Phonological Domains: Universals and Deviations 9783110217100, 9783110205404, 2009021226

Citation preview

Phonological Domains

≥

Interface Explorations 16

Editors

Artemis Alexiadou T. Alan Hall

Mouton de Gruyter Berlin · New York

Phonological Domains Universals and Deviations

edited by

Janet Grijzenhout Barıs¸ Kabak

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

앝 Printed on acid-free paper which falls within the guidelines 앪 of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication Data Phonological Domains : universals and deviations / edited by Janet Grijzenhout, Barıs¸ Kabak. p. cm. ⫺ (Interface explorations ; 16) Includes bibliographical references and index. ISBN 978-3-11-020540-4 (hardcover : alk. paper) 1. Prosodic analysis (Linguistics) 2. Grammar, Comparative and general ⫺ Phonology. 3. Metrical phonology. I. Grijzenhout, Janet. II. Kabak, Barıs¸. P224.P735 2009 4141.6⫺dc22 2009021226

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.

ISBN 978-3-11-020540-4 ISSN 1861-4167 ” Copyright 2009 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book 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 in writing from the publisher. Cover design: Christopher Schneider, Laufen. Typesetting: PTP-Berlin Protago-TEX-Production GmbH, Berlin. Printed in Germany.

Contents List of Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii ix

Prosodic Phonology: An appraisal . . . . . . . . . . . . . . . . . . . . Janet Grijzenhout and Barı¸s Kabak

1

Part 1.

Prosodic Hierarchy and the Nature of Prosodic Constituents

The status of the Clitic Group . . . . . . . . . . . . . . . . . . . . . . . Irene Vogel

15

The distribution of phonological word domains: A probabilistic typology Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

47

Part 2.

Adjunction, Recursion, and the Nature of Syntax-Phonology Mapping

Phrase-level and word-level syllables: Resyllabification and prosodization of clitics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 Anna Cardinaletti and Lori Repetti An interface approach to prosodic word recursion . . . . . . . . . . . . 105 Barı¸s Kabak and Anthi Revithiadou The extended prosodic word . . . . . . . . . . . . . . . . . . . . . . . 135 Junko Ito and Armin Mester Multiple spell-out, assembly problem, and syntax-phonology mapping Yoshihito Dobashi

195

The role of Weight-by-Position in the prosodic development of Spanish and German . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Conxita Lleó and Javier Arias Decomposition of question intonation: The structure of response seeking utterances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 Aslı Göksel, Meltem Kelepir, and Aslı Üntak-Tarhan

vi

Contents

Phonological domains in Modern Icelandic . . . . . . . . . . . . . . . 283 Kristján Árnason Two phonologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Harry van der Hulst Index of subjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 Index of dialects and languages . . . . . . . . . . . . . . . . . . . . . . 356

List of Contributors Javier Arias University of Hamburg, Germany Kristján Árnason University of Iceland, Iceland Balthasar Bickel University of Leipzig, Germany Anna Cardinaletti University of Bologna and University of Venice, Italy Yoshihito Dobashi Kitami Institute of Technology, Japan Aslı Göksel Boˇgaziçi University, Turkey and School of Oriental and African Studies, United Kingdom Janet Grijzenhout University of Konstanz, Germany Kristine Hildebrandt University of Manchester, United Kingdom Harry van der Hulst University of Connecticut, United States of America Barı¸s Kabak University of Konstanz, Germany Junko Ito University of California, Santa Cruz, United States of America Meltem Kelepir Boˇgaziçi University, Turkey Conxita Lleó University of Hamburg, Germany Armin Mester University of California, Santa Cruz, United States of America Lori Repetti State University of New York at Stony Brook, United States of America

viii

List of Contributors

Anthi Revithiadou University of the Aegean, Greece René Schiering University of Münster, Germany Aslı Üntak-Tarhan Cornell University, United States of America Irene Vogel University of Delaware, United States of America

Acknowledgements We are grateful to all the participants of the workshop on Phonological Domains: Universals and Deviations (DGfS-2007, Siegen, Germany) from which the contributions of this volume have been drawn. We are especially grateful to Anthi Revithiadou for her help in literally every step of this enterprise, to Ursula Kleinhenz for her help in the technical issues involved in the publication of this volume, as well as Tanja Reckenziegel and Wen-Hsuan Chiao for their invaluable editorial assistance. Special thanks also go to our colleagues, Heidi Altmann, Aditi Lahiri and Frans Plank, and to SFB 471 “Variation and Evolution in the Lexicon”, funded by the German Research Foundation (DFG). Finally, we would also like to extend our gratitude to our reviewers, (in alphabetical order) Luigia Garappa, Tracy Hall, Ben Hermans, William Idsardi, Marian Klamer, Jaklin Kornfilt, Martin Krämer, Astrid Kraehenmann, Claartje Levelt, Judith Meinschaefer, Marc van Oostendorp, Vassilis Spyropoulos, Hisao Tokizaki, Marina Tzakosta, Ruben van de Vijver, Jeroen van de Weijer and Ngee Thai Yap, without whom this volume would not become possible.

Prosodic Phonology: An appraisal Janet Grijzenhout and Barı¸s Kabak

1. Introduction Nearly three decades after first full-scale models on Prosodic Phonology were advanced, this volume presents a collection of original papers that approach various challenges that the theory has faced and offer new insights into morphosyntax-phonology interface. It not only extends and formalizes some established notions such as prosodic adjunction, but also critically reviews and questions other well-established ones, such as prosodic recursion. Moreover, new constituents (e. g., the Phrasal Syllable) are introduced and old ones, which previous studies were perhaps too eager to dismiss, are re-introduced (e. g., a constituent which is located between the Phonological Phrase and the Prosodic Word), thus showing how much controversy on the nature of the Prosodic Hierarchy still exists. Besides issues pertaining to the internal organization of the Prosodic Hierarchy, controversy still lingers on whether all of its constituents should be unified under the same umbrella, or whether they are needed at all. The volume also addresses the nature of the algorithm on the basis of which prosodic constituents are constructed and, finally, it explores whether there is evidence for such constituents from the early stages of language acquisition. In the following, we review a number of theoretical and empirical issues in Prosodic Phonology, and introduce the ways in which the contributions to the volume approach and address them. It is our hope that the range of topics presented in this volume and the novel prosodic data and empirical findings from diverse languages will stimulate further theoretical, empirical, and experimental research in various aspects of prosody and morphosyntax-phonology mapping.

2. Three decades of Prosodic Phonology Prosodic Phonology, as conceptualized in Selkirk (1980a, b, 1984) and Nespor and Vogel (1986), maintains that grammatical utterances are endowed with a hierarchical prosodic structure and that prosodic constituents such as the Prosodic

2

Janet Grijzenhout and Barı¸s Kabak

Word, the Phonological Phrase, the Intonational Phrase and the Utterance are defined on the basis of the morphosyntactic structure of sentences. Prosodic constituents may – but need not – be isomorphic to morphosyntactic constituents. In contrast to so-called “direct reference” theories (e. g., Kaisse 1985, Odden 1987, 1990), Prosodic Phonology assumes that the domains of application of phonological rules are expressed by the dictates of phonology (i.e., prosodic constituents) rather than morphosyntax. Prosodic Phonology makes it explicit that the structure of the prosodic hierarchy and its constituents are universal although the precise shape of the constituents is left to be filled in depending on language specific wellformedness conditions and constraints. The theory as such, with explicitly testifiable hypotheses, has paved way for numerous explorations across the languages of the world, which not only made it possible to substantiate the core arguments of the theory itself (see Nespor and Vogel 2007), but increased our understanding of the ways in which phonology interacts with syntax (see Truckenbrodt 2007 for an overview). In-depth investigations on individual languages within the past two decades (e. g., among others, Hayes and Lahiri 1991 for Bengali, Hannahs 1995 a, b for French, Kleinhenz 1998 and Hall 1999 for German, Green 2000 for Irish, Kang 1992 and Jun 1993/1996 for Korean, Frota 2000 and Vigário 2003 for Portuguese, Kabak and Vogel 2001 for Turkish), as well as dissertations focusing on cross-language comparisons (e. g., Inkelas 1990, Peperkamp 1997), yielded a wealth of observations, albeit with sometimes incompatible conclusions. Below, we focus on the issues and controversies surrounding the theory, based on which the chapters of this book have been arranged. Prosodic Hierarchy and the nature of prosodic constituents The spine of Prosodic Phonology is the skeleton that holds the prosodic constituents in a hierarchical fashion, namely the Prosodic Hierarchy. As in any theory that makes universal predictions, Prosodic Phonology also faced quite a number of challenges through in-depth investigations into typologically diverse languages. To remedy some of the problems and handle language-specific phenomena, researchers extended the range of prosodic domains (e. g., Condoravdi’s 1990 Minimal Phrase – also called constituent ‘z’ – which is located between the Clitic Group and the Phonological Phrase, Kanerva’s 1989, 1990 Focal Phrase, which is higher than the Phonological Phrase and lower than the Intonation Phrase; Ladd’s (1986) Major Phrase and Tone Group; Selkirk’s (2000) Minor Phrase and Major Phrase; Pierrehumbert and Beckman’s (1988) Accentual Phrase and Intermediate Intonational Phrase; and Cardinaletti and Repetti’s (this volume) Phrasal Syllable which comes in the prosodic hierarchy below the Phonological Phrase and above the Prosodic Word). It appears that

Prosodic Phonology: An appraisal

3

the proliferation of prosodic domains reflects the choice of language and phonological phenomena for investigation, and subsequently creates the problem of cross language compatibility and generalizability. For instance, the Accentual Phrase has been proposed as a result of extensive research on tonal phenomena in languages such as Japanese and Korean (e. g., Beckman and Pierrehumbert 1986, Jun 1989, 1993/1996, 1998, de Jong 1989), while research on prosodic phenomena above the word level on languages such as Bengali (Hayes and Lahiri 1991; Truckenbrodt 2002), German (Truckenbrodt 1995, Wiese 1996, Kleinhenz 1998), Greek (Revithiadou 2006), Hungarian (Vogel 1988, Kenesei and Vogel 1989), and Turkish (Kabak and Vogel 2001) typically use the constituents such as the Phonological Phrase. As such, it is often not clear whether the variation we observe is due to distinct phonological phenomena, typological diversity, or different research practices in the field. Diametrically opposing to the view that there should be more domains to account for typological diversity, we also observe various attempts to eliminate particular phonological domains from the hierarchy. The Clitic Group (Nespor and Vogel 1986, Hayes 1984, 1989, Vogel 1989, McHugh 1990) is perhaps the most notorious of all prosodic domains that raised a great deal of controversy in the world of Prosodic Phonology. Inkelas (1990), Selkirk (1995), Inkelas and Zec (1995), Booij (1996), Peperkamp (1997), Kleinhenz (1998) and Gerlach and Grijzenhout (2000), among others, strictly argued against this constituent on several empirical and theoretical grounds. The most serious blow to the validity of the Clitic Group came from arguments based on clitic adjunction at constituents higher than the Prosodic Word, namely the Phonological Phrase and the Intonational Phrase, ultimately weakening the liability of a constituent that is supposed to handle all clitic-related phenomena in a unified manner. Instead, language-specific subcategorization frames encoded in the lexical representation of clitics have been held accountable for the behavior of clitics. The Clitic Group was dismissed based on other arguments such as (i) the lack of evidence for this constituent in particular languages, (ii) encliticisation-procliticization asymmetries in different languages (e. g., Booij 1996, Hall 1999, Peperkamp 1997, Vigário 2003), as well as (iii) the subminimal nature of clitics, which had to be given the Prosodic Word status in Nespor and Vogel (1986) due to (then unviolable) Strict Layering Principle (see Vogel 1999 for a discussion). Nevertheless, some researchers continued to employ this constituent in an insightful manner to account for prosodic phenomena in different languages (e. g., Vaux 1998 for Armenian, Hayes 1989 for English, Hannahs 1995b for French, Kabak and Vogel 2001 for Turkish) and questioned the validity of clitic adjunction at higher levels of the prosodic hierarchy (e. g., van der Leeuw 1997). In this volume, Irene Vogel discusses various challenges to the Clitic Group, reviews

4

Janet Grijzenhout and Barı¸s Kabak

alternative analyses, and highlights psycholinguisitic evidence for this domain. With the assumption that a prosodic constituent must have unambiguously defined properties, she shows the necessity for the Clitic Group on the basis of cross-linguistic evidence and refutes recursive prosodic structures since they create insurmountable problems. Balthasar Bickel, Kristine Hildebrandt, and René Schiering take a typological route to ask what kinds of possible domains can be empirically substantiated by a sample of languages. Their approach is in striking contrast with several contributions in this volume as well as the bulk of the literature on Prosodic Phonology, and is thereby unique in that instead of assuming a finite and universally fixed set of phonological domains, the authors set out to explore principles that explain the distribution of the attested domains. Focusing on the kinds of domains found at the word level in a sample of languages, the authors employ quantitative methods to establish the kind of principles that guide the distribution of phonological word domains across a sample of languages. What seems to emerge from this line of research, and needs further typological investigation, is whether domains for tonal processes and vowel harmony on the one hand, and prosodic phenomena, namely stress and intonation, on the other hand, constitute distinct levels in phonology, an observation that is also explored by van der Hulst in this volume (see below). Adjunction, recursion, and the nature of syntax-phonology mapping As research introduced constraint-based analyses into prosodic phonology (e. g., van der Leeuw 1995, Selkirk 1995), various assumptions had to be revised, and there was more room for cross-linguistic variation. For instance, the weakening of the Strict Layer Hypothesis (Ito and Mester 1992) allowed the adjunction of subminimal elements such as clitics to Prosodic Words at the cost of the violation of the constraint Non-recursıvıty (Selkirk 1995). With the Clitic Group from the Prosodic Hierarchy, recursive structures could now be proposed to handle phonological processes or constraints that have previously been attributed to the Clitic Group. This move has indeed been on a par with an earlier observation that long-standing problems in intonational phonology can be solved through recursion at the higher levels of the prosodic hierarchy (Ladd 1986). Since then recursive Prosodic Words, for example, have been successfully applied to clitic as well as compound-related phonological phenomena in many languages such as Dutch (Booij 1996), Italian (Peperkamp 1997), Irish (Green 2000), and European Portuguese (Vigário 2003). One recurrent problem with the notion of recursivity in phonology is, however, the lack of a proper definition of recursion insofar as it relates to the generation of recursive phonological structures. Since recursion primarily arises through constraint violation, it is not clear, for

Prosodic Phonology: An appraisal

5

instance, what a well-formed recursive Prosodic Word can be, and what exactly can or cannot invoke recursive structures in morphosyntax-phonology mapping. This volume presents four contributions which specifically deal with the nature of syntax-phonology mapping in various languages and offer a fresh look at concepts and analytical tools which have long been taken for granted by linguists. Anna Cardinaletti and Lori Repetti present data from Northern Italian dialects where verbs plus enclitics display special phonological behaviour which suggests that these entities should not be analysed as Prosodic Words, but rather as a sequence of a verb and a ‘free clitic’ which together form a Phonological Phrase. They suggest a model of the prosodic hierarchy in which there is an additional phrasal syllable level (sPP ) nested under the Phonological Phrase. They cast their proposal in the framework of Optimality Theory and demonstrate how constraints and rankings account for the phenomena in question. Barı¸s Kabak and Anthi Revithiadou formulate a new approach towards recursion in phonology. In their view, recursive structures in phonology mirror recursive structures in morphosyntax. As extended domains, recursive prosodic constituents crucially inherit properties of their mother nodes, and, since they constitute extra layers in the derivation, they may optionally develop unique phonological properties. Such unique phonological properties are to a large extent rhythmic readjustment rules, which are expected under the assumption that an extended domain such as the Prosodic Word involves an additional layer of structure on the metrical grid to which rhythmic rules are naturally sensitive. Prosodic adjunction is also a focal point in Junko Ito and Armin Mester’s contribution to this volume. The authors take a reductionist approach towards prosodic representations, thereby restricting recursive structures that simply mirror the morphological form. They assign a pivotal role to adjunction as a representational devise for various types of subminimal elements (function words, clitics, prepositions, etc.), and call attention to certain, less explored aspects of the prosodic hierarchy, especially in relation to the way functional elements are parsed into prosodic constituents. The authors review the evidence put forward in previous analyses of English and German in favour of Phonological Phraseattachment of functional elements and arrive at the conclusion that the facts were fragmentary. Instead, Itˆo and Mester assume a rich projection structure with adjunction at its heart and argue that function words are not as impoverished as they previously have been regarded, but rather constitute dependent elements within extended Prosodic Word structures. Yoshihito Dobashi approaches the syntax – phonology interface from the angle of the Multiple Spell-Out Hypothesis, adopting the phase-based framework proposed by Chomsky (2001, 2004). Two major issues in relation to the mapping of the phase-material onto prosodic constituents, namely the Mismatch Prob-

6

Janet Grijzenhout and Barı¸s Kabak

lem and the Assembly Problem, have been addressed. The behaviour of subjects in certain languages is an instructive example for the first problem since subjects may resist forming a single Phonological Phrase with the verb even if they belong in the same phase. Dobashi offers an innovative proposal whereby the first element of the Spelled-Out phase-material, in this case the subject, is only linearized and not visible to the prosodization rules. As such, the subject functions as the element around which the linearization of the two phases is fixed. Since it can only be prosodified after root Spell-Out, the Assembly Problem, and consequently the Mismatch Problem, is solved in a straightforward fashion.

Word-level vs. utterance-level phenomena Should we assume one single phonological model, not necessarily isomorphic to other components of the grammar such as syntax and semantics, which unites both the segmental as well as intonational phenomena? From an acquisition point of view, Conxita Lleó and Javier Arias argue that children differentiate between prosodic boundaries that demarcate word edges and boundaries that form the edge of an intonation contour. With respect to primary word-stress, the authors find that in two typologically different languages, i.e. German and Spanish, children produce initial closed syllables within a trochaic foot with a longer duration than final syllables, even when the final syllables also mark the boundary of an Intonational Phrase. Moreover, Spanish children often assign more weight to the primary stressed syllable by adding an extra coda consonant, even though Spanish is a Quantity Insensitive stress system. From the observation that children tend to lengthen the final vowel within an Intonational Phrase, Lleó and Arias infer that from an early age onwards, the right-hand boundaries of larger prosodic domains are marked by vocalic length. They thus indirectly suggest that the acquisition of phonology involves a phonological component which distinguishes between the Prosodic Word, whose margins can be manipulated by adding segmental material, and higher prosodic constituents, whose margins are indicated by means of a longer duration, and that this phonological component is independent of a grammatical one, i.e. one which includes syntax and semantics. The question whether or not segmental as well as intonational phenomena should be united into one phonological model opens up new research venues on the intonational structure of individual languages whose segmental systems have so far been relatively well studied but the links to intonational structure and its interaction with syntax and semantics are crucially missing.Aslı Göksel, Meltem Kelepir and Aslı-Üntak Tarhan in this volume investigate Turkish intonation with a focus on constructions that prompt for a response, and suggest that utterance-

Prosodic Phonology: An appraisal

7

level phonological partitioning is sensitive to semantic/pragmatic functions but not to syntactic information such as boundaries. In a similar fashion, Kristján Árnason provides a fairly detailed description and analysis of prosodic phenomena in Modern Icelandic, coming to the conclusion that syntax-phonology relation is only indirect. Árnason convincingly shows that prosodic reorganization mirrors style and pragmatic function. He furthermore makes a distinction between cohesive (sandhi) and demarcative (boundary signals) phonologies, the latter of which does not demand the marking of phrase boundaries via symbols or brackets but rather the use of grid marks. On a par with this view, Harry van der Hulst proposes a two-way partitioning of phonology into phonotactic structure and utterance structure. The former regulates categorical and obligatory phonological phenomena, whereas the latter is used for phonological phenomena that depend on speech rate and speech style, i.e. on the intended context of use. Thus, van der Hulst argues that, e. g., primary word accent should be accounted for on the phonotactic level with reference to domains such as the Prosodic Word or Phonological Phrase, whereas rhythmic structure should be dealt with at the utterance level in terms of metrical grid structure. In this book, we aimed to include various approaches and stances towards phonological domains. Our point of departure has been to re-visit various assumptions that have long been taken for granted by researchers working on prosody and to offer a fresh look at the nature of prosodic organization and the universals and deviations in the mapping from morphosyntax to phonology. We believe that this could best be accomplished by focusing on typological diversity and bringing together researchers who deal with prosodic phonology from different perspectives and methodologies. We hope that our collection of papers mirrors this intention.

References Beckman, Mary E. and Janet Pierrehumbert 1986 Intonational structure in Japanese and English. Phonology Yearbook 3: 255–309. Booij, Geert 1996 Cliticization as prosodic integration: the case of Dutch. The Linguistic Review 13: 219–242. Chomsky, Noam 2001 Derivation by phase. In: Michael Kenstowicz (ed.), Ken Hale: A Life in Language, 1–52. Cambridge, Massachusetts: MIT Press.

8

Janet Grijzenhout and Barı¸s Kabak 2004

Beyond explanatory adequacy. In: Adriana Belletti (ed.), Structures and Beyond: The Cartography of Syntactic Structures Volume 3, 104– 131. Oxford: Oxford University Press.

Condravdi, Cleo 1990 Sandhi Rules of Greek and Prosodic Theory. In: Sharon Inkelas and Draga Zec (Eds.), The Phonology-Syntax Connection. CSLI Publications and the University of Chicago Press. De Jong, Kenneth J. 1989 Initial tones and prominence in Seoul Korean. Paper presented at the 117th meeting of the Acoustical Society of America, Syracuse [Published (1994) in Ohio State University Working Papers in Linguistics 43, 1–14.] Frota, Sónia 2000 Prosody and focus in European Portuguese. Phonological Phrasing and Intonation. New York: Garland Publishing. Gerlach, Birgit and Janet Grijzenhout 2000 Clitics from different perspectives. In: Birgit Gerlach and Janet Grijzenhout (eds), Clitics in Phonology, Morphologyand Syntax, 1–29. Amsterdam/Philadelphia: John Benjamins Publishing Company. Green, Antony D. 2000 The prosodic representation of clitics in Irish. In: Birgit Gerlach and Janet Grijzenhout (eds), Clitics in Phonology, Morphology and Syntax, 181–218. Amsterdam/Philadelphia: John Benjamins Publishing Company Hall, Tracy Alan 1999 German phonotactics and the prosodic structure of function words. In: Tracy Alan Hall and Ursula Kleinhenz, Studies on the Phonological Word, 99–131. Amsterdam/Philadelphia: John Benjamins. Hannahs, S. J. 1995a The phonological word in French. Linguistics 33, 6, 1125–1144. 1995b Prosodic Structure and French Morphophonology. Tübingen: Niemeyer Verlag. Hayes, Bruce 1984 The Phonology of Rhythm in English. Linguistic Inquiry 15.1:33–74. 1989 The prosodic hierarchy in meter. In: P. Kiparsky and G.Youmans (eds), Rhythm and Meter, 201–260. Orlando: Academic Press. Hayes, Bruce and Aditi Lahiri 1991 Bengali intonational phonology. Natural Language and Linguistic Theory 9: 47–96. Inkelas, Sharon 1990 Prosodic Constituency in the Lexicon. Outstanding Dissertations in Linguistics Series. New York: Garland.

Prosodic Phonology: An appraisal

9

Inkelas, Sharon and Draga Zec 1995 Syntax-phonology interface. In: John Goldsmith (ed.) The Handbook of Phonological Theory, 535–549 . Cambridge, Mass. and Oxford: Blackwell Publishers Ltd. Ito, Junko and R. Armin Mester 1992 Weak Layering and Word Binarity. Manuscript. Cowell College, UCSC: Linguistic Research Center. Jun, Sun-Ah 1989 The accentual pattern and prosody of the Chonnam dialect of Korean. In: Susumu Kuno (ed) Harvard Studies in Korean Linguistics III, 89– 100, Cambridge, Massachusetts: Harvard University. 1993 The Phonetics and Phonology of Korean Prosody, Ph.D dissertation, The Ohio State University. [published in 1996 by Garland Inc., New York]. 1998 The Accentual Phrase in the Korean Prosodic Hierarchy. Phonology 15: 189–226. Kabak, Barı¸s and Irene Vogel 2001 The phonological word and stress assignment in Turkish. Phonology 18, 315–360. Kaisse, Ellen 1985 Connected Speech: The Interaction of Syntax and Phonology. San Diego: Academic Press. Kanerva, Jonni M. 1989 Focus and Phrasing in Chichewa Phonology. Ph.D. dissertation. Standford University. 1990 Focus on Phonological Phrases in Chichewa. In: Sharon Inkelas and Draga Zec (Eds.), The Phonology-Syntax Connection, 145–161. CSLI Publications and the University of Chicago Press. Kenesei, Istvan and Irene Vogel 1989 Prosodic Phonology in Hungarian. Acta Linguistica Hungarica, 39 (1–4), 149–193. Kleinhenz, Ursula 1998 On Words and Phrases in Phonology: A Comparative Study with Focus on German. Ph.D. dissertation. Universität Tübingen, Germany [ZAS Papers in Linguistics, 12]. Ladd, D. Robert 1986 Intonational phrasing: the case of recursive prosodic structure. Phonology Yearbook 3: 311–340. Leeuw, Frank van der 1995 Cliticization, stress and phonological words in European Portuguese: An Optimal(ity) approach. Probus 7: 31–68. 1997 Clitics: Prosodic Studies. The Hague: Holland Academic Graphics. Liberman, Mark and Alan S. Prince

10

Janet Grijzenhout and Barı¸s Kabak

McHugh, Brian D. 1990 The phrasal cycle in Kivunko Chaga tonology. In: Sharon Inkelas and Draga Zec (Eds.), The Phonology-Syntax connection, 217–242. Chicago: University of Chicago Press. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. 2007 Foreword to the second edition. In: Marina Nespor and Irene Vogel, Prosodic Phonology with a new foreword, xiii-xxx. Berlin/New York: Mouton de Gruyter Odden, David 1987 Kimatuumbi phrasal phonology. Phonology 4: 13–36. 1990 Syntax, lexical rules, and postlexical rules in Kimatuumbi. In: Sharon Inkelas and Draga Zec (eds.), The Phonology-Syntax Connection, 259–278. Chicago: Chicago University Press. Peperkamp, Sharon 1997 Prosodic words. Ph.D. dissertation, University of Amsterdam. Pierrehumbert, Janet and Mary Beckman 1988 Japanese Tone Structure. Cambridge, Massachusetts: MIT Press. Revithiadou, Anthi 2006 Prosodic filters on syntax: An interface account of second position clitics. Lingua 116: 79–111. Selkirk, Elisabeth O. 1978 On prosodic structure and its relation to syntactic structure. Distributed by the Indiana University Linguistics Club. [Published in (1980) In Thorstein Fretheim (ed.), Nordic Prosody II, 111–140. Trondheim: TAPIR]. 1980a Prosodic domains in phonology: Sanskrit revisited. In: Mark Aronoff and M.-L. Kean (eds.) Juncture, 107–129. Saratoga: Anma Libri. 1980b The role of prosodic categories in English word stress. Linguistic Inquiry 11: 563–605. 1984 Phonology and Syntax: The Relation between Sound and Structure. Cambridge, Massachusetts: MIT Press. 1995 The Prosodic structure of function words. In: Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), University of Massachusetts Occasional Papers in Linguistics 18: Papers on Optimality Theory, 439–469. University of Massachusetts, Amherst: Graduate Linguistic Student Association. 2000 The interaction of constraints on prosodic phrasing. In: Merle Horne (ed.) Prosody: Theory and Experiment. Dordrecht: Kluwer.

Prosodic Phonology: An appraisal

11

Truckenbrot, Herbert 1995 Phonological Phrases: their Relation to Syntax, Focus and Prominence. Ph.D. Dissertation. Massachusetts Institute of Technology. 2002 Variation in p-phrasing in Bengali. In Pica, Pierre and Johan Rooryck (eds.), Linguistic Variation Yearbook, Volume 2, 259–303. 2007 The syntax-phonology interface. In: Paul de Lacy, The Cambridge Handbook of Phonology, 435–456. Cambridge: Cambridge University Press. Vaux, Bert 1998 The Phonology of Armenian. Oxford: Clarendon Press. Vigário, Marina 2003 The Prosodic word in Portuguese. Berlin/New York: Mouton de Gruyter. Vogel, Irene 1988 Prosodic constituents in Hungarian. In: P. M. Bertinetto and M. Loporcaro (eds.), Certamen Phonologicum. Papers from the 1987 Cortona Phonology Meeting. Torino: Rosenberg and Sellier, 231–250. 1989 The Clitic Group in prosodic phonology. In: Joan Mascaro and Marina Nespor (eds.) Grammar in Progress: GLOW Essays for Henk van Riemsdijk, 447–454. Dordrecht: Foris. 1999 Subminimal constituents in prosodic phonology. In: Stephen J. Hannahs and Mike Davenport (eds.), Issues in Phonological Structure, 251–269. Amsterdam/Philadelphia: John Benjamins Wiese, Richard 1996 The Phonology of German. Oxford: Clarendon Press.

Part 1. Prosodic Hierarchy and the Nature of Prosodic Constituents

The status of the Clitic Group Irene Vogel

1. Introduction Of the prosodic constituents proposed in Nespor and Vogel (1986), the Clitic Group (CG) is the one that has most frequently been challenged, or more accurately, omitted from the prosodic hierarchy in subsequent research. This is not to say that the types of phenomena that the CG was intended to account for have been neglected. To the contrary, they have received a great deal of attention in the research on prosodic domains. In this paper, the main objections to the CG are discussed, and it will be shown that some aspects of the prosodic hierarchy clustering around the CG and the Phonological Word (PW) are indeed problematic. It will be demonstrated, however, that the problem is not the CG itself but rather a component of the geometry of prosodic structure resulting from the Strict Layer Hypothesis (SLH). A common reaction to the problems in question has been to exclude the CG from the prosodic hierarchy and to introduce recursivity at the level of the PW, an approach that will be shown to have a number of undesirable consequences (e.g. the loss of the relatively flat structure that distinguished the phonological and morpho-syntactic components of grammar, and the loss of the concept of constituent as a linguistic string defined on the basis of a particular property or properties). I will propose instead that the problems introduced by the SLH are more adequately and appropriately resolved by the opposite approach: Retention of the CG as a constituent of the prosodic hierarchy, and the continued exclusion of recursivity. In this way, we maintain both the fundamental distinction between phonological and morpho-syntactic structures and the basic concept of constituent as a string with unambiguously defined properties. Following a brief review of some basic premises and assumptions about language and linguistic theory that are relevant for the subsequent discussion, the challenges to the CG and the SLH are examined. Alternatives are discussed, and arguments for retaining the CG in the prosodic hierarchy are presented.

16

Irene Vogel

These are based on data from several languages as well as general theoretical issues regarding the definition of the linguistic constituent, the proliferation of constituents and the implications of the absence of evidence for a given constituent. Finally, several psycholinguistic experiments are examined. It will be shown that while they were originally presented as evidence for the PW as the crucial unit of linguistic processing, the CG offers a more satisfactory account of the findings. General conclusions follow. 2. Basic premises and assumptions It is generally accepted, at least in generative linguistics, that a fundamental goal of linguistic theory is to provide an account of what is and what is not a possible natural language. It is thus desirable that a theory be as restrictive, not as permissive, as possible. Such a theory, furthermore, must be testable and falsifiable. Should a particular hypothesis, or component of a theory, be disconfirmed, one might reject the entire theory; however, more typically the problematic hypothesis is eliminated, or at least modified or refined, in order to account for new findings. Prosodic phonology, as developed in Nespor and Vogel (1986) provides a theory of the interface of phonology with other components of grammar. Two aspects of the model in Nespor and Vogel (1986) that contribute to its restrictiveness are the claims that a) the proposed set of prosodic constituents, listed in (1), is universal, and b) the geometry of the prosodic hierarchy is constrained by the SLH. (1)

Prosodic Constituents (Nespor and Vogel 1986) Syllable < Foot < Phonological Word < Clitic Group < Phonological Phrase < Phonological Utterance1

Crucial to the theory is thus the concept of constituent: A linguistic string that can be consistently identified on the basis of particular properties it shares with other such constituents, regardless of their internal structure. With regard to testability and falsifiability, caution must be exercised especially with regard to claims about the inventory of prosodic constituents. Certain phonological phenomena are easily observable; however, others may be more subtle and thus missed without extensive examination of a language. Furthermore, analysis of the larger prosodic constituents requires a thorough 1

It should be noted that the mora was not widely studied until after Nespor and Vogel (1986), and thus is not included in the original prosodic hierarchy.

The status of the Clitic Group

17

understanding of the syntax of a language and the possibility of constructing and testing various and complex types of sentences. Consequently, the absence of evidence of a constituent does not necessarily mean that it is not present in a language; it might simply be that the evidence has not yet been found. Even if no phenomena apply in relation to a particular prosodic constituent, the role of this constituent may be observable indirectly. For example, it might be the case that grouping the constituents of the lower level directly to the higher level – skipping the one in question – results in incorrect predictions about the phenomena at the higher level. Should even this type of evidence be absent, we must consider the more general question of the universality of the prosodic hierarchy. If it is assumed to be part of universal grammar, the lack of overt manifestation of a constituent in a language does not in itself mean that the constituent is not present – just as the lack of overt tense marking in a language such as Chinese does not mean that tense is not present in the language. Caution must be exercised in the opposite direction as well, and any addition of constituents to the prosodic hierarchy must be made within the context of the general theory. If the constituent structure is universal, new constituents must be components of all languages; otherwise, we no longer have a testable theory of the prosodic hierarchy. That is, if languages are able to include any type and number of constituents, we can no longer make claims about what is, or is not, a possible natural language. Thus, before introducing additional constituents, it is important to first exhaust other possibilities. For example, the phenomena on which the additional constituents are based may represent relatively isolated patterns, applying only in relation to specific morphemes or constructions. In this case, it would seem more insightful to treat the phenomenon as an idiosyncrasy or exception rather than making claims that affect the core of the theory.

3. Challenges to the Clitic Group The Clitic Group (CG) was originally proposed by Hayes (1989) as a constituent located between the Phonological Word (PW) and the Phonological Phrase. Additional arguments were provided in Nespor and Vogel (1986), where the CG is defined essentially as follows: (2)

Clitic Group (cf. Nespor and Vogel 1986) The domain of the CG consists of a PW plus a) a DCL (directional clitic), or b) a CL (plain clitic) such that there is no possible host with which it shares more category memberships.

18

Irene Vogel

Despite the arguments for the CG, objections were raised for two main reasons, a) the frequent overlap of the CG with the PW, and b) the over-assignment of PW status to morphemes such as affixes and clitics, as illustrated in (3a) and (3b), respectively. (3)

a.

b.

Overlap of CG and PW structures CG CG CG CG CG | | | | | PW PW CG CG CG Foreign green ideas sleep fast.2 Over-assignment of PW status CG PW PW PW PW te lo ri seleziona ‘(he) re-selects it for you’

It should be noted that the overlap between two constituents in some instances, as in (3a), does not necessarily mean that there will be overlap in all instances. Even if it turned out that some language always exhibited such overlap, it might still be desirable to retain the CG if the prosodic hierarchy is considered part of UG. The structure in (3b) is more problematic, and indeed appears to be incorrect. In this Italian sentence, the only reason the two clitics (te, lo) and the prefix (ri-) labeled as PWs is because the SLH, in particular the component known as “Strict Dominance”, only permitted a constituent to dominate constituents of the immediately lower level in the prosodic hierarchy. Thus, while structures such as (4a) were considered well-formed, those in (4b) were not. The consequence was that many items that would not otherwise seem to be PWs were nevertheless labeled as such. (4)

a. Well-formed structure Cn Cn−1 Cn−2

2

Cn−1 Cn−2

b. Ill-formed structures i. *Cn ii. * Cn iii. Cn | | | Cn+1 Cn−2 Cn

Cn−2

The original version of this sentence is not used since the words colorless and furiously would have been analyzed as consisting of two PWs each, and thus not illustrate complete overlap of the CG and PW.

The status of the Clitic Group

19

A number of researchers have proposed to resolve the problem with the SLH by modifying the strict dominance component, rather than rejecting the SLH altogether. One widely proposed modification was the introduction of structures such as (4biii) in which a constituent is permitted to dominate constituents more than one level lower in the prosodic hierarchy (cf. among others Ito and Mester 1992; Selkirk 1996; Vogel 1994, 1999, 2008a,b). Another modification was the introduction of recursive structures. Both (4bi) and (4bii) represent recursive structures, however, only the first type, where a constituent dominates another constituent of the same level, was introduced as an additional option for prosodic tree structures (cf. among others, Ito and Mester 1992; Selkirk 1996; Booij 1996; Peperkamp 1997; Hall 1999; Anderson 2005). With the modifications in (4bi) and (4biii), the sentence in (3b) would have a structure such as (5a) or (5b); “ σ ” is a syllable. (5)

a.

PW’

σ te

σ lo

σ ri

b.

PW’

PW seleziona

PW’ PW’

σ te

σ lo

σ ri

PW’ seleziona

When recursive PW structures are introduced, the basic constituent is labeled PW, while the larger ones that include certain affixes and clitics are often labeled as PW’, as shown. In many such analyses, furthermore, the CG is eliminated from the prosodic hierarchy. 4. How many constituents do we need? Since analyses with recursive PWs require a distinction between the innermost PW and those dominating it, we must consider the implications of this approach. Given that the notion of constituent requires the existence of a set of identifying properties, a problem arises if distinct sets of properties are associated with different PWs. 4.1. Defining prosodic constituents In syntax, an NP has certain properties regardless of its internal composition, so both [Mary]NP and [the particularly cheerful girl with glasses]NP represent the same type of constituent. By the same token, in phonology we can consider

20

Irene Vogel

both the Italian words [te]PW ‘tea’ and [regol-ar-izza-zione ]PW ‘regularization’ to represent the same type of constituent, the PW, despite the differences in their internal structure. Identifying the defining properties of a constituent, however, may not be simple. With respect to prosodic constituents, these properties are phenomena such as phonological rules and phonotactic constraints associated with a particular type of string.3 In addition, since the prosodic constituents at the PW and higher levels involve interfaces with other components of grammar, the mapping relations with these components also serve as a means of identifying prosodic constituents. There are, however, several potential “risks” involved in establishing constituents: a) circularity, b) proliferation of domains, and c) lack of overt manifestation. The risk of circularity arises since the application of a phonological phenomenon is often used as the means of delimiting the string that constitutes the domain of the phenomenon. This, by itself, would indeed be circular. In practice, however, this problem is generally overcome by establishing independent evidence for the constituent in question, in particular, other phenomena that apply in precisely the same domain. The second risk, the proliferation of domains, can arise as a result of the circularity problem. If we establish a prosodic domain each time a different phonological pattern is observed, in the worst case, a language would have as many constituents as phonological phenomena. This problem is reminiscent of earlier SPE-type analyses, where large numbers of different types of boundaries were introduced to account for the application of various phonological phenomena (e.g. Dakota: 11 [Carter 1974], 4 [Shaw 1980]; Danish: 5 [Basbøll 1975, 1981]; Italian: 13 [Bertinetto 1999; Loporcaro 1999]). A similar proliferation is also observed in the number of levels in lexical phonology in some analyses (e.g. Dakota: 4 [Shaw 1985; Kaisse and Shaw 1985], 11 [Patterson 1990]). In general, however, the proliferation of prosodic domains is avoided in the same way circularity is avoided. In a number of cases, however, it has been argued that an additional constituent is needed. Downing (1999), for example, proposes the addition of the “Prosodic Stem” as a constituent in Bantu languages. Furthermore, those analyses that include both the PW and PW’, in effect introduce another constituent in various other languages (cf. among others Selkirk 1996; Peperkamp 1997; Booij 1996; Hall 1999; Anderson 2005). While the addition of a single constituent cannot be considered “proliferation”, 3

A rule-based approach is adopted in this paper, however, the generalizations and proposals advanced here have the potential to be recast within an OT framework, since it, too, relies on the fundamental assumptions about linguistic theory and universals of grammar mentioned above.

The status of the Clitic Group

21

other analyses introduce larger numbers of distinct PW constituents (i.e. PW 1 , PW2 . . . , PWn ), for example: 4 in Limbu, 11+ in Lhasa, Burmese, Kham (Bickel and Hildebrandt 2007; Schiering et al. 2007). The large number of constituents arises from the procedure of assigning a new constituent (i.e. PW type) each time a phonological phenomenon is observed that appears to behave differently from another one – the circularity problem. The result is a constellation of observations that no longer qualifies as an appropriate linguistic theory according to the criteria discussed earlier. That is, if any number of constituents is possible, and these may be defined on an ad hoc basis – as opposed to a mapping from other constituents of grammar, we are not able to test or falsify the resulting model of phonology. It is not possible to establish what is, or is not, a possible language, since the possibilities are not restricted in a principled way. Furthermore, we can no longer claim that the prosodic hierarchy is a part of universal grammar, since each language arranges phonological phenomena in its own way.4 The occurrence in a language of a large number of phonological phenomena with apparently different domains of application nevertheless raises an interesting issue. As mentioned, it may be difficult to identify the domain of certain phonological phenomena, so it is possible that at least in some cases additional information could eventually clarify the situation. It is also possible that the identification of certain prosodic constituents may have been based on incomplete or inaccurate analyses of the linguistic components on which they are based. It is unlikely, however, that such considerations will account for all of the phonological phenomena that have led to the proliferation of boundaries or constituents. Instead, it seems that many cases may in reality reflect idiosyncratic phenomena or exceptions, which would be most appropriately treated as such.5 In fact, admitting the possibility of exceptions has several advantages. It allows us to distinguish between phenomena that represent general patterns and those that do not. We are thus able to maintain a model of phonology that fits the criteria of a well-articulated theory, and thus has the potential of being part of UG. Furthermore, since all languages have exceptions of various sorts, such a view is not inconsistent with what is known about language in general. In fact, it is proposed below that certain exceptions are actually characteristic of two constituents of the prosodic hierarchy, the PW and the CG.

4

5

In fact, Bickel’s typological approach does not claim to offer a theory of universal grammar, and indeed would appear to be incompatible with such a theory, as well as with the usual concept of theory as a consisting of a set of testable hypotheses. See Simon and Wiese to appear for detailed discussion of exceptions.

22

Irene Vogel

Finally, let us consider the third risk encountered in identifying prosodic constituents, the possibility that there may be no evidence of a particular constituent. Indeed, Schiering et al. (2007) have proposed that Vietnamese does not have a PW constituent. This is the opposite of the problems we have just considered, however, the same cautions apply. That is, before concluding that a particular constituent is not present in a language, one must be sure that potentially subtle phenomena have not been overlooked, and that the analysis of the morpho-syntax of the language is accurate since this forms the basis for the establishment of prosodic constituents. A case in point is stress in French. It is often claimed that French does not have word level stress, but rather assigns stress to larger strings. Indeed, this might appear to be correct unless particular types of constructions are examined. For example, let us compare the pronunciation of the two names Marie and MarieClaude. In the first, stress is on the final syllable of Marie, but this could be attributed to fact that when a word is spoken in isolation its last syllable is also phrase final.6 In M arie-Claude, however, there is a perception of greater prominence on the first syllable of Marie. This is similar to the Rhythm Rule of English, and other languages. Thus, just as we observe thirtéen mén → thí rteen mén, we can observe Maríe-Cláude → M árie-Cláude.7 If there were no word level stress in French, there would be nothing to constitute a clash in such cases, and also nothing to exhibit a Rhythm Rule phenomenon. It is, however, easy to miss this type of evidence since it can only be observed in a limited set of constructions and since the identification of stress is often quite challenging. The general problem still remains. How do we know when we have considered all the necessary data? This is essentially the philosophical problem of – having seen numerous white swans – how we know if the next swan we see will be white and not black. While there is no real answer to this type of question, we may still come to some conclusions about prosodic structure based on general principles of linguistic theory. 4.2. The PW and PW’ In analyses that eliminate the CG, but include recursive PWs, the question that arises is whether the resulting prosodic hierarchy has actually been reduced by one constituent. Peperkamp’s (1997) analysis of Italian dialects provides an interesting set of data that allows us to systematically consider this question.

6

Throughout the paper, when a syllable is stressed, only the vowel will be indicated in bold. 7 Cf. among others Hoskins (1994); Mazzola (1992, 1993).

The status of the Clitic Group

23

This analysis adopts the modification of the SLH that permits a constituent to dominate items more than one level lower in the prosodic hierarchy. In addition, Peperkamp crucially makes use of recursive PWs. The way in which clitics are grouped with their host into a PW may vary across dialects, as in (6a) and (6b) for the Lucanian and Neapolitan dialects, and (6c) for standard Italian. Peperkamp motivates the different structures on the basis of stress phenomena, which involve clitics in different ways in the three varieties of Italian. (6)

Recursive PWs in Italian and Dialects (Peperkamp 1997) a. Lucanian PW

b. Neapolitan PW PW

host (CL) (CL) vínn´ ‘sell’ v´nní ll´ ‘sell it’ vinn´ mí ll´ ‘sell me it’

host (CL) (CL) cónt´ ‘tell’ cónt´ l´ ‘tell it’ cónt´ tí ´ ‘tell yourself it’

c. Standard Italian (Peperkamp 1997) PPh PW host véndi véndi véndi

(CL) lo me

(CL)

lo

‘sell’ ‘sell it’ ‘sell me it’

In both Lucanian and Neapolitan, stress shifts rightward from the verb form in the presence of clitics. In Lucanian, stress always shifts to the penultimate syllable of an encliticized string regardless of the number of syllables. According to Peperkamp (1997), this means that enclitics are part of the PW with the host. By contrast, in Neapolitan a single clitic does not affect the stress pattern; however, if two clitics are present, an additional stress appears on the first one. In this case, it is argued that a recursive PW structure is needed – precisely to account for a phonological behavior that is not found in either the basic PW or the Phonological Phrase (PPh). Finally, in standard Italian the stress remains fixed regardless of the presence of clitics. This is structurally represented by grouping the clitics directly into the PPh with the host. At first glance, this

24

Irene Vogel

three-way distinction in PW structures appears to account for the facts in a simple, straightforward manner. Closer examination, however, reveals several problems with such an analysis.8 First, while it is true that clitics behave differently with regard to stress in the three varieties of Italian under consideration, by introducing three different PW structures, we miss the fact that all three varieties actually share a significant generalization with regard to stress assignment. In standard Italian, as well as Lucanian and Neapolitan, and indeed numerous other dialects, primary word stress falls on one of the last three syllables of a word, the so-called “Three Syllable Window” (TSW), as illustrated in (7) (cf. among others Cardinaletti and Repetti 2008; Peperkamp 1997). (7)

Lucanian rá ‘give’ [vín:´] ‘sell’ [jén:´r´] ‘brother-in-law’

Neapolitan fá ‘do’ cónta ‘tell’ péttina ‘comb’

Standard Italian fá ‘do’ véndi ‘sell’ péttina ‘comb’

The mechanism for this type of stress assignment involves a combination of phonological and morphological considerations. Its domain consists of a root plus suffixes, and possibly some prefixes, the usual PW for Italian and various dialects (cf. among others Nespor and Vogel 1986; Peperkamp 1997). Such a domain is, in fact, present in the structures in (6), the “host”, although the definition and status of this host are not specified. In standard Italian and Neapolitan, the host coincides with the innermost PW, but in Lucanian it does not. Furthermore, if we take the PW as the domain for the TSW, we would incorrectly expect all PWs to exhibit this property. For example, in Neapolitan, while stress is penultimate when two clitics are present, when only one clitic is present the TSW is regularly violated in certain verb forms (e.g. péttinal´ ‘comb them’) (Peperkamp, 1997: 180).9 Moreover, if it is claimed that stress is penultimate in the PW in Lucanian and Neapolitan, we would expect all PWs to exhibit this property. This is only the case, however, for

8

9

It should be noted that the types of problems discussed here regarding the inclusion of clitics within PWs arise in other analyses that treat clitics similarly. Peperkamp’s (1997) analysis is examined in detail here since it is particularly systematic and clear. In standard Italian there are two forms in which stress appears on the pre-antepenultimate syllable of a small set of verbs, the third person plural of the present indicative and present subjunctive (e.g. péttinano ‘(they) comb’, péttinino ‘(that they) comb’). It should be noted, however, that such forms are rather rare, and have nothing to do with the presence or absence of clitics.

The status of the Clitic Group

25

PWs that include any number of clitics in Lucanian, and for PWs that include two clitics in Neapolitan. Thus, introducing the recursive PW fundamentally compromises the concept of linguistic constituent, defined as a type of linguistic string that is identified by certain properties and assumed to exhibit the same properties, regardless of its size or internal structure. The properties of the inner PW or host are different from those of the outer PW (or PW’) in Lucanian and Neapolitan, with the result that it is not possible to identify a PW constituent on the basis of a specific set of properties. If, however, we wish to maintain the concept of constituent, and we consistently observe two distinct sets of identifiable phonological properties, it follows that we must have two constituents. In a sense, this is achieved by referring to the type or level of PW, in effect a diacritic that accounts for the observed differences in behaviors in different strings, although it is obscured by the fact that “PW” appears in the labeling of the different structures. 4.3. The nature of the PW domain It was suggested above that caution must be exercised in identifying constituents to avoid circularity and a potential proliferation of constituent types. We must thus ask whether the conclusion just reached – that two constituents are involved in the Italian stress patterns suffers from such weaknesses, or whether it can be independently motivated. It was seen that the domain of the TSW and the pattern of stress assignment that combines phonological and morphological considerations are the same across dialects. Indeed, this coincides with the PW mapped from morphosyntactic structure onto phonological structure, following Nespor and Vogel (1986), as in (8).10 (8)

Phonological Word Domain (cf. Nespor and Vogel 1986) a. The PW domain consists of i) a stem, ii) any element identified by specific phonological and/or morphological criteria, or iii) any element marked with a diacritic (e.g. [+W]). b. Any unattached elements within the morpho-syntactic word form part of the adjacent PW containing the stem (or other item specified in (a)).

10

The obligatory formation of a PW for certain affixes and clitics in Nespor and Vogel (1986) due to the SLH is omitted here.

26

Irene Vogel

This definition explicitly permits the PW constituent to include affixes, but not clitics, accounting for the presence of the same domain for the TSW across Italian dialects. The same PW delimits the domain of other phonological phenomenon such as Intervocalic s-Voicing (ISV), in northern Italian. Roughly formulated as /s/ → [z] / V V, ISV applies within a single morpheme, as well as across the juncture with a suffix, as in (9). By contrast, it does not apply between a stem and clitics or prefixes, which are usually excluded from the PW, or at the juncture of two clitics, as in (10) (cf. among others Nespor and Vogel 1986). (9)

Intervocalic s-Voicing (applies) a. [muzeo]PW ‘museum’ ‘(he) was causing’ b. [cauz-ava]PW c. [studi-oz-issimo]PW ‘extremely studious’

(10)

Intervocalic s-Voicing (does not apply) a. ri- [saluta]PW ‘(he) re-greets’ ‘greeting each other’ b. [salutando]PW si c. ci si [saluta]PW ‘(they/one) greet each other’

Finally, the same PW also appropriately excludes the application of other phenomena that apply specifically in relation to clitics. For example, a rule of Clitic Vowel Change (cf. Vogel 2008b) causes /i/ to become [e] in certain sequences of clitics, as in (11a); it does not apply within the PW, as in (11b). (11)

a. b.

mi la [da]PW → me la [da]PW → *[mela]PW [mila]PW

‘(he) gives it to me’ ‘thousands’

There is thus a series of facts of Italian supporting the identification of a PW that excludes clitics. By the same token, there are other phenomena that do involve clitics, and it is necessary to provide the appropriate domain for these phenomena.

5. The Clitic Group as a prosodic constituent (part 1) 5.1. Tree geometry Given the need for two distinct phonological constituents, one that excludes clitics (and some affixes) and one that includes them, the question that arises is to what extent the recursive PW analysis is similar to an analysis that refers to the CG, since both the PW’ (or outermost PW) and the CG are advanced as the

The status of the Clitic Group

27

domain for phonological phenomena involving clitics. Is the difference merely one of labeling, or is it more substantive? An obvious first answer to the question of whether the PW’ and CG analyses are essentially equivalent is “no” – while the former involves recursivity in the prosodic hierarchy, the latter does not. One of the original distinctions between syntax and phonology was the assumption that recursivity is a property of the former, but not the latter (cf. Jackendoff and Pinker 2005; Pinker and Jackendoff 2005; Neeleman and van de Koot 2006 for more recent discussion of this issue), so an analysis that introduces recursivity in the phonological component is crucially different from one that does not.11 Thus, in comparison with potentially unlimited depth in a prosodic tree of the type seen in (5b), represented with brackets in (12b), the CG provides a flatter structure, as in (12a). It should be noted that this structure is more similar to that in (5a), shown again in (12c). (12)

a. c.

[. . . PW. . . ]CG [σ σ σ [PW] ]PW’

b.

[σ [σ [σ [PW] ]PW’]PW’]PW’

Since the structures in (12a) and (12c) are quite similar, we must ask whether introducing only one level of recursive structure produces the same results as the CG, aside from the general considerations regarding recursivity. The answer is again “no”. By considering both the innermost PW and the PW’ to be the same type of constituent, we predict that the same phonological phenomena will be observed in all cases of this (PW) constituent. It was seen in relation to several phenomena of Italian that this prediction is incorrect. Another type of phenomenon allows us to distinguish further between the recursive PW and CG analyses. In the case of the recursive PW, we predict that the left (or right) edges of all PWs will exhibit the same phonological properties. A phonotactic constraint of Italian shows that this is not the case. In Italian, the palatal lateral [λ ], orthographically “gl”, is found within words (e.g. figlio ‘son’), but not at the beginning of lexical items. Thus, while we find the word libro ‘book’ in Italian, we do not find *glibro. Since the beginning of a lexical item typically coincides with the beginning of a PW, we can formulate the constraint in (13). 11

It should be noted that Ladd (1996) and others have argued that recursive structures are needed in order to account for intonation phenomena. It is not clear, however, that the intonational constituents in these analyses are part of the same prosodic hierarchy presented in Nespor and Vogel (1986), and assumed here, where the prosodic constituents are mapped from the morpho-syntax, and serve as the domain for specific phonological rules, as opposed to different types of intonational contours (cf. Vogel 2008b).

28

Irene Vogel

(13)

Italian Phonotactic Constraint: *[λ ] / [PW

...]

The lateral [λ ], however, does occur at the beginning of clitics (e.g. gli ‘to him’),12 so a recursive structure as in (14) would predict the applicability of the phonotactic constraint to both instances of the left edge of PW. Instead, [ λ ] is possible in the first instance but is excluded in the second. (14)

[gli [leggevo]PW ]PW

‘(I) was reading to him’

By contrast, a structure in which a CG constituent dominates a PW with its associated clitics, as in (15), provides a straightforward account of the phonotactic constraint. The exclusion of [λ ] at the left edge of the PW does not apply to gli since it is not at the left edge of a PW. (15)

[gli [leggevo]PW ]CG

‘(I) was reading to him’

While it may be possible to state the constraint so that it applies only to the innermost PW, the presence of the CG constituent predicts, without further qualification, that we should observe different phonological behaviors at the left edge of the PW and the left edge of the CG.13 It should be noted that both recursive PW and the CG analyses involve skipping levels in the prosodic hierarchy, following modification of the SLH. This modification is independent of the introduction of recursivity and it is, in fact, the only modification in tree geometry required by analyses involving the CG. Since any modification of a theory, or one of its components, in addition to fixing the problem for which it was developed, can have far-reaching implications, we must evaluate the impact of proposed changes on the theory as a whole. In this regard, it can be seen that permitting the skipping of levels in the prosodic hierarchy does only that. By contrast, introducing recursivity not only involves skipping levels, it changes the basic structure of the prosodic hierarchy, and indeed of the phonological component of grammar more generally. If such extensive modifications are not strictly necessary, it would seem preferable to avoid them. Even with regard to the possibility of skipping levels, there are less extensive and more extensive options. If the prohibition on skipping levels is simply 12

13

[λ ] is also absent in final position, however, this is due to the avoidance of final consonants in general in Italian. It should be noted that we could formulate the constraint in (13) in terms of the grammatical word instead of the PW, as suggested by a reviewer; however, there are other instances in which the edges of phonological constituents do not behave similarly to edges of grammatical constituents.

The status of the Clitic Group

29

removed, it would be possible for all constituents of the prosodic hierarchy to dominate material more than one level lower, and furthermore to dominate material at any number of levels lower. This would introduce a wide variety of additional prosodic tree structures, which would then be expected to exhibit different phonological properties. This does not, however, seem to be the case. For example, at first glance, it might appear that the auxiliary “s” in English attaches to a PW in (16a), but to larger constituents, the CG, PPh and even Intonational Phrase, in (16b)-(16d), respectively. (16)

a. b. c. d.

Dean’s talking. The dean’s talking. The former dean’s talking. The person standing next to the former dean’s talking.

In all of these items, however, the “s” behaves in the same way: It is pronounced as [z] since it is preceded by a voiced consonant, despite the fact that it is followed by a voiceless one. The possessive “s” exhibits the same phonological behavior as well, and may appear in the same types of constructions as the auxiliary “s”, as can be observed by considering talking to be a noun in the examples in (16). The phonological behavior of the auxiliary and the possessive “s” is also observed with the plural “s” (e.g. The deans talked). Given that the plural may follow “level 2” affixes (e.g. The danc#ers talked), which are excluded from the PW, we may also exclude the plural from the PW, grouping it instead under the CG as a sister of the PW. Since all three types of “s” exhibit the same phonological behavior, there would be no motivation for grouping the plural into a CG and other “s” segments into various other constituents. A similar argument can be made regarding well-known “ordering paradoxes”. For example, while the plural marker in an item such as scare crows must morphologically attach to the entire compound, it is only the vowel at the end of crow that determines the voicing of the “s”. In cases such as transformational grammarian, where the suffix –ian is attached to a phrase (or compound), the phonological interaction is again only with the element to which it is directly adjacent. In this case, the suffix is part of the PW since it is a stress-shifting, or “level 1” affix. There would be no motivation for phonologically associating the suffix with a larger constituent, for example the PPh, when it attaches to transformational grammar, while it forms part of the PW when it attaches to grammar. In the absence of differences in phonological behavior between material attached at levels at or below the CG and those that might at first appear to be attached at a higher level, it seems unnecessary to extend the possibility of skipping levels beyond the CG. By thus restricting the constituents that permit

30

Irene Vogel

skipping levels, we furthermore restrict the number of levels that may be skipped to a maximum of two. If the CG directly dominates syllables, as seen above, two levels are skipped: The PW and the Foot.14 It is also possible to skip only one level, for example the PW, when a CG dominates a function word that constitutes a Foot but not a PW (e.g. in Italian [[[u] σ [na]σ ]F [[se.ra]]F/PW ]CG ‘an evening’).15 We may also skip one level within a PW in the case of an extrametrical syllable that is not included in a Foot (e.g. [[so] σ [[rel.la] ]F ]PW ‘sister’).16 It appears therefore that no additional skipping of levels is necessary. 5.2. Italian stress revisited It has been argued thus far that the CG offers several general advantages with regard to the geometry of the prosodic hierarchy and the nature of the interface between the phonological component and other components of grammar. We must not, of course, compromise the ability to account for specific phonological phenomena in favor of such general considerations. In this section, the stress phenomena of Italian are re-examined, and it is shown that the CG, in fact, offers a simple and insightful analysis of the data. First, we retain the PW as the domain of only word level stress in Italian, characterized by the “Three Syllable Window”, as well as for segmental rules such as Intervocalic s-Voicing in northern Italian. Clitics, which are excluded from the PW, are instead grouped with PWs into CGs. This permits us to posit identical structures for the Lucanian and Neapolitan dialects as well as standard Italian, as shown in (17) and (18), respectively, where brackets are used instead of tree structures. (17)

14

a. Lucanian [ [ PW ] (CL) (CL) ]CG vínn´ ‘sell’ v´nní ll´ ‘sell it’ vinn´ mí ll´ ‘sell me it’

b. Neapolitan [ [ PW ] (CL) (CL) ]CG cónt´ ‘tell’ cónt´ l´ ‘tell it’ cónt? tí ll´ ‘tell yourself it’

It should be noted that the foot level was not indicated in the examples above since this was not relevant to the points under investigation. 15 It can be seen that una ‘a’ does not behave as a PW containing a lexical item since it undergoes a rule that deletes the final –a not observed with PWs (e.g. una isola → un isola ‘an island’ vs. (la) nona isola → *(la) non isola ‘(the) ninth island’). 16 Where the syllables are not relevant, the brackets have not been shown.

The status of the Clitic Group

(18)

31

Standard Italian [ [ PW ] (CL) (CL) ]CG véndi ‘sell’ véndi lo ‘sell it’ véndi me lo ‘sell me it’

Contrary to Peperkamp’s (1997) analysis in which the varieties of Italian are distinguished in terms of their basic prosodic structures, the present analysis unifies the treatment of the different varieties. It explicitly represents the same constituent for the TSW across dialects, which captures the generalization that this phenomenon is indeed characteristic of most, if not all, varieties of Italian. Furthermore, the present analysis only requires one arrangement of clitics with relation to the associated PW, again, providing a unifying property across Italian dialects in their basic structures. Of course, different varieties of Italian show different stress patterns in the presence of clitics, so we must now determine whether an analysis in which the basic prosodic structures are identical can account for the observed differences. In fact, this can be done in a simple and straightforward manner by attributing the differences to the stress assignment rules themselves, as opposed to the prosodic structures. Once the PW stress rules have applied, all that is required is an appropriate CG stress rule for each variety of Italian, as shown in (19). 17 It should be noted that CG stress is not distinct from PW stress in standard Italian, so if a CG stress rules is present in the language, all it does is re-assign prominence to the position of PW stress. (19)

a. b. c.

Lucanian: Neapolitan: Standard Italian:

σ → [+stress] / syll ]CG σ → [+stress] / ]PW syll ]CG σ → [+stress] / . . . ]PW . . . ]CG [+str]

This approach is consistent with the theoretical preference for introducing only those modifications or distinctions that are crucially required. In this case, by introducing relatively small differences in the CG stress rules, we avoid more extensive and fundamental differences with regard to the constituent structures. Thus, the CG analysis has general theoretical advantages as well as empirical advantages. It does not require the introduction of recursivity into phonology, permitting us to retain this fundamental difference between the components of grammar. It also permits us to retain the core concept of constituent as a string 17

The rules are stated informally to show how they operate; no claims are made about their appropriate formulation.

32

Irene Vogel

defined on the basis of a consistent set of properties. With regard to the facts of Italian and its dialects, the CG analysis also allows us to explicitly represent the domain for the TSW, and the fact that it is the same across dialects. Dialects only differ in the assignment of stress in the presence of clitics, and this directly expressed by the differences in the CG stress rules themselves. 6. The Clitic Group as a prosodic constituent (part 2) In this section, the CG analysis is extended, and further support is provided for it. Specifically, its role in accounting for idiosyncrasies and compounds is addressed, and it is shown that the proposal can easily be extended to an unrelated language, Fijian. 6.1. CG as a domain of idiosyncrasy It is well-known that many (lexical) phonological phenomena apply in restricted or very specific contexts. For example, a word such as divide in English exhibits a variety of behaviors when different suffixes are added, as shown in (20). (20)

divide →

a. b. c.

[ˇz] (e.g. divi[ˇz]-ion) [z] (e.g. divi[z]-ible) [s] (e.g. divi[s]-ive)

The differences in the treatment of [d] are not predictable on strictly phonological grounds (cf. recede → rece[ˇs]-ion, not *rece[ zˇ ]-ion). By the same token, the behaviors are not completely idiosyncratic; the patterns in (20a, c) are also observed with certain other words (e.g. corrode → corro[ zˇ ]-ion; corro[s]-ive). Such idiosyncratic phenomena are characteristic of “+ boundary” or “level 1” phonology in earlier models of phonology, and of the PW in prosodic phonology. Thus, a certain degree of exceptionality appears to be typical of the PW constituent. Clearly some type of exception mechanism is required, such as individual diacritics or co-phonologies within the lexicon – as opposed to the proliferation of boundary symbols or other constructs, shown earlier to be disadvantageous. It should also be noted that simply excluding a constituent such as the PW from the prosodic hierarchy on the grounds that its rules do not apply “across the board” does not resolve the problem.18 An analogous situation is observed within the CG domain. As mentioned above, the Clitic Vowel Change (CVC) rule of Italian changes the /i/ of certain 18

See Simon and Wiese to appear for extensive discussion of the treatment of linguistic exceptions.

The status of the Clitic Group

33

clitics to [e]. Specifically, the /i/ of a dative pronominal clitic (mi, ti, ci, etc.) changes to [e] when immediately followed by a direct object clitic (lo, la, li, etc.), as illustrated in (21). (21)

Clitic Vowel Change (CVC) a. mi racconta la storia b. mi la racconta → me la racconta c. raccontando mi la → raccontando me la

‘(he) tells me the story’ ‘(he) tells it to me’ ‘telling it to me’

A second rule, the Clitic Affrication Rule (CAR), changes the “s” of the impersonal clitic si to “c”, phonetically the affricate [ˇc], when it is followed by the reflexive clitic si, as in (22). It should be noted that the second instance of si is then subject to the rule changing /i/ to [e]. The first instance of si, however, remains unchanged, shown in (22c). It is clear that the change is not due to the segmental sequence of si si itself, since no change takes place in si simula ‘one simulates’. (22)

Clitic Affrication Rule (CAR) a. si manda un messaggio b. si si manda un messaggio → ci si manda. . . c. si si lo manda → ci se lo manda d. lo si si manda → lo ci si manda

‘one sends a message’ ‘one sends oneself a message’ ‘one sends oneself it’ ‘one sends it to oneself’

At first glance, the CVC and CAR appear to be problematic in that they do not apply throughout the CG to all relevant segmental sequences, but only to specific elements, and only when these elements occur in specific contexts. Further consideration, however, reveals instead that this behavior provides additional insight into the nature of the prosodic hierarchy. While the CVC and CAR do not apply across the board within the CG, they nevertheless apply in relation to the CG. As in the cases of idiosyncrasy in the PW, all that is necessary is a means of representing the CG rules’ exceptionality in a principled way. For example, we could formulate relatively simple rules that explicitly include the conditions for their application, as in (23), where “C” is a consonant. (23)

a. b.

Clitic Vowel Change: Clitic Affrication Rule:

/i/ → [e] / [. . . [C ]dat [CL]acc . . . ]CG /s/ → [ˇc] / [. . . [ i]impers [si]refl . . . ]CG

Such an approach does not require numerous boundary symbols or additional prosodic constituents. Moreover, it makes use of the same types of prosodic rules that are used for other phenomena, in this case domain juncture rules (cf. (28) below).

34

Irene Vogel

Rather than representing a weakness in the phonological analysis or structure of a language, the presence of idiosyncrasies within the CG can now be viewed positively, as a property of this constituent. The CG and PW are the two prosodic constituents that involve an interface with morphology, so it follows that these constituents will be the ones to exhibit idiosyncratic phonological phenomena. Similar idiosyncrasies do not appear to be characteristic of the phonology of larger constituents involving an interface with syntax. In fact, it seems that the degree of exceptionality decreases as one moves up in the prosodic hierarchy. While words and affixes may behave differently in different contexts within the PW, the CG rules appear to apply without exception in the contexts for which they are specified. We can thus predict that if there are idiosyncrasies in larger constituents, these will be minimal and apply consistently in their specified domains. 6.2. The CG = Composite Group Beyond the issue of exceptionality, a potential challenge to the CG proposal remains with regard to the exclusion of recursivity. In addition to structures involving affixes and clitics, another type of construction has been treated with recursive PWs. That is, it has been proposed that compounds consist of multiple PWs contained within a larger PW, as in (24) (e.g. Peperkamp 1997; Wheeldon and Lahiri 1997, 2002; Vigário 2003). (24)

[ [cat]PW [food]PW ]PW

It is clear that some intermediate grouping is needed for compounds, so that the PWs are not directly grouped into the next prosodic constituent, the Phonological Phrase. This can be seen in relation to the well-known phenomena of compound and phrasal stress in English. In the former, stress falls on the first element of the string and in the latter it falls on the last element. The possibility of representing this difference is lost if the PWs of a compound, like those of a phrase, are grouped directly into a PPh, as in (25a). The question that arises again, however, is whether the appropriate structure for the compound must involve the recursive PW, as in the structure in (25b). (25)

a. [ [hot]PW [dog]PW ]PPh

b. [ [ [hot]PW [dog]PW ]PW ]PPh

The general concerns about introducing recursivity into the prosodic hierarchy discussed above hold in this case as well. Furthermore, a structure that incorporates two constituents bearing the same label (PW) but exhibiting different phonological behaviors means that we again lose the ability to specify a set of

The status of the Clitic Group

35

properties that clearly identifies the PW. That is, we observe that stress is assigned to the inner PWs on the basis of both phonological and morphological characteristics of the word, while it is assigned to the outer PW only in relation to its left edge. By contrast, the CG offers the possibility of an additional grouping of the PWs in a compound. Thus, the difference between compounds and phrases can be represented as in (26a) and (26b), respectively (cf. also Vogel and Raimy 2002). (26)

a. [ [ [hót]PW [dog]PW ]CG ]PPh

b. [ [ [ [hot]PW ]CG [ [dóg]PW ]CG ]PPh

These structures permit stress to be assigned to the PWs on the basis of the usual word stress assignment rules of English, and to the CGs in terms of their left edge. At the level of the PPh, stress continues to be assigned in relation to the right edge. A somewhat similar situation is also observed in Turkish (cf. Kabak and Vogel 2001). While word stress regularly falls on the final syllable, compound stress falls on the leftmost element, and phrasal stress applies further prominence to the leftmost element. If compounds are treated as (recursive) PWs, it would not be possible to account for the difference between compound and word stress. That is, the word stress rule would predict final stress in all PWs. Furthermore, as in English, compounds and phrases do not exhibit the same stress patterns, so if PWs are grouped directly into PPhs in both cases, we are not able to capture this difference. By first grouping the members of a compound into a CG, however, we are able to account for the difference, as seen in (27). (27)

a.

x PPh b. x CG x x PW aç-i ölç-er ‘protractor’ (compound)

x x x x x aç-i ölç-er ‘(it) measures (an) angle’ (phrase)

Thus Turkish, like English requires three stress domains, since there are three distinct stress assignment phenomena.19 At first glance, it may appear problematic to use the same constituent to account for strings consisting of PWs plus affixes and/or clitics as well as for strings consisting of compounds. Indeed, the term Clitic Group does not accurately reflect the nature of the constituent in the latter case. It has therefore 19

Alternative analyses tend to involve a number of complex mechanisms (cf. Kabak and Vogel 2001 for discussion).

36

Irene Vogel

been proposed that a more appropriate name for this constituent is Composite Group (cf. Vogel 2008a, b). Such a terminological solution fits better with the practice of naming prosodic constituents in relation to their composition, however, the substantive issue is whether the new grouping appropriately reflects the necessary phonological facts. A first advantage of using the Composite Group for constructions with affixes and clitics as well as compounds is that it does not require the introduction of an additional constituent in the prosodic hierarchy, assuming the model in Nespor and Vogel (1986). It also allows us to avoid recursivity both in the case of clitics and affixes and in the case of compounds. Crucially, it also permits us to retain the concept of linguistic constituent, and thus to capture differences in phonological phenomena that apply in different types of structures; no changes to the theory are required with regard to the definition of constituents. Finally, there is an additional advantage of including compounds in the CG along with constructions involving affixes and clitics. It was proposed above that the CG is a constituent that permits some degree of idiosyncrasy due to its interface with morphology. The cases of idiosyncrasy seen thus far have involved affixes and clitics. Interestingly, compounds may also exhibit some idiosyncrasy. For example, if we assume that names (of places and people) are compounds, while the Compound Stress Rule assigns stress to the first element of a compound, there are certain cases in which this is not observed. For example, stress falls on the first word in Fí fth Street, but on the last word in Fifth Ávenue, as in most other names, although there is no phonological or morphological reason for this difference. If it is assumed that the CG permits some degree of idiosyncrasy however, the observed stress difference is no longer puzzling: Names, as CGs, may be expected to exhibit different stress patterns. It should also be noted that the combination of compounds with affixes and clitics in the same constituent is not new. In fact, in the other studies discussed, the same combination of structures was treated within a single constituent type, the recursive PW. 6.3. The CG across languages If prosodic constituents are to be considered components of UG, it is assumed that they should have similar properties cross-linguistically. In the present case, we would expect to find evidence of both the PW and the CG in at least most of the world’s languages, of course bearing in mind the issues raised above about the possibilities of missing the relevant data and the ultimate absence of relevant data in a given language. With regard to the PW, we would expect to find phonological phenomena that apply in relation to a constituent including

The status of the Clitic Group

37

a root and some or all affixes. With regard to the CG, we would expect to find phonological phenomena that apply in relation to a constituent that comprises a PW as well as affixes excluded from the PW and associated clitics. In Dixon and Aikhenvald (2002a), an impressive set of languages is analyzed with regard to their definitions of grammatical words (GWs) and phonological words, providing a fertile testing ground for the CG proposal. In a number of cases, there are elements (certain affixes and clitics) that are not included with the basic PW, but are grouped into additional PWs instead. While the result is not the same as the original arrangement in Nespor and Vogel (1986), where each such element formed a PW on its own, we do find cases of individual elements and combinations of elements being labeled as PWs that have little to do with the core understanding of a PW as consisting of a root and possibly some affixes. In other cases, it is explicitly recognized that a single prosodic constituent, the PW, is not adequate for handling all the data, and an additional constituent is posited, for example, the “PW-”, in Woodbury’s analysis of Cup’ik (Woodbury 2002). It is beyond the scope of this paper to reanalyze all the data, however, it seems that the CG offers a means for accounting for the types of considerations that have led to the positing of atypical PWs or additional types of PW constituents. As a test case, data from a language far removed from those seen thus far, Fijian, will be considered. Dixon and Aikhenvald (2002b) provide an analysis of the PW and GW in Fijian in which we find, alongside more typical PWs, other PWs consisting of multi-mora affixes and sequences of elements such as preposition + article + derivational prefix of the following word (cf. Dixon and Aikhenvald 2002b: 37). While it is clear on the basis of certain phonological phenomena that such items cannot be included in a PW with the associated root, labeling such varied items as PWs is inconsistent with a more universal definition of the PW. An alternative analysis in which these items are grouped with the PW into a CG will be shown to avoid this problem. Fijian has a very simple syllable structure: (C)V. There is a minimality requirement of two moras in the PW, and this can be achieved by having at least two syllables or by having a bimoraic syllable (i.e. CV: or CVG). A Diphthongization Rule is crucial in identifying the PW: Sequences of ai, ei, oi, au, eu and ou are realized as diphthongs when they are within a single PW, but as V-V otherwise (Dixon and Aikhenvald 2002b: 36). Primary stress falls on the syllable containing the second mora from the end of a PW, and secondary stresses appear on syllables containing the fourth and sixth moras from the end. Compounds and words formed by full reduplication consist of two PWs, since each component exhibits a primary (and potentially secondary) stress in relation to the right edge of its own PW, as illustrated in (28).

38

Irene Vogel

(28)

Fijian: 2 PWs (Dixon and Aikhenvald 2002b: 36) a. Compound: [sára]PW [vanúa]PW ‘tourist’ (lit. ‘look+at.place’) b. Reduplication: [butá]PW [butá’o]PW ‘steal constantly’ (< butá’o ‘to steal’)

Other types of constructions are also considered PWs. While a monomoraic affix combines with the root into a PW, a longer affix, it is claimed, forms a PW on its own, yielding a structure like that of compounds or reduplications, as in (29a). Similarly, strings of function words and affixes may combine to form PWs, as shown in (29b,c). (29)

Fijian: Affixes and Function Words as PWs (Dixon and Aikhenvald 2002b) a. Multi-moraic Suffix: [tàlanóa]PW [ta’ína]PW ‘tell stories’ (< talanoa ‘tell’) b. Function Words: [í+na]PW ‘to/at the’20 c. Function Word + Affix: [a+j]PW [sele]PW ‘the knife’

In (29c), a is an article and sele is a root meaning ‘to cut, slice’; the nominalizing prefix i- derives a noun from the root verb. It is argued that the prefix must combine with the article in a PW since the Diphthongization Rule applies, yielding what is represented here as [aj] (Dixon and Aikhenvald 2002b: 29–30). If the article a is not present, the prefix i- is included in the PW with the root. Based on this type of argument, even more complex combinations of nonroot elements are combined into PWs depending on the application or nonapplication of the Diphthongization Rule, as shown in (30), where “=” indicates a clitic juncture. (30)

Fijian: more complex structures (Dixon and Aikhenvald 2002: 37) a. [ i = na + j –]PW [talanoa]PW ‘in the story’ prep article nom pref ‘tell’ [ i – talanoa]PW ‘his story’ b. [ a = ona ]PW article poss nom pref ‘tell’ [i – talanoa]PW ‘in his story’ c. [ i = na ]PW [ona ]PW prep article poss nom. pref. ‘tell’

The assignment of PW structure to such varied types of elements yields the correct results with respect to certain phonological phenomena of Fijian. It 20

It should be noted that the article a becomes na when it is preceded by a vowel.

The status of the Clitic Group

39

raises a serious problem, however, since it would no longer allow us to base the construction of prosodic constituents, at least the PW, on general mapping principles that relate the morpho-syntactic structure to phonological structure. That is, there does not appear to be a straightforward algorithm for arriving at the PWs, and indeed, in some cases the prefix i- is part of the PW with the verb root and in others it is not. If prosodic constituents are part of UG, it is important that they share fundamental properties across languages, as opposed to representing specific groupings established for each individual situation. The question now is whether the type of analysis proposed above that includes the CG can allow us to retain general mapping principles for the construction of the PW, and still account for the phonological phenomena of Fijian. It seems that the answer to this question is positive. Based on Dixon and Aikhenvald’s (2002b) data, it could be argued that the possessor is a PW, as it contains two syllables and thus satisfies the minimality requirement. We can also, uncontroversially, assume that the verb root constitutes a PW. The only additional assumption required is that the prefix i- and the other function items in the examples join with an available PW to form a CG. This would mean that Fijian is similar to Italian, and presumably most, or all, other languages. The items in (30) would thus have the structures in (31); the original diacritic boundary symbols are removed. (31)

Fijian: Reanalysis with CG a. [i na j−]PW [talanoa]PW ]CG ‘in the story’ prep article nom pref ‘tell’ [talanoa]PW ]CG ‘his story’ b. [a [ona ]PW ]CG [i− article poss nom pref ‘tell’ [talanoa]PW ]CG ‘in his story’ c. i na [ona ]PW ]CG [− i prep article poss nom.pref. ‘tell’

In order to account for the Diphthongization Rule, all that is necessary is to assign it to the CG domain, as opposed to the PW domain. Thus, it applies in (31a) where the ai sequence is in the same CG, but not in (31b) and (31c) where it is split into different CGs. It should be noted that this analysis still accounts for diphthongization within the PW, as this constituent is contained within the CG, and is thus subject to phonological phenomena that apply to the CG. For example, diphthongization is correctly predicted in [[bojca ]PW ]CG ‘smell’ (cf. Dixon and Aikhenvald 2002b: 15), since the relevant sequence oi is contained within a single CG, which in this case also happens to be a PW. It should also be noted that with the present analysis, it is possible to retain the view of the PW advanced above according to which affixes do not form

40

Irene Vogel

PWs on their own. Dixon and Aikhenvald treat multi-moraic affixes as separate PWs, as seen in (29a), however, this is not necessary. If such affixes are excluded from the PW with the root, like the mono-moraic prefix i-, they will simply be grouped with the relevant PW into the CG (e.g. [ [talanoa] PW ta’ina]CG ‘tell stories’).21 Finally, it can be seen that the Fijian data are also consistent with the proposal to include compounds in CGs along with constructions containing affixes and clitics. It was seen in (28) that full reduplication results in the same type of structure as compounds, where each element is in its own PW. These PWs are taken to constitute a single CG, and we might thus expect the Diphthongization Rule to apply in such cases. It turns out, however, that this rule does not apply across the members of a reduplication (e.g. ilo-ilo / *ilo-jlo ‘glass’ [Dixon and Aikhenvald 2002b: 15]). Closer examination shows that this is not a problem in fact. Instead, we must only formulate the Diphthongization Rule so that it may apply across a single PW boundary within a CG but not a double one. This brief examination of Fijian demonstrates that the proposal developed above can be extended fruitfully to other languages. The inclusion of the CG in the prosodic hierarchy, and the possibility of skipping levels so as to combine affixes and clitics directly into a CG, allows us to simplify the definition of the phonological constituents in question and still account for the relevant facts. Specifically, we are able to capture a difference in behavior between phonological phenomena that apply in relation to distinct domains, the PW and the CG, and this, in turn, allows us to maintain the concept of linguistic constituent as a string exhibiting a particular property, or set of properties. Furthermore, we are able to retain a core aspect of prosodic phonology according to which prosodic constituents represent universal components of phonological structure, constructed via a mapping from the morpho-syntax onto phonological structure in fundamentally the same way across languages. 7. Psycholinguistic evidence of the C(omposite) G(roup) The analyses and proposals presented thus far have been based on descriptive data and general theoretical considerations. The ultimate test for a linguistic analysis, however, is what is referred to as “psychological reality”. It will be 21

The penultimate syllable in the suffix is stressed, which Dixon and Aikhenvald take as evidence of the PW status of the suffix. It is beyond the scope of the present paper to re-examine all the stress facts of Fijian, but it would seem possible that the stress on the suffix in question could be due to a different CG stress rule, while the PW stress rule applies as seen in (23a).

The status of the Clitic Group

41

shown here that there is psycholinguistic evidence for the present proposal as well. Specifically, it will be demonstrated that the Composite Group, which includes constructions with clitics as well as compounds, correctly accounts for response latency data reported in two recent studies (Wheeldon and Lahiri 1997, 2002). In fact, the CG resolves a problem found in the original studies, where it was argued that the basic constituent for speech production is the PW – and also the recursive PW. Wheeldon and Lahiri (1997, 2002) argue that the basic unit of speech encoding is a prosodic rather than a morpho-syntactic constituent (cf. also Levelt 1989, 1992). As a speaker prepares to speak, a reply involving more units will result in a longer delay, or response latency, than one involving fewer units. The question is which constituent serves as the crucial unit in predicting the response latencies. Wheeldon and Lahiri used a prepared speech paradigm to elicit predetermined responses from Dutch speakers. In each experiment, the subjects saw a word or phrase on a computer screen. Following three beeps, they either a) responded to a question (Study 1), or b) produced a simple sentence (Study 2) using the word or phrase they had just seen, and the response latencies were measured. In both studies, a series of experiments was conducted; the present discussion generalizes across them. The response types tested in Studies 1 and 2 are shown in (32) and (33), respectively. The brackets indicate PWs; het in (32c) is a full word, not a clitic, due to the way in which it was elicited. (32)

(33)

Response Types (Study 1) a. clitic sentence [Ik zoek het] [water] b. non-clitic sentence [Ik zoek] [vers] [water] c. pronoun sentence [Ik zoek] [het] d. control [ik zoek]

‘I seek the water.’ ‘I seek fresh water.’ ‘I seek it.’ ‘I seek.’

Response Types (Study 2) a. compound sentence [Het was] [oog] [lid] b. A-N phrase [Het was] [oud] [lid] c. simple: initial stress [Het was] [orgel] d. simple: final stress [Het was] [orkaan]

‘It was eyelid.’ ‘It was old member.’ ‘It was organ.’ ‘It was hurricane.’

In Study 1, it was found that structures such as (32a) and (32c) exhibit the same latencies. This was taken as evidence that the PW is the appropriate unit underlying speech production, as both structures consist of two PWs. In Study 2, it was predicted that (33a) and (33b) would exhibit the same latencies, the assumption being that they both contain three PWs. Instead, (33a), the compound, behaved like the simple items (33c, 33d). These, in turn, were comparable to the items with two PWs in the first study.

42

Irene Vogel

Since Wheeldon and Lahiri (1997) had argued that the PW is the relevant unit for speech production, it was necessary to reconcile the findings of the two studies, which appeared to be contradictory. To do this, Wheeldon and Lahiri reanalyzed the compounds as consisting of recursive structures, with the internal PWs included in a larger PW (e.g. [ [oog]PW [lid]PW ]PW ‘eye lid’). It was then argued that “[. . . ] the prosodic word status that counts for the encoding is at the level of phrasal prosodic structure and [. . . ] the processes that compute phrasal prosody are blind to word internal structure” (Wheeldon and Lahiri 2002: B38). In other words, the implicit claim is that there are two types of PW: a) one that accounts for the usual PW phenomena, such that a compound has two of these, and b) another one that accounts for response latencies, this being the maximal PW (or PW’) in the case of a recursive structure. Closer examination of the constructions considered by Wheeldon and Lahiri as evidence for the PW being the basis for speech production reveals that these are exactly the possibilities included in the Composite Group, without further qualification. This set of structures constitutes Wheeldon and Lahiri’s “fast group”, as seen in (34). It should be noted that the rates are expressed in relative terms, as opposed to the actual number of milliseconds, to permit comparison across the different experiments. (34)

a. Study 1 b. Study 2 non-clitic sentences slow A-N phrase clitic, pronoun sentences fast simple words, compounds control fastest N/A

In sum, while Wheeldon and Lahiri (2002) require an analysis that involves recursive structures as well as the additional stipulation of which PW counts for which purposes, the present analysis only requires the existence of the CG. Furthermore, whereas it might originally have seemed somewhat questionable to include compounds and constructions with clitics in the same prosodic constituent, we now have independent psycholinguistic evidence that such groupings behave in the same way. 8. Conclusion In this paper, several problems with the original Clitic Group were discussed. It was shown that the problems were not due to the CG constituent itself, but to restrictions on the geometry of the prosodic hierarchy as a result of the Strict Layer Hypothesis. Several solutions were considered, and it was argued that all that is required is a slight weakening of the SLH such that a prosodic constituent

The status of the Clitic Group

43

may dominate material more than one level lower in the hierarchy; it is assumed that the CG is still part of the prosodic hierarchy. The introduction of a recursive PW, which has also been proposed in some cases, was shown to be unnecessary, and to have a number of undesirable consequences. The arguments for the present proposal are based on data from several languages as well as general considerations about linguistic theory and structure. It was shown that the CG, which has been renamed the “Composite Group” to include compounds, allows us to maintain the relatively flat structure of the phonology, in contrast with that of the morpho-syntax. Furthermore, use of the CG, as opposed to a recursive PW, allows us to maintain the basic notion of constituent: A linguistic string defined on the basis of a particular property or set of properties. Specifically, the CG provides explicit recognition of the fact that across languages, different types of phenomena are observed where clitics, and often “level 2” affixes, are involved. The recursive PW surreptitiously introduces a similar distinct constituent since the PW’ is fundamentally different from both the PW and PPh, but it fails to explicitly recognize that the additional constituent has its own special phonological properties and interface relations with the morpho-syntax. Finally, several psycholinguistic experiments conducted with Dutch speakers were considered. While these were originally presented as support for the PW as the crucial unit of linguistic processing, it was demonstrated that the results require a somewhat complex interpretation in terms of the PW. By contrast, if they are viewed in terms of the CG, the interpretation becomes quite simple, thus providing independent psycholinguistic motivation for the proposals advanced in this paper. References Anderson, Stephen 2005 Aspects of the Theory of Clitics. Oxford: Oxford University Press. Basbøll, Hans 1975 Grammatical boundaries in phonology. Aripuc 9: 109–135. 1981 On the function of boundaries in phonological rules. In: Didier Goyvaerts (ed.), Phonology in the 1980’s, 245–269. Ghent: Story-Scientia. Bertinetto, Pier Marco 1999 Boundary strength and linguistic ecology. Folia Linguistica 33: 267– 286. Bickel, Balthasar and Kristine Hildebrandt 2007 Word domains. Cross-linguistic challenges for the prosodic hierarchy: Evidence from word domains. Ms. University of Leipzig.

44

Irene Vogel

Booij, Geert 1996

Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13: 219–242. Cardinaletti, Anna and Lori Repetti 2008 Preverbal and postverbal subject Clitics in Northern Italian dialects: phonology, syntax and cross-linguistic variation. Linguistic Inquiry 39: 523–563. Carter, Richard T. Jr. 1974 Teton Dakota Phonology. PhD. Dissertation. University of New Mexico. Published as University of Manitoba Anthropology Papers 10. Dixon, Robert M. W. and Alexandra Y. Aikhenvald 2002a Word. Cambridge: University Press. 2002b Word: A typological framework. In Robert M.W. Dixon, and Alexandra Y. Aikenvald (eds.), Word, 1–41. Cambridge: University Press. Downing, Laura J. 1999 Prosodic stem gprosodic word in Bantu. In: Tracy A. Hall, and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 73–98. Philadelphia: John Benjamins. Hall, Tracy A. 1999 The phonological word: A review. In Tracy A. Hall, and Ursula Kleinhenz (eds.) Studies on the Phonological Word, 1–22. Philadelphia: John Benjamins. Hayes, Bruce 1989 The prosodic hierarchy in meter. In: Paul Kiparsky and Gilbert Youmans (eds.), Rhythm and Meter, 201–260. Orlando, Florida: Academic Press. Hoskins, Steven R. 1994 Secondary stress and clash resolution in French. In: Michael Mazzola (ed.), Issues and Theory in Romance Linguistics: Selected Papers from the Linguistic Symposium on Romance Languages XXIII, 35–47. Washington, DC: Georgetown University Press. Ito, Junko and Armin Mester 1992 Weak layering and word binarity. Ms, University of California at Santa Cruz. Updated version 2003. Jackendoff, Ray and Steven Pinker 2005 The nature of the language faculty and its implications for the evolution of language. (Reply to Fitch, Hauser, and Chomsky). Cognition. 97: 211–225. Kabak, Bari¸s and Irene Vogel 2001 The phonological word and stress assignment in Turkish. Phonology 18: 315–360. Kaisse, Ellen M. and Patricia Shaw 1985 On the theory of lexical phonology. Phonology Yearbook 2: 1–30.

The status of the Clitic Group

45

Ladd, D. Robert 1996 Intonational Phonology. Cambrdige: University Press. Levelt, Willem J. M. 1989 Speaking: From intention to articulation. Cambridge, MA: MIT Press. 1992 Accessing words in speech production: Stages, processes and representations. Cognition 42: 1–22. Loporcaro, Michele 1999 Teoria fonologica e ricerca empirica sull’italiano e i suoi dialetti. Fonologia e morfologia dell’italiano e dei dialetti d’Italia. In: Paola Benincà, Alberto Mioni, and Laura Vanelli (eds.), Atti del 31˚ Congresso della Società di Linguistica Italiana, 117–151.Roma: Bulzoni. Mazzola, Michael 1992 Stress clash and segment deletion. In: Christiane Laeufer, and Terrell Morgan (eds.), Theoretical analyses in Romance Linguistics, 81–96. Philadelphia: John Benjamins. 1993 French rhythm and French segments. In: William J. Ashby, Marianne Mithune, Giorgio Perissinotto, and Eduardo Raposo (eds.), Linguistic Perspectives on the Romance Languages: Selected Papers from the XXI Linguistic Symposium on Romance Languages, 113–126. Philadelphia: John Benjamins. Neeleman, Ad and Hans van de Koot 2006 On syntactic and phonological representations. Lingua 116: 1524– 1552. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Patterson, Trudi A. 1990 Theoretical aspects of Dakota morphology and phonology. Ph.D. Dissertation. Univerersity of Illinois at Urbana-Champaign. Peperkamp, Sharon 1997 Prosodic Words. The Hague: Holland Academic Graphics. Pinker, Steven and Ray Jackendoff 2005 The faculty of language: What’s special about it? Cognition 95: 201– 236. Schiering, René, Kristine Hildebrandt, and Balthasar Bickel 2007 Cross-linguistic challenges for the prosodic hierarchy: Evidence from word domains. Ms. University of Leipzig. Selkirk, Elisabeth 1996 The prosodic structure of function words. In: James L. Morgan, and Katherine Demuth (eds.), Signal to Syntax: Bootstrapping from Syntax to Grammar in Early Acquisition, 187–213. Mahwah, NJ: Erlbaum.

46

Irene Vogel

Shaw, Patricia 1980

Theoretical Issues in Dakota Phonology and Morphology. New York: Garland Press. 1985 Modularization and substantive constraints in Dakota lexical phonology. Phonology Yearbook 2: 173–202. Simon, Horst and Heike Wiese (eds.) to appear Expecting the unexpected: Exceptions in grammar. Trends Linguistics: Studies and Monographs Series. Amsterdam: Mouton de Gruyter. Vigário, Marina 2003 The Prosodic Word in European Portuguese. Berlin: Mouton de Gruyter. Vogel, Irene 1994 Phonological interfaces in Italian. In: Michael Mazzola (ed.), Issues and Theory in Romance Linguistics: Selected Papers from the Linguistic Symposium on Romance Languages XXIII, 109–125. Washington, DC: Georgetown University Press. 1999 Subminimal constituents in prosodic phonology. In: S. J. Hannahs, and Mike Davenport (eds.), Issues in phonological structure, 249– 267. Amsterdam: Benjamins. 2008a The morphology-phonology interface: Isolating to polysynthetic languages. In: Ferenc Kiefer, and Peter Siptar (eds.), Selected Papers from the 12th International Morphology Meeting. Special Issue of Acta Linguistica Hungarica 55. Budapest: Akadémiai Kiadó Hungary. 2008b Universals of prosodic structure. In: Sergio Scalise, Elisabetta Magni, Edoardo Vineis, Antonietta Bisetto (eds.), With more than Chance Frequency. Forty Years of Universals of Language. Amsterdam, Springer. Vogel, Irene and Eric Raimy 2002 The acquisition of compound vs. phrasal stress: The role of prosodic constituents. Journal of Child Language 29: 225–250. Wheeldon, Linda and Aditi Lahiri 1997 Prosodic units in speech production. Journal of Memory and Language 37: 356–381. 2002 The minimal unit of phonological encoding: Prosodic or lexical word. Cognition 85: B31–B41. Woodbury, Anthony C. 2002 The word in Cup’ik. In: Robert M.W. Dixon, and Alexandra Y. Aikenvald (eds.), Word, 79–99. Cambridge: University Press.

The distribution of phonological word domains: A probabilistic typology Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

1. Introduction The past few decades have seen various attempts to develop descriptive frameworks that capture the range of phonological domains expected across languages. Most prominent among these is still the Prosodic Hierarchy framework (Nespor and Vogel 1986; Peperkamp 1997; Selkirk 1984). The framework posits a finite range of possible domains and predicts that phonological patterns universally converge on exactly this range of domains, e. g. there will be exactly one domain between foot and phrase: the phonological word (‘ω ’). We call this the Clustering Hypothesis. Another prediction of the framework is known as Strict Succession or Proper Headedness and states that each level L of the Prosodic Hierarchy dominates at least one level L-1 until the terminal level L=0, e. g. there will always be at least one phonological word level ω dominated by a phonological phrase level. Proper Bracketing, finally, formulates the expectation that no language will exhibit non-stacking domains in its prosodic structure, e. g. the edge of a phonological phrase cannot be situated inside a phonological word. As Inkelas and Zec (1995: 548) pointed out, although the theoretical predictions of the Prosodic Hierarchy framework are straightforward, the empirical evidence is often less so. When analyzing languages, one often encounters challenges to the predictions made by the Prosodic Hierarchy model. In principle, there are two routes for facing these challenges. One route is to adjust and refine the model, e. g. by extending the range of domains posited (e. g. admitting one or more additional levels such as the Clitic Group: Vogel, this volume), by weakening some claims (e. g. allowing violations of the Strict Succession requirement under certain conditions), or by restricting the scope of the framework (e. g. by limiting it to suprasegmental patterns only). An alternative route is typological: Instead of assuming a finite and universally fixed set of phonological domains, one may ask what kinds of domains are

48

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

empirically evidenced by a sample of languages and explore the principles that explain the distribution of the attested domains. This chapter takes this second route and reports on the kinds of domains found at the word level – i. e. domains larger than the foot but smaller than the phrase – in a sample of languages. Rather than making an a priori assumption that there must be exactly one level of word, we instead define a typological variable whose values are the domains identified between foot and phrase. We begin by briefly reviewing the kinds of empirical challenges that individual data present (Section 2), and from this we develop a universally applicable method of measuring the diversity observed (Section 3). In order to analyze our measurements, we adopt standard statistical tools, specifically Multi-Dimensional Scaling (MDS) (Section 4). The key result of this analysis is that stressrelated domains tend to be larger than domains referenced by other phonological patterns (including any kind of phonology, from vowel harmony to consonant dissimilation). In Section 5 we submit this result to regression modeling and show that the impact of stress is independent of areal and genealogical affiliation, and is therefore a good candidate for a truly universal principle. Section 6 summarizes the findings and discusses their consequences.

2. Cross-linguistic challenges Consider the prosodic domains in Limbu, an Eastern Kiranti (Sino-Tibetan) language of Nepal (van Driem 1987; Ebert 1994; Hildebrandt 2007; Schiering, Hildebrandt and Bickel 2007; Tumbahang 2007; Weidert and Subba 1985). At first sight, phonological patterns reference at least four distinct domains between the foot (ϕ ) and the phonological phrase (P) (Figure 1). The data in Figure 1 appear to challenge all predictions of the Prosodic Hierarchy framework: prosodic domains cluster on more domains than provided by the Prosodic Hierarchy, i. e. at the word level, we have to distinguish four distinct layers. With respect to Strict Succession/Proper Headedness, Limbu shows that a level ω may dominate another level ω , i. e. it does not directly dominate the next lower level, a foot. In this case, the domains ω 1 and ω 2 constitute non-stacking domains, i. e. overlapping domains, and thus violate Proper Bracketing. However, under closer inspection, not all four domains pose an equal challenge. First, the Prosodic Hierarchy model only makes claims about purely phonological rules that apply generally across the lexicon, and makes no predictions about lexically limited phonological patterns. This discards the evidence for ω1 and ω2 because the processes which reference the morphological string

The distribution of phonological word domains: A probabilistic typology

49

Figure 1. Prosodic domains in Limbu (Phedappe dialect; Hildebrandt 2007)

in question are observable in only two individual affixes, and we cannot exclude the possibility that these are lexically specified. However, word domains ω3 and ω4 remain an issue because they are referenced by general rules. Since the source for the violations of the Prosodic Hierarchy predictions is prefixes, one possible solution would be to analyze the prefixes as postlexical. However, since prefixes are obligatory elements of inflectional morphology and can only ever appear as parts of verb forms, there is no independent evidence that they could be analyzed as postlexical. More importantly, prefixes (and, for that matter, suffixes) contrast with genuine postlexical elements (“clitics”) in Limbu, which syntactically attach to phrases and do not subcategorize for a specific part of speech or stem type. Clitics of this kind, e. g. the additive focus

50

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

clitic =aN in Figure 1, are always included in both ω 3 and ω 4 . Since prefixes are included in ω 4 but not in ω 3 , analyzing them as postlexical would entail that two distinct postlexical domains need to be posited. As a result, one could at best re-label ω 3 and ω 4 as, for example, “Clitic Group 1” and “Clitic Group 2”, but the challenge for the Prosodic Hierarchy framework remains. Another analysis that could in principle be applied in order to bring Limbu in line with the Prosodic Hierarchy framework is in terms of recursion (e. g. via a low ranking of the non-recursivity constraint proposed by Selkirk 1995). The larger domain ω 4 would then be analyzed as a recursive instantiation of the smaller domain ω 3 . Following standard definitions of recursion, such an analysis would predict that the two domains in question have identical phonological properties (for examples, see Peperkamp 1996, 1997). However, this prediction is not borne out by the Limbu data, since the domains are motivated by a number of phonological patterns which are clearly distinct (Hildebrandt 2007). Primary stress assignment, for instance, operates at ω 4 only, and ω 3 has no stress-related properties at all. In some cases, relativizing prosodic structure to different phonological tiers allows to account for data which otherwise seem to counter the predictions of the Prosodic Hierarchy framework. In Luganda (Bantu), for example, the domain for the presence of only one high-to-low pitch drop and the domain for a rule of final vowel shortening overlap in such a way that Proper Bracketing is violated (Hyman, Katamba and Walusimbi 1987). If we allow prosodic structure to be construed independently on different phonological tiers, in this case tone and quantity, the theoretical predictions are still borne out in a more fine-grained perspective on the architecture of prosodic structure. For Limbu, however, such an approach cannot solve the problem because the various phonological patterns which motivate each word domain come from such diverse phonological tiers and types as phonotactics, segmental, and suprasegmental phonology. In Figure 1, we illustrated ω 4 by stress assignment patterns. But the same domain is also referenced by a segmental rule of Coronal-to-Labial Assimilation: (1)

a.

b.

/mE-n-mEt-paN/ [mEmmEppaN] nsA-NEG-tell-1s>3.PST ‘I did not tell him’ /hEn=phElle/ [hEmbhElle] what-QUOT ‘What?’

The rule states that the coronal phonemes /n/ and /t/ anticipate the place of articulation of the bilabial phonemes /m/ and /p/ within a phonological word.

The distribution of phonological word domains: A probabilistic typology

51

The relevant phonological word domain includes both prefix-stem and stemsuffix boundaries, as illustrated by (1a). Like the domain for Primary Stress Assignment, the domain also includes host-clitic boundaries, as shown in (1b). In sum, regardless of the approach of adjustment or refinement that one may take, languages like Limbu provide unresolved challenges for the predictions contained in the Prosodic Hierarchy Hypothesis. In the following, we turn to an alternative way of looking at the data. 3. Measuring diversity in word domains An alternative route to addressing challenges posed by languages like Limbu is typological: Instead of reducing the observed diversity to a single universal model, we measure the actual distribution of word domains across languages and look for universal principles explaining the observed distribution. For this, we need a universally applicable working definition of word domains and use this to create a database which, for each language, specifies the precise nature of its phonological domains. 3.1. A working definition of phonological word domains Our working definition of phonological word domains includes all sound pattern domains that are delimited by some morphological structure but do not include more than one lexical stem. Thus, we concentrate exclusively on specific combinations of stems, affixes and clitics/particles, and we exclude feet or syllable structure (where definitions do not reference morphological structure) as well as any larger domains like phrases and compounds. In line with the Prosodic Hierarchy framework, we also limit all our analyses to lexically general phonological patterns, i. e. we exclude data that are positively limited to a subset of affixes or stems, or only to loanwords.1 For convenience, we refer to the patterns as “pw-patterns” as defined here.

1

However, we systematically entered such data into our database in order to allow more comprehensive analyses later. Interestingly, the general results presented in this paper do not depend on whether lexically-specified patterns and domains are included or excluded from the analysis. Some statistical signals become stronger, some weaker, but the overall findings stay.

52

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

3.2. The database Our database2 currently covers 70 typologically diverse languages, but for seven of these languages, we have not found any evidence for pw-patterns that are strictly sub-phrasal and are at the same time fully general across the lexicon. This reduces the dataset to 63 languages containing a total of 382 pw-patterns. In one component of the database, we collect, for each language, all pwpatterns, such as stress, tone, segmental rules and phonotactic constraints, with an exact description of their phonological properties. All 63 languages have between 1 and 19 distinct pw-patterns, and more than half of them have between 1 and 5 pw-patterns. For each pw-pattern we code the kind and number of morphemes which are included in its domain. In order to calibrate this information against the morphological set-up of the language, we also keep track of the different morpheme types that are relevant for a given language, such as “suffixes” (defined as postposed grammatical markers that are subcategorized for a stem class), “proclitics” (defined as preposed grammatical markers that are not restricted to a stem class or a part of speech), “stems”, etc. This information is included in a separate component of the database. Most languages have between 2 and 7 morpheme types, and almost 80% of them between 2 and 4 morpheme types. 3.3. A measurement of word coherence The intuition behind the Prosodic Hierarchy is that, independent of the morphological make-up of a language (e. g. whether a language happens to have prefixes or not), domains will distribute in a hierarchy of similarly or identically-sized levels – starting at the top with large domains that include all morphological material that a language has and ending at the bottom with narrowly defined domains like stem-suffix combinations that exclude prefixes and clitics, for example. Such a view suggests two probabilistic expectations across languages: (i) we expect some kinds of domains to be recurrently larger than others, so that, within languages, the larger domains will properly contain the smaller ones; (ii) these hierarchies of domains will tend to cluster on universal “attractors” that are defined by some shared property. For example, there could be a cross-linguistic trend towards three nested sizes – say, “large”, “medium” and “small” – each characterized by the kind of phonology involved, e. g., stress placement vs. vowel harmony vs. consonant assimilation. 2

For a project description and ancillary material, including analyses of individual languages, see www.uni-leipzig.de/~autotyp > projects and > language reports. The web site also makes available the database itself and a bibliography of its sources.

The distribution of phonological word domains: A probabilistic typology

53

To find out, we need a way of measuring the size that domains have in relation to each other. A straightforward way of doing this is by determining how many of the morpheme types that a language has are included in a given phonological word domain. In principle, this could be the stem alone, the combination of stem, suffixes and clitics, as in the Limbu ω 3 domain, or even the combination of all available morpheme types, e. g., the prefix-stem-suffix-clitic string in the Limbu ω 4 domain. Since the number of morpheme types included in a domain depends on the number of morpheme types available in a given language, we define the relative word size c of a pw-pattern p in a language L as (where c is a mnemonic for “coherence”): (2)

c(p, L) =

N (morpheme types in domain referenced by p) N (morpheme types in language L)

For example, a domain like ω 4 in Limbu is defined, among other patterns, by the Primary Stress Assignment listed in Figure 1 and the segmental rule of Coronalto-Labial Assimilation discussed in example (1). The relative coherence c of these patterns is measured as follows: (3)

a. b.

/mE-"thaN-e=aN/ ‘they come up and . . . ’ 3ns-come.up-PST=and /"ku-la:p/ ‘its wing’ 3POSS-wing 4 (prefix-stem-suffix=clitic) – 4 (prefix-stem-suffix=clitic) → c (Limbu Primary Stress Assignment) = 1

(4)

a. b.

/mE-n-mEt-paN/ [mEmmEppaN] ‘I did not tell him’ nsA-NEG-tell-1s>3.PST /hEn=phElle/ [hEmbhElle] ‘What?’ what-QUOT 4 (prefix-stem-suffix=clitic) – 4 (prefix-stem-suffix=clitic) → c (Limbu Coronal-to-Labial Assimilation) = 1

In (3a), only one primary stress is assigned to the combination of prefix-stemsuffix=clitic. In the given form, this primary stress is realized on the stem, but, as shown in (3b), in other forms the stress shifts to the prefix (cf. Hildebrandt

54

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

2007; Schiering, Hildebrandt, and Bickel 2007 for discussion). This suggests that the stress assignment rule of Limbu references all four available morpheme types of the language, and has therefore coherence c = 1. The examples in (4) repeat those of (1) and illustrate Coronal-to-Labial Assimilation. Combining the evidence from (4a) and (4b), the assimilation spans all four morpheme types available in the language and is thus coded as having coherence c = 1. Of course, prosodic word domains often do not reference all available morpheme types of a given language, i. e. have c < 1. This point can be illustrated with the Limbu word domain ω 3 outlined in Figure 1. This domain is motivated by two phonological processes, i. e. Glottal Stop Insertion and the [l] ∼ [r] alternation, and applies to only three of the four morpheme types available in the language. (5)

a. b. c. d.

/ku-e…k/ [ku/e…k] (3POSS-back) ‘its/his/her back’ /am-phu-e…/ [/amphue…] (1POSS-brother-VOC) ‘Brother!’ /nu-ba=i…/ [nu-bai…] (be.alright-NOM=Q) ‘Is this OK?’ /ai…rE/ [/a/i…rE] (1-wander-PST) ‘We wandered.’

3 (stem-suffix=enclitic) – 4 (prefix-stem-suffix=enclitic) → c (Limbu Glottal Stop Insertion) = .75 (6)

a. b. c. d.

/nElEt/ [nErEt] ‘heart’ /pha-le siN/ [pha-re siN] (bamboo-GEN wood) ‘the wood of bamboo’ /pe…g-i=lo…/ [pe…g-i=ro…] (go-p=ASS) ‘Come on, let’s go!’ /kE-lO// [kE-lO/] (2-say) ‘you say’

3 (stem-suffix=enclitic) – 4 (prefix-stem-suffix=enclitic) → c (Limbu [l] ∼ [r] domain) = .75 In (5a), a glottal stop is inserted between the prefix and the vowel-initial stem. In the segmentally identical hiatus situation at the stem-suffix boundary in (5b), the process does not apply to the vowel-initial suffix. The process also fails to apply with vowel-initial enclitics, as is shown in (5c). From this we can conclude that Glottal Stop Insertion targets the left edge of a vowel-initial prosodic word which references the stem, its suffixes and enclitics. Accordingly, three of the four available morpheme types are included in the domain. The resulting coherence value for this domain is therefore c = .75. Example (5d) demonstrates that the vowel-initial prefix constitutes its own prosodic word for the sake of this process.

The distribution of phonological word domains: A probabilistic typology

55

The data in (6) illustrate the domain structure of the [l] ∼ [r] alternation. Syllable-initial /l/ always surfaces as [l] in word-initial position, while in wordmedial position it is realized as either [r] or [l], depending on the structure of the preceding syllable. The realization is [r] if the preceding syllable is open or ends in a glottal stop. In (6a), /l/ appears within a monomorphemic word and surfaces as [r] following an open syllable. For the sake of this alternation, the stem-suffix boundary in (6b) and the host-enclitic boundary in (6c) also provide an appropriate context for the realization of the allophone [r]. On the basis of these observations, we can formulate the generalization that, parallel to the case of Glottal Stop Insertion, the [l] ∼ [r] alternation applies within a prosodic domain which encompasses the stem, its suffixes and enclitics, i. e. three of the four available morpheme types. By dividing 3 by 4 we again get a coherence value of c = .75 for this domain. The data in (6d) show that the prefix lies outside this domain. Our definition of c abstracts away from the distinction between the kinds of morpheme boundaries involved. If a hypothetical language has prefixes, stems and suffixes as morpheme types, a prefix+stem domain and a stem+suffix domain will each score the same c-value of 2/3 although these involve different types of morpheme boundaries. However, both domains are smaller than an all-encompassing prefix-stem-suffix domain, and for finding out whether there are trends towards hierarchical arrangements of larger vs. smaller domains, the general size difference is more important than the exact alignment of domains. 3 This justifies our abstracting away from morpheme boundary types. Still, we checked whether equating prefix- and suffix-oriented domains might have a distorting effect. They do not: A comparison of the means and variances of c values when calculated only to the left of the stem (stem and prefix/proclitics) vs. when calculated only to the right of the stem (stem and suffix/enclitic) revealed no 3

In fact, the hypothetical case violates the principle of Proper Bracketing. The few languages that have this pattern in our dataset invite research beyond the scope of this chapter. Independently of this, however, we explored an alternative way of comparing domains, based on determining whether or not given morpheme boundary types (e. g. stem-suffix, prefix-stem, etc.) are each included in two domains (a criterion we use in Figure 2 below). This approach, suggested to us by an anonymous reviewer, seeks clusters with regard to specific boundary types and exact domain definitions, and not with regard to general size differences. In contrast to what we report below on c, a multidimensional scaling analysis based on such domain definitions did not reveal any systematic clustering (and Kruskal Stress values below 10% required at least 4 dimensions). It is possible that the exact regulation of domains in terms of boundary types is free of universal preferences, but we must again leave this question to future research. (The coding is available in the published database.)

56

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

statistical differences in our data (Monte-Carlo permutation 4 t-test, t = −1.79, p > .05; Fligner-Killeen test, χ 2 < .01, p > .05). The measurements of c across pw-patterns and across languages define a typological variable with values starting near 0 (where only one out of many morpheme types is included) and an upper limit of 1 (where all available types are included). In our database, c ranges from .14 to 1 in eighteen intervals spaced between .02 and .14, which we take as approximating a continuous scale for statistical purposes. 3.4. Data summary and discussion One key prediction of the Prosodic Hierarchy framework is that in each language, phonological patterns converge on a single domain between foot and phrase. After exhaustively searching the literature and, in many cases, having undertaken additional studies on available phonetic data or having queried fieldworkers, we observe that the majority of the 63 languages under study have more than a single domain.5 Figure 2 plots how many languages (y-axis) have a given number of non-isomorphic domains (x-axis), where isomorphism is defined as identity in morpheme types and morpheme boundaries. Figure 2 shows that only 9 out of the 63 languages considered here show the predicted clustering of word-related phonological patterns on one word domain. All other languages violate Clustering by having more than one non-isomorphic domain. In other words, Limbu is not an idiosyncratic isolated case of deviation but instead exemplifies a cross-linguistic trend for prosodic domains to multiply at the word level. 4. Probabilistic clusters A question arising from the observation in the preceding section is whether, instead of a categorial cluster of size 1 (i. e. with exactly one domain on which all pw-patterns converge), there are probabilistic clusters of similarly-sized domains. If there are, it is likely that they depend on various factors. In this paper, we explore the possibility that the coherence of pw-patterns depends on the kind

4

5

Since our data are not randomly sampled, we assess statistical differences by permutation methods throughout (cf. Everitt and Hothorn 2006; Janssen, Bickel and Zúñiga 2006; R Development Core Team 2007). All approximations are based on 10,000 random permutations. For one language, Khalkha Mongolian, the search was not exhaustive and there may be more non-isomorphic domains than what we have coded in our database.

57

0

5

10

15

The distribution of phonological word domains: A probabilistic typology

1

2

3

4

5

6

7

Figure 2. Number of non-isomorphic domains referenced by lexically general phonological patterns (data from 63 languages). The x-axis represents the number of nonisomorphic domains, the y-axis the number of languages with that many non-isomorphic domains.

of sound pattern involved, e. g., it might be the case that across languages, tonal patterns target domains with relative coherence degrees that are systematically different from the coherence degrees of segmentally-defined domains. This is much in line with what Hyman, Katamba, and Walusimbi (1987) suggested for Luganda, but re-interpreted as a probabilistic trend rather than any categorical constraint. To find out whether there are such trends, we conducted a non-metrical MultiDimensional Scaling analysis (MDS, e. g. Cox 1994; Everitt and Hothorn 2006; Venables and Ripley 2002) on our dataset of pw-patterns. 4.1. Coding and methods In preparation of the MDS analysis we coded each individual pw-pattern into a taxonomy of sound pattern types at a chosen level of resolution. For instance, the segmental pw-patterns can be broken down into three basic types: Allophony, phonotactic constraint, and alternation process. Alternation processes can, in turn, be broken down into subtypes like assimilation, deletion, dissimilation, etc. While we tested various levels of resolution and ways of setting up the taxonomy, a relatively shallow taxonomy of 15 types revealed all structure that we found with other taxonomies as well. Figure 3 shows the taxonomy that entered into the analysis.

58

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

allomorphy

segmental

allophony

constraint

size-related

process

suprasegmental

quantity stress tone

rhythm

assimil. V-harmony deletion dissimilation insertion strengthening weakening other Figure 3. Pw-pattern taxonomy

The terms of the taxonomy are defined as follows. (Numbers in brackets indicate how often the pattern is encountered in the database, again only counting lexically general pw-patterns): – Allomorphy (4): Some constraint is resolved by choosing a specific allomorph, e. g. an anti-hiatus constraint within (but not between) words is satisfied by systematically choosing C-initial allomorphs. – Allophony (4): The surface form of a phoneme depends on the position of the segment within the word, e. g. the Limbu [l] ∼ [r] alternation discussed above. – Assimilation (41): The realization of a segment is dependent on the segmental context within the word domain, e. g. the Limbu nasal /n/ assimilates to a following velar segment and surfaces as [N], as discussed above. – Vowel harmony (7): All vowels within a domain share one or more feature specifications. – (Phonotactic) constraint (123): The prosodic word imposes a restriction on the distribution of segments within the domain, e. g. a ban on velar nasals in word-initial position. – Deletion (25): A segment is deleted in a position specified by the prosodic word domain, e. g. a vowel is deleted in word-final position.

The distribution of phonological word domains: A probabilistic typology

59

– Dissimilation and metathesis (3): The realization of a segment is dependent on the segmental context within the word domain, such that the co-occurrence of similar segments is avoided; this also includes metathesis. – Insertion (28): Some constraint is resolved by inserting a vowel or consonant, e. g. an anti-hiatus constraint within (but not between) words is satisfied by inserting an epenthetic consonant. – Other process (2): Segmental processes which cannot be grouped with the other segmental categories in the taxonomy, e. g. re-syllabification across morpheme boundaries within a word domain. – Quantity (9): Generalizations related to the suprasegmental feature length, e. g. vowel lengthening in word-final position. – Rhythm (1): Phonological patterns which result in the rhythmic structuring of a given domain, e. g. alternating stresses within the word yielding trochaic or iambic feet under the word node. – Size-related (35): Constraints which specify the maximal or minimal word size in counts of syllables or moras, e. g. minimal or maximal disyllabicity required for stems, or for stem-affix combinations, but not, e. g. for prefixstem or host-clitics combinations. – Strengthening (25): Processes which result in the phonological strengthening of a segment, e. g. fortition. – Stress (36): Generalizations related to the suprasegmental feature stress, e. g. word-final stress placement. – Tone (10): Generalizations related to the suprasegmental feature tone, e. g. generalizations such as only one high-low pitch drop per word. – Weakening (36): Processes which result in the phonological weakening of a segment, e. g. lenition. The next methodological step is the construction of a distance matrix that measures the Euclidean distance between the relative coherence c of each pair of the 382 pw-patterns in our dataset. These are 382 · (382−1) = 72,771 pairs, of which 2 a randomly selected handful is shown for illustration in Table 1. Like in geographical distance charts, identical pw-patterns have a distance of 0; but unlike in a geographical distance chart, non-identical pw-patterns can also happen to have a distance of 0. This is the case whenever two pwpatterns happen to target domains of the same relative coherence degree. In Table 1, this is illustrated twice, once within a language (Kinnauri), where two distinct pw-patterns (one of final devoicing and one phonotactic constraint) converge on the same domain, and once between languages, where these two patterns happen to target a same-sized domain as a certain assimilation pattern in Burmese.

0

0.25

0

0

0.25

0.1

0.5

assimilation Burmese 624

stress Burushaski 799

strengthening Kinnauri 294

constraint Kinnauri 299

constraint Kinnauri 300

insertion Persian 705

constraint Semelai 890

assimilation Burmese 624

0.25

0.35

0.5

0.25

0.25

0

stress Burushaski 799

0.5

0.1

0.25

0

0

0.5

0.1

0.25

0

strengthening constraint Kinnauri Kinnauri 294 299

0.75

0.15

0

constraint Kinnauri 300

0.6

0

insertion Persian 705

0

constraint Semelai 890

Table 1. Illustration of distance matrix. Numbers behind language and pw-pattern type name are unique identifiers pointing to the specific processes involved (and described in detail in the database).

60 Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

The distribution of phonological word domains: A probabilistic typology

61

The general problem of large distance matrices like the one obtained from our database is that it is virtually impossible to detect, by mere “hand-inspection”, whether certain kinds of elements are closer to each other than others. Therefore, we applied Multi-Dimensional Scaling, which projects the 382 dimensions of the observed distance matrix into lower-dimensional space, keeping the necessary distortions at a minimum. Kruskal’s non-metrical algorithm (as implemented by Venables and Ripley 2002) converged on a two-dimensional solution, with a low stress value of 4.6%. 4.2. Results Figure 4 plots the results of the two-dimensional solution. We add pie chart insets displaying the internal composition of two selected clusters because clusters are very dense where they occur. There is an impression of clusters at regular intervals from left to right, but these are artifacts of the possible distances between coherence degrees which are not fully continuous. Closer inspection by zooming into any of these smaller clusters does not reveal any trends. For example, the third cluster from the left in Figure 4, highlighted by the pie chart inset on the left shows a 40%-preference for pw-patterns defining phonotactic constraints, but such pw-patterns also show up at any other position. This is different with the dense cluster set off to the far right in Figure 4. This cluster shows a 35%-preference for stress-related patterns that is not repeated anywhere else. (The only other pattern with increased frequency in this cluster is again phonotactic constraints.) 4.3. Discussion From a probabilistic point of view, the Prosodic Hierarchy Hypothesis leads one to expect that pw-patterns would tend to cluster into groups of similar coherence so that in any given language the pw-patterns form natural hierarchies of increasing size. Our analysis suggests no such trend in the data. Moreover, we noted earlier that the Prosodic Hierarchy hypothesis does not find statistical confirmation when interpreted in a categorical way, i. e. as predicting exactly one cluster per language. The only pattern that emerges from the MDS analysis is the increased proportion of stress-related pw-patterns in one cluster that is clearly set off from all other patterns. Inspection of the actual coherence degrees of these pw-patterns suggests that the reason for their increased cross-linguistic similarity is most probably that they tend to be systematically larger than those of other pwpatterns. Figure 5 plots the density distributions of word coherence degrees for each pw-pattern. Pw-patterns of types “allomorphy”, “allophony”, “dissim-

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

Figure 4. Multidimensional Scaling results

62

0

1

2

3

0

1

2

. 0

0 .5

streng thening

1 .0

1 .5

. 0

0 .5

stress

constraint

.0 1

1 .5

. 0

.5 0

tone

ee d l tion

1 .0

1 .5

coherence

. 0

0 .5

vowel_ harm ony

insertion

.0 1

1 .5

. 0

.5 0

weakening

antity u q

1 .0

1 .5

size_ related

Figure 5. Density of relative word coherence per phonological type (N = 368). (The x-axis of each panel shows the relative coherence, the y-axis the density of pw-patterns with that coherence degree.)

Density of pw−patterns

3

assim ia l tion

The distribution of phonological word domains: A probabilistic typology

63

64

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

ilation”, “rhythm” and “other process” were removed from the plots because their respective total frequency in the entire dataset was below 5. The panel for stress-related pw-patterns in Figure 5 shows a density peak at higher coherence levels than in any other pw-pattern type, including tone or vowel-harmony patterns. It seems to be fairly common across languages that stress-defined phonological domains are like the one illustrated by Limbu in (3), with c = 1. What seems much less common are stress-related pw-patterns which have a coherence degree c < 1 and which thereby show up as similar to non-stress-related patterns in the Multi-Dimensional Scaling analysis. To be sure, there are cases of stress-related pw-patterns with c < 1, but they are far less common. One example comes from Mon (Austro-Asiatic; see Bauer 1982). This language exhibits five morpheme types: Proclitics, prefixes, infixes, stems, and enclitics. Word-final stress placement singles out the combination of a stem and its prefix or infix: (7)

a. b. c. d.

"lac ‘break down’ p´-"lac (CAUS-break.down) ‘tear down’ k-´-"lÅ/ (cross-CAUS-over) ‘take across’ "pa/ "kÅ "klÅ/ (do CAUS cross.over) ‘make cross over’

In (7a) the monosyllabic stem receives stress as a main syllable. The morphological derivations in (7b) and (7c) result in disyllabic words which take one primary stress on the word-final syllable. (7d) contains a proclitic causative marker kÅ: It is not an affix because it allows gapping of its host stem (e. g. in question-answer pairs), and it is not a separate grammatical word because it doesn’t have the syntactic possibilities of a full word. At the word level, kÅ receives independent stress. This translates into higher acoustic prominence whenever the stress mark coincides with an intonational peak in information structure.

5. A probabilistic universal The findings from the MDS analysis and the inspection of the density distributions of each pw-pattern suggest that there might be a universal trend (i. e. a probabilistic or statistical universal): (8)

Stress-related domains tend to be universally larger than other domains.

A genuine universal must hold independently of genealogical stocks and linguistic areas, i. e. it must be observable to a similar degree in any stock and any area (see Bickel 2008; Dryer 1989). In order to control the potential impact

The distribution of phonological word domains: A probabilistic typology

65

of stock and area onto the distribution of relative word coherence, we submit our data to a multiple regression analysis of a stratified sample of 40 SinoTibetan, Indo-European and Austroasiatic languages, containing a grand total of 246 pw-patterns. The reason for selecting just these stocks is that their geographical distribution allows areal stratification in the sample since all three stocks overlap in South Asia and two stocks (Sino-Tibetan and Austroasiatic) also overlap in Southeast Asia – both regions with well-established linguistic areality. Formulated as a multiple regression problem, (8) predicts that coherence degrees are systematically affected by a binary factor distinguishing between domains that are defined by stress and other domains, independent of area and stock. If there is such an effect, stress-related pw-patterns will have, on average, higher coherence degrees than other pw-patterns, i. e., the mean coherence will systematically differ between the two types. The coherence degree may also be affected by factors of stock and area, but to the extent that (8) is a genuine universal, these factors will not interact with the difference between stress and other pw-patterns. This prediction is captured by the following regression model, where PW-PATTERN is a binary factor defined by the contrast between stressrelated vs. other pw-patterns: (9)

μ (c) = α + β [PW-PATTERN] + γ [STOCK] + δ [AREA]

The factors STOCK and AREA both enter the design as three-level factors, as explained in the following section.

5.1. Coding of stock and area factors In order to achieve sufficient sample sizes for the STOCK factor, we included 11 Austroasiatic, 12 Indo-European and 17 Sino-Tibetan languages. In each case, we chose one representative per sub-branch of the major branches of the stock (e. g. one West Germanic and one North Germanic language), assuming the standard genealogical trees available in AUTOTYP (Bickel and Nichols 1996ff). Where we had access to sufficient data and also knew that there are considerable within-subbranch divergences in the historical phonological developments, we included two languages per sub-branch. We did this in four cases: (i) from the Tibetan sub-branch of Bodish (Sino-Tibetan) we included both Kyirong (Nepal, Central Tibetan) and Dege (Tibet, Eastern Kham); (ii) from the Aslian subbranch of Mon-Khmer (Austroasiatic) we included both Jahai and Semelai; (iii) from the Italic sub-branch of Italo-Celtic, we included both (Colloquial) French and Spanish; and (iv) from the Indo-Aryan sub-branch of Indo-Iranian,

Map 1. Genealogical distribution of sample languages (circles: Indo-European, squares: Sino-Tibetan, diamonds: Austroasiatic; numbers refer to the key in Table 2 on page 68)

66 Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

Map 2. Areal distribution of sample languages (circles: Europe, squares: South and Southwest Asia, diamonds: Southeast Asia; numbers refer to the key in Table 2 on page 68)

The distribution of phonological word domains: A probabilistic typology

67

68

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

Table 2. Key for Maps 1 and 2 key 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

language Irish Spanish French (colloquial) Dutch Swedish Polish Greek (modern) Lithuanian Romani (Sepecides) Armenian (Eastern) Persian Kinnauri Kham Manange Kharia Nepali Tibetan (Kyirong) Hayu Newar (Dolakha) Santali Limbu Garo Khasi Meithei (Manipuri) Burmese Kayah Li (Red Karen) Tibetan (Dege) Mon Jahai Khmu Qiang Semelai Cambodian Pacoh Chrau Mandarin

stock Indo-European Indo-European Indo-European Indo-European Indo-European Indo-European Indo-European Indo-European Indo-European Indo-European Indo-European Sino-Tibetan Sino-Tibetan Sino-Tibetan Austroasiatic Indo-European Sino-Tibetan Sino-Tibetan Sino-Tibetan Austroasiatic Sino-Tibetan Sino-Tibetan Austroasiatic Sino-Tibetan Sino-Tibetan Sino-Tibetan Sino-Tibetan Austroasiatic Austroasiatic Austroasiatic Sino-Tibetan Austroasiatic Austroasiatic Austroasiatic Austroasiatic Sino-Tibetan

The distribution of phonological word domains: A probabilistic typology

69

we included both Romani and Nepali as representatives. Map 1 shows the sample languages coded for genealogical affiliation.6 For the AREA factor, we coded each language of the sample as belonging to either Europe, South Asia or Southeast Asia. We followed the standard area definitions from AUTOTYP7 , except that we assigned Lithuanian to Europe on the account that we have no separate category for northern Eurasia, Sepecides Romani to Europe on the account of the overall distribution of Romani, and Armenian and Persian to South Asia on the account of the long-standing historical links between the Caspian region and South Asia. Map 2 illustrates the areal coding. However, none of these areal coding decisions had any impact on our results. We performed all analyses reported below also on a sample without Romani, Armenian, and Persian, and there were no differences in the results except for a weakening of effects due to the reduced sample size. 5.2. Results For the actual analysis, two languages, Lahu (Sino-Tibetan; Matisoff 1973) and Vietnamese (Austroasiatic; Schiering, Hildebrandt, and Bickel 2007) were excluded because we have found no evidence for pw-patterns in the sense defined here, i. e. phonological patterns that are lexically general and that reference a morphological domain smaller than a phrase or a compound. We first tested the additive model in (9) against a model with interactions between all factors (PW-PATTERN × STOCK × AREA). A Monte-Carlo permutation test8 suggests that there is no significant interaction between any factors. There are significant main effects of STOCK (F(2) = 11.40, p < .001) and PW-PATTERN (F(1) = 21.80, p < .001). AREA, by contrast, has no significant effect (F(2) = 1.64, p = .30). The number of stress-related pw-patterns (19) is much smaller than the number of other pw-patterns (227), and this makes it important to subject the PWPATTERN factor to a Reliability Analysis (Janssen, Bickel, and Zúñiga 2006). This analysis tests the significance of PW-PATTERN in 10,000 randomly permutated samples, where the largest, then the second-largest, then the third-largest etc. c-value from stress-defined pw-patterns is replaced by the grand mean, and 6

The full dataset with all coding and all references is available as an electronic appendix from our website (see Note 2). 7 Again, see our web site for exact definitions and maps. 8 As noted earlier, since our data are not randomly sampled, we applied permutation testing throughout. All results are based on 10,000 random permutations (cf. Janssen, Bickel, and Zúñiga 2006).

70

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

where the smallest, then the second-smallest, then the third-smallest, etc. c-value from the other pw-patterns is replaced by the grand mean. The results of this suggest that PW-PATTERN is retained as a significant factor at an α -level of .01 up to replacing the 5 stress-defined pw-patterns with the largest c-values and up to replacing at least 19 of the other pw-patterns with the smallest c-value.9 Thus, the results obtained here would be retained at a .01 α -level even if up to 5 of the largest stress-defined pw-patterns turned out to be misanalyzed or if they had not ended up in our sample; and the results obtained would be retained if our data grossly underestimated the coherence degrees of at least 19 other pw-patterns. This leaves us with a simple additive model including PW-PATTERN and STOCK as factors. In order to assess the overall model fit, we calculated the coefficients in the model and compared the predictions derived from them to the observed density distributions. For the purposes of coefficient estimations, the STOCK factor was parameterized into two binary factors with Austroasiatic (AA) as the (arbitrary) baseline, i. e. coefficients are estimated as contrasts between Sino-Tibetan (ST) and Austroasiatic and between Indo-European (IE) and Austroasiatic. The resulting model estimates are: (10)

μˆ (c) = .69 + .27 [STRESS vs OTHER] − .30 [IE vs AA] − .14 [ST vs AA]

Figure 6 plots the predictions of this model, together with 95% confidence intervals, over the observed density distributions for the two types of pw-patterns, separately for each stock. While the overall fit of model to data is fairly good, the fit is slightly reduced in the stress-related pw-patterns. This is not surprising given the fact that the number of datapoints is considerably smaller. 5.3. Discussion The regression and reliability analyses provide robust evidence for the hypothesized probabilistic universal. The trend for stress-related pw-patterns to target larger domains than other pw-patterns is independent of linguistic area and stock. While the evidence from our dataset is strong, the sample is clearly limited, and before the effect of phonological type can be fully accepted as a genuine universal, the hypothesis needs to be tested against further samples in other parts of the world. 9

We ran the simulations up to the replacement of 19 pw-patterns, which is the maximum for stress-related patterns. However, it is likely that the findings are retained after replacing many more than 19 non-stress pw-patterns.

Figure 6. Predictions of the model in (10) (black dots for the means, lines for the 95% confidence intervals) and the observed densities of coherence degrees (grey-shaded) across stocks and pw-pattern types.

The distribution of phonological word domains: A probabilistic typology

71

72

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

The statistical results also suggest that genealogical affiliation significantly contributes to the distribution of phonological domains as well. To a significant extent, the kinds of domains found in a language are determined by the individual phonological “signature” that the language has inherited from its family. Since there is no significant interaction between the PW-PATTERN and the STOCK factor however, this means that at the same time, stress-defined pw-patterns are always larger than others across families (as is also evident in Figure 6). In other words, the distribution of pw-patterns is best predicted by appeal to both their phonological type and the historical phonology from which they have developed. Interestingly, the AREA factor has no effect. This is a surprising result because especially Southeast Asia is known to be prosodically fairly homogenous across families. For example, Sino-Tibetan and Austroasiatic have similar tone systems, known to have arisen through language contact in many cases (Matisoff 2001, among others). Apparently, relative word coherence is more faithful to the individual families and has escaped the areal assimilation pressure. The reasons for this are an issue for further research.

6. Conclusions Prosodic Phonology assumes a finite list of prosodic domain types which are hypothesized to be construed in accordance to the Strict Layer Hypothesis. Prosodic domains are conceived as pre-existing entities which can be discovered by phonological patterns which reference them within or across languages. The theory thus states a number of absolute universals and makes numerous predictions with respect to the manifestation and architecture of prosodic structure. From a cross-linguistic perspective, strict adherence to the methodological procedure provided by the theory yields contradictory results in a number of cases (see Schiering, Hildebrandt and Bickel 2007 for extensive discussion). In this study, we applied quantitative methods in order to establish what, if any, principles guide the distribution of phonological word domains across languages. In sharp contrast to the predictions of the Prosodic Hierarchy framework, we find a cross-linguistic trend for languages to multiply prosodic domains between the foot and the phrase. However, the observable diversity is not without limits. We find tentative statistical support for the following probabilistic universal, which remains to be tested in other samples from other parts of the world: (11)

Stress-defined domains tend to be significantly larger than other domains.

The distribution of phonological word domains: A probabilistic typology

73

Compared to the huge body of absolute universals enshrined in the Prosodic Hierarchy, this universal seems to have small scope and to be rather local. However, this probabilistic universal is empirically founded and, at least in our sample, stands the test for genealogical and areal bias. No other pw-pattern has a systematic impact on domain size. This is particularly noteworthy with regard to pw-patterns of vowel harmony which, like stress, are intrinsically “relational” in the sense that they both involve syntagmatic relationships between smaller units (syllables, morae, vowels). Despite this, vowel harmony pw-patterns do not cluster with stress patterns, nor do they show similar degrees of coherence. Tonal pw-patterns are also different from stress with regard to coherence: Although both are “suprasegmental”, there is no trend for the two pw-patterns to show similar coherence degrees. The finding that stress-defined domains behave differently from other wordrelated domains is compatible with pre-generative conceptions of prosodic structure. Pike (1945), for example, used the hierarchically organized domains phoneme, syllable, stress group, pause group and breath group in his analysis of American English. The last three of these are designated to provide domains for stress and intonation phenomena which are necessarily included in hierarchical structures. It is a task of further research to flesh out a theory of hierarchical structure in strictly prosodic domains. In this light, the present study illustrates how quantitative methodologies that are standardly used in other disciplines yield testable cross-linguistic generalizations. These generalizations in turn form a robust empirical foundation for theory construction that accounts for diversity as much as for universality. Hopefully, future investigations into deviations and universals of prosodic structure will further acknowledge the typological variation to be found in the languages of the world. Abbreviations 1 first person; 2 second person; 3 third person; A subject of transitive clause; ADD additive focus; ASS assertive; CAUS causative; GEN genitive; INF infinitive; NEG negative; NOM nominative; ns non-singular; p plural; P patient; PASS passive; POSS possessive; PST past; PTPCL participial; Q interrogative; QUOT quotative; s singular; S subject; VOC vocative Acknowledgements We gratefully acknowledge our student assistants’ help in data collection and databasing: Thomas Goldammer, Franziska Crell, Sven Siegmund, Taras Zakharko, Jenny Seeg, Sebastian Hellmann, and Josh Wilbur. Much of our understanding of the Prosodic Hierarchy has profited from discussions with Tracy Hall. We are also grateful to useful comments on an earlier draft by two anony-

74

Balthasar Bickel, Kristine A. Hildebrandt, and René Schiering

mous reviewers, and to Mathias Jenny for helping us with the Mon data. This research was funded by the DFG under research grants BI 799/2 and 799/3. All statistical analyses and plots were done in R 2.6.1 (with added packages cluster, coin, lattice, MASS, maps and trotter; R Development Core Team 2007ff). Previous versions of this paper were presented at the Workshop “Prosodic Domains: Universals and Deviations“ at the 29th Annual Meeting of the Deutsche Gesellschaft für Sprachwissenschaft, Siegen, February, 2007, at the 15th Manchester Phonology Meeting, May, 2007 and at the 7th Biannual Conference of the Association for Linguistic Typology, Paris, September, 2007.

References Bauer, Christian 1982 Morphology and Syntax of Spoken Mon. London: SOAS. Bickel, Balthasar 2008 A refined sampling procedure for genealogical control. Sprachtypologie und Universalienforschung 61: 221–233. Bickel, Balthasar and Johanna Nichols 1996ff The AUTOTYP database. Electronic database; http://www.uni-leipzig.de/~autotyp. Cox, Trevor F. 1994 Multidimensional Scaling. New York: Chapman and Hall. Driem, George van 1987 A Grammar of Limbu. Berlin: Mouton de Gruyter. Dryer, Matthew S. 1989 Large linguistic areas and language sampling. Studies in Language 13: 257–292. Ebert, Karen H. 1994 The Structure of Kiranti Languages. Zürich: ASAS Press. Everitt, Brian S. and Torsten Hothorn 2006 A Handbook of Statistics Using R. Boca Raton, Florida: Chapman and Hall. Hildebrandt, Kristine 2007 Prosodic and grammatical domains in Limbu. Himalayan Linguistics Journal 8: 1–34. Hyman, Larry M., Francis Katamba, and Livingstone Walusimbi 1987 Luganda and the strict layer hypothesis. Phonology Yearbook 4: 87– 108. Inkelas, Sharon and Draga Zec 1995 Syntax-phonology interface. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 535–549. London: Blackwell.

The distribution of phonological word domains: A probabilistic typology

75

Janssen, Dirk, Balthasar Bickel and Fernando Zúñiga 2006 Randomization tests in language typology. Linguistic Typology 10: 419–440. Matisoff, James A. 1973 The Grammar of Lahu. Berkeley: University of California Press. 2001 Genetic vs. areal linguistics in Southeast Asia: Prosodic diffusibility in Southeast Asian languages. In: Aleksandra Y. Aikhenvald, and Robert M.W. Dixon (eds.). Areal Diffusion and Genetic Inheritance, 291–327. Cambridge: Cambridge University Press. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Peperkamp, Sharon 1996 On the prosodic representation of clitics. In: Ursula Kleinhenz (ed.), Interfaces In Phonology, 102–127. Berlin: Mouton de Gruyter. 1997 Prosodic Words. Amsterdam: Holland Academic Graphics. Pike, Kenneth L. 1945 The Intonation of American English. Ann Arbor: University of Michigan Press. R Development Core Team 2007 R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing (www.r-project.org). Schiering, René, Kristine Hildebrandt and Balthasar Bickel 2007 The prosodic word is not universal. Ms., http://www.uni-leipzig.de/ ~autotyp/download/SchieringHildebrandtBickel2007.pdf. Selkirk, Elisabeth O. 1984 Phonology and Syntax: The Relation between Sound and Structure. Cambridge, Mass.: MIT Press. 1995 The prosodic structure of function words. In: Jill N. Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.). Papers in Optimality Theory, 439–469. Amherst: GLSA. Tumbahang, Govinda Bahadur 2007 A descriptive grammar of Chhatthare Limbu. Ph. D. diss., Tribhuvan University. Venables, William N. and Brian D. Ripley 2002 Modern Applied Statistics with S. New York: Springer. Weidert, Alfons and Bikram Subba 1985 Concise Limbu Grammar and Dictionary. Amsterdam: Lobster Publications.

Part 2. Adjunction, Recursion, and the Nature of Syntax-Phonology Mapping

Phrase-level and word-level syllables: Resyllabification and prosodization of clitics Anna Cardinaletti and Lori Repetti

0. Introduction Prosodic Theory and, in particular, the Strict Layer Hypothesis (SLH), maintain that each level of the prosodic hierarchy is fully contained within the level dominating it (Nespor and Vogel 1986; Selkirk 1995). However, in some languages a syllable is not necessarily contained within a single Prosodic Word (PW). In this paper, we investigate two such cases of violation of the SLH, and offer a new analysis of the facts. First, we examine data involving the prosodization of Romance “free” clitics, or clitics that are incorporated into prosodic structure at the level of the Phonological Phrase (PP) (Selkirk 1995). The northern Italian dialect of Donceto has free subject clitic pronouns. As we can see in (1) within Donceto clitic + host units, the syllable boundaries are clear: there is a single syllable in each case. (1)

/l/ + /E/ > [lE] /E/ + /l/ > [El]

‘he is’ ‘is he?’

But where are the Prosodic Word (PW) boundaries in (1)? Since free clitics, by definition, are not part of the same PW as the host, the clitic + host unit does not form a PW. None of the representations in (2)–(3) capture the fact that clitic + host form a single syllable, but the clitic /l/ is not part of the PW. 1

1

Although /E/ ‘is’ might not be considered to be a PW, we use it to illustrate our point. The same syllabification issues arise with other vowel-final verbs with enclitics (for ex., /be:v´l/ ‘does he drink?’), but we are not able to test our claims using other vowel-initial verbs since there are no vowel-initial verbs in this dialect. Common Romance vowel-initial verbs are consonant-initial in this dialect: It. amare = /vu "le bEN/ ‘to love’, It. uscire = /na føra/ ‘to exit’, It. incontrare = /tru "va/ ‘to meet’, etc.

80

Anna Cardinaletti and Lori Repetti

(2)

PP

PW

PW

σ

σ

E

l (3)

PP

l

E

PP

PP

PW

PW

σ

σ

E

l

E

l

Second, we examine phrasal resyllabification, or the situation in which a single syllable contains elements belonging to two different words. 2 For example, a word-final consonant may be resyllabified as the onset of the following vowelinitial word, as illustrated in the data from Spanish in (4).3 This particular process is attested in many Romance languages, Turkish, Korean, Arabic, Indonesian, etc. (4)

club elegante ‘elegant club’ [klu.ße.le.Gan.te]

While the syllable boundaries are uncontroversial, the Prosodic Word (PW) boundaries are not so clear. Some possibilities are suggested in (5). (5)

2

[klu.ß#e.le.Gan.te] [klu.#ße.le.Gan.te] [klu.ße.#le.Gan.te]

We do not address the topic of (re)syllabification within morphologically complex words, only resyllabification across word boundaries. 3 In addition to speaker intuition, segmental processes in Spanish varieties provide evidence of resyllabification. For example, in the Chinato dialect of Western Spain, certain consonants are aspirated in coda position. Those consonants aspirate in wordfinal position before a consonant-initial word (among other contexts), but not before a vowel-initial word (Hualde 1991). These facts have been interpreted as evidence that a word-final consonant before a vowel-initial word is resyllabified to onset position and, therefore, is not subject to aspiration. Similar patterns are attested in the Spanish dialect of Coria (Cummins 1974) and in some varieties of Buenos Aires Spanish (Goldsmith 1981: 6, Kaisse 1999).

Phrase-level and word-level syllables

81

We propose that the resyllabification facts and the prosodization of free clitics can be best accounted for by adopting a model of the prosodic hierarchy in which there is an additional phrasal syllable level nested under the PP-level. In other words, there are two syllable representations: the syllable that is embedded under the PW (σ PW ) and an additional “phrasal” syllable (σ PP ). (6)

a.

clitic prosodization

b.

resyllabification

PP

clitic

PP

PW

PW

PW

σ PW

σ PW

σ PW

host

σ PP

X

X

σ PP

X

X

X

σ PP

We conclude by analyzing some restrictions on the types of syllables that can be formed in resyllabification and clitic prosodization contexts. In neither of these situations can a consonant resyllabify to form part of a complex onset or coda. Compare (7a) and (8a) in which a word-final consonant syllabifies into onset position, with (7b) and (8b) in which a word-final consonant cannot syllabify to form part of a complex onset. We argue that these facts, too, can be best captured by making reference to phrasal syllables. (7)

Catalan clitic prosodization a. b.

(8)

reb-ho ‘receive it (indef.)’ reb-la ‘receive her’

/rEb/ + /o/ > [rE.ßu] /rEb/ + /la/ > [rEb.la] *[rE.ßla] (N. B. [ßl] is an acceptable onset)4

Spanish resyllabification a. b.

club elegante ‘elegant club’ club lindo ‘pretty club’

[klu.ße.le.Gan.te] [kluß.lin.do] *[klu.ßlin.do] (N. B. [ßl] is an acceptable onset)

82

Anna Cardinaletti and Lori Repetti

This paper is organized as follows. First, in section 1, we discuss the prosodization of free clitics. We show that the traditional prosodic hierarchy cannot handle all the data, and we introduce a new ‘phrasal’ syllable level which is independent of the PW and nested directly under the Phonological Phrase (PP). In section 2, we review previous analyses of resyllabification, point out problems with these approaches, and show that phrasal syllables can account for these facts as well. We then investigate the restrictions on clusters in these two contexts (section 3). We conclude the paper in section 4. 1. Free clitics Clitics may be prosodized differently in different languages and in different contexts (Selkirk 1995). Four types of prosodized clitics are represented below. We do not consider the ‘clitic group’ in this paper (Nespor and Vogel 1986), although it bears a similarity to the model presented in this paper, which we will point out in section 4. We use examples involving proclitics, although the same structures are possible with enclitics. (9)

“free clitic”

clitic

“internal clitic”

PP

PP

PW

PW

host

“affixal clitic”

clitic

“prosodic word”

PP

clitic

host

PP

PW

PW

PW

PW

clitic

host

host

The representation of free clitics presents a challenge to the prosodic hierarchy since a free clitic may form a syllable with its host, but it cannot be part of the same PW as the host (2)-(3). We investigate this puzzle using data from the northern Italian dialect of Donceto. We begin by showing that the dialect 4

/b/ > [ß] in onset position after a vowel (Bonet and Lloret 2005a).

Phrase-level and word-level syllables

83

of Donceto has free subject clitics (section 1.1), which, we claim, can best be described using a new model of the prosodic hierarchy which includes phrasal syllables (section 1.2). 1.1. Donceto subject clitics The dialect of Donceto, which is typical of many northern Italian dialects, has three preverbal subject clitics in declarative sentences, and six postverbal subject clitics in interrogative sentences (see Cardinaletti and Repetti 2004, 2008). We illustrate this using the present indicative forms of the verb ‘drink’ in (10). (10) 1sg 2sg 3sg 1pl 2pl 3pl

declarative "be:v @t- "be:v @l- "be:v´ bu"vum bu"vi i- "be:v@n

interrogative "be:v -j@ "be:v -@t "be:v@ -l bu"vum -j@ bu"vi: -v "be:v@n -j@

Do constraints applying to the PW in Donceto affect clitic + host units? If they do, we can conclude that Donceto subject clitics are “internal” or “affixal”. If they do not, we conclude that Donceto subject clitics are “free”.5 An obstruent voicing assimilation rule is active in the Donceto dialect: two adjacent obstruents of different voice features assimilate to the voicing of the second obstruent. In the following examples we see that certain unstressed vowels may be deleted, leaving the consonants flanking it adjacent. If the adjacent consonants are obstruents, they must have the same voicing feature. (11)

["vEtS] / [(@)f"tSaja] (*[(@)v"tSaja]) ["peza] / [(@)b"zæ] (*[(@)p"zæ])

‘old’/‘old age’ ‘scale’/‘to weigh’

This type of assimilation is found within PWs (11), but crucially not across PWs (12) or at clitic boundaries (13).6

5

6

Subject enclitics in other northern Italian dialects similar to those in (10) have been analyzed by some scholars as verbal affixes (Benincà and Vanelli 1982; Fava 1993; Goria 2004; Poletto 2000; Rohlfs 1968; Zamboni 1974; etc.), implying that the verb + enclitic unit is a PW (i. e., subject clitics are “internal” or “affixal”). Cardinaletti and Repetti (2008) use various diagnostics to show that the postverbal material in (10) cannot be considered as an inflectional suffix, i. e., as word-internal. In rapid speech in Spanish and French, word-final obstruents agree in voicing with the following word-initial segment (Harris 1980; Casagrande 1984).

84

Anna Cardinaletti and Lori Repetti

(12)

/gat/ + /bEl/ > ["gat "bEl] (*["gad "bEl])

‘beautiful cat’

(13)

/t/ + /be:v/ > [@t "be:v] (*[@d "be:v])

‘you:sg drink’

Another example of the difference between PWs and host + clitic units involves consonant clusters. Donceto PWs may contain onset clusters and coda clusters; however, in verb + clitic units they are restricted, as illustrated in the chart in (14). (See section 3 below for further discussion.) (14) PW clitic + host

onset clusters [tri] ’three’ *[t-rõ:f] ‘you:sg snore’ ([@t-rõ:f])

coda clusters [ust] ‘August’ *[pas-t] ‘do you:sg pass?’ (["pas-@t])

Evidence that the subject clitic in the examples in (14) is lexically /t/ and not /@t/ comes from the fact that when adjacent to a vowel, the clitic surfaces as [t], suggesting that the schwa in (14) is epenthetic. (15)

t-e bu"vi:d e-t bu"vi:d

*@t-e bu"vi:d *e-@t bu"vi:d

‘you:sg have drunk’ ‘have you:sg drunk?’

Furthermore, verbal suffixes affect stress assignment, but enclitic pronouns do not, as seen in the following data from Padua (a related northern Italian dialect spoken in the Veneto region). (16) verb stem + suffix YES máñ+i ‘you:sg eat’ mañ+émo ‘we eat’

a.

Does the postverbal material affect stress?

b.

Is the ‘three YES syllable window’ teléfon+i ‘you:sg call’ respected? telefon+émo ‘we call’ (*teléfon+emo)

verb + enclitic pronoun NO máñi ‘you:sg eat’ máñi-to ‘do you:sg eat?’ (*mañí-to) NO teléfoni-to ‘do you:sg call?’ (*telefóni-to; *telefoní-to)

In (16a) we see that inflectional suffixes affect the position of stress on the verb, while the addition of enclitic pronouns does not. In (16b) we see that the familiar ‘three syllable window’ of stress assignment (whereby stress falls on one of the

Phrase-level and word-level syllables

85

final three syllables) is respected within verb stem + suffix units, but not within verb + enclitic pronoun units.7 If the clitic pronouns were part of the same PW as the verb, we would expect the processes described above to apply to the verb + clitic unit; however, this is not what we find. We conclude that clitics are not part of the same PW as the host, therefore, they cannot be internal clitics or affixal clitics. Furthermore, Donceto subject clitics cannot be independent Prosodic Words (PW) since they do not meet word minimality requirements, they cannot be stressed, they cannot appear in isolation, etc. In Cardinaletti and Repetti (2008) we conclude that they are “free clitics”, meaning that they are adjoined to their host at the level of the PP.8 1.2. Prosodic representation of free clitic + host units We have determined that Donceto subject clitics can be described as free clitics. How can we represent the prosodization of these clitics? In (17) (using the 2sg forms) we represent one possibility inspired by Peperkamp (1997) in which the PW boundaries are redrawn when a clitic is added. (See § 2.1 for further discussion of Peperkamp 1997.) (17)

σ

@

t

b

PP

PP

PW

PW

σ

σ

e:

v

b

σ e:

v

@

t

The model in (17) takes the PW to be either [be:v] (with proclitic) or [be:] (with enclitic), and can handle the representation of these data. However, the 7

8

Note that in Italian the ‘three syllable window’ is violated with some verb stem + 3pl suffixes: telèfonano ‘they telephone’. However, this is not the case in Paduan, Donceto, or any of the other northern Italian dialects that we have considered. Some scholars have analyzed Romance object clitics in the same way, for example, Italian object clitics are analyzed as “free” clitics by Peperkamp (1997) and Bafile (1991–1992). In other works, however, Romance object clitics are taken to be “internal clitics” (Bonet and Lloret 2005a for Catalan; Bafile 1991–1992 for Neapolitan; Monachesi 1996 for Italian and Neapolitan single clitics, but not clitic clusters, which are taken to be PWs), “affixal clitics” (Peperkamp 1997 for Neapolitan; Loporcaro 2000 for all Romance object clitics), or part of the “clitic group” (Nespor and Vogel 1986).

86

Anna Cardinaletti and Lori Repetti

prosodic representation of other clitic + host structures cannot be described in as straightforward a manner. Consider the Donceto data in (1). Using the “traditional” model, neither the representation in (2) for the form with the proclitic ([lE]), nor the representation in (3) for the form with the enclitic ([ El]) captures the fact that clitic /l/ is part of the same syllable as the host verb / E/, but it is not part of the PW (since it is a free clitic; see also footnote 1.) We propose that in phrasal contexts, there are two syllable representations: the syllable that is a daughter of the PW (σ PW ), and an addition ‘phrasal’syllable (σ PP ) that is nested directly under the PP level (18).9 (18)

PP PW

σPW l

E σPP

This structure captures the fact that free clitic prosodization is a phrasal phenomenon, and it correctly shows that the clitic /l/ is PW-external, and that [.l E.] constitutes a syllable. We will now see that the representation in (18) does not violate the basic principles of the Prosodic Hierarchy. The Prosodic Hierarchy is subject to the Strict Layer Hypothesis (Nespor and Vogel 1986) (19), which can be restated as a series of constraints on Prosodic Domination (Selkirk 1995) (20). (19)

Strict Layer Hypothesis a. a given non-terminal unit is composed of one or more units of the immediately lower category b. a unit of a given level is exhaustively contained in the superordinate unit of which it is part

(20)

Prosodic Domination Layerdness: no Ci dominates Cj, j > i Headedness: any Ci must dominate Ci-l Exhaustivity: no Ci immediately dominates Cj, j < i-l Non Recursivity: no Ci dominates Cj, j=i

9

The two syllable levels provide domains of certain constraints, as will be illustrated in section 3. For reasons of space, we ignore the question of foot structure.

Phrase-level and word-level syllables

87

Peperkamp (1997:35–36) points out that none of the Prosodic Domination constraints in (20) concern the second clause of the Strict Layer Hypothesis. She suggests that the second clause of the Strict Layer Hypothesis can be interpreted as alignment constraints. For example, in order to require a syllable to be fully contained within a PW, she proposes an alignment constraint requiring PW edges to align with syllable edges. (21)

Proper Nesting Align(PW, L; σ , L) Align(PW, R; σ , R)

How does the representation in (18) fare against the principles and constraints outlined above? Since the phrasal syllable level is a daughter of the PP level (not the PW level), it does not violate any of the principles in (19) nor any of the constraints in (20) or (21). While this model is a change from the standard model, it allows us to capture the process of free clitic prosodization – a phraselevel process – without violating the time-honored principles of the Prosodic Hierarchy. Furthermore, as we will see in the next section, it also allows us to account for certain facts having to do with resyllabification that cannot be accounted for within the traditional model. 2. Resyllabification In this section we will see that many previous approaches to resyllabification have not addressed the question of the prosodic representation of the resyllabified forms. Peperkamp (1997) is a notable exception. However, her solution – that PW boundaries are redrawn in resyllabification contexts – runs into a number of problems, namely that PW constraints do not necessarily apply to the ‘redrawn’ PWs (section 2.1). We then show that the new model of the prosodic hierarchy which includes phrasal syllables can handle these facts (section 2.2). 2.1. Previous analyses The term “resyllabification” implies a derivational approach to syllable structure since there must be an initial “syllabification” in order for there to be a subsequent “resyllabification”. Within a derivational approach to phonology, phraselevel syllabification rules (i. e., “resyllabification”) are ordered after word-level syllabification rules (Kenstowicz 1994). (“Resyllabification” is commonly held to be bound by the phrasal domain [Nespor and Vogel 1986].) (See footnote 2.) (22)

club elegante

kluß # e.le.Gan.te klu.ße.le.Gan.te

(lexical level) (post-lexical level)

88

Anna Cardinaletti and Lori Repetti

More recent work on resyllabification is framed within the formalism of Optimality Theory (OT). Typical of this approach is Face (2002) who proposes the following OT analysis. Three well-attested constraints – Onset, Align, NoCoda – ranked in a particular order allow us to account for the resyllabification of a word-final consonant as a simple onset (as well as other resyllabification facts). (23)

Onset – syllables must have an onset Align(PD, σ ) – every left edge of a phonological domain should align to the left edge of a syllable (Face’s alignment constraint refers to “phonological domains” in order to account for the (re)syllabification of prefixes.) NoCoda – syllables must not have codas

(24)

Onset  Align(PD, σ )  NoCoda

(25)

club elegante a. [kluß# .e.le.Gan.te] b. [klu.ß# e.le.Gan.te] 

Onset *!

Align *

NoCoda ** *

In (25a) the non-resyllabified form violates high-ranked Onset, while the candidate in (25b) does not, even though it violates lower-ranked Align. While both the derivational and the OT type of approach successfully account for resyllabification, neither approach deals directly with the prosodic representation of the resyllabified form. Face’s analysis suggests that in the winning candidate in (25) the PW boundary and the syllable boundary are misaligned, but he does not offer an explanation of how the diacritics (i.e., the period representing syllable boundaries, and the pound sign representing PW boundaries) would translate into a prosodic representation. The representation in (26) is one possible interpretation, but it is ill-formed since the middle syllable is not contained within a PW. (26)

PP PW

PW

σ

σ

σ

klu

ße

le . . .

Peperkamp (1997) directly addresses the prosodic representation of resyllabified structures. She proposes various possible analyses, represented in (27)–(29):

Phrase-level and word-level syllables

(27)

89

PP PW

σ

σ

σ

klu

ße

le . . .

(28)

PP PW

PW

σ

σ

σ

klu

ße

le . . .

(29)

PP PW

PW

σ

σ

σ

klu

ße

le . . .

Peperkamp points out the problems in these structures. In (27) she notes that one PW is eliminated, but there are still two main word stresses implying the existence of two prosodic words, and in (28) Strict Layering is violated since the second syllable is not exhaustively contained within a single PW. Instead, Peperkamp proposes that PW boundaries are readjusted when resyllabification takes place, so that the resyllabified consonant (an onset) is part of the second PW. The result is that PW boundaries do not coincide with morphological word boundaries (29). Note that the representation proposed by Peperkamp in (29) does not violate any of the principles of the SLH or the constraints on Prosodic Domination or Proper Nesting. However, there are some problems with this structure. Peperkamp (1997:30) notes there is no longer a direct correspondence between the morphological word and the prosodic word, but states that since “morphological structure and prosodic structure are represented on separate planes . . . this does not result in an ill-formed representation.” A more serious problem, however, is that constraints applying within the domain of the PW do not necessarily apply to the “resyllabified” PWs. In the following paragraphs we discuss two constraints applying to Italian PWs which are violated in “resyllabified”

90

Anna Cardinaletti and Lori Repetti

structures.10 The first constraint is that PWs cannot end in an obstruent, 11 and the second is that PWs must be minimally bimoraic. Consider the following Italian phrase: piatto sporco ‘dirty dish’ [pját.tos. p´Or.ko] in which the initial /s/ + consonant cluster of sporco resyllabifies as a coda + onset sequence. Following Peperkamp, the PW boundaries would be analyzed as [pját.tos. # .p´Or.ko]. However, the first PW violates the well-formedness constraint banning word-final obstruents in Italian. A similar situation arises in the well-studied phonological process known as raddoppiamento sintattico in which a word-initial consonant is lengthened if the preceding word ends in a stressed vowel (Bullock 1991; Loporcaro 1997; Repetti 1991): città vecchia ‘old city’ [tSit.táv.v´Ek.kja]. Again, if we analyze the PW boundaries as readjusted to coincide with syllable boundaries [t Sit.táv. # .v´Ek.kja], the first PW violates the ban on final obstruents. Another example involves the five Italian consonants that are always long in intervocalic position (word-internally and across word boundaries), and short elsewhere: /S/, /L/, /ñ/, /ts/, /dz/. We illustrate these patterns using words with /ts/. (30) a. b. c. d.

short /ts/ long /ts/ utterance-initial [tsio] *[tstsio] post-consonantal [martso] *[martstso] intervocalic (word-internal) *[pitsa] [pitstsa] intervocalic (word-initial) *[miotsio] [miotstsio]

zio marzo pizza mio zio

‘uncle’ ‘March’ ‘pizza’ ‘my uncle’

These consonants are commonly represented as being underlyingly long or moraic (Burzio 1989; Chierchia 1986; Davis 1990; Vanelli 1992), and their mora (or length) can be realized only in intervocalic position (30c–d), but not in absolute initial position (30a) or after a consonant (30b) since Italian syllables cannot begin or end with a long consonant or certain consonant clusters such as /rts/ in [30b]. How are we to analyze the last example in (30)? If, following Peperkamp (1997), we analyze the PW divisions as [mi.ots. # .tsi.o], this representation is ill-formed because the first PW ends in an obstruent. 10

For more on the PW in Italian, see Nespor (1985); Nespor and Vogel (1986); van Oostendorp (1999); and Peperkamp (1997). 11 The constraint banning word-final obstruents refers to PWs and not lexical words. Evidence of this comes from the fact that in the past, as well as in some modern varieties, borrowings ending in an obstruent were/are generally adapted by epenthesizing a PW-final vowel: beef steak > bistecca, stop [stOp]/[stOppe] in Florentine Italian, etc. (See Bafile 2002.) However, many relatively recent neologisms are obstruent-final (prof, gas, jeep, etc.).

Phrase-level and word-level syllables

91

Can the constraint on bimoraic word minimality be violated in resyllabification contexts, such as in the phrase bar elegante ‘elegant bar’ [bá.re.le.gán.te]?12 If, following Peperkamp, we posit the word boundaries as – [bá. # .re.le.gán.te] – the first PW incurs a violation of the bimoraic word-minimality constraint. Evidence that the first PW ([ba]) is not bimoraic comes from the fact that it does not induce raddoppiamento sintattico (*[bár.re.le.gán.te]) and the vowel of the first syllable is not lengthened (*[bá:.re.le.gán.te]).13 Within a constraint-based model of phonology, like Optimality Theory, we might be able to account for these phonological violations by ranking the constraints requiring Proper Nesting (21) – i.e. the constraints requiring alignment of PWs and syllables – higher than the constraints requiring alignment of prosodic words and lexical words and the markedness constraints banning monomoraic prosodic words and PW-final obstruents. (31)

Align(PW, σ ) – align the L/R edge of the PW to the L/R edge of a syllable Word-Alignment – align the L/R edge of every lexical word to the L/R edge of every prosodic word (Selkirk 1995) Bimoraic Word Minimality – a PW must not be monomoraic *Final-Obs – a PW must not end in an obstruent

(32)

Align(PW, σ )  Word-Align, Bimoraic-Word-Min, *Final-Obs

We illustrate this approach using the phrases bar elegante (33) and cittá vecchia (34). (The symbol  after a candidate indicates that it is incorrectly chosen as the winner.) (33) -------

bar elegante Align(PW, σ ) Word-Align Bi-Word-Min. a. [ba.r # e.le.gan.te] *! b. [ba. # .re.le.gan.te]  * * 12

13

Other examples of this type include tir italiano [tí.ri.ta.ljá.no] ‘Italian truck’, CAR aretino [ká.ra.re.tí.no] ‘Aretino military training center’, etc. We only found examples of monosyllables ending in /r/ which pattern this way. Polysyllablic oxytones ending in /r/ and all other consonant-final lexical items undergo final consonant lengthening Er.to] ‘open bazaar’, tram in this context, for example, bazar aperto [badz.dzár.ra.p´ Et.tri.ko], etc. (see also Chierchia 1986). elettrico [trám.me.l´ However, we do not generally expect vowel lengthening of word-final stressed vowels. See Repetti (1989) for the bimoraic stressed syllable requirement in Italian, and D’Imperio and Rosenthall (1999) for the phonetics and phonology of stress in Italian.

92

Anna Cardinaletti and Lori Repetti

cittá vecchia a. [t∫ it.tá. # .vEk.kja]  b. [t∫ it.táv. # .vEk.kja]

Align(PW, σ )

Word-Align *!

-------

(34) *Final-Obs *

In (33a) the PW is not aligned with a syllable, a form which cannot be represented using the traditional prosodic hierarchy (see [26] above), and is correctly ruled out. The candidate in (b) with the redrawn word boundaries is the winner. This ranking makes the wrong prediction in the other tableau. Since candidate (34a) incorrectly emerges as the winner. Any other high-ranked constraint proposed to rule out candidate (34a) – such as a constraint on faithfulness to input moras or a constraint requiring bimoraic stressed syllables – would then incorrectly rule out the winning candidate in (33b). Given the problems with the prosodic representations of these resyllabified phrases, we will explore the new approach involving a “phrasal” syllable level. 2.2. New analysis We propose that in resyllabification contexts (which, as we have noted, are phrase-level contexts), there are two syllable representations: The syllable level that is a daughter to the (σ PW ), and an additional “phrasal” syllable level that is nested directly under PP (σ PP ). The structure in (6b) captures the fact that resyllabification is a phrasal phenomenon, it does not violate the basic principles of the Prosodic Hierarchy, and it allows us to solve the problems discussed in section 2.1. How are phrasal syllables formed? Phrasal syllables are identical to the PW syllables with the following differences: Violations of Onset are avoided through “resyllabification” at the σ PP level, adjustments to syllable structure due to segmental or metrical constraints (such as the Italian examples discussed above) are represented here, and an unprosodized element (such as a free clitic) is incorporated into the prosodic hierarchy at the σ PP level. Crucially, word-level constraints do not apply to phrasal syllables. Using this model, we can account for the Spanish and Italian data discussed above. The Spanish resyllabification data can be handled using this model since, for example, the /b/ in club elegante is a coda at the σ PW level, but an onset at the σ PP level.14 14

In some varieties of Spanish we have spirantization of coda /s/ in resyllabification contexts, while in other varieties we do not. These facts can be handled as follows: spirantization of coda /s/ may apply at the σ PW level (for example mes es [meh

Phrase-level and word-level syllables

(35)

93

PP PW

PW

σPW

σPW

σPW . . .

e

le . . .

klu

ß

σPP

σPP

σPP

In addition, the Italian data which constituted problems for Peperkamp’s analysis are readily accounted for using this model. Since the σ PW level remains independent of the σ PP level, none of the Italian PW-level constraints are violated in the ‘resyllabified’ contexts. For example, in (36a) bar elegante we see that the first PW satisfies the word minimality requirement, even though at the phrasal level the word-final /r/ is resyllabified as an onset. In (36b) città vecchia we see that at the σ PP level the /v/ spreads to form part of the coda of the first word, but at the σ PW level the first word does not end in an obstruent (although it may end in an empty mora, see Bullock 1991 and Repetti 1991). (36)

a.

bar elegante PP PW

PW

σPW

σPW

σPW . . .

e

le . . .

ba

σPP

r

σPP

σPP

eh] ‘month is’) or at the σ PP level (for example, mes es [mes eh]). The constraints banning coda /s/ in the various domains (σ PW and σ PP ) would be ranked relative to each other and to input-output faithfulness constraints. For dialect variation in Spanish spirantization, see Kaisse (1999) and footnote 3.

94

Anna Cardinaletti and Lori Repetti

b.

città vecchia PP PW

PW

σPW

σPW

tSit

ta

σPP

σPP

v

σPW

σPW

Ek

kja

σPP

σPP

3. Clusters By examining data from various Romance languages, we have determined that the process of resyllabification and the incorporation of free clitics into prosodic structure are both phrasal phenomena that can best be characterized by adopting a model of the prosodic hierarchy with two orthogonal syllable tiers. Another characteristic shared by these two processes (i. e., resyllabification and clitic prosodization15 ) has to do with restrictions on the types of syllables that result. Within resyllabification contexts, a word-final consonant can resyllabify as the onset of the following word, but not if the result would be a complex onset (37). (In this and all subsequent examples, all candidate onset and coda clusters are acceptable in that language.) (37)

CVCα #Cβ VC > *CV.Cα Cβ VC

For example, a word-final /b/ cannot resyllabify as the onset of the following consonant-initial word, even if the resulting onset cluster is acceptable, as illustrated in the following data from Spanish. (38)

club lindo ‘pretty club’ *[klu.ßlin.do]; [kluß.lin.do]

This restriction on complex onsets has been accounted for within derivational approaches by saying that post-lexically only simple onsets can form. So, while

15

In this section, the type of clitic is irrelevant. The constraints on syllabification seem to apply to all clitics equally, regardless of whether they are characterized as free, internal, or affixal.

Phrase-level and word-level syllables

95

the word-final /b/ of club can resyllabify as the simple onset of a following vowel-initial word, it cannot resyllabify as part of a complex onset (Kenstowicz 1994). (39) lexical level post-lexical level (resyllabification to form simple onsets, not complex onsets)

club elegante kluß # e.le.Gan.te klu.ße.le.Gan.te

club lindo kluß # lin.do –

This restriction on clusters has also been successfully accounted for within Optimality Theory. Using the constraints and ranking in (23) and (24), Face (2002) shows that clusters are avoided in Spanish resyllabification. In (40), Onset is not violated in either candidate, so the alignment constraint eliminates the second candidate, leaving the non-resyllabified candidate as the winner. (40)

club lindo a. [kluß# .lin.do]  b. [klu.ß# lin.do]

Onset

Align *!

NoCoda ** *

Bonet and Lloret (2005a, 2005b) notice that the syllabification of Catalan clitics (which they analyze as ‘internal’ clitics, meaning that the clitic + host unit forms a single PW) follows the same general principles as resyllabification: the consonantal portion of a clitic can syllabify as a simple onset or coda, but it does not adjoin to another onset/coda to form part of a complex unit. The facts are illustrated below.16 (Catalan data are from Bonet and Lloret 2005a, Harris 1993, and Francisco Ordóñez.) Within a clitic + host unit, a consonant will syllabify as a simple onset as in (41) and (43), however, not as part of a complex onset as in (42) and (44). (41) (42)

16

proclitic + host (V)Ca - + #VC > (V).CaVC /t/ + /imita/ > [ti.mi.ta] ‘s/he imitates you’ (V)Ca - + #CbV > *(V).Ca Cb V /t/ + /rius/ > *[tri.us] ‘you:sg laugh’ ([@t.ri.us] with epenthetic schwa)

The syllabification of Catalan clitics is similar to data described for other Romance languages (see also Cardinaletti and Repetti 2004 and to 2009).

96

Anna Cardinaletti and Lori Repetti

(43) (44)

host + enclitic CVCa # + -V > CV.Ca V /rEb/ + /o/ > [rE.ßu] ‘receive it’ CVCa # + -CbV > *CV.Ca CbV /rEb/ + /la/ > *[rE.ßla] ‘receive her’ ([rEb.la])17

In addition, asyllabic enclitics (C and CC) may syllabify into coda position. An asyllabic (C) enclitic will not syllabify as part of a coda cluster (45), but it will syllabify as a simple coda (46), and an asyllabic enclitic with a CC structure will form its own coda cluster (47). (45)

host + enclitic CVCa # + -Cb > *CVCa Cb /tirin/ + /s/ > *[ti.Rins] ‘throw (pol., pl.) to yourselves’ ([ti.Rin.s@] with epenthetic schwa)

(46)

CV# + -Ca > CVCa /miri/ + /s/ > [mi.Ris] ‘look (pol.) at yourself!’

(47)

CV# + -Ca Cb > CVCa Cb /tiri/ + /ns/ > [ti.Rins] ‘throw (pol.) to us!’

The generalization appears to be that a consonant can be resyllabified (or an asyllabic clitic can be syllabified) as an entire subsyllabic constituent – an onset or a coda – but not as part of a subsyllabic constituent, for example, not as part of a complex onset or coda. Although these patterns are similar to the resyllabification facts, the constraints discussed above to handle resyllabification cannot account for the clitic data. As we see in the following tableaux, highranked Onset needed to account for the output in (48) incorrectly eliminates the actual winner in (49). (48)

(49)

17

/t/ + /imita/ a. [@t.i.mi.ta] b. [ti.mi.ta]  /t/ + /rius/ a. [@t.ri.us] b. [tri.us] 

See footnote 4.

Onset **!

Align

NoCoda *

*

Onset *!

Align *

NoCoda *

Phrase-level and word-level syllables

97

Bonet and Lloret (2005a) propose the Alignment constraint in (50) to account for the Catalan clitic patterns. (50)

Align(lex, sub-s) – the edge of a lexical word has to coincide with the edge of some subsyllabic constituent: a margin (M) or nucleus (N). (The constraint refers to lexical words and not to PWs, since they analyze clitic + host units as PWs).

This constraint says that the edge of a lexical word cannot be the internal member of a subsyllabic unit, such as an onset, but must be at the edge of such a unit. The constraint does not rule out resyllabification completely, but only resyllabification (or clitic prosodization) that would add to a subsyllabic unit. The tableaux below show how this constraint, along with two other wellattested constraints in (51), ranked as in (52), allow us to explain the Catalan clitic syllabification facts. (51)

Final-C – every PW ends in a consonant Align-R(lex, s) – the right edge of a lexical word is aligned with a syllable

(52)

Align-R(lex, sub-s)  Final-C  Align-R(lex, s)

In (53) we show that an asyllabic clitic can form a simple coda, and in (54) an asyllabic clitic with a CC structure can form its own complex coda. In all of the candidates in these two tableaux, the right edge of the lexical word (the verb) is aligned with the edge of a subsyllabic unit (the nucleus). As a result, the high-ranked alignment constraint is not violated, leaving Final-C to choose the winner. (53)

/tir@/ + /n/ > [ti.R@n] tira’n ‘throw some!’ /tir@/ + /n/ Align-R(lex, sub-s) Final-C Align-R(lex, s) a. ti.R@n  * b. [email protected]@ *!

(54)

/tiri/ + /ns/ > [tí.Rins] tiri’ns ‘throw (pol.) to us!’ /tiri/ + /nz/ Align-R(lex, sub-s) Final-C Align-R(lex, s) a. ti.Rins  * b. ti.Rin.z@ *! *

98

Anna Cardinaletti and Lori Repetti

However, an asyllabic clitic cannot encliticize to a host ending in a consonant to form part of a complex coda (55). In candidate (55a) the right edge of the lexical word /tirin/ is not aligned with the edge of the coda, so this form violates the Align(lex, sub-s) constraint, leaving candidate (55b) as the winner. (55)

/tirin/ + /s/ > [ti.Rin.s@] tirin-se ‘throw (pol., pl.) to yourselves’ /tirin/ + /s/ Align-R(lex, sub-s) Final-C Align-R(lex, s) a. ti.Rins *! * b. ti.Rin.s@  *

The Align(lex, sub-s) constraint can also be used to account for the Spanish resyllabification facts. In (56) the /b/ is either a simple coda (in candidate [56a]) or a simple onset (in candidate [56b]), so the high-ranked Align(lex, sub-s) is not violated, and Onset chooses the correct winner. In (57), the /b/ is a simple coda in candidate (57a), so it does not violate the alignment constraint. However, in candidate (57b) the /b/ is part of a complex onset, meaning that the right edge of club is not properly aligned with the edge of the onset, nor is the left edge of lindo. As a result, this candidate violates the Align(lex, sub-s), leaving candidate (57a) as the winner. (56)

club elegante a. [kluß.e.le.Van.te] b. [klu.ße.le.Van.te] 

Align(lex, sub-s) Onset *!

(57)

club lindo a. [kluß.lin.do]  b. [klu.ßlin.do]

Align(lex, sub-s) Onset *!

While the Align(lex, sub-s) constraint does allow us to account for restrictions on syllable structure in these two contexts (clitic prosodization and resyllabification), the approach taken in this paper with two syllable levels (sPP and sPW ) might be able to explain these facts in a more straightforward way. We can capture the insights of the Align(lex, sub-s) constraint through a series of identity constraints: sPW subsyllabic units must be faithfully represented at the phrasal syllable level. (58)

Ident-sPW -sPP -Onset – the sPW onset must be identical at the sPP level Ident-sPW -sPP -Coda – the sPW coda must be identical at the sPP level

Phrase-level and word-level syllables

99

These constraints require the sPP syllables to be identical to the sPW syllables. The two sets of syllables are compared at the subsyllabic level, and violations are categorical: if the sPW has an onset/coda, it must be faithfully reflected at the sPP level (i. e., no additions or deletions); if the sPW does not have an onset/coda, the constraint does not apply to the candidate outputs. The familiar Onset constraint can apply to either of the two syllable levels (sPW or sPP ). (59)

Onset-sPW – the sPW must have an onset Onset-sPP – the sPP must have an onset

The two Ident constraints in (58) are crucially ranked relative to the Onset-sPP constraint, and all three are ranked higher than DEP. (60)

Ident-Onset  Onset-sPP  Ident-Coda  DEP

In the following tableaux each candidate output consists of two forms which represent the syllabification of the two syllable tiers: the first is the sPW -level, and the second is the sPP -level. We only include candidates in which the PW-level syllabification constraints are satisfied (implying that sPW -bound constraints are higher ranked). In the tableaux, the Ident-Onset constraint compares the sPW -level onsets with the sPP -level onsets. If the sPW -level onsets are identical to the sPP -level onsets, no violations are incurred, as in (61), (62), (63), (64) and (66). Notice that in each of these tableaux there are candidates with new onsets on the sPP -level that do not exist on the sPW -level, but these do not violate the Ident-sPW -sPP Onset constraint. Notice also that the first candidates in (65) and (67) violate the Ident-Onset constraint since the sPW -level onsets (/r/ in [65] and /l/ in [67]) are not faithfully reproduced at the sPP -level, where instead we find a CC cluster (/tr/ in [65a] and /bl/ in [67a]). The Ident-Coda constraint works in the same way as the Ident-Onset constraint. The first candidate in (63) incurs a fatal violation of the IdentCoda constraint since the PW-level syllable contains a coda /n/, while the PPlevel syllable adds a consonant, resulting in a coda /ns/ cluster. In (66a) and (67a) sPW coda /b/ is not present at the sPP -level resulting in a violation of the Ident-Coda constraint. The Onset-sPP constraint evaluates all sPP regardless of the PW syllabification. Violations are found in tableaux (64), (65), and (66), and those violations are fatal in (64) and (66). Finally DEP eliminates the losing candidates in (61) and (62).

100 (61)

Anna Cardinaletti and Lori Repetti

/tir@/ + /n/ > [ti.R@n] tira’n ‘throw some!’ /tir@/ + /n/ a. ti.R@ / ti.R@n  b. ti.R@ / [email protected]@

(62)

Ident-Onset Onset-sPP Ident-Coda Dep *!

/tirin/ + /s/ > [ti.Rin.s@] tirin-se ‘throw (pol., pl.) to yourselves’ /tirin/ + /s/ a. ti.Rin / ti.Rins b. ti.Rin / ti.Rin.s@ 

(64)

*!

/tiri/ + /ns/ > [tí.Rins] tiri’ns ‘throw (pol.) to us!’ /tiri/ + /nz/ a. ti.Ri / ti.Rins  b. ti.Ri / ti.Rin.z@

(63)

Ident-Onset Onset-sPP Ident-Coda Dep

Ident-Onset Onset-sPP Ident-Coda Dep *! *

/t/ + /imita/ > [ti.mi.ta] ‘s/he imitates you’ /t/ + /imita/ Ident-Onset Onset-sPP Ident-Coda Dep a) i.mi.ta / ti.mi.ta  b) i.mi.ta / @t.i.mi.ta/ **! *

(65)

/t/ + /rius/ > *[tri.us] ‘you:sg laugh’ /t/ + /rius/ a) ri.us / tri.us b) ri.us / @t.ri.us 

Ident-Onset Onset-sPP Ident-Coda Dep *! * ** *

(66) club elegante Ident-Onset Onset-sPP Ident-Coda Dep a) kluß#e.le. . . / klu.ße.le. . .  * b) kluß#e.le. . . / kluß.e.le. . . *!

(67) club lindo a. kluß#lin.do / klu.ßlin.do b. kluß#lin.do / kluß.lin.do 

Ident-Onset Onset-sPP Ident-Coda Dep *! *

Phrase-level and word-level syllables

101

4. Conclusion We have seen that the prosodization of free clitics and the resyllabification facts – both phrasal phenomena – can be best accounted for by adopting a model of the prosodic hierarchy in which there is an additional phrasal syllable level (sPP ) nested under the PP-level. We have also been able to account for many of the puzzles involving resyllabification and clitic prosodization, including the restriction on complex onsets and codas, and constraints on PWs that do not seem to apply in these contexts, using this model. We conclude the paper with some questions. Perhaps the approach introduced in this paper could be extended to other prosodic phenomena that have represented puzzles to the traditional approach, for example, the fact that syllable requirements seem to be relaxed at word edges. This has often been accounted for by positing extrametrical segments. Can extrametrical consonants be thought of as outside the sPW but included in the sPP ? In addition, the similarity between the approach suggested in this paper and other proposals – ambisyllabicity (Gussenhoven and Jacobs 1998, etc.), the “clitic group” (Nespor and Vogel 1986, etc.), and stratal OT (Kiparsky 2000, etc.) – is striking. Can the insights of ambisyllabicity, the “clitic group”, and stratal OT be captured by phrasal syllables? We think that the answer to each of these questions is “yes”, and we hope to address these topics in future research. Abbreviations indef (indefinite), PD (Phonological Domain), pl (plural), pol (polite), PP (Phonological Phrase), PW (Prosodic Word), sg (singular), SHL (Strict Layer Hypothesis) and OT (Optimality Theory). Acknowledgements We would like to thank the audiences of the Workshop on Phonological Domains: Universals and Deviations, University of Siegen (February 28–March 2, 2007), the SUNY Stony Brook Brown Bag series (March 14, 2007), Irene Vogel, José Elias-Ulloa, two anonymous reviewers, and Francisco Ordoñéz for help with the Catalan data.

References Bafile, Laura 1991–1992 2002

Fonologia prosodica e teoria metrica: accento, cliticizzazione e innalzamento vocalico in napoletano. Ph.D. diss., Università di Firenze. Le consonanti finali nel fiorentino e nel napoletano. Rivista Italiana Di Dialettologia 26: 149–178.

102

Anna Cardinaletti and Lori Repetti

Benincà, Paola and Laura Vanelli 1982 Appunti di sintassi veneta. In: Manlio Cortelazzo (ed.), Guida Ai Dialetti Veneti IV, 7–38. Padova: CLEUP. Bonet, Eulália and Maria-Rosa Lloret 2005a More on alignment as an alternative to domains: The syllabification of Catalan clitics. Probus 17: 37–78. 2005b Against serial evaluation in Optimality Theory. Lingua 115: 1303– 1323. Bullock, Barbara E. 1991 Mora-bearing consonants in coda position and related quantity effect. In: Pier Marco Bertinetto, Michael Kenstowicz, and Michele Loporcaro (eds.), Certamen Phonologicum II: Papers from the 1990 Cortona Phonology Meeting, 105–120. Torino: Rosenberg and Sellier. Burzio, Luigi 1989 Prosodic reduction. In: Carl Kirschner, and Janet De Cesaris (eds.), Studies in Romance Linguistics (LSRL 17), 51–68. Amsterdam and Philadelphia: John Benjamins. Cardinaletti, Anna and Lori Repetti 2004 Subject clitics and functional vowels in northern Italian dialects. [ms, Ca’ Foscari University of Venice and SUNY, Stony Brook]. 2008 Preverbal and postverbal subject clitics in northern Italian dialects: phonology, syntax and cross-linguistic variation. Linguistic Inquiry 39. Casagrande, Jean 1984 The Sound System of French. Washington, DC: Georgetown University Press. Chierchia, Gennaro 1986 Length, syllabification and the phonological cycle in Italian. Journal of Italian Linguistics 8: 5–34. Cummins, John 1974 El habla de Coria y sus cercanias. London: Tamesis. Davis, Stuart 1990 Italian onset structure and the distribution of il and lo. Linguistics 28: 43–55. D’Imperio, Mariapaola and Sam Rosenthall 1999 Phonetics and phonology of main stress in Italian. Phonology 16: 1–28. Face, Timothy 2002 Re-examining Spanish ‘resyllabification’. In: Theresa Satterfield, Christina Tortora, and Diana Cresti (eds.), Current Issues in Romance Languages, 81–94. Amsterdam and Philadelphia: John Benjamins.

Phrase-level and word-level syllables Fava, Elisabetta 1993

103

Sulla pertinenza della pragmatica nell’analisi grammaticale: un esempio dalla cosiddetta coniugazione interrogativa nel dialetto altovicentino. Omaggio a Gianfranco Folena, 2495–2520. Padova: Editoriale Programma.

Goldsmith, John 1981 Subsegmentals in Spanish phonology: An autosegmental approach. In: William Cressey, and Donna Jo Napoli (eds.), Linguistic Symposium on Romance Languages IX, 1–16. Washington, DC: Georgetown University Press. Goria, Cecilia 2004 Subject Clitics in the Northern Italian Dialects. A Comparative Study Based on the Minimalist Program and Optimality Theory. Dordrecht: Kluwer. Gussenhoven, Carlos and Haike Jacobs 1998 Understanding Phonology. London: Hodder Arnold. Harris, James W. 1980 Spanish Phonology. Cambridge, MA: MIT Press. 1993 Integrity of prosodic constituents and the domain of syllabification rules in Spanish and Catalan. In: Kenneth Hale, and Samuel Jay Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 177–193. Cambridge, MA: MIT Press. Hualde, José Ignacio 1991 Aspiration and resyllabification in Chinato Spanish. Probus 3: 55–76. Kaisse, Ellen 1999 Resyllabification precedes all segmental rules: Evidence from Argentinian Spanish. In: Jean-Marc Authier, Barbara E. Bullock, and Lisa A. Reed (eds.), Formal Perspectives on Romance Linguistics, 197–210. Amsterdam and Philadelphia: John Benjamins. Kenstowicz, Michael 1994 Phonology in Generative Grammar. Cambridge, MA: Blackwell. Kiparsky, Paul 2000 Opacity and cyclicity. The Linguistic Review 17: 351–367. Loporcaro, Michele 1997 L’origine del Raddoppiamento Fonosintattico. Saggio di Fonologia Diacronica Romanza. Tübingen: Francke Verlag. 2000 Stress stability under cliticization and the prosodic status of Romance clitics. In: Lori Repetti (ed.), Phonological Theory and the Dialects of Italy, 137–168. Amsterdam and Philadelphia: John Benjamins. Monachesi, Paola 1996 On the representation of Italian clitics. In: Ursula Kleinhenz (ed.), Interfaces in Phonology, 83–101. Berlin: Akademie Verlag.

104

Anna Cardinaletti and Lori Repetti

Nespor, Marina 1985

The Phonological Word in Italian. In: Harry van der Hulst, and Norval Smith (eds.), Advances in Nonlinear Phonology, 193–203. Dordrecht: Foris. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Oostendorp, Marc van 1999 Italian s-voicing and the structure of the phonological word. In: S. J. Hannahs, and Mike Davenport (eds.), Issues in Phonological Structure, 195–212. Amsterdam and Philadelphia: John Benjamins. Peperkamp, Sharon 1997 Prosodic Words. Den Haag: Holland Academic Graphics. Poletto, Cecilia 2000 The Higher Functional Field: Evidence from Northern Italian Dialects. Oxford: Oxford University Press. Repetti, Lori 1991 A moraic analysis of raddoppiamento fonosintattico. Rivista di Linguistica 3: 307–330. 1989 The bimoraic norm of tonic syllables in Italo-Romance. Ph. D. diss., UCLA. Rohlfs, Gerhard 1968 Grammatica Storica della Lingua Italiana e dei suoi Dialetti. Vol. II: Morfologia. Torino: Einaudi. Selkirk, Elisabeth 1995 The prosodic structure of function words. In: Jill N. Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), Papers in Optimality Theory (UMOP 18), 439–469. Amherst: GLSA. Vanelli, Laura 1992 Da ‘lo’ a ‘il’: Storia dell’articolo definito maschile singolare nell’italiano e nei dialetti settentrionali. Rivista Italiana di Dialettologia 16: 29–66. Zamboni, Alberto 1974 Veneto. Pisa: Pacini.

An interface approach to prosodic word recursion Barı¸s Kabak and Anthi Revithiadou

1. Introduction This paper addresses the issue of recursivity in prosodic phonology with special emphasis on the prosodic word (alias phonological word; henceforth PW) and brings attention to several conceptual and empirical problems and inconsistencies with the notion of recursivity as it has been applied to this phonological constituent.1 Our purpose is to explore the nature of recursive prosodic words and how they are generated at the morphosyntax-phonology interface. In recent phonological theories, the constraint NonRecursivity has been employed to account – presumably via its violation – for the emergence of recursive prosodic structures. It is not clear, however, under which conditions, where, and why recursive structures arise in prosody. For instance, it is still an open question whether there is recursivity below the level of the PW (see also Itˆo and Mester 2007).2 In this paper, we argue that recursion is not an inherent property of phonology but arises by the mirroring of recursive structures at the morphosyntactic level. More specifically, we take a less explored approach towards prosodic

1

The very focus of this paper, namely recursivity, suggests that we adopt a Weak Layering approach (Ito and Mester 1992, 2003) to prosodic organization. 2 Although occasionally recursive feet have been proposed (e. g., Selkirk’s 1980a dactylic foot), they are always subject to re-analysis (Dresher and Lahiri 1991; Kager 1994). Furthermore, the place of phonological units lower than the PW in the Prosodic Hierarchy (Selkirk 1980b, 1981; Nespor and Vogel 1986; Hayes 1989; among others) has been questioned. For instance, Inkelas (1989: 47) calls attention to an ordering paradox that results from the assumption that phonological rules construct moras, syllables and feet but, at the same time, they apply within a prosodic domain, which, strangely, is built on the very same constituents, namely moras, syllables and feet. In other words, these lower level constituents feed parsing into higher prosodic constituents which, nevertheless, are constructed on the assumption that moras, syllables and feet exist. Therefore, she proposes that constituents below the PW should be excluded from the Prosodic Hierarchy, which is thus confined only to constituents that serve as domains of phonological rules.

106

Barı¸s Kabak and Anthi Revithiadou

recursivity whereby the grammar requires recursive morphosyntactic structures such as complex predicates and adjuncts to be mirrored in phonology in the most parsimonious way possible. Due to its genuine interface character, our approach limits recursivity to the PW as well as to higher prosodic levels such as the phonological phrase (henceforth PPh) since these are the main prosodic constituents that are involved in the morphosyntax-phonology interface. This is essentially the view proposed in Selkirk (1995) where the variety of PW structures falls out either from different constraint rankings (i. e., grammars) or from differences in morphosyntactic structures. Here, we explore and expand the latter source in an attempt to properly define and restrict the notion of recursivity in prosodic phonology. Furthermore, we draw attention to various functional problems with the constraint NonRecursivity (henceforth, NonRec; Selkirk 1995: 443) as it stands. First, we show that it does not do the job that it is intended to do because recursive structures do not emerge through the violation of this constraint but through the interaction of different constraint sets (e. g., Alignment  Exhaustivity) (see Section 3). Second, NonRec is not crucially needed to evaluate optimal recursive structures nor to derive Selkirk’s (1995) clitic typology. Its effects can again be subsumed by other independently motivated constraints. Hence, we propose that NonRec should be dispensed with. Instead, we exploit already existing interface constraints such as Alignment and Wrap (McCarthy and Prince 1993; Selkirk 1995; Truckenbrodt 1995, 1999) to derive prosodic recursion. In our analysis, recursion is essentially imposed in phonology by morphosyntax. Since it is not an intrinsic property of phonology, but rather is imported from its interface with another component of the grammar, we assume that recursive structures must be marked prosodic constructions. Formally, they result from the extension of an already existing prosodic constituent, yielding a two-segment prosodic category. Specifically, such an “extended” PW exhibits an ambiguous behavior because (a) it inherits the properties of its mother (head PW), and (b) by being an new entity, it may develop properties of its own. We show that since an additional layer of structure is created, the new properties of this constituent have the form of rhythmic re-adjustment rules. As such, our account provides an explanation as to why the majority of rules that have been observed to apply within recursive domains often tend to be optional rules and very frequently rules related to rhythmic processes. Another positive aspect of this approach is that it constrains recursivity in a principled way and explains why recursion below the level of the PW and above the PPh is less motivated because, as we will show, (i) prosodic recursion derives from the interface with morphology and syntax, and (ii) the PW and the PPh are precisely the sites for this interface.

An interface approach to prosodic word recursion

107

The paper is organized in the following way. First, we discuss the notion of recursivity in phonology. We specifically focus on arguments that motivate recursive PWs in Selkirk (1995) and Booij (1995, 1996) and point out some inconsistencies in the way these arguments have been applied to motivate recursivity in various languages. Next, we lay out the fundamental assumptions of our proposal and define constructions that generate recursive structures in the phonological component. Our proposal is empirically substantiated with detailed evidence from Greek and Turkish. At the same time, we demonstrate how our approach maintains cross-linguistically attested types of PWs in the prosodization of function words. Finally, we discuss some cases of recursion that derive from lexical specification and propose some directions for future research. 2. Recursivity in phonology We identify two major sorts of arguments for recursion at the PW level. The first type of argument is phonological in nature and is primarily the view advocated in the work of Booij (1995, 1996) as well as many others (e. g., Peperkamp 1997; Vig´ario 1999, 2003). According to this view, the main motivation for the existence of a recursive PW is the blocking or the optional application of a PW-level phonological process. For instance, the evidence that proclitics in Dutch are parsed into a recursive PW (henceforth PWrec ) structure with their host comes from prevocalic schwa deletion. This rule normally applies within a PW (1a–b), but not across PWs. Crucially it is not enforced in proclitic plus host strings, (1c).3 For Booij (1996), the blocking of schwa deletion in this environment suggests that the proclitic recursively adjoins to its host, as shown in (2). As such, the left edge of the innermost PW serves as a buffer to block the application of the deletion rule in question. (1)

3

Rule-blocking as evidence for PW rec a. /rom@-ein/ [romEin] ‘Roman’ [haldIk] ‘took I’ b. /hald@ Ik/ c. /d@ avOnd/ [d@ avOnt]/ ‘the afternoon’ *[davOnt]

(Booij 1996: 226) (Booij 1996: 231)

Dutch clitics have been the topic of investigation of many researchers (Berendsen 1983; Gussenhoven 1985; among others). Grijzenhout and Kr¨amer (2000) propose an analysis where clitics are parsed by the PPh.

108

Barı¸s Kabak and Anthi Revithiadou

(2)

Recursive PW in Dutch proclisis PW PW d@

avOnt

Employing a morphosyntactic view towards recursivity, Selkirk (1995: 458– 460) proposes that nested syntactic structures such as the one given in (3) are translated by phonology into a PWrec . (3)

a. syntactic structure V

b. phonological structure PW

PW σ [ [need] ’m] PW PW (cf. Needham PW [nid m]) ˙ In this case, recursion is imposed by syntax forcing phonology to mirror the nested morphosyntactic structure by respecting the prosodic boundaries of the lexical word, i. e., the verb. This is ensured by a high-ranked alignment constraint (McCarthy and Prince 1993) which requires every edge of the morphological word to be matched by a PW boundary. Unfortunately, the morphosyntactic argument was not really pursued fervently in the prosodic phonology literature and, as a result, it gradually lost strength and, eventually, it was silenced. Essentially, the two views are not only diametrically different but also incompatible. For instance, the non-application of a rule has received different interpretations by different researchers. To give an example, for Selkirk (1995: 451–452), the failure of the PW-level rule of aspiration to apply to the initial voiceless stop in a phrase such as [[t]oσ [London]PW ]PPh suggests that the function word in this position does not initiate a PW. Thus, it must be parsed at the level of the PPh, thereby constituting a free clitic.4 On the contrary, for Booij (1995, 1996), the non-application of such a PW-level rule in similar cases is taken as evidence that the function word is adjoined to the PW forming an affixal clitic construction. Yet another veil of uncertainty that surrounds the notion of recursivity lies in the precise phonological behavior assumed to be displayed by recursive structures: Should the application of the new set of rules be legitimately used to argue V pro need ’m

4

Alternatively, one could claim that the constructions here as well as the ones in (3–4) are subsumed under the clitic group (henceforth CG) a` la Nespor and Vogel (1986).

An interface approach to prosodic word recursion

109

for recursive structures, or should it constitute evidence for the existence of another domain (e. g., CG (Nespor and Vogel 1986; Hayes 1989), or any other domain within the Prosodic Hierarchy; see Vogel [this volume] for a discussion)? Analyses differ on this issue. For instance, Peperkamp (1997) makes a crucial distinction between the innermost and the outermost PW in Neopolitan, where the latter is reserved for post-lexical phonological allomorphy. In particular, while prosodic structure up to the level of the PW is built at the lexical level, cliticization in this Italian dialect operates at the post-lexical level. Since prosodic structure built at the lexical level cannot be modified, Peperkamp has to resort to recursion for constructions involving clitics. Another set of argument for the assumption that the relevant domain must be an extended PW, but nothing else, comes from the fact that enclitic clusters develop their own stress under the pressure of the three-syllable window restriction, which crucially holds at the PW level (see Peperkamp 1997: 187–191 for complete argumentation and further details). As a result, the word appears to have two stresses, one assigned at the lexical level and the other one at the post-lexical level. Furthermore, Peperkamp observes that these clitic combinations exhibit consonantal lengthening and an unpredictable quality of the stressed vowel, as shown in (4b). We take this to suggest that the surface form of enclitic clusters is subject to an allomorphic rule, which crucially takes place within the outermost PW structure. This analysis then implies that the recursive PW is not only an extended domain that inherits PW-specific rules, but also the one that initiates its own unique set of rules. As such, Peperkamp has to assume that the PW rec hosts the allomorphic rule in question. Strikingly, the same type of evidence could lay support for approaches that adopt an extra prosodic constituent above the level of PW (e. g., the CG), or even a higher prosodic constituent such as PPh, where phrasal allomorphic rules have been argued to exist (Vig´ario 2003). Likewise, the emergence of another primary stress in (4b) can fall out from some form of binarity requirement (Hayes 1995) which forces the two function words to form a (trochaic) foot. Peperkamp does not address these alternative analyses. However, we believe that such language-specific cases raise significant questions as to under which circumstances we need to assume recursive PWs and what sort of phonological properties they should exhibit, which we aim to address in the remainder of this paper. (4)

Stress and consonantal allomorphy in Neapolitan (Peperkamp 1997: 177–178) a. p´Orta t@ ‘bring-imp yourself’ (cf. purtát@ ‘you (pl.) bring’) b. p´Ortaténn@ ‘bring-imp yourself of-them. . . ’

110

Barı¸s Kabak and Anthi Revithiadou

To sum up, recursivity is primarily used in the literature to distinguish affixal clitics from free clitics, as well as for rule blocking. The primary emphasis so far has been on showing that a clitic may not be part of a PW and the PW rec primarily arose from a necessity to prosodify such elements. Moreover, it is often the case that the phonological argumentation is characterized by circularity and is surrounded by a veil of vagueness. Much less attention has been paid to defining the properties associated with this constituent. 3. Deriving recursive prosodic constituents in Optimality Theory In Optimality Theory (Prince and Smolensky 1993), the source for recursion is standardly assumed to fall out form the violation of the constraint NonRecursivity (NonRec), which militates against the propagation of recursively built prosodic structures. Closer inspection, however, reveals that there are several technical and conceptual problems with the way NonRec is employed in the literature. First, recursive structures in fact result not from the violation of NonRec but crucially from a ranking which requires top-ranking of some other constraint. According to Peperkamp (1997: 189), for instance, this constraint is Faithfulness which states that lexically built structure should not be modified. On the other hand, for Selkirk (1995), what derives recursivity is an alignment constraint (McCarthy and Prince 1993) such as Align-Lex (Lex, L/R; PW, L/R), which ensures that some edge of the lexical word will be matched by some edge of the PW, ranked above Exhaustivity (Exh), a constraint that bans skipping of levels.5 The tableau in (5) illustrates the interaction of these constraints with an abstract example. (5)



/Fnc Lex/ a. [Fnc [Lex]PW ]PW b. [Fnc Lex]PW c. [Fnc [Lex]PW ]PPh

Align-Lex

Exh

*! *PW ! *F

As can be seen in the above tableau, it is not necessary to have NonRec. Rather, the interaction between other constraints has the desired effect of yielding recursive structures. If the violation of NonRec does not render recursive structures, then what is the function of this constraint in the grammar? When top-ranked, this constraint ensures that an internal clitic, that is, a clitic that fully incorpo5

We concur with Selkirk (1984, 1986, 1995) and Selkirk and Shen (1990) that the syntax-phonology mapping is blind to function words.

An interface approach to prosodic word recursion

111

rates into the PW of its host, [Fnc Lex]PW (essentially a non-recursive PW), will be optimal and hence chosen to surface compared to an affixal clitic, [Fnc [V]PW ]PW , which constitutes a recursive structure. The following tableau shows this result. (6)



/Fnc Lex / a. [Fnc Lex]PW b. [Fnc [Lex]PW ]PW c. [Fnc [Lex]PW ]PPh

NonRec

Exh

*! *!*

The primary job of NonRec is to expel candidate (6b) from the competition. However, the same result can equally be achieved if an alignment constraint such as Align-pw (PW, L/R; Lex, L/R) is introduced in the system and ranked below Exh. Hence, NonRec has no vital importance here.6 One could argue that NonRec does the crucial work in rendering free clitics, that is, functional elements that are directly parsed by the PPh. However, the effects of NonRec can once again be covered by Align-pw. Appropriately ranked above Exh, this constraint makes sure that a PW can never start with a function word, as shown in (7). (7) 

/Fnc Lex/ a. [Fnc Lex]PW b. [Fnc [Lex]PW ]PW c. [Fnc [Lex]PW ]PPh

Align-Pw *! *!

Align-Lex *

Exh

**

Furthermore, NonRec is also not crucial to evaluate the optimal recursive structure among possible output forms. For example, the difference between the multiply nested structure [[[Lex]PW Fnc]PW Fnc]PW and the flat structure [[Lex]PW Fnc Fnc]PW can be decided on the basis of other independent constraints without the need for NonRec. *Struc is one such constraint that mitigates against the creation and, by extension, the repetition of any form of structure (Prince and Smolensky 1993; see also Ito and Mester 2007, this volume). Finally, the existence of NonRec suggests that avoidance of recursion is natural for phonology, that is to say, an unmarked pattern across the languages 6

An anonymous reviewer raises the question as to whether Exhaustivity could be dispensed with in favor of NonRec. We argue that Exhaustivity is independently needed primarily because it ultimately forces segmental structure to be parsed into prosodic constituents. Hence, for any theory that assumes the Prosodic Hierarchy, this constraint is essential. See Ito and Mester (this volume) for the decomposition of Exhaustivity into various Parse constraints, which essentially organize segmental input into prosodic constituents.

112

Barı¸s Kabak and Anthi Revithiadou

of the world, to which in essence we agree. In a system where constraints are violable, however, the violation of NonRec can yield recursive outputs for all phonological entities (e. g., features, moras, syllables, feet, etc.). This in turn goes against the original intention of the constraint to ban or restrict recursion in phonology. There have been attempts to show that recursion is indeed compatible with phonology, albeit confined to higher prosodic levels (Schreuder and Gilbers 2004; Schreuder 2006). In summary, we have argued that there are fundamental inconsistencies and functional problems with the way recursion is arrived at through the violation of NonRec. Below, we propose an alternative approach to recursion that draws on the morphosyntax-phonology interface. Furthermore, we discard NonRec as a universal constraint and derive its effects through already existing and well-established interface constraints. Our approach employs a morphosyntactically driven recursivity, which is on a par with Selkirk’s (1995) original motivation to derive differences in phonological structure through differences in morphosyntactic representations. Needless to say, the phonological mirroring of morphosyntactic recursion advanced here does not necessitate per se the abolishment of NonRec from the constraint set. In the remainder of this paper, however, we will rely on the complementarity of these two assumptions to shed light on the nature of recursion, its function and limitations in phonology. 4. An interface perspective on recursion We argue that recursion is not an inherent property of phonology, but rather the by-product of its interface with morphosyntax. Essentially, recursive morphosyntactic structures should correspond to recursive phonological structures, and vice versa. This view builds on van Oostendorp’s (2002, 2006) integrity family of constraints as well as Kaye’s (1974) morphological recoverability which states that phonological structure mirrors morphological structure as closely as possible. We assume that morphosyntactically “recursive” structures are those created outside the main lexical7 or syntactic cycle. In such inherently recursive structures, the category of the whole construction is the same as at least one of its members. This category includes compound constructions as well as complex predicates where the whole construction inherits the properties of its head (8a). Under the same rubric fall also constructions that contain function words that are adjoined to syntactic heads such as clitics (8b), as well as extra-cyclic elements 7

This is naturally predicted by all Lexical Phonology models (Kiparsky 1982; Mohanan 1986; Kaisse and Shaw 1985).

An interface approach to prosodic word recursion

113

such as adjunct modifiers, i. e., pieces of structure elements that are assembled parallel to the main derivation and merge with the rest of the derivation at a later stage (Uriagereka 1999; Nunes and Uriagereka 2000) (8c). (8)

Morphosyntax-phonology mapping syntactic structure phonological structure 0 a. X PW X0

X0

PW

X0

PW

b.

X0

clt c.

PW

σ

XP

PW

Y

PPh XP

PW

PPh

An important consequence of this approach is that one cannot possibly have recursion at the level of the foot or the syllable since the interface gives us primarily two main sites: the PW and the PPh. We achieve this mirroring effect through the use of well-established interface constraints such as Alignment and Wrap (McCarthy and Prince 1993; Selkirk 1995; Truckenbrodt 1995, 1999) which all together strive for grouping X0 as PWs and XPs as PPhs. Below, we show that these constraints, along with Selkirk’s (1995) well-known constraints on prosodic domination, namely Layeredness, Headedness, and especially Exhaustivity, can derive all the attested prosodic structures without the necessity for NonRec. With respect to the phonological properties of PWrec , we assume that adjunction at the level of PPh and PW creates a two-segment category that corresponds to an innermost and an outermost phonological layer at the PF, as shown in (9). The outermost layer inherits properties of the mother constituent which may reiteratively or optionally apply within that layer of structure. Metrical calculations may be sensitive to this extra layer of structure, leading to the development of rhythmic structure building/ re-adjustment rules strictly associated with the PWrec . k

(9)

inherent properties of k(k, g)

k [[ a

properties inherited from k(a, b) b ]k

g ]k

114

Barı¸s Kabak and Anthi Revithiadou

Below we will illustrate our approach by using data primarily from Greek and Turkish. We will focus only on the lowest interface site, namely the PW, leaving high-level interface sites for future research. The structures we explore come from roughly three morphosyntactic processes, compounding, complex predication/ noun-incorporation and cliticization. 5. Deriving PWREC from the interface 5.1. Compounds and complex predicates forming recursive PWs: Evidence from Turkish We first explore various types of composite constructions in Turkish that essentially project an X0 bearing the same category label as at least one of its sisters, a precursor for prosodic recursion in our approach. These constructions are nominal compounds containing bare nouns (10a) and verbal compounds which convey temporal/aspectual meanings (10b). Furthermore, Turkish exhibits particular constructions where an immediately preverbal bare noun forms a special unit with its verbal head. These are complex predicates with light verbs and verbal predicates that are often cited as exhibiting noun-incorporation (10c) (e. g., ¨ urk Knecht 1986; Kornfilt 1995, 2003; Aydemir 2004; among others; see Ozt¨ 2005 for a review).8 What is common to these three types of constructions is that the bare noun and the verb form a verbal complex.9

8

9

Based on standard assumptions on incorporation (Baker 1988, 1996), one can claim that the moved element adjoins to the V. Thus, the analysis proposed here correctly predicts that this element will be parsed recursively to the head category. ¨ urk (2005) argues for a pseudo-incorporation analysis for such constructions, the Ozt¨ details of which are beyond our focus. What remains unchanged in this analysis is the ¨ urk, forms a complex fact that the preverbal noun, which is an NP according to Ozt¨ predicate together with the verb head ([NP+V]V ).

An interface approach to prosodic word recursion

(10)

115

Mapping of various types of composite structures yielding [[word]PW [wordPW ]PW REC a. N N N kırk ayak forty foot b.

‘centipede’10

V V d¨us¸-e fall-cvm

V yaz lv

‘to almost fall’

gid-e go-cvm

dur lv

‘to continue to go’

c.

V N red reject

V et do

kitap book

oku read

‘reject’

complex predicate with light verbs

‘read a book’

noun-incorporation

Under our approach, all such constructions should automatically be mapped onto recursive PWs at the phonological level due to high ranked interface constraints, namely Wrap defined in (11) and Align-Lex. The tableau in (12) shows how these constraints evaluate a set of output forms.

10

In our Turkish examples, we mostly comply with the Turkish orthographic conventions. Accordingly, ı, ü, and ö stand for the high back unrounded (IPA: [W]), high front rounded (IPA: [y]), and low front rounded (IPA: [ø]) vowel, respectively. The alveopalatal voiceless fricative ([S]) and the affricate [tS] are indicated by s¸ and c¸ , respectively. The symbol y stands for the palatal glide. The so-called soft-g is represented with gˇ and stands for a velar glide in Modern Standard Turkish (or a velar fricative in various Anatolian dialects). This segment is most often not pronounced in colloquial Standard Turkish. Following the Turkish linguistic tradition, we use capital letters to indicate those underlying segments that undergo phonological processes such as vowel harmony. Accordingly, [I] stands for an underlyingly high vowel and [E] for an underlyingly low vowel.

(11) (12)

Barı¸s Kabak and Anthi Revithiadou

Wrap: Each X0 / XP is contained in its own PW/ PPh, respectively (based on Truckenbrodt 1995, 1999; Peperkamp 1997). [X0 X0 ]X0 a. [X0 X0 ]PW b. [[X0 ]PW [X0 ]PW ]PPh  c. [[X0 ]PW [X0 ]PW ]PW d. [[X0 ]PW X0 ]PW e. [[X0 ]PW X0 ]PPh

Wrap *!* *!

*!*

Align-Lex **

*! ***

..............................

116

Exh

*

Wrap ensures that every X0 / XPs, regardless of whether it is the lowest or the highest in the tree, is contained in a single prosodic constituent. The constraint is violated when one of the X0 s fails to form a prosodic constituent of its own, as demonstrated by the candidates in (12a, b, e). A double violation of Align-Lex occurs in candidate (12a) since the lowest X0 s lack innermost PW boundaries. Furthermore, Align-Lex is violated by candidate (12d) because the rightmost X0 lacks a left PW-boundary. Finally, the candidate (12e) also violates AlignLex because the edges of the second X0 and the right edge of the upper X0 are not aligned with a PW-boundary. 5.2. Function words forming PWrec with their host: Evidence from Greek The Greek pronominal system has a set of weak forms of pronouns which are prosodically dependent on an adjacent host and constitute clitic elements. Such object clitics have certain distributional properties that vary according to the dialectal variety. In Standard Greek, they always precede the non-imperative verb form which serves as their prosodic host (13a). With imperative forms and gerunds, however, they are always postverbal (13b). (13)

a.

b.

mas málose clt-1.pl.gen scold-past.3.sg ‘(S)he scolded us.’ ðóse tó mu give-2.sg.imp clt-3.nt.sg.acc clt-1.m.sg.gen ‘Give it to me!’

Philippaki-Warburton and Spyropoulos (1999) have convincingly shown that object clitic pronouns do not have the properties of affixes.11 They also provide 11

For a different view on the issue see Joseph (2003) and Condoravdi and Kiparsky (2001).

An interface approach to prosodic word recursion

117

ample syntactic evidence that object clitics behave as the arguments proper of the clause and participate in certain syntactic operations. In more technical terms, the distribution of object clitics with respect to the verb form is derived by means of a cliticization movement rule. The clitic is base-generated as an argument at the relevant theta-position inside the vP and then it targets the IP layer to which it moves overtly. Let us assume that it targets the T head to which it adjoins. Given that in Greek the verb overtly moves to the T functional head in order to license its tense and subject-agreement features, the clitic ends up being adjoined to the verb form as a proclitic (Philippaki-Warburton and Spyropoulos 1999; Philippaki-Warburton et al. 2004).12 (14)

Syntactic structure of proclitic object pronouns in Greek TP T0 max T0 max

clt

T0

v-V

... vP tv−V t cl

mas málose ‘(S)he scolded us.’ In this respect, cliticization does not extend the structure, but rather creates a recursive syntactic constituent which contains the clitic and the functional category that hosts the verb form. In this paper, we claim that these kind of structures are translated in the phonology as recursive PWs. On the basis of Selkirk’s (1995) typology, therefore, Greek proclitics qualify as affixal. (15)

Prosodic structure of proclitic object pronouns in Greek PW PW [mas

12

[málose]PW ]PW

See Revithiadou and Spyropoulos (2008) for a detailed typological survey of object clitic pronouns in Greek.

118

Barı¸s Kabak and Anthi Revithiadou

On the other hand, the fact that object clitics follow the imperative verb forms results from the morphosyntactic licensing of the imperative mood in Greek. More specifically, the imperative is the only affixal mood in Greek, hence it requires the overt movement of the verb form to a Mood functional head above the T head (Philippaki-Warburton 1998, among others): (16)

Syntactic structure of enclitic object pronouns in Greek MP M0 max V

TP M0

T0 max T0

clt tV

T0

nP tV

tcl

málosé mas ‘scold us’ Interestingly, in these constructions object clitics are left behind by the movement of the verb to the c-commanding Mood head and, consequently, they belong to its complement domain. Since they are not adjuncts, they should not be expected to be parsed at the outer layer of the PW any more. Instead, phonology should translate this structure as one unit in a way analogous to the prosodification of a verb plus a complement string into one prosodic constituent (in this case, though, the PPh). This expectation is indeed met since enclitics, unlike their proclitic counterparts, are incorporated into the PW of their verbal host (see Section 6.1. below, and Revithiadou and Spyropoulos 2008 for detailed argumentation). In sum, we have reviewed various structures where phonology mirrors morphosyntactically recursive structures as recursive prosodic domains. This analysis has the advantage to also account for different prosodic patterns in one language by means of one and the same constraint ranking making use of already existing tools. More specifically, the ranking Wrap  Exhaustivity, Align-Lex gives us the desired PWrec pattern for proclitics and, at the same time, it favors the internal pattern for enclitics, as illustrated. This is because Wrap remains inert in non-recursive morphosyntactic constructions such as the one associated with enclisis. The following tableaux demonstrate the selection of the optimal prosodic forms for each clitic order:

An interface approach to prosodic word recursion

(17)

(18)

[T mas [T málose]] adjuction  a. [mas [(málo)se]PW ]PW b. [mas (málo)se]PW

Wrap

[MP málose [TP mas. . . ]] head − complement domain a. [[(málo)se]PW mas]PW  b. [(málo)(sé mas)]PW

119

Align-Lex

*!

Exh ** **

Wrap

Exh

Align-Lex

*

*!* *

6. Phonological properties of PW REC 6.1. Segmental rules and optionality Below, we demonstrate various segmental rules that apply to the extended PW constituent that arise from the interface, which we called the PW rec . Crucially, the processes we demonstrate are lexical rules and hence specifically related to PWs. Closer inspection, however, reveals that their application is extended, either optionally or obligatorily, to the recursive domain as well. We start with the Turkish constructions described in (10) above and demonstrate that consonant cluster resolution, geminate simplification, and long vowel shortening apply primarily to morphosyntactic constructions forming a PW as well as a PWrec , and not to syntactic phrases which are arguably parsed into PPhs. It is a well established fact of Turkish phonology that underlying consonant clusters of a particular type, geminates, and long vowels in closed syllables cannot surface at the end of a word. Such illicit syllable types give rise to processes such as epenthesis, geminate simplification, and vowel shortening. Syllable repair strategies are, however, blocked when the choice of resyllabification is available. This is the case when a following vowel-initial element is part of the same PW or PWrec that arise from the kinds of constructions discussed above (a- and b-examples), and crucially not when it is contained in a syntactic phrase (c-examples). (19) a. /haps-(y)I/ prison-acc

[hapsi]PW *hapisi (cf. hapis, *haps)

‘prison’

/kayd-(y)I record-acc

[kaydı]PW *kayıdı (cf. kayıt, *kayt)

‘record’

b. /haps et-mEk/ prison LV-inf

[[haps]PW [etmek]PW ]PW *hapis etmek

‘to imprison’

120

Barı¸s Kabak and Anthi Revithiadou /kayd et-mEk/ record LV-inf c. /haps iste-mEk/ prison want-inf /kayd a¸c-mEk/ record open-inf

(20) a. /redd-(y)I/ rejection-acc b. /redd et-mEk/ rejection LV-inf

[[kayd]PW [etmek]PW ]PW ?/*kayıt etmek

‘to register’

[[hapis]PW [istemek]PW ]PPh *haps istemek

‘to ask for imprisonment’

[[kayıt]PW [a¸cmak]PW ]PPh *kayd a¸cmak

‘to open a registration’

[reddi]PW *redi (cf. red, *redd)

‘rejection’

[[redd]PW [etmek]PW ]PW *red etmek

‘to reject’

c. /redd al-mEk/ [[red]PW [almak]PW ]PPh rejection receive-inf *redd almak (21) a. /hara:m-(y)I/ forbidden-acc

‘to receive a rejection’

[hara:mı]PW ‘forbidden’ *haramı (cf. haram, *hara:m)

b. /hara:m et-mEk/ forbidden LV-inf

[[hara:m]PW [etmek]PW ]PW *haram etmek

‘to take the pleasure out of something’

c. /hara:m et/ forbidden meat

[[haram]PW [et]PW ]PPh *hara:m et

‘religiously forbidden meat’

Furthermore, the absence of consonant cluster resolution is also observed in N+N compounds, which are subsumed under the PWrec in our analysis (22). (22) a. /kayn-(y)I/ in.law-acc

[kaynı]PW *kayını (cf. kayın, *kayn)

‘in-law; brother-in-law’

b. /kayn+ana/ in.law+mother

[[kayn]PW [ana]PW ]PW *kayınana

‘mother-in-law’

c. /kayn+ata/ in.law+ancestor

[[kayn]PW [ata]PW ]PW *kayınata

‘father-in-law’

It is important to note that compounds that carry an explicit compound marker on the head (N+N-CmpM) (23a) bear a striking similarity to genitive-possessive constructions (23b), which are truly syntactic phrases. The so-called “compound marker” and the genitive 3rd person singular suffix are essentially the same affix. According to Kornfilt (1984: 62–66), the only clearly identifiable difference between the two constructions is that the genitive-marked NP is definite, hence referential, while in the compound case the same NP is non-referential and/or generic. Indeed, contrary to N+N compounds as given in (22b–c), N+N-CmpM compounds behave like phrases with respect to the syllable repair rules. That is to say, syllabification does not apply between the first and the second member of the compound, giving rise to syllable repair rules, as shown in (24b) (see

An interface approach to prosodic word recursion

121

Kabak and Vogel 2001: 350 for further examples). This can straightforwardly be accounted for by the fact that [N+N-CmpM] compounds form PPhs under our analysis, rather than PWrec .13 (23) a. o¨ gˇ retmen ev-i teacher house-CmpM

‘a locale for teachers’

b. o¨ gˇ retmen-in ev-i teacher-gen.3sg house-poss.3sg (24) a. /karn-(y)I/ stomach-acc

‘the house of the teacher’

[karnı]PW ‘stomach’ *karını (cf. karın, *karn)

b. /karn + aˇgrı-sı/ [[karın]PW [aˇgrısı]PW ]PPh ‘stomach-ache’ stomach pain-CmpM *karnaˇgrısı

Moving on to the Greek facts, we see that the rule of s-voicing before sonorants and voiced fricatives obligatorily applies to the PW (25). The fact that the rule in question also applies to a domain created by the adjunction of the object clitics to a verbal host (26) suggests that this extended domain also exhibits PW-level properties, hence yielding empirical support for recursion at the PW level.

13

Likewise, we also make a distinction between postpositional phrases that are constructed by the so-called “genuine” (primary) postpositions as opposed to those that are formed by “fake” (secondary) ones, which bear possessive morphology thereby making the whole phrase similar to genitive-possesive constructions (Lewis 1967; Kornfilt 1997: 423). (i) a. kitap i¸cin ‘for a book’ bı¸cak ile ‘by a knife’ b. kitap arka-sı ‘behind a book’ masa u¨ st-¨u ‘on/above a table’ Being genuine postpositional phrases, the constructions in (ia) arguably behave like bare compound constructions illustrated in (22). Therefore, genuine postpositional phrases are expected to form a PWrec , while those that bear the rubric of genitivepossesive constructions (ib) are presumably grouped as a PPh, similar to the N+NCmpM construction given in (24b). The phonological reflection of this distinction is clearly observed in constructions where genuine postpositions such as i¸cin ‘for’ and ile ‘with’ optionally do not give rise to syllable repair rules (iia). The vowel-initial “fake” postpositions, however, almost obligatorily give rise to the application of the syllable repair rules such as degemination and epenthesis (iib). ‘for an increase’ (ii) a. [[zamm]PW [i¸cin]PW ]PW ‘with pleasure’ [[keyf ]PW [ile]PW ]PW ‘for connection’ [[hatt]PW [i¸cin]PW ]PW b. [[zam]PW [arka-sı]PW ]PPh *zamm arka-sı ‘after an increase’ [[keyif]PW [¨once-si]PW ]PPh *keyf o¨ nce-si ‘before pleasure’ *hatt ara-sı ‘between connection(s)’ [[hat]PW [ara-sı]PW ]PPh

122

Barı¸s Kabak and Anthi Revithiadou

(25)

a. b.

/pros-méno/ pro.zm´eno ‘anticipate’ /Geras-ménos/ Ge.ra.zm´e.nos ‘aged’

(26)

a.

/mas málose/ clt-1.pl.gen scold-past.3.sg maz. málose ‘(S)he scolded us.’ /mas ðjavázi/ clt-1.pl.gen read-3.sg maz. ðJavázi (S)he reads for us.’

b.

It is important to note that [z] syllabifies together with the following morphological element (affix or root) in (25). On the contrary, in (26) the consonant in question is ambisyllabic since complete resyllabification is blocked between the proclitic and the verb: maz.má.lo.se./ *ma.zmá.lo.se. This suggests that the consonants are separated by a PW boundary, hence showing once again that the construction in question indeed forms a PWrec. Another PW-rule that applies to resolve hiatus in a derived environment is vowel deletion. More specifically, a less sonorous vowel deletes when a vowel of greater sonority follows. This rule obligatorily applies within the PW (27) but optionally in the PPh domain (28). Interestingly, vowel deletion is almost obligatorily enforced in constructions which we argue to form PWrec domains (29). The process is blocked only when the construction is produced in very careful speech or rendered with emphatic stress (30). (27)

a. b. c.

/kse-alázo/ ksalázo ‘change’ /kse-onomázo/ ksonomázo ‘un-name s.o.’ /ipo-ánTropos/ ipánTropos ‘sub-human’

(28)

a.

/to ómorfo aGóri/ the handsome-nom.sg boy-nom.sg ‘the handsome boy’

b. c. (29)

a.

Possible outputs (variation): tómorfaGóri (uttered in normal to fast speech rate) tómorfo aGóri ∼ to ómorfo aGóri (uttered in careful speech) /mu to afíni/ clt-1.sg.gen clt-3.nt.sg.acc leave-3.sg mutafíni ‘(S)he leaves it to me.’

An interface approach to prosodic word recursion

b.

c.

(30)

A. B.

123

onomázi/ /me clt-1.sg.acc name-3.sg monomázi ‘(S)he names me.’ /me elénxi/ clt-1.sg.acc control-3.sg meléNc¸ i ‘(S)he controls me.’ Pjó íne to próvlima me ton Perséa? ‘What is the problem with Perseas?’ E, pjó náne? [Sinéxia me afíni]F . ‘Well, what could it be? he constantly abandons me.’

In conclusion, PW-rules are also observed in the outermost domain although there may be some level of optionality involved due to other reasons such as focus, sentence stress, as well as the type of register. It is important to emphasize that none of these segmental rules are specific to this domain. Since we are dealing with an extended domain, hence an extra layer of metrical structure, it is only natural that the inherited properties may show a lesser degree of pertinacity. Unique segmental rules are hard to attest in a constituent between the PW and the PPh, which in turn makes it difficult to substantiate the CG (Nespor and Vogel 1986; Hayes 1989). Below, however, we show that the domain we claimed to be the PWrec induces special rhythmic phenomena, which could arguably be used as evidence for the existence of an independent constituent such as the CG or the composite group (see Vogel this volume). Nevertheless, we argue that this is only epiphenomenal, and crucially arises from the way in which the rhythmic algorithm interprets the extra layer of structure associated with the PWrec . Another related factor is also the length of the resulting constituent, which often calls for further prosodic organization by means of footing, for instance. To the best of our knowledge, the recursive domain does not induce any unique segmental rules in the languages under examination, as well as cross-linguistically. 6.2. Rhythmic structure-building and re-adjustment rules: Unique PWREC processes? Here we demonstrate that the extended PW domain may show spurious metrical structures or serves as a domain where special rhythmic re-adjustment processes take place. We first look into a straightforward case of the emergence of secondary/rhythmic stress in this domain. In several varieties of Greek,

124

Barı¸s Kabak and Anthi Revithiadou

especially those spoken in Northern Greece, a sequence of proclitics provide enough material to form a foot of their own. Such footed elements develop secondary/rhythmic stress, as shown in (31).14 (31)

Rhythmic stress in proclitics to a. /mu clt-1.sg.gen clt-3.nt.sg.acc (mù to) ðJávase ‘(S)he read it to me.’ b. /mas tus clt-1.pl.gen clt-3.m.sg.acc (màs tuz) ðJálekse ‘(S)he chose them for us.’ tus c. /mas clt-1.pl.gen clt-3.m.sg.acc (màs tuz) maJirévi ‘(S)he cooks them for us.’

ðjávase/ read-past.3.sg

ðjálekse/ choose-past.3.sg

maGirévi/ cook-past.3.sg

Although it looks like this rhythmic stress is a unique property of the extended constituent, its triggering force should be sought in the binarity of the adjoined material which naturally calls for grouping into trochaic feet (Hayes 1995) and, consequently, the development of rhythm. The effects of such size constraints can be detected within the PW domain as well: pàrakatjanós ‘of lower descend’, ànTropistikós ‘humanitarian’. Likewise, compounds in Turkish, unlike corresponding phrases, display specific prosodic structure, where non-primary stress is conflated in this domain, as shown in (32). The result is that the second PW surfaces with secondary stress in syntactic phrases while the stress on the same PW is perceptually reduced in compounds. (32)

Leftmost prominence in Turkish nominal compounds a. k´ırk ayak * * * * kırk ayak

14

‘centipede’ PPh PWrec PW

b. k´ırk ay`ak * * * kırk ayak

‘forty feet’ PPh PW

The same phenomenon is also observed in dialects of German (see Kabak and Schiering 2006).

An interface approach to prosodic word recursion

125

Similarly, another type of composite structure, namely verbal compounds ([V+V]) that we analyzed as yielding PWrec in Section 5.1, also shows stress differences in comparison to comparable syntactic phrases. While the syntactic phrases in (34) can bear secondary stress on the second PW, the same makes their homophonous V+V-compound counterparts in (33) sound awkward or illicit. (33)

V+V verbal compounds (no secondary stress) a. ban-a bak-´ıp dur-du (?/* dur-dù) I-dat look-cvm aux-past ‘(S)he kept looking at me.’ b. ban-a at-´ı ver-di (?/* ver-dì) I-dat throw-cvm aux-past ‘(S)he suddenly threw (it) at me.’

(34)

Phrases (secondary stress) a. ban-a bak-´ıp dur-dù I-dat look-conj stop-past ‘(S)he looked at me and stopped.’ b. ban-a at-´ı ver-dì I-dat horse-acc give-past ‘(S)he gave me the horse.’

Again, this could be used as evidence for the existence of unique phonological processes of the rhythmic type taking place in the PW rec . Yet another piece of evidence pointing towards this direction comes from Greek word+word compounds, which exhibit rightmost stress prominence in compounds but not in phrases. Just like in the case of Turkish compounds, these constructions, listed in (35), also form a PWrec in our analysis. Interestingly, such word+word compounds are essentially different from comparable PPhs consisting of an adjective modifying a noun, where the opposite prominence pattern is observed (36). (35)

Adj+N compounds: rightmost stress prominence a. b.

(36)

pirinikòs pólemos ‘nuclear war’ ðiplomatikò sóma ‘diplomatic delegation’

Adj+N phrases: leftmost stress prominence a. b.

irakinós pòlemos ‘Iraqi war’ ðiplomatikó àsilo ‘diplomatic asylum’

Closer examination of both the Turkish and Greek data above reveals, however, that what appears to be unique is nothing more than an extension of word-based

126

Barı¸s Kabak and Anthi Revithiadou

restrictions to this domain. More specifically, accent resolution in Turkish is always left-oriented, including within words (Inkelas 1994, 1999; Kabak and Vogel 2001). This is especially evident in cases where two (lexical) stresses are contained within a morphological word and the leftmost one always wins (e. g., /Ávrupa-lı-la¸s-árak/→ [Ávrupalıla¸sarak] ‘by becoming European’). Similarly, we see a reincarnation of word-related stress prominence resolution in the Greek data as well. The compounds, as well as several other instances of compounding in Greek, indeed follow the three-syllable window restriction, a truly word-level constraint, which prohibits stress from appearing on a syllable further than the antepenult. Yet again, the size of the morphological elements participating in a particular construction lies behind this bias towards rightmostness. In compounds, longer words are at the risk of violating the window requirement, and therefore, they develop modes to keep their prominence closer to the right edge. We see similar tactics being developed within the PW, for example, when enclitics are incorporated into their verbal host, as shown in (37). In this example, the unfooted syllable of the verb forms, together with the clitic, a post-lexical foot which, under the pressure to keep stress within the last three syllables, is assigned primary prominence. (37)

/ðj´avase to/ read-2.sg.imp clt-3.nt.sg.acc ðJàvasé to ‘Read it!’

To sum up, our analysis of compound and clitic+host constructions into the extended constituent of PW, namely PWrec , correctly accounts for the transfer of purely PW-level restrictions and mechanisms to this domain. We showed that such constituents share the same segmental rules with the PW, and this is further supported with evidence from rhythmic phenomena. Assigning to this constituent an independent status simply misses this important generalization. Finally, we argued that the forcefulness and often the uniqueness in which rhythmic rules apply in this domain have been argued to be intimately related to the size of the resulting constituent. 7. Other sources of prosodic recursion Besides the purely morphosyntax-driven account of recursivity, we argue that prosodic recursion can also arise through lexical specification. More specifically, morphosyntactic elements can have a templatic specification as to where they need to be assembled in the prosodic structure. For instance, certain affixes

An interface approach to prosodic word recursion

127

in languages like Dutch (e. g., -achtig ‘like’; -loos ‘-less’) fall outside the PW of their morphological host to which they adjoin recursively (e. g., van Oostendorp 2002, and the references cited therein). This by no means weakens our proposal primarily because such idiosyncratic elements are arguably fossilized instances of the morphosyntax of the past. It has been argued that the origin of such markers is composite constructions (van Beurden 1987: 24) of the sort discussed in Section 5.1, which in the course of history lost their independent lexical status and became bound elements. Similarly, Kabak and Revithiadou (2009) argue that bound elements that do not cohere with the word stress pattern of otherwise a regular (edgemost) stress language seem to have emerged from the morphologization of unstressed elements in composite structures, which essentially correspond to recursive prosodic constituents under the present analysis. It seems that PWrec structures provide a breeding ground for the genesis of bound morphemes with idiosyncratic stress properties at least in edge-oriented stress languages. 8. Conclusions We have argued that recursion is not an inherent property of phonology but the result of its interface with morphosyntax. Recursive constituents as such arise primarily from a requirement to mirror recursive morphosyntactic structures (e. g., complex predicates, adjuncts, etc.). We reviewed different instantiations of recursivity in the prosodic phonology literature and highlighted various conceptual and functional issues related to the notion of recursivity and how it is translated into the OT framework via the constraint NonRec. We concluded that there is no need for NonRec since already available interface constraints such as Alignment and Wrap together with other well-established constraints on prosodic domination suffice to yield recursive structures in phonology. We claimed that differences in morphosyntactic structure are responsible for the differences in which lexical as well as functional elements are prosodified. The need for recursion primarily arises from the principle of mirroring. By definition then, recursion is only confined to the level of the PW and the PPh since these are the main interface sites. Furthermore, we have shown that the same grammar (i. e., constraint ranking) gives rise to different prosodic patterns. It should be noted that this line of reasoning is in accordance with basic premises of OT, and also on a par with Selkirk’s (1995) assumptions about the different patterns of prosodization of function words in English:

128

Barı¸s Kabak and Anthi Revithiadou

“. . . the variety in prosodization of function words, can come about in just two different ways, given an optimality theoretical perspective: through differences in the morphosyntactic input structure in which the Fnc is located and/or differences in the ranking of the relevant constraints.” (Selkirk 1995: 464).

An in-depth examination of two typologically distinct languages, namelyTurkish and Greek, revealed that inherently recursive morphosyntactic structures give rise to extended prosodic constituents in the phonological components whose phonological reflections reincarnate segmental as well as prosodic phenomena observed at the level of the PW. Our analysis also accounted for several instances of the ‘left-right asymmetry’ problem (e. g., proclitics vs. enclitics in Greek), which is born out from an asymmetry at the relevant morphosyntactic representations. Our focus here was primarily on recursion at one constituent, namely the PW. Further research is necessary to explore the phonological properties of recursive structures at the level of PPh, or even higher constituents. It also remains to be seen under what conditions phonology is non-isomorphic to morphosyntax and why. Are there specific morphosyntactic structures or processes that call for non-isomorphism? Do all instances of non-isomorphism result from morphosyntax? Here we already hinted at how diachronic change may lead to recursion in phonology, but we left unexplored the restructuring effect that performance-based phonological constraints, pertaining to the size and length of lexical material, may exercise on the output of morphosyntactic derivation. Future research should address such historical and psycholinguistic aspects of phonological recursion. List of Abbreviations acc clt cvm CmpM conj dat gen ımp lv m

accusatıve clıtıc converb marker (aka gerund) compound marker conjunctive datıve genıtıve ımperatıve lıght verb masculıne

n nom nt past pl poss sg v 2 3

noun nomınatıve neuter past plural possessive sıngular verb second person third person

An interface approach to prosodic word recursion

129

Acknowledgements We would like to thank Junko Ito and Armin Mester for fruitful discussion, and Vassilios Spyropoulos for updating us about recent syntactic notions and representations. Special thanks also go to Miriam Butt for her useful input on the syntactic and semantic aspects of complex predicates. We also wish to thank two anonymous reviewers for providing useful comments. Finally, we would like to thank the audiences of the 29th Annual Meeting of the German Linguistics Society, Siegen, Germany (28.02–02.03.2007) and the 15th Manchester Phonology Meeting, Manchester, United Kingdom (24–26.05.2007) for useful feedback. This research is made possible by funds from (i) the German Research Foundation (DFG) given to Project D9 within the SFB 471 at the University of Konstanz, and (ii) the Pythagoras Project No. 1347 (Action: ‘Operational Program for Reinforcement of Research Groups at Universities’, University of the Aegean). This project was sponsored by the Second Community Support Framework (EPEAEK II) and co-funded by the European Social Fund (75%), the European Regional Development Fund and National Resources (25%).

References Aydemir, Yasemin 2004 Are Turkish preverbal bare nouns syntactic arguments? Linguistic Inquiry 35: 465–474. Baker, Mark 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago: Chicago University Press. 1996 The Polysynthesis Parameter. Oxford: Oxford University Press. Berendsen, Egon 1983 Final devoicing, assimilation, and subject clitics in Dutch. In: Hans Bennis, and W.U.S. Kloeke van Lessen (eds.), Linguistics in the Netherlands, 21–30. Dordrecht: Foris. Beurden, Lisan van 1987 Playing level with Dutch morphology. In: Frits Beukema, and Peter Coopmans (eds.), Linguistics in the Netherlands 1987, 21–30. Dordrecht: Foris. Booij, Geert 1995 The Phonology of Dutch. Oxford: Clarendon Press. 1996 Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13: 219–242. Condoravdi, Cleo and Paul Kiparsky 2001 Clitics and clause structure. Journal of Greek Linguistics 2: 1–40. Dresher, Elan B. and Aditi Lahiri 1991 The Germanic foot: Metrical coherence in Old English. Linguistic Inquiry 22: 252–86.

130

Barı¸s Kabak and Anthi Revithiadou

Grijzenhout, Janet and Martin Kr¨amer 2000 Final devoicing and voicing assimilation in Dutch derivation and cliticization. In: Barbara Stiebels, and Dieter Wunderlich (eds.), Lexicon in Focus, 55–82. Berlin: Akademie Verlag. Gussenhoven, Carlos 1985 Over de fonologie van Nederlandse clitica. [On the phonology of Dutch clitics.] Spektator 15: 180–200. Hayes, Bruce 1989 The prosodic hierarchy in meter. In: Paul Kiparsky, and Gilbert Youmans (eds.), Rhythm and Meter, 201–260. Orlando: Academic Press. 1995 A Metrical Theory of Stress: Principles and Case Studies. Chicago, IL: University of Chicago Press. Inkelas, Sharon 1989 Prosodic constituency in the lexicon. Ph.D. diss., Stanford University. 1994 Exceptional stress-attracting suffixes in Turkish: Representations vs. the grammar. Paper presented at the Workshop on Prosodic Morphology, Utrecht University. [ROA39, http://ruccs.rutgers.edu/roa.html] 1999 Exceptional stress-attracting suffixes in Turkish: Representations versus the grammar. In: Ren´e Kager, Harry van der Hulst, and Wim Zonneveld (eds.),The Prosody-Morphology Interface, 134–187. Cambridge: Cambridge University Press. [Revised version of Inkelas, Sharon. 1994.] Ito, Junko and Armin Mester 1992 Weak layering and word binarity. Linguistic Research Center Working Paper LRC-92–09, UC Santa Cruz. 2003 Weak layering and word binarity. In: Takeru Honma, Masao Okazaki, Toshiyuki Tabata, and Shin Ichi Tanaka (eds.), A New Century of Phonology and Phonological Theory. A Festschrift for Professor Shosuke Haraguchi on the Occasion of his Sixtieth Birthday, 26–65. Tokyo: Kaitakusha. [Revised version of the 1992 LRC working paper.] 2007 Categories and projections in prosodic structure. Paper presented at the Old World Conference in Phonology 4, January 18–21, 2007, Rhodes, Greece. this volume The extended prosodic word. Joseph, Brian D. 2003 Defining word in Modern Greek: A response to Philippaki-Warburton and Spyropoulos. In: Geert Booij, and Jaap van Maarle (eds.), Yearbook of Morphology 2002, 87–114. Dordrecht: Kluwer. Kabak, Barı¸s and Anthi Revithiadou 2009 From edgemost to lexical stress: Diachronic paths, typology and representation. The Linguistic Review 26(1): 1–36. Kabak, Barı¸s and Ren´e Schiering 2006 The phonology and morphology of function word contractions in German. Journal of Comparative Germanic Linguistics 9: 53–99.

An interface approach to prosodic word recursion

131

Kabak, Barı¸s and Irene Vogel 2001 The phonological word and stress assignment in Turkish. Phonology 18: 315–360. Kager, Ren´e 1994 Ternary rhythm in alignment theory. Ms., Utrecht University. [ROA35, http://ruccs.rutgers.edu/roa.html] Kaisse, Ellen and Patricia Shaw 1985 On the theory of lexical phonology. Phonology Yearbook 2: 1–30. Kaye, Jonathan 1974 Opacity and recoverability in phonology. Canadian Journal of Linguistics 19: 134–149. Kiparsky, Paul 1982 Lexical morphology and phonology. In: In-Seok Yang (eds.), Linguistics in the Morning Calm, 1–91. Seoul: Hanshin. Knecht, Laura Ellen 1986 Subject and object in Turkish. Ph. D. diss., Massacusetts Institute of Technology. Kornfilt, Jaklin 1984 Case marking, agreement and empty categories in Turkish. Ph.D. diss., Harvard University. 1995 Scrambling and incorporation in Turkish. In: Artemis Alexiadou, Nanna Fuhrhop, Paul Law, and Sylvia L¨ohken (eds.), ZAS Papers in Linguistics 1(4), 29–42. Berlin: Zentrum f¨ur Allgemeine Sprachwissenschaft. 1997 Turkish. London: Routledge. 2003 Scrambling, subscrambling and case in Turkish. In: Simin Karimi (ed.), Word Order and Scrambling, 125–155. Malden and Oxford: Blackwell. Lewis, Geoffrey N. 1967 Turkish grammar. Oxford: Oxford University Press. McCarthy, John J. and Alan Prince 1993 Generalized alignment. In: Geert Booij, and Jaap van Maarle (eds.), Yearbook of Morphology 1993, 79–153. Dordrecht: Kluwer Academic Publishers. Mohanan, Karuvanuur P. 1986 The Theory of Lexical Phonology. Dordrecht: Reidel. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Nunes, Jairo and Juan Uriagereka 2000 Cyclicity and extraction domains. Syntax 3: 20–43. Oostendorp, Marc van 2002 The phonological and morphological status of the prosodic word adjunct. In: Gisbert Fanselow, and Caroline F´ery (eds.), Resolving Con-

132

Barı¸s Kabak and Anthi Revithiadou

2006 ¨ urk, Balkız Ozt¨ 2005

flicts in Grammars: Optimality Theory in Syntax, Morphology, and Phonology, 209–235. Hamburg: Buske. ‘The theory of Faithfulness.’ Ms., Meertens Institute/KNAW.

Case, Referentiality and Phrase Structure. Amsterdam/Philadelphia: John Benjamins. Peperkamp, Sharon 1997 Prosodic words. Ph.D. diss., HIL Dissertation Series 34 (HIL/University of Amsterdam), Holland Academic Graphics, The Hague. Philippaki-Warburton, Irene 1998 Functional categories and Modern Greek syntax. The Linguistic Review 15: 158–186. Philippaki-Warburton, Irene, and Vassilios Spyropoulos 1999 On the boundaries of inflection and syntax: Greek pronominal clitics and particles. In: Geert Booij, and Jaap van Marle (eds.), TheYearbook of Morphology 1998, 45–72. Dordrecht: Kluwer. Philippaki-Warburton, Irene, Spyridoula Varlokosta, Michalis Georgiafentis, and George Kotzoglou 2004 Moving from theta-positions: Pronominal clitic doubling in Greek. Lingua 114: 963–989. Prince, Alan and Paul Smolensky 1993 Optimality Theory: Constraint interaction in Generative Grammar. Report no. RUCCS-TR-2, Rutgers University Centre for Cognitive Science, New Brunswick, NJ. Revithiadou, Anthi and Vassilios Spyropoulos 2008 Greek object clitic pronouns: A typological survey of their grammatical properties. Universals and Language Typology. Greek in Typological Perspective 61: 39–53. Selkirk, Elisabeth 1980a The role of prosodic categories in English word stress. Linguistic Inquiry 11: 563–606. 1980b Prosodic domains in phonology: Sanskrit revisited. In: Mark Aronoff, and Mary-Louise Kean (eds.), Juncture, 107–129. Saratoga, CA: Anma Libri. 1981 On prosodic structure and its relation to syntactic structure. In: Thorstein Fretheim (ed.), Nordic Prosody 2, 111–140. Trondheim: Tapir. 1984 Phonology and Syntax: The Relation Between Sound and Structure. Cambridge, MA: MIT Press. 1986 On derived domains in sentence phonology. Phonology Yearbook 3: 371–405. 1995 The prosodic structure of function words. In: Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), University of Massachusetts Occasional Papers in Linguistics 18: Papers on Optimality

An interface approach to prosodic word recursion

133

Theory, 439–469. University of Massachusetts, Amherst: Graduate Linguistic Student Association. Selkirk, Elisabeth and Tong Shen 1990 Prosodic domains in Shanghai Chinese. In: Sharon Inkelas, and Draga Zec (eds.), The Phonology-Syntax Connection, 313–338. Chicago and London: The University of Chicago Press. Schreuder, Maartje 2006 Prosodic processes in language and music. Ph. D. diss., Rijksuniversiteit Groningen. Schreuder, Maartje and Dicky Gilbers 2004 Recursive patterns in phonological phrases. In: Bernard Bel, and Isabelle Marlien (eds.), Proceedings of Speech Prosody 2004, 341–344. Nara, Japan: ISCA (SproSig). Truckenbrodt, Hubert 1995 Phonological phrases: Their relation to syntax, focus, and prominence. Ph.D. diss., Massachusetts Institute of Technology. 1999 On the relation between syntactic phrases and phonological phrases. Linguistic Inquiry 30: 219–255. Uriagereka Juan 1999 Multiple Spell-Out. In: Samuel David Epstein, and Norbert Hornstein (eds.), Working Minimalism, 251–282. Cambridge, MA: MIT Press. Vig´ario, Marina 1999 On the prosodic status of stressless function words in European Portuguese. In: Tracy Alan Hall, and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 253–293. Amsterdam: John Benjamins. 2003 The Prosodic Word in European Portuguese. Berlin: Mouton de Gruyter. Vogel, Irene this volume The status of the Clitic Group.

The extended prosodic word Junko Ito and Armin Mester

1. Introduction In an extended prosodic word, a word-sized unit a has combined with additional material b to form, by adjunction, a larger recursive structure [w [w a] b] or [wb [wa]]. Such structures have often been posited, following Inkelas (1989) and others, when b is a loosely bound derivational or inflectional affix, replicating (more or less precisely, due to the pressures of syllabification constraints) morphological structure in prosody. This is a way of capturing the closure effects associated with the word cycle studied by Borowsky (1993) and others, resulting in sing#er without [g] (vs. finger with [g]), trisyllabic kind[l­ ]#ing V (vs. disyllabic kindling N ), etc. Even though sometimes seen as a virtue, the duplication here is worrisome. Paradigm constraints and other Output-Output (OO-) conditions, an important and widely accepted part of current Optimality Theory (OT), cover much the same ground, and the idea of writing closure effects into the phonological representation itself raises the suspicion that the structures posited are little more than a legacy from the days of the heavily representation-focused approach to phonology of the 1970’s and 80’s, a way of encoding effects that are better accounted for elsewhere in the grammar, by non-representational means. We are in basic agreement with this view, which removes much of the motivation for recursive prosodic tree structure that simply mirrors morphological form. What, then, is the status of extended prosodic words, and of adjunction in prosody? We argue that they still have a significant role to play, not as replicas of morphological structure within phonology, but rather at the post-lexical level, as one important option for the parsing of functional items, such as prepositions and determiners. This claim, which takes up earlier proposals by Booij (1996), Vig´ario (1999), and others, is in conflict with the direct phrase attachment of functional material favored by Selkirk (1996), whose influential work was followed by many researchers. The two opposing views are juxtaposed in (1). (1)

a. phrase-attached: [f the [dinosaurs]]

b. word-adjoined: [w the [dinosaurs]]

136

Junko Ito and Armin Mester

The difference might appear small, but it has far-reaching consequences that this paper sets out to explore. Besides the status of the function word itself, an important point of divergence between the two structures lies in the prosodic status of the lexical word dinosaurs: It is a full and independent prosodic word in [f the [w dinosaurs]], but not in [w the [w dinosaurs]], where it is only a segment of a larger prosodic word. In terms of the approach developed in Ito and Mester (to appear), dinosaurs is a maximal projection of w in (1a), but not in (1b). This makes a difference to the extent that there are processes that are specific to maximal prosodic words. Insofar as the evidence to be reviewed here is concerned, we will see that the w-adjoined structure offers the desired structures for English and German.

2. Basic properties of prosodic representation A significant advance in the study of phonological form since the mid-1980’s has been the discovery that speech has its own kind of constituent structure, related to, but in many ways non-isomorphic with, the syntactic structure of utterances. In this general area, a specific line of research has emerged as the main-stream approach, based on precursors including Halliday (1960) and Pike (1967), and developed into a full-scale model in the work of Selkirk (1981), Nespor and Vogel (1986), and others. It conceives of phonological constituent structure as a hierarchy of a small number of well-defined prosodic constituents along the lines of (2), the so-called prosodic hierarchy. (2)

Prosodic Hierarchy:

utterance (u) | intonational phrase (i) | phonological phrase (f) | phonological word (w) | foot (f) | syllable (s)

The hierarchy sometimes contains another level subordinate to the syllable, the so-called “prosodic skeleton”, consisting of moraic (m) or segmental (X, C/V) units and expressing quantity and weight distinctions. Following Ito and Mester (2003 [1992]) and McCarthy and Prince (1993a), among others, we assume that

The extended prosodic word

137

skeletal units are not part of the prosodic hierarchy, and that quantity and weight are properties of syllables and segments, not prosodic constituents (see Ito and Mester 2003 for discussion.) While there are open questions, both of detail and of principle, regarding the exact nature and number of the categories above the prosodic word – indeed, regarding their very existence as separate categories, each with its own label and associated set of defining properties, as opposed to a simple string of unlabeled nodes in a metrical tree structure (Wagner 2005) – it is here convenient to take something like (2) as a broadly shared starting point. Prosodic representations are instances of the tree structures familiar from syntax and other parts of linguistics. Formally speaking, we can conceive of a prosodic representation as a directed graph consisting of a finite set of nodes, where each node is an ordered pair whose first member is a sequence of natural numbers , its address, and whose second member is a label L. An example is (3), given both in its algebraic form (a) and in a representation where precedence and dominance are graphically expressed (b). The latter is usually simplified to (c). (3)

a.

{(, w), (, f), (, f), (, s), (, s), (, s), (, s)} (, w)

b.

(, f)

(, f)

(, s) c.

(, s)

(, s)

(, s)

w f s

f s

s

s

The (unique) root of the tree has the address ; if a given node has the address , its immediate daughters have addresses , . . . , . Node A dominates node B if A’s address is and B’s address is ; if, in addition, b = i, for some number i, A dominates B immediately. Given a pair of nodes, A with address , and B with address such that j > i, A and every daughter of A precedes B, and every daughter of B. This means that prosodic trees are non-tangling in the sense of Wall (1972: 148).

138

Junko Ito and Armin Mester

Prosodic trees have a number of basic properties, as listed in (4). (4)

Basic properties of (well-formed) prosodic trees (see Wall 1972: 144– 152 for the formal background) a. Rootedness: There is exactly one node that dominates every other node. b. Linear order: The nodes immediately dominated by a node are linearly ordered from left to right. c. No tangling: For any nodes x and y, if x precedes y, then all nodes dominated by x precede all nodes dominated by y. This excludes both line crossing and improper bracketing (since nodes do not precede themselves). d. Labeling: Each node bears a label, an element of the ordered set PH = {u > i > f > w > f > s > m}, the prosodic hierarchy, whose elements stand in a relation of containment, as indicated. e. Containment: Each immediate dominance relation respects the containment structure of the prosodic hierarchy, in the sense that lower-ranked elements do not immediately dominate higher-ranked elements. f. Headedness: Every (non-terminal) prosodic category dominates a head, a prosodic category at the next lower level in the prosodic hierarchy.

In an optimality-theoretic conception of phonology, here assumed throughout, it is convenient to assign the conditions in (4) to Gen, given their basic nature and the fact that there is little reason to suspect any of them to ever be violated.1 They are no more than a minimum consensus on prosodic representations and admit a wide variety of structures, many of them never encountered. What else needs to be assumed in order to more precisely characterize 1

Some of them, such as Containment, clearly belong in Gen–syllables containing feet, etc., are simply not part of the toolbox of phonology. For others, this is more a question of convenience, not of principle. Thus, the precise constituency of the prosodic hierarchy is obviously not ready to be carved in stone at this point, and still other conditions might be violable, such as the proper bracketing aspect of NoTangling (as argued by Hyde 2002) and Headedness. To avoid misunderstandings, we note that the one-to-many links between segmental structure and the prosodic tree, as in geminates or ambisyllabic consonants, do not violate the proper bracketing of the tree itself (see Ito and Mester 2003: 35 for discussion).

The extended prosodic word

139

the form of prosodic representations? Building on previous work, we assume four kinds of requirements, which we will take up one-by-one in what follows. (5)

Constraints on prosodic representations a. b. c. d.

2.1.

Parsing constraints Interface conditions Size and shape requirements Tree form restrictions

Parse-into-X

Prosodic parsing is the assignment of constituent structure at each of the levels of the prosodic hierarchy. In OT the basic parsing imperative is implemented as a set of specific constraints, often crucially dominated, requiring parsing at each level. This is the strategy taken by Prince and Smolensky (2004 [1993]: 64), who refer to “Parse-m, which must be distinguished from the other members of the Parse-element family”, such as Parse-segment (Prince and Smolensky 2004, note 40). In another context, they note “the effects of a principle of exhaustive metrical analysis, familiar from the earliest work in the area [. . . ]. This principle is part of the parsing theory, and we will call it Parse-s, omitting from the name the information that s is parsed into F” (Prince and Smolensky 2004: 62). Overall, the family of Parse-element constraints plays a fundamental and ubiquitous role in their work. Extending this approach, with some modifications, we conceive of prosodic parsing as driven by a family of OT constraints Parseinto-X, where X is an open parameter standing for the levels of the prosodic hierarchy (2). (6)

Parse-into-X: {Parse-into-s, Parse-into-f, Parse-into-w, Parse-into-f, Parse-into-i, Parse-into-u} “Every element of the terminal string is parsed at the X-level”

The elements of the terminal string are the phonological segments. Parseinto-X only requires that segmental strings belong to X, and is crucially not “Parse-into-wellformed-X”. For example, Parse-into-f and Parse-into-s are themselves not concerned with the wellformedness of feet and syllables. These units have their own intrinsic defining properties stated as separate constraints (for feet, binarity and rhythmic requirements; for syllables, sonority requirements, coda restrictions, complexity, etc.).

140

Junko Ito and Armin Mester

Full parsing can be seen in an example like (7). We often use equivalent bracketed representations, as in (8), where periods mark syllable boundaries, parentheses show foot boundaries, and brackets with subscripts indicate higherlevel constituents. (7)

(8)

[u [i [f [w (."daI.n@.)(.­sOôz.)]][f [w (.ô"oUmd.)] [w (.­æ.ôI.) (."zoU.n@.)]]]]

Full parsing at every level is enforced by the family of Parse-into-X constraints in (6). To see how these constraints work, consider various parsings of the utterance dinosaurs! given in (9): The totally unparsed (a), syllabified (b), syllabified and footed (c), etc. The fully parsed (g) shows all the levels of the prosodic hierarchy. Given the simple formulation of the parsing constraints in (6), the target is in each case the terminal string of segments, and it is at this level that violations are reckoned, as shown in (9). (9) Parseinto-i daIn@sOôz daIn@sOôz daIn@sOôz daIn@sOôz daIn@sOôz

Parseinto-u daIn@sOôz daIn@sOôz daIn@sOôz daIn@sOôz daIn@sOôz daIn@sOôz

-------------------

Parseinto-f daIn@sOôz daIn@sOôz daIn@sOôz daIn@sOôz

-------------------

Parseinto-w daIn@sOôz daIn@sOôz daIn@sOôz

-------------------

Parseinto-f daIn@sOôz daIn@sOôz

-------------------

Parseinto-s daIn@sOôz

-------------------

/daI n@ sOôz/ ‘dinosaurs’ a. daI n@ sOôz b. [email protected]ôz. c. (.daI.n@.)(.sOôz.) d. [w (.daI.n@.)(.sOôz.)] e. [f [w (.daI.n@.)(.sOôz.)]] f. [i [f [w (.daI.n@.)(.sOôz.)]]] g. [u [i [f [w (.daI.n@.)(.sOôz.)]]]]

The constraints are unranked in this tableau (indicated by vertical broken lines), and the less structure the candidates have at the various levels, the less harmonic

The extended prosodic word

141

they are. There are also output candidates where parsing has skipped a level, either partially or entirely, as illustrated in (10).

daIn@sOôz daIn@

Parseinto-w daIn@sOôz

Parseinto-f

Parseinto-i

-------------

[u [i [f [w (.daI.n@).(.sOôz.) ]]]] [u [i [f (.daI.n@).(.sOôz.) ]]] [u [i [f [w [email protected]ôz. ]]]] [u [i [f [w .daI.n@.(.sOôz.) ]]]]

Parseinto-f

-------------

a. b. c. d.

Parseinto-s

-------------

/daI n@ sOôz/ ‘dinosaurs’

-------------

Partially parsed candidates -------------

(10)

Parseinto-u

Candidate (10b) does not parse its segments into w, and (10c) does not parse them into f (feet). There can also be candidates where parsing into a level is partial, as illustrated in (10d) for f. In this example, none of these partially parsed candidates is of much interest, since the fully parsed (10a) is the clear winner, fulfilling all Parse-into-X constraints. Depending on the segmental make-up, however, there can be situations where the non-fully parsed candidate will be the winner. This is a result of interactions with other constraints, as will be shown next. 2.2. Interactions of Parse-into-X with other constraints A basic question concerns the fact that prosodic structure is generally maximally articulated. Since Parse-into-X demands only that the terminal string be dominated by X, why build more than the minimally required single X? For example, why does the whole sentence not become one gigantic prosodic word? This question does not arise in the same way at the lower levels because there are well-known substantive constraints on feet and syllables that prevent oversized constituents. The whole sentence cannot be one gigantic foot because constraints on foot form allow maximally two syllables per foot (FootBinarity (FtBin): “Feet must be binary under syllabic or moraic analysis”, McCarthy and Prince 1993b: 46). Similarly, forming one gigantic syllable is out of the question because constraints on syllable form and sonority sequencing prevent incorporating a consonant into a syllable after a vowel as a coda, and then adding a second vowel to the same syllable. However, at higher levels of structure, such answers are not readily available. Consider a parse like (11) (where syllable and foot structure is assumed to be present). (11)

[u [i [f [w Dinosaurs roamed Arizona]]]]

All Parse-into-X constraints are fulfilled in (11), and English words can clearly contain more than seven syllables, as shown by formaldehydetetramethylamido-

142

Junko Ito and Armin Mester

fluorimum, or the notorious floccinaucinihilipilification ‘estimation of something as worthless’. So what prevents a parse as in (11)? The answer lies in constraints on the syntax-phonology interface demanding, generally speaking, that a substring that forms a (certain kind of) syntactic constituent must simultaneously also form a (certain kind of) prosodic constituent.2 Prince and Smolensky (2004: 45) propose a family of constraints Lx≈Pr (MCat) that directly requires a member of the morphological category MCat to correspond to a prosodic word. A slightly different approach, originated in Chen (1985, 1987) and Selkirk (1986) and later generalized in McCarthy and Prince (1993a), Selkirk (1996), and Truckenbrodt (1995, 1999), sees the imperative of correspondence as a more indirect one: One edge, left or right, of certain syntactic constituents must match an edge of a specific prosodic constituent.Taking up this line of analysis and writing Lex for “lexical (non-function) word” and LexP for a corresponding maximal projection, we formulate the relevant interface constraints in (12). (12)

Lex-Mapping a. Lex-to-w(L/R):

Align(Lex, Left/Right, w, Left/Right) “Every lexical word is left/right aligned with a prosodic word.”

b. LexP-to-f(L/R): Align(LexP, Left/Right, f, Left/Right) “Every lexical maximal projection is left/right aligned with a phonological phrase.” For English, Lex-to-w(L) and LexP-to-f(R) are high-ranking. Tableau (13) shows these constraints in action. (13) S VP NP a. b. c. 

2

Dinosaurs [f [w Dinosaurs [f [w Dinosaurs] [f [w Dinosaurs]]

V

NP

Lex-to-w(L) LexP-to-f(R)

roamed Arizona roamed Arizona]] ** [w roamed] [w Arizona]] [f [w roamed] [w Arizona]]

* *

We are using “syntactic” here to include both syntactic and morphological structure (i.e., “grammatical”, in traditional parlance).

The extended prosodic word

143

OT constraints are violable, and whether and to what extent they are fulfilled depends on their ranking. The parsing constraints are no exception, and if a constraint on foot form such as FtBin is ranked above Parse-into-f, nonexhaustive parsing is the result. For example, in (14a) we find a syllable immediately dominated by the word, without an intervening foot (i.e., a violation of Parse-into-f). The competing structure (14b), with a foot on [f b´], violates FtBin, the higher-ranking constraint on foot form which requires two moras and is violated by the monomoraic b´. (14)

banana

FtBin

Parse-into-f

a. w b@

f s s s  [b@(næ n@)] b. w f

f

*!

s s s [(b@)(næ n@)] In the analysis of Latin stress in Prince and Smolensky (2004), followed by many others employing exactly this ranking, the tacit assumption is that other Parse-into-X constraints are unviolated in the winning candidate. This is then an example of a size constraint ranked above a Parse-into-X constraint. (15)

Size Constraint  Parse-into-X (e.g., FtBin) (e.g., Parse-into-f)

Selkirk (1996) argues that function words in English exhibit Parse-into-w violations (her Exhaustivity-phrase). We see from (14) that ranking the size constraint FtBin above Parse-into-f leads to output structures in which Parse-into-f is violated by a single light syllable directly dominated by w. There is one situation, however, where such Parse-into-f violations lead to ungrammatical results, namely, when the form contains no other material except for the (unstressed) light syllable: A word that just consists of [b?], the first part of [b @(næn@)], is illicit. What is the difference between free-standing [w b@] (16b) and [b@] in [b@(næn@)] (16a)?

144 (16)

Junko Ito and Armin Mester

a.

w

b.

*w

f s s s [b@(næn@)]

s [b@]

On the empirical side, we know from minimal word effects and similar phenomena in language after language that a single light syllable is often not sufficient to form a prosodic word. The general explanation, originated by Prince (1980), lies in the Headedness requirement (4f), repeated here in (17). (17)

Headedness:

Every (non-terminal) prosodic category dominates a head, a prosodic category at the next lower level in the prosodic hierarchy.

Thus Headedness is fulfilled in [b@(næn@f )w ] because w dominates the foot (næn@f ), but it is violated in [b@w ]. The alternative parse, [(b@f )w ], fulfills Headedness but violates FtBin. The combined result of the two constraints is the generalization that a minimal word must be longer than a single light syllable. The Parse-into-X constraints demand that strings be assigned phonological constituent structure, and Lex-to-w and LexP-to-f prevent underparsing. What stops strings from being overparsed then? An example would be (18), where the last two syllables are parsed not only as a foot, as in (16a), but in addition as a w. (18)

[w b@[w (næn@)]]

In terms of the parsing constraints, both violate Parse-into-f in terms of their first syllables, and are tied in this respect. The overparsed (18) contains an extra internal w, and the string næn@ is parsed twice by w, once by the internal w and again by the external w.3 With its two w-constituents, one dominating the other, (18) violates a specific tree form constraint, namely, No-Recursion. Different 3

A general constraint often invoked for such cases is *Struc (Prince and Smolensky 2004) which militates against structure of any kind. As noted by a reviewer, ranking *Struc higher than the Parse-into-X constraints will lead to unattested rankings where the winning candidate contains unprosodified strings. Stipulating a universally fixed ranking Parse-into-X » *Struc-X, (e.g., Parse-into-f » *Foot) is a covert way of restating the classical metrical principle that parsing involves maximization of structure. In the context of OT, Gouskova (2003) has shown that economy constraints of the *Struc family that penalize structure as such produce unattested patterns under reranking and should be excluded from CON.

The extended prosodic word

145

formulations of No-Recursion have been proposed, either as a categorical constraint directly banning self-domination (Selkirk 1996: 190), or as a gradiently violable constraint requiring any two prosodic constituents of the same category that are not disjoint in extension to be identical in extension (Truckenbrodt 1999: 240). Continuing the ‘string-and-parsing’ approach pursued so far, we formulate No-Recursion in (19) as a constraint against multiple parsing of a string by the same category. Informally speaking, (19) militates against one-to-many relations between strings and instances of categories. (19)

No-Recursion: An element is parsed only once into a given category. Assign one violation mark for each additional parse of an element into the same category.

As McCarthy and Prince (1993b) note, recursion of the categories foot and syllable is impossible because “the various principles of foot theory and syllable theory license a very limited set of expansions of foot and syllable,” but “there is no theory placing comparable limits on the expansion of the prosodic word”. This distinction is also made in Ito and Mester (2007a) in terms of intrinsically versus extrinsically defined prosodic categories. The former are governed by substantive constraints defining their size and shape. Our working hypothesis is that the most basic structural requirements on syllables as well as fundamental rhythmic conditions on feet are universally unviolated (in OT terms, perhaps part of Gen) and preclude recursion. On the other hand, prosodic words and larger prosodic units do not have intrinsically defined shapes but are largely governed by syntax-phonology mapping constraints, which may therefore lead to recursive structures being optimal, depending on the ranking of No-Recursion (19) (see Ladd 1986 for an argument that intonational phrasing can be recursive). The constraints discussed so far play an important role in determining the structure of complexes consisting of function words and lexical words. Previewing a fuller discussion and motivation in later sections, we note here that the independently motivated constraints give us the following options for such structures.

(ð@)

[Lex d

No -Recursion

daIn@sOôz

Parseinto-w

ð@

-------------

a. [f [w (ð@)] [ (daIn@)(sOôz)]] b. [f [w ð@ (daIn@)(sOôz)]] c. [f [w ð@ [w (daIn@)(sOôz)]]] d. [f ð@ [w (daIn@)(sOôz)]]

Lexto-w(L)

-------------

FtBin

-------------

[fnc the] [lex dinosaurs]

-------------

(20) Parseinto-f ð@ ð@ ð@

146

Junko Ito and Armin Mester

Making the determiner an independent w, (20a) violates FtBin, since a single monomoraic syllable cannot make a foot.4 Fully incorporating the determiner (20b) violates Lex-to-w, since no prosodic edge marks the left Lex-edge of dinosaur. Adjoining the determiner (20c) to the following w violates NoRecursivitiy; and attaching the determiner directly to the f (20d) violates Parse-into-w.5 Each one of these structures is potentially optimal in some grammar, although some are more likely than others. We will argue that (20c), with its recursive w-structure, is the correct one for English, contra Selkirk (1996), who has argued for (20d). 2.3. Exhaustivity effects and strict layering Proposed systems of prosodic hierarchy constraints typically include a third type of constraint whose task is to enforce strict layering. Definitional details aside, “strict layering” describes a situation where full parsing is achieved at every level, so all of the Parse-into-X constraints are fulfilled, as is No-Recursion. This was once the standard view of prosodic form. Tentatively suggested, and still remaining nameless, in Selkirk (1981: 381) as something generally true of prosodic representations, the idea acquired a more assertive status in subsequent work, together with its name. (21)

Strict Layer Hypothesis (Selkirk 1984: 26) “We have proposed that a category of level i in the hierarchy immediately dominates a (sequence of) categories of level i-1 [. . . ]. We will call this the strict layer hypothesis, and will take it as a useful working hypothesis here”.

The same work, however, by adopting a grid-only framework (originated by Prince 1983), abolishes the very core of the prosodic hierarchy – the foot together with the prosodic word and the phonological phrase – and replaces phrasal prosodic constituent structure by silent grid beats occupying the junctures between words and phrases. With the remaining endpoints of the old hierarchy, the intonational phrase and the syllable, now in immediate domination (see Selkirk 1984: 31), the spirit (though not the letter) of strict layering seems violated, and the principle is in any case left with very little work to do.

4

5

Leaving the determiner unfooted but parsed as a prosodic word [w ð@] violates the universal Headedness requirement, and is ruled out as in (16) above. Note that (20d) does not violate Lex-to-w since the determiner the is a function word.

The extended prosodic word

147

The prevailing approach, however, as represented by Nespor and Vogel (1986), retained a richer conception of the hierarchy, and here strict layering gained general currency, resulting in representations where prosodic trees can always be “sliced” into the various prosodic levels: Levels are never skipped, nor repeated.6 Subsequent analyses, adopting strict layering without discussion, were faced with two chronic problems: “Too little structure” and “too much structure”. The “too little structure” problem arises because strict layering caps the maximum depth of prosodic trees at n, where n is the number of categories in the prosodic hierarchy. The empirical evidence showed, however, that prosodic structure is more articulate than what can be captured by a closed system of this kind. While it clearly does not have the degree of recursivity seen in syntax, Ladd (1986) presented strong evidence for recursion at the intonational phrase level, arguing that the theory widely overshoots its mark by ruling out all recursivity whatsoever (e.g., as in Selkirk 1986: 384: “[P]rosodic constituents of a same category are not nested”). The “too little structure” problem prompted researchers to posit new categories whenever the prosodic parses required more structure than what was provided by the existing categories. This led to a proliferation of categories (“clitic group”, “intermediate phrase”, “accentual phrase”, “minor phrase”, “major phrase”, “superordinate minor phrase”, etc.; see Ito and Mester 2007b for further discussion). With each new category, any hope of substantiating the hierarchy as a truly universal one receded further into the distance. Ironically, the strict layering doctrine is also confronted with the opposite problem: “Too much structure”. In order to fulfill strict layering, representations have to be padded to fill every layer, leading to the positing of categories emptied of their intrinsic meaning. This line of argument was laid out in detail in Ito and Mester (2003) with evidence from Japanese showing that a form like anime is structured as [w (f s˘ s˘ )s˘ ], with an unfooted last syllable directly dominated by w, not as [w (f s˘ s˘ )(“f” s˘ )], with a degenerate second foot “f”, or as [w (F (f s˘ s˘ )s˘ )], with a superordinate “superfoot” F.

6

Interestingly, in their earlier work Nespor and Vogel (1983) had still made crucial use of recursive prosodic structures that violated strict layering. As Ladd (1996) points out, strict layering seems to have quickly gained the status of a self-evident truth, so much that it was literally built into the formalization of prosodic structure proposed in Pierrehumbert and Beckman (1988) without so much as reference either to the name of the hypothesis (“strict layering”) or to the work it was first introduced (Selkirk 1984).

148

Junko Ito and Armin Mester

Empirical evidence for strict foot-layering would be a language similar to English where words like ba"nana are actually parsed as [w (f ­beI)(f "næn@)] with a secondary stress (and a full vowel) in the first syllable that is exclusively due to the extra foot required by strict layering. It turned out that this kind of outcome is rarely, if ever, attested, and language after language instead presents us with outcomes like the one in English, well captured by the representation [w b@(f "næn@)] with an unfooted first syllable in violation of strict layering. Positing a representation like [w (f b@)(f "næn@)], to live up to the facts while upholding the letter of strict layering, is counterproductive because it deprives the category foot of its intrinsic meaning for a language like English, leading to constructs like “stressless foot”, “headless foot”, etc. In such approaches, strict layering has become a mere convention of how to draw prosodic trees, and few, if any, empirical claims are connected with the constituency posited. The weak layering theory of prosodic structure of Ito and Mester (2003) expresses this kind of view, and presents it as a response to the “too much structure” problem. “Strict layering” is no longer a monolithic requirement, but is reduced to its more basic components, including constraints enforcing headedness, exhaustive parsing, and left alignment of morphological word to foot. The main interest of this work was on word-internal prosody and prosodic morphology, and the goal was to find a theoretical framework that would make sense of the empirically justified distribution of feet, as described earlier. This is the approach that was later incorporated into OT. In the formalizations of weakly layered conceptions that appeared in the mid-90’s (most prominently Selkirk 1996, a study of function words; see also Peperkamp 1997, among others), other constraints enforcing strict layering were also proposed as violable constraints. In particular, Selkirk proposes a specific constraint (dubbed Exhaustivity) that enforces strict layering, replicating earlier proposals in pre-OT work. (22)

Exhaustivity (Selkirk 1996: 190): No Ci immediately dominates a constituent Cj , j < i-1 [. . . ]

At first glance, this seems to capture the essence of the weak layering conception in a very straightforward way. On second thought, such direct transfers of earlier proposals into OT-phonology are often problematic. Here, it turns out that Exhaustivity (split into subconstraints such as Exhaustivity-f, Exhaustivity-w, etc.) is redundant with respect to Parse-into-X. The perspective is certainly different: Whereas Parse-into-X starts with the terminal string and asks whether it has been parsed into all the levels of the prosodic hierarchy, Exhaustivity looks at the tree structure itself, and scrutinizes the daughters of every node.

The extended prosodic word

149

However, it is still true that a candidate that entirely fulfills Parse-into-X, for all levels X of the hierarchy, cannot help also fulfilling Exhaustivity, showing that a theory that contains both groups of constraints is redundant. We follow the basic approach to parsing theory in OT laid out in Prince and Smolensky (2004) (see also section 2.1 above) and interpret exhaustivity not as a separate constraint, but as an effect of Parse-into-X. 3. Function word complexes in English and German We turn now to the main topic of this paper, namely, the prosodic form assigned to syntactic structures headed by functional categories as in (23). (23)

a.

DP

b.

IP

c.

PP

D

NP

I

VP

P

NP

the fnc

N dinosaurs lex

could fnc

V live lex

on fnc

N earth lex

Such function word complexes, illustrated in (24) for a variety of languages, typically have a characteristic prosodic form where the function word is tightly bound to its lexical host and in some sense subordinated to it. (24)

til Rhodos a Rodi nach Rhodos to Rhodes a Rhodos

Danish Italian German English Spanish

a` Rhodes naar Rhodos rodosu-e eic Rodon

French Dutch Japanese Greek

Generally speaking, prosodic subordination is reflected in the exemption of functional categories and their projections from mapping constraints (see e.g. Truckenbrodt 1999: 226). In our terms, D, I, P, and other functional categories are not subject to Lex-Mapping (12).7 7

A reviewer points out that “function word” may not be a well-defined and independently motivated notion. For example, Wagner (2005) argues that the distribution of function words reflects the fact that such words occur with a high frequency in prosodically subordinated environments and therefore tend to develop reduced allomorphs over time. As we will argue in the next section, even prosodically unreduced function words like supposeta pattern with function words and not with lexical words, which indicates that the frequency-based explanation might not be sufficient. See

150

Junko Ito and Armin Mester

This much is standard practice – but what exactly does it mean for a functional element to be prosodically integrated with its lexical host? Earlier work on the prosodic hierarchy subscribing to strict layering (Hayes 1984, 1989, further developed in Nespor and Vogel 1986) introduced a specific constituent, the “Clitic Group”, to parse exactly such combinations. The Clitic Group as a category different from word and phrase has remained controversial, 8 however, and in line with our general strategy favoring a sparse hierarchy with essential appeal to adjunction and relational structure (Ito and Mester 2007ab), we follow Selkirk (1996) in identifying four possible sites for function words in a prosodic structure without any categories intervening between w and f. (25)

Prosodic sites for function words a. full-w fnc b. f

c.

w

w

fnc

lex

w-adjoined fnc f

amalgamated fnc f w fnc

d.

lex

f-attached fnc f

w s

w

s

w

fnc

lex

fnc

lex

These four structures are exactly those predicted to arise from the interaction of the parsing constraints and other relevant constraints discussed in section 1.2.

8

also Keenan and Stabler (1994) for an interesting approach distinguishing open and closed class items on a formal basis. Perhaps the strongest case for its existence involves Italian troncamento (Nespor 1990), with a distinction between obligatory application within the Clitic Group (dar-gli, *dare-gli ‘give them’) and optional application in higher domains (far lezione∼fare lezione ‘to teach’). Taking a new look at the relevant facts and generalizations, however, Meinschaefer (2005) argues that a superior analysis is available which makes no use of the Clitic Group and instead assigns troncamento to the phonological phrase, with all optionality of application reduced to optionality of restructuring.

151

The extended prosodic word

The (unranked) tableau is repeated here in (26), using a German example of a Det-N sequence with a reduced form of the indefinite determiner (’ne instead of eine), see Hall (1999: 104).

> [Lex ts

NoRec

> tsigarEt@

Parseinto-w

n@

----------------

(n@)

Lexto-w (L)

----------------

FtBin

----------------

(Willste) [fnc ’ne ] [lex Zigarette ]? ’(Would you like) a cigarette?’ > a. . . . [f [w (n@)] [w (­tsiga)("rε t@]] > b. . . . [f [w n@ (­tsiga)("rε t@)]] > c. . . . [f [w n@ [w (­tsiga)("rε t@)]]] > d. . . . [f n@ [w (­tsiga)("rε t@)]]

----------------

(26) Parseinto-f n@ n@ n@

The optimal parsing of fnc-lex structures thus depends on the ranking of the constraints FtBin, Lex-to-w, No-Recursion, and Parse-into-w/f. Several questions arise at this point. Which structure is correct for a particular language, such as English or German? Is each of the four structures instantiated in some language? In other words, are all four constraints violable? Or are some of them undominated in every grammar and thus potentially assigned to GEN, together with the basic representational constraints in (4)? For example, FtBin has sometimes been argued to be universally unviolable. A number of crosslinguistic studies (including Selkirk 1996 for English and Serbo-Croatian and Peperkamp 1997 for various Italian dialects) have addressed these questions, with the goal of matching the empirically attested typology of fnc-lex structures with a factorial typology like the one in (25) (see also the papers collected in Hall and Kleinhenz 1999). Our goal here is the more modest one of (re)asking the question regarding the prosodic structure of such fnc-lex combinations in English and German, two of the more well-discussed cases in the literature. In an influential paper, Selkirk (1996) opts for the f-attached structure (25d) for English proclitics, and this position has been adopted for German proclitics in Hall (1999), as well as in a subsequent study by Kabak and Schiering (2006). We will review the previous arguments here, and consider new evidence along the way. From the perspective of the Parse-into-X approach advocated here, it turns out that for both English and German, the overall evidence favors wadjunction (25c) over f-attachment (25d), echoing a line of analysis familiar from earlier work (such as Booij’s (1996) treatment of proclitics in Dutch).

152

Junko Ito and Armin Mester

3.1. Between independence and fusion Past research has uncovered a significant amount of phonotactic and prosodic evidence showing that in many languages (including English and German) function words are neither structured as independent w’s (25a), nor simply prosodically fused with neighboring lexical words into a single w (25b). Informally speaking, (25a) gives too much independence and weight to the prosodically deficient function word, whereas (25b) gives too little prosodic independence to the lexical word. The structures in (25c) and (25d) avoid these problems: The function word is by itself not given prosodic word status, while the lexical word keeps its prosodic integrity. The unavailability of fully independent (27a) and fully fused (27b) parses in a language follows from high-ranking FtBin and Lex-to-w, respectively: The first constraint militates against establishing a subminimal fnc as a w, the second against turning a lexical word into a prosodic non-entity.

FtBin [w (fnc)]!

Lex-to-w (L) [Lex !

-----------

[w fncw ] [w lex w ] [w fnclex w ] [w fnc [w lex w ] w ] [f fnc [w lex w ] f ]

-----------

fnc status: a. independent b. fused c.  w-adjoined d.  f-attached

-----------

(27) No-Rec

lex

Parse-into-w

fnc

In English (28a) and German (28b) (both standardly analyzed with trochaic foot structure and a right-aligned quantity-sensitive head trochee, see Pater 2000 and Alber 1998), single pretonic syllables remain stressless in initial position. > ‘allege’ (28) a. [˘@("lEdZ)] [b˘@("næn@)] ‘banana’ a("drEs@)] b. [˘ Adresse ‘(the) address’ a("Si:n@)] Maschine ‘(the) machine’ [m˘ Foot parsing is non-exhaustive in (28) because Parse-into-f is dominated by FtBin, as shown in (29) (see also section 2.2).9 (29)

a. 

9

[lex allege] > [w (­@)(­lEdZ)] > [w ˘@ ("lEdZ)]

In addition, because of a constraint against stress clash.

FtBin

Parse-into-f

(@)!

˘@

The extended prosodic word

b. 

[lex Adresse] ‘(the) address’

FtBin

[w (­a)("drEs@)]

(a)!

[

˘a ("drEs@)]

153

Parse-into-f a˘

On the other hand, we find secondary stress on the initial syllable in morphologically related forms in (30). (30)

a. b.

­alle"gation ­phone"tician ­adres"sieren ­protes"t´ıeren

cf. a˘ l"lege cf. ph˘o"netics ˘ "dresse cf. A cf. Pr˘o"test

‘(to) address, (the) address’ ‘(to) protest, (the) protest’

This is because Parse-into-f has a free hand to assign foot structure to initial sequences of syllables (31), forestalling initial lapses.10 (31)

a.

[Lex allegation] 

b.

FtBin

[w (­æl@)("geISn)] " [w ˘@l˘@ ("geISn)] " [Lex adressieren] ‘to address’



Parse-into-f

@˘ l˘@! FtBin

Parse-into-f

[w (­adrE)("si:r@n)] [w ˘ adr˘E(si:r@n)]

a˘ dr˘E!

Although morphologically complex, the forms in (30) constitute single prosodic words and are rhythmically different from the function word–lexical word complexes in (32). The latter, but not the former, show initial lapses: A sequence of two unstressed syllables constituted by a subminimal function word (the indefinite a and its German equivalent ’ne) followed by the unfooted initial syllable of the lexical word. (32)

10

fnc-lex complexes a. [˘@ l˘@ ("gu:n@)] [˘@ m˘@ ("Si:n)] [˘@ m˘@ ("sa:Z)] [˘@ b˘@ ("kwEst)]

a laguna a machine a massage a bequest

Odd-numbered initial sequences of light syllables receive left-aligned footing (McCarthy and Prince 1993a).

154

Junko Ito and Armin Mester

b.

a ("gu: n@)] [n˘@ l˘ [n˘@ m˘ a ("Si: n@)] a ("sa: Z@)] [n˘@ m˘ [n˘@ b˘@ ("dI NUN) ]

’ne Lagune ’ne Maschine ’ne Massage ’ne Bedingung

‘a laguna’ ‘a machine’ ‘a massage’ ‘a condition’

Here, the monosyllabic function word fails to form a foot together with the unfooted initial syllable of the lexical word because the high-ranking syntaxprosody mapping constraint Lex-to-w(Left) plays a crucial role. (33) a. b. c.  d. 

fnc status independent fused w-adjoined f-attached

[ fnc a ][lex laguna] [w (­@)] [w l@ ("gu:n@)] [w (­æl@) ("gu:n@)] [w˘@ [w l˘@ ("gu:n@)] ] [f˘@ [w l˘@ ("gu:n@)] ]

FtBin (@)!

fnc status independent fused w-adjoined f-attached

[fnc ne ][lex maschine] a("SI:n@)] [w (­n@)] [w m˘ [w (­n@ ma) ("Si:n@) ] [w n˘@ [w m˘ a ("Si:n@)] ] a("Si:n@)] ] [f n˘@ [w m˘

FtBin (n@)!

Lex-to-w (L)

Parse-into-f l@

[Lex l!

@ l@ @ l@

(34) a. b. c.  d. 

Lex-to-w (L)

Parse-into-f m˘a

[Lex m! n˘@ m˘a n˘@ m˘a

The overall result is that in both English (33) and German (34), the w-adjoined (c) and f-attached (d) candidates trump the remaining candidates since they alone fulfill the top-ranked constraints: Parsing the function word as an independent w (a) violates FtBin,11 and the amalgamated fnc-structures (b) violate Lex-tow(Left).12 The important point here is that without Lex-to-w, the candidates with secondary stress on the function word a (33b) and ’ne (34b) would have been the winners (compare the situation in (31), where Lex-to-w is irrelevant). Seen in the light of (30) and (32), we need to ask why it is that the wadjoined (33c)/(34c) and f-attached (33d)/(34d) structures have two adjacent 11

There is no necessity here to appeal to a specific requirement on prosodic words (such as Hall’s (1999: 106) minimal word requirement): the independently needed requirements on foot size, parsing requirement, and headedness are fully sufficient. Although Hall’s bimoraic requirement on prosodic words correctly describes the facts, it simply stipulates the descriptive generalization. Note also that another one of Hall’s (1999: 114) prosodic word constraints (“A prosodic word must contain at least one full vowel”) is also a consequence of independent factors (headedness, and a constraint on foot heads ruling out schwa in a stressed syllable). 12 The parse [w˘@ l˘@ ("gu:n@)] violates both Lex-to-w(Left) and Parse-into-f.

The extended prosodic word

155

unstressed syllables, violating Parse-into-f, as schematically indicated in (35a). Why does the grammar not instead select the structure [(­æ[l@) ("gu:n@)] ], where the two syllables are footed? The answer is that the resulting structure (35b) violates proper bracketing (see (4b) above), a basic requirement on prosodic tree structures that, we assume, is properly a part of GEN. (35)

The upshot is that a sequence of unstressed syllables is allowed to arise in adjoined structures, but not in prosodically simplex words like (31). For German, the w-initial Parse-into-f effect has an interesting repercussion in the allomorphy of the past participle of verbs. The past participial form adds, besides a suffix (-(e)t for weak verbs, -en for strong verbs), an unstressed prefix ge- (lieben, ge-lieb-t, ‘love, loved’; stehen, ge-stand-en ‘stand, stood’), a combination sometimes regarded as a “circumfix”. The prefix ge- appears only when the following syllable carries the main word stress (36a), and is absent otherwise (36b). When the location of main stress varies, the presence/absence of ge- varies accordingly (36c). (36)

The prefix ge- in participles (hat) *"predig-t a. (hat) g˘e"predig-t (hat) *"kiebitz-t (hat) g˘e"kiebitz-t (hat) st˘u"dier-t b. (hat) *g˘e-st˘u"diert (hat) *g˘e-schm˘a"rotz-t (hat) schm˘a"rotz-t (hat) ­lieb"kos-t c. (hat) g˘e-"lieb­kos-t

‘(has) preached’ ‘(has) spied’ ‘(has) studied’ ‘(has) sponged’ ‘(has) caressed’

Details depend on the analysis (see Kiparsky 1966 for the first formal treatment, and Giegerich 1985, Neef 1996, Wiese 1996, and Geilfuß-Wolfgang 1998 for further developments), but it seems promising to interpret the rule of ge-distribution as part of the w-initial Parse-into-f effects, strengthened so

156

Junko Ito and Armin Mester

as to allow attachment of the unstressable prefix only to main stress. 13 Thus, [g˘e("predigt)] with an initial unstressed syllable is allowed, but [g˘est˘u( "diert)] with two initial unstressed syllables is disallowed. Rather than forcing a stress on the schwa syllable ge by footing it together with the initial syllable of the stem [(­ge.stu)("diert)], the allomorph without the ge-prefix is chosen [st˘u("diert)].14 Besides the prosodic (rhythmic) evidence showing that every lexical word must initiate a w, there is also segmental phonotactic evidence in American English against prosodifying the functional complex into a single w. In identical prosodic environments (stress vs. unstress), the juncture between function word and lexical word acts phonotactically as a w-juncture. In fnc-lex combinations (a"[R]Annie’s) and lex-lex combinations (ea"[R]apples), onsets of stressed syllables show flapped [R]. This is in sharp contrast with the word-internal situation, where such flapping is impossible (a"[th ]ack, *a"[R]ack).15 Similarly in unstressed environments, clearly w-internal /t/ is flapped (pi[R]˘y, capi[R]˘alism, see Steriade 2000 and Davis 2005 for different analyses of the additional paradigm uniformity factors at work here) but lex-initial /t/ is aspirated in fnc-lex combinations (a [th ]˘omato) just as in lex-lex combinations (grow [th ]˘omatoes).

13

The ge-rule extends to the nominal ge-prefix in pejorative nouns (Ge-sing-e ‘singing’, Getanz-e dancing, etc.), which is significant because the latter can be attached quite productively in colloquial speech. Harden (2003) shows that speakers’ judgments when presented with nonce-forms where the prefix is not followed by main stress essentially follow the rule, with some interesting variation in judgments between fully unacceptable and marginal forms (*Ge-be-wirt-e ‘hosting’, ?Ge-telefonier-e ‘calling’, ?Ge-studier-e ‘studying’). Overall, the very existence of the ge-rule is an indication that the schwa-containing prefixes ge- and be-, different from prosodically non-cohering unstressed prefixes like ent- and ver- (which do not allow resyllabification), are prosodically integrated with the stem in a simplex w, and are not part of a recursive structure (such as argued by Booij 1996 for Dutch). Otherwise (cf. (29)), there is no systematic reason why ge- should be absent in (36b). 14 Some examples provided by a reviewer with a sequence of otherwise unstressed verbal prefixes (e.g., ver-) show the robustness of Parse-into-f in prosodically simplex words. In order to avoid stressing the ver-prefix, the past participle forms of ver"haften ‘arrest’ and ver"kaufen ‘sell’ also choose the ge-less allomorph: ver("haf.tet), not *(­ver.ge)("haf.tet), and ver("kauft), not *(­ver.ge)("kauft). However, when the stem itself begins with a stressless syllable (st˘u"dieren ‘study’) or with a stressless prefix (B˘e"amter ‘civil servant’), ver- can receive secondary stress, as in (­verstu)("diert) ‘mis-studied’, (­verbe)("amten) ‘make into a civil servant’. In this case, there is no stem allomorph that improves the prosody: *ver(diert), *ver(amten). Similarly, un"glaublich ‘unbelievable’ but with secondary stress in ­unver"antwortlich. 15 See Ito and Mester (2007) for further discussion regarding the word-onset requirement forcing ambisyllabicity.

The extended prosodic word

157

The syllabification patterns of German also argue against full w-integration of the functional word complex. Contrary to what full prosodic fusion would predict, there is no re- or ambi-syllabification to onset in fnc-lex complexes (.ein."Au.to. ‘a car’, .ein.A."the.ner. ‘an Athenian’, *ei.nAu.to. *ei.nA.the.ner.), just as in lex-lex combinations (.kauft.Au.tos. ‘buys cars’ *.kauf .tAu.tos.). This even holds for function words reduced to a single consonant, such as ’n for ein: ’n Auto and ’n Athener have syllabic nasals unconnected to the onset of the following syllable: [.n].Au.to., not *[.n.n]Au.to. or *[.n]Au.to.16 " " Overall, the case against fully independent [w fnc][w lex] and fully incorporated [w fnclex] parses of function word complexes is strong because of the variety of counterevidence, including both prosodic and phonotactic facts and generalizations. On the other hand, all the evidence considered is compatible with structures of the general form [x fnc[w lex]], whether x=w (w-adjoined fnc) or x=f (f-attached fnc), in both of which the lexical word has prosodic word status but the function word does not. We are left, then, with the choice between the two structures repeated here in (37). As indicated, the difference is whether Parse-into-w or No-Recursion is violated. Convincing evidence for either choice is harder to come by, and many researchers have simply assumed one or the other structure. The w-adjoined structure has been explicitly argued for by Booij (1996) for Dutch, Vig´ario (1999) for European Portuguese, and Ito and Mester (to appear) for English, among others. On the other hand, Selkirk (1996) and Hall (1999) have presented arguments for f-attachment of function words (in English and German, respectively). (37)

16

It is true that in Standard German lack of resyllabification goes hand-in-hand with systematic [P]-insertion at the beginning of vowel-initial words (.ein .[P]Au.to., etc.), but it would be wrong to conclude from this that everything can be reduced to the phonetic presence or absence of a glottal word onset. Among dialects without systematic [P]-insertion, some allow resyllabification (Swiss German) and some do not (Bavarian German, see also Gutch 1992), indicating that this is a separate parameter of variation (Kabak and Schiering 2006: 65).

158

Junko Ito and Armin Mester

Before laying out our case for w-adjunction, we first review the evidence that has been presented for f-attachment, which will turn out to be inconclusive. 3.2.

f-attachment: reassessing the arguments

The evidence presented in Selkirk (1996) and Hall (1999) in favor of f-attachment has a negative form: Instead of positively identifying some property of x in [x fnc[w lex]] that clearly belongs to f and not to w, hence x = f, x is claimed to lack a certain w-property and is for this reason alone set equal to f – faute de mieux, so to speak. Selkirk (1996: 197-198) argues as follows: In w-initial position, voiceless stops are aspirated even when the first syllable is unstressed and not foot-initial: grow th o˘ matoes, grow ph e˘ tunias, grow ch a˘ lendula. Recall that this was one of the phonotactic pieces of evidence in favor of the w-hood of lex in fnc-lex complexes [a [w th omato]], which both of the structures in (38) predict. However, in terms of the aspiration constraint on [w , the two structures make different predictions regarding the initial aspiration of the larger constituent. (38)

Pointing to examples as in (39), where the fnc-initial t’s can remain unaspirated (and be realized as flaps instead), Selkirk (1996) concludes that w-adjunction (38a) makes the wrong prediction and that (38b) is correct, where the function word is not w-parsed and immediately attached to f. (39)

a. b. c.

Take Grey [R]o London. They grow [R]o the sky. We’re going to fly [R]o Seattle on Monday.

The argument here implicitly makes a specific claim about the way domaininitial fortition works in the prosodic hierarchy: The left boundary of w is a position of fortition, but not the left boundary of f. This claim encounters some

The extended prosodic word

159

problems. First, we find aspiration marking at the left boundaries not only of f 17 and w, but also of i and u – in particular, as (40) shows, when a functional complex occupies initial position. (40)

a. b.

[uWhere will you take Grey? ] – [uWhere are you going to fly?] – [u [i I will take Grey,] [i she said,] [u [i I will fly,] [i she said,]

[u [th ]o London. ] [u [th ]o Seattle. ] [i [th ]o London.] ] [i [th ]o Seattle.] ]

Taken at face value, this leaves us with the overall picture in (41), where all categories from f upwards mark their left edges by aspiration – except for f. (41)

category f w f i u

aspiration at left edge? yes yes no yes yes

This should give us reason to pause: The observations in (39) are not in doubt, but do they actually prove that, of all left edges in the hierarchy, [ f is not a position of fortition? Category-by-category stipulations of specific processes were a hallmark of the early days of prosodic hierarchy theory, when they were taken as a positive result since they seemed to provide evidence for the existence of the hierarchy in the first place. In this case, however, even if one accepted the claim at face value, one would, at a minimum, still wonder why it is precisely f that is the odd one out. At the other extreme, some might take the picture in (41) as a good reason to doubt the very existence of f. We remain persuaded by the overwhelming evidence for a level of phrasing corresponding to f, and instead argue that [f is in fact a position of fortition, just like the other categories in (41), and that something else is going on in (39). In a general vein, this is in line with recent research which has accumulated a significant amount of evidence (see Wagner 2005: 131–136 for an overview) showing that the differences between prosodic boundaries at the various levels of the hierarchy do not in general concern the type of effect but rather its strength, which is hierarchically cumulative, as a broad generalization. Category-specific stipulations, such as the putative exemption for f in (41), have no place in this 17

Or alternatively, “stressed s”.

160

Junko Ito and Armin Mester

much more restrictive conception of the effects that are associated with the hierarchy. For the case in point, fortition requirements such as aspiration are not distributed in an on-again, off-again fashion over the prosodic hierarchy, with each category being free to make its own stipulations independent of other categories. Rather, experimental work has demonstrated hierarchical cumulativity. Thus, Fougeron and Keating (1997) conducted a systematic study of articulatory strengthening in American English at different levels of the prosodic hierarchy. Using reiterant speech, with each syllable of a model sentence replaced by [no], linguopalatal contact was measured for consonants and vowels initially, medially, and finally in w, f, i, and u. The general finding was that “the higher the prosodic domain, the more linguopalatal contact the consonant has” (Fougeron and Keating 1997: 3728). A later study (Cho and Keating 2001) replicates and extends these results for Korean, and Flack (2006) collects a large amount of cross-linguistic evidence showing that fortition requirements are upwardly uniform, and proposes a constraint format that succinctly expresses this kind of patterning. In light of these findings, the explanation for the facts in (39) simply cannot be that f for some reason opts out of the otherwise uniform hierarchical patterning of aspiration. Regarding an alternative account for the facts seen in (39), a small additional observation provides an important clue. A handful of lexical words starting with to- show similar unexpected initial flaps, namely today, tonight, and tomorrow. These are not function words like to, but still are very much prone to flapping. Goldsmith (2001), who provides the examples in (42), points out that these cases of flapping cannot simply be ascribed to word frequency – other frequent words (e.g., tomato)18 show no such tendency. (42)

a. b.

What are you going to see [R]onight? Who will you see [R]omorrow?

Utterance-initially, all of these words receive aspiration ([t h ]omorrow is another day; [th ]oday we are going to hike up Mount Daimonji). If aspiration were a sine qua non for w-hood, the left edges of such to-words would be in the same boat as the left edges of to London, to the sky, and to Seattle in (39) and could not be w-edges. But the consequences of this line of thought are hard to live with: Separating to- from the rest of the word and analyzing it as f-attached in [f to[w morrow]] is synchronically unwarranted in any other respect (i.e., besides 18

Tomato is a member of the Oxford 3000 list of the 3000 most frequent words of the English language (see http://www.oup.com/elt/catalogue/teachersites/oald7/ oxford 3000/oxford 3000 list?cc=gb), but still never lenites to *[R]omato in any context.

The extended prosodic word

161

flappability) and unlikely to be the correct analysis, as it violates high-ranking (undominated) Lex-to-w(Left). With respect to other function words besides to, there are only a very limited number that have the requisite properties, namely, reduced forms of till [tl] and can/could [kn, k´d]. We are unaware of any phonetic " " see Martha [t]ill Monday vs. give studies that compare examples like I won’t Martha a [t]omato or Martha [c]an ask her mother vs. give Martha [c]anola oil). In looking for a solution, let us consider the underpinnings of the phonotactic generalizations. Structurally speaking, aspiration is not a hallmark of w-initial segments per se, but occurs when it clearly marks the beginning of a structural unit, here w. In terms of Ito and Mester (1999), it is a crisp-edge property. In both structures in (43), y is w-initial: w-initial segments are aspirated at crisp (w-)edges (43a), and flapped when there is no crisp edge (43b), i.e., when the segment is ambisyllabic (Kahn 1976). (43)

The crisp-edge structure also accounts for the doubled [t] pronunciation in I forgo[t] [t]o call.19 Clusters of adjacent identical consonants in English arise across word boundaries (as in compounds, cf. ho[t] [t]ub), but not w-internally, and are one of the indicators for w-junctures. What is special about the non-crisp cases under discussion, then, is not just the phonetic fact that the segments are flapped, but rather that (onset-)ambisyllabification has here applied to a limited set of to-words.20 Goldsmith (2001) observes that the to in fly to Seattle (39c) has a close phonological relationship with the word to its left. There is no left-alignment constraint for function words 19

Goldsmith (2001) notes the interesting contrast between I’ve go[R]a leave soon vs. I forgo[t] [t]o leave. 20 A reviewer points out that this option might arise out of the availability of reduced allomorphs for these items, which tend to be deaccented and attached or “encliticized” to the preceding intonational phrase when occurring sentence-finally. The combination of being frequent and occurring in this “weak” prosodically subordinated environment might make a development of a reduced allomorph likely. This line of thought is an interesting alternative to the standard approach based on the notion “function word”.

162

Junko Ito and Armin Mester

(no Fnc-to-w constraint), and contraction facts are also indicative of such a close relationship. Besides the celebrated wanna contraction, we find lexicalized encliticization in cases like gotta (got to), gonna (going to), supposeta (supposed to), hafta (have to), and oughta (ought to). The cases in (42) show lexical words (adverbs) showing a similar kind of behavior, which might be related to the fact that they are highly predictable in such conversational routines.21 While further investigation is needed to determine the exact environment of ambisyllabification, and a unified explanation of to-words and other details await further phonetic studies, we here conclude that the flapping evidence does not constitute a strong argument for f-attachment of fnc. In fact, some of the crisp-edge evidence might be construed in favor of w-adjunction. The argumentation for f-attachment in German in Hall (1999: 124–126) proceeds along lines similar to Selkirk’s (1996). To account for the limited distribution of initial @-syllables, Hall posits two constraints, one (Hall 1999: 114) barring schwas from occurring w-initially (*[w @ . . . ]), and another one (Hall 1999: 126) “that ensures that pwords do not begin with a consonant other than [b] or [g] plus schwa” (e.g., B[@]ginn ‘beginning’ and g[@]nau ‘exactly’). (44)

a. b. c.

[@]s kommt ts[@] Hause ’n[@] Machine

‘it comes’ ‘at home’ ‘a machine’

Citing forms such as those in (44),22 Hall points out that the w-adjoined [w fnc[w lex]] structure (as well as the fused [w fnclex] structure) would not fulfill the posited schwa requirements on prosodic words: (44a) violates the w-initial schwa constraint, and (44b,c) the constraint barring w-initial C´ except for b´ 21

See Jurafsky, Bell, Gregory, and Raymond (2000) on the role of frequency and predictability from neighboring words in reduction processes like flapping. 22 There appear to be dialectal differences, and some speakers do not produce onsetless schwa-syllables in phrase-initial position. For such speakers, *[@]s kommt ‘it comes’ is not a possible pronunciation even in a speech register where other fnc reductions take place: [E/*@]s aß ’n[@] Banane ‘it ate a banana’. The dialectal difference cannot be attributed to whether fnc is w-adjoined or f-attached (as argued by Selkirk 1996 for dialects of Serbo-Croatian, and by Peperkamp 1996 for dialects of Italian), since Hall’s (1999) schwa-requirements would favor the w-adjoined structure for [w @s [w aß]] and the f-attached structure for [f ’n@[w Banane]]. Such a situation is also quite problematic for the constraints determining the prosodic structuring of fnc-lex complexes: The ranking of Parse-into-w and No-Recursion would have to vary from function word to function word, depending on a particular segmental property (V- or C-initial), a situation not otherwise encountered.

The extended prosodic word

163

and g´. The fnc must, so the argument goes, be directly attached to f, bypassing w (where these schwa-related requirements hold). The strength of the argument here depends on the analysis on which it rests, in particular, on the constraints that drive it. The constraint specifically barring w-initial C´, with a built-in exemption of the elements b´ and g´, is awkward as a phonological markedness constraint since it is unlikely to find cross-linguistic support. In German, it encodes the morphological fact, arbitrary in terms of phonology, that b´- and g´- but not, for example, k´-, d´-, n´-, or l´-, happen to exist as prefixes. The restriction should therefore be regarded as a distributional generalization on lexical words, rather than a requirement on the shape of prosodic words. A possible formal analysis might invoke constraints on prominent positions, as developed in the work of Smith (2004), where the initial position of content words plays a central role, reflecting its psycholinguistic importance. Functional material, irrespective of its prosodic parsing, is not prominent in this way, and hence not subject to such constraints. The explanation therefore does not depend on whether the fnc is w-adjoined or fattached. In sum, both Selkirk (1996) and Hall (1999) provide negative evidence against w-adjunction rather than positive evidence for f-attachment. There is no intrinsic reason, connected with known and recognizable properties of f, why a proclitic fnc such as to must be f-attached. Rather, some property associated with lex in w-initial position is absent in proclitic fnc, and relegating it to f seems to remove the problem. In general, this is not the case however, once we look at the properties of f in the context of the whole hierarchy. The solution, we have argued, is to move away from constraints that directly encode some descriptive generalization about w (such as beginning with at most one unstressed syllable), which are useful as preliminary diagnostics for investigation, but not genuine elements of the theory. Explanations should rather be sought in the interaction of several general, well-motivated, and independent constraints, such as Lex-to-w, FtBin and Parse-into-f, two of which do not even mention w. For the cases under discussion, we have suggested that the non-crisp alignment of functional material makes a difference at w-edges, as well as constraints on prominent positions, which already incorporate a psycholinguistically grounded distinction between content words and function words. On this basis, we can understand in a more principled way why function words and content words can both occupy w-initial position and still differ in some respects, and the evidence is fully compatible with the w-adjoined structure.

164

Junko Ito and Armin Mester

4. Evidence for w-adjunction In this section, we will lay out our case for w-adjoined function words. We first show that the special prosody of f-final function words provides a crucial argument for adjunction, and then turn to other evidence (binarity, maximal w-projections, and longer function words). 4.1. The argument from f-final function words In the early days of Prosodic Hierarchy Theory, the special behavior of f-final fnc words provided one of the key arguments in favor of Strict Layering (Selkirk 1984: 366–382). The generalization to be captured is illustrated in (45): At the right end of a phrase, a monosyllabic function word must appear in its strong form, not in its destressed form.23 (45) a. b. c. d. e.

Phrase-final They boxed the crowd [In]. Did you let the cat [In]? I’d like [tu], but I’m not sure I [kæn]/[SUd]. What were you thinking [Ov] last night? What did you look [æt] yesterday?

Non-phrase-final They fought [n]/[@n] the ring. Would you sit "[n]/[@n] the car? " but I’m not I’d like [t@] see you, sure I [kn]/[S@d] attend the party. " I was thinking [@v] Paris. Take a look [@t] Tiffany’s.

No special provision for phrase-final function words is needed, so the argument goes, because their promotion to the full status of an independent prosodic word follows directly from Strict Layering (Selkirk 1984: 366; Selkirk and Shen 1990: 332–335; Selkirk and Tateishi 1988). (46) a. b. c. d.

(fWhat did you look at) (f yesterday)? *(f . . . [w look @t]) (f [w yesterday] ) *(f . . . [w look] [w @t) (f yesterday] ) *(f . . . [w look] @t) (f [w yesterday] ) (f . . . [w look] [w æt]) (f [w yesterday] )

The general proclisis pattern of English means that fnc cannot cliticize to the left, making (46a) impossible,24 and proper bracketing prohibits cliticizing across f, 23

This generalization holds also when the function word is a phrase by itself, where it also occupies the right end of a phrase. 24 As is well known, the overwhelming default in English is proclisis, not enclisis, in line with the general syntactic pattern of the language. Pronominal object enclitic

The extended prosodic word

165

ruling out (46b). The f-final fnc must therefore be prosodified on its own, leaving either (46c) or (46d) as options. Strict Layering demands parsing at all levels, hence immediate domination of the fnc-s by f (46c) is ruled out. This leaves only a full w-parse for fnc (46d), implying a strong unreduced form in this position. This explanation carries over straightforwardly to the w-adjunction view of fnc with undominated Parse-into-w, which amounts to the assertion of one aspect of Strict Layering (where all Parse-into-X constraints are undominated). Since adjunction to the left and to the right is independently ruled out, Parse-into-w here requires fnc to be a w by itself. Headedness then requires this w to be headed by f, and this means stress and irreducibility. It is a consequence of the w-adjunction theory as it stands that in f-final position only a structure with full f-status and w-status for a function word is possible. On the other hand, the f-attachment view of fnc, where Parse-into-w is lowranking, has no direct explanation for the non-reducibility of f-final function words.After all, its key idea is to allow fnc-s to become immediate daughters of f, in violation of Parse-into-w. Everything else being equal, the most natural way of parsing a phrase-final fnc should therefore be as in (46c), with the unstressed syllable directly attached to f. In order for the correct (46d) to be chosen, an extra stipulation is needed for right edges of f, insisting on w-parsing in this specific location. Just such an edge alignment constraint, Align-Right (f, w), is posited by Selkirk (1996). (47)

By requiring the element circled in (47) to be present, the statement reiterates, for a particular position, what is independently known – the order of categories of the prosodic hierarchy. This redundancy reveals a deeper problem: What we are looking for is an explanation of why right edges of phrases are special in terms of parsing, not an assertion, in alignment language or otherwise, that things are this way.25 High-ranking (inviolable) Parse-into-w provides such forms like gimme, got’m, or need’m go against this general proclisis imperative and have a special morphosyntactic status (Selkirk 1984: 383–406, 1996: 459–460). We return to this point in section 4.3. 25 While phrase-final lengthening is clearly a widely attested phenomenon, as a reviewer reminds us, it should not be confused with prominence or strengthening. On the other

166

Junko Ito and Armin Mester

an explanation, preserving a central insight of Strict Layering Theory. On the other hand, direct f-attachment of function words as a default, by passing the prosodic word level, means that this result is lost. 4.2. The argument from structural binarity Even though less prominent than in syntax, structural binarity considerations play an important role in phonology, schematized in Ito and Mester (2007b) as in (48).26 (48)

Bin(k), where k is an element of the prosodic hierarchy

Bin(f) is the familiar FtBin constraint, and Bin(s) the constraint against superheavy trimoraic syllables. Versions of Bin(w) and Bin(f) have been recruited to account for, e.g., truncation patterns and phonological phrasing (Ito and Mester 2003, Kubozono 1995, Selkirk 2000, etc.). Of relevance are binarity factors related to the parsing of function words discussed by Selkirk (2000) in accounting for the possible variations in prosodic phrasing, as in (49). (49)

She loaned her rollerblades to Robin. a. → ( f )( f) b. → ( f)

(50)

She loaned her rollerblades to Robin’s sister. a. → ( f )( f) b. ( f)

According to Selkirk (2000), phrasal variation in a neutral non-focused context is permitted in (49), because neither phrasal choice fulfills strict binarity ((49a) has a unary f, (49b) a ternary f). On the other hand, the candidate parse (50a) fulfills f-binarity exactly and therefore forestalls the co-selection of the nonbinary variant (50b). The structure in (51) illustrates how (50a) fulfills f-binarity (each f is binary).

26

hand, iambic lengthening is a phenomenon strictly caused by the rhythmic structure of the iambic foot (Hayes 1995), and not linked to the phrase level. It is therefore unclear whether the alignment constraint mentioned in the text can be independently motivated. Further distinctions between minimally binary and maximally binary are often also necessary (see the reference cited in the text).

The extended prosodic word

167

(51)

This binarity argument is not as cogent as one might wish since (50b) is also longer than (49b), but we will accept it here at face value. In order for the argument to be sustained, it is crucial that the function words are w-adjoined: The (bolded) prosodic words of (51) serve an important grouping function, which results in an overall binary structure. On the other hand, if proclitic fncs’s are directly dominated by f, this feature of phrasal rhythm is lost, resulting in a loss of binarity, as an inspection of (52) reveals. (52)

Could one say that it is all a matter of definition? For example, the constraint could be formulated as in (53), and binarity would be fulfilled in (52), despite appearances. (53)

Bin(f) (rejected formulation): f requires exactly two w.

This formulation tells us to only look at w-daughters of f and to ignore everything else, so that branches leading to other daughter nodes simply do not count. But just stating the constraint reveals its artificiality – the necessity to have recourse to a complication like (53) is a counterargument in itself. More importantly, the constraint has the flaw that it repeats a section of the prosodic hierarchy. For example, it is just an accident that it mentions f and w, and not f and s. The upshot is that the simple everyday conception of binarity based on branching works fine – provided the prosodic structure is understood as w-adjunction. A different way out, pursued in Selkirk (2000: 244), is to postulate an additional prosodic category, MiP (minor phrase) composed of fnc-lex combinations and intervening between w and f (= major phrase), and to formulate the binarity constraint as Bin(MaP) “[a] major phrase consists of just two minor/accentual phrases”. As illustrated in (54), this makes the overall prosodic form isomorphic to the w-adjunction structure (51).

168

Junko Ito and Armin Mester

(54)

Occam’s razor, however, militates against such additional categories as long as the existing ones are sufficient to represent the prosodic groupings manifest in the data, and no solid independent evidence for the new categories is available. Positing the minor phrase raises the additional question of how it would differ from the Clitic Group, which has been argued against as a separate category in earlier work (see the beginning of section 3 above). Whatever the final resolution of these questions is, it remains true that the binarity facts here follow immediately from the w-adjunction view of function words, whereas the f-attachment view needs to introduce an additional phrasal category to specifically create prosodic forms isomorphic to the structures that are the hallmark of w-adjunction. 4.3. The argument from maximal w-projections What other factors, then, might be brought to bear on the decision between the two possible structures for fnc-lex complexes in (55)? (55)

Since the two structures have in common that the lexical word is a w (Lexto-w is fulfilled), the difference might seem to solely lie in the placement and status of the function word. Indeed the arguments that we have seen above all focus on this point (e.g., Is fnc w-initial? Is it dominated by w?). However, another important difference between the two structures lies in the prosodic status not of the function word to but of the lexical word Rhodes: It is a full and independent prosodic word in (55b) [f to [w Rhodes]] but not in (55a), where

The extended prosodic word

169

it is only a segment of the larger prosodic word [w to [w Rhodes]]. In terms of the approach developed in earlier work (Ito and Mester 2007b, to appear), [w Rhodes] is a maximal projection of w in (55b), but not in (55a), where it is dominated by a higher w. This makes a difference to the extent that there are phonological processes that are specific to maximal prosodic words. Such processes indeed exist. An example is linking and intrusive r in non-rhotic varieties of English, whose natural interpretation, pace McCarthy (1993), as hiatus breaker / onset filler can only be maintained if maximal and non-maximal w are distinguished. After a brief review of the basic analysis of English r-sandhi in Ito and Mester (to appear), we pursue the consequences of such an approach in terms of the adjunction status of various types of function words. Comparison with German (Kabak and Schiering 2006) provides further arguments for wadjoined structures, and thus offers evidence for high-ranking Parse-into-w in these languages. The literature on English r-sandhi (r-loss and r-insertion at certain morphological and syntactic junctures) is extensive, both in descriptive and theoretical terms (see McCarthy 1991, 1993, as well as McMahon 2000, and works cited there). The hiatus-breaking -r- occurs after the non-high vowels [@, O:, A:], as in the idea-r-is this or law-r-and order, and is productively inserted, as Wells (1982: 226) points out, in acronyms (as far as NAFTA-r-is concerned), loanwords (schwa-r-insertion, Lufthansa-r-officials) and even in r-intruders’ pronunciation of foreign languages (ich bin ja-r-auch fertig, j’´etais d´ej`a-r-ici, Gloria-r-in excelsis Deo, viva-r-Espa˜na, etc.). The phenomenon is widespread, and a virtually identical process exists, for example, in Bavarian German (wie-r-ich gesagt habe [wia>ri gsAgd hab] ‘as I said’). An argument that the w-adjoined structure has advantages over f-attachment comes from an at first glance puzzling restriction on the process, which is otherwise fully automatic and productive: Intrusive -r- does not appear after function words. This was first noted in Kahn (1976), and McCarthy (1993), the source of the examples in (56), provides an analysis (see also McCarthy 1991). (56)

a. b. c. d.

Why do Albert [d@ ælb@t] and you quarter o(f) eight [@ eIt] the apples [ði æplz] to add [t@ æd] to" (h)is [t@ Iz] troubles,

*[d@ r ælb@t] *[@ r eIt] *[D@ r æpl z] *[t@ r æd]," *[t@ r Iz]

In (56d), we find unresolved vowel hiatus both in the familiar fnc-lex complex (to add) and in the fnc-fnc sequence (to (h)is). The summary table in (57) gives an overview of the presence and absence of the hiatus breaker with various combinations of lex and fnc.

170

Junko Ito and Armin Mester

(57) Second word: lex Second word: fnc First word: lex Pamela-r-Anderson law-r-and order ← with hiatus breaker First word: fnc do(-*r-)Albert to(-*r-)’is troubles ← no hiatus breaker Ito and Mester (to appear) argue that if proclitic function words form w-adjoined structures in English,27 the absence of the hiatus breaker correlates with a specific prosodic property: The following material does not constitute a maximal prosodic word. This is illustrated in (58a,b) vs. (58c,d), where the relevant maximal w’s have been circled. (58)

We define (see also Ito and Mester 2007b, to appear) maximal and minimal projections (or “extensions”) of prosodic categories as in (59).28 (59)

Maximal and minimal projections of category k kmax(imal) =def k not dominated by k kmin(imal) =def k not dominating k

In w-adjunction structures, the top w-projection is referred to as wmax (“maximal prosodic word”), its innermost w-subconstituent as wmin (“minimal prosodic word”). In the absence of adjunction, w is both maximal and minimal, since it is neither dominating, nor dominated by another w.29 In Ito and Mester (2007b), w-adjunction for fnc-lex complexes is also the assumption made in McCarthy (1993). In prosody, it is convenient to make use of the term “projection” in adjunction structures, deviating from current syntactic usage. As a reviewer points out, syntactic adjunction does not “project” a structure because it simply creates a segment of an already existing category. 29 A reviewer points out that the identification of the two in this situation might not be without problems if there are processes specific to wmax that must be prevented from 27 28

The extended prosodic word

171

we show that reference to maximal and minimal projections of w and f is crucial in understanding the phonological behavior of complex compounds in Japanese (wmin as domain of rendaku, wmax as domain of junctural accent, fmin as domain of deaccentuation, etc.).30 In the context of English function words and intrusive -r-, reference to wmax plays an important role in two situations, viz., for wmax versions of lexical word alignment (60a) and of the onset requirement (60b). (60)

a. b.

Lex-to-wmax Onset-wmax

Lex-to-wmax (60a), a more specific version of the general alignment constraint Lex-to-w discussed above in section 2.2, requires alignment of lex to a maximal w. As shown in (61), ranking the right-handed version of the constraint over the left-handed version favors proclisis, as in English, and the opposite ranking yields enclisis.31

30

31

applying in cases where wmax is also a wmin . We leave this as an interesting question for future research. Kleinhenz (1996) also makes use of a distinction between maximal and minimal prosodic words, and argues that the maximal prosodic word is the domain of syllabification and that the right edge of the minimal word is relevant for the intervocalic obstruent voicing found in Rhinelandian/Franconian dialects of German. Kleinhenz’s use of the max/min distinction is different from ours in that the word-internal prosodic categories (such as the syllable) are not integrated into the same hierarchy, i.e., the rightmost consonant of the minimal prosodic word must be syllabified as an onset, violating proper bracketing [w [w .dor.d]aus.] “von dort aus” ‘from there on’. As pointed out by one reviewer, this would necessitate a distinction between domains of application of phonological rules and prosodic constituents, similar to that of the distinction made in Inkelas (1989). PCat-GCat alignment, as in Selkirk (1996) – here, Align-w-to-Lex(R), mapping edges of prosodic words to edges of lexical words – is an alternative way of prohibiting enclitics, with slightly different effects. We find it significant that reference to wmax makes it possible to achieve this strictly within GCat-PCat alignment, the standard direction of syntax-phonology mapping, keeping open the possibility of a more restrictive theory to be carved out of the very broad generalized alignment theory of McCarthy and Prince (1993a).

172

Junko Ito and Armin Mester

(61)

A second important role of the maximal prosodic word lies in the fact that its onset is a position of particular prominence. This is the driving idea of the analysis of English -r- in Ito and Mester (to appear): The appearance of both linking and intrusive r in external sandhi is a manifestation of the requirement for maximal prosodic words to have an onset. The process results from the interaction of two constraints, each a positionally restricted version of a more general one: Onset-wmax (62a), a special version of Onset (62c) for a prominent position (here, the beginning of maximal w-projection), and Dep-init-w (62b), a positional faithfulness constraint that is a special version of Dep (62d) ruling out the insertion of a root node at the beginning of w (any w, not just a maximal one).32 (62)

Constraints driving r-intrusion a. Onset-wmax *[wmax V b. Dep-init-w An output root node in w-initial position has an input correspondent. c. Onset *[w V d. Dep An output root node has an input correspondent.

Ranked as Onset-wmax » Dep-init-w, the interaction of these two constraints derives the correct distribution of intrusive -r-, as shown in (63). We find -r- in lex-lex and lex-fnc configurations (63a,c) at the beginning of a maximal prosodic 32

The occurrence of intrusive -r- in compounds (spa-r-experience, schwa-r-epenthesis, etc.) bears out a prediction made by Lex-to-wmax (60a) (not shared by the alternative in note 31): Each member of a compound is a maximal word, i.e., the whole compound is prosodically some kind of phrase. Following previous researchers including Bakovic (1999), we assume that intrusive -r- involves the filling of an epenthesized root node by spreading from the preceding vowel when no glide is available, hence there is no phrase-initial intrusive -r-.

The extended prosodic word

173

word because of the action of the dominant constraint Onset-wmax .33 On the other hand, in fnc-lex and fnc-fnc configurations (63b,d) we are dealing with the onset of a non-maximal prosodic word, where this constraint is irrelevant. Subordinate Dep-init-w therefore has a chance to assert itself, preventing insertion, and the hiatus is not resolved by -r-.34

33

A special case of lex-fnc (see note 24 above) is found in pronominal object enclitic forms like gimme, got’m, or need’m, which, we assume, following previous researchers, have a special morphosyntactic status. The host verb and the enclitic object pronoun together form a single lexical word, which is in turn mapped by Lex-to-w alignment into a single prosodic word. Intrusive -r- is found in this context ([w drawrit], [w sawrus], [w subpoenar(h)im]) because the pronoun is not w-initial, and hence not subject to Dep-w-init. The general constraint Onset is responsible for r-insertion, just as it is in other w-internal contexts, such as [w subpoenaring], [w withdrawral], etc. In a detailed analysis of Dutch clitics, Booij (1996) provides a variety of arguments that the prosodic structure for host+enclitic sequences is different from proclitic+host sequences. Just as in English, the former is prosodically integrated into a single prosodic word, whereas the latter forms an adjunction structure. Hall (1999) makes a similar point for German. 34 Even though intrusive -r- is impossible in fnc-lex complexes, underlying r is obligatory (McCarthy 1993: our answer, were eating, for eating, either apples or oranges, for any reason, under any circumstances). The distinction is therefore still part of the synchronic grammar (see Ito and Mester to appear for discussion) and creates a serious obstacle for any attempt to reduce the whole r-sandhi phenomenon to allomorphy, i.e., by listing variants with and without final -r for all relevant lexical items (/spA:, spA:r/, etc). The allomorphy approach has a basic problem with the productivity and rule-governed character of the phenomenon (Why would all relevant items have a variant with -r? Why would intrusive -r- appear in new loanwords, and in the pronunciation of foreign languages? etc.) and cannot even deal with the basic fnc-lex distinction. One might think that gonnafnc , different from Wandalex and from under fnc , simply lacks a variant with -r. But this does not explain why gonna is suddenly able to sponsor intrusive -r- when it stands in phrase-final position (I said I was gonna,f -r-and I did f , etc., see McCarthy 1993 and section 4.1). One could resort to further listing and add on a “precompiled” phrase-final variant (/gOn@/, /gOn@r/ / f ]), but this would mean giving up on any attempt to explain the generality of the phenomenon.

174

Junko Ito and Armin Mester

(63)

The crucial prerequisite for this explanation of the lack of r-insertion after a function word, as in to Anne (63b), is the different prosodic status of Anne in this position – as a subpart of a maximal w, not a maximal w all by itself as in (63a), a difference directly due to the w-adjunction structure for function words. Direct f-attachment (64b), on the other hand, by having the function word skip w-parsing entirely, turns both Anne’s into maximal w’s and misses the crucial contrast. (64)

f-attached fnc

The extended prosodic word

175

The Onset-based account of r-intrusion in (62) and (63) rests on the optimality of w-adjoined structures for English function word complexes, which is in turn explained by the ranking Parse-into-w » No-Recursion (65), in terms of the prosodic constraints introduced in section 2. (65) a. b. 

[f to [w Anne] ] [f [w to [w Anne] ] ]

Parse-into-w to!

No-Recursion Anne

Selkirk (1996), on the other hand, follows McCarthy (1993) in ascribing r-intrusion to a Final-C constraint, which requires every w to end in a consonant. In such an approach, the dominating nodes do not make a difference: (66)

Whether fnc-lex structures are w-adjoined (66a) (McCarthy 1993) or f-attached (66b) (Selkirk 1996), fnc is not placed in a w-final position, hence Final-C does not require the presence of -r-. Although successful in capturing the absence of r in these positions, Final-C, in requiring w-final codas, directly conflicts with the universal markedness constraint dispreferring codas in general. As shown above in (62)–(63), once the onset requirements for different levels of the prosodic hierarchy are properly understood, in conjunction with appropriately assigned w-adjoined prosodic structures (65), there is no need to appeal to the intrinsically problematic Final-C constraint (for further discussion and motivation, see Ito and Mester to appear). Looking beyond Final-C’s specific problems, the next section shows that any analysis relying on constraints on w-endings instead of w-onsets, with an assumed across-the-board absence of w-boundaries after function words, becomes untenable when we consider the full spectrum of function word complexes, including prosodically more independent function words such as disyllabic gonna or trisyllabic supposeta.

4.4. The argument from longer function words Previous research on the status of function words has mostly been concerned with the monosyllabic case, prototypically represented by lone light syllables

176

Junko Ito and Armin Mester

such as to or the which are intrinsically prosodically undersized, unfootable by themselves, often reduced, and unable to constitute an utterance. There is usually little detailed discussion of function words that are large enough to be prosodically full-sized (e.g., disyllabic feet such as after or during), and how they are to be incorporated into the overall prosodic structure. Although often relegated to footnotes or silently omitted from discussion altogether, such prosodically full-sized function words are neither rare nor atypical, as can be gleaned from the illustrative lists of monosyllabic, disyllabic, and polysyllabic prepositions in English (67) and German (68). Note that even the monosyllables include heavy (footed) items such as down and through. (67)

English prepositions a. monosyllabic: at, by, down, for, from, in, like, of, on, out, round, since, through, till, to, up, with b. disyllabic: about, above, across, after, against, along, among, around, before, behind, below, beneath, between, beyond, during, except, over, towards, under, until, within, without c. polysyllabic: underneath

(68)

German prepositions a. monosyllabic: ab, an, auf, aus, bei, bis, durch, f¨ur, in, mit, nach, seit, statt, trotz, um, von, vor, zu b. disyllabic: anstatt, aufgrund, ausser, entlang, gegen, hinter, infolge, neben, ohne, u¨ ber, unter, w¨ahrend, wegen, zwischen c. polysyllabic: angesichts, ausserhalb, gegen¨uber, innerhalb

Verbal functional elements (auxiliaries and modals) are mostly monosyllabic in English (69), but forms with the inflectional ending -ing and the contracted negative ending -n t show that disyllabic forms are by no means an anomaly. (69)

Verbal fnc in English a. monosyllabic: be, am, is, are, was, were, been, do, does, did, have, has, had, can, could, may, might, must, ought, shall, should, will, would, can’t, won’t, don’t, aren’t, weren’t b. disyllabic: being, having, isn’t, wasn’t, hasn’t, haven’t, hadn’t, doesn’t, didn’t, couldn’t, mustn’t, oughtn’t, shouldn’t, wouldn’t

The extended prosodic word

177

In German, inflectional endings make many forms disyllabic.35 (70)

Verbal fnc in German (from the paradigms of sein ‘be’, haben ‘have’, werden ‘become’) a. monosyllabic: sein, bin, bist, ist, sind, seid, sei, seist, war, warst, wart, hast, hat, habt, wirst, wird b. disyllabic: waren, w¨are, w¨arest, w¨aren, w¨aret, seien, seiet, haben, habe, hatte, hattest, hatten, hattet, h¨atte, h¨attest, h¨atten, h¨attet, werden, werde, werdet, wurde, wurdest, wurden, wurdet, w¨urde, w¨urdest, w¨urden, w¨urdet

The short overview of different-sized function words in (67)–(70) shows that prosodically full-sized forms are not uncommon as functional elements. The difference between foot-sized function words and (monomoraic) syllable-sized function words is that the former but not the latter can be independent prosodic words, as far as the prosodic constraints (Headedness, FtBin, Parse-into-f) are concerned. Therefore the direct prosodic argument that the fnc cannot be an independent w does not hold for foot-sized cases, and the structure in (71a) is a real contender to (b) and (c) for collocations like under water, haven’t eaten, unter Wasser, haben gegessen, etc. (71)

Previous research has, however, provided evidence against independent w-hood of foot-sized fnc for both English (McCarthy 1993) and German (Kabak and Schiering 2006). In what follows, we will review these arguments against independent w-hood, and show that a broader survey of the same type of facts also turns out to provide key evidence against f-attachment, leaving w-adjunction as the only viable option. 35

Different from their English counterparts, German modals (d¨urfen, k¨onnen, m¨ogen, m¨ussen, sollen, wollen, etc.) act as independent predicates.

178

Junko Ito and Armin Mester

4.4.1. English McCarthy 1993 observes that portmanteau function words such as shoulda (< should have), gonna (< going to), didja (< did you), lotta (< lot of ) do not allow r-insertion (72a), and in this way behave like their monosyllabic counterparts (72b), not like certifiable prosodic words (72c). (72)

a.

*gonna-r-eat *lotta-r-apples

b.

*to-r-eat *the-r-apples

c.

Pamela-r-eats California-r-apples

Since these portmanteau function words have the shape of a trochaic foot (stressed s followed by unstressed s), it is natural to assume that they are footed and fulfill Parse-into-f. Their footing immediately explains why disyllabic function words do not have reduced variants (cf. monosyllabic have [hæv]∼[h@v] vs. disyllabic haven’t [hævnt], *[h@vnt] (Selkirk 1984: 355)). " "w’s as in (73a), r-insertion If such foot-sized fnc’s (72a) are independent is incorrectly predicted in either the onset-based approach (section 4.3 above and Ito and Mester to appear) or the coda-based analysis (McCarthy 1993, and Selkirk 1996). In the former, the post-fnc word eat is a maximal w, hence Onswmax enforces r-insertion. In the latter, the function word gonna is a prosodic word on its own and receives a final consonant by Final-C. (73)

We are then left with the familiar choice between w-adjoined (73b) and fattached (73c). The onset-based approach requires the w-adjoined structure (eat is a maximal w in (73c) but not in (73b)), whereas the coda-based approach makes no distinction between the two (gonna is not a prosodic word in either (73b) or (73c)). Previous discussion of r-insertion has not made explicit the exact prosodic status of the function words in such examples. McCarthy (1993) notes that

The extended prosodic word

179

“[w]hether the portmanteaus shoulda, gonna, or dija [. . . ] are also proclitic has not been discussed in the literature, but by parity of reasoning they should procliticize too, giving structures like (shoulda eaten)PrWd , (gonna eat)PrWd , and (dija eat)PrWd ,” and Selkirk (1996) refers to the same types of examples but does not provide specific discussion or structures indicating foot status.36 A closer inspection of the varieties of English portmanteau function words, however, turns up decisive evidence in favor of w-adjunction and against fattachment. The relevant cases are trisyllabic forms, such as those in (74). (74)

Trisyllabic portmanteau function words supposeta eat (< supposed to) shouldn’ta eaten (< shouldn’t have, should not have) couldn’ta eaten (< couldn’t have, could not have) oughtn’ta eaten (< oughtn’t have, ought not have)

These function words, consisting of a disyllabic foot and an unfooted light syllable, must be grouped into some kind of prosodic unit – in fact, the prosodic word, given the ranking FtBin » Parse-into-f.37 (75)

Plugging these w-sized portmanteaus into the familiar three structures, we end up with the configurations in (76).

36

37

One unresolved issue for either analysis is the mixed behavior of the postverbal reduced pronoun ya (cf. Selkirk 1996: 159–160, notes 14 and 15). In f-final position, it behaves as a regular object pronominal (see note 33 above). Forming a single prosodic word together with its host verb, it can therefore appear in the reduced form in this position (seeya, gottya). On the other hand, in non-f-final position, ya seems to act as a proclitic to what follows, like other fnc’s, and does not allow r-insertion (give ya up, give ya all, give ya everything, give ya it, saw ya on TV, get ya Amanda.) Recursive foot structure is in principle another option, but would require, besides a new category “superfoot” and concomitant FtBin violations, the admission of both amphibrachs and dactyls.

180

Junko Ito and Armin Mester

(76)

Two conclusions emerge. First, the failure of r-insertion after trisyllabic portmanteaus is highly problematic for the coda-based approach (relying on FinalC) since it does not predict the blocking of r-insertion in this context, whichever of the three structures is adopted. The deeper reason for the problem lies in a view of prosodic structure that does not sufficiently distinguish structural subordination and structural reduction. The two often correlate, as in monosyllabic and disyllabic function words, which are non-w’s subordinate to another w (77a,b). But they diverge in longer function words, which are prosodically subordinate (non-heads within a larger maximal w), but still w’s in themselves, as expressed in the right-headed adjunction structure ((77c), where the head is underlined). The correct generalization about r-insertion lies in prosodic subordination, not reduction: It is blocked not only after non-w-sized function words, but also after w-sized function words that are prosodically subordinate. This is correctly captured in the onset-based approach, whose pivotal notion is the onset of the maximal prosodic word. Secondly, we now have evidence against f-attachment of proclitic function words in English: For trisyllabic function words, the f-attached structure (76c) turns out to have merged with the independent-w structure (76a), and consequently f-attachment wrongly predicts r-insertion both for the coda-approach and for the onset-approach. For the latter, advocated here, the w-adjoined structure (76b) makes the correct prediction: Since eat is not a maximal w, there is no r-insertion. Whether the adjoined prosodic unit is a syllable (77a), a foot (77b), or a prosodic word (77c), the prosodic status of the lexical word eat remains non-maximal. (77)

This is strong empirical support for w-adjunction over f-attachment, as well as for the onset-based approach. Significantly, similar evidence exists in German, and a parallel argument can be made, as we will see next.

The extended prosodic word

181

4.4.2. German Contractions of function word complexes, such as auf dem Kanal > auf’m Kanal ‘on the canal’ or wenn es geht > wenn’s geht ‘if it’s OK’, are a productive feature of many varieties of contemporary German (see e.g. Duden 2005). They are optional and show significant variation depending on register, degree of formality, and speech rate. They differ in this respect from the small number of obligatory preposition + article portmanteaus such as am, ans, im, ins, vom, zum, zur (from the prepositions an, in, von, zu ‘on, in, from, to’ and the definite articles dem, das, der), which are obligatory in the sense that they block the non-contracted versions at any level of speech (i.e., unless a demonstrative meaning is intended). For example, zum Mond ‘to the moon’blocks *zu dem Mond, but colloquial an’n Mond (id.) exists alongside an den Mond. The productive optional contractions are not limited to the prototypical preposition + definite article constellation (78a), but occur with all kinds of fnc-fnc sequences, including conjunction + definite article (78b), auxiliary + indefinite article (78c), wh-word + pronoun (78d), or conjunction + pronoun (78e). (78) German fnc-fnc contractions a. mit’m [.mI.pm.] Rad mit dem Rad b. weil’s Wetter " weil das Wetter c. ich will’n Buch lesen ich will ein Buch lesen d. wie’s geht wie es geht e. wennze [.ven.z´.] will wenn sie will

‘with the bike’ ‘because the weather’ ‘I want to read a book’ ‘how it goes’ ‘if she wants’

In a detailed study of the prosodic form of such fnc-fnc complexes in several German dialects, Kabak and Schiering (2006) (henceforth K&S) make several interesting points, summarized in (79) and briefly reviewed below.38 (79)

i. ii. iii. iv.

38

The two function words contract to form a trochaic foot [ f fnc fnc]; this foot provides the context for specific phonological processes and allomorphs in several dialects; the foot-sized fnc complexes cannot be independent prosodic words; the fnc-fnc foot is directly attached to f.

Our discussion here is limited to questions of synchronic prosodic form; see K&S (2006) on its historical genesis as well as additional morphological developments (such as reanalysis and analogical extension) that have arguably led to the reinterpretation of the contracted forms as inflected function words.

182

Junko Ito and Armin Mester

While we are in agreement with (i)–(iii), we will argue that a closer look at the full range of contracted forms shows that w-adjunction is preferable to fattachment (iv).39 The fnc-fnc complexes in question have the surface prosodic shape of a trochaic foot, either as a stressed heavy syllable (80a) or as sequence of stressedunstressed syllable (80b) where “the initial syllable [. . . ] has more prominent stress and more substance than the subsequent Fnc in the same complex” (K&S2006: 79).40 (80)

fnc-fnc feet

K&S discuss several phonological processes (flapping, consonant deletion, consonant intrusion) that occur only in the context of such fnc-fnc feet in various dialects. In Ruhrdeutsch (K&S 2006: 75), for example, disyllabic feet can become monosyllabic through medial consonant deletion in [f fnc-fnc] (81a), but not in [f lex] (81b). (81)

a.

b.

39

40

Consonant deletion in [f fnc-fnc]:41 [aufm] > [aum] auf’m " ] > [nam] nach’m [naxm mit’m [mIpm] > [mIm] " No consonant deletion in [f lex]: [kaUfm] > *[kaum] kaufen flachem [flaxm" ] > *[flam] > *[vIm] wippen [vIpm" ] "

‘on (m./n.)’ ‘to (m./n.)’ ‘with (m./n.)’ ‘buy’ ‘flat (m.dat.)’ ‘swing’

K&S (2006) also briefly consider w-adjunction as an option, but adopt Hall’s (1999) position (for the monosyllabic function words) that they are directly attached to f. Glosses and corresponding full forms: wenn es [vEn Es] ‘when it’, wenn du [vEn u:] ‘when you’, wenn er [vEn E5] ‘when he’, f¨ur den [fy:5] ‘for the (m.acc.)’, weil ein [vaIl aIn] because a (m./n.)’, weil eine [vaIl aIn@] ‘because a (f.)’, vor dem [fo:5 " the’, wenn sie [vEn zi:] ‘when she’, weil einen [vaIl aIn@n] ‘because de:m] ‘before a (m.acc.)’.

The extended prosodic word

183

K&S’s (2006: 75) basic point is that in order to single out [ f fnc-fnc] as a specific reduction site, it cannot constitute an independent prosodic word (82a) since it would then be prosodically indistinguishable from [f lex], where reduction is banned. Its special status for reduction (81), as well as for several other fnc-fnc-specific processes, such as flapping and consonant insertion, is clearly represented by a prosodic subordination of the foot, whether w-adjoined (82b) or f-attached (82c).42 (82)

K&S opt for f-attachment of [f fnc-fnc], following Hall’s (1999) view regarding monosyllabic function words. As we have seen above however, while the evidence for f-attachment of [s fnc] is not overwhelming for either English or German (section 3.2), for English there are indeed facts (trisyllabic cases like supposeta, section 4.4.1) that seem to give the edge to w-adjunction. It turns out that an exactly parallel argument for w-adjunction can be made on the basis of German trisyllabic fnc-fnc complexes. The relevant cases are consonant intrusions (K&S 2006: 69–73). Intrusive -R- in Middle Frankish, the dialect spoken in and around Nuremberg (Northern Bavaria), is restricted to the environment [fnc1 fnc2 ] between a vowel-final fnc1 and a vowel-initial fnc2 (83a), and is not found in other hiatus situations 41

These cases might be considered as listed allomorphs, but K&S (2006: 68) are careful to point to the existence of a complete sequence of forms at each step of reduction: [mIt de:m] → [mId@m] → [mIpm] → [mIm] mit dem ‘with the (m./n.)’, “showing " the gradual processes of phonological reduction and assimilations”. This makes a strict allomorphy approach less attractive: Not only would all possibilities have to be listed, there would also be no explanation why every level of reduction happens to be lexically precompiled and encoded (what is included in an allomorph list, and what not, is inherently arbitrary). 42 An alternative approach, which we will leave unexplored here, ascribes the resistance of content words to reduction processes that affect function word complexes not to a specific kind of prosodic representation singling out the latter, but rather to higherranking faithfulness constraints associated with the former.

184

Junko Ito and Armin Mester

(83b). Crucially, there is no difference between disyllabic and trisyllabic fnc-fnc complexes, undermining an exclusively foot-based analysis. (83) a.

Middle Frankish (K&S 2006: 69-70) ss :fnc-R-fnc sss :fnc-R-fnc

b.

*internal-R*lex-R-lex *fnc-R-lex *lex-R-fnc

[voU-R-I] [voU-R-5] [gEç@-R-@n] [tsU-R-@n5] [nev@-R-@s] *[ka-R-Os] *[audO-R-InSEnjø@] *[dI-R-OöOSn] *[SaU-R-I]

wo ich bin wo er arbeitet gegen den Klaus zu einer Schule neben das Haus Chaos Autoingenieur die Orangen schaue ich

‘where I am’ ‘where he works’ ‘against Klaus’ ‘to a school’ ‘beside the house’ ‘chaos’ ‘car engineer’ ‘the oranges’ ‘I watch’

K&S argue that we are not dealing here with an odd, idiosyncratic property of specific items in one specific dialect. Crediting Ortmann (1998) for facts and generalizations, they observe that the same kind of distribution holds for intrusive n (“Binde-n”) in Swiss German dialects (“Higher Alemannic”), as seen in (84). Especially noteworthy is the sharp contrast between the last examples in (84a,b), with intrusion before unstressed pronouns (zu-n-ere), but not before emphatically stressed pronouns (zu-*n-IRE). Again, and significantly for our argument here, trisyllabic fnc-fnc complexes pattern with disyllabic ones. (84) Higher Alemannic (K&S 2006: 72–73) a. ss :fnc-n-fnc Gr¨osser wie-n-i gr¨oßer als ich wo-n-er ko isch als er gekommen ist sss :fnc-n-fnc bi-n-ene bei ihnen gang zu-n-ere geh zu ihr b.

*lex-n-fnc *fnc-n-lex *fnc-nemph.pron.

China-*n- un Japan wo-*n-irgendeine ko isch gang zu-*n-IRE

China und Japan als irgendeiner gekommen ist geh zu IHR

‘taller than I’ ‘when he comes’ ‘with them’ ‘go to her’ ‘China & Japan’ ‘when someone comes’ ‘go to HER’

The argument for w-adjunction given for English in section 4.4.1 applies here with the same force: Given trochaic foot structure, and FtBin » Parse-intof, the trisyllabic fnc-fnc must be parsed out as an w. Only w-adjunction, not f-attachment, can distinguish the fnc-fnc complex in the appropriate way.

The extended prosodic word

185

(85)

As in the English case, the crucial property of the prosodic configuration is structural subordination to a following head-w (underlined), not (or not necessarily) an intrinsically reduced status as a non-w.

4.5. Further consequences We have seen so far that fnc-fnc contractions in English (lotta, etc.) and German (mit’m, etc.) are best understood as w-adjoined. What, then, is the prosodic structure of non-contracted sequences of function words, i.e., those that are not phonologically merged into foot-sized (or in some cases, w-sized) units? Here, there are two possibilities: one in which the w-adjoins each fnc separately (86a), and another in which the two fnc’s are parsed into a foot before w-adjoining them together as a foot (86b). Although further detailed investigation is necessary, our observation is that English in general prefers the former, and German the latter. (86)

The difference in preferred structure captures the following facts: Portmanteaus √ are more common in German (f¨ur das → f¨urs) than in English (for the →  *forth ), because fusing the two fnc’s segmentally does not disrupt the already existing foot structure. In German, even in non-emphatic/non-focused contexts,

186

Junko Ito and Armin Mester

the first fnc, as the initial member of a foot, has prominence (f¨ur das ["fy5.dAs] . . . ), whereas strings of stressless fnc-syllables are found in English (f˘or th˘e [fr.D@] . . . ).43 " Besides the doubly w-adjoined structure in (86a), we briefly consider alternative structures for the unfooted string of syllables, as systematically diagrammed in (87) and (88). In the first three structures (87a–c), two adjacent syllables are attached as sisters to the same higher prosodic node. (87)

Alternate structures I

The next group of structures (88a–c) shows stepwise adjunctions: Two at the w-level in (88a) (=(86a), the structure argued for here), one each at the w-level and the f-level in (88b), and two at the f-level in (88c).44 (88)

Alternative structures II

Among these, (87a–c) are ruled out as gratuitious violations of Parse-intof,45 and although (88a–c) are all viable options for representing a sequence of

43

44

45

Both languages have trochaic feet. Note that there is no systematic vowel reduction in German, so unstressed syllables retain their vowel quality. (88c) violates No-Recursion at the f-level. The mixed adjunction structure (88b) is an interesting alternative to consider, since it might even suggest that each level is limited to a single adjunction. But no such restriction seems to be empirically valid (cf. a˘ nd f˘or th˘e house, etc.). But see Peperkamp (1996) who makes a distinction between lexical and postlexical foot parsing.

The extended prosodic word

187

two unstressed syllables,46 the intrusive-r- evidence turns out to favor (88a), as shown in (89) with examples of hiatus sites in different positions. (89)

Only the doubly w-adjoined structure (89a) declares the entire fnc-fnc-lex complex to be a maximal w (circled), and correctly predicts r-insertion in law[r-˘of th˘e land], but not in [t˘o (h)˘ıs troubles] nor in [f˘or th˘e apples], where the hiatus sites lie before nonmaximal w’s. On the other hand, the mixed adjunction structure (89b) declares the inner fnc-lex complex to be a maximal w, wrongly predicting r-insertion in t˘o-[*r-˘ıs troubles]. The f-adjunction structure (89c) does the same for the innermost lex, yielding the incorrect f˘or th˘e-[*r-apples]. Thus, both rhythmic form (unstressed sequence of syllables) and segmental form (r-intrusion) point to the doubly w-adjoined (90a), which is left to compete with (90b) where the two fnc’s are first parsed into a single foot and w-adjoined together. (90)

Both structures fulfill Parse-into-w, but (90a) (with fnc2 parsed twice, and lex parsed three times) violates No-Recursion to a greater degree than (90b) (lex is parsed only twice). So, all else being equal, (90b) should always be preferred. What forces the doubly adjoined structure, we suggest, is a Headto-Lex constraint (91) requiring heads of prosodic categories to be contained in lexical (not functional) material. 46

As discussed in section 3.1 (see (35)) proper bracketing blocks cross-boundary footings in (88a-c).

188 (91)

Junko Ito and Armin Mester

Head-to-Lex: Prosodic heads must be contained in lex.

Head-to-Lex is fulfilled in (90a), where both fnc’s avoid headhood by adjunction, and remain weak unstressed syllables.47 On the other hand, it is violated in (90b), where the head of the adjoined fnc-fnc foot lies outside of lex (the violation would be especially notable if the fnc-fnc foot constituted a subordinate prosodic word, as in the trisyllabic cases (77c)). The choice of optimal structure will then depend on the ranking: Head-to-Lex » No-Recursion will yield (90a), while the reverse ranking will make (90b) optimal. The Head-to-Lex constraint straightforwardly captures the preferred non-prominence of function words, and might be considered part of the rationale behind their general invisibility. 5. Conclusion We have presented some thoughts towards a new conception of the constraints concerning layering in prosodic structure. Much is to be gained, in our view, by adopting a slimmed down version of the prosodic hierarchy beyond the word level, with fewer categories but with a richer projection structure where adjunction plays an important role. We have put these ideas to work in considering some questions regarding the phonology of clitics. At least for the languages closely considered in this work, English and German, there is a large amount of evidence that the prosody characterizing a typical function word is not as an impoverished item that has not attained prosodic word-hood and is directly attached to the phonological phrase, but rather as a dependent element within an extended prosodic word structure, where it has adjoined to a core prosodic word, its head. Acknowledgements. Portions of this work have been presented at the 4th Old World Conference in Phonology (Rhodes, Greece, Jan. 2007), the 2nd Atami Phonology Festa (Atami, Japan, Feb. 2007), the 3rd workshop on Prosody, Syntax, and Information Structure (Indiana University, Sept. 2007), in colloquia at UC Santa Cruz (Oct. 2007) and at UCLA (Jan. 2008), and in class lectures at Kyoto University (Spring 2007) and at UC Santa Cruz (Fall 2008). For discussion and helpful comments, we are grateful to the audiences at these occasions, in particular to Outi Bat-El, Mary Beckman, Ricardo 47

A straightforward alignment constraint aligning left edges of fnc to w also produces the doubly w-adjoined structures, but goes against the generally accepted view that syntax-phonology mapping constraints do not refer to functional categories (Selkirk’s 1984: 343 “Principle of the Categorial Invisibility of Function Words”). Alternatively, one might appeal to some requirement that prosodic structure should closely mirror the syntactic adjunction structure (cf. van Oostendorp’s 2002 Mirroring principle).

The extended prosodic word

189

Berm´udez-Otero, Stuart Davis, Caroline F´ery, Steven Franks, Osamu Fujimura, Shosuke Haraguchi, Bruce Hayes, Shin Ishihara, Pat Keating, Ed Keenan, Yoshihisa Kitagawa, Haruo Kubozono, Ingvar Lofstedt, Ania Lubowicz, Michael Marlo, Jim McCloskey, Bruce Mor´en, Jaye Padgett, Anthi Revithiadou, Curt Rice, Shin-ichi Tanaka, Matt Tucker, Christian Uffmann, Marc van Oostendorp, Rachel Walker, Moira Yip, and Kie Zuraw. For constructive criticism and detailed comments on an earlier version of this paper, we are indebted to two anonymous reviewers, as well as to Ryan Bennett, Andrew Dowd, Barı¸s Kabak, and Jeremy O’Brien.

References Alber, Birgit 1998

Bakovic, Eric 1999 Booij, Geert 1996 Borowsky, Toni 1993

Chen, Matthew 1985

Stress preservation in German loan-words. In: Wolfgang Kehrein, and Richard Wiese (eds.), Phonology and Morphology of the Germanic Languages, 113–141. T¨ubingen, Germany: Niemeyer. Deletion, insertion, and symmetrical identity. Ms. Cambridge, MA: Harvard University. [ROA-300-0299]. Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13: 219–242. On the word level. In: Sharon Hargus, and Ellen Kaisse (eds.), Phonetics and Phonology vol. 4: Studies in Lexical Phonology, 199–234. New York: Academic Press.

The syntax of phonology: Xiamen tone sandhi. Ms. University of California, San Diego. 1987 The syntax of Xiamen tone sandhi. Phonology 4: 109–150. Cho, Taehong and Patricia A. Keating 2001 Articulatory and acoustic studies on domain-initial strengthening in Korean. Journal of Phonetics 29: 155–190. Davis, Stuart 2005 Capitalistic v. militaristic: The paradigm uniformity effect reconsidered. In: Laura J. Downing, T. Alan Hall, and Renate Raffelsiefen (eds.), Paradigms in Phonological Theory, 107–121. (Oxford Studies in Theoretical Linguistics. 8.) Oxford, U.K.: Oxford University Press. Duden 2005 Grammatik der deutschen Gegenwartssprache. Series Der Grosse Duden. Vol. 4. Mannheim: Bibliographisches Institut. Flack, Kathryn 2006 Restrictions on the edges of prosodic domains. Ms. Amherst: University of Massachusetts, Amherst.

190

Junko Ito and Armin Mester

Fougeron, C´ecile and Patricia Keating 1997 Articulatory strengthening at edges of prosodic domains. JASA 101: 3728–3740. Geilfuß-Wolfgang, Jochen ¨ 1998 Uber die optimale Position von ge. Linguistische Berichte 176: 581– 588. Giegerich, Heinz 1985 Metrical Phonology and Phonological Structure. Cambridge, U.K.: Cambridge University Press. Goldsmith, John 2001 What is phonology? http://humanities.uchicago.edu/faculty/goldsmith/phono1/Flaps.html. Accessed May 11, 2007. Gouskova, Maria 2003 Deriving economy: Syncope in optimality theory. Ph. D. diss., Dept. of Linguistics, University of Massachusetts, Amherst. Gutch, Donald 1992 Linking and Intrusive -r in English and Bavarian. In: Claudia Blank (ed.), Language and Civilization: A Concerted Profusion of Essays and Studies in Honour of Otto Hietsch, 555–611. Frankfurt am Main: Peter Lang. Hall, Tracy Alan 1999 Phonotactics and the prosodic structure of German function words. In: Tracy Alan Hall, and Ursula Kleinhenz (eds.), Studies on the PhonologicalWord, 99–131.Amsterdam, The Netherlands, and Philadelphia, USA: John Benjamins. Hall, Tracy Alan, and Ursula Kleinhenz (eds.) 1999 Studies on the Phonological Word. Amsterdam, The Netherlands, and Philadelphia, USA: John Benjamins. Halliday, Michael A.K 1960 Categories of the theory of grammar. Word 17: 241–292. Harden, Theo 2003 Die Tanzerei und das Gesinge. Einige Verdachtsmomente zur Ableitung pejorativer Nomina im Deutschen. Linguistic online 13. http://www.linguistik-online.com/13 01/harden a.html. Hayes, Bruce 1984 The prosodic hierarchy in meter. Ms. UCLA. 1989 The prosodic hierarchy in meter. In: Paul Kiparsky, and Gilbert Youmans (eds.), Rhythm and Meter, 201–260. Orlando: Academic Press. 1995 Metrical Stress Theory: Principles and Case Studies. Chicago: The University of Chicago Press. Hyde, Brett 2002 A restrictive theory of metrical stress. Phonology 19: 313–359.

The extended prosodic word Inkelas, Sharon 1989

191

Prosodic Constituency in the Lexicon. Ph. D. diss., Stanford University. Published 1990, Outstanding Dissertations in Linguistics Series. New York: Garland Press. Ito, Junko and Armin Mester 1999 Realignment. In: Ren´e Kager, Harry van der Hulst, and Wim Zonneveld (eds.), The Prosody-Morphology Interface, 188–217. Cambridge: Cambridge University Press. 2003 Weak layering and word binarity. In: Takeru Honma, Masao Okazaki, Toshiyuki Tabata, and Shin-ichi Tanaka (eds.), A New Century of Phonology and Phonological Theory. A Festschrift for Professor Shosuke Haraguchi on the Occasion of His Sixtieth Birthday, 26–65. Tokyo: Kaitakusha. [Reprint of: 1992. Weak layering and word binarity. Santa Cruz, CA: UC Santa Cruz. LRC Working Paper 92–9.] 2007a Adjunction in prosodic phonology. Talk presented at OCP4, Rhodes, Greece, January 2007. 2007b Prosodic adjunction in Japanese compounds. In: Formal Approaches to Japanese Linguistics 4, 97–111. MIT Working Papers in Linguistics 55. Cambridge, Massachusetts: MIT Department of Linguistics and Philosophy. to appear The onset of the prosodic word. In: Steve Parker (ed.), Phonological Argumentation: Essays on Evidence and Motivation. London: Equinox. Jurafsky, Daniel, Alan Bell, Michelle Gregory and William D. Raymond 2000 Probabilistic relations between words: Evidence from reduction in lexical production. In: Joan Bybee, and Paul Hopper (eds.), Frequency and the emergence of linguistic structure, 229–254. Amsterdam: John Benjamins. Kabak, Barı¸s and Ren´e Schiering 2006 The phonology and morphology of function word contractions in German. Journal of Comparative Germanic Linguistics 9: 53–99. Kahn, Daniel 1976 Syllable-Based Generalizations in English Phonology. Ph. D. diss., MIT. Published by Garland Press, New York, 1980. Kiparsky, Paul ¨ 1966 Uber den deutschen Akzent. Studia Grammatica 7: 69–97. Berlin: Akademie-Verlag. Kleinhenz, Ursula 1996 The prosody of German clitics. Papers on Clitics: ZASPIL Vol. 6: 81–95. Potsdam, Germany. Kubozono, Haruo 1995 Gokeisei to On-in-kouzou [Word Formation and Phonological Structure]. Tokyo: Kurosio.

192

Junko Ito and Armin Mester

Ladd, D. Robert 1986 Intonational phrasing: The case for recursive prosodic structure. Phonology 3: 311–340. 1996 Intonational Phonology. Cambridge, UK: Cambridge University Press. McCarthy, John J 1991 Synchronic rule inversion. In: Laurel Sutton, Christopher Johnson, and Ruth Shields (eds.), Proceedings of the Seventeenth Annual Meeting of the Berkeley Linguistics Society, 192–207. 1993 A case of surface constraint violation. Canadian Journal of Linguistics 38: 127–153. McCarthy, John J., and Alan S. Prince 1993a Generalized alignment. In: Geert Booij, and Jaap van Marle (eds.), Yearbook of Morphology, 79–153. 1993b Prosodic morphology I: Constraint interaction and satisfaction. Ms.: University of Massachusetts, Amherst, and Rutgers University. McMahon, April M. S. 2000 Change, chance, and optimality. Oxford linguistics. Oxford; New York: Oxford University Press. Meinschaefer, Judith 2005 The prosodic domain of Italian troncamento is not the Clitic Group. Ms. Konstanz, Germany: Universit¨at Konstanz. http://www.ub.uni-konstanz.de/kops/volltexte/2005/1584/. Neef, Martin 1996 Wortdesign. Eine deklarative Analyse der deutschen Verbflexion. T¨ubingen: Stauffenburg Verlag. Nespor, Marina 1990 Vowel deletion in Italian. The organization of the phonological component. The Linguistic Review 7: 375–398. Nespor, Marina, and Irene Vogel 1983 Prosodic structure above the word. In: Anne Cutler, and D. Robert Ladd (eds.), Prosody: Models and Measurements, 123–140. Berlin, Heidelberg, New York, Tokyo: Springer. 1986 Prosodic Phonology. Dordrecht: Foris. Oostendorp, Marc van 2002 The phonological and morphological status of the prosodic word adjunct. Linguistische Berichte Sonderheft 11: 209–236. Ortmann, Albert 1998 Consonant epenthesis: Its distribution and phonological specification. In: Wolfgang Kehrein, and Richard Wiese (eds.), Phonology and Morphology of Germanic Languages, 51–75. Tuebingen: Niemeyer.

The extended prosodic word Pater, Joe 2000

193

Nonuniformity in English secondary stress: The role of ranked and lexically specific constraints. Phonology 17: 237–274. Peperkamp, Sharon 1996 On the prosodic representation of clitics. In: Ursula Kleinhenz (ed.), Interfaces in Phonology, 102–127. Berlin: Akademie Verlag. 1997 Prosodic Words. Ph. D. diss., Universiteit van Amsterdam. Den Haag: Holland Institute of Generative Linguistics. Pierrehumbert, Janet and Mary Beckman 1988 Japanese Tone Structure. LI Monograph Series No. 15. Cambridge, MA: MIT Press. Pike, Kenneth 1967 Tongue root position in practical phonetics. Phonetica 17: 129–140. Prince, Alan 1980 A metrical theory for Estonian quantity. Linguistic Inquiry 11: 511– 562. 1983 Relating to the grid. Linguistic Inquiry 14: 19–100. Prince, Alan and Paul Smolensky 2004 Optimality theory: Constraint interaction in generative grammar. Malden, Mass.: Blackwell. [Reprint of: 1993. Optimality theory: Constraint interaction in generative grammar. Brunswick, New Jersey, and Boulder, Colorado: Rutgers University and University of Colorado, Boulder. Technical report RuCCS-TR-2] Selkirk, Elisabeth 1981 On the nature of phonological representation. In: John Anderson, John Laver, and Terry Myers (eds.), The Cognitive Representation of Speech, 379–388. Amsterdam: North Holland. 1984 Phonology and Syntax: The Relation between Sound and Structure. Cambridge, MA: MIT Press. 1986 On derived domains in sentence phonology. Phonology 3: 371–405. 1996 The prosodic structure of function words. In: James L. Morgan, and Katherine Demuth (eds.), Signal to Syntax, 187–213. Mahwah, NJ: Lawrence Erlbaum. 2000 The interaction of constraints on prosodic phrasing. In: Merle Home (ed.), Prosody, theory and experiment: Studies presented to G¨osta Bruce, 231–261. Dordrecht, The Netherlands: Kluwer Academic Publishers. Selkirk, Elisabeth and T. Shen 1990 Prosodic domains in Shanghai Chinese. In: Sharon Inkelas, and Draga Zec (eds.), The Phonology-Syntax Connection, 313–337. Chicago: University of Chicago Press.

194

Junko Ito and Armin Mester

Selkirk, Elisabeth and Koichi Tateishi 1988 Constraints on minor phrase formation in Japanese. Papers from the 24th Annual Regional Meeting of the Chicago Linguistic Society, 316– 336. Chicago: Chicago Linguistic Society. Smith, Jennifer L. 2004 Making constraints positional:Toward a compositional model of CON. Lingua 114: 1433–1464. Steriade, Donca 2000 Paradigm uniformity and the phonetics-phonology boundary. In: Michael B. Broe, and Janet B. Pierrehumbert (eds.), Papers in Laboratory Phonology V: Acquisition and the Lexicon, 313–334. Cambridge: Cambridge University Press. Truckenbrodt, Hubert 1995 Phonological phrases: Their relation to syntax, focus, and prominence. Ph. D. diss., Department of Linguistics and Philosophy, MIT. 1999 On the relation between syntactic phrases and phonological phrases. Linguistic Inquiry 30: 219–255. Vig´ario, Marina 1999 On the prosodic status of stressless function words in European Portuguese. In: Tracy Alan Hall, and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 255–294. Amsterdam Studies in the Theory and History of Linguistic Science. Amsterdam and Philadelphia: John Benjamins. Wagner, Michael 2005 Prosody and recursion. Ph. D. diss., Department of Linguistics and Philosophy, MIT. Wall, Robert 1972 Introduction to Mathematical Linguistics. Englewood Cliffs, New Jersey: Prentice-Hall, INC. Wells, John C. 1982 Accents of English. Cambridge, U.K.: Cambridge University Press. Wiese, Richard 1996. The Phonology of German. The Phonology of the World’s Languages. Oxford: Clarendon Press.

Multiple spell-out, assembly problem, and syntax-phonology mapping Yoshihito Dobashi

1. Introduction In this paper, I discuss some theoretical aspects of syntax-phonology interface within the framework of the Minimalist Program incorporating the Multiple Spell-Out Theory (Chomsky 2004; Uriagereka 1999). I propose that a string mapped to the phonological component by Spell-Out corresponds to a phonological phrase. The proposed theory is a null hypothesis in the sense that it requires no particular mapping algorithm that creates a phonological phrase by referring to syntactic information such as maximal projections (cf. Nespor and Vogel’s [1986] Relation-Based Theory and Selkirk’s [1986, 1995] End-Based Theory). This paper is organized as follows. In section 2, first I address two fundamental problems with the Multiple Spell-Out Theory of syntax-phonology mapping. One is that Multiple Spell-Out in phase theory does not give a correct prosodic phrasing under the standard assumption that Spell-Out applies to the sister of phase heads, C and v. I will call this the Mismatch Problem. The other is a problem with linearization between spelled-out units, which I will call the Assembly Problem. Then, I show that the Mismatch Problem is resolved by solving the Assembly Problem, without changing any current syntactic assumptions concerning phases. In section 3, I give a basic analysis of phonological phrasing in Ewe and Chichewa within the proposed theory. I show that the difference in phonological phrasing between these languages is largely due to the difference in their syntax. In section 4, I discuss some theoretical implications. Section 5 is the conclusion.

196

Yoshihito Dobashi

2. Two problems 2.1. The Mismatch problem In the phase theory, it is assumed that CP and vP are phases, and that Spell-Out applies to their complements (Chomsky 2004: 107–108). Given that Spell-Out is the operation that connects syntax with phonology, it is natural to assume that the output of the operation corresponds to a significant domain in the phonological component. I assume that it is a phonological phrase:1 (1)

A phonological string mapped to the phonological component by SpellOut corresponds to a phonological phrase.

This is a null hypothesis since the unit of Spell-Out corresponds to a phonological phrase as it is without any additional assumption. Given (1), let us consider the derivation of (2). Here, the verb has moved to the little v, and the vP-internal subject has moved to the spec of IP. 2 (2)

As the derivation goes on from the bottom, the sister of the little v, namely VP, is spelled-out, and then the sister of C, namely IP, is spelled-out. Under assumption (1), we would get a phonological phrasing shown in (3), where the subject is

1

2

In this paper, I will not discuss the other levels of prosodic constituents such as intonational phrases. However, given (1), it would follow that the prosodic domains larger than phonological phrases are created by assembling phonological phrases. That is, the representational nature of the strict layer hypothesis (Nespor and Vogel 1986; Selkirk 1984) can be reduced to the properties of derivation. For further discussions, see Dobashi (2007). I assume that the movement of V to v is universal for theta-theoretic reasons (cf. Chomsky 1995: 315). The copy of the moved element is indicated by angle brackets.

Multiple spell-out, assembly problem, and syntax-phonology mapping

197

phonologically phrased with the Infl and the verb, while the object is not phrased with the verb. (3)

( Subj Infl V )f ( Obj )f

In many languages, however, it is observed that the subject is not phonologically phrased with the following verb, as schematically shown in (4). For example, the environment for English Iambic Reversal does not hold across the word boundary between the subject and the verb. So, there is mismatch between a unit of Spell-Out and a phonological phrase.3 (4)

a.

b.

3

(Subj)f (V)f (Obj)f English (Iambic Reversal: Hayes 1989 ; Nespor and Vogel 1986), Ewe (M-tone Raising: ANlO dialect: Clements 1978), Italian (Raddoppiamento Sintattico: Nespor and Vogel 1986; Frascarelli 2000), Colloquial French (Liaison: Nespor and Vogel 1986; Selkirk 1974, 1984), etc. (Subj)f (V Obj)f Chichewa (Tone Retraction: Bresnan and Kanerva 1989; Bresnan and Mchombo 1987; Kanerva 1990) Chi-mwi:ni (Vowel Shortening: Kisseberth and Abasheikh 1974)

The readers are encouraged to consult each of the references listed in (4). See also Dobashi (2003) for the detailed discussions of the data. In some of the (4a)-type languages like Italian and English, the object may be phrased with the verb optionally in a certain environment. Following Nespor and Vogel’s (1986: 173) insights, I assume that this kind of optional phrasing is a result of the restructuring of phonological phrases (see also Dobashi 2006). That is, the verb and the object are phrased separately by Spell-Out, and then they undergo restructuring for purely phonological reasons (prosodic branchingness in the sense of Inkelas and Zec [1995], weight balancing in the sense of Ghini [1993], length requirement in the sense of Prieto [2005], etc.). An anonymous reviewer points out that such an assumption undermines the predictive power of the analysis and the postulation of an “original” phrasing is in many cases both difficult and ad-hoc since it is not clear how we decide which phrasing derives from the other (also see Tokizaki 2006). However, it seems to me that it is not impossible to postulate an original phrasing. In many cases the distinction between original and derived phrasings is made when there is an ambiguous phrasing for a syntactic structure. That is, if two syntactically identical structures have different phonological phrasings, then one of the two is the original phrasing. If a theory states, as a principle of grammar, that rephrasing combines two or more phrases into one and that it may not divide a phrase into two or more, then it can define the original phrases unambiguously: The smaller one is the original.

198

Yoshihito Dobashi

Kimatuumbi (Vowel Shortening: Odden 1987, 1990, 1996), Kinyambo (High Deletion: Bickmore 1989, 1990), etc. One crucial difference between the predicted phrasing (3) and the observed phrasing (4) is that the subject is phrased with the verb (and Infl) in (3) while it is phrased alone in (4). That is, even though (1) is a null hypothesis, it does not give a desired result. I call this problem the Mismatch Problem. 2.2. The assembly problem In this section, I point out another problem concerning the Multiple Spell-Out Theory of syntax-phonology interface. It is concerned with the linearization of the spelled-out units. I assume that Spell-Out defines linear order since it connects the narrow syntax, where there is no linear order among lexical items, with the phonological component, where phonological strings are linearly ordered. Let us consider the derivation in (5) in detail. Here, I omit the vP-internal subject and the movement of V to v, as it does not affect my argument here. (5)

First, the VP is spelled-out, and the linear order between the large V and the object is defined, as in (6a). Then, the IP is spelled-out, and the linear order among the subject, Infl and the little v is defined, as in (6b). (6)

a. b.

Spell-Out (Sister of v) ⇒ V