Linguistics in Netherlands [Reprint 2021 ed.] 9783112419502, 9783112419496

Citation preview

LINGUISTICS IN THE NETHERLANDS 1986

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

Still available Linguistics in the Netherlands 1977-1979 Wim Zonneveld and Fred Weerman (eds) Dfl. 20,—/US $10. Linguistics in the Netherlands 1983 HansBennisand W.U.S. van Lessen Kloeke(eds) Dfl.25,—/US$12.50. Linguistics in the Netherlands 1985 Hans Bennis and Frits Beukema (eds) Dfl. 38,-/US $19.

LINGUISTICS IN THE

NETHERLANDS L> 1 M l i B n L ^ l l t l J w 1986 Frits Beukema Aafke Hulk (eds)

1986

FORIS PUBLICATIONS Dordrecht - Holland/Riverton - U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Sole distributor for the U.S.A. and Canada: Foris Publications U.S.A. P.O. Box C-50 Riverton N.J. 08077 U.S.A.

CIP-DATA Linguistics Linguistics in the Netherlands 1986 / Frits Beukema, Aafke Hulk (eds). - Dordrecht [etc.] : Foris. - (AVT Publications ; 2) Papers presented at the 17th annual meeting of the 'Algemene Vereniging voor Taalwetenschap' (Linguistic Society of the Netherlands), which was held in Utrecht on January 25, 1986. ISBN 90-6765-254-7 SISO 805 U D C 801 (492) Subject heading: linguistics ; The Netherlands.

ISBN 90 6765 254 7 ® 1986 Algemene Vereniging voor Taalwetenschap. seer.: A. Hulk, Dept. of French, Free University, P.O. Box 7161, 1007 MC Amsterdam. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. Printed in the Netherlands by ICG Printing, Dordrecht.

Contents

Preface List of contributors Renée van Bezooijen Lay ratings of long-term voice-and-speech characteristics

1

Geert E. Booij Icelandic vowel lengthening and prosodie phonology

9

Gosse Bouma Grammatical Functions and Agreement in Warlpiri

19

Peter Broeder, Guus Extra & Roeland van Hout Acquiring the linguistic devices for pronominal reference to persons: a crosslinguistic perspective on complex tasks with small words

27

Norbert Corver On promotion in contrastive left dislocation constructions

41

Frank Drijkoningen On the order of (auxiliary) verbs and syntactic affixes

51

Wim de Haas A CV analysis of vowel hiatus in Kasem

61

Liliane Haegeman Passivization in Flemish dialects

71

Vincent van Heuven Phonetic correlates of stress patterns in Dutch (compound) adjectives

79

Eric Hoekstra Precedence in Jacaltec

89

VI Teun Hoekstra Passives and participles Harry van der Hulst, Maarten Mous & Norval Smith The autosegmental analysis of reduced vowel harmony systems: the case of Tunen

95

105

Emmy Jacobs Polarity-sensitivity and generalized quantifiers

123

Bob de Jonge Vos- nos- + otros; one or two phenomena?

131

René Kager & Ellis Visch Conditions on metrical adjunction

141

Simone Langeweg "Roots", "words" and the stress behaviour of non-native words in Dutch

151

Làszló Maràcz On transitivity in non-configurational languages

161

Christine ter Mors Syllable reduplication

171

Frans van der Putten Deverbal adjectives: grammatical or thematic relations

179

Pie ter Th. van Reenen The vocalization of / l / in standard Dutch, a pilot study of an ongoing change

189

Bob Rigter Time intervals and identity relations across boundaries of intensional domains

199

Engin Sezer & Leo Wetzels On the interaction of backness and rounding harmony

209

I.H. Slis & R.J.H. van den Berg Phonetic context effects and assimilation: perception and production

217

VII A.M. Slootweg Rhythm and reason in stress rankings Norval Smith Reduplication, Spreading and/or empty suffix slots in Sierra Miwok associative morphology Wiecher Zwanenburg X-bar structure and argument structure in morphology

225

235 245

Preface

This volume contains a selection of the papers presented at the seventeenth annual meeting of the Society, held in Utrecht on January 25, 1986. The aim of the annual meeting is to provide members of the Society with an opportunity to report on their work in progress. At this year's meeting fifty-five papers were presented. The twenty-six papers contained in this volume present an overview of research in different fields of linguistics in the Netherlands. As in previous years, the authors have submitted their articles in camera-ready form. We are grateful to them for their cooperation. May 1986

Frits Beukema Aafke Hulk

List of contributors

R. van den Berg University of Nijmegen, Department of Phonetics, P.O. Box 9103, 6500 H D Nijmegen R. van Bezooijen University of Nijmegen, Department of Phonetics, P.O. Box 9103, 6500 H D Nijmegen G. Booij Free University, Department of General Linguistics, P.O. Box 7161,1007 MC Amsterdam G. Bouma University of Groningen, Department of General Linguistics, Grote Rozenstraat 31, 9712 TG Groningen P. Broeder Tilburg University, Department of Language and Literature, P.O. Box 90153, 5000 LE Tilburg N. Corver Tilburg University, Department of Language and Literature, P.O. Box 90153, 5000 LE Tilburg F. Drijkoningen University of Utrecht, Department of French, Drift 15, 3512 BR Utrecht G. Extra Tilburg University, Department of Language and Literature, P.O. Box 90153 5000 LE Tilburg W. de Haas University of Nijmegen, Department of French, Erasmusplein 1,6525 G G Nijmegen L. Haegeman University of Geneva, Faculté des Lettres, CH-1211 Genève 4, Switzerland V. van Heuven University of Leiden, Phonetics Laboratory, P.O. Box 9515, 2300 RA Leiden E. Hoekstra University of Groningen, Department of General Linguistics, Grote Rozenstraat 31, 9712 TG Groningen T. Hoekstra University of Leiden, Department of General Linguistics, P.O. Box 9515, 2300 RA Leiden

XI R. van Hout Tilburg University, Department of Language and Literature, P.O. Box 90153, 5000 LE Tilburg H. van der Hülst University of Leiden, Department of General Linguistics, P.O. Box 9515, 2300 RA Leiden E. Jacobs University of Utrecht, Department of General Linguistics, Institute A. W. de Groot, Trans 14, 3512 JK Utrecht B. de Jonge University of Leiden, Department of Latin American Studies, P.O. Box 9515, 2300 RA Leiden R. Kager University of Utrecht, Department of Dutch, Institute De Vooys, Padualaan 14, 3854 CH Utrecht S. Langeweg University of Leiden, Department of General Linguistics/Phonetics, P.O. Box 9515, 2300 RA Leiden L.K. Maracz University of Groningen, Institute of General Linguistics, Grote Rozenstraat 31, 9712 TG Groningen C. ter Mors University of Groningen, Institute of General Linguistics, Grote Rozenstraat 31, 9712 TG Groningen M. Mous University of Leiden, Department of African Linguistics, P.O. Box 9515, 2300 RA Leiden F. van der Putten University of Leiden, Department of General Linguistics, P.O. Box 9515, 2300 RA Leiden P. van Reenen Free University, Department of General Linguistics, P.O. Box 7161,1007 MC Amsterdam B. Rigter University of Leiden, Department of English, P.O. Box 9515, 2300 RA Leiden E. Sezer University of Nijmegen, Department of French, Erasmusplein 1,6525 G G Nijmegen I.H. Slis Ufiiyersity of Nijmegen, Department of Phonetics, P.O. Box 9103, 6500 H D Nijmegen A.M. Slootweg University of Nijmegen, Department of Phonetics, P.O. Box 9103, 6500 H D Nijmegen

XII N. Smith University of Amsterdam, Department of General Linguistics, Spuistraat 210, 1012 VT Amsterdam E. Visch University of Utrecht, Department of Dutch, Institute De Vooys, Padualaan 14, 3584 CH Utrecht L. Wetzels University of Nijmegen, Department of French, Erasmusplein 1,6525 G G Nijmegen W. Zwanenburg University of Utrecht, Department of French, Drift 15, 3512 BR Utrecht

Renee van Bezooijen

Lay ratings of long-term voice-and-speech characteristics 0. Summary A n experiment is described w h i c h aimed at gaining insight into the nature of lay judgments of long-term voice-and-speech characteristics. This was done by having two trained speakers read the same text w i t h 26 production settings (lip rounding, high pitch, etc.) and examining the effects of these manipulations on 15 voice-and-speech scales rated by 29 phonetically untrained listeners. In the present paper an inventory is presented of those cases in which a scale was affected by a particular setting in a similar way for both speakers. The problem of interactions between the original voices of the speakers and the effects of settings is discussed as well as the problem of the interdependency of different settings. 1. Introduction In several studies (e.g. Fagel, Van Herpt, & Boves 1983; Scherer 1978) use has been made of phonetically untrained people to obtain a description of longterm voice-and-speech characteristics. Those descriptions usually consist of ratings o n equal-appearing interval scales, such as low-high, warm-sharp, and full-thin. The reason to opt for untrained rather than trained listeners m a y be a practical one, e.g to save time or to avoid the problems involved in training. A more theoretically based reason may be that one wishes to obtain a description of 'proximal cues', defined by Scherer (1979) as the lay percepts of 'distal cues , i.e. voice-and-speech characteristics w h i c h can be assessed objectively, either by means of acoustic analysis, auditory judgments by expert raters, or physiological measurements. Very little is known about the meaning of lay judgments of long-term voiceand-speech characteristics. Attempts at finding acoustic correlates of lay ratings have been disappointing; correlations were generally low (Boves 1984). Comparisons of lay ratings w i t h expert ratings have been very incomplete (cf. Scherer 1979). And the physiological basis of lay ratings has hardly been investigated. In the present study an attempt was made to explore the meaning of lay ratings on 15 long-term voice-and-speech scales by manipulating the speech material to be rated. The manipulations were effected by having two trained speakers read one standard text with 26 different production settings, such as lip rounding and high pitch. In this way the input material was more or less controlled. By examining which scales were affected by which setting(s), or, more precisely, by comparing the scale ratings for a particular setting w i t h those for the 'normal' voices of the speakers, I hoped to gain insight into the nature of the perception of long-term voice-and-speech characteristics by phonetically untrained people.

2 2. Method 2.1 Speech material Two speakers were used, one male, a research associate of the Institute of English at the University of Nijmegen, and one female, a graduate student of the same department. Both had tutored students in English pronunciation. They were given the demonstration tape that accompanies Laver's 'phonetic description of voice quality' (1980) and extensive explanations of the settings they were to perforin. There were 26 settings (see Table 1), most of which have been derived from the vocal analysis protocol given in Laver, Wirz, MacKenzie, and Hiller (1981). Three groups of settings may be distinguished, supralaryngeal ones (e.g. lip rounding, dentalization, denasal), laryngeal ones (e.g. breathy, tremulous, soft) and prosodic ones (e.g. monotonous, repetitive pitch patterns, disrupted) (for theoretical information on the notion of setting in general and acoustic, physiological, and auditory details on many of the settings, see Laver (1980)). After thorough practising, recordings were made in a sound-insulated audio studio, using a Sennheiser condensor microphone type MKH 415, a Studer audio console, and Studer tape recorders type A80/R. Table 1.

Production settings

normal lip rounding lip spreading labiodentalization close jaw protruded jaw dentalization retroflexion nasal denasal imprecise articulation breathy whispery

creak tremulous harsh loud soft high pitch monotonous expressive emphat ic repetitive pitch patterns fast slow disrupted

Each setting (see Table 1) was produced as the 'Northwind passage 1 , a text of about 35 seconds, was read. While producing the different settings, intonation and other vocal characteristics were kept as constant as possible. Each setting was produced until both speaker and experimenter were satisfied with the result, or felt that no further progress could be made. The realizations of the different settings may be characterized as fairly extreme but within the range of what may be found in a non-pathological voice. Two stimulus tapes were prepared. Both tapes started with two practice stimuli, followed by the 'normal' settings of the two speakers. The rest of the settings were presented in a random order. The second tape differed from the first in that the second half of the stimuli was placed before the first half. On both tapes, female and male realizations were put in an alternating order. The stimuli were lengthened to between 50 and 59 seconds by repeating the first couple of phrases at the end of the text. There was an interstimulus interval of about 9 seconds. The stimuli were preceded by numbers corresponding with the numbers on the rating sheets. 2.2 Procedure The ent 32, lus

stimuli were judged by 29 students (14 males and 15 females) from differdepartments at the University of Nijmegen. Their ages ranged from 19 to with a mean of 23.5. Seven males and nine females judged the first stimutape, seven males and six females judged the second stimulus tape. None of

3 the subjects had any experience with the description of speech. They were recruited on a voluntary basis and were paid for their participation. The experiment took place in a small lecture room. A Revox A77 and loudspeakers KEF model 104 were used. All subjects were given a booklet containing the typical semantic differential instructions (see Osgood, Suci, & Tannenbaum 1957:82 ff.), five rating sheets for the rating of their own voice, the typical male voice, the typical female voice, the ideal male voice, and the ideal female voice, respectively, two rating sheets for the rating of the practice stimuli, and 52 rating sheets for the rating of the experimental stimuli. Table 2. Rating scales (English translation from Dutch) and reliability (for an explanation of the Ru-coefficient, see Section 3.1) Scales

Ru

The 1. 2. 3. 4. 5. 6. 7. 8. 9.

voice is: warm-sharp smooth-rough low-high soft-loud nasal clear-dull trembling hoarse full-thin

.96 .89 .97 .92 .90 .89 .93 .92 .92

The 10. 11. 12. 13. 14. 15.

speech is: precise-slurred fast-slow accentuated-unaccentuated expressive-flat fluent-halting staccato

.94 .94 .92 .93 .92 .38

The 15 scales included in the rating sheets are given in Table 2. There were 11 bipolar 7-point scales, for those parameters that form a continuum that can easily be represented by two contrasting terms, and 5 unipolar 4-point scales, for parameters that tend to be interpreted in terms of absence versus presence.1 The order of the scales as well as the direction of the polarity was identical for all subjects.5 No explanations were given of the meaning of the scale terms. The subjects started noting down their ratings as soon as they heard a particular voice. On an average, they had about three-and-ahalf second for the rating of each scale, which proved to be sufficient. The total experiment took about 80 minutes, including a break. 3. Results and discussion 3.1 Reliability The reliability of the ratings given by the 29 raters was determined by means of the so-called Ru-coefficient (Asendorpf & Wallbott 1979), which is a measure of the reliability of the means of the ratings given by a panel of raters. This coefficient is of the form 1 -(Mean Square within items/Mean Square between items), and yields values that are roughly comparable to Cronbach's alpha (with a maximum value of 1.00). Ru was computed for each scale separately over all 52 stimuli. The results are given in Table 2. For most scales the Ru-coefficient is high. From this it can be deduced that the listeners - without any preparatory training - have succeeded in giving

4 the scales a consistent and meaningful interpretation and that the voice-and-speech characteristics which were rated occurred in varying degrees 1 in the material. The only exception is 'staccato . The low reliability for this scale appears to have been caused mainly by a low Mean Square between items, i.e. a lack of variation of this feature in the stimuli. Apparently, none of the 26 production settings resulted in a clear auditory staccato effect. Since all other scales were rated reliably, in the subsequent analyses the individual ratings were replaced by the means. 3.2 Effects of settings on scales For each of the 52 stimuli the mean value for each of the 15 voice-and-speech scales was computed. In Figure 1 the 15 values for the 'normal' settings of the two speakers are shown. It may be seen that the two scale profiles are very different, i.e. in the perception of the lay judges the original voices of the two speakers have few characteristics in common (product-moment correlation of .13). The female voice has been perceived as somewhat sharp, high, and hoarse, and quite nasal, while her speech has been perceived as somewhat fast and staccato; on the other hand, the male voice has been perceived as somewhat clear and hoarse, quite loud and very warm, whereas his speech has been perceived as quite expressive and fluent, very accentuated, and extremely precise. In general, the male voice appears to have received much more extreme ratings on various scales than the female voice. It would be interesting to know whether these differences are specific for these two speakers or reflect general sex-related tendencies. This holds especially for the voice characteristics, as the speech characteristics have probably been induced to a great extent by personal differences in reading style.

Nontn.

n

Ss 4 3

2 1 0

1

2

3

4

6

6

7

a

9 to 11 12 13

7

ts KilM

14

HOMRL F

Ss 4 3

2 1 0

1—2

S

4

S

6

Figure 1. Scale profiles for the female, m = male)

7

a

9

10 11 12 13 14 IS mcmlmm

'normal' voices of the two speakers

(f =

The main purpose of this study was to assess in which respects the perception of the 'normal' voices of the speakers changed as a result of the production of the 25 other settings listed in Table 1. In order to gain insight into these changes, the 'normal' scale profiles were subtracted from all other scale profiles. In this way, for each setting deviation values were obtained for each of the 15 rating scales. An effect of a particular setting on a par-

5 ticular scale was considered to be stable whenever a deviation of at least one scale position in the same direction was found for both speakers. The stable effects are listed in Table 3. Table 3. Scales with settings that cause a deviation in the same direction for both speakers of at least one scale position Scales

Settings

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

sharp rough high loud nasal dull

+ + +

7. 8. 9. 10.

trembling hoarse thin slurred

+ +

11. slow 12. 13. 14. 15.

-

+ +

+ +

unaccentuated + + flat + halting staccato

harsh, loud harsh high pitch whispery denasal denasal, whispery, creak, nasal, imprecise articulation, retroflexion tremulous whispery, harsh nasal, imprecise articulation, dentalization, retroflexion creak, slow, disrupted fast imprecise articulation, monotonous imprecise articulation, monotonous disrupted

Note. + = deviation towards the right scale term (the one given in the Table), - = deviation towards the left scale term It may be seen from Table 3 that in most cases the scales have been affected by just one or two settings. Many of the effects speak for themselves. For example, one would indeed expect rough1 to be affected by 'harsh', 'high' by 'high pitch', 'trembling' by 'tremulous', and 'halting' by 'disrupted'. It may be observed that 'unaccentuated' and 'flat' show identical patterns, and indeed the correlation between the two scales is very high (r = .94). There are three scales that show less specific patterns, namely 'slurred', which has been affected by four different - supralaryngeal - settings; 'dull', which has been affected by six settings, four supralaryngeal and two laryngeal ones; and 'slow', which has been affected by four settings, three prosodic ones and one laryngeal. Apparently, some scales have a much more diffuse meaning than others . There are two scales for which no consistent effect of any setting was found, namely 'staccato' and 'thin'. For 'staccato' none of the settings resulted in a deviation of one or more scale positions for neither of the two speakers (cf. the remarks made in connection with the low reliability of this scale). For 'thin' the situation is quite different. For the male speaker there were eight settings which caused a deviation of at least one scale position towards the thin end of the scale; for the female speaker there was only one, and that one was towards the full end of the scale. This finding should be interpreted in the light of differences between the original voices of the two speakers (mean 'thin' ratings of 2.1 and 4.4 for the male and female speaker, respectively). It could be hypothesized that possibly consistent 'thin' effects of different settings were perceived much more clearly against the 'full' background voice of the male speaker than when they were superimposed on the voice of the female speaker, which sounded somewhat thin to begin with. A similar reasoning can be applied to account for the absence of a consistent effect of the setting 'loud' on the scale 'loud'. It is quite probable that the original voice of the male speaker was already so loud that the 'loud' setting could not manifest itself anymore.

6 We touch here upon an important point, namely the perceptibility of different settings as a function of the original voice characteristics of a speaker. In Table 3 those cases were listed in which for both speakers a non-trivial deviation from their original voices was found. However, it has to be explicitly noted that this list need not be exhaustive. As we have seen for 'thin', a possibly consistent effect of a setting on a particular scale may be obscured in those cases in which the original voice of the speaker was already characterized by an extreme position on that scale. Another point I want to mention is the interdependency of the various settings. Ideally, trained speakers should be able to vary one setting while keeping all other settings constant. As stated by Laver (1980, p.18), himself a very competent performer, in practice this often turns out to be an impossible task. An example of an unintended but uncontrollable effect of one setting upon another in the present study is the influence of creak on rate: both speakers declared that they were unable to produce a sustained creak without at the same time slowing down their speech. And it appears from Table 3 that indeed the production of creak has led to high ratings on the 'slow' scale. 4. Conclusions This study, in which 26 settings produced by two trained speakers were rated by 29 untrained listeners on 15 long-term voice-and-speech scales, was conducted with the aim of further exploring the nature of lay judgments of vocal aspects of speech. Some clear relationships were found between the production of particular settings and the height of the ratings on particular scales. Obviously, the number of speakers was very small, so that one should be very cautious in generalizing the results. Also, the study should be replicated with several gradations of each setting rather than one extreme realization and with combinations of settings rather than each setting separately. However, in view of the scarcity of trained speakers it is altogether questionable whether it is realistic to think such a study feasible. I had a very hard time finding two speakers who were able to perform the task at hand. An alternative is to use synthesis or resynthesis to manipulate various parameters and assess the effects of these manipulations on lay judgments of speech. A great advantage of this approach is that the input stimuli are completely defined, at least in acoustic terms. However, I think that such studies have little value for increasing our knowledge of the perception of human speech unless the acoustic parameters themselves are understood in terms of speech production mechanisms. And this situation is far from being attained. Notes *

This research was supported by the Foundation for Linguistic Research which is funded by the Netherlands Organization for the advancement of pure Research, ZWO. The inventory of voice-and-speech scales used in the study presented here was developed in cooperation with Loe Boves, of the Institute of Phonetics at the University of Nijmegen, and Klaus Scherer, of the University of Geneva/Justus Liebig University. I am very much indebted to the two speakers, Carlos Gussenhoven and Anne-moon van Hest, without whose efforts this study could not have taken place.

1. It is sometimes suggested that the polarity of the scales be varied in order to avoid a left or right oriented rating bias. However, from previous experiments I know that this may lead fairly frequently to subjects ticking the non-intended end of the scale (cf. similar experiences reported in Guilford, 1954: 267-268). This may seriously affect the reliability

7 of the ratings. From the same experiments it appeared that it is not much use varying the scale order since subjects do not stick to the order anyway; they tend to rate those parameters first that they feel are most salient for the voice at hand. 2. There is no unambiguous criterion to decide whether a particular parameter should be represented by means of a unipolar or a bipolar scale. After the experiment had been conducted, I found 'smooth-flowing' as a possible contrast for 'staccato*. (For practical as well as theoretical reasons, it is best to only have one type of scale.) References ASENDORPF, J. & WALLBOTT, H. 1979 Masse der Beobachteruebereinstimmung - Ein systematischer Vergleich, in Zeitschrift fuer Sozialpsychologie 10, p. 243-252 BOVES, L. 1984 The Phonetic Basis of Perceptual Ratings of Running Speech , Dordrecht/ Cinnaminson, Foris FAGEL, W., VAN HERPT, L., & BOVES, L. 1983 Anaysis of the perceptual qualities of Dutch speakers' voice and pronunciation, in Speech Communication 2, p. 315-326 GUILFORD, J.P. 1954 Psychometric Methods (2nd edition), Bombay, McGraw-Hill LAVER, J. 1980 The Phonetic Description of Voice Quality, Cambridge University Press LAVER, J., WIRZ, S., MACKENZIE, J., & HILLER, S. 1981 A perceptual protocol for the analysis of vocal profiles, in Work in Progress, Dept. of Linguistics, University of Edinburgh OSGOOD, C., SUCI, G., & TANNENBAUM, P. 1957 The Measurement of Meaning, Urbana, University of Illinois Press SCHERER, K.R. 1978 Personality inference from voice quality: the loud voice of extroversion, in European Journal of Social Psychology 8, p. 467-487 SCHERER, K.R. 1979 Personality markers in speech, in K.R. Scherer and H. Giles (eds) Social Markers in Speech, Cambridge University Press, p. 181-254

Geert E. Booij

Icelandic vowel lengthening and prosodic phonology 1. Introduction: some basic tenets of prosodic phonology It is an important insight of prosodic phonology that sequences of sound segments are structured according to two hierarchies: the morphosyntactic hierarchy and ^ the prosodic hierarchy. These hierarchies may provisionally be stated as follows: 2 (1)a morphosyntactic hierarchy: b prosodic hierarchy: morpheme (M) syllable () phrase phonological phrase (

In (19a) PRO bears the predicative external role. In (19b), w h i c h illustrates a passive sentence, PRO bears the internal role. In (19c), w h i c h illustrates the situation w i t h an ergative verb, PRO also bears the internal role. In (19d) we have the same situation as in (19c), but w i t h a different choice of the auxiliary. The example is relevant because it shows that one cannot derive

55 the sentences by a deletion rule sensitive to the auxiliary. It shows that the absence of the auxiliary plus the INFLection is dependent on thematic information. In (19e) PRO bears the 'true' external role, and therefore the auxiliary plus the INFLectional marker -ant may not be absent in our view. As a conclusion we may define the following reach for INFLection: (20) Reach of INFL: Case and 0-role of the subject. Once we have defined the "reach" of the affixes, one can define these reaches in terms of domains, in terms of government. The domains correspond to the ones given in structure (21): (21)

The domains correspond to the domains definable by government according to the definitions given in Chomsky (1985). In particular, VP is a barrier protecting the object from Case-influence by INFL and protecting the subject from Case-influence by PERF or PASS. VP also protects the object from 8-influence by INFL. The fact that PASS may influence the subject in terms of 9-roles is accounted for if one assumes that PASS is close enough to the Verb to influence the whole 9-grid. The external role percolates from the V to the VP-node, and this is precisely what is blocked by the PASS morpheme. Inside the VP, however, we have PERF and PASS which both govern the object in the large definition of government. In order to solve these types of ambiguity, Chomsky proposes the Minimality Condition, which states that if two elements govern the same element, the closest one is relevant. The Verb with the PASS morpheme is a closer governor than PERF. In this way PERF cannot influence the Case of the object. In conclusion, the affixes INFL, PERF and PASS may have influence on the arguments of the clause. The influence they have corresponds in practice to some level of affixation in syntax. The various levels of affixation correspond to independently motivatable domains in terms of the theory of government. 3. The insertion of auxiliaries In the preceding section I showed that the order of the affixes can be explained by their intrinsic properties. The order of the auxiliaries is a direct reflex of this order of affixes. To be concrete, we assume that the merger of the affix with the verb is defined cyclically such that the verb will always merge with the affix which it will first encounter. The first encountered affix will be the one that is the lowest in a structure such as (21), where the lowest is, of course, a relative notion. After the first pass of the cyclic merger then, the following situations arise:

56 (22)

(a)

(b)

I NFL' NP INFL

(c)

INFL' NP.1 INFL'

r

INFL PERF

Taking up the conditions in (13) again, w e note that in (22) w e are confronted w i t h several violations of clause (b). In order to "repair" the constructions, verbal elements must be inserted, auxiliaries in m y view. I directly insert the auxiliaries on the position of the affix w h i c h violates clause (b) of (13), such that I derive the following surface orders, w h e r e I have inserted VP-adverbs in order to show the correctness of the ordering facts. (23)

(a) Jean est souvent b a t t u par son copain. (b) John is often b e a t e n b y his partner. INFL

(c) Jean a souvent lu ce livre. (d) John has often read this book. INFL''

(e) Jean a souvent été battu. (f) John has often b e e n beaten. As far as French is concerned, n o special results are achieved, because the merger of INFL and V is upwards (cf. Emonds (1978)), as can be seen in (24):

57 (24) (a) a) NP

f I INFL" INFL' P ADV JV \y

(b) Jean dort souvent. However, in English the merger of INFL and V is downwards, such that (25) corresponds to (24). (25) (a)

INFL"

N INFL

VP ADV

V

(b) John often sleeps. The split between the adverbial positions in (23) and (25) is generally explained by the rule of have/be raising (Jackendoff (1972)). In this standard analysis INFL is 'lowered' to V if there is no auxiliary in the structure, such that adverbials precede the finite verb (as in (25)). However, the leftmost auxiliary is 'raised' to INFL by the rule of have/be raising, such that adverbials follow the finite auxiliary (as in (23)). In my view this split is a consequence of the base-generation of verbs versus the insertion of auxiliaries on the position of the affix. As can be observed in (22) and (25) the first affix lowers to V, leaving the higher affixes without verb to occur on. Direct insertion of auxiliaries in the position in which they are needed automatically entails a higher position for these auxiliaries. It seems therefore that the rather typical English rule can be (at least partially) abandoned in favour of an approach in which only auxiliaries are inserted relatively late. Two other arguments may be adduced in favour of this position. Clause (b) in (13) also covers another typical English phenomenon, do-insertion. It is only after the eventual application of wh-movement that one can see if do must be inserted or not. This suggests that do-insertion is best accounted for after wh-movement, i.e. at S-structure, at the same level as auxiliary insertion. We have not discussed the choice of the auxiliary in this paper, which can also be predicted. Now, theories predicting the auxiliary like the one of Burzio (1981) generally operate after the establishment of the binding relations, in particular after NP-movement. Auxiliary selection then is also relatively late. To us this indicates that the selection of the auxiliary takes place at the same moment as the insertion of auxiliaries, thus leading to a generalized theory for auxiliaries applying at S-structure. Besides this empirical evidence in favour of the conditions in (13) operating on S-structure, I think that there is also a more theoretical advantage in using (13). Recall that clause (13c) essentially explains the one-to-one correspondence between affixes and verbs, in stating that at most one affix may appear on a verbal element. Now, while the clauses (a) and (b) do not seem liable to parametric variation, clause (c) may well be. Classical Latin has the following verbal paradigm for the elements discussed in this paper:

58 (26) (a) (b) (c) (d)

INFL INFL+PERF INFL+PASS INFL+PERF+PASS

voco vocavi vocor vocatus sum

'I 'I *I 'I

call' have called' am called' have been called'

Classical Latin then has synthetic forms up to two affixes, and inserts one auxiliary if there are three affixes. For Latin then clause (13c) may be taken to refer to maximally two affixes on one verbal element instead of maximally one for English and French. Such a restriction to two could also be completely absent, as for example in Ancient Greek. The problem for this view, however, lies in the difficulty of isolating the correct morphemes corresponding to PERF, PASS or INFL. While the -v in (b) is typical for PERF, it is not traceable in (d). While the -o in (a) and (c) is typical for Praesens first person, it is not traceable in (b). So the least we would need in order for our idea to work is a more abstract notion of affix, closer to some type of abstract unit with a specific function (and a specific meaning) rather than to a uniquely identifiable morphological unit.

4. Conclusion In this paper I have argued that the order of the affixes is predictable on the basis of the influence they have on the arguments of the clause in which they are generated. Their syntactic affixation level corresponds to otherwise definable domains. The order of the auxiliaries is a direct reflex of the order of the affixes. I argue moreover that the English and French one-to-one correspondence between affixes and verbal elements is an S-structure phenomenon, and that auxiliaries are inserted at this level, thus explaining a split in adverbial positions in English.

Notes 1. This research was supported by the Foundation for Linguistic Research, which is funded by the Netherlands Organization for the advancement of pure research, ZWO. 2. The same line of argument may also give substance to the claim that e.g. passive morphology is the head of a small clause. The passive morphology then is essentially similar to INFL in the sense that both are the heads of the more traditional node S. Analyses making use of this line grant the auxiliaries the status of main verbs. Our primary concern in this paper is the order of the auxiliaries with respect to other auxiliaries and the verb. But we see no way in which we can predict the order of main verbs other than the fact that they must observe the directionality requirement of the X' base rules (as illustrated in (1) and (2)). On auxiliaries as main verbs, see Hoekstra (1986). 3. Thus, for the sake of this paper, we disregard the discussions about the categorial problems centering around the affix in (16), particularly interesting for the morphemes I called PERF and PASS, which clearly generalize as to their morphological shape. 4. Implicitly, I argue that the Minimality Condition is relevant for Case. 5. The type of parametric variation accounting for the difference between Latin and French or English might perhaps be extended to a difference

59 between French and English. While French uses synthetic forms for the future tenses, English uses auxiliaries.

References BURZIO, L. 1981 Intransitive Verbs and Italian Auxiliaries. MIT Ph.D.Dissertation. CHOMSKY, N. 1985 Barriers. Ms. EMONDS, J. 1978 The Verbal Complex V'-V in French, in Linguistic Inquiry 9, pp. 151-175. FABB, N. 1984 Syntactic affixation. MIT Ph.D.Dissertation. HOEKSTRA, T. 1986 Verbal Affixation, in this volume. JACKENDOFF, R. 1972 Semantic Interpretation in Generative Grammar. MIT Press Cambridge Mass. 1977 X' L. Syntax: A Study of Phrase Structure. LI Monograph, MIT Press. RIZZI, 1982 Issues in Italian Syntax. Foris Publications, Dordrecht.

Wim de Haas

A CV analysis of vowel hiatus in Kasem

0. Introduction The description of Kasem nominals received attention in the earlier linear approaches. Among these are Chomsky and Halle (1968), Halle (1978) and Phelps (1975, 1979). Kasem is a West African language spoken in the northern regions of Ghana and Upper Volta. It belongs to the Grusi division of the Gur languages. Like its surrounding languages it is a tone language, and a vowel-harmony language, but both aspects of the language will be left out of consideration, since they bear no relevance to the points being made. Apart from the harmony feature 'ATR' the vowel system can be characterized as in (1): (1)

i high + low round back -

e -

a +

-

+

o + +

u + + +

The organization of this paper is as follows. In section 1 an overview of the relevant data within' previous linear analyses of Halle (1978), and more in particular Phelps (1975, 1979) is followed by various critical comments. Next, in section 3, an alternative nonlinear analysis is proposed, which does not encounter the problems of the previous analyses. 1. Previous analyses Kasem nominals can roughly be divided into the five classes in (2) below. The classification is based primarily upon various forms of the singular and plural suffixes, and represent the general cases: (2) class A B C Dj E

singular iid+u nakw+i fal+a iud+u sug+u

plural ffid+i nakw+a fal+i ?ud+du su+ni

gloss 'witch1 'elder' 'white man' 'neighbor' 'guinea hen'

Besides the regular forms in (2) Kasem distinguishes class D nominals that show vowel length alternations in the singular and plural stems: (3) singular /pi:+u/ -» [piu] /ti:+u/ -» [tiu]

plural [pi:+du] [ti:+du]

gloss 'mountain' 'piece of soil'

In Phelps (1975:314) the Vowel Shortening rule (4) is proposed. This rule shortens long vowels in prevocalic position: (4) Vowel Shortening; (VS) [+syll, -low]-» [-long] / —

[+syll]

The data in (5) show that two additional processes apply:

62

(5) class A B C D F

singular - » [tu] /tu+u/ /buko+u/ -> [buko] - » [ni] /ni+i/ -* /pi+a/ [pia] - » [pia] /pe+a/ —» [kua] /ku+a/ /vo:+u/ -> [vo:] /ku:+u/ - » [ku:] - > [sua] /su+a/

plural —» /tu+a/ /buko+a/ —> -> /ni+a/ -* /pi+i/ -» /pe+i/ -* /ku+i/ /vo:+du/ —> /ku:+du/ —» -» /su+a/

gloss 'corpse' 'daughter' 'mouth' 'yam' 'sheep' 'bone' 'leaf' 'bone' 'guinea-hen chick

[twa] [bukwa] [nia] [pi] [pe] [kwi] [vo:du] [ku:du] [swa]

The /o, u/ ~ /w/ alternations in the plural forms in (5) are the result of the Glide Formation rule (6), that changes round vowels into glides: (6) Glide Formation-. (GF) [+syll, +round] -» [-syll] / —

[+syll]

Additionally, if the stem vowel and the suffix vowel agree in backness and roundness, the suffix vowel will be deleted by the Vowel Truncation rule (7) proposed by Phelps (1975:313): (7) Vowel Truncation:. (VT) [+syll, aback, ground] é / [+syll, aback, ground]

—

Finally, Phelps discusses the data in (8), in which the quality of the vowels in hiatus is changed: class B C

D

singular /zwa+i/ -•[zwe] /i*wa+i/ ->[5we] /yo+a/ -•[yua] /ku+a/ -»[kua] -•[pia] /pe+a/ -•[pia] /pi+a/ /se:+u/ ->[sio] /foga+u/ -»[fogo]

plural /zwa+a/ -»[zwa] /Swa+a/ -»[Cwa] /yo+i/ -»[ywe] /ku+i/ -»[kwi] /pe+i/ -»[pe] /pi+i/ -»[pi] /se:+du/ -»[se:du] /foga+du/-»[fwadu]

gloss 'ear' ' liver' 'hair' 'bone' 'sheep' 'yam' ' shea-nut' 'die, dice

Phelps accounts for the data in (8) by positing a (special) type of metathesis, that is the Vowel Height Exchange in (9): (9) Vowel Height Exchange-. (VHE2 +syll [+syll] _ i / [+low], [-high] [+high] -high 1 1 Actually, VHE (9) is a fairly complex rule. It is an abbreviation of three disjunctively ordered subrules. The first of these deletes the left vowel and at the same time characterizes the right vowel as [-high]; the second one deletes a [+low] vowel, if a vowel follows; and the third one raises the left vowel in a sequence of two. In (10) I present some relevant derivations, showing the interaction of these processes. Notice that the processes should be extrinsically ordered in the order given: (10) /zwa+i/

[zwe]

/zwa+a/ Í

/yo+a/

/yo+i/

[zwa]

[yua]

w [ywe]

/pe+a/

[pia]

/pe+i/

/pi:+u/

[pe]

[piu]

VT VHE VS GF output

63 2. Critical Comments In this section, the explanatory value of the analysis proposed by Phelps will be questioned. Various arguments will be brought up against this analysis. My first remark cornerns the VHE (9). This rule is formalized as a phonological trans formation, and the feature [+high] is added in one of its subcomponents. However, the addition of [+round] or whatever feature would have no consequences for the markedness of the process. In other words, by allowing phonological transformations the distinction between possible and impossible, and marked versus unmarked phonological changes cannot be decided upon in principle, one of the reasons why phonological transformations are excluded from phonology. Secondly, notice that VHE (9) is both a feature-changing rule and a deletion rule. However, I believe there are good reasons for not doing this. I think that it is useful to separate the feature-changing part of the rule from deletion, since unlike deletion, the feature-changing subrules are for the most part language independent. Secondly, I will show that we need, for independent reasons, a language-specific vowel truncation rule with the same properties. Therefore, maintaining the deletion part of VHE would only be an unwarranted duplication in the phonology of Kasem. Furthermore, Phelps' VT (7) is in a way unmotivated. Halle (1978) posits a slightly different rule that only deletes the second of two identical vowels. Phelps increases the power of this rule to the sequences e+i -» e, o+u -* o. Such an extension is necessary, since otherwise the ill-formed items in (11) would be derived by VHE: (11)

/zwa+a/ i

/tu+u/ i

/pi+i/ i

[zwa]

[tu]

[pi]

/buko+u/ — u o *[bukwo]

/pe+i/ — i e *[pie]

VT (Halle) VHE output

By adopting the extended formulation in (7), the ill-formed items in (11) can be excluded. But, since VT (7) is in a 'bleeding order' with respect to VHE, such an extension is quite suspect. Finally, both Halle (1978) and Phelps (1975, 1979) propose three rules altering the quantity of vowel sequences, that is VS (4), VT (7), and the deletion part of VHE (9). Now, a closer look at the structural descriptions of these rules shows that all of them affect vowels in hiatus - which fact has gone unnoticed in the previous analyses.

3. Resyllabification and Vowel Coalescence I adopt here the basic principles of CV Phonology of Clements and Keyser (1983), particularly, I assume an independent tier consisting of strings of timing units (C or V slots) that mediates between the segmental core and other tiers. Recent research has led to a reappraisal of the syllable. In this section, I will argue in favour of the representation of the syllable in phonological theory from a completely different perspective. I will suggest that resyllabification is crucial for a proper characterization of processes involving vowel hiatus, that is vowel coalescence. The first rule I propose is Vowel Lowering (12) below. Apart from its nonlinear formalization, it resembles the first subrule of VHE (9):

64 (12) Vowel Lowering: (VL) V

+

V

[-high] [ - l o w ] Steriade (1982) convincingly argues that assimilation processes are autosegmental operations. As a consequence of this assumption the output of VL (12) will undergo some version of the universal Entity-Coalescence Principle, as formulated in (13): 2

(13) Entity-Coalescence Principle: (ECP) ^

A ! [aB]

^ [aB] [aB]

w h e r e A = feature (complex), timing unit, or morpheme

syllable,

The application of VL (12) creates shared matrices for the feature [ - h i g h ] , and as a consequence all identical, but separate features are merged into shared ones by the ECP. Some illustrative derivations are given in (14): (14) a:

a

a

A, A\

[ - h ] [+h]

o

1 V V 1 | o u

VL

C V C | | | b u k

r V 1 +r +b -1

V | +r +b -1

ECP

/ \

A\

C C V I I I z w a

VL

ECP

a

/7\/\/

C C

I I

//bukoo//

V . -h +r +b

-h' .-r

b: t-h] [+h] .4= C C V V I 1 1 i i 1 z w -r -r +b -b -1 +1

V

M l

V

I

V

= //zwae//

I

The intermediate structures in (14) show that heterosyllabic vowels share at least one feature. I will assume that these structures activate the application of resyllabification, since in cases like (14a) a heterosyllabic long vowel is generated, thereby violating a universal well-formedness condition. M o r e o v e r , the assumption of resyllabification in cases of adjacent heterosyllabic vowels has at least two other attractive consequences. First of all, consider the language-specific GF (6) of Phelps (1975). This rule is necessitated by the fact that the syllable is not a primitive concept in linear phonology. But the properties of GF (6) are typically syllable-based, i.e. segments that are specified as [+high] act as a vowel in syllable-nucleus position, but. as a glide in syllable-peripheral position. If w e assume GF (6) to be a subrule of resyllabification, all properties of this process follow directly: before resyllabification the rounded high vowel occupies the nucleus position, but ends up syllable-peripherally after resyllabification. A second implication of resyllabification is that we do not have to posit language-specific vowel raising rules. Subsequently, I w i l l assume that resyllabification involves the raising of mid vowels. Highly similar processes in Chicano Spanish (Clements and Keyser (1983)) and Old Portuguese (de Haas (forthcoming)) support this view. The first expansion of Resyllabification is given in (15):

65 (15) Resyllabification-. (first version) g a g Condition: X = C iff. [+round], and V^ is plural suffix V V„

A

The condition in (15) is needed to account for the attested differences in (16): (16) a: /tu+a/ -• /ni+a/ -» b: /su+a/ -» /su+a/

[twa] [nia] [sua] [swa]

'corpse' (pi.) 'mouth' (pi.) 'guinea-hen chick' (sg.) ibid. (pi.)

In (17) some sample derivations show the application of resyllabification: (17)

/tu+u/ a a

/se:+u/ a a

A ' C V V I I I t u u

/ N ' C V V V ' V s e

a

A I C V V I I I p e a

' o

V u

/pe+i/

p e e

/ K C V V

A C V V

/f^ C C V V

p l a

' p

A

K I C V V I I I w a e

VL

o

o

C V V V I v I s i o

/zwa+i/ a a / C I z

a

/ K c v y ' t

/pe+a/ a a

Res. + ECP

V e

The examples indicate that two additional processes need to be posited: Vowel Contraction which will merge two tautosyllabic nonhigh vowels into a long vowel, and Vowel Truncation which will shorten long vowels. The former is formalized in (18): (18) Vowel Contraction: [-high]

(VC)

In (19) below the derivations of /zwa+i/ 'ear' and /foga+u/ 'die, dice' are given. Both derivations start with the intermediate forms //zwa+e// and //foga+o//, where VL (12) and the ECP have already applied: (19)

r-hi

c c V II 1 z w r + bi r-b b l L+iJ L-iIj

VC -

c c V V 1 1 -J z w r+bi [ -bl .•ij -Ij

-h +b c V c I I I fog

V

c c v y I I V

M V

VC

C V c I I! fog

V^

V

A / N C V c V V I I I v fog o

66 Finally, Vowel Truncation (20) is needed to derive the correct forms in (17) and (19) above: (20) Vowel Truncation: V V H /

[]

4 Thusfar all Kasem nominals discussed are accounted for. However, the suggested analysis suffers from one apparent drawback which concerns the class C nominals in (21): (21) singular plural gloss /mala+a/ /ziza+a/ /kaba+a/ /da+a/ /ga+a/

-» -» -• -» -»

[mala:] [ziza:] [kaba:] [da:] [ga:]

/mala+i/ /ziza+i/ /kaba+i/ /da+i/ /ga+i/

-» -» -» -» -*

[male] [zize] [kabe] [de] [ge]

'chameleon' 'small grass roof' 'slave' 'stick' 'difference'

In all the singular forms of (21) a short vowel is predicted by VT (20). To account for the vowel—length alternations in (21), I assume that the singular suffix +a must be marked as extrametrical, rendering it invisible for all proposed processes. At the end of the derivation the stray suffix will be adjoined by a universal Stray Adjunction Principle (SAP) known from recent literature on extrametricality (cf. Hayes (1981, 1982), Harris (1983)). This analysis allows us to solve the problem of the underlying representation of the stems in (21). Chomsky and Halle (1968) assume that the stem vowel is short, while both Halle (1978) and Phelps (1975, 1979) assume that the stem contains a long vowel. If the singular suffix +a is marked [+ex], all arguments in favour of underlying long stem vowels disappear, since iterative application of VT would result in a short vowel in the plural as opposed to the singular, in which VT may not apply (cf. note 3). This approach reflects why the singular of these forms in (21) consists of what Callow (1965b) calls "morphologically complex syllables". In our analysis, after Stray Adjunction, the stem vowel and extrametrical suffix vowel are merged into a long vowel. Finally, the assumption that the singular suffix +a is marked [+ex] has another attractive consequence. Earlier we proposed VL (12) and Resyllabification (14). For VL we had to characterize the target vowel as [-low], to prevent data like /pe+a/ 'sheep' (sg.) from undergoing the rule. Resyllabification on the other hand had to be conditioned morphologically to account for the minimal pair [sua] ~ [swa] 'guinea-hen chick'. If, however, extrametricality is assumed both rules can be reformulated in more general terms as shown in (22): (22) a: Vowel Lowering: V

+

V

[-high] b: Resyllabif ication: 0 0 1 I V V

o -

X

Condition: X = C iff. [+round] V

The consequences of the suggested analysis will be demonstrated in (23):^

67 (23)

/zwa+i/ a

a 1

A C V V C M i l z w a i

/zwa+a/

A\ o

0

/ku+a/ a

A

/ku+i, o

o

a

A1 A

i C C V V C V(V) C V V C V(V) 1 1 1 III z1 w1 a1 a 1 k1 u1a 1 k u i daa a a

Jv.

C C V V 1 1 1 1 z w a e c

11 z w

V AV

/AN C C V V 1 1 zwe

C C V V 1 1 zwa

a

a

A i

Syl1

C V V III dai

C V V III sua

+ Extr.

A

C V(V) III sua —

A

ccv III s w a

a

cvv I V d e

—

a

/N CVV

a

—

[kua]

/N CVV

~

[da:]

VL + Res.

—

VC + VT

/K CVV

—

SAP

III

daa [kwi]

—

'

a

III

kua [zwa]

a

/J\

III [zwe]

a

cvv III d a e

1 1 1 kwi

a

a

Ai A

sua [de]

[sua]

[swa]

output

4. Conclusions A reanalysis of Kasem nominals is presented within the framework of CV Phonology. The processes discussed involve the resolution of vowel hiatus. I have argued that a proper characterization of such processes as resyllabification allows us to simplify the grammar of Kasem substantially. First of all, we do not have to posit language-specific glide formation and vowel raising rules. It is assumed that these are subrules of resyllabification. Secondly, it turned out that Phelps1 VS (4). VT (7), and the deletion part of VHE (9) are instances of one and the same Vowel Truncation rule (20) above. Moreover, it is evident now that marking the singular suffix +a extrametrical has several advantages. Finally, the discussion of vowel hiatus in Kasem nominals is of theoretical importance, since additional evidence for the syllable as a primitive concept is put forward. In case of two adjacent heterosyllabic vowels resyllabification is conceived to be the cause of vowel coalescence, and not the other way around.

Notes * I am indebted to Harry van der Hulst, Haike Jacobs, Engin Sezer, Mieke Trommelen and Leo Wetzels for their careful reading, and criticism of an earlier draft. This paper is part of the research program entitled "Boundaries and Domains in Generative Grammar" (KUN-L-16-12-84). 1 The velar stop is deleted by a Velar Elision rule. For a detailed discussion of this and other consonant rules the reader is refered to Phelps (1975, 1979) and Halle (1978). 2 The ECP is a generalized version of four principles proposed in the recent nonlinear literature, the Obligatory-Contour Principle (Goldsmith (1976)),

68 the Shared-Feature Convention (Steriade (1982)), the Twin-Sister Convention (Clements and Keyser (1983)), and finally the Nuclear-Fusion Principle (Wetzels (1986), de Haas (1986)). 3 The processes under discussion are (cyclic) lexical processes, that is they only apply in derived environments. To be able to derive forms like [piu] from underlying /pi:+u/ we have to assume that resyllabification creates a derived environment in the cases at hand. 4 Harry van der Hülst (personal communication) has suggested an interesting generalization, if a one-valued feature approach is assumed. In his analysis the vowel system can be characterized by the privatives [i] = 'front', [u] = 'round', and [a] = 'low'. If this view is adopted VL (12) and VC (18) can be collapsed into A-Spveading (i) below. All regular cases of Kasem nominals can be explained in a straightforward way as shown in (ii): (i)

A-Spreading [a] V

V

(ii) a: [a] I V b: [a] I V I [u]

V

(i)

[a] r - ^ V V I

[i]/[u] V I [i]

[a] VT "

[i]/[u] (i)

[a] V I [u]

V I [i]

V

a+i

V I

e

a+u -» o

[i]/[u] VT

[a] V V I I [ u ] [i]

=

o+i -» ue (=we)

5 I assume that the SAP creates monosyllables. However, I have to admit that good arguments are hard to find. The changes /pe+(a)/ -» [pia], and /da+(a)/ -» [da:] seem to support this view, but the absence of glide formation in the singular /su+(a)/ -» [sua] seems to be a counterargument. I will leave this unresolved topic for further research. References CALLOW, J.C. 1965a Kasem nominals; a Study in Analysis, in Journal of West African Languages 2, p.29-36 1965b Collected Field Reports on the Phonology of Kasem. The Institute of African Studies, University of Ghana CHOMSKY, N. and M. HALLE 1968

The Sound Pattern of English. Harper and Row, New York

CLEMENTS, G.N. and S.J. KEYSER 1983 CV-Phonology: A Generative Theory of the Syllable. M.I.T.-Press, Cambridge, Mass. GOLDSMITH, J.A. 1976 Autosegmental Phonology. Indiana University Linguistics Club, Bloomington

69 HAAS, W.G. de 1986 Partial Syllabification and Schwa Epenthesis in Dutch, in Gramma 10 forthcoming An Autosegmental Approach to Vowel Coalescence HALLE, M. 1978

Further Thoughts on Kasem Nominals, in Linguistic Analysis 4, p.167-185

HARRIS, J. 1983

Syllable Structure and Stress in Spanish. M.I.T.-Press, Cambridge, Mass.

HAYES, B. 1981 A Metrical Theory of Stress Rules. Indiana University Linguistics Club, Bloomington 1982

Extrametricality and English Stress, in Linguistic Inquiry 13, p.227-276

PHELPS, E. 1975 Simplicity Criteria in Generative Phonology, Kasem Nominals, in Linguistic Analysis 1, p.297-332 1979 Abstractness and Rule Ordering in Kasem: A Refutation of Halle's Maximizing Principle, in Linguistic Analysis 5, p.29-69 STERIADE, D. 1982 Greek Prosodies and the Nature of Syllabification. Ph.D. diss. M.I.T., Cambridge, Mass. WETZELS, W.L.M. 1986 Phonological Timing in Ancient Greek, in W.L.M. Wetzels and E. Sezer (eds.) Studies in Compensatory Lengthening. Foris Publications, Dordrecht

Liliane Haegeman

Passivization in Flemish dialects

0.

Introduction

It is well-known that while indirect object passives are possible in English, they are ungrammatical in standard Dutch: (la) (lb) (2a)

(2b)

John gave Mary the book her Mary was given the book by John She Jan gaf Marie het boek haar Jan gave Marie the book her ''Marie werd het boek gegeven door Jan Zij k a r i e was the book given by Jan She

This contrast between English and Dutch has been related to the Oblique Case Parameter (OCP) Roberts (1983): English lacks oblique case, all case assignment is at S-structure. Dutch has oblique case and it is taken that oblique case is assigned at D-structure. Passivization is standardly taken to entail the loss of the case assigning properties of the V (absorption) but it does not affect the inherent oblique case. In Dutch, the underlying structure of (2b) would then be: ec "

(2c)

NP1

Marie haar OBL uNP2

het boek

gegeven

werd

- Case HP3

NP1 is empty since passivization suppresses the external theta-role, NP2 has oblique case, NP3 lacks case. In order to avoid violating the case f i l t e r , NP3 may move to the NP1 position and be assigned Nominative or alternatively NP3 is assigned Nominative in situ by a form of chain government (cf Den Besten 1980). To obtain the surface structure we assume of move-o

in

(1),

4 - 2 + be + en-

taken

from

3 - by + 1

This is what we now call a construction specific rule. It is well-known that this complex rule has undergone quite some changes over the years: it has b e e n broken up into several more elementary operations, the properties of which are determined by general principles such as binding, ECP etc. The introduction of be+en, however, remains unaccounted for: the appearance of the participle is still regarded as a k i n d of by-product, without it being clear what the real product is. The particular morphological form of the verb in passive constructions is thus regarded as coincidental. It is therefore not surprising to find a distinction being made between a perfect participle and a passive participle: there is no unity in these forms. They happen to be selected by the construction or by an auxiliary verb a n d the formal identity is again coincidental. I shall try to explain why these forms are identical and why this particular morphological form occurs where it occurs. There is another leftover from the classical analysis of passives. The transformation in (1) moves the subject NP to the right, placing it in a PP w i t h by. This phrase could be deleted if the NP was unspecified, but the interpretation of the DS-subJect was clear. What happens to the DS-subJect in passives under more recent analyses w i t h i n the framework of GB is less clear. I shall refer to the active DS-subJect as the external argument, borrowing a term, b u t not its content, from Williams (1981). Consider the so-called Burzio generalization as formulated in slightly different terminology in (2). (2) The verb does not project the external argument role on [NP,S] iff it assigns no structural Case Apart from the fact that (2) is a generalization and not an explanation, the formulation raises a fundamental question regarding the Projection Principle and/or the thematic criterion. If the external role is not projected onto [NP,S], what happens to this role? Suppose that it Is not assigned at all: then, if the role is still part of the verb's argument structure or thematic grid, a violation of the thematic criterion results. Hence, we should assume that the role is no longer part of the thematic grid, possibly as a result of some lexical rule of passive. However, this result is empirically untenable, since passives contrast w i t h e.g. ergatives and middles in ways that suggest that the external argument is present in passives in a w a y in w h i c h it is not present in the other constructions. Arguments that illustrate this for English can be found in Roberts (1985). Parallel arguments can be b a s e d on the following examples from Dutch.

96 (3) a. de tuin werd weloverwogen uitgebreid the garden was delibarately extended b."de tuin breidde zieh weloverwogen uit the garden extended itself deliberately (4) a. de boter werd opzettelljk gesmolten the butter was intentionally melted b."de boter smolt opzettelljk the butter melted intentionally (5) a. de boeken werden verkocht om ruimte te maken the books were sold for room to make b."de boeken verkochten goed om ruimte te maken the books sold well for room to make In the a-examples of (3)-(5) we find a passive which contrasts with the b-examples in licensing an agent-oriented adverb in (3)-(4), which is impossible in a reflexive middle and an ergative and in licensing a rationale clause in (5), which is excluded in middles. These examples indicate that the external argument is grammatically represented in passives, but not in e.g. middles, which, Just as much as passives, imply the participation of an agent. Hence, the external role remains part of the thematic grid in passives and, by the thematic criterion, it must be borne by a legitimate argument expression, i.e. an argument expression which is visible to the thematic criterion. We are then in a position to explain why the correlation that is stated in (2) holds in the case of passives: the Case which is normally assigned to some NP governed by V is used to make the external argument visible to the thematic criterion, i.e. the external role and the internal Case are not absorbed as it has been called, but the internal Case is assigned to the external argument. The argument expression, i.e. the expression which bears the external role and which is assigned Case is the participial morphology, henceforth referred to as PM. This hypothesis is formulated in (6). (6) PM bears the external argument role iff it has Case Unlike Roberts I do not regard PM to be a subject clitic, i.e. I do not assume that the external argument role is ever projected onto [NP,S]. First of all, such a projection is irretraceable at S-structure after the internal NP has been moved into its position. Secondly, PM does not exhibit any person and/or number distinctions, as one would expect from a regular clitic. Consider the following Indonesian examples. (7) a. All mem-baca buku ltu Ali TRANS-read book that b. Buku itu di-baca book that PASS-read c. Buku itu di-baca oleh All book that PASS-read by Ali d. Buku ltu ku-baca Book that I-read e. Buku itu di-baca oleh saja Book that PASS-read by me In (7a) we find a regular transitive active clause, which is signalled by the presence of the prefix mem. The prefix di is usually glossed as PASS (cf. Chung 1976 and references cited there). This gloss seems to be correct, as we can see in (7c), where the passive character is brought out by the occurrence of the

97 Indonesian analogue of the by-phrase. However, cU might also b e considered a subject clitic: it is the clitic counterpart of the third person singular pronoun dia and can be replaced by other subject clitics, like ku in (7d), the clitic counterpart of the first person singular pronoun aku. However, the clitic di clearly has developped into a general passive morpheme as is evidenced by (7e), where the oleh-phrase has sa.1a rather than a full NP. Historically, then, the morpheme of passive may develop from a subject clitic. This does not mean that it still is a subject clitic.

2. The passive participle Let us now examine whether the hypothesis in (6) can be supported with further evidence. Participles are used in three different types of construction: a. as prenomlnal modifiers (cf. (8a)). b. in passives (cf. (8b)). c. in perfect constructions (cf. (8c)). Usually, the participle is assigned to three different categories: a. an adjective (cf. Wasow 1977, 1980, Bresnan 1980, Williams 1981), b . a passive participle and c. a perfective participle. (8) a. een geschreven manuscript a written manuscript b. het manuscript werd geschreven the manuscript was written c. hij heeft het manuscript geschreven he has written the manuscript In Hoekstra (1984) I have argued that there is no motivation to call the participle an adjective in (8a) [1]. The arguments offered by Wasow and Williams to distinguish between a lexical and a transformational analysis of passives were shown to be either untenable or insufficient or superfluous (for further criticism of these arguments I refer to Levin & Rappaport 1984 and Dryer 1985). In fact, the properties of the participles in (8a) and (8b) do not differ In any way, except that the participle has a more restricted occurrence in prenomlnal position. These limitations are accounted for by independent principles which have nothing to do with the participle or its categorial status. In accordance with the hypothesis in (6), then, these participles would be analyzed as in (9). (9)

V'

V PMe fGe,0ll yi [_+CASE_|----'' The verb write has two thematic roles, 9e and 91, the former of which is assigned to PM in accordance with (6), and the latter to the NP object. Where NPi will end up depends on its environment. In (8b), the NP ultimately moves to the matrix subject position of the verb be, which takes the passive V-proJection as its small clause complement. In both cases, PM is an argument, and, as such, visible to the thematic criterion because it is assigned the Case of the verb. In this respect, the participle of a verb like perish differs from that of write: PM is not theta-marked by the verb since perish does not have an external role, i.e. it is an ergative verb. This is harmless since ergative verbs also fall to have a Case property.

98 3. The perfect participle Turning now to the participle in (8c), we note that this use of the participle is an innovation in the development of the Indoeuropean family. Greek a n d Latin h a d a synthetic perfect. To be sure, there was a verb habere in Latin that could be combined w i t h a participle, as in (10a), b u t the interpretation of (10a) differs from the interpretation of the synthetic perfect in (10b). (10) a. habemus oppidum obsessum have-we town besieged b. obsedimus oppidum besiege-PERF-we town The interpretation of (10a) is stative and does not necessarily imply that we are the besiegers. In fact, the sentence may very well m e a n that some other party has b e s i e g e d our town. (10b), on the other hand, can only m e a n what the English sentence "we have besieged the town" means: we are the besiegers. A similar difference can be made in present-day English, as is illustrated b y the pair in (11). (11) a. we have fugitives hidden b. we have hidden fugitives (10a), then, w o u l d be analyzed as (11a) would: have/habere is a m a i n verb taking a small clause complement, the analysis of which differs in no essential way from the analysis in (9). Interestingly, and quite expected from the present perspective, habere could not b e combined w i t h participles of one-place verbs. Only after the construction in (10a) gradually replaced the construction in (10b) a n d started functioning as a perfective auxiliary, the verb could enter into this combination. What, then, is the essential difference between habere as a main verb and habere as a perfective auxiliary? I w o u l d like to argue that habere functions as an auxiliary w h e n it provides Case for PM. In these terms, then, the examples in (11) can be analyzed as in (12). (12) a. we have [SC fugitives

l

C

j

hlde+PM]

L.J

b. we have hide+PM fugitives

L

4::LJ

In b o t h cases, PM receives Case and is therefore an argument according to (6). In the SC-structure, this Case is provided by the verb, whereas it is provided by the auxiliary in the perfect construction. This difference explains why habere can only be combined with participles of intransitives after it has become possible for habere to assign Case to PM: it is Impossible to have a SC complement which solely exists of a predicate, cf. "*I want laughed, "I want off my ship" etc. The shift in Case marking has a further consequence, as we saw above: the external argument of the participle is necessarily identical to the subject of the auxiliary. I shall assume that these two properties constitute the essence of the notion auxiliary: (13) if a verb assigns Case to PM, its subject and PM constitute an argument chain

99 We shall consider (13) in more detail below. It explains why a construction such as "I have laughed" cannot mean, analogous to (10a), that persons other than me laughed.

4. Auxiliary

selection

A further confirmation of this analysis comes from Dutch and other languages that make a distinction between two perfective auxiliaries, one a cognate of habere, the other a cognate of essere (be). In these languages, the participles behave as they should, given (6). As I have argued in Hoekstra (1984), the verbs selecting the essere-cognate zl.1n in Dutch are ergative verbs, i.e. verbs that do not have an external argument as a lexical property. Participles of such verbs distribute like participles of transitive verbs (cf. (8)) and do not combine with the perfective auxiliary hebben. Thus, (14) is ungrammatical. (14) *de soldaten hebben gesneuveld the soldiers have perished The ungrammaticality of (14) is expected since PM would receive Case from hebben and would, by (6), be an external argument, which is impossible as sneuvelen is ergative. Although this confirms the hypothesis in (6), it raises a problem at the same time, because, as we can see in the translation of (14), the English counterpart of sneuvelen freely combines with have and is therefore in a Case marked position. This is true for ergative verbs in general, i.e. the distinction between have and be as perfective auxiliaries no longer exists in modern English. It is interesting that such paradigmatic levelling is always to the advantage of the habere cognate [2]. Why, then, is the English translation of (14) grammatical? One answer would be that English makes no distinction between intransitive verbs, i.e. that all intransitive verbs have an external argument, possibly lexically derived from an internal argument. This is the position taken by Keyser ft Roeper (1984). However, although this position would make it possible to explain why all intransitives in English combine with have, it makes it impossible to explain the difference in the use of the participle in prenominal position. As in Dutch, some intransitive verbs allow their participle to occur in prenominal position, while others do not. It is precisely the set of counterparts to Dutch ergative verbs that allow this distribution [3]. (15) a. ergatlves the escaped prisoners the fallen leaf a descended testicle the degenerated society curdled milk our departed ancestors a stranded boat a diseased person

b. unergatlves *a worked housewife *a laughed pupil "a yawned student "the walked children "the shouted fishermen »the slept baby "a jumped contestor "a sung choir

If we adopt the hypothesis in (6), we can straightforwardly account for the ungrammaticality of the examples in (15b). Since these verbs take an external argument, PM of these verbs is in a position where it receives a thematic role.

100 Hence, b y (6), it must receive Case, but there is no Case assigning category available. The same property explains why these verbs select hebben as their perfective auxiliary in Dutch. Returning to our original question, then, we note that we have a category which may bear a thematic role, which Is Case marked, but w h i c h fails to have a thematic role assigned to it. These properties together are a definition of a dummy argument, i.e. an expletive. The paradigmatic levelling can thus be regarded as a consequence of allowing expletives. Notice that this account straightforwardly explains w h y paradigmatic levelling could never favour the essere cognate without resulting in a split ergative system. The existence of an expression that is Case m a r k e d but not theta m a r k e d is harmless. If the essere cognate won, this w o u l d result in theta-marked expressions (PM), which fail to be visible. Before turning to further evidence corroborating (6), I want to demonstrate that there is a distinction between a main verb and an auxiliary use of the Dutch verb hebben too. Parallel to (10) and (11) we have a D u t c h example like (16). (16) is ambiguous between the readings of (11a) and (lib). There are two ways in w h i c h we can disambiguate (16): either we a d d yet another form of hebben, as in (17a). This results in a perfect of the stative interpretation of (16), i.e. it is the perfect of the structure w i t h the m a i n verb hebben taking a small clause complement. The other way of disambiguating is to shift the order of hebben and the participle. This only leaves the perfective interpretation, i.e. parallel to (lib). (16) dat w e onderduikers verstopt hebben that we fugitives hidden have (17) a. dat we onderduikers verstopt hebben g e h a d that we fugitives hidden have had b. dat we onderduikers hebben verstopt The results in (17), especially in (17b), are as expected. The structure in (16) can be analyzed as either (18a) or (18b), parallel to (12). Clearly, the analysis in (18a) is not applicable to (17b), due to the order of hebben and the participle. Our analysis predicts these results. (18) a. dat we [SC onderduikers verstop+^M] hebben C ' C C

""

_J

b. dat we onderduikers ers verstop+PM , hebben C

t

?

t . .JJ

5. Impersonal passives Let us now turn to a second piece of evidence supporting (6), bearing in m i n d our conclusion above that English differs from Dutch in allowing expletives. Consider the well-known contrast between English and D u t c h relating to the possibility of Impersonal passivization, illustrated in (19). (19) a. dat er gelachen wordt b. "that it is laughed

101 The ungrammatlcality of (19b) follows Immediately from the hypothesis in (6): the verb lachen/laugh has an external thematic role, as is evidenced by the impossibility of using its participle in prenominal position (cf. (15b)). Hence, PM is in a theta position and should therefore receive Case. Since the verb is intransitive, however, no accusative Case is available and (19b) is out by the thematic criterion. Why, then, is (19a) grammatical? Given our reasoning, PM should receive Case. As in (19b), this cannot be a Case assigned by V. There is another possibility, the nominative Case assigned by INFL. This possibility does not exist in English, since it already receives and requires Nominative Case. The Dutch expression er does not require Case: it is an adverb and not a pronominal element like ^t. As we saw, English does allow expletives. In Dutch, then. Nominative Case is not always assigned to a specific structural position. In (19a) it is assigned to PM. This is certainly not the only instance of INFL assigning Case within VP. Various people have argued that Nominative Case assignment in Dutch and German inside VP is possible, e.g. Den Besten (1982) who proposes a notion of chain-government [4]. The difference between languages with respect to impersonal passivization seems to be related to yet a further difference. It was demonstrated above that the external argument appears to be grammatically present in passive sentences because it can control the subject of certain adjuncts, specifically rationale clauses. This result is to a certain extent surprising from the point of view of the so-called Visser Generalization (cf. Bach 1980). This generalization accounts for the patterns of grammaticallty in the examples in (20)-(22), which generally are taken to show that the controlling argument must be present. In passives this would seem not to be the case if the external argument is taken to be the controller. (20) a. John tried [PRO to open the door] b.*It was tried [PRO to open the door] (21) a. John promised Mary [PRO to open the door] b.»Mary was promised [PRO to open the door] (22) a. John forced Mary [PRO to open the door] b. Mary was forced [PRO to open the door] The ungrammatlcality of the b-examples in (20)-(21) is thus explained by appealing to the theory of control. This analysis does not provide a basis for the difference in grammaticallty with the Dutch counterparts of these examples, shown in (23)-(24). (23) Er werd geprobeerd [PRO de deur open te doen] There was tried PRO the door open to do (24) Er werd beloofd [PRO de deur open te doen] There was promised the door open to do The explanation we just gave to account for the grammaticallty of impersonal passives carries over straightforwardly to these cases: in Dutch, the designated argument that functions as the controller is visible, because it can be assigned Nominative Case, but In English it is not [5].

6. Further confirmation A final piece of evidence in favour both of our analysis of participles and of perfective auxiliaries derives from French. Drijkoningen (1986) notes that the

102 present participle of the perfective auxiliary être in (25) is optional as it is w i t h plain adjectives, whereas the present participle of the perfective auxiliary avoir in (25d) is not. He explains this by assuming that - a n t instantiates INFL and that INFL is a necessary ingredient for the assignment of the external role. The latter assumption strikes me as unmotivated a n d w o u l d seem to run into problems w i t h infinitival complements of perception verbs. The difference between étant a n d ayant w o u l d be h a r d to explain if the two perfective auxiliaries are regarded as variants w h i c h are lexically selected b y the m a i n verb. The analysis defended here explains the difference in a straightforward fashion. PM in (25a) bears the external argument role and receives Case from the verb, w h i c h is transitive. In (25b), PM does neither receive a thematic role, nor Case since the verb arriver is ergative. In (25d), i.e. a verb that assigns an however, we have an intransitive unergative verb, external thematic role. Hence, PM is in a theta-position a n d therefore requires Case. Since the verb is intransitive. Case cannot be provided b y the verb itself. Hence, some Case assigning category is called for and ayant functions as such. Thus our account not only explains the difference in choice of the perfective auxiliary, it also explains why ayant m a y not be left out. (25) a. (le soldat),(étant 1 t u é

I

fi |

b. (le roi),(étant 1 arrivé

l fi I

c. (le professeur), (étant!malade

i 0 1

d. le professeur,Jayantlparlé

N I fi )

)

e. le bateau,[ayanticoulé

It is well-known that the selection of the perfective auxiliary in F r e n c h does not bring out the difference between ergative a n d non-ergative verbs as clearly as the choice between hebben and z l j n in Dutch or avere and essere in Italian i.e. some ergative verbs in French select avoir. In this respect, the process of paradigmatic levelling, which has taken place in e.g. English a n d Spanish, is operative in French to a limited extent as well. So, the verb couler is ergative, but selects avoir nonetheless. However, as avoir is not required to Case mark PM, we predict that ayant can b e left out. This prediction is borne out, as is evidenced by (25e).

7. Conclusion The conclusion of this paper is that the hypothesis in (6) is well supported. This made it possible for us to explain why the verb in passives shows up in participial form and what the unity of participles is. The verbal affix functions as an argument under the same conditions as normal arguments, i.e. it may be assigned a thematic role and must therefore be Case marked. In certain languages it may be Case marked without receiving a thematic role. Under these circumstances the participial affix is in fact a dummy argument. It was shown that English allows dummy arguments in both positions where the external argument is projected if there is one, i.e. in the syntactic subject position and in the participial slot.

103 Notes *

The research for this article was part of the Leiden University Project "Word Order and Paradigmatic and Syntagmatic Structure".

Research

1.

I do not exclude the possibility of lexicalized adjectives derived from verbal participles. It is unclear whether the hypothesis concerning participles developed in this paper applies to these adjectives as well.

2.

If a language forms compound tenses b y means of a participle and an auxiliary a n d there is no correlate to habere. we w o u l d predict that this w o u l d result in a split ergative system, i.e. in these compound tenses the subject of intransitives w o u l d be treated as the object of transitives, due to the fact that such compound tenses w o u l d necessarily be "passive" in the sense outlined here. Such systems do Indeed appear to exist.

3.

Some of the participles of ergative verbs sound a little awkward in prenominal position, but this awkwardness disappears if an adverb is added, as is shown b y the examples in (i). In this respect these participles do not differ from the participles of transitive verbs which predicate over a n o u n corresponding to their object (cf. (ii)). ( i) a (recently) arrived shipment the (generally) slackened zeal the (lately) expired treaty the (recently) perished soldiers (il) a (widely) r e a d book a (recently) erected pavillion a (much) debated issue a (well) loved comedian

4.

It can be demonstrated that it is indeed the presence of Nominative marking w h i c h is relevant for the occurrence of impersonal passives. So, impersonal passives can not occur In infinitival constructions or in nominalizations, as is shown by the examples in (1), (cf. H o e k s t r a & Wehrmann 1985). (i) a. »gelachen w o r d e n is storend laughed being is distracting b. "het is storend om gelachen te worden It is distracting for laughed to be

5.

This explanation applies to control structures that are k n o w n as obligatory control structures, i.e. the structures of w h i c h the Visser generalization has b e e n argued to hold. Control is also possible b y Implicit arguments. The status of such implicit arguments vis-a-vis the Projection Principle a n d the 0-criterion is as yet unclear.

References Bach, Emmon 1980 In defense of passives, in Linguistics and Philosophy 3, p. 297-342 Bennls, Hans forthc. Gaps a n d Dummies, diss. Tilburg

104 Bresnan, Joan 1980 Passive In Lexical Theory, paper MIT Chomsky, Noam 1957 Syntactic Structures. Den Haag, Mouton Chung, Sandra 1976 On the subject of two passives in Bahasa Indonesia, in C.N. Li (ed.), Sub ject and Topic, New York, Academic Press, p. 57-98 D e n Besten, Hans 1980 Government, Syntaktische Struktur u n d Kasus, in M. Kohrt & J. Lenerz (eds), Sprache, Formen u n d Strukturen, Tubingen, Max Niemeyer Verlag Dryer, Matthew 1985 The role of thema:tic relations in adjectival passives. In: Linguistic Inquiry 16:2, p. 320-326 Drijkoningen, Frank 1986 Over de volgorde v a n (hulp)werkwoorden en syntactlsche affixen. Paper presented at the TIN-meeting, January 25, 1986 in Utrecht. Hoekstra, Teun 1984 Transitivity, Dordrecht, Foris Publications. Hoekstra, Teun & Pim Wehrmann 1986 De nominale infinitief, in GLOT 8:3, p. 257-274 Keyser, Jay & T. Roeper 1984 On the middle and ergative constructions in English, in Linguistic Inquiry 15:3, p. 381-415 Levin, Beth & M a l k a Rappaport 1984 The formation of adjectival passives. Paper presented at the XVth NELS meeting, Providence November 1984 Roberts, Ian 1985 The representation of implicit a n d dethematlzed subjects, diss. USC Vasow, Thomas 1977 Transformations and the lexicon, in P. Culicover, T. Wasow & A. Akmajian (eds), Formal Syntax. New York, Academic Press, p. 327-360 Williams, Edwin 1981 Argument structure a n d morphology, in The Linguistic Review 1, p. 81-114

Harry

van der Hülst,

Maarten

Mous & Norval

Smith

The autosegmental analysis of reduced vowel harmony systems: the case of Tunen 1. Introduction In the "simplest case" of vowel harmony there are two non-overlapping sets of cooccurring vowels (harmonic classes) such that each element in one set has a harmonic counterpart in the other set which differs from it only with respect to its value for the harmonic feature. Consider the following hypothetical vowel systems, in which vowels have been arranged according to the harmonic class they belong to: (1)

a.

[+ATR] i e

u o

[-ATR]

I E

b.

U O a

A

[-B]

[+B]

c.

[+R]

i i i i u e o A o ae a

u o

[-R]

u o 0

4 e

We refer to these systems as full systems and to vowel harmony systems as full harmony. 1

i A a based

on

such

The treatment of full harmony systems is trivial and not very interesting. Fortunately, most cases, of vowel harmony are not so simple. Usually the complications have arisen due to the fact that some combinations of phonological properties tend to be avoided, which eventually has lead to a language specific constraint against these combinations and, consequently, mergers among v o w e l s . 2 Let us consider possible examples of what are called reduced systems. Next to a full Advanced Tongue Root (ATR) system, as in (la) one commonly finds systems as in (2): (2)

a.

[+ATR] i e

u o

Dashes indicate gaps systems are:

[-ATR] I E

b.

U O in

[+ATR] i

u e

the

system.

[-ATR] o

Similarly,

E

O

common

reduced

palatality

106 (3)

[-B]

b.

[+B]

[-B]

[+B]

i e ae Reduction in rounding systems is not exemplified here. Stewart (1971) provides a functional explanation for the fact that the systems as exemplified in (2) occur so frequently. The vowel /A/ o n the one h a n d a n d the vowels /I, U/ on the other, are the most commonly eliminated b y sound changes, because they involve "the most awkward of the combinations of points o n the low/mid/high scale w i t h points on the root-unadvanced/root-advanced scale as the root naturally tends to be pushed backwards w h e n the highest part is low a n d pulled forwards w h e n the highest part is high." (Stewart,1971:199). W i t h respect to reduced palatality systems the validity of the same reasoning is obvious, given the markedness of unrounded back nonlow vowels. Seductions in the vowel inventory lead to complications in the harmony systems in that alternations are either neutralized or changed in character. To illustrate this let us consider in some detail the reductions in [ATR] systems. There are various ways in which the vowels /A/ a n d /I, U/ may be lost. The fact that they involve awkward combinations of phonological properties does not y e t imply that they should change according to one set rule. One expects that the m a r k e d vowles will merge with vowels w h i c h are either articulatorily or acoustically "close" to them. For the low vowel this implies that it will merge w i t h some non-high vowel, and for the h i g h vowels that they will merge with some non-low vowel. Williamson (1973, 1984) who describes a variety of reduced systems, represents a full ten vowel system as follows:

In Williamson's papers a n d others dealing w i t h reduced systems, routes b y which /A/ may merge w i t h other segments are reported:

the

following

a As for the high vowels /I/ a n d /U/, the following routes are found to occur:

E

/E/, 101 /I/, /U/ -> /e/, /o/

Alternation /a/ - /E/ (a) /a/ - /o/ (b) /i/ - /e/ (c) /u/ - /o/ (d)

Change in class No change in height /A/ -> /a/ /I/. /U/ -> /i/, /u/ No change in class

Change in height /A/ -> /e/, loi /I/, /U/ -> /E/, /O/

neutralized (e) neutralized (f)

/a/ /a/ /i/ /u/

-

/e/ /o/ /E/ /0/

(g) (h) (i) (j)

The above scheme implies a classification of the different routes into three groups. If an awkward vowel merges, it either falls together with a vowel within the other harmonic class, or it falls together with a vowel belonging to its own class (necessarily of a different height). In the first case it either falls together w i t h its harmonic counterpart or with a vowel of a different height. In (8) we illustrate the possibilities; taking the m a r k e d vowel /I/ as an example. (8)

III III

-> 111 -> lei

/I/ - > IE/

(f) (c) (i)

Alternation /I/ - /i/ is neutralized Alternation /I/ - /i/ is changed to /e/ - /i/ i.e. the ATR difference is replaced by a height difference. Alternation /I/ - /l/ is changed to /E/ - /i/ i.e. a height difference is added to the ATR difference

That the complications for the harmony system are different is easy to see. In the first case we will get disharmony effects, since the instance of the vowel 111 w h i c h derives from */I/ will now cooccur with [-ATR] vowels. In the third case no disharmony results, but we do get the complication of having an "extra" change in height. The second case combines both complications, i.e. disharmony results since some /e/*s will cooccur with [-ATR] vowels and the e x t r a height alternation is present as well.1* The first case (in w h i c h an alternation is neutralized leading to disharmony effects) has explicitly been noted in the descriptive literature on vowel harmony. Vowels w h i c h have been merged with their harmonic counterpart have often been called neutral. One must be careful, however, to distinguish two kinds of neutral vowels. In one type of case the neutral vowels appear to be transparent in the sense that the harmonic requirement, as it were, looks right through them, i.e. vowels occurring to the left of these transparent vowels must harmonize w i t h vowels to the right (and vice versa), Just as if the neutral segments were not there. Suffix vowels w h i c h are adjacent to a transparent vowel occurring in the final syllable of the stem harmonize with the first non-transparent vowel to its left, ignoring the fact that the transparent vowel intervenes. In the other type of case the neutral vowels are opaque, i.e. it is not necessary that vowels occurring on either side harmonize with each other. Also, suffix vowels w h i c h are adjacent to an opaque vowel harmonize w i t h it. 5 The second and third case have received less explicit attention, know, not even at the terminological level.

as far as

we

108 2. Theoretical

framework

The issue of dealing with disharmony effects (i.e. neutral vowels) is discussed at length in V a n der Hülst and Smith (1986a), hencefort HS. The focus of this paper is on the treatment of (extra) height alternations, although our examples also involve disharmony effects. Before we study this case in some detail we will give an outline of the theoretical framework adopted b y us.

2.1

The representation of neutral vowels

From the outset, opaque and in particular transparent vowels have been problematic for earlier versions of autosegmental phonology. As an essential characteristic of this model we take the fact that each feature is represented on one tier only. The Impression that vowels agree in harmonic value across a third vowel which has an opposite value has seduced several phonologlsts into abandoning this essential characteristic, a n d this, in our view, obscures one of the most fundamental insights o n w h i c h autosegmental theory Is based. In many publications, including some of our own (Van der Hulst and Smith 1982, Booij 1984, E w e n and V a n der Hulst 1985, Vago 1984, Lieber 1985) it is p r o p o s e d that we should allow for the possibility of specifying a single feature on more than one tier. Let us refer to this as the possibility of having tier duplication. The details of the various proposals certainly differ, b u t they all boil down to the Idea that vowels w h i c h are transparent are segmentally specified,such that a morpheme level autosegment can spread across them. 6 We believe that tier duplication must be avoided for two reasons. Firstly, we believe that, all things being equal, we should prefer the phonological model which comes closest to phonetic reality. This is in fact what Postal's Naturalness Condition says. From this perspective It seems that as long as human beings have one tongue root, we should allow just one tongue root tier in our model. Secondly, it will be obvious that tier duplication Increases the descriptive power of the model a n d thus should only be allowed If no other possibilities remain. In Van der Hulst and Smith (1986a) an account is 7 proposed which makes no use of tier duplication. H S adopt a single-valued feature approach in which the universally m a r k e d pole of phonological features is represented as the lexical value. The opposite value results from phonetic interpretation, and is called the default value H S furthermore adopt a view that all features which are not involved in an alternation must be underlylngly specified. In accordance w i t h this view, neutral segments which have the lexical value require the presence of this value in their morphemic representation. In Hungarian for example, w h i c h has backness harmony, transparent segments would be specified in the lexical stratum as associated to the harmonic feature and in this sense they are accessible vowels, Just like the normal harmonizing vowels. The difference between harmonizing a n d transparent vowels is that the latter are inherently specified. A crucial property of transparent segments is that they do not spread, i.e. they do not impose their value on neighbouring vowels. HS therefore adopt an idea first advanced by Halle and Vergnaud, namely that the universal association conventions (AC's) only apply to floating autosegments. If we accept this, we can represent words with a neutral vowel as in (9).

109 (9)

[i] I r A d i r + n A k 'eraser' dative

[i] i [radirnak] and not

[i] N [radirnek]

The two vowels straddling the neutral vowel will acquire the default value a n d the correct surface form will be derived. H S explain why transparent vowels in suffixes "let through" the [i] value if they are preceded b y a floating Instance of [i] b y appealing to the Obligatory Contour Principle (OCP). The relevant situation arise in the following example (cf. 10a). The configuration in (10a) is interpreted as an OCP violation and HS assume a convention which turns it into (10b) 9. (10)

a. [i]

[j]

b.

[i]

[i]

T I b + i + n A k T I b + I + n A k (petname on the bals of the name Tlbor).

->

[tibinek]

Observe that this proposal seems to presuppose that the A C ' S can distinguish between derived a n d underlying association lines, but this is not necessary if we assume that AC's apply post-cyclically. It is of some Importance to realize that neutral vowels which have the lexical value m a y under special circumstances act harmonically. This happens for example in so called neutral vowel morphemes where w e f i n d the following contrast: (11)

a. [i]

[i]

v I z + n A k 'water'

-> [viznek]

b. [i]

(11) m a y arise since Independent from the inherent vowels roots are specified as either front ([i]) or the logical possibility of the existence of front vowels w h i c h cannot be represented as in (12 a or b) in (11a) as required b y the OCP: (12)

a.

[1] [i] I v i z

[i]

I h i d + n A k 'bridge'

I -> [hidnak]

specification of neutral back (0). This results in roots w i t h only neutral but must be represented as

[1] [i] I v i z

The representation in (11a) will trigger application of the AC'S, w h i c h means that the /I/ in /viz/ sets off its own harmonic domain a n d can thus not be regarded as transparent. The lesser formal complexity of (11a) suggests that morphemes having any neutral vowels should preferably behave as non-transparent (i.e. harmonic) rather than as transparent. Although neutral segments w h i c h have the lexical value can behave either transparently or opaquely neutral segments having the default value seem to behave consistently opaquely.8 Let us now turn to the formal representation of opacity.

110

Opacity Includes, among others, the representation of dlsharmonlc roots. current approaches disharmonic roots are represented as follows: (13)

[i]

I burO

[i]

+ nAk

->

[i]

I [buronak]

I kOsztum + nAk

In

[i]

K

-> [kosztumnek]

As shown, the [i] of the second example must spread to the suffix. Yet we represented it as underlyingly linked. The inevitable conclusion is that if we use lexical association lines for transparent vowels, we cannot use the same mechanism for opaque vowels. In order to solve the problem of representing dlsharmonlc roots, we should like to say that the [i] autosegment is floating, but includes only one of the root vowels in its scope. It has been shown that there are cases where the autosegments are bound to particular prosodic categories other than the phonological word without being associated to elements in these categories. Examples Involve autosegments spreading within the syllable or the foot. We will claim that the smallest prosodic category is a category which comprises a single skeletal point. Ve call it the segmental domain. Since the prosodic hierarchy forms an Independent plane in a three-dimensional phonological representation, imposing limitations on the spreading of an autosegment involves the projection of a prosodic category P onto the relevant tier. Hence we will say that in dlsharmonlc roots the segmental domain is projected onto the harmonic tier. This of course results in a situation in which an autosegment may both be floating and segmentall.y bound. We represent this situation as follows: (14)

[i] ( X )

( X )

[i]"

—

( X )

X

u

—

X

Prosodic domains projected onto an autosegmental opacity: you cannot go In, but you can get out.

X tier

create a "one-way"

The mechanism of segmental binding gives us a complete formal account of opacity. We can use it for example to deal with the opacity of low vowels in Akan. Low vowels in Akan are always [-ATR], which is the default value. In the lexical derivation the following constraints holds (cf. 15a): (15)

a.

"ATR

I V

b.

ATR

We assume that (15a) automatically imlies (15b) Just in the case the vowel at issue belongs to the set of harmony-bearing units. If a particular harmony-bearing unit cannot become associated to the harmonizing feature, it will automatically be opaque in the sense Just proposed. In other words it is inaccessible to the lexical value. In cases of this type opacity is predictable on phonological grounds, which is not the case in disharmonic roots where the opacity is an idiosyncratic property of particular morphemes.

111 2.2

Underspeclflcatlon

As said above, we assume a single-valued trldlrectlonal feature system (cf. Even and Van der Hulst 1985). For our analysis we only need to mention a subset of the features which are relevant for vowels: (16)

VOCALIC FEATURES [i]. [u], [a], [A]

A skeletal point associated to the feature [A] (i.e. [A(dvanced tongue root)]) Is phonetically interpreted as tongue root advanced, whereas skeletal points not associated to [A] are Interpreted as tongue root unadvanced. He use the term default value for the phonetic interpretation of the absence of a phonological feature. A point associated to "pronounceable" segment. (17)

[i] I /X/ -

each of these features separately results in a We will refer to the vowel segments as /i/, /u/ etc..

i

A ten vowel system comprising five advanced and five unadvanced vowels is represented as in (18). Since no confusion is possible, we represent features In diagrams without square brackets: (18)

A ~i

x /I/

i

u

x

x

/E/

/a/

x /0/

I 1

u

x

x /U/

A I

A

i

x /i/

A

u

x /e/

A — [ A ] tier I

x

x

/A/

/o/

I U — [ i ] / [ u ] tier

/u/

The notation used here to represent phonological representations resembles that of Kaye, Lowenstam and Vergnaud (1985). Tiers are piled on top of each other, which is purely for graphical conveniences, and not meant as expressing the idea of coplanar features (cf. Archangeli 1985). Notice that the feature [i] and [u] occur on the same tier. We take over an idea of Kaye, Lowenstamm and Vergnaud (1985) that tiers may be conflated in particular languages. In (18) the features [i] and [u] occur on the same tier, which excludes the possiblity of associating a point to both [i] and [u]. If the [i] and [u] are not conflated we create the possiblity of specifying the vowels /o/ and /vi/. If on the other ahnd all three lines are conflated we get three vowel systems only having the vowels /i/, /u/, and /a/. Cf. Van der Hulst and Smith (1985), Rennlson (1985). With regard to the issue of underspecification, we hold the view that unpredictable and predictable features are specified lexically unless they are involved in an alternation. Predictability (and corresponding "low cost") is captured by formulating redundancy rules. These redundancy rules may apply in the course of a derivation if their structural description is met.

112 Underspecifled segments acquire a full representation then in either of two ways. They may get associated to a feature present in the context (either floating or associated to som eother skeletal point) or a RR may apply a n d fill in a feature.

3. Tunen In this section we will analyse the vowel harmony system of Tunen. This language is spoken in Cameroon in the area surrounding Ndlklnlmeki (NE of Duala) by approximately 35,000 people. It is a Bantu language, A.44 in Guthrie's classification (see Guthrie, 1967). The available data on Tunen are contained in a wordllst a n d a grammar b y Dugast and analysed within a linear framework in Mous (1986).

3.1. Basic facts The vowel harmony system of Tunen is of the cross-height type, w h i c h is characteristic for West Africa. The harmonic feature Is advanced tongue root, [ATR]. A language w i t h classic cross-height vowel harmony has t e n vowels. These can be divided on the basis of harmony into two matching sets: (19)

+ ATR: 1 - ATR: I

e E

A a

o 0

u U

All the vowels w i t h i n a w o r d are of one set. The dominant value is [+ATR]. A vowel harmony system in w h i c h affixes contain only recessive, [-ATR] vowels, is referred to as root control. In other words, only affixes will display alternations in terms of the feature [ATR]. In Tunen, however, suffixes as well as prefixes can be dominant, I.e. can contain [+ATR] vowels. Tunen lacks /I/ a n d /U/. Moreover, /e/ does not occur in roots. It does occur in certain prefixes, where it alternates w i t h /1/ in a dominant (+ATR) environment. In roots a n d suffixes (and in some prefixes) /E/ alternates w i t h /i/. As a consequence of the merger of /U/ w i t h /o/ there are two possible alternations involving /o/. One is the regular alternation /0/ /o/, which occurs in roots; another is the irregular alternation of /o/ in a recessive (-ATR) environment w i t h /u/ in a dominant (+ATR) environment, parallel to the /e/ - /1/ alternation. This /o/ -/u/ alternation occurs in roots, prefixes a n d suffixes. Roots that contain no other vowels than /o/ fall in two categories: a) Dominant: roots which are themselves invariant, and cause affixes to be [+ATR]. b) Recessive: roots w h i c h do not cause affixes to alter and in which /o/ becomes /u/ in a dominant (+ATR) environment. The alternation /a/ - /A/ is regular and occurs in roots and prefixes, but there are a few words w i t h /e/ as an optional variant of /A/, especially before /y/. Thus the harmonic sets of Tunen vowel harmony are: (20)

dominant: 1 A o u recessive: e A E a 6 o

We will now give some examples of these alternations.

113 3.1.1. The alternation /a/ - /A/ In verb roots /a/ changes to /A/ If followed by the dominant

(21)

falab tal bakon

causative

suffix

fwAlAbi "to cause to build" tAli "to cause to put down" bwAkuni "to cause to separate"

"to build" "to put down" "to separate"

(labial consonants are rounded before /A/) The emphatic demonstrative dominant pronominal prefix. (22)

(nearby)

root

motana "this!, for noun class 1" batana "this!, for noun class 2"

/tana/

mutAnA mitAnA

changes to /tAnA/ after a

"this!, for noun class 3" "this!, for noun class 4"

The noun class prefix /ma-/ (class 6) changes to /mwA-/ stems and similarly /ba-/ (class 2) to /bwA-/, e.g. (23)

mabat mahOk

"clothes" (class 6) "axes" (class 6)

bangOt "lizards" (class 2) balEhan "advisers" (class 2)

before

dominant

noun

mwAbll "oil palms" (class 6) mwAkAnu "bellows" (class 6) bwAsAlun"witnesses" (class 2) bwAllmwAn"servants" (class 2)

3.1.2. The alternation /0/ - /o/ The vowel /0/ does not appear in affixes and therefore /Of alternates with /o/ only in stems; in verb stems before the causative suffix /-l/, and in the stem /-mOtE/ "one, some" after certain dominant numeral prefixes. (24)

kOf bOt OmOtE

"to close a door" "to start" "one, class 1"

kofl botl omoti

"to cause to close a door" "to cause to start" "one, class 3"

3.1.3. The alternation /E/ - /i/ The direct [+ATR] counterpart of /E/, /e/, does not appear in roots, with a few exceptions, where it is an optional variant of /A/. The [+ATR] vowel that alternates with /E/ in roots, however, is always /i/. The /E/ in the stem /-mOtE/ "one, some" changes to /1/ after dominant prefixes (cf. (24)). In verb roots /E/ changes to /I/ if followed by the dominant causative suffix /-i/. For example: (25)

fEf

"to blow"

fifi

"make blow"

In affixes /E/ becomes /I/ after [+ATR] verb stems in the applicative suffix, /-En/, in the stative suffix /-Em/, in the neuter suffix /-E/, in the reciprocal suffix /-Enan/, and before [+ATR] class 19 noun stems In the nominal prefix /hE-/.

114 (26)

falabEn "to b u i l d for somebody" fOlEn "to borrow for somebody"

fAnin hukln

"to exchange for somebody" "to blow for somebody"

hEbOb

hibll

"bracelet" (noun class 19)

"ring" (noun class 19)

3.1.4. The alternation /e/ - /i/ The vowel /e/ only appears in certain alternates w i t h /i/ before [+ATR] stems. (27)

ebak enEnE

prefixes

"tree of savanne" (class 7) "occult power" (class 7)

ibwA ilik

before

[-ATR]

stems.

It

"husband" (class 7) "orphan" (class 7)

There Is a tendency towards reharmony in that these n o u n prefixes containing /e/ have a variant w i t h /E/ before stems w i t h /E/ as first stem vowel, to a lesser extant before /0/, and still more rarely before /a/.

3.1.5. The alternation /o/ - /u/ In verb roots /o/ alternates w i t h /u/ if followed b y a [+ATR] suffix. (28)

bol hon kolah

"to "to "to in

disappear" awake" enlarge a hole a palm wine tree"

bull hunl kulAhl

In noun prefixes /o/ is replaced by /u/ (29)

mokas

"branch" (class 3)

"to cause to disappear" "to wake up" "Idem, w i t h causative suffix"

before [+ATR] stems. munA

"grave" (class 3)

The verbal separative suffix /-on/ changes to /-un/ after a [+ATR] stem or before the dominant causative suffix /-i/. (30)

talon tEkon

"to lift" "to throw down"

tinun bwAkuni

"to untie" "to cause to separate"

That this /o/ belongs to the recessive category can be shown lack of influence on [-ATR] vowels. (31)

hE+bak+on+En — > hE+loNom+En — >

hEbakonEn hEloNomEn

"hibAkonin »hiloNomin

by

its

complete

"separating Instrument" "telephone receiver"

There are some exceptions to the rule about the vowel of the separative suffix. (32)

fat+on+i —> alobot+on+i — >

fatoni alobotoni

"cause to open" "cause to reply to proposal of marriage"

115 3.1.6. Non-alternating /o/ In some stems /o/ does not alternate. If other vowels cooccur with a non-alternating /o/ in a single stem these vowels are [+ATR] and the stem takes [+ATR] prefixes or suffixes. (33)

ebok "mortar" (class 7) m e n d o n g a "pall" (class 4) lomEn+1 — > lumini "to cause to send somewhere"

In Just one word, become [+ATR]: (34)

OmOtE

ibo "nine" (class 7) imboli "loans" (class 4) tolin+i — > tolini "to drown"

a dominant prefix containing /o/ causes the stem

"one, for class 1 nouns"

omoti

vowels

to

"one, for class 3 nouns"

In roots containing only /o/'s, this vowel is clearly represented by two types, as can b e seen from (33), i.e. /o/'s that behave as recessive vowels, and non-alternating /o/'s that behave as dominant vowels.10 (35)

kol

"to go and buy protective medicine" "idem, with applicative suffix" "idem, with causative suffix"

kolEn kuli

kol

"to create"

kolln

"idem, with applicative suffix" "idem, with causative suffix"

koli

3.2. Analysis The vowel system of Tunen is represented as follows: (36)

A 1

I i

u

A

I i

A

A

| i

A—[A]

tier

I u—[i]/[u]

tier

-[a] tier I

I

/E/

/a/

/0/

/i/

/e/

/A/

/o/

/u/

The distribution of features in this diagram shows an interdependence w h i c h we can express in the following redundancy rule (RR): (37)

"[A] [ ]

->

[a] / X

This rule says that the occurrence of [a] is predictable, if there is no floating instance of [A] which contains the relevant skeletal point within its scope. As stated in section 2, rule (37) will apply whenever its structural description is met. It is essential, as will be shown below in sect. 3.2.4., that an occurrence of [A], which is associated to the skeletal point, will not prevent the RR from applying.

116 In following sections, we will discuss the data presented In section 3.1. subsection here corresponds to the appropriate subsection of 3.1.

Each

3.2.1. The alternation /a/ - /A/ Roots or affixes may contain an instances of a floating [A], w h i c h association conventions is associated w i t h vowels from left to right: (38)

a-f

X

'

A

[A] tier

ì

[i]/[u] tier

a'-'

l

X

I

b

by

the

[a] tier +

i

-

Observe that in this case the RR in (37) Instance of [A].

>

fwAlAb+1 (cf. 21)

Is

not

met.

There

is

a

floating

In the corresponding cases where no morpheme possessing [A] is involved, no spreading takes place and the underlying representations are directly phonetically interpreted.

3.2.2. The alternation /0/ - /o/ This cases is no different from the preceding one. affixes showing this alternation. It occurs only In dominant affix: (39)

There are, however, no roots combined w i t h a tier

— \-rrr ri—u

I

X

"1--1

[i]/[u] tier [a] tier

a

+

m

I

X

t

X

->

o+moti (cf. 24)

The alternations discussed so far are straightforward a n d involve nothing but the feature [A]. In the following sections, we deal w i t h cases where the alternation involves a height difference, alongside or Instead of a difference In tongue root position.

3.2.3. The alternation /E/ - /i/ The root in (39) alternates w i t h [motE]. The /E/ - /1/ alternation is also found in several suffixes and one prefix. In this case the vowel is left unspecified for the feature [a], and specified w i t h [i] only.

117

If this vowel occurs in an environment where there is no floating instance of [A], RR (37) will apply and insert [a]. This happens in the lefthand example. If, however, there is a floating Instance of [A] no [a] is inserted, as in the rlghthand example: (40)

-A il

i

[A] tier [u]/[i] tier [a] tier

h ->

X + hE+bOb

h ->

X + hi+bil

b

X 1 (cf. 26)

The circled [a] is Inserted before spreading has taken place. Presumably this ordering is universal. W e will assume that RR's, being everywhere rules, always apply before other processes.

3.2.4. The /e/ - /i/ alternation In this type no alternation with respect to tongue root position takes place. In the framework assumed, the feature [A] is specified underlyingly in this case, since it it not Involved in an alternation. Since there is a height alternation, as in the preceding case, we underspecify the vowel for [a]: In (41) we represent the two relevant cases: (41)

A

[T|-

i

ill

-[A] tier -[i]/[u] tier -[a] tier

X - >

+ b e+bak

X ->

+ 1 i+lik

X k (cf. 27)

The b o x e d [A] in the right-hand example will disappear as a result of the OCP; cf. section 2. In this case, RR (37) will be checked, and found not to be applicable. In the left-hand example, on the other hand, RR (37) is m e t due to the fact that there is no FLOATING Instance of [A]. An [a] is therefore Inserted here. This example shows the crucial importance of the way in w h i c h we have formulated RR (37).

3.2.5. The /o/ - /u/ alternation and a transparency effect Unlike the /e/ - /i/ alternation, which only occurs in prefixes, the present alternation occurs in all morphemic classes. The analysis is parallel to the /e/ - /i/ case. The relevant vowels is left unspecified for [a], and specified for [u] only.

118 (42)

— [Tj

-_-A—[A] tier i—[i]/[u] tier -[a] tier

k X -> kolah

1

k X 1 X h -> kulAh+i (cf. 28)

The alternations in this and the preceding section involve segments which are underlyingly associated. In Van der Hulst and Smith (1986a) it is shown that this representation is the right one for so called transparent vowels. The term "transparent", as is pointed out, is appropriate to the extent that underlyingly associated features disappear in the context of an adjacent floating occurrence of the same feature due to the OCP. This is shown in the following example: (43)

--(T}

-A -

•*"""* ~

[A] tier

I

i

[i]/[u] tier [a] tier

+

i

-> bwAkuni (cf. 30)

When occurring without an adjacent floating feature, underlyingly associated features stay put. They do not spread since the AC'S only apply to floating features. Vowels to their left and right surface with the default value. 3.2.6. The dominant /o/ In the preceding section we have seen that some instances of /o/ alternate with /u/. In section 3.2.2. we saw cases where /o/ alternated with /0/. In both cases the vowels occurred in a recessive morpheme, i.e. the choice of the alternant depends on the presence of a dominant morpheme. We represented the two vowels as in (44a) and (44b): (44)

a.

/0/ - /o/

b.

/o/ - /u/ A

u

I u

c.

/o/ A u

I I X a

X In (44c) we give a third with an invariable vowel height alternation, and alternation with respect

X

underlying source for /o/. In this case we are dealing which is specified as [a], because it does not show a being in a dominant morpheme, it does not show an to tongue root position.

In (45) we give a minimal pair which demonstrates the difference between case b and case c:

119 (45)

A

A

'

^

[A] tier Il-Vi

[i]/[u] tier

1 k X 1 -> kolEn - - 0 - -

+

X

n

k X -> kolin

A

Q

i

u'SSS

[ a ] tier 1 + X (cf. 35) -A i

n [A] tier [i]/[u] tier [a] tier

k ->

X kuli

1

k ->

X koli

1 + X (cf. 35)

The left-hand examples involve a root with a recessive /o/ which is intrinsically advanced, and unspecified for height. The right-hand examples Involve a root with a dominant /o/. 4. Conclusion We have seen that (extra) height differences can be handled by assuming that RR's fill In features left unspecified, which means that these e x t r a height changes and disharmony effects can be dealt w i t h without making use of diacritics or abstract vowels. This seems adequate in view of the fact that ATR harmony systems with additional height alternations survive the dlachronls changes and appear to be relatively stable. Notes 1. Harmony systems b a s e d on the feature [Advanced tongue root], [Back] and [Round] are quite common. Harmony can also be b a s e d on nasality, and, less commonly, on height, retroflexion, etc. We use traditional binary-valued features here, but will adopt another system shortly hereafter. 2. The research that we report here is part of a broader study on the synchronic consequences of vowel shifts on harmony systems. This paper is a fusion of two presentations offered at the 17th Annual Meeting of the Dutch Linguistic Society.

3. One and the same language may show the result of different strategies such that one occurs in affixes and another in roots, while different startegies may even be followed as between prefixes and suffixes as in Avatime (Ford 1973). The alternations mentioned in (7) are present in the following african languages: a - E in Ika, a - 0 in Turkana, i - e in Tunen, u - o in Tunen, a - e in Leleml, a - o in Masaal, i - E in Leleml, u - 0 in Lelemi.

120

4. Changed vowels continue to behave as if they have not changed. The disharmony remains confined to heteromorphemic disharmony if the merged vowel occurs in a monosyllabic root, or if it coocurs with other merged vowels in a polysyllabic root. One might wonder why disharmony should result from merger? It is a fact that if vowels change class, affixes attached to the roots in which they occur frequently fail to harmonize with the merged vowel. For example, if /I/ goes to /i/ in a monosyllabic root affixes will fail to show up advanced. Similarly, if /A/ changes to /E/ affixes may still show up as unadvanced. In this type of case then the disharmony arises across morpheme boundaries only. If the marked vowel occurs with tautomorphemic and unmarked vowels, and changes class, a situation of disharmony Inside the morpheme will arise , in addition to heteromorphemic disharmony if the merged vowel occurs in a peripheral syllable. Again it appears to be a fact that if one of the root vowels changes class, other vowels usually remain as they were, i.e. no harmonic adjustment is made. 5. A remark in order. Neutrality may occur even if the harmonic counterpart still exists. We know cases in which harmony systems are "obscured" by the presence of vowels which, although they do not have a predictable value for the harmonic feature, still fall to harmonize, either in particular morphemes, or everywhere, and again such segments may behave as transparent or opaque, Vowels of this type then are neutral without there being a neutralization of an opposition. For want of a better term we will refer to such vowels as pseudo-neutral segments in those passages where we explicitly want to refer to them. It seems reasonable to assume that pseudo-neutrality either results if a merger is taking place, but has not yet effected all occurrences of the relevant vowel, or when the lost vowel reappears marginally in loans. 6. In addition to this Vago argues that the treatment of opaque segments calls for tier duplication.

also

7. As has been pointed out by Lleber, Tier Duplication, has not only been proposed in the study of vowel harmony. From work in the area of non-concatenatlve or associative morphology, we know that tiers can be multlplicated if they correspond to different morphemes. Clearly then, our dismissal of tier duplication in phonology is meaningless, If we cannot also show that TD in morphology can be dispensed with. We believe that this is possible, and refer to Van der Hulst and Smith (1986c) for an exposition of our view. 8. A comparable assymmetry as that observed In ATR systems may exist in front-back harmony, i.e. neutral segments may be [-BACK] or [+BACK]. The data available to us suggests that In segments of this type front neutral segments may be either transparent or opaque, whereas back neutral segments may only be opaque. 9. It is clear that conventions which repair OCP violations should be stated explicitly. HS propose that floating autosegments absorb bound autosegments rather than vice versa ("packman convention"). 10. Due to this double nature of /o/'s there is some confusion in the as can be seen in (33) and in the following examples. lobon sokom

"to weed" "to work in vain"

system,

lobon+En — > lobonEn, but: lobon+i — > lobuni sokom+En — > sokomin, but: sokom+1 — > sukuml

121 References ARCHANGELI, D. 1985 Yokuts harmony: evidence for phonology. LI 16, p. 335-372

coplanar

representation

in

Non-linear

BENNIS, H and F. BEUKEMA (eds.) 1985 Linguistics in the Netherlands 1985. Foris, Dordrecht BOOY, G.E. 1984 Neutral vowels a n d the autosegmental analysis of Hungarian vowel harmony. Linguistics 22, p. 629-664 DUGAST, I. 1967 Lexique de la langue tunen. Langues et Litt'eratures de l'Afrique noire 2, Klincksieck, Paris 1971 Grammaire du Tunen. Langues et Litt'eratures de l'Afrique noire 7, Klincksieck, Paris EWEN C. and H.G. v a n der HULST 1985 Single-valued features a n d the non-linear analysis of vowel harmony, in Bennls a n d Beukema (1985), p. 39-48 FORD, K.C. 1973 O n the loss of cross-height vowel harmony, Research 50-80, Institute of African Studies, Legon

Review suppl.4,

p.

GUTHRIE, M. 1971 Comparative Bantu 2, Gregg Int. Publ, Westmead HALLE, M and J.R. VERGNAUD 1981 Harmony processes, in W. Klein and W. Levelt (eds.)Crosslng the boundaries in linguistics. Reldel, p. 1-22 HULST, H.G. v a n der a n d N. SMITH 1982 Prosodie domains and opaque segments In autosegmental theory, in H.G. van der Hulst and N. Smith (eds.) The structure of phonological representations 2, Foris, p. 311-464 1985 Vowel harmony in DJinglll, Nyangumarda and Warlpirl, in The Phonology Yearbook2 1986a O n Neutral Vowels, in K. Bogers, H. van der Hulst and M. Mous (eds.).The phonological representation of suprasegmentals, Foris, p. 233-279 1986b O n some non-African ATR-based harmony systems ms. 1986c Against tier duplication ms. KAYE, J., J. LOWENSTAMM a n d J.R. VERGNAUD 1985 The internal structure of phonological elements : A theory of charm and government.Phonology Yearbook (1985) LIEBER.R. 1985 A n Integrated theory of autosegmental processes, ms. MOUS, M. 1983 Vowel harmony in N e n (Bantu A.44), ms.

12?.

MOUS, M. 1986 Vowel Harmony In Tunen, In K. Bogers, H. van der Hulst and M. Mous (eds.) The phonological representation of suprasegmentals, Forls, 281-295 RENNISON, J. 1985

Trldlrectlonal vowel features and vowel harmony, ms.

STEWART, J.M. 1971 Niger-Congo, Kwa, In Current trends In linguistics 7: Sub-Saharan Africa, Mouton, p. 179-212 SVANTESSON, J.O. 1986

Linguistics In

Vowel harmony shift In Mongolian, In Lingua 67, 283-327

VAGO, R. 1984

Morpheme level harmony In a multi-leveled autosegmental framework, ms.

WILLIAMSON, K. 1973 Some reduced vowel harmony systems. In Research Notes, p. 145-169 1984 Vowel merger In harmony languages, In JOLAN 2, p. 61-82

Emmy Jacobs

Polarity-sensitivity and generalized quantifiers

0. Introduction Most words and iaioms occur in both affirmative and negative sentences. However, there are elements which might be termed polarity-sensitive, in that they seem to occur only in affirmative sentences (e.g. wel degelijk. 'indeed', inderdaad 'really') or in negative sentences (hoeven 'need', ook maar 'even'). But a negative element solely cannot account for the difference in grammaticality between sentences (1) and (2). (1) (2)

*Geen 'Not Niet 'Not

mens heeft de UFO Wel degelijk gezien anybody has really seen the UFO alleen mannen hebben de UFO wel degelijk only m e n have really seen the UFO

gezien

Zwarts (1981) has argued that the distribution of negative polarity items is dependent on the semantic properties of quantifiers. In this paper I intend to show that the semantic properties of quantifiers play an important role in the distribution of positive polarity items (PPI's) too, but that these properties are not sufficient to account for differences in grammaticality between sentences containing PPI's. First I shall give a brief sketch of the nature of generalized quantifiers. Then I will show that PEI's in sentences with NP's that have specific semantic properties are grammatical, whereas if the wP's have other properties the sentences are ungrammatical or at least marginal. In the main part of this paper I will argue that the distribution of PPI's is dependent on semantic properties of quantifiers and the position these quantifiers occupy in the structure.

1. A sketch of the nature of generalized quantifiers 1.1. NP's as generalized quantifiers Universal and existential quantifiers can be analysed as second-order predicates. Where first-order logic allows quantification over objects in E (where E is an arbitrary non-empty set of things: individuals, entities, etc.), second-order logic allows quantification over sets of things or functions from things to things. Barwise & Cooper (1981) related generalized quantifiers to natural language expressions. NP's like most rabb'ts, all spiders and no fishes will be analysed as generalized quantifiers. They denote collections of sets. Determiners like most, all and no combine with a set expression to produce a quantifier. The structure of the quantifier is then semantically equivalent to the structure of an NP. NP

quantifier det.

set expr.

124

Quantifiers are used to assert that a set has a property. 3x(x) asserts that the set of things which satisfy iji(x) is not empty. That is, the set of individuals having property f Board (=Y) to Juice (=X) in (5a), resulting in the ungrammatical contour 2-3-4-1. (No theoretical status should be attributed to the numbers in the examples; they are only a shorthand notation for prominence rank.) Comparable cases lead Hayes to the observation that right branches generally cannot be analyzed as X. The Maximality Principle correctly excludes RA in (5a), as X and Y are not sister nodes, nor is Y a sister of D T E and c-commanded by X. In (5b) the unconstrained v e r s i o n of R A would, incorrectly, permit th£ choice of law as X and 11 a s Y , leading to the ungrammatical output contour 2-4-3-1. Again, M a x i m a l i t y correctly blocks RA, as X and Y are not sisters, nor is Y a sister of DTE.

143

2

2 X

3

3 Y

3

4

1

2

1 DTE

3

2

4

4

3

1

1

Turning our attention to Dutch, Maximality appears to be v i o l a t e d consistently in analogous examples. In (6a) and (6b), analogues of (5a) a n d (5b) respectively, adjustment applies happily as indicated by the contours 2-3-4-1 and 2-4-3-1. (6) (a)

->

Mendelsohnstraat 2 3 4

tien 1

(b)

vakopleidingspersbericht > vakopleidingspersbericht 2 3 4 1 2 4 3 1 X Y DTE We are now confronted w i t h a n interesting puzzle, seemingly caused by the m o r e extensive possibilities of RA in Dutch, as compared to English. To be m o r e precise, Dutch differs from English in allowing rightward as w e l l as leftward RA. For this reason it is as yet unclear w h e t h e r the examples in (6) are derived b y leftward RA, rightward RA or a n interaction of both. First, let u s consider some simple cases of rightward RA. It w i l l be clear that for these the formulation of RA in (1) has to be adapted into its m i r r o r image: ...DTE...Y X..., where Y adjoins to X. In (7a) w e illustrate rightward shift, in (7b) rightward strengthening.

144 (7) (a) sí ^ h ) DTE bloedarmoede dwangarbied bloedlichaam woordvolgorde noodtoestand

(b}

s- " ^ ^ ( w ) DTE Y X 1 anaemia' waterpomptang 'forced labour' golfslagbad 'blood corpuscle' onderzoeksmethode 'word order' wachtgeldpremie 'state of emervoorhoofdsholte gency )/

X

»• s

/

0*Nî

'pincers' 'whirlpool' 'research method' 'halfpay premium' 'frontal sinus'

It is important to note here that the examples in (7) conform to the formulation of the Maximality Principle (4). In (7a) X and Y are sisters and in (7b) Y is the sister of DTE and c-commanded by X. In any case, the formulation of (4) seems to be an improvement upon Hayes' original Right Branch Constraint, Intended to handle cases like (5a) where RA would be excluded if X were a right branch. Maximality allows that X be a right branch and therefore can be considered as relevant as well for rightward RA. It is not clear to us whether Hayes had considered this possibility when formulating (4). The existence of rightward RA in Dutch offers us a unique chance to gain a deeper understanding of rhythmic phenomena by considering cases with more than one application of RA. (8) gives an example of double application of shift to the left, (9) an example of double application of shift to the right. The same could be demonstrated for double application of strengthening to the left as well as to the right. And examples could be given of interaction of leftward shift and rightward strengthening, etc. Lack of space, however, forces us to leave it as an interesting exercise for the reader to work out.

3 minister j|upra jkjfc'prime 'supra

3 2 president Lubbers segmenteel kenmerk minister Lubbers segmental feature'

s w w 2 3 4 minister president Lubbers^ supra segmenteel kenmerk

(9) s

i

Y Y

3 2 3 hoofdafvoerkanaal 'main drain' hoofdafvoerkanaal fietsuitleensysteem 'bicycle lending system' vaatafwasmachine 'dishwasher' More interesting are those cases where specific types of interactions cannot occur. Before we go into these, however, we will have to pay attention to a second constraining principle, which will have severe consequences, especially for our analysis of these interactions. This restriction is applicable in cases of shift in embedded structures, i.e. the encircled structures in (10a) and (10b).

145 (10)

(a)

X Y DTE 3 2 3 1 Piet Heinstraat tien klaverjas club Utrecht atletlekclub Heracles

2 3 3 1 Plet Heinstraat tien (address) 'card club Utrecht' 'athletics club Heracles'

(b)

3 3 2 minister president Lubbers inter nationale druk

minister president

Lubbers

Note that RA is not allowed despite M a x i m a l i t y , while RA can be a p p l i e d embedded structures such as those in (10c), w h i c h are w e l l - k n o w n cases 'internal rhythm'.

opera componist Verdi vice president Bush

in of

o p e r a componist Verdi 'opera composer Verdi'

Comparing (lOa/b) and (10c), we see that R A is b l o c k e d in structures/ constituents dominated by a strong node. This blocking seems to be related to a condition on metrical defooting rules that w a s proposed in Hayes (1981): (11) 'No foot in strong metrical position may be deleted' (Hayes,

1981:178)

The effect of this condition is that the strong or only syllable in a strong foot cannot be extracted to be adjoined elsewhere in the tree (and consequently be reduced). Both RA and defooting rules belong to the class of rules manipulating tree structure, or 'prosodic transformations'. We w i l l propose a condition on this type of rules that will account both for the b l o c k i n g of R A in (lOa/b) and for the effects of the 'strong foot principle' (11). In this w a y , (11) becomes a subcase of the more general condition (12), w h i c h we w i l l call the Strong D o m a i n Principle: (12) Strong D o m a i n Principle

(SDP)

No prosodic transformation may apply to the head of a strong domain. The head of a constituent is the strong or only element of that constituent. SDP correctly b l o c k s the application of RA in (lOa/b) as w e l l as the extraction of the head of strong feet. As far as we know, isolated instances of (10a) and

146 (10b) have been noted in the literature, but other or to the strong foot principle (11).

left

unrelated

either

to

each

Our next examples illustrate that SDP is also relevant to right directional shifts. The structures (13a) and (13b) are the mirror images of (10a) and (10b) resp. Again, RA is not allowed to apply, and this is explained by SDP. (13) (a)

/

/ w

W] Is Y X 1 3 2 3 hoofdrijkswegennet rij kskinde rbij slag stadsschooladviesdienst

s DTE

*

1

3

3

2

hoofdrijkswegennet 'national highway 'children's allowance' system' 'city school advisory board'

(b)

DTE Y X 1 2 3 3 hoofdafvoerpijp topvoetbalclub rijksinvoerstop

, 1 3 2 3 hoofdafvoerpijp 'top soccer club' 'national import prohibition'

We are now able to return to our discussion of relevant types of interaction. The SDP allows us to determine the order of two rhythmic adjustments as in the examples of (8) and (9) above, repeated here as (14a) and (14b). (14)(a)

minister president Lubbers

/£*b

minister president L.

minister president L.

s s' w QO DTE Y X hoofdafvoerkanaal hoofdafvoerkanaal 1 2 4 3 hoofdafvoerkanaal First, RA applies to the highest pair of nodes and then, internally, to the lowest pair. Internal adjustment first would violate the SDP, therefore, these derivations avoid it by rhythmic adjustment on a higher pair of nodes. We find a comparable situation for the strong feet principle (11) which can be avoided by a word-internal application of adjustment before the application of Defooting as is shown in (15).

147 (15)

or

W V

w

—HsJwAs

X Y DTE explanation

RA

ir

>A k

ir «É "w DEF. "s

explanation

^

^

explanation

Until n o w we have only discussed double applications of RA in one direction. As w e turn to the interaction between left a n d right directional adjustment, we find quite surprising facts. We have seen that the derivations in (14) could be explained b y a n application of RA to a higher pair of nodes before an application to a lower pair. In the examples (16), however, this leads to completely unacceptable results. (16) (a)

DTE Y X 1 4 2 3 kleurenteeveetoestei 'colour TV set'

kleurenteeveetoestel

kleurenteeveetoestel

Nieuw Zeelandse tijd Nieuw Zeelandse tijd Nieuw Zeelandse tijd 'New Zealand time' The derivations in (16) have to be blocked after the first adjunction of the highest pair of n o d e s , and the question arises immediately w h y this should be so. In this respect, we may observe a n important difference b e t w e e n the derivations in (16) and those in (14). In (16), the D T E required for the second adjunction is derived by a previous adjunction. This m e a n s that the domain of the second adjunction w h i c h contains the nodes X , Y and DTE,is a subdomaln (i.e. 'teeveetoestel') of the domain w h i c h contains the nodes X, Y a n d D T E of the previous adjunction (i.e. 'kleurenteeveetoestel'). Those cases in w h i c h a second adjunction Is possible, on the other hand, as in (14), make use of the same D T E for b o t h adjunctions and are thus applied in the same domain. Apparently, adjunctions apply from bottom to top in the tree, e a c h time in a larger domain. W i t h i n one domain, however, we m a y apply several adjunctions, also from top to bottom. Let u s state this in the following principle of rule application (17). (17) Bottom-to-Top Principle (=BTP) 1 A m e t r i c a l adjustment rule m a y not be applied to a subdomain of a domain in w h i c h a metrical adjustment rule has applied. We can n o w explain the difference between (14) and (16). In (14) w e have two applications of RA w i t h i n one single domain. The BTP is irrelevant to the order of application in this case, being determined b y the SDP. In (16), however, after the application of RA in a wider domain, returning to a subdomain is excluded by the BTP.

148 At this point we are in the position to show that the violations of Maximality in Dutch are only so superficially. The derivation of 'Mendelsohnstraat tien' has two applications from bottom to top, respectively, right and left directional RA, as in (18).

Mendelsohnstraat tien

Mendelsohnstraat tien

Mendelsohnstraat tien

The Maximality Principle allows both applications. In the first step of the derivation X c-commands Y and in the next step, the nodes X and Y are sisters. English lacks the possibility of rightward RA and the output of (18) can only arise by a direct application of 'strengthening' to the left as in (5a), which implies a violation of Maximality. Although Dutcjj differs from English w i t h respect to the facts in (18) and (5a) (compare ^pple Juice Board lobbyist), it resembles somewhat surprisingly - English with respect to another class of examples, which Hayes gives as an illustration of Maximality (see 19). (19) (a) *2 3 4 1 overdone steak blues (b) *2 3 4 1 Amsterdam Oost Bijlmerlijn 'East Amsterdam-Bijlmer railroad' (c) * 2 3 4 1 Waterlooplein een 'address' (d) * 2 3 4 1 Watergraafsmeer polderschap 'Watergraafsmeer Polder Board' These examples are comparable to the examples in (18) with regard to branching, but differ in sw-labelling. In Dutch (19b,c,d) with the same contour 2-3-4-1 as (19a) are also excluded. The BTP explains these facts without hesitation. In (20) the derivation proceeds correctly to the input structure of (18) (=the third step in the derivation). (20)

X Y DTE Amsterdam Oost Bijlmerlijn

Amsterdam Oost Bijlmerlijn

/ w @ ^ s BTP DTE ^ Y X Amsterdam Oost Bijlmerlijn

?

sAr

(wAs) ^ s fXX YY DDTTEE ^ Amsterdam Oost Bijlmerlijn

149

X

5w

S

*.Amsterdam

'2

Oost Bijlmerlijn 3 4 1

Wa ter looplein een However, if we want to go on as in (18), we will have to return to a subdomain in (20) and this is blocked by BTP. The only remaining possible application of RA would be 'strengthening' to the left - a direct conversion of the third step into the fifth step - but this would mean a violation of Maximality. Therefore, we cannot proceed beyond the third step in the derivation, just as in English. The formulation of the BTP allows us to draw exactly the right distinctions between structures identical with respect to constituency but different with respect to labelling. The Maximality Principle of Hayes is supported also by Dutch facts, although the effects are obscured at first sight by the possibility of bi-directional adjustment. This situation became clear only after formulating two additional principles needed to constrain adjunction. The first was the Strong Domain Principle which states that the head of a strong constituent is immune to adjustment. This principle seems to be justified by Dutch as well as English facts. The second was the Bottom-to-Top Principle which constrains adjustment rules in their orders of application. Notes *

We thank Wim Zonneveld for helping us in various ways. 1. By mentioning the notion 'domain' in (17) we assume that there is more than one DTE. Every constituent forms a domain in which one element is the strongest. This is the DTE of the domain under consideration.

2. It seems to us probable that our BTP is in fact an instance of 'strict cyclicity', but developing this idea would bring us to the highly uncertain topic of the cycle above word level. We leave this for future research. References HALLE, M. and J.R.VERNAUD forthcoming Grids and Trees: An Essay on Stress HAYES, B. 1981 A Metrical Theory of Stress Rules. IULC HAYES, B. 1984 'The Phonology of Rhythm in English', in LI 15, pp. 33-74 KAGER, R. and E. VISCH 1985 On Trees and Grids. Interne publlcatie Inst. De Vooys, RU Utrecht, nr. 85-NE-01 AFd. Mod. Taalk. PRINCE, A.S. 1983 'Relating to the grid', in LI 14, pp. 19-100 SELKIRK, E.O. 1984 Phonology and Syntax. The Relation between Sound and Structure. MIT Press, Cambridge Mass.

Simone

Langeweg

"Roots", "words" and the stress behaviour of non-native words in Dutch

0. Introduction In this paper I will deal w i t h the stress pattern of words derived b y nonnative prefixation, e.g. synchroon 'asynchronous', lnclusief 'inclusive' and root compounds, e.g. telefoon 'telephone', aristocraat 'aristocrat'. This research has b e e n done as part of a project on automatic stress assignment in Dutch, which, in turn, is part of a project on text-to-speech conversion. The question that I started from was whether these words, as the non-native sufflxatlons (cf. Langeweg (1985)), can be treated like monomorphemes, or whether morphological structure plays a role for metrical structure assignment. Following BooiJ (1977), who proposes that all prefixes of Dutch are followed by a strong morpheme boundary, one w o u l d expect the latter to be the case. In section 1, I will discuss this proposal in more detail a n d show that it leads to problems w h e n considering the stress behaviour of the non-native prefixatlons. Next I will present an analysis of the different types of stress patterns that were f o u n d for the prefixed words. I will follow Selkirk (1982) in showing that the stress behaviour of the prefixes correlates w i t h their subcategorisation frame, i.e. what morphological category they attach to. This analysis will furthermore be applied to words that are composed of b o u n d morphemes from the learned vocabulary (root compounds). a n d which are sometimes difficult to distinguish from the non-native preflxations. Finally, I will show how the proposed analysis can b e applied in a system for automatic stress assignment if a certain approach to text-to-speech conversion is taken.

1. The Class I/Class II approach Following Siegel (1974), BooiJ (1977) classifies all affixes of D u t c h as belonging to one of two classes. The main criterion for classification is syllabification and, in cases in which syllabification cannot decide, stress assignment. Thus, suffixes like - l t e l t a n d -lsme were g r o u p e d together as Class I, i.e. they are preceded b y a weak morpheme boundary, since syllabification rules operate across their morpheme boundary (i-den-tl-telt 'identity'). Furthermore, in the case of these suffixes we find Instances of the deletion of a stem-final vowel before a vowel-initial suffix (AmerlkjU-aan 'American'). These derived words, then, can be treated like monomorphemes since they share a number of phonotactic properties. W h e n we now consider the non-native prefixes in this respect, Booij notes that they must be followed by a strong morpheme boundary (they are Class II affixes). Syllabification rules and homorganic vowel deletion appear to be b l o c k e d b y the morpheme boundary.

152 (1)

a.symbool+isme ->sym-bo-lis-mg vs. tlyper+emie ->hy-pe-re-mie b.Amerika+aan ->Amerikaan vs. co+operatie - > cooperatie

In those cases in which we do find resyllabification (e.g. abortus 'abortion'), BooiJ (1977:103) claims that the strong prefix or w o r d boundary 'has b e e n w e a k e n e d to a morpheme boundary due to a loss of semantic transparency of the complex word'. The implication of classifying all prefixes as w o r d prefixes is that they have to b e lexically m a r k e d for their behaviour with respect to stress placement. Since they are strong affixes, we assume that they are themselves subject to the Main Stress Rule (MSR). The Compound Stress Rule (CSR) then decides the position of the primary stress, i.e. main stress assignment depends on syntactic category information ( [ A B ] ^ [ s w ] , [AB]/p[ws]). However, this does not seem to b e the correct generalisation for the non-native suffixes: they are not necessarily main stressed (cf. reageer 'react'). BooiJ (1977) therefore marks the prefixes as either always or never receiving stress. In m y opinion this classification in terms of the stress behaviour of the prefixes is incorrect. I f o u n d b o t h native and non-native prefixes to receive main stress either never or sometimes. I will return to the latter group of prefixes below. As for the native prefixes, I wish to remark that the stress pattern of the sometimescases appears to be in accordance with the observation made above, that if prefixes are strong affixes, it is the CSR that accounts for the stress placement; cf. the noun ontrouw 'unfaithfulness' vs. the adjective ontrouw 'unfaithfull'. Apparently then the proposal made in BooiJ (1977) does not make the correct predictions for the stress behaviour of prefixes. The arguments used by BooiJ for his classification of the affixes support the view that all prefixes are phonological words. This is contradicted however by the stress p a t t e r n in this type of complex words. A similar kind of observation can be made w i t h respect to the suffixes and their classification. In BooiJ (1977) a number of vowelinitial suffixes were classified as Class I — b e c a u s e they attract the stemfinal consonant across the morpheme boundary (wandel+aar ->wan-de-laar)— whereas their stress behaviour — t h e y are stress n e u t r a l — w o u l d classify them as Class II. Our conclusion could be then that syllabification is not the correct criterion for affix classification.

2. Prefixations In this section, I will discuss the stress patterns that I f o u n d for the nonnative prefixations. The theoretical framework within w h i c h they will be analysed is that of Selkirk (1982). Thus, stress behaviour will b e dealt with in structural terms, i.e. in terms of the place the affixes occupy in w o r d structure. Affixes may attach eiter at root or at word level. Note that 'the status of an item as a root does not imply that it is b o u n d (that is, not unassociated with a sister in some word structure)' (Selkirk 1982:98). If an affix attaches to a root, this structure itself forms a root. Every w o r d then is redundantly a root (Word->Root). The root is the domain w i t h i n w h i c h the MSR applies. An item of the category w o r d has already been assigned w o r d stress then, i.e. it is a phonological word. The following rewrite rules account for root and word prefixes respectively:

153 (2)

a. Root -> Prefix Root b. Word -> Root c. Word -> Prefix Root

Let us now consider the non-native prefixations and the stress patterns that we find. It will be clear that the stress contours that are possible may be of two types. Stress may fall either on the prefix or on the base. This will be accounted for in three ways, each of which corresponds to a subsection below. 2.1. The monomorphemlc type We often see that the prefix does not receive main stress, independent of whether it attaches to a base that does or does not occur as a lexical item. Some examples are given in (3). (3)

aabadananteantiapocatacondeininterperiprere-

ablose 'abiosls' abdlcatie 'abdication' adhesie 'adhesion' anarchfe 'anarchy' antecedent 'antecedent' antipathie 'antipathy' apotheek 'apothecary' catacombe 'catacomb' concentreer 'concentrate' deficient 'deficient' inclusief 'inclusive' interessant 'interesting' perifeer 'peripheral' prelude 'prelude' reageer 'react'

amoralfsme 'amoralism' abnormáál 'abnormal' antedateer 'antedate' catastrofe 'catastrophe' confederatile 'confederation' deblokkéér 'unblock' improductief 'unproductive' interactie 'interaction' peri-optica 'peri-optics' prefabricâtie 'préfabrication reanimatie 'reanimation'

In these examples stress can be assigned by the rules that account for stress placement in monomorphemlc words. Within the framework described above, this Implies that the prefixation as a whole must be of the type root, which, as was mentioned, is the domain of the MSR. This can only be the case if the prefix attaches to an item that is also of the category root. It does not make any difference whether the root to which the prefix attaches is bound, e.g. reageer or unbound, e.g. abnormaal. (4)

a.

Word

b.

Word Root

Pfx

Pfx'

I

ab

nor maal w

154 When the prefix attaches to a root, the whole structure is itself a root and therefore subject to the MSR. Stress assignment in these words then occurs as in monomorphemic words. The prefix will be assigned secondary stress because of Its maximal distance from the main stressed syllable. Sometimes, however, the prefix of a prefix-root combination does receive main stress. This may only be the case if the MSR assigns stress to a syllable that happens to be (part of) the prefix. For instance, in ¿xodus (5), stress is assigned to the antepenultimate syllable, w h i c h in this case is the prefix, because the intrinsically lesser weight of the penultimate VV-syllable prohibits its being labelled strong when it is followed by the intrinsically heavier VC-syllable In a weak position (cf. van der Hulst ft Langeweg (1984)). The same holds for catalogus.

/\

(5)

xx

M

s ^ w ex o

M

|W dus

f W s A , f c a t a lo g u s

2.2. The real compound-like type There are more cases like exodus. in which the prefix part is stressed. They are not of the same type, however. In exodus, we saw that the prefix was stressed by the MSR. This is not so for cases like antlheld 'anti-hero', where the prefix Is stressed by another rule. Consider the words in (6). (6)

anticoconcontradesinternonperpreprosemisubsuper-

antlstof 'antibody' có-auteur 'co-author' conrector 'vice principal' cóntra-gewicht 'counterweight' deslnteresse 'disinterest' interfaculteit 'interfaculty' nón-betallng 'non payment' péroxyde 'peroxide' prehistorie 'prehistory' pro-vitamine 'provitamin' sémi-finale 'semi-finals' sübfaculteit 'subfacuity' Supermarkt 'supermarket'

antlheld 'anti-hero' có-partnerschap 'co-partnership'

deslllusle 'disillusion' intercellulose 'intercellulosis' pér-verbinding 'per-compound' prémedicatie 'premedication' pro-secretaris 'prosecretary' semi-bungalow 'semi-bungalow' subcomlte 'subcommittee' superstructuur 'superstructure'

Note that the non-native prefix may also attach to a base form that is of native origin, as in antlstof, supermarkt. This base form must then have phonological w o r d status; it is itself a lexical Item. Since the prefix attaches at w o r d level, stress rules (MSR) have already applied. The prefix itself, therefore, must also be assigned metrical structure. Thus, the w o r d consists of two phonological words, triggering the CSR to assign m a i n stress. Since the CSR Is sensitive to syntactic category information, w e expect to find different stress patterns in complex nouns and adjectives. The words in (6) are all nouns, so that the CSR assigns main stress to the first morpheme, i.e. the prefix part. The adjectives in (7), though of the same morphological structure, have main stress on the last morpheme when used predicatively and on the prefix when used attributively.

155 (7)

interparapostsemi-

intercontinentâal 'intercontinental' paramédisch 'paramedic' postkoloniaal 'postkolonial* semi-direct 'semi-direct'

These prefixations, then, behave like real compounds, such as voordeur 'front door', llchtgréén 'light green'. Instead of two morphological words, w h i c h are b o t h phonological words, we have to do with a prefixation at w o r d level, also resulting in two phonological words. In (8), the structures for prefix-word combinations are g i v e n for nouns and adjectives respectively. (8)

Word

b.

Word ord Pfx pa ra s w

V

I L

Root

h me disch w

F

2.3. The phrasal compound-like type This leaves us w i t h some words that, in m y opinion, have the same morphological structure as words of the type antlstof, i.e. we have to do w i t h prefix-word combinations. However, the CSR does not account for their stress pattern; they have m a i n stress on the base. Among these words are a large number that have the prefix antl-. Note that the difference in stress placement involves a clear difference in meaning. If we compare antlstof and antlstof 'anti-dust', the former concerns a body that counteracts something, whereas the latter implies something that prevents dust. We also find these structures as part of a larger compound like antl-raket raket 'anti-missile missile' or postconcentratiekampsyndroom 'postconcentrationcamp syndrome'. I w o u l d say that these more

156 complex compounds are of the same type as wlldedferentemmer 'tamer of w i l d animals', onderwaterflora 'underwater flora' a n d rodekrufsverpleegster 'red cross nurse'. These are [[AB]C] compounds, In w h i c h the [AB] part is not labelled according to the CSR since it is of phrasal origin a n d therefore labelled b y the Nuclear Stress Rule (NSR). (9)

Root

Root

rode

kruls

Root ver pleeg ster

Pfx

Root

Root

anti

r a ket

r a ket

In cases such as antlstof vs. antlstof we have to do then w i t h words which have morphologically the same structure, b u t of which one is a real compound whereas the other is a compound of phrasal origin. Thus it appears that the stress behaviour of the non-native prefixes is decided by their subcategorisation, i.e. whether they attach at root or at w o r d level. There are two options. Either the prefix attaches to a root morpheme or the prefix attaches to a word. In the former case the structure is t r e a t e d as monomorphemlc because it is itself a root; in the latter case b o t h prefix and w o r d are a phonological w o r d since the MSR has already applied in the base word, so that a compound-like structure is the result.

3. Root compounds In this section I will deal w i t h stress placement in the so-called root compounds, since it is often difficult to distinguish them from the prefixations. It will be seen that the analysis proposed in section 2 accounts for the stress patterns in these words as well. In the literature, the distinction between non-native prefixations a n d root compounds has sometimes not b e e n made. Quasl-preflxal forms like neo-, clrcum-, electro-, etc. were grouped together with the real prefixes in-, con-, re-, etc. Similarly, forms like -loog, -craat, -flel, etc. could be termed suffixes. In this way, however, words like arlstocraat 'aristocrat', fonoloog 'phonologist' w o u l d have a prefix-suffix structure, thus lacking a stem, w h i c h w o u l d be a highly undesirable result. Arguments c a n also be provided against the affix-status of these b o u n d morphemes. Thus, some of the "suffixes" also occur in prefix-position or the other way round, e.g. telefoon 'telephone'- fonoloog - logopedle 'speech training'- pedagogle 'pedagogy'. A better solution w o u l d be then to consider these b o u n d morphemes roots, some of which can also b e used independently (e.g. foto 'photo', homo 'homo'). If, following Selkirk (1982), we postulate that the category root is a recursive category, the rule Root->Root Root accounts for the structure of the root compounds. A similar analysis for these words is proposed in Hayes (1982), who also provides a number of supporting arguments. Thus, in English, as well as in Dutch, we f i n d co-ordination reduction as in real compounds, cf. appel en kersentaart 'apple and cherry pie' vs. hydro- en aerostatlca 'hydro a n d aero-

157 statics'. BoolJ (1985) provides arguments for the phonological w o r d status of such deletable items. We w o u l d then expect the CSR to account for stress assignment in these root compounds since they consist of (at least) two phonological words. For English, this indeed holds; cf. homonym, spectrograph, telephone. Compare these words, however, with their Dutch equivalents, homonlem. spectrograaf. telefoon. In Dutch, the root compounds do not show the compound stress pattern, but monomorphemic stress. Some more examples are prov i d e d in (10). (10)

bioscóop 'cinema' protoplasma 'protoplasm' lsomórf 'isomorph'

geolóóg 'geologist' leukoplast 'leucoblast' hypofyse 'hypophys i s'

The stress pattern of these words can be explained from their structure. As in the case of the root prefixatlons, the root-root structure itself forms a root, w i t h i n w h i c h main stress assignment takes place in accordance w i t h the rules for monomorphemes, as is illustrated in (11). Word

(11)

Root Root bi bli o I s w

Iy

Root theek

However, the first part of the root compound sometimes receives m a i n stress. If we follow the analysis presented in section 2, our expectation w o u l d be that these are instances of root-word compounds. This expectation appears to be correct g i v e n the fact that the non-native root can also attach to native stems. (12)

amfitheater 'amphitheatre' hydrothérapie 'hydrotherapy' âudloslgnaal 'audiosignal'

bibliobus 'library van' cârdiochirurg 'cardiosurgeon' métataal 'metalanguage'

The root, when attached to a morpheme of the category word, has to b e assigned phonological w o r d status. The CSR then accounts for the stress pattern.

158

The stress behaviour of root compounds can be explained in terms of morphological structure. When we have to do with a real root compound, I.e. a structure that consists of two roots, the compound Is dominated by a root that itself is a word; stress assignment takes place in accordance with the MSR. In other cases the compound is more like a real compound (like voordeur 'front door') since the first morpheme, w h i c h is a root, attaches to a morpheme of the type word. The resulting structure consists of two phonological words, thus triggering the CSR.

4. Conclusion In this paper we have seen that affix classification b a s e d on syllabification does not make the correct predictions where stress assignment is concerned. Apparently, although they are rules that operate within the same domain, the behaviour of syllabification and stress assignment does not follow from each other. If one is chosen as criterion for classification, we w o u l d expect, given the arguments in Siegel (1974) and Selkirk (1982), the other to follow automatically. This, however, is not so. Maybe then w e should agree with Strauss (1982), who argues that stress behaviour is the only criterion for affix classification (cf. also KooiJ & van der Niet (1985)). Syllabification etc. must be explained in other terms. With respect only to stress, then, it was observed above that the stress patterns in the non-native part of the Dutch vocabulary may be reduced to two cases, i.e. monomorphemlc and compound-like. Which of the stress patterns is chosen is determined b y the morphological structure. The two types can be described by the following generalisations, where (for completeness) the nonnative suffixations have also been Included. (16)

a. W o r d Word Word Word b. W o r d Word Word

-> -> -> -> -> -> ->

Root Root -> Root Suffix Root -> Prefix Root Root -> Root Root Word Word Prefix W o r d Root Word

Note that non-native suffixes and prefixes apparently behave differently. Suffixes only attach at root level, whereas prefixes may be "word-like". Syllabification,

then,

will have to be explained from other

principles.

For

159 Instance, it may be the case that the prefix is followed b y a boundary, which, however, is not a phonological word boundary. This boundary is respected by rules of syllabification but ignored by rules of stress assignment. Thus the question whether morphological structure plays a role for metrical structure assignment or not can be answered as being dependent on the level of derivation. If morpheme attachment takes place at root level, morphological structure is ignored; if it takes place at word level, metrical structure is assigned in accordance with morphological structure.

5. Application One of the modules w i t h i n a text-to-speech conversion system for D u t c h must be the assignment of stress. For monomorphemic words, relatively simple rules can be written that refer to syllable weight. Suffixes can b e specified as being neutral or stress-determining. In the case of the prefixations and root compounds discussed above, w e have a problem. Sometimes the MSR and in other Instances the CSR may be triggered by a morpheme that in b o t h cases is segmentally the same, b u t in the one case has phonological w o r d status and in the other case not. How can this ambiguous behaviour b e accounted for? There are two w a y s to do morphological parsing in a text-to-speech system. The one is by means of rules that refer to spelling (e.g. w i t h i n a morpheme we do not find the consonant sequence obstruent-voiced obstruent) and affixes. It will be clear that in this way no difference can b e made b e t w e e n for instance 1 lnterferentle 'interference' and fnterfacultelt 'interfacuity . Secondly, a morpheme lexicon can be used supplemented by rules for morphological parsing. Two strategies can now be adopted. The difference in stress rule application is caused by a difference in level of attachment of the prefix, i.e. root or word. We could suggest that the morpheme lexicon consists of all lexical items plus all affixes. In that case, in lnterferentle the prefix inter- might be recognised, but the root is not in the lexicon, since it is a b o u n d morpheme. The w o r d cannot be parsed and is therefore treated as if it is monomorphemic. The other possibility Is to list all b o u n d morphemes in the lexicon alongside the lexical items and affixes. It will then have to be specified whether the morphemes are b o u n d or not. In that case the rules can refer to whether the morpheme following the prefix/root is bound, and thus necessarily a root, or not. In this way, however, i.e. by means of the morpheme lexicon approach, not all of the cases discussed above can be accounted for. For instance, those cases in which the prefix attaches to a root that may also be a word, and Is therefore in the lexicon (e.g. lnteractle 'interaction') will be treated like compounds. Clear instances of the root/word attachment, however, can be distinguished where stress placement is concerned, e.g. lnterferentle-fnterfacultelt. blbllotheek-bibllobus.

Notes "

The research reported on here is part of a project on automatic stress assignment in Dutch, supported by the Faculty of Letters, University of Leyden (Pfon 11.6). I thank Colin Ewen, Vincent v a n Heuven, Harry van der Hulst, Jan KooiJ and Sieb Nooteboom for their comments.

160 References BENNIS, H. a n d F. BEUKEMA, eds. 1985

Linguistics In the Netherlands 1985, Foris, Dordrecht.

B00IJ, G.E. 1977

D u t c h morphology. A study of w o r d formation In generative grammar, Forls, Dordrecht. 1985 Coordination reduction In complex words: a case for prosodlc phonology, In v a n der Hülst & Smith, eds., pp. 143-160. HAYES, B. 1982 Extrametrlcallty and English stress, In Linguistic Inquiry 13, pp.227276. v a n der HÜLST, H. 1984

Syllable structure and stress In Dutch. Forls, Dordrecht,

v a n der HÜLST, H. a n d S. LANGEWEG 1984

Nederlandse klemtoon, ongelede woorden, INL Working paper 4.

v a n der HÜLST, H. and N. SMITH, eds. 1985

Advances In nonlinear phonology, Forls, Dordrecht.

K00IJ, J. a n d M. v a n der NIET 1985 Stress, stress shift, a n d morphology: der Hülst LANGEWEG, S.J. ft Smith, eds., pp. 101-116. 1985

the case of Dutch - b a a r ,

Non-native suffixes and stress In Dutch, In LIN 1985, pp. 101-110.

SELKIRK, E.O. 1982

The syntax of words, MIT Press, Cambridge Mass.

SIEGEL, D. 1974 Topics In morphology, MIT-dlssertation. STRAUSS, S.L. 1982 Lexlcallst phonology of English and German. Forls, Dordrecht.

in v a n

Laszlo Maracz

On transivity in non-configurational languages

0. Summary It has been observed that non-configurational languages lack the rule of NPmovement. For example, it is difficult to show that passivization, where it exists in a non-configurational language, involves the operation of a movement transformation. In the Government and Binding framework (cf. Chomsky (1981)) passivization is forced by the interaction of Case Theory with Theta Theory moving an NP generated in a D-structure caseless theta-position to a cased nontheta-position. In Hale (1983) it has been demonstrated that the grammar of nonconfigurational languages is characterized by a dichomoty of Lexical Structure (LS) and Phrase Structure (PS). In that concept of non-configurationality the lack of NP-movement is accounted for by the fact that neither at LS, which consists of theta-positions only, nor at PS, which contains only non-thetapositions, the conditions of NP-movement are fulfilled. The question arises then': how are transitivity alternations derived in non-configurational languages? In this paper we will argue that such alternations can be accounted for within a well-articulated theory of the Lexicon (cf. Hale&Keyser (1985)) combined with the universal principles of the grammar (cf. Chomsky (1981)). In section 1 we will prepare the ground by sketching some specific properties of the Lexicon, Theta Theory and Case Theory in non-configurational languages. This will be illustrated by examples from Hungarian. In section 2 and 3 we are ready to analyze some transitivity alternations in Hungarian, resp. Warlpiri and Basque. Finally, in section 4 we will draw some conclusions based on the results obtained in the preceding sections.

1. On the Lexicon, Theta Theory, and Case Theory in Non-Configurational Languages 1.1. The Lexicon Following Hale&Keyser (1985) we will assume that a lexical entry of a verb contains in addition to the registration of its morphophonological and categorial features two parts which are relevant for its syntactic realization: the Lexical Conceptual Structure (LCS), roughly its dictionary meaning, from which the theta-grid can be derived, and a Lexical Structure (LS) which represents the subcategorization properties of the verb. The latter is to be associated in non-configurational languages with the Case frame selected by the verb. For example, the Hungarian dyadic verb esuk 'close> of the agent-theme semantic class selects a NOM-ACC Case frame: (1)(a) LCS for Hungarian 'close' (simplified): x carries out an action on y such that y CLOSE (b) theta-grid for Hungarian 'close': (agent, theme) (c) LS of the Hungarian 'close': [arg, arg, v] (d) Case frame of the Hungarian 'close': NOM-ACC The entities in the theta-grid which belong to a universal set of thematic roles (cf. Gruber (1965), Fillmore (1968), Jackendoff (1972)) are introduced by the

162

participants involved in the action denoted by the verb. An argument in LS may evaluate the associated variable in the LCS of the verb. This can be represented as follows: (2) [ arg^, arg^, v] For verbs of the agent-theme semantic class the agent is always connected to the nominative Case, whereas the theme is connected to the accusative. We can encode this regularity as in (3): (3) [ nom , acc , v] x y LCS and LS are related by linking or mapping conventies in the sense of Carter (1976), and Ostler (1980). Following Guerssel et al. (1985) we will call the LCS of a verb linked to LS its Predicate Argument Structure (PAS). It has been argued that, although the PS representation is flat, the LS arguments are ordered hierarchically in non-configurational languages (cf. Hale (1983)) reflecting the universal shape of the subject-predicate partitioning (cf. Chomsky (1981), Rothstein (1983), Marantz (1984)): (4) [ arg [ ,arg, v] ] v" 3 v Note that the [arg, v"] functions as the external argument, i.e. subject, in the sense of the Extended Projection Principle (cf. Chomsky (1982)), and the [arg, v'] as the internal argument, i.e. object. 1.2. Theta Theory From subsection 1.1. it is clear that assignment of thematic roles to the arguments of a verb depends on the meaning of the verb. In a sense both the agent and theme role of asuk, for example, are assigned by the verb itself. We expect then that 'compositional' theta-assignment of the VP to the subject as in configurational languages will not occur in non-configurational languages. The following tests show that this prediction is born out. (a) In Chomsky (1981) the following sentences are presented: (5)a b

John broke the window John broke his arm

In (5a) John is understood as the agent, while in (5b) it may be understood as the agent, or, preferably, as the patient. The possibility of the latter reading is related to the presence of other material than the verb in the VP (cf. Chomsky (1981)). In Hungarian the two readings of (5b) are disambiguated. They are associated with two different lexical forms of the verb 'break' which select their own PAS. The active reading is expressed by the form eltdr which is of the agent-theme class and hence associated with a NOM-ACC Case frame, whereas for the passive reading the intransitive variant eltdi'ik is selected. Its argument, in (6b) a possessive NP, is assigned the theme role and is marked nominatively: (6)a b

JSnos(nom) eltfirte kezet(3sg.agr-acc) Janos keze(3sg.agr-nom) eltorott

163 (b) Consider the following sentences: (7)a b

Mari oli az embert Mary-nom kills the man-acc "Mary is killing the man" A szomjusag oli Marit the thirst-nom kills Mary-acc "Mary suffers from the fact that she is thirsty"

In Marantz (1984) it has been argued that in English object selection determines the theta-role assigned to the subject, but not the reverse. From (7a,b) we observe that in Hungarian both cases are possible. In (7a) the selection of an agent subject implies a theme object, whereas in (7b) the selection of a theme subject implies an experiencer object. (c) In E. Kiss (1985) it has been observed that the free variable in idiom frames is not restricted to the subject argument, but can be any direct argument of the verb. Consider an example in which the free variable is the object: (8)

¿so, kapa valassza el oket spade, hoe separate away them "Only death shall separate them"

From the tests above we observe that in non-configurational languages thetaassignment does not lead to subject-object asymmetries as in configurational languages (cf. Chomsky (1981), Rothstein (1983), Marantz (1984)). Rather the arguments of the verb in non-configurational languages are assigned their theta-role by the verb itself. Each verb is associated with a particular set of thetaroles which constitute an essential part of its meaning. Hence, all the theta-roles associated with an argument position at LS are properly saturated in the sense of Rothstein (1983), that is, are assigned a theta-role directly by the verbal head. Formally, taking government (cf. Aoun&Sportiche (1983)) as a necessary condition on theta-assignment, we may say that the verb assigns its theta-roles to its arguments at LS under government. Note that although the biuniqueness condition on thetaassignment as elaborated in Chomsky (1981) seems to be violated in non-conf igurational languages the Theta-Criterion is not. Several authors (cf. Jackendoff (1972), Ostler (1980), Carrier-Duncan (1985)) have presented arguments in favor of a thematic hierarchy with universal validity. Following the above authors we will assume the thematic hierarchy in (9) : (9)

agent > theme > path (goal, source, location)

1.3. Case Theory Hungarian has an extensive Case system consisting of 18 morphologically marked Cases. Nearly all the Cases can be associated with inherent arguments of the verb. These Case categories, however, are not allowed to appear freely. There are linear and hierarchical restrictions on their appearance: (a) Linearly Cases have to occur in 'Case patterns', the Case frames. A few of them are listed in (10) (see for an extensive list Molnar (1969)): (10)

number of verbal arguments 1 2 3

Case frames occuring NOM NOM-ACC; NOM- DAT; NOM- INS TR; NOM-ADV NOM—ACC-DAT; NOM-ACC-ADV

(b) In E. Kiss (1985), and in Maracz (1985) arguments have been presented that Cases are not only restricted linearly as in (10) but that they obey a hierarchy:

164

(11)

NOM > ACC > DAT > INSTR > ADV

In order to relate Grammatical Functions to Case categories we use a binary feature system consisting of the mnemonic labels [ subj] , and [ obj] (cf. Hale et al. (1977)). For example, the 'subject', and the 'object' can be represented by means of the following disjunctions [+subj, -obj] , respectively [-subj, +obj] . Hence, we can set up the following equations: (12)a b

Subj = NOM Obj = ACC, DAT, INSTR, ADV

(12) has to be read as follows. The subject is the nominatively marked argument. The object is the accusative argument, if there is one'; otherwise the dative argument, if there is one, etc. In Hungarian the verbal morphology (infl) is construed with the nominatively marked argument (cf. Maracz (1985)). This observation can be incorporated by taking LS as the projection of the verb and infl together. Hence, extending (4) into: (13)

LS for Hungarian [ v,i] " arg

[ v,i] '

It is clear that the LS configuration alone cannot account for the differential Case assignment. The choice of the Case frame is determined by the semantic properties of the verb. In that sense Hungarian can be said to display inherent Case assignment. However, once a particular Case frame has been chosen the associations of Case to the argument positions in LS is done in structural terms obeying the Case-assignment rules of Chomsky (1981) . 2. Some Transitivity Alternations in Hungarian Consider the following sentences: (14)a b

(15)a b

(16)a b

Mari szalad Mary-nom run-3sg "Mary runs" egy ember erkezett a man-nom arrive-past-3sg "There arrived a man" Mari becsuk-ja az ajto-t Mary-nom close-3sg the door-acc "Mary closes the door" az ajto becsuk-od-ik the door-nom close-unacc-3sg "The door closes" az ellenseg megver-t-e a sereg-et the enemy-nom beat-past-3sg the army-acc "The enemy beated the army" a sereg megver-et-ett az ellensSg-tol/Sltal the army-nom beat-pass-past-3sg the enemy-abl/by "The army was being beaten by the enemy"

The pairs in (14)-(16) represent transitivity alternations. In (14a) we have an unergative (active intransitive) monadic verb. Its single argument is active,

165

i.e. of the agent type, and is assigned nominative Case. In (14b), on the other hand, we have a canonical case of an unaccusative (ergative) verb. The single argument denotes a passive participant in the event depicted by the verb. This type of participant is referred to as theme. From the unmarked form of the noun we observe that it is assigned nominative. In (15) we have a transitive/ unaccusative pair. The transitive alternant (15a) is of the agent-theme semantic class and selects the NOM-ACC Case frame. In the unaccusative construction (15b) the root of (15a) is augmented by the unaccusative suffix -6d. Its theme argument is assigned nominative. In (16) we have an instance of the active transitive/ passive alternation. The passive variant is indicated by special verbal morphology glossed as PASS. According to Karoly (1982) the use of the passive construction is passing out of use but continues to be entirely productive. In both variants an agent and a theme are selected. In the active alternant (16a) these are associated with the NOM, resp. ACC, whereas in the passive variant (16b) with the ablative Case (ABL) or a by-phrase, resp. NOM. We can use Reflexivization to illustrate the fact that in both patterns the Cases are selected inherently. Reflexivization in Hungarian takes LS as its target and obey s the restriction that the antecedent must be higher on the Case hierarchy than the reflexive maga "self" : (17)a

a sereg megverte magat "The army beat itself" b * maga megverte a sereget c a sereg megveretett magatol/maga altal ••"The army was being beaten by itself" d »maga megveretett a seregtol/sereg altal

In order to derive these transitivity alternations we have to set up their PAS. The PAS of the unergative (14a), resp. unaccusative (14b) has the following form: (18)a

PAS for unergative run

t V, i]

LS:

b "

PAS for unaccusative arrive [ v,i] "

LS: nom

I v, ij ' I [ v,i] y come to be ARRIVE

[ v, i] LCS:

X RUN

[v'i]

LCS:

Note that the variables in the LCS are mapped onto the nominative argument of LS both in (18a) and (18b). Consider the PAS of the transitive/unaccusative pair (15a,b): (19)a

PAS for transitive close LS:

[v,i]"

b

PAS for unaccusative close

LS: v, i] •

LCS:

acc I v, ij x CLOSE V\

[v,i]" [ v,i] •

LCS:

[J,i] y come to be CLOSE

From the comparison of (19a) and (19b) we can isolate the effect of adding unaccusative morphology to the basic alternant of this pair, i.e. the transitive one, deriving the unaccusative alternant. The agent role is delected from the LCS in (19a). The PAS of the transitive/passive alternation (16a,b) has the following shape:

166

(20) a

PAS for transitive beat

b

LS:

LS:

[v,i]"

PAS for passive beat

nom

nom acc

LCS:

[v,i] "

x BEAT y

[v,i]

[ V,i] ' abl/PP

LCS:

iv.il '

x was BEATEN by

In (20) the agent is mapped onto the nominative position, whereas the theme is mapped onto the accusative. Passivization, i.e. adding passive morphology, falls out in two components: (a) demotion of the ACC in the Case frame of the transitive verb into ABL/PP; and (b) the realization of the agent role internally. In order to associate theta-roles and Cases in the lexical entry of the verb we can set up the following Linking Conventions for Hungarian: (21)a b

Link the agent to the external position at LS Link the theme to the internal position at LS

The conventions in (21) represent the unmarked case (cf. Fillmore (1968), Ostler (1980), Hale&Keyser (1985)). From (21) it follows that the matching of Thetahierarchy (9) with Case hierarchy (11) is mediated by LS. The rules in (21) will handle the unergative (cf. (18a)), and the transitives (cf. (19a, 20a)). Incase there is no agent present the EPP will force the linking of a theta-role in order of the precedence on the Thematic hierarchy to the external position in LS. This accounts for the associations in case of the unaccusatives (cf. (18b, 19b)). In order to derive the passive alternant (20b) we have to postulate a marked linking rule which as noticed above is a consequence of adding passive morphology: (22)

link the agent to the internal position at LS

Rle (22) applies before the regular linking procedure starts. General principles of the grammar will then associate the theme role with the nominative Case in (20b). 3. Some Transitivity Alternations in Warlpiri and Basque In the light of the discussion in the preceding sections we will attribute the following general properties to PAS in non-configurational languages: (23)a b c

thematic roles are drawn from a universal set (agent, theme, etc.) and are hierarchically ordered as in (9) Linking procedures are a function of thematic roles; the unmarked linking procedures are given in (21) LS is structured as in (4); its external position must be filled

Warlpiri and Basque have the following language specific properties. Both languages display an ergative system of Case marking. Following Levin (1983) we will assume that they are only morphologically ergative. The Cases ergative (ERG), absolutive (ABS), and dative (DAT) function as grammatical Cases, i.e., are direct arguments of the verb. They can be represented in the following hierarchy (cf. Hale (1983), Levin (1983) for details)): (24)

ERG > ABS > DAT

Further, the Case marking of arguments is fully determined by the thematic roles the arguments bear (inherent Case assignment):

167

(25)a b c

all agent arguments of a transitive verb will be assigned ERG, of other verbs ABS (Warlpiri) all agent arguments will be assigned ERG (Basque) all theme arguments will be assigned ABS (Warlpiri, Basque)

3.1. Warlpiri In Guerssel et al. (1985) it is reported that Warlpiri lacks transitivity alternations as the causative, middle, passive, and anti-passive. These alternants involve the manipulation of a basic PAS by means of marked lexical operations. Hence, the PAS of Warlpiri can be said to be closed, i.e. the variables introduced in the LCS of a given verb must be mapped onto its LS according to the principles in (23). The system outlined above and the languagespecific properties of Warlpiri, however, do not block the occurence of the transitive/unaccusative alternation: (26)a

b

Japanangka-rlu rdilyki-pu-ngu karli Japanangka-erg broke-3subj-3obj the boomerang-abs "Japanangka broke the boomerang" Karli rdilyki-ya-nu boomerang-abs broke-3subj "The boomerang broke"

From (26) we observe that this alternation involves two distinct verbs in Warlpiri, each of which incoporates the form 'rdilyki' suffixed with 'pu' for the transitive variant, and with 'ya' for the intransitive one. In (26a) the agent is linked to the external position at LS, and the theme to the internal position according to (23b). Case is assigned by (25a), resp. (25c). In (26b) the theme is mapped onto the external argument because of (23c). This is indicated by the agreement facts: in (26a) the theme argument is construed with object agreement, whereas in (26b) with subject agreement just as the agent argument in (26a).

3.2. Basque Consider the following sentences: (27)a

b

c

d

e

ni-k Miren ikusi d-u-t I-erg Miren-abs seen 3sg.abs-AUX-lsg.erg "I see Miren" gizona ettori d-a the man-abs come 3sg.abs-AUX "The man comes" gizon-a-(k) aundi-a-(k) d-a (d-ira) man-def.art-(pi)-abs large-def.art-(pi) 3sg.abs-AUX "The man (men) is (are) large ni-k Miren ikusu nahi d-u-t I-erg Miren-abs seen want 3sg.abs-AUX-lsg.erg "I want to see Miren" ni-k ettori nahi d-u-t I-erg come want 3sg.abs-AUX-lsg.erg "I want to come"

(3pl.abs-AUX)

In (27a) we find an instance of a transitive sentence. This is indicated by the transitive variant of AUX. The theta-roles and Cases are assigned in such structures by the rules (23b), resp. (25b,c). In (27b) a sample of an intransitive sentence is given which can be seen from the intransitive form of

168

AUX. Following Levin (1983) we will assume that all intransitive verbs in Basque are unaccusatives. Hence, in (27b) the theme argument will be mapped onto the external argument position because of (23c). This is supported, firstly, by the fact that the absolutive argument agrees in person and number with the predicative adjective (cf. (27c)); and, secondly, by the fact that the controlled argument in an embedded transitive sentence is the ergative argument, whereas in an embedded intransitive sentence it is the absolutive one (cf. (27d), (27e)). This follows from the Binding Theory which allows only external arguments to be controlled by an argument of the matrix sentence. 4. Concluding remarks In this paper we examined some transitivity alternations in the non-configurational languages Hungarian, Warlpiri, and Basque. We were able to handle the facts by relying on the general principles of the grammar as EPP, ThetaCriterion, Case-assignment rules combined with the properties of PAS in nonconfigurational languages (cf. (23)) and language-specific rules and principles such as inherent Case-assignment and the Case hierarchies (11) and (24). Notes I am indebted to the work and help of the linguists participating in the MIT-lexicon project (cf. Guerssel et al. (1985)) which stimulated me to write this paper. All errors are mine. This study was supported by the Netherlands Organization for the Advancement of Pure Research (Z.W.O.) under grant R 37-26, which is gratefully acknowledged. References AOUN, J. and D. SPORTICHE 1983 On the Formal Theory of Government, in The Linguistic Review 2, p. 211-236. CARRIER-DUNCAN, J. 1985 Linking of Thematic Roles in Derivational Word Formation, in Linguistic Inquiry 16.1., p. 1-34. CARTER, R. 1976

Some Linking Regularities, Ms., University of Paris VIII.

Chomsky, N. 1981 Lectures on Government and Binding, Dordrecht, Foris. 1982 Some Concepts and Consequences of the Theory of Government and Binding, Cambridge, MIT Press. FILLMORE, C. 1968 The Case for Case, in E. Bach and R.T. Harms (eds) Universals in Linguistic Theory, New York, Holt, p. 1-88. GRUBER, J. 1965 Studies in Lexical Relations, MIT dissertation. GUERSSEL, M., K. HALE, M. LAUGHREN, B. LEVIN, and J. WHITE EAGLE. 1985 A Cross-Linguistic Study of Transitivity Alternations, Ms., MIT. HALE, K. 1983 Warlpiri and the Grammar of Non-Configurational Languages, in Natural Language and Linguistic Theory 1, p. 5-47.

169

HALE, K., L. JEANNE, and P. PLATERO 1977 Three Cases of Overgeneration, in P. Culicover, T. Wasow, and A. Akmajian (eds) Formal Syntax, New York, Academic Press, p. 379-417. HALE, K. and J. KEYSER 1985

Some Transitivity Alternations in English, Ms., MIT.

HALE, K. and M. LAUGHREN 1983 The Structure of Verbal Entries: Preface to Dictionary Entries of Verbs, Ms., MIT. JACKENDOFF, R. 1972

Semantic Interpretation in Generative Grammar, Cambridge, MIT Press.

KAROLY, S. 1982 Intransitive-Transitive Derivational Suffixes in Hungarian, in F. Kiefer (ed) Hungarian Linguistics, Amsterdam, Benjamins, p. 185-245. KEYSER, J. and T. ROEPER 1984 On the Middle and Ergative Constructions in English, in Linguistic Inquiry 15.3., p. 381-416. E. KISS, K. 1985

Is the VP Universal? Ms., University of Budapest.

LEVIN, B. 1983

On the Nature of Ergativity, MIT dissertation.

MARACZ, L.K. 1985 Some Notes on Weak Crossover in Hungarian, in H. Bennis and F. Beukema (eds) Linguistics in the Netherlands 1985, Dordrecht, Foris, p. 129-139. 1985 The Interaction of Lexical and Syntactic Structure in Hungarian, Ms., University of Groningen. MARANTZ, A. 1984

On the Nature of Grammatical Relations, Cambridge, MIT Press.

MOLNAR, I. 1969 The Verbal Group, with particular reference to the Case Frames (in Hungarian), in Altalänos NyelvSszeti Tanulmänyok 6, p. 229-270. OSTLER, N. 1980 A Theory of Case Linking and Agreement, Bloomington, Indiana University Linguistics Club. REBUSCHI, G. 1982

Structure de l'enonce en basque, These pour le Doctoral D'Etat, Paris VII.

ROTHSTEIN, S. 1983 The Syntactic Forms of Predication, MIT dissertation.

Christine ter Mors

Syllable reduplication

1. Introduction This paper presents a n analysis of syllable reduplication w i t h i n the A f f i x - t o - X framework of Ter Mors (1984), w h i c h in its turn is one of the fields of application of CV-Phonology as developed by Clements & Keyser (1983). In the above analysis a large number of reduplication cases could be described as affixation of a CV-morpheme, the phonemic filling of w h i c h is derived from the phonemic melody of the affixed word. Agta: CVCCy ¿¿Mi

'leg', vs. CVC

+ CVCCV Ikiii

=

CVCCVCCy taitai-lci

'legs'

A n essential point is that the empty C's and V's are filled w i t h any proper element available. Another form of reduplication we find in certain languages is determined at the syllabic level. This can be represented as the attachment of empty cr - nodes. If w e w a n t to fill in these nodes by means of the same m e c h a n i s m as that used for empty CV-nodes, w e have to face the p r o b l e m that these o * - nodes can - h o w e v e r - n o t be filled by all proper elements available for association o n the next tier. Yindiny:

ff cT 'house', vs. cr«+ cvcvcV i l IJ I I gimuru

=

CVCVCT Willi dimuru

'houses' CVCVCVCVCV L! I I I! It W gimugimuru

*cr £ ff-

Fa

Ta

=

Paa

t 24-1-1986

=

Taa

+ 1-1-2000

Suppose now, that the sentence in (1) is truthfully uttered on 25-1-1986, the day of this year's Linguistics in the Netherlands conference. The dream is then located on 24-1-1986, and, as we have seen above, Taa is stated to be located on 1-1-2000. All this is indicated under (10). Since it follows from (10) that Taa lies in the past of Pa, it seems as if my analysis implies that 1-1-2000 lies in the past of 25-1-1986, and is therefore logically inconsistent. The discussion of identity relations across domain boundaries in section 2 can, however, also be applied to time intervals, which can be regarded as individuals that construct discourse addresses. Using the same notation as in (8) and (9), I will now show that the analysis of the chronology of (1) given in (10) is not logically inconsistent. (11)

t

Ta(t) =(yesterday,t) =(Fa,t) >(Pa,t)

>DREAM>

t

Taa(t) on(t,1-1-2000) =(Paa,t)

In (11), t is a discourse address for one specific time interval. This address is entered both in the primary domain and in the embedded domain which is introduced by the world-creating verb DREAM. In the primary domain, t is an argument of the functions listed in the left-hand column. In the embedded domain, t is an argument of the functions listed in the right-hand column. Now, the seemingly conflicting entries = (yesterday,t) and on(t,1-1-2000) do not lead to a logical inconsistency, because these predications are entered in different domains, which have different truth conditions. Cf. (8) and (9). Time intervals in discourse-domain chronologies can thus be regarded as individuals that construct discourse addresses. Like other discourse addresses, these addresses for time intervals can show identity relations across domain boundaries. Conflicting predications can be sent to these addresses, provided that these conflicting predications are not entered in one and the same domain. This is how discourse-domain theory can deal with the linguistically encoded deceptiveness of dreams.

204 4. The construction of chronologies in discourse domains I have argued that the sentence in (1) should construct the chronology in (10), and that the adverbials in (1) do not point to a logical inconsistency in my analysis. In this section, I will present a fragment of the lexicon of a government and binding grammar of English that constructs the sentence in (I) and its chronology in (10) via a free ride on various accepted principles such as c-command, government and the compositionality principle. Since (1) does not contain a perfect, I leave the syntax and semantics of the lexical entry for HAVE completely undiscussed. Since (1) does not contain a present tense I leave the semantic specification of the -PAST tense undiscussed as well. For discussion of these details, the reader is referred to Rigter & Beukema (1985) and Rigter (1986). (12) Domain tenses: +PAST [t=P>F]

,

INFL: [

V]

-PAST [...]

,

INFL: [

V]

(13) The domain-shift tense: +PAST [+=P]

,

INFL: [

V]

The domain-shift tense must be inserted iff the P of the domain constructed by this INFL projection, or the P of any higher domain, lies in the past of the P of a next-higher domain. (14) Verbs: DREAM , V: N [ [+=T] +6

INFL] +G

DIE , V: N t [t=T] +0

]

In (12) - (14), each lexical entry is followed by a comma. After the comma the syntactic category is stated, followed by a colon. The colon is followed by a syntactic subcategorization frame. Under the tenses and the verbs, their chronological specification is given in square brackets. The significance of the arrows in these specifications will be discussed below. Let us assume that the lexicon also supplies nominals like Pete and the president, which can be inserted in 9-role bearing N slots, and let us take the grammar of the ifezt-complementizer and the adjunction of time adverbials to INFL projections for granted. Under the assumption of an X-bar theory of syntax, the fragment of the lexicon given in (12) - (14) allows the construction of the tree in (15), which is built up by fitting together the subcategorization frames of lexical items.

205 (15) Adv

INFLl INFLl

YESTERDAY INFLO +PAST [+=p>F]

V2 N2

VI

PETE

Vo DREAM [+=T]

INFL 2 TH

+PAST

N2

VI

THE PRESIDENT Vo DIE [+-T] The tree in (15) is in accordance w i t h the insertion restrictions imposed by the lexical entries in (12) - (14). The highest INFL projection subcategorizes a V projection. W h e n DREAM is the verb inserted in this V projection, it projects a 9-role o n an external nominal argument, PETE, and, as a world-creating verb, it projects a 9-role on a subcategorized INFL projection. The embedded INFL projection subcategorizes a V projection again. W h e n D I E is the verb in this V projection, it projects a 9-role on an external nominal argument, T H E PRESIDENT. Note that the tenses under the INFL projections structurally govern the external arguments of the v e r b s w h i c h they subcategorize and can thus assign Case to these nominals. The tenses thematically govern the verbs w h i c h they subcategorize, and can thus induce the correct morphology. For details see Rigter & Beukema (1985). Now let us look at the chronological specifications of the tenses. In (12) and (13) there are two +PAST tenses. The +PAST tense inserted under the highest INFL projection in (15) can only be the domain tense in (12). It cannot be the domain-shift tense in (13), because the insertion restriction on the +PAST tense in (13) implies that the domain-shift tense can only be inserted in an INFL that constructs an embedded domain. Conversely, the +PAST tense under the embedded INFL projection can only be a n instance of the domain-shift tense in (13), because the P under this INFL lies in the past of the P of the next-higher INFL. The difference between the chronological specifications of the two +PAST tenses in (15) is that the domain tense constructs an F in the past of the P of its domain, whereas the domain-shift tense does not construct an F. For detailed arguments underlying the proposal for the lexical entries for tenses and verbs in (12) - (14), the reader is referred to Rigter (1986). Syntactic trees like the one in (15) construct (increments in) discourse domains in accordance w i t h the compositionality principle. 9-role bearing arguments construct discourse addresses. Tenseless verbal projections construct addresses for propositions. The verb constructs the function in such a proposition, and the arguments are constructed by the 9-role bearing complements and specifiers on the relevant v e r b a l projection. World-creating verbs such as DREAM open up a

206 subdomain which is constructed by the part of the tree that is c-commanded by the world-creating verb. To understand the construction of the chronology of the discourse domain built up by the syntactic tree in (15), we must know what the arrows in the chronological specifications of tenses and verbs stand for. The * in these specifications can be read as an instruction to replace this by the right-hand term in the first c-commanding chronological specification. When we apply this instruction to the tree in (15), we see that it constructs the chronology in (10): the arrow of DIE points up to the P of the tense in the embedded INFL; the arrow of this tense points up to the T of the world-creating verb DREAM; the arrow of DREAM points up to the F of the tense in the highest INFL. Note, however, that the arrow in the highest tense finds no c-commanding chronological specification, and is therefore not replaced by anything. This means that, more precisely, the chronology constructed by the tree in (15) is not that in (10), but that in (16).

(16) + = P > F = T

=

P = T

I propose that the unrepl^ced arrow in (16) — i.e. the arrow in the highest INFL projection in (15) — is deictic, and that it points to the present of the model M into which the finite, partial model m represented by the discourse domain is mapped. This proposal has an interesting implication: it explains why headclauses are generally tensed, or, in other words, it explains why the highest projection in a sentence is generally an INFL projection. The explanation goes as follows. 1. The only lexical entries that contain a P — for present — in their chronological specification are the tenses, which are members of the INFL category. 2. The only position in which an uppointing arrow can remain unreplaced is the highest chronologically specified position in the tree. From 1. and 2. it follows that the only way in which the P of the primary discourse domain can be deictically anchored onto the present of the model M is to make the highest projection in the tree a tensed INFL projection. Thus, what used to be stipulated by the first rewrite rule in the days when generative grammar had not yet discarded PS rules appears to be semantically motivated by the need for the deictic anchoring of the P of the primary discourse domain on the present of the model M into which this discourse domain can be mapped. Within the space limitations imposed on this article, the grammar of the time adverbials in (15) cannot be discussed in any detail. Suffice it to say that the adjunction of these adverbials is such that the designators of the time intervals on which they place certain limits are within their scope. Thus, in the primary domain, the focal episode F and the time T of the dream are in the scope of YESTERDAY, and in the embedded domain the time T of the death of the president is in the scope of ON 1-1-2000. As we saw in section 3, it is important to see that the adverbial ON 1-1-2000 and the chronological specifications of the tense and verb in the subclause are in the scope of the world-creating verb DREAM, and thus construct the chronology of the intensional subdomain that represents the speaker's conceptualization of Pete's dreamworld, while the adverbial YESTERDAY and the chronological specifications of the tense and the verb in the headclause are not in the scope of DREAM, and construct the chronology of the primary domain. The chronology constructed by the tree in (15)

207 is thus precisely the one argued to be correct in section 3, and represented by the notation in (11). 5. Conclusion In Rigter (1986) I have proposed a tense theory executed in terms of an integration of government & binding theory with discourse domain theory. The core of this theory is given in the chronologically specified lexical entries for three tenses (i.e. the two domain tenses +PAST and -PAST, and one domain-shift tense +PAST), for the perfective auxiliary HAVE, and for all other verbs. Given these five types of chronological specifications on lexical entries, the rest of my theory is a free ride on the principles of government & binding theory and discourse domain theory. X-bar syntax, subcategorization frames, c-command and government relations ensure the correct insertion of lexical entries in syntactic trees, and they ensure the correct construction of the syntax-driven chronology of discourse domains and subdomains. The insertion condition on the use of the domain-shift tense ensures the correct construction of sequence of tenses in even the most complicated sentences. The deictic anchoring of the present of the primary discourse domain onto the present of a model M into which this discourse domain can be mapped is the semantic motivation for what used to be stipulated by the first rewrite rule in the days when generative theory had PS rules. The tense theory in Rigter (1986) explicitly interrelates the chronologies of embedding and embedded domains. Adverbial specification of certain time intervals in these chronologies may result in the construction of conflicting predications about one and the same time interval. This is why the sentence in (1) looks like a counterexample to my theory. To show that it is not, I have argued that time intervals, too, are individuals that can construct discourse addresses, which, like other discourse addresses, can be entered into more than one domain. Seuren (1985) has shown that discourse representation structures can be incremented with conflicting predications about one and the same discourse address, as long as these conflicting predications are sent to entries of this discourse address in different (sub)domains. This means that the construction of the chronology of the sentence in (1) in accordance with the theory in Rigter (1986) — cf. (10), (11), (16) — does not yield a logical inconsistency. Notes *

This article was inspired by a question posed by Martin Stokhof, who also contributed the example in (1). I thank Frits Beukema and Ron Verheijen for their comments on the first draft. 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.

1. As will be shown in section 4, the chronology in (10) is constructed by the syntax in (15). The question as to whether the adverbial on the subclause constructs an Faa in the subdomain — such that Paa = Faa = Taa — is bypassed here, but does not affect the argument. For discussion see Rigter (1986).

208 References FAUCONNIER, G. (1979) M e n t a l spaces. Unpublished paper. Université de Paris VIII, Vincennes. JACKENDOFF, R. (1975) O n belief-contexts. In Linguistic KAMP, H. (1981)

Inquiry

6, 53-93.

A theory of truth and semantic representation. Reprinted i n Groenendijk, J., Janssen, T. & Stokhof, M. (eds), Truth, pretation and information, GRASS 2. Dordrecht: Foris.

inter-

RIGTER, G. H. (1986) Focus matters. In Lo Cascio, V. & Vet, C. (eds), Temporal in sentence and discourse, GRASS 5. Dordrecht: Foris. representation RIGTER, G. H. & BEUKEMA, F. H. (1985) A government and binding approach Apeldoorn: V a n Walraven. SEUREN, P. A. M. (1985) Discourse

semantics.

to English

sentence

structure.

Oxford: Blackwell.

VAN EIJCK, D. J. N. (1985) Aspects of quantification university of Groningen.

in natural

language.

Doctoral dissertation,

Engin Sezer & Leo Wetzels

On the interaction of backness and rounding harmony

1. Introduction In this paper we will comment on the theoritical implications of a type of Vowel Harmony (henceforth VH) attested in Hungarian and Uygur. We will advance the claim that in those cases where VH involves incomplete harmonic sets, the spreading of harmonic features has to crucially interact with certain independently motivated well-formedness rules discussed by Stanley (1967) and Kenstowicz & Kissebirth (1977). Furthermore, we will demonstrate that in both languages one type of harmony (roundness or backness) feeds into the other type and that therefore the tacit assumption in autosegmental phonology that backness and roundness harmony are independent from one another must be abandoned. The reader should be cautioned that we do not intend a full-fledged analysis of the harmony process in either language, but that we simply point out the theoretical implications of the interaction of roundness and backness harmony with well-formedness rules.

2. The Hungarian case Vago (1980) mentions a number of suffixes in Hungarian which show both backness and roundness harmony. Consider the following examples (taken from Vago, 1980:18): (1)

2nd plural (a) hoz-tok (b) foz-tok (c) n6z-tek

dative 'bring' 'cook' 'see'

(d) hàz-hoz (e) fiil-hoz (f) fej-hez

'house' 'ear' 'head'

In the linear framework of Vago (1980),the vowel of these suffixes is underlyingly /o/. It is fronted by rule after front vowels and is further unrounded after front unrounded vowels (lc,lf). Such an analysis is not directly available in current versions of autosegmental phonology One might suggest that the suffix vowel in question is to be underlyingly specified as /o/, and that, in this particular case, spreading is a rule that changes feature values. Such a strategy, however would clash with the treatment of non-alternating (opaque) /o/ in suffixes like kor, ko, and iroz, mentioned by Vago (1980) 2 . A more recalcitrant problem for an autosegmental account of the facts stated in (1) is how to constrain the spreading of the [ - r o u n d J feature of h&z 'house' (Id) to the dative suffix hoz. We wish to claim that the apparent problem of accounting for cases as (Id) in the autosegmental framework will be resolved by the interaction of a redundancy rule with harmonic spreading.

210 We will assume that root vowels are specified autosegmentally for backness and roundness, and that the suffix vowels in (1) are not underlyingly specified for the features 'back' and 'round'. If we now look at the harmonic distribution of (1), we observe in the crucial case of h6z-hoz that the suffix vowel is round if back. This observation can be formally stated as: (2)

+ syll -high +back

[+round ]

Rule (2) is a segment-formation redundancy rule in the sense of Kenstowicz and Kissebirth (1977). It expresses the observation that Hungarian lacks the mid,back, unrounded vowel Ja] , or that a vowel which is [-high ] and [ +back ] is predictably [fround ]. We consider rules like (2) to be everywhere rules, in that they apply whenever their structural description is met. We will moreover have to stipulate that in Hungarian spreading on the backness tier precedes spreading on the roundness tier 3 . This stipulation is the formal expression of the fact that in Hungarian roundness depends on backness. In (3) below it is shown how hdz-hoz 'house-dat' is derived from / h^z-hEz/: (3)

+B

1

hSz-hEz 2

-R

+R

As stated above, in Hungarian the spreading of the feature 'back' precedes the spreading of the feature 'round'. Once this spreading is accomplished (3 1 ), the segment-formation rule is immediately triggered (3 2 ). The marking of the suffix vowel as +R preempts the spreading of the - R autosegment of the root. 3. The Uygur case1* We now turn to a similar case in Uygur,where again a well-formedness condition interacts with harmonic spreading, but this time it is the roundness value that determines the specification for the backness value. Standard Uygur has the following eight vowels:

(4)

i

li

u

e

o

o

a

a

211

As reported by Pritsak (1956) this language has three types of harmonic suffixes: (5)

(a)

A

{a,a}

(b)

U

= {u,ii}

(c)

I

= {i, li, u}

gol-larr imarat-lar kal-gu-si yaz-gu kor-up tur-up var-ip

'arm-plur' 'soup kitchen-plur' 'his coraing-fut' 'writing-fut' 'see-and' 'stand-and' 'arrive-and'

Ordinarily, suffix vowels are back after back vowels, except that /!/ is neutral and appears where /i/ would be expected: (6)

(a)

qayi-da

'sometimes'

(b)

zil-lar

'years'

In (6a), the suffix vowel agrees in backness with the a_ preceding the neutral vowel In (6b), we typically have a neutral vowel root. It would take us too far afield to present and justify a theory of neutrality in such a short paper as this one. We will therefore concentrate on examples like (5c), but will briefly come back to the problem of neutrality below. What is crucially important for the treatment of cases like (5c) is the languagespecific segment-formation rule (7) (7)

+syll +high -round

[-back]

Rule (7), above, expresses the redundant information that, since the /i/ is nonexistent in the phonological inventory of Uygur, a segment that is [+highj and [-round J is predictably [ - b a c k ] . As in the case of rule (2), we assume that (7) is an everywhere rule. Now, the derivation of the first two examples in (5c) is quite straightforward, assuming, as we have done in the case of Hungarian, that stem-vowels are marked for both roundness and backness, and that the suffix vowels are not specified for the harmonic values 'back' and 'round'. (8)

-B > \

V

(a)

kor-Ip

[korup]

-R +B \

(b)

tur-Ip

+R

[turup]

212

As for varip 'reach-and', it is not clear how in current versions of the autosegmental theory the suffix vowel is to be marked as back] . We will assume for Uygur that spreading on the backness tier follows spreading on the roundness tier. Hence, the derivation of varip is as follows: (9)

+B

-B 2

var- Ip / / i 4

-R The spreading of the - R aotosegment, which must preceed the spreading on the backness tier, triggers the redundancy rule (7), and thereby blocks the spreading of the [+back] specification. 4. Neutrality Before we go into the discussion of the implications of the types of harmony processes we have discussed so far, we would briefly like to comment on examples like (6) above. First, from the examples given in Pritsak (1956) it is rather evident that i^ in the dialect under discussion behaves like a neutral vowel, although the exact nature of this neutrality is somewhat obscure.For cases like (6a), we may assume, after Booij(1984), that the neutral vowel is segmentally, rather than autosegmentally, specified for [-back] and it therefore does not interfere with the spreading of the preceding [+back] value to the vowel on its right. This strategy involves the specification on two different tiers, and for this reason, it is questioned in the literature by v.d. Hulst (1985). Let us, as an alternative,consider the possibility that neutral vowels, by rule, gain to their right a floating copy of the preceding autosegment (call this the Hopping Rule) which then spreads to the following unspecified vowels. (10)

+lf-B "+B tanver- nfik

-R -R

t&ny6rnak

'plate-dat'

-R

Notice that this solution is not subject to the problems pointed out by v.d.Hulst (1985) concerning Booij (1984), since it will no longer be necessary to set up two different tiers for the specification of harmonic features. We claim that the Hopping Rule represents the unmarked case, and that it would be very highly marked, if at all possible, for a neutral vowel, as in (10),above, to behave opaque in some roots. Notice that in such cases, the Hopping Rule must be diacritically blocked. Let us, now, look at the situation in monosyllabic roots. It is well known that Hungarian has two types of monosyllabic roots with and namely, those that behave as neutral and those that behave as opaque. Observe the following:

213 (11)

-B «\

(a)

«

szin-nEk -R

szlnnek

'color-dat'

-»R

(b)

\

hid-nEk

*hidnek

'bridge-dat'

- R -R In neither case in (11), above, the Hopping Rule we have alluded to can apply, since there is no autosegment to the left of -B. In such cases, we claim that (11a) represents the unmarked case and (lib) is the highly marked case, since children learning the language have no phonetic evidence whatsoever, that a root like hid induces C+backJ harmony on the following vowel. We assume that the correct representation for neutral roots as in (lib) is like (12),below: (12)

-B hid

+B -

nEk

hidnak

'bridge-dat'

Here, +B is a lexically specified floating feature located at the right periphery of the root. We would predict that cases like hidnak are rare and unproductive. Finally, we consider bisyllabic roots that contain only neutral vowels. It is noted by Vago (1980) that polysyllabic roots taking f-back] vowels are predominantly opaque; that is, they induce ^ b a c k j harmony on the following vowels. We would, indeed, predict this to be the case. In other words, given (13), below, (13)

-B

-B

benzin-nEk -R

-R

benzinnek

'gasoline-dat'

-R

The Hopping Rule will apply to produce (14), below: (14)

-B

-B

-B

benzin - nEk -R

-R

-R

214 Then, spreading will apply as indicated. This mechanism seems to adequately express the observation that native speakers of neutral-vowel languages tend to interpret sequences of neutral vowels as opaque. Exceptions to this generalization exist, but at a very high cost, since not only must they be lexically assigned a floating autosegment, as in the case of monosyllabic neutral roots, but the Hopping Rule must be diacritically blocked^ 5. Discussion In this paper, we have discussed two cases of vowel harmony involving incomplete harmonic sets—{e,o,o}for Hungarian and {_i,.u,u. }for Uygur. An incomplete harmonic set is one in which not all permutations of the distinctive harmonic features are represented. Both sets are missing [+back, -round J segments. This is why both sets give rise to disharmonic sequences under h a r m o n y — Hungarian h6zhoz 'house-dat' and Uygur varip 'arrive-and'. The question was how the suffix vowel was to be marked [+roundJ for Hungarian and [-back] for Uygur. We believe that such values are supplied by the relevant redundancy rules that apply everywhere when their context is met. It should be pointed out that in our analysis both the redundancy rules in question and spreading are functionally akin, in that they supply the missing harmonic values. The kind of interaction between MS redundancy rules and regular phonological processes has a long history in generative phonology. Stanley (1967) discusses what he calls "segment structure rules, " and points out the need for such rules to supply redundant information in the P(honological) rule component of the grammar. Instead of duplicating the relevant MS rules in the P rule component, Stanley (1967:404) suggests to adopt the convention that "the output of each P rule is automatically subjected to the segment structure rules." The issues relating to the everywhere status of morpheme structure rules is also discussed by Kenstowicz and Kisseberth (1977). The position we are taking here is in agreement with Stanley's original suggestion. Another point of interest concerns the formulation of the redundancy rules, (2) and (7). Notice that the two segment structure rules in question could have alternatively been formulated as (15a) and (b) for Hungarian and Uygur, respectively. (15)

(a)

-high -round.

[-back]

Cf. (2)

(b)

+high _+back.

[+round]

Cf. (7)

H >H *H -i-l M f-< fi en ^ r-l « 3 >. t-l in rÜ r-l C flj 3 O. O. O 4J- r-l

238 (7)

kelelli-:

keela-

k1

1

[c V C V c c ] - V I I I cititti-:

kojojji-:

cittak-

kojoopi-

e

a

l

c1 t k 1 r , - V [c V C V c^ cj 1 ! I i k

a j

l

kelalli-

*citakki-

p

[C VI C VI C^fc] - I V o o 1

"kojoppi-

The reason for the incorrect derivation in these cases is clear - no segment from the source w o r d apart from the first two consonants and the first vowel ever appears in a colour word. A similar problem w o u l d show up in the case of option (6b) - the reduplicative template [cvc+vcc] - because of McCarthy's condition regarding the avoidance of morphological opacity (McCarthy, 1982): (8)

All elements of the root must appear at least once in the output

In fact, as is m a d e clear in the above article, this condition is somewhat less strict, and can be effectively interpreted as follows: (9)

All elements of the root must appear at least once in the output if possible

The output in the three cases w e are considering w o u l d then be as follows: (10)

kelelli-:

keela-

k 1 I I C V C

cititti-:

kojojji-:

cittak-

kojoopi-

k 1 II +

1 ©

V C C

kelakli-

©"a

cI tV ® c\© Ik C V C + V C C l A r \ i (a) (i/ a k j @ C V C

- V

- V I

*citacki-

k © p +

V C C

i © © © " o

©

- V 1 1

kojokpi-

We have assumed here that the segmental melodies are attached from left to right in b o t h elements of the reduplicated structure. If, as is claimed in Marantz (1982) and Ter M o r s (1984), the unmarked case of reduplication is that whereby a reduplicated element on the right is associated from the right, and one o n the left from the left, the unmarked case, in conjunction w i t h McCarthy's above-mentioned condition, w o u l d give us respectively: (11)

*kelakli'citatki"•kojijpi-

I therefore conclude that there is no reduplicative template involved in this case. The third option (6c) - a morphological template [CVCj w i t h a suffix - V C C i - will give the correct output, however, if spreading is assigned to this

239 morphological process: (12)

kelelli-:

keelaLc V I

V~"C C V kelelli-

cititti-:

cittak[C V C] - V C~C V1 i ©

kojojji-:

i

cititti-

kojoopikojojj i-

®G>

I have demonstrated in (5) that association does not take place across a morpheme boundary. If we accept this as the correct analysis, we may ask ourselves if other cases with apparent reduplication, involving at most the repetition of a single vocalic melodic element and a single consonantal melodic element, are susceptible to such an analysis. A case in point would seem to be the case of the (nominalized) adjectival formative suffix, segmented by Broadbent (1964) as -m:a-. Relevant examples would be the following: (13)

kaalliwwatekmhelaajjelly-ccjoloQ-Qeputkal-

"kick with heel" "talk" "kick with toe" scare "stop talking" "be worried" "guts"

kally?y-mma1iwwa ? a-mmatekkymy-mmahellaja-mmajellycy-mmajolloQo-inmaputtaka-mma-

"kicking horse" "talkative" "kicker" "robber" "quiet person" "easily worried" "fat person"

This segmentation causes a problem in that it appears that the spreading parameter - the value of which is morpheme-specific in Sierra Miwok - has a positive value in some cases for vowels - e.g. liwwa?a-mma- - while in other cases it has a negative value - e.g. tekkymy-mma-, where we observe two instances of the default vowel y. As far as consonants are concerned, the value of the spreading parameter is consistently negative - e.g. 1iwwa?a-mma-, where we oberve an instance of the default consonant ?. As in all other cases in Sierra Miwok the spreading parameter has a consistent setting for each morphological process, it is clearly desirable for this to be the case here too. This means that a reanalysis is called for. The solution to this problem would seem to be to segment these structures as CVCCVC-Vmma-, in other words to claim that what we have here is a morphological process involving a suffix -Vmma- which selects a morphological template [CVCCVC] (derived template Type 5): (14) i.

kally?-ymma-:

kaal-

ii.

liwwa?-amma-:

liwwa-

k 1 m 1' i\ , .. A , r [ c v c c v c Tj - v c cA v 1

w

m

ri v c \ v cl - y (A; v I I a a i

240 (14)

iii.

tekkym-ymma-:

tekm-

iv.

hellaj-amma-:

helaaj-

puttak-amma-:

putkal-

t k m m l\ I, A. [ c v c c v c j - v c c v l l rl

h 1 i m A ri l\ I, C V C C V C - V C C V l U ^ I e a a ; K u [ C V C C V C ]

0

-

. V CA :

We observe that the form jolloQ-omma- provides a n illustration of the general principle that the OCP is not operative intramorphemically in Sierra M i w o k , in contrast to the situation intermorphemically (see Smith (1985) and below). (15)

jolloi^-omma-:

joloij-ije-

J

1 m 0 A IN C C V ¿J- V C C V

Note that the double medial consonant in this derived template is caused by a n Initial Association Rule (IAR) operating prior to the association conventions. See Smith (1985) for further details on this. In Smith (1985) I set up a restriction on the second element of a vocalic melody preceding the second element of the corresponding consonantal melody. This restriction h a d no clear basis, other than that it was required to produce the correct output, and was therefore wholly undesirable. A n example of a case where this restriction was operative was the following: (16)

?eeppuÇfahywaat-

"spread out (fire)" "bubble u p "

?epeppepppututtuÇthywattatt-

"spread out (liquid)" "bubble up fast" "run around"

The problem was that the final a_ in the second form did not appear in either the second or the third v o c a l i c slot in the derived structure. Clearly this problem can be got around by positing a [cvc] morphological template w i t h a suffix consisting w h o l l y of empty slots of the form - V C C V C C - . However if w e were to adopt this solution the wrong output w o u l d be obtained in the case of hywaat-. The iterative f o r m w o u l d turn out as * h y w y w w y w w - . The answer in this case w o u l d appear to involve a complex template w i t h a first element of v a r i able length, depending o n the number of consonants in the root. If there are two consonants in the root then the first p o r t i o n of the template is CCVC]; if there are three consonants in the root then the first portion of the template is [CVCVCJ. Following an internal morpheme boundary we then have whatever length of suffix is required to bring the total length of the template up to [CVCVCCVCC]. The complex template required m i g h t be represented as follows: (where a ^ b) (17) U C V C ( V C ) ] + (VC),CVCCl a b The last two examples in (16) w o u l d then be derived as follows:

241 (18)

pututçuçt-

putta-

hywattatt-:

hywaat-

Now the noil-association of the a in puta can be explained as a consequence of the fact that association does not take place across morpheme boundaries in Sierra Miwok. Note that this concept of complex templates with elements of variable length gains support from other cases in Sierra Miwok. Consider the following data: (19)

huk-sicuj-ije?okiihwyllyçhitpciile-

"smell" "salt" "get sweet" "beg for food" "hot" "get cold" "red pepper"

huks is-sekojkoj-jecujcuj-je?okhih-hewyliyç-tehitpyp-pecilcil-le-

"be smelly" "taste salty" "taste sweet" "be pitiful" "be hot" "be cold" "taste peppery hot"

The cases with two consonants in the root seem to be derived via a complex template involving reduplication: (20)

[[CVC] +CVC]

Those with three consonants in the root seem, on the other hand, to be derived via a template of the same overall length but with a longer first portion: (21)

[fcvccvl+c]

The case of hitp- demonstrates that what is involved here is not simply spreading, as the case of ?okiih- might suggest. The result would in that case have been hitpip-pe-. However the second vowel is not a spread i but a default y suggesting the presence of a morpheme boundary. Some sample derivations follow: (22)

kojkoj-je-:

kojjo[fc v c] - c v c] o ®

cujcuj-je-:

cuj-