YEARBOOK OF MORPHOLOGY 2003
Yearbook of Morphology Editors:
Geert Booij Jaap van Marle
Consulting Editors:
Stephen Anderson (Yale) Mark Aronoff (Stony Brook, N.Y.) Mark Baker (New Brunswick, N.J.) Laurie Bauer (Wellington) Rudie Botha (Stellenbosch) Joan Bybee (Albuquerque, N.M.) Andrew Carstairs-McCarthy (Christchurch) Greville Corbett (Guildford, U.K.) Wolfgang Dressler (Wien) Martin Haspelmath (Leipzig) Jack Hoeksema (Groningen) Rochelle Lieber (Durham, N.H.) Peter Matthews (Cambridge, U.K.) Franz Rainer (Wien) Sergio Scalise (Bologna) Henk Schultink (Utrecht) Andrew Spencer (Colchester, U.K.)
Editorial address:
Editors, Yearbook of Morphology Faculteit der Letteren, Vrije Universiteit De Boelelaan 1105 1081 HV Amsterdam, The Netherlands e-mail:
[email protected]
YEARBOOK OF MORPHOLOGY 2003 Edited by
GEERT BOOIJ Vrije Universiteit, Nederland Amsterdam, The Netherlands
and
JAAP VAN MARLE Open Universiteit, Nederland Heerlen, The Netherlands
KLUWER ACADEMIC PUBLISHERS NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW
eBook ISBN: Print ISBN:
1-4020-1513-5 1-4020-1272-1
©2002 Kluwer Academic Publishers New York, Boston, Dordrecht, London, Moscow Print ©2003 Kluwer Academic Publishers Dordrecht All rights reserved No part of this eBook may be reproduced or transmitted in any form or by any means, electronic, mechanical, recording, or otherwise, without written consent from the Publisher Created in the United States of America Visit Kluwer Online at: and Kluwer's eBookstore at:
http://kluweronline.com http://ebooks.kluweronline.com
Table of Contents
Preverbs ((guest) editors: Geert Booij and Ans van Kemenade) GEERT BOOIJ and ANS VAN KEMENADE / Preverbs: An introduction FARRELL ACKERMAN / Aspectual contrasts and lexeme derivation in Estonian: A realization-based morphological perspective MONIQUE DUFRESNE, FERNANDE DUPUIS and MIREILLE TREMBLAY / Preverbs and particles in Old French ALICE HARRIS / Preverbs and their origins in Georgian and Udi ANS VAN KEMENADE and BETTELOU LOS / Particles and prefixes in Dutch and English ANDREW MCINTYRE / Preverbs, argument linking and verb semantics: Germanic prefixes and particles EVA SCHULTZE-BERNDT / Preverbs as an open word class in Northern Australian languages: synchronic and diachronic correlates JOCHEN ZELLER / Moved preverbs in German: displaced or misplaced?
1
13
33 61 79
119
145 179
Other articles MARCO BARONI / Distribution-driven morpheme discovery: a computational/ experimental study CAROL FEHRINGER / Morphological ‘gangs’: constraints on paradigmatic relations in analogical change
213
249
Book reviews Jochen Zeller (2001), Particle verbs and local domains (Geert Booij)
273
Morphology 2000. Selected Papers from the 9th Morphology Meeting, Vienna, 24–28 February 2000 edited by S. Bendjaballah et al. (Geert Booij)
277
This page intentionally left blank
Preverbs: an introduction GEERT BOOIJ and ANS VAN KEMENADE The notion ‘preverb’ is a traditional descriptive notion in Indo-European linguistics.1 It refers to morphemes that appear in front of a verb, and which form a close semantic unit with that verb. In many cases, the morpheme that functions as a preverb can also function without a preverbal context, often as an adverb or an adposition. Most linguists use the notion ‘preverb’ as a cover term for preverbal words and preverbal prefixes. The preverb may be separated from the verb whilst retaining its close cohesion with the verb, which is called ‘tmesis’. It may also develop into a bound morpheme, that is, a prefix inseparable from the verb, with concomitant reduction of phonological form in some cases. If the preverb has become a real prefix, we may use the more specific notion of ‘complex verb’, whereas we take the notion ‘complex predicate’ to refer generally to multi-morphemic expressions with verbal valency. That is, we make a terminological distinction between complex predicates and complex verbs. The latter are multi-morphemic, but behave as single grammatical words. For both complex predicates in general (cf. Spencer 1991, Ackerman and Webelhuth 1998) and complex verbs (cf. Miller 1993) in particular, the question has been raised how and where in the grammar they should be accounted for. Well-known examples of complex predicates are auxiliary-verb sequences, serial verb constructions, the coverb-verb combinations as in Australian languages (Schultze-Berndt, this volume), similar light verb constructions in other languages, and verb raising constructions in Germanic languages. These different types of complex predicates represent various kinds of mismatches in the syntactic and morphological coding of complex events and verbal valency, and thereby challenge our view of the architecture of the grammar, and the relation between syntax, morphology, and the lexicon. Complex predicates of the preverb-verb type occur in most European languages, both the Indo-European languages (Watkins 1963, 1964) and those of the Finno-Ugric family (Ackerman and Webelhuth 1998, Ackerman (this volume), and in Georgian and Caucasian languages (Harris, this volume). A number of mostly descriptive articles on preverbs in the languages of Europe can be found in Rousseau (ed., 1995). In particular, particle verbs in Germanic languages have received a lot of attention in the recent literature (Ackerman and Webelhuth 1998, Lu¨deling 2001, McIntyre 2000, 2001, 2002, this volume), Booij 2002a;b, Dehe´ and Wanner (eds.) 2001, Dehe´ et al. (eds.) 2002, Zeller 2001; this volume, van Kemenade and Los, this volume, and references in these publications). The history of particles and prefixes in Latin and French is discussed in Vincent (1999), and Dufresne et al. (this volume) respectively. It is the aim of the collection of articles in this thematic section of the Yearbook of Morphology on preverbs to provide in-depth empirical investigations of preverbs in a number of typologically diverse languages and to discuss Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 1–11. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
2
Geert Booij and Ans van Kemenade
the consequences of their behaviour for a proper theory of the architecture of the grammar. It is striking indeed that this phenomenon is widely attested crosslinguistically, which suggests that the grammatical and historical mechanisms responsible for the rise of a class of preverbs are universal. In many cases, the development of preverbs and prefixes represents a clear case of grammaticalization, and this thematic section therefore also focuses on the diachrony of preverb constructions. For the preverb situation in Indo-European, Kuryłowicz (1964) and Watkins (1964) remain the authoritative sources. In the early stages, preverbs seem to have been independent constituents. Kuryłowicz notes that, since in many of the daughter languages preverbs behave both as preverbs and as prepositions, it is thought that the origin of both preverbs and prepositions is adverbial (cf. also Baldi 1979). The basis for the divergence in word class in the daughter languages is in the potential for variation between various syntactic modification relations. When a particle appeared with a transitive verb, it was ambiguous between a modifier of the verb (in which case it was interpreted as an adverb) and a modifier of the object (and was interpreted as a preposition/predicate). In addition, the particle could modify other adverbs and be positioned accordingly. For a list of preverbs with cognates in the various languages, the reader is referred to Beekes (1995). Kuryłowicz (1964) gives a brief discussion of some developments in the early Indo-European languages. According to Watkins, preverbs could appear in two basic positions in Sanskrit: a sentence-final one left of the verb it modifies, which is called the contact position and is exemplified in (1); and a sentence-initial one where it is not adjacent to the verb, which is illustrated in (2). This latter position of the preverb in which is does not precede the verb directly is called tmesis. The examples are from Delbru¨ck (1893–1900): (1) # ... P V# dasvasam upa gachatam (RV I, 47, 3) worshipper to come ‘come to the worshipper’ (2) #P ... V# ati tr¸s¸t¸am vavaks¸ita (RV III, 9, 3) ‘you have grown beyond the harmful (smoke)’ 2 Preverb and verb are thought to be a kind of syntactic unit. The argument for this comes from the fact that the preverb is stressed only in main clauses (as in (3), where stress is marked by an acute accent), while in subordinate clauses, in the position preceding the verb, stress shifted to the verb, as in (4). The examples are again from Sanskrit:
Preverbs: an introduction
3
(3) pra´ gacchati (he) forth goes ‘he goes forth’ (4) ya´h pra ga´cchati who forth goes ‘who goes forth’ This stress shift is thought to be the result of what Watkins calls ‘univerbation’, resulting in a syntactic unit. According to Kuryłowicz, a consequence of this univerbation was either that the verb was encliticized to the preverb (as in Sanskrit and Greek), or that the preverb was procliticized to the verb (as in Old Irish, Germanic and Balto-Slavic). In the daughter families/languages, the preverb maintains its status as an independent constituent for quite a long time in some cases, while in others it follows various stages in a classical grammaticalization path from preverb>prefix>ultimate disappearance (cf. also Pinault 1995). As cases in point, we can cite here some developments in Romance and Germanic respectively (see Dufresne et al. (this volume), and van Kemenade and Los (this volume)). Vincent (1999) discusses some interesting cases in Latin from which it is clear that, while in the early Latin prayers, preverbs/prepositions must be assumed to have independent constituent status, they later become members of compound verb stems, later developing into prefixes. This applies to the following words: (5) sub ‘under’; trans ‘across; in ‘in’; ab ‘from’; ob ‘against’; cum ‘with’; ex ‘out of’; pro ‘for’ To contrast the two stages, consider the following examples of Latin preverbs (Vincent 1999: 1118): the grammarian Festus makes two remarks on the language of the early prayers: (6)
a.
Sub vos placo, in precibus fere cum dicitur, significat id, quod supplico ‘when people say, mostly in prayers, sub vos placo, it means the same as supplico’
b.
ob vos sacro, in quibusdam precationibus est, pro vos obsecro, ut sub vos placo, pro supplico ‘ob vos sacro in certain prayers stands for vos obsecro, just as sub vos placo stands for supplico’
What seems to be the case here is that the preverb in the early prayers is in tmesis, with the personal pronoun encliticized to it by the Wackernagel effect. This indicates that the preverb is an independent constituent in first constituent
4
Geert Booij and Ans van Kemenade
position. The same preverbs form part of compound verb stems in Classical Latin and later become prefixes, as in: (7) submittere ‘to put underneath’; permittere ‘to let through’; transmittere ‘to send across’; transferre ‘to carry across’, perferre ‘to carry through’; obligare ‘to bind’ A similar phenomenon can be observed in Gothic, where the aspectual preverb ga occurs in first constituent position with sentence particles encliticized to it: (8) ga-u-hva-sehvi ga – wh particle – anything – saw ‘whether he saw anything’ (Mark, VII, 23) This preverb is attested in the Old West-Germanic languages as the past participle prefix ge-, which disappeared in English but is still widely used in presentday Dutch and German. It is thought to be cognate with Latin cum, and would thus be clearly a locative or circumstantial item in origin (although this is not uncontroversial). Phenomena parallel to the preverb-enclitic pronoun/particle ... V pattern in (6) and (8) have been observed in Hittite and Old Irish (Hopper 1975). These patterns represent instances of preverbs that follow a grammaticalization path as in (9) (9) independent preverb>left member of verbal compound >prefix>(zero). A different type of development seems to be represented by the preverb system that is still very productive in the present-day Germanic languages, in particular in West-Germanic. In the older stages of these languages, there is still a clear differentiation of word class status between adverb and preposition, as observed for Indo-European by Kuryłowicz (1964). For instance, Hiltunen (1983) makes a distinction for Old English between those phrasal adverbs that cannot occur as prepositions and include adun ‘down’; aweg ‘away; forð ‘forth’; niðer ‘down; up ‘up’; ut ‘out’, and prepositional adverbs, which can be used as either preposition or adverb and include beforan ‘before’; æfter ‘after’; to ‘to’; ofer ‘over’; ongean ‘toward’. It is probably fair to say that this differentiation lives on to a certain extent into the present-day language. A similar differentiation is suggested by studies on the early stages of other Germanic languages such as Eytho´rsson (1995) and Ferraresi (1997) on Gothic. An appropriate term for the preverb-verb combination in these languages is: separable complex verb, since this abstracts from the divergent syntactic development that took place mostly during the recorded history (as discussed in van Kemenade and Los, this
Preverbs: an introduction
5
volume). When we consider the history of the West-Germanic languages in particular, it is especially striking that the old preverb-verb system was regimented anew as a syntactically circumscribed and often lexicalized system of aspectual marking. During this process, it became immensely productive, as the very lively recent history of Dutch and German separable complex verbs and English phrasal verbs testify. The analysis of separable complex verbs in Dutch and German has been a focus of interest in the recent literature on preverbs (cf. the references given above), and it is therefore appropriate to provide the reader with some essential discussion concerning the analytical and theoretical issues involved. Preverbs in Modern Dutch and German are quite similar in their behaviour. Most of them derive from adpositions and adverbs. In addition, there are some nouns and adjectives that pattern in the same way as preverbs, in the sense that the N-V or A-V combination behaves as a separable complex verb. Preverbverb sequences in these languages differ from prefixed verbs and verbal compounds in that the preverb is separable from the verb. Dutch and German have two different word orders, XvSOV in main clauses (where v stands for the finite verb), and SOV in embedded clauses. This difference in word order has the effect that preverbs can be stranded at the end of the main clause, as a result of finite verb movement to second constituent position of the verbal part of the separable verb complex. Let us first illustrate the separability of the preverbs with some examples from Dutch (Booij 2002a: 205): (10) Verb-final clause
Verb-second clause
Hans zijn moeder opbelde
Hans belde zijn moeder op ‘Hans phoned his mother’ de fietser neerstortte De fietser stortte neer ‘The cyclist fell down’ Jan het huis schoonmaakte Jan maakte het huis schoon ‘John cleaned the house’ Rebecca pianospeelde Rebecca speelde piano ‘Rebecca played the piano’ dit resultaat ons teleurstelde Dit resultaat stelde ons teleur ‘This result disappointed us’ In the first example, the word op ‘up’ that combines with the verb, is also used as an adposition. In that case, the non-verbal element is also referred to as a particle, and the combination is referred to as a particle verb. Particle verbs form a productive class of separable complex verbs (SCVs). In the second example, the word neer ‘down’ is also used as an adverb. The next two examples show that adjectives (like schoon) and nouns (like huis) can also occur in SCVs. In the last example, the word teleur ‘sad’ does not occur as an independent
6
Geert Booij and Ans van Kemenade
word. The fact that SCVs are felt as word-like units is reflected by Dutch orthography, which requires SCVs to be written as one word, without internal spacing, if the two constituents are adjacent. The separability of SCVs also manifests itself in the position of the infinitival particle te that occurs between the two constituents of SCVs, as in op te bellen, and in the form of the perfect/passive participle, with the prefix ge- in between the particle and the verbal stem: op-ge-beld. In derivational morphology, SCVs behave similarly; for instance, the ge-nominalisation of opbellen is opgebel, with the prefix in between the particle and the verbal stem. A number of these particles correspond to bound morphemes with an identical phonological form; these are real prefixes that cannot be separated from the verbal stem. These prefixed verbs carry main stress on the verbal stem, not on the prefix, whereas the SCVs carry main stress on the non-verbal constituent. Thus we get minimal pairs like the following: (11) SCV do´orboor ‘to go on drilling’ o´mblaas ‘to blow over’ o´nderga ‘to go down’ o´verkom ‘to come over’ vo´orkom ‘to occur’
prefixed verb doorbo´or ‘to perforate’ ombla´as ‘to blow around’ onderga´ ‘to undergo’ overko´m ‘to happen to’ voorko´m ‘to prevent’
Similar facts can be cited for German (Lu¨deling 2001): the German preverbs can be stranded and they can be separated from the verb by means of zu ‘to’ and by the participial prefix ge-. Like phrasal verbs in English (cf. Brinton 1988), the meaning of the preverb-verb combination (PV-V) in Dutch and German is often not fully predictable, and this implies that at least these combinations are lexical units of some sort. Typically, the preverbs contribute to the aspectual properties of the PV-V, in particular lexical aspect (Aktionsart) such as telicity or duration, and thus they may also influence the syntactic valency of the verb. For instance, the Dutch verb lopen ‘to walk’ is intransitive, whereas the SCV aflopen can be used as a transitive verb, as in the VP de straten aflopen ‘to tramp the streets’. In this respect, preverbs are quite similar to some of the verbal prefixes that similarly influence the aspectual and syntactic properties of a verb, as illustrated below. A second domain in which the unitary character of the PV-V combination manifests itself, is that of word formation: PV-Vs can feed word formation, both compounding and derivation., as illustrated by the following examples from Dutch with SCVs in the left column (from Booij 2002a: 209): (12) deverbal suffixation aanbied ‘to offer’
aanbieder ‘offerer’, aanbieding ‘offer’
Preverbs: an introduction
deverbal prefixation: invoer ‘to introduce’ uitgeef ‘to publish’
7
herinvoer ‘to reintroduce’ heruitgeef ‘to republish’
compounding with verbal left constituent: doorkies ‘to dial through’ doorkiesnummer ‘direct number’ doorkijk ‘to see through’ doorkijkbloes ‘lit. see through blouse, transparent blouse’ PV-V sequences form a challenge for our view of the relation between syntax and morphology. On the one hand, PV and V do not form a syntactic atom, as is clear from their separability in various syntactically defined contexts. Yet, their behaviour is similar to that of complex morphologically derived verbs in the sense that they form lexical units of some sort, expressing aspectual notions and having derivational effects like affecting the valency of the verb. Thus, the transitivizing effect of the particle af in aflopen as discussed above, competes with a bound prefix such as be- which has a similarly transitivizing effect. This is illustrated by the following examples: (13) op straat lopen intransitive (lit. on street walk) ‘walk in the street’ de straten aflopen separable, transitive (lit. the street off-walk) ‘roam the streets’ de straat belopen inseparable, transitive (lit. the street be-walk) ‘walk the streets’ These facts raise some intriguing questions with respect to the question of how to model the relation between syntax and morphology. It seems appropriate to view the semantically transparent cases of preverbverb as syntactically defined cases of secondary predication (we would restrict analyses along such lines, e.g. Neeleman (1994), Den Dikken (1995) to the semantically transparent cases). The facts from the history of English as discussed by van Kemenade and Los (this volume) suggest that the secondary predicate configuration may well be the historical origin of the construction. They also suggest, however, that there are many cases in Dutch and German, and probably even more in English, where the phrasal combination is not (no longer) semantically transparent. In addition to this, the construction is immensely productive in all three languages. The particles in these very productive constructions may have highly specific meanings that do not correlate regularly with the range of meanings of the same word when used in nonpreverbal contexts. Therefore, PV-V combinations call for a definition in syntactic as well as lexical terms. It is this that provides us with the possibility of interpreting PV-V combinations as derivationally related to the verbal part. Since PV-V combinations express various aspectual notions, and have gained in
8
Geert Booij and Ans van Kemenade
frequency at the expense of the older bound aspectual prefixes, we might think of them in terms of a derivational type of periphrasis. Are we justified in extending the notion ‘periphrasis’ to word formation? Let us point out that, at a more general level, there are good arguments for locating certain syntactic patterns in the lexicon, although they are productive. These are the so-called constructional idioms or idiomatic patterns, syntactic constructions formed according to the syntactic rules of the language, but with a specific meaning that cannot be derived compositionally. These are the kinds of configurations that are the focus of interest of Construction Grammar (Goldberg 1995). Periphrasis may then be seen as a specific subcategory of construction, since the periphrastic forms compete directly with synthetic morphological forms: they must be used instead of a synthetic form for the expression of specific kinds of information (Booij 2002c). This is clearly the case for inflectional periphrasis. Is it also the case for PV-Vs: do they compete with derivational morphology? Indeed, in languages such as German and Dutch the use of affixes to form derived verbs is very restricted. Dutch has only one productive verbalizing suffix, the suffix -iseer that is exclusively attached to non-native stems. The only more or less productive verbalizing prefixes are be-, ver-, and ont- ‘de-’. These prefixes are on the whole incompatible with the use of a particle. In other words, it appears that preverbs have taken over the function of verbalizing prefixes. A very telling detail in this respect is that preverbs are also employed to derive a verbal stem from a nominal or adjectival root: the PV-V opleuken ‘to embellish’ is a combination of the preverb op and the adjective leuk ‘nice’; the verb leuken does not exist by itself. Another example is the SCV uithuwelijken ‘to marry off’, which is a combination of the preverb uit and the noun huwelijk ‘marriage’, used as a verb. Again, there is no verb huwelijken in Dutch. In the terms of Bybee et al. (1994), the overall development of PV-V combinations in the West-Germanic languages represents a good example of a grammaticalization development. In the older system aspectual bound prefixes loose their aspectual function (this is particularly clearly the case with ge-, which first became a past participle marker, was grammaticalized as such, and retained this status in Dutch and German while it was lost altogether in English). This function is then taken over on a large (and on the face of it, increasing) scale by the aspectual particles, which are bounders in the sense of Bybee et. al. (1994). In a general sense, the development seems to warrant quite clearly the notion of derivational periphrasis introduced above. Let us conclude with some remarks on the types of grammaticalization that we seem to be looking at here. If the above suggestions are correct, they add to the evidence for grammaticalization in the lexical/derivational domain. One scenario here is the regimentation of formerly autonomous preverbs to mark aspect periphrastically, following on the weakening and/or loss of aspectual
Preverbs: an introduction
9
bound prefixes. Both scenarios seem to involve at least one of the core characteristics of ‘grammaticalization processes’: semantic bleaching as evidenced in the case of the West-Germanic SCVs by the rise of metaphorical and idiomatic meanings for the PV-V combination. We would suggest, however, that the details of the historical developments show up a delicate interplay of independent syntactic and morphological (derivational as well as inflectional) developments, which, if given close scrutiny, may add considerably to our insight in the balance between syntax, morphology and the lexicon. The articles on preverbs in this thematic section of the Yearbook of Morphology are revised versions of papers selected from those presented at a workshop on preverbs at the Catholic University of Nijmegen in January 2001, organized by Ans van Kemenade in cooperation with Nigel Vincent and Geert Booij. They demonstrate that the systems of preverbs that are familiar to us in some of the Indo-European languages as sketched here, and the historical developments giving rise to systems of preverbs, are parallelled in languages as diverse as Caucasian languages and Northern Australian languages. This serves to further underline the challenge that preverbs pose to our views of the organization of the grammar, in particular the relation between syntax and morphology. The contributions in this collection rise to this challenge in a variety of ways: by extending our empirical basis; by suggesting various ways of modelling the relationship between morphology and syntax as instantiated by the preverb problem; and by showing how a diachronic perspective will help us to understand their behaviour.
NOTES 1
We would like to thank Nigel Vincent for his helpful comments on an earler version of this paper. 2 We are grateful to Mark Hale for his help with the translation of this example.
REFERENCES Ackerman, Farrell (this volume). Aspectual contrasts and lexeme derivation in Estonian: A realization-based morphological perspective. Ackerman, Farrell and Webelhuth, Gerth (1998). A Theory of Complex Predicates. Stanford: CSLI. Baldi, Philip (1979). Typology and the Indo-European prepositions. Indogermanische Forschungen 84, 49–61. Beekes, Robert (1995). Comparative Indo-European Linguistics, an Introduction. Amsterdam etc.: Benjamins. Booij, Geert (2002a). The Morphology of Dutch. Oxford: Oxford University Press.
10
Geert Booij and Ans van Kemenade
Booij, Geert (2002b). Separable complex verbs in Dutch, a case of periphrastic word formation. In Dehe´ et al. (eds.), 21–42. Booij, Geert (2002c). Constructional idioms, morphology, and the Dutch lexicon. Journal of Germanic Linguistics 14, 301–329. Brinton, Laurel (1988). The Development of English Aspectual Systems: Aspectualizers and Post-verbal Particles. Cambridge: Cambridge University Press. Bybee, Joan, Revere Perkins, and William Pagliuca (1994). The Evolution of Grammar. Tense, Aspect, and Modality in the Languages of the World. Chicago and London: Chicago University Press. Dehe´, Nicole and Dieter Wanner (eds., 2001), Structural Aspects of Semantically Complex Verbs. Berlin etc.: Peter Lang. Dehe´, Nicole, Ray Jackendoff, Andrew McIntyre and Silke Urban (eds., 2002). Verb Particle Explorations. Berlin: Mouton de Gruyter. Delbru¨ck, Berthold (1983–1900). Vergleichende Syntax der Indogermanischen Sprachen, 3 vols. Strassburg: Tru¨bner. Dikken, Marcel den (1995). Particles: on the Syntax of Verb-Particle, Triadic and Causative Constructions. New York: Oxford University Press. Dufresne, Monique et al. (this volume). Preverbs and particles in Old French. Eytho´rsson, Th. (1995). Verbal Syntax in the Early Germanic Languages. Diss. Cornell University. Ferraresi, G. (1997). Word Order and Phrase Structure in Gothic. Dissertation Stuttgart University. Goldberg, Ade`le (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Harris, Alice (this volume). Preverbs and their origins in Georgian and Udi. Hiltunen, R. (1983). The Decline of the Prefixes and the Beginnings of the English Phrasal Verb: The Evidence from some Old and Middle English Texts (Annales Universitatis Turkuensis, Series B, 160). Turun Yliopisto (University of Turku, Finland), Turku. Hopper, Paul (1975). The Syntax of the Simple Sentence in Proto-Germanic. The Hague: Monton. Kemenade, Ans van, and Bettelou Los (this volume). Particles and prefixes in Dutch and English. Kuryłowicz, Jerzy (1964). The Inflectional Categories of Indo-European. Heidelberg: Carl Winter Verlag. Lu¨deling, Anke (2001). On Particle Verbs and Similar Constructions in German. Stanford: CSLI. McIntyre, Andrew (2000). German Double Particles as Preverbs. Morphology and Conceptual Semantics. Tuebingen: Stauffenburg Verlag. [Studien zur deutschen Grammatik, 61] McIntyre, Andrew (2001). Argument blockages induced by verb particles in English and German: Event modification and secondary predication. In: Dehe´ and Wanner (eds.), 131–164. McIntyre, Andrew (2002). Idiosyncracy in particle verbs. In Dehe´ et al. (eds.), 95–118. McIntyre, Andrew (this volume). Preverbs, argument linking and verb semantics: Germanic prefixes and particles. Miller, D. Gary (1993). Complex Verb Formation. Amsterdam etc.: Benjamins. Neeleman, Ad (1994). Complex Predicates. OTS Dissertation Series, University of Utrecht. Pinault, George (1995). Le proble`me du pre´verbe en indo-europe´en, in Rousseau (e´d.), 35–59.
Preverbs: an introduction
11
Rousseau, Andre´ (e´d., 1995). Les pre´verbes dans les langues d’Europe. Introduction a` l’e´tude de la pre´verbation. Lille: Presses Universitaires du Septentrion. Schultze-Berndt, Eva (this volume). Preverbs as an open word class in northern Australian languages: synchronic and diachronic correlates. Spencer, Andrew (1991). Morphological Theory. Oxford: Blackwell. Vincent, Nigel (1999). The evolution of c-structure: prepositions and PPs from Indo-European to Romance. Linguistics 37, 1111–1154. Watkins, Calvert (1963). Preliminaries to a historical and comparative analysis of the syntax of the Old Irish verb. Celtica 6, 1–49. Watkins, Calvert (1964). Preliminaries to the reconstruction of Indo-European sentence structure. In Horace Lunt (ed.), Proceedings of the 9th International Congress of Linguists The Hague: Mouton, 1035–1042. Zeller, Jochen (2001). Particle Verbs and Local Domains. Amsterdam etc.: Benjamins. Zeller, Jochen (this volume). Moved preverbs in German: displaced or misplaced?
(Geert Booij) Vrije Universiteit, Faculteit der Letteren De Boelelaan 1105 1081 HV Amsterdam The Netherlands
[email protected]
(Ans van Kemenade) Katholieke Universiteit Nijmegen, Faculteit der Letteren Erasmusplein 1 6525 HT Nijmegen The Netherlands
[email protected]
This page intentionally left blank
Aspectual contrasts and lexeme derivation in Estonian: a realization-based morphological perspective FARRELL ACKERMAN
Exactly where preverb-compounding takes place is a vexed question. Simpson (1992: 115)
1. INTRODUCTION Many languages have morphological means for distinguishing between atelic/telic aspectual contrasts associated with related transitive predicates.1 For example, it is well-known that the lexicons of Slavic languages contain pervasive patterns of paired predicates. This is exemplified by Russian in (1): (1)
a.
Vcˇera ja kosil travu yesterday I cut.past grass.acc ‘Yesterday I cut/was cutting (the) grass.’
b. Vcˇera ja skosil tselij gektar yesterday I cut.perfective.past whole hectare.acc ‘Yesterday I cut the whole hectare.’ Roughly speaking, the situation denoted in (1a) need not be construed as fully completed, hence the possibility for a progressive reading, nor need the action be interpreted as directed at a specifically delimitable quantity denoted by the obj(ect) argument. In contrast, the situation denoted in (1b) is construed as completed, affecting the entirety of the entity denoted by the obj argument. For the time being, I will assume that (1a) corresponds to an atelic reading and (1b) to a telic one (see section 4 for more detailed discussion). As can be seen, while the predicates in each of these sentences vary, they share the same verbal root: this root is unprefixed in (1a), but prefixed in (1b). Crucially, this variability in predicate marking and semantic interpretation is correlated with invariance of obj marking: the obj arguments in both (1a) and (1b) bear acc case. Thus, the observable meaning differences correspond to alternative forms of related predicates. The obvious formal difference in the morphological shape of semantically related predicates has led traditional descriptive linguists, as well as many constraint-based lexicalists, to assume a lexicality status for these constructions. That is, they are analyzed as composed in the lexical/ morphological component of the grammar, rather than as being products of syntactic word-formation.2 Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 13–31. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
14
Farrell Ackerman
The Finnic language Estonian exhibits a cross-linguistically less common strategy for marking similar atelic/telic contrasts. Instead of systematically creating pairs of morphologically related predicates, as in Slavic, it differentially case marks obj arguments, as in (2): (2)
a.
Ma ehitasin endale suvilat I-nom built-past-1sg myself-all cottage-part ‘I built/was building the cottage for myself.’
b. Ma ehitasin endale suvila I-nom built-past-1sg myself-all cottage-gen ‘I built the cottage for myself.’ (adapted from Erelt et al. 1997: 36) This strategy yields a single verbal root form, in the present instance ehitama ‘to build’, which participates in contrasting case government patterns for obj arguments. As can be seen from the glosses, these alternate marking options correspond to distinct, but familiar, semantic interpretations: the presence of part marking correlates with an atelic reading and the presence of gen marking with a telic one. Lexically governed case and/or grammatical function alternations frequently occur in the languages of world and are straightforwardly associated with the lexical representations for various (classes of) predicates in all lexicalist frameworks.3 This, of course, suggests an easy and principled assimilation of the telic/atelic distinction as distinguished by differential object marking into the class of lexical constructions. In addition, perhaps under Baltic, particularly German, Sprachbund influence within the domain of prefixal complex predicates (see Hassellblatt 1990), Estonian has innovated a system of phrasal predicates with separable preverbs which contrasts with simple verbal stems: such pairs convey the same atelic/telic contrast as that found in (2). This is exemplified in (3): (adapted from Lavotha 1960: 104) (3)
a.
Ku¨tt laskis a¨nest hunter shoot.past.3sg rabbit.part ‘The hunter shot (a/the) rabbit.’
b. Ku¨tt on ja¨nese maha lasknud hunter cop-3sg rabbit.gen pv shoot.perf ‘The hunter has shot down the rabbit.’ c.
Ku¨tt laskis ja¨nese maha hunter shoot.past.3sg rabbit.gen pv ‘The hunter shot down the rabbit.’
Aspectual contrasts and lexeme derivation in Estonian
15
As in the Russian examples in (1) and the Estonian examples in (2), (3a) and (3b) differ with respect to telicity, where, as in (2), the atelic predicate correlates with objPART and the telic predicate with objGEN . In addition, as in Russian, the Estonian predicates in (3) differ formally from one another: the atelic predicate in (3a) is expressed by a simple verb root, while the predicate in (3b) is expressed by the same verbal root accompanied by a preverb. These Estonian predicates, consequently, are consistent with basic assumptions concerning their lexicality, given the formal morphological contrast they share with Russian predicates, the obvious semantic relatedness between the contrasting predicates, and the case-government properties they share with simple Estonian predicates, such as those in (2). Finally, as demonstrated by (3c) the pv is separable from the verbal stem under language particular specifiable syntactic conditions. This suggests that despite their lexicality, such predicates, contrary to their Russian analogues, possess the property of surface phrasality. Research has revealed that there are certain cross-linguistically common properties of complex predicates consisting of verb stems and preverbs or particles.4 First of all (classes of) such predicates can exhibit predictable and systematic or somewhat idiosyncratic lexical semantic and/or syntactic differences, i.e., valence, grammatical function status of arguments, case government, etc., relative to their simple predicate bases. Within the present context this will be interpreted as indicating that phrasal and simple predicates can differ with respect to what I will refer to as lexical or alternatively, lexemic properties. Secondly, such predicates generally become synthetic morphological entities when they undergo category-changing derivation such as nominalization.5 Finally, the pieces of phrasal predicates exhibit their own language particular syntactic distributions depending on the properties of specific syntactic constructions within the languages in which they appear.6 All of the properties converge to suggest that phrasal predicates represent an ‘‘analytic paradox’’ for linguistic theory.7 Their essentially problematic nature is remarked upon by many linguists of previous generations. For example Watkins (1964: 1037) says the following concerning phrasal predicates in Indo-European: ‘‘PV V compositions constitute ‘‘single semantic words’’, comparable to simple lexical items; yet they permit tmesis, or syntactic separation, suggesting that internal parts are independent syntactic entities. This observation permits one to identify two independent problems. The present article explores proposals which address each problem. The first problem can be stated a follows: How can the lexicality of phrasal predicates with respect to their unithood as ‘‘single semantic words’’, their relatedness to simple predicates, and their participation in category preserving and changing derivation, be reconciled with the syntactic independence of their parts.8 The proposal I entertain,
16
Farrell Ackerman
following Aronoff (1994), Stump (2001), Booij (2002), among others, is a realizational and paradigm-based one which claims that Lexeme-derivation operations relate the lexical properties of a (class of) lexeme l to a (class of) lexeme l’ yielding (networks of) derivational paradigms. Thus, the present answer to the question posed by Simpson in the quotation at the beginning of this article is that phrasal predicates are formed in the lexicon. This of course raises the issue as to what view of the lexicon is to be assumed: this will be considered in section 2. The second problem relates to the specific nature of a paradigm-based proposal. It can be formulated as follows: How will such an analysis impact on the simplest interface assumption between words (simple or complex) and their syntactic expression? Much research in lexicalist frameworks appears to be predicated on the following default assumption concerning this interface: A word w is a synthetic member of category X and w is inserted as the head of XP. I address this issue by adopting the Periphrastic Realization Hypothesis for both derivation and inflection (Ackerman and Stump to appear): Periphrastic Realization Hypothesis (General Formulation)9 Rules that deduce the forms occupying a paradigm’s cells from the lexical and morphosyntactic property sets associated with those cells include rules defining periphrastic combinations as well as rules defining synthetic forms. In sum, the present article argues that phrasal predicates in Estonian are periphrastic lexical constructions created by lexeme-formation operations within a Realization-based Lexicalist perspective. Moreover, the general assumptions and mechanisms of realizational models will be shown to straightforwardly extend to account for phrasal predicates when periphrasis (multi-word expression) is permitted to be a possible kind of exponence in lexeme-formation, as has been independently argued to be the case for inflection (see Robins 1959, Sadler and Spencer 2001, Spencer 2001, Spencer to appear, Ackerman and Stump to appear, Harris to appear, Brassil to appear, among others.) Section 2 presents some of the theoretical and conceptual preliminaries which collectively combine to provide the realizational and paradigm-based backdrop for the present proposal. Section 3 demonstrates how these assumptions apply in the analysis of Estonian atelic/telic contrasts with finite transitive verb forms. Section 4 summarizes the results. 2. THEORETICAL AND CONCEPTUAL PRELIMINARIES As observed in Aronoff (1994) there are two conceptions of the notion ‘lexical’ that are profitably, but infrequently, distinguished in the theoretical literature.
Aspectual contrasts and lexeme derivation in Estonian
17
He proposes that Lexical1 denotes the repository of what is idiosyncratic in language, following the tradition of Bloomfield (1933), DiSciullo and Williams (1987), among others: ‘‘The term lexicon should therefore be understood ... as referring only to the permanent lexicon: the list of all idiosyncratic signs, regardless of their category or complexity.’’ Aronoff (1994: 22) Lexical2 , in contrast, bears more centrally on word and paradigm based variants of morphology and has ‘‘to do with lexemes’’, following the tradition of Sapir (1921), Matthews (1972), among others. Aronoff (1994: 22) says that, ‘‘The endless list of all lexemes, by contrast, will remain nameless.’’ 10 It is reasonable to interpret this view as implicating a notion of the lexicon as the locus of derivational and inflectional operations affecting lexemes and exhibiting gradient degrees of regularity; this accords better with the explanatory assumptions of lexicalist theories than the view expressed by lexical1 . Within realization-based models of morphology as typified by Stump (2001), Spencer and Sadler (2001), Blevins (to appear), operations modifying properties associated with lexemes can be legitimately interpreted as lexical in this second sense, and will be so here. This obtains, following the standard lexicalist line, whether such modification has to do directly with lexical semantics and correlated lexical properties such as valence, the grammatical function status of arguments, case government requirements of complements, etc., and thus with lexeme-derivation, or with morphosyntactic features such as tense, aspect, agreement, etc., and thus with inflection. This yields an interpretation of Lexicalism compatible with Lexical2 which can be referred to as the Realization-based Lexicalist Hypothesis (following Blevins 2001): Realization-based Lexicalist Hypothesis Lexicalism is a hypothesis about the correspondence between contenttheoretic aspects of lexemes (associated with lexical and/or morphosyntactic property sets) and the surface forms that realize them. In line with this, Ackerman and Stump (to appear) propose that lexical representations are pairings of lexemic (l) and morphosyntactic (s) content associated with a root or stem (r) which are realized on the surface by some phonological form (y). This can be represented schematically as:
r y.11 In Ackerman and Webelhuth (1998) this separationist view of lexicalism (Beard 1995), sharply distinguishing content from form, was further constrained by the Principle of Lexical Modification (also known as the Principle of Lexical Adicity in Ackerman and Webelhuth (1998))12
18
Farrell Ackerman
Principle of Lexical Modification Only lexical (not syntactic) operations can alter or affect information associated with lexemes. This principle was intended to function as a sufficient condition for determining the lexical status of constructions by hypothesizing a substantive restriction on the modification of contentive information, i.e., if there is evidence of lexical2 effects, then the responsible operation is lexical/morphological, not syntactic, irrespective of whether the surface expression is synthetic or periphrastic. It is generally assumed within morphological theories that the inflectional morphology of a language defines sets of inflectional paradigms. It has been hypothesized, in addition, that the derivational morphology of a language, as typified by lexeme formation operations, defines sets of derivational paradigms (see Booij 2002, among others) and that these, in turn, constitute networks of related lexical representations.13 Following Ackerman (to appear) I will assume that for each derivational category l available to a lexeme L14 with root R, there is a function wderl such that wderl R=R∞ , where R∞ is a cell in the derivational paradigm of L occupied by R∞. Thus, if l=causative and is available to the Estonian lexeme move with a root liigu, then Qcaus <move> liigu=<make move> liiguta, where liigutama is the infinitival form of this verb. Focusing on the domain of Estonian lexeme-derivation I will, additionally, adopt the distinction between two types of lexeme-formation operations for predicates articulated in Ackerman and Moore (2001).16 The first type, Morphosemantic Operations, alters lexical semantics with possible consequences for aspects of lexical property sets such as grammatical function assignment and case-government.17 Causativization as exemplifed by the previously cited Estonian contrast between the non-causative infinitival form liikuma ‘to move, circulate’ versus the causative infinitival form liigutama ‘to move, set in motion’ represents a Morphosemantic Operation.18 In section 3 I will argue that this type of operation relates atelic and telic variants of Estonian predicates. The second type, Morphosyntactic Operations, does not affect lexical semantics, but, like Morphosemantic Operations can alter other aspects of lexical property sets such as grammatical function assignment. Passive, which is generally assumed to simply reassign grammatical functions to a predicate with unaltered lexical semantics, exemplifies this latter class of operations. Now, given the sharp separation between content and form within realization-based proposals, the diagnostics employed to argue for the lexicality of an entity are strictly independent of the synthetic versus periphrastic form by which that element is realized. In particular, I will assume with Ackerman and Stump (to appear) that there are two realization principles for expressions in paradigm cells, where, once again, the pair represents a lexeme and all of its
Aspectual contrasts and lexeme derivation in Estonian
19
morphosyntactic properties, while R represents a root form associated with the pair. Synthetic Realization Principle (=Morphological Expression of Ackerman and Webelhuth 1998): Where the realization w of L,sR is a synthetic member of category X, w may be inserted as the head of XP. Periphrastic Realization Principle: Where the realization of w1 w2 of L,sR is periphrastic and w1 and w2 belong to the respective categories X and Y, w1 and w2 may be inserted as the heads of the respective nodes X(P) and Y(P). The structural relationship between X(P) and Y(P) in the Periphrastic Realization Hypothesis is determined by language-specific empirical evidence concerning constituency. Additionally, the surface distribution of periphrastic lexical constructions is keyed to (classes of) syntactic constructions and, consequently, to the identification of the inventory of syntactic construction types in a language.19 In sum, the preceding assumptions, all motivated for other concerns, make it possible to formally address the analytic paradox concerning the semantic unithood of phrasal predicates, despite the syntactic separability of their pieces.20 The basic idea is that phrasal predicates occupy cells in derivational paradigms and are related to simple predicates as well as other words via derivational paradigm functions involving morphosemantic lexeme-formation operations. Lexical restrictions on the application of specific lexeme-derivation operations to (classes of) predicates, as well as lexical idiosyncrasy associated with certain predicates within even regular derivational paradigms, argue for a lexical/morphological treatment. Participation of phrasal predicates in category preserving and changing derivations, likewise, argues for a lexical/morphological treatment. Realization-based models provide all of the relevant ingredients for the analysis of phrasal predicates in terms of morphology, if, as per The Periphrastic Realization Hypothesis, periphrasis is permitted to be form of morphological exponence.
3. ESTONIAN PHRASAL PREDICATES AND DERIVATIONAL PARADIGMS: ACCOUNTING FOR THE ATELIC/TELIC CONTRAST21 Researchers generally propose that there are two classes of Estonian predicates which participate in atelic/telic telic contrasts. First, Type 1, referred to as Aspectual Predicates in some of the literature (Erelt et al. 1997), contains simple
20
Farrell Ackerman
finite predicates which participate in distinct obj case-government patterns: when the obj appears in the partitive case, the predicate has an atelic interpretation, while when it appears in the genitive case, it has a telic one.22 (4) a.
Ma ehitasin endale suvilat kaks na¨dalat. I-nom build-past-1sg myself-all cottage-part (two-nom week-part) ‘I was building the cottage for myself (for two weeks).’
b. Ma ehitasin endale suvila kahe na¨dalaga. I-nom build-past-1sg myself-all cottage-gen two-gen week-com ‘I built the cottage for myself (in two weeks).’ (adapted from Erelt et al. 1997: 36) The durative adverbial in (4a) is standardly argued to co-occur with atelic predicates, while the time span adverbials, such as that in (4b), co-occur with telic predicates. The operative notion of telicity for present purposes is based on Krifka (1998):
telic predicate: A lexical predicate P is telic iff for every event e and e∞, such that P(a1 , ..., an , e) and P(a1 , ..., an , e∞), and where e∞ is a subevent of e, e and e∞ have the same boundaries (end-points). With respect to this definition, the Estonian phrase for build the cottage, with the genitive suvila in (4b), is associated with a telic interpretation because any subevent that qualifies as build the cottage shares the same end-points. In this instance, the only relevant subevent is the whole event itself. Thus, in most cases, a telic predicate will have a quantized quality; that is, no proper subevent will represent an instance of the predicate. The lexical predicate that selects a genitive object (e.g., suvila) is a telic predicate, thus entailing that one of its arguments will denote an end-point. Given that the end-point of build the cottage corresponds to the entire cottage, the genitive, whole object suvila denotes this end-point. As will be further discussed in sections 3.1 and 3.2 an object denoting endpoints in this manner will be associated with a property which Ackerman and Moore refer to as bounding entity. The Estonian phrase for building the cottage with the partitive form suvilat in (4a), on the other hand, receives an atelic interpretation. There is at least one proper subevent of the cottage building event that itself qualifies as a cottage-building event. Figure (1) provides additional exemplars of Type 1 predicates: avastama parandama kujundama keetma
‘discover’ ‘improve’ ‘shape’ ‘cook’
looma saavutama koostama voltima
‘create’ ‘attain’ ‘put together’ ‘fold’
Figure 1. Type 1 Predicates
Aspectual contrasts and lexeme derivation in Estonian
21
Type 2 consists of simple verbal stems paired with phrasal predicates containing the simple verbal stem and some separable preverb. This is exemplified in (5) (5)
a.
Ma loen *raamatu/raamatut. I.nom read.1sg.pres book.part ‘I am reading the book’
b. Ma loen raamatu/*raamatut la¨bi. I.nom read.1sg.pres book.gen/*book.part pv ‘I will read the book’ (Kippasto and Nagy 1995: 224) In Type 2 the simple predicate is atelic and its obj argument is governed for the partitive case, while the related phrasal predicate is telic (sometimes with additional meaning changes) and its obj is governed for the genitive case. Figure (2) contains a list of additional members of Type 2. arvama kasutama seletama va¨sitama
‘think, guess’ ‘use’ ‘explain, expound’ ‘tire’
a¨ra arvama a¨ra kasutama a¨ra seletama a¨ra va¨sitama
‘puzzle out, solve’ ‘take advantage of’ ‘clear up, account for’ ‘wear out’
Figure 2. Type 2 Predicates Looking at Type 1 and Type 2 a basic descriptive generalization becomes apparent. Predicates in both classes exhibit differential object marking: Atelic predicates govern objPART , while telic predicates govern objGEN . Thus, a difference in meaning corresponds to a difference in lexical property sets for related lexical representations with respect to case government. This kind of meaninginduced manipulation of lexical property sets is diagnostic of morphosemantic lexeme-formation operations as provided previously. In what follows I will argue for two related, but independent, points concerning the analysis of aspectual alternations exhibited by members of Type 1 and Type 2. First, the Estonian atelic/telic contrast reflects a morphosemantic lexeme-formation operation which provides distinctive entailments for atelic and telic variants of predicates. Second, the correspondence between atelicity and objPART and telicity and objGEN follows from the Paradigmatic Argument Selection Principle as formulated in Ackerman and Moore (2001). The point of departure for addressing these issues is the Proto-Property proposal of Dowty (1991). On this account the entailments of predicates are interpreted in terms proto-agentive and proto-patientive properties associated with co-occurring arguments. In addition, the relative numbers of proto-agentive and protoagentive properties associated with distinct predicate arguments determine the correspondence between such arguments and their grammatical function and/or
22
Farrell Ackerman
case marking encoding. This can be illustrated as follows for a generic predicate a which selects for two arguments and the grammatical functions subj and obj: if arg1 has the preponderance of proto-agentive properties in contrast to a preponderance of proto-patientive properties associated with arg2 , then arg1 will align with the subj function, while arg2 will align with the obj function:
On this view, predicates are associated with both semantic entailment sets for their arguments and grammatical function inventories. Dowty’s essential insight is that a syntagmatic argument selection principle represents a reliable crosslinguistic hypothesis concerning how co-occurring arguments with specified entailment set inventories of basic, i.e., non-derived, predicates typically align with grammatical function inventories: the most proto-agentive set aligns with subj, while the most proto-patientive set aligns with obj. The approach motivates why the basic predicate meaning ‘build’ in English (and analogous predicates in other languages) exhibits the following alignment of entailments and grammatical functions: most Proto-Agentive: Subject
most Proto-Patientive: Direct Object
build< x,
y >
volitional (proto-ag) sentient (proto-ag) causing change of state (proto-ag) movement relative to obj (proto-ag) exists independently (proto-ag)
undergoes change of state (proto-pat) incremental theme (proto-pat) causally affected (proto-pat) stationary relative to subj (proto-pat) lack of independent existence (proto-pat)
3.1. The Lexeme-derivation of telic predicates On the assumption that the arguments of predicates are associated with entailment sets, as in Dowty (1991) described above, two predicates can be related in terms of additions and deletions of various proto-property entailments. Ackerman and Moore (2001) argue for the postulation of a new protopatientive argument which they call bounding entity. For Estonian they propose that this property is associated with the obj argument of the base atelic predicate yielding a derived telic predicate with an obj which possesses more proto-patientive properties than the obj of the related base predicate. The
Aspectual contrasts and lexeme derivation in Estonian
23
essential elements of this proposal can be schematized as follows, where a telicizing operation applies to an atelic predicate and results in a telic predicate with bounding entity among the proto-properties associated with the obj argument of the related derived predicate.
This schema can be instantiated in the following way for the predicate ehitama ‘build’:
As previously mentioned, the notion bounding entity as motivated in Ackerman and Moore (2001) is a proto-patientive property which refers to an endpoint. Applied to (4b), the end-point of build the cottage corresponds to the entire cottage where, the objGEN suvila ‘cottage’ denotes this end-point, and is, therefore a bounding entity. As is standard in realizational accounts, the root y in the lexeme-formation operations can be identical to x, and this assumption will be shown to be important below. Clearly, such an account presumes a notion of Lexical Relatedness and this can be characterized as follows within the realization-based assumptions adopted here: A lexeme l∞ is related to a lexeme l iff l∞ is an l derivative of l. Thus, the atelic and telic variants of Estonian predicates are interpretable as related, since the telic variant is a derivative of the atelic variant via the application of the lexeme-formation operation which alters semantic properties of the predicate. This is a morphosemantic operation, according to the taxonomy of lexical operations assumed here, because it has the effect of manipulating and modifying lexical semantic properties of the relevant predicate. So far, I have focused on how the lexeme-formation operation alters the contentive information associated with predicates: specifically, on the difference in entailed properties and their reflection in case government requirements for obj arguments. If one shifts attention to the formal realization or exponence of such contentive information it becomes clear that telic realizations can be either phrasal pv-v compounds (Type 2) or simple predicates (Type 1). This independence of content and form is a foundational assumption of realizational proposals and, hence, is predicted to occur within such proposals: that the same content
24
Farrell Ackerman
can be expressed in numerous ways is a fundamental assumption of realizational approaches. Instructively, the realization rule for Type 1 predicates is of the sort which Stump (2001) calls the Identity Function Default (ifd): it is the realization rule that defaults to the root of the base lexeme without modification, and Stump has shown that it applies quite broadly within morphology. Indeed, the application of this default in Type 1 explains why identical surface forms can be associated with different entailments sets and case-government demands. For Type 1 predicates there is, thus, a semantic change effectuated by a lexemeformation operation which is not matched with a formal change in the exponence of the predicate. Indeed, given both the present adoption of realizational assumptions and the hypothesis that the Identity Function Default is a general realizational option within such realizational proposals, it would be suspicious if the ifd did not obtain in derivation, as it does in inflection. As a consequence, the data from Type 1 can be viewed as following straightforwardly from independently motivated and required realizational assumptions.
3.2. Explaining the correlations between meaning and the case encoding of obj s It was previously stated as a descriptive generalization that telicity correlates with gen and atelicity with part. While these correlations are stipulated in the telic lexeme-formation operation presented above, they do not appear to follow from any particular principle of lexeme-formation operations. Correlations between lexical semantics and grammatical encodings can be stipulated in other ways as well; for example, in terms of highly articulated tree structures as in Ramchand (1997) for Scots Gaelic or as in Nelson (1998) for Finnish. While such stipulations in either the present lexeme-formation account or the two cited syntactic accounts ensure descriptive accuracy, their obvious drawback is that nothing really prevents one from having stated the correlations inversely so that telicity correlates with part case-marking and atelicity with gen casemarking. In recognition of this problem, Ackerman and Moore (2001, following the sorts of functionalist results concerning patterns of argument encoding and the lexical semantics of predicates as found in Tsunoda (1981), Hopper and Thompson (1981), among others, argue that the observed correlations can, in fact, be (at least probabilistically) predicted. In particular, we posit the existence of a Paradigmatic Argument Selection Principle which regulates the correspondence between the semantics of predicate arguments and their surface encodings, when there is a semantic contrast and an alternation in (grammatical function or case) encoding. On such an account the alignments between lexical semantics of related predicates and the specific differential object marking of their arguments can be shown to follow, rather than being simply stipulated. I now turn to how this works.
Aspectual contrasts and lexeme derivation in Estonian
25
Ackerman and Moore (2001: 95) state the Paradigmatic Argument Selection Principle as follows: Paradigmatic Argument Selection Principle: Let p (..., argi , ...) and P∞ (..., arg∞i , ... ) be lexically related predicates, where argi and arg∞i are corresponding arguments. If argi and arg∞i exhibit different grammatical encodings and argi is more prototypical with respect to a particular proto-role than arg∞i , then the encoding of argi encoding will be less oblique than the encoding of arg∞i . This principle constrains the grammatical encoding for the ‘‘corresponding arguments’’ of related predicates in accordance with their relative degree of prototypicality for entailed property sets. In order to see how this principle operates it is crucial to provide a definition of obliqueness: Obliqueness An element b is more oblique than another element a iff b appears to the right of a on either the grammatical function or case hierarchy.23 Obliqueness can be calculated in terms of the standard grammatical function hierarchy, as well as the less standard, but nonetheless justifiable, case hierarchy in:
gf hierarchy:
subj>do>io>obl
Case hierarchy:24 nom>acc>gen>part>dat>loc>abl/inst>other The potential effects of this principle become clear by examining a schematic diagram for the three lexically related transitive predicates Preda , Predb , and Predc below:
26
Farrell Ackerman
As can be seen, if we identify arg2 as the corresponding argument of the three related predicates preda, predb, predc, it is possible to compare the relative number of proto-properties of the ‘same’ argument across related predicates in derivational paradigms. While each difference in an entailed proto-patient property necessitates the postulation of a distinct predicate, the Paradigmatic Selection Principle insures that the actual encoding of corresponding arguments will follow from either the grammatical function hierarchy or the case-hierarchy. We have already seen that atelic and telic predicates are interpretable as lexically related and are, consequently, candidates for the application of Paradigmatic Argument Selection. In fact, Paradigmatic Argument Selection applies straightforwardly to motivate the attested alignment of differential case marking with arguments possessing constrasting proto-patientive property sets. This can be represented diagrammatically, where, given the case hierarchy, predTELIC is predicted to govern the gen case for its obj argument by virtue of possessing more proto-patientive properties than its related predATELIC variant:
Thus, the Paradigmatic Argument Selection Principle can been construed as functioning as a well-formedness condition on the lexical encodings of related predicates which exhibit semantically induced encoding contrasts, as in the Estonian atelic/telic contrast. Lexical representations are constrained to exhibit alignments/correspondences between semantic arguments and their grammatical encodings, e.g., atelicity/partitive and telicity/genitive, in accordance with the Paradigmatic Selection Principle, and thus these correspondences need not be stipulated on the present account. Of course, these correspondences are expected to obtain independent of whether a particular predicate is of Type 1 or Type 2, since these types differ simply with respect to surface exponence, not with respect to the semantic nature of their contrast. Moreover, the hypothesis that predicates participate in derivational paradigms can be seen as facilitating the kind of comparisons required by paradigmatic selection. This follows since the proto-property information that needs to be compared between predicates is associated with the lexical representations of those predicates related via derivational paradigms.
Aspectual contrasts and lexeme derivation in Estonian
27
4. CONCLUSIONS In this article I have argued that Estonian phrasal predicates are lexical constructions created by lexeme-formation operations, but that they participate in derivational paradigms. As is typical within realization-based accounts, the content of these constructions is demonstrably separate from its form. The consequent expectation is that similar content can be expressed by distinct forms and, moreover, that some of those forms may even be phrasal. The phrasality of form is sanctioned by the Periphrastic Realization Hypothesis. In Estonian, the postulation of derivational paradigms provides a straighforward characterization of the notion ‘related predicate’. This, in turn, facilitates the operation of Paradigmatic Argument Selection, providing a principled correspondence between the entailments of predicate arguments and their grammatical encodings in lexical representations. In general, the proposed analysis of Estonian atelic/telic predicates demonstrates that central assumptions of realization-based morphological theories dovetail explanatorily in the domain of lexeme-derivation with Dowty’s independent line of research concerning lexical semantics and grammatical encoding. NOTES 1
I focus here solely on transitive finite predicates, though intransitives in many languages exhibit such pairing as well. See section 4. for a definition of telicity following Krifka (1998, 1992); see also Smith (1997), Filip (1999), among others. 2 See Rappaport-Hovav (2002) for a critique of syntactic analyses of aspectuality and Julien (2002) for a representative approach to syntactic word-formation. 3 ´ preszja´n and Pa´l One need only look at a typical valence dictionary such as that by A (1982) for Hungarian and Russian to see how pervasive and characterized by smaller and larger subsets of systematicity such patterns of case government really are. 4 See Dahlstrom (1996) on Fox, Rice (2000) on Athapaskan languages, Simpson (1992) on Walpiri, Stiebels and Wunderlich (1994) on German, among others. Simpson (1992: 115) identifies a typical profile for Walpiri: Preverbs vary in compositionality. Some are clearly non-compositional (and thus should be formed in the morphology), but many are highly productive and form quite transparent compounds (and thus are candidates for syntactic word formation). Phonologically, preverbs compounded with verbs act like parts of words and not full words, since they do not have to obey the morphological structure constraint that all words in Walpiri must end in a vowel. In terms of word-formation, preverbs compounded with verbs can have nominalizing suffixes attached, and then derivational suffixes ... some preverbs, notably the preverb jangkardu ‘against’ add Dative arguments to verbs. (Simpson 1992: 115) 5 See Ackerman (1987) and Ackerman and LeSourd (1997), however, for instances where separability is maintained with some deverbal adjectives when such derivates are used predicatively.
28 6
Farrell Ackerman
There have been numerous recent attempts to argue that the syntactic behaviors of phrasal predicates in several languages are explicable in terms of syntactic movement operations and thereby participate in a shared structure-based explanation. See Koopman and Szabolcsi (2000) (and references cited there) for the analysis of Dutch and Hungarian. Despite the prevalence of such syntactocentric accounts (see Jackendoff 1997) there is much reason to believe that this increasingly abstract direction of analysis is essentially misguided (Culicover 1999, Jackendoff 2002, among others). 7 This is the term used in Nash (1982), a neglected and insightful investigation into preverbs. 8 A recent effort to address this paradox has been offered within an optimality theoretic perspective by Ackema and Neeleman (2001). They note some similarities in spirit between their proposal and that in Ackerman and LeSourd (1997). The present proposal shares some of the central intuitions guiding these proposals. 9 See below for a more specific formulation of this hypothesis. 10 Following Aronoff (1994:10), lexeme will be restricted to (grammatical) words and thus, excludes vp idioms. 11 The arrow signals the result of applying realizational rules within the class of morphological theories assumed here. See Ackerman and Stump (to appear) for discussion. 12 This is, in effect, amounts to a Generalized Direct Syntactic Encoding Principle following the lead of lfg’s Direct Syntactic Encoding Principle which specifically addresses grammatical function alternations. 13 This adapts ideas from Ackerman and Stump (to appear), and builds on ideas from Robins (1959), Matthews (1972), Aronoff (1976, 1994), Zwicky (1989), Anderson (1992), Stump (2001), Sadler and Spencer (2001), Spencer (2001), among others. 14 More specifically, l itself represents lexemic information consisting of the triplet [lexical meaning (m), lexical category, lexical property set], where the lexical property set is taken to include valence, grammatical functions, case, government, etc. 15 Of course, one can assume here a null set of morphosyntactic properties appropriate to this lexeme, making these representations identical to the pairings for inflection above. 16 See references in Ackerman and Moore (2001) for similar distinctions posited by other researchers. 17 This corresponds to operations on lexical conceptual structures in Levin and Rappaport-Hovav (1998). See Booij (2002) and Baayen and Lieber (1993) for verbal prefixation as modifications of lexical conceptual structures. 18 Throughout this article I ignore as irrelevant to present issues the sort of stem gradation phenomena typified by the contrast between non-causative liiku and causative liigu. 19 Recurrent syntactic construction types, i.e., overarching cross-linguistic generalizations, can be modeled within the grammatical archetype architecture of Ackerman and Webelhuth 1998, Malouf 2000, Ginzburg and Sag 2001. 20 Of course, I am not claiming that these assumptions provide the only means to address this paradox, but simply one that is consistent with realizational models of morphology as they have been independently developed for inflectional phenomena. In fact, there are several analogues to the assumptions made here in other traditions. For example, in some ways the claim that morphosemantic lexical operations create ‘semantic words’ resembles certain claims concerning abstract incorporation in lf which permits syntactically separated heads to form a kind of semantic complex, as in Zeller (2001). I forego detailed comparison between alternatives in this article.
Aspectual contrasts and lexeme derivation in Estonian
29
21
This analysis represents an realizational morphology interpretation of the proposals made in Ackerman and Moore (2001). 22 This discussion follows the more detailed presentation in Ackerman and Moore (2001: Chapter 5). 23 Given this definition of obliqueness it is possible that two elements exhibit opposite obliqueness properties in terms of case and grammatical relations. We leave these cases aside in this discussion. 24 This hierarchy is adapted from Blake (1994: 157). See Ackerman and Moore (2001) for motivation of relative positions on this hierarchy.
REFERENCES Ackema, P. and A. Neeleman (2001). Competition between morphology and syntax. In G. Legendre et al. (eds.), Optimality Theoretic Syntax. Cambridge MA: MIT Press, 29–60. Ackerman, F. (1987). Miscreant morphemes: Phrasal predicates in Ugric. U.C. Berkeley PhD Thesis. Ackerman, F. (to appear). Morphosemantic mismatches and Realization Based Lexicalism. In E. Francis and L. Michaelis (eds.), Linguistic Mismatch: Scope and Theory. Stanford: CSLI Publications. Ackerman, F. and J. Moore (2001). Proto-properties and Grammatical Encoding: A Correspondence Theory of Argument Selection. Stanford: CSLI Publications. Ackerman, F. and G. Stump (to appear). Paradigms and periphrastic expression: A study in Realization-based Lexicalism. Projecting Morphology, eds. A. Spencer and L. Sadler. Stanford: CSLI Publications. Ackerman, F. and G. Webelhuth (1998). A Theory of Predicates. Stanford: CSLI Publications. Ackerman, F. and P. LeSourd. (1997). Toward a lexical representation of phrasal predicates. In J. Bresnan and P. Sells (eds.), Complex Predicates. Stanford: CSLI Publications, 67–106. Anderson. S. (1992). A-Morphous Morphology. Cambridge: Cambridge University Press. ´ preszja´n, J.D. and E. Pa´l (1982). Orosz-ige-Magyar-ige. Budapest: Tanko¨nykiado´. A Aronoff, M. (1976). Word Formation in Generative Grammar. Cambridge, MA: MIT Press. Aronoff, M. (1994). Morphology by Itself. Cambridge, MA: MIT Press. Baayen, H. and R. Lieber (1993). Verbal prefixes in Dutch. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1993. Dordrecht: Kluwer, 51–79. Bauer, L. (1997). Derivational paradigms. In G. Booij and J. van Marle (eds). Yearbook of Morphology 1996. Dordrecht: Kluwer, 243–256. Beard, R. (1995). Lexeme Morpheme Base Morphology. Albany: SUNY Press. Bierwisch, M. (1990). Verb cluster formation as a morphological process. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1990. Dordrecht: Foris, 173–199. Blake, B. (1994). Case. Cambridge UK: Cambridge University Press Blevins, J. (2001). Realisation-based lexicalism. Journal of Linguistics 37, 317–327. Blevins, J. (to appear). Word and Paradigm Morphology. Stanford: CSLI Publications. Bloomfield, L. (1933). Language. New York: Holt, Rhinehart, and Winston. Booij, G. (1997). Autonomous morphology and paradigmatic relations. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1996. Dordrecht: Kluwer, 35–53. Booij, G. (2002). The Morphology of Dutch. Oxford: Oxford University Press.
30
Farrell Ackerman
Brassil, D. (to appear). Periphrasis, standard lexicalism, and realizational lexicalism. Paradigms and periphrastic expression: A study in Realization-based Lexicalism. In A. Spencer and L. Sadler (eds.), Projecting Morphology. Stanford: CSLI Publications. Culicover, P. W. (1999). Syntactic Nuts: Hard Cases in Syntax. Oxford: Oxford University Press. Dahlstrom, A. (1996). Morphology and Syntax of the Fox (Mesquakie) language. Ms. University of Chicago. DiSciullo, A-M. and E. Williams (1987). On the Definition of Word. Cambridge, MA: MIT Press. Dowty, D. (1991). Thematic proto-roles and argument selection. Language 67, 547–619. Erelt, M., T. Erelt, and K. Ross (1997). Eesti Keele Ka¨siraamat. Tallinn: Eesti Keele Sihtasutus. Filip, H. (1999). Aspect, Eventuality Types, and Noun Phrase Semantics. New York: Garland Publishing. Ginzburg, J. and I. Sag (2000). Interrogative investigations: The form, meaning, and use of English interrogatives. Stanford: CSLI Publications. Harris, A. (to appear). Unexpected periphrasis in Udi and Georgian. In F. Ackerman, J. Blevins, and G. Stump (eds.), Periphrasis and Paradigms. Stanford: CSLI Publications. Hassellblatt, C. (1990). Das Estnische Partikelverb als Lehnu¨bersetzung aus dem Deutschen. Wiesbaden: Vero¨ffentlichungen der Societas Uralo-Altaica. Hopper, P. and S. Thompson (1981). Transitivity in grammar and discourse. Language 56, 251–299. Jackendoff, R. (1997). The Architecture of the Language Faculty. Cambridge, MA: MIT Press. Jackendoff, R. (2002). Foundations of Language. Oxford: Oxford University Press. ´ szt Nyelvko¨nyv. Budapest: Bibor Kiado´. Kippasto, A. and J. Nagy (1995). E Koopman, H. and A. Szabolcsi (2000). Verbal Complexes. Cambridge MA: MIT Press. Krifka, M. (1998). The origins of telicity. In S. Rothstein (ed.), Events and Grammar. Dordrecht: Kluwer, 197–235. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In I. Sag and A. Szabolcsi (eds.), Lexical Matters. Stanford: CSLI Publications, 29–53. ´ szt Nyelvko¨nyv. Budapest: Tanko¨nyvkiado´. ¨ . (1960). E Lavotha, O Malouf, R. (2000). Mixed Categories in the Hierarchical Lexicon. Stanford: CSLI Publications. Matthews, P.H. (1972). Inflectional Morphology. Cambridge, UK: Cambridge University Press. Matthews. P. H. (1991). Morphology. Cambridge, UK: Cambridge University Press. Nash, D. (1982). Verb roots and preverbs in Walpiri. Working Papers of SIL-AAB Series A, 6. Nelson, D. (1998). Grammatical Case Assignment in Finnish. New York: Garland Publishing. ´ szt Nyelvko¨nyv. Budapest: Tanko¨nyvkiado´. ¨ do¨n, L. (1960). E O Ramchand, G. (1997). Aspect and Predication: The Semantics of Argument Structure. Oxford: Oxford University Press. Rappaport Hovav, M. (2002). Review of C. Tenny and J. Pustejovsky (eds.), Events as Grammatical Objects: the Converging Perspectives of Lexical Semantics and Syntax. Stanford: CSLI Publications. Journal of Linguistics 38, 697–703. Rappaport Hovav, M. and B. Levin (1998) Morphology and lexical semantics. In A. Zwicky and A. Spencer (eds.), Handbook of Morphology. Oxford: Blackwell, 248–271. Rice, K. (2000). Morpheme Order and Semantic Scope. Cambridge, UK: Cambridge University Press. Robins, R.H. (1959). In defence of wp. Transactions of the Philological Society 116–144.
Aspectual contrasts and lexeme derivation in Estonian
31
Sadler, L and A. Spencer (2001). Syntax as an exponent of morphological features. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 2000. Dordrecht: Kluwer, 71–97. Simpson, J. (1992). Walpiri Morpho-syntax. Dordrecht: Kluwer. Spencer, A. (2001). The Word-and-Paradigm approach to morphosyntax. Transactions of the Philological Society 99, 279–313. Spencer, A. (to appear). Periphrastic paradigms in Bulgarian. Ms. University of Essex. Sapir, E. (1911). The problem of noun incorporation in American languages. American Anthropologist 13, 250–282. Sapir, E. (1921). Language. New York: Harvest Books. Smith, C. (1997). The Parameter of Aspect. Dordrecht: Kluwer [2nd edition]. Stiebels, B. and D. Wunderlich (1994). Morphology feeds syntax: The case of particle verbs. Linguistics 32, 913–668. Stump, G.T. (1993). On rules of referral. Language 69, 449–479. Stump, G.T. (2001). Inflectional Morphology. Cambridge, UK: Cambridge University Press. Stump, G.T. (2002). Morphological and syntactic paradigms: Arguments for a theory of paradigm linkage. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 2001. Dordrecht: Kluwer, 147–180. Tsunoda, E. (1981). Split case-marking patterns in verb types and tense/aspect/mood. Linguistics 19, 389–438. Watkins, C. (1964). Preliminaries to the reconstruction of Indo-European sentence structure. In H. Lunt (ed.), Proceedings of the 9th International Congress of Linguists. The Hague: Mouton, 1035–1042. Zeller, J. (2001). Particle Verbs and Local Domains. Amsterdam: Benjamins. Zwicky, A.M. (1985). How to describe inflection. In M. Niepokuj, M. Van Clay, V. Nikiforidou and D. Feder (eds.), Proceedings of the Eleventh Annual Meeting of the Berkeley Linguistics Society, 372–386. Zwicky, A.M. (1989). Quicker, quickly, *quicklier. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1989. Dordrecht: Foris, 139–173.
Dept. of Linguistics University of California at San Diego La Jolla, CA 92093 U.S.A. e-mail: [email protected]
This page intentionally left blank
Preverbs and particles in Old French* MONIQUE DUFRESNE, FERNANDE DUPUIS and MIREILLE TREMBLAY
1. INTRODUCTION Preverbs are intriguing grammatical objects. Semantically, they form a lexical unit with the verb they modify, sometimes behaving like an affix and sometimes more like an independent word. In all cases, however, preverbs appear to form a complex predicate with the verb they modify. While the formation of such complex predicates in Germanic languages has always received a lot of attention, very few studies have been devoted to the description of complex predicates in Romance languages. This is partly due to the fact that most modern Romance languages appear to lack verb-particle constructions, using derivation (prefixation) as a means of modifying the semantic and morphosyntactic properties of the stem verb. Consequently, with the notable exceptions of Roeper and Keyser (1992) and of Di Sciullo (1996, 1999), very few researchers have tried to provide a unified analysis of particles and prefixes within a broader theory of complex predicates. The goal of this paper is twofold. First, we establish the existence of a productive system where a prepositional element (a particle or a prefix) forms a complex predicate with the verb in Old French. Old French was characterised by two competing aspectual systems based on prepositions: prefixes, which could be either aspectual or locative, and particles, which were mainly locative but could take an aspectual reading. These two systems are illustrated in (1) and (2), respectively. (1)
quant li rois entent ceste parole, si s’apensa meintenant when the king hears this talk, (pro 3ps) A+think now que ce iert Lancelos that it was Lancelot ‘when the king heard these words, he realised that it was Lancelot’ Artu: 16
(2)
Et cil vont arriere a la nef And these go back to the ship. Partonopeu, 5901, cited in Buridant (2000, § 441)
In (1), the prefix a- adds an inchoative reading to the verb. Thus the verb apenser can be paraphrased by ‘to begin to think’. In (2), the locative particle arriere forms a semantic unit with the verb aller, with the meaning ‘to return, to come back’. Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 33–60. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
34
M. Dufresne, F. Dupuis and M. Tremblay
The two systems are of Indo-European origin. Not only are they found in other Indo-European subfamilies such as Germanic or Hellenic, they are also found, according to Vincent (1999), in earlier stages of Romance. The examples in (3) and (4) illustrate the use of prefixes and particles in Latin. (3)
(4)
a.
exercitum Ligerim traducit. he leads his army across the Loire. B.G. 7, 11, 9, cited in Hale and Buck (1988: 203)
b.
ne quam multitudinem hominum amplius trans Rthenum traducere. that he should lead no more crowds of men across the Rhine B.G. 1, 35, 5, cited in Hale and Buck (1988: 204)
si calvitur pedemve struit, manum endo iacito if he plays tricks or runs off, lay hand on (him) the Twelves Tables, cited in Vincent (1999: 1119)
The existence of particles and preverbs in both stages of the language (Latin and Old French) provides further support to the claim that preverbs are an Indo-European phenomenon, and are not simply found in Old French as a result of language contact with the Germanic language family during the Frank invasions. The second goal of this paper is to contrast the morphosyntactic characteristics of Old French prefixes and particles. In recent work, Buridant (1995, 2000) provides a detailed description of such constructions in Old French, which he classifies into two distinct categories. He distinguishes preverbs, which can be separable or non-separable, from particles. In this paper, we study these two systems, from both a synchronic and a diachronic perspective. We will focus on the morphosyntactic status of preverbs and particles, their distribution, and the relationship they entertain with the verb. In doing this, we will provide further arguments in favour of the distinction established by Buridant. More precisely, we show that Romance prefixes and particles show surprising similarities, yet that they have a very different grammatical status: prefixes are morphological objects, in the sense that they seem to form a word with the verb, while particles are syntactic objects because the particle and the verb are independent words. Our study is based on the Base de franc¸ais me´die´val (BFM) constituted by Christiane Marchello-Nizia of the E´cole Normale Supe´rieure de Lyon. This computerised database contains texts in verse and prose from the earliest stages of the language (the Strasbourg Oaths in 842) to the 16th century. In this paper, we only used data from the 12th century to the 15th century. The discussion is organised as follows. Section 2 presents an overview of the morphosyntax of Old French, introducing the morphological case system and the verb second phenomenon. Section 3 focuses on preverbs, particularly the
Preverbs and particles in Old French
35
distinctive properties of separable and non-separable prefixes. Section 4 contrasts the properties of Old French particles with those of prefixes, both separable and non-separable. Some arguments in favour of a unified analysis of prefixes and particles will be outlined in section 5. Finally, the conclusion discusses the consequences of this analysis for the concept of preverb as a theoretical construct. 2. OLD FRENCH AS A V-SECOND LANGUAGE In this paper, we rely strongly on verb movement within the clause to distinguish the morphosyntactic status of preverbs and particles. It is a well-known fact that Old French was a verb second language:1 in main clauses, the verb is raised to the second position, while it remains in its base position in subordinate clauses. This is illustrated in (5): the main verb pensa is raised to a position beyond the nominative subject li rois Artus, while the embedded verb (the auxiliary avoit and the past participle dit) remains in its base position and is thus preceded by its nominative subject Agravains.2 (5)
Cele nuit pensa li rois Artus assez a ce que Agravains That night thought the king Arthur enough to that which Agravains li avoit dit him-DAT had said ‘That night, King Arthur thought about what Agravains had told him’ Artu: 5
Within the Principle and Parameter framework (Chomsky 1981, 1986, 1995), the verb second phenomenon is accounted for by two movements:3 first, some constituent (XP) is moved to the specifier position of some functional category in the left periphery of the clause (such as CP or TopicP), then the verb is moved to the head position of the functional category, to the right of the XP constituent. Movement of the verb accounts for the order Verb-Subject we find in (5), assuming that the subject remains in its base position in the main clause as shown in (6). (6) [TopicP XPi Vj] [IP Subject tj ti] When the main clause also contains an auxiliary, only the auxiliary is moved and the past participle remains in its base position between the subject and the object. In (7), for example, the auxiliary avons occupies the second position, between the adverb la in first position and the subject nous. The past participle atendues remains in its base position and thus appears after the subject. (7)
La vous avons nous atendues / et porwardees par les rues There you have-1PP we waited-FP / and searched-FP in the streets ‘We waited for you there and looked for you in the streets.’ Feuille´e, 855–56, cited in Buridant (1995)
36
M. Dufresne, F. Dupuis and M. Tremblay
It is worth noting that the pronoun vous is not taken into consideration when establishing the V2 order since, as a clitic, it is part of the verb and is moved along with the verb. We will come back to this property of clitics in the next section. 3. PREVERBATION IN MEDIEVAL FRENCH Aspectual prefixes were very productive in Old French: prepositional prefixes, such as a-, de-, en-, sur-, por-, par-, were frequently used to modify the aspectual class of the predicate, as illustrated in (8). (8)
A: De: En: Sur: Pour:
porter parler amer saillir aller
‘to carry’ ‘to talk’ ‘to love’ ‘to jut out’ ‘to move, to leave’
a-porter de-parler en-amer sor-saillir por-aler
Par:
geter
‘to throw’
par-jeter
‘to bring’ ‘to speak ill of’ ‘to fall in love’ ‘to jump’ ‘to go all over, to look after someone’ ‘to throw away’
Old French preverbs possess a number of very interesting properties, both synchronically and diachronically. First, preverbs were characterised not only by their productivity, but also by their wide semantic range. For example, the prefix a- changes a durative activity such as river ‘to sail along the shore’, into an accomplishment ariver ‘to reach the shore’. The prefixation of a- in ariver adds an endpoint to the predicate, which then becomes a bounded event. As a result, the prefixed verb, unlike the bare verb, takes a new directional object, a port in example (9). (9)
Qui en mer entre ultreement se paine d’ariver a Who in sea enters afterwards REFL works to arrive at port, U ilh vuet port, where he wants ‘He who goes to sea struggles afterwards to get to the port where he wants to go’ Eles 517, cited in Tobler: 532
Moreover, the prefix a- can also change a durative activity such as parler ‘to talk’ into an inchoative: the verb aparler in (10) means ‘to address someone’ and thus the focus is on the onset of the event. (10)
II. homes de blans dras vestus Qui unt Anna si aparle´e two men of white clothes dressed who have Anna thus addressed ‘two men dressed in white that have addressed Anna’ Wace, Marie 28, cited in Tobler: 427
Preverbs and particles in Old French
37
Finally, when the prefix a- is added to a continuous durative such as emplir ‘to fill’, the verb acquires an intensive reading and means to ‘to fill up, to fulfil’, as in (11). (11)
Ademplir voeill vostre cumandement. To fill up want (pro) your command ‘I want to carry out your orders’ Roland, XXII, 30
The examples in (12) through (16) show that other prefixes can also have a wide semantic range. For example, the prefix par-4 can be either used as an intensive (12), or transform an activity into an accomplishment (13). (12)
amer destraindre
‘to love’ paramer ‘to love with passion’ ‘to hold (tight)’ pardestraindre ‘to hold tight’ Dictionnaire de l’ancien franc¸ais, Larousse: 439–440
(13)
atendre dire
‘to wait’ paratendre ‘to wait until the end’ ‘to say’ pardire ‘to finish saying’ Dictionnaire de l’ancien franc¸ais, Larousse: 439–440
As for the prefix con-, it can have either an intensive reading (14), or be used to modify the argument structure of a predicate5 (16) (14)
brisier ‘to break’
combrisier
geter
congeter
‘to make someone go/ ‘to throw away’ tencier ‘to quarrel/to blame’
‘to break completely/ to destroy’ ‘to banish’
contencier ‘to fight/to make war’ Examples from Hamacher (2002: 74)
(15)
E sor il pissa li goupiz. And on him urinated the fox. ‘And the fox urinated on him.’ Fables de Marie de France, XIV, 16, cited in Hamacher (2002: 101)
(16)
Et conpissa toz mes loviaus and urinated all my wolf-cubs. ‘And he urinated on all my wolf cubs.’ Le roman de Renart, Branche I, v. 37, cited in Hamacher (2002: 101)
3.1. The loss of productivity of aspectual prefixes Diachronically, a change took place in the system. In earlier work (Dufresne, Dupuis and Longtin (2001), Dufresne and Dupuis (2001)), we showed that
38
M. Dufresne, F. Dupuis and M. Tremblay
aspectual prefixation decreased over the centuries and is no longer productive in Modern French. Tables 1 and 2 illustrate the decrease in the creation of new verbs formed with the prefixes a- and sur- between the 13th and the 20th centuries.
Table 1: Decrease in the productivity of the prefix a- (DDL 2001, DDT 2000)
Table 2: Decrease in the productivity of the prefix sur- (DD 2001)
Interestingly, not all aspectual prefixes disappeared at the same time. While our preliminary study of the prefix a- showed a sharp decrease in productivity during the 16th century, a recent analysis of the prefix con- (Hamacher 2002) dates the loss of productivity of the prefix during the 13th century. It is worth noting that, while most of this morphological process is no longer available in the language, the prefixes themselves did not completely disappear, due to the lexicalisation of some prefixed verbs. In Modern French, detection of this morphological process is only possible when both members of a pair of verbs are still in existence. Thus the verb arriver is no longer perceived as a morphologically complex verb, because river no longer exists, and neither are the verbs investir ‘to invest’, de´valer ‘to tear down’ or avaler ‘to swallow’. On the other hand, the verbs endormir, apporter, parfaire and de´battre are still perceived as derived forms, since they can be contrasted with the bare form. ≠ ar+river ‘to arrive’ ≠ in+vestir ‘to invest’ ≠ de´+valer, a+valer ‘to tear down, to swallow’
(17)
a. b. c.
arriver investir de´valer/avaler
(18)
a. b. c. d.
dormir/en+dormir porter/ap+porter faire/par+faire battre/de´ +battre
‘to sleep / to put someone to sleep’ ‘to carry / to bring’ ‘to do / to perfect’ ‘to beat / to debate or struggle’
To summarize, while prefixes in Old French were frequently used to affect the aktionsart of the verb, in Modern French they are no longer used productively.
Preverbs and particles in Old French
39
3.2. Non-separable prefixes Buridant (2000) points out that Old French prefixes belong to one of two classes: non-separable prefixes such as a-, con-, de-, en-, entre-, and separable prefixes such as par-, re-, and possibly tres-. In the next two sections, we provide further arguments in favour of this distinction, despite the fact that both types of prefixes share similar properties. In Old French, non-separable prefixes behave like regular prefixes: they always appear attached to the verb and form a syntactic unit with the verb. Numerous arguments show that prefixed verbs in OF are formed in morphology, and are thus syntactic atoms. The first argument uses verb movement: non-separable prefixes remain attached to the verb when the verb undergoes movement to fulfil the verb second requirement as in (19). (19)
Molt aloient tost li ceval, / Much went immediately the horses-NOM / si s’entrefierent li vasal / ... thus SE-entre-fought the vassals-NOM / ... ‘The horses rushed forward immediately / thus the vassals fought each other ...’ Le bel inconnu: 81
Moreover, in Old French, infinitive verbs, unlike finite verbs, could also be moved to occupy the first position, in which case it is the whole verb (prefix+stem) which is moved, as illustrated in (20). (20)
Ademplir voeill vostre cumandement. To-fulfil want-I your order ‘I want to carry out your orders.’
Roland (XXII, 30)
The second argument is morphological: prefixed verbs can be nominalised, for example departir in (21). Given that derivation processes only apply to morphological objects, such examples indicate that the prefixed form can be the input of a word formation process. (21)
Au departir li redemande, / La belle Yseut, anuit At-the departure to-him RE-ask / The beautiful Yseut day viande food ‘Before leaving / Beautiful Yseut asks him food for the day’ Be´roul: 121
40
M. Dufresne, F. Dupuis and M. Tremblay
The last argument is lexical. As shown in Di Sciullo (1996, 1999), prefixation can affect the argument structure of the predicate. For example, Dufresne, Dupuis and Longtin (2001) have shown that in Old French the prefix a- could not only change an imperfective verb into a perfective verb, but could also modify the argument structure of the verb. A comparison of the argument structure of porter and apporter in five texts in prose from the 13th century shows that only 38% of the 231 occurrences of the verb porter appear with a locative or a dative argument, whereas this percentage increases to 81% for the 151 occurrences of the prefixed form. In short, the presence of a dative argument, optional with porter, becomes compulsory with aporter. This contrast is shown in (22) and (23). (22)
(23)
... il porta un escu vermeill et couvertures autretelles ... he wore a shield red and blankets matching ‘... he was holding a red shield and matching blankets’
Artu: 27
Lors demande ses armes, et len then (pro 3ps) asks his arms, and someone (object pro 3pp) li aporte maintenant to-him brings immediately ‘He then ask for his arms and someone brings them to him immediately’ QGraal: 94
Such prefixes appear to have been lexicalised with the verb, as the meaning of prefixed verbs is not always compositional. For example, the verb abaisser in (24) had two meanings, a compositional reading ‘to lower’ or ‘to pull down’, and a non-compositional reading ‘to pacify, to turn off’. Given that semantic drift is characteristic of morphological objects which have been lexicalised, the fact that prefixed verbs can undergo semantic change shows that they are lexical units. (24)
Maintien les bonnes coustumes de son royaume, et maintains (pro, 3ps) the good customs of his kingdom, and les mauvaises abaisse the bad eliminates ‘He keeps the good customs of his kingdom and eliminates the bad ones.’ Joinville, cited in Dictionnaire Larousse d’ancien franc¸ais: 2
To conclude, the aspectual morphemes a-, de-, en-, entre- and con- behave like real prefixes in Old French. The prefixed verbs are formed in morphology, where they can undergo subsequent morphological processes such as change of category, or affect the argument structure of the predicate. Prefixed verbs can also become lexicalised and, as lexical items, can be subject to semantic drift. Finally, the prefix of such verbs cannot be separated from the stem by verb movement.
Preverbs and particles in Old French
41
3.3. Separable prefixes At first, the separable prefixes par- and re- appear to behave like other prefixes.6 Semantically, they also affect the aspectual value of the predicate. For example, a verb prefixed with par- can acquire an intensive interpretation, as in paraimer meaning ‘to love with passion’, or it can result in the durative aspect of a verb becoming terminative, as in pardire ‘to finish saying’ and in parattendre ‘to wait until the end’. As for the prefix re-, it can indicate repetition, going back to a previous position, iterativity (i.e. the existence of a similar process), or intensive aspect.7 The prefixes par- and re- also appear to behave like other prefixes with respect to verb movement: they directly precede the verb as in (25) and move with the verb to satisfy the verb second requirement (26). When prefixed to an infinitive form as in (27), they can move with it to occupy the first position. Finally, they can also be part of a nominalised infinitive verb, as in (28). (25)
a.
et la fains tant le partormente and the hunger much him PAR-torment ‘and hunger torments him so much’ Renart II, 3374, cited in Buridant (1995)
b. Et li doi cierge s’en revont / Ariere el palais dont And the two candles return / back to-the palace from-which il sont they are ‘And the two candles return to the palace they come from’ Partonopeu, 1097–98, cited in Buridant (2000, § 441) (26)
(27)
a.
le mers reportoit le nef ariere the sea pushed the ship back
Clari: 73
b.
et s’en revient messire Gauvains a Kamaalot, ... and CL-CL return Messire Gauvains_NOM to Camelot Artu: 73
c.
La sus amunt parjettent tel luiserne On-it above PAR-throw such light
a.
Retorner voldrent arriere en lor contre´e return want-3pp back in their country ‘They want to return to their country’
Roland, CXC, 2634
AmiAmil: 112
b. ... / Ki parjurer les an lairoit, / ... / who to-be-a-traitor-to-one’s oath them of-it let-COND, / ... / who would let them betray their oath about it./ Escoufle: 105
42
(28)
M. Dufresne, F. Dupuis and M. Tremblay
a.
Ainz qu’il fuissent al pareissir del gue´ Before that they were at the PAR-exit of-the ford Guillaume, 2772, cited in Buridant (1995)
b. Au redrescier i corrent plus de cent At-the get-up, to-her run more than hundred ‘As she gets up, more than one hundred people run to her’ AmiAmile: 100 c.
Au revenir, dist: «La vostre ame, ... At-the returning, he-says: ‘‘Your soul ...
Escoufle: 85
Interestingly however, the prefixes par- and re- differ from other prefixes in that they can also undergo tmesis, i.e., they can appear separated from the verbal stem they modify. First, they can be separated from the verb by an auxiliary. In example (29), re is separated from the past participle eu ‘had’ by the auxiliary a ‘has’ and by the direct object enfant ‘child’. In (30), par is separated from the past participle dist by the auxiliary avez ‘have’. (29)
Ma femme ra enfant eu¨ My wife re-has child had ‘My wife had another child’ La complainte, Rutebeuf, 9698, cited in Gosselin (1999)
(30)
Vos en par avez dist trop mal You of-it PAR AUX said too-much bad ‘You were blathering’ Renart, XIV, 14715, cited in Moignet (1979)
Moignet (1979) notes that par and re can also be separated from the verb by a clitic, such as the pronoun i ‘there’ in (31) and (32). (31)
Ne cuidiez ja qu’i par i viegne Not you-believe never that he PAR there come ‘Never believe that he will come’ Eracle, 5604, cited in Moignet (1979)
(32)
Uns r i vint qui ravoit perie One re there came that re-had perish ‘One who had perished again came back there’ La vie de saint Martin, 5573, cited in Gosselin (1999)
In fact, par- and re-, when separated from the verb, seem to become part of the clitic cluster. The most striking argument in favour of such an approach comes from restructuring constructions. It is a well-known fact that in Old
Preverbs and particles in Old French
43
French, as in many other Romance languages such as Italian and Spanish, an object pronoun can appear before a modal verb.8 In example (33), the clitic la appears before the modal verb porent. (33)
Au chief de la sale ot un banc / ou la dame s’ala In front of the room was a bench / where the lady CL-went seoir / la ou tuit la porent veoir ... sit-INF / there where all her can-3PP see, ... ‘At the front of the room, there was a bench where the lady went to sit, and where all could see her’ Ch. Lyon, 2072–2074
Interestingly, the distribution of par- and re- appears to follow the same rule. For example, in (34) from Moignet (1979) the preverb par- appears before the modal verb puet ‘can’, despite the fact that it modifies the verb ataindre ‘to reach’. In examples (35) and (36), from Morin and Saint-Amour (1977: 148), the prefix re- is found before the modal verb devroies ‘should’ and before the factive verb fet ‘made’. (34)
et si ne le par puet ataindre and then not him PAR can reach ‘and he could not easily reach him’ Ch. Lyon, 887, cited in Moignet (1979: 130)
(35)
Tu me redevroies dire you me re-should tell ‘You should tell me again’
(36)
Ch Lyon 356 (Tobler VIII 365, 15), cited in Morin and Saint-Amour (1977: 148)
Une dolors ... lors refet lor joie oblier a sadness them re-made their joy forget ‘A sadness made them forget their joy again’ Ch. Lyon 3819 (Viii 365, 17), cited in Morin and Saint-Amour (1977: 148)
Finally, the preverbs par- and re- behave like clitics in that they are never found in first position of a V2 clause, and they do not count as an independent word with respect to the V2 requirement, as shown in (37). (37)
Li chevaliers par fut tant biax The knight PAR was very handsome.
Perceval: 74
To conclude, while the examples in (25) to (28) seemed to indicate that parand re- are prefixes, the examples in (29) to (37) showed that par and re are
44
M. Dufresne, F. Dupuis and M. Tremblay
sometimes independent from the verb although remaining part of the clitic cluster. Based on the above data, we argue that par- and re- are involved in two distinct processes. On the one hand, just like the non-separable prefixes, they are part of a morphological process, in which a prefix forms a compound with the verb. On the other hand, par- and re- can undergo tmesis and thus behave more like aspectual morphemes9 independent from the verb, very much like preverbs in Germanic languages as noted by Booij and Van Kemenade (this volume). However, separable prefixes are unlike Germanic preverbs in that they cannot be stranded by movement of the finite verb, a property which also sets them apart from the system of particles coexisting in the language, as argued below.
4. PARTICLES IN OLD FRENCH According to Buridant (2000), Old French had a system of separated particles (locative adverbs) which could modify the meaning of the verb. A non-exhaustive list of particles includes arriere, sus, ensus, amont, contreval, avant, devant, aval, hors, fors, and jus.
4.1. Particles are not affixes Particles are unlike prefixes, whether separable or non-separable, in that they do not appear to form a word with the verb. Words, including compounds, are syntactic atoms in the sense that syntactic rules cannot target parts of words. The examples in (38) illustrate the phenomenon: the two components of the compound word essuie-glace ‘wiper’ cannot be separated from the verb by an adverb, or by movement such as caused by passivisation of the verb. (38)
a.
*essui-bien-glace wipe well windshield
b. *[glace-essui-bien] windshield wipe well c.
*Glace a e´te´ re´pare´ cet [essuie e] par Jean. windshield has been repaired this wipe by Jean Examples from Di Sciullo and Williams (1987: 80)
In the case of particles, however, the fact that both the verb and the particle can move independently shows that both items have independent morphological status. First, Old French particles, like their Germanic counterparts, can be stranded by verb movement. This is illustrated in (39) to (44). In (39) to (42),
Preverbs and particles in Old French
45
the finite verb moves to the first or second position to satisfy the so-called verb second requirement, while in (43) and (44) an infinitive verb moves to the first position. In both cases, the particle remains stranded within the verb phrase. (39)
Gitez en donc ma part sus Throw of-it thus my share down ‘Thus throw my share of it down.’ RenartR, IIIb, 4998, from Buridant (2000, § 4400)
(40)
Voist s’en arriere sans nule demoree Goes REFL-EN back without no stay ‘He goes back immediately’ Aliscans, 1949, from Buridant (2000, § 441)
(41)
Ains descent tout maintenant jus de son ceval ... So descends all now down from his horse ... ‘He then dismounts from his horse ...’
Trispr: 127
(42)
Is t’an fors contre moi Exit you of-it outside against me ‘Get out from there and confront me’ Saisnes, L, 3637, cited in Buridant (2000, § 441)
(43)
... retorner voldrent arriere en lor contree ... return want-3PP back in their country ‘... they want to go back to their country’
(44)
AmiAmile: 112
Et si recrut en tel maniere / And so be-exhausted in such way / Qu’aler ne pot avant n’arriere That go-INF NEG can forward nor backward ‘He was so exhausted that he was unable to go any further or back’ Guillaume A, 789–90, cited in Buridant (2000, § 440)
Moreover, the particle can move by itself to the first position,10 leaving the verb in place, as illustrated in (45) to (47): (45)
Je suis mehaingnier / Je ne puis mes avant aller / I am wounded / I NEG can more forward go / Arrier me covient retorner Back me suit return ‘I am wounded / I can’t go on / I have to go back’ RenartR, IX,9238–38b, cited in Buridant (2000, § 440)
46
M. Dufresne, F. Dupuis and M. Tremblay
(46)
Sus le voldrent faire lever On him-ACC wanted-3PP make stand ‘They wanted to force him to get up’ VieSGre´gI, ms. A2, 2411, cited in Buridant (2000, § 440)
(47)
Puis comande qu’an li amaint / Then command that one him-DAT brings / son cheval, et l’an li amainne; / is horse, and someone him-DAT brings; / sus est sailliz de terre plainne on is arised from ground flat ‘He ordered that his horse be brought to him, and someone brings him his horse; He gets on his horse.’ Erec: 22
Example (48) shows that it is possible to co-ordinate two particles, while example (49) shows that it is possible to cumulate two particles (sus and en contremont): (48)
(49)
Assez le quistrent sus et jus Much him-ACC seek-3PP up and down ‘They look for him everywhere.’
Artu: 38
Il resaut sus en contremont He re-jumps up upwards ‘He jumps up in the air.’ Gormont, 279, cited in Buridant (2000, § 440)
To conclude, we have seen that the complex predicate verb+particle does not form a syntactic atom. In the next two sections, we will see that, while the verb and the particle do not form a compound word, they still appear to form a lexical unit. 4.2. On the distribution of particles While particles are syntactically autonomous, their distribution in non-V2 clauses shows that they are still closely related to the verb. Interestingly, if one considers such sentences, one notes that the particle often appears adjacent to the verb. Although in most cases (in fact, mainly in 13th century prose), the particle follows the verb, as illustrated in (50) to (52), there are a number of cases (mainly in 12th century verse) in which the particle precedes the verb (53 to 55).11 (50)
... quant les Griu les virent traire ariere, ... ... when the Greeks them-ACC saw-3PP draw back, ... ‘... When the Greeks saw them withdraw, ...’
Clari: 70
Preverbs and particles in Old French
(51)
Quant il furent bien ale´ avant en le mer ... When they were-3PP well gone forward into the sea ‘When they advanced well into the sea ...’
47
Clari: 20
(52)
.. que che ne fu se menchoingne non que on ... that this NEG was either a lie which (pro-3PS) him le avoit mis sus had not put on ... ‘... that it was either a lie which he had been accused of ...’ Clari: 20
(53)
Voldrent le faire sus lever / Want-3PP him-ACC make up stand / Mais il ne pot sor piez ester But he NEG can on foot be ‘They wanted him to stand up / But he couldn’t stand on his feet’ VieSGre´g1,ms.A1,2537–38, cited in Buridant (2000, § 440)
(54)
Li se¨aus est jus avalez The pail is down taken ‘The pail is taken down’
(55)
RenartR, II, 3641, cited in Buridant, § 440
Ou bois se sont arriere remis Or wood REFL are back went ‘They went back into the woods’ RenartR, VI, 5440, cited in Buridant (2000, § 440)
Examples in which the particle is found before the verb appear to put particles on a par with other preverbs, i.e., (separable and non-separable) prefixes. However, such examples cannot be considered representative of the canonical word order, since they are found mainly in poetry (12th century verse), where additional stylistic constraints such as rhythm, rhyme and caesura may come into play. In prose, the distribution of particles appears more stable: a preliminary analysis of the distribution of particles sus and arriere with respect to the infinitive verb or the past participle (which do not undergo V-movement) reveals that the particle is clearly postverbal in most cases.
Table 3: The distribution of particule sus in our 13th century prose corpus
48
M. Dufresne, F. Dupuis and M. Tremblay
Table 4: The distribution of particule arriere in our 13th century prose corpus
The fact that particles mostly follow the verb in the prose corpus constitutes a strong argument in favour of a distinct analysis of preverbs and particles. This conclusion is supported by the observation that the distribution of particles is generally quite different from that of other preverbs. First, particles do not appear to be part of the clitic cluster (they can occupy the first position, they never appear in clitic position, etc.). Second, particles usually appear after direct objects (56a, b), but before indirect objects (57a, b). (56)
a.
... qu’il est menterres de metre tel chose avant ... that it is lie OF put such thing forward. ‘... that it is a lie to say such a thing.’ Artu: 37
b. le mers reportoit le nef ariere the sea RE-brought the ship back ‘the sea pushed the ship back’ (57)
a. Si keurent il sus as Franchois If run they-NOM up to-the French’ ‘If they attack the French’ b. ... retorner voldrent arriere en lor contree ... return want-3PP back in their country ‘... they want to go back to their country’
Clari: 74
Clari: 106
AmiAmile: 112
Although we found a few examples where a particle appears before a direct object (58), or after an indirect object (59), such examples were clearly not quantitatively significant. (58)
a. Que l’en li amaint avant la reı¨ne that one him-DAT bring forward the queen. ‘That someone bring the queen before him’ b. Turnez ariere les estrees Turn back the roads ‘Turn around and go back’
Artu: 122
Gormont: 32
Preverbs and particles in Old French
(59)
a.
49
A grant joie fu remenez / devant l’empereor arriere, To great joy was RE-brought before the emperor back ... ‘To her great joy, she was brought back before the emperor ...’ Dole: 153
b. ... et tant fist puis k’il vint en la Petite ... and such (pro 3ps) did since that he came in little Bretaingne ariere. Brittany back ‘... and he did so much that he came back to Little Brittany.’ Trispr: 232 Obviously, an individual quantitative description of the distribution of each particle is required before a detailed analysis of Old French particle constructions can be provided. A few conclusions can be reached however. First, while section 4.1 showed that particles are autonomous lexical items, section 4.2 showed that they are base-generated postverbally within the verb phrase. This conclusion is supported by the tendency observed for most indirect objects to appear after the particle.12 In the next section, we will see that there is further morphosyntactic evidence showing that the particle is within the lexical domain of the verb, and forms a complex predicate with it.
4.3. On the lexicalisation of particles Syntactic operations target constituents. Interestingly, we found a limited number of examples where the subject seems to follow the complex predicate. This is illustrated in (60) and (61). In (60), the subject Girflez bears nominative case, and follows the complex predicate saut avant. Similarly in (61), the subject li rois, also in nominative case, follows the complex predicate cort sus. (60)
(61)
Lors saut avant Girflez et dist ... Then jumps forward Girfle´ and says ‘Then Girfle´ comes forward and says ...’
Artu: 319
Lors li cort sus li rois ... Then him-Dat run on the king-NOM ‘Then the king attacks him ...’
Artu: 152
The position of the subject with respect to the verb can be accounted for by verb movement: the adverb lors occupies the Topic position, i.e., the first position, and the complex predicate moves to occupy the second position. The fact that the complex verb moves as a whole indicates that it is a syntactic unit (a constituent).
50
M. Dufresne, F. Dupuis and M. Tremblay
The example in (62) is even more convincing as the complex predicate precedes both the subject Messire Gauvains and the indirect object de son ostel. In this case, the position of the verb with respect to the subject cannot be explained by a rule of subject postposition. (62)
A ces criz et a ces noises issi hors Messires At these cries and at these noises, came out Messire Gauvains de son ostel ... Gauvains-NOM from his house ‘When he heard the cries and the noise, Messire Gauvain came out of his house ...’ Artu: 130
We also considered the order verb+particle with respect to adverbs. In most cases, the adverb precedes the particle. Assuming Cinque’s (1999) analysis according to which adverbs occur in the specifier positions of functional categories to the left of the VP, this leads us to conclude that the particle is probably generated within the VP.13 (63)
Il se trait un poi ariere et lour dist ... He-NOM REFL withdraw a bit back and them-DAT says ... ‘He steps back a little and tell them ...’ Trispr: 137
(64)
... je lour courui esranment sus ... ... I them-DAT ran immediately sus ‘... I immediately dashed after them ...’
Trispr: 62
We also found a few examples (such as 65 below) in which the whole complex predicate (verb+particle) precedes the adverb. Such examples can be analysed in two ways: either the complex predicate has moved as a whole (as a result of verb movement), or the adverb has been extraposed. The fact that such examples involve fairly large adverbs argues in favour of the latter hypothesis. (65)
... qu’il se trest avant plus et plus. ... that he-NOM REFL goes forward more and more. ‘... that he go forwards more and more.’
QGraal: 85
The lexicalisation of complex predicates constitutes another strong argument in favour of the hypothesis that particles form a complex predicate with the verb. First, as shown in Di Stefano (1991), there are a number of examples where the verb and the particle form a semantic unit.
Preverbs and particles in Old French
(66)
a. b. c. d.
Corre sus a` Metre sus a Metre avant Metre arrere
51
‘to attack’ ‘to accuse’ ‘to produce an argument’ ‘to neglect, to discredit’
The presence of a particle, just like that of a prefix, can affect the argument structure of the predicate. Tremblay (2000) has shown that, in the case of the lexicalised complex verb corre sus, the presence of the particle sus has an effect on the argument structure of the verb corre: while the verb corre takes a dative argument in just 50% of occurrences, this percentage increases to 100% when the particle is present.14
Table 5: Corre vs. corre sus in the 13th century corpus
Examples are still found in Modern French, where we find a dative clitic in very similar constructions.15 (67)
a.
Paul court apre`s Sophie. Paul runs after Sophie.
b. Paul lui court apre`s. Paul runs after her. There is, however, an important distinction to be made between Modern French and Medieval French. First, in Modern French, there is almost a complementary distribution between a simple preposition, such as sur, sous, and dans, and the corresponding complex preposition dessus, dessous, dedans, respectively. This observation leads Roberge (1998) to propose that only the former have Case properties. (68)
a.
Paul est tombe´ sur Sophie. Paul is fell on Sophie ‘Paul ran into Sophie / Paul fell on Sophie’
52
M. Dufresne, F. Dupuis and M. Tremblay
b. Paul lui est tombe´ dessus. Paul DAT-Cli is fallen on pro ‘Paul fell on her / Paul set on her.’ In Old French, the situation is quite different since most particles can also function as prepositions. Examples such as (69) and (70) indicate that the preposition sus can assign Case to its complement: sus sa poitrine, sus son lit. (69)
Sus sa poitrine tenoit ses mains croisant On his chest kept his hands crossed. ‘He was keeping his hands crossed on his chest’ Aliscans, 827, cited in Buridant (2000, § 388)
(70)
Si s’est assise sus son lit If REFL is sitted-FS on her bed. ‘She sat on her bed’ FabliauxNR, II, 10, 299, cited in Buridant (2000, § 388)
When sus is used as a particle, however, it loses its argument structure and its ability to assign Case.16 (71)
(72)
Si corent sus au chevalier (pro 3pp) run (part.) toward the knight ‘they pursue the knight’
Artu: 223
Et toutevoies recort il sus au serpent ... However, he attacks (part) again the snake ... QGraal: 94
(73)
Je lour courui esranment sus ... I them pursued immediately (part) ...
Trispr: 62
In fact, such examples lead us to the conclusion that Old French particles are true prepositions, as opposed to adverbs as proposed in Buridant (2000). His conclusion is supported by a number of observations. First, all particles have a prepositional counterpart. Second, we have seen that particles are generally base-generated postverbally, and thus have a distribution quite different from that of adverbs, which are preverbal. Finally, contrary to adverbs, particles appear to form a complex predicate with the verb they modify. Even though this complex predicate is not a compound word, we have seen that the presence of the particle may affect the argument structure of the predicate, a property which sets them apart from adverbs as adverbs do not have such an effect on the verb.
Preverbs and particles in Old French
53
5. TOWARDS A UNIFIED ANALYSIS OF PREVERBS AND PARTICLES To summarise the discussion so far, we have considered three types of aspectual elements in Old French: non-separable prefixes, separable prefixes, and particles. We reached two conclusions. On the one hand, we have argued that the distinction proposed in Buridant (2000) between prefixes and particles is justified on morphosyntactic grounds. Non-separable prefixes always form a word with the verb and can affect the argument structure of the predicate. Separable prefixes can either form a word with the verb, or behave like independent aspectual heads. Unlike non-separable prefixes, they do not appear to modify the argument structure of the predicate. Particles have a completely different morphosyntactic status: while they form a complex predicate with the verb, they do not form a true ‘‘word’’ with it, at least in most cases. On the other hand, our study has also showed that the distinction between prefix and particle is not as clear-cut as previously thought. In fact, as we look more closely at the different types of prefixes and particles, the distinction becomes less sharp: prefixes sometimes behave like particles, and particles share a number of characteristics of prefixes. Thus, rather than a dichotomy between prefixes and particles, it seems as if we are dealing with a morphosyntactic continuum with non-separable prefixes at one end and separated particles at the other.
The existence of such a continuum constitutes an argument in favour of a unified analysis of prefixes and particles. In fact, this appears desirable given the considerable evidence in favour of such an approach. First, both prefixes (Nyrop 1899, Buridant 1995, Di Sciullo 1996, Wilmet 1998) and particles are prepositional. Second, prefixes and particles share similar semantic characteristics: both can be locative and/or aspectual: (74)
parler ‘to talk’ aparler ‘to start talking’ parler avant ‘to continue talking’
(75)
venir revenir aller ariere
‘to come’ ‘to go back’ ‘to go back’
The third argument is diachronic. In our first study on prefixation (Dufresne, Dupuis and Longtin 2001), we dated the loss of productivity of prefixation back to the end of the 16th century. Interestingly French lost its
54
M. Dufresne, F. Dupuis and M. Tremblay
particle system at about the same time (Buridant 2000). Fourth, some particles can also be used as prefixes17 (Buridant 2000). (76)
a.
forjoster forjugier forjurer
b. fors aller issir fors (77)
a.
‘to hunt’ ‘to sentence’ ‘to give up’ ‘to go’ ‘to go out’
horbannir
b. oster hors chacier hors
examples from Buridant (2000, § 439)
Finally, the prefix system seems to sometimes interact with the particle system. For example, the separable prefix re- can be reinforced with the particle arriere, without any change in meaning. (78) retorner, torner arriere, retorner arriere ‘to return’ This phenomenon appears limited to cases where the re- has a locative meaning, as in (79). (79)
Quant Melyans l’entent, si retorne arriere ... When Melyan him hears then (pro 3ps) returns back ‘When Melyan hears him he comes back’ QGraal, 42,3 cited by Gosselin (1999)
Similarly, the prefix par is often used with an intensive adverb. This is illustrated in (80), (81) and (82), where par appears with mult, tant, and trop: (80)
Devers un gualt uns granz leons li From a forest a large lion to him Mult par ert pesmes e So (part) (pro 3ps) appears nasty and ‘From a forest a large lion comes toward and cruelty’
vient, / comes / orguillus e fiers, / arrogant and cruel him full of rage, hardiness Roland, CLXXXV, 2550–1
(81)
Sur lui se pasmet, tant par est anguissus On him (pro 3ps) faints full (part) (pro 3ps) is anguish ‘He faints because he his so full of anguish Roland, CCV, 2880
Preverbs and particles in Old French
(82)
Eracles, trop par ies estranges Eracle too much (pro 2ps) are strange ‘Eracle, you are so strange ...’
55
Eracle: 192
In our 12th century corpus, out 162 occurrences of particle par, only 7 involved no adverb. This kind of doubling phenomena is generally symptomatic of semantic erosion and is thus probably a first sign of the loss of productivity of the aspectual prefix system. 6. CONCLUSION To conclude, a unified analysis of Old French prefixes and particles sheds doubt on the status of preverb as a primitive in the grammar. First, there does not appear to be a uniform class of preverbal elements which can be analysed independently of postverbal elements. Second, ‘‘preverb’’ is a mixed grammatical object which can be involved in complex predicate formation occurring variously in the lexicon, in morphology or in syntax. Finally, ‘‘preverb’’ cannot be conceived as a categorical primitive, distinct from preposition. Hence, given that preverb is not a theoretical construct, there cannot be a theory of preverbs per se. Rather, assuming that preverbs (and particles) are a subclass of prepositions, albeit intransitive, the grammatical and semantic properties of preverbs must follow from a broader theory of lexical categories. NOTES * We wish to thank the participants of the Workshop on Preverbs held at the University of Nijmegen in January 2001 as well as an anonymous reviewer for useful suggestions and remarks. Part of this work was also presented at the following conferences: PREP 2000, LSRL, Going Romance and ICHL. The research in this paper was partially supported by the Social Science and Humanities Research Council of Canada (grants # 410-96-1445 and # 214-97-0016) and by Queen’s University’s Advisory Research Committee. We also wish to thank E. Hamacher for proofreading the text. 1 For a detailed description of the verb second properties of Old French, see Adams (1987), Dupuis (1990), Thurneysen (1892), and Vance (1997). 2 Old French distinguishes between two genders (feminine and masculine), two numbers (singular and plural) and two morphological cases: nominative case (cas sujet) and object case (cas re´gime). Nominative case is usually marked on masculine nouns, while most feminine nouns do not distinguish between the two morphological cases.
Table I: Declination of a masculine noun ‘the wall’
56
M. Dufresne, F. Dupuis and M. Tremblay
Table II: Declination of a feminine noun ‘the girl’ 3
More recent work such as Koopman and Szabolcsi (2000) proposes to dispense with headmovement in favour of phrasal movement. This theoretical distinction is not relevant here. 4 See Label (2003) for discussion. 5 See Hamacher (2002) for a thorough discussion of the evolution of the semantic and morphosyntactic properties (including productivity) of the prefix con- from Latin to Old French. 6 The figure of the 20th century in Table 2 is explained by a new meaning of the prefix: it has acquired the same meaning ot the iterative external prefix re- (contrer to double, surcontrer to redouble) still very productive in ModF. 7 Note however that contrary to non-separable prefixes, the separable prefixes par- and redo not appear to affect the argument structure of the verb. 8 See Lebel (2003) and Barbaud and Lebel (2000), for a detailled description of the semantic and distributional properties of the prefix par-. As for the prefix re-, the reader should consult Gosselin (1999). 9 Retructuring is no longer possible in Modern French, although Grevisse (1980, § 549) gives examples from literary French showing that the construction has not completely disappeared from the language. 10 We assume that these aspectual morphemes are base-generated under functional projections above VP, possibly ASPP. 11 The position of the particle to the left of the verb should not be attributed to tmesis, which is a more local process and is not the outcome of verb movement. 12 A similar distribution has been observed for the borrowed particle back in Acadian French (see King 2000).
13
i. a.
Puis je voulais pas back aller. And I want-IMP NEG back go ‘And I didn’t want to go back’
ii.
Tu peux aller leur dire back. you can go them-DAT back. ‘You can tell them again.’
b.
(=King 2000: examples 5 and 10)
The tendency observed for most direct objects (and some indirect objects) to appear between the verb and the particle can probably be motivated on Case-theoretic grounds: in the Minimalist Program (Chomsky 1995), movement of the direct object is triggered by feature checking requirements. 14 See however Tortora (2002) for an alternative proposal. Tortora provides evidence from Borgomanerese showing that at least some argument prepositions may move from their base positions within the verb phrase to a functional projection above VP. 15 A further argument showing that the complex predicate corre sus has been lexicalised involves the prefix entr-, derived from the preposition entre ‘between’. In Old French, entr-
Preverbs and particles in Old French
57
was used, together with the reflexive clitic se, to indicate reciprocity, and had an effect on the argument structure of the predicate. (1)
Un chevaler de grant valour / Et une dame de honour / A knight of great worth / And a lady of honour / S’entre amerent jadis d’amour SE-entre-love in-times-past of-love ‘A knight of great worth and a lady of honour loved eachother’. FabliauxNR, IX, 113, 5–7, cited in Buridant (2000: 493)
A similar example involves the complex predicate corre sus, which receives a reciprocal interpretation when prefixed with entr-. In this case, the prefix is interpreted as taking scope over the whole complex predicate, and not just the verb. (2)
Il metent les mains as espees et s’entrecourent sus They-NOM put-3PP the hands to-the swords and SE-between-run up mout ireement much angrily. ‘They take their swords and attack each other with anger.’ Trispr: 214
16
For a detailled description of dative clitics in such constructions, see Gaatone (1983). According to Zeller (2001), the non-case-assigning properties of particles follows from their lacking functional structure. 18 It is interesting to note that the prefixes/particles for and hors are both of Germanic origin (Nyrop, cited in Buridant 1995). The fact that a single element can have been introduced in the language in both morphology and syntax consitute an additional argument in favour of a unified analysis of prefixes and particles. 17
REFERENCES Adams, M. (1987). From Old French to the Theory of Pro Drop. Natural Language and Linguistic Theory 5, 1–32. Barbaud, P. and Lebel, M.H. (2000). Du pre´verbe PAR en ancien franc¸ais. Paper presented at the PREP 2000 Conference. Tel Aviv, Israel. Buridant, C. (1995). Les pre´verbes en ancien franc¸ais. In Rousseau, A. (ed.), Les pre´verbes dans les langues d’Europe: introduction a` l’e´tude de la pre´verbation, Universite´ de Lille III: Presses du Septentrion. Buridant, C. (2000). Grammaire nouvelle de l’ancien franc¸ais. Paris: Sedes. Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1986). Knowledge of Language: its Nature, Origin and Use. New York: Praeger. Chomsky, N. (1991). ‘Some notes on the economy of derivation’. In Frieden, R. (ed.), Principles and Parameters in Comparative Grammar. MIT Press. 417–454. Chomsky, N. (1995). The Minimalist Program. Cambridge: MIT Press. Cinque, G. (1999). Adverbs and Functional Heads: A Cross-Linguistic Perspective, New York: Oxford University Press. Di Sciullo, A-M. (1996). Configurations. Somerville: Cascadilla Press. Di Sciullo, A-M. (1999). Verbal structures and variation. In E. Trevin˜o and J. Lema (eds.), Semantic Issues in Romance Linguistics. Amsterdam/Philadelphia: Benjamins.
58
M. Dufresne, F. Dupuis and M. Tremblay
Di Sciullo, A-M. and Williams, E. (1987). On the Definition of Word. LI Monograph 14. Cambridge: MIT Press. Dufresne, M., Dupuis F. and Longtin C.M. (2001). Un changement dans la diachronie du franc¸ais: la perte de la pre´fixation aspectuelle en a-. Revue que´be´coise de linguistique 29, 33–54. Dufresne, M. and Dupuis F. (2001). On the consequences of the grammaticalisation of aspect in French. Paper presented at the Going Romance Conference, Amsterdam. Dufresne, M., Dupuis F. and Tremblay M. (2000). The role of syntactic faetures in historical change. In S. Dworkin and D. Wanner (eds.), New Approaches to Old problems. Issues in Romance Linguistics. Amsterdam: John Benjamins, 129–149. Dupuis, F. (1990). L’expression du sujet dans les subordonne´es en ancien franc¸ais. The`se de doctorat, Universite´ de Montre´al. Gaatone, D. (1983). Le de´agre´able dans la syntaxe. Revue romane 18. Universite´ de Copenhague: Institut d’e´tudes romanes,. Gosselin D. (1999). Une analyse en morphologie configurationnelle: le pre´fixe re- en ancien franc¸ais. Ph.D. dissertation, UQAM. Hale, W.G. and Buck, C.D. (1988). A Latin Grammar (8th ed.). The University of Alabama Press. Hamacher, E. (2002). La pre´fixation verbale en ancien franc¸ais: le recul d’un proce´de´ morphologique. M.A. dissertation, Queen’s University. King, R. (2000). The Lexical Basis of Grammatical Borrowing. Amsterdam: John Benjamins. Koopman, H. and Szabolcsi, A. (2000). Verbal Complexes. Cambridge: MIT Press. ´ cole Lebel, M.-E. (2003). Par dans l’histoire du franc¸ais. Ph.D. dissertation, UQAM & E normale supe´rieure. Moignet, G. (1979). Grammaire de l’ancien franc¸ais. Paris: Klincksieck, Morin, Y.-C. and Saint-Amour, M. (1977). Description historique des constructions infinitives du franc¸ais. Recherches linguistiques a` Montre´al. Montre´al: Universite´ de Montre´al. Nyrop, K. (1899) Grammaire historique de la langue franc¸aise. Copenhague: Picard. Roberge, Y. (1998). Les pre´positions orphelines dans diverses varie´te´s de franc¸ais d’Ame´rique du Nord, in Brasseur (ed). Franc¸ais d’Ame´rique: variation, cre´olisation, normalisation. Centre d’e´tudes canadiennes, Universite´ d’Avignon. Roeper, T. and Keyser, J.S. (1992). Re: the abstract clitic hypothesis. Linguistic Inquiry 23, 89–125. Thurneysen, R. (1892). Die Stellung des Verbums im Altfranzo¨sischen, Zeitschrift fur romanische Philologie 16, 289–307. Tortora, C. (2002). Romance Enclisis, Prepositions, and Aspect. Natural Language and Linguistic Theory 20, 725–758. Tremblay, M. (2000). Pre´positions, particules et interfaces. Paper given at the 4th Bilingual Workshop in theoretical linguistics. York University, Toronto. Tremblay, M., Dufresne, M. and Dupuis, F. (2002). Les pre´positions dans l’histoire du franc¸ais. Paper presented at Synchro-1 Conference, Paris. Vance, B. (1997). Syntactic Change in Medieval French. Dordrecht: Kluwer Academic Publishers. Vincent, N. (1999). The evolution of c-structure: prepositions and PPs from Indo-European to Romance. Linguistics 37, 1111–1153. Wilmet, M. (1998). Grammaire critique du franc¸ais. Paris: Duculot. Zeller, J. (2001). Particle Verbs and Local Domains. Amsterdam: John Benjamins.
Preverbs and particles in Old French
59
Dictionaries and Grammars Di Stefano, G. (1991). Dictionnaire des locutions en moyen franc¸ais. Grevisse, M. (1980). Le bon usage. Gambloux: Duculot Petit Robert I. (1991). Dictionnaire de la langue franc¸aise. Le Robert (1993). Dictionnaire historique de la langue franc¸aise. Larousse (1992). Dictionnaire de l’ancien franc¸ais. Tobler, A. and Lommatzsch, E. (1925). Altfranzo¨sisches Wo¨rterbuch Berlin: Weidmann.
Texts Consulted aliscans Aliscans. Ed. C. Regnier, Champion, 1990–1991 (CFMA 110–111).
amiamile Ami et Amile. Ed. P.F. Dembowski, Paris, 1969, Champion (CFMA 97).
artu La Mort le Roi Artu. Ed. Jean Frappier, Gene`ve, 1954, Droz.
beroul Tristan, Be´roul. Ed L. M. Defourques, Paris, 1947, Champion (CFMA 12).
ch. lyon Yvain, le chevalier au lion, Chretien de Troyes. Ed. M.Roques, Paris, 1960, Champion, CFMA (89)
clari La Conqueste de Constantinople, Robert de Clari. Ed. Philippe Lauer, Paris, Champion (CFMA 40).
dole Le Roman de la Rose ou de Guillaume de Dole, Jean Renart. Ed. Felix Lecoy, Paris, 1962, Champion (CFMA 91).
eracle Eracle, Gautier d’Arras. Ed. Guy Raynaud de Lage, Paris, 1976, Champion (CFMA 102).
erec Erec et Enide, Chretien de Troyes. Ed. Mario Roques, Paris, 1952, Champion (CFMA 80)
escoufle L’Escoufle, Jean Renart. Ed. F. Sweester, Gene`ve, 1974, Droz (TLF. 211).
fabliauxnr Nouveau Recueil Complet des Fabliaux. Ed. W. Noomen and Van Der Boogard, Van Gorcum, I, 1983; II, 1984; III, 1986; IV, 1988; V, 1990; VI, 1991; VII, 1993; VIII, 1994; IX, 1996.
feuille´e Le Jeu de la Feuille´e, d’Adam de la Halle. Ed. J. Dufournet, Gand, 1977, Story-Scientia.
gormont Gormont et Isembart, Ed. Alphonse Bayot (3rd Edition revue), Paris, 1931, Champion (CFMA 14).
guillaume La Chanson de Guillaume. Ed. D. Mc Millan, Paris, 1949–50.
60
M. Dufresne, F. Dupuis and M. Tremblay
joinville La Vie de Saint Louis, Jean de Joinville. Ed. J. Monfrin, Paris, 1996, Garnier Flammarion
le bel inconnu Le Bel Inconnu, Renaut de Beaujeu. Ed. G. Perrie Williams, Paris, 1967 (2e`me e´dition), Champion (CFMA 38).
partonopeu Partonopeu de Blois. Ed. J. Guildea, Villanova, 1967.
qraal La Queste del Saint Graal. Ed. A. Pauphilet, Paris, 1923, Champion (CFMA 33).
renart Le Roman de Renart. Ed. M. Roques, Champion, 1948–1960.
roland La Chanson de Roland. Ed. Ge´rard Moignet, Paris, 1970, Bordas.
saines La Chanson des Saines, de Jehan Bodel. Ed. A. Brasseur, Droz, 1989, TLF, 369.
trispr Tristan en prose (tome 1), Ed. Philippe Me´nard, Gene`ve, 1987, Droz (TLF 353).
viesgre´g1 La vie du pape saint Gre´goire. Huit versions franc¸aises me´die´vales de la le´gende du Bon Preˆcheur. Ed. H.B. Sol, Rodopi, 1977.
Monique Dufresne De´partement de linguistique et de didactique des langues Universite´ du Que´bec a` Montre´al Montre´al, QC, Canada H3C 3P8 dufresne.monique @uqam.ca Fernande Dupuis De´partement de linguistique et de didactique des langues Universite´ du Que´bec a` Montre´al, Montre´al, QC, Canada H3C 3P8 [email protected] Mireille Tremblay De´partement d’e´tudes franc¸aises Queen’s University Kingston, Ont. Canada K7N 3N6 [email protected]
Preverbs and their origins in Georgian and Udi1 ALICE C. HARRIS
0. INTRODUCTION The structure of the preverb+stem combination is superficially similar in Georgian, a member of the Kartvelian (South Caucasian) family, and in unrelated Udi, a member of the Lezgian group of the North East Caucasian family. Compare (1) and (2) in this regard. (1) (2)
Georgian: mi-v-i-t’an-e thither-1sg-cv-carry-aor Udi:
ta-zu-sˇ-er-e thither-1sg-carry-er-aorII
‘I took it (away)’ 2 ‘I took it (away)’
Of interest here is the fact that in both languages the preverb (mi-, ta-) forms a semantic whole with the verb root (-t’an-, -sˇ-), yet an agreement marker (v-, -ne-) occurs between them. This kind of interruption of a semantic and formal unit is a well recognized problem in linguistics (see, for example, Ackerman and Webelhuth 1998). In this paper it is shown that two quite different histories led to the situation found today and illustrated in (1–2).
1. KARTVELIAN 1.1. What is a Preverb in Kartvelian? In the Kartvelian languages, preverbs have many of the properties they have in other languages, including indication of location or direction of motion. For example, we have the forms in (3). (3)
a-vida ga-vida sˇe-vida c’a-vida
‘she went up’ ‘she went out’ ‘she went in’ ‘she went away’
With some verbs, preverbs have meanings that are not predictable from the meanings of the parts. For example, da-inaxa ‘she saw, noticed, caught sight of it’ contrasts with sˇe-inaxa ‘she saved, kept it’; da- means ‘downwards’, though it has acquired a neutral meaning as well, and sˇe- means ‘in, into’. These are characteristics widely associated with preverbs. In certain other languages, notably Russian, the presence of a preverb is Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 61–78. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
62
A.C. Harris
associated with perfective aspect; to some extent this is true also in Modern Georgian, as we can see in pairs such as ak’etebs ‘she is doing it’ and ga-ak’eta ‘she has done it’. However, this has been grammaticalized, and simple, minimal contrasts of a perfective with and imperfective without a preverb hardly exist today (see Holisky 1981 for more on aspect in Georgian). With regular verbs of Classes 1 and 2, a preverb occurs in the tense-aspect-mood (TAM) categories indicated in (4). (4)
Lacks PV
Requires PV
present future imperfect conditional present subjunctive future subjunctive
Occurs most felicitously with PV aorist (imperative) optative
For many verbs, the only difference between the forms of the first two columns is the absence vs. presence of a preverb. For example, the imperfect vak’etebdi ‘I was making it’ differs from the conditional ga-vak’etebdi ‘if I made it’ only in lacking the preverb. The six TAM categories in the first two columns comprise the traditional Series I verb forms, while those in the third column are the categories of Series II. In the latter set of TAM categories, imperfective forms exist in principle and lack the preverb (see Tschenke´li 1958: 158); in practice, however, the perfective is almost always the form used in these TAM categories. This characteristic provides a diagnostic for preverbs in Georgian: A preverb occurs in the future, conditional, and future subjunctive with regular verbs of Classes 1 and 2 and does not occur in the present, imperfect, and present subjunctive. A second peculiarity of preverbs in Georgian is that the directional preverbs can optionally occur with mo- ‘hither’, also a preverb, forming a ‘‘complex preverb’’. (5)
a-vida ga-vida sˇe-vida c’a-vida
‘she went up’ ‘she went out’ ‘she went in’ ‘she went away’
a-mo-vida ga-mo-vida sˇe-mo-vida c’a-mo-vida
‘she came up’ ‘she came out’ ‘she came in’ ‘she came away’
The opposition illustrated in (5) is often referred to in traditional Georgian linguistics as ‘‘orientation’’, and I adopt that term here. There are, however, two exceptions to the generalization that preverbs can occur with mo-: mi- ‘thither’ and da- ‘downwards’. First, the orientation that opposes mi-akvs ‘she takes it’ is not *mi-mo-akvs but mo-akvs ‘she brings it’. Second, in Modern Georgian dalacks an opposition of orientation; there is no *da-mo- or any equivalent. This is probably best viewed as a result of its change of meaning; da- originally meant ‘downwards’ but its current meaning is very general. Other combinations also occur, especially ga-da- ‘across’, also referring to change.
Preverbs and their origins in Georgian and Udi
63
Note that in the TAM forms that have a preverb, all parts of the complex preverb appear; while in those TAM forms that lack a preverb, all parts of the complex preverb are lacking. For example, the future ga-mo-vak’eteb ‘I will repair it, correct it’ corresponds to the present vak’eteb ‘I am doing it’; the future ga-da-vak’eteb ‘I will re-do it, change it’ corresponds to the same general present. (The preverbs ga-mo- and ga-da- here affect the meaning of the basic vak’eteb ‘I do, make it’, as described for another verb just after example (3).)
1.2. Origins The origin of preverbs in the Kartvelian languages is much like that in IndoEuropean languages. In Georgian and its sisters, a number of preverbs are derived relatively recently from adverbs. Svan, a Kartvelian language, possesses two sets of preverbs.3 The type that represents the input to the formation of preverbs is found as an independent word preceding or following the verb, as in (6). (6)
acˇad sga she.go inside (Svanskie teksty na lasˇchskom narecˇii, cited by Deeters 1930: 17 from A. Oniani’s 1917, tale no. 7)
Sga ‘in(side)’ and other items that occur in the position illustrated in (6) may alternatively be proclitic to the verb, as in (7). (7) sga: cˇad (cited by Deeters 1930: 15 from the same source) in=she.go Sga: cˇad in (7) is from sga=acˇad ‘in=went’. In Svan, sga ‘in(side)’ is a proclitic, not a prefix; evidence to support this is the fact that it may be separated from the verb by a particle, as shown in (8), where sga: y is synchronically from sga=i ‘in=again’. (8)
a.
sga: y acˇad da: v-ı´ ... (cited by Topuria 1967: 60) in.again she.go devi(mythological being)-nom ‘The devi went in again ...’
b. sga ud etqe: rix ce: rbat-s ... (cited by Topuria 1967: 61) in again they.implore God-dat ‘... again they implore God ...’ Svan zˇi ‘up, on’, cˇu ‘down, under’, ka ‘out, from, thither’ behave like sga
64
A.C. Harris
‘in(side)’, occurring as independent words, as in (6, 8), or proclitic to the verb, as in (7). Four other preverbs in Svan – an ‘hither’, ad ‘thither, away from’, es ‘thither’, la ‘in, thither’ behave differently from those in the first group. Members of the second group cannot occur independently, as in (6), or separated from the verb, as in (8). In dialects with umlaut, members of the second set undergo this process when followed by i in the verb, as shown in (9); members of the first do not, as shown in (10). Examples in (9–10) are in the Upper Bal dialect. (9) a¨n-te (
Proclitic sga ‘out’: sga: t’a¨x mo –sga in.return ques –in ‘Did s/he return inside?’ –‘Yes.’ (Topuria 1967: 62)
On the basis of these characteristic behaviors verbal proclitics can be distinguished from prefixes. It is assumed that the Svan adverbial predecessors of an ‘hither’, ad ‘thither, away from’, es ‘thither’, and la ‘in, thither’ first were reanalyzed as optionally separable elements of the verb, forming a single lexical item with the verbal base. It is likely that they often occurred immediately before the verb or separated from the verb by one particle. Members of this set became proclitic to the verb when immediately preceding it, and later became prefixes. Later the adverbs sga ‘in’, zˇi ‘up, on’, cˇu ‘down, under’, and ka ‘out, from, thither’ underwent the same processes, except the last.4 The preverbs of Georgian similarly originate as adverbs or nouns. Two kinds of evidence confirm this: (i) traditional etymologies, (ii) tmesis, or the separability of preverbs in Old Georgian, but not in Modern Georgian. These are considered in turn. According to many sources, the Georgian preverb sˇe- ‘in, into’, like the adposition sˇina ‘in’, is derived from the adverb sˇi-na ‘in’, where -na is a formant of adverbs (Mart’irosovi 1946: 217, Mart’irosovi 1956: 41). Gamq’relize (1959: 54) connects this further with a locative noun *sˇov-a ‘middle, inside’ (Old Georgian sˇova, sˇuva), which could also be used in Old Georgian as an adposition or adverb. According to his analysis, the adverb sˇina is derived from *sˇov-a-na (1959: 54); he derives the preverb sˇe- directly from the noun
Preverbs and their origins in Georgian and Udi
65
*sˇova->*sˇua->sˇa->sˇe-, observing that the a vocalism is preserved in certain dialects (1959: 56; see also Deeters 1930: 10–11). The Old Georgian preverb da- (Common-Georgian-Zan *da-) ‘downwards (on the surface)’ is said to come from the noun *dab- ‘ground, earth’ preserved in the Georgian noun dab-a ‘town’, u-dab-n-o ‘desert’, m-dab-al-i ‘low, short’, as well as in Svan, Laz, and Mingrelian words (Sˇanize 1973: 261 [1953: 268], Mart’irosovi 1956: 41). Old Georgian ac- (Common-Georgian-Zan *ac-) ‘up’ is probably from an independent adverb, preserved as the root in the derived word m-ac-al- ‘high’.5 Many other Georgian preverbs have similar sources (Sˇanize 1973: 261, Mart’irosovi 1956, and other sources cited above). Since many Georgian adpositions are also derived from adverbs or locative nouns, we have preverb/adposition pairs, such as the preverb sˇe- ‘in, into’ beside the postposition -sˇi ‘in’, and the preverb gan- beside the postposition -gan both ‘from’. Beside the preverbs of Old Georgian derived from adverbs and nouns we find unreanalyzed locative nouns and adverbs (see Vesˇap’ize 1967, in which examples (12–13) are cited). In (12), the adverb gare occurs as a separate word; in (13) it forms part of the verb. It is assumed here that gare in (13) is an adverb, not a preverb, since there is no evidence other than its position to suggest that it has been reanalyzed. (12)
deda-y sˇeni da zma-ni sˇenni dganan gare mother-nom your and brother-pl.nom your stand.3pl outside ‘Your mother and your brothers are standing outside.’ (Mt 12:47Ad)6
(13)
gare-c’ar-vakcio salmoba-y sˇen-gan (Exodus 23:25) outside-pv-take.1sg sickness-nom you-from ‘I will take sickness away from you.’
A second kind of evidence that supports the view that Georgian preverbs originated as words separate from the verb roots is the fact that in Old Georgian ‘‘little words’’ (probably second position clitics) could occur between the preverb and the verb stem (see Cherchi 1994, 1997, for a complete analysis). (14)
a.
romel ca igi akundes mi-ve-ecos mis gan which.nom indef det have.3sg pv-even-take.it him from ‘whatever he has (even that) will be taken from him’ (Mt 25:29Ad)
b. sˇe-tu-izinos sopel-i q’oveli da ganik’uas sul-i tvisi pv-if-gain.3sg world-nom all and lose.3sg soul-nom self’s ‘[what does it profit a man] if he gain the whole world and lose his soul’ (Mk 8:36Ad) The origin of some Georgian preverbs is unclear; examples are mi- ‘thither’
66
A.C. Harris
and mo- ‘hither’. There is weak evidence that these, too, originated as adverbs of location, but no precise adverbs have been identified. The evidence is the fact that mi- and mo- occur as postpositions or the preposed portion of circumpositions, as illustrated in (15–17). (Examples in (15–17) are cited by Imnaisˇvili 1957: 324–325.) (15)
rametu q’o cˇem-da-mo did-did-ni zlier-man (L 1:49Ad) because do.3sg me-adv-hither great-pl.nom strong-erg ‘because the strong one has done great things to me’
(16)
upalo moed p’irvel vidre mo-sik’udi-d-mde7 q’rm-isa mis (J 4:49B) lord.voc come first before hither-death-adv-until child-gen his ‘Lord, come before the death of my (lit. his) child.’
(17)
acadet tanad acorzinebad orta ve mi-mk’a-mdis (Mt 13:30Ad) two all thither-harvest-until allow.2pl together grow ‘allow both to grow together until the harvest’
If the preverbs mi- and mo- originated in some other way, not as adverbs, we would not necessarily expect them to occur also as adpositions, as they do in (15–17). On the other hand, it is possible that if they did not originate as adverbs they would nevertheless have been drafted as adpositions on analogy with other preverbs, such as sˇina- ‘in’. Because of the possibility of analogy, it is impossible to be confident of the origins of these preverbs.8 In the earliest Old Georgian examples, preverbs are already proclitic to the verb, but other elements may occur between the preverb and the verbal base (see (14) and Cherchi 1994, 1997); in this respect they resemble the Svan forms in (8). In Old Georgian data, perfective forms are likely to occur with preverbs, but perfectivity does not depend exclusively on the presence of the preverb. That is, forms in perfective paradigms, such as the Old Georgian aorist, are perfective whether or not the preverb occurs. In Old Georgian particular preverbs are already common with particular verbal bases. Thus, the history of preverbs in Kartvelian is similar to that of a number of other languages: adverbs or nouns gradually become part of a verb stem. 2. UDI AND OTHER LEZGIAN LANGUAGES9 2.1. Synchronic description Udi inherited structures of the forms in (18), among others. (18)
a. *PV1-(PV2 )-CM-root ... b. *CM-root ...
Preverbs and their origins in Georgian and Udi
67
Here CM is a class marker, the form of agreement found in languages of the North East Caucasian family. Following the root are the tense-aspect-mood markers, which do not interact with the preverb, and which are not relevant to the topic of this paper.10 It is likely that the Proto-Lezgian structure in (18a) was formed through processes similar to those described above for the Kartvelian languages, but at this time that cannot be proven. Udi has a set of seven preverbs, given in (19). (19)
PV
Meaning
Example
e-
‘hither’
e-ne-cˇer-e
ta-
‘thither’
ta-ne-d-e
la(y)-
‘up’
la-ne-p-e
ci-
‘down’
ci-ne-p-e
ba(y)-
‘in’
ba-ne-d-e
cˇ’e-
‘out’
cˇ’e-ne-bak-e
qa(y)-
‘back, repeat qay-ne-bak-e action’
[hither-3sg-carry-aorII] ‘she brought’ [thither-3sg-lv-aorII] ‘she took’ [up-3sg-lv-aorII] ‘she pulled up, dressed, wore’ [down-3sg-lv-aorII] ‘she poured down, threw down’ [in-3sg-lv-aorII] ‘she thrust s.t. in, baked’ [out-3sg-be-aorII] ‘she passed through, ground’ [back-3sg-be-aorII] ‘she returned, went back’
Here LV is a ‘‘light verb’’; -d- is a light verb to which no particular meaning can be attached. The root -p- earlier meant ‘say’ and continues in this meaning, but it has lost this meaning in many words. (19) illustrates the preverbs in verbs where their directional meaning remains fairly clear; but there are other verbs in which that meaning is no longer clear, such as la-ne-d-e ‘she wiped, spread’. In Udi preverbs do not serve a grammatical purpose such as marking perfective aspect. Each preverb occurs with only a few verb roots, including the verb roots ‘come, go’ (suppletive and irregular) and -sˇ-/-cˇ- ‘carry’.11 From a synchronic point of view, Udi does not combine preverbs as was possible in Proto-Lezgian (see (18a)).
2.2. Origins of the Preverbs Udi preverbs appear to come from a variety of sources – inherited, borrowed, reanalyzed (grammaticalized and degrammaticalized).
68
A.C. Harris
Two Udi preverbs, la- ‘up’ and qa- ‘back, reverse action’ are clearly inherited from Proto-Lezgian preverbs, *l- ‘up, on’ and *q- ‘behind, back’, respectively (Harris, to appear).12 These occurred in the position of PV1 in (18a). The element y, which occurs under certain circumstances with three of the preverbs, may be the reflex of a second preverb of Proto-Lezgian, *ay-, appearing in the position of PV2 in (18a) (see Harris, to appear; compare Alekseev 1985: 48). Some of the other Udi preverbs are probably language-specific innovations. For example, the preverb ci- ‘down’ appears to be related to the adverb cina ‘south, to the south’. (-na cannot be identified with certainty; there are numerous possibilities.) It is likely that ci- was reanalyzed from an adverb, much as in Svan, Georgian, and other languages. The Udi preverb ba- ‘in’ appears to be borrowed from an Iranian language (Wolfgang Schulze, personal communication). In all instances, the preverb occurs in the same position as the inherited preverbs – quite unremarkably – and with them forms a system. Perhaps the most interesting source among the Udi preverbs is degrammaticalization. There is evidence for a Pre-Udi verb root *e or *eC ‘come’. Language-internal evidence that e was once part of the root comes from future stems of verbs in which ‘come’ is used as a light verb, in a way similar to the way bak- is used in the last two examples in (19). Such verbs include be1 x-ec- ‘swell’, nep’ax-ec- ‘fall asleep’ (Jeiranisˇvili 1971: 235, 239), box-ec- ‘boil (intr)’. It might be suggested that the e here might be epenthetic (synchronically), although this does not seem to be consistent with the phonology of the language.13 There are a handful of verbs such as c’oro-ec- ‘drip, drain’, axa-ec- ‘hang down’, qay-ec‘open, undo’ (all three, Jeiranisˇvili 1971: 73), ba1 ca-ec- ‘is found’ 14 which preserve -ec- and where epenthesis is a priori unlikely because of the presence of a vowel to the left. Alternatively it might be suggested that e- is the remnant of an old preverb, but this does not seem to be consistent with evidence from the sister languages. While there is not agreement about the forms to be reconstructed, to the best of my knowledge no one has suggested that the comparative evidence supports the reconstruction of preverbs in the meanings ‘hither, here’ or ‘thither, there’ (leading to verbs meaning ‘come’ or ‘go’). Furthermore, the forms generally reconstructed consist only of a single consonant (Alekseev 1985: 117ff., Nikolayev and Starostin 1994, and Schulze 1988). Thus, the internal evidence and comparative evidence seem to support a Pre-Udi form *ec- ‘come’ or a root of similar composition.15 While originally (part of) the root, e- was reanalyzed as a preverb, parallel in behavior to the others listed in (19). In modern Udi, ‘come’ is multiply suppletive, as illustrated in (20). (20)
Stem Example Gloss
Present stem Aorist stem Future stem Imperative stem
e-%a(-)re-ce-k-
e-ne-sa ar-e-ne e-ne-c-o e-k-e
[hither-3sg-%-pres] [ar-aorII-3sg] [hither-3sg-c-futI] [hither-k-imper]16
Translation ‘she comes, is coming’ ‘she came’ ‘she will come’ ‘come (here)!’
Preverbs and their origins in Georgian and Udi
69
It is likely that the vowel a of the aorist stem is related to the e- preverb, but vowel alternation in Proto-Lezgian is not well understood. In the Nij dialect, the stem form is har-. I assume henceforth that at least e-%-, e-c-, and e-k- in (20), are from a shared source, *e-C-. Evidence to support reanalysis of *eC- as *e-C- (that is, evidence to support reanalysis with a morpheme boundary) is of several kinds. First, there are the parallel positions of e- and the other preverbs in (19). Second, e- occupies the position occupied by other preverbs in the future stems of the six intransitive verbs of directional motion, as shown in (21). (21)
e-cta-clay-cci-cbay-ccˇ’e-c-
‘come’ ‘go’ ‘go up’ ‘go down’ ‘go in’ ‘go out’
(Data from Jeiranisˇvili 1971: 74)
Third, e- in e-ke ‘come (imper)’ occupies the position occupied by other preverbs in the imperative stems of the intransitive verbs of directional motion, such as ta-ke ‘go (imper)’, cˇ’e-ke ‘go out (imper)’. Fourth, e- is in the same position as other preverbs in the transitive verbs of directional motion, with the root -sˇ-/-cˇ- ‘carry’, illustrated in (22).17 (22)
e-ne-cˇ-e ta-ne-sˇ-er-e lay-ne-cˇ-er-e bay-ne-cˇ-er-e
[hither-3sg-carry-aorII] [thither-3sg-carry-er-aorII] [up-3sg-carry-er-aorII] [in-3sg-carry-er-aorII]
cˇi-ne-cˇ-er-e18 [out-3sg-carry-aorII]
‘she brought it’ ‘she took it’ ‘she took it up’ ‘she took it in’ (Jeiranisˇvili 1971: 237–238) ‘she took it out’
Thus there is good evidence to support both the view that *eC- was a single root morpheme historically, and the view that e-%-, e-c-, and e-k- are a preverb e- and a suppletive root today. This means that Udi e- has degrammaticalized from the root in which it was found in Proto-Lezgian.
2.3. The problem with -ne- and other person markers As mentioned above, the preverbs have partly lost their locative/directional meanings in verbs other than the intransitive and transitive verbs of directional motion. Some of those in which both stem morphemes can be identified with certainty are listed in (23).
70
(23)
A.C. Harris
ta-ne-d-e la-ne-d-e la-ne-p-e ci-ne-p-e ba-ne-d-e ba-ne-k’-e ba-ne-p-e cˇ’e-ne-bak-e cˇ’e-ne-p-e
[thither-3sg-lv-aorII] [up-3sg-lv-aorII] [up-3sg-lv-aorII] [down-3sg-lv-aorII] [in-3sg-lv-aorII] [in-3sg-lv-aorII] [in-3sg-lv-aorII] [out-3sg-be-aorII] [out-3sg-lv-aorII]
‘she gave’ ‘she wiped, spread’ ‘she pulled up, dressed’ ‘she poured down, threw down’ ‘she thrust s.t. in, baked’ ‘she poured’ ‘she poured in, filled with’ ‘she went through, crossed over’ ‘she undressed, took off’
Evidence that the preverbs in (19) are not productive is that Jeiranisˇvili, a native-speaker linguist, states repeatedly that the first element (which he seems to have considered a morpheme) in these words could not be identified or glossed (1971: 78, 231, 232, 233, 234). Thus the meaning, for Jeiranisˇvili, belongs to the combination of preverb and light verb and is not predictable from the meanings of the parts. This is confirmed by the fact that the light verbs -d- and -k’- have no discernable meaning, while the meaning of -p- is no longer directly related to its basic meaning, ‘say’, in these and other examples. It is not unusual to have preverb-verb combinations with lexicalized (‘‘listed’’ in the terms of Di Sciullo and Williams 1987) meanings and with structures in which an agreement marker intervenes between the preverb and the verb root. This is, after all, what we find in Georgian (1). It is easily explained in terms of the preverb being added to a form in which some agreement markers already preceded the verb root. What makes Udi unusual and problematic is the fact that the agreement markers found in the modern language in the position between the preverb and the root are recently formed. The inherited markers would probably have had forms similar to those in (24), which are reconstructed for Proto-Lezgian. (24) I II III IV
Singular
Plural
*w *r/y *b *d
*b *b *d *d
(Schulze 1992, see also Alekseev 1985)
The b- marker of class III is, in fact, preserved in fossilized form in Udi (Jeiranisˇvili 1956). Comparing the Proto-Lezgian class markers in (24) with the Udi person markers (PMs) in (25), it is easy to see that they are not cognate.
Preverbs and their origins in Georgian and Udi
(25)
71
Person Markers (PMs) Independent Pronouns 1.Sg 2.Sg 3.Sg 1.Pl 2.Pl 3.Pl
-zu -nu -ne -yan -nan -q’un
zu un yan va1 n, efa1 n
The independent pronouns are given in the ergative/absolutive form. It is easy to see that the first and second person PMs are derived from independent pronouns. The third person PMs are derived from independent deictic pronouns, but they are not provided here because their derivation is more complex and not directly relevant (but see Harris 2002: 181–182). The forms in (24) and (25) show that the Udi PMs are not inherited from Proto-Lezgian class markers and that the Udi PMs are relatively new. This means that, at least for the inherited preverbs la- ‘up’ and qa(y)- ‘back, reverse action’, the agreement markers assumed their current positions after the structure PV-root already existed. How can this be?
2.4. How Udi PMs came to occur between the preverb and the root The historical process of univerbation in Indo-European languages is one that is familiar through the work of Watkins (1963, 1964), among others, and a similar schema has been sketched above in this paper and in the work of other linguists. This occurred historically also in Udi and can be seen most clearly in recent noun incorporations, such as (26a), which occurs beside (26b), which lacks incorporation. (26)
a.
merab-en zavod-a asˇ-ne-b-sa Merab-erg factory-dat work-3sg-do-pres ‘Merab works in a factory.’
b. merab-en zavod-a asˇ-l-ax-ne b-esa Merab-erg factory-dat work-ext-dat-3sg do-pres ‘Merab does the WORK in a factory.’ Although in (26a) asˇ occurs in the absolutive case, with a % ending, and in (26b) it is in the dative case, it is easy to see that a structure similar to the latter could have been reanalyzed as (26a), ‘‘trapping’’ the PM (here -ne-) between the incorporated noun and the light verb, -b- ‘do’. It may be that something similar occurred with the innovative preverbs, such as ci- ‘down’, though it is impossible to determine the relative order of the
72
A.C. Harris
change of independent ci- to a preverb and that of independent pronouns to PMs. Thus, it is possible that we had something like *ci-ne p-e [down-3sg sayaorII]>ci-ne-p-e [down-3sg-lv-aorII] ‘she poured down, threw down’. Nearly all of the light verbs in the modern language are single consonants (bak- ‘be, become’ is the only exception), and I assume that status was reached before the change discussed in this paragraph. One way a speaker might analyze the structures resulting from univerbation (with the exception of those employing bak-) is as (27). (27) X-PM-C ... , where X is any incorporated material (including nouns, such as asˇ ‘work’, and possibly innovative preverbs, such as ci- ‘down’), PM is the person marker ‘‘trapped’’ through univerbation, and C is a consonant (the light verb), and where the ellipsis indicates other, irrelevant morphemes following the root. This pattern was extended to monomorphemic verb roots, with X now representing whatever precedes the final consonant in the root. This change is indicated in (28), where the double arrow represents extension, and where bok’- ‘burn’ is a monomorphemic root reconstructed by both Schulze (1988) and Nikolayev and Starostin (1994), and where the forms used for illustration are the modern ones.19 (The pattern was never extended to the two monomorphemic verb roots that end in a vowel.) (28)
Pattern Extension asˇ-ne-b-e work-3sg-do-aorII X-PM-C ...
& X-PM-C ... bo-ne-k’-e burn1-3sg-burn2-aorII
Another root in which this kind of extension must have occurred is berx- ‘grind, mill’, and we find forms such as ber-ne-x-e ‘she ground’ today. For these and certain other verbs it is hypothesized, then, that the PMs derived in Pre-Udi from independent pronouns, as shown in (25), moved inside monomorphemic verb roots by extension of (27), the pattern created through the process of univerbation. For some other monomorphemic verb roots, however, a problem arises: Comparative evidence suggests that the inherited class markers occupied the position in the verb root that is later occupied by the reduced pronouns. For example, both Schulze (1988: 135) and Nikolayev and Starostin (1994: 1025) reconstruct a Proto-Lezgian ancestor for Udi aiz- ‘stand’. Nikolayev and Starostin (loc cit) list the following reflexes in the sister languages indicated,
Preverbs and their origins in Georgian and Udi
73
where the equal-sign marks the position of the inherited marker of class: Rutul (l-u=zfa-), Tsakhur (ul’-o=zar-), Kryts (q:-u=zur-), and Archi (o=c:i-).20 In these sister languages the marker of class occurs immediately before the reflex of the fricative, just as it does in Udi, as illustrated in (29). (29) ai-ne-z-o stand1-3sg-stand2-futI ‘she will stand up’ Given the comparative evidence, we are obliged to reconstruct the position marked by ‘‘=’’ as the position of the marker of class in Proto-Lezgian. This raises three questions: First, what does this indicate about how the innovative Udi person markers got into the position illustrated in (29)? Second, how did the class markers get into this position in Proto-Lezgian or before? What does this have to do with preverbs? Examples like Udi aiz- ‘stand’ show that the position of the inherited class marker is relevant to the position later assumed by the Udi person marker. If we deny this relationship and claim instead that person markers came to their present intramorphemic position through extension (or analogy), we would be claiming that it was a coincidence that in the Proto-Lezgian etymon of aiz- and certain other modern Udi monomorphemic verb roots person markers happen to occupy the position class markers occupied in Proto-Lezgian. We must assume that coincidences of this sort do not occur in root after root.21 There are several ways the position of the Proto-Lezgianclass markers could be related to that of the Udi person markers. First, it is entirely possible that the person markers were innovated before the class markers were lost.22 Second, in Udi, especially Udi of the 19th century, other morphemes occur – when they do occur – with the person marker. It is possible that these were present before the loss of the class markers and marked the ‘‘slot’’ until person markers were positioned. A third possibility is that the position of the class marker may have been dependent on the location of word stress, and that person markers may have been similarly positioned relative to stress. Although it is only a tendency for the person marker to immediately follow stress in modern Udi, at an earlier stage this may have been required. Thus, we must assume that in verb roots of this type, the person marker moved into the slot reserved for it by one of the means outlined in this paragraph; I refer to this below as ‘‘movement to a slot’’. The second question posed above is how the class markers got into monomorphemic verb roots in Proto-Lezgian. It is likely that this occurred through some combination of univerbation and pattern extension, just as happened later in Pre-Udi. The last question, and the one that brings us to the point of this paper, is how PMs got into the positions they occupy with respect to preverbs. In § 2.2 I identified four kinds of origin for Udi preverb: (i) inheritance (e.g. la- ‘up’,
74
A.C. Harris
qa(y)- ‘back, repeat action’), (ii) reanalysis of an adverb (e.g. ci- ‘down’), (iii) borrowing (e.g. ba- ‘in’), and (iv) degrammaticalization, e.g. e- ‘hither’. For each of these the origin of the position of the person marker may be a bit different. (i) The structure inherited for preverbs was (18a), as indicated above. Since the innovative person marker came to occupy the position previously held by the class marker, we must assume that this took place through movement to a preexisting slot. (ii) As noted above, it is possible that the person marker came to follow the preverb ci- ‘down’ through univerbation. On the other hand, the preverb may have been reanalyzed and positioned first, with the person marker later being positioned as with inherited preverbs. This depends on the relative order of changes, which is not at this time recoverable. (iii) Similarly with regard to borrowed preverbs, if we assume that the borrowing took place first, the positioning of the person marker would have been as with inherited preverbs. (iv) Most interesting is the positioning of the person marker relative to the degrammaticalized preverb e- ‘hither’. We may assume that the reanalysis of eas a preverb was motivated in part by the paradigmatic parallels described in § 2.2, but also in part by the position of the person marker relative to it. It is possible that the inherited root *eC- came to be interrupted by the person marker through the mechanism of extension, as indicated in (30). (30)
Pattern Extension asˇ-ne-b-e work-3sg-do-aorII X-PM-C ...
bo-ne-k’-e burn1-3sg-burn2-aorII X-PM-C ... & X-PM-C ... e-ne-c-o come1-3sg-come2-futI
After the pattern of (27) was extended to *eC-, the e- was reanalyzed as a preverb, as indicated in (31), where the triple arrow represents reanalysis. (31)
e-ne-c-o come1-sg-come2-futI X-PM-C ...
e-ne-c-o come1-3sg-come2-futI z PV-PM-root ...
Thus, the single surface form, e-ne-c-o ‘come’, in pattern (27) was reanalyzed as a realization of pattern (18a).
3. CONCLUSION I have shown that agreement markers may come to occur between a preverbroot combination by a variety of diachronic means. The mechanism followed in
Preverbs and their origins in Georgian and Udi
75
Svan and Georgian is a familiar one, involving reanalysis of an adverb or noun, cliticization to an existing agreement-root sequence, and eventually further reanalysis of the proclitic preverb as a prefix. In Udi we find a variety of pathways to this end, including univerbation (a kind of reanalysis), pattern extension, reanalysis of an adverb, and degrammaticalization. Perhaps most surprising is the realization forced upon us by the data, that the innovative person marker of Udi in some instances moves to the position previously occupied by the inherited markers of class. NOTES 1
I am grateful to Wolfgang Schulze for calling my attention to the origin of ba- ‘in’, as noted in the text. I also want to thank both to Wolfgang and to Lyle Campbell for commenting extensively on a related manuscript and for discussion of many issues relating to Udi. The research reported here was supported in part by the National Science Foundation under Grants No. SRB-9710085 and BCS-0091691 and by the International Research and Exchanges Board under the ACLS-Academy of Sciences Exchange with the Soviet Union (1989) and in earlier years. I am grateful for the support of both organizations. My research on Udi is entirely dependent upon the help of my consultants, especially Luiza Nesˇumasˇvili, Dodo Misk’alisˇvili, Nana Agasisˇvili, and Caco Cˇik’vaize, to whom I am very grateful. 2 The following abbreviations are used in this paper: adv adverbial case, aor aorist, c consonant, cm class marker, cv character vowel, dat dative, det determiner, er suffix er/r with no identified function, erg ergative, ext extension, fut future, gen genitive, imper imperative, indef indefinite, lv light verb, nom nominative, pl plural, pm person marker, pres present, pv preverb, ques question particle, sg singular, tam tense-aspect-mood, voc vocative. In Udi examples, the copular is glossed be when it functions as a light verb, and the verb ‘make, do’ as do when it serves as a light verb. Tense-aspect-mood categories in Udi are named and numbered according to the system in Pancˇvize (1974), as described in Harris (2002). 3 Except as noted, examples are from the Lasˇx dialect. 4 It cannot be shown with certainty that the set of prefixal preverbs originated as adverbs, but it is generally assumed that their origin was parallel to that of the adverb/proclitics (see, for example, Schmidt 1969). The description and analysis of preverbs in Kartvelian languages is based on that in Harris and Campbell (1995: 94–96). 5 The etymologies given here can also be found in Klimov (1998) and Penrixi and Sarjvelaze (1990). The derivational morphology is described in detail in (Jorbenaze et al. 1988); note especially the circumfix m–al- ‘characterized by’, which forms adjectives and participles (Jorbenaze et al. 1988: 272–274). 6 Old Georgian examples cited are from the Bible, with the following abbreviations used: Ad the Adisˇi codex (the oldest, 897 A.D.), J John, L Luke, Mk Mark, Mt Matthew. 7 On the formation of the adverbial case of nouns in -il, see Imnaisˇvili (1957: 36–40). 8 Another hypothesis is that the origins of mi- ‘thither’ and mo- ‘hither’ are in some way connected with the first person singular objective prefix, m- (Klimov 1964: 135, 136, Klimov 1998: 122, 124). Problems of this hypothesis include the origins of the vowels, especially in view of the fact that m-i- often has meanings such as ‘for me’, ‘to me’, ‘with respect to me’. Another problem is that, given that the meanings of mi- and mo- are ‘away from speaker and
76
A.C. Harris
hearer’ and ‘toward speaker and hearer’, rather than just ‘away from speaker’ and ‘toward speaker’, one might expect the prefix gv- ‘us inclusive’ (‘me and you’) to have been used, rather than m- ‘me’. 9 Except as noted, data are from the village of Okt’omber, where the Vartasˇen dialect of Udi is spoken. 10 There is some evidence for an aspectal marker *-l-, which would have preceded the root; but this is likewise irrelevant to the present study. 11 As discussed below, e- occurs only with these two roots; and ci- ‘down’ evidently fails to occur with ‘carry’, though it does occur with other roots, as illustrated in (23). 12 Alekseev (1985: 117) reconstructs *al ‘on’, but Schulze (1988: 49) points out that the evidence suggests that the vowel preceding l in several of the sisters is relatively recent. 13 Generally, e syncopates in the environment VC+__sV (see Harris 2002: 82–83). 14 Finding this verb difficult to elicit in any form, I have the form cited here only from two consultants. 15 Nikolayev and Starostin (1994: 269) identify this Udi verb as a reflex of Proto-Lezgian *?irq˙a¨r ‘reach, come, get, hit, overtake, find’, but I do not consider this a firm reconstruction. 16 Although I have glossed the morpheme -e as imperative, it is not the regular imperative formant, which is -a. Even the division of ke into two morphemes is undertaken with little concrete evidence; doing so makes the stem, e-k- structurally similar to the future stem, e-c-, and the putative imperative ending -e similar to the regular -a. 17 See Schulze (1988: 129) on the etymology of this root. Nikolayev and Starostin (1994: 254) give a different view, relating the Udi verb e-cˇ- ‘bring’ to forms in two other Lezgian languages. This etymology would, in any case, not shed much light on the question of whether e- was inherited as part of the root in this word too, thus forming part of the basis for separating off e- as a preverb at a later date, or whether the derived preverb was extended to this root. 18 I assume that *cˇ’e- became *cˇi- in this verb on analogy with the root, cˇ-. I have no explanation for the change in the vowel. 19 On the notion ‘‘extension’’ in diachronic syntax, see Harris and Campbell 1995. For proof of the monomorphemic status of these roots in Udi, see Harris 2000; and for more details of the change, including details of the roots discussed here, see Harris 2002. 20 Nikolayev and Starostin (1994) rely almost entirely on a single source for information about Udi (they make little use of sources in German or Georgian), and they evidently have no information about the location of PMs in Udi. We may assume, therefore, that their analysis was not in any sense constructed with the issue discussed here in mind. 21 All regular monomorphemic verb roots in Udi have endoclitic PMs. For some, only extension is possible; for some it is clear that this explanation is not satisfactory because of evidence from the sister languages; for some the facts are not all known. 22 Plank (1980) and Abraham (1997) have argued that in other instances one set of markers was retained until a replacement system was in place.
REFERENCES Abraham, Werner (1997). The interdependence of case, aspect and referentiality in the history of German: The case of the verbal genitive. Parameters of Morphosyntactic Change, eds. Ans van Kemenade and Nigel Vincent. Cambridge: Cambridge University Press, 29–61.
Preverbs and their origins in Georgian and Udi
77
Ackerman, Farrell, and Gerd Webelhuth (1998). A Theory of Predicates. Stanford: Center for the Study of Language and Information. Alekseev, M.E. (1985). Voprosy sravnitel’no-istoricˇeskoj grammatiki lezginskix jazykov: Morphologija. Sintaksis. Moscow: Nauka. Blake, Robert P. (1974). The Old Georgian version of the Gospel of Mark. Patrologia Orientalis 20, fascicle 3. (First edition, 1928.) Blake, Robert P. (1976). The Old Georgian version of the Gospel of Matthew. Patrologia Orientalis 24, fascicle 1. (First edition 1933.) Blake, Robert P. and Maurice Brie`re. (1950). The Old Georgian version of the Gospel of John. Patrologia Orientalis 26, fascicle 4. Brie`re, Maurice. (1955). La version ge´orgienne ancienne de L’e´vangile de Luc. Patrologia Orientalis 27, fascile 3. Cherchi, Marcello (1994). Verbal tmesis in Georgian. Annali del Dipartimento di Studi del Mondo Classico e del Mediterraneo Antico Sezione Linguistica 16, 33–115. Cherchi, Marcello (1997). Verbal tmesis in Georgian, Part II. Annali del Dipartimento di Studi del Mondo Classico e del Mediterraneo Antico Sezione Linguistica 19, 63–137. Deeters, Gerhard (1930). Das khartwelische Verbum. Leipzig: Markert & Petters. Di Sciullo, Anna-Maria, and Edwin Williams (1987). On the Definition of Word. Cambridge, MA: MIT Press. Gamq’relize, T. (1959). Sibilant’ta sˇesat’q’visobani da kartvelur enata uzvelesi st’rukt’uris zogi sak’itxi. [Sibilant correspondences and some questions of the most ancient structure of the Kartvelian languages.] Tbilisi: Ak’ademia. Giginejsˇvili, B.K. (1977). Sravnitel’naja fonetika dagestanskix jazykov. Tbilisi: Universitet. Harris, Alice C. (2000). Where in the word is the Udi clitic? Language 76, 593–616. Harris, Alice C. (2002). Endoclitics and the Origins of Udi Morphosyntax. Oxford: Oxford University Press. Harris, Alice C. (to appear). The prehistory of Udi locative cases and locative preverbs. Festschrift for Howard I. Aronson, ed. Dee Ann Holisky and Kevin Tuite. Harris, Alice C., and Lyle Campbell (1995). Historical Syntax in Cross-Linguistic Perspective. Cambridge: Cambridge University Press. Holisky, Dee Ann (1981). Aspect theory and Georgian aspect. Tense and Aspect (Syntax and Semantics, vol. 14), ed. P.J. Tedeschi and A. Zaenen. New York: Academic Press, 127–144. Imnaisˇvili, I. (1957). Saxelta bruneba da brunvata punkciebi zvel kartulsˇi. [The declension of nouns and the functions of cases in Old Georgian.] Tbilisi: Universit’et’i. Jeiranisˇvili, Evg. (1956). Gramat’ik’uli k’lasis gakvavebuli nisˇnebi udur zmnebsa da nazmnar saxelebsˇi [Fossilized markers of grammatical class in Udi verbs and deverbal nouns]. Iberiul-k’avk’aziuri enatmecniereba 8, 341–362. Jeiranisˇvili, Evg. (1971). Udiuri ena [The Udi language]. Tbilisi: Universit’et’i. Jorbenaze, B., M. K’obaize, and M. Berize (1988). Kartul enis morpemebisa da modaluri element’ebis leksik’oni. Tbilisi: Mecniereba. ` tymologicˇeskij slovar’ kartvel’skix jazykov. Moskva: Akademija. Klimov, G.A. (1964). E Klimov, Georgij A. (1998). Etymological Dictionary of the Kartvelian Languages. Berlin: Mouton de Gruyter. Mart’irosovi, Aram (1946). Tandebuli kartulsˇi. [Adpositions in Georgian.] Iberiul-k’avk’aziuri enatmecniereba 1, 203–246. Mart’irosovi, A. (1956). C’indebulisa da tandebulis ist’oriuli urtiertobisatvis kartulsˇi. [On
78
A.C. Harris
the historical relationship between preverbs and postpositions in Georgian.] Iberiulk’avk’aziuri enatmecniereba 8, 39–46. Nikolayev, S.L., and S.A. Starostin (1994). North Caucasian Etymological Dictionary. Moscow: Asterisk Publishers. Pancˇvize, Vl. (1974). Uduri enis gramat’ik’uli analizi. [A grammatical analysis of the Udi language.] Tbilisi: Mecniereba. Penrixi, Hainc (=Fa¨hnrich, Heinz), and Zurab Sarjvelaze (1990). Kartvelur enata et’imologiuri leksik’oni. [Etymological dictionary of the Kartvelian languages.] Tbilisi: Universit’et’i. Plank, Frans (1980). Encoding grammatical relations: Acceptable and unacceptable nondistinctness. Historical Morphology, ed. Jacek Fisiak. Berlin: Mouton de Gruyter, 289–326. Sˇanize, Ak’ak’i (1973 [1953]). Kartuli enis gramat’ik’is sapuzvlebi. [Fundamentals of the grammar of the Georgian language.] Tbilisi: Universit’et’i. Schmidt, Karl Horst (1969). Zur Tmesis in den Kartvelsprachen und ihren typologischen Parallelen in indogermanischen Sprachen. Giorgi Axvledians. Tbilisi: Universit’et’i, 96–105. Schulze, Wolfgang. ms (1988). Studien zur Rekonstruktion des Lautstandes der Su¨dOstkaukasischen (Lezgischen) Grundsprache. Bonn. Schulze, Wolfgang (1992). Zur Entwicklungsdynamik morphosyntaktischer Subsysteme. Caucasologie et mythologie compare´e, ed. Catherine Peters. Paris: 335–362. Topuria, Varlam (1967 [1931]). Svanuri ena, I: Zmna [The Svan Language, I: The Verb]. [Published as volume I of his Sˇromebi [Works].] Tbilisi: Mecniereba. Tschenke´li, Kita (1958). Einfu¨hrung in die georgische Sprache, Band 1. Zu¨rich: Amirani. Vesˇap’ize, Irak’li (1967). Zmnisc’ini zvel kartul enasˇi. [The preverb in the Old Georgian language.] Tbilisi: Universit’et’i. Watkins, Calvert (1963). Preliminaries to a historical and comparative analysis of the syntax of the Old Irish verb. Celtica 6, 1–49. Watkins, Calvert (1964). Preliminaries to the reconstruction of Indo-European sentence structure. Proceedings of the Ninth International Congress of Linguists, ed. H.G. Lunt. The Hague: Mouton, 1035–1045.
Department of Linguistics SUNY Stony Brook Stony Brook, NY 11794-4376 U.S.A. [email protected]
Particles and prefixes in Dutch and English ANS VAN KEMENADE and BETTELOU LOS1
1. INTRODUCTION This paper charts the historical development of two sets of verbal prefixes in the West-Germanic languages, which appear to show a large degree of functional equivalence, although they have rather different morphosyntactic properties. The first set is inseparable, as found in the Dutch verbs verbranden ‘burn’, beschrijven ‘describe’, ontmoeten ‘meet’; while the second set is separable, as found in the Dutch verbs opbellen ‘call up’, afzeggen ‘call off’, wegblazen ‘blow away’. The two sets of verbs, whose properties we describe in some detail below, are functionally equivalent in the sense that they denote complex events that involve a change of state in a resultative construction. This reflects their assumed common historical origin. We claim that the morphosyntactic differences between the two sets result from the fact that the verbs with inseparable prefixes have undergone a different morphosyntactic development, whilst retaining the semantics of a complex event. We will distinguish four ways in which the complex events denoted by the verbs under discussion can be encoded morphosyntactically. These four stages to some extent represent a historical sequence: at the first stage, the particle represents a genuine predicate in a secondary predicate construction, and is constructed syntactically as a morpheme independent from the verb. At the second stage, the prefix/particle is part of a separable complex verb (SCV) which, though constructed from free morphemes and separable by syntactic processes, operates as a single lexical unit in other respects. At the third stage, the preverb, though constructed from bound morphemes, is separable from the verbal stem by other bound morphemes. At the fourth stage, which may either develop from the third stage, or develop independently according to morphosyntactic circumstance, the prefix is part of an Inseparable Complex Verb (ICV), which is a bound morpheme inseparable from the verbal stem. The predicate/particle/prefix typically encodes a change of state, and adds telic aktionsart, resulting in an accomplishment or an achievement. The predicate may grammaticalize further into an aspectual marker or inflectional morphology, in which case it no longer encodes a change of state, and is no longer a resultative predicate. The diachronic development of these resultative predicates, then, affects aspectuality (Boogaart 1999): the inventory of expressions of lexical aspect (or aktionsart), and, in time, grammatical aspect. The structure of this article is as follows: in section 2, we describe in some detail the properties of separable verbs, concentrating on present-day Dutch and present-day English. In section 3, we present an analysis of the semantic Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 79–117. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
80
A. van Kemenade and B. Los
core shared by the two sets of verbs, by formulating the lexical conceptual structure (LCS) they have in common. We will then see how this LCS can be variously encoded morphosyntactically. In section 4, we discuss the inseparable verbs and in section 5 we describe the historical development in some detail, developing the historical stages outlined above. In section 6 we link the decline of preverbal particles in English to a change in underlying syntax. Section 7 sums up the results and outlines questions that must be pursued further.
2. PARTICLE VERBS IN PRESENT-DAY DUTCH AND ENGLISH 2.1. The paradox of particle verbs (SCVs) Separable complex verbs (SCVs) in the present-day West-Germanic languages typically consist of a verbal base, and a nonverbal part, often but not always a ‘particle’. By way of example, let’s consider the Dutch verb opbellen ‘to call up’ (from Booij 1990). (1)
a.
Jan zegt dat hij morgen zijn moeder opbelt John says that he tomorrow his mother up-rings ‘John says that he will phone his mother tomorrow’
b. Jan belt zijn moeder morgen op John rings his mother tomorrow up ‘John will phone his mother tomorrow’ In (1a), a nonroot clause, the particle op precedes the non-finite verb, while in (1b), it is left stranded in clause-final position as a result of the Verb Second constraint (V2). Although particle and verb are separated by V2, suggesting that SCVs like opbellen are constructed in the syntax, the meaning of the combination, ‘to phone’, is not completely predictable from its constituent elements bellen ‘ring’ and op ‘up’, which points to an analysis of the SCV as a phrasal verb, stored in the lexicon as one unit (Booij 1998). Thus, these SCVs are hybrid in nature, and Germanic SCVs are not alone here; they seem to descend quite straightforwardly from the preverb+verb combination discussed for Indo-European in Watkins (1964). Similar phenomena are discussed for languages as diverse as Hungarian in Ackerman (1987), the Brazilian language Nade¨b in Weir (1986), Georgian and Udi in Harris (this volume), Northern Australian languages as in Schultze-Berndt (this volume). What such constructions have in common is precisely this paradox between apparently being a lexical combination on the one hand, and being syntactically constructed on the other hand. To resolve this paradox, our hypothesis is that the nonverbal part of the
Particles and prefixes in Dutch and English
81
SCV is a secondary predicate in origin. We will motivate this assumption in the next subsection. Tracing the historical origin from this perspective, we will see that the syntactic behaviour of the particle reflects this provenance. 2.2. Particles in present-day Dutch The erstwhile predicate status of particles is supported by various word order phenomena. If we regard particles as (derived from) predicates, the separation of verb and particle by the operation of the verb-second rule follows naturally, as other predicates are also stranded by verb movement in the root clause. Let’s first elaborate further on Dutch: the two different particle positions shown in (1a-b) above are exactly matched by the position of ‘regular’ secondary predicates like groen ‘green’ in (2a-b): (2)
a.
Jan zegt dat hij de deur morgen groen verft John says that he the door tomorrow green paints ‘John says that he will paint the door green tomorrow’
b. Jan verft de deur groen. John paints the door green ‘John paints the door green’ Another positional quirk of the particle is that it can be separated from the verb by auxiliaries in verb-raising constructions, as in (3a-c) and this too is paralleled by well-established predicates in (4a-c): (3)
a.
dat Jan zijn moeder op probeert te bellen that Jan his mother up tries to phone ‘that John tries to phone up his mother’
b. dat Jan zijn moeder op heeft gebeld that Jan his mother up has phoned ‘that John has phoned up his mother’
(4)
c.
dat Jan zijn moeder op zal bellen that Jan his mother up will phone ‘that John wants to phone up his mother’
a.
dat Jan de deur groen probeert te verven that Jan the door green tries to paint ‘that John tries to paint the door green’
b. dat Jan de deur groen heeft geverfd that Jan the door green has painted ‘that John has painted the door green’
82
A. van Kemenade and B. Los
c.
dat Jan de deur groen wil verven that Jan the door green wants paint ‘that John wants to paint the door green’
There are, however, some additional particle positions from which other predicates are barred: particles can occur inside the verb cluster in verb-raising constructions such as (5a-b), unlike resultative predicates (6a-b), which are only allowed similarly close to the verb in the Infinitive-finite verb order as in (7a-b): (5)
a.
dat Jan zijn moeder heeft opgebeld that John his mother has up-phoned ‘that John has phoned his mother up’
b. dat Jan zijn moeder morgen zal opbellen that John his mother tomorrow will up-phone ‘that John will phone up his mother tomorrow’ (6)
a.
*dat Jan de deur heeft groen geverfd that John the door has green painted ‘that John has painted the door green’
b. *dat Jan de deur zal groen verven that John the door will green paint ‘that John will paint the door green’ (7)
a.
dat Jan de deur groen geverfd heeft that John the door green painted has ‘that John has painted the door green
b. dat Jan de deur groen verven wil that John the door green paint wants ‘that John wants to paint the door green’ It looks as if the particle has been incorporated into the verb in (5a-b). The reality of such a special ‘particle order’ is further supported by the existence and behaviour of SCVs in which the nonverbal part is of a different word category, i.e. AP-verb combinations as in (8a) and (fossilized) PP-verb combinations as in (8b). Some of these PPs are no longer synchronically analysed as PPs but either as APs (e.g. tevreden ‘content’, lit. ‘at peace’) or as unanalysable particles (e.g. tegemoet, tekeer): (8)
a.
leegscheppen ‘empty’ (lit. ‘empty-scoop’), goedkeuren ‘approve’ (lit. ‘good-judge’), volgooien ‘fill up’ (lit. ‘full-throw’), losmaken ‘loosen’ (lit. ‘loose-make’), schoonmaken ‘clean’ (lit. ‘clean-make’), blootstaan
Particles and prefixes in Dutch and English
83
‘be exposed to’ (lit. ‘naked-stand’), vreemdgaan ‘have an extra-marital affair’ (lit. ‘foreign-go’), etc. b.
teloorgaan ‘be lost’ (lit. ‘to loss go’), te gronde richten ‘ruin’ (lit.‘to ground force’), tewerkstellen ‘employ’ (lit. ‘to work put’), tegemoetkomen ‘go to meet’ (lit. ‘to meet come’), tekeergaan ‘rave, storm, wreak havoc’ (orig. ‘parry blows, attack’ from the Middledutch noun keer ‘turn, parry’), terechtstellen ‘execute’ (lit. ‘to justice put’), etc.
The secondary-predicate origins of such adjective-verb combinations are more transparent than those of the particle-SCVs: adjectives are the typical nonverbal predicate category, and postulating a secondary predicate (SP) analysis in synchronic Dutch for these SCVs is rather less controversial than for the particle SCVs. Nevertheless, as Koopman (1995) observes, these adjectival SCVs behave like SCVs with a particle in that they allow both word orders: the one in (9a), cf. (4) above, but also that in (9b) from which ‘nonincorporated’ predicates are barred, cf. (5a-b): (9)
a.
dat Jan zijn auto schoon zal maken that John his car clean will make ‘that John will clean his car’
b. dat Jan zijn auto zal schoon maken that John his car will clean make ‘that John will clean his car’ A telling difference emerges when the adjective in (9) is modified, as in (10), in which case the order in (10a) emerges as the only possibility. The order in (9b) and (10b) apparently requires a ‘bare’ head, suggesting incorporation. (10)
a.
dat Jan zijn auto erg schoon zal maken that John his car quite clean will make ‘that John will clean his car very well’
b. *dat Jan zijn auto zal erg schoon maken that John his car will quite clean make ‘that John will clean his car very well’ We propose to account for these data by assuming that SCVs derive historically from secondary predicates, with a (simplified) Small Clause structure as in (11).
84
A. van Kemenade and B. Los
The linear proximity of predicate and (clause-final) verb, and the special semantics associated with the construction (which will be discussed in greater detail in section 3) in time led to varying degrees of lexicalization, involving a rebracketing of the structure in (11), repeated in (12a), into a structure like (12b). (12)
a. b.
[de auto [schoon]] maken, [zijn moeder [op]] bellen de auto [schoonmaken], zijn moeder [opbellen]
It is this reanalysis (through incorporation and/or subsequent lexicalization) which allows the particle-verb combination to serve as input for word formation on a par with simplex verbs, and to behave as a word with respect to VerbRaising, as in (5b) and (9b). Most intriguingly, the reanalysis represented in (12) did not lead to any loss in phrasal behaviour, as we saw in (2)–(4) above: separation by V2, and by verbal clitics like infinitival te, inflectional geremained, as did the heavy stress on the particle. The reanalysis led to the rise of a new category ‘particle’, which comes with its own SCV-package of syntactic behaviour on the one hand, inherited from its earlier predicate status, and lexical behaviour on the other hand, which is the result of reanalysis.
2.3. Particles in present-day English The issue of separability is also relevant to present-day English particles and predicates. English VO syntax makes two positions available for English predicates: an unmarked one with the secondary predicate following its subject as in (13a), and a marked one with the positions of subject and predicate reversed, which is only acceptable with long and complex Small Clause subjects. Verb and predicate become adjacent.
Particles and prefixes in Dutch and English
(13)
a. b.
85
She stuffed [NP all the documents containing incriminating evidence] [PP into her briefcase] She stuffed [PP into her briefcase] [NP all the documents containing incriminating evidence]
Particles show the same two positions, but there is no length or complexity requirement, apart from the fact that pronouns are generally barred from the ‘extraposed’ position, unless they have contrastive stress, as in (15). (14)
a. b. c. d.
He phoned his mother up He phoned up his mother He phoned her up *He phoned up her
(15) ‘If you force your confidence upon me, Mr. Headstone, I’ll give up every word of it. Mind! Take notice. I’ll give it up, and I’ll give up you. I will!’ (Dickens, 1919 [1865]. Our Mutual Friend, London: Dent, 673) The latter point was noted at least as far back as Fairclough (1965: 61, quoted in Lipka 1972: 25). There is, then, a word order difference between regular predicates and particles in that the latter lack any length or complexity requirements on the NP in the ‘extraposed’ order. As a result, the particle ends up adjacent to V far more frequently than the regular predicate, which paves the way for the verb and particle developing greater morphological unity than the verb and predicate, reminiscent of the adjacency of the Dutch verb and particle under Verb-Raising. We saw in (8a-b) that word order phenomena in Dutch show that SCVs are not restricted to particle-verb combinations only but also include AP-verb and PP-verb combinations; the same is true of word order phenomena in the English phrasal verb illustrated in (14), which similarly extends to V+AP (16a) and V+PP combinations (16b) (see also Quirk et al. 1985: 734, 1167; Claridge 2000: 66–70, 153; Denison 1981: 36–37). (16)
a.
break/blow/blast/cut/fling/push/whisk open, cut/stop short, bleach white, blow/keep/make clear, put straight, let/set free, think fit, cast/let/pry/shake/wrestle loose, strip naked etc.
b.
bring to light, put in execution, take in hand, call to mind, call in question, take into consideration etc.
As in Dutch, the list in (16b) can be extended almost indefinitely with items in which the PP has been grammaticalized into an AP or particle. On in such PPs tends to be reduced to a:
86
A. van Kemenade and B. Los
(17) carry aloft (>on loft), set alight (>on light), take apart (>on part), put awry (>on wry), carry around (>on round), keep asunder (>on sunder), set afoot (>on foot) etc. We saw in the previous section that the AP in a Dutch SCV needs to be bare, and the same appears to be true for the AP in an English multi-word verb, which increasingly resists modification as it lexicalizes (see for some descriptive discussion Claridge 2000: 68, 157). The need for bare heads may also account for the fusion of the PP into a single word as in the examples in (17). Finally, as with Dutch opbelbaar, English phrasal verbs may also be input to derivational processes that do not build on phrases, as in come-at-able, get-at-able or lookerson (all from OED). All this points to a close historical connection between particle and secondary predicate.
2.4. Particles as grammaticalized predicates Secondary predicates (in traditional grammars referred to as ‘object complements’ (Quirk and Greenbaum 1973) or ‘object attributes’(Aarts and Aarts 1982, Aarts 1989)) are verbal constructions in which an embedded predicate denotes the result of the action of the verb. Typical examples usually contain an adjective phrase as predicate and a ‘light verb’ as its verb, as in (18a-b). (18)
a. b.
He made his papers available on the internet He kept the doors open
The predicate and the preceding NP are in a subject-predicate relation: as a result of his action, his papers are available on the internet (18a); the doors are open (18b). Predicates are by no means restricted to adjective phrases: NPs or PPs are also possible. Likewise, the verb is not restricted to a light verb but may be any verb that can indicate the means or manner by which the result was reached, or even a verb that lexicalizes the predicate itself. The construction is extremely productive, witness these real-life examples collected by Rappaport Hovav and Levin (2001), with a ‘means’ verb and an adjective phrase in (19a), a ‘manner’ verb and an adjective phrase in (19b) and a ‘lexicalized predicate verb’ with a PP in (19c): (19)
a.
Last night, the dog poked me awake every hour to go outside (The Toronto Sun, 27 Nov. 1994, p. 6)
b.
Sudse cooked them all into a premature death with her wild food. (P. Chute. 1987. Castine. Garden City, NY: Doubleday, p. 78)
Particles and prefixes in Dutch and English
c.
87
She might employ it [her body] as a weapon – fall forward and flatten me wafer-thin. (Delia Ephron. 2000. Big City Eyes, New York: Bantam, p. 92)
The observation that particles share many characteristics with predicates has a long history in the literature, from at least the early fifties onwards (Anthony 1953: 86), and can be found in, e.g., Fraser (1965: 82ff), Legum (1968: 55ff) and Bolinger (1971: 37ff). The parallels between the two structures are semantic (resultative meaning), syntactic (same verbs, same word orders) and intonational,2 and have led some scholars to posit a predicate analysis for all particles (e.g. Grewendorf 1990, von Stechow 1993, Den Dikken 1995). The similarities between English, German and Dutch particles are striking. They all exhibit the same characteristics: a tendency to transitivize, the phenomenon of ‘unselected’ or ‘transferred’ objects and the problem of seemingly ‘redundant’ particles. In spite of the claim made in Neeleman and Weerman (1993) that English phrasal verbs are syntactic structures and Dutch SCVs morphological, there are sufficient parallels to warrant subsuming them under the label ‘SCV’: a object poised on the interface between syntax and morphology with properties of its own, some inherited from the syntactic structure that spawned it, some newly acquired as part of its special lexicalised status as SCV. The different ordering of the particle in Dutch/German on the one hand and English on the other is due to the vicissitudes of syntactic development: particles are no longer preverbal in English because of the loss of OV order in ME; this means that predicates, and the particles deriving from them, are no longer on the left of the verb. Dutch and German have a special ‘particle-syntax’ too. In Dutch, particles may remain adjacent to the verb in Verb Raising constructions, unlike genuine secondary predicates (see (5)–(6) above; for details see Booij 2002: 206). In German it is just the other way around: particles insists on strict adjacency to the verb, whereas genuine secondary predicates may, but need not, be adjacent (Dehe´ et al. 2002: 5). Further support for the predicate origin of the particle is that some SCVs are synchronically recoverable and could warrant a syntactic derivation. To be synchronically recoverable, the meaning of the particle should be transparent, and the meaning of the verb+particle combination compositional. In such cases, the particle can be construed as the predicate in a copula construction (Booij 1998: 8). Examples of such SCVs are afmaken ‘finish’, lit. ‘off-make’ and opeten ‘eat up’. Both qualify as synchronically recoverable: af and op have a predictable predicate meaning: (20)
a.
Mijn huiswerk is af ‘My homework is finished’ (lit. ‘off’)
b.
Het eten is op ‘The food is gone’ (lit. ‘up’)
88
A. van Kemenade and B. Los
There is evidence in at least two other Germanic languages of such a distinction between predicative (i.e. particles that are independently attested in predicate function) and non-predicative particles. Swedish non-predicative particles only allow the V – Prt – NP order, even with pronominal NPs, whereas predicative particles also allow the V – NP – Prt order (Vinka 1999 quoted in Den Dikken 2002, Toivonen 2001). A similar distinction has been reported from language acquisition facts in Sawyer (1999). It is therefore tempting to grant the particle full predicate status in constructions as in (20). This is further supported by the fact that in these transparent cases, it appears to be possible to move the predicate to first position in the root clause, indicating that it acts as a constituent: (21)
´ f maakte hij het boek niet a. A off makes he the book not ‘He did not finish the book’ ´ p at hij zijn eten niet b. O up ate he his dinner not ‘He did not finish his dinner’
The possibility of topicalization of the particle in (21) would seem to suggest a syntactic difference between SCVs with a compositional meaning, and those with a noncompositional meaning: there is a contrast in this respect between opeten ‘eat up’on the one hand, as in (21b), and e.g. opbellen ‘call up’ on the other hand: (22)
´ p belde hij zijn moeder niet *O up phoned he his mother not ‘He did not phone up his mother’
It should be noted on the other hand that this distinction is not entirely without problems. If there is a syntactic distinction between (21b) and (22), we would expect to see this reflected in facts about Verb-raising as discussed above: if op in (21b) is a constituent, we might expect it to be different from op in (22) with respect to V-raising, but it isn’t: (23)
a.
dat hij zijn eten niet wil opeten that he his dinner not wants up-eat ‘that he doesn’t want to eat his dinner up’
b. dat hij zijn moeder niet wil opbellen that he his mother not wants up-phone ‘that he doesn’t want to phone his mother’ If we want to account for the contrast in (21b) and (22) in terms of constituent
Particles and prefixes in Dutch and English
89
status for op in (21b) and nonconstituent status in (22), we would, in order to account for the lack of contrast in (23), have to make the additional assumption that op in (23a) undergoes head-incorporation in the syntax, or assume two different structures for opeten and opbellen.3 Recoverability of the meaning of the particle is complicated by the fact that many predicate counterparts of particles have more than one meaning, again depending on the context: om ‘about’ may mean ‘amenable to a change of plan’ when said of the members of a boardroom, but ‘entailing a detour’ when said of a route; aan ‘on’ may mean ‘going’ when said of a relationship but ‘on’ when said of an electrical appliance, etc. This complicates any synchronic or diachronic picture. Although we postulate an origin for the particle as a secondary predicate, analysed as a SC-head, we have also quoted good evidence that such an analysis is not tenable in its entirety for present-day English and Dutch. Although the SCV (and ICV, as we will argue below) construction started out as a syntactic object, some aspects of its behaviour (Verb-Raising facts as in (5b) above, and its ability to serve as input for word formation) clearly indicate that they have become a separate category. This means that new SCVs are quite likely to be formed on the SCV-template without going through a ‘syntactic’ stage first. The same is true, as we will see, with Dutch ICVs, which originally came into being as grammaticalizations of Small Clauses and SCVs, but have clearly acquired a dynamic of their own allowing new ICVs to be produced from scratch, as if the verbal prefix is a derivational affix, without any intervening SC or SCV stages. The syntactic template of the SC has remained available to the West-Germanic languages throughout, which means that there are, at least for Dutch, three syntactico-morphological templates in synchronic use to express change of state. Before we discuss ICVs, we will turn to the semantic core shared by various SCVs.
3. THE LEXICAL CONCEPTUAL STRUCTURE OF SECONDARY PREDICATES 3.1. Introducing the LCS Particles in SCV-constructions have long been known to share the ‘resultative’ meaning of adjectival and prepositional secondary predicates. Visser employs the term ‘effective adverb’ (1963: I,597), taken from Curme (see Denison 1981: 64); Lipka observes that both German and English particles indicate the result, and often function much like adjectives (Lipka 1972: 115–116). We will see in section 4 that there is a third category with resultative meaning: the prefixes of Inseparable Complex Verbs (ICVs). We will assume a diachronic connection with predicates and particles here: prefixes have become bound morphemes
90
A. van Kemenade and B. Los
although they, too, once started out as syntactic objects. One way to discuss these morphosyntactically so different categories is to focus on the semantic features that they have in common. We will adopt the semantic representation of resultative predicates in the form of the Lexical Conceptual Structure (LCS, Jackendoff 1990) in the simplified form of (24) from Spencer and Zaretskaya (1998: 6): (24) [CAUSE[ACT (x)], BECOME[W(y)]], BY [V(x)]] For a typical secondary predicate as in (2) the variables may be filled in as follows: (25)
a. b.
He painted the door green [CAUSE[ACT (he), BECOME[GREEN(door)]], BY [PAINT(he)]]
As the parallels between secondary predicates and particles stress ‘result’ as a typical common element, we assume as a first hypothesis that SCVs started out as resultatives expressing the resultative LCS of (24) (henceforth ‘R-LCS’).4 A number of quirky characteristics of SCVs fall out automatically: one is the shifting meaning of the secondary predicate (W) which may vary on a scale between extremely specific and extremely abstract. It is the variability of the content of W that accounts for the wide range of constructions that may encode the LCS in (24): from syntactically defined secondary predicates where W has a specific, transparent meaning to SCVs and phrasal verbs where the meaning of W may become so bleached that it merely conveys an endpoint to the activity. The degree of bleaching of W correlates strongly with the closeness of the bond between W and the verb. If the morphosyntax of the language allows it, W may in time develop into a verbal prefix, though still encoding a change of state. The second point to note is the mismatch between syntactic and semantic embedding: the ‘core predicate’ W, although semantically primary and rendered prominent by stress, is the most deeply embedded constituent syntactically; similarly, the V, prominently encoded syntactically as a verb and therefore in theory expected to play an important role as licenser of arguments, is tucked away in the LCS in a peripheral adjunct position. This mismatch accounts for the phenomenon of the ‘unselected object’ which is so frequently associated with the SCV-construction. We will discuss these points in greater detail in the next sections.
3.2. The interpretation of the predicate W The core predicate W in the R-LCS template marks the endpoint of the activity, as the activity will stop when the variable y has reached a certain state – either
Particles and prefixes in Dutch and English
91
a quality, expressed by an Adjective Phrase such as groen in (2) or schoon in (9), or a position, expressed by a Prepositional Phrase, as in (26a), with an LCS as in (26b): (26)
a.
He took the ring off his finger
b.
[CAUSE[ACT (he), BECOME[OFF HIS FINGER(ring)]], BY [TAKE(he)]]
A very prominent characteristic of the R-LCS template is the massive tendency for the constructions encoding it to lexicalize, with their syntactic origins becoming less and less transparent over time. When W is expressed by a PP, its NP complement tends to disappear. This may happen by fusion to the prepositional head, as in items like awry and aloft in (17) in which the obsolescent nouns wry ‘tortuous movement’ and loft ‘sky’ are no longer recoverable and the synchronic categorial status of the phrase itself is in doubt (particle? adjective? adverb?). Alternatively, it may remain implicit, as is possible with example (26) which could just as well read He took the ring off or He took off the ring. As the minimal requirement of the core predicate W is to indicate a change of state, a prepositional head on its own is enough to express a change in position. The fact that these bare heads are used as a convenient shorthand for a change of state, with the precise meaning to be negotiated pragmatically in interaction with the verb, leads to the development of a range of idiosyncratic meanings. This means that particles in particular (as opposed to adjectives which have more robustly lexical meaning) are unlikely to have a single semantic interpretation, and any attempt at identifying a core meaning will usually yield a highly abstract one. The meaning of many SCVs, then, unlike other predicates, is often noncompositional, and hence not always recoverable synchronically. This seems to preclude a synchronic SP analysis for such SCVs in Dutch and English, even though their separability, as we argued in section 2, points to the SP as its diachronic origin. 3.3. Unselected objects The R-LCS offers helpful insights into another phenomenon that unifies SPs, SCVs and ICVs: their well-known transitivizing effect. Although many of the ‘light’ verbs like make, get, give, keep, let, put, set and the like that typically form SPs, SCVs and ICVs are transitive to start with, intransitive verbs, particularly unergatives, are also robustly attested. We give English examples for the SP and SCV-constructions, and resort to Dutch for ICV-examples: (27)
a.
The small band ... played the company into the supper-room (OED1898)
92
(28)
A. van Kemenade and B. Los
b.
Handel ... asked the organist to permit him to play the people out (OED 1823)
c.
They accused the minister of playing down the number of the unemployed
d.
Handel bespeelde het orgel Handel be-played the organ ‘Handel played the organ’
a.
He worked his fingers to the bone
b.
He worked his way up
c.
Hij bewerkte het toneelstuk voor de televisie He be-worked the play for the television ‘He adapted the play for television’
Although the examples are not as spectacular as Spencer and Zaretskaya’s They drank the pub dry (Spencer and Zaretskaya 1998: 2), or the ones in Jackendoff (1997), the principle is the same: the objects the company, the people, the number of the unemployed, the organ in (27) and his fingers, his way, the play in (28) do not fit the selectional restrictions of the higher verbs, play/spelen and work/werken. The relative position of the verb and the object in the R-LCS show that the V, although prominently encoded syntactically as verb, is in an adjunct position. The object y is not theta-marked by V but by the predicate in a secondary predicate construction (in effect the exact parallel of the ‘Exceptional Case-Marking Construction’ in GB theory, with its mismatch between case- and theta-role assignment). This accounts for the ease with which secondary predicate constructions build on intransitive verbs. All objects in an SP, SCV or ICV construction are in fact ‘unselected’, although the loss of prefixation has obscured this in English.5 The unselected nature of the object shows through in quirks like (29a-b): (29)
a.
das Wasser la¨uft aus/der Eimer la¨uft aus (Lipka 1972: 94)
b.
John poured out the bucket/John poured out the water (McIntyre 2000)
c.
Clear out a river (by removing mud)/clear out mud (from a river) (Lipka, ibid.)
The adjunct-like ‘instrumental’ role of the verb in the LCS template goes some way to account for the often-observed fact that it appears to be the particle that selects the verb rather than the other way around (see Lipka’s (1972) discussion of out and up; note also the V+AP combinations of (16a) above which centre
Particles and prefixes in Dutch and English
93
around a limited list of adjectives). The verb in those cases is not a ‘light’ verb but far more specific, either (a conversion of) the instrument used in causing the object y to reach the state W (i.e. ‘means’), as in (30), a description of the specific action required to reach the state W (i.e. ‘manner’) as in (31), or a conversion of the predicate W itself (32). (30) boot out, bowl over, branch out, brick up, brush up, buckle up, elbow out, fork out, hammer out, hand over, hem in, knuckle down, pan out, poke about, patch up, peg down/out/away, rake up, rule out, sally forth, seal off, tick off, tide over, top up, trail off, worm out, zip up (31) blot out, bob up, butt in, chew up, chime in, chip in,6 chuck out, crop up, dole out, edge away,7 eke out,8 pare down, peter out,9 point out, polish up, root up/out, rub out, snap up, trot out, veer round, wind up (32) back off/away, brazen out, cheer up, clean up/out/off/away, clear up/out/off/away, crack up, free up, gloss over, open up/out, parcel out, pretty up, round up/off (Similar groupings are made by Lipka 1972: 98–114). Although one may quibble over the classification of individual items (e.g. is hammer out a conversion of the noun hammer or was the SCV built on a already existing conversion of that noun, or is the verb hammer the direct descendant of OE hamerian, hamorian ‘to hammer’ with loss of its derivational morphology?, etc.), the overall tendencies are clear. With the SCVs of (30)–(32) it is the particle that provides the template rather than the verb, giving rise to combinations containing verbs that are never or rarely used without the particle, many of them conversions. It seems that some new verbs in turn derive from the SCV rather than the other way around. The OED lists pretty up first (first attestation: 1916), while pretty used as a verb on its own is not attested until 1953. The same lists as in (30)–(32) could be made for Dutch, even though conversion is not as widely available to Dutch as it is to English morphology.10 The proliferation of such conversions is made possible by the central role of the predicate W, and by the fact that the R-LCS template gives a complex event even before any of the variables have been filled in; ‘CAUSE’, ‘ACT’, ‘BECOME’ are already there by default.
3.4. Leaving the resultative LCS There are two groups of SCVs that participate fully in the existing syntactic SCV-patterns without R-LCS semantics: the extremely productive durative particles like German los, Dutch door and English on that create intransitive rather than transitive combinations (McIntyre 2001), and the group of ‘postpositions’ in German, Dutch and Old English (on the model of he cwæþ him to ‘he spoke
94
A. van Kemenade and B. Los
to him,’ Lit. ‘he spoke him to’). Whereas the first group can be shown to derive historically from SCVs with R-LCS semantics (McIntyre 2001), this is not true of the second group, whose diachrony still remains to be investigated. Assuming the R-LCS template for SCVs, then, helps to identify groups that may be the result of different developments than the ones discussed here.
4. INSEPARABLE COMPLEX VERBS So far, we have established an analysis for SCVs in terms of a resultative LCS, which is realized syntactically as a Small Clause, on a par with other resultative secondary predicate constructions. There is a third type of construction to which the resultative LCS is relevant: the ICV. There are a number of cognate verbal prefixes in German, Dutch and Old English which derive from prepositions or adjectives (including past participles); the system is almost completely moribund in Present-day English. Many of these prefixed verbs conform to the R-LCS, as is shown by their transitivizing effect and their telic aktionsart. We will discuss two of them at length: be-, from bi ‘around’, and for-/ver-, from various origins (*fer-, fra-, fur-), representing various ablaut-grades of the same root. The transitivizing effect of be- has long been noted (e.g. Hoekstra, Lansu and Westerduin (1987); the object is invariably fully affected (Booij 1992: 56). A comparison of the prefixed form with its simplex from German is given in (33): (33)
a.
Er gießt Wasser auf die Blumen ‘He pours water on the flowers’
b.
Er begießt die Blumen ‘He waters the flowers’ (Dutch: gieten/begieten, Old English geotan/begeotan)
This is the equivalent to the locative alternation in English (load hay onto the waggon, load the waggon with hay), which is no longer marked by a prefix. English be- has a very limited use, mainly in past participles: bespectacled, becardiganed. Be- is fairly productive in German and Dutch, and its meaning appears to be more unified than in Old English, where we also find it as a privative verb (beheafdian ‘behead’, behorsian ‘deprive of horses’, befotian ‘cut off someone’s feet’) where German and Dutch use ent-/ont-(enthaupten, onthoofden), as a pejorative verb (belædan ‘to lead astray’) where German and Dutch use ver- (verleiten, verleiden) and with adjectives (‘a conversion of the predicate W itself’, i.e. the group of verbs as in (32) above): benac(od)ian ‘lay bare’, from nacod ‘naked’; the productive prefix in German and Dutch for these
Particles and prefixes in Dutch and English
95
predicate-conversions is ver- for Dutch (verkorten ‘shorten’, verkoelen ‘cool off/down’) and ver- or er- in German (verku¨rzen, erkalten, idem). These specializations of function are probably a later development, as fossilized privatives etc. can be found in German and Dutch, e.g. benehmen/benemen ‘take away’ (beside entnehmen, ontnemen, idem). The R-LCS is clearly present in pejorative and privative verbs, as is also evident from their SCV-counterparts with away/weg or off/af (take away/off, weg/afnemen; lead away/off, weg/afleiden), but also recoverable in ‘purely transitivizing’ be- of (33b), which is a W expressing the very abstract goal ‘completely affected’ (cf. also Booij’s LCS for be- in Booij 1992: 56): (34) [CAUSE[ACT (he), BECOME[be(flowers)]], BY [pouring (he)]] The verbs are the typical R-LCS verbs we noted above: either ‘light verbs’, ‘means’ or ‘manner’, with a few examples of the fourth group of verbs that lexicalize the predicate (as in (32)). The fact that the R-LCS is still recoverable suggests that the prefix – W – is not completely devoid of lexical content, however abstract this content may have become. In some fossilized formations the R-LCS is no longer synchronically recoverable (cf. Present-day English become, beget, begrudge).11 The second prefix, for-/ver-, has many functions (Leopold 1977[1907], Lieber and Baayen 1993), the most prominent one common to all three languages (that is, German, Dutch and Old English; for- has practically disappeared from Present-day English apart from isolated fossils like forlorn) is ‘to rack and ruin’ or ‘away’, as in (35). (35) Ger./Du. rotten ‘rot’, verrotten ‘rot away’; OE rotian, forrotian; lassen/laten ‘let’, verlassen/verlaten ‘abandon’; OE lætan, forlætan; werfen/werpen ‘throw’, verwerfen/verwerpen ‘reject’; OE weorpan, forweorpan An extremely productive use in Modern Dutch is as a ‘predicate lexicalizer’ as in (36), creating verbs of the type described in (32) above. German uses er- here: ermu¨den ‘tire out’, erhitzen ‘heat up’, erleuchten ‘illuminate’ etc. (36) Du. verarmen ‘become impoverished’ (<arm ‘poor’), veraangenamen ‘sweeten, make more pleasant’ (
96
A. van Kemenade and B. Los
object, as with be-, fully affected. As with be-, the R-LCS is for the most part still in place for Dutch and German. Other prefixes of the same diachronic age include German er-, from uz ‘out’, as in erarbeiten ‘work out’ and its cognate a- in OE, full-/ voll-/voll- from the adjective fulla ‘full’, mis- from missa, originally a past participle meaning ‘changed, turned’, on-/a-/ont-/ent- from anda ‘against’ (OE oþ- derives from a variant unþa), and to-/te-/zer- of uncertain origin, meaning ‘in pieces’ or perhaps ‘in two’ (cf. Gothic dis-). Ge- from ga also appears to have once belonged to this list, but developed into a marker of grammatical (in this case perfective) aspect as a past participle marker in German and Dutch, having disappeared altogether in English. Its former membership of the class of resultative prefixes is evident from the fact that in Dutch it is in complementary distribution with the other members: e.g. the past participle of the Dutch verb ontmoeten ‘meet’ is ontmoet, not *geontmoet.
5. THE ORIGIN OF SCVs AND ICVs Gothic is a representative of East Germanic, a branch of Germanic of which no descendant has survived. As English and Dutch belong to West Germanic, a different branch altogether, Gothic is not a direct ancestor of Modern English or Modern Dutch. Its importance lies in the fact that it represents the earliest extant text in a Germanic language (apart from isolated runic inscriptions on spears, stones or other artefacts), and it is the only Germanic language that is ancient enough to show evidence that ICV-prefixes were once independent words. Wulfila’s translation of the Greek bible in Gothic dates from the middle of the fourth century AD. The extant texts comprise a fragment from the Old Testament (from Nehemiah) and about three-quarters of the New Testament. The second early Germanic dialect that we will investigate in this section is Old English, which is unique among the Germanic languages in the sheer size of extant documents for the period concerned (ca. 700–1100). We will see that Old English ICVs are already solidly prefixal, and its SCV system is fully operational. The Old English data also show how robustly the two systems have been kept distinct in West Germanic, even from the earliest times.
5.1. Gothic Let us first examine Gothic ICVs. With Gothic, we are dealing with considerable diachronic depth, yet the typical characteristics of the resultative LCS are there: prefixes occur with the same four groups of verbs we identified above as typical for the resultative LCS, i.e. light verbs and the verbs in (30)–(32) above; it is the
Particles and prefixes in Dutch and English
97
secondary predicate rather than the verb that theta-marks the object (‘unselected objects’), hence the prefix has a ‘transitivizing’ effect when it combines with an intransitive verb; the secondary predicate denotes a ‘change of state’. The change of state often develops into an endpoint reached, making the event telic. W is, as we found with SCVs in the present-day languages, expressed either by (a cognate of) a preposition, e.g. fra/for, bi/be, uz/a, af/of, or by an adjective, e.g. full, mis (originally a past participle) and possibly Gothic dis (exactly parallel to, though apparently not cognate with, OE to ‘apart’, for which there is no accepted etymology). Adjective-conversions often appear to lack a simplex form, just like the verbs in (32) that lexicalize the predicate. (37)
afdobnan bairan greipan fraþjan hlahjan leiþan letan niman maitan qiþan satjan sigqan sitan slahan sneiþan standan swairban tiuhan wairpan wandjan
‘be silent’ (dumbs ‘dumb’) ‘carry’ versus frabairan ‘tolerate, endure’ (lit. ‘carry forth’) ‘reach for’ versus undgreipan ‘reach’ (lit. ‘grasp against’) ‘think’ versus fulla-fraþjan ‘be in full possession of one’s faculties’ ‘laugh’ versus bihlahjan ‘laugh to scorn’ ‘go’ versus bileiþan ‘leave, leave behind, forsake’ ‘let, leave, permit, suffer’ versus afletan ‘dismiss, forsake, put away, let alone, forgive, fraletan ‘set free’ (lit. ‘let away’) ‘take, take away, receive, accept’ versus afniman ‘take away’ ‘cut, hew’ versus bimaitan ‘circumcise’, afmaitan ‘cut off’ ‘say, tell, name, speak’ versus afqiþan ‘renounce, forsake’ ‘set, put, place’ versus bisatjan ‘beset, set round anything’, afsatjan ‘divorce’ ‘sink’ versus dissigqan ‘sink under’ ‘sit’ versus bisitan ‘sit about, sit near’, andsitan ‘take care, shun’ ‘strike, smite, beat’ versus afslahan ‘kill, slay’ ‘cut, reap’ versus afsneiþan ‘cut off, kill’ ‘stand, stand firm’ versus afstandan ‘ stand off, depart’ ‘wipe’ versus biswairban ‘wipe dry’, afswairban ‘wipe out’ ‘lead, draw, guide’ versus bitiuhan ‘go about, visit’ ‘throw, cast’ versus afwairpan ‘throw away, put away’ ‘turn, turn round’ versus biwandjan ‘shun’, afwandjan ‘turn away’
The Gothic prefixes differ from their counterparts in various stages of English, Dutch and German in that they may be separated from the verbal stem by other morphemes. There are some clear examples to show that ga-, the lexeme which developed into the inflectional past participle morpheme ge- in present-day Dutch and German, and which has disappeared altogether in English, can be
98
A. van Kemenade and B. Los
separated from the verb root by the interrogative particle u or þau ‘then’ which insist on the second position.12 (38)
ga-
u-
laubeis (John 9:35)
ga – int. – leave2sg ‘do you believe’ (39)
ga-þau-laubidedeiþ (John 5:46) ga – then – leave.pret.subj.2pl ‘you then would believe’
As this position can generally be identified as the position after the first word of the clause, this indicates that Gothic prefixes have retained some of their earlier status of full word, even though some of them ( ga-, uz-, fra-) are never found as free morphemes. There may be as many as three particles intervening between the ga- and verb (e.g. (40)) and such particles may be clitic pronouns (as in (41)). (40)
gah-þan-miþ-sandedidum imma broþar (II Cor. 8:18) ga – then – with – send1pl him brother ‘and then we are sending with him a brother’
(41)
ga-u-hwa-sehwi (Mark 8:23) ga – int – anything – see3sg.subj ‘did he see anything?’
According to the literature, ga- is an independent morpheme here.13 However, we would question this putative independent morphological status and point out that Gothic has a more elaborate clitic system, in which illocutions and modalities may be expressed by clitic particles (interrogative -u, realis -uh). It may also be observed that the Gothic situation is widely attested across languages. The historical development giving rise to the situation in Gothic compares well with that described in Harris (this volume) for Udi, and the preverbsystem of the East Caucasian language Akusha Dargi, which has given rise to combinations of spatial prefixes and ‘light’ verbs (again, reminiscent of the resultative LCS) in which the preverbs can be separated from the verb root by other morphemes. They are not SCVs, however, because they are always bound morphemes. An example is (42) (from Van den Berg 2002): (42)
ka – e – b – ik – ib down neg neut light-verb aorist3 ‘it did not fall’
Clitic order is often inherited from earlier syntactic orders, and the peripheral
Particles and prefixes in Dutch and English
99
position of the Gothic preverb in structures such as (38)–(41) may be due to their position in the clause when they were still independent words, much like the SCV-particle in the later West-Germanic dialects which, as we saw above in (3)–(6), may be separated by infinitival te/zu, participial ge and by negator ne/en, in those dialects that preserve it. Ge- is a bound morpheme, ne/en and te/zu (OE to) are clitics, and the resulting strings (Dutch in (43), German in (44)) resemble the ones in (38)–(41); some of them are written as one word: (43)
a.
op te bellen up to phone ‘to phone up’
b. opgebeld up-ge-phoned ‘phoned up’ (past participle)
(44)
a.
Anzuschauen at to look
b. angeschaut at ge looked (past participle) ‘looked at’
‘to look at’
c.
op en belt (Flemish) up neg phones ‘does not phone up’
The possibility of intervening clitic elements may well have led to a reanalysis from full word to bound morpheme if learners did not encounter evidence for the full word status of the particle elsewhere in the grammar. The Gothic data point to a system in which such a reanalysis has apparently taken place, although the transition from SCV to ICV can still be traced. W is reanalysed as a bound morpheme which appears in the left periphery of the verbal compound. A development from SCV to ICV parallel to that in Gothic would not be likely for the continental West-Germanic dialects, because the nonverbal part of the SCV robustly displays a fair degree of independence. Recall that the particle is stranded by V2-movement, which gives it the status of an independent morpheme, and in clause-final position it carries primary stress. The rise of V2 is foreshadowed in Gothic, and lends further support to our analysis of the Gothic complex verbs as ICVs rather than SCVs. 5.1.1. Evidence for the rise of V2 in Gothic Eytho´rsson (1995: 25) asserts confidently that ‘in Gothic the verb seems to be systematically fronted in cases where we might hypothesize an operator element in SpecCP’. At first blush, there does seem to be systematic verb fronting after a typical operator such as the negator ni. Table 1 presents our data from a corpus search in about a quarter of the extant Gothic material and shows the figures of root clauses – the typical V2 environment – with fronted and nonfronted verbs. Not included are clauses consisting of the negator ni and a finite verb (this sequence contains too few constituents to diagnose V2) or of ni, a finite verb and an embedded clause (where the final position of the clause is
100
A. van Kemenade and B. Los
Table 1: V2 after clause-initial ni
probably due to ‘extraposition’, which likewise does not allow of a firm diagnosis of V2). The position of discourse adverbs like nu and auk was ignored, as they do not seem to count for establishing the second position (cf. Weerman 1989: 220, Ferraresi 1997: 112ff). These results look like promising, even exciting and solid evidence that V2 was associated with increased auxiliation, as claimed by e.g. Delbru¨ck (II, 1911: 15) –, until we realize that we have unearthed a peculiarity of the syntax of the New Testament Greek Vorlage here. Gothic reflects Greek word order very faithfully, the only instances of V2 in this corpus that do not reflect V2 in Greek to be found in Kol. 1:23 and II Cor. 2:11.15 Metlen (1933) argues that the Gothic text is too slavish a translation to allow any claims about Gothic syntax, apart from the position of certain discourse particles.16 Metlen shows that even those few instances in which the Gothic text appears to deviate from the Greek Vorlage either look suspiciously like variants found in other mss. of the Greek text, or in the Latin text. There are some instances, however, in which single Greek verb forms require a periphrastic construction in Gothic. As the order of the elements of the periphrasis cannot be prompted by anything in the Vorlage, they probably do reflect authentic Gothic syntax. Meillet (1908) shows that in those cases in which Gothic has to translate a single passive Greek verb form periphrastically with a form of wisan ‘be’ or wairþan ‘become’ followed by a participle, the participle regularly follows the auxiliary in imperatives (suggesting V-movement) but precedes it elsewhere (suggesting OV-syntax). Imperatives, then, may well have spear-headed the development of V2. The imperative (‘V1’) theory is attractive in that it provides a starting point for the generalization of V to C movement to other contexts beside imperatives, given that a V to C movement analysis for imperatives is uncontroversial. The generalization of V2 could then involve the movement, first of wh-phrases and then of negative adverbs to SpecCP. This order of development is suggested by the situation in early OE, where V2 is general with wh-movement, but not yet completed in negative-initial clauses (van Kemenade 2000). Topicalization to SpecCP is more difficult to account for, but may call for different treatment anyway, given the fact that V-movement does not target C here (see the discussion in Fischer et al. 2000: chapter 4, and references cited there), although van Kemenade (1997) shows that topicalization in Old English targets SpecCP.
Particles and prefixes in Dutch and English
101
Table 2: the position of the adverbial particle inn ‘in’ in Gothic
Other deviations that suggest that verb-second is an innovation in Gothic concern particles.17 The asymmetric behaviour of inn and ut in root and subclauses is strongly reminiscent of the behaviour of SCVs in the Dutch examples (1a-b) above: the particle is postverbal in V2-environments (finite forms in root clauses), but preverbal elsewhere (data based on Delbru¨ck (1910), supplemented by a further 29 examples resulting from a corpus search). The asymmetry is not perfect, but there is a clear pattern: inn generally precedes the verb, but may be ‘stranded’ when the verb moves away in the root clause. With ut the pattern of Table 2 is less clear (see again Delbru¨ck 1910: 360).18 A example of a preverbal particle is (45), which shows the particle separated from the preverb+verb combination by the negation ni: (45) saei inn ni atgaggiþ þairh daur in gardan lambe (John 10:1) he-who in not preverb-goes through door in sheepfold qui non intrat ’ o‘ mg` eı’serxo´ menoz ‘he who does not enter the sheepfold by the door’ The semantics of inn, as well as its positions are best explained under a predicate analysis. Examples like (45) are close to the core of the R-LCS. Predicates precede the verb in an OV language, but are stranded by V2 movement. The position of the negator ni in (45) – after the particle – is reminiscent of the WestGermanic negator ne/en in (43c) above. 1.2. Doubling Inn and ut are often found doubled. These doublings apparently act to reinforce (often cognate) preverbs, e.g. ut and us- in (46). If inn and ut are predicates, and express W in the resultative LCS, this probably means that us in (46) no longer has this function: (46) þanuh modags warþ jah ni wilda inngaggan, iþ atta is usgaggands ut bad ina. (Luke 15:28) then – and angry became and not wished in-go, but father his out-coming out asked him
102
A. van Kemenade and B. Los
v’ rci´sqg de` kai` / ou’ k gqelen ei’selqei˜n. o’ de` patg` r au’ tou˜ e’ jelqv’ n pareka´ lei / au’ to´ n. ‘But he was angry and refused to go in. His father, coming out, pleaded with him.’ Usgaggan and utgaggan are both used to translate Greek e’ jeleu´ oetai (e.g. Mat. 8:34, John 10:9) so there was probably no great difference in meaning. Whenever they both appear on one verb, ut is always on the periphery, so utus-V (e.g. utusiddjedun ‘they went out’, Mat. 9:32), never *us-ut-V. This supports an analysis in which us is a prefix and ut a predicate or particle. The emergence of V2 may have acted as a watershed: it highlighted an existing difference between the syntactically defined SP (with predicates or particles like inn and ut) and an ICV-like preverb-system which represents a grammaticalization of an earlier syntactic SP. We must assume that they still represented W in the R-LCS, as do their descendants (see section 4 above); it is only when we find doubling that we must assume that the prefix in that particular instance was no longer felt as a resultative W expressing a change of state and that that function was taken over by the predicate or particle. Examples in which an inseparable prefix is still firmly within the R-LCS include ga- as in (47), which allows of a resultative interpretation, in which gahas a completive sense and qualifies as W conveying telicity: (47)
hausjan beidan brikan fulljan swiltan
‘hear’ versus gahausjan ‘learn’ ‘wait for something’ versus gabeidan ‘put up with’ ‘break’ versus gabrikan ‘break to pieces’ ‘fill’ versus gafulljan ‘fill up’ ‘lie dying’ versus gaswiltan ‘die’
But there are also instances in which ga- appears to have dropped out of the R-LCS: no change of state is expressed, the verb is not a typical member of the four groups characteristic of the R-LCS (the light verbs and those outlined in (30)–(32) above). It seems that ga in instances like (48) (from Streitberg 1920: 196ff) has developed into a marker of perfective aspect: (48)
þiudanos wildedun saihwan þatei jus saihwiþ jah ni gasehwun kings wished3pl see that you see yet not saw3pl ‘kings wished to see what you now see, yet never saw.’ (L. 10:24)
Ga- seems to be changing from a derivational to an inflectional affix, a process complete in Modern Dutch and German, where ge- attaches to the entire category V, which is a typical characteristic of inflection. As noted in section 4, there is one striking set of contexts where ge- is not used as a past participle prefix in present-day Dutch and German and that is when the verb combines with
Particles and prefixes in Dutch and English
103
another inseparable prefix like be-, ver- ont-. An example from Dutch is hij verdeed zijn tijd (lit. he for-did his time) ‘he wasted his time, but Hij heeft zijn tijd verdaan/*ge-verdaan (lit. he has his time for-done) This complementary distribution testifies to the fact that ga-, fra-, und-, etc. once had the same status (that of a derivational morpheme encoding W in the resultative LCS). Ga- is occasionally found doubled. These doublings cannot have the same analysis as that of ut+us, because ga- is only attested as a bound morpheme and cannot be a predicate or particle. As derivational processes are generally not recursive (i.e. they do not apply to their own outputs; Booij 2002: 92), the two ga’s must have different status. The first, outer ga- in (49), then, is probably an aspectual marker, and the inner one a derivational prefix: (49) ga-ga-leikon sik (II Cor. 11.14) ‘change oneself’ 19 A further prefix in Gothic which needs to be mentioned in this context: the preverb fra-, which, like ga-, is no longer attested as a free lexeme, but is still evident in some formations, as in (50a), and has acquired a negative meaning in others (e.g. 50b). It is also attached to verbs that are already negative in meaning (as in (50c)). Here the semantic contribution of fra- is unclear. Both forms are found translating the same Greek form (a’ pole´ sai in L. 9:56, Mat. 10:28, etc.). (50)
a.
bugjan ‘buy’ versus frabugjan ‘sell’, dailjan ‘divide’ versus fradailjan ‘divide up’, giban ‘give’ versus fragiban ‘give away, grant’, letan ‘let’ versus fraletan ‘liberate, let free, leave, let down, permit’.
b.
qiþan ‘say’ versus fraqiþan ‘curse’, kunnan ‘know’ versus frakunnan ‘despise’, waurkjan ‘do, work’ versus frawaurkjan ‘sin’, wilwan ‘rob’ versus frawilwan ‘take forcibly’ 20
c.
lewjan ‘betray’ versus fralewjan ‘betray’, qistjan ‘ruin’ versus fraqistan ‘ruin’
Subsequent generations may well reanalyse such ‘pleonastic’ uses as in (50c) as intensifiers, as in OE, where the same prefix is found as an intensifier with adjectives: fræhræde ‘very fast’, fræfætt ‘very fat’ with no trace of resultativeness. But if learners find it difficult to assign a meaning to the prefix, or feel that this meaning must be shored up by a particle, the prefix is likely to disappear.
5.2. Old English Reflexes of the various prefixes discussed above for Gothic are amply attested in Old English as well. Let us first consider a number of examples:
104
(51)
A. van Kemenade and B. Los
geænan ærnan geæðelian faran flowan frignan gan gieman gnidan grindan habban hawian healdan heawian gemæstan, amæstan amyrðr(i)an, for-, ofamyrgan niman seon settan springan wyrcan
‘unite oneself to’ (an ‘one’) ‘run, ride’ versus geærnan ‘run to, reach, gain by running or riding’ ‘make noble’ (æðel ‘noble’) ‘go’ versus forfaran ‘pass away, die’ ‘flow’ versus beflowan ‘flow around NP, over NP’ ‘ask, inquire’ versus befrinan ‘question NP, learn NP’, gefrignan ‘learn NP by asking’ ‘go’ versus forgan ‘abstain from, lose’, began ‘traverse, surround, practise’ ‘care for, heal’ versus forgieman ‘neglect, transgress’ ‘rub, grind together, crumble’ versus begnidan ‘rub all over’, forgnidan ‘crush’ ‘grate, grind together’ versus forgrindan ‘grind down, ruin’ ‘have’ versus forhabban ‘hold in, keep back, draw back’ ‘gaze on, view, look at’ versus behawian ‘see clearly, take care, consider’ ‘hold’ versus forhealdan ‘forsake, abuse, defile’ ‘hew’ versus forheawian ‘hew to pieces, cut down, kill’ ‘fatten, feed on mast’ (mæst ‘mast’) ‘murder’ (morðor ‘murder’) ‘delight, cheer’ (myrge ‘pleasant, sweet’) ‘take’ versus forniman ‘take away’ ‘see’ versus forseon ‘overlook, despise’ ‘set’ versus forsettan ‘hedge in, obstruct’, besettan appoint, own, surround’ ‘spring’ versus tospringan ‘spring apart’ ‘do, make, perform’ versus fullwyrcan ‘fulfil, complete’
There is clear evidence that prefixes in Old English as in (51) are in an advanced state of grammaticalization. In particular, they are doubled by a particle quite frequently, e.g. the combination of ut ‘out’ and a- ‘out’ (>uz; Kluge 1901: 476, § 283.4, anm; Lehmann 1906): (52)
leoran a-leoran ut-a-leoran
‘go, depart, vanish, die’ ‘depart, flee away’ ‘(cause to) depart, flee away’
(53)
sellan a-sellan ut-a-sellan
‘give, furnish, lend; surrender, give up, betray’ ‘give up, hand over; expel, banish’ ‘grant outright’
Particles and prefixes in Dutch and English
(54)
tynan a-tynan ut-a-tynan
105
‘hedge in, fence, enclose, shut’ ‘shut off, exclude’ ‘exclude’
In Brinton (1988), separable and inseparable prefixes are regarded as functionally equivalent, but the frequency of doublings as in (52–54) casts considerable doubt on this. Rather, it looks as though the inseparable prefixes, phonologically weak as they are, are in the process of losing their distinctive meaning and cease to encode W in the R-LCS. At the same time, the system of separable prefixes is gaining in robustness, and it has all the hallmarks of the R-LCS. The particle system is treated at length in Hiltunen (1983), and some of the syntactic characteristics of Old English particles are discussed in Koopman (1985); van Kemenade (1987), Pintzuk (1991); Fischer et al. (2000). The Old English particles can be separated from the verb in various ways, as illustrated in (55): (55)
a.
Negation forðæm hio nanne swetne wæsðm forð ne bringð because she no sweet fruit forth not brings ‘because it does not produce any sweet fruit.’ 21
b. Infinitive marking þæt him wære alyfed ut to farenne that him was allowed out to go ‘that he was allowed to leave’ c.
Modals in verb clusters ær he ut wolde faran to gefeohte before he out wanted go to fight ‘that giants would raise up a city’
d.
Preposition stranding ealond .. ðæt we ær ut of gongende wæron island .. that we before out from going were ‘island .. from which we had previously put out.’
e.
Verb Second þa sticode him mon þa eagan ut then stuck him someone the eyes out ‘then his eyes were gouged out’
These examples also show that particles in Old English are clearly amenable to a secondary predicate analysis. In all the examples, the particles clearly denote an end state. Hiltunen (1983) makes some important observations in this respect, noting in particular that the meaning of the verb+particle combination
106
A. van Kemenade and B. Los
is always compositional and transparent. This sits well with the analysis in section 2 in which the particle is the predicate of a resultative Small Clause. For present-day Dutch, we saw that a Small Clause analysis has its attractions for semantically transparent cases, although it is not unproblematic. For Old English, where all the cases of Verb and particle combination seem to be semantically transparent, and clearly resultative in meaning, such an analysis gains in force, and may well provide the historical origin of the syntactic template of the present-day construction. An analysis along these lines is presented in Fischer et al. (2000: chapter 6). This analysis is further supported by the fact that particles frequently occur in a topicalized position, which supports the idea that they represent a syntactic constituent: (56)
<MCharm 11, 31>
a.
Forð ic gefare, frind ic gemete ‘forth I go, friends I meet’
b.
Forð þa eode Wistan, þurstanes sunu <Maldon 297> ‘forth then went Wistan, Thurstan’s son’
All the relevant facts show that particles in the Verb+particle combination show the same kind of syntactic robustness as they do in present-day Dutch: (55e), for instance, shows that the particle is stranded by V2-movement. It can also be shown that particles carry primary stress, as they occur in alliterating positions in Old English alliterative poetry: (57)
a.
/ Sie sio bær gearo, | ædre geæfned, / þonne we ut cymen | be this byre ready speedily made when we out come ‘Let the byre be made ready, speedily wrought, when we come out’ Beowulf, 3105–6
b. / He mid Eotenum wearð | he among Eotens became on feonda geweald / forð forlacen, | Beowulf, 902–3 in enemies’ power ‘he was betrayed among the Eotens into the power of the enemies’ The facts in Old English give clear support then, to an analysis in which the particle as the end state W is represented syntactically as the predicate of a resultative Small Clause. Let us now turn to the development in Middle English. 6. SYNTACTIC CHANGES AS A DIAGNOSTIC: THE OV/VO CHANGE IN ENGLISH English underwent some important syntactic changes after the Old English period. One important change was that OV word orders were gradually lost (for discussion, see Fischer et al. 2000 and references cited there).
Particles and prefixes in Dutch and English
107
If the particle construction is still derived syntactically from a fullblown secondary predicate when this change got underway, we would expect the particle position to change, because secondary predicates ceased to occur preverbally as a result of the loss of OV orders. Such a wholesale shift to postverbal position is reported by Hiltunen (1983). The position of the object is, of course, highly relevant, as this marks the position of the Secondary Predicate. To illustrate with two examples of different word orders, (58), with preverbal particle in an early, ‘conservative’ OV text from the south-east can be assumed to have the (simplified) tree structure of (60); the postverbal particle in (59) in a more ‘progressive’ VO dialect from the West Midlands the tree structure of (61). (58) All ðis woreld was þes dieules hus ær Crist come, ðe him ut warp. (Holthausen, Vices and Virtues 111) All this world was the devil’s house before Christ came, who him out cast (=‘who cast him out’) (59) þet is, wið unlust, warpeð hit eft ut (d’ Ardenne, Hali Meidenhad, Bodley, 50) which means, with nausea, throw hit again out (=‘throws it up again’)
108
A. van Kemenade and B. Los
The Middle English facts show that particles ceased to occur in preverbal position with the loss of OV order. The following facts are from a corpus search in the Middle English part of the Helsinki Corpus, for the particles on, forth and out. These particles were selected because they make a fairly transparent contribution to the compositional meaning of the entire verbal complex, which makes a syntactic derivation likely. The figures show a steady decline of preverbal particles:
Table 3: particle-V and V-particle orders in the Helsinki Corpus for on, forth and out
On closer examination, it turns out that only the preverbal particles of the first period follow the syntactic pattern of Old English: they occur in non-V2 contexts. An example is (62), in an embedded clause. In all later periods, the order is extremely marked in that it occurs either in verse texts to meet the demands of rhyme, as in (63), or in slavish translations from Latin, as in (64). In the later periods, some appear to have been reanalysed as ICVs, e.g. (65) (see also the OED under outtake). (62)
Augustinus cwæð þæt þt festen ... ut ascyfð þa yfele Augustine said that that fasting out pushes the evil þohtæs (Cmbodley 46)22 thoughts ‘Augustine said that that fasting pushes the evil thoughts out’
(63) Kyng Alisaunder is out yride – þre noble knizttes ben went hym myde (Cmalisau I, 231) ‘King Alexander has ridden out – three noble knights have gone with him’ (64) And þat he out-kest her sede in terþes and departed hem in kyngdomes (cmearlps, Psalm 105 (106), 26) ‘And that he cast out their seed on the earth and dispersed them in [different] nations’
Particles and prefixes in Dutch and English
109
(65) I oute-take not o creature (Cmcloud 24) ‘I exclude not a single creature’ The idea that the development of the postverbal particle order (the phrasal verb) is due directly to the loss of OV orders is not uncontroversial. Denison (1981) adduces examples like (66) as evidence against a direct relationship between the rise of postverbal particles and the loss of OV order. The point here is that (telic) up is not only found in postverbal position, but is accompanied by a preverbal object. The combination of a postverbal particle and OV order would be unexpected if postverbal orders were due to the loss of OV, according to Denison. (66)
he suor ... þat he alle his castles sculde iiuen up he swore that he all his castles should give up ‘he swore that he would give up all his castles’ (Clark, Peterborough Chron. 1140, 41)
If we assume an underlying OV order, the particle must have been extraposed. Particles are not generally extraposed in the West-Germanic languages, which is why Pintzuk (1996: 250) and Kroch and Taylor (2000) take postverbal particles to be diagnostic for underlying VO order. In early English, however, particles as well as other predicates are extraposed quite comfortably, especially in late OE, as (67a) and contrast between (67b) and (67c) shows:23 (67) a.
and þæt biþ hire miht ... þæt heo gesewen ne beo ute and that will be her strength that she seen not be outside ‘and that will be her strength, that she is not seen outside’ <ÆLS (Martin) 1101–2>
b. and het hine utgan <ÆLS (Martin) 914> and bade him outgo ‘and bade him go out’ and het [NP hine [PRED ut]] gan c.
and het hi gan ut <ÆLS (Martin) 214> ‘and bade them go out’ and het [NP hi] gan [PRED ut]
Another way to account for word orders such as (66) is along the lines of Fischer et al. (2000: chapter 6). They motivate an analysis in which object and particle are generated in postverbal position as a small clause, with the NP his castles as
110
A. van Kemenade and B. Los
in (66) as the subject. The word order as in (66) could readily be derived by scrambling of the NP his castles to a preverbal position. This analysis would be entirely parallel to derivations that must be independently assumed for the positioning of bare infinitives in late Old English and early Middle English. Brinton (1988) argues for a very different contrast between the word order development and the rise of postverbal particles, probably based on the data in Denison (1985: 55–56). However, Denison’s finding that it is ‘‘highly unusual’’ for completive up to occur in preverbal position (Ibid.) does not automatically entail that particles like up could only develop into a marker of telicity once they occurred in postverbal position. There are too many crosslinguistic examples of telic markers in preverbal position to assume that preverbal order is incompatible with such meanings, not only in Dutch, which even developed the telic meaning as a genuine predicate in a copula construction, but also in Russian (Spencer and Zaretskaya 1998) and Hungarian (Ackerman 1987). W in the resultative LCS very easily develops highly abstract meanings. If languages keep their OV orders, their W will always be preverbal. The intimate connection between preverbal W’s and OV orders, and postverbal Ws and VO orders, is illustrated well by the development of the Scandinavian languages. We conclude that the case for the assumption that the rise of postverbal particle position goes hand in hand with the loss of OV word order must be considered convincing. We leave the late Middle English development of particles for further research.
7. SUMMING UP AND CONCLUSION This paper argues that both inseparable (ICV) and separable complex verbs (SCV) in the West-Germanic languages represent grammaticalizations of secondary predicates, constructed in the syntax. What they both have in common with genuine syntactically constructed predicates is their semantics: the majority of ICVs and SCVs still encode a resultative Lexical Conceptual Structure (R-LCS). Typical features arising from the R-LCS are: (i) the fact that the predicate (or particle, or prefix) encodes a change of state; (ii) their transitivizing effect; (iii) the phenomenon of unselected objects and (iv) the types of verbs that combine with the predicate: ‘light’ verbs, ‘manner’ verbs, ‘means’ verbs and verbs that lexicalize the predicate itself. Syntactically, SCVs still betray their predicate origins by their separability and the clear evidence for phrasal status of at least some of its particles; that they have grammaticalized to some extent is shown by 1) a special word order (verb raising in Dutch, NP-‘extraposition’ in English) not found with genuine syntactically-constructed predicates and by 2) the fact that they can serve as input to morphological derivation.
Particles and prefixes in Dutch and English
111
The prefix of an ICV has grammaticalized even further, although the typical hallmarks of the resultative LCS are still recoverable: the prefix adds telic aktionsart (showing its origins as a Change-of-state predicate) and transitivity; and it tends to combine with the four groups of verbs outlined in iii) above. Some prefixes have grammaticalized beyond the R-LCS and either acquired a new function (as in the case of ge- which came to express perfective aspect) or found themselves superseded by the more recent SCV-system, the earliest beginnings of which were charted in section 5, or became meaningless and were lost, as in English. The ICV and SCV systems continued to co-exist in Dutch and German. The loss of preverbal SCV-particles in English we argued to be due to the change in underlying syntax, from OV to VO; the fact this change affected preverbal particles must mean that SCVs were still syntactic rather than morphological objects in Old and Early Middle English.
NOTES 1
The second author’s work reported in this paper was made possible by a grant from the Netherlands Organization for Scientific Research (NWO), grant number 360-70-051. 2 It has been noticed, for German at least, that the intonation patterns for particles differ from regular adverbs: Er versuchte mitzusprechen versus Er versuchte laut zu sprechen (Lipka 1972: 20, quoting Hundsnurscher 1968: 6ff who is in turn quoting Admoni 1966: 51–53). The same point is made in Winkler (1997: 303), quoted in Lu¨deling (1999: 41), who notes that the ambiguity of Bill hat den Laden leer gekauft is resolved by intonation: one reading is Bill bought the shop in an empty condition (depictive) versus Bill bought everything in the shop, so that it was empty (resultative). Winkler finds that particle constructions are stressed like resultatives. 3 The constraints restricting such topicalizations include not so much the requirement that the particle must be used literally, as has often been suggested in the past (e.g. Fraser 1976: 58–59), but semantic transparency of both particle and verb, see Capelle’s 2002 corpus-based study. Note that af in (20a-21a) too, although not literal/directional, is semantically transparent. For German SCVs, contrast has been put forward as a possible factor: the failure of Topicalization in (i) below could be due to the fact that there is no antonym to aufbrechen (*zubrechen), unlike aufmachen in (ii) which contrasts with zumachen (see Stiebels 1996): (i)
*Auf hat er das Schloss gebrochen.
(ii) Auf hat sie die Tu¨r gemacht. But Zeller (2001) shows that contrast is not enough to license (iii): (iii) *An/Ab hat er den Pullover gezogen. See also Zeller (this volume). 4
Cf. also Lipka’s semantic analysis of German auf/zu/los/festbinden (schrauben) and the like: ‘durch Binden (Schrauben) auf-, zu-, los-, fest- machen’ (Lipka 1972: 117).
112 5
A. van Kemenade and B. Los
The English counterpart of (27d), for instance, has no problem using the object with the simplex verb: Handel played the organ where the simplex verb in Dutch would require a PP: Handel speelde op het orgel, lit. ‘Handel played on the organ’. We are not aware of any in-depth study on the effects of the loss of prefixation in English on argument structure and our evidence is so far only anecdotal, based on general observations. 6 chip meaning ‘chop, cut’ 7 edge meaning ‘move edgeways’ 8 eke meaning ‘increase, lengthen, supplement’ 9 peter (origin unknown, but first used in mining) meaning ‘run out and disappear (as a stream, a vein of ore)’ (OED). 10 Dutch opleuken ‘to fun up’ (from the A. leuk ‘funny’) is a case in point. See Booij and van Kemenade (this volume). 11 It is tempting to hypothesize a link between meaning and prosodic status: although beand ver- do not have a full vowel, and therefore cannot form prosodic words of their own, they still do not fully integrate prosodically into their stem (Booij 2002: 170–171). Perhaps it is not accidental that the few examples of full integration that we do find (bleiben/blijven ‘keep, remain’, cf. OE belifan; OE blinnan ‘leave off’ from *be+linnan ‘desist, lose’) appear to have left the R-LCS; they have become ‘aspectualizers’ (Brinton 1988), or activities, as in the case of vreten/freßen/fretan ‘eat like an animal’, from ver/for+eat. 12 The origins of ga- are obscure. It has been related (controversially) to lat. cum-, con‘with’ (Cf. Lehmann 1986). 13 The argument is as follows: word-final obstruents are as a rule devoiced in Gothic. The fact that final obstruents in preverbs are devoiced could point to their word-status: us ‘out’, as in usgaggan ‘go out’ emerges with -z- when there are intervening clitics like interrogative u or the sentence connector –uh: uz-uh-iddja ‘he went out’ (John 16: 28). If -u, -uh form a prosodic word with the word which they follow, as suggested in Hopper (1975: 27) this means that the underlying form is uz, devoicing to us in usgaggan etc. because preverbs are still words (see also Schmidt 1883; Meillet 1908: 95–97; Eytho´rsson 1995: 52, 124). 14 Including instances of skuld/mahts wisan which we take, with Eytho´rsson (1995: 27–28), to be fixed collocations. 15 We are indebted to Lidewij van Gils (Classics Department, Vrije Universiteit) for untangling the Greek text. 16 And even there, the point can be made that these discourse particles are further evidence of the unidiomatic quality of the Gothic text, as they translate Greek ‘gap-fillers’ that in the translation into other languages are usually left untranslated (Metlen 1933: 543). 17 Verb-movement is apparently not restricted to root clauses: Eytho´rsson (1995: 105–106) has some remarkable data about V-movement in indirect questions, marked deviations from the Greek Vorlage, and also markedly deviant from the operation of V2 in the modern WestGermanic dialects. 18 The asymmetry is somewhat obscured by slavish adherence to Greek orders in some cases or the distinct possibility in other cases (notably inngaggan ‘go in’) that the particle+verb have become an ICV. 19 Alternatively, the second ga- may have lost its resultative meaning altogether. This possibility is suggested by the fact that g- in glı´kr (got. ga-leiks) ‘same’ is one of the few survivals of *ga- (or of any other prefix, for that matter) in Old Norse. The fact that g- was completely incorporated into the prosodic word may indicate that it was no longer analysed as belonging to the regular ga-system with the usual ga-semantics (whatever they may have
Particles and prefixes in Dutch and English
113
been). If the same is true of ga- in galeikon, this might explain the doubling in (49) (Cf. the participial form *gegloofd in Dutch Child Language (<*gegeloofd instead of geloofd ‘believed’). 20 But also ‘rob’ in Mat. 11:12, John 10.12 and there used to translate a’ rpa´ fein, which is translated in John 6:15 by the simplex wilwan. 21 Throughout this paper, the reference to an OE text is enclosed in <> and follows the system of short titles as employed in Healey and Venezky (1985). 22 For the ME short references in (62–65), see Kyto¨ (1993). 23 Another argument in Pintzuk, as well as in Kroch and Taylor (2000) is that they take particles to be light elements, like personal pronouns. Given the fact shown above that particles carry primary stress, this cannot be correct.
REFERENCES Aarts, B. (1989). Verb-preposition constructions and small clauses in English. Journal of Linguistics 25, 277–290. Aarts, F. and J. Aarts (1982). English Syntactic Structures. Oxford: Pergamon. Ackerman, F. (1987). Miscreant Morphemes: Phrasal Predicates in Ugric. Diss. Stanford University. Admoni, V.G. (1966). Stroj Sovremennogo Nemckogo Jazyka/Der deutsche Sprachbau. 2nd edition. Moscow/Leningrad. Anthony, E.M. Jr. (1953). Test Frames for Structures with UP in modern American English. Ann Arbor: University Microfilms. Berg, H. van den (2002). Spatial prefixes in Akusha Dargi (East Caucasian). Paper delivered at the 10th International Morphology Conference, Szentendre, Hungary, 9–12 May 2002. Bolinger, D. (1971). The Phrasal Verb in English. Cambridge, Mass.: Harvard University Press. Boogaart (1999). Aspect and Temporal Ordering: A Contrastive Analysis of Dutch and English (LOT Dissertations 27). The Hague: Holland Academic Graphics. Booij, G.E. (1990). The boundary between morphology and syntax: Separable complex verbs in Dutch. In: G. E. Booij and J. van Marle (eds.), Yearbook of Morphology 1990. Dordrecht: Foris, 45–63. Booij, G.E. (1992). Morphology, Semantics and Argument Structure. In: I.M. Roca (ed.), Thematic structure: Its role in grammar. Berlin/New York: Foris, 47–64. Booij, G.E. (1998). Samenkoppelingen en grammaticalisatie. Morfologiedagen 1996, edited by E. Hoekstra and C. Smits. Amsterdam: Meertens Instituut, 6–20. Booij, G.E. (2002). The Morphology of Dutch. Oxford: OUP. Booij, G.E. and T. van Haaften (1988). The external syntax of derived words, evidence from Dutch. In: G. E. Booij and J. van Marle (eds.), Yearbook of Morphology 1988, 29–44. Booij, G.E. and A. van Kemenade (this volume). Introduction. Brinton, L.J. (1988). The Development of English Aspectual Systems: Aspectualizers and postverbal Particles. Cambridge: Cambridge University Press. Cappelle, B. (2002). And up it rises: Particle preposing in English. In: Dehe´ et al., 43–66. Claridge, C. (2000). Multi-word Verbs in early Modern English: A Corpus-based Study. (Language and Computers 32). Amsterdam/Atlanta: Rodopi.
114
A. van Kemenade and B. Los
Dehe´, N., R. Jackendoff, A. McIntyre and S. Urban (2002). Verb-particle Explorations (Interface Explorations 1), Introduction. Berlin/New York: Mouton de Gruyter, 1–20. Delbru¨ck, B. (1910). Zur Stellung des Verbums im Gotischen und Altisla¨ndischen. Beitra¨ge zur Geschichte der deutschen Sprache und Literatur 36, 359–365. Delbru¨ck, B. (1911). Germanische Syntax II: Zur Stellung des Verbums. Abhandlungen der Ko¨niglichen Akademie der Wissenschaften, philologisch-historische Klasse, 28.7. Leipzig: Teubner. Denison, D. (1981). Aspects of the History of English Group-verbs, with particular Attention to the Syntax of the Ormulum, diss. Oxford University. Denison, D. (1985). The origins of completive up in English. Neuphilologische Mitteilungen 86, 37–61. Dikken, M. den (1995). Particles: On the Syntax of Verb-particle, Triadic, and Causative Constructions. Oxford: Oxford University Press. Dikken, M. den (2002). Review of Zeller, J. (2001). Particle Verbs and local Domains. (Linguistik Aktuell/Linguistics Today), Amsterdam/Philadelphia: Benjamins. Journal of Comparative Germanic Linguistics 4, 145–169. Eytho´rsson, Th. (1995). Verbal syntax in the early Germanic Languages. Diss. Cornell University. Fairclough, N.L. (1965). Studies in the Collocation of lexical Items with Prepositions and Adverbs in a Corpus of spoken and written Present-day English. MA thesis, University College London. Ferraresi, G. (1997). Word Order and Phrase Structure in Gothic. Dissertation Stuttgart University. Fischer, O., A. van Kemenade, W. Koopman, and W. van der Wurff (eds.) (2000). The Syntax of early English. Cambridge: Cambridge University Press. Fraser, B. (1965). An Examination of the Verb-particle Construction in English. Dissertation MIT. Fraser, B. (1976). The verb-particle combination in English. New York: Academic Press. Grewendorf, G. (1990). Verbbewegung und Negation im Deutschen. Groninger Arbeiten zur germanistischen Linguistik 30, Groningen: Germanistisch Instituut, University of Groningen, 57–125. Harris, A.C. (this volume) Preverbs and their origins in Georgian and Udi. Healey, A.D. and R.L. Venezky (1985[1980]). A Microfiche Concordance to Old English. Toronto: The Pontifical Institute of Mediaeval Studies. Hiltunen, R. (1983). The Decline of the Prefixes and the Beginnings of the English Phrasal Verb: The Evidence from some Old and Middle English Texts (Annales Universitatis Turkuensis, Series B, 160). Turun Yliopisto (University of Turku, Finland), Turku. Hoekstra, T., M. Lansu and M. Westerduin (1987). Complexe verba. GLOT 10, 61–79. Hopper, P. (1975). The syntax of the simple sentence in Proto-Germanic. The Hague etc.: Mouton. Hundschnurscher, F. (1968). Das System der Partikelverben mit AUS in der Gegenwartsprache. Dissertation Universita¨t Tu¨bingen. Jackendoff, R. (1990). Semantic Structures. Cambridge, Mass.: MIT Press. Jackendoff, R. (1997). Twistin’ the night away. Language 73, 534–559. Kemenade, A. van (1987). Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Kemenade, A. van (1997). V2 and embedded topicalisation in Old and Middle English. In
Particles and prefixes in Dutch and English
115
A. van Kemenade and N. Vincent (eds.), Parameters of Morphosyntactic Change. Cambridge: CUP, 326–352. Kemenade, A. van (2000). Jespersen’s cycle revisited: Formal properties of grammaticalization. In: S.Pintzuk, G.Tsoulas, and A. Warner (eds.), Diachronic Syntax: Models and Mechanisms. Oxford: Oxford University Press, 51–74. Kluge, F. (1901). Vorgeschichte der altgermanischen Dialekte. In: Paul, H. Grundriss der germanischen Philologie, Band 1. Strassburg: Karl J. Tru¨bner, 320–496. Koopman, W. (1985). The syntax of verb and particle combinations in Old English. In: H. Bennis and F. Beukema (eds.), Linguistics in the Netherlands 1985. Dordrecht: Foris, 91–99. Koopman, H. (1995). De plaats van geincorporeerde hoofden in de werkwoordscluster. Tabu 25, 174–179. Kroch, A. and A. Taylor (2000). Verb-object order in early Middle English. In: S. Pintzuk, G. Tsoulas and A. Warner (eds.), Diachronic syntax: Models and Mechanisms. Oxford: OUP, 132–163. Kyto¨, M. (ed.) (1993). Manual to the Diachronic Part of the Helsinki Corpus of English Texts: Coding Conventions and lists of Source Texts. 2nd edition. Helsinki: University of Helsinki, English Department. Legum, St. E. (1968). The verb-particle constructions in English: basic or derived? In: B.J. Darden, Ch.-J.N. Bailey and A. Davidson (eds.), Papers from the Fourth Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago University Press, 50–62. Lehmann, W. (1906). Das Pra¨fix uz- besonders im Altenglischen. (Kieler Studien zur englischen Philologie. N.F. 3). Kiel: Cordes. Lehmann, W.P. (1986). A Gothic Etymological Dictionary. Leiden: E.J. Brill. Leopold, M. (1977 [1907]). Die Vorsilbe ver- und ihre Geschichte. Germanistische Abhandlungen 27. Reprint Hildesheim: Olms. Lieber, R. and H. Baayen (1993). Verbal prefixes in Dutch: A study in lexical conceptual structure. In: G.E. Booij and J. van Marle (eds.), Yearbook of Morphology 1993. Dordrecht: Kluwer, 51–78. Lipka, L. (1972). Semantic Structure and Word Formation: Verb-particle Constructions in Contemporary English. Mu¨nchen: Wilhelm Fink. Lu¨deling, A. (1999). On particle verbs and similar constructions in German (Sprachtheoretische Grundlagen fu¨r die Computerlinguistik: Arbeitspapiere des Sonderforschungsbereichs 340). Diss. Universita¨t Tu¨bingen. (Published by CSLI, Stanford, 2001). McIntyre, A. (2000). German Double Particles as Preverbs: Morphology and Conceptual Semantics. Tu¨bingen: Stauffenburg. McIntyre, A. (2001). Argument blockages induced by verb particles in English and German: Event modification and secondary predication. In: N. Dehe´ and A. Wanner (eds.), Structural Aspects of Semantically Complex Verbs. Berlin: Peter Lang, 131–164. Meillet, A. (1908). Notes sur quelques faits gotiques. Me´moires de la Socie´te´ de Linguistique de Paris 15, 95–103. Metlen, M. (1933). What a Greek interlinear of the Gothic Bible text can teach us. Journal of English and Germanic Philology 32, 530–548. Neeleman, A. and F. Weerman (1993). Case theory and the diachrony of complex predicates in Dutch. Folia Linguistica Historica 13, 189–217.
116
A. van Kemenade and B. Los
The Oxford English Dictionary on CD-ROM (1989). 2nd edition. Ed. by J.A. Simpson and E.S.C. Weiner. Oxford: Oxford University Press. Pintzuk, S. (1991). Phrase structures in Competition: Variation and Change in Old English Word Order. Dissertation University of Pennsylvania. Pintzuk, S. (1996). Old English verb-complement word order and the change from OV to VO. York Papers in Linguistics 17, 241–264. Quirk, R. and S. Greenbaum (1973). A University Grammar of English. London: Longman. Quirk, R., S. Greenbaum, G. Leech and J. Svartvik (1985). A Comprehensive Grammar of the English Language. London: Longman. Rappaport Hovav, M. and B. Levin (2001). An event structure account of English resultatives. Language 77, 766–797. Sawyer, J.H. (1999). Verb Adverb and Verb Particle Constructions: Their Syntax and Acquisition. (Boston University PhD Dissertation). Michigan: UMI Dissertation Service. Schmidt, J. (1883) Die Germanischen Praepositionen und das Auslautgesetz. Zeitschrift fu¨r Vergleichende Sprachforschung 26, 20–43. Schultze-Berndt, E. (this volume) Preverbs as an open class in Northern Australian languages: Synchronic and diachronic considerations. Spencer, A. and M. Zaretskaya (1998). Verb prefixation in Russian as lexical subordination. Linguistics 36, 1–39. von Stechow, A. (1993). Die Aufgaben der Syntax. In: J. Jacobs, W. Sternefeld, A. von Stechow, and Th. Vennemann (eds.), Syntax: Ein internationales Handbuch zeitgeno¨ssischer Forschung. Berlin/New York: Walter de Gruyter, 1–88. Stiebels, B. (1996). Lexikalische Argumente und Adjunkte: Zum semantischen Beitrag von verbalen Pra¨fixen und Partikeln. Berlin: Akademie Verlag. Streitberg, W. (1920). Gotisches Elementarbuch. Heidelberg: Carl Winter’s Universita¨tsbuchhandlung. Streitberg, W. (1965). Der gotische Bibel. Vol. I and II. Heidelberg: Carl Winter Universita¨tsverlag. Toivonen, I. (2001). The Phrase Structure of Non-Projecting Words. Ph.D. Dissertation, Stanford University. Vinka, M. (1999). On Swedish verb-particle constructions, ms. McGill University, Montre´al. Visser, F. Th. (1963–73). An Historical Syntax of the English Language, Vols. 1–3b. Leiden: E.J. Brill. Watkins, C. (1964). Preliminaries to the reconstruction of Indo-European sentence structure. In: H.G. Lunt (ed.), Proceedings of the 9th International Congress of Linguists. The Hague: Mouton, 1035–1042. Weerman, F. (1989). The V2 Conspiracy: A Synchronic and a Diachronic Analysis of verbal Positions in Germanic Languages. Dordrecht: Foris. Weir, E.M.H. (1986). Footprints of yesterday’s syntax: Diachronic developments of certain verb prefixes in an OSV language. Lingua 68, 291–316. Winkler, S. (1997). Focus and Secondary Predication. Berlin/New York: Mouton de Gruyter. Zeller, J. (2001). Particle Verbs and Local Domains. (Linguistik Aktuell/Linguistics Today 56), Amsterdam/Philadelphia: Benjamins. Zeller, J. (this volume) Moved preverbs in German: displaced or misplaced?
Particles and prefixes in Dutch and English
117
TEXT EDITIONS d’Ardenne, S.T.R.O. (1977). The Katherine Group, edited from MS. Bodley 34. Paris: Socie´te´ d’Edition ‘Les Belles Lettres’. C. Clark (1970). The Peterborough Chronicle 1070–1154. Oxford: Clarendon Press. Holthausen, F. (1967[1921]). Vices and Virtues: A soul’s confessions of its sins with reason’s description of the virtues, 2 vols. (EETS, original series 89, 159). London etc.: Oxford University Press.
Department of English University of Nijmegen Erasmusplein 1 6525 HT Nijmegen The Netherlands [email protected]
This page intentionally left blank
Preverbs, argument linking and verb semantics: Germanic prefixes and particles ANDEW McINTYRE
1. INTRODUCTION This study discusses the semantic and argument-structural behaviour of Germanic preverbs (understood as a cover term for verb particles and prefixes). It is based on detailed research on most German and English particles, and prefixes such as re-, out-, over- and pre-. My goal is to cover the most important phenomena which form the basis for a theory of the argument-structural effects of Germanic preverbs, concentrating on little-known data where possible. Among others, we discuss the following phenomena. (1) illustrates the phenomenon of unselected objects, i.e. objects of complex verbs which do not correspond to the selection restrictions of the simplex verb. The ability of preverbs to license unselected objects is well known, but the interaction between preverbs and objects has various other, less well known manifestations. (2) gives examples where particles allow linking of either the theme or the reference object of the prepositional relation expressed by the particle, even if one or both of the possible objects is not selected by the verb. (3) shows some cases where an obligatorilly transitive verb becomes optionally transitive in the presence of a particle. (4) gives examples where preverbs disallow the objects selected by the verb. (1) work off a debt (cf. *work a debt), vote governments out, think out a plan, dream up a solution, chat someone up, rework an essay, outstay one’s welcome, counteract an order (2) pour out {the bucket/the water} (cf. *pour the bucket); fill in {the form/the information} (3) lock up (the house) vs. lock *(the house); roll/light up (a cigarette) vs. roll/light *(a cigarette) (4) a. read (*articles) on, sing (*a song) along, type away *(at) the essay b. overpurchase (*jewelry), oversmoke (*cigarettes), overeat (*cakes) I keep the study reasonably theory-neutral, in the hope that it will be serviceable as an introduction to an area of grammar which has, I fear, given rise to a number of misconceptions. Our investigation of the semantic and argumentstructural effects of preverbs will proceed as follows. Section 2 gives some basic information about the argument structure of conflation (resultative) structures which will be needed in the rest of the study. Section 3 discusses complex verbs Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 119–144. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
120
Andrew McIntyre
which are a subspecies of the transitive resultative construction. The resultative analysis is shown to be applicable in many cases where intuition at first speaks against such an analysis, notably so-called ‘aspectual’ preverbs. Section 4 discusses scope-bearing prefixes. I argue that these are constrained to have minimal scope, a phenomenon which surprisingly turns out to be the explanation for hitherto unstudied cases like (4b) where, for many English speakers, overprefixation seems to ‘block’ a verb’s direct objects. Section 5 notes some conclusions which can be drawn from the generalisations discussed in the earlier sections, and some problems needing further attention.
2. CONFLATION AND ARGUMENT LINKING To understand the semantic representations in the following sections, it is necessary to discuss the semantics of VP’s which express more than one potentially freestanding event. It is well known that Germanic languages allow constructions like (5), in which a causing event and a result event can be expressed in a VP with a single verb stem and some type of overt result predicate. The causing event is represented by the agent and the verb stem, while the result event is represented by other material in the VP. (5)
a.
Particle verbs (some): Dave scratched a sticker off do(dave, scratch) &CAUSE go(sticker, off [Thing ])
b.
Resultative constructions: Dave ate himself sick do(dave, eat) &CAUSE become(sick(dave))
In the above semantic representations, &CAUSE indicates that the event on the left causes the one on its right. (Other notations might use cause here.) Various authors note that the causal relationship is not the only relationship which can subsist between events. Rappaport Hovav and Levin (2001) and McIntyre (2002b) argue that the subevents in unaccusative resultative structures such as walk into the house and break open are conceptualised as being exactly the same event rather than as two causally related events. These structures are not discussed in this study because the behaviour of preverbs in such contexts is quite straightforward. The ability of a VP with a single verb stem to express two causally related events as a productive option (possible e.g. in Germanic languages but not in e.g. Romance) is variously known as lexical subordination (Levin and Rapoport 1988, Spencer and Zaretskaya 1998a,b), template augmentation (Rappaport Hovav and Levin 1998), lexical adjunction (Wunderlich 1997a), event composition (Pustejovsky 1991) and conflation (Talmy 1985). I will use Talmy’s term here. Most of these authors assume that conflation is the result of a productive
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 121
lexical (semantic) operation over a verb. Others (McIntyre 2002b, Mateu 2001) argue that the (im)possibility of conflation in a given language results from certain (morpho)syntactic properties of the language. The discussion here is independent of this issue. Another important property of &CAUSE structures is that secondary predicates introduced by the go or become conjuncts end up either within the verbal complement domain or (with prefixes and arguably particles) as morphological sisters to the verb. Some linguists are reluctant to assume that all particles appear in the verbal complement position, the rationale being apparently that many such elements (e.g. ‘aspectual’ particles, which are treated as result predicates in section 3) are not subcategorised by the verb. Such reasoning underlies Jackendoff’s 2002 belief that aspectual particles are not arguments and Lu¨deling’s (2001) claim (argued against in McIntyre 2001c) that German aspectual particles with an ‘adverbial’ meaning are adjoined to V∞. Lu¨deling and Jackendoff fail to consider the possibility that the particle, while not selected by the verb’s entry in the permanent lexicon, becomes an argument of the verb once it has undergone conflation. To my knowledge, the following descriptive observations hold of all cases involving conflation with &CAUSE : (6)
If a causing event e1 and a resulting situation e2 are combined in a structure involving a single verb stem with the semantic respresentation [e1 &CAUSE e2], then: a.
e1 contributes the verb stem and its external (usually: agent) argument. No other arguments of a predicate in e1 may be linked unless they are also arguments of a predicate in e2.
b.
e2 may be overtly represented by the remaining material in the VP, i.e. direct objects, resultative AP’s (they trampled the grass flat), directional PP’s (the police wrestled the criminal to the floor), prefixes (offload the goods), particles (pull off the sticker).
These observations are in concert with the widely assumed Direct Object Restriction (DOR) which says that resultative predication can only apply to underlying direct objects (Carrier and Randall 1992, Haider 1997, Hoekstra 1988, Levin and Rappaport Hovav 1995, Li 1999, Simpson 1983, Winkler 1994; critically assessed in Rappaport Hovav and Levin 2001 and McIntyre 2002b). Result predicates and preverbs involved in &CAUSE structures may predicate over a direct object, regardless of whether it happens to be one selected by the verb when no result predicate is present. (7) shows both possibilities. The object in (b), but not in (a), might be seen as an argument of eat. However, in both cases, the object is part of the subevent which introduces the particle. What we
122
Andrew McIntyre
do not find are structures like those in (c) where the verb provides a direct object not licensed by the secondary predicate. (7)
a.
Basil ate the cupboard empty do(basil, eat) &CAUSE become(empty(cupboard))
b.
Basil ate the food up do(basil, eat) &CAUSE become(not(exist(food))
c.
*Basil ate the cupboard the food empty. *Basil ate the food empty.
Of course, (6), while uncontroversial, is purely descriptive. However, offering an explanation for it would take us far away from the topic of preverbs. McIntyre (2002b) and Wunderlich (1997a,b) are two very different theories aiming to account for these generalisations.
3. RESULTATIVE PREVERBS 3.1. Resultative analyses of apparently non-resultative preverbs Perhaps the majority of preverbs are able to be analysed as mapping onto a predicate in a result conjunct introduced by &CAUSE in a conflation structure. They thus have the same function as normal resultative predicates. I wish to present some data to illustrate the widespread applicability of a resultative analysis for preverbs. I concentrate on cases where the resultative analysis might seem counterintuitive at first sight, to show what the resultative analysis does when pushed to its limits. Other studies stressing the resultative character of preverbs are Stiebels (1996, esp. section 7.3) and Spencer and Zaretskaya (1998a,b). In out-prefixed verbs of the type in (8a), outV y means ‘V better/more than y, surpass with respect to V’. As in resultatives, we find direct objects which are obligatory and unselected. I express out’s contribution as a monadic function outdone, cf. (8b). outdone introduces a new subevent into the semantic representation involving an entity which is asserted to be outdone, whether or not this entity happens to correspond to the verb’s normal object selection requirements. (8)
a. b.
outcompete, outdance, outfight, outguess, outrun, outthink, outvote Fred outdrank Stan do(fred, drink) &CAUSE outdone(stan)
The resultative character of these out-verbs becomes clear if we compare (8b) (an unselected object construction: *Fred drank Stan) with the metaphoric resultative Fred drank Stan under the table, which exploits the idea that Fred’s
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 123
drinking causes drinking competitor Stan to become so paralytic that he passes out, falling under the table. The PP under the table has recently undergone semantic bleaching, the result of which is a meaning similar to out-. This allows it to combine with verbs other than drink. Examples I encountered are Page would play today’s guitar heros under the table and I could have mixed that dj under the table (mixing is an activity performed by discjockeys); more combinations (with e.g. talk, smoke, sleep) can be found in an internet search under www.google.com targeting ‘‘anyone under the table’’). Thus, the argument structure and semantics of out-prefixations and under the table resultatives are basically the same, the major difference, irrelevant to our concerns, being that the way in which out- developed its present meaning is not synchronically discernible. Otherwise, the two result predicates differ merely in the fact that one of them happens to insist on morphological unity with the verb. The fact that he is under the table and he is out cannot mean ‘he has been outdone’ does not show that the resultative analysis is wrong. It just shows that the relevant readings of out- and under the table are ‘construction-specific’ in that they have stipulations in their lexical entries to the effect that they are only licensed in a particular construction (in the pretheoretic sense of the term), namely the resultative construction (for under the table) and as morphological nonhead of a complex verb (for out-). McIntyre (2002a) gives more evidence for the need to assume construction-specific senses of lexical items. The result expressed by the particles in (9a,b) is a decremental effect on the direct object (Stiebels (1996: 133–143) treats some similar German preverbs). These can be represented as in (c), or, in a localistic analysis which tries to capture the unity of the senses of the particles, as in (d). (9)
a. b. c. d. e.
I slept/danced/read the afternoon away I slept off a hangover, I worked off a debt, I exercised off excess weight sleep off/away do(x, sleep) &CAUSE become(nonexist(y)) sleep off/away do(x, sleep) &CAUSE go(y, from deictic_centre)) [ VP v np away] ‘waste time V-ing’ (Jackendoff 1997: 555)
With the ‘time’-away construction in (9a), the causative analysis claims that a similar conceptualisation is at work to that seen in in order to pass the time, I slept/drank. It also claims that the passage of time can be conceptualised in the same way as the using up of some substance, which is independently attested by parallels like I have no {time/food} left and I ran out of {time/food}. Jackendoff’s (1997) study of the ‘time’-away construction does not describe away as part of a predication over the time phrase appearing as object. He stipulates that away is part of an idiomatic lexical item, as in (9e), unimpressed by the fact that the argument selection features of the construction are the same as those seen in resultatives. He argues against a resultative analysis like mine (p. 549ff). For
124
Andrew McIntyre
instance, he claims that the object in normal resultatives is a patient, his diagnostic of which is the do to test, which is inapplicable with away: *What I did to Monday was sleep it away. But this is not a good diagnostic for resultative semantics: *what I did to myself was laugh myself silly, *what I did to the remark was edit it into the text, *what he did to air was breathe it into the bag, *what he did to the drug reference was read it into the Beatles song. Jackendoff also claims that the ‘time’-away construction requires a volitional subject while uncontroversially resultative uses of away do not: the wind blew {the papers/*the night} away. I see this fact as an argument for my analysis. The resultative analysis assumes that we can conceptualise time as being a resource which can be depleted because its possessor engages in some activity, a conceptualisation independently attested in watching TV used up/wasted up my time; I had no time yesterday because I was working. But inanimate entities cannot be conceptualised as having time at their disposal (cf. how does grandma spend her time? vs. *how does the wind spend its time?). It is thus logically impossible for inanimates to deplete their time by performing some activity. Jackendoff is content with stipulating the volitionality requirement, while a resultative analysis can explain it. One also finds numerous obscure instances of resultative particles which combine with a limited set of verbs. For instance, the source argument expressed by off can be an abstract possessive domain (sell/auction/hive/flog off ). Another example (cf. Lindner 1983: 80–87, 125–138) is furnished by some verbs with up and out where the result could be called ‘cognitive availability’ (search/seek/ point/pick someone out, find/work/tease out the answer, look up an address, dream/think up an idea, summon/call up a memory, play up an issue, bring out/up the main issue). This class is, like many preverb uses, idiosyncratically restricted. However, the basic resultative structure is still detectable, the hallmarks of resultativity being unselected objects and an accomplishment structure (seek somone {for/*in} 10 mins vs. seek someone out {in/*for} 10 minutes). In the verbs in (10), German ein (basic meaning: ‘in’) indicates that the direct object (sometimes a reflexive coindexed with the subject) enters into a state of readiness for the activity expressed by the verb. The object is often not selected by the simplex. Traditional sources saw ein as an inchoative marker, analysing cases like sich einspielen as ‘begin playing’. This analysis makes no sense of the fact that the argument structure of the particle verbs exactly parallels that seen in normal resultative constructions (compare sich einspielen ‘refl. in-play’ with sich warm spielen ‘refl. warm play’, both of which mean ‘get warmed up (in sport)’). For more on these ein-verbs, see Lindemann (1998: 125–132). (10)
a.
sich einspielen ‘‘refl. in-play’’, ‘get warmed up (in sport/music)’, sich einarbeiten ‘‘refl. in-work’’, ‘work one’s way into something’, sich einsingen ‘get warmed up in singing’
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 125
b.
Schuhe einlaufen ‘‘shoes in-walk’’, ‘wear shoes in’, ein Pferd einreiten ‘‘a horse in-ride’’, ‘break in a horse’
The rest of this section deals with items often called ‘perfective’, ‘completive’ or ‘telicising’ preverbs. I wish to show that, at least in Germanic, such preverbs are not a category unto themselves, but are simply result predicates, differing from the ones seen above at most in the degree of specificity of their semantic contribution. Like resultatives, ‘perfective’ preverbs mostly co-occur with direct arguments, sometimes with unselected objects. One may distinguish two main subclasses: Type 1: the preverb does not add a new result event to the simplex, so that a transitive verb may retain its selection restrictions. A good example is eat up. The preverb may appear redundant at first glance, since the simplex is already telic (provided the theme is quantised1 ). Both Gwen ate the cakes and Gwen ate up the cakes assert the entire consumption of the cakes. I propose that both sentences have the same semantic structure, viz. (11). Eat and eat up are not identical, however. Compare intranstive uses like I ate up (asserts the entire consumption of a specific portion of food) vs. I ate (no implication that none was left over). This is explicable under the assumption that intransitive eat lacks the become conjunct, while intransitive eat up has a become conjunct in which the theme has been suppressed. The latter suppression is presumably a PF deletion of a pronominal (‘pronoun zapping’, Jacobs 1993), found in cases like Don’t touch thati it’si hot! (11) Gwen ate the cakes up/ Gwen ate the cakes [ Event do(gwen, eat)] &CAUSE [ Event become(nonexist(cakes))] Type 2: the preverb contributes a result predication not present in the simplex verb. One subclass introduces unselected objects (cf. (12)), while the other class does not affect the transitivity of the simplex (cf. (13)). The latter class imposes an incremental theme reading upon the object which is not necessarily present in the simplex (I used the resources {for/in} 3 days vs. I used up the resources {in/*for} 3 days. (12)
Fred chatted Mary up (‘he got Mary in a desired state by chatting to her’) [ Event do(fred, chat)] &CAUSE [ Event become(ready(mary))]
(13)
Cecil used up the resources [ Event do(cecil,) [by_means_of (resources)]] & [ Event become(nonexist(resources))]
The argument linking properties of ‘completive’ complex verbs are the same as
126
Andrew McIntyre
those of resultative constructions. In both cases, intransitive verbs can become transitive (chat people up, talk people senseless). This happens because the preverbs or result APs/PPs are licensed by the addition of a new subevent via conflation. This extra predication always contains an entity appearing as direct object. ‘Completive’ preverbs deserve no special theoretical status. Rather, they express information about the culmination of an event by indicating the way in which the event affects an entity. This explains the overwhelming tendency of completive preverbs to transitivity, which tendency is also evident in resultatives. There do not seem to be completive preverbs that telicise an activity expression by making it mean ‘finish Ving’. Rather, they instantiate a schema [become(<state>(<entity>))]. Some potential objections to this analysis stem from the problem that, taken in isolation, complex verbs often look more idiosyncratic than they are. When a preverb does not clearly have the same semantics as a related nonpreverb (e.g. a spatial preposition), an inadequate data base may tempt one to deny that the preverb has any meaning beyond some vague ‘completive’ or ‘telic’ effect. But study of a representative set of verbs formed with the preverb in question usually reveals clear patterns in the particle’s semantic contribution (as is argued in McIntyre 2002a). Consider up, a classic ‘completive’ particle (e.g. Denison 1985). Note for instance that the uses of up in (11)–(13) occur in various other complex verbs: (14)
a. b.
Decremental up: drink up, dry up, mop up, take up (time/space), burn up, smoke up (cigarettes) up signifying that the direct argument reaches a state considered sufficient for a particular purpose: paint/heat it up, fix it up (=till it is usable), frighten up (actors for a horror scene), starve up (people in order to make them compliant), roll up (a joint), chat someone up
See Lindner (1983: 3.4, esp. 151f, 164–67) on these two particle uses and on a larger network of interrelated senses. If the distinctions between these subsenses are psychologically real, then up clearly has a more specific function than just ‘perfectivising’ or ‘telicising’ the verb. It telicises a verb in the same way as any result predicate does, namely by expressing a result state achieved through the simplex activity. (15) illustrates this parallel. (15)
a.
the dog chewed the paper up [ Event do(dog, chew)] &CAUSE [ Event become(nonexist(paper))]
b.
the dog chewed the paper soggy [ Event do(dog, chew)] &CAUSE [ Event become(soggy(paper))]
One could perhaps describe the unity of completive up uses by underspecifying
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 127
the predicate in the scope of become in the lexical entry, allowing an inference based on the verb meaning to supply the appropriate predicate. This may or may not be the right course. Giving up this general a meaning would generate many non-existent combinations without further mechanisms, but it could be argued that up is semiproductive and/or that its productivity is confined to certain tokens and classes of simplex verbs (McIntyre 2002a). If we do give completive up an underspecified meaning like [become(<state>(x))], where the state is contextually inferred, the representation would still be structurally parallel to that of a resultative. Moving on from up, consider some uses of German ver- and er-, seen in traditional treatments as instances of ‘completive’ or ‘intensifying’ prefixes. Partly drawing on Stiebels (1996), we can see that these are also resultative predicates. ver- for instance has a productive decremental reading, parallel to up in (15a) and away in (9a): vertelefonieren/-feiern ‘use up (money, time) by telephoning/celebrating’, verbrauchen ‘use up’2. In a second use (Stiebels 1996: 151–155), ver- specifies a result state describable as together (verlo¨ten ‘solder together’, vermixen ‘mix together’, verkoppeln ‘couple’). Some vague category ‘telicity marker’ does no more justice to the meaning of ver- than it would to that of the English gloss together. One meaning of er- is ‘to death/dead’ (erha¨ngen/-schlagen/-dru¨cken ‘kill by hanging/hitting/pressing’). Simply calling er- a perfective marker cannot explain the contrast between ermorden ‘murder’ (obligatorily transitive) and the simplex verb morden ‘commit murder’ (intransitive with an implicit generic object, unless extreme inhumanity is connoted). It makes no sense to intensify or perfectivise morden. The resultative analysis can capture ermorden by assuming that er- is the overt manifestation of a semantically redundant result predication which is added to circumvent the problem that the simplex verb normally disallows the linking of a patient3. The idea that preverbs primarily contribute telicity is probably connected to the idea that Germanic prefixes and particles always yield telic VP’s. Advocates of this are e.g. Dehe´ (2000), Tenny (1994, section 2.2.2), Keyser and Roeper (1992: 113); the latter authors write ‘‘Like all particles, up appears to require a delimited interpretation.’’) This myth is refuted by examples like the following, where the in-time test is passed without any event iteration effects: (16) Grandma held down the wrestler for an hour; They kept out the enemy for a day; She slagged off her husband for 10 minutes; I scrubbed down the table for 10 minutes; They beat him up for five minutes; He stared down his opponent for a minute; Fran chatted up Stan for ten minutes Summarising the main points from this section, I emphasise that there are many more resultative preverbs than a superficial examination would suggest. So-called ‘completive’ preverbs often turn out to be subinstances of the resultative construction, sometimes with underspecified result states. These result
128
Andrew McIntyre
states may or may not be overt spellouts of a result component already present in a transitive verb’s meaning.
3.2. Theme and reference object in prepositional preverbs The above description of transitive complex verbs needs to be augmented by some observations on the types of arguments contributed by prepositional preverbs. We note that a prepositional element often expresses a relation between two entities, a theme (also called a trajector, figure or locatum) and a reference object (landmark, ground), where the latter is used in specifying the location of the former, cf. (17a). Normally, particle verbs leave their reference object implicit: interpreting (17b) involves supplying a reference object with contextual help (McIntyre 2001b: 275–279 gives details and references). Here the overall semantic structure is resultative: the verbal actitivity leads to a change of position of the direct object, a theme. However, (17c) gives a third possibility: the reference object of the particle is linked at the expense of the theme4, a phenomenon which most researchers fail to note. Table 1 gives a list of the possibilities of (non-)linking of themes and reference objects in complex verbs. (17)
a. b. c.
I wiped [the dust]Theme off [the table]Reference Object I wiped [the dust]Theme off I wiped [the table]Reference object off
The main distinction is between types A (reference object not linked) and B (reference object linked). The research has largely ignored Type B, but the examples given in the table are a small subset of those I have found. Wunderlich (1983) noted that German has a regular (but, I add, not exceptionless) pattern whereby prefix verbs fit into type B and particle verbs into type A; (18a,b) gives a typical minimal pair. (18)
a.
er durchfuhr den Park he through-drove the park
b.
er fuhr durch he drove through
c.
er fuhr durch den Park he drove through the park
McIntyre (2001b), Zeller (2001b) and Svenonius (1996) seem to be the only modern studies on particle verbs which have noted type B. Svenonius sees type B particles as unaccusative elements. This makes sense if one assumes a syntactic derivation of particle verbs where the direct argument originates
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 129
Type
Subject
Object
Examples
A1
Agent
Theme
I pumped the water out, I offloaded the books
A2
Theme
B1
Agent
Reference Object
The doctor pumped his stomach out. She ran him through (with a sword). I filled the hole in (with cement) I overlaid/inlaid the table with gold I underscored the word I struck the word through
B2
Theme
Reference Object
The river overflowed its banks I overstepped the line
B3
Reference Object
I ran in, I walked off
The pot overflowed, boiled over. (pot metonymically represents its upper rim) The pen ran out (of ink) (applicable if conceptually like ‘the ink ran out of the pen’) Der Tank sickerte/lief aus ‘‘the tank leaked/ran out’’, ‘the tank leaked/emptied’ (auxilliary: ‘be’)
Table 1: Themes and reference objects in complex verbs
within a small clause or some other projection which includes the particle and not the verb. Other writers miss the basic semantic properties of the prepositional relations. For instance, den Dikken (1995: 54f) assumes that all particles are unaccusative (ergative) prepositions. This would mean that the theme in (17b) originates in the complement position of the prepositional element, although in (17a) this position is reserved for the reference object. This does not sit easily with den Dikken’s adherence to the Uniformity of Theta Assignment Hypothesis (p. 19ff). Various writers analysing the semantics of complex verbs have failed to take the existence of Type B structures into account. Stiebels’ (1996) otherwise authorative account of complex verb semantics does not discuss the possibility that the formalisation of the German pattern equivalent to (17c) should be related to that of (17b). She (p. 159) analyses (17c) using a result predication become(clear(y)). Note that a number of particle uses show a similar ‘alternation’ where either the theme or the reference object is linked, cf. (19). McIntyre (2001b: 275–279) gives many German examples; the pattern is less well attested in English.
130
Andrew McIntyre
(19) pour out {the bucket/the water}; fill in {the form/the information}; strip off {a person/their clothes} Cases where the reference object is linked often show a ‘holistic’ effect. Thus, pour out the bucket means not just that something gets poured out of the bucket, but that the bucket is emptied completely. Anderson (1971) noted various cases where the realisation of an entity as direct object gives it a holistic reading. I am not aware of a satisfactory explanation for this effect, but refer to Tenny (1994: 49ff) and Brinkman (1997) and their references for some proposals. Type B preverbs might again tempt one to posit a ‘completive’ or ‘telic’ meaning for the particle, but the data known to me suggest that this is epiphenomenal of the holistic effect. The use of through in examples like (20) (called an ‘aspectual particle’ by Jackendoff 2002) can be given the semantic analysis in (20). The simplex verbs are ‘path accomplishments’ (Wechsler 1989) or ‘performative verbs’ (Jackendoff 1996: 332f), verbs whose object does not undergo change, but acts like an incremental theme because the event progressively relates to different parts of it. This can be conceptualised as a progression through the object. The particle expresses this overtly. That it suggests a completive reading can be derived from the holistic effect. Note also that we find transitivisations (think the issue through/*think the issue), where the verb meaning is enriched by a new subevent housing the particle and object. (20)
a.
think the matter through, work the problem through, sing the song through
b.
Vladimir played the sonata through, [ Event do(vladimir, play)]i & [ Event go([event]i ,(through(sonata))]
A complete account of Type B complex verbs, including the ‘aspectual’ through verbs, would involve discussion e.g. of the factors licensing the holistic effect and the suppression of the theme argument in cases like (17c). This must be reserved for separate article.
4. SCOPE-BEARING PREVERBS The last section dealt with preverbs whose argument-structural effects are easy to explain since they reduce to patterns observable with resultative predicates. The present section takes up more challenging cases of scope-bearing, nonresultative preverbs, which, perhaps significantly, are all prefixes. These sometimes have the effect of blocking direct objects selected by the verb. We will find that the data bespeak the correctness of the following (at this point purely descriptive) generalisation:
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 131
(21) Minimal Scope Requirement: Scope-bearing preverbs always have the narrowest possible scope which does not yield semantic deviance. This claim is similar to claims made in Stiebels (1996: ch. 8). Let us look at some preliminary evidence for it. The narrow scope of re- is well documented (e.g. Wechsler 1989, Dowty 1979: 256f). A verb like reseal means ‘cause to become again sealed’ rather than ‘seal again’, cf. (22)5. Secondly, consider the type of prefix exemplified in (23), which pretheoretically has a reversative meaning. This phenomenon, coupled with the non-occurrence of wide-scope negative prefixes, suggests that reversative prefixes are simply negative operators which are forced to assume narrow scope by the Minimal Scope Requirement, cf. the representations in (23). Note that treating dis and un as negative operators has the advantage that we can capture the negative meanings they exhibit when combined with non-verbs (unclear, disinterest). This assumes that the etymological non-relatedness of verbal and adjectival un- is synchronically irrelevant. On un-, see Dowty (1979: 256–260); on the narrow scope of its German equivalent in complex adjectives, see Lenz (1995). (22)
reseal
do(x) &CAUSE become(again(sealed(y)))
(23)
a. b. c.
do(x) &CAUSE become(not(stable(y))) do(x) &CAUSE become(not(tangled(y))) do(x) &CAUSE become(not(locked(y)))
destabilise disentangle unlock
4.1. The scalar reading of over- and (in)transitivity Risch (1995) used the term ‘scalar’ for the German equivalents of over/under and out which include elements related in meaning to those in comparative structures. The relevant uses are seen in I over-/undercooked the food ‘cooked it to a degree greater/less than is good’. Scalar uses of prefixes are to be distinguished from the uses like overlay x with y, overstep the line, overrun a country, where direct objects are the conceptual reference object of the prefixal preposition (sometimes with a metaphoric interpretation). We look at over- in the scalar sense, where it integrates a modifier into the semantic representation for an event. Here are sample representations for verbs showing two main instantiations of scalar over-, a degree interpretation and a frequency interpretation: (24)
Mavis overtightened the bolt (=made it too tight/overtight) [ Event do(mavis)] &CAUSE [ Event become(over(tight(bolt)))]
(25)
Dave overwashes his clothes (=washes them so frequently or intensively that they are damaged) [ Event do(dave, wash)] &CAUSE [ Event over(affect(event, clothes))]6
The contribution of over is left undecomposed because further decomposition
132
Andrew McIntyre
will presumably not affect the remarks below about argument linking and the prefix’s scope. Risch (1995) gives a decomposition of the German equivalent u¨ber- embedded in a larger theory of gradation, but is less concerned with explaining argument-structural matters. A complete discussion of over would take a whole article. The remarks below concentrate on the effects of the Minimal Scope Requirement. Basically, we see that over seems always to take scope over the predication containing the direct object if there is one, and that cases where this narrrow scope is semantically deviant force the neutralisation of the subevent containing the object. 4.1.1. When over- and objects are mutually exclusive In the sentences in (26), over- is incompatible with direct objects selected by the verb in some English varieties. (26)
a. b. c.
Max overate (*cakes). You overbuy/overpurchase (*jewelry). Does your theory overgenerate (*unacceptable sentences like this one)? d. They overbuild (*houses)7. e. Mervyn oversmokes (*cigarettes). f. You overdrank (*beer).
I was initially unaware that the judgements above are variety-specific. Farrell Ackerman informed me that he accepts some of the objects in (26), and T. Deacon in a talk in Leipzig (2.3.2001) utterred the sentence the brain overproduces connections. However, my judgements in (26) have been confirmed with several other speakers. I checked this by embedding such sentences in short texts8, and having native speakers read them under the pretext that I needed their help in checking the English of a non-native speaker. Several native speakers (if not all) adjudged VP’s like overeat chocolate as unacceptable, without any leading questions on my part. Both the ‘inflexible’ and the ‘liberal’ varieties deserve study as manifestations of Universal Grammar. I offer an explanation for the differences below. In the meantime, all judgements refer to an inflexible variety like mine. In (27), I give sample representations for the transitive uses of the simplex verbs corresponding to the prefix verbs in (26). Other representations are imaginable, but what seems uncontroversial is that the simplex verbs all express a non-gradable result state. A cake can be either eaten or not eaten, there are no degrees of eatenness. Certainly, a cake can be half-eaten and a house partly built, but here the portion of the theme to which the property applies is relativised, not the property itself.
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 133
(27)
a.
Max eats cakes [ Event do(max, eat)] &CAUSE [ Event become(nonexist(cakes))] (parallel: drink, smoke)
b.
You buy/purchase jewelry [ Event do(you, buy)] &CAUSE [ Event become(possess(you, jewelry))]
c.
They build houses [ Event do(they, build)] &CAUSE [ Event become(exist(houses))] (parallel: generate)
Contrast these verbs with a verb like tighten (cf. (24)). Tighten is clearly a change-of-state verb, but the gradability of its result state allows for atelic readings not permitted by verbs with verbs with non-gradable result states (I tightened the bolt for a minute vs. *I ate the cake for a minute). I now begin my explanation for why over appears to ‘block’ objects. I will argue that integration of the prefix into the subevent licensing the object, demanded by the Minimal Scope Requirement, yields semantic deviance. Consequently, the result conjunct must be eliminated from the representation, yielding an intransitive structure, as in (28). (28) Max overeats [ Event over(do(max, eat))] The representations below for *Max overeats cakes give the logically possible ways of integrating the preverb into a transitive structure. (The representational format will be revised later in this section.) (29)
a. b. c.
[ Event do(max, eat)] &CAUSE [ Event become(nonexist(over(cakes)))] [ Event do(max, eat)] &CAUSE [ Event become(over(nonexist(cakes)))] [ Event do(max, eat)] &CAUSE [ Event over(become(nonexist(cakes)))]
In the representation in (29a), over has scope only over the object. This would be logically possible if it were a quantifier over entities. The structure would then mean ‘Max eats too many cakes’. But in the inflexible English varieties, over is a modifier of events, not a quantifier of entities. Liberal dialects seem to be able to use over as a quantifier over entities. This is confirmed by the intuition of a liberal informant I found, who indicated that the sentences I starred above are interpreted such that overV x’s means ‘‘V too many x’s’’. The informant disallows sentences with explicitly quantified NP’s like *she overdrank seven glasses of wine or *they overbuilt five supermarkets. The idea that verb prefixes can be quantifiers of entities, and particularly the provisional representation of this in (30a), require more attention, but this must be left for future work. I note merely that the basic idea is not without precedent. Spencer and Zaretskaya
134
Andrew McIntyre
(1998b) discuss critically some analyses of Slavic prefixes which see prefixes as quantifiers over entities. (29b) is impossible because a non-gradable result predicate cannot be modified by an element like over-, which specifies a degree. In (29c), over has scope over the whole change of state. As far as I know, scope over become in the verbs of the eat-class would make semantic-pragmatic sense only under the following conditions: (30)
a.
A frequency reading (=‘too often’) could be assigned to over-.
b.
The object could be interpreted such that the repetitions of events entailed by the frequency reading of over could in successive events apply to different tokens of the type referred to by the object NP.
Condition (b) is not met. I will show below that this can be traced back to another recognised problem. That condition (b) plays a role can be seen in the fact that, to a speaker of the inflexible variety, he overeats cakes sounds bad precisely because it forces the world-knowledge-wise deviant reading where the same cakes are consumed more than once. This reading is the only one available in inflexible dialects because the other readings permitted by the Minimal Scope Requirement, namely (29a) and (b), are semantically uninterpretable, whereas (29c) is, although pragmatically deviant, at least interpretable. To begin explaining why condition (30b) is not met, or, for short, why the object fails to receive what I call a token-differentiated (=sloppy) reading when under the scope of the prefix, it is important to note that the problems attending the frequency reading of over- are strongly reminiscent of another prefix capable of expressing repetition, re-. *re-eat and *rekill are impossible because something prevents their objects from being interpreted in a token-differentiated way, forcing a pragmatically deviant reading in which the same entity is eaten/ killed twice. Thus, *he rekilled people (unlike he killed people again) forbids a reading in which people refers to different individuals in successive killings. In (31), we see that both re- and over- are incompatible with token differentiation while semantically parallel adverbials allow it, provided that the object’s determiner does not intrinsically force reference to specific tokens (cp. (31b)). This suggests that preverbs, unlike adverbials, are insensitive to the semantics of determiners. (31)
a. b.
*I {reate/overate} {cakes/chocolate/the cakes/that piece of chocolate}. I ate {cakes/chocolate/*the cakes/*that piece of chocolate} {again/too often}.
My explanation for why token differentiation is impossible with the prefixes will draw on an idea first presented in Carlson and Roeper (1980: 143f; henceforth: C&R), who present the data and decomposition in (32). They note that
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 135
again is able to have scope over the existential quantifier, while re- cannot. They hold that prefixation occurs in the lexicon, and cannot hold scope over material added in syntax. It can have scope over the direct object variable, but not the NP that binds the variable. Wunderlich (2001) independently uses the same assumptions in his discussion of the scope of re- and the interpretation of direct objects. (32)
a. b.
Bob resigned a card: (Zx) (card(x) & again’ (Bob sign x)) [the same card as before] Bob signed a card again: again’ ((Zx) (card(x) & Bob sign x)) [possibly a different card]
Note that the assumption that re-verbs are morphological objects, even if correct, is not a necessary condition for the scope and non-token-differentiation effects seen with re-, since the German adverb wieder ‘again’, when positioned after an object as in (33), shows the same effects. (33a) and (b) are bad because one cannot kill or bear the same child twice (i.e. *rebear/rekill a child)9. Wieder does not combine the verb morphologically in (33), since full PP’s intervene between wieder and the verb, even in the base word order. (33)
a.
*da sie ein Kind wieder zur Welt gebracht hat since she a child again to.the world brought has (since she had a child (=the same child) again)
b.
*da sie ein Kind wieder ums Leben gebracht hat since she a child again around.the life brought has (since she killed a child (=the same child) again)
c.
da sie ein Kind wieder nach Hause gebracht hat since she a child again to home brought has (since she took a child (=the same child) back home)
Thus, affixes are not the only items that force non-token-differentiated readings of verbal objects. The correct empirical generalisation covering re-, over- and wieder in (33) seems to be as follows: (34) Items whose scope domain consists of (part of) a result state of an event do not allow a token-differentiated reading of the object. (The Minimal Scope Requirement predicts that prefixes will be among such narrow-scope items.) To explain why (34) should hold, I will enlist the basic assumptions made by C&R and Wunderlich (2001) about the relative scope of existential quantifiers and prefixes. I will make this more explicit using the
136
Andrew McIntyre
following examples, enriching the semantic representations assumed above with quantifiers and restrictors: (35)
(36)
a.
*Soldiers rekilled prisoners Zx: soldiers(x), Zy: prisoners(y) & [do(x) &CAUSE BECOME(AGAIN(DEAD(Y))]
b.
Again, soldiers killed prisoners again [Zx: soldiers(x), Zy: prisoners(y) & [do(x) &CAUSE become(dead(y))]]
c.
Soldiers reopened windows Zx: soldiers(x), Zy: windows(y) & [do(x) &CAUSE BECOME(AGAIN(OPEN(y))]
a.
*Soldiers overate cakes Zx: soldiers(x), Zy: cakes(y) & [do(x, eat) &CAUSE OVER(BECOME(NONEXIST(Y)))]
b.
Soldiers ate cakes too often over [Zx: soldiers(x), Zy: cakes(y) & [do(x, eat) &CAUSE become(nonexist(y))]]
The bold-typed segments of (35a,c) and (36a) do not contain a full representation of the object’s semantics but only a variable. The interpretation of these segments is already fixed before the variables are bound by anything which could allow token differentiation. Thus, these segments are interpreted such that some entity, y, respectively becomes dead again or too often stops existing. This interpretation is pragmatically deviant in (35a) and (36a), but not in (35c). I should comment on the assumptions about quantification and variable binding. The above representations follow C&R and Wunderlich (2001) in assuming that existential plurals can in principle allow a token-differentiated interpretation, cf. (35b) and (36b), but binding by some kind of generic operator would also be sufficient. Of course, I make crucial use of the assumption – taken for granted by many linguists – that we need a basic representation of an event where the participants are represented as yet-to-be-bound variables (where ‘event’ may correspond to VP, Diesing 1992). An analysis which dispenses with this assumption is Wechsler (1989), who would with reference to (35c) talk in terms of re- having scope over windows but not soldiers, the logic being that the same windows, but not necessarily the same soldiers, were involved in both the original and the restored situation. However, this is problematic: in (37), it is uncontroversial that again has wider scope than re-, since the causing event must be repeated. Wechsler would therefore have to say that both the subject
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 137
and the object are included in the scope of again. However, token-differentiated readings of both NP’s are possible in this case, contrary to what Wechsler would predict. (37) Again, soldiers opened windows [may be different windows and soldiers from those in prior opening event] 4.1.2. Variable transitivity and the scope of overWe now examine cases where the scope of over differs according to whether the verb is used transitively or not. These offer different evidence for the Minimal Scope Requirement. (38) and (39) give representations for two uses of overheat. (In the representations, heat* stands for the unspecified activity causing something to be heated.) The Minimal Scope Requirement predicts that the transitive use of overheat can only mean that the theme entity becomes too hot. It cannot mean that it becomes hot too often. In the detransitivised use of overheat in (39), there is no implication that anything becomes too hot. Instead, it could be used for instance in stating that the heating costs too much. As suggested by the Minimal Scope Requirement, this meaning is impossible when overheat is used transitively. (38)
Max overheats the furnace [ Event do(max, heat*)] &CAUSE [ Event become(over(hot(furnace))])]
(39)
Max overheats (in sense ‘he heats his room too much’, not the unaccusative sense ‘he becomes too hot’) [ Event over(do(max, heat*))]
We next look at overcook. (40) offers two representations of transitive cook, which could be seen either as a change of state verb or an affect verb of the type mentioned in footnote 6. I prefer the latter option, but either representation will, coupled with the Minimal Scope Requirement, predict the intuition that transitive overcook means that the object is overly strongly affected by the cooking process, the only possible interpretation of transitive overcook. Notice that if we intransitivise it, as in (41), overcook can have a very different reading in which the subject does too much cooking. One entailment of this is that the subject cooks too much food, or cooks for too many people. Fred overcooked food cannot receive this reading, since the Minimal Scope Requirement only allows the readings in (40), where the prefix is part of the result event. (40)
Fred overcooked food a. [ Event do(fred, cook)] &CAUSE [ Event become(over(cooked(food)))] b. [ Event do(fred, cook)] &CAUSE [ Event (over(affect(event, food))]
138
(41)
Andrew McIntyre
Fred overcooked [ Event over(do(fred, cook))]
4.2. Conclusions The discussion of scope-bearing prefixes has illustrated the workings of the Minimal Scope Requirement. In particular, I have shown how the Minimal Scope Requirement gives rise to a set of circumstances which create the impression that the prefix over- ‘blocks’ direct objects with verbs with non-gradable result states. This arises in contexts where the prefix cannot be part of the same subevent as the object without yielding deviant interpretations. It seems appropriate to mention a couple of phenomena superficially akin to those found with over- to which my analysis should not be extended. Firstly, the phenomena found with English over- do not correspond to those found with its German equivalent u¨ber-. I merely comment on one difference, namely the fact that German attests cases like (42a), where the prefixed verbs have an unselected reflexive object, where idiomatic English glosses mostly use an intransitive frame (impossible in German). Risch (1995) contents herself with stipulating that the verbs in (42a) have a semantically vacuous argument position. A possibility for a more principled analysis emerges if we entertain the idea that the reflexive is an instance of the unselected reflexive found in resultative constructions like (42b). An implementation of this would be to assume that the prefix contributes information something like (c). On this analysis, the prefix provides a result state which specifies only that some pragmatically inferred property predicated over an entity (appearing as the reflexive coindexed with the subject) is asserted to hold to too great a degree. I must leave this analysis as a speculation at this point, however. (42)
a.
b. c.
sie haben sich u¨berarbeitet/ u¨bergessen/ u¨berkauft/ u¨berhoben they have themselves overworked/ overeaten/ overbought/ overlifted ‘they overworked/overate/overpurchased; they hurt themselves lifting something’ sie haben sich zu Tode gearbeitet ‘they worked themselves to death’ Scalar u¨ber-: become(over(<property>(y)))
It is also worth noting that there is a series of verb particles such as (43) which, like over-, appear to ‘block’ the direct object of the verb. I believe that the factors causing such blockages are entirely different from those affecting over-. I will not discuss these particles here, as I have already done so in McIntyre (2001a, 2002b). (43) read (*notes) on, sing (*songs) along, play (*a silly game) around, type away *(at) the essay
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 139
The main question which remains is that of what the Minimal Scope Requirement should follow from. One answer would be that, as suggested by Carlson and Roeper (1980) or Wunderlich (2001), the prefix’s combining with the verb in the morphological component greatly reduces the range of material over which it may take scope. However, this does not guarantee minimal scope, for it does not rule out prefixes which have scope over a causing event in a resultative structure, and these do not appear to exist. Unfortunately, I have no adequate solutions to this problem and leave the Minimal Scope Requirement as a descriptive generalisation which can hopefully be motivated in the future.
5. SOME CONCLUSIONS AND RESIDUAL MATTERS I now mention some conclusions and consequences flowing from the discussion of the semantic and argument-structural facets of preverbs reported on above. Firstly, cases where preverbs have the effect of introducing unselected objects (chat someone up, vote someone out, stare someone down, think an issue through) result when conflation introduces an extra subevent to the verb’s meaning, an effect also seen with standard resultative constructions (talk people senseless, stare someone into submission). Transitivisation is not a sign that the constructions are non-compositional. Transitivisation is certainly no proof for the morphological status of particle verbs, unless we want to argue that resultative constructions with full PP’s and AP’s are morphological objects, which has to my knowledge not been attempted. Furthermore, I query whether transitivisation by prefixes necessarily supports Zeller’s (2001a) thesis that German prefixes head a functional projection responsible for licensing direct objects. Even if we agree that such a projection is needed, the resultative nature of the preverbs’ semantics means that they will always license a direct argument anyway, whether or not they originate in a syntactic head responsible for transitivity. Unselected objects have been a primary motivation for the assumption that preverbs are small clause (SC) predicates and for other theories assuming that internal arguments of complex verbs originate in a projection of the preverb (e.g. Hoekstra 1988, 1992, den Dikken 1995, Stechow 1995, Svenonius 1996, Zeller 2001b). The underlying configuration can be represented in a schematic fashion as in (44). (44) work the debt off [ VP work [ SC/PP/Particle P the debt off]] The SC approach is reasonable for resultative preverbs, but it makes no sense to claim that scalar over and the other scope-bearing preverbs discussed in section 4 take a direct object as argument. At least in these cases, we are forced to adopt some variant of the other major type of analysis for complex verbs, in
140
Andrew McIntyre
which the preverb and verb form a constituent to the exclusion of the object, whether this consituent is a syntactic projection (Zeller 2001c, Haider 1997, Lu¨deling 2001, Mu¨ller 2000, Booij 1990, Keyser and Roeper 1992) or a morphological object (Dehe´ 2002, Neeleman and Weerman 1993, Olsen 2000, Stiebels and Wunderlich 1994). Another observation is that direct objects are not characterisable in terms of any type of thematic role or relationship to some predicate in a semantic representation. For instance, any kind of theory which links direct objecthood to affectedness or change of state (e.g. Pinker 1989, Levin and Rappaport Hovav 1995) runs into problems with some complex verbs which allow linking of the reference object, especially cases where the reference object indicates a distance travelled by a theme (e.g. (18a) and cases like circumnavigate the world, overstep the line. It remains to draw attention to some problems not touched on here. Firstly, I must note that there are certain transitivisation effects which do not comfortably fit into the picture presented above. (These mostly involve preverbs which have semantic effects parallel to temporal adverbials.) I cannot yet say why rehas a transitivising effect on the simplex verbs in (45a). It is not obvious that the prefixation involves the addition of a new subevent to the verb’s meaning, nor is it clear whether these three examples, even if unique, should be written off as exceptional. Nor is it clear how to capture the transitivising effect seen in (45b–d). The objects in (b) and (c) can be seen as reference objects of bound prepositional preverbs (cf. paraphrases with before humans, before ours, with Schumann), but the mechanisms involved in deriving the argument structure cannot be derived from what has been said in the rest of this essay. The German particles vor ‘beforehand’ and nach ‘afterwards’ transitivise verbs in some rare cases, cf. (45d), involving intransitive arbeiten. There is no sense in which the unselected object is a conceptual argument of nach. However, all these cases are extremely rare. (45)
a. b.
I reworked the essay, I rethought the plan, I relived my youth dinosaurs preexisted humans, their innovation antedated/predated ours c. Horowitz ... believed he had coexisted Schumann in some way (www.arbiterrecords.com/museum/horowitz.html) d. den Schrank nacharbeiten ‘‘the cubboard after.work’’, ‘put the finishing touches on the cupboard’
Finally, I mention one other riddle that temporal preverbs pose. The prefix preand the adverb beforehand mean ‘before x’ where x is some time or event which is identified from the context. With the prefix, but not the adverb, x is obligatorilly identified with direct objects which have a temporal value (e.g. event nominals), cf. (46a) and (b). I have no solution for this puzzle.
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 141
(46)
a. b.
I got to the partyi late because we were pre-rehearsing the performancek . (rehearsal before k/*i) I got to the partyi late because we were rehearsing the performancek beforehand. (rehearsal before k/i) NOTES
1 NP’s are said to be ‘quantised’ or ‘bounded’ if the entities they refer to are linguistically represented as being finite, as having an inherent quantity. Bare mass nouns and bare plurals (water, books) are examples of non-quantised NP’s; examples of quantised NP’s might be the water, the book(s), seven books. If an incremental theme is not quantised, the event will be atelic: he ate the cakes {in/*for} an hour vs. he ate cakes {for/*in} an hour. For more information, see e.g. Krifka (1998), Jackendoff (1996), Tenny (1994) and their references. 2 Verbrauchen and aufbrauchen ‘use up’ may be related to gebrauchen ‘use’ by prefix truncation. Alternatively, brauchen, which normally means ‘need’, has a second, contextsensitive sense ‘to use’, which is licensed in particular contexts, like these complex verbs and when under the scope of a possibility operator: ich kann’s nicht brauchen ‘I can’t use it’. 3 Similar quasi-redundant uses are found with ‘completive’ up in (3), except here the particle serves to license the suppression of an argument rather than the expression of one. Argument suppression seems to be the purpose of the preverbs in give away and its German equivalents vergeben, verschenken and abgeben. The simplex normally requires expression of the beneficiary. To circumvent this, the possession change subevent in the representation (i) is replaced by a suitable path which does not require mention of the beneficiary, as suggested in (ii).
(i) (ii) 4
Mary gave books to John: Mary gave books away:
do(mary, give) &CAUSE go(books, to john) do(mary, give) &CAUSE go(books, from deictic_centre)
Some speakers reject (18c), but the existence of (mainly American?) speakers who accept it, and a parallel alternation with German abwischen ‘wipe off’, show that it is a possible particle verb. 5 Certainly, I sealed the window badly, so he resealed it implies a repetition of a sealing event rather than a mere return of the window to a sealed state, but I take this to be contextual inference (cf. Dowty 1979: 256f). I cannot find or construct sentences where the prefix superficially appears to have non-minimal scope which cannot be explained as contextual inferences. 6 affect is used with verbs of indeterminate telicity (I swept the floor for/in ten minutes). I assume that the telic reading comes through the inference that the activity brought about its canonically intended result state. Rappoport Hovav and Levin (1998: 114) formalise the relevant verb class, represented by sweep, as [x act<SWEEP> Y ]. They assume that the internal argument is introduced by the idiosyncratic constant rather than by a closed-class predicate. 7 I have twice encountered uses like the house is overbuilt, where overbuilt seems to mean ‘overly elaborately constructed’. This looks like prefixation to an adjectival participle. 8 For instance: Yesterday we had planned to go to the beach with the children. I´t was a perfect day. The sun was shining brightly and there wasn’t a single cloud in the sky. But unfortunately, we ended up having to stay at home. The reason was that Peter felt sick because he had overeaten chocolate. That didn’t exactly make our day.
142
Andrew McIntyre
9
If wieder precedes the object, it has wide scope and allows token differentiation of the object, compare (i) with (34a). See Stechow (1995, 1996) for attempts at an explanation. (i)
da sie wieder ein Kind zur Welt gebracht hat since she again a child to.the world brought has (since she had another child)
REFERENCES Anderson, S. (1971). On the role of deep structure in semantic representation. Foundations of Language 6, 197–219. Booij, G. (1990). The boundary between morphology and syntax: Separable complex verbs in Dutch. In: G. Booij and J. van Marle (eds.), Yearbook of Morpholog 1990. Dordrecht: Foris, 45–63. Brinkmann, U. (1997). The Locative Alternation. Amsterdam: Benjamins. Carlson, G. and T. Roeper (1980). Morphology and subcategorisation. In T. Hoekstra, H. van der Hulst and M. Moortgat (eds.), Lexical Grammar. Dordrecht: Foris, 123–164. Carrier, J. and J. Randall (1992). The argument structure and syntactic structure of resultatives. Linguistic Inquiry 23, 173–235. Curme, G. (1914). The development of verbal compounds in germanic. Beitra¨ge zur Geschichte der deutschen Sprache und Literatur (Halle). 39, 320–61. Dehe´, N. (2000). English particle verbs: particles as functional categories. In: Hero Janßen (ed.), Verbal Projections. Tu¨bingen: Niemeyer. Dehe´, N. and A. Wanner (eds.) (2001). Structural Aspects of Semantically Complex Verbs. Berlin: Peter Lang. Dehe´, N. (2002). Particle Verbs in English. Amsterdam: Benjamins. Dehe´, N., R. Jackendoff, A. McIntyre and S. Urban (eds.) (2002). Verb Particle Explorations. Berlin: Mouton de Gruyter. den Dikken, M. (1995). Particles. Oxford: Oxford University Press. Diesing, M. (1992). Indefinites. Cambridge, MA: MIT Press. Denison, D. (1985). The origins of completive up. Neuphilologische Mitteilungen LXXXVI, 37–61. Dowty, D.R. (1979). Word Meaning and Montague Grammar. Dordrecht: Reidel. Haider, H. (1997). Precedence among predicates. Journal of Comparative Germanic Linguistics 1, 3–41. Hoekstra, T, (1988). Small clause results. Lingua 74, 101–139. Hoekstra, T. (1992). Aspect and theta-theory. In: I.M. Roca (ed.), Thematic Structure. Berlin/New York: Foris, 145–174. Jackendoff, R. (1990). Semantic Structures. Cambridge, MA: MIT Press. Jackendoff, R. (1996). The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14, 305–354. Jackendoff, R. (1997). Twistin’ the night away. Language 73, 534–559. Jackendoff, R. (2002). English particle constructions, the lexicon, and the autonomy of syntax. In Dehe´ et al. (eds.), 67–94. Jacobs, J. (1993). The lexical basis of optional complements. Theorie des Lexikons. Arbeiten des Sonderforschungsbereichs 282. Nr. 53.
Preverbs, argument linking and verb semantics: Germanic prefixes and particles 143
Keyser, S. and T. Roeper (1992). Re: The abstract clictic hypothesis. Linguistic Inquiry 23, 89–125. Krifka, M. (1998). The origins of telicity. In S. Rothstein (ed.), Events and Grammar. Dordrecht: Kluwer. Lenz, B. (1995). un-Affigierung. Tu¨bingen: Stauffenburg. Levin, B. and T. Rapoport (1988). Lexical subordination. Proceedings of the Chicago Linguistics Society 24, 275–289. Levin, B. and M. Rappaport Hovav (1995). Unaccusativity. Cambridge, MA: MIT Press. Li, Y. (1999). Cross-componential causativity. Natural Language and Linguistic Theory 17, 445–497. Lieber, R. and H. Baayen (1993). Verbal prefixes in Dutch. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1993. Dordrecht: Kluwer, 51–78. Lindemann, R. (1998). Bedeutungserweiterungen als systematische Prozesse im Systerm der Partikelverben mit ein-. In: Olsen (ed.), 105–148. Lindner, S. (1983). A Lexico-Semantic Analysis of English Verb Particle Constructions. Bloomington: Indiana University Linguistics Club. Lu¨deling, A. (2001). On Particle Verbs and Similar Constructions in German. Stanford: CSLI. Mateu, J. (2001). Unselected Objects. In: Dehe´ and Wanner (eds.), 83–104. McIntyre, A. (2001a). Argument blockages induced by verb particles in English and German. In: Dehe´ and Wanner (eds.), 131–164. McIntyre, A. (2001b). German Double Particles as Preverbs: Morphology and Conceptual Semantics. Tu¨bingen: Stauffenburg. McIntyre, A. (2001c). Review of Lu¨deling (2001). The Linguist List (www.linguistlist.org/ issues/12/12-1680.html) McIntyre, A. (2002a). Idiosyncrasy in particle verbs. In: Dehe´ et al. (eds.), 95–118. McIntyre, A. (2002b). Event paths, conflation, argument structure and VP shells. To appear in Linguistics. Mu¨ller, S. (2000). Complex predicates. Habil. Thesis, Universita¨t des Saarlandes, Saarbru¨cken. Neeleman, A. and F. Weerman (1993). The balance between syntax and morphology: Dutch particles and resultatives. Natural Language and Linguistic Theory 11, 433–475. Olsen, S. (ed.) (1998). Semantische und konzeptuelle Aspekte der Partikelverbbildung mit ein-. Tu¨bingen: Stauffenburg. Olsen, S. (2000). Against incorporation. Linguistische Arbeitsberichte 74, 149–172. Pinker, S. (1989). Learnability and Cognition. Cambridge, MA: MIT Press. Pustejovsky, J. (1991). The syntax of event structure. Cognition 41, 47–81. Rappaport Hovav, M. and B. Levin (1998). Building verb meanings. In: M. Butt and W. Geuder (eds.), The Projection of Arguments. Stanford: CSLI, 97–134. Rappaport Hovav, M. and B. Levin (2001). An event structure account of English resultatives. Language 77, 766–797. Risch, G. (1995). Verbpra¨figierung des Deutschen. Dissertation, Universita¨t Stuttgart. Simpson, J. (1983). Resultatives. In: L. Levin, M. Rappaport and A. Zaenen (eds.), Papers in Lexical Functional Grammar. Bloomington: Indiana University Linguistics Club, 143–157. Spencer, A. and M. Zaretskaya (1998a). Verb prefixation in Russian. Linguistics 36, 1–39. Spencer, A. and M. Zaretskaya (1998b). Pri-prefixation in Russian. Journal of Slavic Linguistics 6, 107–125.
144
Andrew McIntyre
Stechow, A. von (1995). Lexical decomposition in syntax. In: U. Egli et al. (eds.), Lexical Knowledge in the Organization of Language. Amsterdam: Benjamins, 81–117. Stechow, A. von (1996). The different readings of wieder ‘again’. Journal of Semantics 13, 87–138. Stiebels, B. (1996). Lexikalische Argumente und Adjunkte. Berlin: Akademie-Verlag. Stiebels, B. and D. Wunderlich (1994). Morphology feeds syntax: The case of particle verbs. Linguistics 32, 913–968. Svenonius, P. (1996). The verb-particle alternation in the Scandinavian languages, Ms. Tromsø. (www.hum.uit.no/a/svenonius) Talmy, L. (1985). Lexicalisation patterns. In: T. Shopen (ed.), Language Typology and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon. Cambridge: Cambridge University Press, 57–149. Tenny, C. (1994). Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Wechsler, S. (1989). Accomplishments and the prefix re-, Proceedings of the North Eastern Linguistics Society 19, 419–434. Winkler, S. (1994). Secondary Predication in English. Diss. Tu¨bingen. Wunderlich, D. (1983). On the compositionality of German prefix verbs. In: R. Ba¨uerle et al. (eds.), Meaning, Use, and Interpretation of Language. Berlin: De Gruyter, 452–465. Wunderlich, D. (1987). An investigation of lexical composition: The case of German be- verbs. Linguistics 25, 283–331. Wunderlich, D. (1997a). Argument extension by lexical adjunction. Journal of Semantics 14, 95–142. Wunderlich, D. (1997b). Cause and the structure of verbs. Linguistic Inquiry 28, 27–68. Wunderlich, D. (2001). Prelexical syntax and the Voice hypothesis. In: C. Fe´ry and W. Sternefeld (eds.), Audiatur Vox Sapientiae. Berlin: Akademie-Verlag, 487–513. Zeller, J. (2001a). Prefixes as transitivisers. In: Dehe´ and Wanner (eds.), 1–34. Zeller, J. (2001b). How syntax restricts the lexicon: particle verbs and internal arguments. Linguistische Berichte 188, 461–494. Zeller, J. (2001c). Particle Verbs and Local Domains. Amsterdam: John Benjamins.
Institut fu¨r Anglistik Universita¨t Leipzig Beethovenstr. 15 04109 Leipzig Germany e-mail: [email protected]
Preverbs as an open word class in Northern Australian languages: synchronic and diachronic correlates* EVA SCHULTZE-BERNDT
1. INTRODUCTION Preverbs constituting a distinct part of speech are found in languages of different genetic affiliation throughout Northern Australia. In a large part of the linguistic area defined by the presence of preverbs, they are used to form complex predicates which at first sight bear striking similarities to the separable complex verbs of Germanic languages: the preverb is an uninflecting element which takes primary stress if it appears in preverbal position, but its position with respect to the inflecting verb is variable. Its meaning may be of a spatial or aspectual type. Thus, the Jaminjung examples in (1) and (2) have straightforward translation equivalents in English.1
(1)
jag yirr-ijga-ny binka-bina down 1pl.excl-go-past river-all ‘we went down to the river’
(2)
mangarra burrb nganthi-w-iya! plant.food finish 2sg:3sg-pot-eat.impf ‘you should have eaten up your food!’
However, there is a crucial difference between the preverbs of Germanic languages and the preverbs of Northern Australian languages. The former constitute a closed class, but the latter form an open class with hundreds if not thousands of members, including recent loans. In contrast, in most (though not all) of the Australian languages in question, inflecting verbs form a closed class, often with 30 or less members. At this point, the reader may object that the terms preverb and verb are not appropriate for these two parts of speech, since the term preverb is usually associated with a closed class and the term verb with an open class. This terminological problem is also reflected in the literature on Northern Australian languages: while the terms preverb and verb have been employed by some authors, various other terms can also be found, and none is generally accepted to date. A selection of the terms employed in the Australianist literature is provided in (3), with references. Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 145–177. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
146
E. Schultze-Berndt
(3) Terms employed for the inflecting and uninflecting constituents of Northern Australian complex verbs Verb Verb/Auxiliary Verb Inflecting verb Finite verb Auxiliary Auxiliary
Preverb Verbal particle Coverb
Nash 1982, 1986, Simpson 1991 Hoddinott and Kofod 1976, Merlan 1994 Kofod 1996b, Wilson 1999, Carr 2000, Schultze-Berndt 2000, 2001 Uninflecting verb McGregor 2002 Participle Cook 1988 Base Capell 1979 (Main) verb Reid 1990, Walsh 1996
In this paper, I will – somewhat tentatively and deviating from the terminology I have used in other publications – adopt the term preverb for the uninflecting open-class elements of Northern Australian languages. I hope to show that contrasting their properties with those of more typical (i.e. closed-class) preverbs, both from a synchronic and a diachronic perspective, sheds light on the nature of preverbs in general, since in a sense, Northern Australian preverbs represent an extreme case of what preverbs may look like. The syntactic and semantic differences between Northern Australian and European preverbs are in direct correlation with differences in the size of the preverb class and differences in the division of labour between preverbs and verbs. In the discussion of the synchronic aspects in Section 2, I will focus on preverbs forming separable complex verbs. In this section, all the data used for illustration come from my own fieldwork on Jaminjung,2 a language belonging to the Western Mindi branch of one of the currently recognized Non-PamaNyungan language families, the Mindi family (cf. Chadwick 1997). Some other languages possessing separable complex verbs similar to those in Jaminjung are listed in Table 1. As can be seen from the information provided in the table, the languages mainly differ in the number of (inflecting) verbs.3 These languages, though belonging to different language families, are found in a contiguous geographical area. Inseparable complex verbs of various types also exist in languages of Northern Australia. Although unfortunately, historical data on Australian languages are virtually nonexistent, the comparative perspective permits some conclusions on the diachronic development of preverbs and complex verbs. It can be shown that complex verb formation in at least some of the languages has undergone several cycles whereby former complex verbs have become single, unanalyzable verb roots, which could in turn function as the verbal part of newly arising complex verbs. The comparative and diachronic perspective on preverbs in Northern Australia will be taken up in Section 3. In Section 4 I turn to the question of the analysis of preverb-verb structures. I will argue that the synchronic and diachronic properties of the Northern
Preverbs as an open word class in Northern Australian languages
147
Language family
Language
Number of verbs
References
Wardamanic? Gunwinyguan?
Wardaman Wagiman
>130 >40
Mindi
Jaminjung/ Ngaliwurru Nungali
35 22/28
Miriwoong Gajirrabeng Gija Ungarinyin Wunambal Worrorra Gunin Nyulnyul Nyikina Warrwa Yawuru Bardi Gurindji Ngarinyman Bilinarra
19 24 21 >500 14/11 ~10 >18 >200 >150 >50 >82 >200 ~35 ~35 ~35
Merlan (1994) Cook (1988), Wilson (1999) Schultze-Berndt (2000) Bolt et al. (1971)/ Harvey and Schultze-Berndt (in prep.) Kofod (1996a) Kofod (1996a) Kofod (1996a, b) Rumsey (1982) Va´szolyi (1976)/Carr (2000) Silverstein (1986) McGregor (1993) McGregor (1996a) Stokes (1996), McGregor (2002) McGregor (1994) Stokes (1996), McGregor (2002) Metcalfe (1975) McConvell (to appear) Jones (1994) Nordlinger (1990)
Jarragan
Worrorran
Nyulnyulan
Ngumbin (Pama-Nyungan)
Table 1. Northern Australian languages with separable complex verbs and open classes of preverbs.
Australian complex verbs are best accounted for by a construction-based analysis which allows them to be simultaneously represented as lexicalized expressions and as instantiations of a productive construction type. Diachronic change can be represented as a change in the nature of the construction from a syntactic to a morphological construction, eventually followed by loss of constructional status.
2. PREVERBS AS AN OPEN WORD CLASS: SEMANTIC AND SYNTACTIC CORRELATES As already indicated in Section 1, several properties of preverbs in Jaminjung and languages with separable complex verbs of a similar nature are in direct correlation with the status of preverbs as an open word class. First, preverbs cover a semantic range which corresponds much more closely to that of verbs
148
E. Schultze-Berndt
Map 1. Approximate location of Australian Aboriginal languages or language families referred to in the paper. Key to Map 1 Bininj G.-w. Goon Jam Jarr Jaru Ma Nganki Ngar Nung Nyul Warl Wg Wmb Worr Wrd
Bininj Gun-wok (Gunwinyguan) Gooniyandi (Bunuban) Jaminjungan (Western Mindi): Jaminjung, Ngaliwurru, Nungali Jarragan: Gija, Miriwoong, Gajirrabeng Jaru (Ngumpin) Mangarrayi Ngankikurungkurr and Ngan’gityemerri Ngarinyman and Gurindji (Ngumpin) Nunggubuyu Nyulnyulan: Nyulnyul, Nyikina, Bardi, Warrwa, Yawuru Warlpiri (Ngumpin-Yapa) Wagiman Wambaya (Eastern Mindi) Worrorran: Worrorra, Wunambal, Ungarinyin, Gunin Wardaman Pama-Nyungan/Non-Pama-Nyungan boundary
Preverbs as an open word class in Northern Australian languages
149
than to that of preverbs in the familiar European languages (Section 2.1). Correspondingly, verbal loans are also integrated into the class of preverbs (Section 2.2). Moreover, preverbs exhibit a restricted set of derivational morphology (Section 2.3). Finally, Northern Australian preverbs not only function as constituents of complex verbs, but have a wider range of syntactic functions, e.g. as main predicates in dependent clauses (Section 2.4). In discussing the properties of preverbs in Northern Australian languages, it first needs to be demonstrated that preverbs indeed constitute a distinct part of speech. Preverbs can easily be distinguished from verbs by the absence of verbal inflections (which are obligatory on verbs). Preverbs can also be distinguished from nominals: unlike nominals, they cannot form a noun phrase together with determiners or nominal modifiers, they cannot take the whole range of nominal case markers (although some case markers with preverbs have subordinating function; see Section 2.4.2), and they are not used as arguments. Furthermore, in several languages with gender or noun class systems, for example the Jarragan languages (Kofod 1996b, 1997), preverbs do not exhibit gender marking, and verbs only show gender agreement with core arguments, but never with preverbs. The boundary between preverbs and adverbs, on the other hand, is difficult to draw; while there are minor distributional differences between preverb-like elements with a more adverbial semantics such as ‘slowly, carefully’ and preverbs proper, both are best regarded as subcategories of a larger preverb category.4
2.1. Semantics of preverbs and preverb-verb combinations From the open class status of preverbs in Northern Australian languages, it follows that they cover a much wider semantic range than preverbs in European languages: the semantics of preverbs in the former corresponds closely to that of verbs in the latter. For example, in addition to expressing a spatial path ( jag ‘down’ in (1)) or completion (burrb ‘finish’ in (2)), preverbs may encode manner of motion (yugung ‘run’ in (4)), change of state (ning ‘break off; finish’ in (5) and (6)) and impact (barr ‘smash’ in (6)). Preverbs of these types can typically combine with more than one verb. For example, yugung ‘run’ also collocates with other motion verbs, e.g. -ijga ‘go’, -uga ‘take’ and -arrga ‘approach’, and ning ‘smash’ also combines with other verbs of contact and impact such as -wa ‘bite’, -ina ‘chop, hit with an edge’ or -angu~-angga ‘get, handle, manipulate’ (the last combination is used to describe breaking something off with the hands). The use of different verbs can also influence the transitivity of the complex verb. For example, the intransitive reading of ‘break off, finish’ is rendered by the combination of ning with the verb –ijga ‘go’, as shown in (5). In order to render the transitive reading, the preverb has to combine with one of
150
E. Schultze-Berndt
the verbs specifying the manner and/or instrument involved in contact and impact. In (6) this is the general impact verb -ma ‘hit’. (4)
mulurru=biya yugung wirib-gu janju-wu ngiya old.woman=now run dog-dat dem-dat dem ga-ruma-ny=nu 3sg-come-past=3sg.obl ‘the woman then came running to that dog here’
(5)
ning=biyang ga-w-ijga=wunthu, kikap-mayan buny-b-iyaj break.off=now 3sg-pot-go=cond kick.up-cont 3du-pot-be ‘if it stops, the two will be kickstarting it (a motor boat)’
(6)
ning=biji yirri-ma gurunyung barr smash break.off=only 1pl.excl:3sg-hit.past head ‘we just finished (=killed) it (a flying fox), smashing its head’
Other preverbs only combine with a single verb, which often has a very general meaning. These preverbs often encode specific actions such as ‘scraping’ (wij in (7)) or communicative events such as ‘showing’ (yurrg in (8)). (7)
nganthan wij-wij nganth-angga-m? rdp-scrape 2sg:3sg-get/handle-prs what ‘what are you scraping?’ (addressee was scraping a carrot)
(8)
mulurru-ni gagawuli yurrg gan-garra-ny Gilwi-ni old.woman-erg long.yam show 3sg:1sg-put-past [place.name]-loc ‘the woman showed me yam in Gilwi’
With regard to the semantic range of preverbs, it is perhaps interesting to note that in Jaminjung, and all the neighbouring languages that I am aware of, deixis, which is so prominent in the preverb system of some European languages (cf. German hin/her), is not encoded by preverbs, but rather as semantic component of a verb root (‘come’, ‘bring’), as in Jaminjung, or as a derivational clitic (as in Warlpiri; cf. Nash 1982); often, deixis is not at all encoded in the complex predicate. As the examples above show, the semantic relationship between preverb and verb may vary considerably, in correlation with the wide range of meanings encoded by preverbs. The preverb may specify the manner of the event encoded by the verb, as in (4), or the result, as in (1), (2), and (6). The relationship between the constituents may also be a classificatory one, that is, the verb encodes the general type of event to which the subevent encoded by the preverb belongs. For example, in (7) ‘scraping’ is characterized as a specific type of manipulation by the use of the general manipulation verb -angu~-angga ‘get,
Preverbs as an open word class in Northern Australian languages
151
handle, manipulate’ (compare also (9b) and (10) below). In (5) and (8), the preverb again conveys specific semantics, while the verb only encodes one very general semantic feature of the event. In (5) this is the feature of change of state (encoded by the motion verb –ijga ‘go’ in a secondary sense); in (8), –arra ‘put’ indicates that the event has a component of transfer (in this case, transfer of information).5 Several authors, including Capell (1979), Silverstein (1986), McGregor (1990, 2002) and Schultze-Berndt (2000) have suggested that all kinds of semantic relationships found within complex predicates in the languages of the area can be subsumed under an analysis according to which the semantically generic verbs function as verbal classifiers, although the authors differ in the details of their analysis and their definition of classification.
2.2. The integration of loans as preverbs Considering the open-class status of preverbs in Jaminjung and surrounding languages, it comes as no surprise that recent verbal loanwords are consistently integrated as preverbs, and never as verbs. In a large area in Northern Australia, an English-based creole language, Kriol, is now the language of daily interaction, and the first language of many younger people (Harris 1991). A Kriol verb, bayim ‘buy’ is illustrated in (9a), in a periphrastic conjugation with the non-specific past auxiliary bin (<Engl. been). Kriol verbs borrowed into Jaminjung as preverbs retain the transitivity marking suffix -im, as shown in (9b) and (10). Just like other preverbs, they form complex verbs with one of the Jaminjung verbs. The choice of verb depends on the semantic type of the event that is described; this again reflects the classificatory nature of the generic verbs. For example, the generic verb of manipulation -angu~-angga ‘get, handle, manipulate’, already illustrated in (7), is used with preverb loans encoding both physical (9b) and non-physical manipulation (10). (9)
a.
en waya na, wa thet waitfella bin bayim and wire foc subord dem whitefellow aux:past buy:tr ‘and wire, which the white man bought’
b. mangarra gurrany guny-ngangga-m bayim! plant.food neg 2du:3sg-get/handle-prs buy:tr ‘you two don’t (i.e. never) buy any food’ (10)
mugmug-ni=gun braitenim gan-angu owl-erg=contr frighten 3sg:3sg-get/handle.past ‘the owl frightened him’
152
E. Schultze-Berndt
2.3. Morphological properties of preverbs Corresponding to their open-class status and their wide semantic range, preverbs in Northern Australian languages also have morphological and syntactic properties going beyond those of preverbs in European languages. With regard to their morphological properties, they can form the input to a few types of derivation, including reduplication, aktionsart-changing derivation, and nominalization, illustrated below for Jaminjung. In some languages, including the Jarragan languages (Kofod 1996b), preverbs can be productively derived from nominals. In Jaminjung, preverbs cannot be derived from members of other parts of speech, although there is some degree of overlap between preverbs and nominals in that a few lexical items (discussed in Schultze-Berndt 2000: 73–75) exhibit the distributional properties of both parts of speech. 2.3.1. Reduplication Reduplication of preverbs serves to express extended duration, repetition or intensity of events, as well as multiplicity (or an aggregate) of participants. Usually, this involves full reduplication, although word-initial partial reduplication is also found. An example of reduplication of a preverb encoding an inherently repetitive event, ‘scraping’, is (7). 2.3.2. Continuous marking The suffix -mayan derives preverbs with the aktionsart of activity from preverbs of other classes. The resulting forms exhibit a striking functional resemblance to English present participles in –ing (the latter of course formed from verbs, not from preverbs). They can be used in a complex predicate of the type she came running (where alternatively, the participle may be analyzed as a subordinate clause). The continuous forms also function as main predicates in a progressive construction formed with the verbs -yu ‘be’ or -ijga ‘go’ in auxiliary function, illustrated in (11) (for another example, consider the second clause in (5)). (11)
janyungbari buliki burlug-mayan ga-yu gugu another cow drink-cont 3sg-be.prs water ‘the other cow is drinking water’
For verbs, on the other hand, no derivation affecting the aktionsart exists, at least in Jaminjung. The only type of derivation available to verbs is the transitivity-changing reflexive-reciprocal suffix. 2.3.3. Nominalization Jaminjung preverbs may undergo a few processes of nominalization, such as
Preverbs as an open word class in Northern Australian languages
153
quality nominalization (‘N has the quality of ’), illustrated in (12), and the non-specific ‘associative’ nominalization (‘N is associated with ’), illustrated in (13). Note that verbs cannot be nominalized at all. (12) a. mangurrb-bari ‘black’<mangurrb ‘be black’ b. dum-bari ‘full (of food)’
2.4. Syntax of preverbs 2.4.1. Complex verbs The main function of preverbs in Northern Australian languages, like that of more prototypical preverbs, is complex verb formation. In languages of the Jaminjung type, preverbs constitute independent phonological and syntactic words within the complex verb, as the examples in Section 2.1 have shown. Usually (in about 90% of complex verb tokens) they are found in a position immediately to the left of the verb. But they may also be separated from the verb by other constituents, as in (4), and appear postverbally, as in (9b). Moreover, occasionally more than one preverb is found in a single clause, under a single intonation contour, as in (6). Similar findings hold for the other languages listed in Table 1. The syntactic analysis of Northern Australian complex verbs will be further discussed in Section 4. 2.4.2. Preverbs in case-marked subordinate clauses Quite unlike separable preverbs in Germanic languages, separable preverbs in Northern Australian languages may often function as the main predicate in nonfinite subordinate clauses. Subordination is usually achieved by the use of a case marker (from a subset of the nominal case markers), attached either to the predicate (i.e. the preverb) alone, or to all constituents of the subordinate clause. The case marker attaches either directly to the preverb, as in Jaminjung, or follows a special subordinating suffix on preverbs, as in Wagiman (Wilson 1999: 85). For example, the dative case is often used to mark purposive adverbial
154
E. Schultze-Berndt
clauses, illustrated in (14). In many of the languages, the allative case indicates that the controller of the subordinate clause is an undergoer; in other words, in (15) only be the pig (the undergoer) can be interpreted as standing up, not the actor. (14)
guyug=biyang nganji-bili=rrgu [wujuwuju fire=now 2sg:3sg-pot:get/handle=1sg.obl small wirrigaja-wu] cook-dat ‘you should get fire(wood) for me now, for cooking the small (fish)’
(15)
mung gani-ngayi-m=biyang pigibigi [gurdij-bina] look.at 3sg:3sg-see-prs=now rdp:pig stand-all ‘it is looking at the pig that is standing up’
Some of the languages, e.g. Wardaman (Merlan 1994: 276–290), but not Jaminjung, also have non-finite forms of verbs which may be used as predicates of case-marked subordinate clauses. The use of non-finite inflecting verbs is more common in languages like Warlpiri where most preverbs are inseparable (see Section 3). 2.4.3. Preverbs as semi-independent predicates A characteristic of the separable preverbs in Northern Australian languages which they share with separable preverbs in Germanic languages is the possibility to occur as the sole predicate in a separate intonation unit. Usually they constitute the only word in an intonation unit, as in (16), but occasionally they are also found with arguments or adjuncts, as in (17). In this use, preverbs are independent predicates in the sense that they are not part of a complex predicate, but they are dependent on the linguistic or nonlinguistic context in their interpretation – hence the term semi-independent predicate. In the following examples, a backslash (c) indicates a falling (final) boundary intonation and a comma a rising (nonfinal) boundary intonation. (16)
ning burra-wa-na=gurra c ngilthig c ning c ngilthig c break.off 3pl:3sg-bite-impf=emph swallow break.off swallow ‘they were biting something off (a piece of flesh) – swallow – off – swallow’ (crocodiles attacking cattle)
(17)
dud hold.one
gani-bila=ma wirib=gun c 3sg:3sg-pot:get/handle=subord dog=contr
Preverbs as an open word class in Northern Australian languages
155
gugu-ngunyi yirr c water-abl move.out ‘(...) when she was going to grab the dog, (pulling it) out of the water’ Expressions of this type are only employed in highly contextualized genres, i.e. narratives and procedural texts, conversations, comments on ongoing situations, and as commands. When asked to repeat an utterance for clarification, or for translation out of context, speakers will supply the appropriate inflecting verb, i.e. produce complex verbs. Preverbs as semi-independent predicates are also absent from isolated elicited sentences, and from other more decontextualized genres. The effect conveyed by this use of preverbs is not unlike that conveyed by the use of separable preverbs in isolation in some registers of spoken language in familiar European languages. An example of the use of preverbs and adverbs without a verb from German (Cologne dialect) is given in (18); the preverbs (with Standard German equivalents given in brackets) are op (auf) ‘open’, eren (herein) ‘inside [dir.]’, zo (zu) ‘closed’ and fott (fort) ‘away’. This short section is the climax of the narrative from which it is taken. This is the account of a traffic accident and the scene described in (18) is that of a car (with a person named Klaus as the driver) driving straight into the center of a roundabout which is planted with a hedge: the hedge opens, Klaus (in the car) goes in, the hedge closes again, and the car is gone (i.e. no longer visible). Because of the absence of finite verbs, the description in (18) is maximally condensed and dramatically more effective than the paraphrase just given. German (Cologne dialect) (18)
jo un da jing dat kladderadaatsch, Heck op, well and then went that [ideophone] hedge open Klaus eren, Heck zo, Auto fott [proper name] inside hedge closed car away ‘well and then it went crash-bang: hedge open, Klaus inside, hedge closed, car gone’ (Bhatt and Lindlar 1998: 56)
In Jaminjung, preverbs as semi-independent predicates may also be used with the illocutionary force of a command, even though imperative inflection on verbs exists as an alternative. Both possibilities are illustrated in (19). (19)
gabardag, gad ba-manggu, gad, jarr:: gurdij! quick cut imp-hit cut put.down stand ‘quick, cut it! cut! down! stand up!’ (to a boy cutting up a kangaroo while being videotaped)
This use of preverbs is similar to that of English preverbs in examples like (20), termed condensed directives by Biber et al. (1999: 1102).
156
E. Schultze-Berndt
(20) a. Down! Down the stairs please! b. Hands off the jug please
2.5. Summary The properties of verbs and preverbs in Jaminjung and other Northern Australian languages with separable complex verbs are summarized in Table 2. As can be seen from Table 2 and the preceding discussion, Northern Australian separable preverbs share with closed-class separable preverbs in other languages their main function of forming complex verbs together with an inflecting verb, but also the ability to occur as semi-independent predicates in stylistically marked contexts. In other respects, though, Northern Australian preverbs clearly have the properties of a major lexical category. Semantically, they cover a wide range of ‘‘verbal’’ notions, and include many recent loanwords; in this respect they resemble verbs in serial verb constructions in languages such as Chinese, and the verbal nouns in light verb constructions in languages such as Hindi. Syntactically, they may form the main predicate in non-finite subordinate clauses, and in this respect are reminiscent of non-finite verb forms in European languages. It should be noted though that preverbs do not occur in some of the contexts that one might expect for non-finite verb forms. For example, complement-taking verbs such as ‘want’, and hence complement constructions, are virtually non-existent in Jaminjung and other Northern Australian languages, and therefore preverbs do not occur in non-finite complement clauses of this type. And although they fulfil the function of English present participles in the progressive construction, preverbs cannot function as
Size of word class Verbal inflections Input to nominalization Non-finite forms Ability to function as (simple) main predicate
Verbs
Preverbs
closed (in a subset of the languages) yes (person, tense, aspect, mood) no no (in a subset of the languages) in independent and finite subordinate clauses (in languages with nonfinite verb forms also in non-finite clauses)
open no yes does not apply (there are no finite forms) only in dependent clauses (as semi-independent predicates and in nonfinite subordinate clauses)
Table 2. Properties of verbs and preverbs in Northern Australian languages with separable complex verbs.
Preverbs as an open word class in Northern Australian languages
157
verbal nouns in argument position or as attributes, in the way participles in most European languages can. In the next section, I will argue that from a diachronic perspective, too, Northern Australian preverbs exhibit both similarities and differences to closedclass preverbs in other known languages.
3. A COMPARATIVE AND DIACHRONIC PERSPECTIVE 3.1. Preverbs as an areal feature As already indicated, complex verbs of the type described in Section 2 – with constituents belonging to distinct lexical categories, one of which is a closed class – are not unique to Jaminjung. Rather, they constitute an areal feature in Northern Australia, shared by almost all non-Pama-Nyungan languages, and in addition by some languages of the Pama-Nyungan family.6 Complex verbs appear to have been a feature of these languages for a considerable time. The location of origin of this construction, and the direction of its spread, cannot be traced with any confidence; Capell (1976: 615) suggests that it spread from west to east. Overviews – mostly rather cursory – of complex verb systems attested in Australia can be found in Capell (1976, 1979), Dixon (1980: 426ff.), Nash (1982: 166–169), Blake (1987: 118ff.), Schultze-Berndt (2000: Ch. 7) and for the Daly River group of languages, in Tryon (1974). An in-depth overview is presented in McGregor (2002). The degree of structural convergence under the influence of multilingualism can be illustrated with the pairs of translation equivalents in (21). These were offered by a speaker who is bilingual in Jaminjung and Ngarinyman, a neighbouring Pama-Nyungan language of the Ngumbin group. The Ngarinyman verb roots and inflections in (21b) are distinctly Pama-Nyungan (note the lack of pronominal prefixes on the verbs), but the system of complex verbs is very similar to that of Jaminjung. The two languages also share a number of cognate preverbs such as bag ‘break’ in (21). Jaminjung: (21)
a.
ga-jga-ny yina-wurla guyug-gu::, bag 3sg-go-past dem-dir fire-dat break gani-bila=ma, 3sg:3sg-pot:get/handle=subord ‘she went over there for firewood, and when she was going to break it off ...’
158
E. Schultze-Berndt
Ngarinyman: b. gamurrwarinyjama jalalang maniny=bawurla gankurla garndi midnight:top hang get/handle:past=3du up wood jawi-wu c bugu bag .. maniny=bawurla c fire-dat just break get/handle:past=3du ‘in the middle of the night the two got a branch for firewood by hanging from it, and they just broke it off’ It is quite possible that the greater susceptibility of preverbs to borrowing may in itself have been a motivating factor for the prominence of complex verb constructions in the multilingual environment in Northern Australia, and would have assisted the spread throughout a larger area. Complex predicate formation with one or few inflecting verbs is a well-known strategy in bilingual interaction (cf. e.g. Romaine 1989: 120–164, Myers-Scotton and Jake 1995: 303) since it reduces the need to learn a large number of verbal paradigms in a second language, while uninflecting components can be easily inserted (as code-switching) and, ultimately, borrowed. Retaining only a closed class of verbs would seem like a logical, if somewhat extreme, extension of this strategy (cf. McConvell and Schultze-Berndt 2001). Although at first sight, this requires of speakers that instead of learning verbal paradigms they learn a large number of complex verb collocations, the burden is in fact reduced by the – attested – possibility of calquing the collocations from their first language.
3.2. Inseparable complex verbs in Northern Australia The types of complex verbs attested in Northern Australia constitute a continuum, ranging from the clearly separable complex verbs of Jaminjung and some of the neighbouring languages, discussed in Section 2, to complex verbs whose components have lost any structural and semantic independence and which may be treated, synchronically, as unanalyzable verb roots. Some languages representing stages in this continuum are surveyed in the following. In a number of languages, the phrasal origin of the complex verbs can easily be recognized; however, they behave like single units distributionally. Their constituents are usually not used or recognized in isolation by native speakers, and their order is fixed. On the other hand, verbal inflections intervene between the uninflecting and the inflecting component of the complex verb; in other words, the uninflecting element (the former preverb) is joined onto the inflecting element (just as in perfect participle formation in German and Dutch). The phrasal origin is also reflected in the stress pattern of these complex verbs, in that word stress is assigned to both components (see e.g. McGregor 1990: 126, 134). Example (22) shows a complex verb of this type, from Gooniyandi, a
Preverbs as an open word class in Northern Australian languages
159
language of the Bunuban family. The boundary between the bound preverb and the verbal element is represented as ‘+’.7 Gooniyandi: (22)
ga´rd+binggu´runi hit+fut:2pl:3pl:hit ‘you (pl) will hit them’ (McGregor 1990: 200f.)
This type of complex verb is also found in the Marran languages of South-East Arnhem land, and in the languages of the Daly River area. For one of these languages, Ngan’gityemerri, a restructuring, within less than 60 years, of separable complex verbs into inseparable complex verbs has been documented (Reid to appear). A stress pattern similar to that found in Gooniyandi is described for Warlpiri complex verbs by Nash (1982, 1986: 111f.). In Warlpiri, for a subtype of complex verbs, the constituent order is variable. However, when the preverb precedes the inflecting verb, it often has reduced word status in terms of phonological shape, and in this sense becomes a bound stem. In Warlpiri, verbs are inflected only by suffixes, and person and number are marked on second position clitics. If the preverb immediately precedes the inflecting verb, the two stems are therefore contiguous (again, the boundary is indicated with a plus sign). For example, wuruly in (23a) is not a well-formed phonological word; it has to be augmented with the final syllable -pa, as in (23b) and (23c), in order to be able to follow the inflecting verb. However, only a subset of preverbs is of the separable type in Warlpiri; most preverbs are inseparable, i.e. only the position in (23a) is available for them. A similar patterns has been described for the related language Jaru by Tsunoda (1981: 177). Warlpiri: (23)
a.
wuruly+ya-ni=rli hide+go-npast=1du.incl b. wurulypa=rli ya-ni hide=1du.incl go-npast c. ya-ni=rli wurulypa go-npast=1du.incl hide ‘let’s go and hide’ (Nash 1986: 52)
Mangarrayi, a Non-Pama-Nyungan language, is unusual in that it has both separable preverbs similar to those in Jaminjung, illustrated in (24), and compound verbs where preverb and verb form a single phonological and morphological word and are jointly flanked by inflectional affixes, as shown in (25) (Merlan 1982: 123ff.).
160
E. Schultze-Berndt
Mangarrayi: (24) barnbayi ja-wurla-wu-yi-n tease 3–3pl-hit-refl-prs ‘they are joking with each other’ (Merlan 1982: 135) (25)
ngiyan-galij-ma-ny 3sg:1pl.excl-report-V-past ‘he reported to us’ (Merlan 1982: 68)
Compound verbs of the type in (25), with no verbal inflections intervening between the two constituents, are particularly prone to univerbation, i.e. lexicalisation to a single prosodic and grammatical word. Indeed, a number of nonPama-Nyungan languages, including Nunggubuyu and languages from the Gunwinyguan family, have verb stems which seem to have their origin in compound verbs. In many cases, the former constituents have lost their independent status. For example, Heath (1984: 470) reports that in Nunggubuyu, a large set of verbs which now are simplex verb roots phonologically and semantically have a recurring second element -bu~-wu. This form on its own is still attested as a verb meaning ‘hit, kill’ in Nunggubuyu; however, the first element (presumably an old preverb) is often not attested as an independent form in the language. Some examples of these forms, termed auxiliary compounds by Heath, are given in (26). (26)
a. b. c. d. e. f.
-dhabid+bu -lhilgi+wu -mad+bu -ya:gi+bu -lham+bu -ru+bu
‘tie up’ ‘break’ ‘complete, do well’ ‘sneeze’ ‘restrain’ ‘cook, burn’
In another type of verb in Nunggubuyu and some neighbouring languages, neither constituent can appear as an independent word. However, the original inflecting verbs (now: recurrent submorphemic elements), described as verbal augments by Heath (1984: 407–422), often identify verbs with a common semantic basis. Some examples from Nunggubuyu are given in (27). Corresponding elements are sometimes treated as conjugation markers in grammatical descriptions of other languages. (27)
a. b. c.
-ROOT-nga- mainly intransitive verbs of position -ROOT-ra transitive verbs of transporting -ROOT-ya– often: verbs encoding physical impingement on objects
Reflexes of univerbation can also be found in Jaminjung. For example, the
Preverbs as an open word class in Northern Australian languages
161
(synchronically simplex) verb root -ina+ma ‘kick/step’ is transparently related to the verb -ina ‘chop, hit with an edge’. As a submorphemic element, +ma is found in no less than four verb roots, listed in (28). (28)
a. b. c. d.
-ina+ma ‘kick/step’ -anja+ma ‘bring’ -maliny+ma ‘make’ -yang+ma ‘fear’
Similar reflexes of complex stems in the inflecting verb class are found in other Non-Pama-Nyungan languages, e.g. Wardaman (Merlan 1994: 173f.), Wagiman (Wilson 1999: 25), and the Jarragan languages (Kofod 1996a). Being diachronically complex, these verbs now function as simple verbs and may enter again into complex verb formation with separable preverbs. Both language-internal and comparative evidence thus suggests that Northern Australian languages have gone through several cycles of complex verb formation, with different stages in this cycle reflected by the synchronically observable types (cf. Capell 1979, Dixon 2001, McGregor 2002). This cyclical development is represented in Figure 1. Table 3 contains examples of languages which synchronically represent the postulated stages in Figure 1. One of the postulated stages in Figure 1 has not been illustrated so far. This is stage IV, represented by languages with an open class of verbs and a closed class of preverb-like elements. Languages of this type are indeed attested, and shed a light on the possible origin of open-class preverbs in Northern Australian languages. This is the topic of the following section.
3.3. The rise of preverbs Preverbs forming a closed class are generally assumed to originate from spatial adverbs, and this is also one of the likely sources for preverbs in Northern Australian languages. For these languages, it has occasionally also been suggested that some preverbs have a nominal origin. However, at least in Jaminjung, there is only a handful of forms which can be traced to a nominal origin, or which have double category membership synchronically. Obviously, the class of preverbs, once formed, has also been massively extended by borrowing; this would have been true for earlier stages just as for the historically more recent Kriol loans discussed in Section 2.2. The remaining potential sources for preverbs are verb roots or stems which would have had to be stripped of their inflections, and sound-symbolic, ideophonic elements. At first sight, these are less plausible candidates, but comparative evidence suggests that both of these may have played a considerable role in establishing a class of preverbs with the potential of becoming an open class.
162
E. Schultze-Berndt
Figure 1. A cycle of complex verb formation in Northern Australian languages.
For some languages of Arnhem Land which have an open class of verbs, including Ritharngu, Nunggubuyu and Ngarndi, Heath (1976, 1984) has identified small classes of preverb-like uninflecting elements, which he calls root forms since some of them can be traced back to attested verb roots. These forms may be used as semi-independent predicates in a separate intonation unit and as commands, but also in combination with a verb: These [...] root forms are often used as adjuncts to the inflected verbs, and serve mainly to add an expressive flavour [...] The nuance [...] can sometimes be expressed in translation by an expression such as ‘all of a sudden’ or even an interjection like ‘Pow!’ or ‘Bang!’. However, the
Preverbs as an open word class in Northern Australian languages
163
Type
Example language(s)
Source
I
Jaminjung Gurindji/Ngarinyman Gooniyandi (Warlpiri) Nunggubuyu Bininj Gun-wok (Jaminjung) Nunggubuyu Nyulnyul Ungarinyin Wardaman Warlpiri
Schultze-Berndt (2000) McConvell to appear McGregor (1990) Nash (1982, 1986) Heath (1984 Evans in press
II III
IV V VI
Heath 1976, 1984 McGregor 1996a Rumsey 1982 Merlan 1994 Nash 1982, 1986
Table 3. Some languages representing the types of complex predicates found in Northern Australian languages.
[...] elements are more clearly related to verbal notions. (Heath 1976: 736ff.) In short, these uninflecting elements, illustrated in (29) and (30), resemble ideophones as described for many other languages (cf. the contributions in Hinton et al. 1994 and Voeltz and Kilian-Hatz 2001). Ideophonic root forms in Nunggubuyu: (29)
ni-waya-wayama-ngi-maga::: guj c ni-warga-n-di / he_kept_going chuck! he_threw_it ‘he went along throwing the cane grass spear at the bushes, guj!’ (Heath 1980: 29f.; prosodic transcription modified by ESB based on the original audio recording)
(30)
girjag ngayawi-nyinyung c no! mine jurg jurg jurg jurg jurg wini-jurjurga-nyji-ny c gudbi:::j; push! they_shoved_each_other grab! girjag ngayawi wara:wa c no mine this ‘ ‘‘No! (He is) mine!’’ ... Push! Push! Push! ..., they started shoving each other, grab! [Gecko grabbed Emu. Emu said,] ‘‘No! This (boy) is mine!’’ ’ (Heath 1980: 41; prosodic transcription modified by ESB based on the original audio recording)
Heath (1976) proposes a diachronic scenario according to which constructions like those in (29) and (30) could have been precursors of complex verbs of the
164
E. Schultze-Berndt
type found in neighbouring languages. Similar observations and suggestions have been made by Alpher (1994) and McGregor (1996b: 359, 2001, 2002: 338–339). Indeed, as shown in more detail in Schultze-Berndt (2001), Jaminjung preverbs have uses in which they show striking similarities to ideophones, both in terms of expressive prosody and in syntactic behaviour. One example of the ideophonic use of a preverb in Jaminjung is given in (31); the exclamation mark stands for expressive prosody, in this case, an extreme rise in pitch. (31) bugu mulurrng ganama-ny bugu !dibard just crash 3sg:3sg:kick/step-past just jump ga-ruma-ny yinthu-wurla=wung 3sg-come-past dem-dir=restr ‘it just crashed on the ground and he just came jump! over here (frightened by the noise)’ If this rather tentative reconstruction of a historical scenario, based on the comparative evidence from Nunggubuyu and other languages, is correct, Northern Australian open-class preverbs differ from European closed-class preverbs not only in the size of the class, but also in their diachronic development. While European preverbs are more or less restricted to a small class originating from adverb-like elements with spatial semantics, classes of preverbs in Northern Australian languages have presumably been extended by expressive forms representing actions and processes semantically, which may in turn have had their origin either in former verb roots (stripped of their inflections) or in uninflecting sound-symbolic elements.
3.4. Lexicalization vs. grammaticalization The reconstructed cycle of complex verb formation in Northern Australian languages sketched in Figure 1 in Section 3.2 represents the lexicalization, or more precisely univerbation, of originally separable complex verbs, followed by a new wave of complex verb formation. The old complex verbs become compounds whose elements eventually loose their morphemic status and fuse into a single unanalyzable verb root. The development, as discussed so far, is not an instance of grammaticalization, since none of the elements of the complex verbs become productive grammatical formatives. A process of univerbation is also attested for complex verbs in European languages (cf. e.g. Lehmann 1995: 97–104, Vincent 1999). In these languages, however, we also encounter instances of grammaticalization of preverbs, i.e. development into transitivity-changing prefixes (e.g. German be-) or aktionsartchanging elements (e.g. English up) which may in turn develop into aspectual prefixes.
Preverbs as an open word class in Northern Australian languages
165
In Northern Australian languages where preverbs form an open class but verbs may constitute a closed class, we would expect verbs to be more prone to grammaticalization than preverbs, in accordance with the principle that ‘‘one element of a compound may become a derivational affix if it occurs in a large number of combinations’’ (Bybee 1985: 106). This expectation is indeed borne out. Derivational morphology, including productive verbalizers for nominals and valency changing affixes such as causativizers, can be traced back to former independent verbs in some Australian languages. For example, the form -wo in Bininj Gun-wok, which is synchronically attested both as a verb root with the meaning ‘give’ and as part of unanalyzable stems, in addition functions as a factitive verbalizer (Evans in press: Ch. 8.2.2). Derivational affixes which still betray their origin as free verbs are also found in Warlpiri (Nash 1986: 42f.). These affixes function as general inchoative or causative verbalizers. Typically, Kriol loans are also integrated in these languages by affixation of one of these verbalizers rather than by combination with an independent verb as in Jaminjung (see Section 2.2). Two other, related, grammaticalization paths involving formerly independent verbs can be identified. First, motion verbs may develop into inflectional affixes of associated motion, as described for the Arandic languages (Koch 1984, Wilkins 1997). Second, a small number of verbs, typically stance and motion verbs, may grammaticalize into auxiliaries which encode distinctions of tense, aspect, modality, and possibly voice, polarity and direction, but do not add to the lexical semantics of the resulting complex predicates. In Jaminjung, two of the inflecting verbs, the general stance verb -yu ‘be’ and the general motion verb -ijga ‘go’, also function as auxiliaries in a progressive construction (see Section 2.3.2), although they do not show any formal signs of grammaticalization. The use of stance and motion verbs as imperfective auxiliaries is also reported for some of the neighbouring languages, e.g. Ngankikurungkurr and Ngan’gityemerri (Hoddinott and Kofod 1988, Reid 1990). The Barkly languages, now also referred to as Eastern Mindi, are of particular interest in this respect since they are distantly related to Jaminjung, which belongs to the Western Mindi branch of the same family (Chadwick 1997). In these languages, former preverbs now function as main verbs, but in some of the languages a small number of verbs have been retained in a purely auxiliary function. For example, the Wambaya auxiliary -amany (Nonpast: -ulama), illustrated in (32), is a plausible cognate of a Jaminjung verb, -ruma ‘come’ (past tense form -ruma-ny), since in addition to tense, aspect and modality it also carries directional information (Nordlinger 1998: 151–153).
166
E. Schultze-Berndt
Wambaya (32)
dirdibulyini-nmanji g-amany magi-nmanji yarru peewee-all 3sg-past.towards camp-all go ‘she came to Peewee(‘s) camp’ (Nordlinger 1998: 246)
The possibility that all members of a small closed class of verbs could end up as a set of grammaticalized bound morphemes should also not be ruled out completely. Indeed, McGregor (1990, 2002) convincingly argues that the former inflecting verbs in Gooniyandi should be regarded as a grammaticalized system of (verbal) classifiers. The class of classifiers comprises only 12 bound forms which cannot occur independently of the former preverb (now the main verb; cf. ex. (22)). Thus, in languages where preverbs form an open class and verbs form a closed class, the tendency is for verbs, but not preverbs, to develop into derivational or even inflectional affixes. In Northern Australia, the only case of preverbs with a mainly grammatical function reported in the literature involves the so-called dative-adjunct preverbs in Warlpiri with a valency-increasing function (Nash 1982: 183). Alternatively, as shown in Section 3.2, preverb-verb combinations may be subject to compounding (univerbation), in other words, to lexicalization.
4. A CONSTRUCTION-BASED ANALYSIS OF NORTHERN AUSTRALIAN COMPLEX VERBS In the literature on complex predicates of the separable type, a major issue is the question whether they are part of the lexicon or productively formed in the syntax; cf. the discussions in Mohanan (1994: 234–236), Goldberg (1996), Ackerman and LeSourd (1997), Hampe (1997), and Booij (2002: Ch. 6.4), among many others. The same question obviously also arises in the case of the separable complex verbs of Northern Australian languages. On the one hand, in languages like Jaminjung both the preverb and the verb retain their phonological and syntactic independence, and spontaneous new combinations are possible (e.g. those involving Kriol loans; see Section 2.2.). On the other hand, most preverb-verb combinations clearly constitute lexicalized collocations. This is evident from the lack of semantic transparency in some combinations, but also from the mere fact that apart from the 30 or so inflecting verbs in the language, all predicates in finite clauses are complex. In this section I propose a construction-based model in order to account for this dual nature of preverb-verb combinations and for the joint contribution of preverb and verb to the argument structure of the resulting complex verb. A construction-based account of argument sharing in Jaminjung is offered in Section 4.1. In Section 4.2, I summarize
Preverbs as an open word class in Northern Australian languages
167
the evidence in favour of regarding the preverb-verb structure as a productive construction, but most actual preverb-verb combinations as lexicalized collocations.
4.1 Argument sharing in Jaminjung Complex verbs of the type found in Jaminjung pose a challenge for the mainstream approach to argument structure. The standard approach is characterized by the view that the syntactic behaviour of relational lexemes – of which simple verbs are seen to be the prototype – is determined by a lexical property of syntactic relationality. This is couched in terms like ‘‘verbs govern their complements’’, ‘‘verbs assign case’’ or ‘‘verbs project their argument structure’’. Complex predicates constitute a problem for this approach because they consist of more than one (potentially) relational lexeme which may influence the syntactic behaviour of the predicate. Within the lexicalist approach to argument structure, three analyses of preverb-verb structures are logically possible. The first possibility is that the preverb is not relational, i.e. it does not have syntactic valency or the potential to govern complements. This means that argument structure is determined by the verb alone. It has been demonstrated that this analysis cannot be maintained for preverb-verb combinations e.g. in Latin (cf. Lehmann 1983) or in Germanic languages (cf. e.g. Booij 2002: 211–213). In Australian languages, likewise, the preverb often influences the valency of the complex verb, as will be shown for Jaminjung below. A second possibility, which is the converse of the first, offers itself for those Northern Australian languages with closed classes of verbs, which in their semantic generality can be compared to light verbs in other languages. The verb is considered to be semantically empty to the extent that it has no or only a ‘skeletal’ argument structure specification. Instead, argument structure is determined by the semantically specific, non-finite element alone, which in this case, of course, has to be relational. This analysis has been suggested for the light verb constructions e.g. of Japanese (Grimshaw and Mester 1988), but has been refuted for Japanese by Matsumoto (1996), and for light verb constructions in Hindi by Mohanan (1994). It is also not tenable for preverb-verb structures in Jaminjung (see Schultze-Berndt to appear, 2000: Chs. 3, 4 for details). A third possibility is to treat the complex predicate as an unanalyzable lexical unit which determines argument structure as a whole, just like a simple predicate. This is one of the possible analyses of lexicalized complex verbs in European languages. However, it does not allow generalizations on the semantic contribution of the individual constituents, in particular on the contribution of preverbs to argument structure. In the case of Jaminjung, too, a number of generalizations about the possible combinations of preverbs and verbs, and about the morphosyntactic behaviour of the resulting complex verbs, can be
168
E. Schultze-Berndt
stated most clearly if one considers each a relational lexeme in its own right. In any case, Jaminjung preverbs have to be regarded as relational predicates because they may function as the only predicate in dependent clauses, i.e. without a verb (see Sections 2.4.2 and 2.4.3), and as the main predicate in a productive progressive construction with a verb in auxiliary function (see Section 2.3.2). An alternative, fourth possibility has been explored for complex predicates in a number of languages. According to this analysis, both constituents of a complex predicate are relational, and jointly determine its syntactic possibilities. This approach necessarily leads to the adoption of a concept of argument fusion (or argument sharing): the relational properties of two (or more) lexemes join forces, as it were, in determining the relationality of the complex predicate. Analyses of this type have been suggested for Latin particle verbs by Lehmann (1983), for light verb constructions in Hindi, Urdu, and Japanese by Mohanan (1994, 1997), Butt (1997), and Shibatani (1996), respectively; for serial verb constructions in a number of languages by Foley and Olson (1985), Durie (1997), and Andrews and Manning (1999), and for the complex verbs of the Northern Australian language Wagiman by Wilson (1999), among many others. Argument sharing can be implemented in any framework that allows for unification. Here, I adopt a Construction Grammar approach to argument structure, as outlined in Goldberg (1995). According to this approach, grammatical constructions – including those representing arguments – are seen as signs in their own right, i.e. their existence does not depend on the valency of lexical items. Predicates are not assigned a syntactic, but only a semantic argument structure (for a justification of the criteria adopted for the identification of semantic participants in Jaminjung see Schultze-Berndt 2000: Ch. 4.1.3 and Schultze-Berndt, to appear). Participants (semantic arguments) can be linked directly to the argument roles of grammatical constructions. Lexical items and constructions may unify on the basis of semantic compatibility, although this does not preclude restrictions in productivity by degrees of conventionalization. In Figure 2, based on example (33), three separate boxes represent three distinct argument structure constructions. (The ordering of the boxes does not represent any hierarchical ordering, since all constructions are simultaneously present). The two upper boxes represent two case marking constructions, consisting of a verb and a case-marked noun phrase, labelled ERG(ative) and ABS(olutive), respectively. The bound pronominal construction, consisting of a verb root and its A(ctor) and U(ndergoer) prefix, is labelled ‘TRANS(itive)’ and represented by a box with double lines.8 It should be thought of as embedded in the V-slot of the case-marking constructions (something that is not adequately captured by the notation). (33)
wirib-di jag gan-ardgiya-ny thanthiya munurru dog-erg go.down 3sg:3sg-throw-past dem bee ‘the dog threw down those bees’ (from a Frog Story narration)
Preverbs as an open word class in Northern Australian languages
169
Figure 2. Argument sharing of a bivalent verb and a monovalent preverb.
Figure 2 also illustrates the construction-based approach to argument sharing. The same argument slot of a construction may represent – by unification – participants of more than one relational lexeme. The directional preverb jag ‘go down’, shown in combination with an intransitive verb of motion in (1), is here combined with a transitive verb of caused motion, in a complex verb which receives a resultative interpretation, roughly ‘cause something to go down by throwing it’. Preverb and verb jointly instantiate the complex verb construction (CV), represented by the third box. Preverb and verb, together with their semantic participants, are placed on separate lines in between the argument structure constructions, and the semantic participants are linked to a noun phrase argument slot and/or to a bound pronominal argument slot. Thus, the ‘causer’ participant of the verb -ardgiya ‘throw’ is linked both to the ergative noun phrase and to the A bound pronominal slot, while the patient, the ‘moving entity’, is linked to the absolutive noun phrase and the U bound pronominal slot. The patient participant of the verb shares both these slots with the only participant of the directional preverb jag ‘(go) down’, which in this case is also labelled ‘moving entity’ (no particular theoretical relevance is assigned to the labels provided for the participant roles). Figure 3, based on example (8), illustrates argument sharing for a complex verb where the preverb determines the overall argument structure. A few preverbs with a meaning component of transfer, including yurrg ‘show, teach’, are
170
E. Schultze-Berndt
Figure 3. Argument sharing of a bivalent verb and a trivalent preverb.
trivalent. This preverb exclusively combines with the bivalent verb -arra ‘put’. With the resulting complex verb, the ‘shower’ participant is linked to the A pronominal prefix and the ergative noun phrase, and the ‘recipient’ participant is linked to the U pronominal prefix (and optionally to an absolutive noun phrase, not present in (8)). The third participant, the theme or ‘entity shown’, is (again optionally) represented by an absolutive noun phrase; in (8) this is gagawuli ‘long yam’. The preverb yurrg thus introduces a third participant (the recipient) to the complex verb, encoded as a core argument. A construction-based analysis thus allows one to represent argument sharing of two relational predicates (in this case, preverb and verb); the two constituents can differ in their semantic valency (i.e. the number of semantic participants), as in the examples discussed here, or have the same valency, in which case all of their semantic participants align. Argument sharing, in this approach, is seen as affecting only the semantic valency of the complex predicate, i.e. the level of semantic participants. Since argument structure on the morphosyntactic level is seen as independent from the semantic level in principle (although naturally there is a strong correlation between the two), there is ample room for variation in the morphosyntactic expression of arguments. In Jaminjung, for example, argument NPs can be freely omitted, transitive agents may take ergative or absolutive marking, or an event itself may fill a semantic but not a syntactic argument role (see Schultze-Berndt 2000: Ch. 4 for details).
Preverbs as an open word class in Northern Australian languages
171
4.2 The dual nature of separable complex verbs Verb and preverb in Jaminjung not only jointly determine the argument structure of the resulting complex verb; they also jointly determine the aktionsart of the combination (see Schultze-Berndt 2000: Chs. 3 and 4 for details). Separable complex verbs in Jaminjung and other Northern Australian languages can thus be regarded as multi-headed, and meet the definition of complex verbs as proposed in the recent literature (e.g. Butt 1997: 108, Andrews and Manning 1999). In Section 4.1 I have argued that a construction-based model is ideally suited to representing argument sharing in complex verbs of this type, in that participants of preverbs and verbs can be independently linked to argument slots in the construction. The combination of preverb and verb itself, likewise, can be regarded as a construction (cf. the representation in Figures 2 and 3). This accounts for the syntactic independence of the constituents and for the productivity of the pattern, which allows, among other things, for the integration of recent loans (see Section 2.2). However, there is no doubt that most of the complex verbs in Jaminjung are highly conventionalized expressions (collocations), which are part of the mental storage of the speakers. It has repeatedly been argued that the traditional lexicographer’s lexicon – which includes all conventionalized expressions – is a more appropriate model of analysis than the lexicon which is the repository only of irregular, non-productive combinations (cf. e.g. Pawley and Syder 1983, Pawley 1986, Langacker 1987). Under this conception of the lexicon, separable complex verbs are thus included in the lexicon. This also accounts for their high susceptibility to univerbation diachronically, which may result in complete non-analyzability, as illustrated in Section 3 for some Northern Australian languages. In the approach taken here, this does not have to be described as a radical change of status of these forms (mutating, as it were, from syntactic entities to lexical entities), but rather as gradual further lexicalization, i.e. tightening of already existing collocational links between preverb and verb. As a correlate of lexicalization, a construction characterized by phonological and syntactic independence (i.e. movability) of its constituents develops into a construction with fixed order of constituents, and further into a compounding construction, until, at the final stage of the development, the constructional status is lost because the former combination has become unanalyzable (see Section 3.2). Alternatively, the nature of the construction may change in that one of the constituents may take on grammatical functions, i.e. we are dealing with grammaticalization rather than lexicalization. In Section 3.4 it was shown that in Northern Australian languages this element is more likely to be the verb, whereas in European languages it is more likely to be the preverb, in line with the original size of the class. Again, in a construction-based model this can be represented as a gradual rather than a radical change.
172
E. Schultze-Berndt
The analysis proposed here is equally applicable to preverb-verb combinations in other languages, but the separable complex verbs of Northern Australia with their large, open classes of preverbs provide an even better case for the validity of a dual model. On the one hand, preverbs (at least in those Northern Australian languages with separable complex verbs) cannot only be shown to influence the argument structure and the aktionsart of the complex verb, but they show a larger degree of syntactic and semantic independence than their closed-class counterparts in European languages, since they may also serve as main predicates in certain contexts, and be subject to nominalization. This can be seen as evidence of the syntactic nature of the complex verb construction. On the other hand, the vast majority of predicates in finite, independent clauses are complex, and thus fulfil the function of single verbs in many other languages. In this respect, these preverb-verb combinations can only appropriately be characterized as lexicalized combinations. Thus, although the applicability of the term ‘preverb’ to the open-class uninflecting elements of Northern Australian languages may be a matter of debate, the complex verbs in these languages exhibit enough similarities with separable complex verbs in European languages, both from a synchronic and a diachronic perspective, to warrant the development of a grammatical analysis that is applicable to both types. NOTES * In developing the analyses put forward in this paper, I have profited very much from discussions with Felix Ameka, Ju¨rgen Bohnemeyer, Melissa Bowerman, Penny Brown, James Essegbey, Nikolaus Himmelmann, Dagmar Jung, Silvia Kutscher, Christian Lehmann, Steve Levinson, Bill McGregor, David Wilkins, and Roberto Zavala, as well as the participants of the Nijmegen Workshop on Preverbs, at which a preliminary version of this paper was presented. I am grateful to all of the above, and to Geert Booij, Bernard Comrie and Ans van Kemenade for helpful comments on a draft version of this paper. I would also like to express my gratitude to the Aboriginal Communities in Bulla Camp, Timber Creek, and Mirima (Kununurra), and to the Jaminjung and Ngaliwurru speakers who have been my language teachers and who have contributed to the data discussed in this paper. Financial support for the fieldwork, which is gratefully acknowledged, has come from the Deutscher Akademischer Austauschdienst (DAAD), the Max-Planck-Gesellschaft, the Australian Institute of Aboriginal and Torres Strait Islanders Studies (AIATSIS) and the Deutsche Forschungsgemeinschaft (DFG). This paper was written while working as a research fellow in a research project on the ‘‘Typology of Secondary Predicates’’ at Bochum University, funded by the DFG, and as a postdoctoral fellow at the Max Planck Institute for Evolutionary Anthropology, Leipzig. 1 The following abbreviations are employed in the interlinear glosses: 1, 2, 3 – 1st, 2nd, 3rd person, abl – Ablative, all – Allative, aux – Kriol auxiliary, cond – Conditional marker, cont – Continuous marker, contr – Contrastive focus, dat – Dative, dem – Demonstrative, dir – Directional suffix, du – Dual, emph – Emphatic clitic, erg(/instr) – Ergative(/Instrumental), excl – Exclusive, foc – Kriol focus marker, fut – Future, imp – Imperative, impf – (Past)
Preverbs as an open word class in Northern Australian languages
173
Imperfective, incl – Inclusive, irr – Irrealis, loc – Locative, neg – Negative particle, npast – Nonpast, obl – Oblique pronominal, orig – Origin case, pl – Plural, pot – Potential, prs – Present, past – Past (perfective), rdp – Reduplication, refl – Reflexive/Reciprocal, restr – Restrictive clitic, sg – Singular, subord – General subordinator, top – Topic marker, tr – Kriol transitivity marker. Bound morphemes are separated by a hyphen; clitics are separated by an equal sign (=). Italics indicate Kriol loans in Jaminjung examples. 2 Jaminjung is used here as a cover term for two closely related dialects, Jaminjung and Ngaliwurru. 3 Since many of the languages under consideration are on the verge of extinction, and documentation is generally scarce, the numbers, especially for the languages with larger classes of verbs, should only be regarded as approximations. 4 Adverb is, of course, another potential term for this larger class, but given the semantic range of preverbs/adverbs, this term would be misleading. For a more detailed discussion of the problem of the preverb-adverb boundary, see Schultze-Berndt (2000: 71–73) and Merlan (1994: 59f.). 5 For a detailed account of preverb classes and the semantics of inflecting verbs in Jaminjung see Schultze-Berndt (2000: Chs. 5 and 6). 6 Complex verbs of various types, including serial verb constructions, have also been described for languages outside the Northern Australian linguistic area, for example Diyari (Austin 1981), Ngiyambaa/Wanggaybuwan (Donaldson 1980), and Western Desert languages such as Yankunytjatjara (Goddard 1985). These will be left out of consideration here. 7 Throughout this section, orthography and glosses for the languages cited are adapted, for ease of comparison. 8 For detailed arguments for regarding both case-marked noun phrases and bound pronominals as morphosyntactic arguments, see Schultze-Berndt (2000: Ch. 4).
REFERENCES Ackerman, Farrell and Philip LeSourd (1997). Towards a lexical representation of phrasal predicates. In Alex Alsina, Joan Bresnan and Peter Sells (eds.), Complex Predicates. Stanford CA: CSLI Publications, 67–106. Alpher, Barry (1994). Yir-Yoront Ideophones. In Leanne Hinton, Johanna Nichols and John Ohala (eds.), Sound Symbolism. Cambridge UK: Cambridge University Press, 161–177. Andrews, Avery D. and Christopher D. Manning (1999). Complex predicates and information spreading in LFG. Stanford CA: CSLI Publications. Austin, Peter (1981). A Grammar of Diyari, South Australia. Cambridge UK: Cambridge University Press. Bhatt, Christa and Markus Lindlar (eds.) (1998). Alles Ko¨lsch. Eine Dokumentation der aktuellen Stadtsprache in Ko¨ln. Bonn: Bouvier. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad and Edward Finegan (1999). Longman Grammar of Spoken and Written English. Harlow: Pearson Education Ltd. Blake, Barry J. (1987). Australian Aboriginal Grammar. London: Croom Helm. Bolt, Janet E., William G. Hoddinott and Frances M. Kofod (1971). An Elementary Grammar of the Nungali language of the Northern Territory. Unpublished Ms., Canberra: Australian Institute of Aboriginal and Torres Strait Islanders Studies.
174
E. Schultze-Berndt
Booij, Geert (2002). The Morphology of Dutch. Oxford: Oxford University Press. Butt, Miriam (1997). Complex predicates in Urdu. In Alex Alsina, Joan Bresnan and Peter Sells (eds.), Complex Predicates. Stanford CA: CSLI Publications, 107–149. Bybee, Joan L. (1985). Morphology. A Study of the Relation between Meaning and Form. Amsterdam: Benjamins. Capell, Arthur (1976). Simple and compound verbs: conjugation by auxiliaries in Australian verbal systems. Rapporteur’s introduction and summary. In R.M.W. Dixon (ed.), Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, 615–625. Capell, Arthur (1979). Classification of verbs in Australian languages. In Stephen A. Wurm (ed.), Australian Linguistic Studies. (=Pacific Linguistics C-54). Canberra: Research School of Pacific and Asian Studies, 229–322. Carr, The´re`se M. (2000). Wunambal: A language of the North-West Kimberley Region, Western Australia. Unpublished Masters (Honours) Thesis, University of New England, Armidale. Chadwick, Neil (1997). The Barkly and Jaminjungan languages: a non-contiguous genetic grouping. In Darrell T. Tryon and Michael Walsh (eds.), Boundary Rider: Studies in the Lexicology and Comparative Linguistics of Australian Languages. (=Pacific Linguistics C-136). Canberra: Research School of Pacific and Asian Studies, 95–106. Cook, Anthony (1986). Participle sentences in Wakiman. In Peter Austin (ed.), Complex Sentence Constructions in Australian Languages. Amsterdam: Benjamins, 69–95. Dixon, R.M.W. (1980). The Languages of Australia. Cambridge UK: Cambridge University Press. Dixon, R.M.W. (2001). The Australian linguistic area. In Alexandra Y. Aikhenvald and R.M.W. Dixon (eds.), Areal Diffusion and Genetic Inheritance. Problems in Comparative Linguistics. Oxford: Oxford University Press, 64–104. Donaldson, Tamsin (1980). Ngiyambaa. The Language of the Wangaaybuwan. Cambridge UK: Cambridge University Press. Durie, Mark (1997). Grammatical structures in verb serialization. In Alex Alsina, Joan Bresnan and Peter Sells (eds.), Complex Predicates. Stanford CA: CSLI Publications, 289–354. Evans, Nicholas (in press). Bininj Gun-Wok: A Pan-dialectal Grammar of Mayali, Kunwinjku and Kune. Canberra: Research School of Pacific and Asian Studies. Foley, William and Mike Olson (1985). Clausehood and verb serialization. In Johanna Nichols and Anthony C. Woodbury (eds.), Grammar Inside and Outside the Clause. Some Approaches to Theory from the Field. Cambridge UK: Cambridge University Press, 17–60. Goddard, Cliff (1985). A Grammar of Yankunytjatjara. Alice Springs: Institute for Aboriginal Development. Goldberg, Adele (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Goldberg, Adele (1996). Words by default: Optimizing constraints and the Persian complex predicate. Berkeley Linguistic Society 22, 132–146. Grimshaw, Janet and A. Mester (1988). Light verbs and Theta marking. Linguistic Inquiry 19(2), 205–232. Hampe, Beate (1997). Towards a solution of the phrasal verb puzzle: Considerations on some scattered pieces. Lexicology 3(2), 203–243.
Preverbs as an open word class in Northern Australian languages
175
Harris, John W. (1991). Kriol – the creation of a new language. In Suzanne Romaine (ed.), Language in Australia. Cambridge UK: Cambridge University Press, 195–203. Harvey, Mark and Eva Schultze-Berndt (in prep.) A Grammar of Nungali. Heath, Jeffrey (1976). Topic E: Simple and compound verbs: conjugation by auxiliaries in Australian verbal systems. North-east Arnhem Land. In R.M.W. Dixon (ed.), Grammatical categories in Australian languages. Canberra: Australian Institute of Aboriginal Studies, 735–740. Heath, Jeffrey (1980). Nunggubuyu Myths and Ethnographic Texts. Canberra: Australian Institute of Aboriginal Studies. Heath, Jeffrey (1984). Functional Grammar of Nunggubuyu. Canberra: Australian Institute of Aboriginal Studies. Hinton, Leanne, Johanna Nichols and John Ohala (eds.) (1994). Sound Symbolism. Cambridge UK: Cambridge University Press. Hoddinott, William G. and Frances M. Kofod (1976). Topic E: Simple and compound verbs: conjugation by auxiliaries in Australian verbal systems. Djamindjungan. In R.M.W. Dixon (ed.), Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, 698–704. Hoddinott, William G. and Frances M. Kofod (1988). The Ngankikurungkurr Language (Daly River area, Northern Territory). (=Pacific Linguistics D-77). Canberra: Research School of Pacific and Asian Studies. Koch, Harold (1984). The category of ‘associated motion’ in Kaytej. Language in Central Australia 1, 23–34. Kofod, Frances M. (1996a). Verb Stems in Jarragan Languages of the East Kimberley. Paper presented at the Annual Meeting of the Australian Linguistic Society, Canberra 1996. Kofod, Frances M. (1996b). Kija Learner’s Grammar. Unpublished Ms, Halls Creek, W.A.: Kimberley Language Resource Centre. Kofod, Frances M. (1997). Gender and Number in Jarragan languages. Paper presented at the Top End Linguistics Circle Meeting, Darwin, 24 April 1997. Langacker, Ronald W. (1987). Foundations of Cognitive Grammar. Vol. I. Theoretical Prerequisites. Stanford CA: Stanford University Press. Lehmann, Christian (1983). Latin preverbs and cases. In Harm Pinkster (ed.), Latin Linguistics and Linguistic Theory. Proceedings of the 1st International Colloquium on Latin Linguistics, Amsterdam, April 1981. Amsterdam: Benjamins, 145–161. Lehmann, Christian (1995). Thoughts on Grammaticalization. Mu¨nchen: Lincom Europa. Matsumoto, Yo (1996). Complex Predicates in Japanese. Stanford CA and Tokyo: CSLI Publications and Kurosio Publishers. McConvell, Patrick (to appear). Gurindji Grammar. In Barry Blake and R.M.W. Dixon (eds.), Handbook of Australian Languages VI. McConvell, Patrick and Eva Schultze-Berndt (2001). Complex verb convergence and bilingual interaction in the Victoria River District, Australia. Paper presented at the Workshop on Language contact and areal convergence, DGfS Meeting, Leipzig, 1 March 2001. McGregor, William B. (1990). A Functional Grammar of Gooniyandi. Amsterdam: Benjamins. McGregor, William B. (1993). Gunin/Kwini. Mu¨nchen: Lincom Europa. McGregor, William B. (1994). Warrwa. Mu¨nchen: Lincom Europa. McGregor, William B. (1996a). Nyulnyul. Mu¨nchen: Lincom Europa. McGregor, William B. (1996b). Sound symbolism in Gooniyandi, a language of Western Australia. Word 47(3), 339–364.
176
E. Schultze-Berndt
McGregor, William B. (2001). Ideophones as the source of verbs in Northern Australian languages. In F.K. Erhard Voeltz and Christa Kilian-Hatz (eds.), Ideophones. Amsterdam: Benjamins, 205–221. McGregor, William B. (2002). Verb Classification in Australian Languages. Berlin: Mouton de Gruyter. Merlan, Francesca C. (1982). Mangarayi. Amsterdam: North Holland. Merlan, Francesca C. (1994). A Grammar of Wardaman. A language of the Northern Territory of Australia. Berlin: Mouton de Gruyter. Metcalfe, C.D. (1975). Bardi Verb Morphology. (=Pacific Linguistics B-30). Canberra: Research School of Pacific and Asian Studies. Mohanan, Tara (1994). Argument Structure in Hindi. Stanford CA: CSLI Publications. Mohanan, Tara (1997). Multidimensionality of Representation: Noun-Verb Complex Predicates in Hindi. In Alex Alsina, Joan Bresnan and Peter Sells (eds.), Complex Predicates. Stanford CA: CSLI Publications, 431–472. Myers-Scotton, C. and J. Jake (1995). Matching lemmas in a bilingual competence and production model. Linguistics 33, 981–1024. Nash, David (1982). Warlpiri verb roots and preverbs. In Stephen Swartz (ed.), Papers in Warlpiri Grammar, in Memory of Lothar Jagst. (=Work Papers of SIL-AAB, Series A, Vol. 6). Berrimah, N.T.: Summer Institute of Linguistics, 165–216. Nash, David (1986). Topics in Warlpiri Grammar. New York: Garland. Nordlinger, Rachel (1990). Bilinara. Unpublished Honours Thesis, University of Melbourne. Nordlinger, Rachel (1998). A Grammar of Wambaya, Northern Territory (Australia). (= Pacific Linguistics C-140). Canberra: Research School of Pacific and Asian Studies. Pawley, Andrew (1986). Lexicalization. In Deborah Tannen and James Alatis (eds.), Georgetown University Round Table on Language and Linguistics. Washington DC: Georgetown University Press, 98–120. Pawley, Andrew and Frances H. Syder (1983). Two puzzles for linguistic theory: nativelike selection and nativelike fluency. In Jack C. Richards and Richard W. Schmidt (eds.), Language and Communication. London: Longman, 191–225. Reid, Nicholas J. (1990). Ngan’gityemerri. A language of the Daly River region, Northern Territory of Australia. Unpublished PhD Thesis, Canberra: Australian National University. Reid, Nicholas J. (to appear). Phrasal verb to synthetic verb: recorded morphosyntactic change in Ngan’gityemerri. In Nick Evans (ed.), Studies in Comparative Non-Pama-Nyungan Languages. Canberra: Research School of Pacific and Asian Studies. Romaine, Suzanne (1989). Bilingualism. Oxford: Blackwell. Rumsey, Alan (1982). An Intra-sentence Grammar of Ungarinjin, North-Western Australia. (= Pacific Linguistics B-86). Canberra: Research School of Pacific and Asian Studies. Schultze-Berndt, Eva (2000). Simple and Complex Verbs in Jaminjung. A study of Event Categorization in an Australian Language. PhD thesis, Katholieke Universiteit Nijmegen. Schultze-Berndt, Eva (2001). Ideophone-like characteristics of uninflecting predicates in Jaminjung (Australia). In F. K. Erhard Voeltz and Christa Kilian-Hatz (eds.), Ideophones. Amsterdam: Benjamins, 355–373. Schultze-Berndt, Eva (to appear) Making sense of complex verbs: on the semantics and argument structure of closed-class verbs and coverbs in Jaminjung. In Melissa Bowerman and Penelope Brown (eds.), Cross-linguistic Perspectives on Argument Structure. Implications for Learnability. Hillsdale NJ: Lawrence Erlbaum.
Preverbs as an open word class in Northern Australian languages
177
Shibatani, Masayoshi (1996). Applicatives and benefactives: A cognitive account. In Masayoshi Shibatani and Sandra A. Thompson (eds.), Grammatical Constructions. Their Form and Meaning. Oxford: Clarendon Press, 157–194. Silverstein, Michael (1986). Classifiers, verb classifiers, and verbal categories. Berkeley Linguistic Society 12, 497–514. Simpson, Jane (1991). Warlpiri Morpho-Syntax. A Lexicalist Approach. Dordrecht: Kluwer. Stokes, Bronwyn (1996). The top then Nyulnyulan verb roots: further evidence for language classification. In William B. McGregor (ed.), Studies in Kimberley Languages in Honour of Howard Coate. Mu¨nchen: Lincom Europa, 175–188. Tryon, Darrell T. (1974). Daly Family Languages, Australia. (=Pacific Linguistics C-32). Canberra: Research School of Pacific and Asian Studies. Tsunoda, Tasaku (1981). The Djaru Language of Kimberley, W.A. (=Pacific Linguistics B-78). Canberra: Research School of Pacific and Asian Studies. Va´szolyi, Eric (1976). Simple and compound verbs: conjugation by auxiliaries in Australian verbal systems. Wunambal. In R.M.W. Dixon (ed.), Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies, 629–646. Vincent, Nigel (1999). The evolution of c-structure: prepositions and PPs from Indo-European to Romance. Linguistics 37, 1111–1153. Voeltz, F.K. Erhard and Christa Kilian-Hatz (eds.) (2001). Ideophones. Amsterdam: Benjamins. Walsh, Michael (1996). Vouns and nerbs: a category squish in Murrinh-Patha (Northern Australian). In William B. McGregor (ed.), Studies in Kimberley Languages in Honour of Howard Coate. Mu¨nchen: Lincom Europa, 227–252. Wilkins, David P. (1997). The verbalization of motion events in Arrernte (Central Australia). In Eve Clark (ed.), The Proceedings of the Twenty-Eighth Annual Child Language Research Forum. Stanford CA: CSLI Publications, 295–308. Wilson, Stephen (1999). Coverbs and Complex Predicates in Wagiman. Stanford CA: CSLI Publications.
Max Planck Institute for Evolutionary Anthropology Inselstr. 22 D-04103 Leipzig e-mail: [email protected]
This page intentionally left blank
Moved preverbs in German: Displaced or misplaced? JOCHEN ZELLER
1. INTRODUCTION* A crucial question that arises in the study of preverb-verb constructions across languages concerns the way the preverb is combined with the verb. One can pose this question from a semantic point of view: Are the preverb and the verb semantically independent or is the preverb-verb construction lexicalized? If the meaning of the preverb can be defined in isolation, how is it combined with the meaning of the verb – is the preverb a semantic argument of the verb, or an aspectual operator etc.? Another line of investigation would be to focus on the structural side of the construction, for example by asking if the combination of preverb and verb is formed in morphology or if it has the status of a phrasal syntactic construction. A particular class of preverb-verb combinations in German (the class of so-called particle verbs) has created a debate with respect to the latter question. The main reason for this controversy is that although particle verbs show a lot of properties typical of complex words formed in morphology, they allow for the preverb (the particle) and the verb to be separated. As is well known and undisputed, the preverb and the verb can be split by moving the verb away from the preverb. A more controversial question is whether separation may also be the result of moving the particle. It is this latter type of movement with which I am concerned in this paper. In the following sections I present and discuss a number of examples of particle- and PP- movement.1 Speakers’ judgements about the possibility of particle movement are often quite delicate and show a great degree of ideolectal variation. I therefore collected judgements on these examples from 16 German mother tongue speakers; the method by which I evaluated these judgements is discussed in section 2. In section 3, I address the issue of separability in the study of particle verbs. I then turn to the various instances of particle movement. In section 4, I re-examine some of the data that have been presented in the literature supporting the view that particles can be topicalized, i.e. moved to SpecCP. I show that neither a purely structural nor a purely semantic approach can fully explain the properties of particle topicalization. I then focus on two types of particle movement that to my knowledge have not yet been sufficiently studied, namely long particle topicalization (=movement from an embedded clause into SpecCP of the main clause), which is discussed in section 5, and particle scrambling (=adjunction to IP or VP), which is the topic of section 6. Finally, in section 7, I discuss the problems that arise with respect to movement of particles and offer possible explanations for some of the observations made in previous sections. I intend to show that some of the characteristics of particle Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 179–212. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
180
J. Zeller
movement are the result of the ambiguous properties of particle verbs, whereas others follow from independent principles of grammar.
2. METHOD Most examples of particle movement that can be found in the literature are examples of ‘‘short’’ particle topicalization (i.e particle fronting within the clause in which the particle verb originates). Since the judgements associated with these examples are notoriously inconsistent and are often based on the ideolect of only one or two speakers, my first objective was to test the reliability of the existing data. Some of the examples which I asked my respondents to evaluate were therefore identical to or based on particle topicalization-examples from the literature. My second objective was to present new data which illustrate two different kinds of particle movement, namely long particle topicalization and particle scrambling. I also collected judgements on these instances of particle movement. My third objective was to compare constructions with moved particles to constructions in which a full PP has moved; I therefore also asked my informants for judgements on this latter type of movement. Most of the data which I present in sections 4–6 were collected on the basis of the judgements of 16 German mother tongue speakers who were all linguists.2 All respondents were presented informally (mostly electronically) with identical, typed examples in the same order, which they were required to judge. Anticipating ambiguous responses to most of the examples, I allowed for graded judgements. All informants therefore were instructed to mark examples not only as grammatical or ungrammatical, but also by using symbols like ? or ?* to indicate marginal acceptance. The examples were designed to test whether respondents detected asymmetries between the three different types of particle movement (i.e. particle scrambling; long and short particle topicalization) and between particle movement on the one hand and PP-movement on the other. Speakers were allowed to use Topic/Focus intonation in evaluating the data. For some examples, I provided an additional sentence which provides a contrastive context and hence facilitates the respective Topic/Focus interpretation. Where this was done, these contrastive sentences are provided in parentheses with the examples in the text. In the examples in the text I mark particles and PPs in bold for ease of exposition, but in the sentences which respondents were asked to judge, these constituents were unmarked. I assigned points to the judgements of the respondents (ok=1, ?=2, ??= 3, ?* (or ???)=4, *=5), then calculated the average number for each example, and retranslated this number into a judgement. For example, an average of 1.8 would correspond to?, 4.5 to *? etc. A concern with this method, discussed in
Moved preverbs in German: displaced or misplaced?
181
greater detail in section 5, is that the calculated ‘‘average response’’ may mask a wide distribution of individual judgements. I attached the distribution of responses to the examples where they are discussed in the text.
3. SEPARATING PREVERB AND VERB In German, particle verbs are obligatorily separated if the verb moves to the sentence initial position (which I assume to be C0 ). As (1b) shows, the particle is stranded when the verb moves (particles are glossed as ‘‘Prt’’): (1) a.
(Er sagt) dass er uns ein Bier ausgibt he says that he us a beer Prt-gives ‘(He is saying) that he is going to buy us a beer.’
b. Er gibt uns ein Bier aus __ he gives us a beer Prt
The kind of separation illustrated in (2) is not what one expects from a morphologically derived complex verb. In the light of data like (1b), particle verbs seem rather to look like syntactic constructions; if the particle is analyzed as a phrasal complement of the verb, separation under verb movement no longer comes as a surprise. However, the property of particle verbs depicted in (2) would only provide conclusive evidence against a morphological analysis if there was an independent principle that forbids the movement of part of a word.3 A number of authors have rejected the existence of such a principle. For example, Neeleman and Weerman (1993) and Neeleman (1994) allow for a word to be split by movement of one of its parts as long as the moved part is the head of this word. If the particle verb is analyzed as a word, then its verbal part counts as its head, and consequently, it is allowed to move, leaving the nonhead (the particle) behind. There is also empirical evidence suggesting that parts of words can be moved. McIntyre (2001) shows that in German, not only particle verbs, but also certain prefix verbs can be split if the verb moves to C0 : (3) a. b.
??Peter u¨berbewertet die Auseinandersetzung P. Pref-Pref-estimates the argument ?Peter bewertet die Auseinandersetzung u¨berP. Pref-estimates the argument Pref ‘Peter overestimates the argument.’
182
J. Zeller
The verb u¨berbewerten is derived from the base verb bewerten and the prepositional element u¨ber-, which is a prefix (shown by the fact that it can also move with the verb to C0 in (3a), an option unavailable for particles). Prefixes are usually unstressed in German. However, since the first syllable of the base verb in (3) is unstressed as well (bewerten is also a prefix verb), u¨ber- is stressed in order to avoid a succession of unstressed syllables. This situation allows for the separation of the prefix verb in (3b), a genuine morphological object. Therefore, given that the splitting of morphologically complex words does not seem to be generally excluded, and in the light of the fact that empirical evidence for a morphological analysis of particle verbs can also be found (see e.g. Neeleman 1994, Stiebels 1996 and Booij and van Kemenade (this volume)), it is at least controversial that examples like (1b) constitute conclusive evidence in favor of a syntactic analysis of particle verbs. More insights into the structural properties of particle verbs might be gained if another question is explored: Is it possible to separate the particle verb by moving away not the verb, but the particle? Here, the two competing approaches to particle verbs make different predictions. The syntactic approach treats the particle as a phrasal complement of the verb. As such, it is expected that it can move away from the verb just like other phrasal complements. For example, German is a verb second (V2) language which allows phrases of any category to undergo topicalization (=movement to SpecCP, a position preceding the finite verb in C0. The syntactic approach therefore predicts that particles can be fronted. In contrast, according to the morphological analysis, movement of the preverb (the nonhead of the particle verb) should be impossible. Therefore, proponents of a morphological approach often reject the frontability of particles (cf. e.g. Neeleman and Weerman 1993; Stiebels and Wunderlich 1994; Neeleman 1994; Haider, Olsen and Vikner 1995; Stiebels 1996; Olsen 1997; Zifonun 1999) and present examples like (4) that challenge the syntactic approach: (4) a.
*Aus gibt er uns ein Bier Prt gives he us a beer ‘He buys us a beer.’
b. *Auf hat er ein Gedicht gesagt Prt has he a poem said ‘He recited a poem.’
Moved preverbs in German: displaced or misplaced?
183
In (4), particle topicalization is impossible. Numerous other examples, where separating the particle verb through movement of the particle yields ungrammaticality, can be found and have been listed in the literature. If one could safely conclude from these examples that particle movement in general is excluded, they would provide a strong argument against a syntactic analysis. However, the view that particles cannot be moved has also been challenged, and it has been shown that, in contrast to examples like (4), some particles can in fact undergo phrasal movement like other XPs. In the next section I take a closer look at some of the data that illustrate this possibility.
4. PARTICLE TOPICALIZATION There are certain criteria that are often identified in the literature as necessary conditions for particle fronting. It is claimed (cf. Lu¨deling 1998; Wurmbrand 2000; Zeller 2001) that particle movement is contingent on the particle verb having a transparent semantics and the possibility of a contrastive reading of the particle (i.e. there must be at least one particle verb derived from the same base verb, but with a different particle). These conditions are based on the assumption that movement to SpecCP automatically triggers a Topic – or Focus reading of a topicalized (non-subject) XP (cf. Bu¨ring 1996: 55; Wurmbrand 2000: 18). The particle verbs in the following examples all meet these criteria: All particle verbs in (5)–(10) have a transparent semantics, and, as is sometimes indicated by the clauses in brackets, every particle verb can be contrasted with at least one other particle verb based on the same base verb. (The data are adopted from or based on similar examples from the literature, as indicated):4 (5) Auf geht die Sonne im Osten (aber unter geht sie im Prt goes the sun in.the east (but Prt goes she in.the Westen). west ‘The sun rises in the east but it sets in the west.’ (aufgehen, ‘rise’, vs. untergehen, ‘set’) (Lu¨deling 1998: 57) (6) (Die hu¨bschen Frauen stiegen alle ein). Aus stiegen eigentlich nur the pretty women climbed all Prt Prt climbed actually only Ma¨nner. men ‘The pretty women all got in. It was only men who got off.’ (einsteigen, ‘get in’, vs. aussteigen, ‘get off’) (cf. Zeller 2001)
184
J. Zeller
(7) ?(Die meisten Leute reisen aus). Ein ist dieses Jahr noch the most people travelled Prt Prt is this year still niemand gereist. nobody travelled ‘Most people left the country. Nobody has entered (the country) this year.’ (ausreisen, ‘leave (a country)’, vs. einreisen, ‘enter (a country)’) (cf. Zeller 2002) (8) ?(Angola fu¨hrt viele Waren ein.) Aus fu¨hrt das Land nur Angola takes many goods Prt Prt takes the country only Kaffee. coffee ‘Angola imports a lot of goods. The country exports only coffee.’ (ausfu¨hren, ‘export’, vs. einfu¨hren, ‘import’) (cf. Hoeksema 1991: 19, for the same example from Dutch) (9) ?Auf laden die Ma¨nner das Heu. Prt load the men the hay ‘The men load up the hay.’ (‘aufladen, ‘load up’, vs. abladen, ‘load off’, einladen, ‘load in’) (cf. Zeller 2002) (10) ??(Lachst du mich aus?) Nein, an lache ich dich. laugh you me Prt No Prt laugh I you ‘Are you laughing at me? No, I’m smiling at you.’ (anlachen, ‘smile at’, vs. auslachen, ‘laugh at’) (Lu¨deling 1998: 57) Table 1 shows the individual judgements for (5)–(10): Sentence Number (5) (6) (7) (8) (9) (10)
Ok (=1) 13 12 8 7 6 5
? (=2) 2 1 4 3 5 3
?? (=3) 1 3 2 2 2 0
?* (=4) 0 0 0 1 0 2
* (=5) 0 0 2 3 3 6
Average 1.3 1.4 2.0 2.4 2.3 3.1
Table 1. Particle topicalization I.
The examples in (5)–(10) are acceptable (although some only marginally). This acceptability has an important consequence for the structural analysis of particle verbs. It means that, at least in these specific cases, particles are phrases. This provides strong support for the syntactic approach, which treats the particle as
Moved preverbs in German: displaced or misplaced?
185
a complement of the verb and therefore analyzes the verb-particle combination essentially as a syntactically complex category (a V∞ or a VP respectively). What does the morphological analysis have to say about these examples? There are two possibilities: It could be claimed that only those particles that can be topicalized are phrases (proposals along these lines can be found in the literature, cf. Le Roux (1988) for Afrikaans, Kratzer (1994) and Stiebels (1996) for German and Neeleman and Weerman (1993) for Dutch). According to this approach, if a particle can be fronted, the respective particle verb receives a syntactic analysis, whereas all other particle verbs are still treated as morphological objects. This, however, is an unfortunate conclusion, since it denies the possibility of giving a uniform account for all particles. Furthermore, the judgements about topicalization of particles do not show the kind of distribution that one expects from a structural difference. If there were really some syntactic particle verbs and some morphological ones, then we would expect that speakers draw a clear line between those particles that can and those that cannot be moved. But this is not what one finds: As Table 1 shows in detail, speakers’ judgements about particle topicalization vary considerably. This clearly suggests that the reasons that determine whether or not a particle can be topicalized cannot be exclusively structural. Alternatively, if the assumption that particle verbs are words was to be maintained in the light of examples like (5)–(10), its proponents would have to assume that (i) these words include phrases and (ii) that these phrases can in fact be moved out of these words. Point (ii), however, is an ad hoc stipulation that lacks independent evidence. Words that include phrases do exist (cf. Toman 1983; Lu¨deling 1998), but moving the phrasal non-heads out of these word is absolutely impossible: (11) a.
Die Wer-war-das-Frage ‘the who was it question’ b. Inspektor Morse beantwortete die Wer-war-das-Frage. Inspector Morse answered the who was it question c. [Die Wer-war-das Frage] beantwortete Inspektor Morse t d. *[Wer war das] beantwortete Inspektor Morse die [t Frage]
(12) a.
Die Schmeckt-gut-macht-Laune-Limo the tastes-well-makes-fun-lemonade b. Nur er kauft Schmeckt-gut-macht-Laune-Limo. only he buys tastes-well-makes-fun-lemonade c. [Schmeckt-gut-macht-Laune-Limo] kauft nur er t d. *[Schmeckt-gut-macht-Laune] kauft nur er [t -Limo]
As (11) and (12) show, the phrasal non-head of a phrasal compound is syntactically invisible and cannot undergo XP-movement out of a word. I cannot see how one would explain why this general restriction does not hold for the phrasal
186
J. Zeller
preverb-part of a particle verb. Since the particles in (5)–(10) are phrases that can be moved away from the verb, I conclude that the respective particle verbs are syntactic constructions. I henceforth assume that the particle verbs in (5)–(10) are represented as V∞- or a VP-nodes that dominate the verb and the particle phrase. How does this assumption explain why particle topicalization is not possible with all particle verbs? As noted above, proponents of a syntactic approach have assumed that particle topicalization is only possible if the particle verb is semantically transparent and if there are possible alternatives to the topicalized particle. Consequently, it has been suggested that impossible cases like those in (4) in section 3 can be excluded on semantic grounds alone. To see how this works, let me take a closer look at the semantics of the examples in (5)–(10). Rooth (1985) assumes that each node in a syntactic tree, apart from its ordinary meaning, denotes a second semantic object, its Focus value. The Focus value of a sentence is a set of propositions which count as contextually plausible alternatives to the proposition expressed by the sentence. This set of alternatives is derived by combining the Background of the sentence (=everything which is not the Focus) with each element of the Focus value of the element marked as the Focus. In (10) and (presumably) in (9) above, the particle is the Focus of the sentence. This means that the Focus value of (10) (=(13)) is derived on the basis of possible alternatives to the particle an. Since there are not many possible alternatives to an in (10), the Focus value of (13) is rather small. It is informally given in (14) (the Focus accent is indicated by capitals):5 (13) [AN]F lache ich dich. Prt laugh I you ‘I’m smiling at you.’ (14) {Ich lache dich aus, Ich lache dich an} ‘I’m laughing at you, I’m smiling at you’ The Focus value in (14) shows that there is at least one proposition which is a possible alternative to the proposition expressed by the sentence in (13). In the context of this proposition, (13), with the particle as the Focus, is acceptable. The topicalized particles in (5)–(8) are what Bu¨ring (1996) calls Topics, not Foci. (For reasons that are not clear to me, particle topicalization is slightly better in Topic-Focus- than in Focus constructions.) Topic-Focus constructions are characterized by a typical rise-fall accent, with a rising pitch accent on the Topic and a falling pitch accent on the Focus, as shown in (15): (15) /[AUF]T geht die Sonne im [OSTEN]F c In Bu¨ring’s (1996) analysis, the meaning of a Topic-Focus construction is determined by yet another semantic level, its Topic value. Whereas the Focus of a
Moved preverbs in German: displaced or misplaced?
187
sentence introduces alternative propositions (its Focus value), the Topic of the sentence introduces alternative Focus values. This means that the Topic value of a sentence is a set of sets of propositions. According to this approach, the Topic value of (15) is (16): (16) { {Die Sonne geht im Osten auf, Die Sonne geht im Norden auf, Die Sonne geht im Westen auf, ...}, {Die Sonne geht im Osten unter, Die Sonne geht im Norden unter, Die Sonne geht im Westen unter, ...} ...} { {The sun rises in the east, The sun rises in the north, The sun rises in the west, ...}, {The sun sets in the east, The sun sets in the north, The sun sets in the west, ...} ...} In order to derive the Topic value of (15), the Background (=the sentence minus Focus and Topic) first has to combine with each alternative defined by the Focus, and each outcome then has to combine with the alternatives defined by the Topic. The Focus in (15) is the NP Osten, hence it is alternatives to this NP which define the direct object-part of the Focus values in (16). The Topic is the particle auf, and the two alternative Focus values listed in (16) are therefore also determined on the basis of the Topic value of this element. The semantics of Topic/Focus constructions explains why only semantically transparent particle verbs with contrastable particles can be separated through particle movement. Only if there are possible alternatives to the particle can a Focus- or Topic value be defined; only semantically transparent particle verbs guarantee that the combination of the Background and an alternative particle yields a meaningful proposition (which becomes an element of the Focus- or Topic value of the sentence). However, the particle verbs in (4) in section 3 are idiomatic; their meanings cannot be derived from a particular meaning of the particle and the meaning of the verb. Furthermore, there are no particles with which the particles aus and auf in (4) could be contrasted. Therefore, it seems that impossible cases of particle topicalization can all be explained by a semantic approach: idiomatic particle verbs simply do not meet the semantic requirements of Topic/Focus constructions; hence their particle-parts are predicted to be immobile. However, there are problems with a purely semantic approach. First, notice that it is possible in German that a moved verbal complement can establish VP-Focus (cf. Krifka 1994): (17) Question: Was hat er gemacht? ‘What did he do?’ ¨ CHE] hat er [t gestrichen] Answer: [Die KU the kitchen has he painted He painted the kitchen.’
188
J. Zeller
In (17), the accent-bearing direct object has been topicalized, but nevertheless, the whole VP bears Focus. Bu¨ring (1996: 64) adds the idiomatic example in (18) (see also Lu¨deling 1998; Wurmbrand 2000; Mu¨ller 2002): (18) [Den GARAUS] hat man ihm NICHT [t gemacht] the G. has one him not made ‘They didn’t finish him off.’ In (18), the NP den Garaus has been topicalized, but, as noted by Bu¨ring, the Topic in (18) is the whole VP. Crucially, the topicalized NP does not even have a meaning in itself; it is only possible as part of the idiom jemandem den Garaus machen, ‘finish someone off’. (17) and (18) pose a problem for the abovementioned claim that impossible instances of particle movement can be excluded semantically. Examples like (18) predict that topicalization of idiomatic particle verbs should be possible, with VP-Topic or VP-Focus established by the topicalized particle. However, acceptable examples that can be analyzed along these lines are rare. For example, the idiomatic particle verbs in (4) mentioned in section 3 do not allow a VP-Focus reading to be established by fronting the particle. In the light of the sentences in (17) and (18), it is unclear how a semantic account would explain this. It has to be noted, however, that some examples where a topicalized particle is neither the Focus nor the Topic have been listed in the literature, and I sought to corroborate their acceptability by asking my informants for their judgements. The sentences in (19)–(23) are adopted from Mu¨ller’s (2002: 255f.) vast collection of corpora data (see Wurmbrand 2000 and Zeller 2001 for presentation and discussion of more examples):6 (19) ?*Auf tritt im blauen Anzug der Ko¨nig. Prt steps in.the blue suit the king ‘The king appears in the blue suit.’ (20) ?*Auf schrie die Zieharmonika Prt screamed the accordion ‘The accordion shrieked.’ (21) ??Es klopfte. Ein trat der Studienrat. it knocked Prt stepped the teacher ‘There was a knock on the door. The teacher came in.’ (22) ?Los ging es schon in dieser Woche. Prt went it already in this week ‘It already started this week.’
Moved preverbs in German: displaced or misplaced?
189
(23) ?Vor hat er das jedenfalls. Prt had he it anyway ‘But he does plan this.’ Sentence Number (19) (20) (21) (22) (23)
Ok (=1) 0 3 4 10 11
? (=2) 5 2 2 3 1
?? (=3) 0 0 3 2 4
?* (=4) 2 2 2 0 0
* (=5) 9 9 5 1 0
Average 3.9 3.8 3.1 1.7 1.6
Table 2. Particle topicalization II.
The judgement’s of my informants are shown in Table 2: Taken out of their original context, not all examples were accepted by my informants. For example, as shown in Table 2, (19) and (20) were possible only for a small minority. It seems that the particle fronting in (19)–(21) does not have the typical semantic effect associated with topicalization; if these data are acceptable at all, then this is only in their original stylistic or poetic contexts. The majority of speakers judged the fronting of the particles in (22) and (23) as permissible. In these examples, the whole particle verb or VP is the Topic or the Focus, and the sentences can probably be analyzed on a par with (17) and (18). However, as Mu¨ller (2002: 258) notes, examples like (19)–(23) are not very frequent. If we wanted to treat particle topicalization semantically rather than structurally, we would predict that examples similar to (22) and (23) occur more often; in the light of (17) and (18), they should be the rule rather than the exception. Notwithstanding the acceptability of (22) and (23), the question why VP-Focus cannot be established by particle topicalization in general remains open. A second problem for a semantic explanation is raised by the observation that the data in (5)–(10) above, although grammatical, are not perfectly acceptable. Importantly, similar examples with full PPs being topicalized are definitely better than the ones with topicalized particles. The following sentences include the same verbs that appear as base verbs in the examples (5)–(10), but instead of particles, full PPs have been moved to SpecCP (the PP in (25) is an adjunct; the other PPs are optional or obligatory arguments of their verbs): (24) Auf die Party geht Hans mit Maria (aber in den Zoo mit Usch). on the party goes H. with M. but in the zoo with U. ‘Hans goes to the party with Maria but to the zoo with Usch.’ ¨ ber dich lache ich. (25) U about you laugh I ‘I laugh about you.’
190
J. Zeller
(26) Aus dem Zug stiegen eigentlich nur Ma¨nner. out the train climbed actually only men ‘Only men got off the train.’ (27) Aus dieser Stadt fu¨hrte der Rattenfa¨nger gestern alle Kinder. out this city led the pied-piper yesterday all children ‘Yesterday, the pied-piper led all children out of this city.’ (28) Auf diesen Wagen laden die Ma¨nner das Heu. on this wagon load the men the hay ‘The men are loading the hay onto this wagon.’ (29) In dieses Land ist dieses Jahr noch niemand gereist. in this country is this year still nobody travelled ‘Nobody has entered this country this year.’ As the judgements in Table 3 show, speakers accept sentences with topicalized PPs without exception. Although particle topicalization in the related examples (5)–(10) was also tolerated, the contrast is notable: PP-topicalization is unequivocally fine, particle topicalization is only marginally acceptable. So far, an account based on the semantics of Topic and Focus does not explain why an example like e.g. (10) is considerably worse than (25). Sentence Number (24) (25) (26) (27) (28) (29)
Ok (=1) 16 16 16 15 16 16
? (=2) 0 0 0 1 0 0
?? (=3) 0 0 0 0 0 0
?* (=4) 0 0 0 0 0 0
* (=5) 0 0 0 0 0 0
Average 1.0 1.0 1.0 1.1 1.0 1.0
Table 3. PP-topicalization
The third and most serious problem comes with the observation that the examples in (30) are perfectly grammatical: (30) a.
Die Ma¨nner laden das Heu nicht AUF, sondern ab. the men load the hay not Prt but Prt ‘The men are not loading the hay off, they’re loading it down.’
b. Ich lache dich nicht AUS, sondern an. I laugh you not Prt but Prt ‘I’m not laughing at you, I’m smiling at you.’ The particles in (30) bear contrastive Focus, as in (9) and (10) respectively. But crucially, the examples are not marked like the ones in (9) and (10). Many
Moved preverbs in German: displaced or misplaced?
191
particle verbs that do not allow topicalization of the particle nevertheless allow a contrastive reading of the particle. For example, although most speakers rejected (19), (31) is grammatical: (31) Der Ko¨nig trat (im blauen Anzug) nicht AUF, sondern ab. the king steps in.the blue suit not Prt, but Prt ‘The king didn’t appear in the blue suit, but left.’ McIntyre (2002) also provides examples which illustrate the lack of a 1:1-correspondence between the contrastibility and the topicalizability of particles. He concludes that semantic transparency cannot be a sufficient condition for particle movement. This conclusion, however, leaves open the question of what other factors influence the (in)ability of particles to undergo topicalization. The crucial difference between (30) and (31) on the one hand, and the examples (9), (10) and (19) on the other, is that the particle has not been moved in the former cases. It would be an oversimplification to assume that particle verbs do not allow for particle movement just because their semantics is incompatible with a Topic/Focus construction. Rather, it seems that the problem only arises if the respective Topic – or Focus feature is assigned as a result of movement of the particle. This observation suggests that not only the semantic, but also the structural, properties of particle verbs have an effect on the acceptability of particle movement. I offer an account which is based on this conclusion in section 7.2.
5. LONG TOPICALIZATION As shown in (32)–(36) and Table 4, PP-topicalization is also possible if the PP originates inside an infinitival clause and moves across the sentence boundary into SpecCP of the matrix clause: (32) ?Auf die Party versprach Hans mit Maria zu gehen. on the party promised H. with M. to go ‘Hans promised to go to the party with Maria.’ ¨ ber dich versuche ich zu lachen. (33) U about you try I to laugh ‘I try to laugh about you.’ (34) ?Aus dem Zug haben eigentlich nur Ma¨nner zu steigen versucht. out the train have actually only men to climb promised ‘Only men tried to get off the train.’
192
Sentence Number (32) (33) (34) (35) (36)
J. Zeller
Ok(=1) 14 16 10 15 11
? (=2) 0 0 3 1 5
?? (=3) 0 0 1 0 0
?* (=4) 0 0 1 0 0
* (=5) 2 0 1 0 0
Average 1.5 1.0 1.8 1.1 1.3
Table 4. Long PP-topicalization
(35) Auf diesen Wagen haben die Ma¨nner das Heu zu laden versucht. on this wagon have the men the hay to load tried ‘The men tried to load the hay onto this wagon.’ (36) Aus dieser Stadt versuchte der Rattenfa¨nger alle Kinder zu fu¨hren out this city tried the pied-piper all children to lead ‘The pied-piper tried to lead all children out of this city.’ Given that topicalization out of infinitives is not significantly worse than short topicalization, and in the light of the possibility of topicalizing particles clauseinternally, we expect that long particle topicalization is possible as well. However, this expectation is not borne out by the judgements: (37) ?*Auf begann die Sonne im Osten zu gehen. Prt began the sun in.the east to go ‘The sun began to rise in the east.’ (38) *An versuche ich dich zu lachen. Prt try I you to laugh ‘I’m trying to smile at you.’ (39) ?*Aus haben eigentlich nur Ma¨nner zu steigen versucht. Prt have actually only men to climb promised ‘Only men tried to get off.’ (40) ??Auf haben die Ma¨nner das Heu zu laden versucht. Prt have the men the hay to load tried ‘The men tried to load up the hay.’ (41) ??Angola fu¨hrte viele Waren ein. Aus versucht das Land nur Angola took many goods Prt Prt tries the country only Kaffee zu fu¨hren. coffee to lead ‘Angola imported a lot of goods. The country is trying to export only coffee’ The data are difficult to interpret. According to the judgements given in
Moved preverbs in German: displaced or misplaced?
193
(37)–(41), some examples of long particle topicalization are marginally acceptable, whereas others are ungrammatical. Four out of the five sentences receive an average judgement somewhere between ?? and ?*; they are not perfect, but also not straightforwardly excluded. For each of these four sentences, at least two of the 16 informants could be found that judged them as grammatical. (38) was the only sentence that was clearly judged as ungrammatical, but here the corresponding sentence with short topicalization is already slightly degraded to begin with ((10) is marked as ??). That judgements about long particle topicalization vary considerably means also that the average may not be the appropriate meassure to describe the data. For example, the average response to sentence (37) is that the sentence is unacceptable. Ideally, ?* should reflect the uniform judgement of all speakers, i.e. all speakers are expected to find (37) (almost entirely) ungrammatical. Sentence Number (37) (38) (39) (40) (41)
Ok (=1) 2 0 2 3 3
? (=2) 4 0 2 2 4
?? (=3) 1 1 3 3 0
?* (=4) 2 1 2 1 3
* (=5) 7 14 7 7 6
Average 3.5 4.8 3.6 3.4 3.3
Table 5. Long particle topicalization
However, the frequency distribution in Table 5 shows that more than a third of the speakers actually accepted (37) (i.e. marked it as ok or with a ?). ?* is the result of a wide distribution of judgements about (37), which a (slight) majority found unacceptable. But what does the judgement given in (37) say about the grammar of those speakers that accept long particle movement? How far can we go in ignoring variation? Despite these problems, there remain some interesting observations that can be made with respect to (37)–(41). As Table 1 has shown, the majority of speakers accepted short particle topicalization with the examples in (5)–(10) above (with the exception of (10), the highest number of judgements is always listed in the ok-column). In contrast, as Table 5 verifies, the majority of speakers finds the data in (37)–(41) ungrammatical (most speakers assign a * to these sentences). It is also interesting to investigate the judgements of each individual speaker with respect to a particular pair of sentences (short vs. long topicalization). For example, one notices that with the exception of one speaker, all informants saw a contrast between short and long topicalization with respect to (5) vs. (37).7 For example, the judgements show that of the nine speakers that found (37) impossible (judgements * or ?*), seven completely accepted (5) (i.e. did not even mark it with a ?), while two marked it with a ?. Even for most speakers that found (37) marginally acceptable (?), there is a contrast between
194
J. Zeller
(37) and (5) with short topicalization (which these speakers completely accepted). Similar contrasts can be observed with respect to the other examples of short vs. long particle topicalization, but no comparable observations can be made with respect to PP-topicalization. In sum, although the data perhaps are not sufficiently strong to conclude that particle topicalization is generally clause bound, they nevertheless indicate a definite contrast between short and long particle topicalization.8 Moreover, long particle topicalization shows even more clearly than short topicalization that (i) particle movement is much more constrained than extraction of full PPs and (ii) judgements with respect to particle topicalization are not uniform, but vary considerably from speaker to speaker.
6. PARTICLE SCRAMBLING The following examples show that scrambling of particles is impossible or at least highly marked. In contrast, full PPs allow scrambling. Consider first the examples in (42)–(45), which exhibit scrambling in front of the subject: (42) *Angola fu¨hrt viele Waren ein, obwohl aus das Land nur Angola takes many goods Prt although Prt the country only Kaffee fu¨hrt. coffee takes ‘Angola imports a lot of goods, although the country only exports coffee.’ (43) * ... weil auf die Ma¨nner noch kein Heu geladen haben. because Prt the men still no hay loaded have ‘because the men still haven’t loaded up any hay.’ (44) ?? ... weil aus eigentlich nur Ma¨nner gestiegen sind. because Prt actually only men climbed are ‘because actually only men got off.’ (45) *... weil ein dieses Jahr noch niemand gereist ist. because Prt this year still nobody travelled is ‘because nobody has entered (the country) this year.’ In (42)–(45), a particle has been moved to a position between the complementizer and the subject. In contrast to the data in section 4, which illustrated that the very same particle can be moved to SpecCP, the scrambling examples are unacceptable (with (44) as a notable exception). (46)–(49) shows that PP-scrambling in comparable examples is possible throughout:
Moved preverbs in German: displaced or misplaced?
Sentence Number (42) (43) (44) (45)
Ok (=1) 0 0 5 1
? (=2) 0 1 2 1
?? (=3) 0 0 3 0
?* (=4) 1 0 1 1
* (=5) 15 15 5 13
195
Average 4.9 4.8 2.9 4.5
Table 6. Particle scrambling; adjunction to IP
(46) ... weil aus dieser Stadt der Rattenfa¨nger gestern alle Kinder because out this city the pied-piper yesterday all children fu¨hrte. led ‘because the pied-piper led all children out of this city yesterday.’ (47) ... weil auf diesen Wagen die Ma¨nner noch kein Heu geladen because onto this wagon the men still no hay loaded haben. have ‘because the men still have not loaded any hay onto this wagon.’ (48) ... weil aus dem Zug eigentlich nur Ma¨nner gestiegen sind. because out the train actually only men climbed are ‘because actually only men got off the train.’ (49) ... weil in dieses Land dieses Jahr noch niemand gereist ist. because into this country this year still nobody travelled is ‘because nobody has entered this country this year.’ Sentence Number (46) (47) (48) (49)
Ok (=1) 12 15 16 15
? (=2) 2 0 0 1
?? (=3) 2 1 0 0
?* (=4) 0 0 0 0
* (=5) 0 0 0 0
Average 1.4 1.1 1.0 1.1
Table 7. PP-scrambling; adjunction to IP
The following examples show the same contrasts with respect to scrambling of particles and PPs in the German Mittelfeld: (50) ?*... weil die Sonne auf im Osten geht (aber unter im because the sun Prt in.the east goes but Prt in.the Westen). west ‘because the sun rises in the east but it sets in the west.’
196
J. Zeller
(51) *... weil viele Ma¨dchen dich an gestern gelacht haben because many girls you Prt yesterday laughed have ‘because many girls smiled at you yesterday.’ (52) *Noch niemand ist ein dieses Jahr gereist. still nobody is Prt this year travelled ‘Nobody has entered (the country) this year.’ (53) *Angola fu¨hrte viele Waren ein, obwohl das Land aus nur Angola takes many goods Prt although the country Prt only Kaffee fu¨hrt. coffee takes ‘Angola imports a lot of goods, although the country only exports coffee.’ Sentence Number (50) (51) (52) (53)
Ok (=1) 2 0 0 0
? (=2) 1 0 0 1
?? (=3) 3 0 0 0
?* (=4) 0 0 0 1
* (=5) 10 16 16 14
Average 3.9 5.0 5.0 4.8
Table 8. Particle scrambling; adjunction to VP
Scrambling has moved the particle to a position following the subject and in front of a PP in (50), in front of an adverb in (51) and (52), and in front of the direct object in (53). This kind of particle movement is even less acceptable than scrambling in front of a subject. Again, there is a strong contrast between scrambling of particles and the scrambling of full PPs: (54) ... weil Hans auf die Party mit Maria geht (aber in den Zoo mit because H. on the party with M. goes but in the zoo with Usch). U. ‘because Hans goes to the party with Maria but in the zoo with Usch.’ (55) ... weil viele Ma¨dchen u¨ber dich gestern gelacht haben because many girls about you yesterday laughed have ‘because many girls laughed at you yesterday.’ (56) ?Noch niemand ist in dieses Land dieses Jahr gereist. still nobody is into this country this year travelled ‘Nobody has entered this country this year.’ (57) ... weil der Rattenfa¨nger aus dieser Stadt gestern alle Kinder because the pied-piper out this city yesterday all children fu¨hrte. led ‘because the pied-piper led all children out of this city yesterday.’
Moved preverbs in German: displaced or misplaced?
Sentence Number (54) (55) (56) (57)
Ok (=1) 13 13 9 13
? (=2) 3 1 3 1
?? (=3) 0 1 2 0
?* (=4) 0 0 1 2
* (=5) 0 1 1 0
197
Average 1.2 1.4 1.9 1.4
Table 9. PP-scrambling; adjunction to VP
(54)–(57) show that the majority of speakers accepts Mittelfeld-scrambling of PPs in German. Notice that some of the data in (42)–(57) require Topic-Focus intonation in order to be acceptable; they are instances of what Neeleman (1994) calls ‘‘focus scrambling’’. Respondents were explicitly asked to use this kind of intonation to improve particle scrambling as much as possible. Despite this option, the examples in (42)–(45) and (50)–(53) were not accepted.9 The data are much clearer than those for long particle topicalization. Although there are also a few exceptions, the vast majority of speakers rejected the scrambling of particles. The contrast between particle topicalization and particle scrambling is quite remarkable, particularly given that no comparable contrast can be observed between topicalization and scrambling of full PPs. Any theory of particle verbs that builds on the possibility of particles undergoing topicalization ultimately needs to say something about their inability to undergo scrambling.
7. DISCUSSION In this section I discuss the observations made in sections 4–6 and offer some possible explanations. Before I analyze particle topicalization and particle scrambling in sections 7.2 and 7.3, I look at the licensing conditions for particle verbs in section 7.1. I argue that the movement of any part of a particle verb must be reconstructed at LF in order to derive a local configuration which is necessary in order for the particle verb to be licensed. In section 7.2 I argue that particle verbs are conceptually represented as heads and phrases simultaneously and that syntactic operations must preferably be compatible with both representations. I suggest that this requirement explains some of the puzzling properties of particle topicalization. In section 7.3 I propose an answer to the question of why particle scrambling is impossible.
7.1 Lexical licensing of particle verbs and reconstruction Although the possibility of particle topicalization shows that particles are phrasal complements of the verb, it is clear that a particle complement is in a
198
J. Zeller
crucial way more closely linked to the verb than ‘‘regular’’ complements like full PPs. Even the meaning of a so-called ‘‘semantically transparent’’ particle verb cannot always simply be reduced to the meaning of the verb and the meaning of the particle. In order to illustrate this fact, consider the examples in (58) (cf. McIntyre 2001): (58) a.
eine Langspielplatte auflegen a record Prt-put ‘put on a record’
b. einen Brief einwerfen a letter Prt-throw ‘post a letter’ c.
aussteigen Ptr-climb ‘get off’
The particle verbs auflegen, einwerfen and aussteigen have a transparent semantics. The particles auf, ‘on’, ein, ‘into’, and aus, ‘out’, express the same directional concepts as the corresponding prepositions. They are just used intransitively; their reference objects (=the places onto which the record is placed, into which the letter is thrown etc.) are left implicit. However, McIntyre (2001, 2002) observes that these implicit reference objects cannot be just anything of the right semantic type. The particle verb auflegen in (58a) is only possible with one particular reading: the reference object of auf must be a record player. (58a) cannot be used to express that the record has been put on the shelf or the table.10 Similarly, ein in einwerfen has only one possible reference object, namely a letter box; (58b) cannot be used if the letter is thrown into a rubbish bin. Finally, aussteigen in (58c) is only possible in combination with some means of transportation. Although both Er stieg aus dem Zug, ‘He got off the train’, and Er stieg aus dem Fenster, ‘He climbed out of the window’, are possible, only the former meaning can be expressed with the particle verb aussteigen.11 The specific idiosyncratic properties of these verbs cannot be derived from the meaning of the verb or the particle in isolation. Instead, the conceptual restriction on the implicit reference object must be associated either with the particle verb as a whole or with a special meaning of the particle which is licensed only in the context of the respective base verb. In Zeller (2001), this link between particles and their verbs is expressed in terms of a theory of lexical licensing (cf. Jackendoff 1997). The idea is that for a prepositional element to count as a particle, it has to occur in a specific local configuration with the verb (see Booij (2002) for a similar proposal based on
Moved preverbs in German: displaced or misplaced?
199
Goldberg’s (1995) construction grammar). I suggest that the lexical representation of a preposition states that its realization as a particle (which is often associated with a particular particle semantics, e.g. the requirement to select a particular reference object) is only possible if the preposition is strictly headgoverned by the verb.12 This condition can be compared to the morphological subcategorization frame of an affix whose lexical representation also includes information about the category of its host and about the structural relation between the two elements (i.e. if the affix is a prefix or a suffix etc.). The claim that ‘‘special’’ particle meanings of prepositions are licensed in the local context of a verb implies that the notions ‘‘semantically transparent’’ versus ‘‘idiomatic’’ are not incompatible when it comes to particle verbs. A particle verb like aussteigen in (58c) is certainly semantically transparent (its meaning is based on the special meaning of the particle and the meaning of the verb), but the special meaning of aus is only licensed in combination with the verb steigen, and the lexical representation specifies that the particle and the verb must bear a particular structural relation in order to license this special meaning. In that sense, the particle verb is a ‘‘phrasal idiom’’ (cf. Marantz 1997 for the claim that the meaning of every terminal node is determined by its syntactic context). McIntyre (2002) provides evidence that such constructionspecific meanings of lexical items are not peculiar to particle verbs, but are also attested elsewhere in the grammar. The view that the particle and the verb must be realized in a strictly local configuration requires that this relation is established at the relevant level of syntax where lexical relations are established and checked. I asssume that this level is LF.13 Crucially, this means that verb- or particle movement that takes place in overt syntax must be ‘‘undone’’ at LF in order to restore the local relation between the verb and the particle. The relevant LF process which achieves this is known as reconstruction. If the base position of a moved element is represented as a trace, then reconstruction means that this element is moved back into the trace position at LF. If the base position is represented as an identical copy of the moved element, then reconstruction refers to the fact that it is this copy which is semantically interpreted and accessed by the operations of lexical licensing. Since the lexical licensing of a particle verb requires the particle and the verb to be in a local relation, verb- or particle movement has to be reconstructed at LF. That verb movement is reconstructed at LF is an uncontroversial assumption. Overt verb movement is quite generally taken to be ‘‘invisible’’ at LF (see e.g. Uriagereka 1995, 98f.; Wurmbrand 2000, 16). Evidence is provided by idiomatic expressions with immobile phrasal parts (i.e. if they are moved, the idiomatic reading disappears). For example, the following (b)-examples show that topicalization of the PP-parts of these idioms is impossible.14 In contrast, the (a)-examples with the verb in C0 show that verb movement does not affect
200
J. Zeller
the acceptability of the idiomatic reading (following Jackendoff (1997), I signal an impossible idiomatic reading through #): (59) a.
Er schla¨gt zwei Fliegen mit einer Klappe he beats two flies with one swatter ‘He kills two birds with one stone’
b. #Mit einer Klappe schla¨gt er zwei Fliegen (60) a.
Er malt den Teufel an die Wand he paints the devil on the wall ‘He tempts fate’
b. #An die Wand malt er den Teufel The structural conditions on lexical licensing of particle verbs require that it must also be possible to reconstruct a topicalized particle into its base position at LF. The assumption that topicalized constituents undergo reconstruction is independently required to account for examples like (17) and (18) in section 4 where a topicalized direct object establishes VP-focus. However, this does not mean that topicalization is really ‘‘invisible’’ at LF. In contrast to verb movement, topicalization has a direct impact on the interpretation of the sentence. As noted above, for a non-subject to move to SpecCP, it must be either a Topic or a Focus. I assume with Bu¨ring (1996) that Topic and Focus are syntactic features which are automatically associated with a topicalized constituent. Then we can assume that movement of a particle to SpecCP marks it as a Focus or a Topic (presumably under Spec-Head agreement with the respective feature in C0 , cf. Wurmbrand 2000), and that this property is preserved after reconstruction. In other words, although the LF representation of a sentence whose overt form includes a topicalized particle finds the particle in its base position, the reconstructed particle is now marked with a Focus or Topic feature as a result of overt movement. 7.2 The multirepresentational status of particle verbs and topicalization In section 4, I showed that neither an exclusively structural nor a purely semantic approach can explain the facts that have been observed with respect to particle topicalization. A morphological analysis of particle verbs that denies that particles are phrases might account for those particles that cannot undergo XP-movement, but fails to explain the fact that some particles can be topicalized. The alternative proposal, which assumes that certain particles are immobile because they lack the semantic potential to be Topics or Foci, cannot explain why many of these particles are licensed in Focus constructions if Focus is assigned to them in situ and not as the result of movement of the particle. I want to propose a possible explanation for the properties of particle topicalization that combines insights of the structural and the semantic
Moved preverbs in German: displaced or misplaced?
201
approach. From the structural approach I adopt the idea that the impossibility of certain particles to be moved has something to do with the ‘‘word-like’’ character of particle verbs. I assume with the semantic approach that particle topicalization establishes Focus- or Topic-marking on the particle and that it is this fact that renders movement impossible in certain cases. My proposal is based on the idea that particle verbs are not conceptualized as either words or phrases, but as words and phrases at the same time. I assume that both representations are activated simultaneously whenever a particle verb occurs in a sentence. Crucially, particle topicalization is based on the syntactic representation of the particle verb (because the particle can only be moved to SpecCP if it is a phrase). Therefore, speakers are forced to interpret the particle verb unambiguously as a syntactic construction if the particle has been moved. Marginality or even unacceptability results from the speaker’s inability to match this syntactic representation with the particle verb’s representation as a word. In Zeller (2002) I argue that particle verbs in the Germanic languages are represented by two parallel structures. In one, the particle is a phrase and the particle verb is a V∞ or VP (structure 1 in (61b)). In the other, the particle is not a phrase, since the verb and the particle have been reanalyzed as a complex head (structure 2): (61) a. ein Bier ausgeben, ‘invite (somebody) for a beer’ b.
The syntactic representation in structure 1 is motivated (among other things) by the fact that verb movement can leave the particle behind in V2 and (Dutch)
202
J. Zeller
Verb Raising constructions like (62b).15 Evidence for structure 2 is provided by (62a), which shows that particle verbs can also move as complex heads in Dutch Verb Raising constructions (for more details of this analysis, see Zeller 2002):16
(62) a.
dat Jan zijn moeder wil opbellen that Jan his mother wants Prt-phone
[[ ti ] matrix verb [particle verb]i]
b. dat Jan zijn moeder op wil bellen [[particle ti]matrix that Jan his mother Prt wants phone verb verbi] ‘that Jan wants to call up his mother’ (Neeleman 1994: 24)
Syntactic rules can trigger either of the two structures in (61). For example, particle topicalization is possible if PrtP in structure 1 is moved, whereas a sentence like (62a) is derived if head movement triggers the complex V0 in structure 2. Before I analyze particle movement in the light of the structure in (61b), let me first address some common objections that have been leveled against the reanalysis proposal and multi-representational structures like (61b). For example, one reviewer raises the question to what extent the double structure in (61b) is merely an ad hoc way of capturing the ambiguous character of the verbparticle construction. Others have criticized multi-representational approaches for enriching the theoretical formalism, and some reject reanalysis as too powerful a tool and as being too unrestricted. I think that these objections are unwarranted. It is an observable and undisputed fact that the rules of grammar ‘‘see’’ the particle verb sometimes as a word and sometimes as a phrase. Representing this situation by two parallel structures, one in which the particle verb is a phrase and one in which it as a head, is a straightforward approach to this problem. Both structures in the twosided diagram in (61b) are formed according to the general rules of syntax. Specific rules which license the occurence of two parallel structures and which determine the contexts for reanalysis have been stated in the literature. As discussed in Van Riemsdijk (1998), one such condition for reanalysis is linear adjacency of the two elements which are reanalyzed (here: the particle and the verb). If the verb and the head of its complement in structure 1 were separated by a complement of the particle (and therefore were not adjacent), then reanalysis could not apply. Another condition is proposed in Zeller (2001). There I suggest that a verb and the lexical head of its phrasal complement can only be reanalyzed if no functional structure intervenes between the two heads. Assuming that particle phrases lack functional structure (in contrast to ‘‘regular’’ NP-, AP- or PP-complements, which have functional extended projections), it follows that a double representation like (61b) is only available for particle
Moved preverbs in German: displaced or misplaced?
203
verbs, but not for similar constructions with intransitive prepositions or resultative adjectives (see Zeller (2001, 2002) for more details of this proposal). Conditions like these limit the scope of reanalysis and hence avoid overgeneralization. They might increase the complexity of the theory, but this cannot seriously be regarded as a problem; if it was, then the same objection would also have to be raised with respect to e.g. the introduction of abstract functional heads in syntax or possible worlds in formal semantics. It should also be pointed out that a particle verb is by far not the only construction which simultaneously shows properties of two different structural representations. Interestingly, for many constructions with similar ambiguous properties, a reanalysis rule and/or multi-representational structures have been proposed in the literature. Such proposals include (among others) Hornstein and Weinberg’s (1981) account of preposition stranding, Sadock’s (1985) analysis of noun incorporation in Greenlandic, Zubizaretta’s (1985) study of Romance causative constructions and Haegeman and Van Riemsdijk’s (1986) treatment of verb cluster formation in various Germanic languages (cf. also the theory of coordination proposed in Moltmann (1992) and the proposal made in Van Riemsdijk (2000) regarding free relatives). For all these phenomena, alternative proposals based on standard tree representations exist as well (cf. e.g. the head movement analysis proposed in Baker (1988)). However, in Zeller (2002) I provide a discussion of particle verbs in the context of a comparison between some of these alternative theories and the multi-representational approach, and the discussion shows that only the latter can fully account for the heterogeneous properties of the verb-particle construction. The hybrid status of particle verbs is a challenge for tree-representability, but as long as one does not want to give up hierarchically structured syntactic representations when confronted with problematic data, a multi-representational theory can be considered an adequate and well-motivated approach. Let me return to the representation of particle verbs at LF. As noted above, both particle- and verb movement are reconstructed at LF in order to establish the syntactic configuration required for lexical licensing. This configuration, which I expressed as head government in section 7.1, can now be reinterpreted as the requirement that in both representations, the syntactic relation between the verb and the particle must be strictly local. This locality requirement is met if the verb governs the particle, but also if both the verb and the particle form a complex head.17 I assume that the interpretation of particle verbs involves the checking of structure 1 against the properties of structure 2 at LF. What has to be checked is whether all semantically interpreted features of one representation are also present in the other. Only if the feature specifications of both structures are the same is LF fully interpretable. This assumption does not predict that verb movement causes any problems. Since verb movement is ‘‘invisible’’ at LF, both structure 1 and structure 2 have
204
J. Zeller
the same properties at LF after reconstruction, regardless of whether verb movement has actually been initiated from structure 1 (V2 and (62b)) or structure 2 ((62a)). However, particle movement has a different effect. Only structure 1 allows for this option. As noted above, a topicalized particle receives a Focusor Topic-feature which is preserved in structure 1 after reconstruction. But this feature is not present in structure 2. If the speaker checks structure 1 against the properties of structure 2, the interpretation fails. However, it might be possible to ‘‘save’’ the structure by suppressing the word-like conceptualization of the particle verb. In order to accept particle topicalization, a speaker is forced to abandon the multirepresentational format of particle verbs and to treat them basically as phrasal constructions. Therefore, only if the phrasal representation of a particular particle verb is dominant for a speaker does topicalization of the respective particle become acceptable. I assume that this is what explains the problematic properties of particle topicalization. For some speakers, the conceptualization of particle verbs as words might be too strong for an exclusively phrasal representation to be acceptable. For others, the phrasal representation of particle verbs might be strong enough to ignore the V0 -structure in particular contexts. The more dominant a speaker’s representation of the particle verb as a V∞ or a VP, the more acceptable will (s)he find the examples with topicalized particles. The ‘‘conceptual weight’’ of the syntactic representation differs from particle verb to particle verb and from speaker to speaker, causing inconsistent judgements and ideolectal variation. It is an interesting question of what determines the conceptual strength of the two representations. One aspect that certainly favors the V0 -representation (and therefore renders particle topicalization more difficult) is the nontransparent character of certain particles. Of course, lexical listedness of a construction cannot be taken as evidence that this construction is a word. However, as McIntyre (1998: 26) notes, ‘‘listedness can carefully be reinstated as a sign that a structure is a morphological object rather than a phrase under the important proviso that the phrase has a ‘plausible morphological structure’.’’ The point is that a phrasal idiom like jemandem den Garaus machen, ‘finish somebody of’, is unlikely to be analyzed as a word, because its properties are too obviously syntactic. For example, the idiom includes articles and other functional elements that are typically absent from morphological structures. In contrast, particle verbs clearly have a ‘‘plausible morphological structure’’; I even assume that this structure is represented in tandem with their phrasal representation. The more idiomatic the properties of the particle verb, the more pronounced is its V0 -representation at LF. I take this to be the reason for the ban on particles in SpecCP that are part of a particle verb with a non-compositional semantics. The idiomatic status of the particle verb makes it impossible for the speaker to overwrite the word-like representation of structure 2. Therefore, even though topicalized phrases can sometimes establish Topic or Focus on the
Moved preverbs in German: displaced or misplaced?
205
whole VP, an LF with a topicalized particle simply becomes uninterpretable if the particle verb is not semantically transparent. Notice that it is not the fact that the particle receives a Focus- or Topicfeature, but the fact that this feature is assigned as a result of topicalization of the particle, which causes interpretative problems at LF. Since particle movement is only possible with structure 1, it is only in this structure that the particle phrase bears that feature. In contrast, if a particle is assigned this feature in situ, it can be assigned to the particle node in both representations. Consequently, focused particles without topicalization are perfectly acceptable. Certainly, a number of questions remain open. For example, it is not clear how the proposed analysis of LF-interpretation and licensing can account for there being more diverse judgements about long particle topicalization than about short particle topicalization. However, the data support the general idea behind my proposal: since particle verbs are represented as heads and phrases, the fact that particle movement requires a unique analysis of the particle as a phrase causes difficulties. This explains why the acceptability of particle topicalization often cannot be evaluated on the basis of a sharp distinction between ‘‘grammatical’’ and ‘‘ungrammatical’’, but is rather a matter of degree.
7.3. Scrambling and lexical licensing Finally, let me turn to the observation that for most speakers, even those particles that can be topicalized cannot be scrambled, whereas scrambling of full PPs is readily acceptable. One way of explaining this would be to assume that, in contrast to topicalized constituents, scrambled phrases do not reconstruct at LF. This would imply that particle verbs cannot be licensed if the particle has been scrambled because the local configuration required for lexical licensing is not established at LF. In contrast, the combination of full PPs and verbs does not require lexical licensing. Therefore, the fact that scrambled PPs are not in their base position at LF does not affect the acceptability of these examples. Bu¨ring (1996) argues explicitly that scrambled elements do not reconstruct. However, his only argument in support of this assumption is that according to Bu¨ring, scrambling is A-movement (which is generally taken not to be subject to reconstruction.) Apart from the fact that an analysis of scrambling as A-movement is itself controversial, Grewendorf and Sabel (1994: 301) provide evidence that scrambled constituents do in fact reconstruct at LF. For example, in (63), the infinitival clause has been scrambled into the matrix clause. It includes the pronoun sie. Given that a bound-pronoun reading is available, the infinitive must be in its base position at LF because in its adjoined position, the pronoun would not be c-commanded by the quantifier jeder/keiner Frau:
206
J. Zeller
(63) weil [siei nicht zu vergessen] Peter jeder/keiner Fraui t versprochen since she not to forget Peter every/no woman promised hat has ‘since Peter promised every/no woman not to forget her’ Given the problematic character of the claim that scrambled categories do not undergo reconstruction at LF, we might consider another explanation for the impossibility of particle scrambling. Notice that a contrast between scrambling and topicalization has also been observed with respect to so-called ‘‘remnant movement’’ (cf. Grewendorf 1992; Grewendorf and Sabel 1994: 265f.): (64) a.
[t zu fu¨ttern] hat [den Hund] keiner t versucht to feed has the dog nobody tried ‘Nobody has tried to feed the dog’
b. *dass [t zu fu¨ttern] [den Hund] keiner t versucht hat that to feed the dog nobody tried has c.
dass [den Hund zu fu¨ttern] keiner t versucht hat that the dog to feed nobody tried has
In a first step, the direct object of the infinitival verb zu fu¨ttern, the DP den Hund, has undergone long scrambling in front of the main subject DP keiner in both (64a) and (64b). In a second step, the infinitive from which the DP has been extracted (the remnant [t zu fu¨ttern]) is moved as well. In (64a), the remnant is topicalized, in (64b), it is scrambled. (64a) is grammatical, (64b) is not. (64c) shows that scrambling of the infinitive is possible if the DP-object has remained in situ. The contrast between (64a) and (64b) is interesting, because it corresponds to the contrast between particle scrambling and particle topicalization observed above,18 and it seems in fact that there is a solution that accounts for both sets of data. Grewendorf and Sabel’s (1994) explanation of the contrast shown in (64) is based on their specific account of long scrambling. They argue that in long scrambling constructions, the infinitival TP has moved to SpecCP of the embedded clause, (65). Grewendorf and Sabel assume that in order for a constituent to leave the TP and move into the matrix clause, the head of the TP has to undergo abstract incorporation into the matrix verb. Abstract incorporation, as introduced by Baker (1988), takes place at LF and is therefore not visible in overt syntax. However, like overt incorporation, abstract incorporation ‘‘opens barriers’’. If the embedded T-head incorporates covertly into the matrix verb, the two heads count as non-distinct even in overt syntax. Nondistinctness of the two heads prevents the TP in the embedded SpecCP from being a barrier
Moved preverbs in German: displaced or misplaced?
207
between an XP moved into the matrix clause and its trace. Therefore, long scrambling of the direct object in (64a) is possible because the embedded T-head incorporates into the matrix verb at LF. (65)
This assumption provides the basis for Grewendorf and Sabel’s (1994) account for the ungrammaticality of (64b). Without discussing the details of Grewendorf and Sabel’s proposal, let me note their crucial claim (Grewendorf and Sabel 1994: 288):19 (66) [A]djoined categories constitute barriers for heads contained within them As mentioned above, Grewendorf and Sabel show that scrambled constituents reconstruct at LF (and therefore are not adjoined categories at the level where the ECP is checked). However, they assume that if a category is adjoined to another one in overt syntax, it receives a feature [+ barrier], and crucially, this feature is preserved under reconstruction. Because of (66), scrambling an infinitive into the matrix clause (which is an instance of adjunction) creates a barrier between the infinitive and the matrix verb and hence prevents the embedded T-head from incorporating at LF. But if no abstract incorporation takes place, movement of a direct object out of the infinitive crosses a barrier (the TP). In (64b), scrambling of zu fu¨ttern has the effect that long scrambling of the direct object out of the infinitive crosses a barrier. In contrast, long topicalization of the infinitive (which is not subject to (66)) still allows abstract incorporation. Therefore, (64a) is grammatical. If we adopt this proposal, it is clear that particle phrases are barriers at LF if they have been scrambled. Now recall that the licensing condition for particle verbs that I formulated above requires the relation between the particle and the verb to be local in both structures at LF. In their phrasal representation, particles are only licensed if they are strictly head-governed by the verb. However,
208
J. Zeller
although a scrambled particle phrase is in its base position at LF, the particle is not head-governed by the verb, because the particle phrase still bears the feature [+ barrier]. Particle scrambling is excluded, because this particular kind of movement prevents the particle from being licensed at LF.
8. CONCLUSION In section 3 I discussed the controversial status of particle verbs in German, pointing out that the investigation of particle movement can yield new insights into the structural properties of these peculiar preverb-verb constructions. Three major observations about particle movement were made in this study: (i) (ii) (iii)
Particle topicalization is possible if the particle is the Topic or the Focus of the sentence. Speakers’ judgements about particle topicalization vary; most speakers find the acceptable cases slightly worse than the topicalization of full PPs. Particle scrambling is impossible.
What conclusions about the structure of particle verbs can be gained from these observations? First, I tried to show that (i) supports the ‘‘syntactic’’ view that particles can be represented as phrasal complements of their verbs, a view in line with many proposals made in the literature on particle verbs. However, I emphasized that a syntactic approach also has to take (ii) into account, i.e. the contrast between particle topicalization and PP-topicalization that is notable for most speakers. Therefore, instead of opting for a syntactic analysis of particle verbs that treats the preverb just like a regular (intransitive) PP-complement of the verb, I argued that the marked character of particle fronting is the result of the double-structure representation of particle verbs. I suggested that a particle verb can be represented as a V∞ or VP-construction which has a parallel representation as a V0 at LF. I argued that the conflicting properties of these two representations are responsible for the heterogeneous judgements of speakers about particle fronting. Finally, (iii) provides evidence that the phrasal representation of particles is subject to licensing conditions that require a strictly local relation between the particle and the verb at LF. Since local relations between heads cannot be established if a phrase intervening between these heads is an adjunct at some level of syntax, scrambling of particle phrases (an instance of adjunction to IP or VP) is impossible. The observation that the contrast between particle topicalization and particle scrambling can be compared to similar contrasts observed with respect to remnant movement of infinitives in German shows that not all properties of particle verbs are caused by the exceptional hybrid character of this construction. Some of the restrictions on the movement of
Moved preverbs in German: displaced or misplaced?
209
particles follow directly from general principles that regulate syntactic operations. NOTES * I thank Geert Booij, Andrew McIntyre, Stefan Ploch, Joachim Sabel and an anonymous reviewer for their comments and my informants for the time they took to provide me with their judgements. A special thanks goes to Dori Posel for her help with this article. 1 I restrict myself to an examination of prepositional particles, and I use the term ‘‘particle’’ with respect to these elements only. A number of authors (cf. Booij 1990, 2002; Stiebels and Wunderlich 1994; Zeller 2001, 2002) also discuss examples of nominal particles. Movement of these elements is much less restricted than movement of prepositional particles. 2 Included in the sample were also Austrian-German mother tongue speakers. I sampled only linguists because they were likely to be more sensitive to nuanced differences illustrated in the range of examples. 3 This principle is known as the Principle of Lexical Integrity, which has developed out of Chomsky’s (1970) lexicalist hypothesis (cf. Lapointe 1980; di Sciullo and Williams 1987; Bresnan and Mchombo 1995). See Booij and van Kemenade (this volume) for some discussion. 4 (5), (6), (8) and (10) are judged as perfectly grammatical in the original publications. In Zeller (2002), I mark an example similar to (9) as?? and one similar to (7) as ? (on the basis of judgements of only four informants). 5 The Focus value of a particle is not only restricted by the context, but also by the number of possible particle verbs that can be derived from the base verb that is part of the Background. In addition, the thematic and grammatical properties of these particle verbs have to match those of the original particle verb. For example, although auflachen, ‘give a laugh’, and zulachen, ‘laught at’, are particle verbs formed from the same base verb as auslachen, auf and zu cannot be elements of the Focus value of aus: auflachen is intransitive and hence does not have a h-role which it could assign to the direct object which is part of the Background in (13), and zulachen, although semantically transitive, takes a dative object (in contrast to anlachen and auslachen, which take accusative objects). It therefore seems that an is the only possible alternative to aus in (13). 6 Mu¨ller’s examples in (22) and (23) are taken from newspaper articles, example (19) appeared in a weekly magazine, (20) and (21) were found in novels. In (19) and (21), I corrected the wrong spelling in the original data. 7 The tables do not provide the data on these individual responses to each example. 8 The data in (37)–(41) show particle extraction from control infinitives. I did not collect judgements with movement in ECM- or raising constructions. According to my own judgements, perception verbs allow for long topicalization, raising verbs do not: (i) (ii) 9
??Auf sehe ich die Prt see I the *Aus scheinen nur Prt seem only
Sonne im Osten gehen sun in.the east go Ma¨nner zu steigen men to climb
I only tested one example with long scrambling of a particle which (unsurprisingly) was entirely ungrammatical:
210
(i)
J. Zeller
*... weil auf die Sonne [im Osten t zu gehen] begann because Prt the sun in.the east to go began
(i) was judged as?* by two speakers; the remaining 14 speakers found it entirely ungrammatical (*). 10 A different meaning of auflegen as in den Ho¨rer auflegen, ‘put down the phone’, is also specific in this regard. This particle verb cannot be used to express that the receiver is put on a table or a chair, but only that it is put back on the phone. 11 The particle verb aussteigen can also be used with the special meaning ‘opt out of society’ which is presumably based on a metaphorical extension of its transparent meaning in (58c). 12 In Zeller (2001) I call the local configuration between the particle and verb ‘‘structural adjacency’’, which is defined as the relation between a head and the head of its complement. This definition implies that the particle is head governed by the verb. 13 This notion of LF corresponds to the levels of Lexico-Logical Form (LLF) in Brody (1995) and Syntactic Structure in Jackendoff (1997). This correspondence and the relation between (L)LF and lexical licensing are discussed in detail in Zeller (2001, chapter 1). 14 Examples like (59) and (60) therefore contrast with idioms like (18), discussed in section 4, whose meanings are preserved when one of its parts is topicalized. 15 I assume that verb movement that leaves the particle behind is based on structure 1, but recall that structure 2 is not necessarily incompatible with verb movement (see section 3). 16 In Zeller (2002) I argue that the inflectional properties of particle verbs prevent the whole verb from appearing in Comp0 . Therefore, the two options illustrated by (62) are only attested in Verb Raising constructions, but not in V2-contexts. 17 According to Roberts (1985), the verb in structure 2 ‘‘morphologically governs’’ the particle. See Roberts (1985) and Zeller (2001) for a discussion of the parallels between morphological government (inside X0 ) and head government (in syntax). 18 Thanks to Joachim Sabel who first brought this correspondence to my attention. 19 The definition of barrier that Grewendorf and Sabel (1994) adopt defines a category C as a barrier if ‘‘the maximal projection that immediately dominates C does not include C’’ (p. 277). If a phrase is adjoined to XP, it is immediately dominated by (one segment of) XP, but not included in it. Adjoined categories are hence barriers.
REFERENCES Baker, M. (1988). Incorporation. A Theory of Grammatical Function Changing. Chicago/London: The University of Chicago Press. Booij, G. (1990). The boundary between morphology and syntax: separable complex verbs in Dutch. In: G. Booij and J. v. Marle (eds),Yearbook of Morphology 1990. Dordrecht: Foris, 45–63. Booij, G. (2002). Separable complex verbs in Dutch: A case of periphrastic word formation. In: N. Dehe´, R. Jackendoff, A. McIntyre and S. Urban (eds), Verb-Particle Explorations, Berlin and New York: Mouton de Gruyter, 21–41. Booij, G. and A. van Kemenade (this volume). Preverbs, an introduction. Bresnan, J. and S.A. Mchombo (1995). The lexical integrity principle: Evidence from Bantu. Natural Language and Linguistic Theory 13, 181–254. Brody, M. (1995). Lexico-Logical Form. MIT Press, Cambridge.
Moved preverbs in German: displaced or misplaced?
211
Bu¨ring, D. (1996). The 59th Street Bridge Accent. Ph.D. dissertation, University of Tu¨bingen (SfS-Report 05-96). Chomsky, N. (1970). Remarks on nominalization. In R. Jacobs and P. Rosenbaum (eds), Readings in English Transformational Grammar. Waltham, Mass.: Ginn., 184–221. Di Sciullo, A.M. and E. Williams (1987). On the Definition of Word. Cambridge: MIT Press. Goldberg, A. (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago & London: The University of Chicago Press. Grewendorf, G. (1992). German – A grammatical sketch. Sprachwissenschaft in Frankfurt Nr. 8. Grewendorf, G. and Sabel, J. (1994). Long scrambling and incorporation. Linguistic Inquiry 25, 263–308. Haegeman, L. and H. van Riemsdijk (1986). Verb projection raising, scope, and the typology of rules affecting verbs. Linguistic Inquiry 17, 417–466. Haider, H., S/ Olsen and S. Vikner (1995). Introduction. In: H. Haider, S. Olsen and S. Vikner (eds), Studies in Comparative Germanic Syntax (Studies in Natural Language and Linguistics Theory 31). Dordrecht, Boston, London: Kluwer, 1–45. Hoeksema, J. (1991). Theoretische aspekten van partikelvooropplaatsing. TABU Bulletin voor Taalwetenschap 21, 18–26. Hornstein, N. and A. Weinberg (1981). Case theory and preposition stranding. Linguistic Inquiry 12, 55–92. Jackendoff, R. (1997). The Architecture of the Language Faculty. Cambridge: MIT Press. Kratzer, A. (1994b). The event argument and the semantics of voice. Ms., University of Massachusetts, Amherst. Krifka, M. (1994). Focus and operator scope in German. In P. Bosch and R. van der Sandt (eds.), Proceedings of the Conference on Focus and Natural Language Processing, Volume 1: Intonation and Syntax, Working Papers of the Institute for Logic and Linguistics, IBM Scientific Centre, Heidelberg, 1994, 133–152. Lapointe, S. (1980). A Theory of Grammatical Agreement. Ph.D. dissertation, University of Massachusetts, Amherst (published 1985, New York and London: Garland). Le Roux, C. (1988). On the interface of morphology and syntax: Evidence from verb-particle combinations in Afrikaans. M.A. thesis, SPIL, University of Stellenbosch. Lu¨deling, A. (1998). On Particle Verbs and Similar Constructions in German. Ph.D. dissertation, University of Stuttgart. [published by CSLI, Stanford, 2001]. Marantz, A. (1997). No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. Penn Working Papers in Linguistics, Volume 4/2 [Proceedings of the 21st Annual Penn Linguistics Colloquium], 201–225, University of Pennsylvania. McIntyre, A. (1998). Double Particles of the Type Heran, Hinzu, Raus and the German Preverb System. Doctoral Dissertation, University of Sydney. McIntyre, A. (2001). German Double Particles as Preverbs: Morphology and Conceptual Semantics. Tu¨bingen: Stauffenberg. McIntyre, A. (2002). Idiosyncrasy in particle verbs. In: N. Dehe´, R. Jackendoff, A. McIntyre and S. Urban (eds), Verb-Particle Explorations. Berlin and New York: Mouton de Gruyter, 95–118. Moltmann, F. (1992). Coordination and Comparatives. Ph.D. Dissertation, MIT, Cambridge. Distributed by MIT Working Papers in Linguistics. Mu¨ller, S. (2002). Complex predicates: verbal complexes, resultative constructions and particle
212
J. Zeller
verbs in German. Manuscript, University of Jena, (downloadable from http:// www.dfki.de/~stefan/PS/complex.pdf). Neeleman, A. (1994). Complex predicates. Ph.D. dissertation, OTS, University of Utrecht. Neeleman, A. and F. Weerman (1993). The balance between syntax and morphology: Dutch particles and resultatives. Natural Language and Linguistic Theory 11, 433–475. Olsen, S. (1997b). Zur Kategorie Verbpartikel. Beitra¨ge zur Geschichte der deutschen Sprache und Literatur 119, 1–32. Riemsdijk, H. van (1998). Head movement and adjacency. Natural Language and Linguistic Theory 16, 633–678. Riemsdijk, H. van (2000). Free relatives inside out: Transparent free relatives as grafts. In: B. Rozwadowska (ed.), PASE Papers in Language Studies – Proceedings of the 8th Annual Conference of the Polish Association for the Study of English, 223–233. University of Wroclaw (downloadable at http://cwis.kub.nl/~fdl/general/people/riemsdh/PASE.PDF.) Roberts, I. (1985). Agreement, parameters and the development of English modal auxiliaries. Natural Language and Linguistic Theory 3, 21–58. Rooth, M.E. (1985). Association with Focus. PhD Dissertation, University of Massachusetts. Amherst. Sadock, J.M. (1985). Autolexical Syntax: A proposal for the treatment of noun incorporation and similar phenomena. Natural Language and Linguistic Theory 3, 379–439. Stiebels, B. (1996). Lexikalische Argumente und Adjunkte (Studia Grammatica 39). Berlin: Akademie Verlag. Stiebels, B. and Wunderlich, D. (1994). Morphology feeds syntax: the case of particle verbs. Linguistics 32, 913–968. Toman, J. (1983). Wortsyntax. Tu¨bingen: Niemeyer. Uriagereka, J. (1995). Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26, 79–123. Wurmbrand, S. (2000). The structure(s) of particle verbs. Unpublished manuscript, University of Montreal. Zeller, J. (2001). Particle Verbs and Local Domains. Amsterdam: John Benjamins. Zeller, J. (2002). Particle verbs are heads and phrases. In: N. Dehe´, R. Jackendoff, A. McIntyre and S. Urban (eds), Verb-particle Explorations. Berlin and New York: Mouton de Gruyter, 233–267. Zifonun G. (1999). Wenn mit allein im Mittelfeld erscheint. Verbpartikeln und ihre Doppelga¨nger im Deutschen und Englischen. In: H. Wegener (ed.), Deutsch kontrastiv. Tu¨bingen: Stauffenberg. Zubizaretta, M.L. (1985). The relation between morphophonology and morphosyntax: The case of Romance causatives. Linguistic Inquiry 16, 247–289.
Linguistics Programme Faculty of Human Sciences University of Natal, Durban Durban 4041 South Africa e-mail: [email protected] www.jzeller.de
Distribution-driven morpheme discovery: a computational/experimental study MARCO BARONI 1. INTRODUCTION1 During the process of language acquisition, learners must discover which strings constitute the affixes of their language and which words of the language can be decomposed into affixes and other components. These are prerequisites to morphological acquisition. Ultimately, a learner acquiring a language must discover the syntactic and semantic properties associated with each affix of the language, in order to be able to produce and understand new words. For example, a learner acquiring English must discover that re- is a prefix that attaches to verbs to create other verbs with an iterative meaning. However, in order to learn the morphological properties of an affix, learners must first of all notice the existence of that affix. Moreover, in order to discover the linguistic properties associated with the affix, the learner must inspect the semantic, syntactic and morphological characteristics of a set of words containing that affix. For example, in order to discover the properties of the prefix re-, English learners must first of all, of course, notice that the string re- is a prefix. Moreover, the learners must collect and analyze a number of words containing the prefix re- (redo, rename, remake ...),2 in order to extract the correct generalizations about this prefix. However, not all the words containing a string identical to an affix actually contain that affix. In order to discover the correct generalizations about the properties of the affix, the learners must have a preliminary idea of which of the words containing a string identical to the affix are actually affixed. If an English learner tried to decide what is the meaning and function of re- on the basis of, say, redo, retail and really, the learner would probably come up with the wrong generalizations about the prefix or, more likely, she would not notice any generalization at all and she would conclude that re- is not a prefix. Of course, if the string corresponding to an affix mostly occurs in words which do indeed contain the affix, the learner is probably going to extract the correct generalizations even if there are a few pseudo-affixed words (i.e., words containing the string corresponding to the affix without actually being synchronically analyzable as affixed). However, this is not always the case. For example, Schreuder and Baayen (1994) have shown that, for several common English and Dutch prefixes, the number of pseudo-prefixed words is higher than the number of truly prefixed words (at least in terms of token frequency). Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 213–248. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
214
M. Baroni
Thus, it would not be safe, for a learner, to assume a priori that any word containing a string identical to an affix does indeed contain the affix from a morphological point of view. Consequently, the learner must decide which of the words containing a potential affix are truly morphologically complex, and which are pseudo-affixed, i.e., the learner must assign preliminary morphological parses to the words she hears. While I presented the task of discovering that a certain string is an affix (or more generally a morpheme) and the task of assigning parses to words as separate aspects of morpheme discovery, the two tasks are obviously closely related. A learner is not likely to hear affixes in isolation. Thus, the task of discovering the affixes will typically involve assigning morphological parses to words. A string is an affix of the language if at least one of the words containing the string in the language is parsed as morphologically complex, and the string constitutes one of the morphological components in the parse. Morpheme discovery is a difficult task. Not only does the learner have to consider many possible segmentations of each potentially complex word she hears, but she does not a priori know which meanings and/or syntactic functions are expressed by morphemes in her language, and consequently she cannot a priori know whether a word is morphologically complex or not. Furthermore, the learner does not know which types of morphemes (prefixes, suffixes, circumfixes, infixes, autosegments, templates ...) are present in the language. Thus, even if the learner had some reason to expect a certain word to be morphologically complex, she still would have to determine whether the word should be divided into a prefix and a stem, or into a stem and a suffix, or into consonantal and vocalic templates, or into other morpheme combinations. It is probable that learners follow a number of different morpheme discovery strategies, looking for phonological, syntactic and semantic cues. Moreover, distributional evidence, i.e., evidence based on the frequency and co-occurrence patterns of words and their substrings, provides potentially useful cues that learners can exploit. While each of these approaches can help the learner in the morpheme discovery task, none of them is likely to be sufficient by itself. The primary goal of this study is to contribute to a better understanding of how language learners perform morpheme discovery. In particular, the study provides evidence in favor of the hypothesis that distributional cues play a significant role in this process. Some recent studies have provided new support for the idea that distributional information plays an important role in language learning (see Redington and Chater 1998 for a review of both the classic objections to distributional approaches and recent distribution-driven learning models). Thus, another goal of the present study is to provide further support for the general claim that language learners make crucial use of distributional cues. As I will shortly discuss in the conclusion, another goal of this study is to
Distribution-driven morpheme discovery
215
provide a (partial) explanation for an interesting datum emerging from experimental studies of morphological processing and representation, i.e., that speakers can represent words as morphologically complex even if they lack semantic compositionality (i.e., the meaning of the whole word is not the product of the meanings of the component morphemes). One can argue that this phenomenon (complex representation/treatment of semantically opaque words) is at least in part a by-product of distribution-driven morpheme discovery, and the empirical evidence presented here provides some support for this hypothesis. In order to assess the potential role of distributional cues in morpheme discovery, I designed an automated learner, which performs a simplified version of this task on the sole basis of the distributional evidence it can extract from a corpus of untagged words. The most obvious simplification in the task performed by this computational model is that it only looks for prefixes and stems, and not also for other kinds of morphemes. The strategy followed by the automated learner in its search for prefixes and stems is based on a simple fact about the distributional nature of morphemes: morphemes are independent linguistic units, and as such, they occur in a number of different words where they combine with other morphemes. The nonrandom distribution of morphemes makes them detectible, in many cases, by statistical methods. Given an input corpus of English words, the automated learner, equipped with a small number of simple distributional heuristics, is able to discover a large set of actual English prefixes, finding very few ‘‘false positives’’ (strings which are not English prefixes but are treated by the learner as such). Moreover, the morphological parses (prefix+stem vs. monomorphemic) assigned by the learner to the words in the input corpus are correlated with intuitions of native English speakers about the morphological structure of the same words. Thus, the computational simulation presented here demonstrates first of all that a limited number of simple distributional heuristics can help a morpheme discoverer a great deal, i.e., that there is in principle a large amount of evidence about morphological constituency that children could extract from simple distributional cues. Moreover, I show that the morphological parses assigned by the distribution-driven model to a set of potentially prefixed but semantically opaque words are correlated with morphological complexity ratings assigned to the same words by native English speakers. I argue that this convergence between the model and the speakers, in a domain in which speakers could not have relied on semantic cues, constitutes evidence that humans do indeed rely on distributional cues similar to the ones exploited by my model, when assigning morphological structure to words (see section 5 below for a full discussion of this argument). Notice that in the current project I am modeling morpheme discovery as a purely distribution-driven task because I am interested in trying to determine how much and what kind of information a learner could in principle extract
216
M. Baroni
from distributional evidence alone. I am not trying to argue that this is the only kind of evidence used by human learners. It is plausible that learners would use distributional cues at the earliest stages of morpheme discovery, since distributional information can be straightforwardly extracted from the data, and it can be exploited prior to any linguistic analysis. More sophisticated linguistic information can later be used to refine the coarse guesses on morphological structure made on the basis of distributional cues. The remainder of this study is organized as follows: In section 2, I shortly review some related work. In section 3, I present and discuss the computational model I am proposing. In section 4, I present the results of a simulation in which this model was tested, and I compare the morphological parses assigned by the model to morphological complexity ratings assigned by humans. In section 5, I compare the performance of the model to that of native speakers in parsing semantically opaque (but potentially complex) words. Finally, in the conclusion I briefly discuss future directions that this project could take.
2. RELATED WORK After the pioneering work of Harris in the fifties (see, e.g., Harris 1955) and until very recently, modeling morpheme discovery has been a relatively unpopular research domain. However, in the last few years there has been a resurgence of interest in the topic, and several supervised and unsupervised algorithms performing morpheme discovery or related tasks have been recently proposed: See, for example, most of the papers collected in Maxwell 2002, and the references quoted there. A recent approach that is closely related to the one I propose below is the one of Goldsmith 2001.3 Goldsmith’s model, like mine, is based on the idea that morphological segmentation can be re-phrased as a data compression problem. However, Goldsmith’s model differs from the one presented here in several respects. The most obvious (and probably least interesting) difference is that Goldsmith’s model looks for suffixation patterns, whereas my model focuses on prefixation. More importantly, Goldsmith’s model is based on an information-theoretically rigorous model of data compression constructed using the Minimum Description Length (MDL) principle of Rissanen 1978. The model proposed below, on the other hand, is only loosely inspired by the MDL idea, and the data compression scheme it assumes is not valid from an information-theoretic point of view (see Baroni 2000b: 3.3.7 on why I decided to abandon the more rigorous MDL-based approach I adopted in Baroni 2000a). Furthermore, Goldsmith’s model generates a single analysis using alternative heuristic strategies, and then uses the MDL criterion to refine such analysis.
Distribution-driven morpheme discovery
217
On the other hand, the model presented here uses heuristics to generate a set of alternative analyses, and then applies the maximal data compression criterion to choose the best of these alternatives. From a strictly linguistic point of view, Goldsmith’s model has two desirable properties that are missing in the current model, i.e. it can fully decompose words containing multiple affixes, and it groups stems and affixes into primitive forms of paradigms called signatures. Last but not least, Goldsmith’s main interest seems to lie in the possibility of extracting morphological analysis tools from unlabeled corpora using an automated procedure, whereas I developed the model I describe below because I am interested in testing some hypotheses about the role of distributional learning during human morphological acquisition. Thus, the focus here is less on the technical aspects of the model, and more on how the outputs it produces compare to human intuitions about morphological structure. Another model that is closely related to mine is the utterance segmentation method proposed in Brent and Cartwright (1996). Indeed, my algorithm can be seen as an adaptation of Brent and Cartwright’s lexicon selection and generation methods to the morpheme discovery problem. Thus, my algorithm takes words, rather than unsegmented utterances as input, and it returns maximally binary segmentations. Moreover, my model is biased so that it favors parses in which one element has affix-like distributional properties, and the other element has stem-like properties. Finally, Brent and Cartwright’s model, like the one proposed by Goldsmith, is based on an information-theoretically sound data compression scheme, whereas the model I propose below is only justified by the fact that it captures intuitively plausible morpheme-segmentation heuristics.
3. DDPL: AN AUTOMATED DISTRIBUTION-DRIVEN PREFIX LEARNER In order to assess the effectiveness of distributional heuristics in morpheme discovery, I designed and implemented a learning model which performs a particular aspect of this task – prefix discovery – on the sole basis of distributional evidence. The algorithm presented here takes a corpus of untagged orthographically or phonetically transcribed words as its input and outputs a lexicon composed of a list of prefixes and stems. Moreover, the algorithm assigns morphological parses (prefix+stem or monomorphemic parses) to all the word types in the input corpus.4 The algorithm relies entirely on the distributional information that can be extracted from the input. From here on, I will refer to the algorithm
218
M. Baroni
presented here with the acronym DDPL , which stands for Distribution-Driven Prefix Learner. DDPL is based on a ‘‘generation and selection’’ strategy: a large number of lexica compatible with the input data are generated, a certain measure is computed for each lexicon, and the lexicon with the lowest value of this measure is selected. The formula used to compute this measure constitutes the conceptual core of the algorithm, and it is based on the idea that the best morphological analysis of the input is also the one allowing maximal data compression of the input, given certain assumptions about how the data compression process should work. As we will see, the data compression scheme from which the DDPL lexicon selection formula is derived favors lexical analyses respecting the following three principles, which in turn can be interpreted as plausible morpheme discovering strategies: $
$
$
Substrings occurring in a high number of different words are likely to be morphemes; Substrings which tend to co-occur with other potential morphemes are more likely to be morphemes; All else being equal, low frequency words are more likely to be morphologically complex than high frequency words.
The first two principles should be fairly intuitive: Morphemes – especially affixes – tend to occur in a number of different words. Thus, they will tend to have a high type frequency. Moreover, real morphemes are not simply substrings that occur in a high number of random words, but rather substrings that can co-occur with other morphemes to form complex words. This explains the second principle. The third principle is perhaps less intuitive. Consider, however, the following. At one extreme, if a morphologically complex word is very frequent, the word is likely to have its own lexical entry, distinct from the entries of its component parts (at the very least, for reasons of ease of lexical access). However, once a word has an independent lexical entry, the word can acquire its own semantic features and thus it is likely to lose, over the course of time, its connection with its component parts. In other words, high frequency words are less likely to be morphologically complex because, even if they were complex from an etymological point of view, they will tend to acquire a lexicalized meaning due to heavy usage. At the other extreme, productively formed complex words must be hapax legomena (words with a token frequency of 1), or in any event have a very low token frequency. Indeed, Baayen has shown in several studies (see for example Baayen 1994, Baayen and Lieber 1991) that the number of hapax legomena containing a certain morpheme is a good indicator of the productivity of the
Distribution-driven morpheme discovery
219
morpheme. If a morpheme is productive, then the morpheme is often used to create new forms, and new forms, being new, are likely to have a very low frequency. Thus, all else being equal, it would make sense for a learner to be more willing to guess that a word is complex if the word has a low token frequency than if the word has a high token frequency. While in the following sections I will concentrate on how morpheme discovery can be rephrased as a data compression problem, the reader should keep in mind that I am by no means assuming that morpheme discovery is indeed a data compression problem (indeed, as I discuss in Baroni (2000b), the data compression method proposed here, when seen as an actual data compression algorithm, is far from optimal, and it is not truly implementable). Rather, the interest of the data compression approach lies in the fact that it allows us to derive an explicit formula that, as I will show, favors the same type of lexical analysis that would be favored by distribution-based morpheme discovery strategies such as the ones I just discussed.
3.1. Data compression and morphological analysis: the shortest lexicon+encoding criterion The criterion used by DDPL to select the best lexicon is based on the idea that the lexicon generated by the most plausible morphological analysis is also the best lexicon for purposes of data compression, given certain restrictions on how the compression procedure must work. The rationale behind this intuition is the following: Since morphemes are syntagmatically independent units, which occur in different words and combine with each other, a lexicon containing morphemes is going to be ‘‘shorter’’ (in the literal sense that it can be represented using a small number of characters) than a lexicon containing random substrings, or a lexicon in which no word is decomposed. The advantage of reducing the problem of morpheme discovery to a matter of (constrained) data compression is the following: There are no straightforward ways to decide which one, among a set of possible lexica, is the best one from the point of view of morphology, but it is relatively simple to estimate which lexicon allows maximal data compression. Given that the connection between data compression and morphological analysis is not very intuitive, I will illustrate it with a set of simple examples. Let us suppose that we are given a list of words, and our goal is to find a compact format to store information from which the very same list can be reconstructed. In particular, we take a ‘‘lexicon and encoding’’ approach to this task. We construct a compact lexicon from which all the input words (plus, possibly, others) can be reconstructed. We associate an index to each lexical
220
M. Baroni
unit, and then we rewrite the corpus as a sequence of these indices. I will refer to the rewriting of the corpus as a sequence of indices with the term encoding. As long as some lexical units occur very frequently in the input corpus (and/or many units occur relatively frequently), and the lexical indices are, on average, shorter than the units they represent, the lexicon + encoding strategy will allow us to represent the corpus in a shorter format than the original one. In order to make sure that the second requirement is satisfied (i.e., lexical indices are on average shorter than the input words they represent), I assume here that all indices are exactly one character long (I will revise this in 3.3 below). This is one of the main aspects in which the method presented here is not a realistic data compression scheme. The following example, which has nothing to do with morphological decomposition, is presented to give a first, general idea of how and why the lexicon and encoding strategy works. Suppose that we are given the following input corpus: (1) dog cat dog dog cat cat In order to write this list, we need 18 characters. Following the compression method described above, we can instead write the words dog and cat only once, assigning a one-character index to each of them (this is the lexicon component of the compressed data), and then rewrite the words in the input as a sequence of indices (the encoding): (2)
Lexicon dog 1
cat
2
Length of lexicon: 8 Encoding of (1) 1 (=dog) 1 (=dog) 2 (=cat) 1 (=dog)
2 (=cat) 2 (=cat)
Length of encoding: 6 Total length (lexicon + encoding): 14 To store the word types dog and cat in a lexicon, and then rewrite the word tokens in the input as a sequence of indices is more economical than writing down the list of input word tokens as it is (in the sense that it requires a smaller number of characters: 14 vs. 18). Notice that from the lexicon and the list of indices we can reconstruct the original input. Thus, we can store a corpus in the more economical lexicon and encoding format without any loss of information. Now, suppose that we are allowed to decompose input words into two constituents, in order to further compress the data. We assume the following
Distribution-driven morpheme discovery
221
encoding scheme: if an input word is identical to a lexical entry, then the input word is encoded using the index associated with that lexical entry (as in the example above); however, if a word does not have a corresponding entry, and must be reconstructed by concatenating two lexical units, then the word is encoded as the sequence of the index associated with the first component, a one-character concatenation operator (represented here by the symbol °) and the index associated with the second component. For example, suppose that a corpus contains the word redo. If this word is listed in the lexicon, for example associated with the index 1, then the word is represented by a 1 in the encoded input. However, if redo is not listed in the lexicon, and it has to be reconstructed from the entries re, associated with the index 1, and do, associated with the index 2, then the word will be represented by the sequence 1°2 in the encoded corpus. While it can be convenient to store frequent substrings in the lexicon, in order to make the lexicon shorter, there is a counterbalancing factor, namely the length of the encoding. On the one hand, treating substrings occurring in a number of words as independent lexical entries will make the lexicon shorter. On the other hand, since it takes three characters (two indices plus the concatenation operator) instead of one to encode an input word not listed in the lexicon, any decomposition which makes the lexicon shorter will also make the encoding longer. Thus, only decompositions allowing a decrease in lexical length which more than compensates for the corresponding increase in encoding length are worth performing. From the point of view of morpheme discovery, this trade-off ensures that the only decompositions that will be performed will be those motivated by strong distributional evidence. For example, compare the shortest lexicon + encoding representations of the lists in (3) vs. (6). Consider first the list in (3): (3) redo do remake undo make unmake The shortest lexicon + encoding representation of this list is the following: (4)
lexicon re 1 un 2
do 3 make 4
length of lexicon: 14 encoding of (3) 1°3 (=re°do) 3 (=do)
1°4 (=re°make) 2°3 (=un°do)
length of encoding: 14 total length (lexicon + encoding): 28
4 (=make) 2°4 (=un°make)
222
M. Baroni
In particular, this is a shorter representation than the one in which no decomposition is attempted: (5)
lexicon redo 1 do 2
remake undo
3 4
make 5 unmake 6
length of lexicon: 32 encoding of (3) 1 (=redo) 3 (=remake) 2 (=do) 4 (=undo)
5 (=make) 6 (=unmake)
length of encoding: 6 total length (lexicon + encoding): 38 The representation in (4), which is based on a plausible morphological decomposition of the input, is ten characters shorter than the representation in (5), where no decomposition is attempted. The reason for this is that the analysis of the input upon which (4) is based provides a very compact lexical component, since the units re, un, do and make are all morphemes which occur in at least two input words. Thus, even if the encoding in (4) is longer than the encoding in (5), the lexicon in (4) is so much shorter than the one of (5) that, overall, (4) is the analysis to be selected on the basis of the shortest lexicon + encoding criterion. However, consider now the case of the input in (6): (6) dog tag mug This time, we must choose the lexical analysis of (7), in which no word is decomposed: (7)
lexicon dog 1 tag 2 mug 3 length of lexicon: 12 encoding of (6) 1 (=dog) 2 (=tag) 3 (=mug) length of encoding: 3 total length (lexicon + encoding): 15
The total length of lexicon and encoding in (7) (where the input words are not decomposed) is shorter than the one of (8), where g is treated as an independent lexical unit: (8)
lexicon do 1 mu 3 ta 2 g 4 length of lexicon: 11
Distribution-driven morpheme discovery
encoding of (6) 1°4 (=do°g)
2°4 (=ta°g)
223
3°4 (=mu°g)
length of encoding: 9 total length (lexicon + encoding): 20 While the lexicon in (8) is shorter than the one in (7), the small decrease in lexical length does not justify the increase in encoding length due to the fact that one needs three characters to represent each word not listed as an independent unit in the lexicon. As the previous examples show, the shortest lexicon + encoding criterion favors the representation of substrings as lexical entries only when this approach leads to considerable savings in lexical length. True morphemes, unlike arbitrary substrings, are more likely to lead to such savings, and, thus, to be treated as independent lexical entries.
3.2. Capturing the morpheme discovery heuristics In this section, I present a series of related examples which have the function of illustrating how the shortest lexicon+encoding approach principle favors lexical analyses that are also optimal from the point of view of the distributional morpheme discovery heuristics that I listed in section 3.5 3.2.1. The high frequency heuristic The first example I present shows how the first heuristic (‘‘frequent substrings are more likely to be morphemes’’) is captured by the shortest lexicon + encoding criterion. Consider the data sample in (9): (9) disarray disdain disintegrate disadvantage disaster The analysis in which the word-initial substring dis is treated as an independent lexical entry (10a) allows a more compact lexicon+encoding representation than the analysis in which the words are not decomposed (10b) (10)
a.
lexicon dis 1 integrate array 2 advantage dain 3 aster length of lexicon: 40
4 5 6
224
M. Baroni
encoding of (9) 1°2 (=dis°array) 1°3 (=dis°dain) 1°4 (=dis°integrate)
1°5 (=dis°advantage) 1°6 (=dis°aster)
length of encoding: 15 total length (lexicon + encoding): 55 b. lexicon disarray 1 disdain 2 disintegrate 3
disadvantage disaster
4 5
length of lexicon: 52 encoding of (9) 1 (=disarray) 2 (=disdain) 3 (=disintegrate)
4 (=disadvantage) 5 (=disaster)
length of encoding: 5 total length (lexicon + encoding): 57 The analysis in (10.a) is shorter than the analysis in (10.b) simply in virtue of the fact that dis is a frequent word-initial string (of course, what counts as ‘‘frequent’’ depends in general on the size of the input corpus – in this and the following sections we analyze very small corpora, and, as a consequence, strings occurring in a small number of words will behave as frequent). The lexicon which would be favored by the ‘‘frequent strings are morphemes’’ heuristic is also the lexicon favored by the shortest lexicon + encoding criterion. However, notice also that the difference in length between the analyses is rather small (55 vs. 57). Indeed, it is sufficient to remove one word from the list in (9) ... (11) disarray disdain disintegrate disadvantage ... and the analysis in which dis is treated as an independent entry is no longer shorter than the analysis in which the input words are not decomposed: (12)
a.
lexicon dis 1 integrate array 2 advantage dain 3 length of lexicon: 35
4 5
Distribution-driven morpheme discovery
encoding of (11) 1°2 (=dis°array) 1°3 (=dis°dain)
225
1°4 (=dis°integrate) 1°5 (=dis°advantage)
length of encoding: 12 total length (lexicon + encoding): 47 b. lexicon disarray 1 disdain 2
disintegrate disadvantage
3 4
length of lexicon: 43 encoding of (11) 1 (=disarray) 2 (=disdain)
3 (=disintegrate) 4 (=disadvantage)
length of encoding: 4 total length (lexicon + encoding): 47 As this example shows, the ‘‘store frequent strings as independent units’’ principle plays a secondary role in our data compression scheme. This is also reasonable when we look at our approach as a way to select the best lexicon from a morphological point of view: Morphemes are not simply frequent substrings, but, as the second heuristic stated above says, substrings that frequently co-occur with other syntagmatically independent substrings (i.e., other morphemes) to form complex words. As the following section shows, lexical analyses respecting this heuristic lead to shorter representations than the ones obtained by simply treating frequent substrings as independent units.6 3.2.2. Co-occurrence with other potential morphemes As I just stated, a sensible morpheme discovery strategy should not simply be based on absolute frequency, but on the number of times a string tends to co-occur with other ‘‘potential morphemes’’, i.e., strings also occurring elsewhere in the corpus. In the lexicon + encoding model, independent lexical entries for strings which tend to combine with other independently occurring strings lead to larger savings than simply treating frequent strings as lexical entries. Consider first the sample in (13): (13) redo go remake replug dogs Even if the string re occurs three times in this corpus, the representation in
226
M. Baroni
which this string is not treated as an independent lexical unit (14.b) is shorter than the one in which the words beginning with re are decomposed (14.a): (14)
a.
lexicon re 1 do 2
go 3 make 4
plug 5 dogs 6
length of lexicon: 24 encoding of (13) 1°2 (=re°do) 1°5 (=re°plug) 3 (=go) 6 (=dog) 1°4 (=re°make) length of encoding: 11 total length (lexicon + encoding): 35 b. lexicon redo 1 go 2
remake replug
3 4
dogs 5
length of lexicon: 27 encoding of (13) 1 (=redo) 2 (=go) 3 (=remake)
4 (=replug) 5 (=dogs)
length of encoding: 5 total length (lexicon + encoding): 32 Compare now the input in (13) with the input in (15): (15) redo do remake sprint make The two inputs are similar in that they are both composed of six words of length 4, 2, 6, 6 and 4 respectively. Moreover, the two words containing re in (15) are also present in (13). However, on the one hand the string re only occurs two times in (15) (vs. three times in (13)); on the other hand, in (15) the string re occurs before strings which also occur elsewhere in the list (both do and make also occur as independent words). In this case, the analysis in which re is treated as an independent lexical entry (16.a) is much shorter than the analysis in which the words beginning with re are not decomposed (16.b):
Distribution-driven morpheme discovery
(16)
a.
lexicon re 1 do 2
227
make 3 sprint 4
length of lexicon: 18 encoding of (15) 1°2 (=re°do) 2 (=do) 1°3 (=re°make)
4 (=sprint) 3 (=make)
length of encoding: 9 total length (lexicon + encoding): 27 b. lexicon redo 1 remake do 2 sprint length of lexicon: 27 encoding of (15) 1 (=redo) 2 (=do) 3 (=remake)
3 make 5 4
4 (=sprint) 5 (=make)
length of encoding: 5 total length (lexicon + encoding): 32 The examples from (13) to (16) show how, in our model, strings which co-occur with other strings that also independently occur in the corpus are more likely to be treated as independent units than strings which simply frequently occur in the corpus, even if the latter are more frequent, in absolute terms, than the former.7 The reason for this preference is that, if both elements forming a complex word are already stored in the lexicon, then there is no need to store any extra item in order to be able to reconstruct the word. On the other hand, if only one of the constituents of a word is already stored in the lexicon, we still need to create a lexical entry, complete with an index, for the remainder of the form. 3.2.3. Word frequency and morphological complexity We observed that, all else being equal, learners should be more willing to treat words as morphologically complex if they rarely occur in the corpus than if they are frequent. Again, there is a parallel with the shortest lexicon + encoding approach to data compression, as the following examples show. Consider first the input in (17):
228
M. Baroni
(17) disarray array disobey obey Of course, the shortest analysis of this input is the one in which both disarray and disobey are decomposed (18.a) (both analyses presented here are shorter than the one in which disobey is not decomposed): (18)
a.
lexicon dis 1 array 2
obey 3
length of lexicon: 15 encoding of (17) 1°2 (=dis°array) 2 (=array)
1°3 (=dis°obey) 3 (=obey)
length of encoding: 8 total length (lexicon + encoding): 23 b. lexicon disarray 1 array 2
dis 4 obey 4
length of lexicon: 24 encoding of (17) 1 (=disarray) 2 (=array)
3°4 (=dis°obey) 4 (=obey)
length of encoding: 6 total length (lexicon + encoding): 30 However, consider now the input in (19), which is identical to (17), except that I added four more tokens of the word disarray. (19)
disarray array
disobey
obey disarray disarray disarray disarray
Now, the best analysis becomes the one in which disarray is not decomposed into dis and array (20.b) (both analyses are shorter than the one in which neither disarray nor disobey are decomposed): (20)
a.
lexicon dis 1 array 2
obey 3
length of lexicon: 15
Distribution-driven morpheme discovery
encoding of (19) 1°2 (=dis°array) 2 (=array) 1°3 (=dis°obey) 3 (=obey)
1°2 1°2 1°2 1°2
229
(=dis°array) (=dis°array) (=dis°array) (=dis°array)
length of encoding: 20 total length (lexicon + encoding): 35 b. lexicon disarray 1 array 2
dis 3 obey 4
length of lexicon: 24 encoding of (19) 1 (=disarray) 2 (=array) 3°4 (=dis°obey) 4 (=obey)
1 1 1 1
(=disarray) (=disarray) (=disarray) (=disarray)
length of encoding: 10 total length (lexicon + encoding): 34 These examples show how, all else being equal, in the lexicon + encoding model more frequent words are more likely to be stored in the lexicon in nondecomposed format than less frequent words. The only difference between the distribution of disarray in (17) and (19) is that in (19) this word occurs five times, whereas in (17) it occurs only once. Because of this difference in frequency, disarray gets its own lexical entry in the shortest analysis of (19), whereas in the shortest analysis of (17) it must be reconstructed from the components dis and array. The reason why this model favors independent storage of frequent words, even when both of their components are also specified in the lexicon, is the following: Given that each occurrence of a decomposed word in the encoded corpus requires three indices instead of one, if a word occurs frequently in the corpus, it is more convenient to use some characters to build a lexical entry for it, in order to have an economical way to encode it. On the other hand, if a word is rare, it is more convenient to save the characters required to represent the word in the lexicon, and encode the word in a costly way the few times in which it occurs in the corpus. As observed by a reviewer, the model also predicts that longer words will be more resistant to independent storage, since, being longer, more characters are needed to store them in the lexicon. Future research should try to assess to
230
M. Baroni
what extent a heuristic along these lines (longer words are more likely to be morphologically complex) is empirically founded. The example (19)/(20.b) also illustrates an interesting general property of the DDPL model: Not only is this model flexible enough to allow words to be treated as morphologically simple (i.e., represented in the lexicon as independent units), even if they begin with a substring which is specified as a prefix in the lexicon, but words can be represented as units in the lexicon even if both their component parts are also lexical entries: in the case at hand, both the word disarray and its constituents dis and array are listed in the lexicon. As it is a common feature of many models of lexical-morphological processing (see Schreuder and Baayen 1995 and the other models reviewed there) to assume that words can have an independent lexical representation even if they could be entirely derived from morphemic constituents also stored in the lexicon, I believe that it is a desirable property of our learning model that it can select lexica in which this situation arises.
3.3. Prefix-stem asymmetries in the DDPL model The actual formula used to estimate lexicon + encoding length in the DDPL model assumes a representation strategy slightly different from the one presented in the examples above, in order to account for the fact that prefixes and stems have different distributional properties. In particular, affixes tend to be more frequent units than stems. Given a prefixed input word, it is very likely that its prefix also occurs in a number of other input words, whereas the stem probably only occurs in very few other words. For example, it is plausible that in a corpus of English containing the form reconsider, the prefix re also occurs in hundreds of other words, whereas the stem consider only occurs in this prefixed form, as an independent word and in no more than five or six other derived forms. In order to take the prefix-stem asymmetry into account, a bias against prefixes has been introduced in their lexical representation: prefixes are associated with indices that are slightly longer than the ones assigned to stems. Specifically, prefixes are associated with indices that are 1.25 characters long, whereas stems are associated with indices that are one character long. The value 1.25 was determined empirically, by fitting against the data described in section 4 below. The length of the index needs not be an integer, since it simply represents an arbitrary penalty, and, for our purposes, it does not have to correspond to a plausible encoding scheme. The effect of this representational bias is that prefixes have to be more frequent (and/or more frequently co-occur with potential morphemes) than stems in order to be represented as independent lexical units in the shortest analysis of the input.
Distribution-driven morpheme discovery
231
An important consequence of the bias against infrequent prefixes is that it disfavors analyses in which stems of suffixed forms are mistakenly treated as prefixes. Consider for example the word lovely. Given that, in a reasonably sized English corpus, both love and ly probably also occur in other forms, there is the risk that DDPL could treat love as a prefix and ly as a bound stem. However, compared to a real prefix, love- is likely to occur in a very limited number of forms: for example, in the PHLEX database (see section 4 below) the wordinitial substring love only occurs in a total of 8 forms, whereas even a rare prefix such as para- occurs in 22 words. Thus, given the anti-infrequent-prefix bias, it is unlikely that in the shortest analysis of an input corpus a string such as love will be actually treated as a prefix. Of course, this bias only makes sense in a morpheme-discovery model, such as the one I am assuming, in which prefix- and suffix-discovery are conducted in separate stages (see Baroni 2000b:3.2.1.2 for discussion). 3.4. Finding the best solution The discussion above has laid out the criteria by which the DDPL defines the optimum analysis of the data submitted to it. In addition, it is necessary to find an efficient procedure whereby this solution can be located, through generation and selection. For a detailed description of the algorithm used to generate alternative lexical analyses of the input corpus, see Baroni (2000b:3.3.9). To summarize briefly, the algorithm is based on a greedy strategy similar to the one proposed by Brent and Cartwright (1996) for sentence segmentation. A lexicon constructed with n morphological splits is only evaluated if, of those n splits, n-1 are identical to the ones that were used to generate the shortest lexicon constructed with n-1 splits. In other words, in the exploration of possible, increasingly complex morphological analyses, morpheme breaks can only be assigned, never removed. Moreover, the DDPL lexicon generation strategy uses a series of heuristics to further constrain the set of lexical analyses to be evaluated. In particular, the number of lexica to be evaluated is strongly reduced by the requirement that only word-initial strings that frequently occur in the input before relatively long word-final strings can be treated as prefixes in a DDPL lexicon. The alternative lexical analyses generated in this way are compared, and the one allowing the shortest lexicon + encoding is selected as the best analysis. The length of a lexicon and the corresponding encoding are estimated using the following formula: (21) Formula for estimating lexicon + encoding length dl=
∑
length(ent)+2|stem_entries|+2.25|prefix_entries|
entµentries
+|stem_occurrences|+2.25|prefix_occurrences|
232
M. Baroni
dl: entµentries: length(ent): |stem_entries|: |prefix_entries|: |stem_occurrences|: |prefix_occurrences|:
description length; any entry, prefix or stem, in the DDPL lexicon; length in characters of an entry; total number of entries in the stem list; total number of entries in the prefix list; total number of occurrences of all stem entries in the input corpus; total number of occurrences of all prefix entries in the input corpus.
See Baroni (2000b:3.3.8) for an explicit derivation of (21). In brief, the need to minimize the first three terms of (21) favors distributionally motivated decompositions (as such decompositions will reduce the number of entries in the prefix and stem lists, and they will reduce the lengths of lexical entries), whereas the last two terms will disfavor decompositions (after each decomposition, the total number of occurrences of entries in the encoding increases). The extra weights added to the two prefix-specific factors will insure that, all else being equal, more distributional evidence is needed to postulate a prefix than to postulate a stem. Given that the formula in (21) computes the length of a certain analysis as if it were represented in the lexicon + encoding format, the very same analyses that are selected by the shortest lexicon + encoding criterion (and, thus, by the morpheme discovery heuristics presented in section 3) are also selected when using (21) to select the best lexicon (to be precise, the encoding scheme from which (21) is derived differs from the one described above in that it assigns an extra-character to each entry in a lexical analysis, in order to mark which entries are prefixes and which entries are stems).
4. DISCOVERING ENGLISH PREFIXES WITH THE DDPL MODEL The performance of DDPL was tested with a corpus of untagged orthographically transcribed English words from the PHLEX database (Seitz, Bernstein, Auer and MacEachern 1998) as its input. In this section, I will present the results of the simulation. The PHLEX database contains, among other word lists, a list of the 20,000 most common word types in the Brown corpus (Kucera and Francis 1967), together with their frequency of occurrence in the Brown corpus. I removed from this list all word-types containing non-alphabetic symbols (digits and diacritics such as -). This trimming yielded a set composed of 18,460 word types. The input corpus for the DDPL simulation was generated by multiplying each of the types in this set by its frequency in the Brown corpus. For example, the word kindergarten has a frequency of 3 in the Brown corpus, and thus it occurred
Distribution-driven morpheme discovery
233
three times in the DDPL input used for the simulation presented here. In total, the corpus generated in this way contained 959,655 orthographically transcribed word tokens. 4.1. Prefixes discovered by the DDPL Given the input described in the previous section, DDPL generated an output lexicon containing the following 29 prefixes: (22) prefixes postulated by DDPL adextrapreun-
autojurispsychounder-
cocomininterradio- re-
con- corman- missub- sup-
denonsuper-
disoversur-
exparatele-
A first inspection of this list shows that DDPL was quite successful and accurate in finding (almost) only actual English prefixes. Notice that com-, con- and cor- are actually allomorphs of the same prefix, and so are sub- and sup- (cf. suppress). The purpose of the DDPL model is simply to find the list of strings that correspond to prefixes (and stems) of a language. The model does not attempt to group strings that are allomorphs of the same morpheme into the same entry. I do not think that allomorph grouping is a task which should be performed on the sole basis of distributional cues, as syntactic, semantic and phonological cues would, obviously, be of great help. It is interesting to observe that the DDPL model was able to find prefixes that constitute substrings of other prefixes: co- is a substring of com-/con-/cor-; ex- is a substring of extra-; in- is a substring of inter-; un- is a substring of under-. The list contains only one obvious false positive: the string man-, which is a full noun, not a prefix. As man- often occurs as the first member of compounds, and DDPL does not have access to (morpho-)syntactic information, the model mistakenly treated forms such as manservant and manslaughter as prefixed. Besides man-, there are three ambiguous cases: The strings juris-, radioand psycho- are not classified as prefixes in standard references such as Marchand (1969) or Quirk, Greenbaum and Svartvik (1985) (the MerriamWebster’s dictionary classifies radio- and psycho- as ‘‘combining forms’’). Thus, we should perhaps count them as false positives. However, as these strings correspond to bound word-initial units associated with specific semantic features, even if they might not be prefixes under some definition of what a prefix is, they are ‘‘prefix-like’’ enough that I am reluctant to classify them as ‘‘real’’ false positives. Interestingly, the DDPL did not classify any frequent but linguistically insignificant word-initial string (strings such as pa-, pr- ...) as a prefix.
234
M. Baroni
It seems legitimate, I think, to claim that the DDPL was very accurate in avoiding false positives. On the other hand, the following 19 prefixes listed in Quirk et al. (1985) were not discovered by DDPL (notice that il-, im- and ir- are allomorphs of in-, a prefix that was found by DDPL): (23) prefixes missed by DDPL aanhyper- ilpostpro-
antiimpseudo-
arch- contra- counter- foreirmalminiouttrans- ultra-
Neo-classical prefixes (such as hemi- and paleo-) and conversion prefixes (such as en- and be-) listed in Quirk et al. (1985) but missed by DDPL are not reported in (23). I believe that misses from these classes are not problematic, as they concern prefixes that are very cultivated and/or not very productive. Of the set in (23), the following 6 misses are due to the nature of the input corpus, which did not contain enough forms to motivate their treatment as prefixes: (24)
a-
an- arch- hyper- mini-
pseudo-
The string hyper- never occurs in the corpus, whereas the other five strings in (24) only occur in two or fewer words in which they function as prefixes (even counting completely semantically opaque, highly lexicalized prefixed forms). Of the remaining 13 misses, the following 9 are due to the heuristic constraint on lexicon generation mentioned in 3.4 above, according to which only word-initial strings occurring a certain number of times before long word-final strings can be evaluated as candidate prefixes: (25)
counter- fore- il- im-
ir-
mal-
out-
post-
ultra-
None of these prefixes occurs frequently enough before long word-final strings in the input to avoid being filtered out by this constraint. This leaves us with 4 unexplained misses: (26)
anti- contra- pro-
trans-
The input corpus contains several truly prefixed forms displaying these prefixes in combination with independently occurring stems (although the prefixes contra- and pro- only occur in lexicalized formations with bound stems, such as contraception and proceed). Thus, these misses cannot be attributed to the nature of the input. I plan to explore the issue of why DDPL failed to discover these prefixes in future research.
Distribution-driven morpheme discovery
235
To conclude, the list of prefixes found by DDPL is more accurate than exhaustive, i.e., we can say that the model is better in terms of precision than in terms of recall. On the one hand, the false positives in the list are few and linguistically motivated. On the other hand, even if the prefixes in (24) are excluded from the count, DDPL still missed 13 productive English prefixes. In particular, most of these misses are due to the constraint on lexicon generation requiring word-initial strings to occur a certain number of times before ‘‘long’’ word-final strings. This suggests that the first step in future revisions of DDPL should concentrate on the lexicon generation component of the model.
4.2. Assessing the model against native speaker intuition Besides finding a list of prefixes, DDPL assigns morphological parses (prefixed vs. monomorphemic) to all the words in the input corpus. Of course, only the parses assigned by DDPL to potentially prefixed words, i.e., words beginning with a string identical to one of the prefixes found by the algorithm, are of interest, as all other input words are treated as monomorphemic. Assessing the plausibility of the parses assigned by DDPL to potentially prefixed words is not a trivial task, as the morphological status of many potentially prefixed words is not clear. While probably everybody would agree that the word redo is prefixed and the word red is not, there are many intermediate cases (such as resume, recitation, remove) about whose status morphologists would probably disagree. Thus, rather than trying to decide on my own which of the parses assigned by DDPL were ‘‘right’’ and which ones were ‘‘wrong’’, I conducted a survey in which I collected morphological complexity ratings from native English speakers, and then I tested whether the parses assigned by DDPL to the same words could predict the distribution of such ratings (see Smith 1988, Wurm 1997 and Hay 2000 for other studies using morphological complexity ratings). The idea was to see if speakers’ intuitions on the prefixed status of such words would agree with the parses assigned by the algorithm. 4.2.1. Word list construction, methodology and data collection All the words in the survey corpus begin with a string corresponding to one of the prefixes postulated by DDPL, but half of the words were selected from the set of forms that were treated as complex by the model, the other half from the set of forms which, although they begin with a string identical to a prefix, were not treated as complex by the model. In particular, the survey corpus contained 300 forms that were randomly
236
M. Baroni
selected from the words in the DDPL output that began with one of the prefixes postulated by the algorithm (excluding man-, juris- and radio-). 150 of these forms were randomly selected from the set of words that DDPL treated as prefixed. The other 150 forms were randomly selected from the set of words that DDPL treated as monomorphemic (non-prefixed). The complex-for-DDPL set contained 22 distinct prefixes, the simple-forDDPL set contained 21 distinct prefixes. The two sets shared 17 prefixes. The average length of the words in the complex-for-DDPL set was 9.9 characters, the average length of their potential stems was 7.1 characters. The average length of the words in the simple-for-DDPL set was 9 characters, the average length of their potential stems was 6.5 characters. The average frequency of the words in the complex-for-DDPL set was 3.6, the average frequency of their potential stems was 192.9. The average frequency of words in the simple set was 25.7, the average frequency of their potential stems was 189.2. All words in the complex set had potential stems occurring as independent strings in the corpus; 101 words in the simple set had potential stems that did not occur as independent strings. The word lists used in this and the following survey, together with DDPL parses, average speaker ratings for each word and other statistics are available from http://sslmit.unibo.it/~baroni. A group of eight native English speakers were asked to rate the set of potentially prefixed words (‘‘potentially prefixed’’ in the sense that they begin with a word-initial string identical to a prefix) on a scale from 1 to 5, assigning 1 to words that they definitely felt to be non-prefixed, 5 to words that they definitely felt to be prefixed. In the instructions, the participants were presented with an example of a word that should receive a 1-rating (the word cocoon) and an example of a word that should receive a 5-rating (the word coexist). The 300 words in the survey corpus were presented in a different random order to each participant. Words were presented in list format. To avoid possible ambiguities due to the fact that some of the prefixes under investigation are substrings of other prefixes (e.g., in- is a substring of inter-), each word in the list was followed by the potential prefix that participants were to consider. Participants were given unlimited time to complete the task, but they were asked to write down their ratings as quickly as possible, and to avoid revising a rating once they had written it down. Clearly, in order to take part in the survey, participants had to be familiar with basic notions of morphology (at least, the notions of prefix and prefixed form). Thus, I selected the participants among undergraduate and graduate linguistics students (the participants in the surveys, however, were not aware of the goals of the study).
Distribution-driven morpheme discovery
237
4.2.2. Results and discussion The rating patterns of all eight participants were highly correlated (the Pearson correlation coefficients computed pairwise for each pair of raters were always higher than 0.55; the Spearman correlation coefficients computed pairwise for each pair of raters were always higher than 0.6.). Thus, I computed the per-word average rating across all participants, and I conducted a one-way ANOVA in which these ratings were grouped on the basis of the corresponding DDPL parses (simple vs. complex). The difference between ratings assigned by native speakers to words treated as simple by DDPL and ratings assigned to words treated as complex by DDPL is highly significant (F(1,298)=209.7, p<0.0000). Thus, we can conclude that, besides being able to find a number of actual English prefixes, DDPL also assigned plausible morphological parses to potentially prefixed words. Of course, while the relation between DDPL and the speakers’ ratings is significant, it is by no means ‘‘perfect’’, as there are some discrepancies between the speakers’ ratings and the DDPL parses. Several reasons can explain these discrepancies. First, as there are individual differences in morphological intuitions among speakers (indeed, the ratings of some of the speakers appear to be less correlated with each other than DDPL and the speakers’ average), even if DDPL were a perfect model of human morpheme discovery, we should not expect a 100% correlation between its parses and an average of human intuitions. More importantly, the DDPL is intended to model a hypothesized early stage of language acquisition, in which morpheme discovery is performed. However, the speakers who participated in the survey are adults who successfully completed the task of morphological acquisition, and are aware of the semantic and syntactic properties associated with prefixes and stems. From this perspective, it is actually surprising that the purely distributionally-driven parses assigned by DDPL are as well correlated with adult speakers’ ratings as the results indicate. Interestingly, the discrepancies between DDPL and English speakers appear to be attributable to the fact that DDPL is too ‘‘conservative’’, i.e., DDPL was more likely to treat obviously prefixed forms as simple than obviously simple forms as complex. While it is easy to find obvious misses among the forms treated as simple by DDPL (unconsciously, distrust, subgroups, unavoidable ...), only two of the forms treated as complex by DDPL are obviously non-prefixed (comin – probably a spelling of the colloquial form of coming – and constable). Indeed, the average mean rating across all forms that were treated as complex by DDPL is a rather high 4.05 (recall that speakers had to rate forms on a scale from 1 to 5, assigning 5 to clearly prefixed forms). This indicates that in general speakers largely agree with DDPL on the status of forms that the
238
M. Baroni
algorithm treated as complex. On the other hand, the average mean rating across all forms that were treated as simple by DDPL is 2.11. This is still lower than chance level, but does suggest that there was less overlap between DDPL parses and speakers’ intuitions in the domain of forms that are treated as simple by the computational model. As with the list of prefixes found by DDPL, what emerges here is that the analysis generated by the model is quite accurate (very few ‘‘false positives’’) but not exhaustive (many ‘‘misses’’). Precision is high, recall is low. I observed in the introduction that the purpose of assigning parses to potentially complex words in morpheme discovery is to have a set of forms to analyze in order to discover the semantic and grammatical properties of affixes. In this perspective, it seems that morpheme discovery should indeed favor precision over recall: a relatively small set of words containing a certain prefix is probably more helpful, in identifying the properties of that prefix, than a larger set that also includes many pseudo-prefixed forms (see Snover and Brent 2001 for a similar line of reasoning).
5. TREATMENT OF SEMANTICALLY OPAQUE WORDS The analysis of the results of the DDPL simulation shows that distributiondriven principles such as the ones implemented by this model can be quite helpful in morpheme discovery, both in terms of finding the prefixes of a language and in terms of assigning morphological parses to words. The success of this computational simulation constitutes evidence against the claim that children cannot in principle learn something about morphology from distributional evidence, the claim being that distributional evidence would not provide enough useful cues. Clearly, even the relatively simple distributional cues used by DDPL might be of great help to language learners. However, this does not per se constitute evidence that humans do rely on such cues, since humans, unlike the automated learner, could have used different types of evidence – most plausibly, semantic evidence – in order to discover the same structures found by the automated learner. For example, the learner, exploiting distributional cues only, came to the conclusion that renamed is a prefixed word, composed of the prefix re- and the stem named. Although all the native speakers surveyed shared the intuition that renamed is indeed a prefixed form composed of re- and named, this convergence between the automated learner and humans does not prove that humans exploited distributional cues in morpheme discovery, since humans could have decided that renamed is prefixed simply on the basis of its meaning. However, the comparison of the output parses assigned by the automated prefix learner to English words with morphological intuitions of native speakers can potentially provide a form of more direct empirical evidence supporting the
Distribution-driven morpheme discovery
239
hypothesis that learners resort to distributional cues in morpheme discovery. This evidence would emerge from the analysis of semantically opaque but potentially morphologically complex words such as recitation or remain. Words of this kind are potentially prefixed, at least in the sense that they begin with a string identical to a prefix (re-, in this case). However, the meaning of recitation is not synchronically related to the meaning of the prefix re- nor to the meaning of the stem citation (or cite). In the case of remain, not only is the meaning of the word not related to the meaning of the components, but it is not even clear that the potential stem, the bound verbal form -main, is associated with any semantic content. However, several experimental studies have shown that speakers do treat some of these semantically opaque forms as morphologically complex. For example, the following studies have presented evidence from a variety of experimental tasks that speakers are aware of the morphological structure of words that are (partially or completely) semantically opaque: Emmorey (1989), Bentin and Feldman (1990), Stolz and Feldman (1995) who quote unpublished research by Feldman and Stotko, Roelofs and Baayen (2002), Baroni (2001) and Baayen, Schreuder and Burani (to appear). Now, if it turned out that there is a convergence between the parses assigned by the distribution-driven learner and the speakers’ intuitions about semantically opaque forms, then this would constitute a stronger form of evidence in favor of the hypothesis that speakers used distributional cues to assign morphological structure to words. For example, if it turned out that both the automated learner and the speakers treated recitation as morphologically complex (re+citation), but remain as monomorphemic, then it would be reasonable to conclude that speakers are sensitive to distributional cues similar to the ones implemented in the automated learner, since they could not have assigned a morphological structure to recitation on the basis of its meaning (and, also, it is unlikely that they could have used syntactic or phonological cues to distinguish recitation from remain). Indeed, the results of the second survey I conducted show that, even when only semantically opaque words are considered, there is a significant correlation between the parses assigned by the learner and speakers’ intuitions. Thus, this study provides strong support for the claim that humans use distributional cues in morpheme discovery. Notice that this type of evidence in favor of distribution-driven learning is not available in other domains. For example, even if it has been shown that distributional cues can be very effective for segmenting utterances into words (Brent and Cartwright 1996), there is no clear equivalent to semantically opaque morphemes in the domain of syntactic segmentation. In particular, idiomatic phrases such as kick the bucket are not the equivalent of semantically opaque morphologically complex forms such as permit. First, idiomatic phrases also have a literal, semantically transparent meaning,
240
M. Baroni
and it is unlikely that speakers are not aware of this meaning. Second, words occurring in idioms also occur in non-idiomatic sentences. This is not the case of a bound stem like -mit, which occurs only in opaque forms.
5.1. Constructing the semantically opaque word list Following a standard practice in morphological processing studies (see, for example, Marslen-Wilson, Tyler, Waksler and Older 1994), I first conducted a survey in which three judges were asked to rate a set of forms from the DDPL output for semantic transparency, and then I selected forms that received a low average semantic transparency rating to construct the survey corpus. The DDPL output contains a total of 3,651 forms beginning with strings corresponding to one of the prefixes postulated by the model. Of these, 382 are actually treated by the model as prefixed. Clearly, it was not feasible to ask the semantic transparency judges to assign a rating to all 3,651 forms. Thus, the corpus presented to the judges was constructed in the following way. First, I made a preliminary division of the 382 words treated as prefixed by DDPL into two categories: words that I judged to be obviously prefixed (productively formed, semantically transparent), and words that may or may not be prefixed (this preliminary list included a wide range of types, from obviously non-prefixed words such as adage to only slightly lexicalized forms such as inhumane). The first list was composed of 101 words, the second list of 181 words. I randomly selected 10 words from the first list, and I kept all the 181 words from the second list. From the list of the remaining 3,269 words treated as simple by DDPL, I then randomly selected 10 more words that were obviously prefixed and completely transparent, and 200 words that may or may not be prefixed. The corpus presented to the three judges was composed of the 20 completely transparent words and 381 ‘‘ambiguous’’ words selected in this way. The 20 completely transparent words served both as a control and, more importantly, they were added in order to try to minimize the risk that judges would assign high ratings to some semantically opaque forms merely in order to make use of the whole range of the rating scale. The judges were two graduate students and one postdoctoral fellow in the UCLA Linguistics Department, and were selected because of their strong background in morphology and morphological processing. I selected expert judges because I wanted to make sure that they would understand the task, and in particular that they would understand the distinction between rating forms on the basis of semantic transparency vs. morphological complexity. Judges were asked to rate the words in the corpus on a scale from 1 to 5, assigning 1 to completely opaque words and 5 to completely transparent words.
Distribution-driven morpheme discovery
241
A series of correlation analyses showed that the judges’ ratings were highly correlated (both Pearson and Spearman correlation coefficients in all pairwise comparisons were higher than 0.7). Thus, I computed the average cross-judge rating for each word in the corpus. As expected, the 20 transparent words received very high ratings (the mean rating for this set of words was 4.89). Of the remaining forms, 97 out of the 181 words treated as prefixed by DDPL received an average rating lower than 2.5; 183 out of the 200 words treated as simple by DDPL received an average rating lower than 2.5. Notice the asymmetry between the two sets: just a little more than half of the complex-for-DDPL words that were pre-selected as potentially opaque are indeed semantically opaque, whereas 90% of the simple-for-DDPL words that were pre-selected as potentially opaque are indeed semantically opaque. This suggests that, although DDPL did not have access to semantic information, the model did show a preference for treating semantically opaque words as simple. This is good from the point of view of a general assessment of the DDPL performance, but it made it harder to design the survey presented here. The corpus for the second survey was thus composed of the 97 complex-forDDPL forms that had a semantic rating lower than 2.5, and 97 randomly selected words from the 183 simple-for-DDPL words with a semantic rating lower than 2.5. I decided not to add a control set of semantically transparent forms, as I wanted to maximize the participants’ sensitivity to differences in morphological status among opaque words. If some semantically transparent words had been inserted, speakers would have probably reserved the high values of the rating scale for such forms, ‘‘squeezing’’ the ratings of semantically opaque words within a narrow range at the bottom of the scale. The average semantic rating across the complex-for-DDPL forms in this list was 1.54; the average rating across the simple-for-DDPL forms in this list was 1.21. One of the judges was also asked to rate the 194 forms in the corpus by assigning ratings on a 5 point scale on the sole basis of the degree of semantic transparency of the potential prefix of each form. The average prefix transparency rating across forms treated as complex by DDPL was 1.86; the average prefix transparency rating across forms treated as simple by DDPL was 1.46. Thus, while there is a noticeable and slightly worrisome difference in the degree of prefix transparency between the two sets, it seems safe to state that not only the forms in both sets are semantically opaque when considered as wholes, but also that the potential prefixes occurring in them tend to be opaque. The complex-for-DDPL set contained 17 distinct prefixes, the simple-forDDPL set contained 16 distinct prefixes. The two sets shared 14 prefixes. The average length of the words in the complex-for-DDPL set was 9 characters, the average length of their potential stems was 6.4 characters. The average length of the words in the simple-for-DDPL set was 8.6 characters, the average length
242
M. Baroni
of their potential stems was 6.3 characters. The average frequency of the words in the complex-for-DDPL set was 3.6, the average frequency of their potential stems was 263.4. The average frequency of words in the simple set was 21.1, the average frequency of their potential stems was 102.9. One word in the complex set had a potential stem that did not occur as an independent string in the corpus; 77 words in the simple set had potential stems that did not occur as independent strings.
5.2. Methodology and data collection The same methodology and data collection procedure described in section 4.2.1 above was followed in the second survey. A group of eight English native speakers, all graduate or undergraduate students or post-doctoral fellows in linguistics, took part in the survey. None of them had participated in the previous survey.
5.3. Results and discussion Pairwise Pearson and Spearman correlation coefficients were computed for the ratings of all pairs of participants. The patterns of three participants were poorly correlated with those of the other participants and with each other (for each of these three participants, the correlation coefficient between her/his ratings and those of a majority of other speakers was lower than 0.4). Thus, their data were discarded. As the ratings of the remaining participants were highly correlated (all pairwise Pearson and Spearman coefficients were higher than 0.5), the per-word average rating value across them was computed, and the resulting variable was compared to the parses assigned by DDPL to the same words in a one-way ANOVA in which the average ratings were grouped on the basis of the DDPL parses (simple vs. complex). The results of the ANOVA indicates that, in this case as well, the difference between ratings assigned by native speakers to words treated as simple vs. complex by DDPL is highly significant (F(1,192)=49.2, p<0.0000). If the participants in the survey had mostly relied on semantic cues when assigning ratings to the words in the list, they should have assigned uniformly low ratings to all words. However, this was not the case: as shown by the correlation between the average ratings and DDPL parses, in general speakers assigned higher ratings to words that DDPL treated as complex, lower ratings to words that DDPL treated as simple. The average mean rating across words that were complex for DDPL was 3.78; the average mean rating across words that were simple for DDPL was 2.81.
Distribution-driven morpheme discovery
243
The most plausible explanation for this asymmetry is that the way in which speakers represent potentially complex words is affected by distributional factors such as the ones implemented in DDPL.8 In turn, a plausible hypothesis about why such distributional factors have an effect on speakers’ morphological intuitions is that speakers relied on distributional cues during morpheme discovery. On the other hand, adult speakers are obviously also sensitive to semantic cues, when rating words for morphological complexity. As all the words in the survey corpus were semantically opaque, it is not surprising that the results of this second survey are less clear-cut than those of the previous survey (as shown by the fact that this time there is less of a difference between the average mean ratings assigned to DDPL simple and complex words). I suspect that semantics influenced the results both directly and indirectly. First, the morphological representations of adult speakers are almost certainly affected by the semantic structure of words. Thus, while speakers seem to distinguish words that are complex on purely distributional grounds from simple words, it is likely that such words occupy a middle ground, in terms of morphological complexity, between simple words and semantically transparent words (see Seidenberg and Gonnerman 2000 for similar considerations). Indeed, if no correlation between DDPL and the speakers had emerged, we could not have been sure that the negative result was due to the fact that speakers do not rely on distributional cues such as the ones employed by DDPL during morpheme discovery. The negative result could have instead been due to the fact that, once speakers acquire sufficient evidence about the semantic properties associated with morphemes, they revise their morphological representation of forms, and they change (from complex to simple) the representation of those forms that were originally treated as complex on distributional grounds, but whose complex representation is not supported by semantic evidence. Moreover, as a consequence of the fact that the distinction between semantically opaque but complex forms and simple forms is probably not as clear-cut as the distinction between complex and transparent words and simple words, the participants in the second survey had to provide ratings based on more subtle judgments, requiring more sophisticated metalinguistic introspection skills. Thus, as this was a harder task, it is likely that the participants in the second survey had more difficulty with it than the participants in the first survey, and that the less marked difference between sets is in part due to ‘‘noise’’ in the ratings. However, beyond these considerations, what is truly important from our point of view is that there is a high correlation between DDPL parses and the speakers’ ratings of semantically opaque words. Thus, the survey results provide support for the hypothesis that humans are sensitive to distributional cues to morphological constituency such as the ones used by DDPL.
244
M. Baroni
6. CONCLUSION The results of the simulation reported above provide support for the general hypothesis that distributional information of the kind encoded in the DDPL model can in principle be helpful in morpheme discovery. Moreover, the convergence between the DDPL parses and speakers’ ratings of a set of semantically opaque words provides some preliminary support for the hypothesis that humans rely on distributional cues such as the ones employed by the automated learner when assigning morphological parses to words. A plausible explanation of this finding is that speakers are sensitive to such cues because they employed them in order to assign morphological parses during morpheme discovery. Moreover, these results are also potentially relevant to the theory of morphological processing, in that they could provide the basis for a (partial) explanation of the fact that, as various psycholinguistic studies have shown, speakers treat some semantically opaque words as morphologically complex: They do so because, during morpheme discovery, they used distributional schemes to search for the morphemes of their language, and these schemes lead them to analyze some words as morphologically complex even in the lack of semantic cues supporting the complex analysis. Clearly, while I believe that the results presented here are encouraging, many questions are still open, and much more research has to be done before we can reach safe conclusions about the nature and role of distributional evidence in morpheme discovery. The DDPL model could be improved and extended in various ways. Obviously, the model should be extended to suffixation and other types of affixation. Furthermore, algorithms in which the distributional information used by DDPL is integrated with other types of information (such as syntactic category information) could be developed. Also, alternative lexicon generation algorithms, exploring a larger (or, better, more morphologically sensible) area of the hypothesis space, should be investigated. The reviewers pointed out recent work by Jennifer Hay (see, e.g., Hay 2000) suggesting that what matters in morphological processing is not the absolute frequency of derived forms, but the relative frequency of derived forms and their bases. In short, if a potentially complex form is more frequent than its potential base, the form is more likely to be parsed as a whole, whereas, if the base is more frequent than the complex form, then the complex form is more likely to be decomposed. In this setting, the absolute frequency heuristic used by DDPL can be seen as an approximation of a more realistic relative-frequency-based heuristic. In future research, it will be extremely interesting to test a revised version of the model that takes relative frequency effects into account.9 Finally, a reviewer also suggested that it would interesting to develop a version of DDPL that returns values on a continuous complexity scale, rather
Distribution-driven morpheme discovery
245
than binary complex vs. simple parses. This would allow a more direct comparison with human ratings, and it would correspond, perhaps, to a more realistic model of human morphological processing (see, e.g., Seidenberg and Gonnerman 2000 and Baroni 2001 for arguments in favor of gradient morphological representations). From the point of view of testing the model, we should first of all test DDPL in simulations with other English corpora, both in orthographic and phonetic transcriptions. Furthermore, DDPL should be tested using input corpora from other languages. In terms of collecting empirical evidence, we should first of all collect data from more speakers, possibly re-designing the survey task in order to make it feasible for speakers with no linguistics background. Furthermore, it would be interesting to collect data using other methods (for example, using a morphological priming paradigm), to make sure that the results we obtained are taskindependent. Obviously, it would also be important to collect developmental data from children, to have a more concrete idea of when and how human learners perform morpheme discovery. Last but not least, a more sophisticated analysis of the empirical results obtained should try to assess whether all the cues exploited by DDPL are relevant in predicting the response patterns of the speakers, and/or what is their relative importance as predictors. While all these lines of research should be pursued in the near future, and I am sure that readers will raise other important issues that were not dealt with here, I believe that the current results are already shedding some (weak) light on the role of distributional cues in the domain of morpheme discovery.
NOTES 1 I would like to thank Donca Steriade, Adam Albright, Lynne Bernstein, Harald Baayen, Amy Schafer, Kie Zuraw, the reviewers for the Yearbook of Morphology and, especially, Ed Stabler, Carson Schu¨tze and Bruce Hayes for help and advice. Of course, none of them is responsible for any of the claims I make. A more detailed discussion of several of the issues discussed here can be found in Baroni (2000b), which is downloadable from http:// sslmit.unibo.it/~baroni. 2 In this study, I illustrate my points through examples presented in orthographic transcription. The same points could have been also illustrated by the same examples (or similar ones) presented in phonetic transcription. A preliminary experiment with a corpus of phonetically transcribed words suggests that, because of the different distributional properties of specific morphemes in spoken and written language, the morpheme-discovery algorithm presented here performs in a similar but slightly different way when presented with orthographic vs. phonetically transcribed input. See Baroni (2000b:4.6) for discussion. 3 See also Brent (1993) and Brent, Murthy and Lundberg (1995). The model of Brent and
246
M. Baroni
colleagues represents, as far as I know, the first attempt to apply the Minimum Description Length principle to the problem of morpheme discovery. 4 In Baroni (2000b), I motivate and defend the assumptions about morpheme discovery that went into the design of the algorithm described here, i.e., that it makes sense to model morpheme discovery as a separate task from utterance segmentation, that it makes sense to model prefix discovery as a separate subtask within morpheme discovery, and that it makes sense to consider an approach to the task in which only binary (prefix+stem) parses of words are evaluated. 5 Baroni (2000b) discusses further how the lexicon + encoding criterion reflects morphological heuristics (including, among other things, a discussion of how the heuristics interact and of how such interaction insures that the ‘‘frequent substrings are likely to be morphemes’’ heuristic is interpreted in terms of type rather than token frequency). 6 All else being equal, in the data compression scheme proposed here longer substrings are more likely to constitute independent lexical entries than shorter substrings. For example, at the same frequency of occurrence in the input corpus, a substring like dis- is more likely to be treated as an independent entry than a substring like a-. Again, I would like to claim that this also makes sense from the point of view of morpheme discovery. 7 To keep things simple, I presented here an example in which stems occur elsewhere in the corpus as independent words – i.e., they are free stems. However, the same pattern takes place even if the relevant stems never occur in independent words, but are the product of the parse of other prefixed forms – i.e., they are bound stems. 8 See Baroni (2000b:4.5.3) for a post-hoc analysis that seems to rule out the possibility that the asymmetry can be explained by phonological cues. 9 Interestingly, the earlier model presented in Baroni 2000a did take relative frequency effects into account, at least to a certain extent, by assigning shorter indices to more frequent lexical entries, thus making the likelihood that a form will be parsed as complex dependent not only on the frequency of the form itself, but also on the frequency of its potential base.
REFERENCES Baayen, Harald (1994). Productivity in language production. Language and Cognitive Processes 9, 447–469. Baayen, Harald and Rochelle, Leiber (1991). Productivity and English derivation: A corpusbased study. Linguistics 29, 801–843. Baayen, Harald, Robert Schreuder and Cristina Burani (to appear). Parsing and semantic opacity. In: Morphology and the Mental Lexicon, E. Assink and D. Sandra (eds.). Dordrecht: Kluwer. In press. Baroni, Marco (2000a). Using distributional information to discover morphemes: A distribution-driven prefix learner. Paper presented at the LSA Meeting, Chicago. Baroni, Marco (2000b). Distributional Cues in Morpheme Discovery: A Computational Model and Empirical Evidence. UCLA dissertation. Baroni, Marco (2001). The representation of prefixed forms in the Italian lexicon: Evidence from the distribution of intervocalic [s] and [z] in northern Italian. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1999, 121–152. Bentin, Shlomo and Laurie Feldman (1990). The contribution of morphological and semantic
Distribution-driven morpheme discovery
247
relatedness to repetition priming at short and long lags: Evidence from Hebrew. Quarterly Journal of Experimental Psychology 42A, 693–711. Brent, Michael (1993). Minimal generative explanations: A middle ground between neurons and triggers. Proceedings of the 15th Annual Conference of the Cognitive Science Society, 28–36. Brent, Michael and Timothy Cartwright (1996). Distributional regularity and phonotactic constraints are useful for segmentation. Cognition 61, 93–125. Brent, Michael, Sreerama Murthy and Andrew Lundberg (1995). Discovering morphemic suffixes: A case study in minimum description length induction. Presented at the Fifth International Workshop on AI and Statistics. Emmorey, Karen (1989). Auditory morphological priming in the lexicon. Language and Cognitive Processes 4, 73–92. Feldman, Laurie (ed.) (1995). Morphological Aspects of Language Processing. Hillsdale: LEA. Goldsmith, John (2001). Unsupervised learning of the morphology of a natural language. Computational Linguistics 27, 153–198. Harris, Zellig (1955). From phoneme to morpheme. Language 31, 190–222. Hay, Jennifer (2000). Causes and Consequences of Word Structure. Northwestern University dissertation, 2000. Kucera, Henry and Nelson Francis (1967). Computational Analysis of Present-day American English. Providence: Brown University Press. Marchand, Hans (1969). The Categories and Types of Present-day English Word-formation: A Synchronic-diachronic Approach. Munich: Beck. Marslen-Wilson, William, Lorraine Tyler, Rachelle Waksler and Lianne Older (1994). Morphology and meaning in the English mental lexicon. Psychological Review 101, 3–33. Maxwell, Michael (ed.) (2002). Morphological and Phonological Learning: Proceedings of the Sixth ACL-SIGPHON Meeting. Philadelphia: ACL. Quirk, Randolph, Sidney Greenbaum, Geoffry Leech and Jan Svartvik (1985). A Compresive Grammar of the English Language. London: Longman. Redington, Martin and Nick Chater (1998). Connectionist and statistical approaches to language acquisition: A distributional perspective. Language and Cognitive Processes 13, 129–191. Rissanen, Jorma (1978). Modeling by shortest data description. Automatica 14, 456–471. Roelofs, Ardi and Harald Baayen (2001). Morphology by itself in planning the production of spoken words. Psychonomic Bulletin and Review 9, 132–138. Schreuder, Robert and Harald Baayen (1994). Prefix stripping re-revisited. Journal of Memory and Language 33, 357–375. Schreuder, Robert and Harald Baayen (1995). Modeling morphological processing. In: Feldman (1995), 131–154. Seidenberg, Mark and Laura Gonnerman (2000). Explaining derivational morphology as the convergence of codes. Trends in Cognitive Sciences 4, 353–361. Seitz, Philip Lynne Bernstein, Edward Auer and Margaret MacEachern (1998). The PHLEX Database. Los Angeles: House Ear Institute. Smith, Philip (1988). How to conduct experiments with morphologically complex words. Linguistics 26, 699–714. Snover, Matthew and Michael Brent (2001). A Bayesian model for morpheme and paradigm identification. Proceedings of ACL 39, 482–490.
248
M. Baroni
Stolz, Jennifer and Laurie Feldman (1995). The role of orthographic and semantic transparency of the base morpheme in morphological processing. In: Feldman (1995), 109–129. Wurm, Lee (1997). Auditory processing of prefixed English words is both continuous and decompositional. Journal of Memory and Language 37, 438–461.
SSLMIT University of Bologna Corso della Republica 136 47100 Forli (FC) Italy [email protected]
Morphological ‘gangs’: constraints on paradigmatic relations in analogical change1 CAROL FEHRINGER
1. INTRODUCTION In a wide variety of modern High and Low German dialects, historical phonological developments led to the creation of stem allomorphy in certain inflectional and derivational forms of the same base word. Subsequently, different types of (segmental and suprasegmental) analogical change conspired to maintain the formal relationship between the stems in various subsets of the derivatives concerned. These analogical developments were not motivated by the desire to highlight the relationship between form and meaning, as is often the case with analogical change (see Mayerthaler 1981, Dressler et al. 1987), but by a general tendency to systematise the purely formal connections between the members of the subsets. In this way, the German developments provide a classic example of ‘Morphology by Itself’ (Aronoff 1994), in which morphological properties of words are keyed to each other without any semantic motivation. Indeed, the analogical changes in the German dialects are often ‘unnatural’, in the sense of e.g. Mayerthaler (1981), in that they promote and even create new allomorphy and therefore lead to a complication of the derivational process as a whole. Thus, it is no longer possible to derive certain forms from a base without first referring to other members of the subset concerned, as the formal properties of one member may influence those of others. This type of influence is clearly paradigmatic, but as it apparently cuts across the inflection/derivation divide we need to use the term paradigm in a much wider sense to include derivation as well as inflection (see also Booij 1997).2 Furthermore, given the fact that the analogical changes in the German dialects in question only affect specific subsets of the possible derivatives it is necessary to determine why this is the case, which types of groupings occur and which types of analogical change are favoured in each. In order to account for these developments the concept of morphological ‘gangs’ will be introduced.3 A morphological gang is a subset of the (inflectional and derivational) derivatives of a particular base word, the members of which have come together for historical reasons but, more interestingly, have stayed together due to their high degree of productivity and semantic transparency (i.e. they are perceptually salient, see Chapman 1994), while other derivatives with a similar historical development have not been included in the gang due to a lack of perceptual salience and therefore the tendency towards lexicalisation. It is hoped that once the concept of morphological gangs has been recognised, it will be possible to show how otherwise apparently idiosyncratic analogical developments can often behave quite systematically. Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 249–272. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
250
C. Fehringer
2. THE ADJECTIVE GANG: COMPARATIVES, SUPERLATIVES AND DEADJECTIVAL ABSTRACT NOUNS 2.1. Umlaut German umlaut provides a classic example of an original phonological rule which has now become morphologised. In Germanic stressed back vowels became fronted under the influence of an unstressed */i/ or */j/ in the following syllable, reflected in Old High German (OHG) e.g. lohhir /lœhhir/ ‘holes’<singular loh /lch/ (Braune/Eggers 1987: 27–8), compare modern German Lo¨cher where the original conditioning factor has been reduced to schwa and umlaut functions as a morphological marker, in this instance as a noun plural marker. In the case of WGmc. */a/, an additional process of raising took place whereby the umlauted (i.e. fronted) reflex OHG /æ/ was raised to /e/ in certain lexical items where the stem vowel was not followed by OHG /ht/, /hs/ or, in the south /h/, /xx/, /rw, /rh/, e.g. OHG lenger ‘longer’
/ta:g/ /ta:g/ /kxla:ge/ /hand/
: : : :
/tæ:g/ /tæ:gig/ /kxlæ:gli/ /hændli/
day day lament hand
: : : :
days a day old lamentable little hand
NPL ADJ(denominal) ADJ(deverbal) DIM
Moulton (1971) believes that the analogical extension of /æ/ was due to the fact that the /a/: /æ/ alternation corresponded to the dominant [−palatal]: [+palatal] (i.e. [+back]: [−back]) umlaut opposition characterising other vowel alternations such as /o/: /ø/, /u/: /y/ etc. and that it was this feature opposition, not the phoneme opposition /a/: /e/, which became productive and spread analogically, despite the fact that the /a/: /e/ opposition was originally the quantitatively predominant one for bases in /a/. However, he also notes that this cannot
Morphological ‘gangs’
251
account for the fact that, in adjective comparatives, superlatives and the corresponding deadjectival abstract nouns in -i (e.g. lengi 5 ‘length’
COMP
SUPERL ABST N
/kxa:lt/ /kxelter/ /kxeltist/ /a:lt/ /elter/ /eltist/ /sva:h/ /svexer/ /svexist/
/kxelti/ /elti/ /svexi/ 6
cold old weak
*Contrast deverbal nouns in -i which are nearly always umlautless, e.g. /maxi/ ‘shoddy work’
ADJ
COMP
SUPERL ABST N
a. b. c. d.
/kclt/ /ka:l/ /no:x/ /hauge/
/kœlder/ no comp.7 /nø:xer/ /høxter/
/kœltest/ /kE:lst/ /nø:xst/ /høxste/
/kœlde/* /kE:lt/ /nø:xte/ /høxte/
e.
/nc:x/ /bra:t/
/nœ:xer/ /breitter/
/nœ:xst/ /breitst/
/nœ:xi/ /breitti/
cold (analogical umlaut) cold (analogical umlaut) near (analogical umlaut) high (standard Ger. influence) near (analogical umlaut) wide (etymological diphthong retained)
*Contrast the verb /fer-kElden/ ‘to catch cold’, with the older etymological umlaut reflex. From these examples it is evident that comparatives, superlatives and deadjectival abstract nouns are not simply derived from their base adjectives by
252
C. Fehringer
means of unidirectional rules (contrast Strauss’s 1980, Zwicky’s 1967 and Bach and King’s 1970 rule-based analyses of standard German umlaut) but that the shape of their stems is determined on the basis of paradigmatic relations. Thus, one can predict that if the comparative has a certain type of umlaut, the corresponding superlative and abstract noun will have the same. Such implicational relationships are well-attested within inflectional paradigms (see e.g. Wurzel 1984b, Aronoff 1994, Carstairs-McCarthy 2001) and, to a certain extent, in derivation (van Marle 1985, 1994, Bauer 1997) but, in the cases outlined above, the interdependence between the stem shapes of the complex forms cuts across the inflection/derivation divide (see also Booij 1996, 1997 and van Marle 1996).8 For this reason I have chosen to refer to the members of these categories as a morphological gang rather than use the term ‘paradigm’ which is somewhat problematic. 9 Furthermore, as will become apparent later, only a specific subset of derivatives constitutes a morphological gang: in this case, only the comparative, superlative and derived abstract noun display the characteristics of gang membership (i.e. the formal identity of their stems) while other derivatives of the same adjective, such as deadjectival verbs and other derived nouns (e.g. nouns denoting a person with the quality expressed by the adjective) are ‘outsiders’ in that they have not been included in the analogical change. Returning to the /e/ vs. /æ/ umlaut distinction in the Swiss dialects, Moulton (1971) cannot account for the fact that /e/ is productive in adjective comparatives, superlatives and the corresponding abstract nouns despite the fact that it does not accord with the general [+back]: [−back] feature opposition which is dominant in the dialects. Lu¨ssy (1974: 92) also notes the problem but provides no adequate explanation either. If one acknowledges the relevance of morphological gangs, however, the developments can be explained quite simply. /e/ was largely replaced by /æ/ in e.g. noun plurals but not in adjective comparatives, superlatives and abstract nouns because the adjective gang had more members with /e/ in their stems (i.e. comp, superl and abst noun), thus reinforcing the /e/ vowel as a (secondary) marker of these categories10 while, by contrast, the noun plurals in /e/ did not have any ‘suppport’ from other members of their paradigm and were therefore more affected by the spread of the dominant [+back]: [−back] feature opposition. For instance, diminutives, which may be seen as constituting a morphological gang with their corresponding noun plurals (see 3 below) had /æ/ etymologically due to their phonological shape. In other words, if the [+back]: [−back] opposition found in adjectives such as jung – ju¨nger, hoch – ho¨her etc. spread to the adjectives in /a/ it would not only affect comparatives but also superlatives and abstract nouns. On the other hand, if it spread from plurals such as e.g. Hut – Hu¨te, Loch – Lo¨cher to nouns in /a/ it would only affect the noun plurals. Thus, it appears that the size of a morphological gang (in terms of the number of its members) may affect the direction of analogical change. Indeed, it is often the case that the /e/ umlaut becomes associated with the morphological gang [adj – comp – superl – abst n] to such an extent that it is
Morphological ‘gangs’
253
analogically extended to forms in reflexes of WGmc. long */a:/ where an /e/ versus /æ/ contrast is etymologically unjustified. For instance, in the dialect of Glarn (canton Glarus) the etymological umlaut of WGmc. /a:/ is /æ:/ which appears productively in noun plurals, diminutives and other derivational forms alternating with a base form in /a:/ (Streiff 1915: 20): (4) /na:dl/ : /næ:dl/ needle : needles /na:dl/ : /næ:dli/ needle : litttle needle /na:dl/ : /næ:dlig/ needle : thread on a needle However, in adjective comparatives, superlatives and their corresponding abstract nouns an analogical vowel /e:/ appears (see (5c)) which is identical in quality to the umlaut vowel in the adjective comparatives etc. of forms in WGmc. short */a/ which was lengthened phonologically to /a:/ before certain consonants (see (5b)): (5)
ADJ
COMP
SUPERL ABST N
/xelter/
/xeltst/
a.
WGmc. */a/
/xalt/
b.
*/a/>/a:/ 11
/gra:d/ /gre:der/ /gre:dst/ /sma:l/ /sme:ler/ /sme:lst/
c.
WGmc. */a:/ /ra:ss/ /bla:b/ /spæ:t/ /tsæ:x/
/re:sser/ /ble:ber/ /spe:ter/ /tse:xer/
/re:ssist/ /ble:bst/ /spe:tist/ /tse:xist/
/xelti/
cold
/gre:di/ /sme:li/
straight narrow
/re:ssi/ /ble:bi/* /spe:ti/ /tse:xi/
strong blue late tough
*Contrast /blæ:bæ:le/ ‘blue mark/bruise’, with the etymological umlaut.12 Particularly interesting are the last two examples in (5c) which show the etymological umlaut /æ:/ in the base adjective alternating with the new analogical umlaut /e:/ in the derived forms, since these show that one is not dealing with the analogical extension of an alternation (in this case /a:/: /e:/) but with that of a single phone associated with a well-defined, cohesive group of grammatical functions. Each function is clearly marked by an affix (comparative /-er/, superlative /(i)st/, abstract noun /i/) and the /e:/ vowel appears as a secondary marker expressing the general identity between the members of the morphological gang. Thus, umlaut in the Swiss dialects is not merely a general ‘marker of markedness’, as Wurzel (1984a) describes standard German umlaut, but can also be a marker of morphological gang membership. If one believes, along with Bybee and Brewer (1980: 225), that ‘the formal organisation of a paradigm diagrams the semantic organisation so that forms that are more similar semantically will be more similar in morphohonemic
254
C. Fehringer
shape’ it is not surprising that the comparative and superlative are similar in form, as they both express degrees of gradation. Indeed, it is well-known that, in standard German, umlaut may be present in comparatives and superlatives, e.g. jung/ju¨nger/ju¨ngst ‘young’, or absent, e.g. bunt – bunter – buntest ‘colourful’ but never present in one and absent in another (see e.g. Wurzel 1970: 122 and Becker 1990: 43). However, this does not explain why, in many German dialects, the corresponding abstract noun also follows the same formal pattern, as it is not semantically associated with the comparative and superlative. One could argue that this alliance is merely a result of historical accident: i.e. umlaut occurred phonologically in these categories due to the original presence of /i/ in the suffixes and the identity of the resulting vowels caused speakers to associate the categories with each other psychologically. This, however, is only half of the explanation, as there are other derivational forms of the same adjective which had the same historical umlaut vowel as the comparative, superlative and abstract noun but have not regularly adopted the same analogical vowel as these related forms, e.g. Aldenrade /kclt/ : /kœlder/ ‘cold : colder’ with analogical umlaut but /kclt/ : /fer-kElden/ ‘cold : to catch cold’ with the old historical umlaut (Neuse 1915: 90). Yet such forms are not as productive as deadjectival abstract noun derivation in -i (or -e in central and northern dialects) and are often not semantically transparent, which means that they are more subject to lexicalisation than the abstract nouns. Thus, the formal alliance between comparatives, superlatives and abstract nouns may have originally been due to historical accident but was reinforced over time by the high degree of productivity of all three categories. Particularly interesting is that once these categories became associated with each other due to phonological (and later analogical) developments, further analogical processes other than umlaut conspired to reinforce the formal identity of the related word forms. These are outlined in 2.2 and 2.3 below.
2.2. Analogical vowel lengthening In many Swiss dialects, root vowels in monosyllabic words became lengthened before certain consonants, which often led to a long vowel in a base form alternating with a short vowel in the complex form, e.g. Entlebuch (Lucerne) /gla:s/ : /gleser/ ‘glass : glasses’ (see Chapman 1995 for details). In many inflectional and derivational categories, however, the [+long]: [−long] distinction was levelled in favour of the long vowel. Some complex forms show levelling throughout (e.g. the inflected forms of adjectives, used before nouns: /la:m/ : /lame/>/la:m/ : /la:me/ ‘lame’, /gra:d/ : /grade/>/gra:d/ : /gra:de/ ‘straight’, /ho:l/ : /hole/>/ho:l/ : /ho:le/ ‘hollow’ etc.) while others show levelling more sporadically. In the case of adjective comparatives, superlatives and deadjectival abstract nouns, levelling occurred with most, but not all, adjectives in Entlebuch
Morphological ‘gangs’
255
(although it was more regular in other dialects). What is interesting, however, is that when lengthening occurred in one member of the morphological gang it automatically occurred in the others too, so that one never finds a long vowel in a comparative alternating with a short vowel in the corresponding superlative or abstract noun. Consider the following examples from Entlebuch (Chapman 1995: 5, 10): (6)
ADJ
COMP
SUPERL ABST N
a.
la:m/ /ra:n/
/le:m(m)er/ /le:mst/ /re:n(n)er/ /re:nst/
b.
/gra:d/ /greder/ /sma:l/ /smeller/
/gretst/ /smelst/
/le:mi/ /re:ni/
lame slender
/gredi/* /smeli/
straight narrow
*Contrast /gra:de/, /sma:le/ etc., the inflected forms of the adjective, and /gra:de/ the derived verb ‘to become straight’ which have all analogically lengthened their vowel. Note that the root vowel in the abstract noun /fili/ ‘large amount’ (cf. /fi:l/ ‘much, many’) has not undergone analogical lengthening which is probably due to the fact that it is not influenced by a corresponding comparative (the comparative of /fi:l/ is a suppletive form /me:/ ‘more’) but behaves like most other derivational forms in which lengthening is sporadic (e.g. /œ:l/ : /œ:le/ ‘oil : to oil’, with lengthening, versus /ta:g/ : /tage/ ‘day : to become day’ without). This suggests that deadjectival abstract nouns are formed on the basis of the corresponding comparative/superlative stem where present. Indeed such developments are interesting from a theoretical point of view, as the idea of a traditionally derivational process such as deadjectival noun formation being sensitive to the form of traditionally inflectional comparatives and superlatives does not accord with a ‘split morphology’ theory such as that of Anderson (1992) in which derivational processes are assigned to the lexicon while inflectional ones operate postsyntactically (see also Chapman 1994, Booij 1996). On the other hand, one could argue that the status of adjective comparatives and superlatives is not clear cut. Some linguists claim that they are derivational (see note 8) while others (e.g. Booij 1996) maintain their inflectional status while recognising that they share many properties with derivation. Indeed, in the German dialects, these forms appear to be more akin to (albeit quite systematic) derivation.
2.3. Analogical /n/-insertion Further evidence to suggest that adjective comparatives, superlatives and deadjectival abstract nouns form a cohesive morphological gang can be found in the
256
C. Fehringer
dialect of Zu¨rich (Weber 1948: 87–88) where word-final or pre-consonantal /n/ is elided after a stressed vowel, e.g. /xlaI/ ‘small’, /gmæIsam/ ‘together’, /ufrYntli/ ‘unfriendly’ (cf. standard German klein, gemeinsam and unfreundlich respectively). When a suffix beginning with a vowel is added to base forms in underlying /-n/, the consonant reappears as a linking element, e.g. /grYe/ ‘green’, /e grYe-s platt/ ‘a green leaf’ versus /e grYen-e wi:s/ ‘a green meadow’. As comparatives and deadjectival abstract nouns have suffixes beginning with a vowel (/er/ and /i/ respectively) while superlatives do not (/st/), one would expect /n/ to be present in the former two categories but not in the latter. However, consider the following examples (Weber 1948: 125–129, 345): (7) ADJ
COMP
SUPERL ABST N
/xlaI/ /xlaIner/ /xlaInst/ /bru:/ /bry:ner/ /bry:nst/
/xlaIni/ /bry:ni/
small brown (with umlaut)
Although it is not phonologically justified, /n/ also appears in the superlatives so that the identity between the stems of the complex forms is preserved. This not only applies to adjectives with an etymological linking-n but also to those which have introduced /n/ analogically into the complex forms: (8) ADJ
COMP
SUPERL ABST N
/fraI/ /fraIner/ /fraInst/ /fro:/ /frø:ner/ /frø:nst/ /frYe/ /frYener/ /frYenst/
/fraIni/ free (standard Ger. frei) /frø:ni/ happy (standard Ger. froh) (no noun given)* early (standard Ger. fru¨h)
*Contrast /frYelig/ ‘early bird’, without analogical /n/. Such developments suggest that the etymological /n/ in the stems of comparatives such as /xlaIner/ was spread analogically to a number of other comparatives of adjectives ending in a vowel (and to some ending in a consonant, e.g. /bra:v/ : /brevner/ ‘well-behaved : better behaved, /fol/ : /folner/ ‘full : fuller’) and that the superlatives (and probably also deadjectival abstract nouns, e.g. /brevni/ ‘good behaviour’) were formed on the basis of the corresponding comparative stem.13 3. THE NOUN GANG: NOUN PLURALS AND DIMINUTIVES 3.1. Umlaut In a considerable number of High and Low German dialects, the quality of the root vowel is identical in noun plural and diminutive forms. This was originally
Morphological ‘gangs’
257
due to the fact that umlaut occurred in both categories due to the presence of */i/ in the suffixes, as it did in a variety of other inflectional and derivational forms (see 2.1 above). However later analogical developments show that the formal connection between noun plurals and diminutives was often maintained while other inflectional and derivational forms were not affected as systematically. For instance, in dialects which distinguish between a historical and analogical umlaut, it is sometimes the case that the analogical umlaut vowel characterises noun plurals and diminutives but is not found in other morphologically complex forms to which umlaut would normally apply. Taking the Luxembourg (Central German) dialect of Arlon in south east Belgium as an example (see Bertrang 1921: 125–127), the rounded reflex /c:/ of WGmc. */a:/ (and WGmc. */-age-/) has an analogical umlaut /E:/ (after unrounding) which has become productive in noun plurals and diminutives but not in other inflectional or derivational forms which have retained the etymological diphthong /EI/ (from the MHG umlaut vowel /æ/).14 Contrast the examples in (9a) with those in (9b) (Chapman 1996: 51): (9) a.
N
NPL
/mc:t/ /mE:t/ /brc:t/ /brE:t/ /blc:s/ /blc:zen/ 15 /sc:f/ : /sE:f/ /svc:er/ : /svE:er/
DIM /mE:txen/ /brE:txen/ /blE:sxen/ /sE:fxen/* (dim. not given)*
girl roast bubble/blister sheep brother-in-law
*Contrast /sEIfer/ ‘sheperd’ and /svEIer/ ‘sister-in law’, with the etymological diphthong. b. /brc:den/ /blc:zen/ /gefc:r/ /blc:/
: : : :
/brEIt/ /blEIs/ /gefEIerlex/ /blEI/
to roast to blow (bubbles) danger blue
: : : :
(he) roasts (he) blows dangerous blueness
Similarly, in the Dutch Limburg dialect of Roermond, umlaut is only productive in noun plurals and diminutives (see 10a) while in other derivational categories it is sporadic. It is absent from comparatives and superlatives, and in verbal inflection the stem vowel alternations no longer conform to the original [+back]: [−back] feature opposition due to a number of phonological and morphological developments (see 10b). Examples are taken from Kats (1939: 155–165): (10) a.
N
NPL
DIM
/bc:m/ /ba:rt/ /koop/
/bœ:m/ /bæ:rt/ /kœYp/
/bœ:mke/ /bæ:rtke/ /kœYpke/
tree beard purchase
258
C. Fehringer
b.
/blc:zen/ /slc:n/ /dra:xen/ /loopen/
: : : :
/blø:s/ /sleIs/ /drø:xs/ /lœps/
to blow to hit to carry to walk
: : : :
(he) blows (he) hits (he) carries (he) walks
Indeed the close relationship between noun plurals and diminutives is manifest in a wide range of High and Low German dialects in which the form of the diminutive is directly influenced by that of the corresponding noun plural stem. For example, in many Swiss dialects noun plurals have retained an etymological umlaut /e/ before the plural suffix /-er/ which is introduced analogically into diminutives derived from the same base noun as the plural forms, despite the fact that the etymological umlaut for all diminutives is /æ/. In those cases where /æ/ occurs in the plural (e.g. before the plural suffix /-u/), or the plural form is without umlaut, the corresponding diminutive is marked by /æ/. Contrast (11a) with (11b) from Visperterminen in the canton of Wallis (Wipf 1910: 27): (11) a.
b.
N
NPL
/rad/ /glas/ /ast/
/reder/ /redli/ /gleser/ /glesli/ /est/ /estli/
/fado/ /fædu/ /ba:rt/ /bæ:rt/ /tag/ /tage/
DIM wheel glass branch
/fædi/ thread /bæ:rtli/* beard /tægli/ day
*By contrast, denominal masculine animates in /-li/ are usually without umlaut: e.g. /ba:rt/ : /ba:rtli/ ‘beard : bearded man’. Although it is more common in the German dialects for noun plural forms to exert pressure on diminutives with regard to analogical change and not vice versa, influence in the opposite direction is not unheard of. In some dialects of Alsace, for instance, diminutives with the more productive /æ/ umlaut appear to exert pressure on their corresponding noun plural forms in /e/ (Beyer 1963: 192–196): (12) N
NPL
DIM
/stat/ /stet/>/stæt/ /stætel/ town /nast/ /nest/>/næst/ /næstel/ nest /gast/ /gest/ (dim. not used) guest By contrast, in those Alsatian dialects which have a number of diminutives in /e/, e.g. /stetle/ ‘little town’, there is no tendency to replace /e/ with /æ/ in the noun plural form, despite the fact that the latter is the more productive umlaut.
Morphological ‘gangs’
259
3.2. Vowel lengthening Further evidence of the dependence of diminutive formation on noun plural stems can be found on examinining the process of analogical vowel lengthening in Swiss dialects (cf. 2.2 above). Consider the following examples from Fribourg (Chapman 1995: 6–7 and Henzen 1927: 102): (13) N /ty:ss/ /væ:g/ /ha:s/ /na:s/ /gra:s/ /ba:d/ /ra:d/
NPL
DIM
/tysse/ /væ:ge/ /hase/ /na:se/ /greser/ /beder/ /reder/
/tyssli/ /væ:gli/ /hæsi/ /næ:si/ (dim. not given)* /bedli/* /redli/
table path hare nose grass bath wheel
*Contrast /gra:se/ ‘to graze’ and /ba:de/ ‘to bathe’ with a lengthened vowel. From these data we can see that if analogical lengthening occurs in the noun plural form, it also occurs in the corresponding diminutive. In the last two examples the occurrence of the umlaut vowel /e/ in the diminutive, where it is not etymologically justified, is also due to the influence of the noun plural stem (cf. (11a) above). On the other hand, verbs derived from nouns make no reference to the noun plural form. The verbs mentioned at the end of (13) have a long vowel despite the short vowel in the plural. Conversely, verbs such as /tsale/ ‘to count’ (cf. /tsa:l/ ‘number’) and /spile/ ‘to play’ (cf. /spi:l/ ‘game’) have an etymological short vowel, despite the analogical long vowel in their related noun plural forms (i.e. /tsa:le/ ‘numbers’, /spi:li/ ‘games).
3.3. Intonation In a number of Low German dialects, phonological developments have led to the emergence of special intonation patterns, many of which differentiate morphologically complex forms from their corresponding simplexes. For instance in Roermond (Limburg), a number of noun plurals are marked by a falling intonation (transcribed as /"/) which arose after the loss of a word-final schwa, e.g. MLG /volf/: /vœlfe/ ‘wolf : wolves’>Roermond /volf/ : /vœl"f/). What is interesting is that this intonation is often introduced analogically into diminutive forms in -(s)ke where it is not phonologically justified, so that the noun plurals and diminutives share the same stem (see 14a), while nouns whose plural forms lack the falling intonation due to the presence of a schwa, or simply for lexical
260
C. Fehringer
reasons, do not have the intonation in the diminutive (see 14b).16 Consider the following examples from Kats (1939: 158–159): (14)
N
NPL
DIM
a.
/tœrf/ /dy:j/ /volf/ /da:x/
/tœr"f/ /dy:j"/ /vœl"f/ /da:"x/
/tœr"fke/ /dy:j"ke / /vœl"fke/ /dæ:"xske/
b.
/a:p/ /a:pe/ /æ:pke/ /koup/ /kœYp/ /kœYpke/
peat, turf push wolf day ape purchase
By contrast, Kats (1939) gives no evidence of the noun plural intonation pattern being systematically introduced by analogy into other related derivational forms.
3.4. Analogical /n/-insertion It was argued in 2.3 above that, in the dialect of Zu¨rich, the retention of /n/ in superlative forms where it is not phonologically justified is due to the fact that the form of the superlative is influenced by that of the corresponding comparative stem (e.g. /xlaI/: /xlaIn-er/ : /xlaIn-st/). The only other morphological category in which /n/ is retained without phonological justification is the diminutive, e.g. /ma:/ : /mæn-li/ ‘man : little man’ where /n/ appears despite the following consonant (contrast the genitive singular /ma:-s/). This may be due to the fact that the diminutive form is influenced by the corresponding noun plural, the dative form of which retains the /n/ phonologically before the dative plural marker /e/ (examples from Weber 1948: 110–11, 327–9): (15) N
NomPL DatPL
/tsa:/ /tsæ:/ /bæI/ /bæI/ /ma:/ /ma:/
DIM
/tsæ:ne/ /tsæ:n(d)li/ 17 tooth /bæIne/ /bæIn(d)li/ leg /ma:ne/ /mæn(d)li/ man
These data are rather problematic, however. Although there is much in the literature on noun plurals ‘feeding’ derivation (see e.g. Perlmutter 1988 for Yiddish, Scalise 1988 for Portuguese, Booij 1996 for Dutch and Chapman 1997 for Central German dialects), there is no mention of an oblique member of the paradigm such as the dative plural having such influence. Indeed, it is generally believed that case-marked forms, being purely dictated by syntax, cannot feed derivational processes (see, for example, Anderson 1992, Booij 1996). Similarly,
Morphological ‘gangs’
261
it could be argued that dative plurals are of marginal importance in the paradigm and not salient enough for triggering analogies. If this is the case, there appears to be no explanation for the use of analogical n in diminutives.18 To sum up so far, historical phonological developments (in this case umlaut) led to certain inflectional and derivational forms of the same base word becoming formally identical in their root vowel, while their separate functions were usually clearly marked by suffixes. 19 Later purely morphological developments provide evidence that speakers make a psychological connection between subsets (or ‘gangs’) of these forms and maintain the identitity of their roots through various kinds of analogical change even though there is usually no semantic motivation for linking these forms together (except, perhaps, comparative and superlative). Often the analogical changes appear ‘unnatural’ in that they create, or at least do not eliminate, allomorphy and therefore lead to a complication of the derivational process: e.g. one cannot simply derive a diminutive from its base noun or an abstract noun from its base adjective without first referring to other members of the gang. However, this complication is systematic, as allomorphy is eliminated within the morphological gang, so that one would never find e.g. an analogically lengthened vowel in deadjectival abstract nouns but a historically short vowel in the comparative and/or superlative. Thus, it appears that speakers can best deal with allomorphy if it is systematic and predictable on the basis of general implicational patterns.20 Wurzel (1984b) expresses these relations in terms of ‘Paradigm Structure Conditions’ for inflectional morphology. Cameron-Faulkner and Carstairs-McCarthy (2000), in their treatment of Polish noun inflection, go one step further and argue that implicational paradigm structure conditions apply to non-affixational rather than affixational morphology. This can also be observed in the case of the German dialects, the difference being that these conditions apply to derivational as well as apparently inflectional stem alternants: (16)
a.
Adjective gang Form of comp stem6form of superl stem6form of abst n stem
b.
Noun gang Form of n pl stem6form of dim stem
The use of the term ‘apparently inflectional’ expresses the idea that, although comparatives, superlatives and noun plurals are traditionally regarded as inflectional, they are certainly not prototypical and, certainly in the German dialects, behave more like derivation. The lexical connections between e.g. comparatives, superlatives and abstract nouns are much stronger than the connections between e.g. the inflected form of the adjective, which is a case of prototypical inflection, and the comparative/superlative. The inflected form of the adjective is not at all influenced by the comparative or superlative, or vice versa (see e.g. the vowel lengthening examples in (6)).
262
C. Fehringer
4. THE NON-APPLICATION OF REGULAR PHONOLOGICAL RULES WITHIN A MORPHOLOGICAL GANG Further evidence for the strong formal connection between adjective comparatives, superlatives and deadjectival abstract nouns on the one hand and noun plurals and diminutives on the other is the fact that some phonological rules, which are completely regular in the dialects,21 do not apply within a particular morphological gang if they would otherwise obscure a salient alternation or the formal relationship between the members. For instance in many Swiss dialects, e.g. Zu¨rich (Weber 1948), Glarn (Streiff 1915), Entlebuch (Schmid 1915), /e/ is regularly lowered to /æ/ before nasal consonants, e.g. OHG /enk/>Glarn /ænk/ ‘narrow’. In comparatives, superlatives and abstract nouns, however, /e/ is always retained so that all a-type adjectives alternate with a superlative etc. in /e/. Consider examples (17a) and (17b) from Glarn, the latter of which which have retained /e/ in the derivatives despite the following nasal: (17)
ADJ
COMP
SUPERL ABST N
a.
/alt/ /xalt/
/elter/ /xelter/
/eltst/ /xeltst/
/elti/ /xelti/
old cold
b.
/lann/ /lenner/ /lenst/ /ænk/ /enker/ /enkst/
/lenni/ /enki/
long narrow
Evidence from other Swiss dialects such as Entlebuch (Schmid 1915: 86–87) suggests that the phonological rule may have originally operated in these categories and the outputs in /æ/ were then replaced by /e/ due to analogical extension: (18)
ADJ a.
b.
COMP
SUPERL ABST N
/alt/ /elter/ /eltst/ /lann/ /lænner/ /lænst/ /ænn/ /ænner/ /ænkst/
/elti/ /lænni/ /ænni/
(older speakers lower /e/ to /æ/ before nasal)
/alt/ /elter/ /lann/ /lenner/ /ænn/ /enner/
/elti/ /lenni/ /enni/
(younger speakers extend /e/ analogically)
/eltst/ /lenst/ /enkst/
A similar principle can be observed in the Central German dialect of Arlon, where noun plurals and diminutives appear to be unaffected by a regular monopthongisation rule. In this dialect, /ue/, the diphthongised reflex of MHG /a/ and /o/, and its (unrounded) umlaut /Ie/ regularly alternate in most inflectional and derivational categories affected by umlaut (Bertrang 1921: 125–127, 284–289):
Morphological ‘gangs’
(19) /bruext/ /duext/ /knuet/ /kueref/
: : : :
/brIext/ /dIext/ /knIet/ /kIeref/
brought thought knot basket
: : : :
263
would bring would think knots baskets
However, before nasals /Ie/ is monophthongised to /I/, which leads to an obscuring of the original umlaut alternation (Bertrang 1921: 68, 285): (20) /nuem/ : /ernImen/ /suem/ : /sImex/
name : to name, to mention shame : shameful
By contrast, this phonological rule is blocked in noun plurals and diminutives so that the regular umlaut alternation is maintained: (21) N /nuem/ /huemer/ /ruem/ /kuem/
PL
DIM
/nIem/ /hIemer/ /rIem/ /kIem/
(no dim. given) /hIemertxen/ /rIemtxen/ /kIemtxen/
name hammer frame comb
Unlike the Swiss developments outlined in (17) and (18) above, there is no evidence to suggest that the phonological rule of monophthongisation ever took place in noun plurals and diminutives. Thus, it appears that we are not dealing with a case of analogical extension of /Ie/ but with the simple non-application of a phonological rule within this particular morphological gang.
5. THE DIRECTION OF ANALOGICAL CHANGE WITHIN A MORPHOLOGICAL GANG It is interesting to note that certain types of analogical development appear to be favoured in some morphological categories rather than others. Specifically, the direction of analogical change often differs according to which morphological gang is affected: i.e. intraparadigmatic change or analogical extensions across paradigms. For instance, in noun plurals and diminutives, speakers seek to accentuate the alternation between base and derivative within a paradigm by marking the alternants in a uniform way. This is particularly the case with umlaut (i.e. base vowel [+back]: derivative vowel [−back]) and does not only apply to the replacement of historical umlaut vowels with new analogical ones, as outlined in (9) and (10) above, but also with the exploitation of this alternation by ‘functional ru¨ckumlaut’: i.e. analogical back-formations based on the
264
C. Fehringer
umlaut opposition. For instance, Lu¨ssy (1974: 148) notes that in some Swiss dialects the umlaut alternation /u/: /i/ (with unrounding of the umlaut reflex) in nouns and their corresponding diminutives sometimes serves as the base for the creation of new analogical augmentatives (usually with pejorative overtones). Thus, on the basis of alternations such as /hu:be/ : /hi:bli/ ‘bonnet : little bonnet’, /blu:ze/ : /bli:sli/ ‘blouse : little blouse’ (from Basle), forms such as /vu:be/ ‘awful woman’ are created from the existing diminutive /vi:bli/ (base noun /vi:b/ ‘woman’). This tendency can especially be seen in the analogical creation of augmentative/jocular forms of certain personal names (Lu¨ssy 1974: 184): (22) Gritli Willi Schmid Fischer
: : : :
Grute Wulle d’Schmudle d’Fuschle
Maggie Willie Smith Fisher
: : : :
Big Margaret Big Willie Big Smith Big Fisher
Such developments are analogous to the creation of analogical singular forms opposite frequently occurring plurals with umlaut, which has been attested for a number of High and Low German dialects, e.g. Hessen (Schirmunski 1962: 443): (23)
/fuks/ : /fiks/ (original alternation)
fox
: foxes
(/mik/ >) /muk/ : /mik/ (elimination of umlaut in sg.) midge : midges (/fis/ >) /fus/ : /fis/ (change of /i/>/u/ in sg.)22 fish : fish(es) Thus, the relationship between nouns and their corresponding diminutives is perceived similarly to that between nouns and their corresponding plural forms: i.e. as a productive alternation marked by a [+back] : [−back] feature opposition. By contrast, when producing adjective comparatives, superlatives and abstract nouns the speaker seems less concerned with the nature of the base vs. derivative alternation than with the uniform marking of the derivatives. Thus, in Glarn (Swiss), the /a:/: /e:/ alternation is obscured by the analogical extension of /e:/ across paradigms to the comparative etc. forms of base adjectives in the front vowel /æ/ (examples repeated from (5c) above): (24)
ADJ
COMP
SUPERL ABST N
a.
/ra:ss/ /bla:b/
/re:sser/ /re:ssist/ /ble:ber/ /ble:bst/
b.
/spæ:t/ /spe:ter/ /tsæ:x/ /tse:xer/
/spe:tist/ /tse:xist/
/re:ssi/ /ble:bi/
strong blue
/spe:ti/ /tse:xi/
late tough
What is important here is that the derivatives within the morphological gang all
Morphological ‘gangs’
265
have the same stem vowel. Of course, the fact that the extension of /e:/ in comparatives etc. has been limited to a-type adjectives (i.e. adjectives in /a:/ and /æ:/) indicates that we are still dealing with some sort of alternation, otherwise there would be nothing to stop /e:/ from spreading as a general ‘comparative marker’ to adjectives in /u:/ and /o:/.23 However, these data do seem to indicate that the relationship between base and derivative in the adjectival gang is perceived by speakers to be different to that of nominal bases and their derivatives. Indeed, I have found no evidence of a noun plural or diminutive vowel being spread across paradigms.24 In addition to the type of analogical change occurring within a morphological gang, one must also consider the extent to which the change is realised and the connection between the members of the gang. For instance, most German dialects have comparatives, superlatives and deadjectival abstract nouns based on the same stem, and when any type of analogical change occurs it affects all these categories at the same time. By contrast, the influence of noun plurals on diminutive forms is restricted to a relatively small number of dialects, although it should be noted that the phenomenon is quite widespread and occurs in Upper, Central and Low German. However, in many of these dialects the influence of noun plurals on diminutives is seen as an added complication which tends to be eliminated in favour of the independent application of a more regular diminutive formation process. For example, in St Gallen (Switzerland) older speakers tend to preserve the connection between noun plurals and diminutives while the younger ones tend to form the diminutive independently of the noun plural. This applies both to umlaut and analogical vowel lengthening (Berger 1913: 32): (25) N
NPL
DIM
a.
/naxt/ /stat/ /gla:s/ /ra:d/
/næxt/ /stett/ /gleser/ /reder/
/næxtli/ /stettli/ /glesli/ /redli/
(older speakers – DIM influenced by N PL)
b.
/naxt/ /stat/ /gla:s/ /ra:d/
/næxt/ /stett/ /gleser/ /reder/
/næxtli/ /stættli/ /glæ:sli/ /ræ:dli/
(younger speakers – DIM shows more general [+back] : [−back] alternation with base N and analogical lengthening of DIM vowel when base N has a long vowel)
By contrast, differences between older and younger speakers with respect to adjective comparatives, superlatives and abstract nouns reveal the opposite tendency: younger speakers make new connections between the members of the adjectival gang (see e.g (18) above). The perception of the n pl 6 dim relationship as more of a complication to
266
C. Fehringer
the grammar than the comp 6 superl 6 abst n relationship may be due to the fact that noun plurals are not marked as uniformly as adjective comparatives and this has an important effect on stem allomorphy. That is to say, the occurrence of e.g. /e/ vs. /æ/ umlaut or vowel lengthening in noun plurals often depends on the accompanying plural suffix (e.g. /e/ umlaut usually occurs with -er, /æ/ usually occurs with -e or -ø and umlaut never occurs with -en and rarely with -s). Thus, it is much simpler to derive diminutives directly from the base noun without reference to the complicated lexically conditioned noun plural forms. By contrast, adjective comparatives are not marked by competing suffixes which affect the application of umlaut and lengthening rules. They are uniquely marked by -er (or phonological variants thereof) and, therefore, the occurrence of e.g. umlaut, vowel lengthening, is much more uniform and predictable in this category.
6. SUMMARY AND CONCLUSIONS The examination of analogical change in the morphology of modern German dialects reveals an interesting pattern. Due to historical phonological developments, certain morphological categories become associated with each other formally and, subsequently, various types of analogical change conspire to reinforce this association between some, but not all, of the categories. Specifically, what is affected are subsets of possible derivatives from the same base, referred to to here as morphological ‘gangs’, which appear to be based on purely formal criteria: a case of ‘Morphology by Itself’. The analogical changes that occur often appear to be ‘unnatural’ in that they lead to a complication of the grammar as a whole, e.g. the existence of two productive umlaut variants /e/ and /æ/ for bases in /a/, but are predictable and systematic in terms of the morphological gang in which they occur. These changes often serve to highlight the formal connection between the members of a gang and, at the same time, differentiate them from members of other gangs (e.g. /e/ in adjective comparatives, superlatives and deadjectival abstract nouns versus /æ/ in noun plurals and diminutives). As is the case with most analogical change, the developments outlined here are paradigmatic, yet as the areas of influence cut across the inflection/ derivation divide one must use the term ‘paradigm’ in its wider sense. The fact that certain types of derivational morphology may be influenced by the formal properties of traditionally inflectional stems argues against a theory of the lexicon in which inflection and derivation are strictly separated. It has been noted, however, that the ‘inflectional’ categories noun plural, comparative and superlative, on the one hand, and the ‘derivational’ diminutive belong precisely to those categories which have often been termed ‘transitional’ or ‘non-prototypical’ with regard to inflection/derivation (e.g. Scalise 1988, Dressler 1989,
Morphological ‘gangs’
267
Booij 1996). Thus, morphological gangs operate primarily in this ‘grey area’ of morphology. Once the concept of morphological gangs has been recognised, apparently idiosyncratic analogical developments can often be shown to behave quite systematically. Thus, if some diminutives acquire, for example, an unetymological intonation pattern not present in other derivational forms it is likely to be due to the influence of the corresponding noun plural (see (14) above). Furthermore, one can predict that any analogical change occurring in one member of a morphological gang will automatically spread to the other members. Thus, a change in adjective comparatives will also affect superlatives and abstract nouns, but not necessarily other derivatives of the same base adjectives such as, for example, deadjectival nouns (see (5) and (8)) or deadjectival verbs (see (3)). Similarly, one can predict that a change occurring in noun plurals is more likely to affect the corresponding diminutives than other forms derived from the same base noun, e.g. denominal masculine animate nouns (see (9) and (11)) or denominal verbs (see (13)). Indeed, such developments raise a much more general question: is it possible to predict which types of morphological derivatives are likely to become formally associated with each other over time or is it a purely arbitrary process? In other words, what determines morphological gang membership? With regard to the German dialect data it appears that factors such as productivity and semantic transparency play an important role in determining which forms belong to the gang and which take a different direction. For instance, as comparatives, superlatives and abstract nouns in -i (or -e) in the dialects are extremely productive and semantically predictable, they are less subject to lexicalisation than, for example, Swiss deadjectival nouns in -lig which are often semantically opaque, and deadjectival verbs such as Aldenrade /ferkElden/ whose productivity is restricted by the existence of competing rules for deriving deadjectival verbs using different prefixes and periphrastic constructions.25 Significantly, it is these more lexically-determined forms which usually do not undergo analogical change as speakers do not make such a strong connection between these and their original base forms. This is not to say, however, that non-gang members can be dismissed as isolated lexical items and that the only ‘real’ derivatives are the gang members. Productivity is a matter of degree, and certain non-gang categories can have some degree of productivity, albeit lower than that of the gang categories. For instance, in some Swiss dialects, denominal verbs in -e show no evidence of being systematically influenced by the gang, but are nevertheless quite productive: e.g. /gra:se/ ‘to graze’, /ba:de/ ‘to bathe’, /ta:ge/ ‘to become day’ (Henzen 1927: 102). It must also be noted that some dialects have similar analogical processes but the difference is that the processes are more widespread: i.e. they include all but the most lexically isolated derivatives. For instance, in the Swiss
268
C. Fehringer
dialect of Kesswil in Oberthurgau, derivational forms such as /sœ:ffer/ ‘shepherd’ from /sc:ff/ ‘sheep’ and /blœ:bele/ ‘blue mark/bruise’ from /blc:b/ ‘blue’ have been affected by analogy, while the equivalent forms in dialects such as Arlon (Luxembourg) and Glarn (also Switzerland) have not (see (9) and (5)). This means that in Kesswil, and similar dialects, the concept of gangs is superfluous. Thus, returning to the dialects which demonstrate gang effects, if we assume that some productive and transparent stems are not lexically listed but are predictable in terms of a general morphological rule we can account for the fact that once a German dialect speaker learns how to form the comparative of a particular adjective s/he can automatically produce the related superlative and abstract noun, and any change affecting the former will automatically be applied to the latter, e.g. Entlebuch (Schmid 1915: 113–115) /ra:n/ : /re:nner/ : /re:nst/ : /re:ni/ with umlaut and analogical lengthening versus /gra:d/ : /greder/ : /gretst/ : /gredi/ with umlaut but without lengthening. Similarly, as diminutives are mostly semantically transparent and extremely productive in the dialects (much more so than in standard German) they do not need to be lexically listed and can be predicted on the basis of more general rules. In some dialects this will include reference to the noun plural stem, although it was noted above that this influence is in the process of elimination in some dialects as it is seen as an added complication to the grammar (see (25)). This may be due to the fact that the reference to noun plural stems is more of a complicated matter than reference to adjective comparatives, as stem allomorphy in noun plurals is largely dependent on the co-occurring plural suffix, of which there are at least three competing lexically-conditioned variants (-er, -en, -ø and, in some dialects -e and -s) while, by contrast, adjective comparatives are uniformly marked by -er and stem allomorphy is more predictable. Thus, morphological gangs are more likely to survive if they contain as little arbitrary lexical information as possible. Such developments also tie in with the observation made in section 6 above that analogical changes affecting noun plurals and diminutives tend to highlight the alternation between base and derivative rather than accentuate formal links between derivatives, which suggests that there is more of a general tendency to derive diminutives directly from their base nouns. Hence the use of an umlaut vowel in Swiss noun plurals and diminutives which conforms to a more salient [+back]: [−back] opposition than the less natural /a/: /e/ alternation, the latter of which survived and became productive in the adjectival gang due to the sheer number of members marked by it (i.e. comp, superl + abst n). Thus, the quantitative dominance of particular stem allomorphs within a particular morphological gang must also be taken into account when examining analogical change.
Morphological ‘gangs’
269
NOTES 1
This is a revised version of a paper given at a meeting of the Philological Society in 1999. I would like to thank those people present for their useful comments. 2 Although it will be shown later that the inflectional categories concerned are well-known for displaying characteristics similar to derivation and are therefore not prototypical. 3 The term ‘gang’ is quite common in the psycholinguistic literature and is used in connection with ‘neighbourhood effects’ (usually when referring to word recognition) and ‘community support’ (see e.g. Obler and Gjerlow 1998: 117). 4 For details see Durrell (1989). 5 Deadjectival abstract nouns in /-i/ are extremely productive in the Swiss dialects, much more so than the standard German equivalents in -e, e.g. La¨nge, Ka¨lte, which are outnumbered by forms in -heit, e.g. Scho¨nheit ‘beauty’, Ku¨hnheit ‘boldness’, compare MHG schœn-e, ku¨en-e. See Fleischer/Barz 1995: 147, 158–163 and, for more detail, Wellmann 1993). 6 /e/ replaces etymological /œ/ which appeared before /x/ in OHG. 7 This dialect has an analytic comparative, e.g. /me:r ka:l/ lit. ‘more cold’. 8 Unless, like Blevins (2001), one considers comparatives and superlatives to be derivational, rather than inflectional, forms (see also Robins 1959: 126). 9 Although these forms share the same base they do not constitute a paradigm in the strict sense of ‘a pattern [...] of inflexional realisations for all combinations of non-lexicallydetermined morphosyntactic properties ...’ (Carstairs 1987: 48–49), yet they do in the wider sense of ‘a series of morphologically related forms which share a base or base type’ (van Marle 1994: 2927). For a full discussion of the different interpretations of the word ‘paradigm’ and the problems associated with the term see Ford and Singh (1985) and Bauer (1997). 10 Causative verbs derived from adjectives in /a/ also tend to have /e/, e.g. Kesswil (Enderlin 1913: 53) /werme/ ‘to warm’ /xelte/ ‘to make cold’, /sterke/ ‘to strengthen’, yet this use of /e/ is less systematic than in the comparatives, superlatives and abstract nouns. 11 /a/>/a:/ in monosyllabic words ending in a single lenis obstruent or original sonorant. 12 Note that the corresponding word in the dialect of Kesswil in Oberthurgau is /blœ:bele/ with the analogical umlaut of /c:/ (cf. /blc:b/ ‘blue’). Indeed, this is one of the dialects that systematically have the analogical umlaut for many derivatives outside the morphological gang, except for isolated lexicalized forms. Thus, the concept of gangs only operates in those dialects whose analogical forms have not simply been extended to all derivatives. 13 Note also the analogical inflected adjective /en folne bu:x/ ‘a full belly’. 14 Although Bertrang does not give many examples, he makes it clear that the analogical umlaut is a feature of noun plurals and diminutives: ‘Anstatt /EI/ steht /E:/ als Umlaut von wg. /a:/, Ar. /c:/, im Plural der Substantive und in den Diminutiven’ (Bertrang 1921: 127). 15 The plural suffix -en does not occur with umlaut. 16 If the noun singular has the falling intonation the diminutive usually has it, too. 17 /d/ is often inserted between /n/ and /l/ for ease of pronunciation. 18 Some linguists may wish to argue that since diminutives are used very frequently in Swiss dialects, forms such as /man(d)li/ etc. could be lexicalised. However, other lexicalised forms such as /gmæIsam/ ‘together’ (standard German gemeinsam) do delete -n. 19 Except in the case of noun plurals which are suffixless in some dialects, e.g. Arlon /drc:t/: /drE:t/ ‘wire(s)’. 20 Maiden (1996) refers to this principle as ‘stabilisation’. He demonstrates how some
270
C. Fehringer
apparently ‘unnatural’ alternation patterns in a handful of Romance verbs were extended analogically on the basis of purely formal intraparadigmatic relations. 21 Such rules are different from the n-insertion rule in Zu¨rich which can no longer be regarded as purely phonological due to the analogical extension of -n to derivatives whose base forms do not have -n underlyingly (see (8) above). 22 Here, the /i/ is not a product of umlaut (contrast standard German Mu¨cke ‘midge’ with Fisch ‘fish’). 23 c.f. the spread of /EI/ as a general preterite subjunctive marker to most strong verbs in Luxemburg due to the loss of the corresponding preterite indicative forms and therefore the loss of the indicative vs. subjunctive alternation (see Chapman 1996: 51–52). 24 Although, conversely, I have found isolated examples of back vowels in base adjectives which may be cases of analogical ru¨ckumlaut, e.g. /ra:ss/ ‘strong’ (c.f. MHG ræZe) from Glarn in (24) above and /spc:t/ ‘late’ (c.f. MHG spæte) in Lungau/Salzburg (Mauser 1998: 212). However, this does not appear to be a particularly widespread process. 25 Compare standard German kalt : sich erka¨lten ‘cold : to catch cold’, heiss : heizen/erhitzen ‘hot : to heat’, frei : befreien ‘free : liberate’, trocken – trocknen ‘dry : to dry’, nass – nass machen ‘wet : to wet’ etc.
REFERENCES Anderson, S.R. (1992). A-morphous Morphology. Cambridge: CUP. Arens, J. (1908). Der Vokalismus der Mundarten im Kreise Olpe. Ph.D. dissertation. BornaLeipzig: Noske. Aronoff, M. (1994). Morphology by Itself. Cambridge, Mass.: MIT Press. Bach, E. and R.D. King (1970). Umlaut in modern German. Glossa 4/1, 3–22. Bauer, L. (1997). Derivational paradigms. In: G.E. Booij and J. van Marle (eds.), Yearbook of Morphology 1996. Dordrecht: Kluwer, 243–256. Becker, T. (1990). Analogie und morphologische Theorie. Munich: Wilhelm Fink. Berger, J. (1913). Die Laute der Mundarten des St. Galler Rheintals. Beitra¨ge zur schweizerdeutschen Grammatik 3. Frauenfeld: Huber. Bertrang, A. (1921). Grammatik der Areler Mundart. Brussels: Academie Royale de Belgique. Beyer, E. (1963). La flexion du groupe nominale en alsacien. Paris: Les Belles Lettres. Blevins, J.P. (2001). Paradigmatic derivation. Transactions of the Philological Society 99/2, 211–222. Booij, G.E. (1996). Inherent versus contextual inflection and the split morphology hypothesis. In: G.E. Booij and J. van Marle (eds.), Yearbook of Morphology 1995. Dordrecht: Kluwer, 1–17. Booij, G.E. (1997). Autonomous morphology and paradigmatic relations. In: G.E. Booij and J. van Marle (eds.), Yearbook of Morphology 1996. Dordrecht: Kluwer, 35–55. Born, W. (1983[1978]). Kleine Sprachlehre des Mu¨nsterla¨nder Platt. Mu¨nster: Regensberg. Braune, W. and H. Eggers (1987). Althochdeutsche Grammatik. Tu¨bingen: Niemeyer. Bybee, J.L. and M.A. Brewer (1980). Explanation in morphophonemics: changes in Provencal and Spanish preterite forms. Lingua 52, 201–242. Cameron-Faulkner, T. and A. Carstairs-McCarthy (2000). Stem alternants as morphological signata: evidence from blur avoidance in Polish nouns. Natural Language and Linguistic Theory 18, 813–835.
Morphological ‘gangs’
271
Carstairs, A. (1987). Allomorphy in Inflexion. London: Croom Helm. Carstairs-McCarthy, A. (2001). Umlaut as signans and signatum: synchronic and diachronic aspects. In: G.E. Booij and J. van Marle (eds.), Yearbook of Morphology 1999. Dordrecht: Kluwer, 1–23. Chapman, C. (1994). A diachronic argument against the Split Morphology Hypothesis: analogical umlaut in German dialects. Transactions of the Philological Society 92/1, 25–39. Chapman, C. (1995). Perceptual salience and analogical change: evidence from vowel lengthening in modern Swiss German dialects. Journal of Linguistics 31, 1–13. Chapman, C. (1996). Anomalies in the strong verb paradigms of two West Central German dialects. Zeitschrift fu¨r Dialektologie und Linguistik 63, 49–57. Chapman, C. (1997). Diminutive plural infixation and the ‘West Franconian’ problem. In: J. Fisiak (ed.), Linguistic Reconstruction and Typology. Berlin, New York: Mouton de Gruyter, 73–89. Dressler, W.U., W. Mayerthaler, O. Panagl and W.U. Wurzel (1987). Leitmotivs in Natural Morphology. Amsterdam: Benjamins. Dressler, W.U. (1989). Prototypical differences between inflection and derivation. Zeitschrift fu¨r Phonetik, Sprachwissenschaft und Kommunikationsforschung 42, 3–10. Durrell, M. (1989). Umlaut in Old High German and after. In Flood, J.L. and D.N. Yeandle (eds.), Mit regulu bithuungan. Neue Arbeiten zur althochdeutschen Poesie und Sprache. Go¨ppingen: Ku¨mmerle, 219–232. Enderlin, F. (1913). Die Mundart von Kesswil im Oberthurgau. Beitra¨ge zur schweizerdeutschen Grammatik 5. Frauenfeld: Huber. Fleischer, W. and I. Barz (1995). Wortbildung der deutschen Gegenwartssprache. Tu¨bingen: Niemeyer. Ford, A. and R. Singh (1983). On the status of morphophonology. In: J.F. Richardson et al. Papers from the Parasession on the Interplay of Phonology, Morphology and Syntax. Chicago: Chicago Linguistic Society. Henzen, W. (1927). Die deutsche Freiburger Mundart im Sense- und su¨dostlichen Seebezirk. BSG 16. Frauenfeld: Huber. Kats, J. (1939). Het phonologisch en morphologisch systeem van het Roermondsch dialect. Roermond/Maseik: J.J. Romen and Sons. Lu¨ssy. H. (1974). Umlautprobleme im Schweizerdeutschen. Frauenfeld: Huber. Maiden, M. (1996). The Romance gerund and ‘system-dependent naturalness’ in morphology. Transactions of the Philological Society 94/2, 73–118. Mauser, P. (1998). Die Morphologie im Dialekt des Salzburger Lungaus. Frankfurt am Main: Peter Lang. Mayerthaler, W. (1981). Morphologische Natu¨rlichkeit. Wiesbaden: Athenaion. Meinherz, P. (1920). Die Mundart der Bu¨ndner Herrschaft. Beitra¨ge zur schweizerdeutschen Grammatik 13. Frauenfeld: Huber. Moulton, W. (1971). Der morphologische Umlaut im Schweizerdeutschen. In: M. Bindschedler et al. (eds.), Festschrift fu¨r Paul Zinsli. Bern: Francke, 15–25. Neuse, H. (1915). Studien zur niederrheinischen Dialektgeographie in den Kreisen Rees, Dinslaken, Hamborn, Mu¨hlheim, Duisburg. Deutsche Dialektgeographie 8. Marburg: Elwert. Obler, L.K. and K. Gerlow (1998). Language and the Brain. Cambridge/New York: CUP. Perlmutter, D.M. (1988). The Split Morphology Hypothesis: evidence from Yiddish. In:
272
C. Fehringer
M. Hammond and M. Noonan (eds.), Theoretical Morphology. London: Academic Press, 79–100. Robins, R.H. (1959). In defence of WP. Transactions of the Philological Society, 116–144. Scalise, S. (1988). Inflection and derivation. Linguistics 26/4, 561–583. Schirmunski, V. (1962). Deutsche Mundartkunde. Berlin: Akademie-Verlag. Schmid, K. (1915). Die Mundart des Amtes Entlebuch im Kanton Luzern. Beitra¨ge zur schweizerdeutschen Grammatik 7. Frauenfeld: Huber. Strauss, S.L. (1982). Lexicalist Phonology of English and German. Dordrecht: Foris. Streiff, C. (1915). Die Laute der Glarner Mundarten. Beitra¨ge zur schweizerdeutschen Grammatik 8. Frauenfeld: Huber. Van Marle, J. (1985). On the Paradigmatic Dimension of Morphological Creativity. Dordrecht: Foris. Van Marle, J. (1996). The unity of morphology: on the interwovenness of the derivational and inflectional dimension of the word. In: G.E. Booij and J. van Marle (eds.), Yearbook of Morphology 1995. Dordrecht: Kluwer, 67–83. Weber, A. (1948). Zu¨richdeutsche Grammatik, ein Wegweiser zur guten Mundart. Zu¨rich: Schweizer Spiegel Verlag. Wellmann, H. (ed.) (1993). Synchrone und diachrone Aspekte der Wortbildung im Deutschen. Heidelberg: Winter. Wipf, E. (1910). Die Mundart von Visperterminen im Wallis. Beitra¨ge zur schweizerdeutschen Grammatik 2. Frauenfeld: Huber. Wurzel, W.U. (1970). Studien zur deutschen Lautstruktur. Berlin: Akademie-Verlag. Wurzel, W.U. (1984a). Was bezeichnet der Umlaut im Deutschen? Zeitschrift fu¨r Phonetik, Sprachwissenschaft und Kommunikationsforschung 37, 647–663. Wurzel, W.U. (1984b). Flexionsmorphologie und Natu¨rlichkeit. Berlin: Akademie-Verlag. Zwicky, A. (1967). Umlaut and Noun Plurals in German. Studia Grammatica 6. Berlin: Akademie-Verlag.
School of Modern Languages University of Newcastle Old Library Building Newcastle upon Tyne NE1 7RU UK [email protected]
Book Reviews Jochen Zeller, Particle Verbs and Local Domains. Amsterdam: John Benjamins, 2001, xi+323 pp., Eur 49,46. ISBN 90 272 2762 4. Reviewed by Geert Booij
This book is the published version of a doctoral dissertation defended by Zeller (henceforth Z) at the University of Frankfurt. It deals with the topic of German particle verbs such as anrufen ‘to phone’, a topic that is also the focus of the thematic section of this volume of the Yearbook of Morphology to which the reader of this review is referred for further information on the phenomena in this domain of linguistic research. Z’s basic claim is that particle verbs are not words but have phrasal status and form a V∞-constituent. That is, particle verbs have the structure [[Prt]PrtP V0 ]V∞ . Z adduces a number of arguments for the position that particle verbs do not have the status of complex words (as was claimed by Stiebels & Wunderlich (1994) for German and by Neeleman & Weerman (1993) for Dutch). The basic arguments for assigning particle verbs phrasal status is that the two parts can be split by Verb Second and Topicalization, and that the principle of Lexical Integrity forbids splitting of words by rules of syntax. On the other hand, the existence of particle verbs does not follow directly from the standard syntactic rules of German. Therefore, Z proposes that words that can function as particles have a subentry for their use as particle, in which they are subcategorized for combining with verbs in the structure given above. This position concerning the analysis of particle verbs is similar to the one I defended for Dutch particle verbs in Booij (1990, 2002a;b), where I also argued for the phrasal status of particle verbs. As Zeller rightly emphasizes, the view of the lexicon as defended in Jackendoff (1997) is a perfect framework for such an analysis: the lexicon specifies the correspondence between pieces of phonological, syntactic, and semantic information and it is therefore quite feasible to specify properties of multi-word units in the lexicon. As Z points out, it is necessary to list particle-verb combinations in the lexicon if they have a noncompositional meaning, which is often the case (e.g. aufho¨ren ‘to stop’, and anfangen ‘to begin’. Z’s analysis thus differs from those syntactic approaches to particle verbs that consider particles as predicates of small clauses, and derive particle verbs through syntactic incorporation. This is the approach taken in Den Dikken (1995) for English, and also defended in Den Dikken’s review of Z’s book (Den Dikken 2001). In my opinion, the basics of Z’s analysis and argumentation are quite convincing, and they converge with what I said about particle verbs in Dutch. One Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2003, 273–279. © 2003 Kluwer Academic Publishers. Printed in Great Britain.
274
Book Reviews
difference between his and my analysis is that in Booij (2002a;b) I qualified words that function as particles as ‘non-projecting’, implying that there is no phrase node, but only an X0 node dominating the particle (which can be P, N, A or Adv). Z asssumes that a particle does project a full phrase, because it can be topicalized under certain semantic or pragmatic conditions (see also Zeller, this volume). This reasoning presupposes that being a full phrase is a precondition for being topicalized. On the other hand, by assigning particles an X0 status only, we correctly predict that the normal unconditioned topicalization does not apply. A second argument of Z’s for assigning phrasal status to particles is that they can be modified (p. 100 ff). Z cites, among others, the following German sentences as evidence for this position: (1)
a. b.
Das Kleid da hinten sieht besser aus ‘This dress over there in the back looks better’ [Besser aus] sieht das Kleid da hinten
Sentence (1b), with topicalization of besser aus, is meant to show that besser, which can also be interpreted as an independent constituent of the sentence, can function as the modifier of the particle aus. However, the grammaticality of this sentence is questionable. In other cases where modification of the particle word is possible, the modified word may also be interpreted as a secondary predicate instead of as a particle. That is, there may be structural ambiguity involved. For instance, in the Dutch sentence (2)
a.
Jan maakte zijn huiswerk af ‘John finished his homework’
the word af may be interpreted either as a particle or as a phrase that functions as a secondary predicate (the construction that is a historical source of the verb particle construction), since af also occurs as a predicate without a verb, as in Het huiswerk is af ‘The homework is ready’. In such cases, modification is possible, as in (2)
b.
Jan maakte zijn huiswerk helemaal af ‘John finished his homework completely’
In those cases where a particle interpretation is the only possible one because of the idiomatic nature of the particle verb, modification is excluded, as in: (2)
c.
*Jan belde zijn moeder helemaal op ‘John phoned his mother completely’
where op cannot be interpreted as a secondary predicate with independent
Book Reviews
275
syntactic status. In other words, modification may be taken to indicate the nonparticle status of the word involved. So the problem for Z’s hypothesis of phrasal status for particles is that in most cases modification of particles is impossible, which requires a systematic explanation that a phrasal status of particles does not provide. Interestingly, Toivonen (2001) has argued on the basis of an analysis of Swedish particle verbs that the grammar must allow for non-projecting categories, since in Swedish the postverbal particles can never be modified. A special assumption made by Z is that particle phrases differ from other phrases in that they lack the functional projection that, for instance, turns an NP into a DP. The effect is according to Z that there is a relation of structural adjacency between a particle and a verb: the particle is head-governed by the verb. Z’s definition of this notion reads as follows (p. 36): (3)
Structural adjacency A head X and the head Y of its complement YP are structurally adjacent
According to Z, this structural adjacency implies that particle and verb are together in a local domain, which makes it possible for this word combination to be reanalysed as V0 under certain circumstances, which I will discuss now. If the particle phrase were dominated by a functional projection layer, there would not be structural adjacency anymore between particle and verb, because instead of Prt the head of the Functional Projection of the PrtPhrase is structurally adjacent to V, and hence in the same local domain. Z gives two arguments for the idea of reanalysis of particle verbs into V0 . One is that they can occur in the non-head position of compounds and derived words. Since according to Z derived words cannot contain phrases (the No Phrase Constraint), unlike compounds, the particle verbs must have been reanalysed since they do feed word derivation, as Z rightly points out. For instance, we can derive Anrufer ‘caller’ from anrufen ‘to call’, and the same applies to Dutch particle verbs (Booij 1990). However, I do not consider this a strong argument since not only particle verbs but also other phrases can feed derivation in Dutch, as illustrated by the following examples with the Dutch nominal suffix -er: (4)
derde-klass-er ‘third form pupil’<[derde klas]NP ‘third form’ harde-kern-er ‘hard core member’<[harde kern]NP ‘hard core’
Hence, the No Phrase Constraint is emprically incorrect, and thus no argument for reanalysis of particle verbs as V0. Z’s second argument for reanalysis is based on the phenomenon of Verb Raising in Dutch: not only verbs, but also particle verbs can be raised as units to the right of the verb of the higher clause, as in:
276
(5)
Book Reviews
a. b. c.
dat Jan zijn moeder op wil bellen ‘that John wants to phone his mother’ dat Jan zijn moeder wil opbellen *dat Jan wil zijn moeder opbellen
According to Z, the grammaticality of (5b) in which the particle is raised with the verb, forces us to assume that the particle verb combination is a V0. As Z himself notes in a footnote on p. 278, this creates a serious problem because it implies that the combination of a raising verb and a raised te-infinitive also receives the status of single verb: proberen te zingen ‘try to sing’ would count as one verb! Instead, we can assume that not only verbs can be raised to higher clauses, but also verbal projections (cf. Booij 1990). Note that in Southern Dutch complete VPs can be raised, for instance: (6) dat Jan wil [zijn moeder bellen] ‘that John wants to phone his mother’ Clearly, we do not want to say that because of raising the word sequence zijn moeder bellen must be reanalysed as one V0. An important property of particles is that they seem to have categorychanging power since they also combine with nouns and adjectives, both in German and in Dutch (Booij 1990). This implies that particles can trigger the conversion of adjectives and nouns into verbs. Z points out that his account can handle this. For instance, German has the particle verb anreichern ‘to enrich’, but no established verb reichern. Consequently, the particle an- is specified as occurring before a verb with the structure [[reicher]A Ø]V . In other words, the particle triggers the morphological operation of conversion. A similar analysis is proposed in Booij (2002a,b) where I also observe that it is specific particles that trigger conversion; this can be expressed by a lexical template of the relevant form. In sum, Zeller has written an excellent study of (German) particle verbs, with convincing argumentation for their phrasal status. I am not yet convinced, however, of the full phrasal status of the particles themselves, and of the necessity of reanalysis of particle verbs into V0 constituents. REFERENCES Booij, Geert (1990). The boundary between morphology and syntax: separable complex verbs in Dutch. In: Geert Booij and Jaap van Marle (eds.), Yearbook of Morphology 1990. Dordrecht: Foris, 45–63. Booij, Geert (2002a). The Morphology of Dutch. Oxford: Oxford University Press. Booij, Geert (2002b). Separable complex verbs in Dutch: a case of periphrastic word formation. In: Nicole Dehe´, Ray Jackendoff, Andrew McIntyre and Silke Urban (eds.), VerbParticle Explorations. Berlin: Mouton de Gruyter, 21–41.
Book Reviews
277
Dikken, Marcel den (1995). Particles. On the Syntax of Verb-Particle, Triadic and Causative Constructions. Oxford//New York: Oxford University Press. Dikken, Marcel den (2001), Review of Zeller (2001). Journal of Comparative Germanic Linguistics 4, 145–169. Jackendoff, Ray. S. (1977). The Architecture of the Human Language Faculty. Cambridge, Mass.: MIT Press. Lu¨deling, Anke (2001). On Particle Verbs and Similar Constructions in German. Stanford: CSLI. Neeleman, Ad and Fred Weerman (1993). The balance between syntax and morphology: Dutch particles and resultatives. Natural Language and Linguistic Theory 11, 433–475. Stiebels, Barbara and Dieter Wunderlich (1994). Morphology feeds syntax: the case of particle verbs. Linguistics 32, 913–968. Toivonen, Ida (2001). The phrase structure of non-projecting words. Ph.D. dissertation, Stanford University. Zeller, Jochen (this volume). Moved preverbs in German: displaced or misplaced?
Morphology 2000. Selected Papers from the 9th Morphology Meeting, Vienna, 24–28 February 2000. Edited by S. Bendjaballah, W.U. Dressler, O.E. Pfeiffer and M.D. Voeikova. Amsterdam/Philadelphia: John Benjamins. 2002, viii, 317 pp, ISBN 90 272 3725 5, Eur 105. Reviewed by Geert Booij
This volume contains 24 papers, the focus of which is on comparative morphology (cross-linguistic analysis, including typology, dialectology and diachrony) and on psycholinguistic issues. It is impossible to discuss all these papers in a book notice. Therefore, I will only mention some highlights of this useful and inspiring volume. Mark Baker, in ‘On category asymmetries in derivational morphology’ observes that languages tend not to derive stative or change of state verbs from nouns. For instance, the English denominal verb to dust does exist, but does not receive the meaning ‘to become/to turn into dust’, in which dust receives a predicative interpretation. According to Baker, this has to do with the fact that this kind of word formation requires the referential index that every noun is associated with (in opposition to verbs and adjectives) to be suppressed, a marked process (although sometimes possible). A second piece of comparative morphology is Laurie Bauer’s ‘What you can do with derivational morphology’. Bauer observes a restricted set of recurrent meaning categories in derivational morphology across a large sample of languages, such as agent, instrumental, and diminutive. Another typological contribution, by Grev Corbett and his co-workers, points out how computational linguistics can be used to check the correctness
278
Book Reviews
of linguistic generalizations. This is illustrated by a typological analysis of gender. The authors argue that the assignment of gender to nouns can best be done by making use of, first of all, semantic categories, and if necessary, by additional formal properties such as declination class or phonological properties. The relations between these properties can be modelled in the so-called Network Morphology, a variety of the Word-and-Paradigm approach to morphology. In his contribution ‘How stems and affixes interact’, Andrew CarstairsMcCarthy argues that we can make generalisations about how inflection class systems work by making a fundamental distinction between affixal inflection and stem alternation. Thus, he aims at formulating constraints on how stem alternants can be distributed within paradigms. Wolfgang U. Dressler and Ma´ria Lada´nyi write about ‘On contrastive word formation semantics. Degrees of transparency/opacity of German and Hungarian denominal adjective formation.’ They present a useful survey of semantic factors that may make the literal compositional meaning of a derived word opaque, and hypothesize that an agglutinative language like Hungarian might exhibit less opacity than a language like German which is according to them more of the fusional/inflectional type. Their conclusion is that this hypothesis is incorrect. This should not perhaps come as a surprise because in the realm of derivational morphology there seems to be no clear typological distinction between Hungarian and German, both exhibiting agglutinative patterns. Bozena Cetnarowska argues that in languages such as Polish and English past participles can be converted into adjectives, also those of unaccusative verbs. The gaps that we find, such as ?an appeared book have to do with the Non-redundancy Constraint: such a phrase is possible, and makes sense, if an adverb is added, as in a recently appeared book. This constraint, proposed by Ackerman and Goldberg, is relevant for other word formation processes as well, such as the creation of middle verbs, which also requires an adverb to be present. In some of the psycholinguistic contributions, the issue of symbolic rules versus connectionist models, and the dual mechanism model of Pinker and Clahsen are debated. For instance, Hilke Elsen, in ‘The acquisition of German plurals’ argues that probabilistic models of the German plural noun inflection are more adequate that Clahsen’s dual mechanism account, in which one default plural suffix -s is assumed. Gary Libben and Roberto de Almeida, in their paper ‘Is there a morphological parser?’, argue in favour of the well known dual route model for the recognition of complex words, in which both whole word recognition and morphological parsing play a role. A number of papers deal with diachrony. Comrie, in his contribution ‘Morphophonological alternations: typology and diachrony’ discusses Celtic mutation and raddoppiamento syntactico in Italian, and shows that such
Book Reviews
279
alternations which became partially opaque in the course of time can develop into alternative means of indicating morphological oppositions such as singularplural for nouns. Another interesting paper on the topic of morphological change is Michele Loporcaro’s ‘External and internal causation in morphological change’. In sum, this conference volume gives us a good survey of present day issues and discussions in morphological theory and typology, and of the ongoing debate on the mental representation and processing of complex words.