Linguistics in the Netherlands 1987 [Reprint 2020 ed.] 9783112330364, 9783112330357

Citation preview

LINGUISTICS IN THE NETHERLANDS 1987

AVT PUBLICATIONS Vereniging AVT Publications is a series sponsored by the Algemene voor Taalwetenschap (Linguistic Society of the Netherlands). In addition to the annual publication of Linguistics in the Netherlands further publications, resulting f r o m other activities promoted by the Society, may appear in this series.

Still

available

Linguistics in the Netherlands 1977-1979 W i m Zonneveld and Fred W e e r m a n (eds) Dfl. 3 0 — / U S $21.90 Linguistics in the Netherlands 1983 Hans Bennis and W.U.S. van Lessen Kloeke (eds) Dfl. 3 4 . — / U S $ 2 1 . 9 0 Linguistics in the Netherlands 1986 Frits Beukema and Aafke Hulk (eds) Dfl. 3 8 . — / U S $24.90

Frits Beukema Peter Coopmans (eds)

1987 FORIS PUBLICATIONS Dordrecht - Holland/Providence - U.S.A.

Published by: Foris Publications Holland P.O. Box 5 0 9 3 3 0 0 A M Dordrecht, The Netherlands Sole distributor for the U.S.A. and Foris Publications U.S.A., Inc. P.O. Box 5 9 0 4 Providence R.I. 0 2 9 0 3 U.S.A.

Canada:

CI P-DATA Linguistics Linguistics in t h e Netherlands 1 9 8 7 / Frits Beukema, Peter Coopmans (eds). - Dordrecht [etc.]: Foris. - (AVT Publications ; 3) Papers presented at the 1 8 t h a n n u a l meeting of the ' A l g e m e n e Vereniging voor Taalwetenschap' (Linguistie Society of the Netherlands), w h i c h w a s held in Leiden o n J a n u a r y 24, 1987. ISBN 9 0 6 7 6 5 3 4 0 3 SISO 8 0 5 UDC 8 0 1 (492) Subject heading: linguistics ; The Netherlands.

ISBN 9 0 6 7 6 5 3 4 0 3 © 1 9 8 7 A l g e m e n e Vereniging voor T a a l w e t e n s c h a p seer.: J. Lalleman, Dutch Studies, Univ. of Leiden, P.O. Box 9 5 1 5 , 2 3 0 0 RA Leiden. No part of t h i s publication may be reproduced or t r a n s m i t t e d in any f o r m or by any means, electronic or mechanical, including photocopy, recording, or any i n f o r m a t i o n storage and retrieval system, w i t h o u t permission f r o m t h e copyright o w n e r . Printed in The Netherlands by ICG Printing, Dordrecht.

Contents

Preface List of contributors Hans Bennis & Pim Wehrmann Adverbial arguments Egon Berendsen & Els den Os Glide insertion: domains, speech rate and phonetic prominence

1

13

Lisan van Beurden Playing level with Dutch morphology

21

Saapke Bouma The discourse function of quantifier float: tous in French

31

Josée Coenen & Roeland van Hout Word order phenomena in second language acquisition of Dutch

41

Colin Ewen & Harry van der Hulst Single-valued features and the distinction between [ - F ] and [ 0 F ]

51

Liliane Haegeman Van constructions in Dutch

61

Aafke Hulk & Erna Vermeulen The impersonal construction in French and the Burziogeneralization

71

Harry van der Hulst & Norval Smith Vowel harmony in Khalkha and Buriat (East Mongolian)

81

Koen Jaspaert & Sjaak Kroon The relationship between global language proficiency tests and language loss

91

VI

Bob de Jonge Estar comes of age Johan Kerstens On the functional interpretation of prenominal adjectives in Dutch

101 111

Hans van de Koot On explaining subjacency Simone

Langeweg

Stress assignment in Dutch compound nouns Laszlo

121

131

Maracz

Some notes on the syntax and semantics of modification in Hungarian Martina Noteboom Hit and dative complements in Old English Florimon van Putte Alternating verb forms in Papiamentu

141 151

161

Bob Rigter How do you do do?

173

Ineke Schuurman Incorporation in the Groningen dialect

185

Annemarie S/ootweg Word stress and higher level prosodies

195

Frits

Stuurman

On Chomsky and Jespersen: two approaches to grammar

205

Arie Verhagen On certain functional approaches to word order Fred Weerman The change from OV to VO as a 'possible change' Wim van der Wurff Adjective plus infinitive in Old English

213 223 233

Preface

This volume contains a selection of the papers presented at the eighteenth a n n u a l meeting of the Society, held in Leiden on J a n u a r y 24, 1987. T h e aim of the a n n u a l meeting is to provide m e m b e r s of the Society with an o p p o r t u n i t y to report on their work in progress. At this year's meeting fifty-two papers were presented. T h e twenty-four papers contained in this volume present an overview of research in different fields of linguistics in the Netherlands. As in previous years, the a u t h o r s have submitted their articles in camera-ready f o r m . We are grateful to them f o r their cooperation. We also t h a n k a n u m b e r of colleagues w h o helped us referee the p a p e r s submitted for this volume. May 1987

Frits B e u k e m a Peter C o o p m a n s

List of contributors

H. Bennis University of Leiden, Department of Dutch, P.O. Box 9515, 2300 RA Leiden E. Berendsen University of Utrecht, Department of Phonetics, Trans 14, 3512 JK Utrecht L. van Beurden University of Utrecht, Research Institute for Language and Speech, Trans 14, 3512 JK Utrecht S. Bouma Obiplein 75, 1094 RB Amsterdam J. Coenen University of Brabant, Department of Language & Literature, P.O. Box 90153, 5000 LE Tilburg C. Ewen University of Leiden, Department of English, P.O. Box 9515, 2300 RA Leiden L. Haegeman University of Geneva, Département of English, 22, bd des Philosophes, 1205 Geneva, Switzerland R. van Hout University of Brabant, Department of Language & Literature, P.O. Box 90153, 5000 LE Tilburg A. Hulk Free University, Department of Linguistics, P.O. Box 7161, 1007 MC Amsterdam H. van der Hülst University of Leiden, Department of General Linguistics, P.O. Box 9515, 2300 RA Leiden K. Jaspaert University of Brabant, Department of Language & Literature, P.O. Box 90153, 5000 LE Tilburg B. de Jonge University of Leiden, Department of Latin American Studies, P.O. Box 9515, 2300 RA Leiden J. Kerstens University of Utrecht, Department of Dutch, Institute De Vooys, P.O. Box 80091, 3508 TB Utrecht

X

H. van de Koot University of Utrecht, Department of Linguistics, Trans 14, 3512 J K Utrecht S. Kroon University of Brabant, Department of Language & Literature, P.O. Box 90153, 5000 LE Tilburg S. Langeweg University of Leiden, Department of General Linguistics/Phonetics, P.O. Box 9515, 2300 RA Leiden L. Maräcz University of Groningen, Department of General Linguistics, Grote Rozenstraat 31, 9712 T G Groningen M. Noteboom University of Leiden, Department of English, P.O. Box 9515, 2300 RA Leiden E. den Os University of Utrecht, Department of Phonetics, Trans 14, 3512 J K Utrecht F. van Putte University of Leiden, Department of Latin American Studies, P.O. Box 9515, 2300 RA Leiden B. Rigter University of Leiden, Department of English, P.O. Box 9515, 2300 RA Leiden I. Schuurman University of Groningen, Department of Dutch, Grote Kruisstraat 2-1, 9712 TS Groningen A. S/ootweg University of Nijmegen, Department of Phonetics, P.O. Box 9103, 6500 H D Nijmegen N. Smith University of Amsterdam, Department of General Linguistics, Spuistraat 210, 1012 VT Amsterdam F. Stuurman University of Utrecht, Department of English, Oudenoord 6, 3513 ER Utrecht A. Verhagen Free University, Department of Linguistics, P.O. Box 7161, 1007 M C Amsterdam E. Vermeulen Free University, Department of Linguistics, P.O. Box 7161, 1007 M C Amsterdam F. Weerman University of Utrecht, Department of Linguistics, Trans 14, 3512 J K Utrecht

xi P. Wehrmann University of Leiden, Department of Dutch, P.O. Box 9515, 2300 RA Leiden W. van der Wurff University of Amsterdam, Department of English, Spuistraat 210, 1012 VT Amsterdam

Hans Bennis & Pim Wehrmann

Adverbial arguments 1. Introduction It is a well-known fact that NPs, which generally appear as arguments, may show up as adverbials. In this paper we will investigate the question whether it is also true that adverbials may appear as arguments. At first sight, this question seems rather paradoxical, since adverbials are generally defined by their non-argument status. In order to escape this paradox, we will restrict our attention to the question whether locative PPs may appear in Q-position. There are two obvious candidates for such a situation. The first shows up if a verb is subcategorized for a locative PP, as is the case with verbs like wonen (to live) or zetten (to put). The second one involves the occurrence of R-pronouns in PP [1]. Examples are given in (1). (1)a. Jan woont hier John lives here b. Jan praat daar over John talks there about In (1b) the R-pronoun is not a locative. It is an NP-type argument, which has no locative interpretation at all. The question with respect to the R-pronoun in (1a) is whether it is an argument. The answer to this question is entirely dependent on the definition of argument. Whatever the answer, nothing much is at stake. Let us assume for the moment that the locative in (1a) is not an argument. In order to find a true adverbial argument, we have to look for a locative phrase which occupies a position which is unquestionably an argument position. In this article we will argue that such a situation does not arise directly, but only through intervention of a resumptive pronoun. The article is structured in the following way. First we will discuss several simple cases of locatives that are interpreted as arguments. Then we will discuss the Dutch tough-construction. We have to discuss this construction rather extensively, since there is no satisfactory analysis of the corresponding Dutch construction. This analysis provides us with a final argument in favour of our analysis of locative arguments. 2. HET as a resumptive, locative pronoun In a sentence such as (2a) the neuter pronoun het may be referring to a neuter concept or a proposition, as is indicated in (2b). In another interpretation of (2a) het appears as a kind of dummy pronoun. This interpretation is given in (2c). (2)a. Het is hier leuk It is here nice b. Wat betreft fietsen/dat feest, het is hier leuk maar daar niet Concerning cycling/that party, it is here nice but there not c. Wat betreft Amerika, het is daar leuk Concerning America, it is there nice

2 In (2b) het refers to the phrase in front of the sentence, while in (2c) the locative R-pronoun daar refers to America and het does not seem to have any interpretation at all. This latter construction has the following properties: - the occurrence of het is obligatory, i.e. neither substitution by other pronouns such as dat or er nor deletion of het is allowed (cf.3a,b) - the occurrence of a locative PP is obligatory to arrive at the intended interpretation (cf.3c). [2]. (3)a.*Dat is hier leuk That is here nice b.*Hier is leuk Here is nice c.*Het is leuk It is nice We find this construction in a variety of sentence types. Het cam be the object of a transitive verb (4a), the subject of an ergative verb (4b), a passive construction (4c), a Small Clause (SC) (4d) and a raising verb (4e). (4)a. Ik haat het hier I hate it here b. Het bevalt me hier It pleases me here c. Meestal wordt het hier leuk gevonden Mostly is it here nice found d. Ik vind het hier leuk I consider it here nice e. Het lijkt me hier leuk (te zijn) It seems me here nice (to be) In all these cases the locative can be interpreted as an argument of a verb (4a,b) or an adjective (4c-e). For instance, (4a) is synonymous with a sentence such as I hate this place. Given the fact that het receives no interpretation at all, it seems as if it is a dummy pronoun. However, we will argue that het is syntactically an argument, i.e. an NP with a thematic role and Case, and that it is a resumptive pronoun, i.e. a pronoun c-commanded by and coindexed with a constituent in a non-argument position. Let us first discuss the structure of a sentence such as (5a). (5)a. (dat) dit boek leuk is that this boek funny is b. (dat) dit boekj^ [S(, t^ leuk] is We will adopt a Small Clause analysis for copula constructions (Stowell 1981, Hoekstra 1984), as given in (5b). In Bennis (1985,1986) this analysis is extended to constructions of the type in (6). (6)

Het^ is [ s c t^ leuk] [s,.dat Jan komt] It is nice that John comes

In (6) the S1 introduced by dat is not an argument structurally but only through coindexation with the structural subject of the SC. This subject (het) is moved in order to receive Case. One of the arguments in favour of this analysis concerns an asymmetry in wh-movement from S', as is illustrated in (7). (7)a. Wat is leuk dat Jan t zei What is nice that John said

3 b.*Wat is het leuk dat Jan t zei What is it nice that John said If het is present, the S1 is an adjunct and no wh-movement can take place. If het is not present, the S' is syntactically an argument and wh-movement is allowed. Assuming this analysis to be correct, we may try to extend it to the construction under consideration here. This implies that we assign the following structure to one of the interpretations of (2a), i.e. the interpretation given in (2c). (8)

Het^ is hier± [ t ± leuk]

This analysis implies that hier is an adjunct which is logically, but not syntactically, the subject of the SC, and that het is a resumptive pronoun, as in (6). In order to provide a full account of this construction we have to make two additional assumptions. First, we have to claim that locative PPs may not appear in A-position. This rather uncontroversial assumption accounts for the fact that het cannot be left out (cf.(3b)). This assumption provides an explanation for the difference between (9) and (10) as well. (9) a. Ik haat *(het) hier I hate it here b. Mij bevalt *(het) hier Me pleases it here (10)a. Ik haat (het) dat Jan komt I hate it that John comes b. Mij bevalt (het) dat Jan komt Me pleases it that John comes The ungrammaticality of (9) without het follows from the fact that these verbs require an object while no potential syntactic object is present. Given the fact that S', contrary to PP, may appear as argument, the occurrence of het in (10) is optional. The second assumption we have to make is that only het may function as a resumptive pronoun with a locative interpretation. This is a lexical idiosyncracy. In this respect the relation PP-res.pronoun differs from the relation S'-res.pronoun, as is shown in (11) and (12). (11)a. Ik haat dat.^ dat Jan zingt] I hate that that John sings b. Ik houd er^ van [ [glide"! / [aback]

+voc +long -low aback

M)

[+voc]

In the sentences in (2) glide-insertion may apply. (2)a b c d

Hij haalt de vlo (w) uit het eten he removes the flea from the food Riny (j) aaide Bert over zijn hoofd Riny stroked Bert on the head Die (j) ezel stoot zich niet meer that donkey will not bump itself anymore Het menu (w) Antoinette is aan te bevelen the Antoinette menu can be recommended

2. Method 2.1. Sentences Six groups of sentences were composed in which the syntactic and/or prosodic structures were different in terms of the positions where the application of glide insertion could be expected on the basis of the phonological environment. In (3) each group is represented by one sentence. The possible prosodic structures in the relevant position of each sentence, are indicated as well. PhP means phonological phrase and IP means intonational phrase (see Nespor and Vogel (1982)). (3)a

[PhP die ezel] stoot zich geen twee keer aan dezelfde steen that donkey does not bump into the same stone twice bi hij verkocht [PhP een serie artikelen] aan de krant ii hij verkocht [IP[PhP een serie][PhP artikelen]] aan de krant he sold a series of articles to the news paper

15 c

[IP[PhP ik ski][PhP elk jaar][PhP beter]] I ski better every year di [IP[PhP Rinie][PhP aaide][PhP Bert][PhP over zijn zere hoofd]] ii [IP[PhPRinie]][IP[PhP aaide][PhP Bert][PhP over zijnzere hoofd]] Rinie stroked Berts sore head e [IP Hendrik haalde [PhP de vlo][PhP uit het eten]] Henry removed the flea from the food fi [IP hij ging naar de stad toe om boodschappen te doen] ii [IP hij ging naar de stad toe][IP om boodschappen te doen] he went to the city to shop Groups a, b, c and e contained three sentences, group d contained four sentences and group f two sentences. There were 18 sentences altogether. In seven of them a w might be inserted and in eleven a j. 2.2. Speakers Two male speakers (RK and WO) read the sentences aloud, at a normal and a fast speech rate. Before reading, the speakers heard examples of the speech rates required. The 18 sentences were embedded in a list that also contained 20 fillers. The sentences were recorded with high quality equipment in a soundtreatened booth.

2.3. Measurements of the sentences The duration of the sentences was measured by means of a computer-based segmentation program with auditory feedback. The derived measure for speech rate (syllables per second) was calculated for every sentence. 2.4. Listening procedure Eight subjects listened to the sentences, which were randomized for speaker and speech rate. The listeners had to decide whether they heard an inserted glide or not. They were allowed to listen to the sentences as often as was thought necessary to arrive at a decision. 3. Results 3.1. Speech rates Mean normal speech rate of the sentences spoken by RK amounted to 6.8 syll/sec with a standard deviation of 0.6 syll/sec. Mean fast speech rate amounted to 7.5 syll/sec, the standard deviation being 0.7 syll/sec. Mean speech rate for speaker WO amounted to 6.2 syll/sec (sd 0.5 syll/sec) at normal rate and to 7.5 syll/sec (sd 0.8 syll/sec) in fast speech. For each sentence, speech rate in syll/sec is shown in Table I (see next page). It will be noted that for each separate sentence, fast speech rate was indeed faster than normal rate, although the amount of difference between the two realizations of a sentence may differ. The difference between the normal and fast realization of a sentence ranged from 0.2 to 1.2 syll/sec for speaker RK and from 0.4 to 1.9 syll/sec for speaker W0.

16 Table I: The sentences with speech rates and glide insertion specified per speaker indicated: no = normal speech rate in syll/sec, fa = fast speech rate in syll/sec, g = glide is heard, n = no glide is heard, d = it is doubtful whether a glide is heard. Group i

(a) die ezel stoot zlch geen twee keer aan dezelfde steen that donkey does not bump into the same stone twice (b) wat moet jlj met die aardbeien? what are you doing with those strawberries (c) bij het laatste spelletje had Piet drle azen in zijn handen in the final game, Pete was holding three aces in his hand ii (a) Hans verkocht een serie artikelen aan de krant Hans sold a series of articles to the news paper (b) het menu Antoinette was niet duur maar wel smakelijk the menu Antoinette was not expensive, but delicious (c) deze serie auto's is uitverkocht this series of cars is sold out iii(a) ik ski elk jaar beter I ski better every year (b) ik zie appels aan de bomen I see apples hanging from the trees (c) ik doe alle kwartjes in de automaat I am putting all quarters in the machine iv (a) de televisie is vanavond waardeloos the television program is bad tonight (b) de radio is al drie weken stuk the radio has been broke for three weeks (c) Rinie aaide Bert over z'n zere hoofd Rlnie stroked Berts sore head (d) opoe eet haar bord niet leeg grandma has not finished her dinner v (a) wij maakten een foto op zee we made a picture at see (b) Hendrik haalde de vlo uit het eten Henry removed the flea from the food (c) stiekem aaide ze m'n knie in de bus she furtively stroked my knee on the bus vi (a) hij ging naar opoe om haar te feliciteren he went to grandma to congratulate her (b) hij ging naar de stad toe om boodschappen te doen he went to the city to shop

RK no 6.4 n

fa 7.1 n

WO no fa 5.7 n 7.3 n

7.3 n

7.6 n

6.9 n 8.2 n

6.9 n

7.9 g

6.2 n

8.1 d

6.8 g

7.3 g 6.2 n

7.7 d

6.7 g 7. 8 g

6. 7 d

5. 9 g

6.8 g

6.9 g 5.5 n 6.7 d

5.8 g

6.4 g

5.0 n

5.6 n

7.5 n

8.3 n

6.8 n

8.3 n

7.6 g

8.4 g

6.7 n

8.1 n

7.4 n

8.1 n

6.7 n

7.8 n

6.9 d

7.8 d

6.0 n

8.0 d

6.3 g

6.8 g

5.8 d

7.3 d

5.5 n

6.0 g

6.2 d

6.6 d

6.2 g

7.0 n

6.1 g

7.2 d

6.8 g

7.5 d

6.5 g

7.7 g

6.6 g

7.6 g

6.0 g

6.8 g

7.5 g

8.3 g

6.8 d

8.6 g

6.6 g

7.8 g

6.3 n

8.2 g

17 3.2. The effects In Table I all sentences for the two speakers are given, together with the rate at which they were spoken and with information about whether glide-insertion was heard or not. It was arbitrarily decided that glide-insertion took place (indicated by g) if 6 or more of the 8 listeners perceived a glide, that glide insertion did not take place (indicated by n) if fewer than 3 listeners perceived a glide and that glide-insertion was doubtful (indicated by d) if 3, 4 or 5 listeners heard a glide. A shift from n in normal rate to d or g in fast rate and a shift from d in normal rate to g in fast rate was interpreted as an indication that glide-insertion applied more easily in fast speech than in normal speech. It could be that the difference between application and non-application of glide-insertion in these sentences can be traced back to a difference in domain assignment with respect to Intonational Phrases (IP's), as Nespor (1986) suggests. However, we carefully listened to the sentences, made FO-plots of the doubtful cases and concluded that all sentences with the exception of one were uttered as one IP as defined by Selkirk (1984a; 287-290). Three of the four sentences of group (vi), in which a continuation rise is possible on the syllable (o)poe and toe, indicating a boundary between two IP's, were uttered as one IP. However, only sentence (vib), spoken by WO at a normal rate, turned out to be divided into two IP's. Here, glide-insertion is not applied, which seems to indicate that IP is the domain of glide-insertion. However, other factors must be held responsible for the difference in application of glideinsertion as well. Let us therefore take a closer look at the data in Table I. If we compare the results of group (i), (iii), (v) and (vi), it is quite clear that in those cases in which glide-insertion is applied the second word is mostly a 'stressless' function word, whereas in the cases where glide-insertion is not applied the second word is mostly a content word with a relatively high degree of stress realized on the first syllable. Therefore, it seems at first sight that glide-insertion requires no stress or a low degree of stress on the first syllable of the second word. However, there seem to be some exceptions to this hypothesis. Most of them can be accounted for adequately, if a distinction is made between phonological stress based on weak vs. strong patterns in metrical, phonological structures, and phonetic prominence which is expressed in prosodic factors such as duration, loudness, pitch accent and preciseness of articulation. The sentences in (ii) seem to be counterexamples to the picture given above: a glide is perceived between two content words. In these cases the first syllable of the second word has primary or secondary word stress in phonological structure. However, the realization of phonetic prominence is not isomorphic with the underlying phonological sw-patterns in these examples^. Particularly in fast speech, the syllables ar(tikelen), An(toinette) and au(to's) are not realized as phonetically prominent. Thus, it is not the absence of phonological stress on the first syllable of the second word, but the absence of phonetic prominence which is one of the triggers of glide-insertion. The differences between the two speakers as found at normal speech rates for these sentences are probably also caused by this rate-dependent realization of phonetic prominence. As indicated, WO spoke relatively more slowly than RK in normal speech so that the realization of phonetic prominence by WO is relatively stronger than that by RK, which may result in a more restricted application of glide-insertion in the sentences of this group as spoken by WO. Another apparent counterexample is found in (ic). This sentence differs from the other two sentences in group (1) in that it has a numeral before the noun. In contrast with the other function words in these sentences, the numeral is

18 realized with phonetic prominence, which is even stronger than the prominence on the first syllable of the second word. This was found to occur in particular at the fast speech rate. This again explains the application of glide-insertion in (ic) in fast speech. A comparable explanation can be given for the fact that in (iiia) and (iiic) speaker WO does not insert a glide. He realizes the pronunciation of the words elk and alle with more phonetic prominence than does speaker RK. Thus, in the sentences read by WO, the application of glideinsertion is blocked because of the phonetic prominence of elk and al(le). Another problematic case is encountered in (iva) and (va), where the function words is and op do not trigger glide-insertion in fast speech while they may do so in normal speech. However, in the fast sentences vowel deletion took place, which blocks glide-insertion. This also accounts for the results of sentence (ivb) in which one would expect a clearer application of glide-insertion. However, it does not explain the results of sentence (vb RK),because vowel deletion did not take place here. Thus, almost all sentences receive an explanation with respect to glideinsertion. Only sentences (ivc) and (ivd), (via WO) and (vb RK) remain problematic. With respect to (ivc) and (ivd), a low degree of phonetic prominence could be realized on the finite verbs, although we do not really have any indication for that to occur. With respect to (via WO), there is a difference in application between normal and fast speech. It could be possible that there is an IP-boundary between opoe and toe in normal speech.' However, we did not find a phonetic indication for such a boundary. For the results of (vb RK) we do not have a convincing explanation. 4. Conclusions The results of our experiment lead us to the preliminary conclusion that glideinsertion seems to be an IP-domain process and that the first syllable of the second word must be realized with no or low phonetic prominence in order for glide-insertion to take place. The process of glide-insertion does not occur in fast speech only and it seems that it is not influenced by speech rate itself, but by other factors which are themselves speech rate dependent. These factors are the phonological domain IP and the realization of phonetic prominence on the first syllable of the second word. Of course, our results require a followup in which more variation in speech rate is included, longer and more complicated sentences are used which invite speakers to utter more IP's, and in which the variables stress, phonetic prominence and speech rate are controlled for more systematically. Although our results are tentative, they show that it is necessary to extend Nespor's (1986) proposal regarding this class of 'rate-dependent' processes: these processes may also be influenced by the realization of phonetic prominence which is in its turn influenced by speech rate. This observation provides a good starting point for further research. Both phonological domains and phonetic prominence are prosodic features. The following question now immediately presents itself: is it (only) prosodic features which are influenced by speech rate, and can the rate dependency of phonological processes be traced back to the influence of speech rate on the environmental prosodic conditions of the processes under consideration? Another area on which our results may have a bearing, is where phonological processes which are 'influenced' by speech rate apply in grammar. We have shown that the phonetic realization of prominence influences the application of glide-insertion. Given the usual model of phonology in which the relatively phonetic processes are ordered late (cf. Kaisse (1985), Berendsen (1985)), the influence of phonetically realized prominence on glide-insertion indicates that

19 this process must be ordered late in the phonology. If this holds for other processes of this type as well, Kaisse (1985) seems to be on the right track with her postulation of a late, fast speech module, and Nespor (1986) who argues against such a module, is not. Furthermore, Nespor's argument against a separate fast speech module is built on the ordering of the processes of Vowel Deletion and Stress Retraction in Italian. According to Nespor, Stress Retraction is a rule not influenced by speech rate and it has to be ordered after the fast speech rule of Vowel Deletion. Stress Retraction is a rule changing the prosodic structure in such a way that it will be optimal again, i.e. that it results in a structure without stress clash. In this respect, it is comparable to Resyllabification. However, Resyllabification is usually assumed to be an automatic process applying whenever necessary. It could well be that Stress Retraction is of the same nature, only applying when necessary. If that is the case, Nespor's argument is no longer an obstacle to the postulation of a fast speech module^.

Notes 0. We would like to thank the students who participated in our 1985 advanced course in Phonology and Phonetics. We are grateful to Marcel van den Broecke for his suggestions and his help with our English and to Marina Nespor and an anonymous reviewer for their critical remarks. 1. We tentatively assume here that the realization of phonetic prominence depends on several different factors such as phonological sw-patterns, focus information, the position of the word under consideration in prosodic structure above the word-level and speech rate. If this assumption turns out to be true, it is not surprising that phonological stress and phonetic prominence are not isomorphic. 2. Marina Nespor reported to us that she and Irene Vogel argue in forthcoming work that Stress Retraction and Resyllabification in Italian are not automatic processes. However, the position we take here with respect to resyllabif ication is the one usually found in the l i t e r a t u r e on syllabification (cf. Harris (1983) and Clements and Keyser (1983)), and our position with respect to Stress Retraction seems to be comparable to Selkirk's (1984a) position.

References BERENDSEN, E. 1985 Tracing case in phonology, Natural language and linguistic theory 3, p. 95-107. 1986

The phonology of cliticization, Forls Publications, Dordrecht.

CLEMENTS, G. and S. KEYSER 1983

CV-Phonology, MIT-Press, Cambridge Mass.

HARRIS, J. 1983

Syllable Structure and

Stress in Spanish, MIT-Press, Cambridge Mass.

KAISSE, E. 1985 Connected speech; the interaction of syntax and phonology, Academic Press Orlando. NESPOR, M. 1986 Vowel degemination and fast speech rules, Unpublished paper, University

20 of Amsterdam. NESPOR, M. AND I. VOGEL 1982 Prosodic domains of external sandhi rules, (eds.), The s t r u c t u r e of p h o n o l o g i c a l Publications, Dordrecht, p. 225-256.

H. v.d. Hulst and N. Smith r e p r e s e n t a t i o n s , Forls

SELKIRK, L. 1984a Phonology and syntax; the relation between sound and structure, MIT Press, Cambridge. 1984b (Dis)juncture in sentence phonology as a temporal phenomenon, Paper presented at the Fifth Phonologietagung, Eisenstadt.

Lisan van Beurden

Playing level with Dutch morphology 1. Introduction* Since the publication of The Sound Pattern of English (Chomsky and Halle, 1968) it may be considered well established that English has two classes of suffixes behaving differently with respect to stress rules (among other phonological rules). The behaviour of the class of so-called stress-sensitive suffixes can be demonstrated in data such as in (1), where attachment of -al and -ity to their hosts causes stress shift toward the suffixes: (1) Industry pAtency personal particular

industrial potintial personility particulArity

The suffixes -ly and -less in (2), however, do not influence the stress patterns of their bases: (2) particular instant shidow activity

particularly instantly shAdowless activityless

Siegel(1974) made the important observation that the stress behaviour of these two classes of suffixes coincides with the order in which they attach. The stress-sensitive suffixes ('Class I' in Siegels terminology) never appear rightperipherally with respect to the stress neutral ('Class II') suffixes: (3) person-alj-lyjj activ-ityj-lessjj-

*common-lyjj-alj *f ear-less j-j-ity^

A theory which succesfully explains the interaction between "stress sensitivity" and "affix order" is that of Lexical Morphology (Siegel(1974), Kiparsky(1982) et al.). The lexical model in (4) is one of the possible variants: (4) 1. Class I-derivation 2. Class II-derivation

f

>

stress rules

t,

>

level 2 phonology

The stress rules in (4) have access to words derived at level 1. Class Isuffixed words, therefore, undergo these stress rules. No stress rules apply at level 2. This explains the stress neutrality of Class II-suffixes, along with the affix order, because Class I-derivation occurs before Class II-derivation. According to Kiparsky(1982), a model such as (4) is the result of the mapping of language specific properties (number of levels; contents of levels) onto a universal abstract mould. Although the model in (4) correctly handles the bulk of the relevant observations in English, it cannot as such be transposed to Dutch, in spite of attempts to this effect by Zwarts(1975) and Booy(1977). They assume that the

22 stress behaviour of Dutch affixes resembles that of English in also requiring a subdivision into Class I and Class II suffixation, and intervening (stress)rules. However, the interaction between stress sensitivity and affix order appears to be of a different nature in Dutch. My aim in this paper is to show how stress and affix order phenomena are correlated in Dutch. Thus, I will propose an analysis along the lines of the theory of Lexical Morphology that will account for derivational morphology in Dutch, and I will show which properties of Dutch morphology explain the differences between this language and English.

2. Levels and lexical output categories Just as in English, Dutch suffixes can be divided in a stress-sensitive class (e.g. vljand 'enemy' vs. vijànd-ig 'hostile'; róme 'rome' vs. rom-èin 'roman') next to a stress-neutral class (e.g. vljand vs. vljand-schap 'hostility'). If the model in (4) were to account for the facts in Dutch, two kinds of problems would arise. First, as Schultink(1980) points out, the model in (4) incorrectly predicts the righthand column in (5) to be ungrammatical. (5) dlen-aar^j 'servant' hèer-schapjj 'lordship' mòlen-aarjj 'miller'

d i e n - a a r ' f e m a l e servant' heer-schapjj-ijj 'mastery' molen-aar^j-inj 'female miller'

The suffixes -aar and -schap in (5) are stress-neutral, hence Class II. The suffixes -es, -in, -ij bear stress and are therefore Class I. The model in (A) excludes Class I-suffixes which are outside Class II-suffixes, so the facts in (5) argue against (4). An observation of a different type is provided in Van Santen(1984). She observes that there are some restrictions in affixation among Class I-suffixes which (4) cannot account for. In some way, the suffixes in (6) all have their influence on stress placement which classifies them as I. The righthand column shows that suffixes deriving adjectives (hence: A-suffixes) do not attach to derived nouns: (6) bloem-istjj 'florist' voogd-ij^j 'guardianship' held-in^ 'heroine'

*bloem-istjj-ig, *voogd-ijjj-lijlc^ *held-in-erig^

Although it is impossible for the A-suffixes in (6) to attach to derived nouns, they can attach to non-derived nouns, as is shown in (7): (7) bloem^ 'flower' huis^ 'house' popjj 'doll'

bloem-ig. 'floury' huis-lijK^ 'homy' popp-erig^ 'doll-like'

Nominalizing suffixes, on the other hand, do not distinguish between derived and underived bases: (8) vijand N -elijk A -heid N 'hostility' gier^-ig^-aardjj 'miser'

kwaad A -heid N 'anger' lui^-aard^ sluggard'

The model in (4) appears to be inadequate to account for the aversion of A-suffixes in attaching to derived nominals. Rather than abandoning the idea of a level-ordered lexical model, we will have a closer look at the stress behaviour of Dutch suffixes. Actually, the stress-

23 sensitive class appears to be heterogeneous. In comparison with English, Dutch has a very large class of stress-bearing suffixes. See for some examples (9a) and their English equivalents in (9b). (9) a. indikatief opportunist fundamentèel amerikâan artistiek

b. indicative opportunist fundamèntal amèrican artistic

From this point of view, Dutch has three classes of suffixes: stress-neutral, stress-attracting and stress-bearing, whereas English has only the former two. Let us, for a moment, ignore the stress-bearing suffixes, and confine ourselves to the neutral and attracting ones. We are indebted to Schultink(1980a) for the observation that the stress behaviour of Dutch suffixes coincides with the lexical output categories, as displayed in (10) and (11). All of the stressneutral suffixes in (10) are nominal, contrary to the stress-attracting suffixes in (11), which are adjectival. In the latter case, stress is always on the rightmost full (non-schwa) vowel before the suffix. (10) -heid^ -domjj -schap^j -sel., N

èenzaajn 'lonely' hèrtog 'duke' bindgenoot 'ally' misbakken ' ? '

¿enzaamheid 'loneliness' h£rtogdom 'duchy' bindgenootschap 'alliance' mlsbaksel 'monster'

( I D -ig. -li^ -zaam^ -isch.

vljand 'enemy' hârtstocht 'passion' hèrberg 'inn' prôza 'prose'

vijAndig 'hostile' hartst6chtelijk 'passionate' herbirgzaam 'hospitable' prozciisch 'prosaic'

To describe these facts Schultink proposed a set of stress rules including a 'stress attraction' rule which makes special reference to the lexical category A. Within a level-ordered morphology, the different stress behaviour of the suffixes in (10) and (11) gives us reason to consider a model comparable to the one in (4), but in which the "numbers" identifying each level are replaced by "categories": (12)

A-derivation N-derivation

i-

main stress on the first (full) vowel before the suffix ->

N-level phonology

In addition to stress behaviour, the model in (4) explains the order in which affixes can occur. In exactly the same manner the model in (12) predicts that (stress-sensitive) A-suffixes attach before (stress- neutral) N-suffixes. Consider the grammatical examples in (13) and the ungrammatical ones in (14) (compare also (6) and (7)): (13) vijand-elijk^-heidjj 'hostility' opmerk-zaam.-heid N 'attentiveness' wantrouw-ena,-heid^ 'suspiciousness' gier-ig.-aara^ 'miser' vrijwill-ig^-er^ 'volunteer' heil-ig.-dom., 'sanctuary'

24 (14) gezond-heidjj 'health' probeer-sel^ 'attempt' vuil-nis^ 'rubbish' flauw-te„ 'faith' waar-heid 'truth' huur-ling^ 'mercenary'

*gezond-heidjj-lijk^ *probeer-sel^-lijk^ *vuil-niSjj-lijk^ *flauw-t(e) N -ig A *waar-heidjj-ig^ *huur-ling N -ig.

Thus (12) correctly predicts (a) the order in which the A-suffixes and N-suffixes attach, and (b) the fact that A-suffixes do not occur outside derived nouns, a distinction which would otherwise have to be stipulated. N-suffixes, on the other hand, can take both derived and underived adjectives as their input. The hypothesis that level ordering in Dutch is bound to the lexical output category can be maintained, if we take prefixes into consideration. As noted by Trommelen and Zonneveld(1986) prefixes create verbs in the unmarked case, irrespective of the lexical category of the host. Hence, the words ken„ 'know', groen^ 'green' and lichaam^ 'body' become verbs when preceded by a prefix: ver-ken„ ' to scout', ont-groeny ' to rag' , be-lichaaniy ' to embody'. However, Dutch also has words derived by both a V-prefix and a A- or N-suffix. The fact that these doubly-derived words are either A's or N's leads to the conclusion that V-prefixes are attached before A-or N-formation. Otherwise, it remains unexplained why, for example, ver-draagy-zaam^ 'tolerant' and be-manriy-ing^ 'crew' are an adjective and a noun, respectively, in spite of the verbalizing quality of the prefixes. I therefore claim that Dutch has a level for verbalizing processes preceding the A-level in (12). This will also give us an explanation for the fact that, in the unmarked case, V-prefixes only attach to underived words; A-suffixes attach to underived words and derived verbs (ver-draag-zaam 'tolerant'; ver-antwoord-elijk 'responsible'); and N-suffixes attach to both derived words and underived words of all lexical categories. 2.1. Some recalcitrant data Before I turn to the analysis of stress-bearing suffixes, I will discuss several difficulties arising from the model proposed in (12). First, there are two A-suffixes which behave unexpectedly with regard to stress attraction, i.e. -achtig '-like' and -loos '-less', in that they do not influence the stress pattern of the host: (15) schôolmeisje 'schoolgirl' schiolmeisjes-achtig 'like a schoolgirl' vâderland 'native country' vâderlands-loos 'without a native country' If we consider other properties of these two suffixes, however, their stress behaviour becomes less odd. Actually, words derived by -achtig and -loos share characteristics with compounds rather than affixed structures. As Booy(1977) shows, for instance, -achtig and -loos block word-internal syllabification rules, thereby feeding Final Devoicing (devoicing of voiced obstruents syllable finally) in the left-hand member of the complex forms in (16a). Compounds show the same lack of syllabification, as is shown in (16b). "Normal" suffixes such as -ing and -ig, however, allow the syllabification rules to apply and block therefore Final Devoicing (16c).

25 (16) a. roo[t]-achtig ri[p]-loos b. gou[t]-ader kra[p]-ijzer c. *hou[t]-ing *waar[t]-ig

*roo-[d]achtig 'reddish' *ri-[b]loos 'without bones' *gou-[d]ader 'vein of gold' *kra-[b]ijzer 'scraper' hou-[d]ing 'attitude' waar-[d]ig 'worthy'

Like compounds, but unlike suffixes, -achtig and -loos allow a preceding diminutive or a transitional -.s-. Compare, for example, the derived structures in (17a) with the compounds in (17b). (17) a. vogel-tjes-achtig 'like little birds' vriend-jes-loos 'without friends' dame-s-achtig 'ladylike' vaderland-s-loos 'without a native country' b. vogel-tjes-dans 'bird-like dance' vriend-jes-politiek 'favouritism' dame-s-tas 'lady's bag' vaderland-s-liefde 'patriotism' All these facts, including the stress behaviour in (15), can be easily handled if we assume -achtig and -loos not to be suffixes, but rather righthand members of compounds which are attached at a level later than the N-level. Second, Dutch contains a not infrequent counterexample to the expected affix order A-N: -schap^-elijk^ (litterally '-ship-ly' (see (18a)), although not all -schap formations can be adjectivized by -elijk as is shown in (18b). (18) a. vriendschap 'friendship' vriendschappelijk 'friendly' kameraadschap 'comradeship' kameraadschappelijk 'comradely' maatschap 'partnership' maatschappelijk 'social' wetenschap 'science' wetenschappelijk 'scientific' b. leraarschap 'mastership' pausschap 'papacy' moederschap 'maternity' gezelschap 'company'

*leraarschappelijk *pausschapppelijk *moederschappelijk *gezelschappelijk

A revealing difference between these two types of data is that the nouns in (17a) all have de as their definite article, whereas the nouns in (17b) have het. In Dutch, suffixes are generally associated with one definite article out of two possibilities (a characteristic which Trommelen and Zonneveld(1986) ascribe to righthand head properties of Dutch suffixation). A possible solution (which allows us to maintain the Right Hand Head Rule for Dutch) is to assume that there are in fact two suffixes -schap: one which takes de as a definite article and one which takes het. The latter will be attached at the regular N-level and therefore cannot be followed by an A-suffix; the former is exceptional in that it has to be attached at a level before the A—level, so attaching of a A-suffix is a possibility. This will give us a account for the difference between the words in (17a) and those in (17b).

26 3. The Romance part of derivation in Dutch So far, I have only been concerned with the classes of stress-neutral and stress-attracting suffixes. With respect to these two classes, the model in (12) correctly predicts that (a) stress attraction depends on the lexical category A, and (b) N-suffixes follow A-suffixes. Clearly, (12) in no way accounts for the behaviour of the stress- bearing suffixes. These suffixes bear main stress regardless of the lexical category they create: sport-lef, 'sporting', muzik-£al^ 'musical', muzik-Ant^ 'musician' public-ist^ 'publicist . Moreover, they frequently allow A-suffixes outside N-suffixes: indikatie„-ief^ 'indicative', funda-ment.j-eel^ 'fundamental'. In this s e c t i o n , I will demonstrate that it is possible to incorporate the stress-bearing suffixes in (12) in a way that explains their behaviour, but also maintains the generalizations made up to this point. In fact, the two classes discussed so far largely contain the so-called Germanic affixes, whereas the class of stress-bearing suffixes contains the Romance ones. My claim is that this division is not an ad hoc one. In Dutch, stress behaviour is the principal test for distinguishing Germanic suffixes from Romance ones. To my knowledge, this observation has not been made before, so I would like to explore this issue below. The models in (4) and (12) differ in one respect not yet mentioned. In (4), the stress rules associated with Class I-derivation are applied to Class I-derived words and to underived words. Therefore, Class I-derived words and underived words behave identically stress-wise. In (12), the stress rule restricted to the A-level places main stress on the first full vowel immediately before the attached A-suffix. This stress rule cannot be identified with the stress rule for main stress in underived words. The latter is generally claimed to be sensitive to syllable weight, i.e. underived words receive final stress provided that they end in a "super-heavy" (VVC or VCC) syllable (cf. Van der Hulst(1984) and Kager et al(1987)). The stress rule at the A-level in (12), on the other hand, ignores syllable weight: both light syllables like -ig or -lijk and heavy ones like -baar, -zaam or -isch are treated equally: they are never stress-bearing themselves, main stress is always placed on the first available vowel before the suffix. Therefore, the stress behaviour of the 'Germanic' part of derivation in Dutch can be described by a single stress rule, restricted to the A-level. The stress neutrality of N-suffixes (and V-prefixes) is explained by the absence of a stress rule at the levels where these are attached. The question which now arises is how to incorporate the stress-bearing 'Romance' suffixes in the model in (12). Neither the A-level stress rule nor the absence of stress rules at the other levels can account for main stress on the suffixes themselves. A stress rule which can, however, is the stress rule for underived words mentioned above, for all stress-bearing suffixes consist of heavy syllables. Romance derived words, therefore, share a generalization with underived words. In Lexical Morphology, the way to express a generalization between two processes is to place them at the same level, so that they are submitted to the same group of phonological rules (Kiparsky(1982)). For this reason, I assume the Romance formations to be derived at the level of simplex words (hence: the lexical level), which naturally precedes the levels in (12). This lexical level contains a stress rule for underived words (which is necessary anyway) at the same time providing for the stress pattern of Romance

27 derivations . See (19) for some examples, where (19b) shows that those Romance suffixes that do not consist of VVC or VCC, behave like equivalent underived words as well: (19) a. derived words muzik-int 'musician' stupid-itiit 'stupidity' b. organis-Atie 'organization' special-isme 'specialism' kritik-us 'critic'

underived words diamànt 'diamond' fazànt 'pheasant' tapijt 'carpet' habijt 'monk's habit' kanàrie 'canary' salimi 'salami' nuànce 'nuance' legènde 'legend' làthyrus 'sweet pea'

Once Romance derivations are executed at the lexical level, the above-mentioned discrepancies between Romance and Germanic morphology become clear: stress behaviour of Romance suffixes is not sensitive to lexical category; the Romance affix order is not A-N per se. The categorial levels of model (12) are in fact only concerned with the Germanic part of derivation. As a bonus, the proposed organization of Dutch morphology also accounts for the previously noted fact that Germanic affixes always occur right-peripherally to Romance affixes (see e.g. Schultink(1980b)). The theory outlined so far can be corroborated by two more properties the Romance derivations share with underived words. First of all, Zonneveld(1985) notes that, although a Romance nominalizing suffix can take het as well as de as its definite article (where Germanic nominal suffixes hardly ever fluctuate between jte and het), Romance derivations referring to human beings always take the definite article de. This also appears to be a strong generalization among underived words referring to human beings. We consequently find de dief 'thief' and de heraut 'herald', which are monomorphemic, next to de industri-eel 'industrialist' and de kom-iek 'clown', which are Romance derivations (compare, for instance, het materi-eel 'material' and het mechan-iek 'mechanism', which show that the article is not the only possible option for these suffixes). On the other hand, Dutch has (Germanic) het heer-schap 'lordship' and het schippertje 'little skipper'. This shared feature of Romance derivations and underived words is again easily expressed by assuming both types to be formed at the same level. At this level the generalization holds that words referring to human beings take d£ as their definite article. At the levels later in the model the choice of the specific article is governed by the (right-hand) head of the derived word. Finally, compounding also suggests a parallelism between underived words and Romance derivations. Nominal compounds in Dutch can have the structure NN, VN or AN, although the latter case is marked. However, once the structure is AN, the A has to be underived. Hence, ziek^bed^ 'sick-bed' and klein^kind^ 'grandchild' are both wellformed compounds, but ziek-ig.bed^ (litterally 'sickish-bed)' and klein-er.kind^ (litterally 'smaller-child'7 are completely ungrammatical. It is remarkable though, that there are a few compounds which do take a Romance derived adjective as their left-hand member: speci-aal^zaak^ (litterally 'special shop'), norm-aal•school^ 'normal school'. Again, Romance derivations behave like underived woras. The model in (20) is a revised and expanded version of (12), incorporating Romance affixation.

28 (20) underived words Romance derivations

» m a i n stress

V-derivation A-derivation

» V-level

/

rule.

_generalization: human beings

de

phonology

main stress at the first (full) vowel before the suffix

N-derivation This leaves us with the three stress bearing suffixes mentioned in (5): - ¿ s , - I n and - l j . These are not made up of heavy syllables, so it is not at all clear how the stress rule at the lexical level can provide for the correct stress pattern (underived words ending in - i n or - e s do not bear stress on the final syllable per se: ¿ndes 'the andes', plnguin 'penguin'; underived words ending in - i j always have final stress). In fact, it is impossible for - ¿ s , - I n and - i j to be attached at the lexical level, because they can appear outside stress-neutral Germanic suffixes (see (5)). These three suffixes, therefore, have to be grouped among the stress neutral Germanic suffixes and are therefore expected to be nominalizing, which they are. All we have to do is to assume that - ¿ s , - i n and - l j are marked, in that they are inherently stressed. Their apparently exeptional surface properties follow from this single assumption.

4. Conclusion Although Lexical Morphology provides a successful explanatory model for English derivational phenomena, previous attempts to analyze Dutch morphology in an identical fashion gave rise to several fundamental difficulties. The model was both too restrictive and not restrictive enough. However, once we take into consideration that Dutch deviates from English in that the former possesses three - instead of two - suffix classes, Lexical Morphology appears to be at least as fruitful for Dutch derivational structures. By means of stress behaviour, it is possible to distinguish Romance suffixes from Germanic ones. Romance suffixes behave like underived words in several respects, which can be expressed by attaching them at the first level in the lexical model in (20). Consequently, it is also explained that in the unmarked case Germanic suffixes follow Romance suffixes and not the other way around. Among the Germanic suffixes it is possible to distinguish stress-attracting from stress-neutral ones. This distinction is associated with the lexical output categories for both kinds of suffixes: stress-attracting A-suffixes precede stress-neutral N-suffixes. A-suffixes, therefore, cannot attach to derived nouns, whereas otherwise they do not make a distinction between derived and underived bases. To my knowledge, there is no other variant of morphology which provides an equally satisfactory account for this peculiar property of A-suffixes. Note that the fact that the A-suffixes, rather than the N-suffixes, are the stressattracting ones, is arbitrary as well as the order in w h i c h the V - , A - and N levels occur. Remnants like vluch-t,,-ia, 'volatile' and jach-t..-ig, 'to be in a hurry' may be an indication that in Middle Dutch the order of the A- and N levels was the reverse.

29 In sum, a lexical model as in (20), which is language-specific in number and contents of the various levels, but represents the abstract universal mould made available by linguistic theory, appears to be the appropriate way to account for both stress behaviour and affix order in Dutch.

Note * I would like to thank Det Paulissen and Mieke Trommelen for their comments and stimulating discussions. Special thanks go to W i m Zonneveld. Without his help this article would not exist.

References B00IJ, G.E. 1977 Dutch Morphology. A Study of Word Formation in Generative Grammar, Lisse, Peter de Ridder Press. CHOMSKY, N. AND M. HALLE 1968 The Sound Pattern of English, New York, Harper and Row. HÜLST, H.G. van der 1984 Syllabe structure and stress in Dutch, Dordrecht, Foris. KAGER, R., E. VISCH AND W. ZONNEVELD 1987 Nederlandse woordklemtoon; hoofdklemtoon, reductie en voeten, to appear in Glofr.

bijklemtoon,

KIPARSKY, P. 1982 From Cyclic to Lexical Phonology, in V a n der Hülst and Smith (eds.), The Structure of Phonological Representations I, Dordrecht, Foris. SANTEN, A. van 1984 De morfologie van het Nederlands, Dordrecht, Foris. SCHULTINK, H. 1980a Boundaries, Wordclasses and the Accentuation of Derived W o r d s in Dutch, in W. Zonneveld, F. van Coetsem and O.W. Robinson (eds.), Studies in Dutch Phonology, The Hague, Nijhoff. SCHULTINK, H. 1980b O n stacking up affixes, mainly in Dutch words, in D.J. van Alkemade et al, Linguistic Studies offered to Berthe Siertsema, Amsterdam, Rodopi, p. 229-242. SIEGEL, D. 1974 T o p i c s in E n g l i s h Morphology, MIT diss. TROMMELEN, M. AND W. ZONNEVELD 1986 Dutch Morphology: Evidence for the Right Hand Head Rule, in Linguistic Inquiry 17, p. 147-169.

30 ZONNEVELD, W. 1985 De morfologie van de mens: het hoofd, in Spektator p. 380-385.

14,

ZWARTS, F. 1975 - A a r , - a r i j , - s e i en - t e , reprint in Booij (ed.) Morfologie van het Nederlands, Amsterdam, Huisaandedriegrachten, 1979.

Saapke Bouma

The discourse function of quantifier float: tous in French 0. Introduction* In the linguistic literature of the last fifteen years a great deal of attention has been paid to the phenomenon of Quantifier Float (QF). The phenomenon owes its name to a transformation, assumed by many linguists to represent the relation between expressions like the following: (1) (2)

Tous les étudiants sont bien habillés 'all the students are well dressed' Les étudiants sont tous bien habillés 'the students are all well dressed'

Studies on Q F have focused on describing this relation in terms of transformational or interpretive rules, and on determining the restrictions on Q F (which types of NP can be the source of a Floating Q (FQ)). In these studies, the equivalence of expressions like (1) and (2), and comparable examples in other languages, has been a fundamental, though often tacit, assumption (e.g. Kayne (1975)). In this article, this assumption is questioned on pragmatic grounds. The study of a corpus of contemporary French utterances containing the universal quantifier (UQ) tous/toutes shows that (1) and (2) function in different contexts and with different implications. Besides, as far as frequency is concerned, (2) turns out to represent a 'minor' type of quantification. A description in a transformational format, then, gives us at best a very limited insight into the relation between (1) and (2). Instead, I will interpret the facts of quantification, and especially QF, in French in terms of speaker's strategies and general pragmatic operations^. The conceptual framework of this study is Functional Grammar (FG) (see Dik 1978). Most of the examples I use are from the corpus, details of w h i c h are given in section 3.

1. Syntactic and semantic properties of Q F in

French

In linguistic expressions , the universal quantifier quantify any NP as a part of that NP, irrespective position: (3) (4) (5)

(UQ) tous 'all' may of its function or

Tous les Orléanais vont à l'église 'all the inhabitants of Orléans go to church' Dans le Midi, ils prononcent toutes les syllabes 'in the Midi they pronounce all the syllables' Ces produits viennent de tous les coins de France 'these products come from all the corners of France'

(Subj) (Obj) (Source)

In these cases I use the term internal Q (IntQ). Tous may also quantify NPs at a distance, as a FQ. This is only allowed if the antecedent is a NP

32 bearing Subject function, cf.: (6)

Les Orléanais vont tous à l'église 'the inhabitants of Orléans all go to church'

FQ tous cannot be related Locative,etc.) , cf.:

to

NPs

with

other

functions

(cf. (3))

(Obj,

Recipient,

(7) *J'ai tous lu ces livres 'I read those books all' (8) *Ces produits viennent des coins de France tous 'these products come from the corners of France all' When tous is related to a clitic ptonoun, on the other hand, QF is obligatory, cf.: (9)

O n est tous heureux (Subj clitic) *0n tous est heureux 'we are all happy' (10) Je les ai tous lus, les classiques (Obj clitic) *Je les tous ai lus, les classiques 'I read them all, the classics' (11) On leur donne tous la possibilité de poursuivre leurs études (Dative/Recipient clitic) *0n leur tous donne la possibilité de poursuivre leurs études 'we give them all the opportunity to continue their study' Thus there is a clear distinction between nominal and pronominal antecedents with respect to QF. Within the category of pronouns, moreover, the difference between clitic and strong pronouns is of interest. The plural forms of the two paradigms are given in (12): (12) pi p2 p3

Clitic pronouns on (Subj) nous (Obj/Recipient) vous (Subj/Obj/Recipient) ils (m), elles (f) (Subj) les (Obj), leur (Recipient)

Strong pronouns nous vous eux (m), elles (f)

Benveniste (1965) has shown that the two paradigms are in complementary distribution, the main differences being the following: 1. Clitics can only appear as arguments of a predicate, in the functions of Subject, Object and Recipient; their presence depends on the presence of a verb. Strong pronouns, on the other hand, are excluded from the argument functions of verbs, but they appear in verbless utterances, in left- and right-dislocated position (Theme / Coda), and in semantic functions requiring prepositions (in general: lower functions on the Semantic Function Hierarchy, which runs from Agent via Patient, Recipient, Beneficiary, Comitative to Time and Location (Dik 1978: 69)). 2. Clitics cannot be emphasized, modified, contrasted or coordinated. Strong pronouns can undergo these operations; as far as modification is concerned, their modifiers are mainly adverbs with a specifying or emphasizing function, which in most cases are placed directly after the pronoun, e.g. moi seul 'only me', même eux 'even them', eux aussi 'they/them too', vous tous 'all of you', nous autres 'WE' (contrastive).

33 The difference between clitic and strong pronouns is pertinent w i t h respect to Q F as well: IntQ is excluded with clitic pronouns, whereas Q F is excluded with strong pronouns. Consider: (13)a lis sont tous partis ce matin b *Ils tous sont partis ce matin c *Eux tous sont partis ce matin d ?*Eux sont tous partis ce matin^ 'They all left this morning' This means that clitic pronouns can only be quantified by a FQ. IntQ and FQ agree in number and gender (masc.: tous, fem.: toutes) w i t h their antecedents; FQ may agree in case. The antecedent has to satisfy two more conditions: it must be plural and definite (personal pronouns are inherently definite; other terms must contain a definite article or some other overt definiteness marker). There is one more restriction, which has to do w i t h the plurality of the antecedent. If the set of referents designated by the antecedent has two or three members, it cannot be quantified by tous alone, but has to be quantified by tous les deux/trois respectively. Now, these heavy quantifiers appear almost exclusively as FQ's: (14) ??Tous les deux enfants vont a l'ecole ' both children go to school' (15) Les enfants vont tous les deux a l'ecole 'the children both go to school'

The FQ, especially fem. toutes, may in certain cases show semantic ambiguity (in fact, the same ambiguity shows up in English): (16) (17)

Toutes les tables sont rouges 'all the tables are red' Les tables sont toutes rouges 'the tables are all/completely red'

In (16), only a collective interpretation of toutes is possible, whereas in (17) toutes may have a collective as well as a holistic reading. The exact conditions that determine the interpretation of toutes are discussed in Bouma (to appear); I will not go into them here. What is of interest, however, is that there are clear indications that the ambiguity of toutes may lead speakers to avoid Q F in cases like (17). In summary, then, when speakers use UQ tous, Q F can be expected in three cases: i) when the antecedent is a plural clitic pronoun (obligatory), cf. (9) - (11) ii) w h e n the antecedent is a definite, plural Subject (optional), cf. (6) iii) when the quantifier is heavy (strongly preferred), cf. (14), (15). Q F can be expected to be avoided, on the other hand, w h e n the interpretation of toutes leads to semantic ambiguity, cf. (16), (17). Before discussing the actual distribution of these options in the corpus, I first turn to a closer inspection of the definiteness restriction.

34 2. Tous and definiteness There appears to be a specific and the UQ. Consider: (18) (19)

relation

between

the definiteness

operator

Les enfants vont à l'école 'the children go to school' Tous les enfants vont à l'école 'all the children go to school'

The truth conditions of (18) en (19) are identical; the same assertion is made about the same set of referents. The use of the definite article in (18) involves implicit reference to the totality of the referent set, while the additional effect of the use of tous in (19) is to make this totality explicit (see Fennema & Bouma (1982), see also Brown (1985)). This means that the function of the use of tous is pragmatic in nature: it is an extra indication by the speaker to the addressee of the scope of the referential domain to which the assertion applies. Both (18) and (19) may have a generic interpretation, generic terms being obligatorily definite in French^ . Now compare (19) w i t h (20): (20)

Les enfants vont tous à l'école 'the children all go to school'

Without context, (19) and (20) may have both a generic and a specific reading, but the generic reading strongly imposes itself in (19), whereas in (20) the specific reading is by far the more probable. This m e a n s that (20) will be used in situations in which a subset of children has been marked off from the whole universe of existing/possible children. In other words: (20) can be expected when the term les enfants has topical properties. The definiteness of the term les enfants plays a role in this topicality. It is also a condition for Q F to occur, but not a sufficient condition: terms w i t h IntQ also have to be definite. It is clear, on the other hand, that statements involving pronouns can never be interpreted generically; pronouns necessarily have a specific reading because their reference is context bound. W i t h all this in mind, QF can be conceived of as a strategy to avoid generic reading. In section 4. I will go more deeply into the relationship between topicality of constituents, the occurrence of QF, and the use of UQ's in general. We will have a look at the corpus data first.

3. Q F in a corpus of contemporary

French.

The corpus I have used consists of 1030 utterances containing instances of tous/toutes, collected from transcribed interviews (Corpus d'Orleans; 875 instances) and written texts (scientific and literary works, newspapers, magazines; 155 instances). A small number of utterances were uninterpretable because of performance errors. The remaining instances of tous were in the great majority IntQ's, occurring in terms with a whole range of semantic functions. There were also 145 instances of FQ's; about 30% of these FQ's

35 were heavy quantifiers of the type tous les deux. Table I shows the distribution of FQ and IntQ over noun-headed terms and pronouns with Subj, Obj and Recipient/Beneficiary functions. I have also included in the table 62 Theme's/Coda's , which are in fact extraclausal constituents; I will come back to them later. Table I. Subj

IntQ FQ

Obj

N

Pro

N

Pro

113 14

* 115

143 *

* 12

Recipient/ Beneficiary N Pro 16 *

4(strong) 4(clitic)

Theme/ Coda N Pro 60 *

2 *

There is an almost complementary distribution between the antecedents of FQ tous and IntQ tous: FQ occurs almost exclusively w i t h clitic pronouns (where Q F is the only allowed form of quantification), IntQ occurs almost exclusively w i t h noun-headed terms. There are 4 cases of a pronoun w i t h IntQ, these are strong pronouns (see note 2). The most interesting case is that of the Subjects. Of all constituent types, only Subjects occur more often with a FQ than with an IntQ, but there are only 14 instances of FQ w i t h a nounheaded Subject, the rest being Subj clitics. In percentages: - in 90% of all the Q F cases, clitic pronouns are involved - of these clitic pronouns, 86% are Subj clitics - of all the Subjects in the corpus, 47% are clitic pronouns - of all the Subjects occurring with a FQ, 88% are clitic pronouns The tendency of Subjects to occur as pronouns more often than other constituents has to do with the fact that speakers maintain their discourse topics by means of anaphoric pronouns. In section 2. I suggested a relationship between Q F and a specific, more precisely, a topical interpretation of the antecedent. It seems clear, then, that in 90% of the cases the FQ is related to a topical antecedent, as far as the clitic pronouns are concerned. This interpretation is less obvious in the case of the 14 nominal Subjects occurring with FQ. The question then arises in w h i c h respect(s) these 14 Subjects are distinct from the 113 Subjects with IntQ, and in what res p e c t s ) they are similar to the 115 pronoun Subjects with a FQ. I will address this question by first discussing the notion of topicality in some detail. Subsequently, I examine the extent to which this notion can contribute to an explanation of the distribution of tous.

4. Topicality, givenness and the antecedents of tous In Functional Grammar the pragmatic function of Topic is associated w i t h constituents " 'about' which a certain predication predicates something in the given setting" (Dik 1978:130). And thus Topic is defined on the level of the predication. Nevertheless, it seems difficult, if not impossible, to detach this notion from discourse. What a linguistic expression (and its underlying predication) 'is about' often depends on preceding or following linguistic expressions in discourse. In coherent discourse, then, some topiG is introduced by one of the discourse participants and continued, possibly

36 over several utterances. In the first utterance, the discourse topic is new; in the subsequent utterances it will be given, that is, retrievable for the participants. The topic may also be abandoned, to be reintroduced later in the discourse. This is the position taken in Dik (to appear), where a distinction is drawn between three types of Discourse Topic: New Topic, Given Topic and Resumed ^ r e i n t r o d u c e d ) Topic. It is well known that the prime means for a speaker to treat a certain discourse referent as given is by representing it by a definite term or by an anaphoric pronoun. New Topics, on the other hand, will typically (though not necessarily) be represented by indefinite terms. Thus, there is a close relationship between definiteness and givenness. It seems useful, however, to refine the concept of 'givenness' somewhat more. In this connection, I want to mention two proposals , made within the framework of FG, to clarify the notions of givenness and topicality. One can be found in Bolkestein (1985). She argues that coreference is not the only way by which entities can be considered as given, but that semantic relations between discourse constituents also play a n important role in deciding on givenness. These semantic relations can be of the following type: (near)synonymy, contrast, antonymy, hyponymy, superordinateness, etc. A similar proposal comes from Hannay (1985), who introduces the notion of sub-Topic. A sub-Topic can be inferred from a Discourse Topic if there is a relation of 'inference' between them, and this relation is specified as being (i) part of, (ii) member of, (iii) subset of, (iv) instance of, (v) copy of, (vi) aspect of, (vii) opposite of, (viii) projection of, (ix) associated with. Such sub-Topics may be definite when first mentioned. Thus, a discourse entity may be given in several ways: by the text (coreference), by the situation, and by inference. These distinctions in givenness, together with the distinctions in topicality, appear to be highly relevant with respect to the type of givenness of the antecedents of tous. The 14 cases of nominal Subjects with FQ constitute predominantly clear cases of givenness: three of them are textually given, in ten of the other cases inferrability is an important factor (e.g. six utterances about the children of the speakers or about near members of their families; some cases of part-whole relations). Only one case cannot be classified this way. The Subjects w i t h IntQ, on the other hand, tend to occur as Resumed Topics and, very frequently, in generic statements and presentative constructions (il y a tous ces ... 'there's all these ...'), contexts which do not particularly favour the occurrence of given constituents. In this connection, I also want to mention the 62 Themes and Codas w i t h IntQ, presented in Table I: to a considerable extent, these serve the function of enumerating discourse referents in advance (Theme) or afterwards (Coda). The prototypical QF-case is found under conditions of strict coreference, a s in (21): (21)

...merci pour vos propositions, elles sont toutes r...l interessantes 'thanks for your suggestions, they are all T . . . ] interesting'

where elles is coreferential with vos propositions in the preceding utterance. In some of the less prototypical QF-cases, the antecedents are inferentially given by part/whole and membership relations, as in (22):

37 (22)

tant pis on partira pas les enfants partiront tous si vous voulez 'too bad, we won't go, the children will all go if you please

Pragmatically, this means that the domain of the quantified referent(s) coincides exactly w i t h the domain of a previously introduced referent, or constitutes a subdomain of it. The prototypical use of a generic statement, on the other hand, involves the opposite operation: extending a restricted referential domain to a whole referential universe. Consider: (23)

A B A B

Y a cette abbaye je ne sais plus le nom ... A h ben oui. Toutes les abbayes tripotent plus ou moins euh ... 'there is that abbey I don't remember its name .. 'oh yes. All abbeys swindle more or less ..'

Notice that the transition from cette abbaye to toutes les abbayes would seem to match exactly Hannay's relation of projection (see above); consequently, toutes les abbayes should be considered as given. Yet (24) sounds less natural instead of (23)B: (24) ??Ah ben oui. Les abbayes tripotent toutes plus ou moins It seems that, w i t h respect to quantification, the semantic and pragmatic operations that keep a given domain equal or decrease it favour the occurrence of FQ's, whereas the operations that extend a given domain (superordinateness, projection) favour the occurrence of IntQ. From this point of view, Q F and internal quantification appear to be mirror images. There is another remarkable difference between the antecedents of FQ's and those of IntQ's: the FQ-antecedents (clitics and Subject terms together) designate human beings in over 95% of all cases. Subject terms with IntQ, on the other hand, refer to human beings in only 40% of the cases, and to inanimate, entities in almost 60%. Thus, in general the FQ-antecedents are higher on the animacy.hierarchy, which contributes to their topicality. This may be related to the fact that Q F seems to have a more individuating effect, whereas internal quantification has a collectivizing effect. In summary, speakers tend to use FQ's with antecedents designating human beings and having the pragmatic status of givenness by text, situation or certain kinds of inference. The prototypical antecedent is a clitic Subject pronoun. Noun-headed Subject terms are infrequent; the reason may be that anaphoric pronouns constitute a better means of ascertaining the T o p i c status of discourse entities. IntQ's are found with antecedents designating any entity, but preferably inanimates, in generic, presentative, or enumerating Subjects and Resumed Topics.

5. Tous in discourse For reasons of space, it is impossible to -discuss examples of all the uses of tous with their relevant cbntext. Instead, I will present one stretch of

38 discourse (taken from the Orléans cases of the use of tous. (25) A B A B A

B A

B A

A B A B A

B A

B A

corpus) w h i c h shows

four

characteristic

(A=interviewee, B=interviewer) mais je crois qu'avec le temps euh/-/ le le genre de vie des gens se rapproche mm euh les les gens vivent davantage tous un peu plus de la même façon m m ah oui? enfin tout au moins dans le milieu entre le l'aristocratie la bourgeoisie les les /-/ les ingénieurs les cadres tous ces gens-là évidemment les ouvriers n'ont pas encore le mhm le même genre de vie mais enfin dans le milieu qu'on appelle indépendant si vous voulez qui est un monde assez /:/ étendu je trouve que les différences sociales se font moins sentir oui /-/ et comment est-ce que cela se fait? /-/ ben du fait je crois que les jeunes vont en classe tous ensemble beaucoup plus maintenant dans les mêmes écoles euh il y a beaucoup de jeunes dans les lycées enfin toutes les classes sociales se trouvent mélangées

1

2

3 4

'but I believe that, as time goes by, the kind of life of people will get more and more similar mm people all live more and more a little more the same way oh yes? at least in the environment between the aristocracy the bourgeoisie the /-/ engineers the officers all those people of course the workers still don't have mhm the same kind of life but in the environment you could call independent if you like which is rather extended I think that social differences are not felt so much yes /-/ and how does that come about? /-/ OK by the fact I believe that the young people go all together to the same schools much more often nowadays there are many young people in the high schools all the social classes are being mixed'

The four numbered instances of UQ in (25) can be described as follows: in 1, the antecedent is textually given; 2 is a case of a summarizing Coda; in 3, les jeunes is a subTopic, inferred from the other classes of people m e n tioned before, but the quantifier is heavy as well; in 4, les classes sociales is a superordinate of the specific classes mentioned above, it refers to a non-human entity, and ,in addition, Q F would have led to semantic ambiguity ('all'/'completely'), all factors favouring an IntQ.

6. Final remarks In the foregoing I have disregarded the effects caused by different speaker styles and differences between formal and informal discourse. These may play

39 a role in the distribution as I have found it. As far as I can judge now, Q F is more frequent in less formal styles (spoken language, magazines, literary works). It is .however, impossible to predict the choices speakers will make, so the tendencies described in this article are no more than just tendencies. Moreover, they can only have preliminary value and should be confirmed (or falsified) by a more extensive investigation of more, especially written, material.

Notes *

I wish to thank Daan de Jong for introducing me to the Orleans corpus, and the editors of this volume for their helpful comments.

1. Since the names 'Q-Float' and 'Floating Q' are widely accepted, they will be used throughout the article, but only to indicate a specific behaviour of the UQ and without commitment to any transformational implications. 2. Strong pronouns, appearing in non-argument functions, can be quantified by a n IntQ: (a)

Jean a fait cela pour nous tous 'John has done that for all of us'

(Beneficiary)

The text examples (13c) and (13d) are unacceptable because strong pronouns are excluded as Subjects, at least in standard French. Some dialects seem to be less restrictive: (13d) is acceptable in Walloon. My informants (four speakers of standard French) rejected (13c) a s well as (13d). I found no examples of strong pronouns in Subject function in the corpus. 3. Generic interpretation of a statement is restricted to the unmarked tense: the Present.

References BENVENISTE, E. 1965

'L'antonyme et le pronom en français moderne.' In: Benveniste, E., Problèmes de linguistique générale II. Paris: Gallimard.

BOLKESTEIN, A.M. 1985 'Cohesiveness and syntactic variation: quantitative vs qualitative grammar.' In: Bolkestein, A.M., C. de Groot & J.L. Mackenzie (eds.) (1985a). BOLKESTEIN, A.M., C. D E GROOT & J.L. MACKENZIE (eds.) 1985a Syntax and pragmatics in Functional Grammar. Dordrecht: Foris. 1985b Predicates and terms in Functional Grammar. Dordrecht: Foris.

40 BOUMA, S. to appear 'Over funktie en betekenis van een dobberende kwantor: het Franse tous.' In: N. Corver (ed.)> Grammatikaliteiten. TILL-collection, Tilburg University. BROWN, R. 1985 'Term operators'. In: Bolkestein, A.M., C. de Groot & J.L. Mackenzie (eds.) (1985b). DIK, S.C. 1978 Functional Grammar. Amsterdam: North-Holland, to appear The theory of Functional Grammar. FENNEMA, A., & S.BOUMA 1982 Universele kwantifikatie in het Frans: een funktionele analyse. Unpublished paper, University of Amsterdam. HANNAY, M. 1985 'Inferrability, discourse-boundness, and sub-topics'. In: Bolkestein, A.M., C. de Groot & J.L. Mackenzie (eds.) KAYNE, R. 1975 French syntax. Cambridge, Massachusetts: MIT Press.

(1985a).

Josée Coenen & Roeland van Hout

Word order phenomena in second language acquisition of Dutch 1. Introduction The early stages of adult second language acquisition are marked by the use of simple word order restrictions and intonation as the main tools for expressing meaning (Klein 1986:82). Giv6n (1979, 1984) summarizes the organizational principles in the utterances of early or beginning learners both pidgin speakers and second language learners - under the heading of a 'pre-syntactic mode'. The development from an early language variety towards a native-like syntax goes from a simple, economic syntax, mainly governed by universal discourse principles and pragmatic principles, towards a more language-specific, near-native syntax. In this article the structure of the Dutch utterances of two Moroccan and two Turkish adult immigrants are analysed. In the analysis of these spontaneous speech utterances, the following questions have to be considered: (1) What is the basic pattern in an early learner variety? (2) Which principles determine the structure of the basic pattern? (3) Is there a development over time and if so, does this influence or change the basic pattern? (4) What differences exist between the learners and how can these differences be explained? The formulation of the first three questions reflects the recent orientation in second language research, i.e. the emphasis on universal aspects. This orientation towards universality (cf. Klein & Perdue forthcoming), however, may result in a neglect of possible source language influences. Recent L2 research in the Netherlands, for instance, seems to ignore the role of the source language. Both Appel (1984:133 & ff.), analysing Dutch data from Turkish and Moroccan children, and Lalleman (1986:121-122), analysing Dutch data from Turkish children, find deviations in word order which could be caused by source language patterns. Nevertheless, both authors are rather reticent about source language influence as an explanatory factor. Our data, however, show a pattern which warrants the conclusion that source language patterns play a role in the acquisition of a new syntax by early L2 learners. In section 2 some general principles and constraints are discussed which are mentioned in the literature in relation to early L2 syntax. In section 3 a short description is given of the kind of data used. Section 4 contains a discussion of some methodological aspects of this type of research. In section 5 and 6 the typical word order patterns of each of the four informants are presented, followed by a short description of developmental trends in section 7. In section 8 the most important results are discussed. 2. Principles and constraints Klein (1986:82-83) sets up a list of eight principles which early learners tend to apply in constructing their utterances and which are related to the way information is put in a linear order and to the way human information processing is structured (cf. Levelt 1982). These principles are aimed at the general organisation of discourse and only the principle of orientation makes an explicit prediction about word order within an utterance: "Place orientational elements (place, time, modality, etc.) at the beginning of an utterance" (Klein 1986:82).

42 On the basis of English L2 data of a Punjabi and an Italian speaker Huebner (1986) reformulates these principles as constraints which operate on various linguistic levels (semantic, pragmatic, discourse). These constraints make strong predictions about word order patterns and they refer especially to the position of a NP in relation to the verb, being preverbal (=N^) or postverbal (=N2). The following constraints are of special interest for the analysis of our data (cf. Huebner 1986): Semantic constraints (= SC) SCI: Agents or experiencers assume the N^ position SC2: In two-argument verbal constructions, objects (sources, goals, datives, locatives) assume the N2 position SC3: In one-argument verbal constructions, objects assume N]^ position There are two restrictions: 1. agents never occur with experiencers 2. experiencers never occur with objects Prapnatic constraints (= PC) PCI: Recoverable topics are realized as zero PC2: Introduce new arrivals on the scene in N2 position These constraints may be in conflict, for instance in an L2 utterance like "then comes police" where SCI is overruled by PC2. As these constraints appear to be also applicable to other early L2 data (cf. Klein & Perdue forthcoming) they seem to have fairly universal value. The data of our Turkish informants, however, show a great many utterances in which the verb takes the utterance-final position. In order to account for such patterns it is necessary to reinterpret and adapt the semantic constraints in the sense that N^ and N2 note the order of NPs within an utterance. PC2 then should be reformulated in the constraint that new arrivals on the scene should be introduced in postverbal position. Another problem in applying these constraints is the possibility that the word order predicted by a constraint may coincide with a word order current in the second language. PC2 prescribes a word order which is also a word order frequently occurring in standard Dutch main clauses and, consequently, the occurrence in our data of this word order could also reflect a rather advanced L2 syntax. In interpreting the utterances of our informants it has to be taken into account that the basic word order in Dutch is SOV (Koster 1975) with Verb-second in main declarative sentences. In main declarative sentences the subject (N^) may be preverbal or postverbal (in case another constituent takes the sentence-initial position). 3. Data The analysis is based on the spontaneous Dutch speech of two Turkish and two Moroccan informants.^ The data were collected at two different moments in time, with an interval of about nine months. All four informants are adult migrants who are learning Dutch spontaneously (i.e. they have not taken part in a professional language training course); at the time of the first recording they had been living in the Netherlands for about two years.^ The data to be analysed here come from a specific kind of activity, viz. the retelling in Dutch of a shortened version of the silent movie 'Modern Times' by and with Charlie Chaplin. The informant watched the film while a Turkish or Moroccan speaking researcher was present. Next, the informant was asked to retell the film to a Dutch speaking researcher who had not seen the movie. The time needed for the retelling varied from 15 to 25 minutes. The same procedure was repeated about nine months later. So, for every informant two texts were available.

43 4. Some methodological remarks The analysis of spontaneous data from early second language learners faces a number of serious problems. Utterances may be unfinished, interrupted or even inaudible and it is sometimes hard if not impossible to establish utterance boundaries. These kind of problems may reduce the sample of analysable utterances considerably as can be observed in table

Moroccan Turkish Table 1:

Fatima Mohamed Mahmut Ergtin

number of utterances Text 1 Text 2 102 52 107 106 153 168 173 125

analysable utterances Text 1 Text 2 93 50 104 104 142 157 133 99

Number of utterances and analysable utterances in the filmretelling data

In addition, there is the problem of finding a satisfactory analytic model. A paradox inherent in the analysis of utterances of second language learners is that one has to apply analytic categories of which it will always remain unclear to what extent they match or cover the linguistic notions or categories applied by the language learner. A neutral and at the same time modest approach seems to be most attractive: one could argue for the use of simple categories like N^ (or NP^), V and N2 (or NP2) . However, these categories do not say anything about possible underlying semantic relations and, in addition, they presuppose a distinction between nouns and verbs, whereas an early L2 learner may use other word categories to fill the function of, for instance, a verb (hii wee, 'he away'). We have therefore opted for more semantically oriented categories like predicate, arguments and operators. The use of these categories implies that a predicate may still consist of a verb, but also of an adverb or an adjective. In the case of a two-place predicate two arguments are to be present, respectively called AG^ and AG2. In a learner's actual utterance these arguments may be empty and they can be interpreted as empty slots. For most of the utterances of our informants a one- or two-place predicate structure conveys the intended meaning; in some cases a third argument has to be assumed. However, more complex predicate structures are irrelevant in the sense that we are mainly interested here in the factors that govern the ordering of the predicate in the utterance (the 'verb') with respect to the first argument and to the rest of the arguments (AG2, AG3). The description of the utterance structure also requires the introduction of some other generally occurring categories. First of all, there is the category of interjections, which are especially frequent at the beginning and at the end of an utterance. Secondly, a special group of words (especially adverbs) turns out to appear frequently at the beginning of the utterances of every informant. We call them 'discourse operators' and they mark the time and place orientation of the utterance in the plot-line of the discourse (cf. principle of orientation). The third special category is the so-called tail, which contains "information meant to clarify or modify (some constituent contained in) the predication" (Dik 1978:153). The following example (from Ergtin) illustrates the analytic model we have applied (the plus-sign indicates a pause): (1)

ja yes I INT

dan then I DISC. OP.l

daar 00k die 00k there too that (one) too I I I I DISC. (OP) AG! (OP) OP.2

komen he + ander vrouw come [..] + other woman I I I I PRED INT TAIL

44 The second interjection (he) has no English equivalent. The repeated adverbial operator ook 'too' is put in brackets in this example, because the emphasis in this article is upon the relative positions and functions of AG^, AG2 (and related parts of the utterance) and PRED. It is important to note that many utterances occur in our material in which no AG;l is present. It is the AG^ that is absent, first, because it is the agent argument which does not show up and, secondly, because the agent related to the predicate can easily be recovered from the preceding utterance(s) (= PCI). The category of AG2 may contain various types of arguments: sources, goals, datives. The main problem in analysing the utterances is that one has to interpret the words of the informants in terms of categories and that often doubt remains about the correctness of the interpretation and categorization assigned (the 'matching problem'; see Klein 1986:138 & ff.). A helpful strategy is to compare utterances and especially utterances collected at different moments in time ('matching') and to re-interpret the utterances again and again. The film retellings have the invaluable advantage that the 'facts' spoken about by the informants are known. Although this strategy is rather time-consuming, its fruitfulness can be illustrated by two examples. In several utterances in the first retelling, Mahmut uses the word betalen 'to pay' which is a Dutch verb. However, he uses this word together with the verb peven 'to give' in the first retelling; he uses the same verb in the second retelling, but now in combination with the standard Dutch noun peld 'money', as can be seen from the examples in table 2. The word betalen has disappeared in the second retelling in favour of the noun peld and it seems reasonable to interpret the word betalen as a noun. It follows that the utterance jij betalen geven should be categorized as AG^ - AG2 - Pred. Retelling 1 "ik niet betalen" "I not pay" "j ij betalen geven" zegt "you pay give" says

Retelling 2 - politie [Ind.Obj.] zegt: "jij geld geven'h police [Ind.Obj.] says: "you money give" - "jij geld geven" zegt ti. "you money give" says

Table 2: Mahmut, matching betalen 'to pay': Verb - Noun Another example of a word class category reinterpreted is the word moet. frequently used by Mahmut. In Dutch, this word is a finite form of the modal verb moeten 'must'. If this word in the utterances of Mahmut were interpreted as a auxiliary modal verb, it would be the only auxiliary verb he uses. Moreover, Mahmut rarely uses finite verbs (the two exceptions are zept 'says' and komt 'comes); the normal verb form is the infinitive or the participle. Careful comparison with other utterances (some examples are given in table 3) shows that the distribution pattern in the utterance of this 'verb' is the same as the distribution pattern of adverbials like niet 'not' and misschien 'maybe'. The most natural explanation is that the word moet is used by Mahmut as a modal adverb. - jij moet *hapis* gaan you must *hapis* (- Turk, jail) go - moet betalen must pay Table 3:

- ik niet *hapis* gaan I not *hapis* go - niet betalen not pay

Mahmut retelling 1, matching moet 'must': Verb - Adverb (operator)

45 5. Overview of the w o r d order patterns for the four informants I n table 4 a n overview is given of the number of occurrences of specific w o r d o r d e r p a t t e r n s . T h o s e p a t t e r n s w h i c h are particularly relevant to the relative positions of AG^, AG2 a n d PRED are p r e s e n t e d in the patterns 1 to 6; only those utterances are c o u n t e d w h i c h c o n t a i n a clear predicate. Q u o t e d speech introductions are k e p t separate (pattern 7). The utterances w h i c h render q u o t e d speech constitute the m a i n p a r t of the rest category, while t h e y t u r n o u t to f o r m a r a t h e r h e t e r o g e n e o u s s u b s e t containing more f r e q u e n t l y f o r i n s t a n c e f o r m u l a i c e x p r e s s i o n s a n d questions. The rest category also contains utterances w i t h o u t a predicate, unclear utterances and, i n a few cases, complex utterances (utterances containing subordinated clauses). Mahmut 1 2

Retelling (OP) (AG X ) A G 2 (--) PRED (OP) (AG;L) PRED A G 2 ( - - ) (--) PRED (OP) (AGi) A U X AG2 (OP) (AGx) PRED (--) (OP) PRED A G i (OP) A U X AGi (--) PRED Q u o t e d speech introduction (e.g. 'he says to h i m ...') 8. rest

1. 2. 3. 4. 5. 6. 7.

T o t a l analysable utterances Table 4:

Ergttn 1 2

Fatima 2 1

Mohamed 1 2

29 1 0 42 0 0 9

28 4 0 37 0 0 44

26 12 4 33 2 0 11

10 9 13 23 2 1 9

0 28 0 26 0 0 13

0 19 0 14 0 0 5

0 27 8 17 3 0 13

0 8 13 6 11 1 21

61

44

39

38

26

12

36

44

142 157 133

99

93

50 104 104

Overview number of occurrences of w o r d order patterns in the a n d second film retelling

first

6. W o r d order patterns for the four informants: first retelling

Mahmut

(Turkish)

M a h m u t ' s linguistic repertoire a t the first film retelling is v e r y limited. A s for the verb forms, h e m a i n l y uses the infinitive a n d sometimes a past particle (without auxiliary). There is h a r d l y a trace of inflection a n d no trace at all of a tense system. Function words rarely occur a n d the only pronouns h e uses are ik 'I', j ij 'you', hii 'he' (both deictic a n d anaphoric) a n d m i j n 'my'. The b a s i c p a t t e r n in Mahmut's utterances (apart from the p o s i t i o n of the operators) is clear: the predicate is utterance final (table 4, p a t t e r n 1 a n d 4), w h e r e a s AG2 takes the p o s i t i o n after AG^. There is only one e x c e p t i o n ( e v e n zll g e z i e n d i e b r o o d 'in a glimpse she s e e n the bread'). This c o n c l u s i o n presupposes the interpretation of m o e t as a m o d a l adverb. I n most cases the predicate is a n infinitive. G i v e n this pattern, the order a n d f u n c t i o n o f AG^ a n d AG2 are in line w i t h the semantic constraints as adapted to cover predicate final utterances. Examples are: (2)

(3)

As

is

twee kindje chocolade g e v e n two c h i l d [dim.] chocolate give 'he gives some chocolate to two children' d a n m e n s e n b r i e f lezen h e t h e n people letter r e a d [interj.] 'then the m a n reads the letter'

shown by

(2),

AG^

may

be

left

implicit,

[PCI, SC2]

[SCI, SC2]

as p r e d i c t e d b y

pragmatic

46 constraint 1 if the topic is recoverable. An interesting finding is that AG^ accommodates pronouns. AG2 also functions as a slot for NPs which indicate the Source or the Goal argument of a predicate. Coordinate sentences and sentences rendering quoted speech do not show a deviant pattern. In violation of PC2, no word order change occurs in the case of topic introduction. Ergttn (Turkish) Ergttn is a more advanced learner than Mahmut; several traces of the beginning of inflection are observable and he uses various pronouns in both the nominative and the oblique form. Nevertheless, he is obviously an early learner as well. The most frequent patterns in Ergttn's retelling are predicate final (pattern 1 vs. 2 and 3; patter 4 vs. 5 and 6). He also often uses the infinitive verb form (see (4)) and sometimes the past participle form. The AG2 may also contain Source or Goal arguments. AG2 accommodates pronouns but this seems to be restricted to the verb zeggen 'to say' introducing quoted speech (see (5)). Examples are: (4)

dan daar 00k een *puro* roken then there also one *puro* (- cigar) smoke 'then he is also smoking a cigar' (5) hij zegt tegen hem: [...] he says to him: [...]

[PCI, SC2] [SCI, SC2]

Example (5) also illustrates a word order pattern which regularly appears(see also pattern 2 and 3) and which suggests a development of patterns which are closer to Dutch: AG^ - Pred - AG2. In such cases the verb form mostly turns out to be finite; on the other hand this development is mainly restricted to two verb forms, viz. komt 'comes' (see (6)) and zegt 'says'. Pattern 3 in table 4 is formed by utterances in which Ergttn might be trying to express an aspectual aspect of the predicate by the auxiliary 'is'. An example is (7). (6)

die dinges komt hoofd [SCI, SC2] that thing comes head 'the beam falls on his head' (7) hij is 00k die restaurant gaan [SCI, SC2] he is too that restaurant go 'he goes into the restaurant too' Twice an utterance occurs in which the predicate precedes AG^; both cases are topic introducing utterances. On the other hand, there are more topic introducing utterances in which this change of word order does not occur. In general, Ergttn's basic word order strongly resembles Mahmut's basic structure and the same holds for the constraints. Whenever Ergttn deviates from this pattern, the deviation seem to reflect an early stage of mastering the word order structure of standard Dutch. Fatima (Moroccan) Fatima's linguistic repertoire is very limited: there is hardly any trace of inflection, and marking of tense is absent; she only knows a few function words and hardly any adverbs of time or place. She clearly structures her word order in a fixed form: AG^ - Pred - (AG2) (pattern 2 and 4). The predicate immediately follows the first argument and the function of AG^ correponds to the semantic constraints. The verb has a finite form, either the root form or the root form plus the suffix t. (8)

die man pakt die meisje that man takes that girl (9) politie komt police comes (10) pakt die charlo takes that charlo

[SCI, SC2] 'the man takes the girl away' [SCI] 'there comes a policeman' [PCI, SC2] 'he takes charlie away'

47 A zero AG^ position as in (10) occurs rather frequently; in these cases the topic is recoverable from the preceding context because it is the last AG^ mentioned explicitly. The basic pattern is also found in quoted speech and even in all topic introducing utterances (see (9)). Interesting exceptions are two questions in quoted speech, in which Dutch word order for questions is found: the finite verb form precedes AG^. Mohamed (Moroccan) Mohamed is the most advanced learner, although his length of stay in the Netherlands is the shortest (less than two years). He uses auxiliary verbs and past participles according to the syntactic rules of Dutch. He also uses a tense system (mainly irregular verbs for past tense (e.g. kwam 'came')). He has acquired a relatively extended pronominal system and he uses pronouns both in AG^ and AG2. His basic word order is the same as for Fatima: AG^ - Pred - (AG2) (pattern 2 and 4). However, in general Mohamed makes a correct use of the finite verb form according to standard Dutch and he also succeeds in using auxiliary verbs, while at the same time putting the lexical verb in sentence-final position (pattern 3), as required by the syntax of Dutch (see (11) and (12)). (11) (12)

toen then toen then

een meisje heeft brood gepakt [SCI, SC2] one girl has bread taken 'the the girl stole some bread' gaat snel lopen [PCI] goes fast walk 'then she hurried away'

Example (12) also shows that Mohamed, too, may leave the AG^ position empty, provided the topic is recoverable. A few topic introducing utterances show the order predicted by pragmatic constraint 2 (pattern 5): (13)

toen heeft een vrouw haar gezien then has a woman her seen

[PC2] 'then a woman saw her'

Both the word order pattern with only a finite verb form, and the pattern with a split predicate occur in various types of utterances: quoted speech, and also in subordinate clauses (where Dutch requires that the finite verb shows up in clause-final position). 7. Development The second retelling took place eight to nine months after the first retelling. Three informants show only minor changes in their basic pattern in the sense that they clearly correspond with the word order patterns in the first retelling. Mahmut is still using a AG^ - AG2 - Pred order, but the AGi - Pred - AG2 is gaining ground, however mainly in introducing quoted speech (pattern 7). This last word order pattern is also marked by an increase of finite verb forms. Especially the finite verb form for topic introduction komt 'comes' is frequent. The number of zero AG^s is smaller and pronouns are used in AG2 positions for various verbs. Ergttn uses the same two patterns as in the first retelling, but the pattern that Is more like that of the target language (AG^ - Pred - AG2) is increasing in frequency. In this pattern the predicate contains a variety of verbs with a finite form. The use of pronouns in AG2 position is more frequent. There are two utterances -both introducing a new topic (referent)in a Pred - AG]^ pattern (pattern 5). Fatima's second retelling does not show a change or development in the two basic patterns she used in the first retelling. The most conspicuous aspect

48 is her lexical progress. She now frequently uses the pronoun hij 'he' and, as a consequence, the number of zero AG^s has diminished. No pronouns occur in AG2 position as yet. Mohamed's second retelling shows some remarkable changes. He now uses the Pred - AGi pattern (pattern 5) in a fairly consistent way, especially in introducing new topics (referents) (see (14)), and also in the case of an inanimate topic (see (15)). He sometimes even uses this pattern in cases where the utterance starts with a (discourse) operator. This could mark the beginning of the development of a Verb-second pattern, the pattern that is standard in Dutch main declarative sentences. (14) (15)

toen then toen then

kwam politie bij came police komt een hout van boven comes one wood from above

'then a policeman came' 'then a piece of wood falls down'

His pronominal system is extended to non-emphatic forms (cf. Broeder, Extra & Van Hout 1986). The extension of the pronominal system goes together with a decrease in the number of zero AG^s. A pragmatic constraint (PCI) is replaced by the syntactic constraints of the target language. 8. Conclusion and discussion It is evident that the informants investigated share numerous characteristics as far as the word order in their utterances is concerned. The topic-focus or theme-rheme structure is a paramount factor (cf. Klein 1986: 82-83). The placement of orientational elements (discourse operators) at the beginning of an utterance is also a ubiquitous feature (principle of orientation). The semantic functions and the order of AG^ and AG2 within the utterance match the semantic constraints adapted for predicate final utterances. All informants have zero AG^s (PCI), but their frequency diminishes in the second retelling. This decrease in frequency should be considered in connection with the development of the use of pronouns and the pronominal system. In that sense, the number of zero AG^s is negatively correlated with the language proficiency in Dutch of our informants. Mohamed is by far the most proficient speaker, already displaying near-native syntactic patterns in the second retelling. Fatima, Mahmut and Ergttn (this order reflecting an increasing proficiency in Dutch) are obviously early learners, even in the second retelling. Their lexical development between the first and second retelling is far more conspicuous than their development in syntax. Pragmatic constraint 2 (introduce new arrivals on the scene in N2 position) turned out to be highly disputable. The applicability of this constraint would lead to word order patterns which match the syntax of Dutch. This complicating factor does not hold for English, the language studied in Huebner (1986). Nevertheless, if this constraint held as a universal feature, one would expect a large number of Pred-AG^-structures in the Dutch data. In contrast, the incidence of the specific word-order predicted by pragmatic constraint 2 is low, whereas the increase in number of these kind of utterances seems correlated to an increase in Dutch proficiency. The position of the verb as implied in Huebner's constraints is also worthy of attention with respect to the differences in the word order patterns between the two Moroccan and the two Turkish informants. Our Turkish informants very frequently place the predicate in utterance-final position and use the infinitive (and sometimes the past participle) in this position. This order is never found in the utterances of the Moroccan informants, who in addition use in this position a verb form which is finite (in the case of

49 Fatima it would be better to speak of the verb stem). These differences between Moroccan and Turkish learners of Dutch are supported by the findings of Janssen, Lalleman & Muysken (1981) and, although to a lesser degree, by the findings of Appel (1984) and Lalleman (1986). The explanatory factor could be source language influence. Turkish is a SOVlanguage, wheras Moroccan-Arabic is marked by a (S)VO-pattern.^ The difference in verb position in the two languages corresponds with the difference in verb position in the utterances of our Turkish and Moroccan informants. The reason why this difference is more pronounced in our data than in the data of Appel (1984) and Lalleman (1986) is, very probably, that their informants were children who were far more advanced learners of Dutch than our adult informants. The underlying patterns of the source languages, however, could not explain the difference in the use of finite verb forms and infinitives. One has to assume that, in addition to the source language structure, structural properties of Dutch, the target language, play a decisive role. The sentencefinal verb in Dutch takes the form of the infinitive or the form of the past participle. In their perception of Dutch, then, the Turkish learners of Dutch probably pay attention to the form of the verbs in sentence-final position and apply the forms they note there in their own production of Dutch utterances, whereas Moroccan learners probably pay attention to the verb form that occurs in the initial part of Dutch utterances, after which they apply these finite forms in their own production of Dutch utterances. These strategies will in particular hold for early learners, as is illustrated by the utterances of Mohamed, who is clearly more advanced. This would mean that the differences found between our Moroccan and Turkish informants are not only caused by source language differences, but by an interaction of the source language structure and the ambivalent structure of Dutch regarding the position of the verb. It is this ambivalence which may trigger the emergence of source language structures. Notes 1. These data are part of a larger data base, collected within the framework of a longitudinal study of second language acquisition by adult Moroccans and Turks at the University of Brabant. This study has several international counterparts and is financed by the European Science Foundation (ESF), Strasbourg. See Perdue (1984), for the design, the data techniques and the topics of investigation. 2. See Broeder et al. (1985) for further sociological and biographical characteristics . 3. For operationalising the concept of utterance, syntactic criteria (cf. Hunt (1966) are used. In cases of doubt, however, decisions are made on the basis of suprasegmental characteristics, such as intonation contour and pause length. 4. See Lewis (1967) and Lehman (1975) for a description of the Turkish language and Harrell (1962) for a description of Moroccan-Arabic. References APPEL, R. 1984 Immigrant children learning Dutch: sociollnguistic and psycholinguistic aspects of second-language acquisition, Dordrecht, Foris Publications

50 BROEDER, P., J. COENEN, G. EXTRA, R. VAN HOUT & R. ZERROUK 1985 Spatial reference in L2 Dutch of Turkish and Moroccan adult learners: the initial stages, in G. Extra & T.Vallen (eds.). Ethnic minorities and Dutch as a second language, Dordrecht, Foris Publications, p. 209-252 BROEDER, P., G. EXTRA & R. VAN HOUT 1986 Acquiring the linguistic devices for pronominal reference to persons: a cross-linguistic perspective on complex tasks with small words, in F. Beukema & A. Hulk (eds.), Linguistic in the Netherlands (1986), Dordrecht, Foris Publications, p. 27-40 DIK, S.C. 1978 Functional grammar, Amsterdam, North Holland GIVON, T. 1979 From discourse to syntax: grammar as a processing strategy, in T. Giv6n (ed.), Syntax and semantics, vol. 12: Discourse and syntax, New York, Academic Press, p. 81-109 1984 Universals of discourse, structure and second language acquisition, in W. Rutherford (ed.), Language universals and second language acquisition, Amsterdam, Benjamins, p. 109-136 HARRELL, R.S. 1962 A short reference grammar Georgetown University Press

of Moroccan Arabic, Washington

D.C.,

HUEBNER, Th. 1986 Thematic structure in L2 data of a Punjabi and Italian learner, Nijmegen, Max Planck Institute for Psycholinguistics (manuscript) HUNT, K.W. 1970 Recent measures in syntactic development, in M. Lester (ed.), Readings in applied transformational grammar, p. 187-201 JANSSEN, B., J. LALLEMAN & P. MUYSKEN 1981 The alternation hypothesis: acquisition of Dutch word-order by Turkish and Moroccan foreign workers, in Language Learning 31, p. 315-336 KLEIN, W. 1986 Second language acquisition, Cambridge, Cambridge University Press KLEIN, W. & C. PERDUE (forthcoming) The learners problem of arranging words, Nijmegen, Max Planck Institute of Psycholinguistics (manuscript) KOSTER, J. 1975 Dutch as a SOV language, in Linguistic Analysis, p. 111-136 LALLEMAN, J.A. 1986 Dutch language proficiency of Turkish children born in the Netherlands, Dordrecht, Foris Publications LEHMAN, W.P. 1975 Topic to subject in Indo-European, in. C.N. Li (ed.) (1976), Subject and topic, New York, Academic Press, p. 445-456 LEVELT, W.J.M. 1982 The speaker's linearization of discourse. XIII International Congress of Linguistics, Tokyo LEWIS, G.L. 1967 Turkish grammar, Oxford, Clarendon Press PERDUE, C. (ed.) 1984 Second language acquisition by adult immigrants: A field manual, Rowley Mass., Newbury House

Colin Ewen & Harry van der Hülst

Single-valued features and the distinction between [-F] and [0F] In Ewen & van der Hulst (1985), a number of aspects of underspecification theory are considered, in relation to a system of phonological representation based on single-valued features. We concluded there that the distinction drawn by Archangeli (1984) between complement and default rules was not relevant in a singlevalued approach, in that complement values do not exist for single-valued features: for a binary feature the complement value is simply the reverse, so that C+F], say, has the complement value [-F], while the same strategy applied to a single-valued feature [F] simply yields the absence of the feature in question, as in (1): (1)

Binary

Single-valued

Lexical value

[+F]

[F]

Complement value

[-F]

Thus, in an underspecification approach, the only rules required to handle cases of the sort dealt with by Archangeli are default rules, which simply supply the redundant features which are not present in lexical representations. The validity of single-valued approaches within an underspecification framework has been called into question in recent work by Steriade (e.g. 1986), in which various kinds of apparently non-local phonological processes are dealt with. One of Steriade's claims is that the distinctions which she is able to make rely crucially on the use of a binary feature system. If her arguments are correct, then, this would provide important evidence against a system based on single-valued features. Let us first consider the most important aspects of Steriade's theory. Like Archangeli, she draws a distinction between complement and default values, as in (2): (2) a. Complement value A non-underlying value of F within a segmental class where F is distinctive b. Default value A non-underlying value of F within a segmental class where F is predictable As well as complement value and default value, we will use the term lexical value for the value specified lexically for segments belonging to the class in which F is distinctive. The choice of which value for a particular feature is lexical and which the complement value depends in Archangeli's model on language particular factors such as alternations, economy, and, if these are not decisive, universal markedness. Steriade (1986) provides a simple example involving the binary feature [voice] in Japanese, which has the lexical inventory in (3):

52 (3) a. [+voice]

b

d

g

z

[-voice]

p

t

k

s

b. [+vo iceJ

m

n

r

j

w

vowels

For the obstruents in (3a), [voice] is distinctive; the sonorants in (3b) are predictably voiced. On the assumptions that [+voice] represents the lexical value for the category in which the feature is distinctive, and that voiceless obstruents are lexically unspecified for [voice], we can formulate two rules, as in (4): [-son] + [-voice]

(4) a. CR b. DR

[+son]

+

[+voice]

Steriade formulates a universal complement rule, given in (5), to provide the appropriate complement values: (5)

Universal complement rule (VCR) Given some feature F and a level of representation where only underlying values of F, [aF], are present: find all pairs of segments (si, S2) such that si differs from S2 only in that si is [aF] and S2 is [0F], Specify S2 as [-aF],

Given the UCR, there is no need to specify individual complement rules. Within this approach, then, a feature may have any of the specifications given in (6): (6) a. L-values b. C-values c. D-values We will now consider the analysis of processes involving action-at-a-distance, i.e. apparently non-local processes in which intervening segments are skipped. Assimilation rules of this sort (i.e. harmony rules) involve agreement between two segments for a particular feature F. In an underspecification model, segments intervening between the trigger and target segments may be ignored if they do not have a value for F at the point at which the rule applies; otherwise we would have crossing of association lines, as shown in (7): (7) a. +F X

X

X

b. +F

-F

X

X

r - k

~x

There are three points at which such rules might logically apply, given the various types of value in (6). A rule might apply before the assignment of Cvalues, in which case it can skip any segment which will later be assigned either C-values or D-values. Such a rule can be termed an L-rule. A rule might also apply after C-values have been assigned, but before D-values: in this case only segments which will later acquire a D-value can be skipped. This is a C-rule. Finally, if a rule applies after the assignment of D-values, it can only apply to strictly adjacent segments (unless we assume that certain segment-types never get specifications for particular features): this is a D-rule. The three types of rules are summarised in (8):

53 (8) a. L - r u l e s :

apply before C - and D-values have b e e n assigned

b. C-rules:

apply before D-values have been assigned

c. D-rules:

apply after D-values have b e e n assigned

The three types of rule allow us to establish a typology of potentially skippable and/or opaque intervening segments. We give this typology as (9), w h i c h takes into account the specific value of intervening segments for the feature in question in relation to the spreading or L - v a l u e . If the L - v a l u e is [aF] then the C-value is [-aF], and the D-value can be either [aF] or [-aF]: L-value:

[aF]

L-rule

C-rule

D-rule

OPAQUE

OPAQUE

[-aF] (C-value)

SKIPPABLE

[-aF] (D-value)

SKI PP ABLE

SKIPPABLE

OPAQUE

[aF] (D-value)

SKIPPABLE

SKIPPABLE

OPAQUE

It is of course relevant to consider whether all the cells in (9) are in fact attested. We will return to this shortly; for the moment let us assume that they are Indeed all empirically motivated. It will be clear that the distinctions made by Steriade rely crucially on a distinction between [ - F ] and [0F]. To illustrate this, let us consider one example in detail. This is a spreading phenomenon from the dialect of Scots Gaelic spoken in Barra, discussed by Clements (1986), whose analysis is based on the data presented in Borgstr^m (1937, 1940). In this dialect there is a process in w h i c h a vowel is epenthesised in a non-homorganic consonant cluster of which the first element is sonorant. The inserted vowel is identical to the preceding v o w el, except for the feature [back], which is binary in Clements' model. Barra Gaelic has an opposition between palatalised and non-palatalised or 'plain' consonants, characterised by Clements as involving a distinction between [ - b a c k ] and [+back]. The labials form an exception to this, however, and are lexically u n specified for the feature [back], as in (10):

(10)

Barra Gaelic sonorants a. [+back]

n

[-back]

N

b. [0back]

m

N 1

r

R

L

r' 1' L'

(There is a further classification for the non-labials in terms of 'lenited' (/n r r' 1'/) and 'non-lenited' (/N L R N' L'/) sonorants, which is not relevant to our concerns here.) In general, the vowel gets its value for [back] from the sonorant consonant. However, if the consonant is /m/, the backness value for the epenthesised vowel is acquired from the previous vowel, so that the labial, w h i c h w i l l later acquire the default value [+back ], is skipped. This process, then, gives the six possibilities in (11):

54 -B -B 1 r1 1 c V

-B +B 1 1 i V c

V

-B r 1 V m

+B -B | r1 1 c V

+B +B I ri . 1 V c V

f . +B r1 V m

c.

The first five of these possibilities are illustrated by the data in (12): (12) a. [ken'ep]

'hemp'

b. [Jaerok]

'to fade'

c. [t'imiqal]

'round about'

d. [bul'ik]

'bellows', gen. sg.

e. [balak]

'bellows'

The possibility in (llf) is not recorded in the data given by Borgstr(5m. ever, its absence does not affect the point being made here.

How-

Notice that in Steriade's model the distinctive non-palatalised consonants a c quire [-back] by a C-rule (on the assumption that [+back] is the L - v a l u e ) , while the labials, which are non-distinctive for palatality, acquire their value from a D-rule. Thus the spreading rule is a C-rule: if it were an L - r u l e , distinctive non-palatalised consonants would be skipped, and if it were a D-rule, labials would not be skipped. Both these possibilities would of course be theoretically available in Steriade's theory. A n analysis appear to raise [ - F ] and [0F] (11), given as

of this problem in terms of single-valued features does indeed problems, given that we are unable to make a distinction b e t w e e n in such a framework. Consider the single-valued equivalent of (13):

(13) a. i

d.

i V

C

e. V

f. V

C

V

V

m

V

Here we incorporate a single-valued feature of frontness (cf. Anderson & Ewen 1987 and Goldsmith 1985, among others): this gives a more natural interpretation of the secondary articulation involved, in that palatalised consonants now have a property which the plain consonants lack, and also allows us to avoid having to specify labials as being phonetically back, which at best seems counterintuitive. However, it will be clear that we are now unable to account for the case enclosed in the box in (13): as the C is no longer associated with a feature, there a p pears to be nothing preventing spreading into the second vowel; there is no formal difference between (13b) and (13c). Moreover, given a single-valued feature system, it is unclear h o w the absence of F could block anything. However, we w i l l now show that a single-valued feature system can, contrary to the claim made b y Steriade, provide an adequate analysis of the kind of phenomena we have b e e n looking at. We proceed as follows. First we consider a m o d e l close to Steriade's, w h i c h employs 'empty nodes' in contexts where Steriade makes a p -

55 peal to complement values. This model will prove to be untenable, however. We then propose a more drastic alternative which has various empirical consequences. In both proposals we assume that all spreading involves adjacency. However, this does not mean that spreading necessarily involves adjacent segments. Rather, in terms of an autosegmental model like that of Clements (1985), adjacency refers to the class nodes to which a feature can associate. A segment can be skipped in a spreading process if it lacks the appropriate class node (cf. van der Hulst & Smith, this volume: §3). Let us turn to the first proposal. Suppose we allow empty nodes to be generated in the course of the derivation. Underlyingly we find only those class nodes which bear features, which we will refer to as lexical nodes. In the same way as Steriade differentiates between complement and default values, so we can distinguish complement nodes and default nodes. Complement nodes can be detected by a principle analogous to that in (5), Steriade's Universal Complement Rule: (14)

Universal complement node rule Given some feature F: find all pairs of segments (sj, s 2 ) such that s^ differs from s2 only in that Sj is F-specified and s 2 not. Assign to s 2 the class node which bears F.

This principle gives us a set of (empty) complement node generation rules. Finally, we can define a set of default node generation rules, which creates all nodes not otherwise specified, whether they are filled or empty. Given our assumption about adjacency of nodes, spreading will be blocked not by the presence of a particular feature-value, but by the presence of a particular class node. In demonstrating this, we will make use of the model of segmental representation proposed by Archangeli & Pulleyblank (1986) (cf. also Clements 1985), which incorporates a distinction between a place node, associated with consonantal place features, and a secondary place node (S-place node), associated with vocalic features. Palatalised consonants, however, also involve an S-place specification, as the secondary articulation involved is characterised by the tongue-body feature [back] (cf. also Clements' approach). Spreading will be blocked if the spreading feature cannot associate with the S-place node under one of the conditions in (15): (15)

Opacity Condition1 A segment S is opaque with respect to F iff: a. S has the class node to which F associates and b. F is prevented from associating to S by virtue of i. a filter or2 ii. the SD of the spreading rule

For Barra Gaelic, then, we end up with the analysis in (16). Spreading of the frontness feature applies after C-node generation but before D-node generation:

56 (16) a. i I

V d.

i

c. i

r-

C

0

V

rI --

o

1 0 0 I o 1 I I V m V

S-place tier place tier

f. S-place tier place tier

Intervening non-labials have an empty S-place node. If we assume that they cannot act as targets because the SD of the rule demands a vowel, they will be opaque due to (15), in particular (15b.ii). The second possibility which we would like to consider involves the proposal of a condition which rules out particular associations because they cause neutralisation, i.e. because they assign F to a segment which is distinctively nonF. We do not, of course, wish to eliminate neutralisation rules in general, but we can rule out the neutralisation of (non-target) intervening segments in spreading processes. Let us extend the Opacity Condition as in (17): (17)

Opacity Condition (revised) A segment S is opaque with respect to F iff: a. S has the class node to which F associates and b. F is i. ii. iii.

prevented from associating to S by virtue of a filter or the SD of the spreading rule or the fact that F is distinctively absent while S is not specified as being the target of an association

This approach has the advantage of not needing 'empty nodes'. More interestingly, however, it leads to a more restrictive theory. As opacity is no longer due to an empty node which is generated at some point in the course of the derivation without actually functioning as an anchor point, we can exclude the possibility of such empty nodes, thus eliminating a category of rules allowed under the previous account, i.e. the category of L-rules, which were crucially distinguished from C-rules by applying before the empty C-nodes were assigned. It does indeed seem that this category of rules is idiosyncratic. The examples provided by Steriade (1986) can be shown to be the result of the particular feature system which she adopts, while other conceivable cases are of a rather special type in the sense that they involve rules which refer only to vowels. The first category is exemplified by some varieties of Finnish palatal harmony, where /ii/ and /o/, which are distinctively [-back], are skipped. On the assumption that [+back] is the spreading value and also the L-value, skipping of the two front rounded vowels provides an example of the top left-hand cell in (8). It is easy to see that different claims are made if the spreading value is taken to be [-back] or [+front], or indeed the single-valued frontness feature. Following van der Hulst & Smith (1986), we can characterise the vowels in such a case as transparent rather than skippable, in that they allow the spreading of a feature which is identical to the feature which they themselves bear (or will

57 bear). We return to the second category of so-called L-rules in a moment. Let us first investigate the consequences of our alternative proposal in more detail. There is another reason to prefer the second proposal, besides its greater restrictiveness. The first proposal is inconsistent in that we assumed that the presence of a class node, specifically an S-place node, blocks the spreading of all features which associate to this node if the segment bearing this class node does not constitute a target. Since, in the first proposal, distinctive non-palatalised consonants are provided with an (empty) S-place node, it is not clear how S-place features other than [i ] can skip these consonants, as they must, since the consonants in question can determine only one feature of the epenthetic vowel, i.e. the presence or absence of [i]. Even if we do not provide distinctive non-palatalised consonants with an empty S-place node, the same point applies to distinctive palatalised consonants, which must have an S-node because they bear [i]. In short, non-labials in Barra Gaelic are not opaque with respect to S-place features other than [ij, despite the fact that in our first proposal they have the relevant class node and do not belong to the target class of vowels. In terms of our second proposal, however, S-place features other than [i] can spread to and thus across distinctive non-palatalised consonants without problems. Consider (18), where we show the effect of potential spreading of [ij and [u] (the roundness feature) across a distinctive non-palatalised consonant: (18)

i

-0 i

S-place tier place tier

V Spreading of [i] to the intervening consonant is prevented because association of this feature with the S-place node would neutralise the opposition between palatalised and non-palatalised non-labial consonants, while our strict adjacency assumption prevents spreading to the vowel. However, there is nothing to stop the association of [u] with the consonant, as there is no pair of consonants which is distinguished merely by the presence or absence of this feature. If the intervening consonant had been a labial, then [ ij could be associated with it in the same way: there is no distinctive opposition between palatalised and non-palatalised labials, and so there is no question of neutralisation in this case. In this analysis, then, we assume that vocalic features may associate to the place node of the intervening consonant, which is either underlyingly present (distinctive palatalised consonants) or generated (distinctive non-palatalised consonants), as these features are not contrastively absent. Phonetically, these associated features are interpreted in terms of co-articulation. In this view, it is also no longer necessary to assume that labials are skipped. Rather, we can say that vowel features (including [i ]) are associated to them, and that these associations will also be interpreted as co-articulatory effects. This is in accordance with the observations made by Borgstrrfm (1940: 18-19): 'labials, in the neighbourhood of front vowels or palatal consonants, are usually pronounced with the tongue in a somewhat palatal position, and possibly also with the lips more tightly drawn than in other positions'. The upshot of this move is that we have virtually eliminated skipping from the theory. To conclude this paper, we consider two possible revisions of the framework

58 proposed here. In the first place, our analysis of (18) is based on the assumption that although distinctive non-palatalised consonants are not provided w i t h an empty S-place node by a C-rule, they do generate one. This is necessary, because otherwise (17) cannot be 'implemented'. We must assume then that the spreading rule cannot simply ignore consonants, even though they are not targets for a spreading rule w h i c h demands a vowel. This is not unreasonable, since consonants can be the source of a spreading [i]. Alternatively, we could abandon the idea that a distinction must be made b e tween a place tier and a n S-place tier (cf. Ewen 1986). In that case spreading vowel features simply associate to the place node, unless clause (iii) of the Opacity Condition is violated: (19)

i

place tier root tier skeletal tier rhyme tier Our second revision relates to the category of C-rules mentioned above, i.e. rules, involving only vowels, in which F spreads across distinctive non-F. Such rules contradict the claim that segments which are distinctively non-F are necessarily opaque. Rules of this type are reported for certain dialects of Latvian (cf. Timberlake 1986), in which palatalisation takes place across intervening distinctive palatalised and non-palatalised consonants. We illustrate this type of situation in (20): (20)

i r--. j

J' ~

V

C

place tier

V

in which Ci] spreads across C in spite of the presence of the relevant class node. In our analysis of Barra Gaelic, w e assumed that adjacency with respect to the spreading of a feature F is always defined in terms of F - b e a r i n g class nodes. Such an analysis is highly restrictive, but, given the evidence from Latvian, possibly over-restrictive. Perhaps, rather, adjacency is a parameter which is set differently for different processes, and, rather than being specified in terms of class nodes, makes reference to structural tiers. Consider situations like (16b) and (20) in terms of a model incorporating a rhyme tier and a skeletal tier (or macro tier - cf. again A r c h a n g e l ! & Pulleyblank 1986): (21)

i

1 ! i f T i f 0 Ì

place tier root tier skeletal tier rhyme tier

59 Here w e have added two structural tiers (the rhyme and skeletal tiers). We ignore, however, further details of the representation of the internal structure of the segments. Why, then, does spreading across distinctive non-palatals take place in Gaelic, but not in Latvian? We can associate this with the choice of the adjacency parameter: in Barra Gaelic, where both consonants and vowels are relevant to the process, the adjacency tier is the skeletal tier, where every segment has a representation, whereas in Latvian it is the rhyme tier, where only vowels are present. 3 Thus in Latvian the intervening distinctive non-palatalised consonant ean be skipped in the spreading process because it is not visible to the spreading element: it is not associated w i t h a node on the rhyme tier. On the other hand, in Gaelic adjacency is defined on the skeletal tier. Here the distinctive non-palatalised consonant is associated w i t h an element on the adjacency tier, and so spreading is blocked. Although our second proposal involves the introduction of a new element into our analysis to deal with a restricted class of 'L-rules', we have nevertheless maintained our central claim, i.e. that spreading is bound to adjacency, and that a feature cannot spread across a segment for which the absence of that (singlevalued) feature is distinctive. However, this now only holds if that segment is visible to the feature in question, i.e. if it is associated w i t h a node on the adjacency tier defined for the process in question. 1 *

Notes 1. This approach to opacity derives essentially from Pulleyblank (1985), and is similar to that of A r c h a n g e l ! & Pulleyblank, although their framework differs in that features are binary and empty nodes are ruled out by stipulation. 2. Cases of this type involve for example opaque low vowels in A T R harmony systems (cf. Pulleyblank 1985; van der Hulst & Smith 1986; and note 4 below). 3. The distinction between adjacency being defined structurally in two ways is similar to the distinction made by Archangeli & Pulleyblank between maximal and minimal rules. 4. The question arises as to whether we still need empty D-nodes from the theory. In particular, is it possible for a segment which is non-distinctively non-F to be opaque? Such cases can indeed be found. However, they only appear to occur if the class node which bears F is implied by the presence of other features. This appears to be the case w h e n a predictably [ - A T R ] low vowel is opaque to ATR-spreading, i.e. where it is non-distinctively non-F (cf. the discussion of Akan harmony in v a n der Hulst & Smith 1986). We assume here that the lowness and ATR features are associated w i t h the same class node: ATR

0

o

0

V

V

V

Cases of this type are discussed by v a n der Hulst & Smith (this volume).

60 References ANDERSON, J.M. & C.J. EWEN 1987 Principles of dependency phonology, Cambridge, Cambridge University Press ARCHANGELI, D. 198A Underspecification in Yauelmani phonology and morphology, PhD dissertation, M.I.T. ARCHANGELI, D. & D. PULLEYBLANK 1986 The content and structure of phonological representations, Ms, University of Arizona & University of Southern California BORGSTR0M, C.H. 1937 The dialect of Barra in the Outer Hebrides, in Norsk Tidsskrift for Sprogvidenskap 8, 71-242 1940 A linguistic survey of the Gaelic dialects of Scotland, Vol. 1, Norsk Tidsskrift for Sprogvidenskap, Supplement 1 CLEMENTS, G.N. 1985 The geometry of phonological features, in Phonology Yearbook 2, 225-252 1986 Syllabification and epenthesis in the Barra dialect of Gaelic, in K. Bogers, H. van der Hulst & M. Mous (eds) The phonological representation of suprasegmentals, Dordrecht, Foris, 317-336 EWEN, C.J. 1986 Segmental and suprasegmental structure, in J. Durand (ed) Dependency non-linear phonology, London, Croom Helm, 203-222

and

EWEN, C.J. & H. van der HULST 1985 Single-valued features and the non-linear analysis of vowel harmony, in H. Bennis & F. Beukema (eds), Linguistics in the Netherlands 1985, Dordrecht, Foris, 39-48 GOLDSMITH, J.A. 1985 Vowel harmony in Khalkha Mongolian, Yaka, Finnish and Hungarian, in Phonology Yearbook 2, 253-275 HULST, H. van der & N.S.H. SMITH 1986 On neutral vowels, in K. Bogers, H. van der Hulst & M. Mous (eds) The phonological representation of suprasegmentals, Dordrecht, Foris, 233-279 this volume Vowel harmony in Khalkha and Buriat (East Mongolian) PULLEYBLANK, D. 1985 Tonal and vocalic redundancy rules, paper presented at the Colloquium on Phonologie Pluri-linéaire, Lyon STERIADE, D. 1986 Non-underlying feature values, paper presented at the Workshop on Nonlinear Phonology, Wassenaar TIMBERLAKE, A. 1986 Assimilative-dissimilative phonological change, paper presented at the Generative Phonology Workshop Meeting, Nijmegen

Liliane Haegeman

Van constructions in Dutch 0. Introduction In this paper I will look at the use of van-constructions in Dutch in (1): la. Overal kocht hij van die oude raunten. (Overdiep 1949:434) Everywhere bought he of those old coins, lb. Waar kan ik van die zulke munten krijgen? (ib.) Where can I of those such coins obtain? lc. Er staan in m'n tuin van die hooibergjes. (ib.) There stand in my garden of those haystacks. Id. Hij werkt met van die vervelende mensen. He works with of those boring people. le. De meeste bijdragen waren van die lange teksten over recht. The most contributions were of those long texts about law. In the examples in (1) the van-construction behaves as if it were an NP: in (a) and (b) the construction is a direct object, in (c) it is the subject of an existential sentence, in (d) it is a complement of a preposition and in (e) it is a predicate, in this paper I shall describe the syntactic properties of the van-construction in some detail and I will try to analyse it in terms of the Government-Binding framework. 1. Van—constructions as verb complements Let us first consider the use of van-constructions as complements of verbs (cf. (la) and (lb)). Consider: 2a. Hij houdt veel van die oude munten. He holds much of those old coins. 2b. Hij zag van die oude boeken. He saw of those old books. 2c. Hij nam van die zure appelen. He took of those sour apples. In the discussion that follows it will be shown that the van + NP-string in (2a) behaves like a PP, van being its head. The van + NP group in (2b) is unlike a PP: it has the properties of an NP. The van-construction in (2c) will turn out to be ambiguous between the PP-structure in (2a) and the NP-structure in (2b). 1.1. R—pronouns It is well-known that so-called R-pronouns can pronominalize the complement NP in Dutch PPs (Van Riemsdijk 1973 for discussion). 3a.

Op On 3b. *:0p On 3c. ?0p On

die that t-'M» dat that iets something

p ^ » * .> •

daarop/erop thereon daarop thereon ergens op somewhere on

62 R-substitution is often obligatory (3b). Consider now the possibilities of R-substitution in the various cases of the van-complements in (2): 4a. Hij houdt er veel van/daar ... van/hier ... van/ergens van. He holds there much of/there ... of/here ... of/somewhere of. 4b.*Hij zag ervan/daarvan/hiervan/ergens van. He saw thereof/thereof/hereof/somewhere of. 4c. Hij nam ervan/daarvan/hiervan/ergens van. He took thereof/thereof/hereof/somewhere of. R-substitution isolates (2b) from the other van-constructions. The van-group in (2b) as a whole can be replaced by er, the proform for indefinite NPs (cf. Blom 1977). This is not possible in (2a): 5a.*Zij She 5b. Zij She 5c. Zij She

houdt van honden en hij houdt er ook holds of dogs and he holds there too zag honden hij zag er ook saw dogs and he saw there too (= he saw some too) nam oude boeken en hij nam er ook took old books and he took there too (= he took some too)

The ambivalence of (2c) is confirmed: it participates in the PP-R-pronominalization (4c) and in the er-substitution in (5c). 1.2. Questions with van Returning once more to (2) we see that wh-questions for the van-string also distinguish the construction in (2b) as being NP-like rather than being a PP: 6a. Waarvan Whereof 6b.*Waarvan Whereof 6c. Waarvan Whereof

houdt hij?/%Van wat houdt hij?/*Wat houdt hij? holds he?/ Of what holds he?/ What holds he? zag hij?/*Van wat zag hij?/Wat zag hij? saw he?/ Of what saw he?/ What saw he? nam hij?/%Van wat nam hij?/Wat nam hij? took he?/ Of what took he?/What took he?

Questioning by means of a wh-PP is disallowed in (6b) but is allowed (6a) and (6c). Questioning with a wh-NP is allowed in (6b) and (6c).

in

1.3. The NP following van While the NP following the preposition van in (2a) is subject to few selectional constraints, if any, that in (2b) is clearly subject to some conditions. 7a. Hij houdt veel van die oude munten/oude munten/oude munten those old coins/old coins /the old coins 7b. ... dat oude huis /een oud huis /het oude huis that old house /an old house /the old house 7c. ... die oude wijn/een oude wijn/de oude wijn that old wine/an old wine /the old wine 7d. ... zulke rare mensen such funny people 8a. Hij zag van die oude munten He saw of those old coins 8b. ... *dat oude huis /*een oud huis /*het oude huis that old house / an old house / the old house 8c. ... die oude wijn/*een oude wijn/*de oude wijn/de oudste wijn that old wine/ an old wine/ the old wine/the oldest wine

63 8d. Hij zag van zulke rare mensen such funny people The constraints on the selected NP are not clear. The preference seems to go to {+ definite] and either Q countable, + plural] or countable] NPs. However, as (8d) shows, indefinites are not excluded. I cannot, for the moment, offer an analysis or even a descriptive generalization. One interesting observation is that van selects exactly the nominal elements selected by quantificational al. Also, the NP associated with van refers to something "given", i.e. to "old information" (cf. section 3 below). 1.4. Post-verbal PPs In Dutch subordinate clauses the verb tends to occupy the S-final position but some constituents, specifically PPs, may occur post-verbally: 9a. dat that 9b. dat that

hij [over dat boek] zal he about that book will hij zal praten [over dat he will talk about that

praten talk boek] book

The possibility for the PP to occur post-verbally is limited to the van-groups in (2a) en (2c). The van-group in (2b) is like NP-complements in that it cannot occur post-verbally (cf. 11). 10a. 10b. 10c. 11a. lib.

dat that *dat that dat that dat that •dat that

hij houdt |^an die oude munterQ he holds of those old coins hij zag [van die oude munten] he saw of those old coins hij nam |~van die zure appels] he took of those sour apples hij die oude boeken zag. he those old books saw. hij zag [3ie oude boeken] he saw those old books

1.5. Conclusion So far I have shown that it is necessary to distinguish two van-groups as Vcomplements. Examples like (2a) have a van-PP and all PP properties are applicable. Examples like (2b) are not like PPs: the van-group has NP properties (cf. 1). Finally in (2c) the van-group is ambivalent between the two structural possibilities. 2. French analogues Kupferman (1979) discusses a number of constructions with de in French which bear a strong similarity to the cases discussed above. He shows that we need to distinguish between the de ! NP construction in (12a) and that in (12b). 12a. Je mange de I eat of 12b. II a vu He has seen

la viande. the meat. de la viande. of the meat.

I shall only discuss of the contrasts between (12a) and (12b); for further discussion, the reader is referred to Kupferman's own article. In essence Kupferman shows that the de-group in (12b) is not to be equated with a PP, but has NP-like properties, while that in (12a) is ambivalent between a PP and an NP structure. (12a) thus corresponds to our (2c) and (12b) to (2b). (12a)

64 allows, for instance, for relativization with dont does not. 13a.

La The 13b. *La The

viande dont ..je meat of which-/"I viande dont il meat of whiefi he

('whereof'), while (12b)

mange. eat. a vu. has seen.

Wh-movement for interrogatives also shows that the de-phrase in (12b) is not a PP: 14a.

Il He 14b. *I1 He

m'a has m'a has

demandé asked me demandé asked me

de of de of

quoi what quoi what

j'ai mangé. I have eaten. j'ai vu. I have seen.

On the other hand, non-restrictive relatives introduced by ce que are allowed after both manger and voir in (12 ). 15a. J'ai mangé de la viande jaune, ce que il a aussi mangé d'ailleurs. I have eaten of the yellow meat, which he has also eaten as a matter of fact. 15b. J'ai vu de la viande jaune, ce qu'il a aussi vu d'ailleurs. I have seen yellow meat, which he has also seen as a matter of fact. While (12a) is ambivalent between the NP-and the PP-interpretations of the dephrase, (15a) has only the NP-reading also found in (15b) and (12b). Kupferman uses evidence like that above to argue that in (12b) the de-phrase has NPstatus and that in (12a) is either NP or PP. Exploring the contrast for French and Dutch further we might say that the preposition van in the PP-structure heads the PP and theta-marks its complementNP, perhaps in conjunction with the verb. J.n the NP-like constructions the preposition does not play any part whatsoever in theta-marking the NP following it. 3. Van-NPs So far, I have only dealt with van + NP strings as V-complements, but the examples in (1) show that the construction may have the other Grammatical Functions associated with NPs. This section considers the full range of such van-NPs. 3.1. Indéfiniteness One striking property of the NP-like van-string is that it functions specifically as an indefinite NP. Examples like (lc) confirm this observation: the van-construction is the (VP-internal) subject of an existential sentence, a possibility only allowed for indefinite NPs (the so-called definiteness effect). Similarly in French the de-NP may be used as the subject of an il-existential sentence : 2 16. II is venu de ces gens bizarres. There is come of those funny people. The indefiniteness of the van-NPs is confirmed by the observation that they can be pronominalized by er, the proform for indefinites. 17a. Hij kocht rare boeken en kocht er ook. He bought funny books and I bought er ('some') too.

65 17b. Hij zag van die rare boeken en ik zag er ook. He saw of those funny books and I saw er ( 1 some') too. 3.2. Agreement Consider the following: 18a. dat er van die oude boeken te koop waren /*was/*waart that there of those old books for sale were/*was/*were 18b. de la bière s'est/*se sont/ écoulée/*écoulé of the beer is /are run out In both French and Dutch the de/van-NP agrees for person and number with the finite verb. In addition there is gender agreement with the participle in French. The agreement facts give us another argument for not considering the P van as the head of a P-projection. Consider the representation in (19): 19.

XP

*

,

In order to ensure the agreement observed in (18) it must be the case that the features of the N percolate to XP. If XP were PP, and P the head, then it not obvious how this would be allowed. Under the view that XP is an N-projection, feature percolation from N can be explained. The N-projection represented by the van-group is parallel with other specifierless N-projections such as bare plurals or bare mass nominal groups in being indefinite. In bare nominal groups, where there is no overt NP-specifier, it is not obvious whether the N-projections ought to be interpreted as having a zero-specifier: 20a. [n' Oude boeken] Old books 20b. Oude wijn} Old wine

OR

[ N p 0 Oude boeken]

OR

[ N p 0 Oude wijn]

The role of van seems to be that of turning a N-projection with specifier into a specifierless N-projection. One way of interpreting this is to see van here as the marker of the 'partitive' genitive, which is specifically associated with an indefinite interpretation (cf. Belletti 1987). 4. Analysis 4.0. The main options So far the discussion of the van-phrases in (1) has been fairly informal. I have -argued that there is evidence to say that the vein-string is nominal in nature and is not a PP. The question to be addressed now is how to formalize this proposal in terms of Chomsky's Government and Binding framework (1981). I shall offer some possible analyses, though there is no conclusive evidence to choose among them here. The following options will be discussed: (a) the van-phrase is a genitivized N-projection with the structure in (21):

66

(b) The van-construction is the bare equivalent of the partitive structures: (22) drie een 0

van die oude boeken van die oude boeken van die oude boeken

(three of those old books) (one of those old books) (of those old books)

In the latter option two possibilities remain: (b-i) partitives are NPs with non-overt nominal heads: diagram (23a) (b-ii)partitives are nominal projections in which the quantifier, if present, is a specifier of the van-N : diagram (23b) (cf. Bennis (1979), Coppens (1985), Grover (1986), Klein (1980) for discussions of partitives). (23a) = hypothesis (b-i):

(23b) = hypothesis (b-ii): NP, Q

drie

0

N

van

die oude boeken

These proposals will be discussed briefly in the next section. 4.1. Van as a case—marker If we take the van-NP construction to be a genitivized NP then its nominal functions will follow: like all NPs it can be subject, object, etc. and the fact that van does not act like a genuine PP-head is predicted. This analysis is closely parallel to Kupferman's analysis for de-NPs in French. In order to explicitize the proposal we would have to claim that van, a preposi-

67 tion, is a case-marker like the French dative a. Such a view can be supported by arguments similar to those advanced by Jaeggli (1981:21) and by Vergnaud (1974) for the NP-status of dative phrases in French. The arguments concern, among other things, coordination and relativization. The agreement facts pointed out in 3.2. also follow. However, the role of the (genitive) case is problematic. If van assigns partitive genitive to the NP it is not clear why the NP in question must always appear in a position to which structural case is assigned, as can be seen from the ungrammaticality of the following sentences in all of which the van-NP appears in a position not marked for case: (24a) *dat hij zag van die oude boeken that he saw of those old books (24b) *0m van die jonge studenten dat te doen For of those young students that to do Assuming case is assigned to the left in Dutch, the post-verbal position in (24a) is caseless. In (24b), the subject of infinitivals with om lacks case, being in an ungoverned position. One possible solution would be to say that the partitive genitive case is a realization of any structural case in Dutch and French for a certain range of NPs, but this is not satisfactory. One immediate question will be which NPs will be realized with 'partitive case': as shown above the selectional restrictions on the NP associated with van are far from clear. Note also that in languages with overt partitive case it is indefinite NPs that tend to be able to be marked partitive (cf. Belleti 1987) while in the cases discussed here it is definite NPs that are particularly favoured. Indeed, as shown in (8), indefinite NPs do NOT take the van-construction easily. A further question would be what determines the realisation of a particular structural case as either nominative/accusative or partitive genitive. At a more general level we should ask what the role of the partitive case would be. Does it render NP 'visible' for theta-role assignment, like structural case? If so, what distinguishes it from structural case? If not, why not? Alternatively we could say that the partitive genitive is a and that the van-phrase moves to an A'-TOP-position at over the entire proposition. After movement of the relevant ing trace will be a variable which will be case-marked structural case position. The latter option ties in n^ptly that the van-phrases in (1) refer to something 'given'.

kind of TOP-marker LF to take scope phrase the remainsince it is in a with the intuition

4.2. The partitive analogue The structure in (23b) is very close to that in (21); the only difference is that we assume a SPEC-node with a zero-quantifier in the (23b). Presumably we can say then that van casemarks the lower N-projection and that the resulting group is an N-projection selected by the quantifier. Assuming that the full construction is also nominal and needs case we could argue that the structural case is assigned to NP independently of the partitive genitive assigned to the lower N-projection. Note that, if we assume that the construction as a whole is marked partitive genitive, then the issues concerning the role of partitive/ genitive case and structural case pointed out above reemerge. An alternative analysis is that in (23a), where it is assumed that partitives have empty nominal heads. For French this proposal receives some support from the observation that there is an analogous construction with overt N' (cf. Milner's analysis (1979)):

68 (25) J'ai vu I have seen

(quelques-un) de ces gens bizarres. (some) of those funny people.

One question that needs to be asked is what kind of non-overt nominal this is. Obviously it could not be a trace of movement. This leaves the options of interpreting the empty category as PRO or pro. The former is not possible given that the partitive constructions in general and the bare partitives we are discussing occur in a governed position and under standard assumptions PRO needs to be ungoverned. The alternative is to treat the empty N-projection as pro, a non-overt pronominal. This proposal raises questions too. In the theory we are assuming an empty pronominal must be licenced and locally identified (cf. Rizzi 1986). In the constructions in (23b) we shall assume that a case-governing head (e.g. a verb or preposition) will licence the N-projection. The content of pro is determined locally by the content of the van-NP: the agreement facts in (18) have already shown how the features of the van-string determine those of the dominating N-projection. Though this solution may be fairly attractive it is non-standard: usually the feature content of pro is taken to be determined by verbal inflection or by clitics. Note though that the structural configuration in (23b) would allow local identification of the content of the ec by the van-phrase: the latter c-commands the former locally. Assuming the ec in (23b) is pro raises another problem. So far in GB-theory pro was taken to be the non-overt equivalent of a definite pronominal while in the construction under discussion the empty pronominal is indefinite and quantifier-like in nature (cf. 25) . It would be interesting to see if there are other instances of indefinite or quantificational pro and under what conditions this is allowed (cf. Rizzi 1986 for instances of generic pro). 5. Conclusion In this paper ^E have discussed in some detail the syntactic properties of vanconstructions . I have argued that the van-NP structure has nominal properties rather than prepositional ones. I have provided syntactic evidence for this assertion and I have examined the analysis of the construction within the theoretical framework of GB-theory. The analysis leaves many problems. Further research in the area of partitives and the status of van as a marker of partitive case is clearly necessary. Notes *

I wish to thank the participants of the TIN-dag and the reviewers and editors of the LIN volume for their comments. Special thanks are due to Frits Stuurman.

1. For some speakers the use of van with Prepositions is restricted: (i) met van die oude boeken with of those old books (ii) %over van die oude boeken about of those old books 2. The non-agreement of the de-group in (16) is in line with the lack of agreement in the il-existential in French: (i) il est /»sont venu deux filles there is/ are come two girls (ii) il est/*sont venu de ces gens bizarres.

69 3.

This proposal quantifiers.

is

inspired

on

Pesetsky's

(1982)

discussion

of

Russian

4. For isne semantics oi the construction cf. (Jacobs 1986).

References ANS

BELLETTI, A. 1987 The Case of Unaccusatives. To appear in Linguistic Inquiry. BENNIS, H. 1979 Appositie en de interne structuur 1986 Gaps and dummies, Dordrecht, Foris BLOM, A.

van NP1 s, in Spektator 8, p. 209-28.

1977 Het kwantitatieve er, in Spektator 6, p. 387-395 CHOMSKY, N. 1981 Lectures on Government and Binding, Dordrecht, Foris COPPENS, P.-A. 1985 De aard van het quantitatieve er, in De Nieuwe Taalgids 78,2, p. 149-63 GROVER, C. 1986 An analysis of specifiers in English. Ms. Department of Linguistics University of Lancaster JACOBS, 1986 Partitieven, enkele syntactische en semantische aspecten, in TABU 16,4, p.111-144. JAEGGLI, 0. 1981 Topics in Romance Syntax, Dordrecht, Foris KLEIN, E. 1980 Determiners and Social Sciences. KUPFERMAN, L.

the

category

Q.

Ms.

University

of

Sussex.

School

of

1979. L'article partitif existe-t-il?, in Le français moderne 47,2, p. 1-17. MILNER, J.-C. 1979 De la syntaxe à 1'interpretation : quantités, insultes, exclamations, Paris, Seuil OVERDIEP, G.S. 1949 Stilistische Grammatica van het moderne Nederlands, 2nd impression, G.A. van Es, Zwolle, Tjeenk, Willink PAARDEKOOPER, P.C. 1971 Beknopte ABN syntaxis, 5th impression, Eindhoven PESETSKY, D. 1982 Paths and Categories, MIT diss. RIEMSDIJK, Henk van 1978 A case study of syntactic markedness, Dordrecht, Foris RIZZI, L. 1986 Null Objects in Italian and the Theory of pro, in Linguistic Inquiry 17,3, p. 501-58

70 S C H U T T E R , G. D e e n P . V A N H A U W E R M E I R E N 1983 De structuur van het Nederlands. Taalbeschouwelijke Sikkel VERGNAUD, J.-R. 1974 French Relative Clauses, MIT

diss.

grammatica, Malle,

De

Aafke Hulk & Erna Vermeulen

The impersonal construction in French and the Burzio-generalization 1.

Introduction*

Consider the following impersonal, existential construction in French (1)

il est arrivé deux bateaux 'subject clitic is arrived two boats'

A striking property of this construction is that the verb does not agree with the postverbal NP but with the (impersonal) subject clitic il. This seems to be related to the Case-marking of the postverbal NP. In this article we discuss the problem of Case assignment in the French impersonal construction of the form in (2) [1] (2)

il

V

NP

We compare it to equivalent constructions in English, as in (3) and Dutch, as in (4) which differ from the French one with respect to subject-verb agreement and the status of the expletive element in preverbal position. (3) (4) We argue mechanisms postverbal accusative

There came two boats Er kwamen twee boten that these differences are related to the Case-marking operating in this construction: while in English and Dutch the NP is assigned nominative Case, in French it is marked for Case.

Finally, we propose a modification consider some of its consequences.

of

the Burzio-generalization and

2. Chain formation Pollock (1983) compares the French impersonal construction with the English one with respect to chain formation between the expletive element and the postverbal NP. According to Pollock chain formation between ¿1 and the postverbal NP in French is excluded by two general principles. The first principle states that a chain may contain only one element marked for number. The second is some sort of Case-uniqueness principle; it states that a chain may contain only one element of a specific case K. The subject clitic is inherently marked for (nominative) Case as well as for the features gender, person and number. Chain formation between iJ and the postverbal NP would therefore violate Pollock's "number uniqueness" principle. Furthermore Pollock assumes that V assigns nominative Case to the postverbal NP in the impersonal construction; consequently chain formation would also violate his Case-uniqueness principle. In English, however, the expletive element there is marked neither for Case nor for number. Chain formation with the postverbal NP

72 is therefore not excluded by Pollock's principles. Although we do not think V assigns nominative Case in French (see section 3) and we have some serious doubts about the validity of Pollock's "number uniqueness" principle [2], we agree with him in rejecting chain formation between ¿1 and the postverbal NP in the French impersonal construction. As far as we can see, there are no real arguments in favour of such a chain formation in French, neither with respect to visibility requirements nor in relation to subject-verb agreement. As the ungrammaticality of (5)-(7) demonstrates, the verb in the French impersonal construction (5) cannot agree with the postverbal NP, while the verb in the English (6) and Dutch (7) construction has to agree with it. (5) *il arrivent deux bateaux (6) *there comes two boats (7) *er komt twee boten Now, if we assume that agreement between the inflected verb and an NP shows that the NP has nominative Case, it follows that the postverbal NP is nominative in the English (3) and Dutch (4) construction, but not in the French construction. Consequently, nominative Case transmission cannot be an argument in favour of chain formation in French. This would be problematic anyway, since the features of and the postverbal NP do not necessarily match. In the next section we will adopt the hypothesis that the verb assigns structural accusative Case to the postverbal NP in French, making it visible with respect to theta-role assignment. Pollock (1983) provides another argument in favour of the difference in chain formation between French and English. It concerns wh-movement of the postverbal NP: this is possible in French, whereas it is generally impossible in English, as shown in (8) and (9): (8) *how many boats did there come ? (9) combien de bateaux est-il arrivé ? According to Pollock chain formation would imply that the trace of the wh-moved NP is bound twice: by there/il as well as by the NP in COMP. Hence the ungrammaticality of (8) versus the grammaticality of (9) could be explained by assuming chain formation in English, but not in French. In English chain formation would also be necessary for nominative Case transmission, according to Pollock. However, as pointed out to us by P. Coopmans, the status of this argument is rather weak: it is generally assumed in the literature that binding between an expletive element and an NP does not violate the Binding Theory. Consequently this theory could not block (8). What might be involved is a violation of the Bijection Principle. The situation in English is complicated, however, by the fact that in be-existentials the verb does not necessarily agree with the postverbal NP and moreover, this NP may be extracted by wh-movement, which, according to Pollock, would imply absence of chain formation. The problem of chain formation in the English impersonal construction is beyond the scope of this article and we leave it aside here. For Dutch it has been argued by Bennis (1986) that although the postverbal NP has nominative Case, there is no chain between the expletive er_ and the NP in the existential construction (4). Er is freely adjoined to a V-projection, subject to pragmatic conditions involving definiteness, and the NP receives nominative Case directly by INFL that canonically governs all elements outside T [3], Moreover wh-movement of

73 the NP is possible, just as in French, as shown in (10). (10)

hoeveel boten komen er aan ?

In this section we have rejected chain formation between and the postverbal NP in the French impersonal construction. We have very briefly considered the English construction where chain formation seems sometimes possible. As for the corresponding Dutch construction we have adopted the analysis by Bennis (1986), who argues against chain formation. 3. Case assignment in the French impersonal construction. In the preceding section we have rejected the idea of chain formation and of Case transmission between and the postverbal NP. We will argue here in favour of the hypothesis that V directly Case-marks the postverbal NP in French [4] . A similar idea has been proposed by Pollock (1983) and by Belletti (1986), among others. Belletti assumes that the postverbal NP is inherently marked for partitive Case. This proposal implies an extension of the generally accepted mechanisms of inherent Case-marking. Inherent Case-marking is usually assumed to take place at D-structure and to be linked to theta-role assignment. According to Belletti, ergative verbs and transitive verbs optionally assign inherent partitive Case to the NP they govern at D-structure, if this NP is indefinite. This last restriction seems to be particularly difficult to account for. Moreover, the inherent Case-marking cannot be linked to a specific theta-role either, since in French not only ergative verbs, but also unergative intransitive verbs may occur in constructions with an indefinite postverbal NP (11)

il dort un chat au coin de la cheminée 'there sleeps a cat on the corner of the chimney'

The postverbal NP in this construction has the external theta-role because dormir only takes one theta-role, the external one, whereas the postverbal NP with ergative verbs always has the internal theta-role [5], There is still another problem with Belletti's analysis. In her account, partitive Case assignment is crucially related to the so-called definiteness effect: only NPs that may be inherently marked for partitive Case show this effect. However, in Dutch, as shown by Bennis (1986), the postverbal NP in this construction is marked for nominative Case and it has to be indefinite. Belletti would have to allow a chain with nominative and partitive Case here, as she indeed seems to propose for similar cases in standard Italian. Usually, though, superimposing of inherent and structural Case is excluded by Universal Grammar. Finally, if Bennis (1986) is right in assuming that in Dutch not only the subject but also the direct object sometimes has to be indefinite in expletive erconstructions such as (12), Belletti will have to allow partitive Case to be assigned twice: another novelty in Universal Grammar. (12)a

er heeft niemand iets gekocht 'there has no-one anything bought' b*? er heeft niemand dat gekocht 'there has no-one that bought'

74 The main advantage of Belletti's analysis, as far as we can see, lies in the fact that by assuming inherent Case-marking of the object, a violation of the Burzio-generalization, as in (13), is avoided, which Belletti assumes to apply only to structural Case assignment. (13) if a verb Case-marks its object, it theta-marks its subject Although we feel that on the whole Belletti's analysis raises more problems than it solves, we do agree with her that Case is assigned by in this construction. Contrary to Belletti though, we assume that in the French construction V assigns accusative Case to the postverbal NP structurally. The ungrammaticality of sentences such as (14) has often been cited in the literature as a problem for accusative Case assignment. (14) *il les arrive 'scl them arrives 1 However, this is only apparently so, because an indefinite NP governed by V is always pronominalized by en, as shown in (15) (15)

il mange des petits pains: il en mange 'he eats small buns

: he of them eats'

We therefore expect the same in the impersonal construction: (16)

il 'sclen ofarrive them arrive'

and the ungrammaticality of (14) is due to the definiteness effect only. A minimal condition for the postverbal NP in the impersonal construction to receive (accusative) Case is government by V. This condition is fulfilled straightforwardly as far as ergative verbs are concerned, since these verbs are subcategorized for an internal argument within 7. However, we also find unergative verbs in the French impersonal construction, as was shown in (11), repeted here as (17) (17) il dort un chat au coin de la cheminée 'scl sleeps a cat on the corner of the chimney' These verbs select only an external theta-role, so we do not expect the postverbal NP in (17) to be governed by V. The "quantitative en" test, though, shows that V does govern the postverbal NP, not only if it is ergative, as in (18), but also if it is unergative, as in (19): (18) il en arrive deux 'scl of them arrive two' (19) il en dort un au coin de la cheminée 'scl of them sleeps one on the corner of the chimney' Quantitative en may only be cliticised to a V that governs a quantitative NP with an empty head, just as ne in Italian (see Burzio (1981)). In Italian, however, the "ne test" crucially distinguishes ergative verbsthat govern the postverbal NP - from unergative verbs, which do not govern it. In French, on the other hand, both types of verbs govern the postverbal NP in the impersonal construction. Consequently the minimal

75 condition for accusative Case assignment is ergative and for unergative verbs.

fulfilled

in French, for

Resuming, we tentatively proposed in this section that (in French) all intransitive verbs may optionally assign accusative Case to the NP they govern, thus allowing the postverbal NP in the French impersonal construction to be Case-marked in situ by V. 4. The Burzio-generalization reformulated A problem for our analysis is that it violates the Burziogeneralization: if we assume that V assigns accusative Case in (2), the Burzio-generalization predicts that an external theta-role will be assigned to the preverbal NP/S position, contrary to what we find. Recently however, a modification of this generalization has been proposed in Borer (1986) on the basis of certain data in substandard Hebrew which also seem to violate the "classical" Burzio-generalization. Borer cites the following examples (20) haya 'was (21) haya 'was

katuv'et ha-yedi'a ha-zot ba-'iton written acc the message the-this-f in the paper' ktuva yedi'a xashuva ba-'iton written-f message-f important-f in-the-paper'

In both sentences the same verb appears in its passive form. In (21) the postverbal NP agrees with the verb and therefore we assume that it has nominative Case. On the other hand, in (20) the postverbal NP does not agree with the verb, which takes an impersonal 3rd person singular form and the NP is preceded by an accusative Case-marker. Borer proposes the following generalization: if accusative is assigned and INFL is present, there must be an I-subject (= an NP coindexed with INFL). What she seems to do here is relate accusative and nominative Case assignment. We could rephrase her idea in terms of (22): (22)

V may assign accusative Case only if nominative Case has been realized through AGR

If we assume furthermore that, at least in French, all verbs are optional Case assigners, the fact that V assigns accusative Case in the French impersonal construction is no longer a problem since nominative Case is realized in the form of the subject clitic il. This new formulation of the Burzio-generalization also allows us to explain an intriguing fact about (free) inversion in some northern Italian dialects which seem to violate the classical generalization. In these dialects, contrary to what happens in standard Italian (23), the postverbal NP does not agree with the verb and moreover there is (sometimes) an impersonal SCL in preverbal position (24), just as in the French impersonal construction: (23) hanno telefonato le tue sorelle (It) (24)a *hanno telefonato le tu'sorele (Fiorentino) 'have phoned your sisters' b gl'ha telefonato le tu'sorele 'scl has phoned your sisters'

76 The postverbal NP may be extracted by wh-raovement from this position (25)

quante ragazze gl'ha telefonato ? 'how many girls scl has phoned ?'

As far as ergative verbs are concerned, there is evidence that the postverbal NP is governed by V: quantitative jie may appear (26)

e ne vien tre (Fiorentino) 'scl of it comes three'

Since the NP does not agree with the verb, nor with the SCL, it seems plausible to suppose that V assigns accusative Case, just as we saw in the French impersonal construction. This Case-marking is no longer problematic if we adopt the newly formulated generalization: nominative Case is realized as a SCL , so accusative case may be assigned by V. [6] 5. Some complications. The question may be raised why accusative Case marking of the postverbal NP leads to ungrammatically in the following Dutch sentences: (27)a b

*er komt twee mannen 'there comes two men' *er wordt volksdansen gedanst 'there is folk-dances danced'

This can be explained by the fact that no nominative Case has been assigned in these sentences and that they violate the new generalization in (22). This explanation can only be valid, however, if there is no empty expletive pro marked with nominative Case. We think that there is indeed no such pro and we adopt Bennis' (1986) analysis of this construction in Dutch. Moreover we agree with Bennis that empty dummies, as he calls them, should in principle be excluded in the theory of grammar. This implies, however, that we have to come up with another explanation for the Hebrew sentence in (20), since for such examples Borer assumes an empty expletive pro marked with nominative Case. A reasonable alternative seems to be the hypothesis that in null subject languages AGR has a nominative Case feature. This feature may satisfy the requirement of the new generalization in the Hebrew sentence considered here. Since Dutch is not a null subject language, this solution does not interfere with the explanation given for the Dutch sentences in (27). This hypothesis does predict, though, that a construction similar to the one in Hebrew should be possible in Italian, a null subject language. Apparently, this is not true as (28) shows. (28) *arriva due ragazze 'arrives two girls' The postverbal NP always agrees with the verb and is therefore marked with nominative Case and not with accusative. Let us suppose that Italian shows the genuine pro-drop paradigm. This forces us to abandon the hypothesis that the nominative Case feature in AGR of pro-drop languages may satisfy (22). Consequently the explanation of (28) can be equivalent to the one given for (27): nominative Case has not been

77 realized, consequently accusative Case may not be assigned to the postverbal NP. We are left with the Hebrew sentence in (20) where there is no lexically realized nominative, although the postverbal NP is marked for accusative Case. The crucial question now is: how has this Case been assigned, structurally by V or morphologically by the accusative Case marker et ? Hebrew being a language with morphological Case marking (contrary to e.g. Italian), we will adopt the latter hypothesis [7], In that case (22) will not be involved in the explanation of (20). The plausibility of this hypothesis is strengthened by the fact that in other languages with morphological Case marking, such as German and Middle Dutch, accusative Case may be morphologically realized without the presence of a nominative marked element, as shown in (29) [8], (29) mich friert 'me (acc) is cold* Haider (1984) provides a solution for German that is comparable to the one proposed here by us for Hebrew: according to him, the Burziogeneralization may be circumvented only in languages which have, what he calls, lexical (as opposed to structural) Case indices. However, further research will have to reveal the exact relation between structural and morphological Case assignment in connection with principles such as the Burzio-generalization [9], 6. Conclusion. In this article we have argued that in the French impersonal construction the postverbal NP receives Case from the verb structurally, independent of its theta-role which may be internal or external. This Case-marking violates the so-called Burzio-generalization since no theta-role assignment to the preverbal NP/S position is involved. We have tentatively proposed a modification of the Burzio-generalization in the following terms: nominative Case has to be realized through AGR before accusative Case may be assigned. This newly formulated generalization works out nicely not only for the French impersonal construction but also for inversion in some northern Italian dialects and for the Dutch expletive er-construction. It raises some problems however, with respect to languages such as Hebrew and German which allow Case to be assigned not only structurally, but also morphologically. Further research will hopefully reveal the right solution for this problem.

Notes *The research for this article was part of the RUU/VU Research Project "Formal Parameters of Generative Grammar" financed by the Dutch Ministery of Education. We thank the editors for helpful comments. 1. Impersonal passive constructions and the impersonal construction with reflexive se are left out of the discussion.

78 2. This principle seems to exclude chain formation in clitic doubling constructions. 3. Bennis (1986) also abandons the Extended Projection Principle. 4. The fact that the construction is ungrammatical with an AP instead of a V suggests that it is V rather than a possible eventual chain that is relevant for Case-marking: (i) *il est content trois de mes amis 'scl is satisfied three of my friends' 5. We do not consider here the problem of mapping theta-roles lexicon onto syntactic structure. In forthcoming work (Hulk argues that all theta-roles of X are base generated within internal theta-roles inside Xbar, the external-role outside

from the 1987) Xmax: Xbar.

6. Notice that this construction constitutes a serious problem for Belletti's partitive analysis, since the postverbal NP may be definite contrary to the postverbal NP in the French impersonal construction. This suggests once again that Case assignment cannot be linked directly to the definiteness effect. 7. This was suggested to us by Fred Weerman. 8. This was brought to our attention by Hans Pijnenburg. 9. Sentences such as (i) and (ii) (i) *John dances it (ii) *it has arrived two men can be excluded by the Theta criterion if we assume, following among others Bennis (1986), that ¿t is a referential expression. References BELLETTI, A. 1986 Unaccusatives as Case-assigners, Center for Cognitive Science, Lexicon Project Working Papers 8, MIT, Cambridge Mass. BENNIS, H. 1986

Gaps and Dummies, Foris, Dordrecht

BORER, H. 1986

I-Subjects, in Linguistic Inquiry 17, p. 375-416

BRANDI, L. and P. CORDIN 1983 Dialetti e italiano: un confronto sul parametre del soggetto nullo, in Rivista di grammatica generativa 6, p. 33-89 BURZIO, L. 1981 Intransitive verbs and Italian auxiliaries, PhD dissertation, MIT, Cambridge Mass.

79 CHOMSKY, N. 1981 Lectures on Government and Binding, Foris, Dordrecht HAIDER, H. 1984 The Case of German, in J. Toman (ed.) Studies in German Grammar p. 65-103, Foris, Dordrecht HULK, A.C.J. 1987 Le français: une langue prodrop?, to appear in Recherches Linguistiques. Paris VIII KAYNE, R. 1984 Connectedness and binary branching, Foris, Dordrecht POLLOCK, J-Y. 1983 Accord, chaînes impersonnelles et variables, in Linguisticae Investigationes VIII, 1, p. 131-181 SAFIR, K. 1982 Syntactic Chains and the Definiteness Effect, PhD dissertation, MIT, Cambridge Mass. VERMEULEN, E. 1986 Un inventaire de la problématique d'une construction impersonnelle du français, unpublished manuscript, VU Amsterdam

Harry van der Hulst & Norval Smith

Vowel harmony in Khalkha and Buriat (East Mongolian) 0. Introduction In this paper we analyze certain aspects of the vowel harmony systems of the East M o n g o l i a n languages Khalkha and Buriat. These languages have dual harmony systems, involving both ATR and labiality. In the ATR-system the vowel /i/ functions as neutral, whereas in the system of labial harmony high vowels are neutral. Both harmony systems thus raise interesting questions regarding the behaviour of neutral vowels as either opaque or transparent. This paper is organized as follows. In section 1 we briefly discuss the K h a l k h a vowel system, and its historical development. We will point out that the behaviour of /i/ in the ATR-system is predicted by the theory of neutral vowels proposed in V a n der Hulst & Smith (1986). In section 2 we show that the same vowel's behaviour in the labial system is not, as it is transparent where we w o u l d expect it to be opaque. Expected and attested opacity is demonstrated using vowel harmony in Akan (Kwa) and Bashkir (Turkic). In section 3 we offer a solution, which makes use of recent proposals concerning the internal structure of segments. In section A we w i l l then discuss a problematical aspect of labial harmony in Buriat, which leads to a (tentative) refinement of the model.

1. The Khalkha vowel

system

Our discussion of Khalkha (K) vowel harmony relies on Svantesson (1985), w h o convincingly argues that East Mongolian languages have harmony based on tongue root position (cf. also Hattori 1982, G r e g e r s o n 1976). Svantesson shows that the vowel system and harmony relations in K are those in (la) and not those in (lb), which are presumably correct for Classical Mongolian:

(I)

Vowel system i u Q e e a o

(2)

Harmonic pairs ATR LABIAL u - e - e e - o a - o e - a

(K)

b.

b.

Vowel system (Classical Mongolian) i ii u e o a o

Harmonic pairs PALATAL a - u 0 - 0 e - a

Before we turn to an analysis of the labial system, we w i l l offer an interpretation of the historical development from palatal to ATR-harmony, and point out why the transparency of /i/ is exactly what we expect. The major difference between K and Classical M o n g o l i a n resides in the fact that *ii and *o have been backed to /u/ and /a/ respectively, pushing down, so to speak, the old * u and *o to /a»/ and /o/. A s a result of this, the nature of

82 the relation dimension to feature system Hulst & Smith (3)

a.

between harmonic pairs has involving the height-dim we employ (cf. Ewen & Van 1986, ms.) the two systems

changed from involving the front-back nsion. In terms of the single-valued der Hulst 1985, this vol. and Van der are represented as follows: b.

Khalkha IE 1}

IE

(E A I ) IA}

U> U} IE A UÎ IA U)

Classical M o n g o l i a n

IA I lit {A} IA U>

(Where [E] = advanced tongue root; [A] = open jaw; [I] = front; [U] = round) Several scholars have noted that there is an intrinsic correlation between frontness (our [I]) and advancing of the tongue root (our [E]). If w e assume that vowels specified as [I] are subphonemically provided with [E], we can view the historical development as the phonemicization of [E], and concomitant subphonemicization of [I]. Phonetically the feature [I] is only preserved for /i/ and /e/. In a synchronic analysis of K this feature is quite clearly not present in the underlying representation of /e/, as the rounded counterpart of /e/ is /»/, a back vowel. The feature [I] then will be added by a redundancy rule to derive surface /e/: (4)

a. FCR: ->([E] a-.([I] v [U]))

b.

A-rule: Add [I]

The Feature Cooccurrence Restriction (FCR) says that no segment c a n have the feature [E] without having either [I] or [U]. The rule in (4b), called an AUTOMATIC RULE (cf. Stewart 1983) applies whenever a violation of (4a) occurs. We assume that in the unmarked case A-rules add rather than delete features which have not (yet) been used in the representation of segments. The F C R in (5) will prevent [I] from being added to segments which possess [U]: (5)

FCR: - ( [ I ]

[U])

As for /i/, we could take IE} as its representation and derive /i/ by rule (4b), but we could also assume that the representation is tl}. In that case we would need (6). This doesn't necessarily imply a more costly analysis. (6a) is a valid FCR, ruling out a class of i n f o r m e d segments and as such it is part of the grammar anyway. (6b) represents the unmarked A-rule (cf. supra): (6)

a.

FCR: ->( [ I ] A -"[E])

b.

A-rule: A d d [E]

In this paper we will simply assume that the K /i/ is represented as (El. With regard to the ATR-system, /i/ acts transparently. If preceded by ATR-vowels, suffixes show up with ATR-vowels, but if preceded by non-ATR-vowels suffixes have non-ATR-vowels. This is the behaviour we expect g i v e n the proposals in V a n der Hulst and Smith (1986). In a nut shell, what is proposed there is that neutralization in favour of the spreading value leads to transparency, whereas neutralization in favour of the default value leads to opacity. As /i/ lacks an harmonic counterpart / t / , this case represents a neutralization toward the spreading value and thus /i/ should behave transparently. We realize that there is more to say on the ATR-system and we refer to Van der Hulst and Smith (ms.). Here we choose for discussing the more challenging data involved in the labial system, which appear to provide a problem rather than support for our earlier proposals.

83 2. Labial harmony In K /«/ and /»/ occur non-initially if and only if /e/ or lo / occur in the initial syllable. There is no rounding correlation for high vowels, and these vowels are therefore neutral with respect to labial harmony. On the basis of what was said at the end of the preceding section we would expect /i/ to be opaque, as it represents a neutralization toward the default value (i.e. non-round), whereas /u/ and /»/ ought to act transparently as they represent a neutralization toward the spreading value. Unfortunately, it turns out that the behaviour of both the rounded and the unrounded high vowels is precisely the other way around. We will first look at the rounded vowels /u/ and /«V and then turn to /i/ later. /u/ and /ca/, in whatever position, can never be followed by rounded low vowels. This implies that rounded high vowels do not function as triggers, but it also means that such vowels function as blockers. We can explain this behavior by requiring that vowels whose [U] feature is spread as well as vowels to which this [U] may associate must be low, i.e. [A]-specified. This is a language-specific constraint on [u]-harmony: (7) a.

[E] [A] [U]

o o

o

o— o—

o»»>o (/e/ /u/ + /e/)

[Uj

[U]

(7a) should be read as follows. The little circles represent features associated to a node, which bears all the vocalic features. Adjacent features are assumed to be one specification (due to the OCP). " » > " represents a spreading of [U]. (7a) then expresses the same as (7b). Below our conception of the way in which vocalic features are hierarchically structured will change, but we will keep the notation used in (7a) constant. The spreading in (7) is not allowed since /u/ is not specified for [A]. If such spreading was allowed, the feature [U], characterizing /u/, would be set afloat and associate to the suffix vowel, i.e. /u/ would behave transparently. The opacity of high rounded vowels follows then from the fact that they are P-bearing units for [U], while [U]-harmony does not apply to them because of the extra requirement that a target vowel must be [A]-specified. We now turn to /i/. A central question in this paper is how we can account for the fact that /i/'s appear to be transparent. We expect /i/ to be opaque, given the proposals in Van der Hulst and Smith (1986). Let us explain in" more detail why we have this expectation. The strongest claim concerning assimilation is that it always involves spreading under adjacency. Let us express the adjacency requirement in terms of the following condition: (8)

COLINKING CONSTRAINT Colinking involves adjacent elements.

Assuming (8) as a universal principle we arrive at a quite simple theory of opacity, as was first pointed out by Pulleyblank (1985). Given that some segment S is P-bearing with respect a spreading feature F, S will be opaque if F may not associate to it either due to an extra condition on the spreading

84 rule or due to a FCR (cf. Archangeli and Pulleyblank 1986, E w e n and V a n der Hulst, this vol). The first case is exemplified by the opacity of high rounded vowels in the labial system in K. The second case will now be illustrated with two examples involving A T R and rounding harmony. - Like many African languages, A k a n has ATR harmony (Clements 1981). The low vowel /a/ which belongs to the non-[E] set has no harmonic counterpart, at least not at the phonological level. To account for this gap we assume a FCR which rules out the presence of [E] in the absence of both [I] and [u]: (9)

i e

u o

L

e

o

{E IJ iE A Ii

fE Uj (E A 0 )

ilj il A }

{U> {U A } U J

( 10)

FCR: ->([E] a --([I] v

[U]))

W h e n a low vowel occurs between a [E]-specified stem vowel and a suffix vowel which is harmonic with the stem, it acts opaquely. On the assumption that the low vowel is a P-bearing unit for [E], the expected result is indeed that the low vowel will block the propagation of [E]. The point is that the low vowel may not be associated to [E] due to the FCR in (10) and that, as a P-bearing unit, it may not be skipped because this would violate the condition in (8). - Bashkir has labial harmony. It has a vowel system of the following type. Note that /a/ and /ae/ lack a rounded counterpart. (11)

i e ae

U o -

(12)

FCR:

~»([U] A i[E])

A a

u o -

IE II fE A I) (A If

IE III) (E A I U )

- (E Bj IE A} (E A UJ {A}

Only non-high vowels trigger rounding harmony, but if a low vowel occurs in between a m i d rounded vowel and a mid suffix vowel, the suffix will have an unrounded vowel (Poppe 1962). Hence low vowels act opaquely. T h i s is what we predict if we assume that low vowels are P-bearing units for [U]. to the claim G i v e n the above it is reasonable to ask how we can give content that the opaque vowels in Akan and Bashkir are considered P-bearing units with respect to a spreading feature to which they cannot be associated due to a FCR, as it will be clear that this is crucial to the account of opacity just illustrated.

3. On the difference between transparency and skippability (8) implies that all spreading involves adjacent elements. Clearly, we cannot simply take this to m e a n that spreading involves strictly adjacent SEGMENTS, since in the case of harmony consonants may intervene between vowels. Rather "elements" should be understood as P-bearing units. What does it m e a n for some segment to be a P-bearing unit with respect to some feature F? T o clarify what this means we must provide some background. In recent autosegmental work it is assumed that segments are not represented by an unordered set of phonological features, but rather by an HIERARCHICALLY ordered set of features. Individual features do not associate directly to points on the "skeletal tier" but to non-terminal nodes in the segmental tree called class nodes (cf. Clements

85 1985 (and references there), Sagey 1986, Steriade 1986, Schein & Steriade 1986, Archangeli & Pulleyblank 1986). In an underspecification framework, class nodes which bear no features need not be present in the representation of a segment (cf. especially Archangeli & Pulleyblank 1986). Such nodes however must be generated, if necessary, by a convention. Given this view, we can regard a segment S P-bearing with respect to some feature F iff S is (or can be) provided with a class node C to which F associates. In addition, it follows that a segment S which is not provided by the class node C will be skipped by a spreading F if S is not "invited" to generate the class node in question. Schematically, skipping is illustrated in (13). Observe that (8) is not violated:

(13)

A

B

C

Can we use this notion of skipping to account for the behavior of K /i/? In most proposals concerning the hierarchical structure of segments [Round] is set apart from the other "vocalic" features by associating it to a separate class node. We believe that the key to explaining the skippability of /1/ lies herein. /i/ obviously has no specification [U]. Since [U] is the only feature associating to the [U]-bearing node, /i/ does not have this node underlyingly. Now recall that segments involved in [U]-spread must be [A]-specified. Given this condition, /i/'s will not be considered at all and therefore not be invited to generate the [U]-bearing node. This explanation then is based on the following hierarchical grouping of vowel features: (14)

a.

[E] [A]

[I] [U]

o

o

o

o » » » » o — f/3/ /i/ + /o/ )

Suppose we were to simply say that /i/ must be ignored because it doesn't fulfill the requirement of being [A]-specified, instead of assuming the structure in (14). This implies that we loosen the adjacency requirement by allowing discontinuous col inking as long as no target is skipped. This is more or less the way in which Goldsmith (1985) handles K /i/. It will be clear that this less constrained notion of adjacency broadens the possibility of skipping vowels in other types of harmony as well. Consider again the case of Akan. Since we have observed that the low vowel cannot be associated to [E] we might just as well say that only nonlow vowels are targets. Since the low vowel does not fulfill this condition it will be skipped, but we claim that in systems of this type low vowels are never skipped. Since the presence of a feature obviously implies the presence of the class node which bears it, we explain, given (14), why the low vowel in Akan is

86 opaque with respect to [E] - spreading. Being [A]-specified this vowel is P-bearing for [E]. If we were to assume that /a/ is completely unspecified, it is no longer clear why [E] couldn't pass it, so we will have to rule out complete unspecification in this case. The opacity of Bashkir /a/ and /ae/ with respect to [U]-spread now presents a problem, however. The [I] and/or [A] specifications of these vowels do not imply the [U]-bearing node. We thus predict that these vowels can be skipped, but they are not, as we have seen. To complicate matters even further, it may not even be relevant that K /i/ is not invited to generate the [U]-bearing node. This point is suggested by an aspect of labial harmony in Buriat. In the next section, we first deal with the Buriat case and then we will return to the opacity of low vowels in Bashkir.

4.

The skippability of Buriat /e/

Buriat has a vowel system like that of K, with one difference. Svantesson (1985, 318): "In Buriat the rounding harmony rule is more complicated, since e has been unrounded in certain positions, and has merged with u in others." The development alluded to by Svantesson has the result that short /e/ (schwa in Buriat) lacks a harmonic counterpart. Being [A]-specified, /e/ belongs to the set of targets. Since /e/ lacks a rounded counterpart, [U] should not associate to it. We predict that /e/ will be opaque if we were to assume that segments which are invited to generate the [U]-bearing node behave in the same way as segment*which have this node to begin with (because its presence is required by other features). However, Poppe (1960) makes it clear that lei is not opaque. If lei is preceded by a low rounded vowel, i.e. /ee/, subsequent /AA/-suffixes turn up with /ee/: /xeereld-ee/ "they conversed" (Poppe, 23). Presumably, the lei derives historically from *e. Suffixes with a low short vowel surface with lei after a stem ending in /ee/. Surprisingly a subset of the short /u/'s also trigger labial harmony. Presumably these go back to *e as well. We seem to have two problems then. We must explain why lei does not block RH and we must account for the fact that some /u/'s trigger harmony. Both problems indicate that short lei is not phonologically absent, but only phonetically, i.e. that the initial /u/ which cause labial harmony is in fact an underlying /e/, and also that due to labial harmony /e/'s are derived in non-initial position. To account for the absence of these /e/'s in the surface we assume the following post-lexical FCR and A-rules: (15)

a.

FCR: -.([U] a [A] A [E] A SHORT)

b. A-rules: i. Delete [A] initially ii. Delete [U] elsewhere

(Observe that, apparently, post-lexical A-rules can delete features.) A drawback of this (abstract) analysis is that we have now predict the existence of a nine vowel ATR system in which the low vowel does not block ATR-spreading, since we could assume that the FCR barring the ATR counterpart of /a/ is post-lexical and "correct" the output of the lexical derivation in the post-lexical phonology. We have no reason to assume that such a case should be allowed and we will therefore consider a way of constraining the theory. In van der Hulst & Smith (ms) we defend a view found in various forms elsewhere (e.g. Schane 1971, Archangeli 1985) viz. that the hierachically organization of vowel features is more sophisticated than what we assumed so far:

87 ( 16) CE]

> [A]

> [I]

> [U]

Now suppose that we do not treat nodes which are generated in the course of a spreading process on a par with nodes which are there to begin with, and assume that a generated node will disappear again if the intended association is not brought about. We then explain why the Buriat /e/ does not block [U]-spreading, whereas the Akan /a/ will always block [E]-spreading. Clearly, in the first case [U] is hierarchically lower than any specification of /e/. In the second case however /a/'s feature make-up implies the presence of the [E]-bearing node. There is no way then in which [E] can pass over /a/. This approach is of course still compatible with our treatment of K and Buriat /i/. (17) a.

Akan: opaque /a/ [E]—o»>x [A] o [I]—o o — [U]

b.

Buriat: skipped /e/ [E]—o o o [A]—o o o [I] o [U]—o>»»»o

f/i/ /a/ • /«•/ )

C/e/ / e W a / )

We now return to the case of Bashkir in which [U] fails to spread across low vowels. Again we proceed on the assumption that this is necessarily the case. Here a second proposal advanced in Van der Hulst and Smith (msj is of relevance, viz. that harmony is basically PARASITIC, a notion introduced in Steriade (1981). By this we mean the following. Given that features are hierarchically ranked one might suggest that the spreading of lower features is dependent on the spreading, colinking or adjacency of higher features. A tentative way of expressing this parasitic principle is the following:

(18)

PARASITIC PRINCIPLE Two segments A and B can be colinked on tier T iff their shared specifications on all higher tiers are adjacent

If we apply this to Bashkir it is necessarily the case that low vowels block labial harmony, because the segment*'to be colinked are both specified for [E] without being colinked due to the intervening low vowel which is not [E]-specified. (19)

Bashkir: [E]—o [A]—o [I]—o

opaque /a/ or /ae/ x o o o o o

[Uj—o

(/67 /aeA/e/) Observe that in the Buriat case [U] is not prevented from passing over /e/ in terms of the principle in (18) (cf. 17).

88 5. Conclusion In this paper we have studied the conditions under which neutral vowels which predictably have the default value act opaquely in vowel harmony. We considered a variety of cases, which, put together, present a rather chaotic picture in which everything seems to go. Assuming however that the distribution found is not random, showing the systematic rather than the accidental absence of conceivable other cases, we have attempted to develop a theory which predicts what we find, nothing more nothing less. The account offered in section 4 fulfills this goal. It relies on the the Colinking Condition (11), the Opacity Condition (cf. Ewen and Van der Hulst, this vol.), the feature hierarchy in (16) and the Parasitic Principle in (18). It is of course essential that all these aspects of the analysis can be motivated in terms of phenomena which are independent from the cases considered here, but limitations of space prevent us from doing so here. We refer to Van der Hulst and Smith (ms.).

Bibliography ARCHANGELI, D. 1985 Yokuts harmony: evidence phonology. LI 16/3, 335-372

for

coplanar

representation

in

Nonlinear

ARCHANGELI, D. & D. PULLEYBLANK 1986 The content and structure of phonological representations. MS. UCSD CLEMENTS, G.N. 198 1 Akan vowel harmony: a nonlinear analysis. In G.N. Clements (ed.), Harvard studies in phonology. Vol. II, 108-177. [IULC]. CLEMENTS, G.N. 1985 The geometry of phonological features. PhY 2, 225-253 EWEN, C. & H.G. VAN DER HULST 1985 Single-valued features and the nonlinear analysis of vowel harmony. In Bennis & Beukema ( 1985)(eds.), Linguistics in the Netherlands. Dordrecht: Foris, 39-48 EWEN, C. & H.G. VAN DER HULST this Single-valued features and the distinction between [-F] and [OF]. vol. GOLDSMITH, J. 1985 Vowel harmony in Khalkha Mongolian, Yaka, (1985), 253-277

Finnish and Hungarian.

PhY

2

GREGERSON, K. 1976 Tongue-root and register in Mon-Khmer. In P. Jenner et al. (eds.), Austroasiatic Studies. Part I. Oceanic Ling. Spec. Publ. no 13. Hawaii, THE University Press.

89 HALL, B. & R. HALL 1980 Nez Perce vowel harmony: an Africanist explanation and some theoretical questions. In R. Vago (1980) (ed.), Issues in vowel harmony. Amsterdam, 201-236 HATTORI, S. 1982 Vowel harmonies of the Altaic languages, Korean, and Japanese. Orientalia Academiae Scientiarum Hung. Tomus XXXVI (1-3), 207-2 14

Acta

HÜLST, H.G. VAN DER & N. Smith 1986 O n Neutral Vowels. In K.Bogers, H. van der Hülst & M . Mous (1986) (eds.), The representation of suprasegmentals in African languages. Dordrecht: Foris Publications, 233-279 HÜLST, H.G. VAN DER & N. SMITH ms. Harmony systems. POPPE, N. 1960 Buriat Grammar. Indiana university publications Uralic and Altaic vol 3. Bloomington: Indiana University.

series,

POPPE, N. 1962 Bashkir Manual. Indiana university publications Uralic and Altaic vol 36. Bloomington: Indiana University.

series,

PULLEYBLANK, D. 1985 Tonal and vocalic redundancy rules. [Paper phonologie pluri-lineaire, Lyon 1985] SAGEY, E. 1986 The representation M I T , Diss.

of

SCHANE, S. 1973 [back] and [round]. In S.

presented

at

the

features and relations in non-linear

Colloque

phonology.

Anderson and P. Kiparsky (eds.), A Festschrift

for Morris Halle. New York,

174-184.

SCHEIN, B. and D. STERIADE 1986 O n geminates. LI 17/4, 691-745. STERIADE, D. 1981 Certain parameters of metrical harmony. GLOW

lezing.

STERIADE, D. 1986 Non-underlying feature values. Lecture Phonology Workshop Wassenaar. STEWART, J. 1983 A k a n vowel harmony: the w o r d structure vowels. Studies in African linguistics 14.

conditions and

SVANTESSON, J.-O. 1985 Vowel harmony shift in Mongolian. L i n g u a 67/4, 283-329.

the

floating

Koen Jaspaert & Sjaak Kroon

The relationship between global language proficiency tests and language loss 1. Introduction This article reports on a pilot study carried out in the context of a large scale sociolinguistic research project into processes of language shift and language loss in Turkish and Italian immigrants in the Netherlands and Flanders, the Dutch speaking part of Belgium-*-. A number of methodological issues regarding the measurement of language loss will be discussed. 2. The operationalisation of language loss An important problem in especially primary language loss research is the operationalisation of the concept language loss itself. The central question in this respect is what kind of data should be collected in order to draw conclusions on the extent to which an individual has been 'suffering' language loss. In research projects such as our own, in which the use of spontaneous data is excluded on practical grounds (due to the large number of informants) , the above problem narrows down to the question of how language loss can best be tested. Language loss research needs to make use of tests that are in essence language proficiency tests: they relate to the command of morphological categories, syntactic structures, lexicon, etc. Making decisions on the basis of the results on these tests about the existence or non-existence of language loss is a matter of interpretation. Language proficiency tests, administered to Dutch people in The Netherlands would lead to conclusions about the degree of language proficiency of these informants, and not to conclusions about language loss. In the case of e.g. Italians residing in the Netherlands, the results of tests measuring proficiency in Italian are interpreted in terms of language loss. The observed variation owes its language loss interpretation to its coincidence with the social fact immigration, which caused a rupture between the forementioned group and the Italian speech community to which they adhere. The linguistic evolution of the immigrant group is no longer concurrent with the evolution in the 'home' community, as a consequence of which there will occur differences in language behaviour, potentially leading to communication problems between the immigrant and the home groups^. Language proficiency tests used in language loss research need to meet two main requirements: they have to be reliable as language proficiency tests and they have to be valid measurements of language loss. In their demand for reliability, these tests do not differ from other language proficiency tests. The implications of the validity demand, however, are specifically related to the language loss interpretations of these tests. We will discuss them in some more detail. 1 The choice of a point of reference Language proficiency tests can never be used directly as measures of language loss. Language loss is the difference between language proficiency observed at a certain point in time, and a certain degree of language proficiency that serves as a base line measure, as a point of reference. The choice of such a point of reference is decisive for the scope of thé

92 language loss interpretation that is given to language proficiency data. We have discussed this issue extensively elsewhere (Jaspaert, Kroon & Van Hout 1986), and will, therefore, not elaborate on it here. 2 Interfering factors Language proficiency tests may show a good deal of variation that is in no way relatable to language loss, even when the social conditions for the loss interpretation are fulfilled. Some of the sources of variation are typical of research in which language tests are used. Especially in combination with certain points of reference such factors can lead to misleading results. Metalinguistic knowledge and test skills, for instance, may cause an individual to be attributed a greater language proficiency than he can show in unguided language production. As far as measuring language proficiency is concerned, a certain influence of especially metalinguistic knowledge may not be unacceptable or even undesirable. After all, the existence of (positive) influence of metalinguistic knowledge on language proficiency is generally accepted (and put into practice in, for instance, grammar teaching). The relationship between meta-linguistic knowledge and language proficiency becomes a validity problem, however, when a language loss interpretation of the data is envisaged. When working with a theoretical point of reference - i.e. a point of reference that is not inferred from actual test results, but from an estimate of the former language proficiency of the informants - metalinguistic knowledge and test skills will not influence the point of reference and the proficiency measure in the same way. This unequal influence will result in a lower language loss rate for those informants having extensive metalinguistic knowledge and/or test skills. Imagine, for instance, that while using 100% proficiency as a point of reference, these factors would have a strong effect on the test results. This would have as a consequence that informants with a high level of schooling, the typical group in which metalinguistic knowledge and test skills may be assumed, would automatically be attributed a higher level of proficiency, and thus also a lower degree of language loss than less educated people, although the real language loss situation may be the inverse. An effect along this line was found in two pilot studies of our project with Turkish and Italian immigrants. In both cases the test items under investigation formed an implicational scale; the social correlates of these scales, however, showed that the informant rank order that resulted from it depended more on the ability to handle linguistic categories and tests than on degrees of attrition of language proficiency. This problem can be dealt with by choosing a point of reference relative to the degree of metalinguistic knowledge and/or test skills each informant has - e.g. by measuring these factors independently and using these measures as a weight coefficient for the point of reference measure, or by introducing in the research design a control group in which a similar effect of these factors can be expected. In controlling the effects of factors such as metalinguistic knowledge and test skills in this way, one should take care that the effect itself is not too large. Otherwise the variation in the language loss measure, resulting from a subtraction of the observed language proficiency from the point of reference, will be seriously reduced. 3 The lack of information on language-structural aspects of the loss process In order to measure the degree of language loss adequately, it is important that the language proficiency tests that are used contain the linguistic elements that are most susceptible to loss. It is, however, far from

93 evident w h i c h elements are likely loss candidates: r e s e a r c h o n primary language loss h a s scarcely p r o c e e d e d b e y o n d formulating h y p o t h e s e s on, a n d giving salient examples of structural aspects of language loss (cf. e.g. A n d e r s e n 1982, S h a r w o o d Smith 1983, Gonzo a n d Saltarelli 1983, Campbell 1980, T o s i 1984). There is a n almost total lack of reliable empirical data in this respect. A s a result language loss research runs the risk of w o r k i n g w i t h tests that c a n h a r d l y show v a r i a t i o n due to language loss. Depending o n the p o i n t of reference that is chosen, the v a r i a t i o n that does occur m a y seem to b e due to language loss a n d thus lead to a n illegitimate interpretation. The implicational scales from the p i l o t studies m e n t i o n e d earlier are striking examples in this respect. To a v o i d m i s s i n g language loss altogether and, consequently, m i s i n t e r p r e t i n g the v a r i a t i o n present, we c o n s t r u c t e d a fairly extensive test battery, including v e r y diverse tasks a n d items. A s long as structural language loss r e s e a r c h does n o t provide more insight into the w a y the language loss process proceeds, this seems to b e the safest w a y to measure the degree of language loss. The negative side of this w a y of w o r k i n g is that a considerable investment of time is a s k e d from the informants.

3. The language loss tests I n v i e w of the considerations m e n t i o n e d above w e c o n s t r u c t e d five language p r o f i c i e n c y tests to investigate language loss w i t h Italians living in The Netherlands: a c o r r e c t i o n test, a n editing test, a l e x i c o n t e s t a n d two c o m p r e h e n s i o n tests. We w i l l give a short d e s c r i p t i o n of each^. The correction test consisted of seven sentences incorporating 21 errors. B o t h interference errors a n d errors relatable to the complexity of Italian as such were included (cf. the different hypotheses f o r m u l a t e d in A n d e r s e n 1982). A n example sentence gave a n indication of w h a t h a d to b e done. The editing test c o n s i s t e d of a 189 w o r d text t a k e n from a n I t a l i a n popular magazine. I n r a n d o m places in the text 36 w o r d s from another text in the same magazine w e r e inserted. The informants w e r e a s k e d to strike out the superfluous w o r d s V The l e x i c o n test w a s a vocabulary test of 25 items. The w o r d s w e r e chosen f r o m t h e I B M w o r d frequency list of Italian (Bartolini, Tagliavini & Zampolli 1971). The items were randomly selected from all w o r d s w i t h a frequency of 3 (the lowest frequency class that is i n c o r p o r a t e d in the list), that w e r e n o t inflected or d e c l i n e d forms of more frequent words. The informants w e r e a s k e d to either translate or describe the w o r d s under investigation. C o m p r e h e n s i o n 1 c o n s i s t e d o f 16 s e n t e n c e s , e a c h r e f e r r i n g to o n e photograph. The informants were a s k e d to decide for every sentence whether it was, in v i e w of the photograph, "true" or "false". I n comprehens ion 2 the informants were a s k e d to decide to w h i c h of 15 photographs a c e r t a i n sentence applied. The test c o n t a i n e d 12 items. I n b o t h c o m p r e h e n s i o n tests we tried to keep interference stemming from v i s u a l details d o w n to a minimum. In other words, the test c o n c e n t r a t e d o n a g o o d u n d e r s t a n d i n g o f lexical, morphological a n d syntactic aspects of the sentences. The c o r r e c t i o n a n d editing tests were a d m i n i s t e r e d as w r i t t e n tests, the others as oral ones. The order in w h i c h the tests are p r e s e n t e d h e r e reflects a d e c l i n i n g appeal to m e t a l i n g u i s t i c knowledge a n d test skills. Moreover, the specific test s k i l l s that w e r e n e e d e d differ from test to test. The c o r r e c t i o n a n d editing tests, for example, appealed to the ability to f i n d a n d correct linguistic errors in a w r i t t e n text, w h e r e a s i n the c o m p r e h e n s i o n tests the i n f o r m a n t s n e e d e d to b e a b l e to c o m b i n e o r a l s t i m u l i w i t h v i s u a l

94 information. The tests were administered to 30 informants of Italian origin, all living in the city of Nijmegen, by a bilingual testing assistent^. The language of instruction depended on the informants; 26 chose Italian. In the following we will use the Nijmegen data to comment upon the reliability of the tests, upon the relationships between the tests and between the tests and a number of independent variables. 4. Reliability For every test a reliability analysis was carried out. The results are shown in table 1.

correction editing lexicon comprehension 1 comprehension 2

n items

n reliable items

19 36 21 15/10 8/6

14 34 19 10/8 6/5

Cronbach's Alpha .85 .96 .89 .56/.69 .61/.67

scale mean 8.7 19.6 15.7 11.6/7..6 6.9/5..0

Table 1: Reliability data of the language loss tests A short explanation of the table: -column 1 contains the number of items that show variation; -column 2 contains the number of reliable items; an item is considered reliable if its removal leads to a decreasing Alpha; this criterium coincides more or less with an item-total correlation of .40; -Cronbach's Alpha in column 3 is a measure of reliability in an additive model: it shows the extent to which items can be considered indicators of the same characteristic; -scale mean (column 4) represents the mean score of all informants on all test items in column 1. A first analysis with respect to Comprehension 1 and 2 made clear that a number of items did not show any relationship - or even a negative onewith the rest of the test. Therefore the analysis was carried out once again, leaving out these totally unreliable items. Hence the two rows of figures in table 1 for these tests. Generally speaking table 1 shows that the tests have a fairly high reliability. Only the comprehension tests are somewhat disappointing: apparently some items measure completely different things from the overall characteristic the test measures. The intercorrelations between items that behave in a different way are very low, especially in Comprehension 1. In Comprehension 2 the fall-out is mainly due to the fact that informants score too high on the test (see the scale means in table 1). So, the rather poor reliability does not point towards a multidimensional solution. The other tests also show a rather strong unidimensionality. This means that, if intervening factors occur, they do not manifest themselves as a separate dimension. If metalinguistic proficiency, for example, plays a role, then it does not manifest itself as a dimension independent of language proficiency.

95 5. Intra-test correlations An overview of the relationships between the different tests is presented in table 2. The high correlation between Editing and Correction indicates that both tests measure the same characteristic. Lexicon and Comprehension 1, too, correction correction editing lexicon comprehension 1 comprehension 2

editing

lexicon

compr.l

compr.2

.84 *

.71 * .62 *

.60 * .57 * .48 *

.35 * .23 .45 * - .21

Table 2: Correlation coefficients between tests (Pearson's R) show a rather high correlation with Editing and Correction. These correlations are indicative of the fact that the results of the four tests contain a common core of variation. It is not very likely that this core should be interpreted as testing proficiency: the proficiencies needed in Editing and Correction may well overlap, the ones needed in Lexicon and Comprehension 1 differ thoroughly from those needed in Editing and Correction and from each other. The lack of correlation between Comprehension 1 and Comprehension 2 is surprising. These two tests were intended as just two variations on one theme both of which we wanted to test. Now it turns out that informants who score high on Editing, Correction and Comprehension 1 make as many mistakes on relevant items of Comprehension 2 as informants with low scores. However, the conclusion that Comprehension 2 is simply a poor test, may, in view of its correlation with Lexicon, be premature. The variation in the response pattern of Comprehension 2 does not seem exclusively due to a sort of perceptive skill. If that were the case, Comprehension 2 might correlate with Comprehension 1, and possibly also with Correction and Editing, but not with Lexicon. In short, the correlations point in the direction of at least two dimensions in our results that may not simply be explainable in terms of intervening factors. We will come back to this multidimensional structure in section 7. 6. Correlations between dependent and independent variables Table 3 gives and statistically describes the independent variables dicussed in this section. Table 4 presents an overview of the relationships between the test results and these variables. Apart from the remark that all independent variables are based on selfreport and self-evaluation data, we will not discuss the operationalization of these variables here. The variables level of education, proficiency in other languages and social background are seen as indicators of interfering factors such as metalinguistic knowledge and test skills, the variables length of stay, degree of contact with Italians and nationality of partner as variables that are directly connected with the interpretation of the research results as indicators of language loss. The variable proficiency in Dutch is ambiguous with respect to this division.

96 Variable social background age stay Netherlands education proficiency Dutch contact Italians stay Italy nationality partner proficiency other languages

Mean

Std Dev

Minimum

Maximum

1..69 43,.43 19..67 2..00 3..27 3..97 23..77

.76 8 .41 6 .89 1 .36 .74 .81 7 .71

1..00 17 .00 5 .00 0 .0 2 .00 2..00 0,.0

4,.00 54 .00 35..00 4..00 5..00 5..00 38..00

1,.58 .55

.50 .51

1,.00 0,.0

2,.00 1..00

Table 3: Descriptive statistics for independent variables (above line ordinal/interval variables, below line nominal variables)

correction editing lexicon compr.l compr.2 social background age stay Netherlands education proficiency Dutch contact Italians stay Italy nationality partner proficiency other languages

.17 -.29 - .49 * .72 * .08 - .01 .12

.12 - .28 -.29 .74 * .14 .12 - .05

.14 .09 - .35 * .52 * - .24 - .20 .41 *

.37 .38

.33 .26

.13 .32

*

.27 .00 .01 .53 .23 .24 -.01 .49 .43

*

* *

- .21 -.17 -.45 * .17 -.18 - .21 .22 .26 .35

Table 4: Correlation coefficients between dependent and independent variables (above line Pearson's R; below line eta's) Level of education shows high correlations with Editing and Correction, and rather high correlations with Lexicon and Comprehension 1. These correlations largely confirm our hypothesis about the diminishing importance of metalinguistic knowledge and test skills in the tests (see section 3). In this context the significant correlation between proficiency in other languages and Correction can be mentioned, although the correlations between this variable and the other tests can less easily be interpreted in this way. The significant correlation between proficiency in other languages and Comprehension 1, for example, is striking. Perhaps some form of test skill is showing here. Comprehension 1, just as Comprehension 2 and Lexicon, are tasks that resemble the tasks that language learners in guided language acquisition have to carry out. Length of stay in The Netherlands shows fairly high correlations with all tests but Comprehension 1. From the variables that were considered indicators of the interpretation of variation in terms of language loss, length of stay in The Netherlands is the only one that shows the expected effect. Contact with Italians does not show any significant correlation and nationality of partner even shows a significant correlation with Comprehension 1 contradictory to the hypothesis (and also fairly high contradictory correlations with Editing and Correction): informants with a Dutch partner do better on the tests than informants with a partner of

97 Italian origin. The variable length of stay in Italy is mainly important for Lexicon. Generally speaking level of education and length of stay in The Netherlands, two main factors in the discussion about the interpretability of test results in terms of language loss, appear to be the most important factors, with patterns of influence that correspond to our expectations. As far as the other variables are concerned, it is difficult to give a meaningful and straightforward interpretation. The opaqueness of the correlation patterns with these factors may very well have something to do with the intercorrelations between independent variables and the small number of informants we worked with. In order to obtain a better insight into the patterns of influence we factor-analysed the independent variables. This analysis resulted in two factors (explained variance 61.7%) that could be interpreted as the informant's position between two communities and the informant's social position with length of stay in The Netherlands and level of education respectively as the main variables. This factorial approach appears to be a promising tool for the detection of variation in the tests caused by other factors than language loss. The first factor is directly relatable to the presented language loss interpretation, whereas the second factor points much more towards the presence of other skills and proficiencies. The correlations of these factors with the test results corroborate very well the interpretation in terms of language loss and intervening factors we have presented so far. We will not elaborate on these analyses here. 7. Conclusions and perspectives The tests that were developed seem to be reliable instruments to measure differences in language proficiency with Italian immigrants in The Netherlands®. Most bothersome in this respect are the two comprehension tests. The correlations between the different tests showed that the characteristics that are measured by the tests, resemble each other greatly. Moreover, the correlations seemed to indicate that the test results can be reduced to a limited number of dimensions. The correlations with the independent variables, finally, indicated that length of stay in The Netherlands and level of education are the most important independent variables. The data that were gathered and the results that were obtained suggest that the tests that were developed can be used to measure language loss with Italian immigrants in The Netherlands. In section 2 we discussed the possibility that the variation measured was a reflection of variation in metalinguistic knowledge and test skills rather than variation in language proficiency. The correlations with level of education show that this possibility should be taken seriously, especially as far as Editing and Correction are concerned. However, this influence seems to depend on test skills rather than on metalinguistic knowledge. This can be inferred, among other things, from the fact that the test items that show the highest correlations with the total test results turn out to be exactly those that should not produce substantial problems from a metalinguistic point of view. Items in which, for example, a whole sentence structure has to be considered in order to be able to evaluate the (in)correctness of a word (Correction) or its superfluity (Editing) are less typical of the test as a whole than items like double articles or prepositions (Editing) or literal translations of Dutch (Correction). Since these results point towards test skills as the main intervening factor, an effort should be made to reduce the test difficulty, in a technical sense, of the editing and correction tests. This could be done by leaving out one of the two tests because their

98 intercorrelation is rather high. This would also mean a considerable reduction in the time investment asked from the informants. Test skill demands could also be reduced by shortening the editing test, by spreading its items more and by providing examples. During the administration of the tests it became clear that such adaptations and changes would have a considerable facilitating effect on the test in a technical sense. As far as the other tests are concerned skills other than language proficiency seem to play a less important role. Perhaps a reservation should be made for Comprehension 1. Before using this test as a language loss test, an explanation will have to be found for the large number of unreliable items (among eight variables that were not removed, there were still four with an item-total correlation far below .40). Contrary to those in Comprehension 2 the large number of unreliable items in Comprehension 1 cannot be explained by their high scores. Comprehension 1 also shows a rather odd correlation pattern with the independent variables. In this respect it is also remarkable that most of the unreliable items needed to be called "true". It may very well be that perceptual skills and the informant's disposition to search for visual details incongruent with the text play an important role here. In a test such as Comprehension 1 it may always be possible to find one reason or another, related to some detail in the photograph, to call "true" utterances "false". The correlations with the independent variables, especially with length of stay in The Netherlands and Italy, make it possible to interpret the variation that was found in terms of declining language proficiency under the influence of migration and the coinciding process of loss of intracultural contacts, in terms of language loss in other words. The fact that the results obtained also show some variation that needs to be attributed to other sources is not a serious problem. As has been pointed out before, this variation, provided that it stays within reasonable bounds, can be controlled by using certain points of reference. Making the point of reference dependent on the level of education would already do away with most of the unwanted variation discussed in this paper. In other words, we have strong indications that via the tests a measure (or measures) for language loss can be developed. It is important, however, not to work any longer with five different measures for what in essence is the same phenomenon, but to combine relevant information from these measures into a more limited number of dimensions. When discussing the intra-test correlations we already pointed out that this seems possible. By way of an epilogue we present an instance of such a reduction by means of a multidimensional scale analysis. This analysis results in three dimensions (R^ - .96) which show exactly the correlation pattern with the tests that was expected (see table 5).

dimension 1 dimension 2 dimension 3

correction

editing

lexicon

compr.1

compr.2

-.91 * - .04 .07

-.97 * .13 .02

-.76 * -.37 * - .09

-.61 * .38 * .13

-.34 * -.57 * - .05

Table 5: Correlation coefficients between scale dimensions and dependent variables (Pearson's R) As to the correlations with the independent variables a pattern occurs that can easily be linked with the results of the factor analysis discussed in section 6 (see table 6).

99 social backgr. - .15 .41* -.20

d 1 d 2 d 3

stay Neth.

education

prof. Dutch

cont. Ital.

stay Italy

.24 .33* - .19 .27 - .43* - .08

-.74* .22 .17

- .09 .43* .07

-.03 .33* .15

- .03 .32 - .45* .33 -.39* .36

age

nat. prof. part. lang. .30 .19 .08

Table 6: Correlation coefficients between scale dimensions and independent variables (to the left of the line Pearson's R; to the right eta's) These results loss measures remains that, data presented

also strenghten our confidence in the validity as language of the tests that were constructed. Of course, the fact in order to reach a definite conclusion in this matter, the here will have to be tied to an adequate point of reference.

Notes 1.

The project is being carried out at the Department of Language and Literature of the University of Brabant. It is financed by the Netherlands Organisation for the Advancement of Pure Research (ZWO), project number 300-165-020, the University of Brabant and the Belgian Fund for Fundamental Collective Research (FKFO). An extensive description of the project can be found in Jaspaert & Kroon (1986).

2.

Incidentally, the above view on language loss as a form of language change occurring in a certain social context that serves as a necessary as well as a sufficient condition for the loss interpretation, is not adhered to by all researchers in the field (cf. e.g. the various contributions in Weltens, De Bot & Van Els 1986).

3.

The tests were constructed in cooperation with S. van Volsem, Department of Romance Philology, University of Leuven.

4.

Editing tests turn out to be at least as reliable and valid as the more popular cloze-tests (cfr. e.g. Mullen 1979). Editing scores relate very well to the acceptability scores on cloze-tests, but they are much easier to compute, which in view of the large sample in our project (800 informants) is a very serious advantage.

5.

Names and addressess of potential informants were provided by Kees de Bot, Institute of Applied Linguistics, University of Nijmegen; the data were gathered by Catia Cucchiarini.

6.

The prudent tone of this statement is due to the limitations of our sample.

References ANDERSEN, R. 1983 Determining the linguistics attributes of language attrition, in R. Lambert & B. Freed (eds) The Loss of Language Skills. Rowley, Mass., Newbury House, p. 83-188. BARTOLINI, U., C. TAGLIAVINI & A. ZAMPOLLI 1971 Lessico di frequenza della lingua Italiana contemporania. Pisa, IBM Italia. CAMPBELL, C. 1980 Second generation Australian Italian in Giru, North Queensland, in ITL Review of Applied Linguistics. 49-50, p. 5-22.

100 GONZO, S. & M. SALTARELLI 1983 Pidginization and linguistic change in emigrant languages, in R. Andersen (ed) Pidginization and Creolization as Language Acquisition. Rowley, Mass., Newbury House, p. 131-197. JASPAERT, K. & S. KROON 1986 Onderzoek naar taalverschuiving en heidsgroepen: theoretisch kader en J. Creten, G. Geerts & K. Jaspaert opname van de sociolinguistiek in Amersfoort, ACCO, p. 197-206.

taalverlies bij etnische minderoperationaliseringsproblemen, in (eds) Werk-in-uitvoering. MomentNederland en Vlaanderen, Leuven/

JASPAERT, K. , S. KROON & R. VAN HOUT 1986 Points of reference in first-language loss research, in Weltens, De Bot & Van Els (eds) 1986, p. 37-49. MULLEN, K. 1979 An alternative to the cloze-test, in C. Yorio, K. Perkins & J. Schachter (eds) On TESOL '79. The Learner in Focus. Washington D.C. SHARWOOD SMITH, M. 1983 On first language loss in the second language acquirer, in S. Gass & L. Selinker (eds) Language Transfer in Language Learning. Rowley, Mass., Newbury House, p. 222-231. TOSI, A. 1984 Immigration and Bilingual Education. Oxford, Pergamon. WELTENS, B., K. DE BOT & T. VAN ELS 1986 Language Attrition in Progress. Dordrecht, Foris.

Bob de Jonge

Estar comes of age 1. Introduction In this paper I will discuss the innovative use of estar vs. ser (both verbs translate, in different contexts, as the English verb 'to be 1 ) in the spoken Spanish of Mexico (cf. Silva-Corvalán (1986)) and Venezuela. A first reading of El habla popular de México, Lope Blanch (1971) and El habla culta de Caracas, Rosenblat (1979J1 reveals that the innovative use of estar is far more frequent with expressions of age than with other expressions. The utterances we have in mind are of the following type: (1) He/she is young/old In peninsular Spanish, the use of estar with expressions of as emerges from the Sevilla Culto corpus, where ser is used thus appears that at least in Mexico and Venezuela a change the use of estar vs. ser; it is this process of change that on.

o age is hardly knownt exclusively.^ It is taking place in we will be focusing

2. The process of diffusion Table 1 shows the relation between use of ser and estar with expressions of age in four different corpora.4 Table 1: Use of ser and estar with expressions of age Sevilla Culto México Culto México Popular Caracas Culto

Total N

% ser

% estar

20 40 91 59

100% 55% 41% 32%

0% 45% 59% 68%

The extension that Silva-Corvalan (1986:606-7) observed in the use of estar vs. ser in the two Mexican corpora can also be seen here in Caracas Culto. The fact that estar expands on ser territory is not surprising. For example, Pountain (1982), to whom Silva-Corvalan also refers, describes the historical expansion of estar territory. Pountain claims that, owing to semantic weakening, estar could succesively take over several functions originally fulfilled by ser• Interestingly enough, the preference for estar rather than ser is not equally intense in all corpora, although these are practically contemporaneous. The quantitative increase in the use of estar could reflect a qualitative phenomenon, namely the diffusion of estar into different contexts. In short, the rank-order of corpora shown in table 1 may be seen as reflecting different stages in a process of change. We shall examine whether this is the case and what the change is that these corpora illustrate. In the first place we will discuss a number of examples from the first two corpora that show the innovative use of estar, illustrating this use with expressions of age. This will allow us to identify special features present with

102 estar, but absent from the comparable examples containing ser. We will then propose a hypothesis as to how the use of estar became current with expressions of age, and try to support this by means of a quantitative analysis of the entire corpus. We will check, finally, whether our hypothesis on the expansion of estar is qualitatively confirmed by the use of this verb as observed in the corpus that presumably represents the most advanced stage in the development, namely Caracas. We begin by analysing an example from the México Culto corpus, illustrating the earliest stage of the innovative use of estar: (2) (p.316) Enc.-[...] es muy fino su... A.-SI. Ene.-... su teatro. B.-Bueno, sí; pero se me hace que ella ya está grande para determinados papeles, como que no le quedan. 'Int.-[...] it's very special, her... A.-Yes. Int.-... her theatre. B.-Well, yes; but it appears to me that she is [estar] already old for certain parts, as if they are not right for her.1 Let us compare this example, which is characteristic of the use of estar in this corpus, with an example of ser: (3) (p.441) [Inf.] Todos jóvenes en aquella época, ahora todos son grandes y son... los... el recuerdo grandísimo y agradable de aquella época, y los hijos de ellos también han estado. 'All of them [the founders of the boy scouts club, BJ] young in those days, now they are [ser] all old and are... the... the memory of those days is great and nice, and their children were also members.' We note a number of similarities: the same adjective grande 'big, old' is used, and both verbs are in the present tense. Furthermore, both cases contain a reference to the past: in (2), the actress is too old for certain parts for which she was not too old at an earlier time, and in (3), the persons referred to were first young, and are now old. At first sight both contexts might appear identical, and yet estar is used in the first example, while ser appears in the second one. If we are to understand the preference for estar we must first ascertain what precisely the difference is between the two examples, and secondly, we must relate this difference to the different values of the verbs ser and estar, such as they emerge from their uses in contexts other than those of age expressions. We will begin with the second point, namely the systematic difference in values between ser and estar• We can take it as given (cf. Falk (1979)5) that where ser is used, the attribute is understood to describe a norm, while with estar the attribute signals a deviation from the norm. A norm is by definition, if not entirely static, at least something that is not explicitly dynamic. A deviation from the norm, however, is by definition not static, since it presupposes a process or a change. If we now return to the two examples quoted above, we perceive a significant difference between them. (3) gives a description of the age of certain persons, which is compared with a description of an earlier age. It is not the process of living or ageing that is emphasized here: all we have is a statement that the age of the persons was once 'young', and is now 'old', just as all persons of that age used to be young and are now old. In (2), on the other hand, a norm is presupposed which is relevant to the topic

103 under discussion: in order to play certain parts, actresses must be of a certain age. Informant B thinks that the actress referred to is too old for these parts, and therefore deviates from a relevant norm. The process of ageing is of great importance in this case®, and this makes the use of estar perfectly appropriate. On the basis of these and similar examples, it may be concluded that the México Culto corpus reveals an innovation, namely the use of estar (instead of ser) whenever the age of a person deviates from the norm in some way relevant to the context. In the second corpus, México Popular, we notice a considerable expansion of the use of estar with expressions of age. Many of these examples are as in (2), but a fair number are quite different: (4) (p.293) Enc.-Y usted qué estudió? B.-Pos yo estudié... ora sí que andar picándole la cola a los bueyes en mi tierra! [Risas]. Pues sí... pus yo estudié, porque le voy a decir que tuve estudio. Yo tuve estudio, cuando estuve ya grande. Cuando fui chico, fui un cabezón, pero bueno! 'Int.-And did you study? B.-Well I studied... well, to prick the rear of oxes in the place where I come from! [Laughter]. Yeah... well I studied, because I'm going to tell you that I went to school. I went to school when I was [estar] already big. When I was [ser] small, I was headstrong, but allright! ' This example is much like (3): two points in the life of a person are contrasted by means of two age expressions. Yet estar is used in one of the two cases, while in (3) ser was used in a similar situation. Note that in (4) both verbs are in the past tense. In (4) the situation is as follows: the informant did not study in his childhood. He was rather troublesome, and that apparently interfered with his going to school. Later, as he grew older, he also grew less headstrong, and underwent some schooling. Here, the expression 'when I was small' only describes the person at a certain point in time. The expression 'when I was already big', however, besides describing the person, also alludes to a change in personality. Of the two age expressions, then, the latter is obviously the one more likely to go with estar: it is clear that 'being big' involved a process of change for the person, a deviation from how he used to be, small and headstrong. It is worth noting that while in (2) the norm deviated from is a generally accepted one, in the particular case of (4) the norm is purely individual. 7 From the discussion of these examples we can draw a number of conclusions which are of importance for a quantitative analysis of the phenomenon. We begin with a basic twofold classification of the purpose served by age expressions, relevant to the innovative use of estar• 'Age' can be considered from two points of view: (a) Age as the result (or beginning) of a process: age is not dissociated from the process of living, which takes place between birth and death, and consists of constant change. From this perspective, an age expression indicates the stage reached in the process of life. But the indication of this stage suggests, in turn, a retrospective view of the change that life presupposes. This we will call the chronological view of age. Time adverbs like ya 'already' and todavía 'still' are contextual indicators of this view of age. (b) Age as a characterisation of individuals: an expression of age can be regarded as a snapshot view of a person, a characterisation of a person at a particular moment. In this case, the process gone through by the

104 individual before arriving at this condition is irrelevant. This we will call the momentary view of age. Time adverbs like ahora 'now' and cuando 'when' are contextual indicators for this view of age. There is, however, a second factor that is important for age, and therefore of probable relevance to the view taken of age: this is time itself, the very substance of life and age. Time can basically be divided into two parts: time of which we have, vs. time of which we have no experience. Times of which we have experience are the present and the past; we will limit ourselves to this segment of time, since all the expressions of age found in our corpora refer to the present or to the past. The present age of a person referred to is clearer and more vivid than the age of a person in the past, because the speaker lives in the present at the moment of speech. The difference between the two views of age, i.e., whether age is viewed in chronological or momentary fashion, is therefore also clearer in the present than in the past. The past, in comparison, rather blurs the difference between the two views. The reason for this is simple and lies in the nature of past tense age expressions: these refer to an age that is necessarily different from the actual (i.e., present) age of the person in question. For this reason, any reference to a past age, whether chronological or momentary, per se implies a (subsequent) change, i.e., the one resulting in the current age. As a consequence, a momentary view in the past resembles a chronological view in the retrospective vision that is necessarily implied by the past in relation to the present. The lesser discriminability between chronological and momentary views in the past (vs. present time), then, requires us to take time itself into account, and consequently to distinguish four contextual situations in which estar could theoretically be used: (I) (II) (III) (IV)

a a a a

chronological view of present age chronological view of past age momentary view of past age momentary view of present age

These four possibilities are graphically represented in figure 1: Figure 1: Situations of age expressions momentary view

chronological view ^ ^

present

/

X

past

\

11

IV

\

III

/

Let us now see the relevance of this classification for the (innovative) use of estar, bearing in mind that the use of estar presupposes a deviation from the norm. Example (2), where estar was used, involved the age actresses must be in order to play certain parts. Since this deviation from the norm is the result of the process of living, what we have in this example is, clearly, a chronological view of the actress' age. Accordingly, since this example constitutes an

105 'early' example of estar, the very first usage of estar should occur i) in age expressions which incorporate a chronological view ii) in those contexts where the chronological character of the age reference is most obvious to the speaker Since the view taken of age is generally more obvious to the speaker in the present, the first use of estar is expected in present contexts for chronological views (sector I), i.e., where 'change' is clearest. For exactly the same reason, momentary age references in the present are the ones least likely to receive estar -or likely to receive estar last, i.e., the contexts in which ser should have the longest life expectancy (sector IV). The transition from the use of estar in sector I to its use in II is not a difficult one: the past does not contradict chronology, since of itself it presupposes a retrospective vision. We accordingly expect that from contexts in the present estar will be extended to contexts in the past, but only for chronological age expressions. If the use of estar expands even further®, the fundamental hurdle between chronology and momentariness must be taken. We have argued that the difference between chronological and momentary views of age is more easily blurred in the past, inasmuch as, even with a momentary view of age, the necessarily different (current) age of the person in question implies a change. Since an implication of change favours the use of estar (cf. Falk (1979))past time references will facilitate the use of estar with momentary age expressions, and thus allow the progression from II to III. We should then get the mirror image of the first appearance of estar with chronological age expressions: first in contexts where momentariness is less obvious to the speaker, i.e., in the past (sector III), and, at the very end, in those contexts where it is more obvious to the speaker, i.e., in the present (sector IV). If estar ever comes to be used with any and all age expressions, i.e., even with those that render a momentary view in the present, a new view of age must necessarily be involved, namely one that is compatible with and matches the semantic value of estar. In short, if the use of estar ever becomes pervasive with age expressions, what should prove the primary consideration in the choice of that verb is the fact that age, in itself, is a constantly changing phenomenon. And indeed, if age is not seen as corresponding to a norm, but as standing, by definition, for deviation, it would be very natural for all age expressions to go with estar•

3. Testing the predictions If the different corpora investigated indeed represent different stages in the same process of change, as we have suggested above, the picture we have just drawn of the progressive intrusion of estar in the general field of age expressions should be reflected in our material. We must consequently seek for an objective correlate of the progression of estar through the four sectors defined by the intersection of time reference and view of age. Past and present reference have obvious correlates in the texts that can be measured: the tenses of the finite verb. We accordingly begin by comparing the proportion of estar in the present tense with that of estar in the past.® Our expectation is as follows: since the first uses of estar are expected with chronological age expressions, and then omly in the present, the corpora that represent the initial phase of development will show a clear preference for estar in the present tense. This preference will, however, not rise to a full 100%, because we expect use of estar with chronological expressions only, and not with momentary age ex-

106 pressions in the present. Next we expect an expansion to the chronological age expressions in the past, i.e., into sector II. This should be reflected (in the 'later' corpora) as a rising percentage of estar in the past tense. If the use of estar expands even more, we expect it to take place in sector III, with momentary age expressions in the past, which means that all age expressions in the past -both chronological and momentary- can show use of estar. If we then find that a given corpus shows a marked preference for estar in the past tense, this would mean that the third stage in the development has been reached. The last stage, which amounts to the reinterpretation of the concept of age itself, should be reflected, finally, in the use of estar with momentary age expressions in the present (sector IV). As a result, an overall preference for estar is expected in expressions of age generally, and the use of estar in both past and ultimately present should approach 100%. We begin by measuring the proportion of estar (vs. ser) in the present tense vs. the past tense. The results appear in table and figure 2. Table 2: estar in present and past tense % estar/Total N (estar + ser)

MC

MP

CC

past tense present tense

29%/17 57%/23

69%/26 55%/65

78%/ 32 56%/27

Figure 2: estar in present and past tense 100%|

= estar present tense | - estar past tense

estar

0%-

MP

I i

CC

The corpus that shows the smallest use of estar and thus, presumably, represents the earliest stage in the development, MC, indeed shows a considerably higher percentage of estar in the present than in the past tense, thus confirming our expectation that estar first enters chronological age expressions in the present. However, since even in MC we already find estar in past contexts, this corpus cannot represent the very earliest stage, but must reflect a somewhat advanced early stage in the development. Note now that the percentage of estar in the present is approximately the same in all corpora: this may be taken as a reflection of the fact that only part of these expressions, i.e., the chronological ones, occur with estar, while others -presumably the momentary age expressions- do not. If this is the case, the proportion of chronological to momentary age expressions in the present would lie around 56:44%. Now, since as far as age itself is concerned, the past does not per se impose a particular view different from the one taken for the present^, we have no reason to assume that the general proportion of chronological to momentary age expressions will be drastically different in the two tenses. We may accordingly expect that about 56% of all age expressions in the past are of a chronological nature. Should we then find that the past tense expressions re-

107 veal a percentage of estar significantly higher than 56% (as we in fact do in Caracas Culto) we may assume that the increase is due to the fact that in past contexts the use of estar has expanded from chronological to momentary age expressions . If the due to tively by the

increase in percentage of estar in past tense contexts noted for CC is the use of estar with momentary age expressions, this should be qualitaobservable in the actual examples. That this is indeed the case, is shown following representative example from Caracas Culto:

(5) (p.447) B.-...SÍ, yo me ponía a pensar el otro día, conversando con una ... amiga mía, que a nosotros nos debían llamar "La generación del silencio" . A.-¿Por qué? B.-Porque... cuando yo estaba chiquita, no nos dejaban hablar, como se usa ahora, hablar en la mesa; [...]. 'B.-...yes, when X was talking the other day to a friend of mine, I started to think that they should call us "The generation of silence". A.-Why? B.-Because.. . when I was [ estar] small, we were not allowed to speak, as is usual nowadays, to speak at the table; [...].' In the other corpora, this kind of use occurred exclusively with ser since no chronological view is given; the message is purely momentary: not being allowed to speak at the table is put into a temporal setting by the expression cuando yo estaba chiquita 'when I was small'. Being allowed to speak at the table nowadays is not the result of growing older, but the result of changed opinions on what children are and are not allowed to do. It is clear that this age expression falls within sector III. The relation that the past as such has with the present, and the retrospective vision that it therefore implies is what makes it possible for innovative estar to penetrate in contexts that give a momentary view of age: and indeed, the presupposed connection with the present is explicitly made by the phrase como se usa ahora 'as is usual nowadays'. The entry of estar in contexts of sector III implies that estar is no longer used to convey an exclusively chronological view of age: estar is now triggered by what is common to both chronological and momentary age expressions, especially momentary ones in the past: both views imply a certain (though not the same) relation with time. Now, this merely temporal aspect of age expressions can also be measured independantly. As can be seen in the examples, and as our discussion of the different views on age makes clear, the temporal aspect of age expressions (whether momentary or chronological) is often overtly expressed by time adverbs, and the cooccurrence of time adverbs with ser and estar in age expressions can certainly be measured in our corpora. Since from the very beginning estar is used with age expressions involving a temporal aspect, no drastic change is expected in the use of time adverbs with estar• But in the later stages, once estar has been extended from sector II to III, we expect ser to be confined to those age expressions where the temporal aspect -namely a momentary view in the present- is least obvious. This restriction in the distribution of ser should be reflected in a decline in the use of time adverbs with ser• Moreover, this decline should occur precisely in Caracas Culto, the corpus where, according to our hypothesis, estar has presumably invaded sector III and marks the general temporal aspect of age, whether momentary or chronological. To test whether this is so we have counted the presence vs. absence of time adverbs with age expressions in all corpora. The results appear in table 3.

108 Table 3: presence of time adverbs with ser and estar % + time adverb/Total N (+ adverb)

ser

estar

México Culto México Popular Caracas Culto

55%/22 41%/37 21%/19

83%/18 59%/54 68%/40

p | C a r d (PflQ)=2j. This implies that the numeral modifiers in (1) can only have an intersective reading. However, according to Szabolcsi(1983) even in this triple there is a subtle semantic difference caused by the fact that the head-noun is focussed in (lb,c). The sentence in (la) means that Mary saw two bikes. The sentence may be still true in case Mary saw other things like two cars, one plane, etc. In sentence (lb) Mary saw two things that were bikes. Note that in this case the sentence is false w h e n she saw two things not having the property bike such as two cars, two planes, etc. Of course, she may have seen one car, three planes, etc. The meaning of the sentence (lc) differs from (la,b) in that everything except bikes are barred from the universe. The comma indicates that occasionally there happened to be two bikes. Again, we derive the conclusion that the variants in (1) may have different truth values. Summarizing, in this section we examined the semantics of m o d i f i c a t i o n in H u n garian. It turned out that the triples in (1), (2) respectively have different truth values. The triples have in common that modification in all three cases expresses conjunction representable in a set expression. These results will play an important role in our analysis below.

3.

Derivational versus

Representational

In section (1) we pointed out that intuitively the variants in (1), (2) respectively are related syntactically and semantically. Therefore it seems justified to connect them by means of a single grammatical operation. Such an operation is provided b o t h by a derivational and a representational approach to the grammar. Let us first discuss the derivational analysis to the triples in (1), (2) respectively. In Horvath (1986:29,83) a derivational analysis based on move alpha is proposed, that is, the a-sentences in (1), and (2) are taken as the underlying structures for their counterparts in (b) and (c) respectively. The (b)- and (c)-sentences are derived by applying resp. Quantifier Float and Topicalization, leaving a trace in the modifier position of the NP. The advantage of this approach is that the derivation of split constituent constructions in Hungarian falls under the universal scheme move alpha. However, a number of problems arise w i t h i n a derivational grammar: (A) The derivational analysis violates the requirement that stems are b a s e generated in their fully inflected forms, the so-called Lexical Integrity Hypothesis. Firstly, it is unclear where the case marker on the modifier in split variant comes from. Secondly, the same is valid for the plural marker on the head-noun in (4d). Recall that in the underlying structure only singular h e a d nouns are allowed in case the modifier is in construction w i t h a m o d i f i e r expressing quantity. Thirdly, the derivational analysis allows the creation of new lexical predicative stems after scrambling the attributive modifier out of its NP (cf. ket/kettB, kis/kicsi in (1), (6) respectively). (B) Horvath (1986) refers to the derivation of the split constituent construction as Quantifier Float. This term suggests, however, that a generalization is missed. For we saw that not only numerals, or quantifiers but also adjectives may appear in the split variant. The question is rather w h y only these m o d i fiers m a y be scrambled out of their NPs? A further question concerning the modifiers which is not solved satisfactorily by assuming Quantifier Float to operate here is: why must they be bare?

145 (C) It is a well-known fact that in languages possessing floating quantifiers (cf. Japanese) subject-object asymmetries turn up. As we noted in section (1) this is not the case in Hungarian.3 (D) It remains unclear under a derivational analysis why the split variant may have an intersective reading only. For the NP configuration which is related in case of a single expression to a syncategorematic reading remains unaffected because of the trace in the modifier position. As a starting point for a representational analysis of the split constituent phenomenon in Hungarian we take Hale's (1978) rule of merger: (9)

semantic expressions bearing identical categorial signatures [N, Number, Case] may be merged

Rule (9) was originally proposed to handle split constituents in Warlpiri. Hale remarks in his (1983:39) paper: "perhaps an adequate general theory of grammar will define some principles which forces the merger of discontinuous expressions thereby rendering superfluous any language specific rule of merger." We believe that such a theory, or at least the shape of it, is available now. Below we will argue that "merging" in Hungarian follows from Higginbotham's (1985) theory of thematic discharge. For ease of perception let us first introduce some principles of Higginbotham (1985, 1986). Higginbotham (1985:560) proposes the following redefinition of the theta-criterion: (10a) (10b)

every argument is assigned one and only one thematic role every thematic position is discharged

In Higginbotham (1985, 1986) the following types of thematic discharge are distinguished, among others: (11)

theta-identification, exemplified in simple adjectival modification as in "white wall" interpreted as white(x) and wall(x) autonymous theta-marking, where the value assigned to the open position in the theta-marker is the attribute given by its sister constituent

Consider an example of each. The semantics of the phrase "white wall" is expressed by a simple conjunction: a white wall is a thing that is white and a wall. Bare nominals are open constructions. This is supported by the fact that nominals can serve as predicates in many languages. Adjectives must have an open position as well since they may occur as a predicate. Hence, we may assign nominals and adjectives the following thematic grids as part of their lexical entries: (12a) (12b)

nominal, -V +N, adjective, +V +N,

In a phrase "white wall" some position in the adjective is identified with the nominal position. The thematic structure of "white wall" may be represented in the following diagram: (13)

(N\) (A,) I

The open tions in line. We and then

(N,) 1

position of the adjective is discharged by identifying the open posithe adjective and the head-noun. This is expressed by the connecting can compare its structure to that of building up a compound Fx & Gx identifying x and y.

146 In section (2) we noticed that in the paraphrase of (8a) the head-noun is an argument of the adjective, so that the head-noun in a syncategorematic adjective-noun construction serves to discharge two thematic positions, one by identification and the other by theta-marking, by the adjective, of the very noun itself. This latter mode of thematic discharge is called autonymous theta-marking indicated by an arrow in the diagram: (14)

(N' ,) (A, ) big (

(N,) J butterfly

Let us examine the representations of the Hungarian examples (1,2) starting with (la,2a). In section (2) we noticed that numeral modifiers always have an intersective reading. So, (la) can be analysed along the lines of (13) with the mode of theta-identification. In section (2) we found that in (2a) we have a case of syncategorematic modification. Analogously to (8a), we may handle this sentence by the combination of theta-identification and autonymous theta-marking as in (14). In section (2) we pointed out that the semantics of (lb,2b) is characterized by simple conjunction. From this it follows that the thematic positions in the parts of the split constituent must be identified. As noted above, Higginbotham assumes that theta-identification is restricted by government. It is easy to verify that the parts of split constituents are not in a government configuration. Therefore, we propose to add identity of morphological features (i.e. case, number, etc.) as a licensing condition on thematic discharge by theta-identification. Notice that merging follows now as a concomitant of the theory of thematic discharge. Summarizing, in a representational analysis the properties of split constituent constructions (lb,2b) are derived as follows: (a) Free base-generation accounts for the order of constituents. Note that by doing so the Lexical Integrity Hypothesis is respected and the absence of subject-object asymmetries is predicted; (b) The Projection Principle determines the argument status of the constituents. For example, the head-noun may be taken as the direct argument of the verb forcing the predicative modifier to receive the function of adjunct (cf. Simpson 1983); (c) Nominals and adjectives have an open position in their theta-grid w h i c h m u s t be discharged. This explains then why only certain type of modifiers (adjectives, numerals, some quantifiers) m a y enter the split constituent construction, i.e. exactly those which may function as predicative nominals, and thus may be open structures. The fact that the parts of split constituents must be open structures in order to undergo theta-identification also provides an explanation for the observation that they must be bare.'1 Modification by a definite or indefinite determiner would close the structure, i.e. eliminate its thematic role from its theta grid, by the mode of theta-binding (cf. Higginbotham 1985,1986). Hence, theta-identification would be blocked, (d) Identity of morphological features serves as a condition on theta-identification. This depends on the different functions morphological markers may have in a language. In Hungarian case acts as an attribute relater in split constituents, that is, case indicates that the adjunct is predicated of the head-noun. Note that the optional concord of plural features does not form an obstacle under the representational analysis, (e) The differences in meaning between the single expression and the split constituent construction in (1), (2) respectively are derived in the representational analysis by applying different modes of thematic discharge. These primitive semantic operations are induced by different structural and morphological contexts. If this analysis of noun modification in Hungarian is correct, we make the

147 following predictions about the possibility of "splitting" single n o u n phrase expressions. We w o u l d expect that in case the semantics of modifier-noun comb i n a t i o n cannot be captured by a simple conjunction, or to say the same thing otherwise, if it is not possible to take the combination of adjective and n o u n as having as its denotation the intersection of the set denoted b y the adjective w i t h the set denoted by the noun, the split constituent construction will not be possible. This arises at least in the following two cases: (1) m o d i f i cation is interpreted as a combination of theta-identification and autonymous theta-marking (cf. 8a); and (2) modification is interpreted as a case of autonymous theta-marking only. Examples of the latter are phrases like "former president" and "alleged murder". Adjectives such as "former" and "alleged" cannot have as denotation the intersection of any such sets: e.g. "former president" cannot be analysed as the intersection of the set of presidents with the set of "things that are former". It is easy to see that the latter phrase is meaningless. Compare: (15a)

Mari nagy bolhat latott Mary-NOM big flea-ACC saw-3sg "Mary saw a big flea" (15b)* Mari bolhat latott nagyot Mary-NOM flea-ACC saw-3sg big-ACC (16a)

Mari latta az elBzH elnokot Mary-NOM saw-3sg the former president-ACC "Mary saw the former president" (16b)* Mari elnokot latott eltfzBt Mary-NOM president-ACC saw-3sg former-ACC (17a)

Mari latta az allitolagos gyilkost Mary-NOM saw-3sg the alleged murder-ACC "Mary saw the alleged murder" (17b)* Mari gyilkost latott allitolagosat Mary-NOM murder-ACC saw-3sg alleged-ACC Sentence (15a) means: Mary saw a thing that was a flea and it was big for a flea. This reading is analysed as a combination of theta-identification and theta-marking. In sentence (15b) the modifier is separated from its head-noun. The semantics of such constructions is captured by simple conjunction. Therefore, the sentence should mean: Mary saw a thing that was a flea and it was big. However, this is not in coherence w i t h the properties of fleas. E v e n big fleas are not big creatures. Hence, (15b) is ruled out. In sentence (16a) the phrase "former president" designates a person whose presidency is former. This is not a case of intersective modification (cf. Higginbotham 1985 for suggestions). Hence, as (16b) shows, it is impossible to have the split variant. In (17a) we have a case of autonymous theta-marking (cf. Higginbotham 1985). A n alleged murder is true of things alleged to be a murder. Again, it is not possible to form a split constituent construction (cf. 17b). For this presupposes an intersective reading of the modifier-noun combination.

4.

Concluding Remarks

In this paper I have discussed some aspects of the syntax and semantics of n o u n modification in Hungarian. Modifier-noun combinations have two patterns in Hungarian: a single and a discontinuous expression. It has b e e n argued in sections (1) and (2) that these variants have different semantic and syntactic properties. This makes a derivational analysis of these pairs hard to defend.

148 For this reason, and to satisfy our intuition that the single and discontinuous variant of modifier n o u n combinations are related grammatically we explored a representational analysis based on Higginbotham's (1985) theory of thematic discharge and the special role morphological markers play in the grammar of Hungarian.

Notes I am extremely grateful to A n n a Szabolcsi for split constituents in Maracz (1987) providing favour of the position taken in this account. Hoekstra, Istvan Kenesei and Sjaak de Mey for

her comments o n the section o n some crucial suggestions in I a m also indebted to Eric discussion.

1. There is some uncertainty among native-speakers whether the singular count noun bicikli "bike" is in focus position, in the preverbal modifier position, or may be in both positions in sentences such as (lb,c). In this paper I will follow Szabolcsi (1983) who presents only examples in w h i c h the h e a d - n o u n is focussed. 2. In case to a non-numeral modifier a plural marker is attached in the split constituent construction the head-noun must be plural too: (i)*

(ii)

biciklit lattam nagyokat bike-SING-ACC saw-lsg big-PLUR-ACC "I saw bikes such that they were big" bicikliket lattam nagyokat bike-PLUR-ACC saw-lsg big-PLUR-ACC

From the comparison b e t w e e n (4c,d) vs. (i,ii) we observe that in case of a non-numeral modifier there must be full morphological concord between the parts of a split constituent. The reason for this difference is that numerals are morphologically singular (cf. *kett8k "two-PLUR") in such constructions but are lexically specified for plural (except egy "one"). 3. To the list of syntactic problems concerning the derivational analysis of split constituent construction "Across-the-Board" coordination of split NPs may be added (cf. Szabolcsi 1983:fn.ll): (i)

biciklit es autot lattam kettBt bike-ACC and car-ACC saw-lsg two-ACC

4. There is an interesting exception to this restriction mentioned in Szabolcsi (1985:48). Szabolcsi notes that in case the separated modifier is in the superlative it might be modified by a determiner: (i)

zold loval itt talalkoztam a legszebbel green horse-INSTR here met-lsg the prettiest-INSTR "I m e t a prettier green horse here than anywhere else" *"As for green horses, it was here that I m e t the prettiest of them, i.e. the prettiest green horse that there is"

As we see from the glosses a superlative adjective modified by a determiner may only be separated from the head-noun on the comparative reading. The sentence is ungrammatical on the absolute reading of the superlative. Szabolcsi claims that the superlative phrase in the absolute reading is in the specifier position of the NP, whereas the superlative phrase is NP internal in case of the comparative reading. Her conclusion fits in nicely with our result. For in the absolute reading the NP w o u l d be closed by a binder in the SPEC position. In the comparative reading, although there is a determiner present, the NP

149 counts as an o p e n structure having an undischarged thematic position. Hence, the superlative phrase m a y be available for the split constituent construction only on the comparative reading.

References BOUMA, G 1986 Kategoriale Graimnatika e n het Warlpiri, in GLOT 8, p. 227-256 HALE, K. 1981 O n the Position of Warlpiri in a Typology of the Base, Bloomington Idiana University Linguistics Club HALE, K. 1983 Warlpiri and the Grammar of Non-Configurational Languages, in Natural Language and Linguistic Theory 1, p. 5-47 HIGGINBOTHAM, J. 1985 O n Semantics, in Linguistic Inquiry 16, p. 547-593 HIGGINBOTHAM, J. 1986 Elucidations of Meaning, Ms, MIT HORVATH, J. 1986 Focus in the Theory of Grammar and the Syntax of Hungarian, Dordrecht, Foris KAMP, J. 1975 Two Theories about Adjectives, in E. Keenan (ed) Formal Semantics of Natural Language, Dniversity of Cambridge Press, p. 123-155 MARACZ, L.K. 1987 O n the Status of the Projection Principle in Hungarian Grammar, in I. Kenesei (ed) Approaches to Hungarian II, University Press Szeged NASH, D. 1980 Topics in Warlpiri Grammar, MIT Dissertation QUINE, W.V.A. 1960 Word and Object, Cambridge, MIT Press REULAND, E. 1987 O n the Syntax and Semantics of Existential Sentences, in Proceedings of the West Coast Conference on Formal Linguistics V RIEMSDIJK, H. V a n 1982 Locality Principles in Syntax and Phonology, in The Linguistic Society of Korea (ed) Linguistics in the Morning Calm, Seoul, Hanshin, p. 693-708 SIEGEL, M. 1976 Capturing the Russian Adjective, in B. Partee (ed) Montague Grammar, New York, Academic Press, p. 293-309 SIMPSON, J. 1983 Aspects of Warlpiri Morphology and Syntax, MIT Dissertation SZABOLCSI, A. 1983 Focussing Properties, or the Trap of First Order, in Theoretical Linguistics 10, p. 125-145 SZABOLCSI, A. 1986 Comparative Superlatives, in N. Fukui et al. (eds) MIT Working Papers

150 in Linguistics 8, p. 245-265 VERKUYL, H.J. 1981 Numerals and Quantifiers in X-bar Syntax and Their Semantic Interpretation, in J. Groenendijk et al. (eds) Formal Methods in the Study of Language, Amsterdam Mathematisch Centrum, p. 567-599

Martina Noteboom

Hit and dative complements in Old English 0. Introduction* This a r t i c l e discusses the complementarity of the d i s t r i b u t i o n of Old English (OE) matrix d a t i v e o b j e c t s and the pronoun h i t which anticipates an extraposed clause. This complementarity appears to occur only in passive matrix clauses: in comparable constructions with a matrix a d j e c t i v e , i t is not found. Dative matrix complements are not found in the Modern English (ModE) r e f l e x e s of the relevant constructions. In the development of the language, these datives have been replaced by nominatives or PPs. On the face of i t . there does not appear to be any structural d i f f e r e n c e between the OE examples in ( l a ) and ( l b ) ; yet t h e i r modern equivalents are d i f f e r e n t , as is c l e a r from ( 2 ) and ( 3 ) , r e s pectively. (la) (lb) (2) (3a) (3b)

Him waes He-dat was eow eallum you-all-dat

gesaed baet S (AE1 Horn i i . 3 5 6 . 1 ) told that S is cub t>aet S (Wf N 284.18) is c l e a r that S

He was t o l d that S *He was c l e a r that S I t was c l e a r to him that S

This a r t i c l e aims to demonstrate that the divergent development i l l u s t r a t e d in ( 2 ) and ( 3 ) is the result of a d i f f e r e n c e in behaviour of these constructions in OE: structures l i k e ( l a ) did not allow the pronoun h i t to co-occur with a d a t i v e , whereas structures l i k e ( l b ) d i d .

1. Hit as a dummy subject There are constructions in OE in which a matrix passive verb i s combined with an extraposed S. The f i n i t e matrix a u x i l i a r y ( V f ) in t h i s type of construction may be preceded by an a d v e r b i a l , as in ( 4 a ) ; by a preposed d a t i v e c o n s t i t u e n t , as in ( 4 b ) ; by a preposed PP. as in ( 4 c ) ; or by the pronominal h i t . as in ( 4 d ) : (4a) (4b) (4c) (4d)

6a waes sweotole getacnad baet S (Or 248.6) Then was c l e a r l y indicated that S t>am f o l c e waes behaten baet S (AE1 Horn XX.376) the-people-dat was promised that S be bam is awriten í>aet S (CP 79.2) about that i s w r i t t e n that S h i t waes eac sweotole gesiene baet S (Or 252.29) i t was also c l e a r l y seen that S

I t i s c l e a r from this d i s t r i b u t i o n that the f i n i t e verb finds i t s e l f in verbsecond p o s i t i o n . (For a recent survey of verb-second (V2) l i t e r a t u r e , see Haider and Prinzhorn 1986.) Hit in ( 4 d ) , then, could at f i r s t sight be regarded as a dummy, inserted whenever no other constituent occurs immediately b e f o r e the f i n i t e . A comparable s i t u a t i o n e x i s t s in Modern German, where the pronoun as. ' i t * g e n e r a l l y occurs only when the pre-Vf p o s i t i o n is not occupied by some

152 other constituent. E & is not used in m a i n clauses of w h i c h the pre-Vf is filled. Nor is it used in subclauses, w h i c h are V-final. (5a) (5b) (5c)

position

Es wurde gesagt dass S Ihm wurde (*es) gesagt dass S ...dass (*es) gesagt wurde dass S

It can be concluded, therefore, that efi. is a dummy inserted to ensure that the finite verb is in sentence-second position. OE hit. however, does not function in this w a y . It occurs not o n l y in pre-Vf position, but also a f t S X the finite verb (6a), in root clauses that do not have V2 (6b), and in embedded clauses (6c), w h i c h generally do not have V2 either: (6a) (6b) (6c)

Ys hyt alefed t>aet S? (St M a t t h e w 22.17) Is it allowed that S? ac swa hit an scopleo^um sungen is J>aet S (Or 72.20) but so it in songs sung is that S ^jaet gewrit t>e hit on awriten waes hwaet S (Or 268.22) that writing that it o n written was what S

W e m a y conclude, therefore, that h i t anticipating an constructions is not a dummy inserted to preserve V2.

extraposed

S

in

these

2. Complementary distribution It can be observed that hit only occurs in root clauses w h e n they do not otherwise contain a nominal constituent. T h u s , we find examples like (7a), containing hit; like (7b), containing a preposed dative but not hits and like (7c), containing a postverbal dative but again not hit. On the other hand, we do not find examples like (7d), containing b o t h h i t and a dative. The hypothetical structures in (7d) are therefore asterisked. (7a) (7b) (7c) (7d)

& eft hit is gecueden: 'S' (CP 201.24) and again it is said: 'S' him waes eac forgifen £aet S (AE1 Liv XXIII.234) him-dat was also granted that S eac is aelcum preoste forboden t>aet S (Wf 304.30) also is every-priest-dat forbidden that S *hit is h i m gesaegd ^>aet S *him is hit gesaegd )>aet S

H i t and dative objects therefore appear to be in complementary distribution.* Butler (1980) finds o n l y one example w h i c h seems to show the p a t t e r n w e have marked as non-existent: (8)

ac hit nis n a n u m m e n alefed t>aet S (Cons 135.6) but it is no-man-dat allowed that S

However, there is no reason w h y this example should be regarded as a construction. It can also be analysed as a copula construction. Then it par w i t h adjectival structures. Thus (8) n e e d not be regarded as a example to the observed complementarity of h i t and dative objects in constructions.

passive is on a counterpassive

3. Dative subjects Because of the complementarity b e t w e e n h i t and datives in passive constructions, it could be suggested that t h e y o c c u p y the same structural p o s i t i o n , thus ex-

153 eluding one another. This suggestion has the following grounds. In the development of English, dative objects of passives were reinterpreted as subjects. This process may be interpreted as having two steps: (i) development of a specific structural position; (ii) the position was uniformly marked with nominative case. The complementarity could then be taken to signal the first step in this process. The specific position that developed may, since it later came to be assigned nominative case, have been a position restricted to NPs. This idea is supported by the fact that hit is not incompatible with prepositional objects in the same clause, since PP-objects would not occupy this NP-position: (9)

nu is hit scortlice ymbe baet gesaegd baet S (Or 58.7) now is it briefly about that said that S

Interestingly, the construction exemplified OE, but disappeared in later OE: (10)

in (10) was quite common in early

to \>aem is gecueden ^aet S (CP 191.25) to him is said that S

Although clauses with prepositional objects were not rare in later OE, at that stage PP-objects no longer occurred sentence-initially. This indicates that the pre-Vf position was being interpreted as an NP-position, and not as a non-category-specif ic XP-position. Modern Icelandic dative objects seem to be on their way to reinterpretation as subjects. Zaenen and Maling (1983) discuss a number of criteria on the basis of which they conclude that Icelandic dative objects are in fact (quirky case) subjects. These criteria are: preposed dative objects can be replaced by PRO in a control structure; they can occur in ECM positions; and they can be deleted under identity with a nominative in a coordinated structure. All these criteria are characteristic of nominative subjects. Icelandic dative objects do not control verb agreement. Thus, they show syntactic characteristics of subjects, but not subject coding properties like nominative case and agreement. When we apply these criteria for subjecthood to OE dative objects, we find that only the criterion of ellipsis holds: (11)

Oswold sende f a to Scotlande, ... and baed ba heofodmenn Oswold sent then to Scotland, and asked the chiefs baet S ... and wearb t>aes getibod (AE1 Liv XXVI.47) that S and e_ was that granted

We conclude therefore that the absence of hit in constructions with datives cannot be explained by attributing subject status to these datives in OE. 4. Anticipatory pronouns and presupposition OE not only has hit as a subject pronoun, but also as an object pronoun, anticipating an extraposed object clause: (12a) (12b)

J>aet he hit understandan maeg hwaet S (Wf 175.7) that he it understand can what S swa we hit secgan hierdon ... baet S (Or 286.6) so we it say heard that S ('so we heard it said that S')

154 The anticipatory pronoun subject clauses: (13)

is optional for object

clauses

just as

it

is for

gif bu (>onne wenst ()aet S (Cons 15.23) if you-nom then think that S

Bolinger (1977) argues that it-objects occur only when the proposition contained in the extraposed S is presupposed to be true. The appearance of an anticipatory pronoun in object position, then, could be limited to factive environments (see also Kiparsky and Kiparsky 1970). Hoekstra (1983), discussing the distribution of extraposition hfit 'it' in Dutch, argues that the structural position of the embedded S differs significantly depending on the presence or absence of an anticipatory pronoun. The structural difference derives from the assumption that het is referential. It is therefore assigned a thematic role, which it receives in object position. This means that the extraposed S with which it is coindexed cannot be in an argument position, because otherwise there would be only one 6-role available for two argument positions. The extraposed S must therefore be an adjunct clause, as in (14a): (14a)

V'

^

v \

(14b)

\

V'

^

het

s v s i V i' ' (See also Bennis 1986.) Because of its adjunct status, an S' that co-occurs with hfit will be an extraction island. On the other hand, an S' that does not cooccur with het must have argument status to receive a 0-role. This means that its structural position will be as in (14b), and that it will allow extraction. This analysis holds for 0E as well: examples of the type in (15) show that extraction is possible in the absence of hit. (15) Hwaet wylt bu baet ic be do? (AE1 Horn i.152) What wish you that I you-dat do? ('What do you want me to do for you?')

Although conclusive evidence of the non-occurrence of constructions in OE cannot be provided, it would appear that examples of the type in (16) are non-existent, as predicted. (16)

?Hwaet maeg he hit understandan J>aet ic him do What can he it understand that I he-dat do

Passivisation of the structures in (14) will yield a surface subject in the case of (14), but not in the case of (14b). Evidence for the fact that the anticipatory pronoun in the relevant constructions is an underlying object also comes from examples like (17). In (17). the anticipatory pronoun is haes. the genitive form of demonstrativefeafit,which is an alternative for hit in extraposition contexts.^ (17)

Jjaette naenne mon f>aes tweogan ne ^>earf ^aette S (Cons 26.11) that no-man-acc that-gen doubt not may that S

The pronoun anticipating S is in the genitive case, the oblique case assigned by tweogan 'doubt' to its theme object. The passive near-equivalent of (17) is (18): (18) ac forjion hit maeg beon tweod ... hwaeber S (DG 177.8) but therefore it may be doubted whether S

155 Summarising, we have two constructions, one with hit and one without, both passivisable. We observe, then, that the alternative without hit was selected if the verb also takes a dative complement. However, since numerous verbs in OE have more than one complement, there can be no inherent structural reason for an accusative hit and a dative object to be mutually exclusive. Now recall that it was suggested above that it is used only when the propositional content of the extraposed S is presupposed. The passive constructions that we are discussing usually involve verbs of saying, commanding, writing, etc., i.e. verbs whose propositional complement is being asserted rather than presupposed. Thus, a choice would generally be made for a construction without hit. except in cases in which the embedding verb is factive.

5. Adjectival predicates In the above discussion, we have limited our attention to passives. In recent GB-theory, passive constructions do not result from application of a separate 'passive' rule. The term passive is no more than a descriptive label, used to refer to types of constructions that result from the interaction of a number of subtheories. Passive constructions result from the 'disappearance' of the external 6-role of the passivised verb. As Hoekstra (1986) argues, following Roberts (1985), the external 6-role is assigned to the participial affix. This results in absorption of the verb's Case-assigning property. Therefore, an NP that would receive its Case from the active form of the verb remains Caseless when it occurs with the corresponding passive participle. This means that it must move to a position where it can pick up Case. Because the subject position no longer holds a 6-role, it is available as a landing site for such a moved NP: the subject position, to which nominative Case can be assigned. NPs may also fail to receive Case for reasons unrelated to absorption of Case and 6-role found in passives. Under Stowell's (1981) Small Clause analysis, this is the case for adjectival constructions. Thus, the analysis of (19a) would be as in (19b)s (19a) (19b)

John was ill e_ was [ John ill ] SC

The NP Jtolui.« the subject of the small clause, fails to receive Case, as the copula does not assign Case. Jahll therefore moves to the nominative position e_. When we consider OE constructions with matrix adjectives (see (22) below), they would appear to have the structure in (20):^ (20)

ua be adj NP-dat that S

This pattern is realised in ModE as (21): (21)

it is clear to me that S

The OE adjectival constructions are parallel to the types of passives discussed above (cf. (4a-d)): (22a) fconne bib ful uncuf> hu S (Wf N 264.5) then is completely unclear how S (22b) ^>am yrmingan waere micle betere ... baet S (Wf 161.99) the-poor-dat were much better that S (22c) on byssum halgan is swutel ... baet S (AE1 Liv XXXII.250) in this saint is clear that S (22d) hit bi^> eac geornlic t>aet S (Or 212.27) it is also desirable that S

156 In addition, we find adjectival constructions in which hit and dative combine: (23)

swa swa hit gewunelic is bam i>e S (AE1 Liv XXXII.149) so as it customary is those-dat who S

Given the parallellism between passives and adjectives indicated in (4) and (22), we would expect the two constructions to have developed in a parallel fashion. Clearly, however, they have not. Compare the following structures: OE (24a) Him waes gesaegd |>aet S (24b) *Hit waes him gesaegd baet S

ModE He was told that S It was told to him that S

(25a) (25b)

*He was clear that S It was clear to him that S

Him waes cut> \>aet S Hit waes him cuj> baet S

I propose that the difference in grammaticality between ModE (24a) and (25a) is the result of passive constructions in OE disallowing co-occurrence of hit and datives (modern PP)--in other words, that it results from the difference in grammaticality between OE (24b) and (25b), because passive participles absorb Case, but adjectives do not. Consider once more the structure of (19a) above. In underlying structure, Jolm is the subject of an adjectival small clause. Similarly, (¿lit in constructions with adjectival predicates and S'-complements originates as the subject of a small clause complement tofefi..Hence the underlying structure of (25b) is as in (26):4 (26)

(26) corresponds to OE (27): (27)

[{. [ IP VP

[ hit [ cut) ]] be ] I NFL ] S' AP A1

Now compare the underlying structure for passives follows the verb; in OE, it would precede the verb):

(in ModE, the complement

157 IP

(28)

NP e_

I

VP V be

VP I V'

V+PM

NP

Note that the anticipatory pronoun in (26) is the subject of AP, while in (28) t it is the complement of a main verb. It has been argued in various places (e.g. Koster 1978) that subject sentences do not exist. If this is a correct generalisation, it would follow that S's construed with an adjective can never occur in the subject position of an adjectival small clause. Thus» the anticipatory pronoun is necessary in these constructions. S's construed with a verb of saying» however, would either have an anticipatory pronoun in the complement position of the participle (and not in the subject position), or would occur in that position themselves. Therefore, (29a-c) are possible structures, but (29d) is not. (29a)

[ be [ VP VP

[ i t V+PM ]]] S' V1

(29b)

[ be [ VP VP

[ S' V+PM ]]] V1

(29c)

[ be [ it [ clear ]]] S' VP AP A'

(29d)

*[ be [ S» [ clear ]]] VP AP A»

The ban on sentential subjects is derived in Stowell's (1981) theory from the Case Resistance Principle (CRP): S's resist being in Cased positions. However, it has been pointed out that the CRP makes incorrect predictions in various languages. For example, it would predict that S's cannot occur as complements to prepositions, but they do occur there in Dutch and Swedish (cf. Hoekstra 1984; Holmberg 1986). The CRP would also prohibit S's from occurring as objects to verbs, but, given the present analysis, they do occur there. Nevertheless, the restriction against S' occurring in (nominative) subject position seems to stand. That is why an alternative to the CRP is needed. Let us adopt Chomsky's (1986) proposal that subjects of small clauses enter into a relation of HeadSpecifier Agreement, and postulate the principle in (30): (30)

S' may not enter into an agreement relation

Principle (30) will doubtless prove ultimately to be derivable from some deeper property, but for the moment it suffices for our purpose to postulate it in this admittedly rather crude form. Notice that the principle has scope beyond the specific problems we are discussing. Stowell's Small Clause theory has made it clear that the notion of predicate, like the notion of subject, is not categorially limited. Not only Vs and Ns are capable of functioning as predicates, i.e. of taking subjects, but also Ps and As. But the notable exception to this generalisation is S', which cannot function as a predicate:

158 (31a) he found this a nuisance (31b) he found this ugly (31c) *he found sports that it is detrimental to one's health The inability of S' to function as a predicate follows from principle which prohibits it from entering into a head-specifier relation.

(30),

6. The historical development Let us now return to passives and adjectival predicates and try to explain their divergent historical developments. In the adjectival constructions, since S' may not be a subject of AP, there is an obligatory hit as subject of the small clause. Not receiving case in situ, hit moves to the position governed by INFL, where it receives nominative case: (32)

[ hit£ [ [ ti NP-dat cub ] be ] I NFL ] IP VP AP

After the historical loss of structural (morphological) case, the dative NP came to be realised by means of the preposition tain passive constructions, on the other hand, the S' usually denoted the content of what was being asserted (i.e. not of what was presupposed). Therefore it occurred without an anticipatory pronoun hit, i.e. as an internal argument of the verb. Since there is now no nominal requiring Case, the nominative position remains empty, as in (33): (33)

[ e. [ [ IP VP VP

[ S' NP-dat gesaegd ]] be ] I NFL ] V'

The fact that the subject position remains empty is by no means unusual in OE, which allowed passives of the type in (34): (34)

t>aem maeg beon swibe rabe geholpen (CP 28.5) he-dat can be very soon helped

Any dative complement of the participial verb would receive Case in situ. It was usual for the dative complement to occur in topic position, as in (35), preceding the finite verb after V-second. (35)

[ CP

[ NP-datj] Spec

V+INFLi [ ...ti ...ti ...]] J J IP

Within the confines of this paper we cannot say more about the reinterpretation of the specifier position as an NP position which was a specifier of IP (but see section 3 above, where the narrowing down of the specifier position in COMP to NP status was noted). We take it that in time, there being no competitor for nominative case, the loss of lexically assigned case was compensated for by the assignment of nominative case to the former dative argument.

7. Conclusion The different behaviour of passive vs. adjectival constructions in ModE. allowing (36a) but disallowing (36b) is foreshadowed by a difference in behaviour in OE, disallowing (37a) but allowing (37b): (36a) He was told that S (36b) *He was clear that S

159 (37a) *Hit waes him gesaegd ^aet S (37b) Hit waes him cut» |>aet S In OE passive constructions» the complement S' occurred in internal argument position to the main verb. There generally was no anticipatory pronoun, so that there was no NP requiring Case. Nominative Case was later acquired by the topicalised dative. Notice that the ModE alternative in (38) may have arisen out of the structure which we have not found attested in OE, but which presumably was not downright ungrammatical and may have been more frequent in Middle English. An additional factor in the rise of the structure may have been the requirement, appearing in Middle English, that all subject positions must be filled: hence the need for an anticipatory pronoun regardless of the content of the extraposed S. In (38), i£ would be the complement of the main verb, and would be raised to the subject position to acquire Case. (38)

It was told to him that S

The original dative complement has been replaced by a prepositional phrase. In OE adjectival predicates, on the other hand, the pronoun anticipating the extraposed S is the subject of a Small Clause, as the S itself cannot function as a subject. The pronoun is forced to move to the nominative position to pick up Case. This means that nominative Case could not be reassigned to a preposed dative, and the ungrammaticality of (36b) follows. Notes *

I want to thank Colin Ewen, Robert Lankamp, Bob Rigter, Robin Smith, the editors of this volume and an anonymous reviewer for comments on earlier versions of this paper. Special thanks are due to Teun Hoekstra, without whose help it would not have been completed. The research on which this article is based was supported by the Netherlands organisation for the advancement of pure research, ZWO (project nr 300-163-028), and was carried out in cooperation with the Leiden University research project Word order and svntagmatic and paradigmatic structure.

1. Note that German ea is not incompatible with a matrix object: (i) Es wurde ihm gesagt dass S 2. Although hit did have genitive and dative forms, these were seldom used. Generally, when an oblique form was required, as for instance when the pronoun was governed by a preposition, the oblique form of the demonstrative tafit was used. 3. In all syntactic strings given for OE, the relative word order is irrelevant to our purpose. He assume that OE is head final. 4. The PP object (corresponding to the OE dative object) may be a complement of A, or it may be in a small clause with Chlit as its subject. We leave this problem out of account, as it is not relevant to our discussion.

References BENNIS, H. 1986 Gaps and dummies. Dordrecht, Foris BOLINGER. D. and form. London, Longman 1977 Meaning

160 BUTLER. M.C. 1980 Grammatically motivated subjects in Early English. PhD thesis. University of Texas. Austin CHOMSKY, N. 1986 Barriers. Cambridge. Mass., MIT Press HAIDER, H. and M. PRINZHORN (eds) 1986 Verb second phenomena in Germanic languages. Dordrecht: Foris HOEKSTRA, T. 1983 Government and the distribution of sentential complements in Dutch, in W. de Geest and Y. Putseys Ceds), Sentential complementation. Dordrecht, Foris, p. 105-116 1984 Transitivity: grammatical relations in Government-Binding theory. Dordrecht, Foris 1986 Passives and participles, in F. Beukema and A. Hulk Ceds). Linguistics in the Netherlands 1986. Dordrecht. Foris. p. 95-104 HOLMBERG. A. 1986 Word order and syntactic features in Lhfi Scandinavian English. PhD thesis. University of Stockholm KIPARSKY, P. and C. KIPARSKY 1970 Fact, in M. Bierwisch and K.E. Heidolph The Hague. Mouton, p. 143-173

(eds). Progress in

languages and

linguistics.

KOSTER, J. 1978 Conditions, empty nodes and markedness, in Linguistic Inquiry 9. p. 551-593 ROBERTS, I.G. 1985 The representation of implicit and dethematized subjects. PhD thesis, University of Southern California STOWELL, T. 1981 Origins of phrase structure. PhD thesis. MIT ZAENEN, A. and J. MALING 1983 Passive and oblique case, in Levin, L. et al. (eds), Papers in lexicalfunctional grammar. Indiana University Linguistics Club, p. 159-191

Florimon van Putte

Alternating verb forms in Papiamentu 1. Introduction A characteristic of many Creole languages is the occurrence of so-called 'variable verbs': a verb of which the same speaker sometimes uses a 'long' form and a 'short' one on other occasions. The use of these alternants is by no means transparent. Their distribution is considered "largely predictable" - in the languages of the Indian Ocean, for instance - by the position the verb has in the predicate (Baker 1972:98, Green 1984:6), and although the short form frequently occurs as an auxiliary verb, this does not mean that auxiliary and non-auxiliary verbs are formally distinguished by short vs. long forms (Valdman 1978:223). Moreover, it is not at all clear whether the occurrence of long and short alternants of what is considered to be the same verb is a matter of free, meaningless, variation, or not. Papiamentu too has a small number of verbs with alternating realizations: BINI/ BIN 'to come', KERE/KE 'to believe', KIER/KE 'to want', SABI/SA 'to know', 'to be able to' and TINI/TIN 'to have'. The examples given under (l)-(5) show that both the short and long forms may be chosen by the same speaker in syntactically comparable environments. Furthermore, it can be inferred from the examples that the alternation cannot be identified with the distinction between independent versus not-independent (i.e. auxiliary, modal, etc.) use. (1)a b (2)a b (3)a

b (4)a b

(5)a b

Ma asina eu e tata a bin muri, tur loke nan tabatini a caba (Eoha Cuenta 44/2) 'But as soon as the father had died, everything they had ran out' Ela buta riba e balansa tur loke e tin den su cas. {Eoha Cuenta 47/-8) 'He put in the balance everything he has in his house' Bo sabi eu mi ta maestro tambe den musica? (Juliana. 3/22) 'Do you know that I am professor in music as well?' Bo jîa di con e ta 'sina? (Juliana 4/16) 'Do you know why he is like that?' I loke nan ta papia no tin pia ni kabes. Ta esei nos kier kambia. (Martinus 7/-2) 'And what they say has neither head nor tail. That's what we want to change' Nos ke saka tur lokura for di nos komunidat! {Martinus 11-2) 'We want to banish all foolishness from our society' E mes anochi ei un pader a bini den e kampamentù di Tula na Porto Mari. (Lauffer 53/27) 'That same night a father came into Tula's camp at Porto Mari' Ademas mi por a hole ku el a bin Corsou ku mal intenshon. (Lauffer 62/28) 'Furthermore, I could smell that he had come to Curaçao with bad intentions' Mi ta kere ku boso, hendenan blanku, ta hasi esey di otro manera... (Rosario 57/2) 'I believe that you white people do this differently' Mi ke ku m'a yega di bisabo esey kaba. (Rosario 57/-4) 'I think I might have told you this before'

162 The alternation is mentioned in the literature, but very little information is given w i t h respect to its distribution. The question as to whether all the alternating verbs are part of the same phenomenon is not even raised. Nor has it b e e n discussed whether the choice of alternant is arbitrary or involves a communicatively useful distinction. In this paper w e will try to answer b o t h questions. We will first present a survey of w h a t can be found in the literature about the distribution of the alternants. Subsequently we will add new quantitative data. O n the basis of this we will discuss and test a hypothesis about the use of long versus short forms using a corpus of authentic material.

2. The data The data found in the literature are few and not very exact. To b e g i n w i t h , Goilo (1953 and 1968) - the m o s t extensive textbook o n Papiamentu - does not include all the alternating verbs used in (l)-(5). He gives SABI/SA, BINI/BIN, KIER/KE and KERE/KE, but does not m e n t i o n TINI as alternant of TIN, although both forms do occur in a text used b y Goilo (1968). The m o s t recent dictionary of Papiamentu, Dijkhoff (1980), includes TINI and TIN, but here KE, as an alternant of KERE, is missing. 1 This suggests that the alternation is not equally frequent w i t h all the verbs mentioned. Furthermore, w e find the following observations in Goilo about the distribution of the forms: i. SA, KE (as alternant of KIER) and BIN are used independently as well as p r e ceding another verb.^ ii. SABI and BINI are used only as independent verbs. As such they compete w i t h SA and BIN, respectively. It is not discussed, however, whether the o p t i o n is arbitrary or meaningful. iii. In Goilo (1953) only KIER (in his orthography QUIER) is given, w h i l e in Goilo (1968:163) there appear to be three forms: an independent verb K I E R (or KIER), a form K E which can be used preceding another verb as well as independently, and a form K I E R or KER which is used w h e n the following verb is preceded by the particle a. In other words, w h a t we have here is conditioned variation (E ker/kier/*ke a bai 'He wanted to go'), different from the conditioned v a r i a t i o n as in the case of example (3). Moreover, there is another complicating factor w i t h KIER: given the insecure orthography of Papiamentu it cannot be decided whether it is always the same form. K I E R m a y represent KIER (pronounced [kjer]) as well as KIER (pronounced [kjaer]), the two long forms distinguished by Goilo (1968). Since different and unclear factors are interfering in the case of KIER/KE, we therefore will leave this pair (or triplet?) for further investigation and concentrate o n the remaining alternating verbs.'' From the data g i v e n in the literature very little can be concluded: although position m i g h t be a factor of influence - and not m o r e than that - w i t h some verbs, with other verbs it certainly is not; the distinction 'independent' vs. 'not independent' seems to be relevant w i t h some verbs (SABI/SA, BINI/BIN), but not w i t h all, and again this distinction does not seem to be decisive. It thus cannot be inferred from these data w h a t precisely is going on. Let us therefore take a look at w h a t speakers of Papiamentu actually do in relation to these verbs. We will start w i t h the use as independent vs. not-independent verb, and see to what extent the choice of long vs. short forms is correlated w i t h it. I n order to exclude the interference of dialectal differences w e will limit our in-

163 vestigation to the Papiamentu spoken on Curaçao. The linguistic material we use as a corpus consists mainly of written texts (prose written by well-known authors: Rosario, Lauffer, Juliana and Martinus) and the edited transcription of a couple of anonymous folk-tales (Eoha Cuenta). The authors do not belong exactly to the same generation. The oldest (Eoha Cuenta) is approximately 35 years older than the youngest (Martinus) and differs by about 15 to 25 years from Rosario, Lauffer and Juliana. The relative frequency of occurrence of short and long forms in these texts is presented in Table 1. Table 1: Relative frequency of short and long forms in Eoha Cuenta, Rosario, Lauffer, Juliana and Martinus % short

SABI/SA

KERE/KE

Eoha Cuenta Rosario Lauffer Juliana Martinus Eoha Cuenta Rosario Lauffer Juliana Martinus Eoha Cuenta Rosario Lauffer Juliana Martinus Eoha Cuenta Rosario Lauffer Juliana Martinus

99 100 100 100 100 100 100 100 87 100 40 46 56 50 -

33 15 23 -

40

% long 1 -

13 -

60 54 44 50 100 67 85 77 100 60

Number of oc 163 135 131 63 26 74 86 62 31 26 197 65 46 12 2 15 26 22 5 5

The quantitative data demonstrate that i. the verbs differ as far as their preference for the long or the short form is concerned: as was already inferred from the literature, KERE/KE and BINI/BIN do not clearly prefer one form or the other, while TINI/TIN and SABI/SA show a striking preference for the short form. The long form TINI is but a sporadic alternant and appears only in the oldest text, while SABI occurs exclusively (and infrequently) in Juliana. ii. The observed tendency in the relative frequency of long and short forms is the same with all the authors, even in the case of Martinus where our corpus is rather small. Nor do the texts differ significantly in an absolute sense: only Juliana seems to have a slightly greater preference for long forms than the others. It can thus be concluded that the speakers of Papiamentu show a similar behaviour in their preference for long and short forms. It therefore is highly improbable that the use of alternants is arbitrary. Let us now look at the use of alternants as independent vs. not-independent verbs. The quantitative data, which are given in Table 2, show a number of similarities and differences between the alternating verbs. Table 2: Distribution of short vs. long forms in relation to independent vs. not-independent use in total corpus^

164 TIN

SABI

BIN

BINI

(1) 0,5

(193) 98

(4) 2

(5) 3

(179) 97

0 (-)

100 (57)

0 (-)

99 (136)

TINI

(515) 99,5 100 (2)

SA

1 (2)

KE

KERE

(12) 18

(53) 82

N 7.

50 (4)

50 (4)

Z N

For each verb there seems to be a correlation b e t w e e n its use as an independent vs. not-independent verb o n the one hand and long vs. short forms respectively on the other. The use as independent verb cooccurs m o r e frequently w i t h the choice of the long form than the use as not-independent verb, even in the case of TINI/TIN and SABI/SA where the alternation is very rare. This suggests that the principle motivating the choice is the same for all the verbs and this principle m u s t be highly compatible w i t h the difference b e t w e e n independent vs. not-independent use of a verb. Apart from the striking difference as to the relative frequency of alternation w i t h TINI/TIN and SABI/SA o n the one hand and BINI/BIN, KERE/KE on the other, we observe a notable difference b e t w e e n BINI/BIN and KERE/KE: the distribution of BINI and BIN in relation to the factor investigated (independent vs. not-independent use) is almost complementary, while the alternation b e t w e e n KERE and KE occurs in the same context. Does this m e a n that the alternation is motivated by different factors in the case of BINI/BIN and KERE/KE? The d a t a collected in Table 2 seem to confirm Goilo's observation that there is no choice w h e n BINI/BIN is followed by another verb and that the short form has to be chosen in that case. It could be claimed then that the use of BINI and BIN is at least in part a m a t t e r of position. However, there is no need to set these alternants apart from the other verbs. After all, the distribution shown b y BINI/ BIN differs only in degree from the general tendency. Moreover, there certainly is alternation b e t w e e n BINI and BIN in the context of independent use - as in the case of the other verbs - and this too has to be explained. A n explanation w h i c h covers all the distributional facts w o u l d undoubtedly b e more interesting. Since we are interested in the possibility of a meaningful use of the long vs. short forms, let us approach the alternation from a functional point of view.

3. The hypothesis There is a natural tendency in communication to encode w h a t is considered to be the same conceptually in the same linguistic form (Slobin 1985:238). Thus, w h e n a language has partially different forms for what seems to be the same concept and these forms are m a i n t a i n e d in the language in question, there exists the possibility that the distinction is a communicatively useful one for w h i c h the formal difference is, at least, not an inappropriate means of expression. According to Mayerthaler (1978:34) there is a tendency towards m o r e naturalness in morphological development w h i c h manifests itself in constructional iconicity: those aspects of the m e s s a g e w h i c h are central also tend to be perceptually m o r e prominent than the m o r e marginal elements. To put it differently: what is m o r e prominent morphologically (i.e. 'longer') should be m o r e prominent (central, independent) conceptually as well. This principle seems to hold indeed in the case of verb alternation in Papiam e n t u . Our hypothesis is that w i t h the long form the speaker of this language signals the prominence (centrality) of the verbal concept in the m e s s a g e he wants to convey, while w i t h the short form he does not. The d a t a w e have presented so far can b e understood w h e n looked at from this perspective: independent use suggests m o r e attention for the verbal concept than w h e n it is part of a complex predicate.

165 It is therefore natural that the more prominent long form is preferred when the verb is used independently and not when it is followed by another verb. The position in front of a verb is typically occupied in Papiamentu by the monosyllabic particles of time and aspect and what are traditionally called auxiliary verbs and modals like ser 'become', wordu 'become' and por 'can'. It is also the position of the less central, modifying verb in so-called serial verb-corstructions. Furthermore, it is plausible for the alternation to preferably occur in independent use and not when the verb is followed by another verb, since the use as independent verb does not imply the prominence of the verbal concept in the same way that the occurrence with another verb presupposes the relatively lower prominence of the first one.5 This explains the distribution of long and short forms observed in Table 2. In addition to the data we have seen so far, there are other factors which point in the same direction, i.e. the long form is used to highlight the verbal concept while the short form is not. i. One of the time/aspect particles of Papiamentu is ta. Ta is used when the speaker perceives the event as related to a moment in time which is not anterior to the moment of speaking. When he does not use any particle the event is not seen as related to time but rather as potential or anticipated. This difference is illustrated by (6) and (7), respectively. (6) E ta_ kome. 'He eats/is eating' (7) Si e 0 kome... 'If he eats'6 When we look at the difference between (6) and (7) from the perspective of prominence of the verbal concept, it is evident that the absence of ta rather than its presence corresponds with lack of prominence of the event involved. It therefore comes as no suprise that the verbs which always occur without ta, according to the traditional grammars, are precisely the alternating pairs TINI/ TIN and SABI/SA which in modern Papiamentu seem to have lost their long alternant . 1 Moreover, it is likely that the use of ta. cooccurs more frequently with the long form than with the short one, if our hypothesis is correct. As is shown in Table 3, this expectation is confirmed by the data: ta is used in 42% and 61% of the occurrences with BINI and KERE respectively, while it does not cooccur with KE and is used 32% less with BIN. Table 3: Relative frequency of long vs. short forms with ta, in % ^ TA TINI TIN SABI SA BINI BIN KERE KE

0 0 0 0 42 10 61 0

other particles 100 31 0 18 31 65 33 0

0

N

0 69 100 82 27 25 6 100

1 502 4 226 145 132 49 16

ii. If our hypothesis is correct, it is likely that the long form will be chosen and not the short one when the message to be conveyed is a command. By using an imperative the speaker emphatically brings to the attention the event intended by the verb. Goilo claims that the two forms SABI and SA are used as imperatives, without further comment. In our corpus we did not find examples of the imperative message with this verb. Thus, quantitative validation »ras not possible. However, our expectations are completely satisfied with the other verbs. Goilo (1953 and

166 1968) gives BINI and not BIN as the imperative form when the verb is used independently and we did not find one counterexample in our corpus. TINI/TIN is not used for commands, according to Goilo (1953), but on the other hand he names TENE - a long form, and etymologically related to TIN - as the only possible form in this case. Although data are lacking in the literature about the use of KERE/KE as an imperative, our informants are positive that only KERE and not KE can be used here. The data adduced make it plausible that there is a common motivation for the choice of alternant and they confirm our hypothesis about its nature: by using the long form the speaker of Papiamentu wishes to highlight the concept introduced by the verb. This is not to say that the contrast between long and short forms is equally useful for all the alternating verbs. As was shown in Table 1 the alternation is rare with TINI/TIN and SABI/SA and the short form seems to be more useful than the long one with these verbs, while in the case of BINI/BIN and KERE/KE it is rather the long form which seems to be more useful0 Given this difference, let us take a closer look now at the context in which the infrequent vs. the more frequent forms are used. 4. Infrequent alternation Forms with very low frequency, like TINI and SABI tend to be identified with archaic, hence unusual, speech. It is therefore likely that TINI and SABI are used to achieve certain stylistic effects. However, stylistic use does not imply arbitrariness. On the contrary, given the contrast between long vs. short forms, it is plausible that the unusual long form will also be used when the extra attention drawn to the verbal concept is justified by the context. This is illustrated by the comparatively frequent use of SABI by the 'younger' Juliana (see table 1). The alternant SABI is used exclusively in direct speech when the principal character of the story, the conceited, quasi-learned Dr. Kriki addresses himself to somebody. Dr. Kriki is constantly out to impress other people by using, among other things, ceremonious language interspersed with learned words. In this context the use of long forms is appropriate in general, especially the use of unusual long forms like the infrequent SABI. However, not only SABI, but also SA is used by Dr. Kriki, and what is more, the choice of alternant is again not arbitrary but related to a difference in the respective contexts which can be characterized in terms of the supposedly different value of long vs. short forms: mi 'I', i.e. Dr. Kriki, is the preferred subject when he uses SABI (3 out of 4 occurrences) while it is the subject of SA in only 30% of the 27 cases. Since 'knowing' is made more prominent by the long form than by the short one, according to our hypothesis, it can be expected that a character like Dr. Kriki will be inclined to use SABI more frequently when his own 'knowing' is under discussion than when he is talking about what other people know. 5. BINI/BIN and KERE/KE Although the use of TINI and SABI might be considered stylistic (but not arbitrary), this does not apply to BINI/BIN and KERE/KE. In the case of these verbs the formal contrast is fully exploited to communicate the degree of prominence of the event involved. Let us see now how this is supported by the data. We start with the pair BINI/BIN. It is plausible that the event of 'coming' is more central when the goal implied by the verb is not explicitly mentioned than when it is mentioned. As a con-

167 sequence, the probability that the concept of 'coming' will be introduced with the longer form BINI is higher when a locative complement is missing than when such information is explicitly given. This prediction is not contradicted by the quantitative data in Lauffev, the text where the alternation between BINI and BIN is most frequent. When a locative complement is missing, BINI is the form chosen by Lauffev as shown in Table 4. Table 4: Relative frequency of BINI and BIN in relation with the presence vs. absence of a locative complement (in %) BIN -locative comp. +locative comp

0 19

BINI

N

100 81

6 11

Further proof of the correctness of our hypothesis can be obtained when we look at the kind of locative complement involved. Since we are dealing with a verb of direction, the goal is a more natural, more expected, locative complement than, for example, the place in which the coming occurs. The distinction between the two kinds of complements with verbs of direction formally corresponds in Papiamentu with its introduction in a bare noun phrase versus a prepositional phrase, as is illustrated by the examples (8) and (9). (8) E ta bin 0 kas. (9) E ta bini den e kampamentu.

'He comes home' 'He comes within the camp'

It is natural that the looser relationship between the locative complement and the verb (in the case of a PP) favours the prominence of the event of coming and will thus cooccur more frequently with the long form than when the verb is accompanied by the closely related goal. Again, the quantitative data found in Lauffev - summed up in Table 5 - clearly confirm this. Table 5: Relative frequency of BINI and BIN in relation to the presence vs absence of a preposition preceding the locative complement (in %) locative comp. = 0NP locative comp. = PP

BIN

BINI

N

33 12

67 88

3 8

Although the use as an independent verb is the natural environment of alternation, it is not excluded that the less expected long form BINI should be chosen when the verb is followed by another verb. Compare the minimal pair (10) and (11). (10) E tabata bini stad tur siman bin pidi limosna. (Eaha Cuenta 55/1) 'She used to come up to town every week to beg for alms' (11) Su otro siman ela bolbe bini. Rey bondia! Rey di cune: Bondia wela! Rey m'a bini pidi un limosna. {Eaha Cuenta 55/17) 'The following week she came back. King, good day! The king said to her: Good day, granny! King, I have come to beg you for an aim' According to our hypothesis, the relevant difference in message structure between (10) and (11) should be the different prominence of the event of coming. Example (10) is a description given by the story-teller of what the protagonist of the story (an older woman) used to do every week. Reference is not made in this case to a specific event, but to what usually happens every week. The most essential part of information is 'pidi limosna' ('beg for alms'), that is what the story is all about. In (11) the woman herself is speaking. Furthermore, in (11) - as opposed to (10) - we are dealing with a specific event (as can be inferred also from the use of the perfective particle a), which is aimed at getting something

168 from the hearer ( t h e k i n g ) . I t i s evident that in t h i s s i t u a t i o n (as compared with ( 1 0 ) ) the appearance ( ' c o m i n g ' ) of the woman and her ' b e g g i n g ' deserves as much a t t e n t i o n as p o s s i b l e . To t h i s end, the long form BINI i s an a p p r o p r i a t e means. By foregrounding the event of coming - ' I have come t o you' - in c l o s e r e l a t i o n to the event of b e g g i n g , the request made i s o b v i o u s l y r e i n f o r c e d . The l i n e of reasoning we f o l l o w e d f o r BINI/BIN, i n s p i r e d by our h y p o t h e s i s , can a l s o be applied t o KERE/KE: the more c e n t r a l the concept of ' b e l i e v i n g ' , the more probable KERE (and not KE) i s . Contexts where ' b e l i e v i n g ' i s r e l a t i v e l y more prominent a r e , f o r example, those where the speaker c o n t r a s t s his own b e l i e f s with those of other p e o p l e , versus s i t u a t i o n s where the thing b e l i e v e d i s more important than the event of b e l i e v i n g i t s e l f . Compare example ( 1 2 ) , where the s p e a k e r ' s b e l i e f i s not contrasted (and KE i s used) with (13) where Dindina's lack of b e l i e f - implied by the question - i s contrasted with her f a t h e r ' s b e l i e f (and KERE i s u s e d ) . ( 1 2 ) Mi t a b a t i n miedu, e l a s i g u i . Pasobra m'a s i n t i ku mester t i n mas kos e i t r a s . I mi ke ku m'a hasi bon d i no p a p i a . ( L a u f f e r 17/14) ' I was f r i g h t e n e d , he continued. Because I f e l t that there was more behind i t . And I think I did w e l l in not saying anything' (13) "Dindina, bai buska p o l i s " , Pachi F e l i a p i d i su j u muhe. " S i mi mester s i g u i drumi bow d i e ramada aki un anochi mas, romatisma ta matami." "Papa ta kere ku e s p i r i t u ta s a l i f o r d i kushina ora e mira p o l i s ? " "Tur s p i r i t u t i n miedu d i p o l i s " , Pachi F e l i a d e k l a r t . ( L a u f f e r 9/3) (Context: Pachi F e l i thinks that there are ghosts in h i s house and f o r that reason he sleeps on the porch) 'Dindina, go and g e t the p o l i c e , Pachi F e l i begged of h i s daughter. I f I have to sleep another n i g h t out here under t h i s shed, I w i l l d i e of rheumatism. Do you r e a l l y b e l i e v e , daddy, that the ghost w i l l l e a v e the k i t c h e n when i t sees the p o l i c e ? A l l ghosts are a f r a i d of the p o l i c e , Pachi F e l i s a i d . ' Again, t h i s meaningful use of the long v s . the short form can be g i v e n q u a n t i t a t i v e support. I t i s natural that the concept of ' b e l i e v i n g ' i s g i v e n more a t t e n t i o n when i t stands on i t s own, i . e . when i t i s not f o l l o w e d by a complement. S i m i l a r l y , the concept of ' b e l i e v i n g ' i s more prominent when i t i s the thing questioned, or when i t i s negated v s . when t h i s i s not the c a s e . A l l these p r e d i c t i o n s are c l e a r l y corroborated by the r e s u l t s of the counts we made in Rosapio, Lauffer and Mccpt-Lnus, the t e x t s where KERE/KE a l t e r n a t e . As f o l l o w s from Tables 6, 7 and 8, the long form and not the short one i s chosen with absolute p r e f e r e n c e in the more prominent c o n t e x t , w h i l e t h i s choice i s found in only 76, 77 and 75%, r e s p e c t i v e l y , of the occurrences when the context i s l e s s prominent. Table 6: R e l a t i v e frequency of KE and KERE in i n t e r r o g a t i v e v s . other sentences ( i n %)

i n t e r r o g a t i v e ss. 4 i n t e r r o g a t i v e ss.

KE

KERE

N

0 23

100 77

6 47

sentences

Table 7: R e l a t i v e frequency of KE and KERE in r e l a t i o n to the presence v s . absence of a complement of KE/KERE ( i n %)

verb -complement verb +complement

KE

KERE

N

0 24

100 76

7 46

169 Table 8: Relative frequency of KE and KERE in relation to the presence vs. absence of a negation (in %) +negation +verb -negation +verb

KE

KERE

N

0 25

100 75

9 44

6. Conclusions It seems justified to conclude that the occurrence of alternating forms for what is conceptually the same verb in Papiamentu cannot be dismissed as positionally conditioned or arbitrary, meaningless variation. Although in the case of TINI/ TIN and SABI/SA the use of the long form might be characterized as archaic or stylistic, this does not imply that the principle underlying the use of SABI and TINI differs fundamentally from the one we suggested for the long vs. short form in general. The alternation observed in Papiamentu links up with the natural tendency to formally encode less prominently what is also less prominent conceptually. From this principle the preference of the long form when used as an independent verb and of the short alternant for the use as 'auxiliary' follows. However, there is no absolute correlation, as can be expected when there is free - but motivated - choice. Ultimately, it is the degree of prominence of the verbal concept as perceived by the speaker which justifies the choice made, and not a particular linguistic context. Our analysis is validated by a number of qualitative and quantitative data which, taken separately as well as cumulatively, demonstrate that the choice of alternating forms is far from arbitrary or mechanic and can be related to general principles underlying the shaping and exploitation of natural human language. Notes 1. Given the prescriptive intentions of both Goilo and Dijkhoff it is very unlikely that the alternants are socially marked. The alternation is mentioned also in Birmingham (1971). All the alternating verbs discussed here are given, but no more than that. 2. Bendix (1972), discussing BINI/BIN, claims that only the short form is chosen when the verb followed by another verb is used for the "aspectual, inceptive category" "whereas for direction both bin and bini occur". It is not clear, however, whether it is claimed that the verb preceding another verb always expresses "the aspectual, inceptive category". 3. One of the problems to be investigated is whether and/or to what degree KIER is specializing for the function of bridging the gap between KE and a, while KIER ([kjaer]) is saved for a function comparable with that of the other long forms. It would be interesting, furthermore, to investigate the possible interaction between KE/KIER and KE/KERE. It is not discussed in this paper how the KE of KIER is distinguished from the KE of KERE (minimal pairs do occur). Undoubtedly, the different contexts in which the two verbs appear will have an important part in it. Birmingham (1971:82) claims that "the distinction [between KE=KIER and KE=KERE] is usually made as follows: [mi ke] 'I want', [mi ta ke] 'I believe'". However, this claim is not confirmed by our data. See Table 3. 4. It might be objected that in the corpus on which Table 2 is based different styles (oral and written) and generations are mixed. However, Table 1 already showed that there is no significant difference between Eaha Cuenta (the 'older',

170 oral text) and the other texts as to the distribution of long and short forms, and this is further confirmed when comparing Table 2 with 2a below where Echa Cuenta has been left out. Table 2a: As Table 2, but without Echa TIN (353) 100 100 (2)

TINI (")

0

0 (-)

SA

Cuenta

SABI

BIN

BINI

KE

KERE

(4) 3

(5) 7

(63) 93

(11) 19

(46) 81

N %

100 (1)

Z N

(137) 97 100 (39)

0 (-)

100 (57)

0

0

(")

(-)

The only notable difference between the two tables regards the occurrence of KE and KERE in the context of not-independent use. However, no conclusions can be drawn since the number of occurrences is rather small: N = 8 in the case of Table 2 and only 1 in Table 2a. 5. When in a sequence of two verbs one has to be interpreted as hierarchically inferior to the other, it is natural that the hierarchically inferior one should be toned down and not the other one. 6. Although it is to be expected that the hypothetical si favours the absence of a particle, the use of ta is not excluded in this environment. In the following example, for instance, ta is the appropriate choice and not 0 since the speaker's interest is in a particular, time-related event and not in the act of eating by the subject in general. Bai pidi'e si e por fiami dies florin. Ma si e ta kome (??si e kome), warda te ora e kaba di kome. 'Go and ask him if he will lend me ten guilders. But if he is eating, wait until he has finished his meal' 7. As far as their no cooccurrence with ta is concerned, they are in the company of the modality verbs por 'can', mester 'must', and also ke 'want' the alternant of KIER. 8. An additional argument can be found in the following. If the use of alternants has to do with the prominence of the verbal concept, it can be expected that the distribution of long forms observed in Table 3 will be comparable with that of 'common' transitive verbs. This is indeed so. However, since in the case of other transitive verbs the contrast between long and short forms is lacking, the distribution is less skewed, as is shown by the following data taken from Rosario. Relative frequency of transitive verbs vs. long alternants in relation to _ta vs. other particles and 0 in Rosccrio (in %) transitive verbs long alternants

TA

other particles

0

N

21 52

79 32

8 16

292 194

9. The corpus on which this table is based, consists exclusively of those occurrences where the presence or absence of a particle can be related unquestionably to the verbs studied, i.e. those cases where the alternant is used either independently or as the first verb of a sequence of verbs, 10. It might be asked why the use of the alternation apparently is diminishing in the case of TINI/TIN and SABI/SA. This problem needs further diachronic investigation. Part of the answer could be the competition with other,formal-

171 ly and/or semantically related forms: TENE vs. TINI and KONOSE vs. SABI.

References BAKER, P. 1972 Kreol, London, Hurst BENDIX, E.H. 1972 Serial verbs in the Carribean and West Africa: their semantic Analysis in Papiamentu, Hunter College of the City University of New York, unpublished paper BIRMINGHAM, J.C. 1970 The Papiamentu language of Curaçao, Ph.D. Thesis, University of Virginia DIJKHOFF, M. 1980 Dikshonario Pap-iamentu-Uland.es, Ulandes-Papiamentu, Zutphen, De Walburg Pers ECHA CUENTA 1970 A collection of folk-tales transcribed and edited by E. Juliana, pp. 9-50, Amsterdam, De Bezige Bij GOILO, E.R. 1953 Gramatica Papiamentu, Curaçao, Hollandsche Boekhandel 1968 Papiaments leerboek, Aruba, De Wit GREEN, J.N. 1984 Creoles Registers and popular French, in La Francophonie (Proceedings of the AFLS Symposium, University of Nottingham, 1984), edited by D.E. Ager and R. French JULIANA, E. 1960 Aventura di un kriki, pp. 3-34, Curaçao LAUFFER, P. 1968? Kwenta pa kaminda, pp. 42-80, Aruba, De Wit MARTINUS, F. 1983 Martein Lopap 2 o malesa di semi-lingualismo, Curaçao, Editorial Antiyano MAYERTHALER, W. 1978 Morphologische Natürlichkeit, Berlin, Technische Universität Berlin ROSARIO, G.E. 1969 E rais ku no ke muri, pp. 5-60, Amsterdam, De Bezige Bij SLOBIN, D. 1985 The child as a linguistic icon-maker, in J. Haiman (ed), Iconiaity in Syntax, Amsterdam, John Benjamins

Bob Rigter

How do you do dol 0. Introduction* One of the most remarkable features of the grammar of English clauses is the d i f f e r e n c e in behaviour between f i n i t e main verbs and f i n i t e a u x i l i a r i e s . The most s t r i k i n g l y i d i o s y n c r a t i c items in the class of the English f i n i t e a u x i l i a r i e s are the forms of supportive DO. which a l t e r n a t e with bare t e n s e s . I argue that the grammar of English requires no c o n s t r u c t i o n - s p e c i f i c rules f o r the use of supportive DO. I t s behaviour r e s u l t s from the i n t e r p l a y of the general rules of English syntax and the c a t e g o r i z a t i o n of supportive do/does/did as I - v e r b s ( i . e . f i n i t e verbs generated as the head of an INFL p r o j e c t i o n ) which, apart from t h e i r tense s p e c i f i c a t i o n , are semantically empty.

1. The syntactic f r m n r a i k In t h i s a r t i c l e I s h a l l adopt the v e r s i o n of government & binding syntax proposed in Chomsky (1986).* In t h i s proposal the D-structure of an English clause with a t r a n s i t i v e verb i s as in ( 1 ) . in which CP means Complementizer Phrase, of which C i s the head, and IP means INFL Phrase, of which I i s the head. IP c o r r e sponds t o the t r a d i t i o n a l S node, and CP to S 1 . Following Chomsky's (1986) d e f i n i t i o n of government. I assume that the f i l l e r of a tensed I governs the NP in the s p e c i f i e r p o s i t i o n of the I P . and can thus assign nominative Case t o t h i s NP. This nominative NP accommodates the external argument of the t r a n s i t i v e V, i . e . the argument which i s not subcategorized by the V. Only the NP which accommodates the i n t e r n a l argument of the t r a n s i t i v e V i s c-commanded. governed, subcategorized and Cased by t h i s v e r b . (1)

CP spec

C' C

~~IP NP

I'

V

NP

2. The category I in English I t i s g e n e r a l l y assumed in generative grammar that tense f e a t u r e s are generated in the I p o s i t i o n . In V2 (= verb second) languages such as Dutch. F r i s i a n . German and the Scandinavian languages, the tensed I p o s i t i o n i s l e x i c a l i z e d by moving the V into i t . Modem English distinguishes i t s e l f from these V2 languages by the f a c t that i t has a separate class of overt l e x i c a l IP heads consisting of

174 1) the modals; 2) the finites bare« has. had when used as perfective auxiliaries and in some other, marginal uses; 2 3) the finites am. are. is. was. «ere: 4) the supportive finites do. does, did; 5) the negative variants of all these items. In Rigter (forthcoming) I argue that these I-verbs. as I shall call them, have inherent tense specifications, while the finite forms of main verbs receive their tense under government by a 'bare tense' generated in the I position. If the I-verbs have inherently specified tenses and if the tense morphology of main verbs is induced under government, it can be understood why I-verbs can show idiosyncratic tense restrictions which are totally absent in the class of main verbs. Under the assumption that I-verbs are stored in the lexicon with inherent tense specifications, it can also be understood why the 3rd pers. sing, present form of all these I-verbs is morphologically irregular, whereas the ruleinduced 3rd pers. sing, present form of main verbs is always regular, with only three exceptions, viz. says. has. does, of which the latter two can be regarded as due to the fact that they are of the same origin as the I-verbs has and does. The difference in behaviour between I-verbs and main verbs is illustrated by the data in (2)-(5). (2) (3) (4) (5)

Pete hasn't left Has Pete left? Pete has left, hasn't he? Pete HAS left (capitals indicate emphasis)

-

*Pete *Left *Pete *Pete

leftn't Pete? left, leftn't he? LEFT (cf. Pete DID leave)

Examples (2)-(5) illustrate the 'NICE properties' of the English I-verbs. NICE is used by Huddleston (1976) and Palmer (1983) as an acronym for Negation, Inversion. Code and Emphasis. Referring to Langacker (1974), Palmer (1983) distinguishes between on the one hand the 'modal component' of a sentence, in which the speaker expresses his attitude to what he says, or his degree of commitment to its truth, and on the other hand the 'propositional component', in which the propositional content of the sentence is expressed. Palmer observes that the phenomena referred to under the heading 'NICE properties' do not add to, or alter the propositional content of a sentence, but that they deny it, question it, repeat it. confirm it. Thus the NICE properties are to be seen as properties of elements in the modal component of the sentence. In English, the syntactic correlate of this modal component is the IP. The syntactic correlate of the propositional component is the VP. A main verb functions as the head of a VP, and does not show the NICE properties of the I-verbs. Tenses also belong to the modal component of the sentence, since they are used to express a commitment on the part of the speaker that a tenseless proposition gets a particular truth value for a time interval which either precedes (in the case of a [+PAST] tense) or does not precede (in the case of a [-PAST] tense) the speaker's present. Thus it is in full accordance with Palmer's views that in a finite clause without an I-verb a 'bare tense' is analyzed as heading the IP. It is clear from (2)-(5) that, contrary to what happens in V2 languages, an English main verb cannot move into the I position to lexicalize the tense features in that position: if a finite main verb could move into the tensed I position, the distinction between finite main verbs and I-verbs would be lost. This leads to the conclusion that the class of I-elements in English consists of the overt I-verbs listed above, plus the 'bare tenses' which, though morphologically 0, also belong to the category I, and, just like the other I-elements. subcategorize a VP, and have a chronological specification. Thus bare tenses as IP heads are roughly comparable with pro as an NP head. The overt manifestation of a bare tense is accomplished by induction under government of the relevant morphology on the head of the subcategorized VP. If a bare tense, though morphologically 0. counts as a member of the lexical class of I fillers, this partially explains

175 why the V which heads the subcategorized VP cannot move into an X position which contains a bare tense: the position is already occupied. However, it does not quite explain why main verbs do not alternate with I-verbs as heads of IPs. The generalization that accounts for this appears to be that« in English, only items with a lexical tense specification can be inserted as heads of tensed IPs.

3. V2 phenomena in English In the literature on the V2 phenomenon (cf. Haider & Prinzhorn 1986) it is commonly assumed that the V2 effect is caused by movement of the finite verb from the I position into the C position, and movement of a topicalized constituent into the specifier position of the CP. Kayne (1982) proposed that the highest CP in a sentence is predicative, while an embedded CP is an argument. Under this proposal, the highest C in a sentence has a [+V] specification, which makes it a suitable landing site for V movement. Because the highest C is not properly governed, and has feature content but is not lexical, V movement is necessary to avoid a violation of the ECP (=empty category principle). Weerman (1986) proposes that SAI (=subject-auxiliary inversion) in English interrogative headclauses and in English headclauses with negative topicalization is due to the same factors as movement of the finite verb in V2 languages. Thus in (6)-(ll) the I-verb has moved from the I position to the C position. (6) (7) (8) (9) (10) (11)

Has Jessie left? Did Jessie leave? Who have you seen? Who did you see? Never had she seen anything like it Not a soul did we see in the deserted area

Weerman (1986) provides a systematic treatment of the differences between V2 phenomena in Dutch and the more limited set of instances in which finites appear to move into C in English, in cases of SAI. He attributes the differences between Dutch and English in this area to a difference in the grammar of the feature [AGR]. On the basis of detailed arguments based on Dutch and Frisian examples, De Haan & Weerman (1985) and Weerman (1986) base-generate the feature [AGR] in C, while the tense is base-generated in I. Weerman argues that in Dutch headclauses [AGR] in C has to lexicalize to avoid an ECP violation. In accordance with Kayne (1982) the C is then lexicalized by moving a tensed V from I into C. In subclauses the CP will have a [-V] specification, and then [AGR] lexicalizes as a complementizer. In the theory sketched here, the subject gets its Case from [AGR], which governs the IP, in which the nominative Case can percolate to the subject NP. Weerman then reduces the differences between English and the V2 languages to a single parameter, i.e. whether or not [AGR] is 'absorbed' by IP or not. In the V2 languages, [AGR] cannot be absorbed by IP, and hence the C containing [AGR] has to lexicalize. In English non-interrogative headclauses without negative topicalization, Weerman proposes that [AGR] is not only copied in the IP. but is drained from the CP altogether. The possibility of an ECP violation is then averted because the C no longer dominates any features. The reason why [AGR] absorption is possible in English, but not in V2 languages, must be sought in the existence of the separate class of I-verbs in English, which makes the IP into a lexical projection in its own right. In the examples in (6)-(ll), Modern English is in perfect agreement with regular V2 languages in the sense that in these sentences there is no [AGR] absorption. The difference between English and regular V2 languages lies in the fact that V2

176 ¿nguages never appear to allow [AGR] absorption, whereas Modern English always ,hows this absorption, except in constructions of the type in (6)-(ll). In these constructions the I-verb has moved into the C position, which must be due to non-absorption of [AGR]. Since [AGR] absorption in [+V] clauses is the rule rather than the exception in Modern English, [+V] clauses without [AGR] absorption are marked constructions. In (6)-(9) the [+V] clauses are marked because of their interrogative illocution. In (10) and (11) they are marked because negative topicalization moves the negative constituent into the specifier of the CP. (12) (13) (14) (15) (16)

A blue car and a red one (*... and That boy left but this one stayed (*... but Charles never drank a lot but John did (*... but Mary wanted Pete to go and Sue wanted Charles to (*... and I saw him leave and Mary saw him leave (/do so) too (*... and

a red 0) this 0 stayed) John 0) Sue wanted Charles 0) Mary saw him 0 too)

Before we take a closer look at the behaviour of supportive DO, a general principle concerning the lexicalization of heads in English must be considered. In (12) and (13), the NPs of which the specifiers are lexical must have lexical heads. In (14) John is in the specifier position of the [+tense] IP head did. In (15) Charles is in the specifier position of the [-tense] IP head to. These IP heads cannot be omitted. In (16) hia is in the specifier position of a verbal small clause, of which the head (i.e.leave, or do) cannot be omitted either. Examples (12)-(15) give rise to the generalization in (17). If (17) is a correct generalization, it is to be expected that a C must be lexicalized in all cases in which the specifier of C is lexicalized. (17)

If, in English, a specifier is lexicalized, the head of the projection that takes this specifier must be overt.^

4. The overt manifestation of the [+V] C position in English It is important to observe that the non-absorption of [AGR] in (6)-(ll) is overtly manifested, because the I-verb in the C position is to the left of the subject in the specifier of the IP. In section 2 it was argued that in English a bare tense, though non-overt because it is morphologically 0, counts as an IP head. If the I-verbs do, does, did are semantically empty apart from their inherent tense specifications, the sole difference between bare tenses and the corresponding I-verb forms of supportive DO is that the former are non-overt, whereas the latter are overt. Now let us see what happens when we replace the overt I-verb did in (7), (9) and (11) by the corresponding bare [+PAST] tense, which induces past-tense morphology on the subcategorized verb. The resulting structures, provided in (18), (19) and (20), respectively, are ungrammatical. (18) * [ c p [+PAST]i [ I P John [ ^ e£ [yp left ]]]] (19) * [ c p Whoj [ c . [+PAST]i [ I P you [jt

ei

[yp saw ej ]]]]]

(20) * [ c p [Not a soul]j [ c . i+PAST^ [ I P we [j. e £ [yp saw ej ]]]]] Let us see why (18) is ungrammatical. Because the bare tense (i.e. [+PAST]) in (18) is non-overt because it is morphologically 0, its movement from the I position into the C position is not perceptible. This means that in (18) a structure that should manifest a marked, interrogative illocution is not perceptibly different from its unmarked, non-interrogative variant. The only way in which the marked non-absorption of [AGR] in this [+V] CP can be made physically percepti-

177 ble is by moving an overt I element into the C position to lexicalize the nonabsorbed [AGR] feature. The only way to keep the resulting structure semantically indistinct from (18) is to use the I-verb did« which is semantically empty apart from its inherent tense specification, which is identical with the specification of [+PAST]. The resulting structure is (7). in which the fact that the I-verb is in the C position is physically perceptible, since it is overt and precedes the subject. Similar explanations could be provided to account for the ungrammaticality of (19) and (20). However, in the case of (19) and (20) it might be objected that the markedness of the construction is already physically perceptible because of the initial position of who and not a sonl, respectively. This is where the generalization formulated in (17) comes in. The [+V] C is not a lexical category in its own right, and a bare tense, which is morphologically 0, cannot function as the lexical head of a CP. Yet, this CP must have a lexical head, because it has a lexicalized specifier. Thus (18), (19) and (20) can only be salvaged by using an overt I-verb in the [+V] C position.

5. Ron-extraction from subject positions In the WH-questions and cases of negative topicalization considered in (8)-(ll) and in (19) and (20), the WH-constituent or topicalized negative constituent was always extracted from a non-subject position. Let us now look at (21) and (22), which look like cases of extraction from subject position. If (21) and (22) involve extraction from subject position, their analysis would be as in (23) and (24), respectively. (21)

Who likes John?

(22)

Not a soul was stirring

(23)

[ c p Wh0£ [ c . [ c [-PAST]j ][ I P e£ [j.

(24)

C c p [Not a soul]£ [(jt [(; wasj ][jp e£ [jt [j ej ][yp stirring]]]]]

ej H y p likes John]]]]]

In (23) there are no overt constituents between who and likes, and in (24) there are no overt constituents between was and stirring. This means that in structures of the type in (21) and (22) there is no overt evidence of movement into the CP. With reference to (21). Chomsky (1986s 50) remarks that '...the language learner assumes that there is syntactic movement only where there is overt evidence for it. We might suppose that the unmarked case for a language with overt WH-movement is that it always takes place at S-structure. so that nonmovement of subject in English would have a somewhat marked character, accounting for the persistence of weak island effects even with WH-subjects, as in (105). (105) What do you wonder [ c p who saw t] ' Chomsky argues that what could not have moved out of the embedded CP if who is in the specifier of the CP of the embedded clause. If, on the other hand, there has been no vacuous movement of who in Chomsky's example (105), who is in situ in the subject position of the embedded clause, and what can move through the specifier position of the embedded CP into the matrix specifier position. Chomsky thus postulates that, for lack of positive evidence in the languagelearning process, a WH-constituent is not moved out of the D-structure subject position in sentences like (21) and in his own example [i.e. (105)]. He accounts for the relative awkwardness of that example as due to the influence of structures in which movement of a WH-constituent does occur.

178 If Chomsky's theory that« in English, there is no movement into CP in examples in which there is no overt evidence for it is correct, this not only explains the ungrammatically of (18)-(20). but it also implies that examples (21) and (22) should not be analyzed as in (23) and (24), but as in (25) and (26). (25)

[ c p 0 [ c , 0 [ l p w h o [j, [j. [[-PAST][AGR]]

(26)

[cp 0 [c, 0 [lp [not

a

likes John]]]]]]

soul] [jf Cx was ][yp stirring]]]]]

Chomsky (1986:ch.l0) presents further evidence for the S-structure analysis of WH-subjects in situ, which is based on structures containing parasitic gaps. In (27)-(44) I present a body of evidence drawn from the area of adverbial modification which further supports Chomsky's claim.® (27) (28) (29) (30) (31) (32)

John probably saw Mary John probably DID see Mary John probably will see Mary John will probably see Mary John will see Mary, probably Probably, John will see Mary

(33) (34) (35) (36) (37) (38)

*Who£ 0? probably John ej saw e£? *Who£ probably DIDj John ej see e£? »Whoj probably willj John ej see e£? *Who£ willj probably John ej see e£? Whoj willj John ei see e£, probably? »Probably, whoj willj John ej see ej?

(39) (40) (41) (42) (43) (44)

Who probably 8aw Mary? Who probably DID see Mary? Who probably will see Mary? Who will probably see Mary? Who will see Mary, probably? *Probably, who will see Mary?

In examples (27)-(30) we see that probably can occur between the subject and the IP head, and between the IP head and the VP. In (31) and (32) probably is in clause-final and clause-initial position, respectively. Examples (33)-(36), in which probably occurs before or after the C position, are ungrammatical. The ungrammatically of (33) can at least partly be due to the fact that the bare tense does not overtly manifest the non-absorption of [AGR]. The ungrammaticall y of (34)-(36) can only be due to the place of probably, as is clear from a comparison with (37). The conclusion must be that probably cannot occur in positions between the specifier of the CP and the specifier of the IP. The graramaticality of (37) is on a par with the grammatically of (31), in which probably occurs in clause-final position. The ungrammatically of (38) is on a par with the ungrammatically of (44): probably apparently cannot precede clause-initial WH-constituents. The grammatically of (43) parallels the grammaticality of (37) and (31): structures with probably in clause-final position appear to be always grammatical. The grammaticality of (39)-(42) corresponds with the grammaticality of (27)(30). This leads to the conclusion that in these examples probably finds itself between the subject and the IP head, or between the IP head and the TP. And this, in turn, leads to the conclusion that the WH-forms and the IP heads in (39)-(42) are in situ, i.e. have not moved. For, if movement had occurred, (39)-(42) would have been just as ungrammatical as (33)-(36). Thus the grammaticality of (39)(42), as opposed to the ungrammatically of (33)-(36), constitutes additional and independent evidence for Chomsky's theory that there is no WH-movement if the WH-form is the D-structure subject.

179 The adverbial merer can be used in the same clause-internal positions as probably. Thus the use of never instead of probably in (33)-(36) and (39)-(42) leads to identical conclusions. There is a difference between these two adverbials, however. Merer can be topicalized by moving it into the specifier position of the CP in a non-interrogative headclause. When the specifier position of the CP is thus lexically filled by a negative constituent, the head of CP must be lexicalized by movement of an overt IP head. This leads to structures of the type in (10) and (11). repeated here for the sake of convenience. (10) (11)

Never had she seen anything like it Not a soul did we see in the deserted area

The structural position of never and not a soul in (10) and (11) differs from that of initial clause-modifying adverbials. such as probably in (32). Whereas negative topicalization moves constituents into the specifier of the CP. initial clause-modifying adverbials are adjoined to IP. (45) (46) (47) (48)

Last year they tried (*did they try) to salvage the ship Never did they try to salvage the ship I remember that last year they tried (*did they try) to salvage the ship *I remember that never did they try to salvage the ship

It is clear from (45) that last year is not in the specifier position of CP. If it was. the head of CP would have to have an overt lexical filler, contrary to fact. In (46). the highest CP is predicative (cf. Kayne 1982). and has a [+V] specification. When this is the case, negative topicalization fills the specifier of CP and the C must the have a lexical filler, which is provided by movement of an overt I-verb. In (47). the embedded CP functions as an argument in the higher clause, and thus has a [-V] specification. When this is the case, the C cannot be a landing site for an I-verb. and the complementizer that is placed in the head position of the CP. If. in (47). that occupies the C position. last year cannot be in the C position. Thus the grammaticality of (47) is evidence of adjunction of last year to IP. If. in (48), the embedded CP is specified as [-V]. did cannot move into the C position, which is then occupied by that. This entails that negative topicalization is impossible in the clause that is an argument to reaeri>er in (48). since negative topicalization requires an overt [+V] head of the CP. The position that we have arrived at now can be summarized as follows. In Modern English [+V] clauses. [AGR] is absorbed by IP in all cases except in questions and in cases of negative topicalization. When [AGR] in [+V] clauses is not absorbed. it must be lexicalized by movement of an (overt) I-verb. In cases of WH-questions and negative topicalizations in which the questioned or topicalized constituent is the subject, even overt lexicalization of the C position could not show that movement of the subject had occurred. Because in English absorption of [AGR] in [+V] clauses is the rule rather than the exception, such lack of evidence of subject movement is interpreted as absence of subject movement. As a result. [AGR] is also absorbed in cases in which a WH-subject or a negative subject remains in situ.

6. Bare tenses In (18)-(20) we have seen that a bare tense cannot be used to lexicalize a C with a [+V] specification and with a non-absorbed [AGR] feature, because a bare tense is not overt. In the light of (17). it may therefore be assumed that, at a superficial level of analysis, a bare tense is not be interpretable as the head of an IP that has absorbed the [AGR] feature either. If this assumption is cor-

180 rect. and if, as I argued above» a main verb (i.e. a non-I-verb) cannot move into the I position because only items with an inherent tense specification are admitted to that position» the implication is that in clauses which lack an overt I-verb, induction of tense and agreement morphology on the main verb that is governed by the bare tense results in a reinterpretation of S-structure» in which clause-head status is transferred from the non-overt IP head that dominates the tense and [AGR] features, to the finite main verb that absorbs these features. Under these assumptions, the S-structure of (49) provided in (50) — in which the bare tense does not overtly lexicalize the I position -- is reinterpreted as (51). This reinterpretation obviates the necessity for the IP of the S-structure in (50) to be lexicalized by an overt I-verb. Of course it is always possible to use an overt I-verb in structures of the type in (50). This is illustrated in (52). (49) (50)

John saw Mary reinterpreted as:

(51)

VP NP John

V saw [+PAST] [AGR]

NP Mary

NP Mary (52)

John will see Mary

7. The graaaar of DO/DOES/DID Let us compare the sentences in (53) and (54) with those in (55) and (56). In (53) we use supportive DO. If we don't, we get the ungrammatical sentence in (55). In (53), we do not use supportive DO. If we do. we get the ungrammatical sentence in (56). (53) Who did Jessie see? (54) Who saw Bob? (55) *Who Jessie saw? (56) *Who did see Bob (weak stress on did) / Who DID see Bob? There is a difference in degree of ungrammaticality between (55) and (56). Whereas (55) is strongly ungrammatical. (56) is not. As soon as we put some emphasis on did in (56), it becomes perfectly grammatical. Let us see why this is. In accordance with the discussion in section 4, the analysis of (53) is as in (57). in which the [+V] C position must be overtly lexicalized to avoid the violation of the ECP and of (17) that we see in (55). In accordance with the discussion in sections 5 and 6. the analysis of (54) with vho in situ is as in (58). which is a structure of the type in (50). When [[+PAST][AGR]]. which is a non-overt IP head, induces the correct form so« on the main verb see in (58). (58) can be reinterpreted as in (51). (57)

[gp Whoj [(;» did^ [jp Jessie [i» e^ [yp see ej ]]]]]

(58)

C c p 0 [ c , 0 [ l p w ho [j. [ x [[+PAST][AGR]] [yp see Bob]]]]]]

Now suppose we want to stress the IP head in (58). Stress can only be borne by an overt lexical item. The I-verb did has the same tense & agreement specifics-

181 tion as the IF head in (58). and, apart from this tense & agreement specification it is semantically empty. This means that did is semantically equivalent with [[+PAST][AGR]] in (58), from which it differs only because it is overt. This is why did is used when we stress the IF head in (58). Of course, we can use other I-verbs in that position as well, as in Who will see Bob?, but then we add something to the semantics of the IP head in (58). Only the use of supportive DO can keep the semantics of an IP head which dominates a bare tense constant . The question may arise as to why we do not stress the TP head saw in (54). instead of the overt IP head did. as we do in the grammatical variant in (56). This is where Palmer's (1983) distinction between the modal component and the propositional component of the sentence comes in. which, as we saw in section 2. corresponds with the distinction between the IP and the VP. When we stress saw in (54). we stress a lexical choice with propositional implications, for example to distinguish between SKK(x.y) and HBAK(x.y). But when we stress did in (56). we stress the speaker's desire to be supplied with information allowing him to assign a positive truth value to SEE(x.y). Thus the use of stressed did emphasizes truth-conditional aspects, which are characteristically expressed by the IP as the syntactic correlate of Palmer's modal component. (59) (60)

John didn't see Mary Who didn't see Bob?

The negative variants of (49) and (54) are (59) and (60). The use of supportive DO in these negative variants can be attributed to the fact that not. and. even more characteristically, enclitic n't. obligatorily attach to an overt head. The reason why this must be an overt IP head, and not to an overt VP head must be sought in the truth-conditional rather than propositional quality of the negation in these examples. The fact that in V2 languages negatives like German nieht and Dutch niet attach to finite main verbs does not clash with these observations: as we saw in section 2. the finite main verb in V2 languages moves into the I position, so that nieht. niet. etc. in the corresponding constructions in V2 languages can also be analyzed as attaching to the IP. Finally, an answer is due to the question in the title of this article. The answer is that the grammar of English requires no special rule or rules for the grammar of supportive DO. The only information that the grammar of English needs for the correct deployment of do/does/did in finite clauses is the following: 1) these forms are overt I-verbs: 2) they are semantically empty apart from their inherent tense specifications; 3) do/does have the same tense specification as the [-PAST] tense, and did as the [+PAST] tense; 4) does is compatible with a third-person-singular [AGR] specification, and do with all other [AGR] specifications .

8. ConeInsion I have discussed some general phenomena in the grammar of English that relate to the difference between the grammar of I-verbs and main verbs. The most important of these were [AGR] absorption, overtness requirements for non-absorbed C positions, in situ analysis of WH-subjects and topicalized negative subjects, re interprétât ion of structures with a bare tense in the IP, and negation and emphasis as processes affecting the IP as the syntactic correlate of Palmer's modal component. An analysis of these phenomena along lines discussed in this article leads automatically to the correct deployment of supportive DO in finite clauses, given the categorization of do/does/did as I-verbs which are overt, but semantically identical with the bare tenses.

182 Hotes *

The research on which this article is based was part of the Leiden University research project 'Wordorder and Syntagmatic and Paradigmatic Structure', financed by the Dutch Ministry of Education. Thanks are due to Frits Beukema, Peter Coopmans, Marcel den Dikken and Teun Hoekstra for comments and discussion. The responsibility for the version of the article presented here is entirely mine.

1. Other approaches are conceivable. An alternative to the Chomsky (1986) approach would be to base-generate the subject NP in the specifier position of the VP, and to restrict government to c-command within the X' level. If this approach is adopted, the assignment of nominative Case to the subject NP would be effected through structural government by an outside Case assigner, which would mean that the subject moves to the specifier position of the IP, and gets its Case from the C position on its left. 2. For example, in combinations such as had better, had rather, had sooner, which, for lack of space, I leave undiscussed, and in structures of the type Yon hare a daughter, in which hare can be classified either as a main verb or as an I-verb. Cf. Do yon hare a daughter/Hove you a daughter? 3. In Rigter (1982, 1986, forthcoming) I argue that three tenses must be distinguished: 1) a [-PAST] domain tense; 2) a [+PAST] domain tense; 3) a [+PAST] domain-shift tense. The third of these is responsible for the 'sequence-oftenses' phenomenon. Past-tense epistemic/deontic I-verbs such as m a t . night do not occur in contexts in which a [+PAST] domain tense is used, but they do occur in contexts in which a [+PAST] domain-shift tense is used (cf. *I might «as allowed to) use her typewriter / She said that I might use her typewriter; *John must (had to) go / Father said that John vast go). Similar restrictions are not found in the class of main verbs. Idiosyncratic tense restrictions on I-verbs can be understood if the lexical entries for I-verbs have inherent tense specifications. 4. The phenomenon pointed out in (17) should receive a principled explanation. Rather than indulging in handwaving in a footnote, I will assume without discussion that the principle(s) accounting for (17) applies(/y) at S-structure. 5. The evidence in (27)-(44) was developed in close cooperation with Teun Hoekstra, who also drew my attention to Chomsky's in situ analysis of WH-subjects. 6. This difference had not yet materialized in early Modern English. Consider Shakespeare's Julius Caesar, III, 2, 105-106: I speak not to disprove what Brutus spoke / But here I am to speak what I do know. In this example, the finite main verb speak has moved into the I position, which is clear from the fact that it is followed by not. This indicates that the Modern English difference between I-verbs and main verbs had not yet emerged. In this respect Shakespeare's English still resembled a V2 language. Yet, the sequence here I am shows that here is not in the specifier position of the CP, because that would have resulted in lexicalization of the C by am. Thus Shakespeare's English differed from V2 languages in that [AGR] absorption was already in evidence. While the language was thus in the balance between T2 and non-V2, the verb do — which is weak-stressed in the context in which the example occurs — could already be used as a semantically empty verb to lexicalize the tense in the I position, even before the emergence of I-verbs as a separate lexical class of IP heads.

183 Reference» CHOMSKY, N. 1986 Barriers. Cambridge, Mass., MIT Press DE HAAN. G. & F. WEERMAN 1986 Finiteness and verb fronting in Frisian, in Haider & Prinzhorn (eds), p. 77-110 HAIDER, H. & M. PRINZHORN (eds) 1986 Verb second phenomena in Germanic languages. Dordrecht, For is HUDDLESTON, R. 1976 Some theoretical issues in the description of the English verb, in Lingua 40. p. 331-83 KAYNE, R. 1982 Predicates and arguments, verbs and nouns. Paper given at the GLOW colloquium in Paris LANGACKER, R. W. 1974 Movement rules in functional perspective, in Language 50, p. 630-664 PALMER, F. 1983 Semantic explanations for the syntax of English modals, in F. Heny & B. Richards (eds), Linguistic categories: auxiliaries and related pussies, Dordrecht, Reidel, Vol. 2, p. 205-17 RIGTER. G. H. 1982 Intensional domains and the use of tense, perfect and modals in English, in Journal of Semantics 1. p. 95-145 1986 Focus matters, in T. Lo Cascio & C. Tet (eds). Temporal structure in sentence and diseonrse. Dordrecht. Foris. p. 99-132 forthcoming English tenses, auxiliaries and verbs, and the construction of chronologies in discourse domains, in V. Ehrich & H. Vater (eds). Teaporalität. Tübingen, Niemeyer Verlag WEERMAN, F. 1986 The different faces of verb second. University of Utrecht

Ineke Schuurman

Incorporation in the Groningen dialect 1. Introduction* In the Groningen dialect, spoken in the northeastern Netherlands, constructions with incorporation are very often used. (1)

(te) k o w i e drinken; (te) bouk lezen (to) coffee drink; (to) book read 'to drink coffee'; 'to read a book'

Evers (1984) argues that in all natural languages incorporation only takes place when the verb has undergone V-raising and is marked as [+inf] [-te). But, in spite of his argument, in the Groningen dialect non-verbal elements may be incorporated in every non-finite action verb (Schuurman and Wierenga (1986)). (2) (3) (4) (5)

K heurde, dat zai hom bouk lezen luit 1 I heard that she him book read made 'I heard that she made him read a book' K zag, dat e zat te eerabbels schillen 2 I saw that he sat to potatoes peel 'I saw him sitting peeling potatoes' K wol, dat e ophuil te bloumen plukken I wanted that he stopped to flowers pick 'I wanted him to stop picking flowers' K wait, dat e kezienen vaarfd het I know that he window-frames painted has 'I know that he has painted window-frames'

V-raising, [+inf] [-te] V-raising, [+inf] [+te] Extraposition, [+inf] [+te] V-raising, [-inf]

It is not at all clear whether the process of incorporation takes place in the morphological or in the syntactic component of the grammar. One point of view is that incorporation is a morphological process, albeit the most nearly syntactic of them (Mithun (1984)), another point of view is that incorporation is in fact a syntactic process (Sadock (1985), (1986)). Discussions on this topic are nearly all based on noun-incorporation as it is found in, for example, Inuit and Indian languages. It can be shown, however, that incorporation in the Groningen dialect behaves quite differently in several respects. In the first part of the next section (2.1.) I will examine those aspects that, according to Mithun 3 , favour a morphological approach in order to establish whether they apply in our case. In the second part (2.2.), some other, supposedly morphological, features will be considered. In section 3., arguments for a syntactic approach will be given, partly based on Sadock's analysis. Finally, in section 4., the findings of the preceding sections will be evaluated. 2. Morphology 2.1. Mithun's arguments Mithun (1984) gives the morphological process.

following

arguments

for

treating

incorporation as

a

186 A n incorporated n o u n

1) 2) 3) A) 5) 6) 7)

does not refer, is not m a r k e d for definiteness, is not m a r k e d for number, does not co-occur w i t h demonstratives, does not co-occur w i t h numerals, loses its individual syntactic function, and loses its individual semantic function.

Referring capacity - Incorporated nouns do not refer to specific entities according to Mithun. Seuren (1973) makes the same point, claiming that such elements invariably receive a generic interpretation. In the G r o n i n g e n dialect, however, the following sentences are correct' (see section 4). (6)

(7)

N o u m o t je ophollen te Pieter Moatje pesten Now y o u should stop to Pieter Moatje tease 'Now you should stop teasing Pieter Moatje' En ast n o u nog aln moal lapst te ons kiender ploagen... A n d if-you n o w still one time dare to our children tease... 'And if y o u once more dare to tease our children...'

Definiteness - Articles, both definite and indefinite, are never involved incorporation. Y e t constructions of this kind are frequently interpreted as they contain a n indefinite article (cf (9)).

in if

(8) *Hai zat te t / n bouk lezen He sat to the / a book read 'He w a s sitting, reading the / a book' (9) Hai zat te bouk lezen 'He w a s sitting, reading a book' Mithun expects incorporated elements not to be m a r k e d for definiteness precisely because they fail to have referring capacity. In the Groningen d i a l e c t such elements are indeed not marked for this feature, but they sometimes are nevertheless able to refer (as is shown in (6) and (7)). Number - Mithun's argumentation w i t h respect to number is the same as that for definiteness. Incorporated elements are not marked for number because they do not refer. A s w e have seen before incorporated elements in the G r o n i n g e n dialect can refer and indeed do occur meaningful w i t h plural and singular markings. (10)

(11)

Hai zat te kraant lezen He sat to newspaper read 'He w a s sitting, reading a newspaper' Hai zat te k r a a n t e n lezen 'He was sitting, reading newspapers'

Demonstratives - Incorporated elements may co-occur w i t h demonstratives Groningen dialect, contrary to w h a t M i t h u n predicts. (12)

E n ast n o u nog a l n moal lapst te dei kiender ploagen... A n d if-you n o w still one time dare to those children tease... 'And if y o u once more dare to tease those children...'

Numerals - In our dialect numerals may be incorporated as well. (13)

Zai zat bie t roam te drei eerabbels schillen She sat at the w i n d o w to three potatoes peel 'She w a s sitting at the window, peeling three potatoes'

in the

187 Individual syntactic function - Mithun argues that incorporated elements lose their individual syntactic function. As far as the G r o n i n g e n dialect is concerned this statement is not correct. It does not make any difference w h e t h e r or n o t any of the arguments of the verb is realized as a n incorporated element. They have the same features in both cases. Individual semantic function - Constructions w i t h incorporated elements tend to get a lexicalized meaning of their own, varying f r o m cases in w h i c h the incorporated element simply modifies the verb (to N-V is a special k i n d of Ving) to cases in w h i c h the meaning of the whole is no longer built up in a compositional w a y , resulting in a complete 'new' m e a n i n g (Mithun (1984)). Something like this is not the case in the constructions under consideration, however. Their meaning is compositionally built u p = and hence in fact nearly synonymous w i t h the construction without incorporation. (14) (15)

Hai zat te bouk lezen Hai zat n bouk te lezen 'He was sitting, reading a book'

2.2. Other

arguments

Of Mithun's arguments in favour of a morphological a p p r o a c h only the one concerning definiteness acts in the Groningen dialect more or less the w a y it is supposed to. In all other respects the Groningen dialect contradicts her arguments. In A and B some other features of the construction are considered.

A. A g a i n s t morphology Diminutives - In the Groningen dialect, diminutives m a y be incorporated. V a n Zonneveld (1983) argues that in cases of morphological incorporation a diminutive is only possible if a non-diminutive is excluded. So (16)a is correct because (16)b is bad, but (17)a is bad, precisely because (17)b is good. Exceptions to this rule are rare. (16)a vrouwtjesaap (17)a *diertjestuin

(16)b *vrouwaap (17)b dierentuin

'female monkey' 'zoo'

Such a restriction does n o t hold in the Groningen dialect. struction exist at the same time. (18)

Both types of

con-

Hai zat te bouk / boukje lezen 'He w a s sitting, reading a book / a little book'

Modifiers - It is possible in the Groningen dialect to incorporate a n o u n together w i t h a modifier. This is not common in morphological constructions, in w h i c h a modifier w i l l usually be realized outside, because braching elements can not be incorporated. (19)

(20)

Hai ston doar te dikke leugens verteilen He stood there to big lies tell 'He stood there telling big lies' Zai zat aaltied bie t roam te dikke swaarde w o l l e n k o u z e n braaien She sat always at the w i n d o w to thick black w o o l l e n stockings k n i t 'She used to sit at the window, knitting thick black w o o l l e n stockings'

A series of modifiers is not excluded either, as (20) shows. Though it is not allowed to place a modifier in front of the incorporation construction in the

188 way it is obligatorily done in Inuit languages, in the Groningen dialect there is the possibility to have an external modifier in postposition (see below). More elements - It is quite common in the Groningen dialect to have more than one element incorporated, although it is not necessary at all to incorporate every argument (see (46) below). (21)

Hest volk genog te heu in schuur bringen? Have-you people enough to hay in barn bring? 'Do you have people enough for bringing hay in (the) barn7'

B. In favour of morphology Shape of the incorporated elements - Articles do not occur in incorporated elements, but in spite of this in the Groningen dialect full NP's may occur (for example in constructions like (7) and (12)). Usually, however, N's and N's seem to be affected. In predicative PP's (in which the P is the head) too articles are never realized, but nevertheless a PP may appear, as in (21). This is only possible in those constructions where more than one element is incorporated. If a PP should be the only candidate for incorporation, it has to take the shape of a N (or may be a N)• (22)

Hai ston te (*op) sloatje kaauwen He stood to (at) tobacco chew 'He was chewing tobacco'

If the argument in question does not undergo incorporation, the whole PP must be realized®. (23) Hai ston op n sloatje te kaauwen (24) *Hai ston n sloatje te kaauwen Shape of negation - In the Groningen dialect the negative nait 'not' in combination with the indefinite a in 'a* forms gain 'not a', which can only be used to negate nouns. Verbs are negated with nait. It can be shown that the whole construction, verb plus incorporated element, behaves like a verb with respect to the choice of the negation element (cf.(25)). The incorporated element as such cannot be negated (26), whereas a non-incorporated element of course can (27). (25) (26) (27)

Zai zat nait / *gain te bouk lezen She sat not / not a to book read 'She was not sitting reading a book' Zai zat te *nait / *gain bouk lezen Zai zat *nait / gain bouk te lezen 'She was sitting, reading not a book'

In (27) bouk is treated as non-incorporated (but see note 8). Action verbs - A restriction has to be formulated with respect to the kind of verbs appearing in constructions with incorporated elements. The expressions (28)/(30), for example, are ungrammatical. (28) *te hoezen bezitten (29) *te senten hebben (30) *te peerden hollen (31) te hoezen verkopen (32) te senten uutgeven (33) te peerden fokken

'to 'to 'to 'to 'to 'to

own houses' have money' keep horses' sell houses' spend money' breed horses'

189 The latter verbs are action-verbs, whereas the former are not. This seems to be the reason that these constructions differ with respect to grammaticality. 2.3. Preliminary results Most of Mithun's points favouring a morphological approach of incorporation do not favour such an approach for the Groningen dialect. Only her point concerning definiteness more or less applies to these cases. Furthermore, there are the arguments summed up above ^B), of which the restriction to action-verbs is the least important. Recall that in all other respects the constructions do not behave the way one expects them to, from a morphological point of view.

3. Syntax 3.1. Sadock's arguments Sadock, partly in response to Mithun, argues that at least in Inuit incorporated elements do have syntactic status. In his opinion this is so because such elements 1) can be referred to, 2) can co-occur with external modifiers, and 3) must be arguments. The syntactic status of a part of the incorporation construction makes a morphological treatment of the whole construction impossible. Reference - It is not of vital importance whether or not the incorporation construction introduces a discourse topic. In both cases the incorporated element is a possible antecedent and may be referred to with ordinary anaphoric devices. "We do not expect pieces of words to have independent referential or discourse properties." (Sadock (1986)). Since genuine lexical items in a non-polysynthetic language count as anaphoric islands, even for pragmatically sanctioned anaphora, as Sadock (1986) observed, a syntactic approach to our construction is plausible (see (34)/ (36) below). Note that reference will be impossible if the anaphor does not agree in number and gender with its incorporated antecedent (34a). (34)

K hoop, dat haile femilie blift te eerbeien eten, aans kommen ze aal om haals I hope, that whole family stays to strawberries eat, otherwise come they all about the neck 'I hope that the whole family stays and eats strawberries, otherwise they go bad' (34)a *K hoop, dat e blift te eerbeien eten, aans komt aal om haals 'I hope that he stays and eats strawberries, otherwise it goes bad' The preferred reading of (34) is the one in which ze is anaphorically related to eerbeien, not to (haile) femilie. Reference does not suppose pronouns, however. (35)

(36)

K wait best dat moeke tegenwoordeg al om drei uur zit te eerabbels schillen, want dei kringen willen ja hoast nait goar I know well that mother nowadays already at three o'clock sits to potatoes peel, because those terrible things will almost not be done 'I do know that nowadays mother is peeling potatoes already at three o'clock because those terrible things are hard to get well-done' T wicht ligt op bfer te tillevisie kieken. Zai har heur net n neie koft The girl lies on bed to television watch. She had her just a new bought 'The girl is lying on her bed watching television. She has just bought a new one'

190 Dei kringen in (35) refers to eerabbels, n neie in (36) to tillevisie. In all these cases the antecedent is an incorporated element, not unusual at all in the Groningen dialect. External modifiers - Incorporated elements may be externally modified. Such modifiers always occur in postposition. Some syntactic process has to be involved. (37)

(38)

Wils wel leuven dat e haile doagen niks aans dut as bie toavel zitten te foto's ienplakken van ons leste vekaansle7 Want-you well believe that he whole days nothing else does but at table sit to photo's paste of our latest holidays? 'Would you believe that he does not do anything else but pasting the photo's of our latest holidays?' Hai komt al joaren bie ons te haardbroden kopen, haarde en zachte He comes already years to us to haardbroden buy, hard and soft 'He comes to us to buy haardbroden, hard (ones) and soft (ones)'

In (37) van ons leste vekaansie, a PP, modifies f oto' s. in (38), on the other hand, the modifier is an A (haarde en zachte) belonging to haardbroden (a special kind of bread). I consider the modifier in (38) as an apposition (Klein (1977)) and consequently as part of the syntactic component. However, I am aware of the possibility to consider such a modification as a kind of afterthought as in (38)a, where the modifier is an NP. (38)a Hai komt al joaren bie ons te haardbroden kopen, haarden en zachten 'He comes to us to buy haardbroden, hard ones and soft ones' Adjuncts - Adjuncts are not possible candidates for incorporation, only complements are. Therefore, the following constructions are ungrammatical. (39) *te paark eten (AO) *te elk in aine dansen

'to eat in a park' 'to dance for everyone'

It is not easy to tell which preposition should be used in translating these examples. In this respect the situation in for example (22) is much clearer, precisely because in this construction there is a tight connection between the preposition 02 and the verb kaauwen (the PP is in fact a (possible) complement of the verb), which implies that the preposition is easily recovered. 3.2. Other arguments Sadock's arguments for assigning a syntactic status to incorporated nouns in Inuit seem to hold for our constructions as well. There is just one point concerning the ability of incorporated elements to be referred to which does not substantiate his claim. Sadock (1985) states that such a referential element may even be definite. But above we have argued that in the Groningen dialect the incorporated element is never (in)definite in the strict sense. That is to say, it is never marked for this feature. A. Pro syntax; more arguments Anaphora - Incorporated elements in the Groningen dialect may contain anaphoric or deictic expressions, cf. (41). (41)

te foto's van ons leste vekaansie ienplakken 'to paste photo's of our latest holidays'

191 Neither ons nor leste can be completely Interpreted without further information and, as a consequence, the whole construction cannot be argued to be based on a lexical process. A syntactic approach seems to be preferable because in compounds only in sporadic cases does the first member get a deictic interpretation 7 . Discontinuous internal elements - Between an incorporated element and the governing verb other expressions may intervene, without effecting ungrammaticality. (42)

(43)

Ons jongs eten zo'n bult tegenwoordeg, dak nou al wel hfen goan mag te stoeten begunnen te smeren, as zai nog op b&r liggen Our sons eat so much nowadays, that-I now already well to work go may to bread begin to spread, when they still in bed are 'Nowadays our sons eat so much that I have to start making their bread when they are still in bed' T splet pa tonen oet, dat opoe aingoal bie ons zit k o w i e te drinken It regrets father to toes out, that grandmother all the time at us sits coffee to drink 'Father regrets it very much, that grandmother all the time sits at our place to drink coffee'

In the first sentence (42) the incorporated element stoeten belongs to smeren, in the second one (43) k o w i e to drinken. In the incorporation constructions in the preceding sections the incorporated elements were always located directly to the left of the governing verb, without intervening material. On account of this one should expect constructions like the following. (44) (45)

te begunnen te stoeten smeren te k o w i e drinken

Both types of constructions are equally correct. That in (42) and (43) incorporation is involved can be shown by the behaviour with respect to determiners, negation, Adjuncts, etcetera®. In these particular cases it is easy to show that stoeten and k o w i e are incorporated, stoeten because it is located between the infinitive marker te and the verb begunnen, k o w i e because zitten 'to sit' triggers V-raising, not Extraposition of its verbal complement. Another type of discontinuous elements occurs arguments are incorporated and others are not. (46)

in constructions

in which

some

Zits dei jong nou alweer te sokken braaien? Sit-you that boy now again to socks knit7 'Are you again knitting socks for that boy?'

3.3. Preliminary results The phenomena discussed in section 3 all argue for a syntactic approach of incorporated elements in the Groningen dialect. The more so as the points against morphology can be seen as pro syntax. 4. What kind of phenomenon? In my opinion none of the observations in the sections 2 and 3 is of decisive importance with respect to the question whether the incorporation construction

192 in the Groningen dialect should be understood as being morphological or syntactic. Only when taken together do they support the syntactic approach. It cannot be denied, however, that the construction has several typically morphological characteristics, such as being unmarked for determiners, sometimes missing prepositions, and the shape of negation. It is my claim that the construction in our dialect with incorporated elements behaves like a real phrase and that therefore the node dominating it is a V. Only in this way can the many syntactic characteristics of the construction be accounted for. On the other hand, none of the morphological features can be explained in this way. Nor will it be possible to account for the observation (see Schuurman and Wierenga (1986)) that incorporated elements are interpreted in a not very specific way (cf. (47)), do not bear contrastive accent (48), and cannot be questioned (49)®. These sentences are considered very bizarre. Note that they will be correct if the elements eerabbels. bonen/spruten, and heu are not incorporated. But why7 (47) *Moeke zat al om drei uur te eerabbels schillen, omreden dat dei s'oavends in n sloatje mozzen Mother sat already at three o'clock to potatoes peel, because those in the evening in a salad must 'Mother already sat to peel potatoes at three o'clock, because they must be in a salad in the evening' (48) *Hai zat nait te BONEN doppen, moar te SPRUTEN doppen He sat not to beans shell, but to sprouts shell 'He was not shelling beans, but shelling sprouts' (49) *Woar begunnen zai morgen te heu in bringen? cf. (21) 'Where do they tomorrow begin to bring the hay in?' The following description of the characteristic properties of the incorporation construction in the Groningen dialect illustrates the fact that this construction is nearly synonymous with the construction without incorporation. 4.1. Characteristics of incorporation in the Groningen dialect The productive coexistence of a construction with incorporation and one without leads one to expect a functional distinction between them, which in my opinion there is. The construction with incorporation marks an expression in such a way that a more or less fixed pattern is expressed. Without incorporation, however, a more incidental state of affairs is expressed. Therefore these sentences will get a pattern reading and an incidental reading respectively. The relation between these terms and the more familiar terms gnomic (generic/habitual) reading and episodic reading is depicted in figure 1 (cf. Carlson (1982), Goldsmith and Woisetschlaeger (1982))10. kind of description:

fig. 1

phenomenal gnomic kind of reading:

episodic pattern

structural

-

+

+

-

+

+

The term incidental reading is not found in fig. 1. I suggest this term, instead

193 of episodic reading to express the contrast with respect to the structural description11.

the

pattern

reading

with

The particular properties of our construction can be explained by means of (50). (50)

Zai zat bie t roam te drei eerabbels schillen

cf. (13)

The expression (te) drei eerabbels schillen may be used even if not always exactly three potatoes are peeled (sometimes it may be two big or four small ones), as long as the pattern is three. In this respect, a gnomic reading would be appropriate (or better a habitual one, since a generic reading is excluded (see (50) where drei 'three' makes such a reading impossible)). The moment one actually sees that the number of potatoes is not three, but four or two, the incorporation construction cannot be used, neither with vaaier 'four' or twli 'two', for this is not in conformity with the fixed pattern, nor with drei 'three', for this is not in conformity with the concrete situation. It is the latter requirement concerning the concrete situation which makes a gnomic reading not appropriate. In other words, an incorporation construction can be defined as a construction describing a concrete situation as the realization of a type. It is the 'structural description'-aspect of the construction that prevents the incorporated elements from being referred to in too specific a way, from bearing contrastive accent and from being questioned. It also prevents them from being negated, one of the morphological features of the construction (see 2.2.B above). For a fixed pattern is difficult to combine with this type of relief, which is more appropriate when combined with the more incidental reading of the construction without incorporation. Notes *

I am indebted to Reineke Bok, Dicky Gilbers, Joop Houtman, Theo Janssen, Jan Jullens, Fronique Oosterhof, Annet Wierenga, Ron van Zonneveld and Frans Zwarts for substantial discussions on incorporation and comments on earlier drafts of this paper. Furthermore I would like to thank Sharon Parry, Frans Zwarts and their granny for improving my English and the editors of this volume for their advice.

1. In (2) bouk is an incorporated element. If it were not, the determiners t 'the' or n 'a' would have to occur. 2. That in (3) and (4) V-raising and Extraposition constructions are involved can be shown by adding Adjuncts as veur elk in ain 'for everyone' (see Schuurman and Wierenga (1986)). 3. In this paper I will abstract from arguments which are not relevant for the Groningen dialect, for example those related to Case-assignment. 4. Most examples involve te-constructions, because these clearly show whether there is an incorporated element or not. But note that incorporation is also possible in constructions without te. 5. I am aware of the existence of at least one construction which is not regularly formed, namely eerabbel ruden 'to dig potatoes'. When the object is not incorporated one should use the plural form eerabbels• 6. In LFG this will cause no problems, because verbs are selecting functions, not categories. The function of sloatje (22) is the same as that of op n sloatie (23) (see Bresnan (1982)).

194 7. Some exeptions I am aware of are gisteravond morgenochtend 'to-morrow morning'.

'yesterday-evening' and

8. Some constructions are structurally ambiguous. These therefore have two readings, one with incorporation and one without (see Schuurman and Wierenga (1986)). 9. I abstract from eventual formal grounds for this impossibility. 10.A phenomenal description of the world describes what things happen in the world, a structural description describes how the world is made that such things may happen in it (Goldsmith and Woisetschlaeger (1982)). 11.1 am not aware of this kind of reading in other languages. In the Groningen dialect, however, the aan het-construction, a kind of progressive, can be interpreted analogously. I will return to this matter elsewhere.

References BRESNAN, J. (ed) 1982 The mental representation of grammatical relations, Cambridge/London CARLSON, G.N. 1982 Generic terms and generic sentences, in Journal of Philosophical Logic 11, p. 145-181 EVERS, A. 1984 Are V-headed composita dissolved by the Inflection Factor?, in G.J. de Haan, M. Trommelen and W. Zonneveld (eds) Van periferie naar kern. Dordrecht, p. 63-71 GOLDSMITH, J. and E. WOISETSCHLAEGER 1982 The logic of the English progressive, in Linguistic Inquiry 13-1, p. 79-89 KLEIN, M. 1977 Appositionele constructies in het Nederlands. Nijmegen MITHUN, M. 1984 The evolution of noun incorporation, in Language 60, p. 847-894 SADOCK, J.M. 1985 Autolexical syntax: a proposal for the treatment of noun-incorporation and similar phenomena, in Natural Language and Linguistic Theory 3, p. 379-439 1986 Some notes on noun-incorporation, in Language 62, p. 19-31 SCHUURMAN, I. and A. WIERENGA 1986 Syntactische nomen-incorporatie bij infinitieven en deelwoorden, in C. Hoppenbrouwers, J. Houtman, I. Schuurman and F. Zwarts (eds) Proeven van taalwetenschap (Tabu-special), Groningen, p. 339-350 SEUREN, P. 1975 Referential constraints on lexical items, in E.L. Keenan (ed) Formal semantics of natural language, Cambridge, p. 84-99 ZONNEVELD, R.M. VAN 1983 Affix-grammatika• Een onderzoek naar woordvorming in het Nederlands. Groningen

Annemarie Slootweg

Word stress and higher level prosodies 0. Summary* Incorporating fine prosody control into linguistically oriented text-to-speech systems calls for a more precise description of the interface of the phonological and phonetic modules of generative grammar than is available as yet. This paper deals with the phonetic realisation of 'primary' versus 'secondary' stress in monomorphemic and compound words as assigned by the phonology of Dutch, and with the effect of sentence accent on the acoustic realisation of syllables carrying lexical stress in simplex words. Measurements show a linear phonetic factor interacting with phonological variables to determine the acoustic realisation of lexical stress.l) Also, the important acoustic correlates of stress turn out to react differently to phonological prominence on higher prosodic levels. The unit under investigation is the syllable with 'main' stress, the designated terminal element (DTE for short) in the string of syllables forming the word. This approach reflects our view of the syllable as the domain in which stress is operative, rather than the vowel within it. 1. Word stress within prosodic structure Most phonetic work on stress has involved the search for the acoustic manifestations of lexical stress, and the question of the perceptual 'cue value' of each of a number of acoustic parameters which have come to be associated with stress. Stressed syllables differ from unstressed ones in that they show longer duration, specific pitch movements and more intensity. The difference between stressed and unstressed syllables within one word has been studied extensively. (See Niemi (1984) for an overview.) However, we know of no systematic investigation as to the influence of prosodic levels higher than the word on the realisation of lexical stress. Whenever two words are joined to form a compound, one turns out to be stronger, and the other less prominent. The recently developed 'metrical' framework in phonology captures this generalization by always grouping constituents into binary pairs, within which the strong-weak relation is formally defined. Thus, it is impossible to end up with two 'primary' stresses. Under the shorthand name of 'status', we studied the acoustic difference between identical syllables in the weak parts of compounds or as the DTE of simplex words. This material was embedded in the higher-level prosodic structure known as the phonological phrase. Within this type of word group, there also exists a strong-weak relation between constituents. In accord with Liberman & Prince's (1977) Lexical Category Prominence Rule, the second part of Dutch phrases is more prominent than the first. The rule operates without regard to the internal structure of the constituents, so that the weak element in the compound retains its status of being weaker than the first element. In the experiment reported here, the phonological phrase is the context in which compounds and simplex words appear and within which phonetic conditions were rigidly controlled. The highest level where prosodic prominence relations occur is that of the sentence. Sentence accent distinguishes strong and weak (more prominent or less so) intonational phrases, on pragmatic and contextual bases. The effect of sentence accent on syllables of either status is the second issue that was studied.

196 Traditionally, sentence accent is primarily associated with specific prominence lending pitch movements, while acoustic features like intensity and syllable duration are affected to a lesser degree. In the context of the present experiment, there were two questions to be answered: 1) What is the difference, in acoustic terms, between a syllable assigned phonological main stress and the same syllable bearing 'secondary' stress in the weaker phonological environment formed by the second part of a compound? 2) How does the presence or absence of sentence accent affect these characteristics? To answer these questions, acoustic measurements were performed on some 500 syllables, pronounced in controlled phonological and phonetic contexts. Along with providing insights into the acoustic effects of both status and sentence accent, this setting disclosed the influence of a linear phonetic parameter, namely the position of the syllable within the word, on the acoustic data. 2.0. The experiment: material Speech material was devised to study the influence of status and accent, and to introduce a third independent variable, the position of the lexical stress within the word. The material was based on fifteen tri-syllabic target words with lexical stress on the first, middle or last syllable, equally divided. In the figures below, the phonetic variable of position of the target syllable in the word is referred to as 'type n', the numeral indicating first, second or third position. To minimize acoustic differences caused by syllable make-up, we strove for uniformity in the consonant-vowel structures. Accordingly, (almost) all syllables consisted of one consonant and one long vowel. Five vowels occurred, balanced across word position types. The words are listed in the Appendix. To form two kinds of word groups in which the target word would be either strong or weak, we embedded that word in an adjective-noun phrase, where the target would be strong, and in a nominal compound, where it was the second element and, thus, weak. Thus, the syllable could end up as the DTE of a monomorphemic word ('main stress') or in the weaker part of a compound ('secondary stress'). For example, (the target syllable is the middle syllable of 'kimono'): (DTE, strong): 'die rode kimono'(that red kimono) (secondary stress): 'die mode-kimono'(that fashion-kimono) As the examples show, the phonetic context within the phrases was kept as similar as possible. The number of syllables preceding the stressed one was kept constant, through the introduction of the adjective. The CV-structure of all syllables within the phrase was the same, with identical vocalic segments. In order to study the effect of the presence or absence of sentence accent, we used two different sentences of which the phonological phrase was a part. One sentence prompted a reading with sentence accent on the phrase in question. In the other sentence the phrase was relatively unimportant; it was easy to pronounce it without sentence accent. The sentences were: (accent): 'Vergeet niet die in je toespraak te noemen.' (Don't forget to mention that in your speech.) (no accent): 'Ik geef NIKS om die die ze daar aanprijzen.' (I don't care AT ALL for that they're selling there.) Both sentences were to be pronounced as one intonational block.

2.1. Speakers, instructions and the recording session Speakers were 8 males, who pronounced the set of sentences presented to them in

197 random order. Each speaker produced 60 utterances: the product of 15 words x 2 status possibilities x 2 sentence accent conditions. After some training, they were all able to produce the accent patterns desired. This does not mean they delivered exact replicas of an example: during the recording session, performance was checked on the basis of perceptual equivalence. Recordings were made in a soundproofed studio with semi-professional equipment. 2.2. Measurements and computed values Acoustic measurements were performed on the DTE syllables with and without accent, both in compounds and phrases. 2) For each syllable, a number of values to capture pitch, duration and intensity features were obtained. From these, three new variables were computed, taking into account speakers' speech rate, and melodic and dynamic speech characteristics. The new duration variable expressed syllable duration as a proportion of the sum of the four unreduced syllabes in the compound or comparable phrase. Syllable intensity was expressed in dB above the intensity baseline. This was defined as the lowest mean value observed on unstressed syllables, for each speaker and condition separately. To arrive at a new pitch variable, the aim was to express pitch changes rather than absolute values. These changes were then to be related to speakers' melodic ranges. Pitch range was defined as the difference between highest pitch observed on stressed accented syllables and lowest pitch as found on unstressed syllables in a condition without accent, both mean values for each speaker and condition separately. The pitch change was chosen as the largest pitch movement occurring on a given syllable, and expressed as a percentage of the pitch range. 3) While using the variable names of 'pitch', 'duration' and 'intensity', reference is made to the variables defined as above. Mean values were computed for these new variables, to gain an overview of the acoustic differences in each linguistic condition. Analyses of variance served to determine which of the independent linguistic variables (status, accent and the linear phonetic factor described above) most influenced the dependent acoustic variables. 3.0. Results The results for the three variables are presented separately. Mean values for each variable are in the Appendix, along with standard deviations. An overview of the results of the analyses of variance is in the Summary Table of Effects, given in section 4.

198 3.1. Proportional syllable duration The results for the duration parameter are presented in figure 1. The most striking effect on the durational parameter can be seen on the horizontal axis. As the DTE is situated further back in the target word, its proportional duration increases from 25% via 29% to 30%. This f u l l y agrees with the observations in Nooteboom (1972), where vowel duration is a function of the number of syllables to follow in the word. While that study employed nonsensesequences of identical syllables, the present results show that the same holds for proportional duration of syllables in normal speech: F(2,384)=63.73,p