left” We were not able to elicit any of such reversals. See Section 7 for a discussion. 11. But as we shall discuss later in the paper, Bhatt 2003 provides a number of arguments showing that the left adjoined position of the clause in Hindi correlative constructions is derived through movement. See Section 5.3 below. 12. See Citko (this volume) for a very detailed survey of the properties of correlatives as opposed to full relatives and free relatives. 13. As already mentioned in fn 9, Cecchetto et al. 2006 use a different label for the element we gloss PE, namely PROREL. For sake of simplicity, we reframed their analysis discussed here unifying the terminology and systematically replacing their PROREL with the more descriptive PE. 14. This crosslinguistic generalization has been challenged recently: a.o. see Lipták 2005 for Hungarian and Pancheva 1997 for Bulgarian. Both languages are SVO languages displaying correlatives. 15. We thank Carlo Cecchetto for pointing out this possibility to us. 16. Notice that in (40) the NP modified by PE is obligatorily extraposed. Together with its focus interpretation, this characteristic might advocate for a cleft analysis of the construction. See Branchini 2007 for details. 17. For the position of NOBODY at the end of the clause as a general rule for negative words in LIS see Geraci 2006. 18. Cable (this volume) applies similar tests to a relative-like construction in Tibetan, obtaining opposite results, in line with the predictions of the correlative analysis. 19. See Franchi (2004) for an overview. 20. See Geraci, Gozzi, Papagno, Cecchetto 2008 for an interesting hypothesis connecting this avoidance of center embedding in sign languages to the documented lower short term memory performance displayed by signers. Of course, this kind of processing trigger for clausal extrapo-

 Chiara Branchini and Caterina Donati sition is not extraneous to spoken languages as well, as is well known for German and other OV languages. As for relativization, notice that Lipták 2005 derives the left adjoined status of Hungarian correlative clauses from the same kind of processing strategy. 21. See Cable (this volume) for similar conclusions on correlatives in Tibetan. 22. That what we have here is really two stacked PE-clauses, and not simply two conjoined PEclauses referring to the same antecedent (roughly corresponding to The vase that I saw and that I bought is expensive) is confirmed by looking at the context of utterance: in the context provided for (58) a number of vases that I saw were available. (58) was meant to pick up among them the one I bought. 23. While core cases of relative clauses involve nominal positions, there are a number of adverbial relative clauses attested crosslinguistically. A familiar example is that of ever- free relatives in English, as in (i). (i) I will eat however much you will. Thanks to Anikó Lipták for reminding us about this. 24. See Chomsky 2004, Donati 2006, Cecchetto and Donati in press for a discussion of this ‘projection property’ of head (movement). 25. Notice that Shimoyama 1999 provides a different translation for the example given, making explicit its non restrictive status. Being irrelevant here, we preferred to give a more transparent translation See Section 6 for a discussion of these interpretive facts. 26. Actually the clause is ambiguous, as signaled by Comrie: it can also mean ‘the story you told the girl is very pretty’. 27. For Tibetan, Cable (this volume) discusses a different construction, arguably a correlative one. 28. These data were provided to us by Karine Arnéodo, whom we hereby thank. 29. See § 5.4 above for details. 30. In his 1978 paper, Liddell reports that ASL displays an externally headed strategy together with the internally headed one discussed here. (i) provides an illustration from Liddell 1978. ___________r _nod (i) ask give dog ursula kick that ‘I asked him to give me the dog that Ursula kicked’ However, Liddell has recently admitted that ASL displays a single strategy of relativation, namely, internally headed relative clauses (personal communication). 31. Admittedly, this cannot be the whole story. Recall that both in LIS and in English (and possibly, in Japanese), quantifiers can appear floating in the main clause (see Section 5.3). The interpretation is then invariably the same, one in which the quantifier is restricted in its domain by the relative. (i) a. boysi exam done pei today all pass done ‘All the boys that took the exam today passed’ b. The boys that took the exam all passed While this interpretation is not surprising for English, this fact is puzzling for LIS in light of what we claim here. 32. Notice that by stacking we mean a relative clause modifying another relative clause modifying an antecedent, not just two conjoined relative clauses modifying the same antecedent (which is possible with nonrestrictive relative clauses as well).



Italian Sign Language relative clauses  33. The sentence in (94) is the only case we found where the PE-clause does not occur sentence-initially. Notice, however, that it is not a counterexample to the generalization according to which PE-clauses are always left-extraposed (see § 5.4): here it is indeed at the left periphery of the clause it belongs to, which happens to be embedded under the matrix intentional verb, and thus (rightward) extraposed.

part 2

The derivation of correlatives

The syntax of the Tibetan correlative Seth Cable This paper examines a construction in Lhasa Tibetan that might be identified as a ‘correlative’. I first argue that the construction is indeed best analyzed as a correlative (rather than as a simple conditional). I then explore the syntax of the construction in further detail, particularly the relationship between its key components: the subordinate clause and the demonstrative phrase. Given the properties of this relationship, I argue that the Tibetan correlative is threeway ambiguous structurally, i.e., that Tibetan grammar derives the surface construction in three different possible ways. This provides an interesting contrast with the correlative found in Hindi (Bhatt 2003), and I propose that this may be related to the absence of ‘phi-agreement’ in the pronouns of Tibetan. Keywords: Tibetan, correlative, derivation, binding, island-sensitivity

1. Introduction The Tibetan translation of an English sentence containing a free relative – such as (1a) – seems to employ a structure in which an adjoined CP containing an interrogative phrase is associated with a pronoun or demonstrative phrase in the matrix clause. The Tibetan structure is illustrated in (1b).1

(1) a. I killed whatever yak you bought.

b. [CP Khyodra-s gyag gare nyos yod na ] you-erg yak what buy aux if nga-s de bsad pa yin.2 I-erg that kill perf aux ‘I killed whatever yak you bought.’ (Lit. ‘If you bought what yak, I killed that’.) Somewhat tendentiously, I shall refer to the construction in (1b) as ‘the Tibetan Correlative’. In this paper, I will explore the syntax of this construction, comparing it in particular to the more well-studied correlative construction of Hindi.

 Seth Cable

In the next section, I begin by introducing some basic facts about Tibetan. Section 3 then introduces certain basic facts regarding the Tibetan Correlative. The motivation for referring to this structure as a ‘correlative’ is explained, and certain elementary claims regarding its structure and meaning are defended. The fourth section explores the gross phrasal structure of the Tibetan Correlative. It is argued that the Tibetan Correlative arises from three possible derivations within the grammar of Tibetan. This is an interesting contrast with the correlative found in Hindi, as described in Bhatt (2003), and a possible reason for this difference is proposed in Section 5. The paper ends with a summarizing conclusion.

2. Tibetan: An outline Throughout this paper, I confine my attention to the dialect of Tibetan known as Lhasa Tibetan (DeLancey 2003b), and I use the term ‘Tibetan’ to refer to this particular Tibetan dialect. Lhasa Tibetan, which is native to an area of Central Tibet surrounding the city of Lhasa, is the best-studied living dialect of Tibetan, and functions as a lingua franca and linguistic standard throughout the Tibetan Autonomous Region (TAR). Other major dialects of Tibetan include Khams, spoken in the Eastern TAR, Amdo, spoken in the Chinese province of Qinghai, and the Western and Southern dialects spoken in the Western TAR, Nepal and Pakistan.3 All Tibetan dialects belong to the ‘Bodic’ branch of the Tibeto-Burman family, ultimately a sub-branch of Sino-Tibetan. Tibetan is natively written in the Tibetan script, a Brahmic writing system based upon the seventh-century Gupta script. Like all scripts within the Brahmic family, the Tibetan script is an abugida, and every consonantal symbol has the ‘inherent’ vowel quality of /a/. Famously, the spelling conventions used in present-day Tibetan reflect pronunciations common only as late as 1000 CE, prior to drastic simplifications of the language’s syllable structure. Thus, the root bsad meaning ‘kill’ in (1b) is in Lhasa Tibetan pronounced /se/ (ignoring tone). This paper follows the common convention of representing words of Tibetan by transliterating their Tibetan spelling into a Roman alphabet. In this paper, all Tibetan data are transliterated using the Wylie transliteration system (Wylie 1959). The ‘default’ word order within a Tibetan sentence is SOV, though pre-verbal word order is essentially free. As in many languages, word-order freedom in Tibetan seems to be accomplished through two distinct methods of ‘scrambling’: A-scrambling and Abar scrambling (Mahajan 1990). A-scrambling allows for the re-binding of anaphors.



The syntax of the Tibetan correlative 

(2) a. * [DP Gzugs-gzugs1 gi amalags ]-gi Sgrolma1 mthong pa red. body-body gen mother erg Droma see perf aux ‘His1 mother saw Droma1.’ b. Sgrolma1 [DP gzugs-gzugs1 gi amalags ]-gi mthong pa red. Droma body-body gen mother erg see perf aux ‘Droma1’s mother saw him1.’ A-bar scrambling, however, does not. The sentence in (3) illustrates.4 (3) * Sgrolma1 ni Norbu-s [CP[DP gzugs-gzugs1 gi amalags ]-gi mthong pa Droma top Norbu-erg body-body gen mother erg see perf red ] lap pa red. aux say perf aux ‘Droma1, Norbu said his1 mother saw.’ Although constituents preceding the verb in a Tibetan sentence may be freely ordered, no constituent may follow the verb (aside from the inflectional material within the ‘verbal complex’). Tibetan is thus a ‘rigidly verb-final’ language. (4) * Norbu-s lap pa red [CP nga-s Sgrolma mthong pa red ] Norbu-erg say perf aux I-erg Droma see perf aux ‘Norbu said that I saw Droma.’ The putative existence of correlatives within Tibetan therefore challenges the typological generalization that correlatives only appear in ‘loose’ verb final languages, verb final languages where some constituents – particularly ‘heavy’ ones – may appear to the right of the verb (Downing 1973, Keenan 1985). It should be noted, though, that this generalization is independently challenged by the existence of correlatives in such VO languages as Hungarian (Lipták 2005) and the Slavic languages (Izvorski 1997). In addition to its V-finality, Tibetan has a number of other properties indicative of a head-final alignment, including the use of postpositions, V-Auxiliary order, NP-Determiner order, and Genitive-NP order. No overview of the Tibetan language would be complete without some mention of its complex auxiliary system. The auxiliary system of Tibetan includes a rich array of ‘copular verbs’, such as yin, red, yod, and ’dug, as well as a number of more contentful verbs, such as song, zhag, and byung. The choice of auxiliary is affected by such factors as whether the proposition is part of the speaker’s de se knowledge, the identity of the subject across clauses (switch-reference), volition of the subject, motion towards speaker or addressee, whether the proposition ‘benefits’ the speaker, and evidentiality. Any given auxiliary may cross-reference several of the aforementioned properties, leading to a highly complex and richly expressive system.5 In the interests of simplicity, however, I will simply gloss all auxiliaries throughout this paper with the label aux.

 Seth Cable

3. Basic properties of the Tibetan correlative 3.1

Correlatives in Tibetan

I assume that the following is an uncontroversial definition of the term ‘correlative construction’.6 (5) A correlative construction is any structure containing: – an adjunct CP containing a (wh- or relative) operator – a pronoun or demonstrative phrase, occupying an argument position and ‘associated with’ the aforementioned adjunct CP. I will refer to the adjunct CP of a correlative construction as the ‘Correlative CP’, and I will refer to the pronoun or demonstrative phrase of a correlative construction as the ‘Correlative DP’. Correlative constructions have an impressively broad distribution amongst the languages of the world. However, they are perhaps best known from the literature on Indo-Aryan languages such as Hindi. Sentence (6) presents an example of a Hindi correlative, taken from Bhatt (2003). (6) [CP jo CD sale-par hai ], Aamir us CD-ko khari:d-ega: rel CD sale-on be Aamir that CD-acc buy-fut.msg ‘Aamir will buy the CD that is on sale.’ (Lit. ‘Which CD is on sale, Aamir will buy that CD’) In light of sentence (6) and the definition in (5), consider again the Tibetan sentence in (1b), repeated below. (7) [CP Khyodra-s gyag gare nyos yod na ] nga-s de bsad pa yin. you-erg yak what buy aux if I-erg that kill perf aux ‘I killed whatever yak you bought.’ (Lit. ‘If you bought what yak, I killed that’.) This sentence seems to possess an adjoined CP containing a wh-operator. This adjoined clause, moreover, appears to bear some form of construal-like relationship with a demonstrative phrase in the matrix clause. In light of these properties, sentences such as (7) will be regarded as exemplifying ‘correlative constructions.’7 Collected under (8) are sentences further exemplifying this Tibetan construction. Note the range of interrogative phrases (in italics) that may appear within the Correlative CP. 8 (8) a. [ Khyodrang gis mi su thug na ] nga de thug gi yin. you erg man who meet if I that meet non.past aux ‘I will meet whatever man you meet.’



The syntax of the Tibetan correlative 

b. [ Deb gagi khyodrang gis mthong na ] nga de nyon book which you erg see if I that buy gi yin. non.past aux ‘I will buy whatever book you see.’ c. [ Khyi ga’drazhig khyodrang gis mthong na ] nga de nyon dog what.kind.of you erg see if I that buy gi yin. non.past aux ‘I will buy whatever dog you see.’ d. [ Khyodrang gapa-r ’gro na ] nga de la ’gro gi yin. you where-dat go if I that dat go non.past aux ‘I will go wherever you go.’ e. [ Khyodrang gadus ’gro na ] nga ’ gro gi yin. you when go if I go non.past aux ‘I will go whenever you go.’ f. [ Morang zhabsgyo ga’dra rgyab na ] she dance how do if nga zhabsgyo de’dra rgyab gi yin. I dance that way do non.past aux ‘I will dance however she dances.’ g. [ Khyodrang gi ’oma gatshod ’thung na ] nga-s ’thung you erg milk how.much drink if I-erg drink gi yin. non.past aux ‘I will drink however much milk you drink.’ The interrogative phrases appearing above are representative of all those in the language.9 Thus, it seems that any interrogative phrase in the language may appear within the Tibetan Correlative construction. The English translations in (7) and (8) reflect the fact that the Tibetan Correlative seems to admit of a kind of ‘universal reading’. That is, a sentence such as (8d) has a meaning that may be paraphrasable as ‘I will go anywhere that you go.’ Although this is the reading I have chosen to indicate in the translation, these sentences may also be interpreted in a second way; the Tibetan Correlative also admits of a kind of ‘definite reading’. The sentence in (7), for example, is reported as felicitous in contexts in which there is a single yak which the addressee has purchased, and the speaker wishes to communicate that they have killed that particular yak.10 This ‘definite reading’ of the correlative is pragmatically forced in sentences such as the following.

 Seth Cable

(9) [ Tshagspar sgam ga’di nang la bzhag yod na ] deb de-’i newspaper box what inside dat placed be if book that-gen nang la yod red. inside dat be aux ‘The book is inside whatever box the newspaper is inside of.’ (10) [ Deb cogtse ga’di-’i ’og la yod na ] sgam yang de’i ’og book table what-gen under dat be if box also that-gen under la yod red. dat be aux ‘The box is under whatever table the book is under.’ The availability of both a ‘definite’ and a ‘universal’ reading is also a property reported for correlatives in Hindi (Srivastav 1991, Dayal 1995, 1996), supporting the original diagnosis of these Tibetan structures as true correlatives. Interestingly, as our English translations again indicate, this is also a property of free relatives in English (Jacobson 1995). Indeed, Srivastav (1991) and Dayal (1995) propose an explicit semantic connection between English-style free relatives, in which an interrogative clause behaves like a sentential argument, and Hindi-style correlatives, in which an interrogative clause behaves as a dislocated antecedent to a DP. The reader will note, moreover, that some of the illustrative examples in (8) do not contain overt demonstratives or pronouns within their matrix clauses (8e, 8g). The possibility thus arises that some of the structures claimed above to be correlatives might simply be English-style free relatives. Furthermore, one might wonder whether the Tibetan Correlative itself is simply a structure built upon an English-style argumental free relative, as English does seem to permit structures such as those in (11). (11) a. Whatever box the newspaper is inside of, the book is inside of that. b. Wherever you go, I will go there too. We will presently see, however, that the Tibetan constructions illustrated above do not in any way involve English-style free relatives. 11

3.2

Tibetan correlatives are not free relatives

An interesting pattern of data argues that English-style argumental free relatives are not structures available in Tibetan. First, note that null arguments are a rather common feature of Tibetan discourses. (12) Nga Norbu la dgagi yod. Gyag bsad pa red. I Norbu dat good aux yak kill perf aux ‘I like Norbu. He killed a yak.’ Thus, for sentences such as (8e), (8g) or (13) below, a phonologically null pronoun may simply be obscuring what is ultimately a basic correlative structure.



The syntax of the Tibetan correlative 

(13) Nga [CP khyodrang gapa la ’gro na ] Ø ’gro gi yin. I you where dat go if pro go non.past aux ‘I’ll go wherever you go.’ We are presented, then, with two prima facie possibilities: either the bracketed material in (7), (8) and (13) is an argument, as in English-style free relatives, or it is a CP adjunct as in Hindi-style correlatives. To decide between these two analyses, one should seek to determine how the bracketed material in the Tibetan Correlative behaves with respect to generalizations that seem to otherwise hold of argumental XPs. In this regard, consider that the bracketed material in (13) cannot be marked by the dative postposition. (14) * Nga [ [ khyodrang gapa la ’gro na ] la ] ’gro gi yin. I you where dat go if dat go non.past aux ‘I will go to wherever you go to.’ The dative postposition, however, is otherwise required for any overt directional argument of the verb ’gro ‘go’. (15) Nga [ khyodrang gapa la ’gro na ] de *(la) ’gro gi yin. I you where dat go if that dat go non.past aux ‘I’ll go wherever you go.’ The unacceptability of (14) could not be due to an inability for la to take phrasal material as its complement. Given the possibility of sentences such as (16), all the standard descriptive literature for Tibetan regards la as a phrasal postposition. (16) Nga [ [ [ [ dep ] dmar-po ] de ] la ] dga-po yod.12 I book red the dat good be ‘I like the red book.’ Moreover, the impossibility of (14) could not be due to a violation of the ‘matching effect’ for free relatives, as the case requirements on the embedded interrogative phrase are necessarily identical to those placed upon the argument of the matrix verb. We see, then, that the impossibility of (14) would be difficult to explain under an analysis in which the bracketed material constitutes an English-style free relative. Sentences like (17) are perfectly well-formed in English. (17) I will go [ to [ whatever1 party you go to t1 ] ]. On the other hand, under an analysis in which the bracketed material is necessarily a CP adjoined to the clause, and the argument position of the matrix clause is occupied by a null pronominal, the impossibility of (14) would follow from the inability for pro in Tibetan to be complement to the dative postposition. This is illustrated in (18), below.

 Seth Cable

(18) * Nga Ø la ’gro gi yin. I pro dat go non.past aux ‘I will go there.’ Furthermore, the acceptability of (13) would follow from the apparent ability for pro to escape the requirement that it be marked by the dative postposition when standing as the locative argument of ’gro. Compare sentence (19a) to sentence (19b). 13 (19) a. * Nga de ’gro gi yin. I that go non.past aux ‘I will go there.’ b. Nga Ø ’gro gi yin. I pro go non.past aux ‘I will go there.’ In summary, our original ‘correlative’ analysis of the structures in (7) and (8) uniquely predicts that the subordinate clausal material in these sentences can be directly adjacent to an adnominal marker if and only if that adnominal marker may take pro as its complement. We have found that this prediction is born out in the case of the dative postposition la. The following data show that this prediction is moreover born out for a number of other adnominal markers in the language. 14 (20) Mnyamdu ‘with’ (Can take pro as complement) a. Nga (kho) mnyamdu cham-cham phyin pa yin. I him with aimlessly go perf aux ‘I walked with him.’ b. [CP Khyodrang su mnyamdu gompa rgyag na ] nga-s (de) mnyamdu you who with walk do if I-erg him with gompa rgyag gi yin walk do non.past aux ‘I will take a walk with whoever you take a walk with.’ c.

Nga-s [CP khyodrang su mnyamdu gompa rgyag na ] (de) I-erg you who with walk do if him mnyamdu gompa rgyag gi yin with walk do non.past aux ‘I will take a walk with whoever you take a walk with.’

(21) Nas ‘from’ (Cannot take pro as complement) a. Nga *(de) nas rkang-thang la yong pa yin. I that from by.foot dat come perf aux ‘I came from there by foot.’



The syntax of the Tibetan correlative 

b. [CP Khyodrang zampa ga nas yong na ] nga *(de) nas you bridge what from come if I that from yong gi yin. come non.past aux ‘I will cross whatever bridge you cross.’ c.

Nga [CP khyodrang zampa ga nas yong na ] *(de) nas I you bridge what from come if that from yong gi yin come non.past aux ‘I will cross whatever bridge you cross.’

(22) Nang ‘inside’ (Can take pro as complement) a. Deb (de-’i) nang la ’dug. book that-gen inside dat be ‘The book is inside there.’ b. [CP Khyodra-s deb sgam ga’drazhig gi nang la bzhag yod you-erg book box what.kind.of gen inside dat put aux na ] nga-s (de- ’i) nang la deb cig bzhag gi yin. if I-erg that-gen inside dat book a put non.past aux ‘I will put a book inside whatever box you put a book inside.’ c.

Nga-s [CP khyodra-s deb sgam ga’drazhig gi nang la I-erg you-erg book box what.kind.of gen inside dat bzhag yod na ] (de- ’i) nang la deb cig bzhag gi yin. put aux if that-gen inside dat book a put non.past aux ‘I will put a book inside whatever box you put a book inside.’

(23) ’Og ‘under’ (Can take pro as complement) a. Sgam (de-’i) ’og la ’dug. box that-gen under dat be ‘The box is under there.’ b. [CP Deb cogtse ga’di-’i ’og la yod na ] sgam (de-’i) ’og book table what-gen under dat be if box that-gen under la yod red. dat be aux ‘The box is under whatever table the book is under.’ c.

Sgam [CP deb cogtse ga’di-’i ’og la yod na ] (de-’i) ’og box book table what-gen under dat be if that-gen under la yod red. dat be aux ‘The box is under whatever table the book is under.’

 Seth Cable

In all the sentences above, it is only when the adnominal element may mark pro that that element may appear directly adjacent to the Correlative CP. Furthermore, I know of no adnominal marker which violates this generalization.15 Although this generalization would be rather mysterious under an analysis in which our Correlative CPs are English-style argumental free relatives, it is an immediate consequence of our ‘correlative analysis’.16 Overall, then, we find that there is good reason to analogize these Tibetan constructions to Hindi correlatives and to refer to them both by similar names. Thus, we might look to studies of the Hindi correlative for some insights into the finer syntactic details of the Tibetan structures. Of course, this is simply a heuristic, and it needn’t preclude there being some nuanced variation between the languages’ correlatives. In Section 4, we shall explore in more depth the syntactic relationship between the Correlative CP and the Correlative DP in the Tibetan construction. It will be found that, in certain respects, this relationship in the Tibetan Correlative systematically differs from that in the Hindi correlative. Before we come to this, however, one final aspect of the internal structure of the Tibetan Correlative CP bears some mention.

3.3

Tibetan correlatives are not conditionals

Unlike the Hindi correlative in (6), the Tibetan Correlative contains a particle na glossed as ‘if ’. Although the ultimate status of na is somewhat unclear, its gloss as ‘if ’ connotes the fact that this particle is also productively used within the language’s conditional statements. 17 (24) [ Kyodrang Lhasa la ’gro na ] nga Ø ’gro gi yin. you Lhasa dat go if I pro go non.past aux ‘If you go to Lhasa, I will go there.’ One might wonder, then, whether sentences such as (7), repeated below as (25), aren’t more straightforwardly related to sentences such as (24). In particular, if the interrogative phrases of Tibetan were interpretable as indefinite DPs, the sentence in (25) might be rather directly translated as the English sentence in (26), a sentence which – via ‘donkey anaphora’ – seems to have much the same truth conditions as the ‘universal reading’ of (25). 18 (25) [CP Khyodra-s gyag gare nyos yod na ] nga-s de bsad pa yin. you-erg yak what buy aux if I-erg that kill perf aux ‘I killed whatever yak you bought.’ (26) If you bought a yak, I killed it. Although it might appear attractive at first to link the universal reading of (25) to the ‘donkey anaphora’ witnessed in conditionals such as (26), a direct reduction of the



The syntax of the Tibetan correlative 

Tibetan Correlative to the conditional is found not to be so straightforward.19 First, although they are both adjunct CPs, the Correlative CP and the protasis (i.e. ‘antecedent’) of a conditional differ in the range of interpretations they allow. A theory under which the Tibetan Correlative is simply an instance of donkey anaphora within a conditional would, for example, fail to account for the ‘definite reading’ available to the Tibetan Correlative. Note that although the ‘universal reading’ of (27a) may be paraphrased more-or-less by the conditional statement of (27b), the ‘definite reading’ may not. This is highlighted by the pair in (28); sentence (28b) doesn’t seem to convey the truth conditions – or carry the presuppositions – of sentence (28a). (27) a. I will eat whatever cake you make. b. If you make a cake, then I will eat it. (28) a. The book is under whatever table the paper is under. b. ?? If the paper is under a table, then the book is under it. Indeed, it seems that sentence (28b) is rather anomalous. The semantics of the English conditional appears to force a universal reading which is inconsistent with one’s world knowledge about spatial relationships of the sort described. This is not so for either the free relative in (28a) or the correlative in (10), repeated below as (29). (29) [CP Deb cogtse ga’di-’i ’og la yod na ] sgam yang de-’i ’og la book table what-gen under dat be if box also that-gen under dat yod red. be aux ‘The box is under whatever table the book is under.’ Rather, speakers indicate that the free relative in (28) and the correlative in (29) are both interpreted as definite descriptions bearing an implication of ‘ignorance’. Recent work in the semantics of free relatives obtains such ‘ignorance’ readings from special mechanisms distinct from the mechanisms used to obtain donkey anaphora in conditionals (von Fintel 2000, Tredinnick 2005). A second fact to bear in mind is that, although the interrogative phrase in (25) appears to be translatable as an indefinite, this does not seem to be an interpretation generally available to Tibetan interrogative phrases. For example, the sentence in (30) cannot be interpreted as a statement containing an indefinite, but only as a question. (30) Khyodrang gapa la phyin pa red. you where dat go perf aux ‘Where did you go?’ (* ‘You went somewhere.’)

 Seth Cable

4. The syntactic relation between the correlative CP and the correlative DP The definition of the correlative construction given in (5) indicates that there is some ‘relationship’ between the Correlative CP and the Correlative DP. When one considers the interpretation of these constructions, there certainly feels to be a semantic relation between the two, perhaps one of ‘binding’ or some other form of ‘co-construal’. Of course, one might also ask whether there isn’t a syntactic relationship between these two elements. On the basis of various elementary facts, Bhatt (2003) proposes that there are three prima facie plausible structures for the Hindi correlative. Each posits a different surface syntactic relation between the Correlative CP and the Correlative DP. The first is the ‘(Non-Local) IP Adjunction’ structure, illustrated in (31).20 (31) (Non-Local) IP Adjunction [IP [ which CD is on sale ]1 Aamir bought [ that CD ]1 ] IP wo [ Correlative CP ]1 IP [ Which CD is on sale ]1 6 …[ Correlative DP ]1… Aamir bought [ that CD ]1 Within this structure, the Correlative CP is initially Merged as an adjunct to the IP projection of the clause. The Correlative CP then acts as a generalized quantifier, binding the Correlative DP within the matrix IP. Aside from this binding, no purely syntactic relation holds between the Correlative CP and the Correlative DP. The second structure Bhatt introduces is that of the ‘Demonstrative XP Adjunction’ structure, illustrated below. (32) Demonstrative XP Adjunction Aamir bought [ [ which CD is on sale ] that CD ] IP wo DP … Aamir wo DP V wo bought Correlative CP1 Correlative DP1 [ which CD is on sale ]1 [that CD]1 Under this analysis, the Correlative CP is initially Merged as an adjunct to the Correlative DP it is associated with. Furthermore, the Correlative CP occupies this IP-internal position at Spell-Out. Although perhaps a bit puzzling at first, this structure



The syntax of the Tibetan correlative 

might underlie sentences in which the subject precedes the Correlative CP, such as the Tibetan sentences in (13), (15), (20c), (21c), (22c) and (23c). The last structure that Bhatt proposes is the ‘IP Adjunction via Movement’ structure, diagrammed in (33). (33) IP Adjunction, via Movement [IP [ which CD is on sale ]1 Aamir bought [ t1 [ that CD ] ] ] IP wo [ Correlative CP ]1 IP Which CD is on sale 3 DP … Aamir 3 DP V 3 bought Correlative CP Correlative DP t1 [that CD] Within this analysis, the Correlative CP is again initially Merged as an adjunct to the Correlative DP it is associated with. Subsequently, however, the Correlative CP moves out of this DP projection and adjoins to the matrix IP. At this position, the Correlative CP – just as any moved phrase – binds its trace within the matrix demonstrative phrase. It is argued in Bhatt (2003) that, despite the initial plausibility of each of these structures, only two – ‘Demonstrative XP Adjunction’ and ‘IP Adjunction via Movement’ – are actually made available by the grammar of Hindi. That is, Bhatt argues that properties of Hindi correlatives indicate that they cannot be derived by ‘Non-Local IP Adjunction’.21 On the other hand, Bhatt does find evidence that supports the existence in Hindi of both the ‘Demonstrative XP Adjunction’ structure and the ‘IP Adjunction via Movement’ structure. One might naturally ask then, which of the above structures might stand a chance as proper analysis of the Tibetan Correlative. Interestingly, we will find – based on tests similar to the ones employed in Bhatt (2003) – that all three of these structures are made available by the syntax of Tibetan. This stands as a rather intriguing difference between Tibetan and Hindi, and I will suggest that a certain property of Tibetan morphology may provide a basis for it. First, however, I will defend the claim that the grammar of Tibetan makes use of all three of the above structures.

4.1

The availability of ‘demonstrative XP adjunction’

Bhatt (2003) argues that the ‘Demonstrative XP Adjunction’ structure in (32) is required to account for Hindi sentences such as the following.

 Seth Cable

(34)

Rahul a:jkal [ [ [ jo kita:b Saira-ne likh-i: ]1 vo1 ] aur [ [ jo cartoon Rahul nowadays rel book Saira-erg write-perf that and rel cartoon Shyam-ne bana:-ya: ]2 vo2 ] ] parh raha: hai. Shyam-erg make-perf that read prog be ‘Nowadays, Rahul is reading the book that Saira wrote and the cartoon that Shyam made.’ (Lit. ‘Nowadays, Rahul is reading what book Saira wrote, that and what cartoon Shyam made, that.) In sentence (34), it appears that the coordinator aur ‘and’ is joining together two sequences consisting of a Correlative CP and its associated Correlative DP. Assuming that coordination can only apply to constituents, coordination structures such as the one above suggest that the Correlative CP and its associated Correlative DP can combine to form a single constituent. Consequently, the ‘Demonstrative XP Adjunction’ structure in (32) must be one derivable by the grammar of Hindi. The sentence in (35) demonstrates that identical correlative coordination structures are possible for Tibetan as well. (35)

Nga-s [DP [DP [CP khyodra-s gyag gare njos yod na ]1 de1 ] dang I-erg you-erg yak what buy aux if that and [DP [CP Ø bya gare rkus yod na ]2 de2 ] ] bsad pa yin. pro-erg chicken what steal aux if that kill perf aux ‘I killed whatever yak you bought and whatever chicken you stole.’

In sentence (35), the coordinating element dang seems to be joining together two sequences consisting of a Correlative CP and its associated Correlative DP. The same reasoning outlined above for Hindi would demand that the ‘Demonstrative XP Adjunction’ structure is one made available by the syntax of Tibetan. Additional evidence supporting the existence in Tibetan of the ‘Demonstrative XP Adjunction’ structure comes from the ability for an ergative subject to appear before the Correlative CP. This word order can be seen in sentences such as (20c), (22c) and (35) above, as well as in simpler sentences, such as (36) below. (36) Nga-s [kyodrang gi su la ’o gyed na] ( de la ) ’o gyed gi yin. I-erg you erg who dat kiss give if that dat kiss give non.past aux ‘I’ll kiss whoever you kiss.’ One might initially guess this word order to be the result of scrambling the ergative subject to the left periphery of the clause. However, Tibetan does not generally allow an ergative subject to scramble, not even clause-internally. (37) a. [ Norbu-s bsam pa la ], nga-s kho-’i mogmog bzas pa red. Norbu-erg think perf dat I-erg he-gen momo eat perf aux ‘According to Norbu, I ate his momo.’ (Lit. ‘To Norbu’s thoughts, I ate his momo.’)



The syntax of the Tibetan correlative 

b. * Nga-s [ Norbu-s bsam pa la ] kho-’i mogmog bzas pa red. I-erg Norbu-erg think perf dat he-gen momo eat perf aux The impossibility of (37b) is not due to a general inability for the subject to scramble to this position; absolutive subjects may scramble to and beyond the left-periphery of their clause. (38) a. [ Norbu-s bsam pa la ], nga ril pa red. Norbu-erg think perf dat I fall perf aux ‘According to Norbu, I fell.’ b. Nga [ Norbu-s bsam pa la ] ril pa red. I Norbu-erg think perf dat fall perf aux ‘According to Norbu, I fell.’ Thus, sentences such as (36) indicate that the Correlative CP must be able to Merge at a position lower than Spec IP. 22 The structure in (32) provides such a low position for the Correlative CP. 23

4.2

The availability of ‘IP adjunction via movement’

Bhatt (2003) argues that certain ‘reconstruction effects’ provide evidence for the availability within Hindi of the ‘IP Adjunction via Movement’ structure. Bhatt observes that if a quantificational DP c-commands the Correlative DP, then that quantificational DP can bind a pronoun within the Correlative CP. The following sentence illustrates. (39) [ jis larke-ko vo1 pasand kar-ti: hai ]2 [ har larki ]1 [ us larke-ko ]2 rel boy-dat that like do-hab.f be every girl that boy-dat buddhima:n samajh-ti: hai intelligent consider-hab.f be ‘Every girl considers the boy who she likes to be intelligent.’ (Lit. ‘What boy she likes, every girl considers that boy intelligent.’) Assuming that binding can only be established under c-command, and that Hindi does not allow WCO configurations, the binding relationship in (39) would not be predicted by the ‘(Non-Local) IP Adjunction’ structure. It is, however, consistent with the ‘IP Adjunction via Movement’ structure. Under this latter analysis, a trace-copy of the Correlative CP in (39) can be found within the Correlative DP. By interpreting the lower copy of the Correlative CP at LF, the observed binding relationship in (39) may be established. An argument of exactly this form can also be made regarding the Tibetan Correlative. 24 Just as for Hindi, if a quantificational DP in a Tibetan sentence c-commands the Correlative DP, then that quantificational DP can bind a pronoun appearing within the Correlative CP. Sentence (40) illustrates.

 Seth Cable

(40) [ Ø1 mogmog gare mthong na ]2 [ mi tshangma-s ]1 de2 njo gi red. 25 pro momo what see if man every-erg that buy non.past aux ‘Every man buys whatever momos he sees.’ Again, the reasoning laid out above for Hindi would also require that the ‘IP Adjunction via Movement’ structure be derivable by the grammar of Tibetan. That WCO configurations are not tolerated by the grammar of Tibetan may be seen from sentences such as (41). The null possessive pronoun in (41) cannot be interpreted as being bound by the quantificational DP pugu tshangma ‘every boy’. (41)

4.3

Ø Amalags pugu tshangma dgapo gyed gi red. pro mother boy every love give non.past aux ‘His2 mother loves [ every boy ]1.’ *‘[ Every boy ]1 is such that their1 mother loves them1.’

The availability of ‘(non-local) IP adjunction’

We have seen evidence that the correlatives of both Tibetan and Hindi may have either of the structures in (32) and (33). Thus far, then, the two languages seem to agree on the structures available to their correlative constructions. In this section, however, I will present evidence that these languages differ in whether their correlatives may receive the structure in (31), the ‘(Non-Local) IP Adjunction’ structure. First, let us consider the Hindi correlative. Bhatt (2003) argues that, unlike the structures in (32) and (33), Hindi grammar cannot assign its correlatives the ‘(NonLocal) IP Adjunction’ structure in (31). The strongest evidence to this effect is the fact that the relationship between the Correlative CP and the Correlative DP is islandsensitive. When the Correlative CP is adjoined to the matrix clause, the Correlative DP cannot be contained within a relative clause. This is illustrated in (42) below. (42) * [ jo vaha: rah-ta: hai ]1 mujh-ko [ vo kaha:ni [ jo Arundhati-ne rel there stay-hab be I-dat that story rel Arundhati-erg [us]1-ke-baare-me likh-ii ] ] pasand hai. that-about write-perf like be ‘I like the story that Arundhati wrote about who lives there.’ (Lit. ‘Who lives there, I like the story that Arundhati wrote about him.’) As Bhatt (2003) observes, the unacceptability of (42) cannot be due to a general proscription against the relation between the Correlative CP and the Correlative DP crossing a finite CP boundary. Sentences such as (43) demonstrate that the relation can, in principle, be long-distance.



The syntax of the Tibetan correlative 

(43) [ jo larki: TV-par ga: rah-i: hai ]1 Sita soch-ti: hai ki [ vo ]1 sundar hai. rel girl TV-on sing prog be Sita think be that that beautiful be ‘Sita thinks that the girl on TV is beautiful.’ (Lit. ‘Which girl is on TV, Sita thinks that she is beautiful.’) Furthermore, sentences such as that in (44) demonstrate that the relation between the Correlative CP and the Correlative DP is also sensitive to the Coordinate Structure Constraint. 26 (44) * [ jo kita:b Saira-ne likh-i: ]1 Rahul a:jkal [ [ vo1 ] aur [ [ jo cartoon rel book Saira-erg write Rahul nowadays that and rel cartoon Shyam-ne bana:-ya: ]2 vo2 ] ] parh raha: hai. Shyam-erg make-perf that read prog be ‘Nowadays, Rahul is reading the book that Saira wrote and the cartoon that Shyam made.’ (Lit. ‘What book Saira wrote, Rahul nowadays is reading that and what cartoon Shyam made, that.’) As Bhatt (2003) rightly observes, if ‘(Non-Local) IP Adjunction’ were a structure available to Hindi grammar, then the relation between the Correlative CP and the Correlative DP would be predicted not to be island sensitive. Sentences such as (42) and (44) could easily be derived, simply by initially Merging the Correlative CP as an adjunct to the IP. On the other hand, if the structure in (31) were not available, then all Correlative CPs would have to be initially Merged at the position of the Correlative DP. The observed island-sensitivity of the relation between the Correlative CP and the Correlative DP would then be a necessary consequence. Further evidence against the availability of the ‘(Non-Local) IP Adjunction’ structure comes from the existence of certain Condition C effects. Bhatt observes that if a pronoun c-commands the Correlative DP, then that pronoun cannot corefer with a name contained within the Correlative CP. This is illustrated in (45) below. (45) * [ jo larkii Sita-ko1 pyaar kar-tii hai ]2 [ us-ne ]1 [ us-ko ]2 rel girl Sita-acc love do-hab be that-erg that-acc thukraa di-yaa. reject give-perf * ‘She1 rejected the girl that loved Sita1.’ (Lit. ‘[ Which girl loves Sita1 ]2, she1 rejected her2.’) Again, if ‘(Non-Local) IP Adjunction’ were a structure available for Hindi correlatives, then the impossibility of (45) could not be explained. However, if that analysis were not available, and all Correlative CPs had to be initially Merged at the position of the Correlative DP, then the impossibility of structures like (45) would follow from standard assumptions regarding the evaluation of Principle C (Lebeaux 1998).

 Seth Cable

We find, then, that there is good evidence against the Hindi correlative being assigned the ‘(Non-Local) IP Adjunction’ structure (31). Interestingly, however, the reasoning detailed above demands just as strongly that the structure in (31) is one available to the Tibetan Correlative. Consider first that – unlike the Hindi correlative – the relationship between the Correlative CP and the Correlative DP in Tibetan is not island sensitive. As (46) illustrates, even if the Correlative CP is adjoined to the matrix IP, the Correlative DP may be contained within a relative clause. (46) [ Khyodra-s mogmog gare njos na ]1 nga-s you-erg momo what buy if I-erg [ de1 bzo mkhan gyi bsad mkhan ] de ngozhi gi yod that make agnt gen kill agnt the know non.past aux ‘Whatever momos you bought, I know the person who killed the person who made them.’ (Lit. ‘Whatever momos you bought, I know their maker’s murderer.’) The sentence in (46) illustrates Tibetan’s prenominal relative clause. Tibetan can also perform clausal modification of nouns via a ‘head-internal’ or ‘circumfixal’ relative clause. As the sentence (47) illustrates, the Correlative CP-Correlative DP relation is also free to cross into these latter relative clauses. (47) [ Khyodra-s magyan su thug yod na ]1 nga-s you-erg cook who meet aux if I-erg [ kho-s1 mogmog bzos pa ] de bzas pa yin. he-erg momo make perf the eat perf aux ‘Whatever cook you’ve met, I’ve eaten the momos that he has made.’ Furthermore, sentences such as (48) show that the relation between the Correlative CP and the Correlative DP can cross into verbal adjuncts. (48) [ Khyodra-s mogmog gare bzos yod na ]1 nga [ de1 bzas tsang ] you-erg momo what make aux if I that eat because na byung. sick become ‘I got sick because I ate whatever momos you made.’ (Lit. ‘You make what momos, I get sick because I ate them.’) Recall that the unacceptability of sentence (44) demonstrates the Hindi correlative to be subject to the Coordinate Structure Constraint. Sentences such as (49) equally well demonstrate that the Tibetan Correlative is not subject to the CSC.



The syntax of the Tibetan correlative 

(49) [ Khyodra-s mogmog gare bzos yod na ]1 nga-s [ de1 dang sha’gyas ] you-erg momo what make aux if I-erg that and shamje bzas pa yin. eat perf aux ‘I ate whatever momo you made and shamje.’ (Lit. ‘Whatever momo you made, I ate that and shamje.’) The acceptability of sentences (46) – (49) demonstrates that the ‘(Non-Local) IP Adjunction’ structure must be one available to Tibetan speakers. If only the ‘IP Adjunction via Movement’ structure were available, then sentences (46) – (49) would be as unacceptable as their Hindi counterparts, since they all would require movement of the Correlative CP to cross an island. However, if the ‘(Non-Local) IP Adjunction’ structure in (31) were available in Tibetan, then all the sentences above could be derived by initial Merger of the Correlative CP to the matrix IP. 27 Supporting this conclusion is the absence in Tibetan of the Principle C effects noted above for Hindi. (50) [ Norbu2 mogmog gare mthong na ]1 kho-s2 de1 njo gi red Norbu momo what sees if he-erg that buy non.past aux ‘Norbu2 buys whatever momos he2 sees.’ (Lit. ‘If Norbu2 sees what momo, he2 buys that.’) Unlike what has been observed for Hindi, a pronoun in Tibetan may c-command the Correlative DP and co-refer with a name contained within the Correlative CP. Again, if only the ‘IP Adjunction via Movement’ structure were available, then sentence (50) would be ruled out as a Principle C violation, just as its Hindi counterpart presumably is. Of course, this sentence could be correctly derived if speakers had recourse to the ‘(Non-Local) IP Adjunction Structure’. By initially Merging the Correlative CP to the matrix IP, sentence (50) could be derived without any Principle C effect resulting. In summary, we have observed evidence that the Tibetan Correlative may be derived via any of the three structures illustrated in (31) – (33).28 This is an interesting difference between the Tibetan Correlative and the correlative construction of Hindi, which can only admit of the structures in (32) and (33). In the next section, I offer an ‘educated guess’ regarding one possible source for this micro-parametric difference.

5. The locality of merge and the locality of agree As we have seen, ‘(Non-Local) IP Adjunction’ cannot be a structure available to the Hindi grammar. However, the observed unavailability of this structure immediately begs the question ‘Why?’ Bhatt (2003) argues that it can be seen as a consequence of the following condition, which he names the Condition on Local Merge.

 Seth Cable

(51) Condition on Local Merge (CLM) The structure-building operation of Merge must apply in as local a manner as possible. In brief, Bhatt reasons as follows. The coordination facts presented in Section 4.1 demonstrate that the Correlative CP can be initially Merged in the projection of the Correlative DP. Moreover, initial Merge of the Correlative CP at IP is ‘less local’ than initial Merge of the Correlative CP within the Correlative DP, as the initial relationship between these two constituents is correspondingly ‘less local’. Therefore, the CLM entails that, when given the choice between initial Merger of the Correlative CP to the Correlative DP and initial Merger of the Correlative CP to the IP, Hindi must always choose the former over the latter. Thus, of the three structures in (31) – (33), only ‘Demonstrative XP Adjunction’ and ‘IP Adjunction via Movement’ are ones available to Hindi grammar. Bhatt (2003) adduces additional, typological evidence in support of the CLM. Note that the reasoning above entails that the ‘(Non-Local) IP Adjunction’ structure should be available to any language in which the Correlative CP cannot adjoin directly to the Correlative DP. Bhatt submits that Bulgarian is a language that witnesses the truth of this prediction. In addition, Bhatt (2003) argues that properties of Clitic Left Dislocation in languages such as Greek provide further evidence supporting the CLM. It appears, then, that the Hindi correlative – and perhaps dislocation structures more generally – is governed by the CLM. Recall now, however, that we have also seen evidence that all three of the structures in (31) – (33) are derivable in the grammar of Tibetan. The fact that Tibetan grammar generates both the ‘(Non-Local) IP Adjunction’ structure and the ‘IP Adjunction via Movement’ structure presents a prima facie challenge to the universality of the CLM. If the CLM were active in Tibetan, then – all else equal – the availability of the ‘IP Adjunction via Movement’ structure should block the availability of the ‘(Non-Local) IP Adjunction’ structure, just as it allegedly does for Hindi. In the face of these facts, one might conclude that the CLM is simply not active in Tibetan. Indeed, the activity of the CLM may be exactly what distinguishes Tibetan, which has access to all the structures in (31) – (33), from Hindi, which has access only to (32) and (33). The proposal that languages may differ on the activity of the CLM, seems a plausible one, but such a parameterization of the CLM raises its own questions. Is variation in the CLM a linguistic primitive, or does a language’s sensitivity to the CLM follow from some other property it might have? For all the usual reasons, it would be best not to leave sensitivity to the CLM as a wholly independent, primitive property of a language. What other point of variation then, might the activity of the CLM be tied to? One possibility worth exploring is that the activity of the CLM is tied to the presence of Agreement between the adjoined phrase and the resumptive DP within the matrix clause. Examples such as the following demonstrate that, within Hindi, the Correlative DP and the operator within the Correlative CP agree in number.



The syntax of the Tibetan correlative 

(52) a. [ jo laRkii khaRii hai ]1 vo1 lambii hai. rel girl standing be that tall be ‘The girl who is standing is tall.’ (Lit. ‘Which girl is standing, she is tall’.)

(Srivastav 1991; p. 646)

b. [ jo laRkiyaa khaRii hai ]1 ve1 lambii hai. rel girls standing be those tall be ‘The girls who are standing are tall.’ (Lit. ‘Which girls are standing, they are tall.’) (Srivastav 1991; p. 663) Furthermore, it seems like similar phi-agreement must also hold between the adjoined phrase and the resumptive, matrix DP in the other languages Bhatt (2003) claims witness the activity of the CLM. In Tibetan, however, the demonstrative elements acting as Correlative DPs do not seem to carry any phi-features – aside from person, which for logical reasons is invariably third. Tibetan demonstratives do not indicate the gender of their antecedents. Moreover, although Tibetan does have a kind of plural marker, tsho, that may appear on the demonstrative, this plural marking does not seem to bear an ‘agreement’ or ‘phi-like’ relation with the plural marking of the operator within the Correlative CP. The following sentences illustrate. (53) a. [ Khyodra-s [ bumo su tsho ] la ’o gyed na ]1 nga-s [ de tsho ]1 you-erg girl who pl dat kiss give if I-erg that pl la ’o gyed gi yin. dat kiss give non.past aux ‘I will kiss whatever girls you kiss.’ b. [ Khyodra-s [ bumo su ] la ’o gyed na ]1 nga-s [ de tsho ]1 la you-erg girl who dat kiss give if I-erg that pl dat ’o gyed gi yin. kiss give non.past aux ‘I will kiss whatever girls you kiss.’ c. [ Khyodra-s [ bumo su tsho ] la ’o gyed na ]1 nga-s de 1 la you-erg girl who pl dat kiss give if I-erg that dat ’o gyed gi yin. kiss give non.past aux ‘I will kiss whatever girls you kiss.’ d. [ Khyodra-s [ bumo su ] la ’o gyed na ]1 nga-s de1 la you-erg girl who dat kiss give if I-erg that dat ’o gyed gi yin. kiss give non.past aux ‘I will kiss whatever girls you kiss.’ Although, as (53a) illustrates, both the Correlative DP and the operator within the Correlative CP may be marked with tsho, this marker may appear on one without it

 Seth Cable

appearing on the other, as shown in sentences (53b) and (53c). This is consonant with a wider optionality regarding the appearance of tsho. Although there do appear to be syntactic contexts in which tsho is required for a plural interpretation, it is at least clear from sentence (53d) that the Tibetan Correlative is not such a context. The pattern in (53) suggests that Tibetan does not require phi-agreement between the features of the Correlative DP and the operator within the Correlative CP. Although our pool of languages here is indeed rather small, the following generalization seems to come into view: the CLM rules out high initial Merger of a Correlative CP only if the language requires phi-agreement between the Correlative DP and the operator within the Correlative CP. Given the well-known sensitivity of phi-agreement to strict locality conditions, this is a rather intriguing generalization. It would suggest that the locality condition embodied by the CLM may actually derive from the more basic and general locality conditions that govern Agreement. If such proves to be the case, the result may be that Tibetan differs from Hindi in the structures it can assign to its correlative ultimately because of the morpho-syntactic difference illustrated in (52) and (53).

6. Conclusion We have seen that Tibetan sentences such as (1b) employ a construction that may neither be fruitfully analyzed as a free relative nor a conditional. The subordinate CP in such sentences does not have the external syntax of its component wh-operator, which renders implausible the notion that such constructions are argumental free relatives. Moreover, the construction possesses a definite, ‘ignorance’ reading, which entails that it cannot be analyzed simply as an instance of donkey anaphora within a conditional. Given the definition of a correlative construction assumed in (5), it was concluded that such constructions are best labeled as ‘correlatives’. The Tibetan Correlative was then compared to the Hindi Correlative in terms of the syntactic relation holding between the Correlative CP and the Correlative DP. It was found that the Tibetan Correlative may be assigned all three of the possible structural analyses in (31) – (33). This distinguishes it from the correlative of Hindi, which permits only the analyses in (32) and (33). The ability for the Tibetan Correlative to be assigned the structure in (31) was suggested to be due to the inactivity of the CLM within the language, a property which was itself suggested to follow from the lack of phi-agreement in Tibetan between the Correlative DP and the wh-operator within the Correlative CP.29

Acknowledgements I wish to thank Chashing Bugang for all the assistance he has given me in my study of his language. Beyond his boundless patience in evaluating and translating my often convoluted exam-



The syntax of the Tibetan correlative  ple sentences, he has been a first-rate instructor in the language. His linguistic expertise and insight often guided me towards data and patterns I would otherwise be ignorant of. In addition, a great many errors in transliteration have been caught by his watchful eye. Special thanks are also due to Anikó Lipták, Alec Marantz, David Pesetsky, Norvin Richards, Ken Wexler, and one anonymous reviewer for their many helpful comments upon earlier drafts of this paper. All errors empirical or conceptual remain my own responsibility.

References Arsenijević, Boban. This volume. {Relative {conditional {correlative}} sentences. Bhatt, Rajesh. 2003. Locality in Correlatives. Natural Language and Linguistic Theory 21. 485–541. Bhatt, Rajesh & Roumyana Pancheva. 2006. Conditionals. In Martin Everaert, Henk van Riemsdijk, Rob Goedemans & Bart Hollebrandse. The Blackwell Companion to Syntax I, 638–687. Oxford: Blackwell. Bittner, Maria. 2001. Topical Referents for Individuals and Possibilities. In Rachel Hastings, Brendan Jackson & Zsófia Zvolensky (eds), Proceedings from SALT XI, 36–55. Ithaca, NY: CLC Publications. Branchini, Chiara & Caterina Donati. This volume. Relatively different: Italian Sign Language relative clauses in a typological perspective. Cheng, Lisa L.-S. & C.-T. James Huang. 1996. Two Types of Donkey Sentences. Natural Language Semantics 4. 121–163. Citko, Barbara. This volume. What don’t wh-questions, free relatives, and correlatives have in common? Dayal, Veneeta. 1995. Quantification in Correlatives. In Emmon Bach, Eloise Jelinek, Angelika Kratzer & Barbara H. Partee (eds), Quantification in Natural Languages, 179–205. Dordrecht: Kluwer Academic Publishers. Dayal, Veneeta. 1996. Locality in Wh-Quantification: Questions and Relative Clauses in Hindi. Dordrecht: Kluwer Academic Publishers. DeLancey, Scott. 2003a. Classical Tibetan. In Graham Thurgood & Randy J. LaPolla (eds), The Sino-Tibetan Languages, 255–269. London: Routledge. DeLancey, Scott. 2003b. Lhasa Tibetan. In Graham Thurgood & Randy J. LaPolla (eds), The Sino-Tibetan Languages, 270–288. London: Routledge. Downing, Bruce. 1973. Correlative relative clauses in universal grammar. Minnesota Working Papers in Linguistics and Philosophy of Language 2. 1–17. Minneapolis, Minnesota: University of Minnesota. von Fintel, Kai. 2000. Whatever. In Proceedings of SALT 10. Ithaca, NY: CLC Publications. Garrett, Edward. 2001. Evidentiality and Assertion in Tibetan. Doctoral dissertation. UCLA. Groos, Anneke & Henk van Riemsdijk. 1981. Matching Effects in Free Relatives: A Parameter of Core Grammar. In Adriana Belletit, Luciana Brandi & Luigi Rizzi (eds), 171–216. Theory of Markedness in Generative Grammar: Proceedings of the IVth GLOW Conference, Pisa: Scuola Normale Superiore di Pisa. Hale, Kenneth. 1976. The Adjoined Relative Clause in Australia. In Robert M.W Dixon (ed) Grammatical Categories in Australian Languages [Linguistic Series No. 22. Australian Institute of Aboriginal Studies. Canberra], 78–105. New Jersey: Humanities Press Inc.

 Seth Cable Iatridou, Sabine. 1994. Clitics and Island Effects. In Roumyana Izvorski & Victoria A. Tredinnick (eds), University of Pennsylvania Working Papers in Linguistics 2, 11–30. Philadelphia: University of Pennsylvania Linguistics Club. Philadelphia. Izvorski, Roumyana. 1997. The Syntax and Semantics of Correlative Proforms. In Kiyomi Kusumoto (ed), Procedeings of NELS 26, 133–147. Amherst, MA: GLSA. Jacobson, Pauline. 1995. On the Quantificational Force of English Free Relatives. In Emmon Bach, Eloise Jelinek, Angelika Kratzer & Barbara H. Partee (eds), Quantification in Natural Languages, 179–205. Dordrecht: Kluwer Academic Publishers. Keenan, Edward. 1985. Relative Clauses. In Timothy Shopen (ed), Language Typology and Syntactic Description, Volume 2, 141–170. Cambridge: Cambridge University Press. Lebeaux, David. 1998. Where Does the Binding Theory Apply? NEC Research Institute, Inc. Technical Report 2–092–0079–91020. Lipták, Anikó. 2005. Correlative Topicalization. Manuscript. University of Leiden. Mahajan, Anoop. 1990. The A/A-bar Distinction and Movement Theory. PhD dissertation. MIT. Rebuschi, Georges. This volume. Basque correlatives and their kin in the history of Northern Basque. Srivastav, Veneeta. 1991. The Syntax and Semantics of Correlatives. Natural Language and Linguistic Theory 9. 637–686. Tredinnick, Victoria A. 2005. On the Semantics of Free Relatives with –Ever. PhD dissertation. University of Pennsylvania. Wylie, Turrell. 1959. A Standard System of Tibetan Transcription. Harvard Journal of Asiatic Studies 1959, 261–267.

Endnotes 1. All Tibetan data in this paper are written following the conventions of the Wylie transliteration system (Wylie 1959). Although most polysyllabic words in Tibetan are analyzable to some degree, I only indicate the structure of morphologically complex words where I consider it relevant. 2. Throughout this paper, I use the following abbreviations in my glosses of example sentences: abl ‘ablative’, acc ‘accusative’, agnt ‘agentive nominalizer’, aux ‘auxiliary’, dat ‘dative’, erg ‘ergative’, f ‘feminine’, fut ‘future’, gen ‘genitive’, hab ‘habitual’, m ‘masculine’, non.past ‘non-past’, perf ‘perfective’, pl ‘plural’, prog ‘progressive’, rel ‘relative operator’, sg ‘singular’, top ‘topic marker’. 3. Modern Tibetan dialects can sometimes diverge sharply from one another, and it is not known to me how these dialects might differ in regards to the presence or structure of correlatives. 4. These data were originally discovered by Justin Fitzpatrick (p.c.). 5. See DeLancey (2003b) for a list of works pertaining to the semantics of the Tibetan auxiliary system. See also Garrett (2001) for an extended discussion of the various semantic dimensions that the system encodes. 6. This use of the term seems to be essentially identical to that found in Keenan (1985), Srivastav (1991), Dayal (1996), Izvorski (1997), Bhatt (2003) and several others. 7. Of course, given the ‘literal translation’ I offer in (7), one might wonder whether these constructions should not be treated as a species of conditional. I will later provide some reasons to



The syntax of the Tibetan correlative  avoid a ‘simple’ conditional analysis. Ultimately, however, the distinction between correlatives and conditionals is a slippery one (Hale 1976), especially given the formal semantics each typically receives (Bittner 2001). Moreover, it is quite likely that conditionals are themselves nothing more than a subspecies of correlative (Bhatt & Pancheva 2006, Arsenijević this volume). Thus, the question ‘Are these correlatives or conditionals?’ is to a certain extent meaningless. Again, though, I will later review some important differences between these structures and English conditionals. 8. Note, also, that the interrogative phrases in these examples generally occupy the same insitu position that they would occupy in the corresponding wh-question (though such interrogative phrases can optionally undergo scrambling to the left-periphery in both correlatives and wh-questions). 9. Tibetan does not have a single lexical item corresponding to English ‘why’, but rather uses a complex locution translatable as ‘what reason is there for’. (i)

Khyodrang ’gro dgos don gare red? you go should reason what is ‘Why did you leave?’ (Lit. ‘What reason is there for your leaving?’)

10. A concrete example of such a context is the following: Your friend buys a yak and brings it home. He asks you to slaughter the yak. You do so and wish to report to him that you did what he asked. 11. Similarly, Citko (this volume) shows that correlatives in Polish are quite distinct from free relatives. 12. Note that the element de in Tibetan can function both as a free-standing demonstrative and as an adnominal determiner. 13. It is my guess that this apparent ability for pro to escape the requirement that it be dative marked is actually due to some kind of null pro-form for PPs in Tibetan. In sentences like (19b), for example, the pro probably has the category of a dative PP. 14. The adnominal markers that cannot take pro as complement (la ‘dative’, nas ‘ablative’, gi ‘ergative/genitive’) are in the standard, descriptive literature for Tibetan often referred to as ‘case postclitics’, and those that can take pro as complement are referred to as ‘relator nouns’ (see DeLancey 2003b). Despite the terminology, it is quite difficult to pin down any wider distributional difference between these elements, and the nominal status of the so-called ‘relator nouns’ is certainly up for debate. 15. Other adnominals which have been tested include gi ‘ergative’, gi ‘genitive’, dkyil ‘across/ through/middle’, brgyud ‘along’, and dra-po ‘like/similar-to’. I reiterate that these elements might ultimately differ in their syntactic category. 16. There do, of course, exist analyses of English-style free relatives under which they are hypothesized to be headed by a null pronoun (Groos & van Riemsdijk 1981). If such a view is adopted, then an analysis of the Correlative CP as a free relative would still predict the data above. There is, however, some cross-linguistic evidence against such an analysis of ‘free relatives’. In the case of Tibetan, this analysis would immediately raise the question of why only null pronouns may be modified by CPs of the relevant form. Adnominal clausal modification is typically accomplished in Tibetan by means of either a nominalized clause or a head-internal relative (see the examples in Section 4.3). Overt NPs certainly cannot be modified by CPs of the form of the Correlative CP in the examples above; why, then, should a null pronoun allow such modification?

 Seth Cable 17. In Classical Tibetan, this particle was also used as a subordinator meaning ‘when’ (DeLancey 2003a). In Modern Lhasa Tibetan, this particle can additionally mark certain ‘irrealis’ or ‘subjunctive’ complements. (i) [ Sgrolma ’gro na ] bsam byung. Droma go na think aux ‘I want Droma to go.’ Thus, although na is glossed throughout this paper as ‘if ’, a more apt analysis may be as a kind of irrealis or subjunctive marker. 18. Interestingly, Cheng & Huang (1996) argue that a construction in Chinese that is superficially similar to our ‘Tibetan Correlative’ is actually a conditional, and not a correlative. It is important to note, however, that the most convincing of their arguments (i.e. the fact that grammatical aspect within the Chinese construction cannot be perfective) does not carry over for Tibetan. Moreover, as I remarked in footnote 7, the distinction between correlatives and conditionals is a rather tenuous and perhaps ultimately unreal one. 19. Branchini & Donati (this volume) also demonstrate that correlatives in Italian Sign Language are a distinct construction from conditionals, despite some initial similarities. Likewise, Rebuschi (this volume) demonstrates that Basque correlatives, despite their similarities to Basque conditionals, are a separate construction. Generally speaking, Rebuschi’s discussion of Basque correlatives reveals some rather interesting parallels between the correlatives of Basque and those found in Tibetan. Besides their superficial similarity to conditionals, correlatives in both languages are also rather similar to subordinate wh-questions, and differ from them only in the morphological form of the verb. Furthermore, the correlatives of both Basque and Tibetan are strikingly different from the languages’ adnominal relative clauses, in that they obligatorily contain wh-operators, while such operators are obligatorily absent from true, adnominal relative clauses (cf. Rebuschi, this volume, Section 4.3). 20. Bhatt (2003) notes that this structure was first proposed by Srivastav (1991) and Dayal (1996). 21. Strictly speaking, this is only true of what Bhatt calls ‘simple correlatives’. Bhatt (2003) argues that some Hindi correlatives – those containing multiple operators – are only derivable by NonLocal IP Adjunction. As I have not yet been able to investigate the existence of such ‘multi-operator’ correlatives in Tibetan, let alone their structure, I will ignore them in the present discussion. 22. Note that by this reasoning, any material typically occurring to the left of an ergative subject can also occur to the left of a Correlative CP. To my knowledge, this is indeed the case. 23. Of course, another possibility is that the Correlative CP is Merged to intermediate projections of I, such as the sister to the subject. The possibility of such a structure does not entirely undermine our arguments for the existence of the ‘Demonstrative XP Adjunction’ structure in Tibetan. The ‘Demonstrative XP Adjunction’ structure may still be required for the proper analysis of sentences such as (35). 24. Furthermore, see Branchini & Donati (this volume) for somewhat similar arguments that the ‘IP Adjunction via Movement’ structure is available in Italian Sign Language. 25. The use of pro in this sentence appears to be crucial; speakers reject the sentence if pro is replaced with the overt pronoun kho. However, this seems related to a general distaste for cataphora with overt pronouns. 26. Sentence (44) also demonstrates that the island-sensitivity witnessed in sentence (42) cannot be the result of ‘(Non-Local) IP Adjunction’ targeting the minimal IP containing the Cor-



The syntax of the Tibetan correlative  relative DP, such as in the analysis Lipták (2005) provides for Hungarian correlatives, or the analaysis Iatridou (1994) puts forth for Clitic Left Dislocation structures in Greek and Italian. 27. Norvin Richards (p.c.) points out that the proposals regarding the sentences in (40) and (46) – (49) predict that one cannot in Tibetan ‘reconstruct for binding’ across an island. Although more study is needed, preliminary tests suggest this prediction is accurate. Compare the sentence below to that in (46). (i) ?? [ Ø1 mogmog gare nyos na ]2 [mi tsangma-s ]1 pro momo what buy if man every-erg [ de2 bzo mkhan gyi bsad mhkan ] de ngozhi gi red. that make agnt gen kill agnt that know non.past aux ‘Whatever momos he1 buys, [every man]1 knows the person who killed the person who made them.’ The sentence in (i) was reported to be rather awkward. If this judgment turns out to be stable, this would support the proposal that the reconstruction for binding in (40) relies upon a structure incompatible with that required for the ‘island-crossing’ sentences in (46) – (49). 28. In this context, it is important to note that our conclusion here is not that the available evidence is conflicting and leads to a contradictory result. Indeed, the available data in no way conflict with one another. The binding facts from Section 4.2 show merely that speakers can parse a surface string containing a ‘Tibetan Correlative’ in such a way that the Correlative CP initially occupies a lower position. Meanwhile, the various facts in Section 4.3 show merely that such a parse, while possible, is not obligatory; speakers could also parse such strings in such a way that the Correlative CP doesn’t initially occupy a lower position. Thus, since such surface strings could be parsed in either way, it follows that the grammar of Tibetan allows either structure. Importantly, however, since we (naturally) assume that speakers cannot simultaneously assign two parses to a single surface string, our account makes a clear prediction regarding how the phenomena noted in Sections 4.2 and 4.3 should interact with one another. As noted by David Pesetsky (p.c.), the proposals regarding the sentences in (40) and (50) predict that the following binding configuration should not be possible in Tibetan, since reconstructing the Correlative CP to the position of the Correlative DP for binding would induce a Principle C violation.

(i) [CP … pro1 … name2 … ]3 QuantDP1 … pronoun2 … Correlative-DP3 …

Unfortunately, however, I have not yet been able to construct sentences of Tibetan that would test this prediction. 29. Anikó Lipták (p.c.) and an anonymous reviewer both observe that correlatives of Hungarian might seem to challenge my proposal concerning the relevance of phi-agreement to whether a language permits high initial Merger of Correlative CPs. As both observe, the correlatives of Hungarian allow high initial Merger of the Correlative CP (Lipták 2005), despite the fact that they show Hindi-style phi-agreement between the Correlative DP and the wh-operator. (i) Aki most jött, az a barátom. rel.who.sg now came.sg, that.sg my.friend.sg ‘Who came just now, that is my friend.’

 Seth Cable (ii) Akik most jöttek, azok a barátaim. rel.who.pl now came.pl, that.pl my.friend.pl ‘The people who came just now, those are my friends.’ On the other hand, Lipták (2005, p.c.) also demonstrates that Hungarian correlatives – unlike either Hindi or Tibetan correlatives – only permit high initial Merger of the Correlative CP, a fact that she attributes to differences between the discourse-functional status of Hungarian correlatives and Hindi(/Tibetan) correlatives. Thus, for independent reasons, Hungarian does not permit its Correlative CPs to be initially Merged as adjuncts to the Correlative DP. With this in mind, recall that my proposed account seeks only to explain the permissibility of high initial Merger of the Correlative CP in languages where it should be ruled out by the CLM. That is, since Hungarian does not allow the Correlative CP to be an adjunct to the Correlative DP, the CLM wouldn’t rule out an initial high Merger of the Correlative CP (cf. Bhatt’s (2003) discussion of Bulgarian). On the other hand, since Tibetan does permit the Correlative CP to be an adjunct to the Correlative DP, then the CLM as stated in Bhatt (2003) would incorrectly rule out initial high Merger of the Correlative CP. My proposal is that the effects of the CLM aren’t found in Tibetan because – unlike the languages where the CLM does hold (e.g. Hindi, Hungarian and Bulgarian) – Tibetan does not require phi-agreement between the Correlative DP and the operator of the Correlative CP. Thus, a counter-exemplifying language would be one which (unlike Hungarian) does allow initial low Merger of the Correlative CP, and which (unlike Hindi) does allow initial high Merger of the Correlative CP, but which (unlike Tibetan) also requires phi-agreement between the Correlative DP and the operator of the Correlative CP.

Adjunction, features and locality in Sanskrit and Hindi/Urdu correlatives Alice Davison Earliest Sanskrit and modern Hindi/Urdu have correlative clauses, in which the dependent relative clause is adjoined to the clause containing the correlate. The two related but temporally distant languages differ in the adjunction relation of relatives: Sanskrit has symmetric adjunction of CP to CP, Hindi on the other hand has asymmetric adjunction of CP to TP. Asymmetric c-command rules out stacked relatives, non-local coindexing and locality condition violations in Hindi/Urdu. Symmetric adjunction in Sanskrit permits stacked relatives, nonlocal coindexing and a much looser anaphoric relation between the two finite clauses. Based on these properties and some novel data, the paper makes the case for a base-adjunction analysis, which reveals how the grammar of Sanskrit and Hindi/Urdu differ in small but significant ways. Keywords: adjunction, base-adjunction, asymmetric c-command, Sanskrit, Hindi/Urdu

1. Correlative clauses in Indic languages 1.1

Correlative clauses in Hindi/Urdu

Hindi/Urdu has correlative clauses, like almost all the modern Indic languages. Correlatives in Hindi/Urdu have been much discussed, for example in Subbarao (1984), Dayal (1996), Mahajan (2000), and Bhatt (2003), among others. I will be assuming, descriptively, that correlatives consist of pairs of sentences. One of them, which I will call the correlative clause, is interpreted as a modifier, and contains a relative determiner and an optional NP. The other clause contains a constituent which is modified by the correlative clause. I will call this constituent the correlate, using the term in Grosu (2002). An illustrative example of a correlative construction is given in (1). In this example, and in the examples to follow, the relative determiner is bolded, and the correlate is given in italics. For ease of understanding, the translations are often literal translations that try to reflect the original structure as much as possible.

 Alice Davison

(1) [aap-nee joo kitaab-eeN kal khariid-iiN] vee khoo ga-ii haiN you-erg rel books-fpl yesterday buy.pf-fpl 3pl get.lost go-pf.f are ‘[Which books you bought yesterday] those are missing.’1/‘The books you bought yesterday are missing.’ (2)

[jis (kisii)-nee bhii mujh-see kitaab udhaar lii] us-nee kitaab rel some-erg even I-from book borrow took 3s-erg book waapas nahiiN dii back not give.pf.f ‘Whoever borrowed my book did not return it.’

The correlative typically has a restrictive definite interpretation, as in (1). With the addition of kooii/kisii ‘some’ or the emphatic bhii ‘even’ to the relative in (2), there is a specific indefinite or generic ‘free’ relative interpretation. Finally, two relative expressions are possible within the same correlative sentence (cf. (3)), each corresponding to a correlate. The association between the two is indicated by indexing, in the following example. (3)

[jis-jis sipaahii(i)-kii jis-jis caurahee(j)-par DyuuTii hoo-tii hai] rel-rel policeman-gen rel-rel intersection-on duty be-impf is [woo(i) is(j)-kaa zimmeedaar hai] 3s.nom 3s-gen responsible is ‘Whichever/the policeman(i) who is on duty at whichever intersection(j), he(i) is responsible for it(j).’

1.1.1 Clause internal and clause external relatives There are two major variants on relative constructions in Hindi/Urdu.2 The first is similar to the correlatives in (1), with the difference that the correlative clause follows the clause with the correlate (4). Right adjoined correlatives (like the one in (4)) allow a non-specific indefinite interpretation not possible with (1)-(3). (4)

unhooN-nee apnee irdgird aisaa pardaa khiiNc liyaa [jis-kee piichee 3pl-erg self around such curtain draw take.pf rel-gen behind vee apnee dard chipaa sak-eeN] 3pl self ’s pain.pl hide be.able-cont.3pl ‘She draw around herself a kind of curtain [behind which she could hide her suffering].’ (Jayekar 1992)

The second variant is composed of a relative adjoined to the right of an NP. This construction is the main way that an appositive reading is expressed cf. (5a), but this structure can also have a restrictive reading (5b). Left-adjoined correlatives cannot have an appositive reading, as shown in (6).



Sanskrit and Hindi/Urdu correlatives 

(5) a.

NP adjoined, appositive relative anu [joo khaRii hai] lambii hai. Anu rel standing is tall is ‘Anu, who is standing, is tall.’ (Dayal 1996, 155)

b.

NP adjoined, restrictive relative mujhee [woo aadmii [joo (*aadmii) siitaa-koo acchaa lag-taa I.dat that man rel man Sita-dat good seem-impf hai]] pasand nahiiN hai is liked not is ‘I do not like [the man [who Sita likes]].’ (Mahajan 2000, 203)

(6) Left-adjoined correlative: no appositive reading *[joo khaRii hai] anu lambii hai rel standing is Anu tall is ‘Anu, who is standing, is tall.’ (Dayal 1996, 155) The NP adjoined relative construction in (5) clearly involves subordination of a finite clause within a matrix clause. The relative clause is adjacent to the NP it modifies, forming a constituent with it. The correlatives in (1)-(4) appear to be peripheral to TP, and adjoined to a clausal category. The relative clause can be linearly separated from the correlate which it modifies (1). While Hindi/Urdu allows clause internal finite clauses (5a,b), it is more usually the case that finite relative clauses are at the left or right periphery of the main clause. 1.1.2 A common semantic translation For those accustomed to NP-headed relatives, correlative structures seem less transparent than NP-headed structures. A clearly subordinated relative clause adjoined to its head very unambiguously singles out in syntactic structure what the modifying clause is, in contrast to the matrix clause. It is also completely clear from surface syntax which NP is modified. But these are parsing considerations not necessarily built into Universal Grammar, which allows various syntactic options besides NP-headed clauses, such as the correlatives discussed here, as well as internally head relative clauses. The semantic interpretation of both NP-adjoined and correlative structures is a conjunction of propositions sharing a common referent: (7) a. b.

NP-adjoined restrictive relative woo [laRkii [joo khaRii hai]] lambii hai that girl rel standing is tall is ‘The girl [who is standing] is tall.’ (Dayal 1996: 152) Translation of (7a) tall’ (σx (girl’(x) & stand’ (x))) (Dayal 1996: 156)3

 Alice Davison

c. d.

Correlative clause version of (7a) [joo laRkii khaRii hai] woo lambii hai rel girl standing is 3s tall is ‘[Which girl is standing] she is tall.’ (Dayal 1996: 188–181) Translation of (7c) tall’ (σx(i) (girl’ (x(i) & stand’ (x(i)))) (Dayal 1996: 191)4

Note that the translations of both the NP-adjoined and the clause-adjoined relatives are very similar in structure.5 The clause-adjoined structure of correlatives is closer syntactically to the form of its translation, which in both cases consists of coordinated propositions. I will say more below about how the relative clause, with its internal NP description syntactically linked with its correlate, and what the role is of relative and other determiners. One could argue that there is no inherent or universal preference for NP-headed structures over clause-adjoined correlative structures. Clause adjunction is a free operation of Merge, not necessarily driven by a need to check features or to discharge theta roles. The placement of modifier clauses are an example of Chomsky’s (2004) merger of one XP to another. So I will assume that there is no a priori reason that languages cannot select either NP-adjoined relatives or clause-adjoined correlatives, or both. I will make the case in this paper that it is not necessary to reduce all relative structures in a language to a single base source, from which other variants are derived by movement. The single-base concept would be more attractive in a theory of syntax which allows D-structure, a well-formed complete syntactic projection from the lexicon in which various conditions must be met. But with the absence of D-structure, a Minimalist theory can derive syntactic objects of many sorts, subject to interface conditions such as Full Interpretation.

1.2

Old Indic correlative clauses

In this paper, I will be comparing Hindi/Urdu and Sanskrit, in which the forms of relative clauses are very similar, particularly the lexical forms used for marking relative and interrogative clauses. Correlative clauses are found in the earliest form of Sanskrit, the language of the religious poetry and prose of the Vedic era, and the later Classical period. Correlatives have a restrictive interpretation (8), and a ‘free relative’ interpretation as in (9). There are also multiple relatives within one clause (10): (8)

Restrictive interpretation of correlatives [yás tán ná véda ] kím Wcu karisyati prel.nom that.acc not know.pres.3s what.int.acc verses.inst do.fut.3s ‘[The one who does not know this] what will (he) do with the Verses?’ (R.V. 1.164.39c, Etter 1985, 1676)



Sanskrit and Hindi/Urdu correlatives 

(9)

‘Free’, universal interpretation of correlatives [yád īm uśmási kár-tave] karat tát what.rel.acc he.acc be.eager.pres.1pl do-inf do.pres.3s that.acc ‘[What(i) we are eager for him to do], he does that(i)’ (R.V. 10.74.6, Hettrich 1988, 273)

(10)

Multiple relatives [yah(i) svabhāvo hi yasya(j) syāt] [tasya(j) asau(i) rel.nom nature.nom ptcl rel.gen be.opt.3s that.gen that.nom duratikramah difficult.nom ‘[Who(i) has what(j) nature], it(j) is difficult for him(i) (to overcome).’ (Hit. 3.8, Hock 1989, 96.)

The interpretations found in Hindi/Urdu correlatives (1)-(3) above also have counterparts in Sanskrit. But Sanskrit is less restrictive than the modern language. Among other differences, there is an appositive reading for some left-adjoined correlatives  compare the following example with (6): (11) a. Initial appositive relative clause [ yó gWVatum íd usithā rel.nom sing.part.gen.pl ptcl be.impf.middle.2s āpír ūtw śiváh sákhā ] ally.nom favor.inst auspicious.inst familiars ] sá tvám ná indra mWlaya ptcl you.nom we.dat Indra.voc be.gracious.caus.imper.2s ‘You who have become the good friend of the Singers with your favor to your familiars be merciful to us, O Indra.’ (RV 6.45.17, Hettrich 1988, 639.) There is a striking absence in Sanskrit of evidence for the NP adjoined relative. Examples such as (12a) which could be seen as NP adjoined are just as easily seen as clauses adjoined to the left or right of a full clause.7 The initial relative pronoun is (only) linearly adjacent to the correlate. Hettrich includes a number of examples such as (12a), in which the relative pronoun immediately follows the head N. It is not clear that the relative clause is adjoined to the preceding NP. It could equally well be adjoined to the preceding CP, as indicated by the brackets in (12b). (12) Headed or adjoined relative a. ná m´rsā [śrāntam [yád ávanti devuh]] not useless pain rel.acc favor.pres.3pl gods.nom.pl ‘The [pain [which the gods favor ]] is not useless.’ (Hettrich 1988, 541) b. [ná m´rsā śrāntam ] [ yád avanti devuh ] not useless pain rel.acc favor.pres.3pl gods.nom.pl ‘[The pain is not useless] [which the gods favor.]’

 Alice Davison

Hock (1989: 111–2) offers a compelling argument against the (12a) structure in Vedic Sanskrit. What appear to be examples of NP-adjoined structures are plausibly analyzed as relative adjunction to a clause with an initial nominal phrase (12b). Hock finds that the small number of examples of relative clauses wholly contained within a clause turn out to have other explanations, such as parenthetical phrases in apposition, to the left or right of the correlative element. It is not clear where the ‘head’ N must be in Sanskrit relative constructions. The nominal head can appear freely in either the relative or the host clause, or in both, although there is a clear tendency for N to occur in the first clause in linear order (Hettrich 1988). This pattern would seem to be motivated by discourse conditions rather than syntactic structure. If so, then the NP adjunction source for correlatives is absent and we can assume that correlative clauses are base-adjoined to another clause in Sanskrit.

2. A base-adjunction analysis The base-adjunction view of correlative clauses has been discussed in various analyses, such as Andrews (1976/1985), McCawley (2004) and Dayal (1996). It seems to be something of a minority view, given the rich pool of knowledge of displacement processes and the constraints on them. My discussion of this issue will add a historical perspective, examining the syntax of correlatives in Sanskrit, which represents what we know of the earliest stages of Indic. I will argue that Sanskrit did not express syntactic subordination of finite clauses, though non-finite clauses are subordinate and clause internal. It will also take into account some aspects of correlative clauses which have not emerged from movement analyses based on correlative clause extraction from DP. The main difference between correlatives in Sanskrit and modern Indic languages lies in the syntactic adjunction relation: in the modern language the adjunction syntactically encodes subordination, while in Sanskrit it does not. There are some interesting interpretative and locality differences which, I argue, follow ultimately from this syntactic difference. The discussion will show that the syntax is coupled with the semantic properties of relative determiners, expressed as features building an anaphoric link between the modifier clause and the correlate which it modifies. It does not appear that movement rules are at all involved in the derivation of Sanskrit correlatives, while they might be involved in Hindi/Urdu. But the crucial difference is in the adjunction relation, not in the availability of movement. This section begins with an overview of Sanskrit relative clauses, which are always clause adjoined, not adjoined to a nominal phrase. I then propose that Sanskrit correlative CPs are symmetrically adjoined to CP, in contrast to Hindi/Urdu, which has asymmetric, subordinating adjunction of correlative CPs to TP. This difference of structure is supported by differences of clausal architecture, the range of possible interpretations of correlatives, minimality of question scope, and possible iterations of adjunction. In particular, Hindi/Urdu multiple relatives are locally adjoined and



Sanskrit and Hindi/Urdu correlatives 

constrained by a condition on asymmetric c-command which does not hold for Sanskrit. I outline the case for saying that all correlatives are base-adjoined, rather than extracted from a clause-internal DP. Evidence is given for the fundamentally anaphoric nature of correlatives, which is a central part of my formal analysis using syntactic and semantic features. I conclude with a note on the core difference between Sanskrit and Hindi/Urdu correlatives.

2.1

Symmetric and asymmetric adjunction of correlative clauses

In this section, I discuss the two possible structures for adjoining correlative clauses to a host clause. The difference between the two is in the clausal category of the host clause, CP or TP, and in the projecting category (Chomsky 2004). For Sanskrit, I follow Hock (1989) in giving correlatives the structure in (13). (13) [Symmetric adjunction to another clause] (Hock 1989) a. CP wo CP1[Rel] CP2 Relative XP Correlate XP b.

CP wo CP1 CP2[Rel] Correlate XP Relative XP

It has long been observed that clauses in Sanskrit seem to be linked in a very loose paratactic way, without syntactic encoding of subordination for example by Delbrück (1888), Hermann (1895), Gonda (1975), Hettrich (1988). Hock (1989) uses a formal syntactic representation to express this relation, the symmetric adjunction of a full clausal projection to another full clausal projection, which I express as in (13). A correlative may precede or follow another full clause. It is unclear exactly which CP projects, given that all the categories are CPs. Each CP c-commands the constituents of the other clause: while the relative CP c-commands its correlate, the correlate CP also c-commands the relative phrase. By contrast, for Hindi/Urdu, I propose a different adjunction structure, cf. (14). In this structure, the correlative CP is adjoined to TP, not to CP. The TP is the category which projects, as TP*. The adjunction structure TP* is the complement of its own CP* projection. This is a base-adjoined structure (cf. Dayal 1996, McCawley 2004).There is an asymmetric relation, in that the subordinate correlative clause does not project in category. It c-commands the constituents of TP, including the correlate linked to the

 Alice Davison

correlative. The correlate TP does not c-command the relative XP, for which evidence will be given in Sections 2.2.4-2.2.6 and 2.3 below. (14) Asymmetric adjunction CP* 3 C’ 3 C TP* wo CP[Rel] TP 3 3 relative XP correlate XP I will argue that the two structures explain some systematic and striking differences between correlatives in the two languages. Interestingly, there is evidence from Dutch comparative correlatives (Den Dikken, this volume) that both kinds of adjunction can coexist in one language. In Dutch, embedded examples of comparative correlatives are marked with the finite complementizer dat ‘that’. The presence of the overt complementizer on the two-clause combination shows that the structure is identical to (14), with the correlative adjoined to TP. In unembedded contexts, CP is adjoined to CP (cf. 13), and a complementizer is possible within the host clause as well as the correlative.

2.2

Sanskrit-Hindi/Urdu finite clause combining strategies

In this section I review evidence that Sanskrit and Hindi/Urdu differ in the adjunction relation between the correlative clause and the ‘host’ clause, leading to the conclusion that Sanskrit does not syntactically encode subordination for finite clauses, while Hindi/Urdu does, with consequences for interpretation and locality constraints. 2.2.1 Nonfinite clauses Sanskrit does encode non-finite clauses syntactically. For example, the Raising/ECM construction in (15) shows syntactic subordination. In both Sanskrit (15a) and Hindi/Urdu (15b), the matrix verb ‘see’ selects a participle subordinate clause with an accusative subject: (15) Exceptional Case Marking/Raising to Object a. arūVó [mā sak´rd v´rkah pathu yántam ] reddish.nom 1s.acc once wolf.nom path.inst go.impf.acc dadárśa hí see.pf.3s ptcl ‘On one occasion a reddish wolf saw [me going along the road].’ (RV 1.105.18a, Hettrich (1988), 178.)



Sanskrit and Hindi/Urdu correlatives 

b. eek bheeRiyee-nee [ hameeN jaNgal-meeN jaa-tee hu-ee] deekh-aa one wolf-erg we.dat forest-in go-impf be-pf see-pf ‘A wolf saw [us walking in the forest].’ Sanskrit has control infinitive subordinate clauses of many sorts, including object control in (15a). This must mean that there is clear evidence for case selection, selection of a non-finite verb form, and a controlled null embedded clause subject, very much like in the modern Indic languages. In the next sections, I will note some very different properties of the finite clauses in Sanskrit and Hindi/Urdu. 2.2.2 Sanskrit clause architecture: The clause initial string and markers of subordination In this section, I note the first of several aspects of finite clauses which fail to distinguish subordinate clauses from matrix clauses. A distinctive prefix to finite clauses is found in several very old Indo-European languages, including Sanskrit (Hock 1989, Schaüfele 1990) and Avestan, Old Persian (Hale 1987). This is a string of head positions occupied by particles and pronouns, which occupy up to five ordered positions in the left periphery of the CP (16). (16)

Vedic clause-initial string positions (Hock 1989, 115) Nexus 1 2 3 4 5 conjunction accented unaccented accented enclitic stressed eg. atha ‘so’ word particle particle pronoun pronominal [Rel, Int]

Evidence that this clause initial string involves CP projections comes from the nature of the particles, which are sentence oriented. (17) [Sanskrit] Sentence-oriented particles a. Unaccented: u ‘and’, sma ‘always, indeed’ ha ‘certainly’ b. Accented: tú ‘then’, vaí ‘truly, indeed, now, furthermore, surely’ In the string relative and interrogative determiners must appear as single words. They may be moved away from their DP, leaving a remnant NP (Schaüfele 1990). The clauseinitial string seems to be characteristic of an independent clause, because it contains sentence-oriented particles. Yet the clause initial string is found not just in the independent ‘correlate’ clause, but also in the correlative clause as well (18). The presence of the clause initial string in both clauses of (18) is another indication of symmetric adjunction (Hock 1989). (18)

[Sanskrit] Clause initial string in both clauses [yám(i) u ha evá tát paśávo manusyèsu yam rel.acc ptcl ptcl ptcl that cattle.pl.nom man.pl.loc rel.acc kumam(i) árohams] [tám(i) u ha evá paśúsu desire.acc obtain.pres.3pl that.acc ptcl ptcl ptcl cattle.pl.loc

 Alice Davison

tam kumam(i) rohati] that.acc desire.acc obtain.pres.3s ‘The desire which(i) the cattle obtained among men, he obtains the same desire(i) among the cattle.’ (S.B. 2.1.2.7 Hock 1989, 111). Note that the determiners yam ‘relative’ and tam ‘that’ have been moved away from NP in both clauses, a freely available combination. Determiner movement is marked in main or dependent clauses in Hindi/Urdu.8 2.2.3 No syntactically subordinate complement clauses in Sanskrit The next section offers another example of the absence of syntactic subordination in Sanskrit. Interrogative subordinate clauses cannot be marked as interrogative. In both languages, main clause interrogative sentences are marked by the k- series of determiners, which are distinct in form from the y-/j- series which are specific to relative clauses in both Sanskrit and modern Indic languages. Yes-no questions are prefixed by ‘what.int’ cf. (19). (19)

[Sanskrit] Yes/no question kád aryamVó mahás pathu áti krāmema dūdhyo what.int Aryaman.gen great.gen road.inst surpass.opt.1pl inferior.pl.acc ‘Should we overcome the base people on the path of the great Aryaman? (R.V. I. 105.6cd, Etter 1985, 125.)

(20)

[Sanskrit] Constituent question kó dadarśa [prathamàm juyamānam]? who.int see.perf.3s first.m.acc born.m.acc ‘Who has seen [(him) as first born]?’ (R.V. I. 164. 4a, Etter 1985, 66.)

If a question is in a dependent complement clause, Vedic Sanskrit substitutes a relative y- determiner for the k- interrogative in both yes-no questions (21) and constituent questions (22): (21)

[Sanskrit] Complement yes/no question ná tásya vidma tád u su prá vocata not this.gen know.pres.1pl this.acc ptcl good forth speak.imper.2pl [yúvā ha yád yuvatyuh kséti yónisu] youth.m.nom ptcl what.rel young.girl.gen lie.pres.3s womb.loc.pl ‘We do not know of this, tell us well [whether the young man lies in the lap of the young girl.’ (R.V. 40.11ab, Etter 1985, 210.)

(22)

[Sanskrit] Complement constituent question nuham tád bhagavan veda [yátra gamisyāmi] not.I this.acc Lord.voc know.pres.1s where.rel go.fut.1s ‘I do not know, O Lord, [where I will go] (S.B. 14.6.11.1, Hettrich 1988, 524.)



Sanskrit and Hindi/Urdu correlatives 

This pattern is found in Homeric Greek (Chantraine 1958), and disappears in both Classical Greek and later Sanskrit. Nothing like (21)-(22) is possible in Hindi/Urdu, as (23) shows. (23) ham-nee (yah) puuch-aa [ ki kahaaN /*jahaaN vee aa-eeNgee] we-erg this ask-pf that where.int /*where.rel 3pl come-fut.3pl.m ‘We asked where they will come.’ Sanskrit has several ways of marking sentential complements: by (a) parataxis of the selected complement clause, (b) prefixation or suffixation of the quotative iti ‘thus’ (Hock 1982), or else (c) the interrogative complement is put in relative form, with an interrogative interpretation. The predicate selecting the complement determines whether it is an embedded question or not (Lahiri 2002). Sanskrit does express a selectional relation, but this selectional relation can be expressed in Vedic Sanskrit only by the very general CP-CP adjunction licensed by the relative form of one of the clauses. Hindi/Urdu has an available marker of subordination (ki) which marks interrogative as well as other complement types as syntactically distinct from main clauses. 2.2.4 Minimality violations in questions and relative clauses Minimality violations include violations of the Complex NP condition, in which a questioned phrase cannot be extracted from a relative clause in languages like English and Italian (Rizzi 1990). Analogues of Minimality violations occur in Hindi/Urdu if a question phrase is within a correlative clause, and the whole combination has the force of an interrogative sentence. Well-formed sentences, on the other hand, are found which have the force of a question, but the questioned phrase is in the matrix clause. The correlative clause is within the scope of the question. Such sentences are found in both Sanskrit and Hindi/Urdu. (24) Minimality contrast in Hindi a. Interrogative in main clause [joo kitaab(i) us-nee t(i) likh-ii hai] woo kis-koo sab-see rel book 3s-erg write-pf is 3s who-int.dat all-from acchii lag-ii? good strike-pf ‘Who likes best the book [that he/she wrote]?’ b. Interrogative in relative clause *[joo kitaab(i) kis-nee t(i) likh-ii hai] woo aap-koo sab-see rel book who.int-erg write-pf is 3s.nom you-dat all-from acchii lag-ii? good. strike-pf *Who(j) did you like best the book [that t(j) wrote]?

 Alice Davison

Note that in (24a), the whole clause projects as a question, and the restrictive relative is interpreted inside the scope of the question. Otherwise the correlative would be adjoined to CP[Int] and as a result, the relative meaning would be outside the scope of the question, as in the case of an appositive clause. This contrast in Hindi/Urdu supports the assumption that the relative CP must be adjoined to TP, not the matrix CP, in the asymmetric adjunction structure proposed above in (14). The contrast can be represented structurally in (25), assuming the clause-typing features [Rel] and [Int] on C, standing for relative and interrogative clauses, respectively. (25) a. good result (24a) CP 3 IntDPj C' 3 C[Int] TP wo CP TP* 3 3 RelDPi C' IntDPj 3 C[Rel] TP 3 RelDPi b. bad result (24b) CP wo IntDPj C' 3 3 C[Int] TP 3 CP TP* 2 2 RelDPi C' 3 C[Rel] TP 3 IntDPj RelDPi In the well-formed version, (24a), the relative DP, RelDPi has a copy in the nearest CP projection, which is typed as a relative clause.9 The relative clause is adjoined to TP*, containing an interrogative in situ. The TP structure is the complement of a C[Int], with a copy of the interrogative phrase, IntDPj in its specifier. This specifier is the



Sanskrit and Hindi/Urdu correlatives 

closest such position to the interrogative phrase. This conforms to Rizzi’s (1990) Minimality requirement. In the ill-formed version (24b), both RelDPi and IntDPj originate in the same TP, which projects as a relative clause. Nothing is wrong with the relation between the relative specifier and RelDPi; this is the closest specifier position. But the interrogative, IntDPj is copied to a specifier position which is not the closest to it, as the relative specifier with RelDPi intervenes. Sanskrit appears not to have a Minimality contrast, observe (26). CP is adjoined to CP, not TP, in the symmetric adjunction structure. Either CP could be a question, and either one could be a relative clause. In principle, the normal case in Sanskrit should be like (24a) in Hindi/Urdu, in which one CP is a relative clause, the other is a question, as in (26a). There is, however, a striking example, first noted in Delbrück (1888), in which one clause is both a relative and a question, on the order of the ungrammatical (24b). (26) [Sanskrit] Absence of Minimality contrast a. Interrogative in main clause [yás tán ná véda ] kím Wcu rel.nom that.acc not know.pres.3s what.int.acc verses.inst karisyati do.fut.3s ‘[The one who does not know this] what will (he) do with the Verses?’ (R.V. 1.164.39c, Etter 1985, 167) b.

Interrogative in relative clause: [yát kim ákaram ] tásmād idám upād what.rel what.int.acc do.aor.1s that.abl this.nom befall.aor.3s ‘[Because(i) I have done what.int] therefore(i) this(j) has happened.’/ ‘What [for which reason(i) I did __], for that reason(i) this(j) befell./ ‘That happened [because of what.int that I did].’ (S.B. 4.1.5.4, Delbrück 1888, 550, Hock 1989, 105)

The sentence in (26b) is very marked, and a challenge to translate into any modern language with a Minimality requirement.10 Yet it appears to be grammatical, and with some reservations we can assume it represents a rare but occurring sentence type. It constitutes a Minimality violation shown schematically in (27): (27) Minimality violation: yát(i) kim(j) t(i) t(j) Representing the structure of (26b) is difficult under the older assumptions that both relative and interrogative operators move to the same Specifier of CP, as in Chomsky (1995) and earlier work which assumes a single COMP position. A way to separate the operator positions of relatives and questions can be found in Rizzi’s (1997) proposal which has it that a CP projection may in fact be a series of related but semantically distinct functional projections. The functional heads include Force (marked below as

 Alice Davison

FP), the site of relative operators, and in some languages, the position of an overt marker of subordination like that. They also include Finite (marked below as FinP), where interrogative operators are found. In between are Focus and Topic heads.11 Now, to show CP adjunction to CP, in which one CP has both relative and interrogative properties, let us assume the following structure for (26b): (28) [Sanskrit] FP 3 FP1 FP2 wo wo yáti F’ tásmādi F’ what[Rel] 3 that 3 F(orce) FinP F(orce) TopP 3 3 kimj Fin’ idamk Top’ what[Int] 3 this 3 Fin TP Topic TP 3 3 [yáti kimj ákaram] [ tásmādi idamk u ´pad] do befell There is a Minimality violation in FP1, and also in FP2 depending on whether TopP involves operators or not. These violations seem to be tolerated in Sanskrit. Structurally, the Minimality violations in FP occur internally to FP. (28) is different from (25b), in that the relative maximal CP in (28) (FP1) is not adjoined within the syntactic domain of the interrogative, the relationship which is forced in (25b) by the Hindi/Urdu requirement that CP be adjoined to TP. There is a tempting parallel which I would like to suggest between the Rizzi projections in (28) and the clause-initial string noted above for Sanskrit finite clauses. Note that both of the adjoined clauses have initial strings of pronouns and determiners, recalling the clause-initial string discussed earlier. It is interesting to speculate that the clause-initial string actually is a version of Rizzi’s Left Periphery, with the differences that only the first position may have a maximal projection in its specifier, the following head positions may be filled with only X0 projections, and the head positions are not absolutely specified semantically. A full justification of this idea is beyond the scope of this paper, but if there are such parallels, then the finite clauses of the oldest Indo-European languages are not so utterly unlike the corresponding clauses of modern languages, and there is a resolution to the debate among Hale (1987), Kiparsky (1994) and Hock (1989) about whether the left periphery includes Topic, COMP and wh-phrases. The ancient and modern languages actually differ in clausal structure, in that modern languages do not have the clause-initial string, filled by heads. In the modern



Sanskrit and Hindi/Urdu correlatives 

languages, the Rizzi-type CP projections allow XP specifiers. In any case, we see a difference in grammaticality in relative clause structures between Sanskrit and Hindi/ Urdu which can be explained by a difference of symmetric and asymmetric adjunction. If this difference did not exist, we could expect that full CP relative clauses in Sanskrit could adjoin lower down, within the clause initial string, either on TP or on NP. But if the relative CP as a phrase adjoined to TP, the lowest clausal functional projection, the relative clause would be preceded by the particles and pronouns of the clause-initial string, an impossible order in Sanskrit, see Davison (2009). Relative clauses are peripheral, adjoined to the outermost clausal projection.12 2.2.5 The stacking condition on Hindi correlatives As adjunction is a free possibility of Merge, there is no reason in principle why restrictive relative clauses should not iterate in a ‘stacked’ structure, consisting of two or more clauses modifying the same DP. This is possible for English, and other languages with NP adjoined relative clauses, but not for correlatives. So sentences like (29a), and (29b) as well for some speakers, are ungrammatical. This fact seems to be a general property of correlatives (Grosu 2002, Den Dikken 2005). (29) a.

[Hindi] Stacked relatives on the left *[joo laRkii(i) skuul-meeN mehnat kar-tii hai], rel girl school-in effort do-impf is [joo(i) anu-kii doost hai] woo(i) bahut acchii hai rel Anu.gen friend is 3s very good is ‘That girl is very nice who works hard in school who is a friend of Anu.’ (Grosu 2002)

b. [Hindi] Stacked relatives on the right (*)woo laRkii(i) bahut acchii hai, [joo(i) skuul-meeN mehnat that girl very good is who school-in effort kar-tii hai] [ joo(i) anu-kii doost hai] do-impf is rel Anu-gen friend is ‘That girl is very nice who works hard in school who is a friend of Anu.’ (Ibid) Stacked correlatives are quite ungrammatical in Hindi/Urdu, and can only be saved by inserting an overt conjunction like aur ‘and’ between the two relatives. But stacking seems to be quite usual in Sanskrit, even with restrictive correlatives:13 (30) [Sanskrit] a. Stacked initial restrictive relatives (with verb gapping) [yá§(i) s�ryam] [yá§(i) usásam jajuna] rel.nom sun.acc rel.nom dawn.acc create.pf.3s

 Alice Davison



[yó(i) apum netu] sá(i) janāsa indrah rel.nom water.gen.pl leader.nom that.nom people.voc Indra.nom ‘Who.rel created the sun, who.rel created the dawn, who.rel is leader of the waters, that, O people, is Indra.’ = ‘That one who made the sun, who made the dawn, who is the leader of the waters, O people, is Indra.’ (R.V. 2, 12.7c, Hettrich 1988, 544)

b.

Stacked initial restrictive correlatives (full verbs) [yám(i) agne prtsú mártyam] rel.acc Agni.voc battle.loc.pl mortal.acc ávā vujesu yám(i) junuh aid.aor.2s contest.loc.pl rel.acc urge.on.aor.2s sá(i) yántā śáśvatīr ísah] that.nom leader.nom steady.nom strong.nom ‘[Which mortal(i) you aided in battles, Agni][whom(i) you urged on in struggles] [that one(i) is the leader, steady (and) strong].’ (R.V. 1.27.7, Hettrich 1988, 542.)

c. Initial appositive relative clause [yó gWVatum íd usitha rel.nom.s sing.part.gen.pl ptcl be.impf.middle.2s āpír ūtw śiváh sákhā ] ally.nom favor.inst auspicious.inst familiars sá tvám ná indra mWlaya ptcl you.nom we.dat Indra.nom be.gracious.caus.imper.2s ‘O Indra, who has become the good friend of the Singers with your favor to your familiars, you be merciful to us.’ (RV 6.45.17, Hettrich 1988, 639.) If I am right about the symmetric adjunction structure in Sanskrit, then multiple symmetric adjunction would look like the tree diagrams in (31a,b). (31) Multiple symmetric adjunction a. CP 3 CP[Rel] CP 3 CP[Rel] CP b. CP 3 CP CP 3 CP[Rel] CP[Rel]



Sanskrit and Hindi/Urdu correlatives 

For Hindi/Urdu, I will propose that multiple relative clauses are adjoined differently. Dayal (1996) notes that the structure of adjunction in Hindi is similar to the structure proposed by Jacobson for English DP adjoined multiple relatives which undergo extraposition (Jacobson 1983). That is, instead of adjoining both of the relative CPs to a single TP, each relative is adjoined to the following relative, and the combination is adjoined to TP. This can be seen if we compare single clause adjunction in (32) to successive adjunction in (33a) and (33b). (32) TP wo CP[Rel] TP 3 3 [ relative ] [ matrix ] (33) a. Multiple relative adjunction TP wo CP2 TP 3 3 TP [ matrix ] wo CP1 TP 3 3 TP [ relative 2 ] 3 [ relative 1 ] b. Multiple relative adjunction, alternative possibility TP wo CP1 TP 2 3 TP CP2 TP 3 2 3 [ relative 1 ] TP [ matrix ] 3 [ relative 2 ] Below I will argue that the correct structure for multiple adjunction is the one in (33a). The first relative clause in the sequence is represented in bold in (33a); it is adjoined to the TP of the second relative clause CP. As I propose below, this structure distinguishes between well-formed multiple relatives from ill-formed ones in Hindi/Urdu. What distinguishes the well-formed examples in (30) in Sanskrit from the structure giving rise to ungrammatical sentences in Hindi/Urdu (29)? I will propose a

 Alice Davison

syntactic explanation which follows from the difference of adjunction structures, at least indirectly. In the structure in (32), the correlative CP asymmetrically c-commands the correlate within TP. What I suggest is that this is a condition on correlatives (34) in Hindi/Urdu. (34) Correlative condition  Hindi/Urdu The relative CP must asymmetrically c-command its correlate; correlativecorrelate construal is well-formed only if the correlate is asymmetrically ccommanded by the relative CP.14 The asymmetry of adjunction and projection of TP is a way of encoding subordination as a property of the relative clause. The TP is the matrix clause because the TP category projects, not the relative CP. The semantic consequence is that the relative clause is interpreted as a modifier, not an independent clause. What is wrong with (29) is a failure of (34). For both correlatives in (29a) to meet the condition in (34), they both must c-command the TP with the correlate. The condition is met for CP2, which is adjoined to TP containing the correlate. But it is not met for CP1, which is adjoined to the TP within CP2, and so it c-commands just Relative 2, containing the relative operator as a possible correlate. In general, correlates cannot also be operators; Dayal (1996:  182) notes that interrogatives cannot be correlates modified by a correlative clause. This condition can be extended to relative phrases; neither interrogative phrases nor relative phrases may be the correlate of a relative clause.15 The condition (34) does not hold for Sanskrit. Anaphoric connections may hold across CP clause boundaries, so that two correlatives may be linked to a single correlate, independently of the adjunction structure. See discussion below of anaphoric construal conditions on the relative and its correlate. I conclude this section with a comment on my proposal that adjunction is successively local, as in the structure (33a). The alterative structure, as was shown above in (33b), involve adjunction of both relative clauses to the same matrix TP, in a simplified structure as in (35): (35)

TP 3 CP1 TP 3 CP2 TP

I will argue below that this cannot be the right structure, whether derived by baseadjunction or movement. It does not discriminate between grammatical and ungrammatical iterated relative clauses, or explain the absence of stacked correlatives without a special stipulation ‘don’t adjoin twice’. Jacobson (1983) argues against the analogue of this stacking structure for English on empirical grounds, as I will for Hindi/Urdu: it does not correctly predict grammaticality differences. So far no theoretical constraint rules out (35), but see footnote 7.



Sanskrit and Hindi/Urdu correlatives 

2.2.6 Iterated restrictive correlative clauses The local c-command condition appears also when correlative clauses iterate, a marked but possible option in Indic languages. One restrictive clause finds its correlate in a TP, and in this TP a referentially distinct second relative phrase is introduced. The second relative finds its correlate in another TP. This pattern was noted independently by Dasgupta (1980, 334–349), Hock (1989) and McCawley (2004); below I cite McCawley’s examples. Because these sentences are very complex and not easy to process either by the reader, or by native speakers, I tag each clause with a roman number, which is then repeated in a summary of the sentence showing the coindexing in it. In Hindi/Urdu, there is a contrast of grammaticality between the well-formed sentence (36), schematized as (37), and the ill-formed sentence (38), schematized as (39): (36) [Hindi/Urdu] [I] [joo aadmii(i) caay pii rahaa hai ] which man tea drink prog is [II] [maiN jis makaan(j)-meeN us-kee liyee(i) kaam kar rahaa huuN] I which house-in 3s-for work do prog am [III] raam pichlee saal wahaaN(j) rah-taa thaa Ram last year there stay-impf was ‘Ram was living in the house where I am doing work for the man who is drinking tea.’ (McCawley 2004, 307; ? or?? for some speakers) (37) a. [I] [joo aadmii(i) ] [II] [us-ke liyee(i) b. [II] [jis makaan(j) [III] [wahaaN(j)] (38) [Hindi/Urdu] * [I] [joo aadmii(i) caay pii rahaa hai] rel man tea drink prog is [II] [jis makaan-meeN(j) maiN ab rah-taa huuN] rel house-in I now stay-impf am [III] woo(i) pichlee saal wahaaN(j) rah-taa thaa 3s last year there stay.impf was ‘The man who is drinking tea, he lived last year in the house where I am living now.’ (McCawley 2004, 307; * for all speakers)16 (39) a. *[I] [joo aadmii(i) ] b. [II] jis makaan-meeN(j)]

[III][ woo(i)] [III] wahaaN(j)]

The problem here is that joo aadmii ‘which man’ in CP[I] finds its correlate in TP[III], skipping over CP[II]. Let us assume the structure with multiple clausal adjunctions in (33a) rather than (33b). This means that the initial relative clause is adjoined ‘low’ to the following relative clause, rather than to the matrix TP. The whole structure is adjoined to the matrix TP. In the well-formed combination, the first relative clause adjoins to a TP containing

 Alice Davison

its correlate, so that the first relative clause c-commands the TP with the correlate. A new relative is introduced in the second clause, which c-commands the TP containing the second correlate. This configuration corresponds to the grammatical (36)-(37). In the ungrammatical combination (38)-(39), the first relative clause does not c-command TP containing both correlates. Only the second relative clause c-commands TP. The condition (34) then also guarantees a local relation between the correlative and its correlate. The correlative clause has to be adjoined to the TP containing the correlate.17 There is no contrast like this in Sanskrit. As Hock (1989) pointed out, iteration in Sanskrit is much freer than in Hindi/Urdu. (40) [Vedic Sanskrit] [I] [yo(i) vai śreyasah parivesaVam avavadati] rel.nom ptcl better.gen food.acc deprecate.pres.3s [II] [yayā(k) vai sa(j) tam(i) ārtyā kāmayate] rel.inst ptcl he.nom he.acc affliction.inst afflict.pres.3s [III] tayā(k) enam(i) ninayati that.inst that.acc afflict.desid.pres.3s ‘Who(i) deprecates the food of his(i) better(j), he(i) is afflicted by him(j) with whatever afflictions(k) he(j) wants to afflict him(i) with.’ (J.B. 3.303. Hock 1989, 108.) (41) a. [I] yo(i) [II] tam(i) b. [II] yayā(k) [III] tayā(k) The c-command condition is observed in (40)-(41). Each relative clause finds a correlate in the next clause to the right. But equally well-formed sentences are possible both for the initial relative shown here and in other examples in Hock (1989: 107–9), including both initial relatives (42)-(43), as will as final relatives. Their counterparts are ill-formed in Hindi/Urdu. (42) [I] [yády(i) áhainam if.rel motiveless

pruncam forward

[II] [yáthā(j) párāca so.rel face.other.way upaháret bring.opt.3s

ácaisīr] gather.aor.2s

āsīnāyā sitting.dat

pWsthato ‘nnudyam] back food.etc.acc

[III] tádWk(j) tát(i) that.like that/so ‘If(i) you have gathered it forward without motive, just as if(j) one were to bring food from the back to someone sitting on the other side, so(j) (would) that(i) be’. = ‘If you piled the fire altar forward without motive, that would be as if one were to bring food from the back to someone sitting (and looking away).’ (S.B. 10.5.5.2, Hock 1989, 107.)



Sanskrit and Hindi/Urdu correlatives 

(43) a. [I] yády(i) [III] tát(i) b. [II] yáthā(j) [III] tádWk(j) Here the local linking is not observed; [II] with its relative phrase intervenes between yády ‘if ’ and tát ‘that’. The free iteration of relatives seems to follow from a much less strict condition on relative-correlate construal.  Discourse anaphoric links hold between relatives and their correlates, freely cutting across CP clause boundaries, not constrained by the asymmetric c-command condition in (34). A condition of this sort would have no effect applied to symmetric adjunction structures. Neither clause is syntactically differentiated as the modifying clause. Instead the relative is only marked lexically as subordinate, because it contains a relative determiner. 2.2.7. Syntactic conditions on adjunction structures in Hindi/Urdu I have proposed for Hindi/Urdu a c-command condition in (34) which constrains iterated correlatives of two types: stacked relatives modifying the same DP, and chains of relative and correlative links. This condition rules out both stacked correlatives and ill-formed combinations of two correlatives each linked to the correlate in the matrix clause. In each case, there is a relative clause which does not asymmetrically c-command its correlate in TP. It is interesting that there is also a semantic explanation for the absence of stacked correlatives. The correlative constructions discussed in Grosu (2002) all prohibit stacking. Grosu derives this property from the definiteness feature [MAX] which is added to clauses with predicational effect, such as relative clauses. The [MAX] feature of correlatives should produce exactly one singleton set for each relative – correlate combination. Adding a second property would be semantically vacuous – no additional unique value could be derived. This should be sufficient to rule out stacked correlatives, as only one correlative can contribute information which determines reference. But this explanation does not carry over to the contrast between good and bad cases of iteration of correlatives, which are examples of successive correlative clauses modifying different correlates. The syntactic explanation in (34) seems more general as a condition on well-formed syntactic structures, though clearly [MAX] is involved in the semantic interpretation of correlatives in all these cases.

2.3

Adjunction structures and their consequences

Sanskrit and Hindi/Urdu have correlative clauses, but there are differences of syntax and relative-correlate construal. In this section I have proposed that there is a critical syntactic difference between the relative structures in the two languages, from which many of the differences can be derived. The difference lies in how the relative clause is adjoined to the clause containing the correlate (13)-(14). In Sanskrit, CP is adjoined to CP, forming a categorially symmetric structure, not marking either clause as syntacti-

 Alice Davison

cally subordinate. In general, Sanskrit does not distinguish syntactically subordinate clauses. The clause-initial string of pronouns, determiners and particles is found in both the relative and the correlate clause (18). The relative construction substitutes for interrogative complements. There are fewer constraints than in Hindi/Urdu on interrogative constructions, and on relative clause iterations. In Hindi/Urdu, the correlative clause is adjoined to the TP containing the correlate. The relative CP is distinguished syntactically as subordinate, due to the fact that it does not project its category. Rather, the host or matrix clause projects as TP, and then as matrix CP, which results in an asymmetric c-command relation between the correlative and its correlate. I propose that this relation is a necessary condition for well-formed construal of the correlative as a modifier of its correlate. This condition takes the form of a condition in (34) which guarantees a ‘local’ relation between a correlative and the correlate which it asymmetrically c-commands. This condition rules out relative clauses which do not c-command the correlate, assuming the adjunction structure in (33a), which resembles the structure for stacked relatives in English proposed by Jacobson. This analysis allows multiple relatives within one clause. In Sanskrit, anaphoric links are established across clause boundaries in a very general way. In Hindi/Urdu, one correlative clause c-commands one TP with multiple correlates, cf. (3) (see Dayal 1996 for an account of the semantics of such clauses). Condition (34) rules out a paradoxical structure resembling the ‘Bach-Peters’ sentences in English: (44) [Hindi/Urdu] * [I] [joo paarTii(i) aisaa ummiidwaar(j) DhuuNDh sak-eegii] rel party such candidate search be-able.fut.3fs [II] [ jis-kii(j) usee(i) waaqaaii zaruurat hai] rel-gen 3s.dat real need is ‘The party(i) which really needs him(j) will search for the candidate(j) which(j) it(j) needs.’ In this example, there are two relative-anaphoric links: (45) a. [I] joo paarTii(i)... [II] usee(i) b. [I] aisaa ummiidwaar(j)... [II] jis(j) While relative clauses may be adjoined to the left or to the right of the main clause, the structure of (44) would have to be paradoxical. CP-I would have to be locally adjoined to TP-II, for the anaphoric linking in (45a.). But CP-II would have to be locally adjoined to TP-I for the linking in (45b). Both these requirements cannot be satisfied simultaneously. This kind of paradox is another reason to assume CP adjunction to TP, rather than to another CP, as the paradox would not exist except for the CP-TP difference. This argument predicts that no such paradox should exist in Sanskrit, and indexing relations in (45a,b) should in principle be possible in the same sentence. (In the movement analysis, clause I would be extracted from a right relative clause II extrap-



Sanskrit and Hindi/Urdu correlatives 

osed from aisaa ummiidwaar ‘such candidate’; presumably this sequence of movement would be prohibited in some way.) The condition in (34) also allows relatives on both the left and right, which are common in Hindi/Urdu (46). There are two correlates in the matrix clause, but no violation of (34)18: (46) [Hindi/Urdu] Left and right relatives on the same correlate TP [I] [jis vidyaarthii(i)-kee nambar sab-see acchee haiN] rel student-gen grades all-than good are [II] [usee(i) yah puraskaar(j) mil-eegaa] 3s.dat that prize receive-fut.3s.m [III] [joo(j) mukhyaa mantrii vidyaalay-kee saalaanaa mahootsav-par rel.nom chief minister college-gen annual ceremony-on dee-gaa] give-fut ‘The student(i) who(i) gets the best grades will receive the prize(j) which(j) the chief minister will give at the annual college ceremony.’ Since the relatives are not adjacent to each other, they can adjoin to TP as in (47). (47) TP 3 TP* CP[Rel,III]j 3 CP[Rel,I]i TP# 3 ... DPi ... DPj... The left correlative adjoins to TP#, while the relative on the right adjoins to TP*. Both relative CPs c-command their correlates.19 I propose that the syntax of correlative clauses in Sanskrit and Hindi/Urdu can be characterized by a combination of two factors: (i) the difference of syntactic adjunction and (ii) the asymmetric c-command condition on the relative-correlate construal in (34), which depends on asymmetric adjunction in Hindi/Urdu. These factors explain why some combinations are well-formed, while others are not. This account has taken the adjunction structures as primitives, derived by Merge without movement (as copying and remerging) applying to the relative clause itself. It is in this respect a version of the base-adjunction analysis which has been proposed in various ways for correlatives.

 Alice Davison

3. The placement of correlative clauses: movement vs. base-generation In this section I contrast the base-generation hypothesis I have just proposed above with some proposals which derive correlative clauses by movement. There are two different proposals I will discuss briefly, Mahajan (2000) and Bhatt (2003). They differ considerably in details, but what they have in common is that they derive left correlative clauses by movement.

3.1

The Kaynean analysis of relatives (Mahajan 2000)

Mahajan’s account starts with a DP in which D is the sister of a CP clause, following the derivation of English headed relatives in Kayne (1994). He uses a series of leftward scrambling and deletion operations, which derives the familiar NP-relative structure, and also a number of other well-formed structures in Hindi/Urdu, including the correlative in left position in the sentence. In both cases, the relative clause starts out as a constituent of the modified DP. The leftward position is derived, not base-adjoined as I have suggested. The derivation of correlatives by movement has no obvious application to Sanskrit. If Hock’s (1989) arguments show that there is no unambiguous evidence for nominally adjoined relatives, it would not be available to derive Sanskrit correlatives.20 The Kaynean DP structure would be plausible for Hindi/Urdu, as it can derive NP-internal relatives as well as single correlatives. But as Mahajan notes, his proposal as it stands does not have an obvious extension to multiple relatives. It remains to be seen how the analysis would derive the properties of left-adjoined relatives such as the nostacking constraint, the absence of appositive readings, or the well-formed instances of iteration with interlinked relative-correlate pairs. Within its basic assumptions, however, the analysis is very elegant and derives many well-formed outcomes, which need to be accounted for in other theories by phrasal movement.

3.2

The CP-DP analysis (Bhatt 2003)

Bhatt (2003) derives left-adjoined correlatives from a position left-adjoined to DP, forming a relative clause-DP constituent which is grammatical in colloquial Hindi/ Urdu, cf. (48a).21 (48) a.

rahul [[joo kitaab(i) shiilaa-nee likh-ii] woo(i) (kitaab)] paRh Rahul rel book Shila-erg write-pf that book read rahaa hai prog is ‘Rahul is reading the book which Shila wrote.’ (After Bhatt 2003)



Sanskrit and Hindi/Urdu correlatives 

b.

[joo kitaab(i) shiilaa-nee likhii ](j) [e(j) woo(i) (kitaab)] rahul rel book Shila-erg write.pf that book Rahul paRh rahaa hai read prog is ‘Rahul is reading the book which Shila wrote.’ (After Bhatt 2003)

The relative CP is optionally raised overtly to adjoin to TP by quantifier raising/A'-scrambling (48b). (Some comments on quantifier raising as a motivation are included in Section 4 below). Prosodic conditions account for the options of expressing the noun N in the relative, the main clause or both. Because A'-movement is involved, relative clause constructions should be subject to island constraints, reconstruction effects and the Principle C condition, and Bhatt offers examples which make up the most compelling evidence for his proposal. I will discuss these briefly below. I will focus here on the special conditions which must be stipulated to derive the contrasts of grammatical and ungrammatical sentences for which I have proposed a basegeneration analysis. Bhatt (2003) assumes one application of A'-scrambling which moves the entire relative clause to adjoin to TP. Correlatives with multiple relatives cannot, however, have a single DP source from which the relative clause is raised, so they are derived by base-adjunction to TP. Bhatt proposes an economy principle governing adjunction which places the relative clause as close as possible to the DP which is modified. So single correlatives and multiple relatives have different sources, and should differ in locality conditions (see Bhatt 2003 for examples showing these contrasts).22 To derive the no-stacking condition, Bhatt stipulates that only one relative clause may raise by scrambling or quantifier raising. He does not specify the syntactic domain in which this stipulation would apply, nor how this restriction would be related to general conditions on scrambling and quantifier raising; these processes are normally able to occur multiple times. It is not clear whether the [MAX] feature in Grosu 2002 would be sufficient to rule out both stacked correlatives and the source of movement, stacked relatives on DP. Bhatt rules out all cases of two relative clauses raising and adjoining to TP, on the basis of ungrammatical sentences like (38) above (Bhatt 2003, 508), which I have analyzed as violating the c-command condition in (34).23 Bhatt’s analysis could be extended with additional provisions to derive the grammatical sentence (36), with the structure in (33a). The first relative joo aadmii caay pii rahaa hai is generated as a modifier of us-kee liyee, then the relative clause adjoins to TP, by some operation of quantifier raising. Then this whole TP is turned into a relative clause by merge/adjunction to ‘which house’, by another operation of quantifier raising. So there is some sequence of operations of Merge and quantifier raising which can derive the well-formed (36). The ill-formed version in (38) is ruled out by stipulation, prohibiting two operations of quantifier raising out of the same matrix clause. Bhatt offers no independent conceptual justification for it. Also, if there is no ‘earliest’

 Alice Davison

principle, it’s not clear what prevents the whole complex structure from being generated by adjunction of two modifier DPs in the matrix. Both would undergo quantifier raising, generating the ungrammatical version (38). The analysis I propose for Hindi/Urdu is conceptually simpler, in that it requires a c-command condition filtering the output of an ‘earliest’ adjunction condition, which adjoins clauses in the c-command order closest to linear order. This ‘earliest’ condition is highly speculative but it is what distinguishes two kinds of bad combinations from the one good case of iteration. It may ultimately derive from the predicate properties of relative D. My explanation is based on the c-command relation among finite clauses which is found in Hindi/Urdu, but not in Sanskrit. It offers a conceptually simple explanation for the difference in grammaticality between Hindi/Urdu and Sanskrit with respect to iterating correlatives. The DP adjoined sentences which Bhatt assumes as the source of correlatives would have to be treated in my system as relatively marked innovations in the syntax of Hindi/Urdu. In addition, stacking of relatives with the same correlate would have to be ruled out at the source DP, by a separate stipulation against more than one relative adjoined to DP.24

3.3

The construal relation between the correlative clause and the correlate

Correlatives are adjoined to TP in both Bhatt’s analysis of Hindi/Urdu relatives and the one proposed above. Bhatt’s (2003) account derives the structure by A'-movement. This A’-movement could be optional adjunction, or some kind of overt, optional quantifier raising, suggested by the fact that left-adjoined correlatives have the semantics of a Generalized Quantifier, incorporating the semantics of a determiner (Dayal 1996). The analysis depends on an economy condition, requiring relatives to be adjoined as low as possible, for example to the modified DP, the phrase which corresponds to the correlate in the examples above. The relation between the relative clause and what it modifies is a sister relation in (49a). The external relative is the head of an A'-chain, whose tail is the deleted copy/trace, and subject to constraints on A'-chains as Bhatt shows. (49) a. ... [CP DP]... b. CP [...[CP DP]] There must also be a local predication relation between CP and DP in (49a), and between DP and the reconstructed CP in (49b) (cf. Safir 1986). There is a restriction on adjunction to DP (Bhatt 2003: 493ff): the quantifier in DP must belong to a set which excludes indefinites. Without this constraint, left-adjoined correlatives would be ungrammatical. Left correlatives cannot be directly construed with indefinites (50a) (Subbarao 1984, Dayal 1996: 160ff): (50) a. *[ joo laRkiyaaN khaRii haiN] doo lambii haiN rel girls standing are two tall are ‘[Which girls are standing] two are tall.’ (Dayal 1996: 160)



Sanskrit and Hindi/Urdu correlatives 

b. [ joo laRkiyaaN khaRii haiN] un-meeN-see doo lambii haiN rel girls standing are 3.pl-in-from two tall are ‘[Which girls are standing] two of them are tall/Two girls who are standing are tall.’ (Dayal 1996: 161) In the grammatical sentence (50b), the correlate is a pronoun within a PP. In Bhatt’s derived structure, the relative clause would have to be adjoined within a PP; it would be a sister of a pronoun in PP. In order to rule out ungrammatical correlatives as in (50a), a restriction on quantifiers and DP structure has to be stated which limits possible applications of Merge. Relative clauses are basically anaphoric, as I will say in more detail below. But it seems that the quantifier-anaphora constraint illustrated in (50) can be factored out of the analysis of relatives per se, as Hock (1989) argues. It turns out that non-relative sentences connected only in discourse obey the same constraint on construing an indefinite or weak quantifier with an antecedent. (51) a. shimlaa-meeN meree kaii sahayoogii dakSiN bhaarat-kee thee Simla-in my few associate southern India-gen were ‘In Simla a few of my associates(i) were from South India. b. (52)

un-meeN-see eek-nee kah-aa thaa ki rah-nee-kee liyee 3pl-in-from one-erg say-pf was that stay-inf-for kanaanor bahut acchii jagah hai Kananor very good place is ‘One of them(i) said that Kananor is a very good place to stay.’ (Rakesh 1961: 10)

vee laRkee bahut caalaak haiN, un-meeN-see/* har eek meeraa those boys very clever are 3pl-in-from each one my chaatr hai student is ‘Those boys are very clever, each one of them is my student.’

The problem of coindexing two DPs is not that one of them is relative, but rather that indefinite DPs have restrictions on an antecedent, and a relative antecedent is just one type of unsuitable antecedent, among others. Another argument against the movement analysis addresses the question of whether the moved DP must be identical to what it is in its original site. Anaphoric coindexing does not require absolute identity. We see examples in both Hindi/Urdu cf. (53) and Sanskrit, cf. (54) in which the relative phrase is not quite identical to the correlate. (53) a. [joo chaatr wahaaN khaRaa hai] woo laRkaa meeraa doost hai. rel student there standing is that boy my friend is ‘The student who is standing there, that boy is my friend.’

 Alice Davison

b. *[joo laRkaa wahaaN khaRaa hai] woo chaatr meeraa doost hai. rel boy there standing is that student my friend is ‘The boy who is standing there, that student is my riend.’(McCawley 2004, 300.) (54)

[yád adhyá tvā sūryo pabrávāmahai] rel.acc today 2s.acc Surya.voc speak.about.imper.1pl tám no devu ánu mamsīrata krátum that.acc.m 1pl.dat god.pl.nom ask.for.3pl goal.acc.m ‘[That which we must speak to you about, O Surya,] the gods should ask us for that goal.’ (RV 10, 37,5c, Hettrich 1988, 535.)

In Hindi/Urdu, the common noun laRkaa ‘boy’ is a superset of chaatr ‘student’, and linear order of more specific-less specific is well-formed (53a). The reverse in (53b) is not.25 In Sanskrit, mismatches of the relaive and correlate are common; in (54) there is a mismatch of gender between the neuter relative D and the masculine correlate. The mismatch poses a serious problem for analyses of correlatives as extracted form the main clause, leaving an overt trace in the form of the correlate pronoun, as McCawley (2004) notes.26 Bhatt’s analysis places the correlative within a DP which has an overt head, a pronoun or an NP. There would have to be a separate constraint which would not be motivated by general conditions on discourse, requiring the specific-general order in (53a). The optional CP movement to TP proposed by Bhatt 2003 suffers from a lack of motivation. If movement is an instance of scrambling, it is not clear why the moved version adjoined to TP is the unmarked case. If it is quantifier raising, it is not clear why it is optional, or indeed what motivates it. In languages like Hungarian, quantifiers must be moved overtly to a functional projection below CP to express wide scope over another quantified phrase which c-commands the moved quantifier (Szabolcsi 1997). But in the case of Hindi/Urdu correlatives, there is no c-commanding quantifier in TP over which the correlative must have scope.27 If correlative movement is some other kind of A'movement, then its motivation and optionality need to be further worked out. In this section, I have discussed stipulations which would have to be made to derive the patterns of iterated relative clauses which I outlined above. These stipulations are drawbacks to Bhatt’s (2003) movement analysis for Hindi/Urdu. I propose instead a base-generated analysis which directly adjoins the correlative to a clausal projection in both Sanskrit and Hindi/Urdu. Movement of a correlative from DP is not an option in Sanskrit, as relative clauses are not found sentence-internally, adjoined to NP/DP, and NP adjunction seems to appear relatively recently in Hindi/Urdu. My proposal gives a uniform analysis of both languages, with one specific difference correlated to difference of interpretation and grammaticality.



Sanskrit and Hindi/Urdu correlatives 

4. Feature valuation and anaphora in a base-generation analysis In this section, I work out a feature-based analysis which makes use of an anaphoric link between the correlative clause and its correlate(s), and a semantic feature [PRED]. There is also a feature which motivates movement of the relative DP within the correlative clause. This analysis is based on a base-generated adjunction of the correlative.

4.1

Features common to correlatives in Sanskrit and Hindi

The first feature I propose is a semantic feature, [PRED]. It is a feature of C in correlative clauses which requires the correlative to be a restrictive (intersective, maximalizing) predicate on the correlate in the host clause (Grosu 2002). It is an instruction to the interpretative interface to translate the finite correlative clause as a predicate intersecting with the set defined by the common noun expression. It is in some ways similar to Rizzi’s feature Rel, attached to C, which distinguishes relative clauses from questions and declarative complements (Rizzi 1990: 67). Rizzi’s feature is both a motivation for movement of a wh-phrase and an instruction for interpretation. Below I separate these functions. [PRED] is equivalent to the Λ feature on C in Scottish Gaelic and Irish (Adger and Ramchand 2005). It requires the translation of a CP or the chain of Λ-marked Cs as a predicate (λ expression) on the head NP, at the Conceptual-Intentional Interface. The Λ feature is lexically encoded in Gaelic/Irish by complementizers used only for restrictive relative clauses. It is obligatorily associated with left-adjoined Hindi relatives (correlatives), but this association so far follows from no independent principle (cf. Grosu 2002). Motivation for movement within the relative clause is due to the feature [uRel] on C, which is checked by the relative determiner in the correlative clause. It motivates movement of the relative (phrase) to the CP projection (mostly) obligatorily in Sanskrit, but optionally in Hindi. Within the correlative CP in Sanskrit and Hindi/Urdu, the relative D(P) moves to a left peripheral position in C [Force according to Rizzi 1997]. The feature [uF:Rel] on C is strong (Sanskrit) and either strong or weak (Hindi). It attracts D(P)s which have the relative series of determiners with [iF:Rel]. If the feature is weak, then [uF:Rel] is valued by Agree by the [iF:Rel] in the relative phrase in situ.

4.2

Anaphoric linking

In headed relatives the relative clause is syntactically adjoined to an NP (Safir 1986). A correlative clause and its correlate are anaphorically linked, and are not necessarily

 Alice Davison

adjacent in narrow syntax. The correlate contains a determiner, which can be deictic or anaphoric (55): (55) Deictic/anaphoric determiners a. Sanskrit (i) sás ‘that’, tá ‘that’ (ii) tátra ‘there’ (etc.) b. Hindi (i) woo ‘that’ (distal in contrast to deictic/proximal yah ‘this’) (ii) wahaaN ‘there’ (distal in contrast to deictic/proximal yahaaN ‘here’) An anaphoric relation is established between the correlative clause and its correlate. The deictic pronouns are anaphorically dependent, with an ID feature valued by an antecedent, in this case the correlative clause, which bears the index of the RelP in its specifier (by move or Agree). This relation identifies the correlate to be modified by the correlative clause.

4.3

The derivation of correlative structures

Relative determiners have the feature [iF:Rel]. C is specified [uF:Rel]. If [Rel] is present in Hindi, then the semantic feature [PRED] is possible. In Sanskrit, however, C has only the feature [uF:Rel]. We can represent this difference by saying that [PRED] is present in the lexicon of Hindi/Urdu, and so it can be part of the array in the numeration. In Sanskrit, it is not a lexical item, and as a result, it is only a default value inserted at the interpretative interface. In narrow syntax, relative D(Ps) move to Spec/CP within the CP projection, checking [uF:Rel]. Correlative CP is merged with the host clause, according to what is possible in the two languages. Sanskrit has an independent prohibition on syntactic subordination, preventing asymmetric adjunction to TP. The derivation in both languages proceeds with two differences, (i) the absence of [PRED] in Sanskrit, and its obligatory presence on left-adjoined correlatives in Hindi, and (ii) the different adjunction relation of the correlative CP to its host.28 After merger of the correlative to its host, an anaphoric link is established between the correlative and the (local) correlate, valuing the [uF:ID] feature of the deictic element of the correlate, according to the index of the correlative specifier.29 The operation Agree values the [ID] feature of the correlative. In Sanskrit, the dependence need not be local, allowing for longer distance relations in the stacked and iterative uses of correlatives; their counterparts in Hindi/Urdu are ungrammatical.  It is not clear whether Agree is subject to the Phase Impenetrability Condition (Chomsky 2004). But we can speculate that Agree is not limited by conditions like Phase Impenetrability in Sanskrit, but it may be so limited in Hindi/Urdu. I return to this issue in 4.5 below.



Sanskrit and Hindi/Urdu correlatives 

4.4

The interpretative interface

In Hindi, the feature [PRED] has been part of the derivation from the numeration onwards, and its presence requires a restrictive predicative interpretation of relative CP in conjunction with the correlate, blocking other interpretations. It is interesting to speculate that the translation of the correlative clause to a predicative lambda expression creates the need for an argument, found locally in the host clause, a consequence of the asymmetric c-command condition in (34). But this discussion cannot be conclusive without more independent evidence. In Sanskrit, the absence of [PRED] means that the correlative is free to have an appositive interpretation or indeed others, such as conditional or interrogative (see Davison 2009). Information in the main clause defines the interpretation of the relative. For example, the appositive interpretation is possible if the correlate is a personal pronoun or a proper name, the relative clause is translated as a sentence with a variable. The variable is co-valued with the pronoun or proper name, as in a discourse sequence. See the examples (11), (30c) above for correlatives which have an appositive interpretation because the correlate is a proper name or pronoun, and the relative is co-valued with it, as in (56). (56) Susan, who is coming today, will see you ==> Susan will see you; x is coming today Co-valuation: x = Susan In other instances of correlatives, correlatives in Sanskrit have interrogative interpretation, recall (21)-(22). This interpretation is determined by the matrix predicate if the matrix predicate selects C[Q]. In that case, the relative specifier is reanalyzed as interrogative in (57), because of the clash of the features on C and its specifier. (57)

CP 3 DP[Rel]→[Q] C’ 3 C[Q]

If there is no specification in the correlative structure which requires an appositive or interrogative interpretation, then in Sanskrit the default feature [PRED] is inserted at the interface, yielding the restrictive interpretation. Otherwise the correlative will lack a well-formed interpretation because of underspecification. In sum, I have made use of several syntactic features in the derivation of correlative constructions. [uF:Rel] motivates movement or Agree linking a relative DP to C of the correlative clause. [ID] expresses the referential dependence of the deictic D in the correlate, and its valuation forms an anaphoric chain (Adger and Ramchand 2005), which identifies the correlate. To capture the restrictive use of correlatives, I adopt the privative feature [PRED] (Grosu 2002). It captures the restrictive predicative nature of

 Alice Davison

the relative clause (Safir 1986, Adger and Ramchand 2005). There is empirical evidence for its existence in the correlation between relative clause forms and their interpretation, and in the Sanskrit-Hindi/Urdu difference. The derived structure for Hindi/Urdu is represented in (58): (58)

CP 3 C’ 3 TP wo CP* TP 3 3 RelXP C’ XP* 3 C[Rel] TP 3 ...RelXP...

The chain linking CP* and the correlate XP* is an A'-chain derived by base-generation and indexing, rather than by movement.

4.5

Locality

Locality constraints apply to A'-chains; the tail of the A'-chain cannot be within certain syntactic configurations, such as a Complex NP. Movement-derived A'-chains are clearly subject to this constraint, one of the arguments which Bhatt (2003) uses to support his movement analysis of Hindi correlatives. Here I discuss how a Merge-derived A'-chain could be subject to the CNPC. First I want to note that the movement chain and the Merge chain I propose are somewhat different. Bhatt (2003) moves the relative clause from an adjoined position within DP, leaving an unpronounced copy. I propose that there is a chain of category links (derived by Agree), which is directly composed of the relative CP and its correlate, often realized as a prononimal (55). Demonstratives as pronouns can have the familiar bound variable reading, if a quantified antecedent c-commands them, cf. (59a,b). The correlate or bound pronoun is within a complement clause subordinate to the matrix clause. (59) a. har laRkaa sooc-taa hai [ki woo caalaak hai] each boy think-impf is that 3s clever is ‘Each boy thinks [that he is clever].’



Sanskrit and Hindi/Urdu correlatives 

b.

[joo (laRkaa) wahaaN rah-taa hai] maiN (yah) sooc-tii huuN rel boy there stay-impf is I this think-impf am [ki woo caalaak hai] that 3s clever is ‘[Which boy lives there], I think (it) [that he is clever].’

Relative XPs have quantifier properties, but Bhatt (2003) shows that the relative construal with a correlate is not simply a matter of a quantifier and a pronoun which gets a variable interpretation: the bound variable interpretation as well as a correlate interpretation are possible in subordinate complement clauses (59a,b)-(60a), but not within another relative clause (60b). (60) a.

har laRkee-koo [woo kahaanii [joo arundhati-nee us-kee each boy-dat that story which Arundati-erg 3s-gen baaree-meeN likhii hai]] pasand hai about write-pf is liked is ‘Each boy likes [the story [which Arundhati wrote about him]].’

b. *[joo (laRkaa(i)) wahaaN rah-taa hai] mujh-koo rel boy there stay-impf is I-dat [woo kahaanii(j) [ joo(j) arundhatii-nee us(i)-kee baaree-meeN that story rel Arundhati-erg 3s-gen about likhii]] pasand hai write.pf liked is ‘[Which boy lives there], I like [the story [which Arundhati wrote about (him)]]’ (after Bhatt 2003: 300) These examples show that a quantifier can bind a pronoun inside a complement clause and a complex NP, but a relative cannot have its correlate within a complex NP, though it can be coindexed with a correlate in a plain complement clause. Quantifiers seem to have a weaker condition than correlatives on binding a pronoun in a complex DP. The features I have proposed for relative clauses construed with correlates are essential for defining the kind of relation which involves a relative CP and a correlate. The [PRED] feature in particular requires an argument, the correlate. Note that there are two [PRED] construal relations in (60b), one between the initial relative clause and woo, the other within a DP between joo and the head NP, which are in Safir’s (1986) predication binding relation. Taken literally, the syntactic correlate of the initial clause [joo laRkaa...] in (60b) is the complex DP [woo kahanii(j) joo(j)...], which has a different index and NP description, so it is not modified by the correlative. Rather the intended reading is that the correlate is us(i). The predicate-argument relation is established on the basis of the coindexing described above, provided there is no intervener like the complex DP in (60b). The correlate may be within a complement clause, as in (59b). The complement is linked with a theta position, and as a clausal constituent it has no individual index or common noun description.

 Alice Davison

I explain the ungrammaticality of (60b) as a conflict of descriptions and indices at the interface, rather than a violation of a syntactic locality condition. Interestingly, a relative clause can be construed with a correlate inside another relative clause, as in (36) above, which should also be a violation of the Complex DP island constraint.30 There is a residue of cases involving crossover violations and bound variable binding. Some of the data are complex and subject to speaker variation, others not. The analysis of the conditions depends greatly on how logical form is conceived of in general, which is outside the scope of this paper. I will leave these interesting problems for future research. I will conclude that, aside from this set of examples, there is no convincing evidence which requires movement. I see no reason to prohibit movement, if it is motivated by a need to check formal features, and I have made use of an EPP feature on C which requires movement of relative phrases to the clause periphery in both Sanskrit and Hindi/Urdu.

5. Summary and conclusion The goal of this paper was to make a case for the base-adjunction of correlative clauses to a host clause, in both Sanskrit and Hindi/Urdu (as opposed to base-adjunction in Sanskrit, and movement from DP in Hindi/Urdu, as in Bhatt 2003). In the preceding sections, I have offered evidence that the two languages are sharply different in the way they treat correlatives. First, there is evidence for a difference of adjunction: correlatives are symmetrically adjoined to CP, while Hindi/Urdu has asymmetric adjunction to TP. Sanskrit allows the clause initial string of functional heads on both correlative and host clauses, while Hindi/Urdu strictly limits wide scope questions to the host/ matrix clause. In this respect, Hindi/Urdu follows Minimality, while Sanskrit appears not to. Sanskrit has no syntactic subordination, substituting the relative clause construction for it. Second, iteration of correlatives is free in Sanskrit, allowing discourse dependence between correlatives and a distant correlate. Hindi/Urdu disallows multiple relatives on the same correlate, and allows iteration only if there is no ‘skipping’ of clauses with a correlate. In spite of these differences, I analyze the syntactic derivations in exactly similar terms, with three specific differences: (i) the adjunction relation, (ii) the existence of the restrictive feature [PRED] as a lexical choice in Hindi/Urdu, and as a default interpretative value in Sanskrit and (iii) a c-command condition (34) on correlatives in Hindi/Urdu, which might, if explored, further turn out to be a consequence of the need for an argument for the relative clause as predicate. There is no motivation in Sanskrit for movement of correlatives from DP, as this analysis would require syntactic subordination of DP-internal relative clauses. Sanskrit does not have syntactically coded subordination, nor adjunction of correlatives to DP/ NP. As Bhatt (2003) points out, there is also no motivation for movement of correlatives with multiple relatives. There are several categories of evidence which would suggest



Sanskrit and Hindi/Urdu correlatives 

movement, but the analysis requires a number of stipulations to work. These stipulations block iteration of correlatives, and limit the kind of quantifiers that can occur with the correlate. I derive these limitations from movement-independent factors. The distinction between good and bad iterations derives from the c-command condition on correlatives merged successively in a local relation, offering a more general account of iteration, including cases not discussed by Bhatt. The restrictions on what kind of quantifier can be in the correlate seems to follow from a general condition on discourse anaphora, matching the anaphoric coindexing component I proposed for correlatives; the anaphoric links between correlative and correlate identify the argument of the predication relation. While correlatives have properties of generalized quantifiers, there seem not to be sufficient motivation to raise them. Overt or covert raising derives a scope relation which would not otherwise be possible to express. Some important issues are left unresolved, which have to do with binding theory and reconstruction. Nevertheless, making a case for base-adjunction has revealed alternative explanations for the facts and focused attention on some aspects of Indic syntax which have not been part of the discussion. The base-adjunction analysis captures the continuity between the older language and a modern language, in which changes of syntactic structure seem to be relatively recent. The difference of syntactic structure are ones which can be expressed in Minimalist terms.

Acknowledgements Many thanks to Anikó Lipták and an anonymous reviewer for very detailed and helpful critical comments, which have much improved the first version. Earlier versions of this paper were presented at the University of Iowa Linguistics Colloquium, the 5th Asian Meeting of GLOW, Jawarharlal Nehru University in New Delhi, the Diachronic Generative Syntax Conference, University of Trieste, the Linguistics Colloquium, University of Venice and the Symposium on Historical Linguistics, University of Illinois, Champaign-Urbana. I thank the audiences at these presentations for helpful discussion. I would like to thank the following for their generous help in supplying information and for helpful discussion: Josef Bayer, Elena Bashir, Robert Chametzky, Priyanka Chaturvedi, Probal Dasgupta, Hans Hock, Makur Jain, Arpita Kumar, Servachetan Katoch, Paula Kempchinsky, Roumyana Slabakova, Peter Scharf, Frederick Smith. Thanks also to D.J. Berg, S. Cassivi, J. Ferguson, M. Goodheart, Y. Romero and M. Yao. I am grateful to the Obermann Center for Advanced Study for support for the writing of this paper in 2007.

References Adger, David & Gillian Ramchand. 2005. Merge and move: Wh-dependencies revisited. Linguistic Inquiry 36. 161–193. Andrews, Avery. 1976/1985. Studies in the syntax of relative and comparative clauses. New York: Garland Press.

 Alice Davison Bhatt, Rajesh K. 2003. Locality in correlatives. Natural Language and Linguistic Theory 21. 485–541. Chantraine, Pierre. 1958. Grammaire Homérque. Tome II, Syntax. Paris: Klinksieck. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 2004. Beyond explanatory adequacy. In Adriana Belletti (ed.), Structures and beyond. 104–131. Oxford: Oxford University Press. Dasgupta, Probal.  1980. Question and relative and complement clauses in a Bangla grammar. Unpublished dissertation. New York University. Davison, Alice. 2009. Correlative clause features in Sanskrit and Hindi/Urdu. In Paola Crisma & Giuseppe Longobardi (eds), Historical syntax and linguistic theory. 271–291. Oxford: Oxford University Press. Dayal, Veneeta. 1996. Locality in Wh-quantification. Dordrecht: Kluwer Academic Publishers. Delbrück, Berthold. 1888. Altindische syntax. Verlag der Buchhandlung des Waisenhauses. den Dikken, Marcel. 2005. Comparative correlatives comparatively. Linguistic Inquiry 36 (4). 485–541. den Dikken, Marcel. This volume. Comparative correlatives and successive cyclicity. Etter, Annemarie. 1985. Die Fragesätze im Rgveda. Berlin: Walter de Gruyter. Gonda, Jan. 1975. The original character of the Indo-European relative pronoun io-. In Jan Gonda, Selected studies. Vol. 1, 164–199. Leiden: E.J. Brill. Grosu, Alexander. 2002. Strange relatives at the interface of two millennia. Glot International 6.6. 145–167. Hajati, Abdol-Khalil. 1977. Ke constructions in Persian: descriptive and theoretical aspects. University of Illinois, Champaign-Urbana dissertation. Hale, Mark. 1987. Studies in the comparative syntax of the oldest Indo-European languages. Harvard University dissertation. Hermann, Eduard. 1895. Gab es im Indogermanischen Nebensätze? Zeitschrift für vergleichende Sprachforschung auf dem Gebiete der indo-germanischen Sprachen 33. 481–534. Hettrich, Heinrich. 1988. Untersuchungen zur Hypotaxe im Vedischen. Berlin: de Gruyter. Hock, Hans H. 1982. The Sanskrit quotative: a historical and comparative study. Studies in the Linguistic Sciences 12:2. 39–85. Hock, Hans H. 1989. Conjoined we stand: theoretical implications of Sanskrit relative structures. Studies in the Linguistic Sciences 19 (1). 93–126. Jacobson, Pauline. 1983. On the syntax and semantics of multiple relatives in English. Bloomington: Indiana University Linguistics Club. Jayekar, Pupul. 1992. Indira Gandhi. Excerpt in Hindi translation. India Today. Kayne, Richard. 1994. The Antisymmtry of syntax. Cambridge: MIT. Press. Kiparsky, Paul. 1994. The Indo-European origins of Germanic syntax. In Ian Roberts & Adrian Battye (eds), Clause structure and language change. 140–167. Oxford: Oxford University Press. Lahiri, Utpal. 2002. Questions and answers in embedded contexts. Oxford: Oxford University Press. Lipták, Anikó & Georges Rebuschi. To appear. Types and distribution of northern Basque correlatives. 5th Asian Glow proceedings. New Delhi. Mahajan, Anoop. 2000. Relative symmetries and Hindi correlatives. In Artamis Alexiadou, Andre Meinunger, Chris Wilder & Paul Law (eds), The syntax of relative clauses, 201–230. Amsterdam: John Benjamins. Marlow, Patrick. 1997. Origin and development of the Indo-Aryan quotatives and complementizers: an areal approach. Unpublished dissertation. University of Illinois, Champaign-Urbana.



Sanskrit and Hindi/Urdu correlatives  McCawley, James. 2004. Remarks on adnominal, adsentential and extraposed relative clauses in Hindi. In Veneeta Dayal & Anoop Mahajan (eds), Clause structure in South Asian languages, 291–313. Dordrecht: Kluwer Academic Publishers. Rakesh, Mohan. 1961. aakhirii caTTaan-tak. Delhi: Raajkamal prakaashan Private Limited. Rebuschi, Georges. This volume. Basque correlatives and their kin in the history of Northern Basque. Reinhart, Tanya. 2006. Interface strategies. Cambridge: MIT. Press. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge: MIT. Press. Rizzi, Luigi. 1997. On the fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of grammar, 281–337. Dordrecht: Kluwer Academic Publishers. Safir, Kenneth. 1986. Relatrive clauses in a theory of binding and levels. Linguistic Inquiry, 17 (4). 663–689. Schäufele, Steven W. 1990. Free word-order syntax: the challenge from Vedic Sanskrit to contemporary syntactic theories. Unpublished dissertation, University of Illinois Champaign-Urbana. Subbarao, Karumuri V. 1984. Complementation in Hindi syntax. Delhi: Academic Publications. Szabolcsi, Anna. 1997. Strategies for scope taking. In Anna Szabolcsi (ed.) Ways of scope taking. 109–154. Dordrecht: Kluwer Academic Publishers.

Endnotes 1.

The following abbreviations are used in the glosses in the examples:

1: 1st person; 2: 2nd person; 3: 3rd person; abl: ablative; acc: accusative; aor: aorist; caus: causative; cont: contingent; dat: dative; desid: desiderative; erg: ergative; f: feminine; fut: future; gen: genitive; imper: imperative; impf: imperfective; inf: infinitive; inst: instrumental; int: interrogative; loc: locative; m: masculine; middle: middle voice; neg: negative marker; nom: nominative; opt: optative; part: participle; pf: perfective; pl: plural; ptcl: particle; pres: present; prog: progressive; rel: relative; s: singular; voc: vocative. 2. See papers by Lipták and Rebuschi (to appear), Rebuschi (this volume) for an overview of the various relative clause structures in Basque. They have both clausal and right-headed DPadjoined relatives, with sharply differing properties. 3. This sentence is true if the unique individual at the intersection of girls and standing people is a member of the set of tall people (Dayal 1996: 156). 4. See Dayal (1996) for a discussion of the semantic difference between (7b) and (7c). 5. The translation into a logical language does not distinguish subordinate clauses syntactically, as differences of bracketing with & make no difference in truth value. Logical conjunction is also symmetric both syntactically and semantically. 6. The correlate is the subject of the verb in the correlate clause. It is expressed as morphological features on the verb. 7. Sanskrit freely allows both left and right adjunction of correlative clauses, with restrictive or appositive interpretation. Restrictives tend to precede, appositives to follow, but actually both orders are possible for the two interpretations. Discourse factors seem to be important in which clause is first. See Hettrich (1988) for discussion.

 Alice Davison 8. An anonymous reviewer notes that the following sentence is grammatical: (i) jo tumheeN kitaab caahiyee, woo meeree-paas nahiiN hai rel you.dat book want that me-near neg is ‘I don’t have the book you want.’ 9. Throughout this paper I will use referential indices to make clear which XPs are coreferent. This convention is for clarity of exposition. 10. This sentence seems to be unique in the corpus of Vedic Sanskrit, but it must be presumed to be grammatical.  This assumption is at least partially supported by the existence of other anomalous correlatives, such as the examples cited in Delbrück (1888: 365), with a 3ppl imperative apparently within a relative clause. While there is an alternative construal which places the imperative outside of the relative clause (see Hettrich 1988: 140), it appears that genuine imperatives within correlatives occur in the Classical Sanskrit Mahabharata (H. Hock, p.c.). 11. Rizzi (1997) shows that Focus and Topic follow relatives, but precede interrogatives, and so we have to assume that Force distinguishes relatives from complement clauses, and Finite distinguishes questions from statements. Force is not to be taken as referring to illocutionary force. 12. This would be ForcePhrase in the structure shown in (25), which is abbrieviated as CP in the schematic structure in (13). 13. This kind of stacking persists into Classical Sanskrit works such as Shakuntala; it is normal for ‘praise’ of notable figures. 14. The asymmetry guarantees an unambiguous construal relation. 15. Interestingly, Jacobson (1983) shows that both question and relative DP can be modified by relative clauses in English: (i) Who do you like who/that you invited? (ii) Who who/that you invited do you like? (Jacobson 1983) 16. Sentences of this sort can be improved by using a specific intonation contour. The ‘middle’ clause can be given a lower intonation than the surrounding clauses, somehow making it less of an intervener between the first relative and its correlate in the last clause. 17. We may speculate, that in the ill-formed cases, the ‘low adjoined’ correlative in (35) forms an independent CP phase, which is spelled out separately from the phase containing the corresponding correlate before the relative CP modifier finds its argument, the correlate. 18. The inverse pattern is found in Sanskrit, in which multiple relative DPs are linked to separate peripheral correlate clauses (Hock 1989: 109), which would involve multiple CP adjunctions: (i) [I] sā(i) vai daivī vāg [II] yayā(i) yad yad(j) eva vadati that ptcl divine speech [II] rel-inst if rel ptcl say.pres.3s [III] tad tad(j) bhavati then that become.pres.3s ‘Divine speech is that such that if you speak something by means of it, then that comes about.’ ( BAU 1.3.27, Hock 1989, 109)

(ii) [DP(i).... ] [ Rel(i) Rel(j)... ] [DP(j)...]



Sanskrit and Hindi/Urdu correlatives 

(iii)

CP



wo



CP

CP



3

3



CP

CP

CP

This kind of structure is clearly ruled out for Hindi/Urdu, assuming asymmetric category adjunction. TP-adjoined structures would violate local c-command. The multiple relative clause would be predicated of one sister TP with just a single correlate, but it could not also simultaneously be predicated of the higher TP, with the other correlate. 19. I have not included a discussion of right-adjoined relatives, taking no position here about derivation in Hindi/Urdu by base-generation or rightward movement. They are freely available in Sanskrit, apparently like left-adjoined correlatives in syntax and interpretation. For discussion of these right-adjoined relatives in Hindi/Urdu and how they differ from left-adjoined correlative, see Subbarao (1984) and Dayal (1996). 20. Interestingly, Marlow (1997) proposes a historical derivation of Hindi/Urdu NP-adjoined relatives from a left correlative clause by a process of leftward movement and copying of the N from a correlative, with reanalysis, rebracketing the N and the remnant correlative as a constituent. He uses the same ideas of leftward movement, copying and deletion as Mahajan, in the opposite direction. 21. See Srivastav (1991). This structure seems to have special discourse properties, best exemplified in contexts of emphasis or contrast, as in Bhatt (2003). If so, it is not likely to turn up in the literary texts which are the sole records of earlier Hindi. In this genre of text, relative clauses are either sentence-peripheral correlatives, or (beginning approximately in the 18th century) adjoined to the right of NP (see Davison 2009). 22. From a historical perspective, this suggests a change in phrase structure in early Hindi. The NP-CP structure is known from the 18th century on, presumably reflecting the influence of the Persian NP-CP relative structure with the complementizer ke instead of a member of the set of relative pronouns (Hajati 1977). In Hindi, this construction took on the function of expressing appositives, which otherwise were expressed by a relative clause to the right of the matrix. A further innovation would be left adjunction of CP to D, with only restrictive interpretation. 23. Bhatt (2003: 507) gives several examples of different combinations of two relatives which, with different sentence structure, would be well-formed by condition (34) (his 32), or would violate it (his (34a,b) if both relative clauses were moved to the left. The sentence (33b) in Bhatt (2003: 508) is marked grammatical, but would seem to violate my condition (34). I leave this question for future investigation. 24. To rule in sentences like (36), which Bhatt does not discuss, there would have to be an unusual combination of factors, distinct from the stacking of two relatives on the same DP. To derive (36) relatives adjoined to two DPs would each raise to TP. The first relative to move would have to contain the correlate of the second relative to move. 25. This order is similar to the name-epithet coindexing in discourse (i)-(ii): (i) John is leaving, and I won’t miss the idiot. (ii) *The idiot is leaving and I won’t miss John.

 Alice Davison 26. Adger and Ramchand (2005) argue against a movement analysis of relatives and questions Celtic languages by showing that the head of the chain and the foot may differ in features of definiteness and case. 27. Reinhart (2006) proposes that covert quantifier raising is possible only when the intended scope reading is not otherwise possible; the use of choice functions accounts for many cases of scope relations without movement. 28. An alternative Merge in Hindi/Urdu is to the right of NP, with or without [PRED]. These headed relatives have a restrictive or appositive interpretation. There is also the option of left Merge to DP. 29. Alternatively this indexing could take place at the interpretative interface, for extra-sentential antecedents. In any case, the referential dependence of the deictic elements must be remedied by indexing. 30. All of this discussion is highly speculative, as much remains to be discovered about the relative constructions in Indic languages. The construal into a complement clause in (52b) may be possible because the complement clause is a phase with covert focus movement of the correlate into an edge position from which it accessible to construal with [PRED]. Finite clauses are phases, which are strictly limited for interrogative scope (ia). Non-finite clauses are (obligatorily) transparent to interrogative scope (ib). There are indications that relative operators have the same properties vis-à-vis finiteness. (i) a. aap sooctee haiN [ki vee kyaa kar rahee haiN] you think.impf are that 3pl what do prog are *What are you thinking [that they are doing e]? = ‘You are thinking about what they are doing.’ b. aap [pro kyaa kar-nee]-kii sooctee haiN? you what do-inf-gen think.impf are ‘What are you thinking [of pro doing e]?’ = *‘You are thinking about [what to do e].’

Comparative correlatives and successive cyclicity Marcel den Dikken The topic of this paper is the locality restrictions on Aʹ–dependencies involving the fronted comparative in the two constituent clauses of the Dutch comparative correlative. Long-distance dependencies across a finite clause behave differently in the two clauses. They are grammatical in the relative clause; but there is considerable speaker variation regarding long-distance dependencies in the headclause: about one third of the informants in a detailed questionnaire study found these to be degraded, with the distribution of the judgements being sensitive to the choice of correlative particle used in the headclause. Though the empirical picture may at first sight appear to be chaotic, the facts fall into place once it is realised that whenever the comparative lands in SpecCP (as it does in the relative clause), long-distance dependencies are grammatical, via successive-cyclic movement through intermediate SpecCP positions, but when it lands in an IP–adjoined position (as it often does in the headclause), severe locality restrictions emerge. Besides emphasising the role played by principles of Universal Grammar in the analysis of comparative correlatives, the results of this study also illuminate our general understanding of what constrains longdistance dependencies in the syntax. Keywords: comparative correlative, locality, Aʹ–dependency, successive cyclicity, unambiguous binding

1. Introduction: The problem Comparative correlatives (see (1) for simple illustrations from English and Dutch) belong to the family of correlative construc­tions (Den Dikken 2005a), though their distribution is cross-linguistically wider than that of ‘ordinary’, non-comparative correlatives: present-day English and Dutch have the former but not the latter.

 Marcel den Dikken

(1) a. the more you read, the less you understand b. {hoe/des te} meer je leest, {hoe/des te} minder how/the-gen deg more you read how/the-gen deg less je begrijpt (Dutch) you understand ‘the more you read, the less you understand’ Srivastav (1991) has presented seminal evidence for an analysis of correlative constructions wherein the initial clause functions as a (free) relative clause adjoined to a headclause containing a correlative particle:1 (2) a. standard correlative (Srivastav 1991) [headcl [relcl rel/wh-operator...] [headcl correl-prt...]] b. comparative correlative (Den Dikken 2005a)2 [headcl [relcl rel/wh-operator+cpr...] [headcl correl-prt+cpr...]] For the examples in (1), this entails that the initial clause is an adjoined (free) relative clause, with hoe or des (the latter being an archaic genitive case form of the definite article/demonstrative; see also Citko, this vol., on the ‘frozen relative pronoun im’ in Polish comparative correlatives) in Dutch (1b) serving as a relative pronoun, and that the second clause is the headclause, with hoe or des being the correlative particle. Though in the English example in (1a), neither the relative pronoun nor the correlative particle is overt (English the in (1a) is the equivalent of Dutch te in (1b), a syncategorematic degree particle; see Den Dikken 2005a for dis­cuss­ion), there is support for a correlative structure here as well. Thus, it is easy to show that the clause-initial comparatives in both constituent clauses of comparative correlative constructions end up in clause-initial position as a result of syntactic movement, exactly as in correlative constructions generally: the fronting of the comparatives is subject to locality conditions familiar from the domain of A'–movement (see esp. Culicover & Jackendoff 1999: 555–56; Taylor 2006): (2) the more she {believes/*regrets} (*the claim) that you read, the less she {believes/*regrets} (*the claim) that you understand

(3) what she {believes/*regrets} (*the claim) that you read

In both the sentence-initial clause and the headclause of the comparative correlative, attempts at establish­ing a dependency involving the initial comparative across a nonbridge verb complement or a complex noun phrase fail completely, just as in the case of garden-variety A'–movement dependencies such as the one in (3), involving either free relativisation (as in what she believes that you read is disgusting/interesting/wellwritten) or embedded wh-question formation (what she believes that you read is unclear/uncertain/unknown). For English, it is not very difficult to find ‘real-life’ cases of grammatical long-distance fronting in non-island contexts in comparative correlatives, in either the first or



Comparative correlatives and successive cyclicity 

the second clause. Here are some Googled examples, all with think in the matrix of the long-distance dependency: (4) long-distance dependency in the relative clause a. the less you think that you know, the better you can receive, the more enthusiastic you can be about what you don’t know [www.redbullmusicacademy.com/ tutors.9.0.html?act_session=84] b. the more you think that they intended to bag, the more hubristic seems their undertaking [www.austlii.edu.au/au/journals/mulr/1999/1.html] c. the more people you think you’ve “loved,” the greater the likelihood that... [decime.blogspot.com/2006_02_12_decime_archive.html] (5) long-distance dependency in the headclause a. the more you learn about something, the less you think that you know [www.nsftools.com/blog/blog-05–2003.htm] b. the more things you do the more things you think you can do [www.insearchofheroes.com/internet-heroes/pages/craig%20garber%20 hero.html] c. the more you get done, the more you think that you gotta get done [www.lostinbrooklyn.com/work/writing/smugwriting02.html] It seems clear, therefore, that English readily allows long-distance dependencies involving the comparative in these constructions, in either of the constituent clauses. But for Dutch, the picture is less straightforward. The two sentences in (6), translation equivalents taken from a website containing sermons in a variety of languages, sug­gest that some English long-distance dependencies have grammatical counterparts in Dutch.

(6) a. the more you think you could do, the more you employ your gifts to the very limit [www.tscpulpitseries.org/english/undated/tsfaith.html] b. hoe meer je denkt dat je kunt doen, hoe meer je je gaven how more you think that you can do how more you your gifts tot het uiterste benut to the extreme employ [www.tscpulpitseries.org/nederlands/gf000004.htm] The Dutch example in (6b) involves a long-distance dependency across a finite clause inside the relative clause of the comparative correlative. Such dependencies are accepted by the vast majority of the informants in a questionnaire study I conducted for this paper on long-distance dependencies in Dutch com­parative correlatives (see the appendix for discussion of the questionnaire): my test example in (7) (sentence (11) of the questionnaire) was found to be acceptable by all but one of the respondents.

 Marcel den Dikken

(7)

hoe meer je denkt dat je begrijpt, hoe minder je in how more you think that you understand how less you in feite begrijpt fact understand ‘the more you think you understand, the less you actually understand’

However, examples of long-distance dependencies in the headclause of the Dutch com­parative correl­ative are very difficult to find. The attested example in (8), featuring extraction of hoe minder out of the in­fini­tival complement of denkt ‘think-2sg’ in the headclause, is grammatical; a structurally parallel constructed example that I included in my question­naire (see (9a) = (41) in the questionnaire) was almost universally accepted by the speakers that I queried.3 (8) hoe meer je leert, hoe minder je denkt te hoeven leren how more you learn how less you think to need learn [www.wijnstokgemeente.nl/wordfortoday0444.pdf] (9) a. hoe meer je leest, hoe minder je denkt te begrijpen how more you read how less you think to understand b.

%hoe

c.

%hoe

meer je leest, hoe minder je denkt dat je begrijpt how more you read how less you think that you understand meer je leest, des te minder je denkt dat how more you read the-gen deg less you think that je begrijpt you understand

But speakers are con­siderably less unanimous when it comes to sentences of the type in (9b) (= (21) in the questionnaire), involving extraction across the finite complement of denkt in the headclause. About one third of my informants found (9b) to be degraded in comparison with the corresponding example not involv­ing a long-distance dependency in the headclause; and I have not found any examples of this type on the web either. The sen­tence in (9c) (= (23) in the questionnaire), with des te+cpr in the headclause, triggered an even higher proportion of negative responses from my informants.4 From the brief discussion of the Dutch data in the preceding paragraphs, it already emerges that there are several factors at play when it comes to the (im)possibility of long-distance dependencies in Dutch comparative correlatives. I have summed them up in (10). (10) factors influencing the (un)grammaticality of long-distance dependencies in Dutch CCs a. the relative clause vs headclause distinction → for most speakers, long-distance dependencies are easier in the relative clause than they are in the headclause



Comparative correlatives and successive cyclicity 

b. the finite/non-finite distinction → for most speakers, long-distance dependencies are easier across infinitival clause boundaries than they are across finite clause boundaries c. the choice of (cor)relative particle – hoe [+wh] versus des [–wh] → for some speakers, long-distance dependencies (esp. in the headclause) are easier with hoe+cpr than they are with des te+cpr d. the word order of the headclause (with des te+cpr) → for some speakers, long-distance dependencies in the headclause (and to some extent also in the relative clause) are easier when des te+cpr in the headclause is not directly followed by the subject but instead is separated from it by either the raised finite verb (in a Verb Second construction) or a complementiser Three of these factors we have already seen at work. (10a) is reflected by the contrast between (7) (which is virtually unanimously accepted) and (9b,c), which receive much less favourable responses. The role of finiteness, (10b), is shown by the contrast between (9a) and (9b,c). And the choice of correlative particle is responsible for the fact that (9c) is rejected by more speakers than (9b) is. So far, however, I have not pre­sented the facts in a particularly systematic fashion, and I have not even begun to touch upon factor (10d). Both (10c) and (10d) present complications that are peculiar to Dutch comparative correlatives – whereas English does not exhibit any kind of variation in these domains, the empirical picture in Dutch is complicated, due to the fact that Dutch comparative correlatives exhibit quite a broad palette of possibilities in these areas: (1b) is in fact a simplification of the empirical lie of the land, laid out more fully in (11). (11)

a. hoe meer je leest, hoe minder {(dat) je begrijpt/??begrijp je} b. hoe meer je leest, des te minder {(dat) je begrijpt/begrijp je} c. %des te meer je leest, des te minder {(dat) je begrijpt/begrijp je} d. *des te meer je leest, hoe minder {(dat) je begrijpt/begrijp je}

In light of the fact that standard Dutch has two particles, hoe [+wh] and des [–wh], available for use in its comparative correlatives, there are four logically possible options.5 In its brief discussion of comparative correlatives, the Algemene Nederlandse Spraakkunst ‘General Dutch Grammar’ (the standard reference grammar of Dutch) only mentions the hoe... hoe (11a) and hoe... des te (11b) patterns; des te... des te (11c) is not brought up at all. This is definitely an oversight on the part of the ANS: though certainly the least popular of the three grammatical patterns, comparative correlatives with des te+cpr in the relative clause are quite common and by no means impossible. As far as I am aware, however, the fourth logically possible pattern, des te... hoe (11d), genuinely does not occur: hoe can be used as the correlative particle in the headclause only if hoe is also chosen as the operator in the relative clause, not otherwise (that is, hoe... hoe is good but *des te... hoe is not). This suggests that the basic correlative particle in the headclause is des, a determiner/demonstrative, in line with what the

 Marcel den Dikken

correlative analysis leads one to expect; the use of the wh-element hoe as a correlative particle is licensed only under ‘harmony’ with the relative clause, i.e., if hoe is also chosen as the relative operator. The restrictions on des te... des te (which only about half of my informants accept readily) remain poorly understood at this time, both descriptively (what determines the amount of speaker variation found here?) and analytically; I will leave this issue aside completely in this paper, treating (11c) as a grammatical subspecies of the Dutch comparative correlative throughout. Within the three attested patterns in (11a–c), there is speaker variation concerning factor (10d) outside the realm of locality issues as well, even in simple comparative correlatives of the type in (11): a few speak­ers find subject–finite verb inversion (V2) in the headclause of comparative correlatives with des te+cpr degraded;6 and some do not (readily) accept insertion of a comple­mentiser (dat in (11)) following the comparative in the headclause.7 All these points of speaker variation add to the impression that comparative correlatives are chaotic – which might play right into the hands of those who believe that comparative correlatives are not to be analysed in terms of principles of Universal Grammar. But as I will show in this paper, there is genuine order beneath the surface appearance of chaos – especially when it comes to the distribution of long-distance extraction in comparative correlatives, which is the topic of this paper. The most robust effect on long-distance dependencies found among speakers of Dutch is (10a), the relative clause vs headclause distinction. Almost all of the speakers I checked with find no difference between short and long extraction in the relative clause; but more than one third of my informants report an effect of distance in the headclause, vary­ing in strength from speaker to speaker and, for a subset of speakers, also depending on the other factors listed in (10) (see (10b–d): finiteness, choice of particle, and word order). I myself am a speaker for whom all of the factors in (10) play a role (in ways that will be spelled out in detail later in this paper); and I have found one other speaker who is sensitive to all four factors as well, showing a pattern of judgements that is virtually identical to mine. A number of other speakers report a sensitivity to factor (10a), either with or without concomitant activation of factor (10b) (finiteness). One speaker finds all long-distance extraction degraded, regardless of any of the factors in (10). And finally, several speakers (about half of my group of informants, but showing quite significant individual vari­ation in more microscopic judgement patterns; only one speaker accepted all fifty of my stimuli) show no special sensitivity to long-distance extraction in comparative correlatives: that is, for them the long-distance extraction cases are sys­te­mat­ically as good or as bad as the corresponding short extraction sentences. In this paper, I will con­centrate on those speakers for whom long-distance dependencies in compar­ative correlatives are impacted by the fac­tors in (10) – though of course I will also address the question of why not all speakers are sensitive to them. Theoretically, the influence of the relative/headclause distinction (10a), finiteness (10b), the choice of particle (10c), and the word order in the headclause (10d) on the



Comparative correlatives and successive cyclicity 

grammaticality of long-distance depen­d­encies in comparative correlatives is of great interest. That the success of such dependencies is dependent on which of the two constituent clauses they are located in supports the conclusion (drawn inde­pendently in Den Dikken 2005a) that the two clauses have a different function – I will show that the almost complete absence of a locality effect in the initial clause is directly predicted by the fact that this clause is a relative clause, with the compar­ative generally fronted to SpecCP via an Aʹ–movement operation that can readily proceed success­ive-cyclically through intermediate SpecCP positions; and the emergence of locality effects in some subspecies of the second clause can be made to follow from an analysis according to which fronting of the comparative targets an IP–adjoined position in the relevant cases. Sensitivity to finiteness (10b), which I will translate struc­turally in terms of the presence or absence of a CP in the matrix verb’s complement, strongly confirms that the comparative ends up in clause-initial position via Aʹ–movement. The choice of correlative particle (10c) will turn out to translate structurally into a choice of substitution for SpecCP or adjunction to IP in the headclause – something which in turn influences the grammaticality of long-distance dependencies. And finally, the question of whether the fronted comparative in the headclause is directly followed by the subject (in a V–final surface pattern: [correl+cpr] – subject – Vfin) or is instead followed by an element occupying the C–head of the headclause (a complementiser or a raised finite verb, the latter yielding a V2 pattern in the matrix: [correl+cpr] – Vfin – subject) once again crucially implicates the role of CP in the construction of long-distance dependencies: whenever there is direct evidence (from complementisers or finite verbs surfacing to the immediate right of the fronted comparative) that the comparative has raised into the SpecCP position of the headclause, long-distance dependencies succeed; but in the absence of such direct evidence for movement to SpecCP (i.e., in surface V–final patterns in the headclause), some language users may, to some extent depend­ing on the choice of correlative particle, instead have the fronted comparative land in an IP–adjoined position. Throughout, we find that, whenever the comparative lands in SpecCP (which it always does in the relative clause, but not necessarily in the matrix), long-distance depen­dencies are gram­matical, via successivecyclic movement through intermediate SpecCP positions; but when the comparative lands in an IP–adjoined position, severe locality restrictions emerge. This is revealing when it comes to our understanding of what constrains long-distance dependencies in the syntax. So besides giving us detailed insight into the inner workings of the comparative correlative construction, the discussion in this paper will also inform the syntax of locality. The paper is organised as follows. In Section 2, I will start out by looking at the gross structure of the comparative correlative in Dutch, paving the way for a detailed discussion of the locality effects and the factors impacting them in Section 3. In Section 4, I subsequently return to the theoretical question of how to analyse long-distance dependencies in current syntactic theory, addressing in particular the question of under which circumstances successive-cyclic extraction via an embedded SpecCP

 Marcel den Dikken

position is forced. Sec­tion 5 provides an overview of the main findings, and some general concluding remarks. And the appendix lays out the details of the questionnaire study that underlies most of the empirical discussion in this paper.

2. The macrostructure of Dutch comparative correlatives Let me begin by revisiting the macrosyntactic structure of comparative correlatives universally – the struc­ture in (2b), repeated here for convenience. (2) b. comparative correlative (Den Dikken 2005a) [matrixcl [relcl rel/wh-operator+cpr...] [matrixcl correl-prt+cpr...]] This structure leaves many detailed questions about the syntax of comparative correlatives unanswered, in­cluding the important questions in (12): (12) a. what is the categorial status of the headclause? b. what is the structural relationship between the headclause and the relative clause? c. what is the landing-site of the fronted constituent [(cor)rel+cpr] in each of the two clauses? I will assume as a given that the relative clause in (2b) is a CP – a standard and generally uncontroversial assumption which, as we will see, actually receives empirical support from the Dutch facts under discussion in this paper. It is likely that the landingsite of the fronted comparative in the relative clause will be its SpecCP position – again this is a point that requires little discussion, and we will find it confirmed below. But the categorial status of the headclause is far from self-evident, nor is the relationship between it and the relative clause or the landing-site of the fronted comparative in the matrix. The Dutch facts that this paper focuses on allow us to get a clearer picture of these complex questions. As far as (12a) is concerned, it seems beyond question that the headclause can itself be as large as a full-fledged CP: as we have seen already (recall (11)), the headclause can contain a complementiser or may exhibit a Verb Second word order, with the finite verb fronted to C.8 But what are we to do with (13)? (13) a. hoe meer je leest, hoe minder je begrijpt b. hoe meer je leest, des te minder je begrijpt c. %des te meer je leest, des te minder je begrijpt rel deg more you read correl deg less you understand These sentences feature the verb-final versions of the headclauses in (11). Apart from the question of how to analyse the position of the relative clause such that it never counts for the computation of Verb Second (which I return to presently), the question



Comparative correlatives and successive cyclicity 

these sentences pose is the following: How can the headclause of a root comparative correl­ative fail to exhibit any activity on the part of the C–head? This question is par­ ticu­larly inter­esting because outside the realm of (comparative) correlative constructions,9 Dutch never features root clauses whose C–head position is radically empty, as it seems to be in (13) (after all, neither a lexical complementiser nor a fronted finite verb appears to the immediate right of the fronted comparative here). Suppose that the correlative particle (pied-piping the comparative), unlike topics, wh-constitu­ents and other fronted material, has no designated landing-site at the left edge of the clause, and that its only require­ment is that, whenever it fronts (see fnn. 24 and 25 on the marginal possibility of in situ correl+cpr), it must end up adjacent to the relative clause. Let us take this to be a hallmark of correlative construc­tions in general (cf. also Bhatt & Pancheva 2006, (78), on the position of if-clauses vis-à-vis the correlative particle then in conditionals), and take (14) as our first step towards an understanding of the (non-)inversion facts: (14) the fronted correlative particle and the sentence-initial relative clause must be adjacent The correlative particle can satisfy this requirement by raising to SpecCP, in which case [correl+cpr] is followed by a C–head lexicalisable either by a raised finite verb or by the complementiser dat (as in (15a)). (15) a. [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL [correl+cpr]j [Cʹ C={Vfin, dat} [IP S... tj...]]]] b. [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL C=Ø [IP [correl+cpr]j [IP S... tj...]]]] Alternatively, however, (14) can be satisfied by having [correl+cpr] adjoin to the IP of the headclause (see (15b)), in which case the C–head that separates [correl+cpr]j from the relative clause must remain empty: V–move­ment to C or comple­mentiser insertion in C would preclude satisfaction of the condition in (14). And since there is no head position between the IP–adjoined [correl+cpr] and the subject (abbreviated as ‘S’ in (15)) in SpecIP, the IP–adjunction scenario will never give rise to inversion of the subject and the finite verb. So non-inversion in the headclause of a root comparative correlative results whenever [correl+cpr] adjoins to IP, which is one of the two options that present themselves when it comes to satis­fying the require­ment that says that the correlative particle must be adjacent to the relative clause (14). The relative clause itself is arguably extraclausal (see also Srivastav 1991 and Bhatt 2003 on regular correlatives, and Bhatt & Pancheva 2006 on conditional correlatives): it itself never triggers Verb Second, and behaves in this respect like a variety of other clauses, all of which involve conditional semantics.10 Thus, (i) verb-initial conditional clauses (16a), and (ii) clauses introduced by (ook) al ‘(lit.) also all, i.e., even if ’ (16b), and (iii) clauses introduced by hoe ‘how’ followed by a non-comparative adjective or quantifier (veel in (16c)) all fail to trigger Verb Second, hence are not followed by the finite verb of the main clause.

 Marcel den Dikken

(16) a.

mocht je nog geld nodig hebben, { ik wil/*wil ik} might you yet money needy have I want/want I je wel helpen you surely help ‘should you need money, I am willing to help you’

b. (ook) al gaf je me een miljoen, { ik doe/*doe ik} het niet also all gave you me a million I do/do I it not ‘even if you gave me a million, I won’t do it’ c.

hoeveel je ook leest, { je begrijpt/*begrijp je} how-much you also read you understand/understand you het toch niet it still not ‘no matter how much you read, you still won’t understand it’

The sentence-initial clause in (16c) is very similar to that found in the comparative correlatives in (13a,b), differing only in that the latter features a comparative while the former does not. All of the constructions in (16) behave basically like the comparative correlative as far as their distribution in non-root contexts is con­cerned as well. For comparative correlatives of the type in (11/13a), we find that they can be embedded in finite complement clauses to bridge verbs and also (though somewhat less felicitously to my ear) in finite non-bridge-verb contexts (including adjunct clauses), but not in non-finite contexts. This is illustrated in (17).11 For the example in (16c), which I will use as a representative of the entire set of cases exemplified by (16), we see this pattern reproduced in (18) (with a somewhat stronger degradation in the c–example). (17) a. ik denk dat hoe meer je leest, hoe minder je begrijpt I think that how more you read how less you understand ‘I think that the more you read, the less you understand’ b.

?het is te

betreuren dat hoe meer je leest, hoe minder it is to regret that how more you read how less je begrijpt you understand ‘it is regrettable that the more you read, the less you understand’

c. ?het lijkt alsof hoe meer je leest, hoe minder je begrijpt it seems as-if how more you read how less you understand ‘it seems as if the more you read, the less you understand’ d. *het is vervelend om hoe meer { je leest/te lezen}, hoe it is annoying comp how more you read/to read how minder te begrijpen less to understand *‘it is annoying the more you read/to read, the less to understand’



Comparative correlatives and successive cyclicity 

e. *hoe meer { je leest/lezen}, hoe minder begrijpen is irritant how more you read/read-inf how less understand is irritating *‘the more reading, the less understanding is irritating’ (18) a. ik denk dat hoeveel je ook leest, je het toch niet begrijpt I think that how-much you also read you it still not understand ‘I think that no matter how much you read, you still won’t understand it’ b.

?het is te

c.

??het lijkt

betreuren dat hoeveel je ook leest, je het toch it is to regret that how-much you also read you it still niet begrijpt not understand ‘it is regrettable that no matter how much you read, you still won’t understand it’ alsof hoeveel je ook leest, je het toch it seems as-if how-much you also read you it still niet begrijpt not understand ‘it seems as if no matter how much you read, you still won’t understand it’

d. *het is vervelend om hoeveel je ook leest, het toch niet it is annoying comp how-much you also read it still not te begrijpen to understand *‘it is annoying no matter how much you read, still not to understand it’ e. *hoeveel je ook leest, het toch niet begrijpen is irritant hoe-much you also read it still not understand is irritating *‘no matter how much you read, still not to understand it is irritating’ Having established that the sentence-initial relative clause of comparative correlatives (like the initial clauses in (16)) is an extraclausal constituent, let me assume (with Srivastav 1991 and Bhatt 2003, among others) that it is adjoined to the headclause. The relative clause is arguably a full-fledged CP: relative clauses generally are CPs (perhaps with the exception of English infinitival relatives with a subject gap, as in a man to fix the sink; cf. Law 1991). But what about the head­clause? Clearly, the headclause can be as large as a full CP (see also Citko, this vol.): as I already pointed out earlier with reference to (11), the headclause-adjoined relative clause can occur to the immediate left of a V2–construction or a headclause featuring the com­ple­men­tiser dat. But if the headclause of a Dutch comparative correlative were always as large as a full-blown CP, it would be difficult to account for the relative ease of embedding comparative correlatives inside finite clauses. For finite clauses in the complement of bridge verbs one could imagine a CP recursion structure, with one CP directly embedded in another (see e.g. Vikner 1995, Iatridou & Kroch 1992). Indeed, some attested in­stances of embedded com­parative correlatives exhibit subject– verb inversion in the headclause. But these cases of embedded root word order are not confined to bridge verb contexts: they occur in extraposed subject sentences and adjunct

 Marcel den Dikken

clauses as well, as in (19). To my ear, these cases of embedded headclauses with V2 order (all from academic discourse; (19c) is from a phonetics abstract for a linguistics conference) have an unmistakably anacoluthic ring to them (quite unlike what we find in German, where, as Klaus Abels points out to me, com­parative correlatives always have V2 order in the headclause, even in embedded contexts). (19) a.

het vervelende is, dat hoe meer jij je probeert te verzetten the annoying is that how more you you try to resist tegen blozen des te sneller treedt dit op against blushing the-gen deg faster treads this up ‘the annoying thing is that the more you try to resist blushing, the more quickly it happens’ [www.leidenuniv.nl/ics/ sz/so/persoonlijk/bloosblunder.htm]

b. ten eerste is het zo dat hoe meer bootstrapsteekproeven to-the first is it so that how more bootstrap-tests je gebruikt, des te “betrouwbaarder”/“stabieler” zal het you use the-gen deg reliable-cpr/stable-cpr will the percentieleninterval en daarmee ook het betrouwbaarheidsinterval zijn percentile-interval and therewith also the reliability-interval be ‘first of all it is the case that the more bootstrap tests you use, the more “reliable”/“stable” will be the percentile interval and with that also the reliability interval’ [www.ppsw.rug.nl/~kiers/ Syllabus%20Stat1A%20HS2%20Bootstrap.doc] c.

omdat hoe meer een woordvorm lijkt op de ‘canonieke vorm’, because how more a word-form looks on the canonical form des te makkelijker is het voor de luisteraar om deze te the-gen deg easier is it for the listener comp this to ‘mappen’ op het mentale lexicon map onto the mental lexicon ‘because the more a word form looks like the canonical form, the easier it is for the listener to map it onto the mental lexicon’ [www.fon.hum.uva.nl/FonetischeVereniging/DvdFonetiek/DagvdFonetiekAbstracts2002.html]

Cases like (19) aside, embedding comparative correlatives in non-root contexts gives rise to obligatory verb-final order in the headclause of the comparative correlative (Den Dikken 2005b): (20) a. b.

ik denk dat hoe meer je leest, hoe minder { je I think that how more you read how less you begrijpt/*begrijp je} understand/understand you ik denk dat hoe meer je leest, des te minder {je begrijpt/*begrijp je}



Comparative correlatives and successive cyclicity 

c. ik denk dat des te meer je leest, des te minder {je begrijpt/*begrijp je} This dependency of the word order inside the second clause of the Dutch comparative correlative on the root/non-root asymmetry confirms that the second clause is the head of the construction. But at the same time it suggests that the headclause of the comparative correlative is not always as large as a full-fledged CP: to facilitate regular embedding of comparative correlatives in non-bridge contexts (in a non-anacoluthic way, that is, unlike in (19)), it seems we have to countenance the possibility that the headclause of the comparative cor­relative is a mere IP, as in (21): (21) C [IP [RELCL [rel+cpr]i... ti...] [IP=HEADCL [correl+cpr]j [IP S... tj...]]] In this structure, both the relative clause and the [correl+cpr] constituent in the headclause are adjoined to IP, to the left of the subject. We had already found reason to believe that, instead of substituting for SpecCP, the fronted comparative in the headclause can adjoin to IP – recall (15b). And of course the requirement that the correlative particle be adjacent to the relative clause (14) is met by (21). The structure in (21) is entirely within the range of allowable structures for comparative correlatives, therefore. The embeddability of com­parative correlatives in a broad range of finite complementation contexts is thus accounted for. The struc­ture in (21) arguably is not available in infinitives, however – as witness, for instance, the fact that English does not allow topicalisation in infinitives (cf. *[to Mary, to give a book] would be very nice and [that to Mary, they gave a book] was very nice). So that will continue to rule out the embedding of com­parative cor­relatives in infinitival contexts (recall (17c,d), and also fn. 11 on English, where the same applies, mutatis mutandis). Now that we have (21) in place, we could consider getting rid of (15b) altogether and analyse instances of lack of verb fronting in comparative correlatives in root contexts (i.e., cases of the type illustrated in (13)) in terms of a bare IP with both the relative clause and the fronted comparative adjoined to it. Zwart (1997) (following Travis 1984) argues that Dutch subject-initial root clauses are simple IPs, with the subject in SpecIP and the finite verb raised to I, producing a Verb Second effect. The fronting of the finite verb, however, is precisely what we would like to avoid in the analysis of sentences of the type in (13), where the verb is final, not raised. For (13), it is hard to tell whether the finite verb has raised to I, because of the paucity of additional lexical material. But in slightly more complex examples such as those in (22), the verb clearly must be in final position in the headclause: it cannot surface between the subject and the PP ervan ‘thereof ’. (22) a. hoe meer je leest, hoe minder je 〈*begrijpt〉 ervan 〈✓begrijpt〉 b. hoe meer je leest, des te minder je 〈*begrijpt〉 ervan 〈✓begrijpt〉 c. %des te meer je leest, des te minder je 〈*begrijpt〉 rel deg more you read correl deg less you understand ervan 〈✓begrijpt〉 thereof understand

 Marcel den Dikken

An IP that serves as the root will, in a Verb Second language such as Dutch, always receive the finite verb in its head in the course of the overt-syntactic derivation. By contrast, the head of an IP embedded within CP never serves as the final resting place of a raised verb: terminal V–to–I movement is possible in Dutch only in undominated IPs, and in fact obligatory there. If we were to represent the root of the comparative correl­atives in (13) and (22) as a bare IP, therefore, we would fail to block raising of the finite verb to I, thereby deriving the ungrammatical variants of (22). By having comparative correlatives in root contexts always pro­ject all the way up to CP, as in (15), we avoid this problem: IP is not undominated, hence terminal V–to–I cannot proceed. V–to–I–to–C movement can proceed in (15a); but verb movement is blocked altogether in (15b): verb raising all the way up to C would interfere with (14), and is impossible across IP–adjoined material in any event;12 and verb raising to I and no further is incompatible with IP not being undom­inated in this structure. So (15b) correctly ensures that there will be no verb fronting at all in (22). Since a root version of (21) does not manage to accomplish this, we therefore cannot trade (15b) in: we must keep it. This is all we need to say about the macrostructure of the Dutch compara­tive correlative. Before pro­ceeding to the discussion of the locality effects that are at the core of this paper, let me briefly recapitulate my main findings in this section, which will serve as important background to the dis­cussion to follow: (23) a. comparative correlatives in root contexts are full CPs, with the relative clause CP–adjoined b. comparative correlatives in embedded contexts are IPs, with the relative clause IP–adjoined c. when [correl+cpr] raises to SpecCP (in root contexts only), the C–head can be filled by either a complementiser or a raised finite verb (the latter delivering subject–verb inversion) d. when [correl+cpr] raises to an IP–adjoined position in root contexts, the C–head must remain unfilled for compliance with (14) Viewed against the backdrop of Davison’s (this vol.) discussion of Sanskrit and Hindi/ Urdu correlatives, (23a,b) suggest that CP– vs IP–adjunction is not just a cross-linguistic parameter (as it seems to be in the case of the Indic languages, if Davison is right) but also a possible locus of variation within a single language.

3. Long-distance dependencies in comparative correlatives and successive cyclicity The key to our understanding of long-distance dependencies in comparative correlatives is held by an account of the restrictions on successive-cyclic movement through



Comparative correlatives and successive cyclicity 

the SpecCP position of an embedded clause into an Aʹ–position in a higher clause. But before turning to those restrictions, let me first pro­­vide em­piri­cal support for the existence of suc­cessive-cyclic Aʹ–movement through SpecCP, to set the stage.

3.1

Successive-cyclic movement via SpecCP

That it is possible to Aʹ–move a constituent out of an embedded clause into a higher clause via an inter­mediate touch-down in the lower SpecCP is suggested in a particularly interesting way by certain ­Q–Float facts in Irish English discussed by McCloskey (2000).13 Consider the triple in (24): (24) a. what all did he say (that) he wanted t? b. what did he say (that) he wanted t all? c. what did he say all (that) he wanted t?

(Irish English)

On the assumption (based on Sportiche 1988) that floating quantifiers are stranded in positions that are mem­bers of the chain of a moved constituent that the quantifier is construed with, (24b) is straightforwardly analysed in terms of stranding of the quantifier all in the base position of the wh-phrase. And of course (24a) is unproblematic as well: here the quantifier is simply pied-piped by movement of what into the matrix SpecCP. The interesting case is (24c), which has the quantifier stranded in an intermediate position on the movement path. This position, to the immediate left of that (when it occurs) and to the immediate right of say (which does not appear to have undergone any movement), seems to have to be the SpecCP position of the embedded clause – a position in which standard analyses of long-distance Aʹ–extraction going back to Chomsky (1977) predict there to be an intermediate trace of the extracted wh-phrase. The grammaticality of (24c) thus presents evidence, on a Q–stranding approach to Q– Float, for an intermediate touch-down in SpecCP in the process of long wh-fronting.14 (25) [CP whati [Cʹ did he say [CP [ti all]j [Cʹ (that) he wanted tj]]]] In recent work of my own (see Den Dikken 2006) on long Aʹ–dependencies involving focus fronting in Hungarian, I have argued on the basis of the Case and agreement behaviour of long focus fronting that it can proceed via an intermediate touch-down in SpecCP as well. The key cases under discussion in that work are illustrated in (26). Here the raised focus, though functioning as the subject of the embedded finite verb (which would normally get nominative Case), bears accusative Case, and ­the matrix verb agrees in definite­ness and person with ­­the extractee: in (26a), the raised focus is indefinite, hence the matrix verb shows up in the indefinite agreement form; in (26b) both the focus and the matrix verb are definite; and in (26c), the matrix verb agrees in person with the second-person focus, resulting in specialised morphology (-lak). (26) a. két fiú-t akar-ok, hogy ec jöjjön (Hungarian) two boy-acc want-1sg.indef that come-subjunc-3sg

 Marcel den Dikken

b. a két fiú-t akar-om, hogy ec jöjjön the two boy-acc want-1sg.def that come-subjunc-3sg c. téged akar-lak, hogy ec jöjjél you-acc want-lak/lek that come-subjunc-2sg ‘it is {two boys/the two boys/you} that I want ec to come’ These examples contrast with the likewise grammatical ones in (27), where the raised focus surfaces with (morphologically unmarked) nominative Case and the verb systematically bears definite agreement as a reflex of its Agree relationship with the embedded CP (which, in Hungarian, triggers definite agreement). (27) a. két fiú akar-om, hogy ec jöjjön two boy.nom want-1sg.def that come-subjunc-3sg b. a két fiú akar-om, hogy ec jöjjön the two boy.nom want-1sg.def that come-subjunc-3sg c. te akar-om, hogy ec jöjjél you.nom want-1sg.def that come-subjunc-2sg ‘it is {two boys/the two boys/you} that I want ec to come’ I have argued that both (26) and (27) involve extraction of the focus out of the embedded clause into the matrix clause via Aʹ–movement.15 But while in (27) the focus extracts from the embedded clause without a touch-down in SpecCP (via a derivation along the lines of Rackowski & Richards’s 2005 recent analysis of long Aʹ–movement in Tagalog), in (26) it proceeds successive-cyclically via a stopover in SpecCP. It is this stopover that enables the matrix verb to establish a morphological agreement relationship with the extracted focus (under local Agree between the matrix v and the focus in SpecCP) for Case, (in)definiteness and person. If my analysis of these Hungarian facts (which is spelled out in greater detail in Den Dikken 2006) is on the right track, there are two ways in which a constituent of an embedded finite clause can escape from that clause via A´–movement: either of the scenarios depicted in (28) is available to it in principle. (28) a. XPi... [VP V [CP tiʹ [Cʹ C [TP (...) ti (...)]]]] b. XPi... [VP V [CP C [TP (...) ti (...)]]] For Hungarian, the alternation between (26) and (27) provides direct evidence for the existence of these two strategies – and since the evidence is directly manifest in the morphosyntactic output, the language user will have no difficulty deciding which of the two scenarios is at work in which particular case. But in light of the conclusion that (28a,b) both exist as valid long Aʹ–movement strategies in Universal Grammar, for every language it will have to be decided, for each individual case, whether movement proceeds successive-cyclic­ally through SpecCP or not.



Comparative correlatives and successive cyclicity 

For the case of Dutch, the discussion to follow will make a case that speakers that show a sensitivity to the factors in (10) avail themselves of a derivation of long-distance dependencies in com­parative correla­tives making an intermediate stopover in SpecCP – in other words, (28a) is the way to go for these Dutch speakers. Once I have established this, I will return to the theoretical questions posed by the outcome of the dis­ cussion of the Dutch facts, in Section 4. But first, let me put in place an important restriction on successive-cyclic Aʹ–movement.

3.2

The principle of unambiguous binding (chain uniformity)

In their detailed study of the lack of interaction between scrambling and wh-movement, Müller & Sterne­feld (1993) argue for a condition on Aʹ–dependencies that they formulate as follows: (29) Principle of Unambiguous Binding (PUB) a variable that is α-bound must be β-free in the domain of the head of its chain (where α and β refer to different types of positions) They go on to explain that, since ‘α-bound’ means ‘bound from a position of type α’ and ‘β-free’ means ‘not bound by a position of type β’, it follows that ‘a particular type of Aʹ–movement (e.g., movement to a β-position) may not feed another type of movement (e.g., movement to an α-position)’ (Müller & Sternefeld 1993: 461). They show that this restriction on Aʹ–dependencies finds strong support in the restrictions on scrambling in a broad variety of languages. Central among these restrictions is the fact that scrambling can­not transgress the boundaries of a finite clause: sentences of the type in (30a,b) (from German) are ungrammatical (whereas their counterparts involving clause-mate scrambling are grammatical).16 (30) a. *... daß niemand [VP Puddingi [VP sagt [CP tiʹ daß sie ti mag]]] (German) that nobody pudding says that she likes b. *... daß [IP Puddingi [IP niemand sagt [CP tiʹ daß sie ti mag]]] that pudding nobody says that she likes The problem with these sentences lies precisely in the fact that they violate the PUB in (29). Traces of scrambl­ing are taken to be variables;17 but the variables (ti) in (30) are not unambiguously bound: they are bound ambiguously from two different types of positions, one an Aʹ–specifier position (SpecCP) and the other an Aʹ–adjunction position (the VP– or IP–adjoined position in the matrix clause). Especially interesting for our purposes in this paper is the ill-formedness of structural configurations of the type in (30b), with an IP–adjoined element in the matrix clause unsuccessfully serving as the ante­cedent in a case of successive-cyclic movement through SpecCP. I will show why in the next subsection.

 Marcel den Dikken

3.3

Long-distance dependencies in comparative correlatives: Analysis

In Section 2, I argued that Dutch comparative correlatives provide two potential landing-sites for the fronted comparative in the headclause: the phrase [correl+cpr] may either substitute for SpecCP, or alternatively it may adjoin to IP. The two derivations are depicted in (15), repeated here. (15) a. [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL [correl+cpr]j [Cʹ C={Vfin, dat} [IP S... tj...]]]] b. [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL C=Ø [IP [correl+cpr]j [IP S... tj...]]]] In local contexts, in which the variable left by movement of [correl+cpr] (tj) is a clausemate of its binder, both scenarios in (15) are in perfect conformity with the PUB in (29). But now consider a derivation in which the fronted comparative in the headclause starts out in an embedded finite CP and extracts from it via an intermediate touch-down in SpecCP, as depicted in (31): [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL [correl+cpr]j [Cʹ C [IP S... [CP tjʹ [Cʹ... tj...]]]]]] b. *[CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL C [IP [correl+cpr]j [IP S... [CP tjʹ [Cʹ... tj...]]]]]] (31) a.

This time, the two choices of final landing-site for the fronted comparative do make a vital difference: the derivation in (31b), featuring successive-cyclic movement of [correl+cpr] into an IP–adjoined position in the headclause, is ruled out by the PUB for the same reason that (30b) is; but (31a) is perfectly well-formed. This, in essence, is what is responsible for the ungrammaticality of certain instances of long-distance depen­dencies in Dutch comparative correlatives. But some more work is needed to fill in the details, which, as we have seen, are rather involved: there are four factors that influence the grammaticality of long-distance dependencies in Dutch comparative correlatives, summed up in (10), which is repeated here for convenience: (10) factors influencing the (un)grammaticality of long-distance dependencies in Dutch CCs a. the relative clause vs headclause distinction → for most speakers, long-distance dependencies are easier in the relative clause than they are in the headclause b. the finite/non-finite distinction → for most speakers, long-distance dependencies are easier across infinitival clause boundaries than they are across finite clause boundaries c. the choice of (cor)relative particle – hoe [+wh] versus des [–wh] → for some speakers, long-distance dependencies (esp. in the headclause) are easier with hoe+cpr than they are with des te+cpr d. the word order of the headclause (with des te+cpr)



Comparative correlatives and successive cyclicity 



→ for some speakers, long-distance dependencies in the headclause (and to some extent also in the relative clause) are easier when des te+cpr in the headclause is not directly followed by the subject but instead is separated from it by either the raised finite verb (in a Verb Second construction) or a complementiser

I will address each of these factors in the following subsections, against the background of (31). 3.3.1 The headclause/relative clause distinction Relative clauses are always CPs. And headed relative clauses, as far as I am aware, always involve Aʹ–move­ment of the relative operator into SpecCP. For a language such as Hungarian, which is known to often front its wh-constituents into positions substantially lower in the tree than SpecCP, this can be verified by in­vestigating the (im)possibility of inserting a topic to the left of the relative operator. As Kenesei (1992) points out, Hungarian does not allow topics to be placed in such a position in its headed relatives: (32a,b) are bad with Pétert to the left of aki (though perfectly fine if it stays in clause-internal position). (32) a.

a lány, 〈*Pétert〉 aki meghívta 〈Pétert〉, the girl Péter-acc a-who.nom invited Péter-acc rosszul tette badly did

(Hungarian)

b. az 〈*Pétert〉 aki meghívta 〈Pétert〉, rosszul tette that Péter-acc a-who.nom invited Péter-acc badly did c. Ø 〈?Pétert〉 aki meghívta 〈Pétert〉, rosszul tette Péter-acc a-who.nom invited Péter-acc badly did ‘{the girl/(s)he/Ø} who invited Peter made a mistake’ In (32a) we are dealing with a common or garden headed relative clause, with a lány ‘the girl’ as the head. Though interpretively more like (32c) than like (32a), the sentence in (32b) involves a headed relative as well – albeit that the head in this particular instance is light, just a demonstrative pronoun (az). This is an example of what Citko (2004) calls a ‘light-headed relative’. It behaves, as far as the criterion in question is concerned, just like a headed relative, in disallowing the placement of a topic to the left of the relative pronoun. But free (or ‘empty-headed’) relatives are different when it comes to this: as (32c) shows, Hungarian free relatives differ from headed relatives in allowing topics to precede the relative operator. Topicalisation across the relative pronoun in free relatives is not a staple of the spoken vernacular: rather, it is typical of elevated, literary styles.18 This suggests that it is possible in principle for Aʹ–movement

 Marcel den Dikken

of the relative pronoun to target a position lower than SpecCP – in free relatives, but not in headed relatives; and only in certain registers. Let us take the position of the wh-pronoun in the free relative in (32c) to be an IP–adjoined position.19 Assuming so, these Hungarian facts become of immediate interest for our discussion of Dutch comparative correlatives. In comparative correlatives, the initial clause is a relative clause – more specifically, it is a free relative clause. And whereas in headed relative clauses, as we have just seen, the relative operator must raise to SpecCP, under certain circumstances it is possible for the relative operator to move to a lower, IP–adjoined position in free relatives. Those special circumstances, in the case of Hungarian, are all about register and style. Now, comparative correlatives per se are not restricted to elevated or literary styles: they occur fre­quent­ly in the spoken vernacular. But the comparative correlatives that we are investigating in this paper all involve a long-distance Aʹ–dependency in one of the two constituent clauses. This will certainly have an effect on their distribution across different registers. Though I have not explicitly tested this, the high pro­pensity of comparative correlatives with long-distance dependencies in web-posted texts of an academic or otherwise formal nature (notably including sermons) suggests that indeed, these are characteristic of more elevated registers, not of the everyday spoken language. In light of this, one would expect that the initial clause of the comparative correlative (the free rela­tive clause) will standardly feature Aʹ–movement of the relative operator (pied-piping the comparative) into SpecCP, but that, under certain circumstances, depending on style and register, the [rel+cpr] complex could alternatively target an IP–adjoined position inside the relative clause. Simply put, in the free relative clause, (33a) is the normal case, and (33b) may be available under special conditions. (33) a. [CP=RELCL [rel+cpr]i [Cʹ C [IP S... ti...]]] b. [CP=RELCL C [IP [rel+cpr]i [IP S... ti...]]] The predictions with regard to long-distance dependences are now straightforward, bearing in mind what was said in Section 3.2 and assuming that long-distance movement of the comparative must proceed successive-cyclically through SpecCP (recall Section 3.1): in the standard case, long movement of [rel+cpr] in the relative clause of the comparative correlative should be able to proceed without obstruction as long as the embedded clause is not an island; but whenever [rel+cpr] adjoins to IP, the PUB in (29) prohibits all long-distance dependencies. The facts are basically in conformity with the predictions. The vast majority of speakers find there to be no significant difference between short- and long-distance dependencies in the relative clause of the Dutch comparative correlative. Except for one speaker, all my informants find that long-distance dependen­cies in the relative clause featuring the wh-operator hoe ‘how’ are not (significantly) worse than the corres­ponding short-distance dependencies – i.e., for virtually all speakers, (34a) is grammatical.



Comparative correlatives and successive cyclicity 

(34) a. [RELCL [hoe+cpr]i ... [CP tiʹ... ti]] b. %[RELCL [des te+cpr]i ... [CP tiʹ... ti]] There are four speakers (K, L, O and P in appendix A.4), however, who report a slight degradation for rela­tive-clause internal long-distance dependencies involving des as the relative operator, as in (34b); and two speakers (N and Q) reject such dependencies outright. This suggests that, whereas hoe+cpr almost systematically lands in SpecCP, for some speakers des te+cpr targets an IP–adjoined position in the relative clause – thus inducing ‘PUB effects’ under long-distance extraction. The use of des te in the relative clause is a stylistically marked option: several of my informants simply do not accept des te in the relative clause of the comparative correlative at all, allowing it only in the head­clause. I take it, then, that des te+cpr is a likely candidate for IP–adjunction in the relative clause – and that some speakers allow des te+cpr no other option, thereby ruling out long-distance dependencies in the relative clause altogether in the case des te+cpr. With the facts interpreted this way, the one informant who consis­tent­ly rejected all long-distance depen­dencies in the relative clause (the speaker identified in Section A.4 as ‘Q’) must have generalised the IP–adjunction strategy to hoe+cpr as well. But for the vast majority of speakers, long movement of hoe+cpr in the relative clause is perfectly grammatical, as expected in light of the fact that (a) hoe+cpr is not stylistically marked, and (b) SpecCP is the unmarked landing-site for relative operators, as I showed before. In the headclause, by contrast, there are considerably more rejections of long-distance dependencies involving hoe+cpr – though here as well, des te+cpr consistently scores lower grades when it comes to long extraction. I have annotated the empirical lie of the land (see informants I–Q in A.4) as in (35), where ‘%’ stands quite simply for speaker variation, and ‘%*’ represents a strong but not absolute tendency for rejection. (35) a. %[HEADCL [hoe+cpr]i ... [CP tiʹ... ti]] b. %*[HEADCL [des te+cpr]i ... [CP tiʹ... ti]] Relative clauses are generally forced to be CPs, and movement of the relative operator generally targets SpecCP – for reasons that are entirely independent of anything that is going on in the comparative correla­tive per se. But in the headclause, there are no strong independent reasons that force [correl+cpr] to raise to SpecCP. IP–adjunction would seem to be a perfectly viable option in principle. For all headclauses in which an Aʹ–dependency involving the comparative is formed such that the fronted comparative lands in an IP–adjoined position, the PUB in (29) predicts ungrammaticality under long extraction. The higher rate of rejection of long-distance dependencies in the headclause of the Dutch comparative correlative thus suggests that IP–adjunction is resorted to much more often in the headclause than in the relative clause – because this strategy is available much more liberally than it is in the relative clause, and arguably also because adjunction to IP is ‘cheaper’ than substitution for SpecCP,

 Marcel den Dikken

especially in contexts in which there is no direct empirical evidence from word order to suggest that the fronted comparative has landed in SpecCP. The distribution of the two options (substitution for SpecCP and adjunction to IP) seems to be determined to a significant extent by the nature of correl (hoe or des), and also by the word order in the headclause, following the fronted comparative. This naturally leads us to a discussion of factors (10c) and (10d). But before I turn there, let me briefly address factor (10b), the role of finiteness, which is relatively easy to deal with. 3.3.2 The role of finiteness So far we have seen that long-distance dependencies in the two constituent clauses of the comparative correl­ative are constrained by the PUB in (29): whenever the fronted comparative can be argued on reasonable grounds to have landed in an IP–adjoined position, long-distance extraction thereof via an intermediate touch-down in SpecCP is ruled out. The sensitivity of long extraction of the comparative from a finite complement clause to the nature of the landing-site (which will be further supported in Section 3.3.3) suggests that such extraction does indeed proceed via a touch-down in SpecCP in the case of finite complement clauses. For extraction from infinitival complement clauses (as in (9a), repeated below), by contrast, a stopover in SpecCP is not needed – in fact, it is probably impossible: (9a) and its ilk are verb raising con­structions showing clause union effects that arguably diagnose the absence of a CP–layer for the infinitival clause (cf. e.g. Wurmbrand 2003). (9) a. hoe meer je leest, hoe minder je denkt te begrijpen how more you read how less you think to understand So with CP altogether absent from the complement of the matrix verb in cases like (9a), no touch-down in an embedded SpecCP position will be made in the derivation of sentences of this type. As a result, the PUB in (29) will not reject any instance of a longdistance dependency across an infinitival clause boundary. And indeed, almost all speakers report that in infinitival contexts, long-distance dependencies are no worse than their short-distance counterparts, in both the relative clause and the headclause. One speaker (the one iden­tified in A.4 as ‘Q’) reports consistent degradation (to full ungrammaticality, throughout) in all long-distance contexts, whether finite or non-finite; and another speaker (‘P’) rejects all finite and non-finite long-distance dependencies in the headclause. A possible line of approach would be to say that, for these two speakers, the in­finitival complement of denken is a CP, with successive-cyclic movement through SpecCP forced, and with the output of successive-cyclic movement ruled out by the PUB on the assumption that, for these speakers, the fronted comparative consistently targets an IP–adjoined position (for speaker ‘P’ only in the headclause). The viability of such an approach remains to be investigated. I will not discuss this judgement type in any further detail in this paper, since it is very clearly a minority pattern. For all other speakers, long-distance dependencies across infinitival clauses are generally possible in both clauses of the Dutch com­para­tive correlative: there is no



Comparative correlatives and successive cyclicity 

sensitivity to the nature of the landing-site (SpecCP or an IP–ad­joined position), and, consequently, no sensitivity to the nature of the (cor)relative pronoun or to the word order in the headclause – two factors that do constrain long-distance dependencies across finite clause boun­daries, as I will now go on to discuss. 3.3.3 The nature of the (cor)relative pronoun We have already seen, in our discussion of long-distance dependencies in the relative clause in Section 3.3.1, that the choice of relative pronoun to some extent influences the success of such dependencies there: for a relatively small subset of speakers (only four of my informants), des te+cpr does not easily form long-distance dependencies in the relative clause, while hoe+cpr generally encounters no significant difficulties in this domain. I blamed that on the general stylistic markedness of des te in the relative clause of the com­parative correlative, in line with the discussion of the vicissitudes of IP–adjunction of the relative pronoun in Hungarian free relatives. In the headclause of the Dutch comparative correlative, there are no (strong) stylistic factors at play, however, when it comes to the choice between hoe+cpr and des te+cpr: while (13c), with des te in the relative clause, is marked, both (13a) and (13b) are generally deemed perfectly fine, and to be essentially in free variation. (13) a. hoe meer je leest, hoe minder je begrijpt b. hoe meer je leest, des te minder je begrijpt c. %des te meer je leest, des te minder je begrijpt rel deg more you read correl deg less you understand So the fact that many speakers nonetheless show a distinctive sensitivity to the choice of correlative particle in the headclause of the comparative correlative in long-distance dependency contexts – a sensitivity that is in fact stronger here than it is in the relative clause – calls for a different sort of explanation. I would like to argue here that hoe+cpr, by virtue of being [+wh], can in principle always raise to SpecCP, in headclauses and relative clauses alike. But whenever hoe+cpr raises to SpecCP in the head­clause, raising of the finite verb to C is impossible: if one were to have performed subject–finite verb inversion under fronting of the wh-constituent, one would necessarily have ended up with the semantics of a wh-question or exclamative for the headclause (cf. Postma’s 1995 and Bennis’ 1995 structure-based approach to the inter­pretation of wh-elements): strings of the type wh – Vfin – subject –... in Dutch are either questions (wat kan hij eten? ‘what can he eat?’) or exclamatives (wat kan hij eten, zeg! ‘boy, can he eat!’); but question­hood or exclamativity is obviously an inappropriate illocutionary force for the headclause of a (declarative) correl­a­tive con­struc­tion. In the case of comparative correlatives, things are in fact even worse: the string *hoe+cpr – Vfin – subject –... is not even well-formed – hoe+cpr is not usable in wh-questions and exclamatives alike, exclamatives using wat+A ‘what’ and questions employing hoeveel+A ‘how much’ instead (this is an instance of Corver’s 1997 ‘much-support’, which apparently, for reasons unknown to me, is not applic­able

 Marcel den Dikken

in com­parative correlatives, except for the dialectal cases mentioned in fn. 5, above). There are overriding fac­tors, there­fore, which categorically ban inversion of the subject and the finite verb in the headclause of com­parative correlatives with hoe+cpr. To produce a grammatical output, com­para­tive correlatives with hoe+cpr in the headclause have three options: (i) they can avoid the whole problem of raising to C by adjoining hoe+cpr to IP, as in (36b) (cf. (15b)); (ii) they can raise hoe+cpr to SpecCP (as in (36a)) and plug the C–head up with a complementiser; or (iii) they can raise hoe+cpr to SpecCP and leave the C–position empty. (36) a. [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL [hoe+cpr]j [Cʹ C={dat/Ø/*Vfin} [IP S... tj...]]]] b. [CP [RELCL [rel+cpr]i... ti...] [CP=HEADCL C=Ø [IP [hoe+cpr]j [IP S... tj...]]]] Of these three options, the third is the most marked one: it flies in the face of an otherwise robust property of Verb Second languages, that the C–position of a root clause must be occupied whenever the SpecCP position is occupied. Since the linear output of (iii) is the same as that of (i), one would therefore expect that at least some speakers would prefer (36b) in the case of hoe+cpr, thereby avoiding the selection of the version of (36a) with an empty C–head. But of course, for speakers selecting (36b), we expect there to be no chance of constructing a long-distance dependency: the PUB in (29) rules out movement into an IP–adjoined position via an intermediate touch-down in SpecCP. This, then, takes care of those speakers who reject (35a) in the headclause. For most speakers, however, the fact that hoe+cpr is a wh-phrase, combined with the fact that wh-phrases generally raise to SpecCP in headclauses in Dutch, seems to be sufficient reason to select (36a) even in the absence of any overt material filling C. For those speakers, long-distance dependencies involving hoe+cpr are perfectly wellformed: as we have seen, movement to SpecCP can proceed successive-cyclically via SpecCP, in keeping with the PUB in (29). Though, as we have seen, some speakers reject long-distance dependencies in the headclause in the case of hoe+cpr, there are significantly more speakers who reject (35b), with des te+cpr. Since des te+cpr is not a wh-phrase (unlike hoe+cpr), there is no particular incentive in this case to substitute des te+cpr for SpecCP. A significant piece of data in this connection seems to me to be the following. There is no doubt that hoe+cpr can land in a high SpecCP position: it can be immediately followed by the complementiser of ‘if ’(as in (37a); see (i) in fn. 7 for some examples and discussion). But des te+cpr does not seem to be able to land this high: it is never followed by of in the comparative correlative (i.e., (37b) does not occur, and I strongly reject it myself, while I do accept the sentences of the type in (37a) as representatives of spoken Dutch). (37) a. [HEADCL [hoe+cpr] [(of) (dat)...]] b. [HEADCL [des te+cpr] [(*of) (dat)...]] While it is possible to have des te+cpr in the headclause followed by the complementiser dat ‘that’, as in (37b), it is never followed by of, which is the higher of the two



Comparative correlatives and successive cyclicity 

complementisers (cf. wie of dat ‘who if that’ vs. *wie dat of); hoe+cpr, on the other hand, can be followed by either or both of of and dat (see also Coppen 1997). It is plausible, therefore, to assume that des te+cpr targets relatively lower positions in the tree than does hoe+cpr. Adjunction to IP is a desirable option for des te+cpr in the head­clause – particularly if there is no evidence in the linear string for the ‘activation’ of the CP layer. So in comparative correlatives in which, in the headclause, des te+cpr is immediately followed by the subject (i.e., there is no subject–verb inversion and there is no complementiser insertion in C either), one would expect (36b) to be the structure of choice. With (36b) chosen, the PUB will then rule out all long-distance dependencies involving des te+cpr. And indeed, what we find is that the most strongly rejected case of a long-distance dependency in comparative correlatives in Dutch is the case of such a dependency established in a headclause featuring des te+cpr immediately fol­lowed by the subject, as in (38a). (38) a.

%*...,

des te minder je denkt dat je begrijpt the-gen deg less you think that you understand

When des te+cpr in the headclause is not immediately followed by the subject, however, but instead is separated from the subject by either the raised finite verb or a lexical complementiser, the ban on long-distance dependencies is somewhat less severe: there are speakers (the ones identified as ‘K’ and ‘L’ in appendix A.4, in particular) for whom one or both of (38b,c) are noticeably better than (38a).20 (38) b.

%...,

des te minder denk je dat je begrijpt the-gen deg less think you that you understand

c.

%...,

des te minder dat je denkt dat je begrijpt the-gen deg less that you think that you understand

This, then, leads us to our last factor influencing the vicissitudes of long-distance dependencies in Dutch com­parative correlatives: the word order in the headclause. 3.3.4 Word order in the headclause While des te+cpr cannot be followed by of ‘if ’, as (37b) showed, it can be followed by other material that evidences the activation of the C–domain – fronted finite verbs or the complementiser dat ‘that’, as in (38b,c). Whenever des te+cpr, or hoe+cpr,21 for that matter, is immediately followed by material that signals explicitly that the [correl+cpr] sequence is not IP–adjoined but finds itself instead in a SpecCP position, that should give the language user the necessary cue to allow long-distance dependencies involving the com­parative. Some speakers do indeed report an effect of word order in the headclause on the grammaticality of long-distance dependencies in the headclause. The effect is not very systematic or robust, however (recall fn. 20 on independent factors weakening the effect of complementiser insertion). The analysis predicts a strong effect here – but while

 Marcel den Dikken

there are two speakers in my survey for whom the effect manifests itself to a signi­ficant extent, I must admit at this time that the theory’s predictions are stronger than the judge­ments I have collected so far seem to warrant. This is an issue that remains to be investigated further. 3.3.5 The effect of embedding on long-distance dependencies There is one last thing I would like to briefly bring up before closing this section on locality effects in Dutch comparative correlatives. Recall from Section 2 that comparative correlatives in root contexts must be as large as CP (in a Verb Second language such as Dutch), but that in non-root contexts, the comparative raises into an adjunction position to IP and the relative clause adjoins to IP as well, as in (21), repeated below. (21) C [IP [RELCL [rel+cpr]i... ti...] [IP=HEADCL [correl+cpr]j [IP S... tj...]]] The structure in (21) is generally available, also under non-bridge verbs (so long as the non-root clause is finite). Cases of CP recursion aside (recall (19)), (21) is the structure employed in all instances of comparative correlatives in non-root contexts. So cases of CP recursion aside, in embedded comparative correlatives the [correl+cpr] constituent is systematically adjoined to IP, never substituted for SpecCP. This is interesting in light of our discussion earlier in this section of the constraints imposed by the PUB in (29) on long-distance dependencies. Since successive-cyclic long-distance dependencies involving movement through SpecCP into an IP–adjoined position fail to comply with the PUB, we expect them to be ruled out by the grammar. In root contexts, we have so far found a significant amount of support for this prediction. What are our expectations for embedded comparative correlatives? With [correl+cpr] con­sis­tently adjoining to IP, we predict that it should be entirely unable to move successive-cyclically. Thus, if extraction from a finite CP is to proceed successive-cyclically, it is predicted that long-distance dependencies should generally degrade, for both hoe+cpr and des te+cpr, in the headclause of a comparative correlative when the construction as a whole is itself embedded; but there should be no effect of embedding on long-distance dependencies established in the relative clause (which will always be a CP, regardless of the struc­tural environ­ment in which the construction as a whole appears). Concretely, then, what we may expect to find is that the sentences in (40b–e) should be significantly worse than both (40a) and the corresponding sentences in (39b–e); but that the sentences in (41b–e) should be no worse than their counterparts in (41a) and (39b–e).22 (39) a. hoe meer je leest, hoe minder je begrijpt how more you read how less you understand b. men zegt vaak dat hoe meer je leest, hoe one says often that how more you read how



Comparative correlatives and successive cyclicity 



minder je begrijpt less you understand

c.

ik ben ervan overtuigd dat hoe meer je leest, hoe minder I am thereof convinced that how more you read how less je begrijpt you understand

d. het is alsof hoe meer je leest, hoe minder je begrijpt it is as-if how more you read how less you understand e. omdat hoe meer je leest, hoe minder je begrijpt because how more you read how less you understand (40) a. b. c. d. e.

hoe meer je leest, hoe minder je denkt dat je begrijpt how more you read how less you think that you understand men zegt vaak dat hoe meer je leest, hoe minder je denkt dat je begrijpt ik ben ervan overtuigd dat hoe meer je leest, hoe minder je denkt dat je begrijpt het is alsof hoe meer je leest, hoe minder je denkt dat je begrijpt omdat hoe meer je leest, hoe minder je denkt dat je begrijpt

(41) a. b.

hoe meer je denkt dat je begrijpt, hoe minder je in how more you think that you understand how less you in feite begrijpt fact understand men zegt vaak dat hoe meer je denkt dat je begrijpt, hoe minder je in feite begrijpt c. ik ben ervan overtuigd dat hoe meer je denkt dat je begrijpt, hoe minder je in feite begrijpt d. het is alsof hoe meer je denkt dat je begrijpt, hoe minder je in feite begrijpt e. omdat hoe meer je denkt dat je begrijpt, hoe minder je in feite begrijpt This is something that I have not yet tested in any systematic way. Sentences of the type in (40b–e) and (41b–e) were not part of the questionnaire on which I based the discussion in the previous subsections. For what they are worth, my own intuitions suggest that there is indeed an effect of the expected sort, with (40b–e) degraded compared to all other examples. But verifying the possible pre­diction regarding the embed­dability of comparative correlatives featuring a long-distance dependency in the headclause must be left as a topic for future investigation. Let me add before closing this section that there is an important reason why I have phrased the dis­cussion in the preceding paragraphs cautiously, in terms of possible predictions. The second paragraph of this section sets the tone for the rest of the discussion by tentatively basing itself on the premise that extraction from a finite CP proceeds successive-cyclically – a premise that is by no means a given. As we have seen in Section 3.1, it seems that languages have at their disposal two ways of extracting material from an embed­ded finite CP, one making a stopover in SpecCP and the other involving direct

 Marcel den Dikken

movement across CP. The question of whether the prediction outlined in the preceding paragraphs regarding long-distance dependencies inside embedded comparative correlatives will indeed present itself depends directly on the question of wheth­er the Dutch linguistic community uniformly avails itself of a successive-cyclic movement strategy when it comes to extraction out of a finite clause. At this point, therefore, it will be good to return to this latter question, which is the topic of Section 4.

4. On successive cyclicity At the outset of the discussion in Section 3, I pointed out that there is evidence, in the languages of the world, for successive-cyclic extraction from finite embedded clauses via the edge of CP, but that there are also good reasons to believe that extraction from finite CPs does not necessarily always proceed successive-cyclically. My analysis of the Hungarian facts in (26) and (27) in Den Dikken (2006) indicates that the two extraction scenarios in (28a) and (28b), repeated below, can be present side by side within one language, the two giving rise to different surface outputs. (28) a. XPi... [VP V [CP tiʹ [Cʹ C [TP (...) ti (...)]]]] b. XPi... [VP V [CP C [TP (...) ti (...)]]] I have cast my discussion of the locality sensitivity of Aʹ–dependencies in the Dutch comparative correlative entirely in the mould of (28a). We have seen that (28a), combined with the PUB in (29) and specific assumptions regarding the landing-site of the fronted [(cor)rel+cpr] constituent, makes a number of detailed and largely correct predictions in this empirical domain. It therefore seems entirely plausible that (28a) is indeed exploited in the analysis of comparative correlatives – at least by the speakers for whom we have found the kind of selective locality sensitivity involving (some of) the factors in (10). For speakers showing no locality sensitivity whatsoever (outside island contexts, of course) in the comparative correlative, I have assumed up to this point that they systematically substitute the fronted comparative for SpecCP in both constituent clauses. There is another possibility, however: such speakers could also be users of (28b), which (outside island contexts) will deliver no PUB-induced sensitivity to the final landing-site of the raised com­parative. So how do we find out whether (28a) or (28b) is employed? It is likely that we can independently determine the selection of (28a) or (28b) based on agreement. For Rackowski & Richards (2005), who argue for (28b) on the basis of the facts of Aʹ–extraction out of finite CPs in Tagalog, the C–head of the embedded clause in a case of long-distance Aʹ–depen­dency does not need to (and in fact probably does not) Agree with the extracted constituent: the matrix verb first Agrees with the embedded CP as a whole, after which this CP is ‘opened up’ (thanks to the Principle of Minimal Compliance) so that the verb can establish a second Agree relationship with the constituent that is to extract from CP; the C–head does not mediate



Comparative correlatives and successive cyclicity 

the extraction process in any way. By contrast, in a ‘classic’ successive-cyclic movement derivation à la (28a), the embedded C must engage in an Agree relationship with the extractee: satisfaction of C’s ‘EPP property’ by raising of the extractee into SpecCP is contingent upon the prior establishment of an Agree relationship between the bearer of the ‘EPP property’ (C) and the extrac­tee. We thus derive the following conclusion: (42) a. Agree between C and the extractee is required in (28a) b. Agree between C and the extractee is not required, and perhaps blocked, in (28b) On a strong reading of (42b) (with ‘blocked’ in lieu of ‘not required’), this gives us a correlation between suc­cessive-cyclic extraction and complementiser agreement such that whenever there is morphophonologically mani­fest complementiser agree­ment, there is successive-cyclic extraction through SpecCP. (The correlation does not hold in the other direction because it is entirely possible that in the absence of morphophonological evi­dence for successive-cyclic passage through SpecCP (i.e., lack of overt complementiser agreement), there still is movement through this position: morphological agreement paradigms often contain gaps; null morph­ol­ogy (i.e., morphology with a zero phonological exponent) should certainly be countenanced by the theory. Recall here the discussion in Section 3.1 of Hungarian, which arguably employs (28a) in light of the agree­ment facts in the matrix clause, but does not show morphophonological agreement with C itself.) If this way of looking at the world is on the right track, one might expect, in connection with the cases under investigation in this paper, that whenever a variety of Dutch presents overt evidence for success­ive-cyclic extraction (in the form of complementiser agreement), that variety should show selective local­ity sensitivity effects of the types discussed in Section 3. Varieties, on the other hand, which show no overt evidence for successive cyclicity could in principle be entirely impervious to depth of embed­ding (islands aside, as before): even with the fronted comparative adjoining to IP, such varieties could con­struct long-distance dependencies across a CP–boundary without violating the PUB in (29), exploiting (28b). Research in this area is still in the exploratory phase. In particular, it is not clear at this time exactly what the variation among speakers of Dutch with respect to long-distance dependencies in comparative cor­relatives is correlated with. But it is certainly interesting that all of my Flemish informants have seen their judge­ments land within the ballpark of selective locality sensitivity (group IV in the appendix, Section A.4) – this is interesting because Flemish varieties of Dutch are very well known for their complementiser agree­ment. Complementiser agreement is not in any way confined to Flemish varieties, however: it is found elsewhere in the Dutch-speaking world as well.23 Successive cyclicity should not be confined to Flanders, therefore; and indeed, several of my Northern Dutch informants (myself included) show selective locality sensitivity as well. Speakers who show such locality sensitivity do not necessarily have to exhibit overt morphological complementiser agree­ment: such

 Marcel den Dikken

agreement may be morphophonologically abstract. But in the other direction, one would reason­ably expect, in light of the above line of thought, that there will be a correlation: speakers who have overt morphological complementiser agreement will show selective locality sensitivity. Verifying this is something that remains to be done on a systematic basis. So far the signs are good, as I indicated above, but that is all that can be said at this stage.

5. Concluding remarks Comparative correlatives are ‘peripheral’ constructions that do not carry their syntactic analysis on their sleeve. Prying open their secrets is like opening up a Pandora’s box chock full of mysterious surprises whose existence one would never have suspected in the absence of a principled theory. The central topic of this paper, the locality restrictions on the Aʹ–dependencies involving the fronted comparative in the two constituent clauses of the Dutch compara­tive correlative, is a perfect collection of such prima facie mysterious surprises. What is interesting, and in fact very revealing, however, is that these mysteries start to unravel very quickly once one puts into place an anal­ysis of the comparative correlative built on the principles of Universal Grammar. By adopting the structures in (15), which are supported on the basis of both universal syntactic principles and language-particular but independently empirically verifiable properties, the restrictions on long-distance dependencies in the relative clause and the headclause of the Dutch comparative correlative readily translate into effects of (i) the Prin­ ciple of Unambiguous Binding (a UG principle proposed on entirely independent grounds by Müller & Sterne­feld 1993), (ii) a parameter determining whether or not extraction out of CP proceeds via a stopover in SpecCP (see (28a,b)), and (iii) microscopic lexical properties of the relative and correlative particles em­ployed, plus a small residue of what appear to be mostly stylistic or register-based idiosyncrasies (which, interestingly, can be related to similar stylistic or register-based idiosyncrasies found in Hungarian – these, then, arguably are not truly idiosyncratic either; but our present state of knowledge does not allow us to say much more about them at this time). One assumption that I resorted to in order to get the facts under control is apparently specific to the construction under investigation: the condition in (14), which (in a somewhat more broadly formulated form) says that the correlative particle and the relative clause must be adjacent. This condition argu­ably holds for (simple) correl­a­tive constructions generally, at least at some level of analysis – Bhatt’s (2003) careful dis­ cussion of correl­a­tives reaches the conclusion that ‘the Correlative Clause [must be] Merged as locally as possible with the objects which it is associated with’ (p. 525; original italics). For simple correl­atives in languages like Hindi/Urdu, this means (according to Bhatt; but see Davison, this vol., and Arsenijević, this vol., for alternative ways of deriving Bhatt’s locality restrictions without assuming movement of the correlative clause) that the correlative clause is merged directly with the projection of the



Comparative correlatives and successive cyclicity 

correlative particle (hence is adjacent to the correlative particle in the base), and then raises to an IP–adjoined position (so that, on the surface, the sentence-initial relative clause and the sentence-internal correlative particle need not be adjacent). For comparative correlatives in languages like Dutch, where the relative clause is presumably base-generated in (not moved into) an IP– or CP–adjoined position, Bhatt’s con­clusion entails that the relative clause must be surface-adjacent to the correlative par­ticle.24 Either way, cor­rel­ative constructions cross-linguistically satisfy the condition that the cor­relative particle be adjacent to the relative clause. Bhatt (2003:Section 5) argues that a condition of this sort can actually be derived from conditions on Merge, the syntactic structure-building mechanism. If a reduction of a condition of the type in (14) to general UG principles along Bhatt’s lines is ultimately successful, we can conclude that all of the apparent mysteries of the Dutch comparative correlative in the domain of locality reduce to a com­bination of UG principles, parameters, and language-specific lexical idiosyncrasies – precisely the expected result from the point of view of standard principles-and-parameters theory. Appendix: Questionnaire and speaker judgements A.1 Background The questionnaire used for this study included fifty sentences, all comparative correlatives, arranged into five sets of ten, each with the same internal organisation. The full list of sentences is given below, in the order of presentation in the questionnaire. (Glosses are added here for the benefit of readers unfamiliar with Dutch.) No attempt was made to randomise the order of presen­tation of the test sentences: from the nature of the stimuli, it would be eminently clear to all participants (all professional syntacticians) that locality was one of the key issues that this quest­ionnaire was trying to probe into. Items (#6) and (#0) in each set, involving in situ comparatives in the head­clause, served as filler items: this study was not particularly concerned with the distribution of in situ place­ment of the comparative in the headclause – future research should look into this further: the pattern exists, and for some speakers it is even available in long-distance contexts;25 but a large majority of my informants found the pattern to be entirely impossible (see also Citko, this vol.:sect. 2.2). A.2 Systematic pattern

(#1) (#2) (#3) (#4) (#5) (#6) (#7)

[relcl hoe+cpr... ] [relcl hoe+cpr... ] [relcl hoe+cpr... ] [relcl hoe+cpr... ] [relcl hoe+cpr... ] [relcl hoe+cpr... ] [relcl des te+cpr... ]

[headcl hoe+cpr [headcl hoe+cpr [headcl des te+cpr [headcl des te+cpr [headcl des te+cpr [headcl su V ... [headcl des te+cpr

... ... V] dat ... V] ... ... V] dat ... V] V ... ...] des te+cpr] ... ... V]

 Marcel den Dikken

(#8) [relcl des te+cpr...] (#9) [relcl des te+cpr...] (#0) [relcl des te+cpr...]

[headcl des te+cpr dat ... V] [headcl des te+cpr V ... ...] [headcl su V ... des te+cpr]

A.3 Test sentences A.3.1

Set 0: The baseline cases

(01) hoe meer je leest, hoe minder je begrijpt how more you read how less you understand (02) hoe meer je leest, hoe minder dat je begrijpt how more you read how less that you understand (03) hoe meer je leest, des te minder je begrijpt how more you read the-gen deg less you understand (04) hoe meer je leest, des te minder dat je begrijpt how more you read the-gen deg less that you understand (05) hoe meer je leest, des te minder begrijp je how more you read the-gen deg less understand you (06) hoe meer je leest, je begrijpt des te minder how more you read you understand the-gen deg less (07) des te meer je leest, des te minder je begrijpt the-gen deg more you read the-gen deg less you understand (08) des te meer je leest, des te minder dat je begrijpt the-gen deg more you read the-gen deg less that you understand (09) des te meer je leest, des te minder begrijp je the-gen deg more you read the-gen deg less understand you (10) des te meer je leest, je begrijpt des te minder the-gen deg more you read you understand the-gen deg less A.3.2

Set 1: Long-distance dependency in the relative clause; finite

(11)

hoe meer je denkt dat je begrijpt, hoe minder je in how more you think that you understand how less you in feite begrijpt fact understand

(12) hoe meer je denkt dat je begrijpt, hoe minder dat je in feite begrijpt (13) hoe meer je denkt dat je begrijpt, des te minder je in feite begrijpt (14) hoe meer je denkt dat je begrijpt, des te minder dat je in feite begrijpt (15) hoe meer je denkt dat je begrijpt, des te minder begrijp je in feite (16) hoe meer je denkt dat je begrijpt, je begrijpt in feite des te minder



Comparative correlatives and successive cyclicity 

(17) des te meer je denkt dat je begrijpt, des te minder je in feite begrijpt (18) des te meer je denkt dat je begrijpt, des te minder dat je in feite begrijpt (19) des te meer je denkt dat je begrijpt, des te minder begrijp je in feite (20) des te meer je denkt dat je begrijpt, je begrijpt in feite des te minder A.3.3

Set 2: Long-distance dependency in the headclause; finite

(21) hoe meer je leest, hoe minder je denkt dat je begrijpt how more you read how less you think that you understand (22) hoe meer je leest, hoe minder dat je denkt dat je begrijpt (23) hoe meer je leest, des te minder je denkt dat je begrijpt (24) hoe meer je leest, des te minder dat je denkt dat je begrijpt (25) hoe meer je leest, des te minder denk je dat je begrijpt (26) hoe meer je leest, je denkt dat je des te minder begrijpt (27) des te meer je leest, des te minder je denkt dat je begrijpt (28) des te meer je leest, des te minder dat je denkt dat je begrijpt (29) des te meer je leest, des te minder denk je dat je begrijpt (30) des te meer je leest, je denkt dat je des te minder begrijpt A.3.4

Set 3: Long-distance dependency in the relative clause; infinitival

(31) hoe meer je denkt te begrijpen, hoe minder je in feite begrijpt how more you think to understand how less you in fact understand (32) hoe meer je denkt te begrijpen, hoe minder dat je in feite begrijpt (33) hoe meer je denkt te begrijpen, des te minder je in feite begrijpt (34) hoe meer je denkt te begrijpen, des te minder dat je in feite begrijpt (35) hoe meer je denkt te begrijpen, des te minder begrijp je in feite (36) hoe meer je denkt te begrijpen, je begrijpt in feite des te minder (37) des te meer je denkt te begrijpen, des te minder je in feite begrijpt (38) des te meer je denkt te begrijpen, des te minder dat je in feite begrijpt (39) des te meer je denkt te begrijpen, des te minder begrijp je in feite (40) des te meer je denkt te begrijpen, je begrijpt in feite des te minder A.3.5

Set 4: Long-distance dependency in the headclause; infinitival

(41) hoe meer je leest, hoe minder je denkt te begrijpen how more you read how less you think to understand (42) hoe meer je leest, hoe minder dat je denkt te begrijpen (43) hoe meer je leest, des te minder je denkt te begrijpen

 Marcel den Dikken

(44) hoe meer je leest, des te minder dat je denkt te begrijpen (45) hoe meer je leest, des te minder denk je te begrijpen (46) hoe meer je leest, je denkt des te minder te begrijpen (47) des te meer je leest, des te minder je denkt te begrijpen (48) des te meer je leest, des te minder dat je denkt te begrijpen (49) des te meer je leest, des te minder denk je te begrijpen (50) des te meer je leest, je denkt des te minder te begrijpen A.4 Judgements The table below brings together the judgements reported by the 17 native speakers who responded to my questionnaire. The respondents are set out along the horizontal axis; the test sentences are listed along the vertical axis. All numbers on the vertical axis correspond to the example numbers in A.3. The letters on the hori­zontal axis represent respondents (who will remain unidentified); the columns are arranged in such a way as to provide natural groupings, indicated at the bottom of the table. Significant effects of locality are high­lighted with the aid of shading. The table is cut up into five speaker groups, characterised below the table. Table 1 A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

01

OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK

02

OK OK OK ?

03

OK OK OK OK ?

04

OK OK ?

05

OK OK OK OK OK OK ✱

OK ??

✱

OK OK OK ??

OK ✱

OK

06

OK ??

✱

?*

**

✱

✱

✱

OK OK *?

✱

✱

✱

**

07

OK OK OK OK **

?*

??

✱

??? ??

OK OK ?

OK OK ✱

✱

08

OK OK OK ??

**

?*

??? ✱

??? ?

OK OK ??

?

OK ✱

✱

09

OK OK ✱

OK ✱

✱

✱

✱

??

??

OK OK ?

?

OK ✱

✱

10

OK ??

✱

**

✱

**

✱

✱

✱

✱

✱

✱

**

11

OK OK OK OK OK OK OK OK OK OK OK OK OK ?

OK OK ✱

12

OK OK OK OK OK ?

OK OK ✱

✱

✱

??

✱

OK ?

?

(?)

OK ?

??

(?)

OK ??? OK ?

?*

?

OK ??? OK OK OK ?

(?)

OK ??

OK OK OK ?

OK ✱

✱

OK OK OK OK ✱

✱

OK ✱

OK OK OK ?

?

?

?

OK ok? ✱

✱



Comparative correlatives and successive cyclicity  13

OK OK OK OK ?

??

(?)

OK ?

OK ?

14

OK OK OK ?

??

?*

?

OK ??

OK OK OK ?

15

OK OK OK ?

OK OK ✱

OK ??

✱

OK OK OK ?*

OK OK ✱

16

OK ??

✱

OK **

✱

✱

✱

OK OK *?

✱

✱

??

**

17

OK OK OK OK **

✱

??

✱

??? ??

?

✱

?

✱

✱

18

OK OK OK ?

**

✱

??? ✱

??? ?

OK OK ??

✱

?

✱

✱

19

OK OK OK OK **

✱

?*

✱

?

??

OK ?

?

✱

?

✱

✱

20

OK ??

✱

**

✱

✱

✱

✱

??

✱

✱

✱

✱

21

OK OK OK OK OK OK OK OK ?(?) OK OK OK ?

?

✱

?/* ✱

22

OK OK OK OK OK ?

(?)

OK ?(?) OK OK OK ??

?

✱

✱

✱

23

OK ?

(?)

OK ??

OK ✱

✱

?

✱

✱

✱

✱

24

OK OK OK ?

?*

?

OK ??

OK ?

?

??

✱

✱

✱

✱

25

OK OK OK OK ✱

?

(*)

OK OK ?*

?

✱

?*

✱

✱

✱

✱

26

OK ??

OK ✱

✱

✱

✱

✱

OK ✱

✱

✱

✱

✱

27

OK OK OK OK **

?*

?

✱

?

?*

✱

??

??

✱

✱

✱

✱

28

OK OK OK ?

**

✱

??

✱

??

??

?

OK ?*

✱

✱

✱

✱

29

OK OK OK OK **

✱

*?

✱

??? ✱

?

??

✱

✱

✱

✱

✱

30

OK ✱

**

✱

✱

✱

✱

OK ✱

✱

✱

✱

✱

31

OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK ✱

32

OK OK OK ?

33

OK OK OK OK ?

34

OK OK OK ??

35

OK OK OK OK ✱

36

OK ??

37

OK OK OK OK **

✱

38

OK OK OK ?

**

39

OK OK OK OK **

✱

✱

✱

✱

**

OK OK OK ??

✱

✱

✱

✱

✱

✱

?

✱

**

**

?

?

?

(?)

OK ??

?*

(?)

OK OK OK ?

?*

?

OK ?

OK OK OK ?

OK ✱

OK ?

OK **

✱

?

✱ ✱

OK ?

??

✱

OK OK ?

OK OK OK ?

?

?

OK OK ✱ ?

?

✱

OK OK ✱

OK OK OK OK OK ✱ ?

OK OK ✱

?*

OK OK OK ?

OK OK ✱

✱

✱

OK OK ?*

✱

✱

✱

✱

✱

??

??

?

OK OK ✱

✱

??

✱

??? ?*

OK OK ??

?

OK ✱

✱

✱

✱

?(?) ✱

OK OK ?

?

OK ✱

✱

OK ?

 Marcel den Dikken 40

OK ✱

41

OK OK OK OK OK OK OK OK ?

OK OK OK OK OK OK ?/* NB

42

OK OK OK ??

OK ?

(?)

OK ?

OK OK OK ?

43

OK ?

?

??

(?)

OK ?(?) OK ?

OK OK OK OK ✱

NB

44

OK OK OK ??

??

?*

?

ok? ?(?) OK ?

OK ?

45

OK OK OK OK ✱

OK ✱

OK (?)

?*

46

OK ??

OK **

✱

✱

47

OK OK OK OK **

?*

?

✱

??

48

OK OK OK OK **

✱

??

✱

??? ?

49

OK OK OK OK **

✱

✱

✱

50

OK ✱

✱

**

✱

I group I group II

✱

✱

OK ?

✱

✱

II

✱

✱

**

**

**

✱

**

III

✱

✱

✱

✱

OK ?

✱

?

✱

✱

OK ✱

✱

✱

?

OK ✱

✱

OK OK ?

??

OK ✱

NB

✱

??? OK ✱

✱

✱

✱

✱

??

?

OK OK ✱

✱

OK OK ??

?

OK ✱

✱

OK ??

OK OK ?

??

OK ✱

✱

??? ✱

✱

??

✱

✱

✱

OK ?

IV

✱

✱

V

– no distinctions of any kind: all fifty sentences are acceptable – with the exception of (#6) and (#0), all sentences are generally acceptable or at least passable group III – des te+cpr is generally rejected in the relative clause (cf. (#7)–(#0)); some speakers of group IV and both speakers of group V have an effect of this type as well, but in addition, these speakers make further distictions not made by group III speakers group IV – long-distance dependencies in the headclause, and for some to some extent also in the rela­tive clause are impossible or degraded, especially with des te+cpr but for some also with hoe+cpr; this locality effect manifests itself only in finite contexts: there are no correspond­ing degrations in (31)–(50), and for some speakers there is an effect of word order as well group V – degradation of long-distance dependencies with hoe+cpr in both finite and non-finite contexts (for speaker R in both the headclause and the relative clause, and for speaker Q primarily in the headclause only) nb re: R Speaker R did not judge the examples in (41)–(50) as given but substituted hoe meer/minder and des te meer/minder with hoe meer/minder boeken ‘the more/less books’ and des te meer/minder boeken ‘the-gen deg more/less books’, indicating that (41), (43) and (45) become marginal (‘?’) under such circumstances (and tacitly implicating that, in their given forms, the examples in (41), (43) and (45) are unacceptable).



Comparative correlatives and successive cyclicity 

Acknowledgements I am very grateful to Anikó Lipták for giving me the opportunity to publish this piece in the present volume, and for important discussion of some of the questions addressed in this paper. In addition, I thank both her and the two reviewers of this paper (one anonymous and one clearly identifying himself as Klaus Abels) for their insightful comments on specific points made in the paper. Finally, I would like to acknowledge the huge debt of gratitude I owe to the many colleagues who returned the questionnaire I sent them – in alphabetical order: Bas Aarts, Peter Ackema, Sjef Barbiers, Hans Broekhuis, Peter-Arno Coppen, Jeroen van Craenen­broeck, Liliane Haegeman, Jack Hoeksema, Eric Hoekstra, Hilda Koopman, Jan Koster, Ad Neeleman, Gertjan Postma, Henk van Riemsdijk, Guido Vanden Wyngaerd, and Jan-Wouter Zwart.

References Arsenijevic, Boban. this vol. {Relative {conditional {correlative clauses}}}. Beck, Sigrid. 1997. On the semantics of comparative conditionals. Linguistics and Philosophy 20. 229–71. Bennis, Hans. 1995. The meaning of structure: The wat voor construction revisited. In Marcel den Dikken & Kees Hengeveld (eds), Linguistics in the Netherlands 1995. Amsterdam: John Benjamins. Besten, Hans den. 1977. On the interaction of root transformations and lexical deletive rules. Unpublished paper, University of Amsterdam; published in 1983 in Werner Abraham (ed.), On the formal syntax of the Westgermania. Amsterdam: John Benjamins; reprinted in 1989 in Hans den Besten, Studies in West Germanic syntax. Ph.D. dissertation, Tilburg University. Bhatt, Rajesh. 2003. Locality in correlatives. Natural Language and Linguistic Theory 21. 485–541. Bhatt, Rajesh & Roumyana Pancheva. 2006. Conditionals. In Martin Everaert & Henk van Riemsdijk (eds), The Blackwell Companion to Syntax, Vol. I, Ch. 16. Oxford: Blackwell. Bobaljik, Jonathan. 2001. Floating quantifiers: Handle with care. In Lisa Cheng and Rint Sybesma (eds), The second Glot International state-of-the-article book. Berlin: Mouton de Gruyter. 107–48. Chomsky, Noam. 1977. On wh-movement. In Peter Culicover, Thomas Wasow and Adrian Akmajian (eds), Formal syntax. New York: Academic Press. 71–132. Citko, Barbara. 2004. On headed, headless, and light-headed relatives. Natural Language and Linguistic Theory 22. 95–126. Citko, Barbara. this vol. What don’t wh-questions, free relatives, and correlatives have in common? Coppen, Peter-Arno. 1997. Linguistisch miniatuurtje XXXIV: “Hoe langer hoe gekker”’, NederL. Elektronisch tijdschrift voor de Neerlandistiek. Corver, Norbert. 1997. Much-support as a last resort. Linguistic Inquiry 28. 119–64. Culicover, Peter and Ray Jackendoff. 1999. The view from the periphery: The English comparative correlative. Linguistic Inquiry 30. 543–71. Davison, Alice. this vol. Adjunction, features and locality in Sanskrit and Hindi/Urdu correlatives. Dikken, Marcel den. 2005a. Comparative correlatives comparatively. Linguistic Inquiry 36. 497–532. Dikken, Marcel den. 2005b. Comparative correlatives and Verb Second. In Jan Koster and Henk van Riemsdijk (eds), Germania et alia: A linguistic webschrift for Hans den Besten. Available at http://odur.let.rug.nl/~koster/DenBesten/DenDikken.pdf.

 Marcel den Dikken Dikken, Marcel den. 2006. When Hungarians agree (to disagree): The fine art of ‘phi’ and ‘art’. Ms., CUNY Graduate Center. Gervain, Judit. 2003. Syntactic microvariation and methodology: Problems and perspectives. Acta Linguistica Hungarica 50. 405–34. Gervain, Judit. 2005. Two strategies of focus-raising: Movement and resumption. In Christopher Piñón & Péter Siptár (eds), Approaches to Hungarian 9. Papers from the Düsseldorf conference. Budapest: Akadémiai Kiadó. Iatridou, Sabine & Anthony Kroch. 1992. The licensing of CP-recursion and its relevance to the Germanic verb-second phenomenon. Working Papers in Scandinavian Syntax 50. 1–24. Johnson, Kyle. 1991. Object positions. Natural Language and Linguistic Theory 9. 577–636. Kenesei, István. 1992. Az alárendelt mondatok szerkezete. Strukturális magyar nyelvtan I. Budapest: Akadémiai Kiadó. Koppen, Marjo van. 2005. One probe – two goals. Aspects of agreement in Dutch dialects. Ph.D. dissertation, ULeiden/ULCL. Law, Paul. 1991. Effects of head movement on theories of subjacency and proper government. Ph.D. disser­tation, MIT. McCawley, James. 1988. The comparative conditional constructions in English, German and Chinese. In Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society, BLS, Berkeley. 176–87. McCawley, James. 1998. The syntactic phenomena of English (2nd edn). Chicago: University of Chicago Press. McCloskey, James. 2000. Quantifier Float and wh-movement in an Irish English. Linguistic Inquiry 31. 57–84. Müller, Gereon & Wolfgang Sternefeld. 1993. Improper movement and unambiguous binding. Linguistic Inquiry 24. 461–507. Pesetsky, David. 1989. Language particular processes and the Earliness Principle. Paper presented at GLOW Utrecht; handout available at: http://web.mit.edu/afs/athena.mit.edu/ org/l/linguistics/www/bibliography/pesetsky.html Postma, Gertjan. 1995. Zero semantics. Ph.D. dissertation, University of Leiden/HIL. Rackowski, Andrea & Norvin Richards. 2005. Phase edge and extraction: a Tagalog case study. Linguistic Inquiry 36. 565–99. Rebuschi, Georges. this vol. Basque correlatives and their kin in the history of Northern Basque. Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19. 425–49 Srivastav, Veneeta. 1991. The syntax and semantics of correlatives. Natural Language and Linguistic Theory 9. 637–86. Taylor, Heather. 2006. The syntactically well-behaved comparative correlative. Paper presented at GLOW Barcelona; abstract available in GLOW Newsletter 56. Travis, Lisa. 1984. Parameters and effects of word order variation. Ph.D. dissertation, MIT. Vikner, Sten. 1995. Verb movement and expletive subjects in the Germanic languages. Oxford/ New York: Oxford University Press. Vries, Mark de. 2004. Hoofd-interne relatiefzinnen in het Nederlands: Over vrije relatiefzinnen, pied piping en links-dislokatie. Nederlandse Taalkunde 9. 193–230. Wurmbrand, Susanne. 2003. Infinitives: Restructuring and clause structure. Berlin: Moutin de Gruyter. Zwart, Jan-Wouter. 1997. Morphosyntax of verb movement: A minimalist approach to the syntax of Dutch. Dordrecht: Kluwer.



Comparative correlatives and successive cyclicity 

Endnotes 1. Bhatt (2003) argues that simple correlatives involve a derivation in which the relative clause, which forms a con­stituent with the projection of the correlative particle in the base, ends up in its surface clause-initial adjunc­tion position via move­ment. For comparative correlatives (at least in Dutch) Bhatt’s arguments cannot be reproduced. I will assume base-adjunction for simplicity. 2. Here and throughout the paper, ‘cpr’ stands for ‘comparative’. Note that op/prt+cpr in (2b) forms a constituent, pied-piped to initial position. See Den Dikken (2005a) for detailed dis­cussion of the internal structure of op/prt+cpr, irrelevant in what follows. 3. In my questionnaire, I chose an example not containing the negatively polar auxiliary hoeven found in (7a), whose licensing by the fronted negative comparative minder ‘less’ introduces additional complications blurring the picture. 4. Klaus Abels (p.c.) suggests that the ‘noisiness’ of the speaker judgements on (9) may be due in part to the fact that verbs like understand, and German verstehen ‘understand’, allow unspecified object deletion (I understand; German ich verstehe). Though this could indeed have been an interfering noise factor if the experiment had been conducted on English or German using counter­parts of the sen­tences in (9), for Dutch this factor is actually not in effect: Dutch begrijpen does not allow unspecified object deletion (thus, see ik begrijp *(het) ‘I understand it’). There must be long-distance dependencies involved in the Dutch examples in (9), therefore. 5. Jack Hoeksema (p.c.) has pointed out to me that in his variety of Dutch, there are two further possibilities, illustrated in (ia,b). Note that these show the same alternation as (11a,b), with a wh-element (hoeveel ‘how much’) in the relative clause and either a wh-element or a determiner/demonstrative (des, just as in (11b,c)) in the headclause. I would expect them to pattern like (11a,b), but I have not investigated this in any detail. (i) a.

hoeveel te meer je leest, hoeveel te minder how-much deg more you read how-much deg less je begrijpt you understand

a. hoeveel te meer je leest, des te minder je begrijpt how-much deg more you read the-gen deg less you understand 6. According to the Algemene Nederlandse Spraakkunst, however, inversion after des te+cpr is the preferred pattern – a claim with which I personally concur. The Algemene Nederlandse Spraakkunst also points out that inversion following hoe+cpr in the headclause, while not the preferred word order, is ‘not excluded’. In two earlier papers of mine on comparative correlatives (Den Dikken 2005a,b), I reported that inversion following hoe+cpr is impossible, a claim which, in light of the ANS’s assertion, I now have to slightly adjust. But I should add that none of the many examples of comparative correlatives with hoe+cpr that I found on the web (using Google) involved inversion, whereas many of the headclauses containing clause-initial des te+cpr featured the finite verb in a position immediately following the comparative. (I did not include inversion in headclauses with hoe+cpr in my question­naire because I had assumed that it does not occur; I did not want to lengthen the already quite substantial questionnaire even further.) The least we can say, it seems to me, is that, even if apparently ‘not excluded’, inversion after hoe+cpr is much less common than inversion after des te+cpr; the latter oscillates relatively

 Marcel den Dikken freely between inversion and non-inversion, depending to some extent on whether the relative clause features des te+cpr or hoe+cpr. 7. I only included dat ‘that’ in my questionnaire, but one of my respondents pointed out a preference for of ‘if ’ instead of dat. I did indeed find several examples of hoe meer of ‘how more if ’ on the web; but des te meer of is not attested – something to which I will return. Complementiser insertion between the clause-initial comparative and the subject is possible in principle in both con­stituent clauses of the comparative correlative (though in my questionnaire I only included examples featuring a complementiser in the headclause) – but not if the relative pronoun is des (see Den Dikken 2005b). An interesting example is (i), a rhyming proverb: (i) hoe langer of men slaapt, how longer if one sleeps hoe korter dat men leeft; how shorter that one lives hoe wijder of men gaapt, how wider if one yawns hoe minder dat men heeft how less that one has [www.dbnl.org/tekst/heij007alki01/heij007alki01_0173.htm] Here, systematically, of is used in the relative clause and dat in the headclause, which shows that, in the variety of Dutch that (i) represents, there is a formal means of differentiating the two con­ stituent clauses of the comparative correlative even when the (cor)relative particles used in the two clauses are identical: of ‘if ’ marks the relative clause ([+wh]), and dat ‘that’ marks the headclause ([–wh]), in a way that is perfectly compatible with the correlative analysis of the construction. In light of the fact that combinations of of and dat, in that order, are generally found following wh-phrases in the spoken vernacular (cf. e.g. ik weet niet wie of dat Marie gekust heeft ‘I know not who if that Marie kissed has’), one would expect such combinations to be possible in comparative correlatives as well, with hoe+cpr – and indeed, hoe meer of dat S strikes speakers as grammatical in spoken colloquial Dutch (see also Coppen 1997). 8. The V2 pattern, with the finite verb fronted to C in the headclause, is of course readily expected in light of the fact that Dutch is a V2 language. But the fact that the headclause, when serving as the root clause, can have a lexical complementiser in its C–head raises an immediate question, given that root clauses with complementisers as their heads are not particularly common – in Dutch they are normally confined in their distribution to exclamatives (dat je dat durft, zeg! ‘boy, that you dare to do that!’) or wishes (dat je er nog lang van moge genieten ‘may you enjoy it for years to come’); but a root comparative correlative with a lexical complementiser in the headclause does not have the interpretation of an exclamative or a wish at all: it is illocutionarily on a par with comparative correlatives with a V2 or dat-less headclause. See Den Dikken (2005b) for some relevant discussion of the appearance of dat in the headclause of Dutch comparative correlatives; perhaps related is the mysterious optional appearance of the conjunction eta ‘and’ between the two clauses of the Basque complex correlative (see example (26) in Rebuschi, this vol.). I would like to add here that one of the respondents to my questionnaire pointed out that dat in the headclause is better if dat is also included in the relative clause. This enhances Den Dikken’s (2005b) conclusion that the occurrence of dat in the headclause is a ‘harmony’ effect, with the headclause mimicking the structure of the relative clause. How to formally analyse this effect remains an open question.



Comparative correlatives and successive cyclicity  9. I placed ‘comparative’ in parentheses here because, as Den Dikken (2005b) points out, archaic non-comparative correlatives (preserved in contemporary Dutch in the form of proverbs) show the same peculiar word order. The examples in (ia–d) are cases in point; (idʹ), with its V2 order in the headclause, is grammatical but only has a literal interpretation. (i) a. wat niet weet, †dat/wat niet deert what not knows d-word/what not harms ‘what the eye doesn’t see, the heart doesn’t grieve about’

(proverb)

b. wie dan leeft, †die/wie dan zorgt who then lives d-word/who then cares ‘we’ll cross that bridge when we get to it’

(proverb)

c. wie niet waagt, †die/wie niet wint who not tries d-word/who not wins ‘nothing ventured, nothing gained’

(proverb)

d. wie het eerst komt, †die/wie het eerst maalt who the first comes d-word/who the first grinds ‘first come, first served’

(proverb)

dʹ. wie het eerst komt, die maalt het eerst who the first comes d-word grinds the first ‘who comes first gets the first shot at grinding’

(literal)

10. It is here that the conditional nature of the comparative correlative (noted by McCawley 1988, 1998 and discussed in some detail in Beck 1997) asserts itself and makes it form a natural class with other sentence-initial conditionals. See Den Besten (1977:fn. 3) for early discussion of some of the cases exemplified in (16); and see Bhatt & Pancheva (2006) and Arsenijević (this vol.), as well as references cited there, for the strong links between conditionals and correlatives. 11. On the non-finite cases, see also Culicover & Jackendoff (1999: 550–51), for English. Embedding a comparative correlative in a finite clause under a non-bridge verb or in a finite adjunct clause is reasonably common. The reader may wish to verify this at his/her leisure by for instance typing in the string “alsof hoe meer” ‘as if how more’ or “omdat hoe meer” ‘because how more’ into his/her favourite search engine: these strings produce plenty of hits, all clearly instances of the comparative correlative construction. 12. To see this, consider the English and Dutch examples in (i) and (ii), with the leftmost adverbs assumed to be IP–adjoined.

(i) a. ik denk dat Jan de computer niet op de juiste manier heeft uitgeschakeld

I think that yesterday Jan yesterday the computer yesterday not on the right way has off-switched

‘I think that Jan did not switch off the computer in the proper way yesterday’

b. waarom heeft <*gisteren> Jan de computer niet op de juiste manier uitgeschakeld?

why has yesterday Jan yesterday the computer yesterday not on the right way offswitched

‘why didn’t Jan switch off the computer in the proper way yesterday?’

 Marcel den Dikken

(ii) a. <probably> John <probably> has <probably> not switched off the computer in the proper way



b. why has <*probably> John <probably> not switched off the computer in the proper way?

13. Rackowski & Richards (2005) note that other familiar evidence for successive cyclicity, primarily hav­ing to do with com­plementiser agree­ment, can be taken care of without movement through SpecCP on an Agree-based approach. But the Q–Float facts (which they do not address) are less easy to take care of without a stopover in SpecCP. 14. There are two nontrivial qualifications in order here. First, it is not obvious that Sportiche’s (1988) original Q–stranding approach to Q–Float is literally correct (see e.g. Bobaljik 2001 for relevant discussion). And secondly, it is not immediately evident that the matrix verb, say in the examples in (24), indeed has not undergone any movement: though it is clear that, if it is to have raised, it has not raised very far, it could have undergone short-distance movement to a functional head position immediately outside VP (see e.g. Johnson 1991 and Pesetsky 1989 for arguments to the effect that in standard English the lexical verb does in fact leave its VP). If so, one could argue that all in (24c) finds itself in a position adjoined to the VP (or vP) of the matrix clause, not in SpecCP. 15. That is, it is not the case that the focus in (26) originates in the matrix clause and binds a resumptive pronoun downstairs. Such is in fact the case when the morphologically singular but notionally plural noun phrases in (26a,b) co-occur with a third person plural finite verb in the downstairs clause (i.e., jöjjenek) – something which a subset of speakers of Hungarian allow (see Gervain 2003, 2005 for details). But in (26a,b), where the embedded verb is singular (jöjjön), only a long movement scenario is available. 16. The sentences in (30) are presented here with Müller & Sternefeld’s (1993: 465) original structural annotations. 17. Müller & Sternefeld (1993:  466, fn. 3) duly note that the idea that scrambling is Aʹ– movement is not uncontroversial (giving some key references to seminal contributions to this debate from the late ’eighties and early ’nineties) – famously, scrambling does not give rise to crossover effects, instead feeding binding. But I agree with Müller & Sternefeld that scrambling is indeed Aʹ–move­ment, certainly not garden-variety A–movement. 18. The fifth line of the first stanza of the Hungarian national anthem (‘Hymnusz’) contains a well-known example of this type: (i) bal sors akit régen tép adverse fate.nom a-who-acc long tear.apart ‘(he) whom bad fate has long torn apart’

(Hungarian)

19. It certainly is not a focus position: relative pronouns are not focusable. It could be a topic position – word orders in which the relative pronoun immediately precedes a topic which in turn immediately precedes the subject are possible in Hungarian free rela­tives, even in those featuring a topicalised constituent to the left of the wh-pronoun (e.g.: a könyvet akinek tegnap Péter odaadta ‘the book-acc a-who-dat yesterday Péter(nom) pv-gave, i.e. the person to whom Péter gave the book yesterday’ – Anikó Lipták, p.c.). On the wh-element in the subclause of correlatives being in topic position, see also Arsenijević (this vol.) and Rebuschi (this vol.). 20. The effect is clearer in (38b) than it is in (38c) – but this is due, no doubt, to the fact that complementiser insertion in the headclause of the comparative correlative is generally a marked



Comparative correlatives and successive cyclicity  option, felt to be typical of the spoken vernacular. Since long-distance dependencies are more likely to be found in the written language, a combination of dat-insertion and a long-distance dependency in the headclause thus leads to a bit of a register clash, which is likely to be responsible for the less systematic effect of dat-insertion on long-distance extraction of the comparative. 21. For hoe+cpr, most speakers do not accept a fronted finite verb to follow it in a V2 configuration (recall (11) and fn. 6). So for hoe+cpr the only explicit signal that it occupies SpecCP (besides the grammaticality of long-distance dependencies) is the presence of a lexical complementiser to its immediate right. 22. The glossing of the examples in (40) and (41) is kept to a minimum to save space – the only difference between (39) and (40) is the underlined portion of (39), glossed as in (40a); and in (41) the glosses for all the ingredients of the b–e sentences can be found in either in the gloss for (42a) or in (39). 23. In the Netherlands, complementiser agreement is found, for instance, in the varieties spoken in Katwijk (on the west coast), Aalten (in the east, near the border with Germany) and in Friesland (in the north). I refer the reader interested in complementiser agree­ment, its geographical distribution across the Dutch-speaking world, and its analysis in the minimalist theory of syntax to Van Koppen (2005), where references to the key literature are also provided. 24. Placing the constituent formed by the correlative particle and the comparative in sentenceinternal position (in the ‘Mittelfeld’) in the headclause is possible to some extent for a few speakers (myself included), but as the appendix shows (see esp. fn. 25, below), the overwhelming majority of Dutch speakers robustly reject comparative correlatives with non-fronted [correl+cpr]. (To see this, the interested reader should inspect examples (#6) and (#0) in each of the sets of ten example sentences in the appendix.) 25. No examples of unembedded in situ correl+cpr were found on the web (using Google). Though I did not specifically search for in situ correl+cpr, it is interesting that as part of my search for embedded comparative correlatives I found several unmistakable instances of embedded in situ correl+cpr. I reproduce them below. Anikó Lipták (p.c.) notes correctly that (i), (ii) and (iii) are different from the kinds of comparative correlatives discussed in the paper in involving a ‘bare’ comparative in the headclause, without a correl particle, and that these three examples all feature ook in the headclause. She suggests plausibly that ook is needed to establish the semantic link with the subclause that the correl particle otherwise helps establish. I have nothing to contribute to the semantics of (comparative) correlative constructions in this paper, other than the negative note (prompted by an anonymous reviewer) that it differs from that of contrastive left-dislocation constructions like wie dit leest, die is gek ‘who reads this, dem is crazy’ (which, like comparative correlatives, feature an initial free relative plus a demonstrative in the headclause, but in these left-dislocation constructions the free relative and the demonstrative must be coreferent whereas in comparative correlatives they are not; see De Vries 2004 for detailed discussion of the syntax and semantics contrastive left-dislocation constructions). (i)

tegelijkertijd besef ik dat hoe meer je weet je ook minder at.the.same.time realise I that how more you know you also less gaat weten go know

(ii) bedenk wel dat hoe meer je naar de kant van het laagst toegaat be-think dprt that how more you to the side of the lowest prt-go

 Marcel den Dikken je foto ook steeds your photo also gradually

minder in kwaliteit less in quality

wordt becomes

(iii)

denk eraan dat des te hoger je ramp wordt je ook meer think thereof that the-gen deg higher your ramp becomes you also more aandacht besteed [sic] aan de kracht van het platform en een hekje attention pay to the power of the platform and a fence of zoiets of something.like.that

(iv)

daarnaast is het waar dat hoe meer de overheid supporters there-beside is it true that how more the government supporters behandelt, als collectiviteit treats as collectivity zo’n groep zich ook des te meer als collectiviteit gaat gedragen such-a group refl also the-gen deg more as collectivity goes behave

(v) daarom geldt dat hoe meer groene elektriciteit er wordt therefore holds that how more green electricity there is gevraagd door de consument, demanded by the consumer er des te meer geïnvesteerd wordt in nieuwe installaties voor there the-gen deg more invested is in new installations for groene elektriciteit green electricity en hoe minder milieuonvriendelijkere grijze elektriciteit er geproduceerd hoeft te worden (vi) omdat, hoe meer de mens aan de goederen van deze aarde hecht, because how more the man to the goods of this earth attaches hij des te minder zijne bestemming begrijpt he the-gen deg less his destination understands [from Allan Kardec, Het boek der geesten, 1857, translated from the original French by J.G. Plate, 1875] (vii)

omdat hoe meer verdrukking zij hebben, zij des te meer graden because how more repression they have they the-gen deg more degrees van heerlijkheid zullen hebben of divinity will have [in the Dutch translation of Thomas Watson’s 1692 A Body of Practical Divinity]

part 3

The matching effect

On the matching requirement in correlatives Tommi Tsz-Cheung Leung Correlatives are characterized by a matching requirement, according to which the number of relative morphemes in the relative clause has to equal the number of demonstrative morphemes found in the main clause. This paper offers a novel formalization for the matching requirement of correlatives. We claim that the matching requirement arises out of the syntactic configuration that contains a ‘doubling constituent’ formed by the relative morpheme and the demonstrative morpheme. The function of a doublet is to establish a co-dependent relation between the two morphemes in question. The surface representation of correlative constructions is the result of various derivations such as the application of sideward movement (Nunes 2004) that displaces the two codependent morphemes within the doublet across two clauses. Keywords: single correlatives; multiple correlatives; matching requirement; doubling constituent; sideward movement

1. Introduction Correlative constructions (henceforth CORs) refer to a particular relative construction in which the relative clause is left-adjoined to the main clause (Andrews 1985, Keenan 1985, Srivastav 1991, Dayal 1996, Bhatt 2003, McClawley 2004, among others).1 Typologists discovered that CORs constitute the major strategy of relativization in various Indo-Aryan (Cardona 1965, Berman 1972, Srivastav 1991, Bhatt 2003, etc) and Slavic languages (Izvorski 1996, Citko 2006, this volume). It was also argued that CORs also have left their footprints on Old English (Curme 1912) and aboriginal languages such as Walpiri (Hale 1976, Keenan 1985). It should be made clear to the readers that in another stream of thought (e.g. in functional grammar), CORs can also refer to a dependent relation between two adjacent clauses, for instance in conditionals, causeand-reason adverbial clauses and coordinations, etc. (Quirk et al. 1985). In this paper, we restrict the definition of CORs to left-adjoining relative constructions. Among a number of properties of CORs, in this paper we focus on the ‘matching requirement’ (henceforth MR) as a formal relation between the two clauses in CORs.2 We describe the general typology of MR and propose a formalization that we argue to

 Tommi Tsz-Cheung Leung

be well-motivated, both conceptually and empirically. The outline of the paper is as follows: Section 2 presents the basic properties of CORs. Section 3 discusses Bhatt’s (2003) approach to them. Section 4 discusses the matching relation between the relative morpheme and the demonstrative morpheme. Section 5 proposes a novel formalism for the expression of this matching relation. Section 6 spells out the conditions under which the MR can be violated. Section 7 is the conclusion of the paper.

2. Properties of correlatives As mentioned above, CORs are typical of most Indo-Aryan languages. While the details of each correlative language may vary, CORs can be defined by a list of common properties. These include (i) the leftward adjunction of the correlative clause (Section 2.1), (ii) the demonstrative requirement (Section 2.2), and (iii) the availability of multiple correlatives (Section 2.3). The combinational possibilities between the relative morphemes and the demonstrative correlates in both single and multiple correlatives can be subsumed under the matching requirement (Section 2.4).

Leftward adjunction of the correlative clause

2.1

CORs can be schematized by the following representation:

(1) [IP [Cor-CP …Rel(-XPi)…] [IP …Dem(-XPi)…]]

In this representation, the relative clause is left-adjoined to the main clause at the surface level. For the sake of exposition, we call this left-adjoining relative clause a correlative clause (Cor-CP). The following examples present a brief list of correlative constructions from various languages, starting from the Indo-Aryan family. The correlative clause is indicated by brackets:3, 4 Hindi (2)

[jo CD sale-par hai], Aamir us CD-ko khari:d-ege. rel CD sale-on be.pres Aamir dem CD-acc buy-fut.m.sg ‘Aamir will buy the CD that is on sale.’ (lit. ‘Which CD is on sale, Aamir will buy that CD.’)

Hittite (3) [Kuis-an appa-ma uwatezzi n-za ], apas-at dai. rel.nom-him back-prt bring.3sg prt-prt dem.nom-him take.3sg ‘The one who brings him back takes him for himself.’



On the matching requirement in correlatives 

Bangla (4) [Je mee-Ti okhane daRie ache ], Se lOmba. rel girl-3sg there stand.conj be.pres.3sg 3sg tall ‘The girl who is standing over there is tall.’ Sanskrit (diacritics omitted) (5) [ye ‘ngara asans], te ‘ngiraso ‘bhavan. rel.who coals were these Angiras became ‘Those who were coals became Angiras.’ Gujarathi (6) [Je dhobi maarii saathe aavyo], te DaakTarno bhaaii che. rel washerman my with came that doctors brother is ‘The washerman who came with me is the doctor’s brother.’ Malayalam (7) [Eet-oruvan drooham ceyyunnu], avan paapi aakunnu. rel-one.m evil do.pres he sinner become.pres ‘He who does evil becomes a sinner.’ Nepali (8) [Jun keTilai Ramle dekhyo], ma tyo keTilai cinchu. rel girl.dat Ram.erg see.pst 1sg.nom dem girl.dat know-1sg.prt ‘I know the girl who Ram saw.’ Maithili (9) [Je bidyarthi kailh ae-l rәh-әith ], se biman pәir ge-l-ah. rel student yesterday come-perf aux-pst(3) 3 sick lie go ‘The student who came yesterday got sick.’ Late Middle Indo-Aryan (10)

[jam dinnu saccu taem ti-vara ], tam mai mi rel give.part truth father.instr three-times that I.instr prt dimmu tumha saya vara. give.part you.gen hundred times ‘I gave you hundred times of the truth that my father gave you three times.’ (lit. ‘Which truth (my) father gave (you) three times, that (truth) I gave you hundred times.’)

 Tommi Tsz-Cheung Leung

Tamil (11)

[neerru entap paiyan va-nt-aan-oo], antap paiyan-ai yesterday rel boy come-pst-3s-comp that boy-acc naan inru paar-tt-een. I today see-pst-1s ‘Today I saw the boy who came yesterday.’

Telugu (diacritics omitted) (12) [eppudu akali ayte], appudu tinali. when hungry feels then eat ‘When one feels hungry, then one must eat.’ CORs are also attested in other languages. While some of them (e.g. Russian, Hungarian and Polish) can be considered true ‘correlative languages’ in that correlatives are widely attested, other languages (e.g. Dutch, Bambara, Thai, Lhasa Tibetan, Vietnamese, Korean, Latin) do not favor the use of correlatives productively:5 Dutch (13) [Wie jij uitgenodigd hebt], die wil ik niet meer zien. rel.who you invited have that-one want I no longer see ‘I don’t want to see the one you have invited any longer.’ Russian (14) [Kogo ljublju ], togo poceluju. rel.whom love.1sg that.one will.kiss.1sg ‘I will kiss who I love.’ Hungarian (15) [Aki korán jött ], azt ingyen beengedték. rel.who early came that.acc freely pv.admitted.3pl ‘Those who come early were admitted for free.’ Bambara (16) [N ye so min ye ], cε be o dyf. I pst house rel see man prog it build ‘The man is building the house that I saw.’ Polish (17) [Co chcesz], to dostaniesz. rel want.2sg dem get.2sg ‘You will get what you want.’



On the matching requirement in correlatives 

Korean (18)

[Na-lul ch’otaeha-nun saram-un  nuku-tunchi ], I-acc invite-rcl  person-top rel-who-ever ku-nun John-to ch’otaeha-n-ta. he-top John-also invite-pres-decl ‘Whoever invites me also invites John’

Lhasa Tibetan (19) [khyodra-s gya gare njos yod na], nga-s de bsad pa yin. you-erg yak rel.what buy aux if I-erg that kill past aux ‘I killed whatever yak you bought.’ Vietnamese (20) [Ai nâu ], nây ăn. rel.who cook that-person eat ‘Whoever cooks eats.’ Finnish (21) [Joka kaiken lupaa], se jotakin antaa. rel all.acc promise.3sg it which.part give.3sg ‘S/he who promises everything will provide something.’ CORs were also found in various archaic languages:6 Latin (22) … [terra pulli qui nascentur], eos in terram deprimito. Earth scions rel will.spring these in Earth press.imperf ‘…Press into the Earth scions that will spring from (the ground)…’ Medieval Russian (23)

2.2

I kotoruju zvezdu potrebno bylo nam videt’ and which.acc star necessary was us see.inf tu zvezdu zaslonilo tucheju that star covered cloud.by ‘The star we needed to see was covered by cloud.’

The demonstrative requirement

CORs are also signaled by the obligatory presence of a relative morpheme (Rel) in the Cor-CP, and a demonstrative morpheme (Dem) in the main clause, as indicated in the schema (1). The requirement for the presence of a Dem is salient in CORs. The use of

 Tommi Tsz-Cheung Leung

an indefinite expression is strictly banned, as shown in the Hindi (24) (Srivastav 1991) and the Hungarian (25) (Lipták 2004): (24)

[jo CD sale-par hai], Aamir {us CD-ko /*CD/*koi CD/ rel CD sale-on is Aamir {dem CD-acc/*CD/*some CD/ *har CD} khari:d-ege. *every CD} buy-fut.m.sg ‘Aamir will buy the CD that is on sale.’

(25) [Akivel Mari moziba jár], {az /az a fiú/* egy fiú} illedelmes. rel.who.with Mari cinema.to goes {that/that the boy/a boy} polite ‘The boy Mari goes to the cinema with is polite.’ Other types of relative constructions such as embedded relative clauses and extraposed relative clauses do not observe the demonstrative requirement, indicating that CORs are structurally independent of these types of relative constructions. For illustration, we focus mainly on Hindi. (26) shows an embedded relative clause, and (27) an extraposed one. (26) a. {vo larkii /har larkii /larkii} [CP jo kharii hai] lambii hai. {dem girl /every girl /girl} rel standing is tall is ‘Every girl/The girl/A girl who is standing is tall.’ b. {vo larkii / har larkii / larkii} lambii hai {dem girl /every girl /girl} tall is [CP jo kharii hai]. rel standing is ‘Every girl/the girl who is standing is tall.’ A caveat is in order here. The obligatory presence of Rel and a definite expression is neither necessary nor sufficient as the defining property of CORs. Other relative constructions also readily allow the use of a Rel in the relative clause and a definite phrase in the main clause. Consider the English relative construction in (27):7 (27) John recognized the boy [Rel-Cl who/whom/which I met yesterday]. For this reason it was sometimes argued that CORs can be defined by an alternative condition that bans indefinite NPs in the main clause (e.g. Subbarao 1984).

2.3

Multiple correlatives

Perhaps the most notable property of CORs is that they allow the use of multiple relative morphemes. We refer to cases where this happens as multiple correlatives. Compare the schema of a single correlative in (28a) and that of a multiple correlative in (28b): (28) a. [IP [Cor-CP…Rel(-XPi)…] [IP…Dem(-XPi)…]] b. [IP [Cor-CP…Rel(-XPi)…Rel(-YPj)…][IP…Dem(-XPi)…Dem(-YPj)…]]



On the matching requirement in correlatives 

The schemas of single and multiple correlatives are largely parallel to each other. In single correlatives a Dem is anaphoric to the Rel in the Cor-CP. In multiple correlatives, two Dems (i.e. Dem-XP and Dem-YP) are anaphoric to two Rels (i.e. Rel-XP and Rel-YP). Multiple correlatives are widely attested in various correlative languages, shown in the following list:8, 9 Bulgarian (29) [Komu co Jan dał], temu to Maria zabierze. rel.who.dat rel.what.acc Jan gave dem.dat dem.acc Maria take-back ‘Anything Jan gave to whom, Maria took it back from him.’ Hindi (30)

[jis larkii-ne jis larke-ke-saath khel-aa ], rel.obl girl-erg rel.obl boy-with play-perf us-ne us-ko haraa- yaa. dem-erg dem-acc defeat-perf ‘A girl who played with a boy defeated him.’

Hungarian (31) [Aki amit kér ], az azt elveheti. rel.who rel.what.acc wants dem dem.acc take.pot.3sg ‘Everyone can take what he wants.’ Marathi (32) [jya mula-ne jya muli-la pahila], tya mula-ne tya muli-la pasant kela. rel boy-erg rel girl-acc saw dem boy-erg dem girl-acc like did ‘A boy who saw a girl liked her.’ Polish (33) [Kto co chce ], ten to dostanie. rel.who rel.what wants dem dem gets ‘Everyone gets what he wants.’ Russian (34) [Kto kogo ljubit], tot o tom i govorit. rel.who rel.whom loves he of him and speaks ‘Everybody speaks about the person they love.’

 Tommi Tsz-Cheung Leung

Serbo-Croatian (35) [Kome se kako predstavĭs ], rel.whom refl how present.yourself taj misli da tako treba da te tretira. he thinks that thus should to you treat ‘The way you present yourself, this is how people think they should treat you.’ Note that multiple relative pronouns are specific to CORs. Other relative constructions strictly ban the use of multiple relative pronouns, as can be seen in the Hindi extraposed relative in (36b) and the Hungarian embedded and extraposed relatives in (37b) and (37c): (36) a. [jis laRkii-ne jis laRke-ko dekhaa], rel girl-erg rel boy-acc saw us-ne us-ko passand kiyaa. dem-erg dem-acc liked did ‘A girl that saw a boy liked him.’ Lit. ‘Which girl saw which boy, she liked him.’ b.

*us laRkii-ne us laRke-ko pasand kiyaa dem girl-erg dem boy-acc liked did [jis-ne jis-ko dekhaa].10 rel-erg rel-acc saw Lit. *‘The girl liked the boy, who saw whom.’

(37) a. [Aki amit kér ], az azt elveheti. rel.who rel.what.acc wants that that.acc take.pot.3sg ‘Everyone can take what he/she wants.’ Lit. ‘Who wants what, s/he can take that.’ b. *Az azt [aki amit kér ] elveheti. that that.acc rel.who rel.what.acc wants take.pot.3sg c. *Az azt elveheti [aki amit kér]. that that.acc take.pot.3sg rel.who rel.what.acc wants ‘Everyone can take what he/she wants.’ The number of Rels and Dems in multiple correlatives is not restricted to two. In principle any number of Rel and Dem can be found. Consider (38) from Marathi, where we find three Rels and Dems (Wali 1982): (38)

[jyaa muline jyaa mulaalaa je pustak prezent dila hota], rel girl rel boy rel book present gave had tyaa muline tyla mulaalaa te pustak aadki daakhavla hota. dem girl dem boy dem book before shown had ‘A girl that presented a book to a boy had shown it to him.’ (lit. ‘Which girl presented which book to which boy, she had shown it to him.’)



On the matching requirement in correlatives 

For the sake of exposition, we will focus on multiple correlatives with two Rels and Dems.

2.4

The matching requirement of correlative constructions

The parallels between single and multiple correlatives concerning the number of Rels and Dems can be captured by the matching requirement (MR) (Bhatt 1997, 2003). In this paper, the following two statements concerning MR will be entertained: (39) a. The number of relative morphemes in the correlative clause equals the number of demonstrative morphemes in the main clause. b. The mapping between the set of relative morphemes within the correlative clause and the set of demonstratives within the main clause is bijective, i.e. for every relative morpheme within the correlative clause, there is exactly one demonstrative morpheme within the main clause such that relative morpheme and demonstrative morpheme are coindexical with each other. The two statements in (39) suggest that the notion of ‘matching’ in CORs can have more than one interpretations, though they are not in conflict with each other. (39a) means that the number of Rels and Dems has to be equal. There are n instances of relative pronouns if and only if there are n instances of demonstratives. On the other hand, (39b) involves a more restricted requirement for the presence of Rel and Dem. Accordingly, the set of Rels and the set of Dems should be bijective. In order to satisfy bijective mapping, the mapping function between a Rel within the correlative clause and a ‘corresponding’ Dem within the main clause should be uniform across all RelDem pairs. For instance, in set theory, given two sets of numbers {0, 1, 2} of X and {0, 3, 6} of Y, we can define a bijective mapping function f: X → Y in which f(x)=3x, such that f(0)=0, f(1)=3 and f(2)=6. Note that the meaning of bijective mapping is bidirectional: For all numbers x within X, there exists exactly one number y in Y such that f(x)=y; and for all numbers y within Y, there exists exactly one number x in X such that f(x)=y. By analogy, the similar bijective function can be defined in CORs in the following statement: (40) A Rel that bears an index i maps onto a Dem that bears an index i, a Rel that bears an index j maps onto a Dem that bears an index j, and so on. In usual cases, the violation of the MR is strictly banned, as can be seen in the Hindi examples in (41) and the Hungarian ones in (42) (Dayal 1996, Bhatt 2003, Lipták 2004, 2005): (41) a. [jis larke-ne sports medal jiit-aa], *(us-ne) rel boy-erg sports medal win-pfv dem-erg academic medal-bhii jiit-aa. academic medal-also win-pfv ‘A boy who won the sports medal also won the academic medal.’

 Tommi Tsz-Cheung Leung

b. *[jis larke-ne jis larki-ko dekha], us larki-ko piitaa gayaa. rel boy-erg rel girl-acc saw dem girl-acc beaten was. ‘A girl whom a boy saw was beaten.’ c. *[jo laRkii jis laRke-ke saath khelegii], vo jiit jaayegii. rel girl rel boy-obl with play.f she win.perf.f ‘A girl who plays with a boy will win.’ (42) a. [Akit bemutattál], *(annak) köszöntem.11 rel.what.acc introduced.2sg that.dat greeted.1sg ‘I greeted the person you introduced to me.’ b. [Aki korán jött ], *(annak) ingyen adtam jegyet. rel.who early came that.dat freely gave.1sg ticket ‘Who came early, I have given those a ticket for free.’ The above examples show that the matching requirement applies to both single and multiple correlatives. Example (41a), (42a) and (42b) indicate the significance of the MR in single correlatives, whereas (41b) and (41c) verify the significance of the MR in multiple correlatives. Before we proceed, two things need to be noted. First, the MR stated in (39) is at this point nothing but a descriptive statement. It is not a particular syntactic representation that complies with well-formedness conditions. Second, the MR is not absolutely inviolable. Whether a language have both single or multiple correlatives and whether these relatives comply with the MR depends on a number of factors, including basic word order, the representation of wh-questions, the availability of multiple questions, among others. To the best of my knowledge, we are still far from being able to predict whether any given language can have CORs and whether CORs strictly observe the MR. This, however, does not necessarily undermine the need for formalizing a mechanism that derives the MR for correlative constructions that observe it, and to spell out the conditions in which the MR can be violated. Section 4 will attempt to formalize such a mechanism. Before this, however, Section 3 introduces various syntactic approaches to correlatives.

3. The syntactic representation of correlative constructions In this section we turn to syntactic analyses of correlative constructions that have been proposed in the literature. An influential account by Srivastav (1991) and Dayal (1996) proposed a base-generation approach for Hindi single and multiple correlatives. In this account the correlative clause is base-generated as an adjunct to the main clause. The Dem in the main clause is bound by the adjoined Cor-CP via generalized quantification. An alternative to this account was proposed by Bhatt (2003). According to this proposal, single and



On the matching requirement in correlatives 

multiple correlatives receive different analyses. Single correlatives receive an adjunction-and-movement approach, in which the Cor-CP originates adjoined to Dem-XP. Cor-CP and Dem-XP therefore form an underlying constituent. The leftward adjunction of the Cor-CP to the main clause is argued to be the result of overt A'-movement of the Cor-CP to a left-peripheral position. Multiple correlatives differ from single correlatives in that the Cor-CP in the former is adjoined to the minimal configuration that contains all instances of Dem. The two approaches are represented in (43): (43) The base-generation approach, single correlatives a. [IP [Cor-CP … Rel-XPi…]i [IP … [Dem-XPi] …]] The base-generation approach, multiple correlatives b. [IP [Cor-CP… Rel-XPi… Rel-YPj… ]i, j [IP … Dem-XPi… Dem-YPj…]] (44) The adjunction-and-movement approach, single correlatives a. [IP [Cor-CP … Rel-XPi… ]i [IP … [ti Dem-XPi] …]] The adjunction-and-movement approach, multiple correlatives b. [IP [Cor-CP… Rel-XPi … Rel-YPj…]i, j [IP… Dem-XPi… Dem-YPj…]] According to Bhatt, there are three pieces of evidence that support the adjunctionand-movement approach to CORs in Hindi. First, the Cor-CP and Dem form a constituent and function as a conjunct in coordination. Consider (45) (Bhatt 2003: 504): (45)

Rahul a:jkal [DP [DP [jo kita:b Saira-ne likh-i:] vo1] Rahul nowadays rel book Saria-erg write-perf dem aur [DP [jo cartoon Shyam-ne bana:-ya]2 vo2]] parh raha: hai and rel cartoon Shyam-erg make-perf dem read prog be.pres ‘Nowadays, Rahul is reading the book that Saira wrote and the cartoon that Shyam made’ (lit. ‘Nowdays, Rahul is reading [[which book that Saira wrote] that (book)] and [[which cartoon that Shyam made] that (cartoon)].’

Second, Hindi fragment answers containing a Cor-CP always need to be accompanied by a Dem. This again shows that the two form a constituent (Lipták 2005, data from Rajesh Bhatt p.c.): (46) Question: Who came first? Answer: [jo laRkii khaRii hai]??*(vo) rel girl standing is that ‘The girl who is standing’ Third, overt movement of the Cor-CP to the left-peripheral position observes certain constraints on movement, such as the complex NP island constraint, i.e. it rules out a configuration like that in (47a) instantiated in (47b): (47) a. *[Cor-CP]i [IP…[NP NP [RC …Dem-XPi…]]…]

 Tommi Tsz-Cheung Leung

b. *[jo vaha: rah-ta: hai]i mujh-ko [vo kaha:ni rel there stay-hab be.pres I-dat that story [RC jo Arundhati-ne us-ke-baare-me likh-ii]] pasand hai rel Arundhati-erg dem-about write-perf like be.pres *‘Whoi lives there, I like the story that Arundhati wrote about that boyi’ Another piece of argument for the movement of the correlative clause comes from a Condition C violation, which is observed as a result of the reconstruction of the CorCP to a Dem-adjoined position. An illustrative example is given in (48) (for other examples, see Bhatt 2003): (48) a. *[Cor-CP…Namej…]i [Pronj [ti Dem-XPi ]…] b. *[jo larkii Sita-koj pyaar kar-tii hai]i [us-nek/*j us- koi rel girl Sita-acc love do-hab is dem-erg dem-acc thukraa di-yaa] reject give-perf *‘Shei rejected the girl who loves Sitai.’ Concerning multiple correlatives, Bhatt (2003) points out that the Cor-CP in multiple correlatives underlyingly adjoins to the minimal configuration of the main clause that contains all instances of demonstratives. In (49), the pronoun vo ‘she’ as the subject of the main clause cannot be coindexed with Sita within the Cor-CP. This shows that the Cor-CP actually starts off at a position that is c-commanded by the subject vo, giving rise to a Condition C violation, with the configuration *[Rel-XP Namej Rel-YP]l [Pronj thinks that [tl [Dem-XP…Dem-YP…]]]. (49) *[jis larke-ne Sita-se jis topic ke-baare-me baat ki-i] rel boy-erg Sita-with rel topic about talk did [vo soch-tii hai ki [[vo larkaa us dem think-hab is that dem boy dem.obl topic par paper likh-egaa]. topic on paper write-fut *‘For x, y such that x talked to Sitai about topic y, shei thinks that x will write a paper on topic y.’ To account for the observation that the Cor-CP is adjoined to the minimal configuration of the main clause that contains all instances of Dem, Bhatt (2003: 525) suggests the following locality condition on Merge that applies to two ‘associated’ elements:12 (50) The locality of Merge The structure-building operation of Merge must apply in as local a manner as possible. For the definition of ‘associated’ elements, Bhatt notes that it “is meant to subsume both head-argument relations as well as the relationship that obtains between a modifier and



On the matching requirement in correlatives 

what it modifies. Relative clauses are associated with the noun phrase they modify, the ‘head’ of the relative clause [footnote omitted]. Correlative Clauses are associated with the Dem-XPs they occur with.” (Bhatt 2003: 526, emphasis in origin). Discussions of CORs in other languages brought to light that the principle of ‘the locality of Merge’ may not hold for all correlative languages. Lipták’s (2005) analysis of Hungarian provided counterexamples to the claim that the Cor-CP starts off at the Dem-adjoining position in Hungarian. The first arguments come from fragment answers (e.g. 51), in which Hungarian behaves differently from Hindi (cf. 46). The demonstrative morpheme cannot be overt: (51) Question: Who came first? Answer: [Aki ott áll], (*az). rel.who there stands that ‘The one who stands there’ Second, reconstruction effects of the Cor-CP cannot be observed in Hungarian, e.g. (52). (52) a. [Akit szeret Marii], azt meghívta proi a buliba. rel.who.acc loves Mari that.acc invited the party-to ‘Who(ever) Marii loves, shei invited to the party.’ b. *Azt meghívta proi [akit szeret Marii] that.acc invited rel.who.acc loves Mari *‘Shei invited who(ever) Marii loves to the party.’ Example (52a) shows that there is no reconstruction of the Cor-CP to a position adjoined to azt ‘that’ that would lead to a Condition C violation. On the other hand, (52b) shows that the subject pro c-commands Mari and therefore the two cannot be coindexical with each other due to a Condition C violation. A similar conclusion about Nepali was reached by Anderson (2007). In (53), the subject pronoun un-le ‘he’ can be coindexed with Ram within the Cor-CP, showing that the Cor-CP never starts off at a lower position within the main clause: (53)

[jun siksak-lai Rāmi man parauncha], rel teacher-dat Ram likes un-lei un-lai buddhiman thāncha. 3sg-erg 3sg-dat intelligent considers ‘Rami considers the teacher hei lies to be intelligent.’ (Lit. ‘The teacher who Rami likes, hei considers him to be intelligent.’)

 Tommi Tsz-Cheung Leung

Also, the left-peripheral position of the Cor-CP does not seem to be the result of overt movement from a lower position, as constructions like (54) are not subject to island constraints: (54)

[jun manche tyo gharma basthyo]i, Rāmle kathā rel man dem house-loc lived Ram-erg story sunāyo ki uskoi pharak paricaya thiyo. told that dem-gen different life was ‘Ram told a story that the man who lived in that house had a secret life.’ (Lit. ‘[The man who lived in that house]i, Ram told a story that hei had a secret life.’)

These facts led Lipták and Anderson to conclude that the Cor-CP is left-dislocated in sentence-initial position and is interpreted as a topic.13, 14 Left-dislocation is generally argued to be base-generation in a left-peripheral position and therefore immune to constraints on movement. The following examples from Hungarian (É. Kiss 2002: 258) evidence this: (55) a. János, már dél felé járt az idő, amikor felébredt. János already noon towards went the time when awoke ‘János, it was already about noon when (he) woke up.’ b. Jánost, nincs az az ember aki fel tudná bosszantani. János.acc is.not that the man who vm could angry.make.inf ‘John, there is no man who could make (him) angry.’ In (55a), the left-dislocated János does not violate constraints on movement such as the Condition on Extraction Domains (Huang 1982), and in the same vein, the leftdislocated János in (55b) does not violate the Complex NP Constraint. These facts support the absence of overt movement of the dislocated element. As a result, the parallel shown between left-dislocation and the placement of Cor-CP suggests that the latter is also a type of left-dislocation.

4. The matching relations in correlative constructions While more research is needed to arrive at a parametric approach to the full variation in reconstruction or island constraints in CORs in various languages, we now turn our attention to the matching requirement in correlatives. The abovementioned proposals do not directly hinge on the MR as a major defining property of CORs.15 While there might be more than one adjunction levels available to Cor-CPs in languages (e.g. at the level of Dem as in Hindi, or at the level of CP as in Hungarian and Nepali), correlatives display a matching relation between the Rel and the Dem. It should be pointed out that



On the matching requirement in correlatives 

the matching relation is a referential relation and is also syntactically and morphologically instantiated. Let us start with illustrating the matching of syntactic features between Rel and Dem, followed by the matching of morphological features. Consider the Bulgarian example (56) and the Dutch example (57). We see in these that the syntactic position of the Rel/Dem relative to the Cor-CP/main clause corresponds to each other. Both of them are clause-initial: (56) [Kolkoto pari iska ], toklova misli če (*toklova) rel.how.much money wants that.much thinks that that.much šte i dam (*toklova). will her give.1sg that.much ‘She thinks that I’ll give her as much money as she wants.’ (57) [Wie jij uitgenodigd hebt], die wil ik niet meer zien (*die). rel.who you invited have that want I no longer see that ‘The one you have invited, I don’t want to see him any longer.’ While the left-peripheral position of the Cor-CP can be due to left-dislocation as mentioned before, the clause-initial position of the Rel and Dem is argued to be the result of focus movement in Izvorski (1996). The overt movement of Dem to the clause-initial focused position can be verified by the following Bulgarian example, in which the complex NP constraint is violated in the presence of sluha ‘the rumor’: (58) [Kakto im kazah], taka čuh (*sluha) how them told.1sg that.way heard.1sg rumor če sa postâpili. that are done ‘I heard (the rumor) that they had acted the way that I had told them to.’ The second type of syntactic match concerns the identity of categorial features between Rel and Dem. This can be generalized as the following statements:16 (59) a. In correlatives, if the syntactic category of the Rel phrase is an NP (e.g. what, who, which NP), the syntactic category of the Dem phrase is an NP (e.g. that, that NP). b. In correlatives, if the syntactic category of the Rel phrase is a PP (e.g. where, when, how), the syntactic category of the Dem phrase is a PP (e.g. there, then, in that way). The third property of matching, which we pointed out in (39), concerns interpretation: Rel and Dem agree with respect to the referential features in which Rel and Dem essentially bear the same index.

 Tommi Tsz-Cheung Leung

Identity between Rel and Dem can also be observed from their morphology. In Hungarian, the Rel and Dem essentially bear the same number marking. Consider the following (Lipták 2004): (60) a. [Aki most jött], az bejöhet. rel.who.sg now arrived.3sg that enter.pot.3sg ‘The person(s) who just arrived can enter.’ b. *[Aki most jött], azok bejöhetnek. rel.who.sg now arrived.3sg those enter.pot.3pl ‘The person(s) who just arrived can enter.’ c. [[Aki most jött] és [ aki itt volt]], rel.who.sg now arrived.3sg and rel.who here was az/azok bejöhet/-nek. that/those enter.pot.3sg/-3pl ‘Those who just arrived and those who have been here already can enter.’ Example (60a) is grammatical since both Rel and Dem are singular, whereas (60b) is ungrammatical due to a number mismatch between the Rel and Dem. (60c) shows that when the Cor-CP is formed by coordination, Dem can either be singular or plural. The plural marking stems from the coordinated NPs in the Cor-CP.17 The same plural marking is not grammatical in Hungarian when the coordinated relative clauses occur in a headed context: (61)

Az /*Azok [[ aki most jött] és [aki itt volt]], that /*those rel.who now arrived.3sg and rel.who here was bejöhet/*-nek. enter.pot.3sg/*-3pl ‘Those who just arrived and those who have been here already can enter.’

This provides further evidence to the structural and morphological relation between Rel and Dem in CORs. To sum up this section, we have strong reasons to believe that Rel and Dem match with each other with respect to the list of features stated in (62): (62)

a. b. c. d.

categorial feature referential feature focus feature agreement feature



On the matching requirement in correlatives 

5. The representation of matching relations The observation that Rel and its corresponding Dem agree with each other in CORs brings up the question what the syntactic relation between them is. Given the list of features shared between the two items (i.e. (62)), we postulate that their syntactic relation can be analyzed in terms of the formation of a syntactic constituent. This is schematized in (63), in which the two items share the same feature matrix: (63)

[Rel Dem] [XP] [XP] [+Foc] [+Foc] [index i] [index i]

We call the syntactic constituent formed between Rel and Dem a doubling constituent: (64) The doubling constituent contains two elements that share a formal feature matrix. At first glance, we find little evidence for the existence of such a constituent. For instance, the following Hindi example is strictly ungrammatical: (65) *Aamir [DEM jo us cd-ko] khari:d-ege. Aamir rel dem cd-acc buy-fut.m.sg *‘Aamir will buy the cd that’ While it is true that the doubling constituent [Rel Dem] is not easily attested in natural language, such configuration is conceptually well-founded for at least two reasons. First, the formation of a syntactic constituent usually involves feature sharing or matching of some sort. In spec-head configurations, the specifier usually checks off the uninterpretable feature(s) of the functional head such as case features (e.g. of finite T) or EPP features (e.g. of T or C). In head-complement configurations, the uninterpretable feature (e.g. case features) of the complement is checked off by the case-assigning head (e.g. V). By analogy, the doubling constituent formed by Rel and Dem also involves feature matching and sharing as shown in (62). This provides the first motivation for analyzing the matching relation in terms of the formation of syntactic constituents. Second, and more importantly, the syntactic constituents in the form of [Rel Dem], [Spec Head] and [Head Comp] indicate that the two elements within the constituent are contextually related to each other. The notion of ‘contextual relation’ between elements and its relation with syntactic constituents merits more attention. Starting from Vergnaud (2003), and also in Prinzhorn et al. (2004) and Leung (2007a,b) among others it was proposed that the notion of Merge in Chomsky (1995) can be expressed by the following set of equivalence relations: (66) Merge (X, Y) ≈ (X, [ – Y]) ≈ X ↔ [ – Y] ≈ [X, Y] If X is merged with Y (i.e. Merge (X, Y)), X is placed ‘in the context of Y’ within the derivational space (i.e. (X, [ – Y])). The claim that contextually related elements can

 Tommi Tsz-Cheung Leung

be formalized as members of syntactic constituents is also attested in work on language processing such as Collins (2002) and O’Grady (2005). The essential point is that contextually related elements can be grouped together and be subject to certain constituency tests (e.g. coordination or contraction), whether or not they are ‘typical’ syntactic constituents. Consider for instance the following examples of coordination and contraction: (67) a. [John likes] but [Mary hates] Peter. b. [He’s] (< is) there.

(coordination) (contraction)

In traditional syntactic representation, the subject and the main verb usually do not form a syntactic constituent. However the above coordination and contraction examples show that there is a derivational stage in which the two elements should be grouped together. To be more precise: (68) The set of syntactic objects that are contextually related form a syntactic constituent. We assume that (68) can well apply to the matching elements in correlatives:18 (69) a. (Spec, [ – Head]) b. (Head, [ – Comp]) c. (Rel, [ – Dem]) In recent versions of the minimalist program (e.g. Collins 2002), linguists express contextual relations by means of feature checking. In (69a), the role of the specifier is to check off some formal (uninterpretable) features postulated under the head, whereas in (69b) the complement satisfies some feature requirement of the selecting head (e.g. a subcategorization feature). Note that contextual relations are bi-directional. As a result, one could understand the H-Comp relation alternatively by stating that a head is required by the complement that needs to value its uninterpretable (i.e. case) feature, otherwise the uninterpretable feature remains at LF and leads to a crash. We expect that the same type of feature matching is involved in (69c) (i.e. (63)). Though all three types of configurations are understood in different terms in the minimalist program, they essentially express the same concept of syntactic relation, i.e. the two elements within a syntactic constituent are contextually related to each other. To summarize the discussion so far, one can tentatively state that following statement concerning the matching requirement in correlatives: (70) The matching requirement in correlative constructions stems from the contextual relation between the relative morpheme and the demonstratives, represented by the notation (Rel, [ – Dem]). Now assuming that Rel and Dem can form a syntactic constituent, one ensuing question is what kind of syntactic mechanism derives the right surface order in the correlative construction, especially the left-peripheral position of the Cor-CP in (1). We assume that [Rel Dem], in which Rel is an adjunct to Dem, is generated in the main



On the matching requirement in correlatives 

clause, and this is followed by a number of derivational steps that give rise to the surface representation. The various steps are schematized in (71a-d): (71)

a. b. c. d.

[Dem Reli Demi] → [Cor-CP … Reli …] … [Dem ti Demi] → [Dem [Cor-CP … Reli …]… [Dem ti Demi]] → [IP [Cor-CP … Reli …]j … [IP …I… [Dem tj [ti Demi]]]]

Since Cor-CP and the Dem are not connected, the movement of Rel needs to be sideward in the sense of Nunes (2004). As an adjunct to Dem in (71a), Rel moves sideways to an independent phrase marker that forms the Cor-CP, as in (71b). Such movement could be motivated by the postulation of a Probe (in the sense of Chomsky 2001) represented by a selecting head of the Cor-CP. The sideward movement of Rel is felicitous since its uninterpretable features (e.g. case features) need to be valued by a case-assigning head. In addition, given that Rel already exists in the derivation, it would be accessible for feature matching with the Probe. The Probe and Goal need to satisfy minimality (Rizzi 1990) in the sense that there exists no intervening element that also satisfies the Probe. This corresponds to the assumption that the Cor-CP merges with the main clause.19 As we mentioned before, the level at which the Cor-CP merges with the main clause is subject to parametric variation. Step (71c) shows the case of Hindi where the Cor-CP is adjoined to the Dem (Bhatt 2003), whereas other languages might adjoin the Cor-CP to different levels of the main clause. Finally, the subsequent movement of the Cor-CP to the left-peripheral position results in the surface representation of CORs. Now consider the following Hindi example with the addition of traces that show the sideward movement of Rel. (72b) shows the chain (CH) formed by the sideward movement of jo:20 (72) a. [joi larkii kharii hai], [ti vo] lambii hai. rel girl.f standing.f is dem tall.f is b. CH (Jo) = (*hai Cor-CP, vo) The CH is expressed by a list of occurrences (i.e. positions) (Chomsky 1995) that define the moved item jo. Given the MR, we assume that the Dem vo ‘she’ should be one occurrence of the Rel jo in that these two bear a contextual relation with each other (see the statement in (70)). Note that while hai and vo are two occurrences of the CH formed by jo, they do not belong to the same clausal domain. Therefore it is expected that the CH formed by CORs should be quite different from that formed in other movement cases. In CORs, one occurrence, namely the Dem vo ‘she’, forms its own CH with another list of occurrence(s) given its phonological realization (see also Leung 2007b).21 In the following expression (73), hai ‘is’ within the main clause (not the correlative clause) is a ‘strong’ occurrence (notated by *, see Boeckx 2003) that forces the phonological realization of vo in the main clause: (73) CH (vo) = (*hai IP)

 Tommi Tsz-Cheung Leung

We call a lexical item (e.g. vo ‘she’ in this case) that functions as an occurrence of one CH (i.e. CH (jo)) on one hand, and constitutes an independent CH (i.e. CH (vo)) on the other hand, a pivotal occurrence. The discussion of the theoretical significance of pivotal occurrence in chain formation is beyond the scope of this paper. For more details, see Leung (2007a). It should be pointed out that there were attempts formalizing coindexation by means of the formation of constituents, thus the treatment of the MR in terms of doubling constituents is not particularly novel. Of high relevance is the discussion of floating quantifiers (Sportiche 1988), antecedent-pronoun relation (e.g. Kayne 2002; Zwart 2002), resumptions (Boeckx 2003), and clitic doubling (Uriagereka 1995). Technical details aside, the surface representations of the abovementioned constructions can all stem from a doubling constituent that is followed by the extraction of its complement (e.g. floating quantifiers, resumptions) or its adjunct (e.g. antecedent-pronoun relation, clitic doubling): Floating quantifiers (74) The studentsi have [all ti] left. Resumption (75) Jani sanble [[ li ti]j te [tj renmen Mari]]. Jan seems he pst love Mari ‘Jani seems hei loved Mari.’

[Haitian]

Antecedent-pronoun relation (76) Johni thinks that [ti hei] is intelligent. Clitic doubling (77) Celaj esti [tj il] ti vrai? that is it true ‘Is that true?’

[French]

If it is tenable to maintain the proposal that the MR of CORs stems from the contextual relation between Rel and Dem that constitutes a doubling constituent, we could pursue an alternative pathway through which single and multiple correlatives can be unified. Given the coexistence between single and multiple CORs across correlative languages, we have reasons to believe that both types of CORs stem from the same derivational mechanism. In our model, multiple correlatives are nothing but the multiple instantiation of the MR. The consequence is that the derivation contains more



On the matching requirement in correlatives 

than one doubling constituent formed by Rel and Dem. Consider the following example given by Dayal (1996): (78)

jis DaakTar-nei jis mariiz-koj dekhaa, us-nei us-koj paisa diyaa. rel doctor-erg rel patient see-past he-erg he-dat money give-past ‘The doctori who saw the patientj paid himj. (lit. Which doctori saw which patientj, hei paid himj)

The above surface representation, according to the present proposal, should stem from the following representation that contains two doubling constituents: (79) [IP [Dem-XP jis DaakTar-nei us-nei] [Dem-XP jis mariiz-koj] rel doctor-erg he-erg rel patient-dat us-koj paisa diyaa.] he-dat money give.past What is more intricate in the case of multiple correlatives is that the two Rels jis DaakTar-ne ‘which doctor’ and jis mariiz-ko ‘which patient’ as the adjuncts to the two Dems move sideways to the Cor-CP independently, depending on the placement of the Probe within the Cor-CP. We suggest that our account of MR is superior to Bhatt’s treatment of correlatives. To recapitulate, in Bhatt’s proposal the multiple Cor-CP merges locally with the IP that contains all instances of Dem, repeated as follows: (80) [IP [Cor-CP… Rel-XPi …Rel-YPj… ]i, j [IP…Dem-XPi…Dem-YPj…]] While we agree that (80) exists as a stage in the derivation, such a proposal fails to account for the MR that generalizes across single and multiple correlatives. Moreover it does not provide a way to depict the coindexical relation between Rels and Dems. An alternative proposal would be to say that a multiple Cor-CP merges with all Dems at the underlying representation, followed by the ATB-extraction of the Cor-CP to the left-peripheral position: (81)

[IP [Dem-XP [Cor-CP … Rel-XPi … Rel-YPj…]i, j Dem-XPi]... [Dem-XP [Cor-CP… Rel-XPj…Rel-YPj…]i, j Dem-YPj...]] → [Cor-CP … Rel-XPi … Rel-YPj…]i, j [IP [Dem-XP ti, j Dem-XPi]... [Dem-XP ti, j Dem-YPj...]]

This proposal is problematic for reasons that were outlined in Bhatt (2003: 497–498). First, the underlying representation in which the multiple Cor-CP adjoins to a Dem in not attested at the surface level. Second, ATB-extraction out of a non-coordinated structure is not attested in Hindi. We contend that generating multiple CORs by means of multiple instantiations of the doubling constituent formed by Rel and Dem is more tenable than the above proposals in that it can generalize across single and multiple correlatives at the same time and provide a formal representation for the MR. In addition, our proposal also

 Tommi Tsz-Cheung Leung

reveals a formal relation between Rel and Dem that shares conceptual grounds with current syntactic theory. As Dayal noted about example (78), there exists a further asymmetry between the Dems within the main clause: the subject of the main clause (i.e. us-ne ‘he-erg’) needs to be coindexical with the subject Rel of the Cor-CP (i.e. jis DaakTar-ne ‘Rel doctor’). The following sentence in which the referential indices of the two Dems are switched is semantically weird: (82) #jis DaakTar-nei jis mariiz-koj dekhaa, us-nej us-koi paisa diyaa. rel doctor-erg rel patient see.past he-erg he-dat money give.past ‘The patientj paid the doctori who saw himj.’ (lit. Which doctori saw which patientj, hej paid himi) While Dayal expressed the above asymmetry in terms of a functional dependency, i.e. assuming that the functional relation between the two Rels is ‘copied’ into the main clause, we suggest that this could be equivalently accounted for by some condition on movement, for instance superiority (Kuno and Robinson 1972, Aoun and Li 2003: 11) that is stated in (83), with the contrast shown in (84):22 (83) A wh-word cannot be preposed, crossing over another wh. (84) a. I wonder who bought what. b. *I wonder what who bought. Example (84b) is ungrammatical because the object wh-word what is preposed and crosses over the subject wh-word who. In the same vein, we could assume that superiority also applies to examples (78) and (82), i.e. the subject (or object) Rel needs to move into the subject (or object) position within the Cor-CP, otherwise (83) would be violated and the sentence would become ungrammatical. It should be pointed out, however, that some languages might not observe such a ‘parallelism’ between the ordering of Rels and Dems. Polish (Citko, this volume) is a good example of the latter type:23 (85) a.

Komu co Jan dał, to who.dat what.acc Jan gave dem.acc temu Maria zabierze. dem.dat Maria take-back ‘Maria gave back to everyone what Jan gave to them.’

b. Co kto chciał, ten to dostanie. what.acc who.nom wanted dem.nom dem.acc gets ‘Everyone gets what he wants.’ The linear order ‘Dat>Acc’ within the Cor-CP is not parallel to the ‘Acc>Dat’ order within the main clause in (85a), and likewise, the linear order between the nominative and accusative case marker is not parallel in the two clauses in (85b), either. While further research is necessary in order to establish the variation in this domain, it is



On the matching requirement in correlatives 

possible that the behavior observed in (85) is linked to the fact that Polish allows violation of superiority in general, as Citko (this volume) points it out: (86) a. Co kto chciał? what who wanted ‘Who wants what?’ b. Co kto chciał, ten to dostał. what who wanted dem dem got ‘Everyone got what they wanted.’ More research should be done with respect to superiority and its impact on CORs, also with respect to D(iscourse)-linked ‘which’ phrases which were argued to allow for superiority violations, too (Hornstein 1995, Pesetsky 2000). We leave this topic for further study.

6. The violations of the matching requirement In previous sections, we have claimed that the matching requirement is an important property of CORs. While we took MR to be universal across all correlative languages, there are examples showing that the MR is not inviolable. In this section, we list a few examples and show what morphosyntactic and semantic conditions allow for the MR to be violated. Note that these do not count as counterexamples to the MR. Instead they just strengthen the contention that the MR as observed in CORs should be syntactically represented, subject to subsequent conditions. Probably the most salient violation of the MR is that the Dem can be optionally deleted in particular cases. In Hindi, the Dem is optionally spelled out if its morphological case is the same as the morphological case of the Rel, and their shared case is phonetically empty (Bhatt 2003: 531). In (87), both the Rel (as the subject of the CorCP) and the Dem (as the subject of the main clause) have a nominative case that is phonetically empty. The Dem can be optionally missing: (87) [jo lar.ki: khar.i: hai], (vo) lambii hai. rel girl standing.f is dem tall.f is ‘The girl who is standing is tall.’ This also happens in Hungarian which allows subject and object pro-drop. In (88), the accusative Dem can be dropped, too: (88) [Aki korán jön], (azt) ingyen beengedik. rel-who early comes dem-acc freely pv.admit.3pl ‘Those who come early, will be let in for free.’

 Tommi Tsz-Cheung Leung

Contrary to these examples, the Dem cannot be erased if the morphological case it bears is ergative in Hindi (89) or non-droppable dative in Hungarian (90): (89)

[jis lar.ke-ne sports medal jiit-aa], *(us-ne) rel boy-erg sports medal win-pfv dem-erg academic medal-bhii jiit-aa. academic medal-also win-pfv ‘A boy who won the sports medal also won the academic medal.’ (Lit. ‘Which boy won the sports medal, he also won the academic medal.’)

(90)

[Akit bemutattál ], *(annak) köszöntem. rel.what.acc introduced.2sg dem-dat greeted.1sg ‘I greeted the person you introduced to me.’ (lit. ‘Who you introduced, I greeted that.’)

In the case of multiple correlatives, another set of morphosyntactic conditions apply for optional deletion of Dem. In Hindi, all instances of Dem can be optionally deleted if they bear the same morphological cases with their corresponding Rel in the Cor-CP (Bhatt 2003: 534). Note that multiple correlatives are not entirely identical to single correlatives in that the morphological case that the Dem bears in the former does not need to be phonetically empty for optional erasure. Consider the following example (Bhatt 2003: 534): (91)

[jis-ne jo chahaa ], (us-ne vo) kiyaa. rel-erg rel want-perf dem-erg dem do.perf ‘People did what they wanted.’ lit. ‘Who wanted what, they did that.’

In the second set of cases, the MR can be violated when one Rel does not denote a specific referent. For instance, free-choice wh-words such as whatever in English do not necessarily pick up a specific entity, and the use of Dem in the main clause would be optional. This also explains why wh-words suffixed by –ever can be productively used in free adjunct free relative constructions (e.g. Izvorski 1996) without a Dem in the main clause, since this relative construction does not denote a restrictive modifier to any entity.24 On the other hand, a bare wh-word without the suffixation of -ever cannot be used in free adjunct free relative constructions. This is true at least in English (92) and Polish (93): (92) [Whatever/*What John cooks], he will win the cooking contest. (93) [Cokolwiek /*Co ty zrobisz], Jan nie dostanie pracy. whatever /*what you do Jan not get job ‘Whatever/*What you do, Jan will not get the job.’

(Polish)



On the matching requirement in correlatives 

Similarly to these cases, the MR can also be violated in Hindi, provided that one (or more) Rel has a free-choice reading:25 (94) [jo laRkiii jis-i laRke-ke saath khelegii], voi jiit jaayegii. rel girl rel-ever boy with play.f she win.perf.f ‘lit. Which girl played with whichever boy, she will win.’ Therefore it seems that the MR can be optionally violated provided that some morphosyntactic and semantic conditions are met. This is not only true for CORs, but also for other related constructions as well. For instance, it was generally argued that in ATBextraction the extracted item from the two conjuncts should belong to the same morphosyntactic class. The following shows a case of Polish ATB-extraction (Citko 2000): (95) a. Kogo Jan lubi tacc a Maria podziwia tacc? who.acc Jan likes and Maria admires ‘Who does Jan like and Maria admire?’

(Polish)

b. *Kogo/ komu Jan lubi tacc a Maria ufa tdat? who.acc/ who.dat Jan likes and Maria trusts ‘Who does Jan like and Maria trust?’ Sentence (95a) is grammatical due to the fact that the two object gaps bear the same structural cases. (95b) on the other hand is ungrammatical since there is a case mismatch between the two object gaps. Interestingly, in cases where the wh-word exhibits case syncretism in which the dative and accusative cases are phonetically identical, ATB-extraction from the two morphologically distinctive gaps is grammatical (Citko 2000: 97): (96) Kogoacc/gen Jan lubi tacc a Jerzy nienawidzi tgen? who Jan likes and Jerzy hates ‘Who does Jan like and Jerzy hate?’ As a result, the satisfaction of some morphosyntactic conditions can rescue a sentence that is otherwise considered as ungrammatical. The relation between the use of freechoice wh-words and the violation of the MR is more straightforward, i.e. the MR stems from the obligatory use of a Dem that semantically denotes a specific referent. In the case of free-choice wh-words, no specific entities are modified by the Cor-CP, and the use of a Dem would be infelicitous. Violations of the MR presented in this section do not necessarily undermine the validity of the MR. On the contrary, they rather strengthen the claim that the MR as a formal relation between the Rel and the Dem should be represented by a syntactic configuration. It is the mapping between syntax and semantics on one hand, and the mapping between syntax and morphology on the other hand, that accounts for the full-fledged typology of the MR.

 Tommi Tsz-Cheung Leung

7. Conclusions In this paper, we have focused on the matching requirement (MR) observed in correlative constructions (CORs). We have summarized previous analyses of CORs, including primarily Bhatt’s (2003) attempt to unify single and multiple correlatives. We have argued that while Bhatt’s analysis nicely accounts for the various facts of correlative constructions in Hindi, it does not necessarily extend to other languages. We pointed out that in other correlative languages (e.g. Hungarian and Nepali) the correlative clause (Cor-CP) does not originate adjoined to Dem. Instead the Cor-CP can be adjoined at different levels of syntax, for instance it can be base-generated adjoined to CP. Theory-internally, the Dem-adjunction proposal is unsatisfactory in that it does not attempt to provide a formal link between the relative morpheme within the CorCP and the demonstrative morpheme within the main clause. The account we put forward in this paper suggests that the formal link between Rel and Dem can be properly described by postulating a doubling constituent [Rel Dem] in which Rel functions as an adjunct to Dem, generated in the main clause. Rel undergoes sideward movement that is motivated by the needs of a selecting head within the correlative clause. We assume that sideward movement can establish a proper syntactic relation between two contextually related elements (i.e. Rel and Dem) across two clauses. The relation between Rel and Dem is analogous to that between a specifier and a head, or between a head and a complement, the latter instances also involving feature checking or matching of some sort. In the doubling constituent formed by Rel and Dem, Rel and Dem agree with each other with respect to categorial features, referential features, focus features and agreement features. While [Rel-Dem], [Spec-Head] and [Head-Comp] involve different feature checking mechanisms, they can all be subsumed under the notion of a contextual relation between elements that define the meaning of syntactic constituents. The interaction between elements and contexts defines syntactic derivation and other formal representations such as chains and movement.

Acknowledgement I would like to thank Elena Guerzoni, James Higginbotham, Roumyana Pancheva, Jean-Roger Vergnaud, and two reviewers for useful comments and corrections on the previous draft of this paper. All errors are mine.

References Anderson, Corinna. 2007. A non-constituent analysis of Nepali correlative constructions. Paper presented in LSA 2007, Anaheim.



On the matching requirement in correlatives  Andrews, Avery. 1985. Studies in the Syntax of Relative and Comparative Clauses. New York/ London: Garland. Aoun, Joseph & Yen-Hui Audrey Li. 2003. Essays on the Representational and Derivational Nature of Grammar: the Diversity of wh-constructions. Cambridge, Mass: MIT Press. Asher, Ronald E. & T. C. Kumari. 1997. Malayalam. Descriptive Grammars. London: Routledge. Bagchi, Tista. 1994. Bangla correlative pronouns, relative clause order and D-linking. In Miriam Butt, Tracy Holloway King & Gillian Ramchand (eds), Theoretical perspectives on word order in South Asian languages, 13–30. Stanford, CA: CSLI. Berman, Howard. 1972. Relative clauses in Hittite. In Paul M. Peranteau, Judith N. Levi & Gloria C. Phares (eds), The Chicago which hunt: Papers from the relative clause festival, 1–8. Chicago: Chicago Linguistics Society. Bhatt, Rajesh. 1997. Matching effects and the syntax-morphology interface: evidence from Hindi correlatives. In Benjamin Bruening (ed.), Proceedings of SCIL 8, MIT Working Papers in Linguistics 31, 53–68. Cambridge, MA: MIT Press. Bhatt, Rajesh. 2003. Locality in correlatives. Natural Language and Linguistic Theory 21. 485–541. Bianchi, Valentina. 2000. Some issues in the syntax of relative determiners. In Artemis Alexiadou, Paul Law, André Meinunger & Chris Wilder (eds), The Syntax of relative clauses. 53– 81. Amsterdam: John Benjamins. Boeckx, Cedric. 2003. Islands and Chains. Amsterdam: John Benjamins. Bresnan, Joan & Jane Grimshaw. 1978. The syntax of free relatives in English. Linguistic Inquiry 9 (3). 331–391. Bubeník, Vit. 1998. A Historical Syntax of Late Middle Indo-Aryan. Amsterdam/Philadelphia: John Benjamins. Cable, Seth. 2005. A reply to Bhatt (2003): correlatives in Tibetan as evidence for the parameterization of local merge. Ms. MIT. Cable, Seth. This volume. The syntax of the Tibetan correlative. Cardona, George. 1965. A Gujarati Reference Grammar. Philadelphia: The University of Pennsylvania Press. Chomsky, Noam. 1955. The Logical Structure of Linguistic Theory. Ms, Cambridge, MA: Harvard University. Chomsky, Noam. 1977. On wh-movement. In Peter W. Culicover, Thomas Wasow & Adrian Akmajian (eds.), Formal Syntax, 71–132. New York: Academic Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: a Life in Language, 1–52. Cambridge, MA: MIT Press. Citko, Barbara. 2000. Parallel merge and the syntax of free relatives. Doctoral dissertation, SUNY at Stony Brook. Citko, Barbara. 2006. On the interaction of left branch extraction and across-the-board whmovement. Syntax 9 (3). 225–247. Citko, Barbara. This volume. What don’t wh-questions, free relatives, and correlatives have in common? Collins, Chris. 2002. Eliminating labels. In Sam D. Epstein & T. Daniel Seely (eds.), Derivation and Explanation in the Minimalist Program, 42–61. Oxford: Blackwell. Culicover, Peter & Ray S. Jackendoff. 1999. The view from the periphery: the English comparative correlatives. Linguistic Inquiry 30 (4). 543–571. Curme, George. O. 1912. A history of the English relative construction. The Journal of English and Germanic Philology 11.

 Tommi Tsz-Cheung Leung Dayal, Veneeta. 1996. Locality in Wh Quantification. Dordrecht: Kluwer Academics. den Dikken, Marcel. 2005. Comparative correlatives comparatively. Linguistic Inquiry 36 (4). 497–532. den Dikken, Marcel. This volume. Comparative correlatives and successive cyclicity. Downing, Bruce. 1974. Correlative relative clauses in universal grammar. In Minnesota Working Papers in Linguistics and Philosophy, no 62. Dordrecht: Kluwer. Dwivedi, Veena Dhar. 1994. Topicalization in Hindi and the correlative construction. In Miriam Butt, Tracy Holloway H. King & Gilian Ramchand (eds), Theoretical perspectives on word order in South Asian languages. 91–117. CSLI Publications. É. Kiss, Katalin. 1987. Configurationality in Hungarian. Budapest: Akadémiai Kiadó. É. Kiss, Katalin. 2002. The Syntax of Hungarian. Cambridge: Cambridge University Press. Fillmore, Charles, Paul Kay & Mary C. O’Conner. 1988. Regularity and idiomaticity in grammatical constructions: The case of let alone. Language 64 (3). 501–538. Givón, Talmy. 2001. Syntax: An Introduction. Amsterdam, Philadelphia: John Benjamins. Goldberg, Adele. E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago, London: University of Chicago Press. Groos, Anneke & Henk van Riemsdijk. 1981. The matching effects in free relatives: a parameter of core grammar. In Adriana Belletti, Luciana Brandi & Luidi Rizzi (eds.), Theory of Markedness in Generative Grammar. 171–216. Pisa: Scuola Normal Superiore. Hale, Kenneth L. 1976. The adjoined relative clause in Australian. In Rovert M. W. Dixon (ed.), Grammatical Categories in Australian Languages. 78–105. Australian Institute of Aboriginal Studies, Canberra. Hornstein, Norbert. 1995. Logical Form: From GB to Minimalism. Cambridge, MA: Blackwell. Huang, James. C.-T. 1982. Logical relations in Chinese and the theory of grammar. Doctoral dissertation, MIT. Izvorski, Roumyana. 1996. The syntax and semantics of correlative proforms. In Kiyomi Kusumoto (ed.), Proceedings of NELS 26. 133–147. Amherst, MA: GLSA. Jaggar, Philip J. 2001. Hausa. Amsterdam, Philadelphia: John Benjamins. Kayne, Richard. S. 2002. Pronouns and their antecedents. In Sam D. Epstein & T. Daniel Seely (eds.), Derivations and Explanation in the Minimalist Program. 133–183. Oxford: Blackwell. Keenan, Edward. 1985. Relative clauses. In Timothy Shopen (ed.), Language Typology and Syntactic Description, vol 2. 141–170. Cambridge: Cambridge University Press. Krishnamurti, Bharijaju. 1998. Telugu. In Sanford B. Steever (ed.), The Dravidian Languages. 202–240. London, New York: Routledge. Kuno, Susumu & Jane J. Robinson. 1972. Multiple wh-questions. Linguistic Inquiry 3. 463–487. Kuroda, S.-Y. 1968. English relativization and certain related problems. Language 44. 244–266. [Reprinted in David A. Reibel & Sanford A. Schane (eds), 1969. Modern Studies in English: Readings in Transformational Grammar. 264–287. Englewood Cliffs, NJ: Prentice-Hall.] Lehmann, Christian. 1974. Der Relativsatz. Tuebingen: Gunther Narr Verlag. Lehmann, Thomas. 1993. A Grammar of Modern Tamil. India: Pondicherry Institute of Linguistics and Culture. Leung, Tommi T-C. 2003. Comparative correlatives and parallel occurrence of elements. Ph.D screening paper, University of Southern California, Los Angeles. Leung, Tommi T-C. 2005. Typology and universals of comparative correlatives. Association of Linguistic Typology (ALT VI). Padang, Indonesia.



On the matching requirement in correlatives  Leung, Tommi T-C. 2007a. Correlatives and the conditions on chain formation. Proceedings of the International Conference on Linguistics in Korea (ICLK 2007). Seoul National University, Seoul, Korea. Leung, Tommi T-C. 2007b. Syntactic derivation and the theory of matching contextual features. Doctoral dissertation, University of Southern California, Los Angeles. Lipták, Anikó. 2004. On the correlative nature of Hungarian left-peripheral relatives. In Benjamin Shaer, Werner Frey & Claudia Maienborn (eds), Proceedings of the Dislocated Elements Workshop (ZAS Berlin; November 2003), ZAS Papers in Linguistics 35 (1). 287–313. Berlin: ZAS. Lipták, Anikó. 2005. Correlative topicalization. Manuscript. ULCL, Leiden University. Mahajan, Anoop. 2000. Relative asymmetries and Hindi correlatives. In Artemis Alexiadou, André Meinunger, Chris Wilder & Paul Law (eds), The syntax of relative clauses. 201–229. Amsterdam: John Benjamins. Masica, Colin. 1972. Relative clauses in South Asia. In Paul M. Peranteau, Judith N. Levi & Gloria C. Phares (eds), The Chicago which hunt: Papers from the relative clause festival, 198–204. Chicago Linguistics Society. 198–204. McCawley, James. 2004. Remarks on adsentential, adnominal, and extraposed relative clauses in Hindi. In Veneeta Dayal & Anoup Mahajan (eds.), Clause Structure in South Asian Languages. Boston, Dordrecht, London: Kluwer. 291–312. Nunes, Jairo. 2004. Linearization of Chains and Sideward Movement. Cambridge, MA: MIT Press. O’Grady, William. 2005. Syntactic Carpentry. Mahwah, New Jersey: Lawrence Erlbaum Associates. Pesetsky, David. 2000. Phrasal Movement and its Kin. Cambridge, MA: MIT Press. Pittner, Karin. 1995. The case of German relatives. The Linguistic Review 12. 197–232. Prinzhorn, Martin, Jean-Roger Vergnaud & Maria Luisa Zubizarreta. 2004. Some explanatory avatars of conceptual necessity: elements of UG. Ms, University of Southern California. Probert, Philomen. 2006. Clause boundaries in Old Hittite relative sentences. Transactions of the Philological Society 104 (1), 17–83. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass: MIT Press. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, & Jan Svartvik. 1985. A Comprehensive Grammar of the English Language. London, New York: Longman. Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19. 425–449. Srivastav, Veneeta. 1991. The syntax and semantics of correlatives. Natural Language and Linguistic Theory 9. 637–686. Subbarao, Karumuri V. 1984. Complementation in Hindi syntax. Doctoral disseration, University of Illinois, Urbana-Champaign. Sulkala, Helena & Merja Karjalainen. 1992. Finnish. London/New York: Routledge. Uriagereka, Juan. 1995. Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26. 79–123. Vergnaud, Jean-Roger. 2003. On a certain notion of “occurrence”: the source of metrical structure, and of much more. In Stefan Ploch (ed.), Living on the edge. 599–632. Berlin: Mouton de Gruyter. Wali, Kashi. 1982. Marathi correlatives: a conspectus. In P. J. Mistry (ed.), South Asian Review: studies in South Asian languages and linguistics. 78–88. Jacksonville, Florida: South Asian Literary Association. Yadav, Ramawatar. 1996. A Reference Grammar of Maithili. Berlin, New York: Mouton de Gruyter.

 Tommi Tsz-Cheung Leung Zwart, Jan-Wouter. 2002. Issues relation to a derivational theory of binding. In Sam D. Epstein and T. Daniel Seely (eds), Derivation and Explanation in the Minimalist Program, 269–304. Oxford: Blackwell.

Endnotes 1. McClawley used the term ‘adsentential relative clauses’ to refer to the same construction. As he noted, the term ‘correlatives’ or ‘correlative clauses’ are sometimes used to refer to the main clause of that particular construction. In this paper the term ‘correlative clause’ and ‘main clause’ will be used. 2. The term ‘matching requirement’ should be clearly distinguished from a similar yet independent term ‘matching effect’ as discussed originally in Bresnan and Grimshaw (1978) and Groos and van Riemsdijk (1981) in the discussion of free relatives. 3. The references for the language examples used in this paper are listed as follows (in alphabetical order): Bambara (Givón 2001); Bangla (Bagchi 1994); Bulgarian (Izvorski 1996); Dutch (Izvorski 1996); Finnish (Sulkala and Karjalainen 1992); Gujarathi (Masica 1972); Hausa (Jaggar 2001); Hindi (Bhatt 2003); Hittite (Downing 1974); Hungarian (Lipták 2004, 2005); Korean (Hyuna Byun, personal communication); Late Middle Indo-Aryan (Bubeník 1998); Latin (Bianchi 2000); Lhasa Tibetan (Cable 2005, this volume); Maithili (Yadav 1996); Malayalam (Asher and Kumari 1997); Medieval Russian (Keenan 1985); Nepali (Anderson 2007); Russian (Izvorski 1996); Tamil (Lehmann 1993); Telugu (Krishnamurti 1998); Thai (Kingkarn Thepkanjana, personal communication); Sanskrit (Lehmann 1974); Vietnamese (Thuan Tran, personal communication). 4. Abbreviations are used as follows: acc: acusative; aux: auxiliary; conj: conjunction; dat: dative; decl: declarative; dem: demonstrative; erg: ergative; inf: infinitive; f: feminine; fut: future; gen: genitive; ger: gerund; hab: habitual; imperf: imperfective; loc: locative; instr: instrumental; m: masculine; n: neuter; nom: nominal marker; neg: negation; obl: oblique; part: participle; pfv: perfective; pl: plural; pot: potential; pres: present; prog: progressive; prt: particle; pst: past; pv: preverbal element; rcl: relative clause marker; refl: reflexive; rel: relative; sg: singular; top: topic; vm: verb modifier; 1: 1st person; 2: 2nd person; 3: 3rd person. 5. Some languages have non-productive correlative formations. Consider the following English and Hausa examples, respectively:

(i) [Where there is a will], there is a way.

(ii) [Sadda ta bar makaranta], sai ta yi aure. when she left school then she got married ‘When she left school, she got married.’ The discussion of these correlative constructions is beyond the scope of this paper. 6. It has been argued that CORs were the ancestor of other types of relative constructions such as headed relatives. See Pittner (1995) and Bianchi (2000) for, and Probert (2006) against this view. 7. While the relative pronoun can be omitted in (27), e.g. ‘The boy ∅ I met yesterday’, there are cases in which the relative pronoun cannot be omitted, e.g. when the relative pronoun is the subject of the relative clause:



On the matching requirement in correlatives 

(i) The boy who/which/*∅ met me yesterday.

8. Bulgarian (Citko 2006, also this volume); Hindi (Dayal 1996); Hungarian (Lipták 2005); Marathi (Bhatt 2003); Polish (Citko 2006, this volume); Russian (Izvorski 1996); Serbo-Croatian (Izvorski 1996). 9. One reviewer correctly points out that some languages do not allow multiple correlatives even though single correlatives exist. An example is Dutch: (i) *[Wie jij wanneer uitgenodigd hebt], die dan wil ik niet zien. who you when invited have that then want I not see *‘I don’t want to see the person(s) you invited sometimes then.’ (lit. ‘The person(s) you invited sometimes, I don’t want to see him/those then.’)  e unavailability of multiple correlatives might lead one to doubt if Dutch should be regarded Th as a ‘correlative language’ and if the Dutch example in (13) above should be regarded as an example of a correlative construction. As mentioned before, in this paper we allow for languages to differ in the details of the various correlative constructions they exhibit. For instance, the particular requirements observed in Hindi CORs may not be attested in other ‘correlative languages’. The same is true for multiple correlatives. We are therefore not claiming that all correlative constructions are constrained by the same set of conditions. Instead we are interested in the patterns that are attested in a number of languages which are unlikely to be accidental, including the co-existence of single and multiple correlatives in various languages. While Dutch and English do not generally allow both single and multiple correlatives as Hindi does, they still have correlative constructions in some contexts. A notable example is the comparative correlative construction, exemplified in (ii) for English (see Den Dikken 2005): (ii) The more you eat, the fatter you get. While some linguists (e.g. Fillmore et al. 1988, Goldberg 1995, Culicover and Jackendoff 1999) treated (ii) as an idiomatic expression that should receive a separate analysis, others (e.g. Leung 2003, den Dikken 2005) suggested that this construction is analogous to correlative constructions in various interesting ways. The fact that comparative correlatives are represented by similar means across languages provides another piece of evidence that (ii) should not be treated as an ad-hoc construction (Leung 2005). 10. Srivastav (1991) and Mahajan (2000) noticed that (36b) is grammatical for a number of native speakers of Hindi. Srivatav suggests that for these speakers, we might be dealing with leftward extraposition of the main clause, an analysis indirectly supported by the intonational break between the two clauses. 11. Hungarian and Hindi allow violations of the matching requirement under some morphological conditions. See section 6 for details. 12. Cable (2005, this volume) suggests that CORs in Lhasa Tibetan can take part in Dem-adjunction. The combination of Cor-CP and Dem can pass the coordination test: (i) Nga-s [[[khyodra-s I-erg you-erg [[ ∅ bya pro-erg chicken

gyag gare yak what gare rkus what steal

njos yod na]1 de1 ] dang buy aux if that and yod na]2 de2]] bsad pa yin. aux if that kill past aux

‘I killed whatever yak you bought and whatever chicken you stole.’

 ther tests on the other hand show that Lhasa Tibetan also shows evidence for IP-adjunction. O The left-peripheral Cor-CP is not sensitive to islands:

 Tommi Tsz-Cheung Leung (ii) [Khyodra-s mogmog gare njos na]1 nga-s you-erg momo what buy if I-erg de1 bzo mkhan gyi bsad mkhan] de ngozhi gi yod. that make agent gen kill agent the know non.past aux ‘Whatever momos you bought, I know the person who killed the person who made them.’ 13. The same conclusion is also reached in Dwivedi (1994) for Hindi. 14. It should be pointed out that the left-dislocated element is not necessarily situated above the TopicP. The following sentence shows that Péter as a left-dislocated element is sandwiched between tegnap ‘yesterday’ as a normal topic, and AJÁNDÉKOT ‘present’ as a focus (É. Kiss 1987; Lipták 2004): (i) Tegnap Péter (az) ajándékot kapott Maritól. yesterday Péter that present.acc got Mari-from ‘Péter, he got A present from Mari yesterday (while other people might have got something else.).’ 15. This also extends to similar discussions of relative constructions such as ‘John knows the boyi whoi I met yesterday’ in which boy within the main clause and who within the relative clause are coindexical with each other. To the best of my knowledge, such coindexation is argued to be the result of some semantic interpretation rule (Chomsky 1977) which lacks any structural description. For the original discussion of this, see Kuroda (1968). 16. Another case where categorial matching is relevant is the free relative construction that displays the matching effect (Bresnan and Grimshaw 1978; Groos and van Riemsdijk 1981) between the subcategorization frame of the matrix predicate and the syntactic category of the wh-XP. Consider for instance:

(i) I will buy [NP [NP whatever] you want to sell].



(ii) John will be [AP [AP however tall] his father was].



(iii) I’ll word my letter [AdvP [AdvP however] you word yours].



(iv) I’ll put my books [PP [PP wherever] you put yours].

17. This example is somewhat similar to Hindi where a plural pronoun can be used to agree with two Rels at the same time (McCawley 2004: 302; transcription mine): (i)

jo larki jis larke se batcit kar rahi thi, rel girl rel-obl boy-obl with chat do stayed ve ek sath sinema gaye hein. they together movies went.pl Lit. ‘Which girl was talking with which boy, they have gone to the movies together.’

18. Note that the equivalence relation (X, [ – Y]) ≡ (Y, [ – X]) that suggests that contextual relations are symmetric is a fact about the physical world that is independent of syntactic computation. This being said, either the use of contextual relation (X, [ – Y]) or (Y, [ – X]) but not the use of both should be sufficient for computational purposes. For more detailed discussion, see Prinzhorn et al. (2004). 19. Thanks to Hisatsugu Kitahara (personal communication) for the discussion of this idea. 20. The asterisk * is the strong occurrence that determines the phonological realization of the moved item in the sense of Boeckx (2003).



On the matching requirement in correlatives  21. Leung suggested that the syntactic position of lexical items can be formalized by matching the item with its particular context. As a result, each spell-out item should have its own list of occurrences that state its context. In the usual case, each element can be adequately defined by having one occurrence. In the case of movement, the moved item has more than one occurrence. 22. One could reduce superiority to other minimal conditions on movement, along the line of Chomsky (1995: 296). 23. Thanks to Anikó Lipták for pointing this out to me. 24. We assume that left-adjoining free relative constructions are just correlative constructions. 25. I thank Santosh Tandon (personal communication) for her suggestions.

Matching effects in the temporal and locative domains Rajesh Bhatt and Anikó Lipták The correlative strategy is used in a number of languages to achieve restrictive modification. The commonly discussed cases of ordinary correlatives involve abstraction over an individual variable. However, the correlative strategy is also used to abstract over variables that range over times, locations, manners, and degrees. Our goal in this paper is to investigate the extent to which the semantic treatment suggested for correlatives that involve abstraction over individual variables can be extended to the wider class of correlatives. We will limit our discussion to correlatives that abstract over times and locations in Hindi-Urdu and in Hungarian. In these languages, we find that a kind of matching effect that does not apply to ordinary correlatives applies to a subclass of temporal and locative correlatives. Having demonstrated these facts, we will consider explanations for this divergence between ordinary correlatives and temporal/ locative correlatives. Keywords: temporal and spatial correlatives, matching effects, maximalization, individualization

1. Correlatives 1.1

Correlatives over individuals

Correlative constructions are schematized in Keenan (1985) as consisting of a correlative clause and a main clause. The correlative clause contains a relative phrase and the main clause contains a phrase that is anaphoric to the correlative clause, as represented schematically in (1):

(1) [[relative clause... NPrel...] [main clause...NPana...]]

The schema in (1) leaves many aspects of the analysis of correlatives underdetermined, such as the nature of locality constraints that might hold between the correlative clause and the main clause and the correlative clause and the phrase in the main clause that

 Rajesh Bhatt and Anikó Lipták

is anaphoric to it. Another aspect worth discussing is the derivation of correlative constructions, with respect to whether the relative phrase undergoes correlative clauseinternal movement, and whether the anaphoric element undergoes main clause-internal movement. The proper analysis of these aspects of correlativization has to be done on a language-specific basis, as the correlative schema is realized in different ways in different languages. Consider for example (2), which provides examples of correlatives from Hindi-Urdu and Hungarian.1 (2) a. jo sale-par hai Maya us cd-ko khari:d-egi: rel sale-on be.prs.3sg Maya that cd-acc buy-fut.3fsg ‘Maya will buy the cd that is on sale.’ (Lit. What is on sale, Maya will buy that cd.) b. Aki korán jött, azt ingyen beengedték. rel.who early come.pst that.acc freely pv.admitted.3pl ‘Those who come early were admitted for free.’ In Hindi-Urdu, the relative phrase and the anaphoric demonstrative phrase can both stay in-situ. Hungarian obligatorily fronts both the relative phrase and the anaphoric demonstrative phrase. Our focus here will not be on the proper analysis of correlative constructions within any particular language, rather we will concentrate our attention on a particular type of correlative construction: correlatives that abstract over nonindividual variables.

1.2

Correlatives over worlds, degrees and times

In many languages, correlative clauses are also used to realize when-clauses, since-clauses, until-clauses as well as comparatives and conditionals (on the latter, see Arsenijević (this volume) and Rebuschi (this volume)). The surface variation from construction to construction seems limited to the relative phrase and the proform. The rest of the syntax is essentially identical to that of ordinary correlatives. Like ordinary correlatives, there is a correlative clause that contains a relative phrase and a main clause that contains an anaphoric demonstrative phrase. The relative phrase is obligatorily present but the anaphoric demonstrative phrase can under circumstances be absent.2 The following examples give illustration for all types of correlatives mentioned above: (3) a.

conditionals (Marathi, from Pandharipande (1997)) dzar tyāne abhyās kelā tar to pās hoīl. if he.erg studying do.pst then he pass be.fut ‘If he studies, then he will pass.’



Matching effects in the temporal and locative domains 

b.

comparatives Rodman ke jitne tattoo hẼ, Jordan ke-paas Rodman gen how.many tattoo be.prs.3pl Jordan near us-se zyaadaa khitaab hẼ. that-than more title be.prs.3pl ‘Michael Jordan has more scoring titles than Dennis Rodman has tattoos.’ (Lit: How many tattoos Dennis Rodman has, Michael Jordan has more scoring titles than that.)

c.

equatives John bhautiki-me jitnaa kushal hai, John-kaa John physics-in how.much good be.prs.3sg John-gen bhai gaNit-me utnaa kushal hai brother math-in that.much good be.prs.3sg ‘John’s brother is as good at math as John is at physics.’ (Lit: How good John is at physics, John’s brother is that good at math.)

d.

when-clauses jab mE kamre-me ghus-aa, tab Mona gaa rahii when I room-in enter-pfv.msg then Mona sing prog.f thii be.pst.fsg ‘When I entered the room, Mona was singing.’

e.

until-clauses jab tak John nahii aa jaa-taa, tab tak mẼ yahĩ: when till John neg come hab.msg then till I here rahũgaa stay.fut.1msg ‘I will stay here until John arrives.’ (Lit: Till when John hasn’t come, I will stay here till then)

f.

since-clauses jab-se tum yahã: aa-ye ho, tab-se Mona when-since you here come-pfv.mpl be.prs.2pl then-since Mona khush hai happy be.prs.3sg ‘Mona is happy since you came here.’

 Rajesh Bhatt and Anikó Lipták

The limited surface variation found from construction to construction thus pertains to lexicalization of the relative operator and the main clause anaphoric proform, as (4) sums up. (4)

Construction relativization comparative equative conditional when-clause until-clause since-clause

Relative Phrase [... jo...] who [... jitnaa...] how.much [... jitnaa...] how.much [... dzarMarathi...] if [... jab...] when [... jab-tak...] when-till [... jab-se...] when-since

Proform [... vo...] he/she [... us-se jyaadaa...] that-than more [... utnaa...] that.much [... tarMarathi...] then [... tab...] then [... tab-tak...] then-till [... tab-se...] then-since

Given that the surface syntax of these different kinds of constructions seems to be essentially identical, one might expect the semantics of these different kinds of correlatives to be essentially identical too. Under a view of the syntax-semantics interface where the semantics interprets the syntax, any differences that we find in the semantics of these constructions should follow from the properties of the relative phrase and the demonstrative phrase and from the properties of the variable that is being abstracted over.

1.3

Maximalization in correlatives

Before we move on to examining correlatives of time and space in greater detail, let us take a look at the semantics of correlatives. Srivastav (1991) pointed out that the semantics of correlatives involve maximalization. If the relative phrase is singular, maximalization manifests itself in the form of definiteness over the predicate created by the abstraction in the correlative clause. This can be seen in (5a), which is only felicitous in a scenario where exactly one girl is standing. If no girl is standing or more than one girl is standing, (5a) cannot be used. (5) a. jo laRkii khaRii hai, vo lambii hai rel girl standing be.prs.3sg she tall be.prs.3sg ‘The girl who is standing is tall.’ (Only one girl is standing, and she is tall.) b. jo laRkiyã: khaRii hẼ, ve lambii hẼ rel girls standing be.prs.3pl they tall be.prs.3pl ‘The girls who are standing are tall.’ (All the girls who are standing are tall.)



Matching effects in the temporal and locative domains 

If the relative phrase is plural, as is the case in (5b), the correlative refers to all the girls standing, and the anaphoric phrase ve ‘they’ refers to the maximal ‘sum’ of the standing girls. Assuming that correlatives involve a uniqueness operator that applies to a predicate and picks out the maximal element that the predicate holds for allows for a uniform characterization of singular and plural correlatives. (5a) involves a singular predicate while (5b) involves a plural predicate. In case of (5a), if there is more than one standing girl or if there is none, there will be no maxima and there will be a presupposition failure. With (5b), if there is no standing girl, there will be a presupposition failure but if there is more than one standing girl, the uniqueness operator will pick out the maximal plural entity that the predicate is true of. Here this will be all the standing girls. The semantics of uniqueness/maximalization relevant here seems to be the same as that proposed for the English definite determiner the. We take maximalization to be an important aspect of the semantics of correlatives. Before we move on to the trickier case of temporal correlatives, let us examine how the maximalization semantics suggested above apply to comparatives. (6)

Rodman ke-paas jitne tattoo hẼ, Rodman near how.many tattoo be.prs.3pl Jordan ke-paas us-se zyaadaa khitaab hẼ. Jordan near that-than more title be.prs.3pl ‘Michael Jordan has more scoring titles than Dennis Rodman has tattoos.’ (Literally: How many tattoos Dennis Rodman has, Michael Jordan has more scoring titles than that)

The correlative clause in (6) provides the degree predicate how many tattoos Rodman has. Maximalizing this predicate yields the maximal degree to which Rodman has tattoos, i.e. the total number of tattoos Rodman has. This number is picked up by the main clause demonstrative and the main clause just asserts that Jordan has more tattoos than that number. In other words, maximalization gives us the right semantics. The treatment of equatives is largely the same — the only difference being that the main clause introduces a statement of equality. We will now turn to temporal and locative correlatives whose semantics do not fall out as straightforwardly.

2. Temporal and locative correlatives As discussed earlier, correlatives over non-individual variables are structurally quite similar to correlatives over individual variables. On the surface the only difference resides in the form of the relative and demonstrative pronouns used and the connectives involved. In this section, we will illustrate the connectives and their use in correlative and non-correlative sentences from Hindi-Urdu only. We refrain from illustrating the same examples from Hungarian for reasons of space, and only give a summary of the connectives as well as the types of relative and demonstrative pronouns at the end of

 Rajesh Bhatt and Anikó Lipták

this section. The interested reader might also consult Lipták (2006) on the syntax of temporal correlatives in Hungarian. Let us start the discussion with considering three kinds of connectives: the bare (zero) connective, whose semantic contribution will be the subject of further discussion; the connective that indicates the left boundary of a temporal or spatial path; and the connective that indicates the right boundary of a temporal or spatial path. Each of these is shown in the following examples: (7) a.

at/in: bare Ram baarah baje/ kal aa-yaa thaa Ram 12 o’clock/ yesterday come-pfv.msg be.pst.msg ‘Ram had come at noon/yesterday.’

b.

till: -tak Ram baarah baje-tak ghar-me thaa Ram 12 o’clock-till home-in be.pst.msg ‘Ram was at home until 12 o’clock.’

c.

from/since: -se Ram baarah baje-se office-me hai Ram 12 o’clock-from office-in be.prs.3sg ‘Ram is in the office since noon.’

The left boundary connective and the right boundary connective can also co-occur with each other. In this the Hindi-Urdu left boundary connective -se patterns with English from and not with since.3 (8) Ram baarah baje-se (le-kar) do baje-tak office-me thaa Ram 12 o’clock-from take-cp two o’clock-till office-in be.pst.msg ‘Ram was in the office from noon to 2pm.’ In Hindi-Urdu the left boundary temporal connective -se ‘since/from’ and the right boundary temporal connective -tak ‘till’ also have a spatial/locative usage. The following examples give illustration of all types of locative connectives. (9) a.

at: bare Ram vahã: rah-taa hai Ram there stay-hab.msg be.prs.3sg ‘Ram lives there.’

b.

to: -tak Ram vahã:-tak dauR-taa hai Ram there-till run-hab.msg be.prs.3sg ‘Ram runs to there.’



Matching effects in the temporal and locative domains 

c.

from: -se Ram vahã:-se dauR-naa shuruu kar-taa hai Ram there-from run-inf start do-hab.msg be.prs.3sg ‘Ram starts running from there.’

d.

from X to Y: X-tak Y-se Ram Amherst College-se Umass-tak roz dauR-taa hai Ram Amherst College-from Umass-till daily run-hab.msg be.prs.3sg ‘Ram runs daily from Amherst College to UMass.’

With these connectives in place, we can now construct temporal and locative correlatives. Again, we will only illustrate the Hindi-Urdu cases for reasons of space. Let us start the discussion with temporal correlatives. In these, the correlative clause involves a temporal relative pronoun (jab in Hindi-Urdu) together with a connective, which can be zero (10a), the left boundary marker -se (10b), or the right boundary marker -tak (10c). In the relevant examples, the main clause also consists of a temporal demonstrative (tab) together with the connective found in the correlative clause. Consider the following examples, where lbm stands for the left boundary marker and rbm for the right boundary marker. (10) a.

[when + zero connective], [then + zero connective] jab mE-ne kamre-me pravesh kiyaa, tab when I-erg room-in enter do.pst.msg then Mona-ne naac-naa shuruu kiyaa Mona-erg dance-inf start do.pst.msg ‘When I entered the room, then Mona started dancing.’

b.

[when + lbm], [ then + lbm] jab-se tum yahã: aa-ye ho, when-since you here come-pfv.mpl be.prs.2pl tab-se bhagwaan-kii kripaa-bhii yahã: aa gayii hai then-since god-gen grace-also here come go.pfv.f be.prs.3sg ‘From the time that you have come here, God’s grace has also come here.’

c.

[when + rbm], [then + rbm] jab-tak steshan khulaa thaa, tab-tak when-till station open be.pst.msg then-till Mary vahã: baiThii rah-ii Mary there seated stay-pfv.fsg ‘Mary sat at the station as long as it was open.’

The corresponding locative correlatives are very similar structurally. The difference lies in the choice of the relative pronoun: instead of a temporal relative and demonstrative pronoun (jab/tab), a locative relative and demonstrative pronoun (jahã:/vahã:) is used.

 Rajesh Bhatt and Anikó Lipták

(11) a.

[where + zero connective], [there + zero connective] Ram-ne jahã: zamiin khariid-ii hai, vo Ram-erg where land buy-pfv.f be.prs.3sg he vahĩ: makaan banaa-egaa there.only house make-fut.3msg ‘Where Ram bought land, there he’ll build a house.’

b.

[where + lbm], [there + lbm] jahã:-se Ram dauR-taa hai, mẼ where-from Ram run-hab.msg be.prs.3sg I vahĩ:-se dauR-ũ:gaa run-fut.1msg there.only-from ‘I will run from exactly the location that Ram runs from.’

c.

[where + rbm], [there + rbm] jahã:-tak Ram dauR-taa hai, mẼ where-till Ram run-hab.msg be.prs.3sg I vahã:-tak dauR-ũ:gaa there-till run-fut.1msg ‘I will run till where Ram runs.’

The temporal correlatives in (10) and the locative correlatives in (11) pattern with ordinary correlatives. The parallel with ordinary correlatives goes further – like ordinary correlatives, both temporal correlatives and locative correlatives allow for multiple correlatives. Such correlatives have more than one relative pronoun in the correlative clause and a corresponding number of demonstrative pronouns in the main clause. Consider the following examples, which show a multi-headed temporal and a multiheaded locative correlative.4 (12) a.

[when + lbm, when + rbm], [then + lbm, then + rbm] jab-se (le-kar) jab-tak Ravi Dilli-me thaa, when-from take-cp when-till Ravi Delhi-in be.pst.msg tab-se (le-kar) tab-tak Sita khush thii then-from take-cp then-till Sita happy be.pst.fsg ‘Sita was happy throughout the interval that Ravi was in Delhi.’

b.

[where + lbm, where + rbm], [there + lbm, there + rbm] jahã:-se (le-kar) jahã:-tak Ravi dauR-taa where-from take-cp where-till Ravi run-hab.msg hai mẼ-bhii vahã:-se (le-kar) vahã:-tak dauR-ũ:gaa be.prs.3sg I-also there-from take-cp there-till run-fut.1msg ‘Ravi runs from point A to point B. I will also run from point A to point B.’

This section illustrated the form and use of temporal and locative connectives as well as the relative and demonstrative pronouns that occur in correlatives, which will be



Matching effects in the temporal and locative domains 

relevant for the discussion of matching effects in the next section. Before turning to these matching effects, we would like to provide a note about the Hungarian equivalents of the relevant connectives, as well as the relative and demonstrative pronominals that are formed with the help of these. The temporal domain in Hungarian makes use of the following connectives: -­ kor ‘at’, -ig ‘till’ and óta ‘since’. The locative domain uses partly different connectives: the equivalent of the at connective is –nál/nél; the equivalent of till/to is -ig, and from is expressed as –tól/től, a connective that is also used in the temporal domain in constructions like (8) above. Concerning the form of the relative pronouns that occur in correlatives, these are always built on a wh-item prefixed with the relative marker a- morpheme. The whitem mi (or its variant me) ‘what’ gives rise to all temporal and some locative relative pronouns, and there is a specifically locative wh-pronoun, hol ‘where’ that can be used in the expressions of locative relatives. The demonstrative pronominals that occur in the main clause of correlative constructions in Hungarian have two paradigms. The temporal paradigm is built on the distal demonstrative az ‘that’, and includes akkor ‘that.at’ in the meaning of then, addig ‘that.till.’ in the meaning of till then and azóta ‘that.since’ in the meaning of since then. In the latter meaning, attól ‘that.from’ can also occur (most always in combination with kezdve/fogva ‘beginning’). The locative paradigm also uses attól and addig in the meaning of ‘from a point/to a point in place’. Next to these items, there is an independent paradigm of locative demonstratives, which can be called the o- series. Thus we find ott ‘there’, which consists of the locative pronoun o- and a –tt at connective, oda ‘there.to’ (in the meaning of to there) and onnan ‘there.from’ (meaning from there). In these forms the at, to and from connectives are obsolete, non-productive connectives. Interestingly, oda and onnan can further combine with –ig and –tól/től to give odáig ‘there.to.till’ (meaning to that point) and onnantól ‘there.from.from’ (meaning from that point). We will come back to the peculiarities of the pronominal paradigms in Sections 3.3. and 4.2. below.

3. Matching effects A major point of divergence between ordinary correlatives and temporal/spatial correlatives is that ordinary correlatives do not display ‘matching’ effects. Temporal correlatives in Hindi-Urdu as well as in Hungarian do. Spatial correlatives in HindiUrdu do not display these matching effects but a weaker version of the effect surfaces in Hungarian.

 Rajesh Bhatt and Anikó Lipták

3.1

Matching effects with free relatives and correlatives

Matching effects are typically used to describe cases like the following which involve free relatives (see Grimshaw (1977), Bresnan and Grimshaw (1978) and Groos and van Riemsdijk (1981) among many others.) For a structure containing a free relative to be acceptable, the free relative needs to appear in a position where the case assigned to its wh-phrase (the internal case) is identical to the case assigned to the position where the free relative appears (the external case). Consider for example the case of German in (13). The grammatical (13a) has the free relative appear in a position that receives nominative case. The wh-phrase itself also receives nominative case. In the ungrammatical cases, there is a mismatch between the external case and the internal case. (13) a.

external case = internal case = nominative Wer(nom) nicht stark ist, muss klug sein. who not strong be must clever be-inf ‘Who is not strong must be clever.’

b. external case = nominative ≠ internal case = accusative * Wen(acc)/*Wer(nom) Gott schwach geschaffen hat, muss whom/who God weak create.ptcp has must klug sein. clever be-inf ‘Who God has created weak must be clever.’ c. external case = nominative ≠ internal case = dative * Wem(dat)/*Wer(nom) Gott keine Kraft geschenkt hat, whom/who God no strength give.ptcp has muss klug sein. must clever be.inf ‘Who God has given no strength to must be clever.’ In the above examples, we see an instance of case-matching. Next to case-matching, category matching is also displayed by some languages. In category matching the category of the free relative wh-phrase and the external environment have to be the same. Having seen what matching phenomena are, we can now come back to correlatives. Crucially, matching effects are not displayed by ordinary correlative clauses. Consider for examples the following examples from Hindi (cf. 14) and Hungarian (cf. 15), which all involve distinct case markers on the relative phrase and the demonstrative: (14) a.

jis aadmii-ne mujhe naukrii dii thii which man-erg I.dat job give.pfv.fsg be.pst.fsg us-ko aaj medal mil-aa dem-dat today medal receive-pfv.msg ‘The man who had given me a job got a medal today.’



Matching effects in the temporal and locative domains 

b.

jis aadmii-ke-paas avEdh banduuk thii which man-near illegal gun be.pst.fsg us-ko jurmaanaa de-naa paR-aa dem-dat penalty give-inf have.to-pfv.msg ‘The man who had an illegal gun had to pay a fine.’

(15) a. Akit meghívtunk, annak küldtünk meghívót. who.acc pv.invite. pst.1pl that.dat send.pst.1pl invitation.acc ‘Whom we invited, we sent an invitation to those.’ b. Akinél fegyvert találnak, attól elveszik. who.at gun.acc find.3pl that.from confiscate.3pl ‘With whom they find a gun, they confiscate it from those.’ The absence of matching effects in these constructions makes a certain amount of intuitive sense since matching effects can be thought of as being a language particular way of resolving a morphological conflict: there are two cases, internal and external, but only one location where they can be realized. In correlatives, this conflict does not arise: the internal case is realized on the relative phrase inside the correlative clause (a free relative) and the external case is realized on the demonstrative phrase in the main clause.

3.2

Matching effects with temporal correlatives

Interestingly, unlike correlatives denoting individuals, correlatives denoting other types of entities show matching effects. We start illustrating this with correlatives over times. Of these, Section 1.2. illustrated three kinds of correlative clauses, which we referred to as when-clauses, till when-clauses, and since when-clauses. Corresponding to these we can have three kinds of main clauses: then-clauses, till then-clauses, and since then-clauses. Among these relative and main clauses, one might expect to find 9 possible combinations. In reality only some of these combinations are attested. The matching cases are all grammatical, as the following examples show. The (a) examples are from Hindi-Urdu, the (b) ones from Hungarian. (16) [when...] [then...] a. jab Ram ghar lauT-aa (tab) Sita nahaa rahii thii when Ram home return-pfv.msg then Sita bathe prog.f be.pst.fsg ‘When Ram returned home, Sita was bathing.’ b. Amikor János megjött, (akkor) Mari TV-t nézett what.at János arrive.pst that.at Mari TV-acc watch.pst ‘When János arrived, Mari was watching TV.’

 Rajesh Bhatt and Anikó Lipták

(17) [till when...] [till then...] a. jab-tak Ram yahã: thaa, (tab-tak) Sita khush thii when-till Ram here be.pst.msg then-till Sita happy be.pst.fsg ‘During the duration Ram was here, Sita was happy.’ b. Ameddig János alszik, (addig) Mari TV-t néz what.till János sleep that.till Mari TV-acc watch ‘While János sleeps, Mari watches TV.’ (18) [since when...] [since then...] a. jab-se Ram yahã: aa-yaa hai, when-since Ram here come-pfv.msg be.prs.3sg (tab-se) Sita khush hai then-since Sita happy be.prs.3sg ‘Sita is happy since Ram came.’ b. Amióta János dolgozik, (azóta) Mari rendszeresen sportol. what.since János work that.since Mari regularly work.out ‘Since the time that János has started working, since that time Mari works out regularly.’ Cases involving a till/since when-correlative clause and a mismatching then-clause are ungrammatical in both Hindi and Hungarian.5 (19) *[till when...] [then...] a. *jab-tak Ram yahã: thaa, tab Sita khush thii when-till Ram here be.pst.msg then Sita happy be.pst.fsg ‘Till the time Ram was here, then Sita was happy.’ b. *Ameddig János alszik, akkor Mari hazajön what.till János sleep that.at Mari come.home ‘Till John sleeps, at that time Mary comes home.’ (20) *[till when...] [since then...] a. *jab-tak Ram yahã: thaa, tab-se bhagwaan-kii when-till Ram here be.pst.msg then-since god-gen kripaa yahã: hai] grace here be.prs.3sg ‘Till when John was here, since then God’s grace is here.’ b. *Ameddig János aludt, azóta Mari TV-t néz what.till János sleep.pst that.since Mari TV-acc watch ‘Till the time John was sleeping, since that time Mari is watching TV.’



Matching effects in the temporal and locative domains 

(21) *[since when...] [then...] a. *jab-se Ram yahã: aa-yaa hai, when-since Ram here come-pfv.msf be.prs.3sg tab Sita khush thii then Sita happy be.pst.fsg ‘Since when Ram came here, then Sita was happy.’ b. *Amióta János dolgozik, akkor Mari elkezdett what.since János work that.at Mari begin.pst rendszeresen sportolni regularly work.out.inf ‘Since the time that János is working, then Mari started working out regularly.’ (22) *[since when...] [till then...] a. *jab-se Ram yahã: aa-yaa hai, when-since Ram here come-pfv.msg be.prs.3sg tab-tak Sita so-tii rah-ii then-till Sita sleep-imp.f stay-pfv.fsg ‘Since when Ram came here, till then Sita kept sleeping.’ b. *Amióta János dolgozik, addig Mari rendszeresen sportolt what.since János works that.till Mari regularly work.out.pst ‘Since the time János works, till that time Mari regularly worked out.’ Hindi-Urdu and Hungarian differ with respect to the grammaticality of the two remaining cases — in these the correlative clause is a when-clause and the main clause is a mismatching one: (23) [when...] [till then...] a. jab Ram Dilli-se lauT-aa thaa, when Ram Delhi-from return-pfv.msg be.pst.msg Sita-ne tab-tak tapasyaa kii thii Sita-erg then-till meditation do.pfv.f be.pst.fsg ‘The time when Ram had returned from Delhi, Sita had meditated until then.’ b. * Amikor János megjött, addig Mari szomorú volt what.at János arrive.pst that.till Mari sad be.pst ‘The time that János came, Mari was sad till then.’ (24) [when...] [since then...] a. jab Ram yahã: aa-yaa thaa tab-se when Ram here come-pfv.msg be.pst.msg then-since bhagwaan-kii kripaa yahã: hai god-gen grace here be.prs.3sg ‘God’s grace is here since the day that Ram came here.’

 Rajesh Bhatt and Anikó Lipták

b. *Amikor János megjött, azóta Mari folyton TV-t néz what.at János arrive.pst that.since Mari continuously TV-acc watch. ‘Since János arrived, Mari continuously watches TV.’ The pattern can be schematized as shown in Table 1. It is worth noting that the difference between Hindi-Urdu and Hungarian seen above cannot be reduced to an exceptional property of Hindi-Urdu by which the till/ since on the relative phrase can simply be left unpronounced.6 The idea behind this putative reduction would be that the till/since markers inside the correlative clause are semantically unnecessary, and their presence is purely required by morphological reasons. If this was true, the difference between Hindi-Urdu and Hungarian would be a morphological difference only, with Hungarian imposing a strict matching requirement and Hindi-Urdu a weaker one. The Hungarian requirement would be ‘only likes match’ while the Hindi-Urdu requirement would allow for a bare when to match with any main clause then (bare, till, or since). The reason we do not pursue this line of inquiry is because the presence of the till/ since in the temporal correlative is subject to aspectual restrictions. If we insert the ‘missing’ till/since into the acceptable mismatch structures, the resulting sentences become aspectually odd. (25) a. [till when...] [till then...] #jab-tak Ram Dilli-se lauT-aa thaa, when-till Ram Delhi-from return-pfv.msg be.pst.msg Sita-ne tab-tak tapasyaa kii thii Sita-erg then-till meditation do.pfv.f be.pst.fsg ‘The time when Ram had returned from Delhi, Sita had meditated until then.’ b. [since when...] [since then...] #jab-se Ram yahã: aa-yaa thaa when-since Ram here come-pfv.msg be.pst.msg tab-se bhagwaan-kii kripaa yahã: hai then-since god-gen grace here be.prs.3sg ‘God’s grace is here since the day that Ram came here.’ Table 1.  Matching in Hindi-Urdu and Hungarian temporal correlatives

when till when since when

then

till then

since then

✓

✓Hindi-Urdu/*Hungarian

✓Hindi-Urdu/*Hungarian

* *

✓ *

* ✓



Matching effects in the temporal and locative domains 

The oddness of (25) stems from the aspectual restriction imposed by till/then. Till seems to need a stative predicate and since seems to need a present perfect. When, on the other hand, by itself does not impose any such restrictions. If the till/since elements were present for purely morphological reasons, we would not expect such aspectual restrictions in these cases. Therefore we cannot treat the Hindi-Urdu grammatical mismatch cases of the [when...][till/since then...] sort as being the same at some level as the corresponding matching cases. In other words, the when-clauses here are really when-clauses and the till/since when-clauses in the matching cases are really till/sincewhen clauses.

3.3

Matching effects with locative correlatives

Section 2 has shown that locative correlatives and temporal correlatives look very similar. In fact, in Hindi-Urdu and to a more limited extent in Hungarian, even the connectives (till, from/since) are same. The only difference seems to be in the choice of the relative phrase/demonstrative phrase: when/then vs. where/there. Given these parallels, it is surprising that locative correlatives do not display matching effects in Hindi-Urdu. In Hungarian, the presence of matching effects depends upon the exact locative relative phrase used, as we will show later in this section. The matching cases of locative correlatives in Hindi-Urdu have been considered earlier and are repeated here, with their Hungarian equivalents added to them. (26) [where...] [there...] a. Ram-ne jahã: zamiin khariid-ii hai, Ram-erg where land buy-pfv.f be.prs.3sg vo vahĩ: makaan banaa-egaa. he there.only house make-fut.3msg ‘Where Ram bought land, there he’ll build a house.’ b. Ahol fúj a szél, ott hideg van where blow the wind there cold be ‘Where it is windy, it is cold there.’ (27) [till where...] [till there...] a. jahã:-tak Ram dauR-taa hai, mẼ vahã:-tak dauR-ũ:gaa where-till Ram run-hab.msg be.prs.3sg I there-till run-fut.1msg ‘I will run till where Ram runs.’ b.

Ameddig tegnap elfutottam, addig/odáig what.till yesterday run.pst.1sg that.till there.to.till ma kocsival mentem today car.with go.pst.1sg ‘To the place where I ran to yesterday, I went today by car.’

 Rajesh Bhatt and Anikó Lipták

(28) [from where...] [from there...] a. jahã:-se Ram dauR-taa hai, mẼ vahĩ:-se where-se Ram run-hab.msg be.prs.3sg I there.only-from dauR-ũ:gaa run-fut.1msg ‘I will run from exactly the location that Ram runs from.’ b.

Ahonnantól futni kezdtem, onnantól where.prom.from run-inf begin.pst.1sg that.from.from látszik a lábnyomom.7 show the footsteps.poss.1sg ‘From the place where I started running, my footsteps are showing.’

Other combinations of correlative and main clauses do not display matching effects either in Hindi-Urdu: (29) [till where...][from there...] a. Ram jahã:-tak dauR-taa hai, mẼ vahã:-se Ram where-till run-hab.msg be.prs.3sg I there-from dauR-naa shuruu kar-ũ:gaa run-inf start do-fut.1msg ‘I will start running from the point to which Ram runs.’ b.

[till where...][there...] Ram jahã:-tak dauR-taa hai, Sita vahĩ: Ram where-till run-hab.msg be.prs.3sg Sita there.only rah-tii hai stay-hab.f be.prs.3sg ‘Sita stays at he location till which Ram runs.’

c.

[from where...][till there...] Ram jahã:-se dauR-taa hai, mẼ vahã:-tak Ram where-till run-hab.msg be.prs.3sg I there-from dauR-naa shuruu kar-ũ:gaa run-inf start do-fut.1msg ‘I will run till the location from which Ram starts running.

d.

[from where...][there...] Ram jahã:-se dauR-taa hai, Sita Ram where-from run-hab.msg be.prs.3sg Sita vahĩ: rah-tii hai there.only stay-hab.f be.prs.3sg ‘Sita stays at he location from which Ram runs.’



Matching effects in the temporal and locative domains 

e.

[where...][till there...] jahã: Sita rah-tii hai, mẼ vahã:-tak dauR- ũ:gaa where Sita stay-hab.F be.prs.3sg I there-till run-fut.3msg ‘I will run till where Sita lives.

f.

[where...][from there...] jahã: Sita rah-tii hai, mẼ vahã:-se dauR-ũ:gaa where Sita stay-hab.f be.prs.3sg I there-from run-fut.1msg ‘I will run from where Sita lives.

In Hungarian, the situation is slightly different. The locative paradigm clearly displays matching effects, even if a somewhat less strong sort than the temporal paradigm. We will illustrate this with till where-correlatives. As can be seen in the following examples, the non-matching combinations are degraded.8 (30) a. [till where...] [there...] ??Ameddig tegnap elfutottam, ott van egy fa what.till yesterday run.pst.1sg there be a tree ‘The place where I ran to yesterday, there is a tree there.’ b. [till where...] [from there...] ??Ameddig tegnap elfutottam, onnantól what.till yesterday run.pst.1sg there.from.from ma tovább mentem today further go.pst.1sg ‘From the place where I ran to yesterday, I went further from there today.’ It must be mentioned that next to the above instances of locative correlatives, there are also cases of correlatives that do not display matching effects, as the following examples demonstrate: (31) a.

[where...] [to there...] Ahol a karók vannak, oda paprikát vetek where the sticks be.pl there.to paprika.acc sow.1sg ‘I will sow paprika where the sticks are.’

b.

[where...] [from there...] Ahol a karók vannak, onnan kiszedtem where the sticks be-pl there.from remove.pst.1sg a virághagymákat the flowerbulbs-acc ‘I took out the flowerbulbs from the places where the sticks are.’

c.

[from where...] [there...] Ahonnan elköltözött az iskola, ott nincs könyvtár where.from move.out.pst the school there be.neg library ‘There is no library in places where the school has moved out.’

 Rajesh Bhatt and Anikó Lipták

d.

[from where...] [to there...] Ahonnan elköltözött az iskola, oda nem where.from move.out.pst the school there.to not költöznek új családok move new families ‘New families do not move into places where the school has moved out.’

e.

[to where...] [there...] Ahova leszúrtam egy karót, ott virághagymák vannak where.to push.pst.1sg a stick.acc there flowerbulbs be.pl ‘At the place where I have put a stick, there are flowerbulbs.’

f.

[to where...] [from there...] Ahova leszúrtam egy karót, onnan where.to push.pst.1sg a stick.acc there.from kiszedtem a virághagymákat remove.pst.1sg the bulbs.acc ‘I have removed the flowerbulbs from the place where I have put a stick.’

This class of locative pronominals could be called pointwise locatives, as the location expressed in these examples is not associated with a path, rather with a point. This contrasts with the forms in examples (30) which are clearly path-denoting. The appearance of the forms in (31) is moreover determined by the subcategorizational needs of the predicate at hand. For example, in (31f) the predicate in the correlative clause subcategorizes for the to connective in ahova ‘where. to’ and the predicate in the main clause subcategorizes for from connective in onnan ‘there.from’. The lack of matching effects in these cases is thus on a par with the examples in (15). According to the evidence of these examples, we can conclude that the matching effect in Hungarian seems to be keyed into the expression of an unselected temporal/ spatial path argument. When we can be certain that the path argument is being manipulated as is the case with pathwise locatives in (30), we find matching effects.

3.4

Another difference between locative and temporal correlatives

Related to the fact that we find matching effects in Hindi-Urdu temporal correlatives but not in Hindi-Urdu locative correlatives, we can observe another contrast as well. From/till locative correlative clauses in Hindi-Urdu can combine with a demonstrative phrase to yield a DP that denotes a location. Thus both (32b) and (32c) are well-formed DPs that pick out the location to which Ram runs, and the location from which Ram starts running, respectively, as possible answers to (32a): (32) a. tum Ram-se kahã: mil-oge? you Ram-instr where meet-fut.2mpl ‘Where will you meet Ram?’



Matching effects in the temporal and locative domains 

b. jahã:-tak vo dauR-taa hai, vahã: where-till he run-hab.msg be.prs.3sg there ‘Till where he runs, there’ c. jahã-se vo dauR-naa shuruu kar-taa hai, vahã: where-from he run-inf start do-hab.msg be.prs.3sg there ‘From where he starts running, there.’ But this is not possible with temporal correlatives. Here we cannot pick out the left boundary or the right boundary like we were able to with locative correlatives. (33) a. tum Ram-se kab mil-oge? you Ram-instr when meet-fut.3mpl ‘Where will you meet Ram?’ b. *jab-tak vo dauR-taa hai, tab when-till he run-hab.msg be.prs.3sg then ‘*Till when he runs, then’ c. *jab-se vo dauR-naa shuruu kar-taa hai, tab when-from he run-inf start do-hab.msg be.prs.3sg then ‘From when he starts running, then.’ To actually get at the left/right boundary, we need to switch to a plain when-clause and modify the predicate so that the when-clause picks out the point of culmination/initiation. Merely switching to an externally headed version does not help. This is similar to what we find with Hungarian externally headed temporal and locative relative clauses, which can be considered parallels of (32) and (33). In answers to questions like (32a) or (33a), Hungarian uses headed relative patterns that require matching connectives in the temporal case: (34) a. Mikor találkozol / találkoztál Jánossal? what.at meet.2sg meet.pst.2sg János-with ‘When will you meet János?’ b. *Akkor, ameddig fut. that.at when.till run ‘Till when he runs.’ c. *Akkor, amióta fut. that.at when.since run ‘Since when he is running.’ And, unlike in Hindi-Urdu, the locative pattern is also just as bad as the temporal ones: (35) a. Hol találkozol Jánossal? where meet.2sg János-with ‘Where will you meet János?’

 Rajesh Bhatt and Anikó Lipták

b. *Ott, ameddig fut. there where.to run ‘At the point to where he runs.’ c. *Ott, ahonnantól fut. there where.to.from run ‘At the point where he runs from.’ Stepping back to the Hindi-Urdu cases in (32) and (33), it needs to be mentioned that by modifying the question so that it is about paths instead of points, the contrast between locative and temporal correlatives disappears, as the following show, where both locative and temporal answers (examples (b) and (c) respectively) are possible to the questions in (a). Note that the presence of till/since on the demonstrative is obligatory. (36) a. tum kahã:/kab-tak Ram-kaa piichaa kar-oge? you where/when-till Ram-gen follow do-fut.2mpl ‘Till where/when will you follow Ram?’ b. jahã:-tak Ram jaa-egaa, vahã:-tak where-till Ram go-fut.3msg there-till ‘Till where Ram goes, till there.’ c. jab-tak Ram bhaag-egaa, tab-tak when-till Ram run-fut.3msg then-till ‘Till when Ram runs, till then.’ (37) a. tum kahã:/kab-se Ram-kaa piichaa kar rahe ho? you where/when-from Ram-gen follow do prog.mpl be.prs.2pl ‘From where/when will you follow Ram?’ b.

jahã:-se Ram-ne bhaag-naa shuruu kiyaa where-from Ram-erg run-inf start do.pfv.msg thaa, vahã:-se there-from be.pst ‘From where Ram had started running, from there.’

c.

jab:-se Ram-ne bhaag-naa shuruu kiyaa when-since Ram-erg run-inf start do.pfv.msg hai tab-se be.prs.3sg then-since ‘Since when Ram had started running, since then.’

There are two ways to think about this pattern. The first is to treat the answers to questions as being derived via reduction of a full clause that consists of a correlative clause and a matrix clause. Then the pattern reduces to the previously observed fact that in Hindi-Urdu, temporal correlatives display ‘matching’ while locative correlatives do not. But it has been noted that correlative clauses in Hindi-Urdu can also directly adjoin to the demonstrative phrase they modify (see Dayal (1996), Bhatt (2003)). Given



Matching effects in the temporal and locative domains 

this, it must be the case that the denotation of till/since temporal correlative clauses is such that when it modifies a demonstrative phrase, the resulting object does not denote a point of time. This object should, however, be able to combine with a till/since and then modify another clause. We will take this line of investigation further in our discussion of the temporal matching effect in Hindi-Urdu.

4. Explanations We will attempt to provide a semantic explanation for the matching effect discussed in the previous section. This is in large part because the environments where this matching effect is found are not the kind of environments where one finds the classic morphosyntactic matching effects familiar to us from free relative constructions, as we have shown in (14) and (15). We find matching effects in correlatives and in Hungarian even in headed relatives (cf. fn 5 and 8), both environments where a morpho-syntactic matching effect would be quite surprising. Next to the semantic ingredient, our explanation will also need to have a syntactic component to handle the variation that we found between Hindi-Urdu and Hungarian. Recall that matching effects in Hungarian are stricter than in Hindi-Urdu. Hungarian requires matching in locative correlatives also and disallows the ‘[when...] [till/since then...]’ case allowed in Hindi-Urdu.

4.1

The first attempt: Points of time

We need to start by making an assumption about what a when-clause denotes. Let us begin with the following proposal which is inspired directly by the semantics of plural individual correlatives discussed earlier in the paper. (38) Putative Semantics for when-clauses: a. The when-clause gives the maximal interval/sum of points at which the predicate holds – a point of time or an interval/sum of points depending upon the predicate b. The then picks out this point/interval and the matrix clause must hold at this point/throughout this interval/sum of points Note that this semantics runs into problems right away with durative predicates in when-clauses and with non-durative predicates in then-clauses. The following example shows such a problematic case: the when-clause picks out an interval, but the thenclause is not durative: (39) a.

jab tum so rahe the, tab bagal-vaale when you sleep prog.mpl be.pst.mpl then neighboring ghar-me chori ho gayii house-in theft be go.pfv.fsg ‘While you were sleeping, a theft happened next door.’

 Rajesh Bhatt and Anikó Lipták

b.

jab tum so rahe the Madhu tab when you sleep prog.mpl be.pst.mpl Madhu then aa-yii thii come-pfv.f be.pst.fsg ‘Madhu had come while you were sleeping.’

The source of this problem lies in our treatment of all temporal abstractions as involving an underlying at connective. A more adequate treatment needs to take into account the contribution of aspect. We will not attempt such a treatment here but just note this problem with our proposed semantics of when-clauses. Next let us consider the contribution of till and since. (40) Assuming then refers to a point of time (see Iatridou et al. (2001), von Fintel and Iatridou (2002)): a. till then P is true if there is an interval whose right boundary is set by then and P holds throughout this interval. The left boundary is set by context. b. since then P is true if there is an interval whose left boundary is set by then and P holds throughout this interval. The right boundary is set by tense. (41) Setting the right boundary with until: a. (Talking about a graduate student who graduated in 1999.) Marlyse was at Harvard until 1999. (This does not require that she was at Harvard all her life.) b. I was at IIT Kanpur until 1993. Then I was at Penn until 1999. After that I was in Texas until 2004. (42) Setting the left boundary with since: a. John is in London since 2002. (continues up to now) b. John was in London since 1999. (continues up to a point in the past) (43) Assuming then refers to an interval: a. till then P is defined if then picks out an interval with a linguistically specified right boundary. It is true if P holds throughout this interval. b. since then P is defined if then picks out an interval with a linguistically specified left boundary. It is true if P holds throughout this interval. The next question that we need to answer is what till/since when correlatives denote. We consider two options. The first is that they pick out points of time and the second is that they pick out intervals, as summarized in (44a) and (44b). When it comes to maximalization, the difference between the two approaches is that the interval approach does not need to make reference to the direction of maximization. (44) a. Point of time approach (i) till when P picks out the latest point at which P holds (maximization (to the right) of the right boundary)



Matching effects in the temporal and locative domains 

(ii) since when P picks out the earliest point at which P holds (maximization (to the left) of the left boundary) b. Interval approach (i) till when P picks out the maximal interval bounded on the left by the contextually specified left boundary (ii) since when P picks out the maximal interval bounded on the right by the contextually specified right boundary

Of the two approaches, the point of time approach makes the wrong predictions with respect to matching. It would allow the following to be possible, contrary to facts in Hindi (45a) and Hungarian (45b): (45) *[till when...] [since then...] a. *jab-tak Ram yahã: thaa, tab-se bhagwaan-kii when-till Ram here be.pst.msg then-since god-gen kripaa yahã: hai grace here be.prs.3sg ‘Till when John was here, since then God’s grace is here.’ b. *Ameddig János aludt, azóta Mari TV-t néz what.till János sleep.pst that.since Mari TV-acc watch ‘Till the time John was sleeping, since that time Mari is watching TV.’ There is no reason why the till when-clause could not give us a right boundary which could then be used as a left boundary by the since then-clause. That this is not possible suggests that what we get from the till/since when-clause cannot be a point of time. This conclusion also fits well with the observation that till/since when temporal correlatives do not form time-denoting DPs with a demonstrative phrase. In this they diverge from ordinary when-clauses which can and do combine with demonstrative phrases to form temporal definite descriptions.

4.2

The second attempt: Intervals

Having seen that the point of time approach fails to account for matching effects in both Hindi-Urdu and Hungarian, the conclusion must be drawn that this approach is not viable. We need to see if the interval approach does a better job in this respect. If we assume that till/since when-clauses actually yield temporal intervals, we have greater success with providing an explanation for the matching effect. But to account for the facts, we need more than just plain intervals. What we need for modeling till/ since when-clauses are intervals with distinguished right/left boundaries. In this model, till when-clauses will pick out an interval with a distinguished right boundary and since when-clauses will pick out an interval with a distinguished left boundary. In the case of multi-headed temporal correlatives, we will also need intervals with distinguished left and right boundaries.

 Rajesh Bhatt and Anikó Lipták

Concerning then, we need to assume that it is by itself just a temporal variable that can range over points of time/ordinary intervals but it cannot denote a distinguished interval. If it ranges over points of time/ordinary intervals, it can stand by itself and modify a clause without the help of a temporal connective like till/since. However, when then is associated with a till/since when-clause, it can only denote an interval with a distinguished left/right boundary. In such a case, it cannot stand by itself nor can it directly modify a clause. For it to do so, a till/since connective is needed. Moreover, it has to be the right connective. If the interval has a distinguished right boundary, then must combine with a till and if it has a distinguished left boundary, then must combine with a since. This explains the matching effects observed, by ruling out the combinations [till when...][since then...] or [till when...][ then...] as well as [since when...] [till then...] and [since when...][then...]. While the above sketched interval approach seems promising, it is not clear how it can explain a set of facts that we have not accounted for yet. These concern the distinct behavior of Hindi-Urdu and Hungarian when it comes to allowing for [when...] [till/since then] combinations. As we have shown in Section 3 (recall Table 1), the two languages differ in that while Hindi-Urdu allows for such combinations, Hungarian does not. Consider the following examples repeated from above: (46) [when...] [till then...] a. jab Ram Dilli-se lauT-aa thaa, Sita-ne when Ram Delhi-from return-pfv.msg be.pst.msg Sita-erg tab-tak tapasyaa kii thii then-till meditation do.pfv.f be.pst.fsg ‘The time when Ram had returned from Delhi, Sita had meditated until then.’ b. * Amikor János megjött, addig Mari szomorú volt what.at János arrive.pst that.till Mari sad be.pst ‘The time that János came, Mari was sad till then.’ (47) [when...] [since then...] a. jab Ram yahã: aa-yaa thaa tab-se when Ram here come-pfv.msg be.pst.msg then-since bhagwaan-kii kripaa yahã: hai god-gen grace here be.prs.3sg ‘God’s grace is here since the day that Ram came here.’ b. *Amikor János megjött, azóta Mari folyton what.at János arrive.pst that.since Mari continuously TV-t néz TV-acc watch. ‘Since János arrived, Mari continuously watches TV.’



Matching effects in the temporal and locative domains 

The behavior of Hindi-Urdu is the pattern that we expect on the basis of our model. In these examples the ordinary when-clause denotes a point of time, which serves as the left or right boundary of the distinguished interval denoted by till/since then. How can we account for the presence of matching effects in Hungarian? Since we do not want the semantics to vary across languages, it must be the case that the ungrammaticality of the above Hungarian examples follows from something more language-specific. Unfortunately, at this point, we cannot offer any explanation of the observed facts, but we note that although (46b) and (47b) are ungrammatical, the [when...] [since then...] combination can be expressed in a grammatical way, using a less frequent form of since then: akkortól ‘that.at.from’, which is built with the from connective –tól/től, added to the ordinary then form akkor ‘that-at’. Crucially, this form can surface in correlatives, and when it combines with a when-clause, it does not give rise to matching effects:9 (48)

Amikor a szerver elküldi az üdvözlőlevelet, what.at the server send the welcome.letter.acc akkortól él a tagság. that.at.from live the membership ‘Membership is active from the time when the server sends the welcome letter.’

The only speculation we can offer about this type of matching example is that some kind of parallelism effect can be at play here: the relative phrase amikor and the demonstrative akkor are matching forms in [when…][then...] contexts, which might facilitate the acceptability of sentences like (48).

4.3

Explaining the difference between Hindi-Urdu and Hungarian in the temporal domain

The previous section has shown that the interval approach seems to be a promising approach to account for matching effects in correlatives in the temporal domain. The question is, does this approach carry over to the locative domain, too? If the locative and the temporal domain were the same, we would expect that locative correlatives show the same kind of matching effects as temporal ones. This is, however, not what we find. As Section 3.3 showed, Hindi-Urdu displays no matching effects of any sort in the locative domain. It is tempting to relate the absence of matching effects in Hindi-Urdu locative correlatives to the plausibly greater individuatability of locations over times. This squares well with the finding that from/till locative correlatives can easily combine with a demonstrative to denote a location, consider the following examples, repeated from (32) again: (49) a. tum Ram-se kahã: mil-oge? you Ram-instr where meet-fut.2mpl ‘Where will you meet Ram?’

 Rajesh Bhatt and Anikó Lipták

b. jahã:-tak vo dauR-taa hai, vahã: where-till he run-hab.msg be.prs.3sg there ‘Till where he runs, there’ c. jahã-se vo dauR-naa shuruu kar-taa hai, vahã: where-from he run-inf start do-hab.msg be.prs.3sg there ‘From where he starts running, there.’ The demonstrative moreover can also take the form of the ordinary, non-locative vo ‘that’ phrase, which can be subject of a predicate nominal such as place, as the following example illustrates: (50)

Ram jahaaN-tak/se dauR-taa hai, vo Ram where-till/from run-hab.msg be.prs.3sg that acchii jagah hai good place be.prs.3sg ‘The place till/from where Ram runs, that’s a good place.’

Intuitively, it is correct to say that spatial relations are better at providing points than temporal ones. This no doubt follows from the dimensionality differences between the two domains, the idea being that 3-dimensional space allows for a better definition of points that 1-dimensional time. The explanation behind the lack of matching effects in Hindi-Urdu locatives might then follow from the fact that where/there as well as from/ to where/there can make reference to both points as well as paths, unlike what happens in the temporal domain. Yet, it is also clear that greater individuatability of locations is not a universal property, languages can differ with respect to what extent their syntax allows for it. Hungarian is clearly an example where from/to where/there cannot express a point in place, as was shown in (35) above. Neither can from/to where correlatives associate with an ordinary demonstrative that is subject to a predicate like place (compare the grammatical (50) above): (51) *??Ameddig János futott, az szép hely. what.till János run.pst that nice place ‘The place where János ran to is a nice place.’ This shows that from/to where correlatives cannot express a point, only a path, and as the result of this, Hungarian locative correlatives do show matching effects, similarly to the temporal domain. Greater individuatability does enter the picture, but only to the degree that grammaticality judgments improve a bit compared to the temporal domain. The different behavior of from/to where correlatives in Hungarian vs. Hindi-Urdu when it comes to reference to points and paths is what underlies the difference in matching effects in the locative domain in the two languages. There seems to be a lexico-semantic difference between the two languages in that from/to/till connectives in the locative domain are capable of expressing points in Hindi-Urdu, but not in



Matching effects in the temporal and locative domains 

Hungarian. The generalization that we arrived at, on the basis of these two languages, is that if from/to where correlatives can denote a point, matching effects are missing. While we cannot undertake the checking of the cross-linguistic validity of this generalization in the present article, we round off the discussion by showing that our generalization makes the right predictions for Dutch.10 It seems that the Hungarian pattern is replicated in Dutch. To consider the case of Dutch, note first that temporal clauses show a matching effect. A till-temporal clause can only combine with a tillphrase in the main clause: (52)

Totdat Jan wakker werd, totdan/*sindsdien heeft till.that Jan awake become.pst till.then /since has Marie tv gekeken. Marie tv watch.prtc ‘Until the time that Jan woke up, Marie watched tv.’

Similarly to the temporal domain, non-matching cases in the locative domain sound weird: (53) Tot waar toe Jan gerend heeft, tot daar toe ga ik ook rennen till where to Jan run.prtc has till there to go I also run.inf ‘I will run to the place where Jan has run to.’ (54) ??Tot waar toe Jan gerend heeft, (van) daar (af) begin ik te rennen till where to Jan run.prtc has of there from begin I to run.inf ‘I will begin to run from the place where Jan has run to.’ The problem is that the locative tot waar toe-clause denotes a path and as such it cannot be construed as the starting point which can be picked out by the main clause promominal in (54). As expected, a tot waar toe-clause cannot show up as subject of a nominal predicate denoting a place, either: (55) *??Tot waar toe Jan gerend heeft is een fijne plek. till where to Jan run has is a nice place ‘The place where Jan run to is a nice place.’ The behavior of Dutch thus parallels Hungarian, reinforcing the conclusion above that it is the lexico-semantic property of certain connectives that determines the availability of matching in the locative domain. Given that Dutch does not evidently use correlative structures for the expression of all adverbial clauses (the examples in (52) have a different underlying structure for example), these facts show that matching effects in temporal/locative multi-clausal structures extend beyond correlative constructions.

 Rajesh Bhatt and Anikó Lipták

5. Summary This paper dealt with a particular kind of matching effect in the realm of correlative clauses that shows up in relativization over times and locations. Such a matching effect manifests itself in the fact that not all combinations of temporal/locative connectives are possible in the correlative and the main clause. To offer an explanation about this matching requirement and the variation it shows among two unrelated languages, Hindi-Urdu and Hungarian, we have reviewed properties of path and point denoting temporal and locative expressions and concluded that there are semantic constraints on the combination of clauses containing these in corerlatives. We have also shown that matching effects of this sort are not confined to correlative and headed relative constructions alone, but characterize other types of adverbial clauses as well.

Acknowledgements This paper is dedicated by the first author to the memory of Carlota Smith, whose support was essential for the writing up of the first version of this paper. Thanks are also due to Boban Arsenijević, Marcel den Dikken and István Kenesei for suggestions on the material presented in this article, and NWO (Netherlands Organization for Scientific Research) for material support, in the form of a postdoc grant and a visiting scholar grant. Any errors in this article are our own.

References Arsenijević, Boban. This volume. {Relative {conditional {correlative clauses}}}. Bhatt, Rajesh. 2003. Locality in Correlativization. Natural Language and Linguistic Theory 21 (3). 485– 541. Bresnan, Joan. & Jane Grimshaw. 1978. The syntax of free relatives in English. Linguistic Inquiry 9 (3). 331–391. Dayal, Veneeta. 1996. Locality in wh-quantification: Questions and relative clauses in Hindi. Studies in Linguistics and Philosophy 62. Dordrecht: Kluwer Academic Publishers. Fintel, Kai von & Sabine Iatridou. 2002. If and When If-Clauses can restrict quantifiers. Manuscript. MIT. Grimshaw, Jane. 1977. English Wh-Constructions and the Theory of Grammar. Doctoral dissertation, University of Massachusetts-Amherst. Amherst, MA: GLSA. Groos, Anneke. & Henk van Riemsdijk. 1981. Matching effects in Free Relatives: A Parameter of Core Grammar. In Adriana Belletti, Luciana Brandi & Luigi Rizzi (eds), Theory of Markedness in Generative Grammar: Proceedings of the IVth GLOW Conference. 171–216. Pisa: Scuola Normale Superiore di Pisa. Iatridou, Sabine, Elena Anagnostopoulou & Roumyana Izvorski. 2001. Some Observations about the Form and Meaning of the Perfect. In Michael Kenstowicz, & Ken Hale (eds), A Life in Language. 189–238. Cambridge, MA: MIT Press.



Matching effects in the temporal and locative domains  Keenan, Edward. 1985. Relative Clauses. In Timothy Shopen (ed.), Language typology and syntactic description, volume 2, 141–170. Cambridge, England: Cambridge University Press. Lipták, Anikó. 2006. Relativization strategies in temporal adjunct clauses. In Pierre Pica, Johan Rooryck & Jeroen van Craenenbroeck (eds), Linguistic Variation Yearbook 5, 133–185. Amsterdam: John Benjamins. Pandharipande, Rajeshwari V. 1997. Marathi: a descriptive grammar. London: Routledge. Rebuschi, Georges. This volume. Basque correlatives and their kin in the history of Northern Basque. Srivastav, Veneeta. 1991. The syntax and semantics of correlatives. Natural Language and Linguistic Theory 9 (4). 637–686.

Endnotes 1. Here and in the examples below, we use the kind of Hindi orthography that represents retroflexes by capitalization, nasal vowels by following the vowel by the capitalized nasal, and long vowels by the doubling of the vowel. The glosses are: acc: accusative; cp: conjunctive particle; dat: dative; erg: ergative; f: feminine; fut: future; gen: genitive; imp: imperfective; lbm: left boundary marker; nom: nominative, m: masculine; pfv: perfective; pl: plural; prs: present; pst: past; ptcp: participle; pv: preverbal particle; rbm: right boundary marker; sg: singular. In the Hungarian examples, we only indicate agreement morphemes when these are different from 3 person singular, present tense agreement. 2. Conditionals are an exception to this generalization. In conditionals, the relative phrase can be absent but the anaphoric marker is generally present. A further exceptionality of conditionals is that they do not reliably pattern with correlatives in all Indo-Aryan languages. In Hindi-Urdu, for example, the conditional marker is agar, which is not a relative pronoun. It is possible that conditionals are only diachronically related to correlatives in Hindi-Urdu. We will not consider conditionals further in this discussion. 3. cp in the gloss here stands for conjunctive participle, a term we have adopted from the descriptive literature on Hindi-Urdu. Ordinarily V-CP contributes a meaning along the lines of ‘having V-ed’. Here, however, ‘le-kar’ forms a fixed expression that optionally appears between a left boundary marker and a right boundary marker. See also example (12). 4. The multi-headed temporal and spatial correlatives shown below seem to be the only ones possible (next to cases where a temporal/spatial abstraction combines with an individual abstraction, not illustrated here). Other combinations lead to ungrammaticality. For example it is not possible to have two relativizations with a left boundary marker or two relativizations with a right boundary marker. Further we cannot have a multi-headed temporal/spatial correlative where one of the abstractions involves the bare/zero connective and the other involves a left/ right boundary marker. The fact that it is impossible to combine a bare connective abstraction with a left/right boundary abstraction might indicate that the way bare connective adverbials modify the matrix clause is distinct from the way left/right boundary marker adverbials modify the matrix clause. The intuition is that the former involve reference to points while the latter modify a temporal/spatial interval/path argument of the clause.

 Rajesh Bhatt and Anikó Lipták 5. Temporal correlatives in Hungarian have headed relative counterparts where the main clause demonstrative functions as the head. Matching restrictions apply to these headed relative structures as well, as the following show: (i) a. Akkor amikor János megjött, Mari TV-t nézett. that.at when János arrive.pst Mari TV-acc watch.pst ‘When John arrived, then Mari was watching TV.’ b. *Azóta amikor János megjött Mari TV-t néz. that.since when János arrive.pst Mari TV-acc watch ‘When John arrived, since then Mari is watching TV.’ c. *Addig amikor János megjött Mari TV-t nézett. that.till when János arrive.pst Mari TV-acc watch.pst ‘When John arrived, till then Mari was watching TV.’ The corresponding headed structures are marginal in Hindi. 6. We thank Boban Arsenijević for making us consider this option. 7. Examples like this can be constructed, but are not naturally occurring data. Ahonnantól is almost always used to refer to location in the abstract sense. 8. Like with temporal correlatives, locative correlatives have headed counterparts that display matching effects. (i) a.

Addig/odáig ameddig tegnap elfutottam, that.till/ there.to.till what.till yesterday pv.run.pst.1sg ma kocsival mentem. today car.with go.pst.1sg ‘To the place where I ran yesterday, I went by car today.’

b. *Ott ameddig tegnap elfutottam, van egy nagy fa. there what.till yesterday pv.run.pst.1sg be a big tree ‘The place to where I ran yesterday, there is a big tree there.’ c. *Onnantól ameddig tegnap futottam, ma there. from.from what.till yesterday run.pst.1sg today tovább mentem. further go.pst.1sg ‘The place to where I ran yesterday, from there I went further today.’ The non-matching headed structures seem to be worse than their locative correlative counterparts. 9. It needs to be noted that akkortól and amikortól most usually occur in the headed pattern, but this does not affect our point here, as temporal relatives of the headed type also show matching effects (see footnotes 5 and 8 above). 10. We thank Marcel den Dikken for providing us with the Dutch examples in this section, as well as for calling our attention to the relevance of examples (50)- (51) for our theory of the lack of matching effects in locatives.

Index A across-the-board extraction (ATB)  45, 60, 97, 329, 333 adjacency  17, 108, 271, 275, 292–293 adjunction  12, 13, 15–19, 35–37, 206–214, 276, 283–285, 318 of a correlative clause to CP  12, 229–230, 270–276 of a correlative clause to IP  12, 229–230, 270–276 agreement  161, 163, 169, 213–215, 222, 251–255, 277–279, 290–292, 324 American Sign Language  160, 162, 171, 177, 189 A'-movement/dependency  247, 269, 277–284, 304 anacoluthon  274 anaphor (binding)  63–66, 167, 169–170 Anderson, Corinna  12, 44, 321, 322 Andrews, Avery  25, 27, 46, 228 anti-reconstruction effect  63– 67, 70 apodosis  2, 82–83, 120 approaches to correlatives uniformity approaches  13–15 non-uniformity approaches  15–18 high adjunction & binding approach  15–18, 206–207, 210–213 low adjunction & movement  16–18, 206–209 high adjunction & correlate raising  18 aspect  159–160, 356–357 asymmetric adjunction  229–245 B Bach, Emmon  15, 21 Bambara  82, 312

base-generation (of correlative clauses)  15–16, 248–250, 318–322 Basque  45, 81–129 Berman, Howard  11, 309 Bhatt, Rajesh  5–6, 15–18, 27–28, 68, 147, 206–216, 246–250, 254–256, 318–321 Bianchi, Valentina  78, 174, 338 Bittner, Maria  28, 118, 141–142, 146–148, 151 Borsley, Robert  45, 59, 72, 77 Bresnan, Joan  71, 352 Burushaski  45, 82–83 C Caponigro, Ivano  101, 128 cardinality implication  63 case hierarchy  72, 79 case marking  230 case matching  72–73, 201, 532–353 c-command  14, 229–230, 240–245 centering theory  28, 146, 156 Chain Uniformity see Principle of Unambiguous Binding Cheng, Lisa L.-S.  29, 141–142, 147 Chomsky, Noam  226, 252, 277, 325, 327 coindexation  249, 253, 317–318 comparative clause  344–345 comparative correlative  11, 18–21, 53, 58, 64, 66 comparative phrase  20, 21, 45, 264 complex NP island  15, 254–255 conditional  26–29, 110–119, 204–205, 271, 303 Condition on Extraction Domains  322 Condition on Local Merge  213– 216, 320–322

conjunction  97, 231, 302 conservativity  136 Cooper, Robin  11, 15, 21 coordination  102, 208, 214, 324, 326 coordinator  45, 208 correlative subordinator  1, 82 correlative particle  263–264, 276, 285–287, 305 Culicover, Peter  11, 19–20, 264, 303 D Dasgupta, Probal  11, 15, 241 Dayal, Veneeta  1–7, 8, 12, 15–16, 17, 22–27, 200, 225–226, 248, 318, 362 see also Srivastav degree expression  20–21 demonstrative requirement (in correlative constructions)  2, 4–5, 248–249, 313–314 determiner  8, 24–25, 162, 169, 174–176, 248, 347 den Dikken, Marcel  11, 12, 19–21, 57, 339 dimensionality  368 donkey anaphora  29, 95, 105, 136–137, 167, 204–205, 216 doubling  325–328 Downing, Bruce  7, 10, 15, 168, 197 Dutch  11, 12, 20, 37, 263–306 Dwivedi, Veena Dhar  12, 340 E ellipsis  82, 184 English  2,3, 11, 18–21, 62–63, 178–184, 201, 263–265, 273, 275, 277 EPP  256, 291, 325 equative clause  345–346 eta (in Basque)  95–97, 99, 108, 119–120, 302 existential context  103

 Correlatives Cross-Linguistically extraction  19, 20, 59–63, 67–71, 77, 228, 266–269, 277–279, 283–284, 288–291 extraposed relative  5–8, 13, 44, 90, 166–178, 190, 245, 314, 316 extreme non-specificity  137– 143, 147–148, 150, 153 F Farkas, Donka  137–138 feature checking  251, 325 finiteness  109, 265–267, 284–285 von Fintel, Kai  27, 364 floating quantifier  277, 328 focus movement  18, 262, 272, 323 free choice  22–24, 29, 104, 139, 150, 332–333 G Gajewski, Jon  24, 26–27, 46 Garrett, Andrew  11, 13 Geis, Michael L.  12, 27 generalized quantifier  15, 22–26, 167, 206, 255 generic interpretation  22–24, 28, 89, 110, 135–136, 141, 150, 224 German  20, 72, 79, 134, 190, 274, 279, 301, 352 German Sign Language  171 Grimshaw, Jane  71, 340, 352 Groos, Anneke  51, 219, 340, 352 Grosu, Alexander  8–10, 24–26, 44, 61–63, 101, 237, 243, 247, 253 H Hale, Mark  28, 219, 236, 309 Haudry, Jean  11, 82 head-complement configuration  325–326 head movement  174–176 Hettrich, Heinrich  227, 228 Hindi(-Urdu)  1–10, 12–29, 206–216; 223–262, 343–372 Hittite  11, 13, 310 Hock, Hans H.  228–229, 231, 242, 246, 249, 260 Huang, C.-T. James  29, 141–142, 147, 322 Hungarian  13, 18, 23, 29, 156, 221–222, 277–278, 281–282, 304, 316, 321, 324, 343–372

I Iatridou, Sabine  221, 273, 364 indefinites  4, 6, 14, 25, 29, 46, 204, 248–249, 277, 314 Indo-European  1, 11, 82, 231 intentional verb  183 interval  152, 363–369 Italian  62–63, 101 Izvorski Roumyana  15, 18, 27, 52, 57, 103, 145, 197, 322, 332 see also Pancheva J Jackendoff, Ray  11, 19–20, 264, 303 Jacobson, Pauline  22, 62, 200, 239, 244, 260 Japanese  175–177, 180, 190 Johannessen, Bondi J.  1 K Kayne, Richard S.  14, 67, 78, 174, 246, 328 Keenan, Edward L.  7, 8, 10, 15, 197, 343 É. Kiss, Katalin  85, 322 L lambda-abstraction  21, 253 Landman, Fred  8–10, 24–26, 44 left boundary  348–349, 361, 364–366, 371 left branch extraction  59–63, 66–73, 77 left dislocation  12–13 left periphery  13, 99, 109, 148, 190, 236 Lehmann, Christian  7, 11, 15, 45 locality  16–18, 19, 213–216, 263–306, 320–321 locative clause  89, 343–372

M

Mahajan, Anoop  5, 6, 12, 14–15, 246, 261 Marlow, Patrick  261 maximalization  8–10, 22, 24–27, 44, 346–347, 363–365 McCawley, James D.  45, 182, 228, 229, 241, 250, 303, 340 Merge  171, 174, 213–216, 252–255 minimalist program  326 multiple correlative  17–18, 25–27, 314–320, 328–329, 339, 350

N negation  159–162, 182, 189 negative polarity item  83, 120 Neidle, Carol  160, 162, 171 Nepali  12–13 ‘no matter’ clause see unconditional nominalization  174–177 non-manual marking (NMM)  159, 162, 164, 165, 171, 177, 189 Nunes, Jairo  327 O ordinal  168, 183 OV language  10, 190 P Pancheva, Roumyana  27–28, 271 see also Izvorski Partee, Barbara  21, 120 path (vs. point)  360, 362, 368–370, 371 Pfau, Roland  162, 171, 178 pied piping  150, 184, 271, 282 point (vs. path)  360, 362, 368–370, 371 Polish  49–79, 330–331, 332–333 possible worlds  133–137, 140–143, 146–147 predicate  26–27, 248, 251, 253, 368 Principle C  63–73, 78, 211, 213, 221, 247 Principle of Unambiguous Binding (PUB)  279–293 Principle of Minimal Compliance  290 Probe  327, 329 pro-drop  167, 170, 331 promotion analysis (of relativization)  14, 67–70, 78, 174 proper name (as head of a relative clause)  182 protasis  2, 82 Q quantificational variability effect  22 quantifier raising  247–250 Quirk, Randolph  1, 82 R raised eyebrows  162, 163



Index  raising analysis (of relativization) see promotion analysis reconstruction  16, 63–73, 172 relative clause amount relative  8–10 appositive relative  8, 90–94, 125, 178–184, 224–225 degree relative see amount relative free relative  8, 22–24, 28, 94–97, 140, 156, 190, 204–205, 281–282, 304, 305, 332, 352–353 see also indefinite free relative headed relative  2–29, 67–72, 174, 246, 251, 281–282, 361, 363, 371 indefinite free relative  101– 103, 108–109 internally headed relative  7–8, 44, 174–178 postnominal relative  3, 7, 44 prenominal relative  7, 44  restrictive relative  8–10, 25, 44, 87, 126, 178–184, 224, 226, 241–243, 249–251, 251–256 ‘third kind’ relative  8 van Riemsdijk, Henk  51, 338, 340, 352 right boundary  348–349, 361, 364–366, 371 right periphery  5–7, 15, 44

Rizzi, Luigi  110, 120, 233, 235–236, 251, 327 Romance languages  90 Russian  21, 101, 314 Medieval  10 S Sanskrit (Vedic)  1, 10, 223–262 scrambling  14, 16, 246, 279, 279, 305 sentential adverb  181 Serbo-Croatian  10, 76, 131–156, 316 Shimoyama, Junko  175–177, 180, 190 sign language  157–191 SignStream  158, 188 since-clause  344–372 spec-head configuration  325 Sportiche, Dominique  170, 277, 304, 328 Srivastav, Veneeta  1–7, 261, 264, 171, 314 (see also Dayal) stacking (of relative clauses)  10, 25, 45, 237–240, 243–244, 247–248, 259 Subbarao, Karumuri  13, 223, 261, 314 subordination  27, 228–237 superiority  57–59, 76, 330–331 symmetric adjunction  229–245 symmetric reading  28–29, 141–142, 147

T temporal clause  347–372 tensed eyes  171, 172 then-clause  344–372 Tibetan (Lhasa)  27, 194–222 topic  13, 119, 137–152, 281, 304 topic interpretation of correlatives  13, 44 typology (of relativization)  7–10 U Universal Grammar (UG)  19– 20, 268, 292–293 unconditional clause  103–105, 108–110, 121–122 uniqueness  22–24, 26 until-clause  344–372 V verb second (V2)  267–277, 286 variable bound by the correlative  15, 24 Vergnaud, Jean-Roger  325 de Vries, Mark  7, 10, 11, 305 W Wali, Kashi  13, 17, 316 weak existential commitment  137–138 when-clause  344–372 WorldP  135–136, 138–141, 145–148 Y yes-no question  133 yes-no relative  132–137

In the Language Faculty and Beyond series the following titles have been published thus far or are scheduled for publication: 2 1

Aroui, Jean-Louis and Andy Arleo (eds.): Towards a Typology of Poetic Forms. From language to metrics and beyond. xiv, 401 pp. + index. Expected September 2009 Lipták, Anikó (ed.): Correlatives Cross-Linguistically. 2009. vii, 375 pp.