caroline r. wiltshire, joaquim camps romance phonology and variation selected papers from the 30th...

ROMANCE PHONOLOGY AND VARIATION

AMSTERDAM STUDIES IN THE THEORY AND

HISTORY OF LINGUISTIC SCIENCE

General Editor

E. F. KONRAD KOERNER

(University of Ottawa)

Series IV – CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board

Raimo Anttila (Los Angeles); Lyle Campbell (Christchurch, N.Z.)

Sheila Embleton (Toronto)

John E. Joseph (Edinburgh); Manfred Krifka (Berlin)

Hans-Heinrich Lieb (Berlin); E. Wyn Roberts (Vancouver, B.C.)

Hans-Jürgen Sasse (Köln)

Volume 217

Caroline R. Wiltshire and Joaquim Camps (eds)

Romance Phonology and Variation.

Selected papers from the 30th Linguistic Symposium on Romance Languages,Gainesville, Florida, February 2000.

JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA

ROMANCE PHONOLOGY

AND VARIATION

SELECTED PAPERS FROM THE 30TH

LINGUISTIC SYMPOSIUM ON ROMANCE

LANGUAGES, GAINESVILLE, FLORIDA,

FEBRUARY 2000

Edited by

CAROLINE R. WILTSHIRE

JOAQUIM CAMPS

University of Florida, Gainesville

The paper used in this publication meets the minimum requirements of American

National Standard for Information Sciences — Permanence of Paper for Printed

Library Materials, ANSI Z39.48-1984.

8 TM

Library of Congress Cataloging-in-Publication Data

Linguistic Symposium on Romance Languages (30th : 2000 : Gainesville, Florida)

Romance phonology and variation : selected papers from the 30th Linguistic Symposium on

Romance Languages : Gainesville, Florida, February 2000 / edited by Caroline R. Wiltshire, Joaquim

Camps.

p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV,

Current issues in linguistic theory, ISSN 0304-0763 ; v. 217)

Includes bibliographical references and index.

1. Romance languages--Phonology--Congresses. 2. Romance languages--Variation--Congresses. I.

Wiltshire, Caroline R., 1963-. II. Camps, Joaquim. III. Title. IV. Series.

PC76.L56 2002

440’0415--dc21 2001037886

ISBN 90 272 3724 7 (Eur.) / 1 58811 079 6 (US)

© 2002 – John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other

means, without written permission from the publisher.

John Benjamins Publishing Co. • P.O.Box 36224 • 1020 ME Amsterdam • The Netherlands

John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA

ACKNOWLEDGMENTS

The papers selected for inclusion in this volume originated as presentationsat the Thirtieth Annual Meeting of the Linguistic Symposium on RomanceLanguages (LSRL 30), with parasession on Current Issues in RomanceLanguage Sociolinguistics and Second Language Acquisition, which tookplace February 24-27, 2000, at the University of Florida, Gainesville, Florida.

The editors gratefully acknowledge the scholars listed below for theirgenerous assistance, first, in selecting the papers to be presented at LSRL 30,and later, for reviewing those submitted for inclusion in the selected papersfrom the conference:

Michel Achard, Theresa Antes, Joseph Aoun, Julie Auger, J.-MarcAuthier, Ignacio Bosque, Diana Boxer, Barbara Bullock, AndreaCalabrese, Vicki Carstens, J. Clancy Clements, Heles Contreras,Michel deGraf, Anne-Marie DiSciullo, Donca Farkas, Ralph Fasold,Randall Gess, Grant Goodall, Jorge Guitart, Javier Gutiérrez-Rexach,S.J. Hannahs, Galia Hatav, Julia Herschensohn, Larry Horn, JoséIgnacio Hualde, Haike Jacobs, Richard Janda, Renée Jourdenais, EllenKaisse, Richard Kayne, Paula Kempchinsky, Jurgen Klausenburger,Juana M. Liceras, John M. Lipski, Andrew Lynch, Enrique Mallén,Fernando Martínez-Gil, Diane Massam, Gary Miller, Jean-PierreMontreuil, Richard Morris, Carole Paradis, Ana T. Pérez-Leroux,David Pharies, Eric Potsdam, Lisa Reed, Lori Repetti, Susana Rivera-Mills, María Luisa Rivero, Yves Roberge, Ana Roca, Nuria Sagarra,Mario Saltarelli, Lisa Selkirk, Ester Torrego, Rena Torres Cacoullos,Christina Tortora, Barbara Vance, Co Vet, Daniel Villa, Marie-ThérèseVinet, Dieter Wanner, Ratree Wayland, Ann Wehmeyer, Lydia White,Karen Zagona, Mary L. Zampini, María Luisa Zubizarreta, and Jan-Worter Zwart

Additional thanks to Bill Calin and D.Gary Miller for their Outreach lectures,to Jean and Juanita Casagrande for their tireless service on the LSRL 30organizing committee, to Maritza Bell-Corrales, Jodi Bray, Jodi Nelms, andother UF graduate students in Linguistics and Romance who made theconference work, and to Jodi Bray, Konrad Koerner, and Anke de Looper fortheir assistance in editing this volume.

ACKNOWLEDGMENTSvi

Finally, neither the LSRL 30 conference nor this volume would have beenpossible without the financial support of the following University of Floridaunits:

The Program in LinguisticsThe Department of Romance Languages and LiteraturesThe College of Liberal Arts and SciencesThe Office of Research and Graduate ProgramsThe English Language Institute

Caroline R. Wiltshire & Joaquim CampsProgram in Linguistics & Department of Romance

Languages and LiteraturesUniversity of Florida, Gainesville

Box 115454Gainesville, FL 32611-5454

U.S.A.

April 10, 2002

CONTENTS

Acknowledgments v

List of Contributors ix

Romance Phonology and Variation 1

Caroline Wiltshire & Joaquim Camps

Constraining the Vagaries of Glide Distribution in Varieties of French 11

Barbara E. Bullock

On the Relationship between Comprehension and ProductionData in Codeswitching 27

Paola E. Dussias

Focus, Word Order Variation and Intonation in Spanish and English: 39An OT account

Rodrigo Gutiérrez-Bravo

Morphological Complexity and Spanish Object Clitic Variation 55

David Heap

Catalan Phonology: Cluster simplification and nasal place assimilation 69

Dylan Herrick

The Articulator Group and Liquid Geometry: Implications for 85Spanish phonology present and past

D. Eric Holt

CONTENTSviii

Intonation in Spanish and the other Ibero-Romance Languages: 101Overview and status quaestionis

José Ignacio Hualde

‘Partial Spanish’: Strategies of pidginization and simplification 117(from Lingua Franca to ‘Gringo Lingo’)

John M. Lipski

The Death of French in Medieval England 145

D. Gary Miller

Discourse Context and Polysemy: Spanish casi 161

Scott A. Schwenter

New Insights into French Reduplication 177

Mary Ellen Scullen

Local Conjunction in Italian and French Phonology 191

Bernard Tranel & Francesca Del Gobbo

On the Relation between Quantity-sensitive Stress and Distinctive 219Vowel Length: The history of a principle and its relevancefor Romance

W. Leo Wetzels

Index of Terms & Concepts 235

CONTRIBUTORS

Barbara E. BullockDepartment of FrenchPennsylvania State University325 S. Burrowes BuildingUniversity Park, PA 16802-6203United [email protected]

Joaquim CampsDept. of Romance Languages &LiteraturesUniversity of FloridaBox 117405Gainesville, FL 32611-7405United [email protected]

Francesca Del GobboDepartment of LinguisticsUniversity of California, IrvineIrvine, CA 92697-5100United [email protected]

Paola E. DussiasDept. of Spanish, Italian & Port.Pennsylvania State University352 N. Burrowes BuildingUniversity Park, PA [email protected]

Rodrigo Gutiérrez-BravoDepartment of LinguisticsUniv. of California, Santa CruzSanta Cruz, CA 95064United [email protected]

David HeapDept. of French, Faculty of ArtsUniversity College - Univ. ofWestern OntarioLondon, Ontario N6A [email protected]

Dylan HerrickDepartment of LinguisticsUniv. of California, Santa CruzSanta Cruz, CA 95064United [email protected]

D. Eric HoltUniversity of South CarolinaDept. of Spanish, Italian & Port. &Program in LinguisticsColumbia, SC 29208United [email protected]

CONTRIBUTORSx

José Ignacio HualdeDept. of Spanish, Italian & Port.4080 Foreign Languages BuildingUniv. of Ill. Urbana-Champaign707 S. Mathews Ave.Urbana, IL 61801United [email protected]

John M. LipskiDept. of Spanish, Italian, & Port.Pennsylvania State University352 N. Burrowes BuildingUniversity Park, PA 16802-6203United [email protected]

D. Gary MillerProgram in LinguisticsUniversity of FloridaBox 115454Gainesville, FL 32611-5454United [email protected]

Scott SchwenterDept. of Spanish & PortugueseThe Ohio State University266 Cunz Hall1841 Millikin Road

Mary Ellen Scullen3106 JimenezDepartment of French & ItalianUniv. of Maryland, College ParkCollege Park, MD 20742-4821United [email protected]

Bernard TranelDepartment of LinguisticsUniversity of California, IrvineIrvine, CA 92697-5100United [email protected]

W. Leo WetzelsGroesbeekseweg 406524 DD NijmegenThe [email protected]

Caroline WiltshireProgram in LinguisticsUniversity of FloridaBox 115454Gainesville, FL 32611-5454United [email protected]

Columbus, OH 43210-1229United [email protected]

We dedicate this volume to Professor Jean Casagrande, one of the organizersof the first LSRL, which was held at the University of Florida, Gainesville, in1971, to honor his many contributions to Romance linguistics and thelinguistics community at the University of Florida and around the world.

ROMANCE PHONOLOGY AND VARIATION

CAROLINE R. WILTSHIRE & JOAQUIM CAMPSUniversity of Florida

0. IntroductionThe papers in this volume address phonology, variation, and combinations

of the two. The types of variation include allophonics, allomorphy, polysemy,variation among dialects, across languages, and across time. The collectiondraws upon familiar Romance languages (French, Spanish, Italian,Portuguese), and those less often discussed (Anglo-French, Catalan), andranges from familiar issues discussed since the beginning of modernphonology (Trubetzkoy 1939), to those resulting from the latest theoreticaladvances (Boersma & Hayes 2001). Although many papers combine issues ofphonology and variation, this overview divides them according to those twobasic categories.

1. PhonologyThe phonology papers are roughly subdivided into three main themes:

markedness/correspondence, typology, and representations.

1.1 Markedness/CorrespondenceMarkedness, which rose to prominence in structuralist theory with

Trubetzkoy (1939), has again emerged as a fundamental tool in phonologyfollowing McCarthy & Prince (1994). Their work in Optimality Theory (OT;Prince & Smolensky 1993) secures a role for markedness not only as a way toevaluate the complexity of a phonological system but also as an explanatoryforce. The effects of markedness, however, can be hidden in corners of agrammar, due to higher ranked constraints on correspondence to an input orbase form. Two papers illuminate French phonology and morphophonologyusing markedness constraints in OT (BULLOCK, SCULLEN); both shed new lighton issues that have long been controversial. A third tackles the familiarproblem of opacity in Catalan nasal place assimilation and cluster reduction

CAROLINE R. WILTSHIRE & JOAQUIM CAMPS2

(HERRICK), showing how output-output correspondence overrides markednessto result in opacity.

BULLOCK revisits the question of how to treat glide-vowel (GV) sequencesin French, long a source of conflicting theoretical accounts based on internalevidence (e.g., from Schane 1968 to Noske 1988). BULLOCK addresses fourissues: whether all glides are vowels in underlying form, whether the glide in aGV sequence is in an onset or part of the nucleus, why a consonant plus liquidcan precede some GV sequences but not others, and why some GV sequencesare more common than others. Her answers to all these questions invoke theuse of markedness in an OT grammar. For example, the interaction ofmarkedness constraints provides for the correct surface form as a glide or avowel, regardless of the input form. BULLOCK also addresses different kindsof markedness: that of glides in margin positions, of specific diphthongs, andof CLGV sequences in different varieties. She brings external evidence to bearon her analyses, drawing from observations of child language, language games,and dialectal variation; all of the evidence converges to suggest that speakersanalyze the GV sequences as a true rising diphthong. Rising diphthongs in thenucleus result from the relative markedness of glides in margins over that ofglides in general. Constraints on markedness are also conjoined withcorrespondence constraints requiring identity between a base form and anaffixed form, providing for different results in monomorphemic vs.polymorphemic forms. Different rankings in an OT grammar account for theobserved dialectal variation in the treatment of sequences involving GV.

SCULLEN examines another longstanding problem from French, againresolving it with new data and an OT analysis. Reduplication, found in Frenchbaby-talk (Morin 1972), had been previously analyzed with disappointingresults that raise the issues of whether it is regular, productive and/or worthy ofattention. SCULLEN redeems the investigation of reduplication in French bytesting forms reported in the literature and gathering new data from currentspeakers; her analysis, developed within OT, also shows the important role ofmarkedness. Some of the apparent problems for previous analysis are resolvedwith the additional data, for instance the question of whether reduplicationstarts at the right or left edge of the base. Her survey shows that both new andpreviously reported forms tend to be right-edged, with the exceptional left-edged forms resulting when the base has a suffix at the right edge. Thoughanchoring at the right is the norm, left-edged copying is forced in such cases bya high ranking RED≈ROOT constraint that requires the reduplicant to containlexical material. Her analysis also reveals the importance of markednessconstraints such as NOCODA, and ONSET, whose force is otherwise obscured inFrench phonology due to correspondence with the input. Thus reduplication

ROMANCE PHONOLOGY AND VARIATION 3

reveals the Emergence of the Unmarked (McCarthy & Prince 1994) whencorrespondence to the input is moot, as in reduplication.

HERRICK draws upon OT analytical machinery to solve another familiarproblem: the interaction between nasal place assimilation and consonantcluster simplification in Catalan. On the surface, place assimilation issometimes opaque due to cluster simplification. Previous analyses usedCatalan to argue for cyclic rule application and lexical phonology, so thatopacity resulted from rule ordering (Mascaró 1976, Kiparsky 1985). HERRICKproposes instead a correspondence based account, in which markednessconstraints forcing deletion (*COMPLEXCODA) are limited by constraintsrequiring correspondence to input place (MAX PLACE). Furthermore, HERRICKextends the use of Output-Output correspondence (Benua 1995) to require thatthe output of words in a phrase match that of the word in isolation. Nasal placeassimilation reveals that MAXPLACE is not monolithic, but rather a rankablefamily of subconstraints, with MAX(COR) lowest ranked to allow coronals toassimilate most easily. The interaction of consonant cluster simplification andnasal place assimilation results from simply interranking the constraintsrelevant to each. Hence, the interaction of correspondence with markednessconstraints explain the apparent opacity; the result is a parallel analysis inwhich input maps directly onto output without intermediate stages.

1.2 TypologyLikewise a concern since structuralist times, typological generalizations

continue to play a major role in phonological analyses, including the works ofTRANEL & DELGOBBO, WETZELS, and GUTIÉRREZ-BRAVO.

TRANEL & DELGOBBO combine the themes of markedness and typology.They begin with two markedness constraints that are widely motivated: ONSETand NOCODA. Applying the machinery of locally conjoined constraints in OT(Smolensky 1993), through typology they predict that languages should showevidence for the existence of a constraint {ONSET & NOCODA}. Thisprediction is borne out as TRANEL & DELGOBBO show how {ONSET &NOCODA} explains the unusual acquisition pattern of Dutch syllable structure,the familiar Italian suppletive distribution of the masculine plural definitearticle, and the exceptional behavior of French h-aspiré words with respect tooptional schwa deletion. Crucially, they draw on a typical property of OTconstraints, violability, showing that the effects of {ONSET & NOCODA), whenlow-ranked, emerge only in corners of a grammar. Furthermore, their accountforces them to address questions of the domain of local conjunction and thetreatment of acquisition and variability in OT, for which they adopt the use ofconstraint ranges (Boersma & Hayes 2001) rather than strict rankings.


WETZELS re-examines a typological claim attributed to Trubetzkoy (1939),the principle equating a geminate interpretation of vowel length with the use ofmora counting. Though WETZELS has argued elsewhere for the relevance ofmora count in the location of stress in Brazilian Portuguese, a language withoutcontrastive vowel length, Trubetzkoy’s principle has been used to dismiss sucharguments for weight-sensitivity in Romance languages. WETZELS’s articleexamines the history of this idea, through Trubetzkoy’s numerous worksrefining his ideas about phonological quantity, and concludes that Trubetzkoydid intend to make a strong generalization that all and only languages with ageminate interpretation of long vowels are mora counting. However, WETZELSpoints out that the distinction of syllable vs. mora counting is the result ofTrubetzkoy’s search to find a way to represent distinctive vowel length, ratherthan weight, so that the extension to generalizations about stress systems isunwarranted. Furthermore, with changes in what counts as a phonologicalgeneralization in generative phonology, we cannot simply base typologicalclaims on structuralist generalizations. Therefore, he refutes the use ofTrubetzkoy’s principle to dismiss weight sensitivity in Romance languages,leaving the issue of stress in Romance open to decision by more familiargenerative phonology arguments.

Two articles in this volume address intonation in Romance Languages.First, GUTIÉRREZ-BRAVO examines intonation in focus contexts in both Spanishand English, analyzing the variation using an OT approach of typology throughre-ranking. Many Romance languages are known to have focus-related word-order variation, which is often given a unique structural analysis (Belletti &Schlonsky 1995, Grimshaw & Samek-Lodovici. 1998). GUTIÉRREZ-BRAVO,however, observes that sentential stress plays a role both in languages that havethe focus-related word-order variation (Spanish) and languages that do not(English). His analysis motivates variation in word order using a ranking ofprosodic vs. syntactic constraints in an OT model (following Selkirk’s 1995suggestion for English). The variation between English and Spanish resultsfrom different rankings of the nuclear stress constraint (NSR) on sentencestress relative to constraints on focus and subject position. In English, thesubject position constraint outranks NSR, forcing the focused subject to stay insitu; in Spanish, priority is given to satisfying intonational requirements oversyntactic, so that focused sentences show special word order. His accountshows that both languages share a high ranking of focus, and that the NSR iscrucially violable, rather than absent, both unlike previous accounts. By usingthe same constraints with different rankings, he is also able to avoid languagespecific intonational or syntactic mechanisms to explain the variation.


1.3 RepresentationsThe second article on intonation is provided by HUALDE’s contribution,

which provides both an overview and new evidence related to therepresentation of specific intonation contours in Spanish. HUALDE surveys theprogress made in Spanish intonation research in the last ten years due toadvances in both physical apparatus (computerized phonetic analysis) andtheoretical machinery (Pierrehumbert 1980). HUALDE’s work here focuses onareas of recent controversy in analyzing basic utterances: the phonologicalanalysis of rising pitch accents, the nature of final declarative contours, andintonational phrasing. For rising accents, which show a displacement of thehigh pitch to the syllable following the stressed syllable, HUALDE proposes acontour (L+H)* associated with the beginning of the stressed syllable. For thelast syllable in neutral declaratives, he rejects Sosa’s (1999) proposedcategorical distinction between H* and L* on the final stressed syllable of anutterance, arguing instead that independently motivated accent reductionhandles these cases. Finally, in his review of prosodic phrasing and intonation,he notes that subject-predicate or predicate-subject have a contour differentfrom neutral sentence structures. He attributes the difference to the presence ofa phrasal tone H-, located to coincide with the end of the constituent conveyingold information. Throughout, HUALDE focuses on the system of contrast andthe phonological representation of contrasting elements.

HOLT also focuses on representational questions, bringing together avariety of data relating to liquids to resolve questions about the representationof laterals, in particular in the choice of the value of the feature [continuant].He proposes a development of feature geometry combining Padgett’s (1995)articulator group, in which continuancy is dominated by place features, withWalsh-Dickey’s (1997) representation of laterals, in which these segmentshave two place nodes. Laterals may therefore be simultaneously [-cont] undertheir consonantal place node, and [+cont] under their vocalic place node. Theresult is a representation for laterals which not only can be used to make senseof synchronic spirantization in Spanish, but also can explain historical changes,such as the delateralization of [¥] that results in an affricate in early OldSpanish and the vocalization of Late Latin [l]. HOLT suggests that his modelhas potential applications in a host of other languages, both in the Romancefamily and beyond.

2. VariationTwo types of variation can be distinguishes here: dialectal variation and

variation resulting from contact between languages.


2.1 DialectalHEAP examines variation in the third person object pronoun paradigms in

two varieties of Spanish and proposes a theory of hierarchical relationships thatconstrain the possible combinations of monovalent morphological features,i.e., a feature geometry representation of morphology (c.f., Harley & Ritter1998). Furthermore, the notion of markedness plays a role in predicting thelimits of feature combinations. As in phonology, markedness is multi-dimensional and positionally determined; here it is evaluated in terms ofstructure, including complexity constraints (Rice & Avery 1993). The resultingsystem relies on an interaction between structural markedness and the availableclitic inventory. HEAP concludes that morphology supports the interpretationof markedness as a matter of structure and representation, rather than rules,derivations, or constraints. His model succeeds in predicting the variation thatdoes occur, without the overgeneralization that would result from a system ofbinary phi features allowed to freely combine. Finally, he notes that a widerange of variation is attested in Spanish object clitic systems alone; furtherstudy of these systems would help test and refine the proposed geometry.

SCHWENTER presents a novel analysis of the canonical meaning of casi anda description and analysis of an innovative use of casi in spoken ValencianSpanish. SCHWENTER argues that casi, like almost in English (Horn 2000), hastwo components, proximal and polar; the polar component is entailed but notasserted, while the proximal component is both entailed and asserted. Thevariation found in Spain's Valencian community is called “inverted casi”because it seems to have the opposite entailment. Here SCHWENTER illustrateshow to distinguish a contextually determined interpretation of a lexical itemfrom a conventionalized sense, showing that inverted casi is a case ofpolysemy rather than an ironic use. Irony exploits the proximal componentrather than the polar, and similar expressions that can replace casi in ironicuses cannot replace it in the inverted uses. Thus the inverted casi has aconventionalized interpretation in this variety of Spanish, one that mustexpress “contravention of a temporal expectation”. SCHWENTER also points outthat other extensions of casi extend the proximal and ignore the polar, such asthe illocutionary hedge in Spain. Thus, the polar component is an entailmentbut is transparent/inert, so when extending meaning in innovations, variation islimited to extending the proximal component in some way.

2.2 ContactThe three articles in this section illustrate three possible consequences of

language contact: codeswitching, reduced linguistic systems, and languagedeath.


DUSSIAS looks at the “functional element effect” in codeswitching.Previous research has found a systematic favoritism for switching certaingrammatical categories over others in code-switched productions for a varietyof languages in contact, including Spanish-English (Sankoff & Poplack 1981).DUSSIAS had earlier investigated whether this favoritism occurs in readingcomprehension as well, and found that codeswitched consitituents in whichfunctional elements do not participate in the codeswitching process arepreferred. To investigate how production and comprehension systems for code-switching are related, she conducted an experiment to check whether thecomprehension preferences are replicated in oral production. Using sentencesthat parallel those of her reading comprehension test, she elicited code-switched productions from thirty-four Spanish-English bilinguals. The resultsshow that some linguistic environments resist codeswitching more than others;for example, verbs in Spanish are usually accompanied by the Spanish COMPrather than the English. However, the production results did not always matchthe preferences for ease of comprehension. DUSSIAS discusses possibleexplanations for the preferences in production, including ‘insertability’(Muysken 1997) and psycholinguistic factors (Myers-Scotton 1993), andsuggests that semantic factors should also be investigated.

LIPSKI begins with the observation that native speakers of Spanish have astereotype of a reduced variety of Spanish to use towards and expect fromforeigners; the properties of this variety include misusing mi as subject andoverusing bare infinitives. The origins of these kinds of “simplifications” aremysterious, as infinitives, for example, are not truly simpler than other verbforms in Spanish, and as this foreigner-speak maintains a correct vocabulary,compound verbs, NP-internal agreement, and other complexities. Nonetheless,the stereotypic idea of imperfect Spanish has survived for centuries, despite itsfailure to correlate with any stages of L1 or L2 Spanish acquisition. LIPSKIcircles the world looking at contact of Spanish with other languages and theresulting pidgins and creoles. He concludes that the survival of the stereotyperelies on factors such as literary tradition, inaccurate awareness of early childlanguage, and conscious decisions to simplify Spanish grammar to speak to anL2 group which the Spanish native speakers do not expect to learn Spanishfully. The foreigner talk is thus artificially generated, rather than the result ofnatural language contact, and therefore shows distinct features from naturallyreduced systems.

Sometimes language contact results in death; such was the case of Anglo-French (AF), dead by about 1430 at the hand of English. MILLER’s workallows us to watch the death throes of AF as documented in the LondonGrocers' Company records between 1400 and 1430 (Kingdon 1883-84). The


records provide evidence for the decline of French calques in English and theincrease of French suffixes in English hybrids; both indicate that the recordsare not the product of code-switching bilinguals, who are expected to producecalques but not hybrids. Thus, the records show marked deterioration inknowledge of French in England after 1400. The AF hybrids evidence a typicallanguage death situation, in which the dying language tries to survive in someform by employing extreme measures of conversion.

In addition to explaining the observed variation within the Romancefamily, the researchers whose work appears in this volume have added to ourunderstanding of language change, cross-linguistic variation, and acquisition.The work presented here reflects the importance of data from Romancelanguages on questions of theory, and the importance of theoretical advancesin contributing to our understanding of Romance languages.

REFERENCES

Belletti, Adriana & Ur Shlonsky. 1995. “The Order of Verbal Complements, aComparative Study”. Natural Language and Linguistic Theory 13.489-526.

Benua, Laura. 1995. Transderivational Identity: Phonological RelationsBetween Words. Ph.D. Dissertation, University of Massachusetts, Amherst.

Boersma, Paul & Bruce Hayes. 2001. “Empirical Tests of the GradualLearning Algorithm”. Linguistic Inquiry 32.45-86.

Grimshaw, Jane & Vieri Samek-Lodovici. 1998. “Optimal Subjects andSubject Universals”. Is the Best Good Enough? ed. by P. Barbosa, D. Fox, P.Hagstrom, M. McGinnis & D. Pesetsky, 193-219. Cambridge, Mass.: MITPress.

Harley, Heidi & Elizabeth Ritter. 1998. “Meaning in Morphology: Motivatinga feature-geometric analysis of person and number”. Ms., University ofCalgary & University of Pennsylvania.

Horn, Laurence R. 2000. “Assertoric inertia”. Paper presented at LinguisticSociety of America, held in Chicago, January 2000.

Kingdon, John Abernethy, ed. 1883–84. Company of Grocers of the City ofLondon, A.D. 1345–1463. Facsimile, transcription, and translation [Frenchtext only]. 2 vols. London: Richard Clay & Sons (1886).

Kiparsky, Paul. 1985. “Some Consequences of Lexical Phonology”.Phonology Yearbook 2. 85-138.

Mascaró, Joan. 1976. Catalan Phonology and the Phonological Cycle. Ph.D.Dissertation, MIT. [Bloomington, Ind.: Indiana University Linguistics Club,1978.]


McCarthy, John & Alan Prince. 1994. “The Emergence of the Unmarked:Optimality in Prosodic Morphology”. Proceedings of the 24th Meeting ofthe North Eastern Linguistics Society ed. by M. Gonzalez, 333-79.Amherst, Mass.: GLSA.

Morin, Yves-Charles. 1972. “The Phonology of Echo-Words in French”.Language 48.97-105.

Muysken, Pieter. 1997. “Code-mixing: Constraints and strategies”. Paperpresented at the First International Symposium on Bilingualism, held inNewcastle upon Tyne, U.K., April 1997.

Myers-Scotton, Carol. 1993. Duelling languages: grammatical structure incodeswitching. Oxford: Oxford University Press.

Noske, Roland G. 1988. “La syllabification et les règles de changement desyllabe en français”. La phonologie du schwa français ed. by S. P. Verluyten,43-88. Amsterdam & Philadelphia: John Benjamins.

Padgett, Jaye. 1995. Stricture in Feature Geometry. Stanford, Calif.: CSLI.Pierrehumbert, Janet. 1980. The Phonology and Phonetics of English

Intonation. Ph.D. Dissertation, MIT.Prince, Alan & Paul Smolensky. 1993. “Optimality Theory: Constraint

interaction in generative grammar”. Ms., Rutgers University & University ofColorado, Boulder.

Rice, Keren & Peter Avery. 1993. “Segmental complexity and the structure ofinventories”. Toronto Working Papers in Linguistics ed. by Carrie Dyck12.2.191-202. Toronto: Toronto Working Papers in Linguistics.

Sankoff, David & Shana Poplack. 1981. “A formal grammar of code-switching”. Papers in Linguistics 14.3-43.

Schane, Sanford. 1968. French phonology and morphology. Cambridge,Mass.: MIT Press.

Selkirk, Elisabeth. 1995. “Sentence Prosody: Intonation, Stress and Phrasing”.The Handbook of Phonological Theory ed. by J. Goldsmith, 550-569.Oxford: Blackwell.

Smolensky, Paul. 1993. “Harmony, Markedness, and Phonological Activity”.Handout for Rutgers Optimality Workshop 1, October 23, 1993. [ROA-87]

Sosa, Juan Manuel. 1999. La entonación del español. Madrid: Cátedra.Trubetzkoy, Nikolaj. 1939/1969. Principles of Phonology (trans. by Christiane

A.M. Baltaxe). Berkeley & Los Angeles: University of California Press.Walsh-Dickey, Laura. 1997. The Phonology of Liquids. Ph.D. Dissertation,

University of Massachusetts, Amherst.

CONSTRAINING THE VAGARIES OF GLIDE DISTRIBUTIONIN VARIETIES OF FRENCH

BARBARA E. BULLOCKThe Pennsylvania State University

0. IntroductionThe distribution of glides and their alternation with high vowels is a staple

issue in analyses of French phonology. This distribution, although complex, islargely predictable for standard French or, more accurately, le français deréférence. Despite such regularity, the same language internal data havespawned numerous competing theoretical interpretations over the last fewdecades. Extant analyses rely in varying degrees on a distinction between ‘true’diphthongs which are nuclear and derived or ‘false’ ones in which a glide,derived from an underlying high vowel, is parsed consonantally as part of thesyllable onset. Here, this distinction will be shown to be untenable. In thisarticle, I examine language-external evidence that points to how speakers ofFrench actually treat surface glide-vowel (GV) sequences in linguisticperformance. A great deal of external evidence -- from child language tolanguage games and dialectal variation -- converges on a view of glides inFrench as essentially part of a complex syllable nucleus. In essence, althoughmost linguistic descriptions of the phonetics of the language claim thatdiphthongs do not exist in French, external data based on speakers’ linguisticbehavior show instead that rising diphthongs are quite plentiful and that theyneed not be assumed to derive from their corresponding high vowelcounterparts. Instead, what drives the distinction between dieresis (VV) anddiphthongs (GV) across different language varieties is markedness interactionswith faithfulness constraints.

1. The formal interpretation of syneresis and dieresisThere are three glides in French corresponding to the three high front

vowels. Their essential distributional properties are: all glides can appear wordinitially (e.g. yeux “eyes” [jO], huile “oil” [Áil], oie “goose” [wa]). The coronalglide [j] has a wider distribution than the others in all dialects, appearing

BARBARA E. BULLOCK12

finally (ail “garlic” [aj]) and intervocalically (royal [rwajal]). In northernvarieties of French, syneresis occurs after a single syllable initial consonant(lieu “place” [ljO], puis “then” [pÁi], pois “pea” [pwa]) in both derived andlexically listed forms. In southern varieties of French and in careful speechstyles everywhere, either syneresis or dieresis is an acceptable output after asimple onset in derived forms. In all varieties, syneresis is avoided after aconsonant plus liquid (CL) syllable initial cluster in derived forms (pli+er “tofold” [plije], *[plje]; flu+er “to flow” (tide) [flye], *[flÁe]; écrou+er “to lockup” [ekrue], *[ekrwe]). After a complex onset in lexically listed forms, only GVcombinations of [Ái] (fruit [frÁi]), [wa] (gloire “glory” [glwar]), andinfrequently [wE‚] (groin “snout” [grwE‚]) occur.

All analyses of French glide distribution have proceeded from theassumption that at least some GV sequences are derived from VV ones.However, the question of which ones is very much undecided. Schane (1968)and Dell (1973), working without the benefit of syllable boundaries in earlygenerative phonology, simply posit a massively overgeneralized ruleconverting all high V+V clusters into glides. In more nuanced analyses, Morin(1976) recognizes that syllable conditions (global or derivational constraints)block syneresis after clusters and that certain GV clusters are more likelyoutputs for historical reasons than are others. Kaye & Lowenstamm (1984),Noske (1982, 1988), and Tranel (1987) argue for the monophonemic status ortrue diphthong status of [Ái] and [wa] as distinct from the derived status of allother GV clusters. de Kok & Spa (1978) depart significantly from the otherapproaches by assuming that French has rules of both syneresis and dieresis,deriving vowels from glides and vice versa under different conditions.

Within non-linear phonology, a high vocoid’s syllabicity is dependent onthe segment’s association to a syllable template. Alternations between highvowels and their corresponding glides can be represented as a simple reflex oftheir association to either the onset, where they become consonantalized asglides, or the nucleus, where they remain syllabic. Kaye & Lowenstamm(1984), Noske (1982, 1988) and Tranel (1987) each propose a syllable templatefor French that consists of a maximal onset of two consonants. Their insight isthat since CLG clusters are not permitted as onsets, potential syneresis isblocked by an already saturated syllable margin. The constraint, cast as a filter,is shown in (1).

GLIDE DISTRIBUTION 13

(1) from Kaye & Lowenstamm (1984:138)

*Onset

X [+son] [+son]

Without the blocking onset cluster, the glide maps freely into the availableonset position as shown in (2).

(2) σ

O R

lu + e → l w e louer “to rent”

Support for an analysis in which surface glides derive from mapping vowels toonset positions as in (2) comes from the fact that some word initial glides canbehave like consonants in that they block liaison and elision just as consonantsdo (le yack “the yak” [l´jak], le whiskey [l´wiski]. On the other hand, dieresisoccurs when the filter in (1) is activated. Within this analysis, the residualproblem of the CLGV sequences found in words like fruit, gloire, and groin, isdispatched by assuming that they contain true diphthongs, that is the glideportion falls not into the onset but, instead, into a branching nucleus with thevowel.

2. The inadequacy of the formal treatmentThe formal treatment summarized in the section above describes the

distribution in le français de référence. Unfortunately, the crux of the accountwhich draws a distinction between a select group of nuclear GV sequences andnon-nuclear ones in which the glide surfaces in the onset fails to adequatelyaccount for a wide range of linguistic external data. In fact, the externalevidence points in exactly the opposite direction; that is, speakers consistentlyparse all non-initial GV sequences as nuclear so that the distinctivity of the‘true diphthongs’ amounts to nothing more than a slightly more liberaldistribution in lexical items. Further, the interplay between syneresis anddieresis in actual pronunciation is quite variable and dependent on such thingsas regional variation, affective pronunciation, and speech rate. In what follows,I review a range of external data that points to the nuclear status of GV clusters.


2.1 Child languageA child’s acquisition of phonology is marked in its earliest stages by a

marked preference for CV syllable structure. From the very early stages (lessthan 10 months of age), French children begin to produce CGV syllables asreadily as CV ones. Examples from François et al. (1977:57-60) are shown in(3).

(3) [bwE‚] 10 months [sE‚bjE‚] 17 months[mjamja] 13 months [vavwar] 17 months[mjau] 16 months [abwar] 18 months

Although young children tend to neutralize the anterior labial-velar glide to itsback counterpart [w], their early productions of CGV syllables and the lack ofsimilar CCV or CLV ones indicates that the glide is treated most likely as part ofa branching nucleus rather than as part of a complex onset.

2.2 Language gamesEvidence for adult perception of syllable constituency is provided most

directly by language games in which onsets and codas are manipulated by aspeaker. Verlan, a syllable inverting game, is but one of several syllablemanipulating games stretching back to probable early 19th century origins thatare still in use today. The others include the infixing game, javanais, in whichthe sequence -av- is inserted within a syllable, and a Parisian butcher's slangknown as largonji des louchébèms, where word initial onsets are detached andmoved to the right edge of the base and then affixed with a dummy, preferablypejorative, suffix.

The language game data I include in (4) and (5) are all taken frompublished performance data, elicited by linguists from native speakers.

(4) Performance dataa. Verlan (from Méla 1988; cf. also Leftkowitz 1991, Scullen 1993, Seguin &

Teillard 1996)moi [wam] “me”toi [wat] “you”chien [jEEEE‚S] “dog”bien [jEEEE‚‚ ‚‚b] “good’

poil [walp] “hair’


b. Largonji des louchébèms (Mandelbaum-Reiner 1991, Plénat 1985)pied [ljepE] “foot”chier [ljeS] “to shit”boire [lwarbEm] “to drink”voir [lwavEm] “to see”viande [ljaaaa‚‚ ‚‚dvem] “meat”huit [lÁÁÁÁite] “eight”

(5) Javanais (Plénat 1991)joua [Za.u.a.vwa] “play”, 3rd. sg. pret.fois [fa vwa] “times”poignet [pwa va.¯E] “wrist”atelier [a va ta v´ la vje] “workshop”choir [Sa vwar] “to drop”derrière [da vE rja v´] “behind”vieux [va vjOOOO] “old”poursuivait [pa vur sa vi va vE] “pursue”, 3rd. sg. imp.pointe [pwa vEEEE‚ ta v´] “peak”bien [ba vjEEEE‚] “good”prétoire [pra ve ta vwar] “tribunal”ses yeux [sa vE za vjOOOO] “his/her eyes”point [pa vwEEEE‚] “point”client [kla vi ja vaaaa‚] “client”

Notice that in verlan (4a) and largonji (4b), GV clusters are invariably treated asbelonging to the nucleus. Interestingly, verlanized forms, whether vowel-,consonant-, or glide-initial, always block liaison and elision and thus tell usnothing about the status of surface word-initial glides. However, children’sspelling of verlanized words given in Seguin & Teillard (1996) indicate apreference for orthographic vowels to indicate initial GV clusters althoughthere is some variation. For instance, chien is spelled as ‘yinche’ or ‘ienche’but doigt is spelled ‘oide’ rather than *‘wad’ and joie is written as ‘oije’, not*‘waje’.

The javanais data (5) provided by Plénat's study is more complex. Hisspeaker, a 62 yr. old man with years of practice with javanais nearly alwaysparses surface GV series as nuclear but, in 5 out of the 27 possible occurrencesof such clusters in the base form, he splits the GV with the infix. There is noconsistency, however, in the quality of the cluster; [ja], [je], [wE‚], [Ái], [wa] areall split at times. Compare pointe with point in (5).


(6) [pwa va ta v´] [wa] [ka vç‚ tra va rja ve] [je][pa vur sÁÁÁÁa vi va vE] [Ái] [va va…rja va Bla v´] [ja]

While we certainly expect some inconsistency from performance data, theinteresting thing to note about the split GV clusters in the data in (6) is that thefirst three tokens presumably reflect the so-called only ‘true diphthongs’ in theformal accounts. We might expect that, based on their theorized uniquephonological status as underlying complex nuclei, these sequences wouldsurvive intact to the output. In fact, it seems as if speakers make no distinctionat all between types of surface GV clusters, recognizing them almost uniformlyas diphthongs and, if choosing to split the cluster, they separate the putative‘true diphthongs’ as frequently as they do the ‘false’ ones. This distinction,though theoretically convenient, does not hold of actual speech.

2.3 Regional and dialectal variationDialectal language data show a great range of strategies when confronted

with potential CLGV clusters. One frequently exercised option in many regionalvarieties of French is to delete the liquid, reducing the complex onset to asingleton consonant but keeping the GV cluster intact. Thus, pronunciationssuch as [fwa] froid “cold”, [twa] trois “three”, [pÁi] pluie “rain” are frequentthroughout the French-speaking world. Less usual, but certainly well-attested,are dialectal and regional forms in which a schwa is epenthesized between theconsonant and liquid, again leaving the GV unchanged. Examples of thisphenomenon, known in the phonological literature as Dorsey’s law, are shownbelow in (7) and (8).

(7) Normandy varieties (Brasseur 1980)a. groseillier (à maquereau) “gooseberry bush”

i. [gruezil]ii. [g´rwazel]

b. “pebbles” dialect wordi. [grue]ii. [g´rwa]

c. brouette “wheelbarrow”i. [bruwEt]ii. [b´rwEt] compare with Old French berouette

Each of the (a) forms are from northwestern regions of Normandy. The (b)forms are all from Orne in the area of Saint Fraimbault. While the schwa may


be historic in brouette, reflecting the earlier French berouette, it is notetymological in the form for “gooseberry bush” (Old Dutch croesel).

Morin (1976) records several similar dialect words culled from earliergeolinguistic atlases of France. In some dialects, Dorsey’s law applies to stemonsets under affixation.

(8) a. prier [p´rje] “to beg” c. trouer [t´rwe] “to dig”b. oublier [ub´lje] “to forget” d. truelle [t´rÁEl] “trowel”

In (8a-c), the surface glide is clearly derived from the final high vowel of averbal stem. Each of the cases in (8) mirrors the precise phonetic environmentin which dieresis is expected in standard French. Note, though, that for thedialectal varieties in (7) and (8), there is a distinct preference to keep thediphthong intact in the output regardless of its lack of correspondence in (8a-c)with its bare stem form.

The dialectal forms cited above are reminiscent of the interaction betweenthe appearance of surface glides and schwa in French words such as atelier“studio” [at´lje], bourrelier “harness-maker” [bur´lje] and conditional first andsecond plural verb forms like demanderions “we would ask” [d´ma‚d´rjç‚]. Inthese examples, schwa, which is normally deletable between a C and L, isretained. Most traditional accounts have ordered glide formation to occurbefore, and thus block, the rule of schwa deletion.

In the theoretical sketch of glide distribution and variation in French whichfollows, I will outline an approach to this issue that is almost exclusively basedon markedness constraints on surface outputs. These constraints, ordered withrespect to one another and interacting with constraints demanding identity tostem vowels, will derive a variety of surface outputs.

3. Rising diphthongs in the ouput as an effect of markednessThe overview of external evidence seems to suggest that speakers actually

treat surface GV strings as belonging, as a unit, to the nucleus of the syllable.They are parsed as vowel-like by children and by adults in language games.Various regional and dialectal evidence demonstrates that speakers tend topreserve GV surface clusters intact, rather than resolving them into theirputatively underlying VV forms. Data from other sources, such as hypocoristic(nickname) formation in French has led other researchers, notably Scullen(1993), to also conclude that non-initial glides are joined with their followingvowel into a syllable peak. If external evidence of this sort is taken to be agood proving ground for testing the observational adequacy of a theoreticalanalysis, then the standard approach clearly comes up short.


The standard theoretical approach to the language internal data rests on theassumption that syneresis is, unequivocally, the result of a high vocoidappearing in an onset. However, even on the basis of the internal data alone,belief in this assumption requires that we pay more attention to the lexicallypredicated behavior of consonant-like initial glides than we do to the equallycompelling, and almost exceptionless, fact that the rounded glides, unlike theircoronal counterpart, fail to occur in other consonantal positions (cf. Heap etal.1992 on variation in the production of word initial glides). Yet aside from afew exotic tokens, like the place name Haüy [aÁi], we do not find roundedglides either in codas or in word internal onsets. Such rarity suggests thatrounded glides do not make good consonants. In other words, rounded glidesare very marked consonantal outputs.

The distributional facts alone suggest that there is a fairly high-rankedconstraint in French that penalizes parsing glides, particularly the roundedglides, into syllable margins (9) (cf. Prince & Smolensky 1993).

(9) ...*M/Á » *M/w » *M/j...

The choice of making the labial velar glide more harmonic as a margin than thefront rounded glide is weakly motivated by the slightly more liberaldistribution [w] enjoys in French but, more important, it may simply reflect amore universal markedness of the front round glide:

(10) *Á » *w » *j

According to Ladefoged & Maddieson’s survey (1996:322), 85% of theworld’s languages have the coronal approximant, 76% contain the labial velar,and only 2% of languages contain any other type of semi-vowel. Crucially,some varieties of French even neutralize the labial-palatal approximant to thelabial velar.

(11) *[pÁi] ~ [pwi] (Belgium French, Jersey French (Liddicoat 1994)).

These markedness constraints are separately rankable with respect to othermarkedness and faithfulness constraints in a grammar. If undominated, (10)would prevent surface glides from appearing in the output at all. Theconstraints in (9), if high ranked, more modestly prevent surface glides frombeing parsed as onsets and codas. Since the labial glides are virtually absentfrom these positions, I assume that the markedess constraints penalizing theirappearance in traditionally consonantal syllable positions are undominated.


Hence, a ranking such as that in (12) represents the normal state of affairs inFrench.

(12) *M/Á, *M/w » ONSET

Notice that (12) does not necessarily preclude the appearance of glides in theouput. This ranking only ensures that a labial glide not be parsed as an onset; itwould, however, permit such segments to surface in the output as diphthongsas shown in Tableau 1, where nuclei are represented within brackets.

doigt “finger, digit” *M/w ONSET *wa) dw[a] *!b) d[u][a] *!c)☞ d[wa] *

Tableau 1: Effect of undominated *M/w

The constraint against a coronal glide in a margin position is, however, not ashighly ranked as the one against its labial counterparts. Within words, [j]surfaces intervocalically and is, thus, indistinguishable from an onset (13).

(13) Intervocalic [j]: ONSET » *M/jmayonnaise [majçnEz] “mayonnaise”joyeux [ZwajO] “joyous”

The relatively low ranking of the *M/j constraint with respect to ONSET doesnot, however, entail that [j] will be parsed everywhere as a consonantal onset.In fact, the ranking of these two constraints for forms that already have aparseable consonantal onset is in some sense irrelevant. The best candidate willstill be one that incurs no violations of either constraint, that is, a diphthongalparse.

4. The effect of markedness & faithfulness in derived formsNotice that because markedness constraints alone can deal with these

distributional facts, there is no need to enforce any correspondence between theinput and the output in underived words. This means that the underlyingfeature value of a surface glide can be either syllabic or non-syllabic. Withderived words, however, transderivational faithfulness does play a role.Consider, for example, the effect of the conjunction of faithfulness andmarkedness in verb derivations and the variation it can produce. There are


often regional and stylistic preferences favoring dieresis to syneresis inderivatives depending on the nature of the high vowel of the base form. Tranel(1987:122), reproduced and modified here, notes the following variation.

(14)nier “to deny” nouer “to knot” nuer “to shade”

a. [nje] [nwe] [nÁε] syneresis

b. [nje] [nwe] [ny.e]

c. [nje] [nu.e] [ny.e] dieresis

The variety in (14c) is generally indicative of Meridional Frenchpronunciations with a preference toward dieresis with the rounded frontvowels. The register in (14a), with syneresis regardless of vowel quality, isrepresentative of Paris. The pronunciation shown (14b) is the most widespreadall across northern France.

Each of the infinitives in (14) stands in correspondence with a related highvowel final stem form: [ni-], [nu-], [ny-]. There is, in essence, an identityrelationship between a stem and its affixed form which may or may not behighly ranked depending on the linguistic variety. Following Bakovic(2000:23), such a relationship demands that morphologically related formshave identical feature values (15).

(15) S(tem)A(ffix)-IDENT[F]A segment in an affixed form [Stem + affix] must have the samevalue of the feature [F] as its correspondent in the stem ofaffixation [Stem].

In sum, the variation of (14) can be seen to fall out of the markedness hierarchyadvanced for the surface glides conjoined with Bakovic’s constraint. Theconjunction of constraints, as illustrated by Lubowicz (1998), has the effect ofpromoting a normally low-ranked markedness constraint in a derivedenvironment. Clearly, in French, constraints against output glides are low-ranked. However, when conjoined with faithfulness, the effect of *glide can beelevated within the ranking hierarchy The following interactions of possibleconstraint rankings define the pronunciation variations of (14).


(16) Register a:ONSET » [SAIdent[V] & *Á]: [SAIdent[V] & *w]: [SAIdent[V] & *j]Register b:[SAIdent[V] & *Á] » ONSET » [SAIdent[V] & *w]: [SAIdent[V] & *j]Register c:[SAIdent[V] & *Á]: [SAIdent[V] & *w] » ONSET » [SAIdent[V] & *j]

If ONSET dominates a conjoined faithfulness and markedness constraint, thensyneresis is optimal and the glide will surface. If the local conjunction insteadis dominant, then, the resulting output contains a high vowel identical to that ofrelated lexically listed stem. As expected, attested outputs differ according todifferent ranking hierarchies.

5. The peak fitness of diphthongsAs we have seen in the evidence presented so far, some surface GV

sequences have a more liberal distribution than do others although they are alltreated rather uniformly by native speakers. The putative ‘true diphthongs’surface intact after complex onsets in underived words. I propose that just asall surface glides are evaluated for margin goodness, all GV diphthongs inFrench enter into a markedness hierarchy that evaluates them for peak harmonyalong the lines suggested by Prince & Smolensky (1993). The ‘truediphthongs’, then, are simply the least marked. Because this ranking islanguage specific, we must stipulate its order, placing [Ái], [wa], and [wE‚]lower than all other combinations. A suggestion as to how this might look ismade in (17). Any glide combination to the left of the hierarchy is a worse peakthan those that fall to the right. Vowels are clearly less marked as peaks thandiphthongs, and in French, marks against consonantal peaks are dominant.

(17) ....*P/jE » *P/wa,wE‚ » *P/Ái ...

What the hierarchy in (17) recognizes is that all complex nuclei are not equallyharmonic. Given that the evaluation of each diphthong is separately rankable,the prediction made here is that grammars may differ according to harmonicrankings. In fact, there is good evidence from the history of French that theseries [jE] at one time enjoyed the same liberal distribution as the so-called‘true diphthongs’. The 17th century grammarian, Ménage, attributes thepenchant for dieresis in words like meurtrier “murderer” to Corneille, claimingthat before Corneille, the only speakers to pronounce three syllables in wordslike sanglier “wild boar”, peuplier “poplar”, bouclier “shield” were womenand foppish men (Nyrop 1909). Even in contemporary French, this diphthongis frequently permitted after clusters as a verb ending, as in boucliez “you


should buckle” [buklje] (cf. Tranel 1987). Thus, within certain historicalperiods and, today, under particular morphological impetus, the series CLjVenjoys a more liberal distribution than presumed by current formal analyses.

The implication of the markedness constraints on branching nuclei is thatthere is no global constraint in French that prohibits CLGV series. Instead, thereis a distributional constraint on what kind of branching nuclei may follow acomplex onset. In essence, all surface GVs following a syllable initial onset arerising diphthongs, some are simply less marked in that position than are others.Just as we have done above in (17) to account for syllable structure variationacross regions of French, we can account for the differing distribution ofdiphthongs after complex clusters in underived words via the followinginteractions (18).

(18) a. Prior to the 17th century or “men’s speech” at the time ofMénage: ONSET » *CL[jE]: *CL[Ái]: *CL[wa]

b. After the 17th century or “women’s speech” at the time ofMénage: *CL[jE] » ONSET » *CL[Ái]: *CL[wa]

Because there has been a good deal of indeterminacy concerning thepronunciation of clusters like that in groin, I have left this particular diphthongout of the hierarchies in (18). Needless to say, breaking down the globalconstraint of CLGV into a markedness constraint on diphthong distributionallows for a good deal of variation and change.

Again, however, we must deal with the issue of the non-occurence of CLGVclusters in derived forms. A classic example of the language’s resistance to thiscluster in a derived form is the 3rd. person singular preterite form of the verbtrouer “to dig” - troua [trua] - which could be potentially homophonous withthe word “three” trois [trwa] but is not. Even though [wa] is considered to be a‘true diphthong’, here redefined as a relatively unmarked diphthong, it does notsurface in derivations after a complex onset if it stands in a faithfulnessrelationship to a stem final vowel.

In order to explain this relationship, we again invoke the SA-IDENT[V]constraint which penalizes any deviation from the feature value of, in this case,a stem final vowel. As we have seen above in (16), this faithfulness constraintplays a role in derived forms when conjoined with particular markednessconstraints. The effect of such conjunction is to elevate low-ranked markednessconditions within the hierarchy, prohibiting features which surface in lexicalforms from appearing in derived environments. In the case at hand, diphthongswhich can appear in underived outputs after a cluster are categorically excludedfrom this distribution in derivations.


The form of the conjunction necessary is simply a more specificmarkedness constraint than that mentioned in (16). Conjoining SA-IDENT[V]with a sequential markedness condition *CC[GV] will effectively rule outderived diphthongs after complex onsets as shown in Tableau 2. But, absentsuch conjunction, underived lexical items escape the effect of low-rankedmarkedness (Tableau 3).

trou+abase: tru

[SAIdent[V] & *CC[wa]] ONSET

☞ [trua] * [tr[wa]] *!

Tableau 2: Effect of conjunction in derived forms

trois Onset *CL[wa] [trua] *!☞ [tr[wa]] *

Tableau 3: Non-effect of markedness in the lexicon

At this point in time in the history of the language, all derived diphthongs areruled out after complex onsets. Given the vagaries of linguistic change,however, there is always reason to anticipate that things may change.

6. ConclusionIn this article, I have argued that we can account for a wide range of

evidence in French by considering internal GV clusters to be contained withinthe nucleus. Crucially, we need not be concerned with the underlying form ofsuch clusters as, in fact, there is good evidence to suggest that faithfulness tosyllabic features only enters into consideration when vowel-final stems stand ina correspondence relation with derived forms. Under this approach, underivedforms may contain input vowels or glides; all constraints address output only.Reranking of these constraints leads to linguistic variation.

One striking aspect of contemporary variation is a vogue among journaliststo produce high vowels in lexical items which normally contain onlydiphthongs. In this expressive pronunciation, considered to be more correct ormore chic (Lyche & Girard 1995), the ‘elicit’ dieresis is accompanied by a shiftof stress to the word initial position. So common is this style that I was able torecord several examples by listening to a 10 minute recording of France-Inforadio (e.g. Vienne [viεn], “Vienna”; siège [siEZ] “seat”; souhaiter [suete] “tohope”; violence [violãs]. It is possible to think that this is evidence that high


vowels underlie surface glides. But, so marked and remarked upon is thisjournalistic pronunciation that it is perhaps only in the media -- and inlinguistic-internal phonological analyses -- that high vowels can bereconstructed in underlying form for all surface glides.

REFERENCES

Bakovic, Eric. 2000. Harmony, dominance and control. Ph.D. Dissertation,Rutgers University, New Brunswick, NJ.

Brasseur, Pierre. 1980. Atlas linguistique et ethnographique normand. Paris:Editions CNRS.

Dell, François. 1973. Les règles et les sons: introduction à la phonologiegénérative. Paris: Hermann.

François, Frédéric, Denise Francois, Emilie Sabeau-Jouannet & Marc Sourdot.1977. Syntaxe de l’enfant avant 5 ans. Paris: Larousse.

Heap, David, T. Nadasdi & J. Tennant. 1992. “Elision et semi-voyelles enfrançais”. Phonétique, phonostylistique, linguistique et littérature:Hommages à Pierre Léon ed. by P. Martin, 165-187. Toronto: EditionsMélodie.

Kaye, Jonathan & Jacques Lowenstamm. 1984. “De la syllabicité”. Formesonore du langage ed. by F. Dell, J. Hirst & J.-R. Vergnaud, 123-160. Paris:Herman.

de Kok, A. C. & Jaap J. Spa. 1978. “Semi-vocalisation, diérèse et OLISEM”.Linguistics 213.65-77.

Ladefoged, Peter & Ian Maddieson. 1996. The sounds of the world’slanguages. Oxford & Cambridge, Mass.: Blackwell Publishers.

Leftkowitz, Nathalie J. 1991. Talking backwards, looking forwards: TheFrench language game verlan. Tübingen: Gunter Narr.

Liddicoat, Anthony. 1994. A grammar of the Norman French of the ChannelIslands: The dialects of Jersey and Sark. Berlin: Mouton de Gruyter.

Lubowicz, Anna. 1998. “Derived environment effects in OT”. Ms., Universityof Massachusetts, Amherst. [ROA 239-0198].

Lyche, Chantal & Francine Girard. 1995. “Le mot retrouvé”. Lingua 95(5).205-221.

Mandelbaum-Reiner, Françoise. 1991. “Secrets de boucher et largonji actueldes louchébèm”. Langage et Société 59.21-49.

Méla, Vivienne. 1988. “Parler verlan: règles et usages”. Langage et Société45.47-72.


---------. 1991. “Le verlan ou le langage du miroir”. Langages 101.73-94.Morin, Yves-Charles. 1976. “Phonological tensions in French”. Current

Studies in Romance Linguistics ed. by Marta Lujan & F. Hensey, 37-49.Washington, DC: Georgetown University Press.

Noske, Roland G. 1982. “Syllabification and syllable changing rules inFrench”. The structure of phonological representations ed. by H. G. van derHulst & N. Smith., vol 2.257-310. Dordrecht: Foris.

---------. 1988. “La syllabification et les règles de changement de syllabe enfrançais”. La phonologie du schwa français ed. by S. P. Verluyten, 43-88.Amsterdam & Philadelphia: John Benjamins.

Nyrop, K. R.1909. Grammaire historique de la langue française. Copenhagen:Nordisk Forlag.

Plénat, Marc. 1985. “Morphologie du largongi des loucherbems”. Langages78.73-122.

---------. 1991. “Le javanais: concurrence et haplologie”. Langages 101.95-117.Prince, Alan & Paul Smolensky. 1993. “Optimality Theory: Constraint

interaction in generative grammar”. Ms., Rutgers University & University ofColorado, Boulder.

Schane, Sanford. 1968. French phonology and morphology. Cambridge, Mass.:MIT Press.

Scullen, Mary Ellen. 1993. The prosodic morphology of French. Ph.D.Dissertation, Indiana University, Bloomington.

Seguin, Boris & Frédéric Teillard. 1996. Les céfrans parlent aux français.Paris: Calmann-Lévy.

Tranel, Bernard. 1987. The sounds of French: an introduction. Cambridge:Cambridge University Press.

ON THE RELATIONSHIP BETWEEN COMPREHENSION ANDPRODUCTION DATA IN CODESWITCHING

PAOLA E. DUSSIASThe Pennsylvania State University

0. IntroductionNumerous studies examining the regularity of occurrence of types of

codeswitches in bilingual production data reveal that there is a systematicfavoritism for switches that involve certain grammatical categories over othersduring codeswitched speech (see, for example, DiSciullo, Muysken & Singh1986, Halmari 1997, Klavans 1985, Lance 1975, Lipski 1985, Myers-Scotton1993, Park, Troike & Park 1993, Poplack 1980, Sankoff & Poplack 1981,Treffers-Daller 1995). In this respect, the available literature indicates thatwhereas functional elements tend to appear in one language, their complementsappear in the other language (Muysken 1997). This phenomenon has come to beknown in the literature as ‘the functional element effect.’1 Several studieslooking at Spanish-English codeswitching data, for example, have shown thatwhen switches involve nominal phrases with a Determiner (Det) and a Noun(N), instances in which both Det and N appear in the same language occur lessfrequently than cases in which Det appears in one language and N in the other.This asymmetry has been attested in Sankoff & Poplack (1981), who note thatthe relative propensity of switches occurring between a determiner and a nounwas 13% in their data, compared to the 2.3% figure representing codeswitchedsentences containing the determiner and the noun in the same language.Another example is provided in a recent study by Milian (1996, cited in Myers-Scotton & Jake 1997), who reports that 81% of the switches involving

1 Following Cowper (1992), I define functional elements as categories that lack substantivemeaning, do not assign theta roles, are closed classes (no new words can be created), and do notpermit recursion on X-bar. Lexical elements, on the other hand, are defined as categories thathave substantive meaning, assign theta roles to their arguments, are open classes, and permitindefinite recursion on X-bar. Accordingly, COMP, INFL, and DET are instances of functionalelements, and N, V, and A are instances of lexical elements.

PAOLA E. DUSSIAS28

determiners and their corresponding nouns consist of English nouns andSpanish determiners, but only 6% of the switches involve English determinersaccompanied by English nouns.

The limited participation of Complementizer (Comp) in codeswitchingprovides another example of this functional element effect. To take an example,Sankoff & Poplack (1981) report the propensity for a subordinate conjunctionto be the locus of a codeswitch at less than 0.2% in their data, which contrastswith the 3.9% representing the propensity of switches that occur at the categoryfollowing the subordinate conjunction2. In addition, the observation thatsubordinate conjunctions preferentially appear in the language of the headelement on which they depend (i.e., what DiSciullo, Muysken & Singh 1986,Halmari 1997, and Treffers-Daller 1995 refer to as ‘the governor’ or ‘caseassigner’) suggests that complementizers are subject to minimal participation incodeswitching. Examples (1) through (4) illustrate the types of codeswitchesinvolving Det-NP and Comp-IP clauses most frequently encountered innaturalistic data:

(1) Mis padres van a venir para los holidays.My parents are going to come for the holidays.“My parents are going to come for the holidays.”

(2) Es una little box asina y yaviene…Is a small box like this and alreadycomes…“It is a small box like this and it comes already …”

(3) No, la potato de anoche, you acabaste conella.No, the potato from last night, you finished withit.“No, the potato from last night, you finished it.”

(4) I’m not saying that son chuecos, yo nodigo eso.I’m not saying that are old, I NEGsay that.“I’m not saying that they are old, I am not saying that.”

2 See Lipski (1985) for similar results.

COMPREHENSION AND PRODUCTION IN CODESWITCHING 29

1. The functional element effect in comprehension dataIn a series of studies, Dussias (1997, 1999, 2001) investigated whether the

functional element effect, often observed during sentence production, occurredin comprehension as well. Dussias (1999), for example, conducted twoexperiments in which subjects’ eye-movements were recorded while readingsentences with codeswitches between functional heads and their complements.Twenty-four Spanish-English bilinguals participated in the study. All subjectshad learned the two languages before the age of six and reported using bothlanguages in their daily lives and in a variety of contexts.

Experiment 1 investigated codeswitches involving Comp and its IPcomplement by looking at reading performance in two conditions. In Condition1, the codeswitch occurred at Comp and in Condition 2, it occurred at IP. Asample of each condition is given in (5) and (6) below:

(5) La maestra no sabía that the boy had left.The teacher neg know that the boy had left.“The teacher did not know that the boy had left.”

(6) La maestra no sabía que the boy had left.The teacher neg know that the boy had left.“The teacher did not know that the boy had left.”

Experiment 2 investigated switches that involved Det and its complementNP. Again, two conditions were compared. In Condition 1, the codeswitchoccurred at Det, whereas in Condition 2 the switch occurred at NP. This isillustrated in (7) and (8):

(7) La maestra compró the books for the children.The teacher bought the books for the children.“The teacher bought the books for the children.”

(8) La maestra compró los books for the children.The teacher bought the books for the children.“The teacher bought the books for the children.”

The findings revealed that for both experiments, reading times for thecritical region in Condition 1 (i.e., the condition where the functional head andits complement appeared in the same language) were significantly longer thanreading times for the same region in Condition 2 (i.e., the condition where thelanguage of the functional head and that of its complement were different).These findings corroborated the results of corpus-analysis, replicating in readingtimes the preference patterns found in corpora frequencies: codeswitched

PAOLA E. DUSSIAS30

constituents in which functional elements do not participate in thecodeswitching process seem to be preferred over constituents in whichfunctional elements undergo codeswitching. The convergence of data fromthese disparate sources lends support to the hypothesis that corpora andcomprehension data are somehow interdependent. This, of course, raises thequestion of how the production and comprehension systems are inter-connected. One possibility, discussed in Dussias (2001), is to suggest that theproduction mechanism is constrained in a principled manner but thecomprehension system is sensitive predominantly to information aboutstatistical frequency. Although this seems a viable possibility, one limitation ofthe comprehension studies referred to above is that production data for thegroup of bilinguals that participated in them was not available. Hence,generalizations regarding the interaction between the comprehension andproduction systems in bilinguals are, at best, tentative. To examine thisinteraction in finer detail, the present study collected experimental productiondata from Spanish-English bilingual speakers who belonged to the same speechcommunity as the bilinguals that participated in Dussias (1999). For comparisonpurposes, the two syntactic sites under investigation are, once again, Comp-IPand Det-NP.

2. The present studyThe purpose of the present study is to investigate whether the

comprehension preferences found in Dussias (1999) can be replicated inproduction data.

2.1 SubjectsThirty-four subjects participated in this study. The subjects completed a

language background survey designed to tap into several aspects of languageproficiency and use by self-report. Although only approximately half of thesubjects that participated in Dussias (1999) were available to participate in thisstudy, the language survey revealed that all subjects belonged to the samespeech community and had very similar language histories. The subjects werefluent Spanish-English bilinguals, had learned Spanish before the age of five,and English before seven years of age. Subjects reported using both languagesin their daily lives with family and friends, and in a variety of contexts,including academic and non-academic. All subjects indicated that Spanish-English codeswitching was part of their daily linguistic behavior.

When asked about language dominance, 24% of the subjects indicated thatthey were equally proficient in both languages, 73% were English dominant and3% were Spanish dominant.


2.2 Materials and designForty experimental items representing two conditions were used in this

experiment. Condition 1 was designed to elicit the production of a determinerand Condition 2 to elicit the production of a complementizer. Items in eachcondition consisted of two phrases--one in English and one in Spanish--andeach phrase was between three and seven words long. Given that determinerscan appear in several positions within a sentence (i.e., subject noun phrase,object noun phrase, prepositional phrase, etc.), the stimuli for Condition 1 wereconstructed so that the target determiner always occupied the head position ofan object noun phrase.

To ensure that order of presentation would not bias the results (i.e., if asentence began in English, subjects would be more likely to produce an Englishdeterminer), order of presentation was counterbalanced so that for half of theitems, the Spanish phrase appeared before the English phrase, and for the otherhalf the English phrase appeared before the Spanish phrase. This yielded theresulting experimental items outlined in Table 1.

An additional set of 120 filler items was added to the experimental stimulusto serve as distracters. These sentences were similar in structure and length tothe experimental sentences but required subjects to produce prepositions,question words, adverbs, quantifiers, and the copula ‘be.’

2.3 ProcedureThe instrument used for data collection was an elicited oral production task.

Subjects were seated in front of a computer and, at the sign of a prompt, twophrases appeared on the screen, one below the other. Subjects were instructed toread both phrases aloud and to produce a complete sentence by combining thetwo phrases using only one word.3 Subjects were told that the word could beeither in English or Spanish, depending on their preference, and that there wasno correct answer. For example, for the phrase ‘La enfermera dijo_____thepatient didn’t want to eat,’ subjects were expected to produce either ‘Laenfermera dijo que the patient didn’t want to eat’ or ‘La enfermera dijo that thepatient didn’t want to eat.’ For half of the sentences, the ending phrase wasdisplayed on the screen prior to the beginning phrase, and for the other half theorder was reversed.

Each time a sentence was presented to a subject, it was pseudo-randomlyscrambled. This involved assigning an equal number of experimental and filler

3 Subjects were not given explicit instructions to use a complementizer or a determiner whenconstructing the sentences to ensure that the language samples produced were as natural aspossible.

PAOLA E. DUSSIAS32

sentences to a number of different blocks, with the result that the items werepresented in a different order to each subject, yet the items in each conditionwere evenly distributed throughout the duration of the experiment. Finally, carewas taken so that not more than two sentences representing the same conditionwere displayed sequentially on the computer screen and that six filler sentenceswere displayed between conditions.

The data was recorded in a sound-proof facility using a SONY digitalrecorder._____________________________________________________________Condition 1

Spanish-to-English sample itemEl estudiante se olvidó history book in the library.The student forgot history book in the library.“The student forgot history book in the library.”

English-to-Spanish sample itemThe boy broke plato que estaba en la mesa.The boy broke plate that was on the table.“The boy broke plate that was on the table.”

_____________________________________________________________Condition 2

Spanish-to-English sample itemLa enfermera dijo the patient didn’t want to eat.The nurse said the patient didn’t want to eat.“The nurse said the patient didn’t want to eat.”

English-to-Spanish sample itemMy sister thought su amiga estaba en la universidad.My sister thought her friend was at the university.“My sister thought her friend was at the university.”

_____________________________________________________________Table 1: Experimental design and item sample

2.4 Analysis and resultsIn order to determine the frequency with which subjects produced Spanish

and English functional elements, the percentage of determiners and comple-mentizers produced in both languages was calculated. The results for thedeterminer are given in Table 2 and for the complementizer in Table 3. Table 2shows that when the stimulus began in Spanish, subjects were more or lessequally likely to produce a Spanish or an English determiner.


Determiner

Spanish English

Spanish 56% 44%Sentence

beginning in English 86% 14%

Table 2: Percentage of determiners produced in Spanish and English

Complementizer

Spanish English

Spanish 84% 16%Sentence

beginning in English 82% 18%

Table 3: Percentage of complementizers produced in Spanish and English

The results further reveal that when the stimulus began in English, subjectsproduced significantly more determiners in Spanish than in English (86% vs.16%, respectively). These findings alone indicate that the language of thephrase preceding the determiner did not influence the subjects’ choice oflanguage at the time of production. If that had been the case, we would haveexpected to see a greater occurrence of, say, Spanish determiners when thestimulus began in Spanish.

Turning now to Table 3, we see that when the items began in Spanish,subjects produced the Spanish complementizer ‘que’ significantly more timesthan the English complementizer. Interestingly, the same preference wasobserved when the stimulus began in English, where subjects produced theSpanish complementizer 82% vs. 18% for the English complementizer.

What we see from the findings reported here is that during languageproduction, some linguistic environments resist codeswitching more thanothers. Thus, for example, codeswitches where both Comp and IP appear inEnglish preceded by a Spanish verb (e.g., La enfermera dijo that the patientdidn’t want to eat) are produced significantly less times than codeswitcheswhere a Spanish Comp is preceded by a Spanish verb and followed by anEnglish IP (e.g., La enfermera dijo que the patient didn’t want to eat). This mayvery well reflect the strong bond that exists between verbs and the

PAOLA E. DUSSIAS34

complementizers they subcategorize for.4 The 82% production preferenceobserved for Spanish complementizers preceded by English verbs and followedby Spanish IP complements (as in ‘My sister thought que su amiga estaba en launiversidad’) could simply indicate that the bond between a verb and acomplementizer in English is somewhat weaker, thereby making codeswitchingat this site more permissible. This intuition is partially supported by the fact thatovert English complementizers are optional in the constructions underinvestigation.

In concluding, I return to the question of the nature of the relationshipbetween the comprehension and production systems. One possibility, suggestedabove, is that linguistic, psycholinguistic and discourse principles may underliethe frequency patterns observable in spontaneous discourse for codeswitchedutterances, and that the comprehension of these utterances takes place to a greatdegree independent of these principles. In this case, then, factors such asfrequency of occurrence in the production may be said to be at least partiallyresponsible for preference patterns observed in comprehension. The resultspresented here partly support this hypothesis. We saw, for example, thatSpanish complementizers were produced significantly more times than Englishcomplementizers when preceded by Spanish verbs. This was precisely thepattern of preference observed in the comprehension data discussed above:subjects’ eye fixations were shorter when reading sentences with a Spanishcomplementizer preceded by a Spanish verb than when reading constructionswith an English complementizer preceded by a Spanish verb. It is clear,however, that this account does not do an adequate job of explaining the resultsobtained for Spanish determiners. That is, although codeswitches between aSpanish determiner and an English noun are easier to understand (as measuredby eye-movement fixations), there appears to be no clear preference for onecodeswitch type over the other when it comes to production. A comprehensiveaccount of the variables that may determine comprehension preferences incodeswitching remains to be investigated through future research. There are,however, a number of accounts that have been put forth to account forproduction preferences. For one, the source of the preference found inproduction data could be partly psycholinguistic, and partially rooted in theautomaticity that characterizes function words. Because function words are lessvulnerable to lexical access effects than content words, they are less likely to

4 One may assume that this relationship manifests itself at different levels of linguisticrepresentation. Thus, for example, one could expect to find phonological evidence (i.e.,prosodic patterns) that is consistent with a view that the verb and its complementizer behave asone unit. This is the focus of Dussias (2001).


participate in codeswitching. In addition, function words are not accessedindependently of syntactic information, are less often preceded by pauses thancontent words, are more predictable than content words, and do not count asseparate elements in motor planning units. Combined, these traits makefunction words less prone to undergo codeswitching (Muysken 1997).

An additional account put forth in Muysken (1997) deals with the lowdegree of ‘insertability’ of functional elements vis-à-vis lexical elements into asyntactic frame. The argument goes something like this. A content word, say, aSpanish noun, would be easily insertable in an English frame because both theSpanish noun and the English noun it would replace instantiate the category N.In addition, since it is relatively easy to find a match between two content wordsin different languages (e.g., Spanish casa and English house), nouns are insertedquite effortlessly. Function words, on the other hand, are not easily definablebecause they are made up of highly specific features. Hence, finding a matchbetween two function words is, presumably, more difficult. These factors resultin a decrease in the involvement of function words in codeswitching.

A final account is provided by Myers-Scotton’s Matrix Language FrameModel (1993, 1997; see also Myers-Scotton 1995). The model, which ispsycholinguistically based, tries to account for codeswitching behavior byrelating it to models of monolingual speech production. A key feature of themodel is the distinction between Matrix Language and Embedded Language.The Matrix Language is defined as “the main language in codeswitchingutterances…[It is the language that specifies] the morpheme order and suppliesthe syntactically relevant morphemes in constituents consisting of morphemesfrom both participating languages” (1993:3). The Embedded Language, on theother hand, refers to the other language that participates in codeswitching. Itsrole is less prominent than that of the Matrix Language, since, for the most part,it provides the content morphemes in codeswitched constituents. Drawing fromthe differential behavior of closed-class and open-class morphemes inmonolingual speech production data, Myers-Scotton (1993, 1997) proposes afurther distinction; that between ‘system morphemes’ and ‘content morphemes’.The distinction is similar, but not identical, to the one made between closed-class items (i.e., determiner, complementizer, inflection, etc.) and open-classitems (i.e., content words) in monolingual speech production studies. Myers-Scotton proposes that during codeswitched speech ‘system morphemes’ areexpected to participate in codeswitching in a different way from ‘contentmorphemes’. This is so because ‘system morphemes’ are selected following themorphosyntactic specifications set by the Matrix Language, whereas ‘contentmorphemes’ are supplied by the Embedded Language. The Matrix LanguageFrame Model provides an explanation for why, in codeswitching utterances

PAOLA E. DUSSIAS36

involving Det + Noun ((1) through (3) above) and Comp + IP ((4)), the mostfrequent codeswitches are those where the determiner and the complementizerappear in Spanish and their corresponding complements (i.e., Noun and IP)appear in English. In (1) above, for example, since the determiner is accessedaccording to the morphosyntactic specifications of Spanish (the MatrixLanguage), the Spanish determiner ‘los’ is accessed and inserted into thesyntactic frame. English content morphemes that are congruent with themorphosyntactic specifications set by the Matrix Language are later inserted inthe codeswitched constituent (‘holidays’ in this case), giving rise to acodeswitched constituent that has a Spanish system morpheme and an Englishcontent morpheme. This type of codeswitched constituency is, according toMyers-Scotton (personal communication), the optimal type of constituent froma psycholinguistic perspective because, presumably, it is produced faster andwith less effort than other types of codeswitches (i.e., whole noun phrasecodeswitches).

In closing, we would like to note that we made the accidental discovery thatcertain types of determiners may be more prone to participating incodeswitching than others. In our data, if subjects produced a possessivedeterminer for sentences such as María compró____vestido de novia en NuevaYork’ (“Mary bought____wedding dress in New York”), it invariably appearedin Spanish. However, when subjects produced an article, it appeared in Spanishat approximately the same rate as it appeared in English. Because the number ofsentences of this type was small, no definite statements can be made. Ifsubstantiated with additional data, this finding may suggest that semanticfactors are also at play when dealing with codeswitching at the level of speechproduction. At present, we are collecting additional data to corroborate thishypothesis.

REFERENCES

Cowper, E.A. 1992. A Concise Introduction to Syntactic Theory. Chicago, Ill.:University of Chicago Press.

DiSciullo, Anne-Marie, Pieter Muysken & Rajendra Singh. 1986. “Governmentand code-mixing”. Journal of Linguistics 22.1-24.

Dussias, Paola E. 1997. “Sentence matching and the functional head constraintin Spanish/English codeswitching”. Spanish Applied Linguistics 1.114-150.


---------. 1999. “The function-word effect in Spanish-English codeswitching: Alook at eye-movements”. Paper presented at the Second InternationalSymposium on Bilingualism, held in Newcastle upon Tyne, U.K., April 1999.

---------. 2001. “Psycholinguistic complexity in codeswitching”. TheInternational Journal of Bilingualism: Cross-Disciplinary, Cross-LinguisticStudies of Language Behavior 5.87-100.

---------. in progress. Prosodic patterns in Spanish complement clauses. Ms.,Pennsylvania State University.

Halmari, Helena. 1997. Government and codeswitching: Explaining AmericanFinish. Amsterdam & Philadelphia: John Benjamins.

Klavans, Judith E. 1985. “The syntax of code-switching: Spanish and English”.Selected Papers from the 13th Linguistic Symposium on Romance Languagesed. by L. D. King & C. A. Matey, 213-231. Amsterdam & Philadelphia: JohnBenjamins.

Lance, David. 1975. “Spanish-English code-switching”. El lenguaje de loschicanos ed. by E. Hernández-Chávez, A. Cohen & A. F. Beltrano, 138-153.Arlington, Va.: Center for Applied Linguistics.

Lipski, John. M. 1985. “Linguistic aspects of Spanish-English languageswitching”. Series: Special Studies 25. Tempe: Arizona State University.

Milian, Sarah. 1996. “Case assignment in Spanish/English Codeswitching”.Paper presented at the Symposium on Codeswitching, Linguistic Society ofAmerica, annual meeting, held in San Diego, California, January 1996.

Muysken, Pieter. 1997. “Code-mixing: Constraints and strategies”. Paperpresented at the First International Symposium on Bilingualism, held inNewcastle upon Tyne, U.K., April 1997.

Myers-Scotton, Carol. 1993. Duelling languages: grammatical structure incodeswitching. Oxford: Oxford University Press.

---------. 1995. “A lexically based model of code-switching”. One speaker, twolanguages: Cross-disciplinary perspectives on code-switching ed. by LesleyMilroy & Pieter Muysken, 233-256. Cambridge: Cambridge University Press.

---------. 1997. Duelling languages: grammatical structure in codeswitching(2nd ed.). Oxford: Oxford University Press.

--------- & Janice L. Jake. 1997. “Codeswitching as evidence for two types offunctional elements”. Paper presented at the First International Symposiumon Bilingualism, held in Newcastle upon Tyne, U. K., April 1997.

Park, J., Rudolf Troike & M. Park. 1993. “Constraints in Korean-English Code-Switching: A Preliminary Study”. Journal of the Applied LinguisticAssociation of Korea 6.115-133.

PAOLA E. DUSSIAS38

Poplack, Shana. 1980. “Sometimes I’ll start a sentence in Spanish Y TERMINOEN ESPAÑOL: toward a typology of code-switching”. Linguistics 18.581-618. Also in Spanish in the United States, sociolinguistics aspects ed. by J.Amastae & L. Elias-Olivares. 1982, 230-263. Cambridge: CambridgeUniversity Press.

Sankoff, David & Shana Poplack. 1981. “A formal grammar of code-switching”. Papers in Linguistics 14.3-43.

Treffers-Daller, Jenine. 1995. “Code-switching between a case assigner and itscomplementer: The case of Turkish-German prepositional phrases”. SummerSchool codeswitching and language contact, 14-17 Sept. 1994.Ljouwert/Leeuwarden, The Netherlands: Fryske Akademy.

FOCUS, WORD ORDER VARIATION AND INTONATIONIN SPANISH AND ENGLISH

AN OT ACCOUNT∗∗∗∗

RODRIGO GUTIÉRREZ-BRAVOUniversity of California, Santa Cruz

0. IntroductionFocus-related word order variation is a well-attested phenomenon observed

in a number of different languages. In Italian and Spanish, for example, we seecases of subject inversion like (1) and (2), where the postverbal (i.e.,‘inverted’) subject DP is interpreted as a focus.

(1) ItalianHa gridato Gianni.has screamed John“JOHN has screamed.” (Grimshaw & Samek-Lodovici 1998)

(2) SpanishAyer compró el periódico Juan.yesterday bought the newspaper Juan“JOHN bought the newspaper yesterday.”

Many recent analyses connect this word order variation in Spanish andother Romance languages with some structural condition that requires afocused constituent to occupy a specific syntactic position. Belletti & Shlonsky

∗ I would like to thank Judith Aissen, Daniel Büring, Kazutaka Kurisu, Brian Lindsey, LineMikkelsen, and two anonymous reviewers for helpful comments and discussion of the materialpresented here. I would also like to thank the participants at the Stanford-UCSC OptimalTypology Workshop (10/30/99) and the participants at the UCSC Research Seminar (Winter2000) for their feedback. Also, many thanks to Silvia Bravo, Elisa and Mauricio Gutiérrez,Gilberto and Sofía Jiménez, and Rebeca Mejía for their judgments on some of the Spanishexamples presented in this paper. All errors that remain are my own. This work was supportedin part by the NSF grant SBR-9818177 and by the National Council of Science andTechnology of Mexico (CONACYT), scholarship No. 117325.

RODRIGO GUTIÉRREZ-BRAVO40

(1995) for example, suggest the existence of a Focus Phrase that in languageslike Italian projects its specifier to the right as schematized in (3). In thisanalysis, the focused subject moves to the specifier of FocP, thus deriving thesubject-final order in (1).

(3) AgrOP

AgrO FocP

Foc’ Spec

Foc VP2

On the other hand, in Optimality-Theoretic analyses, subject inversion isaccounted for by suggesting the existence of a constraint ALIGN-FOCUS

(Grimshaw & Samek-Lodovici 1998, Samek-Lodovici 1997) which requiresfoci to appear as the rightmost constituent of the clause.

(4) ALIGN-FOCUS: Align the left edge of focus constituents with the rightedge of a maximal projection. (Grimshaw & Samek-Lodovici 1998)

In languages which show focus-related subject inversion, this constraint isranked above the constraint that requires subject DPs to occupy the specifier ofIP (typically, the SUBJECT constraint of Grimshaw 1997).1

Input: <gridare (x), x=focus, x=Gianni> ALIGN-FOCUS SUBJECT

! a. ha gridato Gianni *

b. Gianni ha gridato *!

Tableau 1: ALIGN-FOCUS and SUBJECT

All these structural analyses, however, overlook the fact that a focusedsubject in the Spanish VOS example (2) and a focused subject in thecorresponding English SVO translation have a property in common despite thedifference in word order, namely, that in both cases the subject DP is the phraseto which sentential stress is assigned, as noted by Jackendoff (1972), Contreras

1 But see Samek-Lodovici (1997) for an alternative formulation.

FOCUS, WORD ORDER VARIATION AND INTONATION 41

(1976), Selkirk (1995), Zubizarreta (1998) to quote just a few references.2Once this property is acknowledged, the alternative to the structural accountsoutlined above are analyses where focus-related word order variation isprosodically motivated.

Zubizarreta (1998) develops a derivational analysis along these lines.Adopting the assumption that constituents which function as foci need to bemarked with sentential stress (Jackendoff 1972, Selkirk 1995, and manyothers) and by redefining the Nuclear Stress Rule (NSR) of Chomsky & Halle(1968), Zubizarreta suggests that the VOS order of (2) is the result of leftwardscrambling the VP that follows the subject DP before SpellOut. As shown in(6), the result of this operation is that the focused subject DP ends up being therightmost constituent of the sentence and thus receives sentential stress.

(5) Nuclear Stress RuleThe rightmost word-level stress of a phrase carries the main stress ofthe phrase. (Chomsky & Halle 1968)

(6)TP

T VP1

comprói VP2 k VP1 bought

el periódico DP V1’the newspaper

[JUAN]Foc ti tkJuan

The problem in this case is that as soon as the NSR is introduced into theanalysis, considerable extra machinery is required to account for the fact that inthe English equivalent of (2) and (6), sentential stress is not assigned to the

2 Costa (1996) also makes use of ALIGN-FOCUS, but differs from the works above in that itmakes this constraint follow from the need for foci to receive sentential stress. In this account,however, it still goes unexplained why focused preverbal subjects in English are stressed. I amgrateful to an anonymous reviewer for bringing this point to my attention.


rightmost constituent of the sentence, but rather to the subject DP JOHNin-situ, as shown in (7).3

(7) JOHN bought the newspaper yesterday.

In this respect, the difficulties entailed by analyses of focus based on theNSR have lead many researchers to abandon it altogether (see for exampleSchwarzschild 1999). Selkirk (1995), however, notes in relation to the NSR thatthe principles governing intonation and focus in English show the typicalpattern of constraint interaction of Optimality Theory. Taking Selkirk’sobservation as a starting point, in what follows I will suggest an Optimality-theoretic account of intonation and the distribution of focus in English andSpanish, which I will argue has the advantage that it provides an explanation ofthe relevant facts through the different ranking of the same constraints. As aresult, I suggest that there is no need for projections like the Focus Phrase ofBelletti & Shlonsky (1995), language-specific intonational mechanisms thatguarantee that the NSR is satisfied, or any specific syntactic position associatedwith focus altogether.

1. Focus and intonationAs a first step in the analysis, I will lay out my assumptions on focus and

its relation to intonation, along with the constraints that govern their behavior.First, I assume that the syllable associated with the sentential stress is moreprominent than any other syllable in the sentence. In essence, this amounts toassuming the Pitch Accent Prominence Rule of Selkirk (1995), which statesthat a syllable associated to a pitch accent has greater stress prominence than asyllable which is not associated to a pitch accent. Consequently, for the focusto be intonationally the most prominent constituent in a sentence, it must bearthe pitch accent that corresponds to the sentential stress. With respect to thelocation of the sentential stress in both Spanish and English, I will assume anNSR constraint, in its informal formulation by Selkirk (1995):

(8) NSRThe most prominent syllable of the rightmost constituent of a phraseP is the most prominent syllable of P.

3 Zubizarreta (1998), for example suggests different NSRs for Spanish and English and thatdeaccented constituents are invisible to the NSR in English. I remit the reader to Zubizarreta’swork for details on her specific formalization of the different NSRs.


This NSR constraint will be violated when any constituent other than thesentence-final constituent bears the sentential stress. Secondly, thecharacterization of focus I will assume throughout this paper is as follows.Throughout the analysis I will deal only with presentational focus, which willbe characterized, following Halliday (1967), as the constituent in the answer toa wh-question that corresponds to the wh-operator in the question. This isexemplified in (9). I crucially distinguish it from correction focus, exemplifiedfor English in (10), which is not necessarily associated with a wh-question andwhich in Spanish has a different distribution and different intonationalproperties than presentational foci (see for example Fant 1984,Zubizarreta 1998, inter alia).

(9) Presentational focusa. Who screamed?b. JOHN screamed.

(10) Correction focusa. John’s mother voted for Bill.b. No, she voted for JOHN. (Schwarzschild 1999)

Specifically, following the sources quoted above, I assume that SpanishSVO cases like (11a), where the subject DP receives the extra-high accent ofcorrection focus4 in-situ, are felicitous answers to a question like (11b), wherethe answer is overtly implied by means of a tag or understood to be implied bythe hearer. As such, I take them to correspond to cases of correction focus, andthey will not be further dealt with in this paper.

(11) a. MARÍA me regaló la botella de vino. María to-me gave the bottle of wine

“MARÍA gave me the bottle of wine.”b. Quién te regaló la botella de vino? (Juan, verdad)?

who to-you gave the bottle of wine Juan right “Who gave you the bottle of wine?(Juan, right?)”

I further assume that constituents that are part of the focus are F-marked(Jackendoff 1972, Selkirk 1995), and concretely, I assume the proposal inSchwarzschild (1999) that F-marking is free (i.e., there is no focus projection).Instead, F-marking is conditioned by the following constraint:

4 This accent is represented in boldface in (11a). See the references above for details on thesyntactic and intonational differences between these two kinds of foci.


(12) GIVENnessA constituent that is not F-marked is GIVEN.Violated when a constituent that is not GIVEN is not F-marked.5

For the purposes of this analysis, it will be enough to note that a nominalexpression x counts as GIVEN if there is a nominal expression y in thepreceding wh-question with which x is coreferential. When this condition is notmet, x must be F-marked.6 More formally, GIVENness can be defined as<Not(GIVEN(x)) → F-marked(x)>. For example, in a question-answer pair, aDP that corresponds to the wh-operator in the question is not GIVEN, since thereis no nominal expression in the question that is coreferential with it.Consequently, John is not GIVEN in (13b), so it must be F-marked to avoid aviolation of GIVENness. In cases of sentence focus, I assume that everyconstituent in the answer needs to be F-marked.

(13) a. Who read the book?b. [JOHN]F read the book.

The other of Schwarzschild’s constraints that I will adopt here is the FOCconstraint, which requires that a Foc-marked phrase contain an accent (where aFoc-marked phrase is an F-marked phrase not immediately dominated byanother F-marked phrase, i.e., the focus of the sentence). Schwarzschild’s FOCconstraint is defined with respect to accents and not specifically with respect tosentential stress. This is necessary in an analysis like Schwarzschild’s, whichincorporates cases of multiple foci. Here, however, I will only consider caseswith a single focus, so for expository purposes I redefine FOC asin (14). Accordingly, FOC will be violated when the focus does not receive theaccent that corresponds to the sentential stress.

(14) FOCA Foc-marked phrase is signaled with sentential stress.

2. EnglishIn providing an analysis of the cross-linguistic differences between Spanish

and English with respect to focus and intonation, it is illustrative to first look atthe syntax of English. This will set up the necessary infrastructure that willlater be used in the analysis of Spanish.

5 See Schwarzschild (1999) for a full definition GIVENness. See also Choi (1999).6 For clarity, I will include the relevant wh-question above the corresponding tableaux.


As a first step, consider briefly the surface distribution of subjects inEnglish. In English, all else being equal, subject DPs occupy the specifier of IPat S-structure. Assuming the VP-internal subject hypothesis, this implies thatthe subject DP moves from Spec-VP to Spec-IP:

(15) a. [IP Johni will [VP ti buy the newspaper]]. b. *[IP will [VP John buy the newspaper]].

Grimshaw (1997) provides an OT account of this fact that follows from therelative ranking of the constraint STAY, which penalizes movement operations,and the constraint SUBJECT which requires the Spec of IP to be filled.Concretely, English is a language where SUBJECT outranks STAY. This isexemplified in Tableau 2, where for expository purposes I further assume thatinputs are D-structure representations.

(16) STAY

No traces.Violated once for every movement operation.

(17) SUBJECT

The highest A-specifier in an extended projection must be filled.Failed by a clause without a subject in the canonical position.

[IP will [VP John buy the newspaper]] SUBJ STAY

! a.[IP Johni will [VP ti buy the newspaper]]. *

b.[IP will [VP John buy the newspaper]] *!

Tableau 2: Canonical subjects in English (Grimshaw 1997)

This is the only part of the syntax of English that will be relevant for theanalysis. At this point, we can turn to the constraints related to discourse andintonational phenomena. Consider now the interaction between intonationaland discourse constraints. In a case of subject focus like (18b), the focused DPcontains the most prominent syllable of the sentence. This shows that inEnglish the FOC constraint outranks the NSR, since it is preferable not to haverightmost intonational prominence than to leave a focused constituent withoutsentential stress. The analysis is presented in Tableau 3.


(18) a. Who bought a book about bats? b. [MARY]Foc bought a book about bats.

Antecedent: Who bought a book about bats? [IP [VP Mary bought a book about bats]] GIVEN FOC NSR

! a. [MARY]Foc bought a book about bats. *

b. [Mary]Foc bought a book about BATS. *!

c. Mary bought a book about BATS *!

Tableau 3: English subject focus

Consider candidate (b) in Tableau 3. This candidate assigns sententialstress to the rightmost constituent of the sentence, and so it satisfies the NSR.However, as a result the focused subject does not receive sentential stress, soFOC is violated. Thus candidate (b) loses to candidate (a), which insteadsatisfies FOC but violates the NSR by assigning sentential stress to the subjectin-situ. Notice that under the opposite ranking, NSR >> FOC, candidate (b)would wrongly emerge as the winner. Consider now candidate (c). SinceF-marking is unrestricted, GEN can generate this candidate, where there is nofocus altogether. FOC is thus vacuously satisfied and the NSR is satisfied, sincesentential stress falls on the rightmost constituent, but this candidate fatallyviolates GIVENness, because the subject DP Mary is not coreferential with anyDP in the wh-question and is therefore not GIVEN. In this respect, the analysisprovides a mechanism to make sure that discourse status of a constituent isrespected, without having to specify focus in the input.

At this point, we can bring all this together and start to relate it to wordorder. English does not resort to movement to resolve a potential conflictbetween the requirements of the FOC and NSR constraints, as shown in (19).

(19) a. Who bought a book about bats? b. * [VP [VP ti bought a book about bats] MARYi].

Example (19b) corresponds to candidate (b) in Tableau 4 and represents thepossibility of satisfying both the FOC and the NSR constraints through right-adjunction of the focused subject to VP. Following the analysis of Grimshaw &Samek-Lodovici (1998), in English this possibility is ruled out by the higherranking of the SUBJECT constraint. But in contrast with their analysis, in the


ranking in Tableau 4, SUBJECT does not outrank an ALIGN-FOCUS constraint butrather the independently motivated NSR.7

Antecedent: Who bought a book about bats? [IP [VP Mary bought a book about bats]] SUBJ FOC NSR

! a. IP[[MARY]Foc [ bought a book about bats]] *

b. IP[ __ [[bought a book about bats] [MARY]Foc ]] *!

Tableau 4: English; SUBJECT >> NSR

Crucially, even though the NSR appears at the very bottom of the ranking inEnglish, we still expect to see its effects when satisfaction of the constraintsabove it is not at stake. This is exactly the case observed in cases of sentencefocus, whose analysis is presented in Tableau 5.

Antecedent: What’s been happening?[IP [VP Mary bought a book about bats]] GIVEN SUBJ FOC NSR

! a. IP[Maryi [ ti bought a book about BATS]]Foc

b. IP[MARYi [ ti bought a book about bats]]Foc *! c. IP[ __ [[ti bought a book about bats] [MARYi]]]Foc *!

Tableau 5: English sentence focus

In cases of sentence focus, none of the constituents of the answer are givenand so the IP (or the CP) as a whole is Foc-marked. As a result, in principle thesentential stress could fall anywhere in the sentence and still satisfy the FOCconstraint. But here is where the NSR comes into play and decides the outcome.In Tableau 5, the winning candidate is candidate (a), which satisfies the NSR byassigning the sentential stress to the rightmost constituent of the sentence.Nothing else is at stake, since candidate (a) satisfies all the other relevantconstraints. Contrast this with candidate (b), which corresponds to aninfelicitous answer to the wh-question under consideration. This candidate alsosatisfies all the relevant discourse and syntactic constraints, but is ruled out as 7 From here onwards, only candidates that satisfy the undominated GIVENness constraint willbe included in the tableaux.


a result of its violation of the NSR. Just as in Tableau 4, the right-adjunctioncandidate (c) in Tableau 5 is ruled out because of its violation of SUBJECT.

3. SpanishI will now extend this proposal to Spanish, where I will develop an analysis

in which Spanish differs from English in that Spanish prioritizes thesatisfaction of intonational requirements over the satisfaction of syntacticrequirements. Drawing on the OT proposal developed so far, this difference canbe captured through a different ranking of the same constraints used in theanalysis of English. As is well known, Spanish shows three different orders ofargument DPs in relation to the verb when there is no Clitic Left-Dislocation ofthe direct object: SVO, VSO and VOS, exemplified below:

(20) a. Ayer Juan compró el periódico.yesterday Juan bought the newspaper (SVO)

b. Ayer compró Juan el periódico.yesterday bought Juan the newspaper (VSO)“John bought the newspaper yesterday.”

c. Ayer compró el periódico [Juan]Foc.yesterday bought the newspaper Juan (VOS)“JOHN bought the newspaper yesterday.”

The basic assumptions I will adopt about the structure of these examplesare laid out in (21), all essentially following Suñer (1994).

(21) a. Spanish is underlyingly SVO, with the subject generatedin Spec-VP

b. V raises to I overtly.8c. The VSO order is derived by movement of V to I, where

the subject stays in Spec-VP. d. The SVO order is derived by movement of the subject to

Spec-IP.e. The VOS order is derived by right-adjoining the subject

DP to VP.

It is also a well-known fact that the VOS order in (20c) is different from theother two in that it is compatible only with an interpretation in which the

8 In the tableaux that follow, violations of STAY resulting from V-to-I movement are notincluded, since they play no role in deciding the output.


subject is focused (Contreras 1976, Zubizarreta 1998). To provide an analysisof these different word orders, recall now that Grimshaw (1997) suggests forEnglish an account in which the obligatory movement of the subject DP toSpec-IP was explained through the interaction of the constraints SUBJECT andSTAY, with the former dominating the latter. This prevents a subject DP fromstaying in its VP-internal position in English. However, under the assumptionsin (21), the Spanish VSO example (20b) is precisely that case where the subjectstays in its VP internal position. The point is that the SVO output satisfiesSUBJECT but violates STAY, whereas the VSO output satisfies STAY but violatesSUBJECT. Consequently, in principle it may be possible to derive both of theseword orders by means of a tie between these two constraints. What isinteresting is that the right-adjunction VOS candidate violates both constraintsand thus would not emerge as a winner even under a constraint tie analysis.This is shown in Tableau 6.

[IP [VP Juan compró el periódico]] SUBJ STAY

! a. [IP Juani compró [VP ti el periódico ]]. SVO *

! b. [IP compró [VP Juan el periódico]] VSO *

c. [IP compró [VP[VP ti el periódico ] Juani]] VOS * *

Tableau 6: SUBJECT and STAY in Spanish

Notice that the same result is achieved if the VOS candidate is analyzed asthe result of scrambling of the direct object (Ordóñez 1998). In this case,scrambling of the direct object results in a violation of STAY, and SUBJECT is alsoviolated, since Spec-IP is left empty.9 Yet even though the VOS order is the lessharmonious of the candidates in Tableau 6, in the dialects of Spanish studiedin Zubizarreta (1998:125-126), it is the only felicitous full-sentence answer tothe question in (22).

9 It is important to note that the constraint tie in Tableau 6 does not imply that SVO and VSOwill have the exact same distribution. Other discourse constraints may eventually favor onecandidate over the other. This is presumably the case in those Spanish varieties, pointed out tome by an anonymous reviewer, where SVO sentences require a topic subject. What theconstraint tie implies is that, all else being equal, SVO and VSO will freely alternate in cases ofsentence focus. As this reviewer points out, this is the case in some varieties of Spanish, andsome of the speakers consulted here share this judgement, though not all of them. Furtherresearch is certainly necessary to identify all the variables at play in these cases.


(22) Quién te regaló la botella de vino? who to-you gave the bottle of wine “Who gave you the bottle of wine?”(23) a. # María me regaló la botella de VINO.

María to-me gave the bottle of wine (SVO)b. # Me regaló María la botella de VINO.

to-me gave María the bottle of wine (VSO)c. Me regaló la botella de vino MARÍA.

to-me gave the bottle of wine María (VOS)“MARÍA gave me the bottle of wine.”

Essentially the same pattern of subject inversion is observed in othervarieties of Spanish, although some differences must be noted. MexicanSpanish, for example, shows OVS and not VOS constructions as the felicitousanswer in this context, as shown in (24)-(25).

(24) Mexican Spanish Quién compró los discos? who bought the records ‘Who bought the records?’(25) a. #Una muchacha compró los DISCOS.

a girl bought the records (SVO)b. #Compró una muchacha los DISCOS.

bought a girl the records (VSO)c. Los discos los compró una MUCHACHA.

the records DO-CL bought a girl (OVS)“A GIRL bought the records.”

The OVS constructions, however, require an analysis of Clitic

Left-Dislocation of their own, and so here I will concentrate on Zubizarreta’sdata only. Returning to (22)-(23), it is worth pointing out that from anOptimality-theoretic perspective, the state of affairs observed in Tableau 6 andthe examples in (22)-(23) is far from being a puzzle, and, in fact, it represents aprototypical OT situation: a candidate that, all else being equal, is predicted notto emerge as the winner actually surfaces as the output by virtue of satisfying ahigher ranked constraint that the other candidates violate. The relevant


constraints in this case are the NSR and FOC, in a ranking like the one I suggestin Tableau 7, where they both outrank SUBJECT and STAY.10

Antecedent: Who gave you the bottle of wine? [IP [VP María me regaló la botella de vino]]. María gave me the bottle of wine

FOC NSR SUBJ STAY

a. [IP [MARÍAi]Foc me regaló [VP ti la botella de vino]]. *! * b. [IP me regaló [VP [MARÍA]Foc la botella de vino]]. *! * c. [IP [Maríai]Foc me regaló [VP ti la botella de VINO]]. *! *

d. [IP me regaló [VP [María]Foc la botella de VINO]]. *! *

! e. [IP me regaló [VP[VP ti la botella de vino] [MARÍAi]Foc]]. * *

Tableau 7: Spanish; NSR >> SUBJECT

Let us now consider the candidates in Tableau 7 one by one. The SVO andVSO candidates (a) and (b), both infelicitous answers to the relevant wh-question, are ruled out because of their violation of the NSR; in the former casesentential stress is assigned to the subject in Spec-IP, and in the latter to thesubject in Spec-VP, but neither of these positions corresponds to the rightmostedge of the sentence. Candidates (c) and (d), which correspond to theinfelicitous examples (23a) and (23b), avoid this situation altogether byassigning sentential stress to their rightmost constituent. But in doing so theyincur a fatal violation of FOC, since the focused subjects do not receive thesentential stress. The VOS candidate (e) thus emerges as the winner; eventhough it fares worse than any of the other candidates with respect to bothSUBJECT and STAY, it is the only candidate that simultaneously satisfies the NSRand the FOC constraints.

The analysis in Tableau 7 is the core of this proposal, and there are tworelevant aspects related to it. The first one is that with a ranking in which theNSR dominates SUBJECT and STAY, we explain why Spanish, but not English,shows focus-related subject inversion. From this perspective, what makes

10 As in the case of English, I assume that GIVENness is undominated, so candidates that do notFoc-mark the subject are ruled out because of their violation of this constraint and are notincluded in Tableau 7.


English and Spanish different with respect to the phenomenon underconsideration is not the different ranking of focus-related constraints, as hasbeen suggested in previous OT analyses; in both languages, the constraintsgoverning focus are at the top of the ranking, and what makes them different isthe relative ranking of the NSR. The second important point is that this analysisprovides a direct explanation for the fact that the VOS order in Spanish is onlycompatible with an interpretation in which the subject is the focus. In thisparticular situation, the VOS candidate emerges as the output because itsimultaneously satisfies the NSR and FOC constraints, as opposed to the SVOand VSO candidates. But when no subject focus is required by the previousdiscourse, the VOS candidate will necessarily lose against either the SVO or theVSO candidates, since it violates both SUBJECT and STAY.

4. ConclusionsIn this paper I have provided an OT analysis of the cross-linguistic

differences between Spanish and English with respect to focus, intonation andword order. This analysis draws on several advantages of OT. First, byformulating the NSR as a violable constraint, I have suggested that there is noneed to abandon it as a tool in the analysis of sentential stress assignment. Inlanguages like English, where it is a low-ranking constraint, we expect itseffects to be overridden by other higher-ranking constraints. However, in thecases of sentence focus, where satisfaction of the constraints that dominate theNSR is not at stake, its effects are still observed. Secondly, an analysis ofFocus-related subject inversion in Spanish is arrived at through the language-specific ranking of the same constraints operational in English. Concretely, inSpanish the NSR dominates the SUBJECT constraint, whereas English has theopposite ranking. Unlike previous analyses, both derivational and Optimality-theoretic, this result is achieved without the need to appeal to a stipulativeALIGN-FOCUS constraint, a Focus Phrase, or language-specific intonationalproperties distinguishing both languages. Lastly, the analysis explains whyVOS sentences in Spanish necessarily have an interpretation in which thesubject is the focus; all else being equal, a VOS sentence will always be lessharmonious with respect to the SUBJECT and STAY constraints than the SVO andVSO candidates. Only when satisfaction of the NSR and FOC constraints is atstake can this candidate be more harmonious than the SVO and VSO candidatesand thus emerge as the winner.


REFERENCES

Belletti, Adriana & Ur Shlonsky. 1995. “The Order of Verbal Complements, aComparative Study”. Natural Language and Linguistic Theory 13.489-526.

Choi, Hye-Won. 1999. Optimizing Structures in Context: Scrambling andInformation Structure. Stanford, Calif.: CSLI Publications.

Chomsky, Noam & Morris Halle. 1968. The Sound Pattern of English.Cambridge, Mass.: MIT Press.

Contreras, Heles. 1976. A Theory of Word Order with Special Reference toSpanish. Amsterdam: North Holland.

Costa, João. 1996. “Word Order and Constraint Interaction”.Ms., HIL\Leiden University.

Fant, Lars. 1984. Estructura Informativa en Español: Estudio Sintáctico eInformativo. Ph.D. Dissertation, Uppsala University.

Grimshaw, Jane. 1997. “Projection, Heads and Optimality”. Linguistic Inquiry28.373-422.

Grimshaw, Jane & Vieri Samek-Lodovici. 1998. “Optimal Subjects andSubject Universals”. Is the Best Good Enough? ed. by P. Barbosa, D. Fox, P.Hagstrom, M. McGinnis & D. Pesetsky, 193-219. Cambridge, Mass.: MITPress.

Halliday, M.A.K. 1967. “Notes on Transitivity and Theme in English(Part 2)”. Journal of Linguistics 3.199-244.

Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar.Cambridge, Mass.: MIT Press.

Ordóñez, Francisco. 1998. “Post-verbal Asymmetries in Spanish”. NaturalLanguage and Linguistic Theory 16.313-346.

Samek-Lodovici, Vieri. 1997. “OT-interactions between Focus and CanonicalWord order: Deriving the Crosslinguistic Typology of Structural ContrastiveFocus”. Ms., Universität Konstanz.

Selkirk, Elisabeth. 1995. “Sentence Prosody: Intonation, Stress and Phrasing”.The Handbook of Phonological Theory ed. by J. Goldsmith, 550-569.Oxford: Blackwell.

Schwarzschild, Roger. 1999. “GIVENness, AvoidF and Other Constraints onthe Placement of Accent”. Natural Language Semantics 7.141-177.

Suñer, Margarita. 1994. “V-movement and the Licensing of Argumentalwh-phrases in Spanish”. Natural Language and Linguistic Theory12.335-372.

Zubizarreta, María L. 1998. Prosody, Focus and Word Order. Cambridge,Mass.: MIT Press.

MORPHOLOGICAL COMPLEXITYAND SPANISH OBJECT CLITIC VARIATION*

DAVID HEAPUniversity of Western Ontario

0. ProblemSpanish has two different types of third person object pronoun paradigms in

different varieties or dialects. Broadly speaking, we can call these two systems theetymological (or case-based) system and the referential system. The latter systemincludes the usages termed loísmo, leísmo, and laísmo in the Hispanic philologicaltradition. These two types of pronominal paradigms make differential use ofmorphological contrasts based on features such as [DATIVE], [FEMININE],[PLURAL], and in some referential varieties, the distinction between [COUNT] and[MASS] nouns. But an analysis which treats these as binary features and allows fortheir free combination in unordered matrices seriously overpredicts the numberof surface contrasts, whereas the attested pronoun paradigms are in fact highlyconstrained. This paper proposes a Feature Geometry account which allows forthe attested range of variation in pronoun paradigms without opening the door tounconstrained variation.

1. DataKlein-Adreu (1981) and Fernández-Ordóñez (1993) contrast the �etymo-

logical� (or case-based) pronoun system of third person object clitics shown in (1)with the �referential� system shown in (2). In cases where the referential usagediffers from the etymological usage, the traditional Hispanic dialectological termsleísmo, laísmo, loísmo are noted in (2) as well.

* Many thanks to audiences at the third Bilingual Workshop in Theoretical Linguistics held atQueen�s University (December 1999), and at the Section d�études en linguistique française,University of Toronto (June 2000), as well as to Jacques Lamarche, Barbara White and especiallyto Susana Béjar, for their comments and questions on previous versions of this work. Usualdisclaimers apply.

56 DAVID HEAP

(1) Etymological system of third person object clitics1

a. Lo conocí.him-CL met-I�I met him.�

b. La conocí.her-CL met-I�I met her.�

c. Le di un regalo.him-DAT-CL gave-I a gift.�I gave him a gift.�

d. Le di un regalo.her-DAT-CL gave-I a gift�I gave her a gift.�

e. Lo compramos.it-CL-MASC-ACC bought-we.�We bought it.�

f. Le cambiamos la tapa.it-CL-DAT changed-we the cover.�We changed the cover on it.�

g. El vino lo tomamos con las comidas.wine, it-CL-ACC drink-we with meals.�Wine we drink with meals.�

h. La cerveza la tomamos con las tapas.beer, it-CL-FEM-ACC drink-we with tapas�Beer we drink with tapas.�

i. Le añaden de todo hoy día.it-CL-DAT add-they everything today.AThey add everything to it nowadays.@

1 Plurals are omitted in both (1) and (2) for ease of exposition; in most cases, plurals can be formedby adding -s, except in the case of non-count nouns, which cannot be pluralized.

COMPLEXITY AND CLITIC VARIATION 57

(2) Referential system of third person object cliticsTraditional term

a. Le conocí.him-CL met-I�I met him.� (animate leísmo)

b. La conocí.her-CL met-I�I met her.�

c. Le di un regalo.him-DAT-CL gave-I a gift.�I gave him a gift.�

d. La di un regalo.her-FEM-CL gave-I a gift.�I gave her a gift.� (laísmo)

e. Le compramos.it-CL-MASC-ACC bought-we.�We bought it.� (inanimate leísmo)

f. Le cambiamos la tapa.it-CL-DAT changed-we the cover.�We changed the cover on it.�

g. El vino lo tomamos con las comidas.wine it-CL-ACC drink-we with meals�Wine we drink with meals.�

h. La cerveza lo tomamos con las tapas.beer it-CL-MASS drink-we with tapas.�Beer we drink with tapas.� (loísmo)

i. Lo añaden de todo hoy día.it-CL-MASS add-they everything today�They add everything to it nowadays.� (loísmo)

The etymological paradigm in (1) has gender syncretism for both animate andinanimate objects, since the masculine and feminine forms are identical in thedative. In the referential paradigm, animate objects distinguish only gender, at theexpense of case: there is one form for the masculine singular and another for thefeminine singular, regardless of whether they are accusative or dative. Withinanimate objects, the referential paradigm has total case and gender syncretism,with one form for all mass nouns and another for all count nouns, regardless ofcase and gender.

Of course, these are two somewhat idealized systems and the situation inreality is much more complex. Standard (Peninsular) Spanish usage corresponds

58 DAVID HEAP

to neither of these systems, but rather to a hybrid or compromise system:essentially, standard Spanish uses the etymological paradigm but with variationbetween the two systems with masculine animate objects, (1a). Animate leísmoas in (2a) is more common in Peninsular Spanish than in Latin America Spanish,while laísmo and loísmo only occur in Peninsular Spanish (Fernández-Ordóñez1994:73). Pure referential usage is relatively rare; most speakers vary, mixingreferential usage with more standard usage, and a number of transitional varietiesexist where one type of usage blends into another (Fernández-Ordóñez 1994).Both loísmo (2h) and laísmo (2d) are often socially stigmatised, as is inanimateleísmo (2e).

2. Binary phi-features?If we attempt to represent the oppositions in (1) using binary phi-features such

as [+ DATIVE] [+ PLURAL] and [+ FEMININE], we obtain a paradigm such as the onein Table 1, for the etymological system:

Feminine Dative -Plural + Plural- - lo los- + le les+ + le les+ - la las

Table 1: Etymological paradigm with binary features

Here we see some syncretism, but not too much; there are six distinct surfaceforms for eight possible cells (identical forms are separated by dotted lines). The(surface) feature combination [+ FEM, + DAT] is not formally distinguishable from[- FEM, + DAT] in this paradigm. (cf. 1c, 1d in the etymological paradigm). In thisparadigm, [- DAT] clitics can show a [+ FEM] distinction as well as [+ PLURAL], but[+ DAT] clitics can only show [+ PLURAL], not [+ FEM] (=les for both genders). The[+ COUNT] distinction is not formally expressed in this system. If, however, we addthe feature [+ COUNT] in order to cover the distinctions made in the referentialsystem in (2), we obtain the paradigm in Table 2.

Here the syncretism reaches more serious proportions: there are only fivedistinct surface forms for sixteen logical cells, i.e., there are far more surfacefeature combinations which remain formally indistinct than there are distinct ones(again, identical forms are separated by dotted lines). There are four blank cellswhich may be deemed to be natural and cost-free, due to the semanticincompatibility of [- COUNT] and [+ PLURAL]. But the other gaps in this paradigm,in particular the fact that perfectly reasonable feature combinations such as


[+ COUNT, + DAT] in (2f), [- COUNT, +FEM] in (2h) or [- COUNT, + DAT] in (2i) allremain formally indistinct, seem rather gratuitous. In this referential paradigm,animates (which are inherently [+ COUNT]) can show [+ FEM] distinctions as wellas [+ PLURAL], but not [+ DAT]. Inanimate count objects show only [+ PLURAL](neither gender nor case), while mass objects are [- COUNT] and thus inherentlynon-plural, and show neither case nor gender.

Count Feminine Dative -Plural + Plural

- - - lo

- - + lo

- + - lo

- + + lo

+ - - le les

+ - + le les

+ + - la las

+ + + la lasTable 2: Referential paradigm with binary features

It is clearly undesirable to stipulate all these apparently idiosyncraticrelationships between features. How then can we express these dependenciesamongst features which characterize these clitic paradigms? In order to do so, wemust abandon the view of morphological representation as unstructured matriceswhich freely combine binary features; such a system will always vastly over-predict the possible paradigms. What we need is a principled way to rule outunattested combinations: a theory which specifies hierarchical relationshipsbetween monovalent features.

3. Proposed feature geometryIn order to reflect the (variable) interaction of Case, number, gender and

countability in Spanish clitics, a Feature Geometry such as the one proposed byHarley & Ritter (1998; see also Harley 1994, Ritter 1997, Béjar 1999, 2000) needsto be enriched by integrating (dative) case as one of the possible values of theClass node. This differs slightly from the Feature Geometry proposed by Bonet(1991, 1995) for Romance clitics, where Case is a distinct branch of the geometry,separate from Agreement (number and gender) features (Béjar & Currie Hall 2000propose yet another view of Case geometry). The resulting structure in (3) shows

60 DAVID HEAP

the proposed relationship amongst these monovalent features in Spanish clitics,with Case being a possible dependent of Other. Note that first and second personclitics have a Participant node, but never Class features (such as feminineGender or dative Case), which are both dependents of Other. Third person clitics,on the other hand, are obtained from different combinations of features under thenode Other.

Such proposals crucially transfer some major assumptions about structuralmarkedness from phonology (see for example Rice forthcoming) to morphology.The most important of these assumptions is that markedness correlates withstructure; more complex structures are more marked, less complex structures areless marked. Furthermore, markedness is not viewed as a single dimension; whichfeatures within a class function as marked depends on a number of factors(including the inventory and the amount of structure required to distinguish thecontrasts within it). And finally, markedness is seen as something which can varypositionally; different features may be unmarked in different positions within thesame inventory.

(3) Internal structure of Spanish object cliticsCL

wowowowo Participant Other gggg wowowowo

[speaker] [group] Class ~ [feminine]

Note that in this Feature Geometry, organizing nodes are in Bold, andfeatures, which are privative, are [bracketed] terminal nodes. The structureproposed in (3) is a modified and simplified2 version of Harley & Ritter (1998)in which the root node is simply CL (for clitic, rather than for all pronouns andagreement markers, as in Harley & Ritter�s proposal). Furthermore, theorganizational node Other, which corresponds to all third persons, replacesHarley & Ritter�s label Individuation (since mass objects are not individuated).Finally, the Class node denotes whatever the marked type of object is within agiven inventory: it corresponds to dative le (as opposed to unmarked accusativelo) in the etymological paradigm, and count le (as opposed to the unmarked masslo) in the referential paradigm. Crucially, in both cases, Class is in complementarydistribution with the feature [feminine]: this embodies the generalisation that it isonly possible to mark [feminine] gender in the unmarked (accusative) Case. To

2 Harley & Ritter (1998) make a number of additional Number and Class distinctions which arenot needed here, and are thus omitted.


achieve this, the terminal feature node [feminine] is represented as a daughter ofOther and a mutually exclusive alternative to Class, and a relationship ofstructural markedness is posited whereby the unmarked case is accusative, and theunmarked gender is masculine.

This proposed account relies critically on the interaction between structuralmarkedness and the available clitic inventory. Thus lo always corresponds to theleast marked Other clitic, which has both unmarked gender (masculine) andunmarked case (accusative), while the Class specifications for feminine la anddative le are in complementary distribution. Some of the general motivation formorphological Feature Geometries can be found in the claims made by Harley &Ritter about their geometric proposal:

i. Cross-linguistic variation and paradigm-internal gaps and syncretisms are constrained bythe hierarchical organization of features in the universal geometry.ii. The interpretation of sub-trees of the geometry may be relativized so that language-specificinterpretation of a given feature will depend in part upon the contrasts available within thefeature system of that language. (Harley & Ritter 1998:1)

The split between Participant and Other in (3) corresponds to a distinctionmade since at least Bloomfield (1938) and Benveniste (1956) between �realpersons�(first and second persons) and �non-persons� (third persons) in manydifferent languages. In the present proposal, this split also reflects the fact that,across Romance pronoun systems, first and second person clitics are not inflectedfor case and gender, while third person clitics are. Furthermore, number markingin third persons is (often) more transparent than in first and second persons: allOther clitics form a plural by adding an -s in Spanish, while Participant cliticshave no clear relationship between their singular (me, te) and plural (nos, os)forms.

It should also be noted that in this proposed geometry, se is the least markedclitic (see Heap 1996, 1998) and thus corresponds to the bare CL node which hasno features of its own and must therefore get them from an antecedent.3 Thebinding properties of the clitic se do not need to be specified by a special ad hocfeature [REFLEXIVE], as in Grimshaw�s (1999) OT account; instead, reflexivity isseen here as a byproduct of morphological underspecification. As discussed inHeap (forthcoming), there is considerable advantage to avoiding a feature such as

3 Grimshaw (1997) also suggests that se is the least specified clitic but rejects this view in laterwork (1999). Kayne�s (1998) suggestion that se may be part of a natural class with me and te iscompatible with the geometry proposed here, wherein these three clitics have in common their lackof the Other node and its dependents.

62 DAVID HEAP

[REFLEXIVE] in the morphological representation of this clitic, since many of theuses of Spanish se do not have really reflexive readings at all.

The general geometry in (3) provides the inventory in (4) for third personobject clitics in those varieties of Spanish which use the etymological paradigm,while the inventory of third person object clitics in varieties using the referentialparadigm is given in (5).

(4) Spanish third person clitics (etymological paradigm):

CL CL CL | | |

Other Other Other | |Class [feminine]

lo le [=dative] la

CL CL CL | | |

Other Other Other/ VVVV VVVV

[group] [group] Class [group] [feminine]

los les [=dative] las

(5) Spanish third person clitics (referential paradigm):

CL |

Other

lo

CL |

Other |Class

le [=count]

CL |

Other |

[feminine]

la

CL |Other

VVVV[group] Class

les [=count]

CL |

Other V V V V[group] [feminine]

las


The differences between (4) and (5) flow from the presence of the defaultsemantic interpretation of third persons as �mass objects� in the latter grammarand its absence in the former. The default pronoun lo always corresponds to theleast marked clitic under the node Other; in (5), the unmarked Other clitic is themass clitic lo, which cannot be pluralized (thus the absence of los in thatparadigm). The semantic interpretation of Class depends on what feature it iscontrasted with; in (4), where the unmarked lo has the default case (i.e.accusative), Class is interpreted as Dative, and where unmarked lo has a massreading, then Class marks count objects.

4. Complexity constraintsThe geometry in (3) has certain built-in advantages over a system of binary

morphological features, in that the hierarchy of features constrains the possiblecombinations. Crucially, only one value for Class can be expressed: either Class(interpreted as either [dative] or [count]) or [feminine] but not both. An additionaladvantage of such a geometry is that it allows us to capture other generalizationsabout the overall amount of morphological specification which a clitic can bear,a notion which cannot easily be expressed in a binary feature system. Inphonology, where feature geometries are well-established, Structural ComplexityConstraints can be used to express just such limits on specification, for example:

(6) Structural Complexity Constraint (SCC):Specified SV [Sonorant Voice] structure implies lack of specified Placestructure; specified Place structure implies lack of specified SVstructure. (Rice & Avery 1991, 1993)

This sort of phonological asymmetry (between different branches of ageometry) is analogous to the morphological asymmetry between Participant andOther clitics noted above, i.e., the fact that first and second persons never showcase and gender distinctions while third persons regularly do. This sort ofcompensatory effect would need to be stipulated in any account of the cliticlexicon which depends on a binary feature matrix (see for example Grimshaw1997, 1999), but flows directly from a geometrical representation. The basicintuition expressed by this SCC is simple: while a geometry may include manypossible distinctions, there is a limit to how many of these distinctions can beexpressed by any one phonological segment. This notion of ComplexityConstraint has been successfully extended to morphological feature geometrieslike the one proposed here by Béjar (1999, 2000), whose Complexity Constraintfor Standard Arabic verbal morphology is given in (7):

64 DAVID HEAP

(7) Morphological Complexity Constraint (MCC):There can be no more than two non-organizing nodes in arepresentation. (Béjar 1999, 2000)

It turns out that this MCC as stated is directly applicable to the inventories in(4) and (5); no clitic in Spanish has specifications for more than two non-organizing nodes. When we recall that these terminal nodes are marked featurevalues, such a constraint simply means that there is a limit to how muchmarkedness can be expressed by a given morphological unit. The clitics in (8) aretherefore the most specified clitics that can occur in Spanish: nos and las, whichhave two terminal features each.

(8) Maximally specified Spanish clitics:

CLwo wo wo wo

Participant Other gggg g g g g [speaker] [group]

nos

CL ggggOther

rorororo[group] [feminine]

las

I do not wish to suggest that the direct applicability of Béjar�s MCCformulated for Arabic to the Spanish inventories studied here is anything morethan a coincidence: it seems implausible that the limit of just two terminal nodeswould be a language universal of any kind. Rather, it seems likely that thepossibility of setting the MCC is provided universally and that grammars canparametrize this MCC at different levels of complexity, depending on the inputthey receive. Under this view, the acquisition of a morphological geometry can beseen as proceeding in much the same manner as the step-by-step development ofa phonological Feature Geometry to represent the contrasts which are present inthe input received (Rice & Avery 1995).4

5. Conclusions and further researchThe analysis proposed here lends morphological support to the idea that

markedness is a matter of structures and representations, rather than a product of

4 While the present proposal is not directly informed by acquisition data, morphologicalgeometries of this type are in principle compatible with what is known about the learnability ofmorphological contrasts (see for example Hanson, Harley & Ritter 2000).


derivations, rules or constraints (Rice forthcoming). Specifically, a geometricrepresentation of Spanish object clitics can directly represent the hierarchicalrelationships amongst morphological features, which would otherwise need to bestipulated. Such a geometry can also represent the variation which is attested inSpanish object clitic paradigms in a direct and natural fashion, without recourseto ad hoc features or stipulations.

In terms of the types of morphological features used, it is also clear thatprivative features are preferred to binary ones, especially since the combinationof [+ α] and [Ø α] (unmarked for feature α, see Rooryck 1994) inevitably leads toternary oppositions which significantly overpredict the typology of potentialparadigms, and thus require extensive stipulation. For example, Grimshaw (1999)uses a clitic lexicon in which first and second person clitics are specified as[Ø REFLEXIVE] while third persons are [- REFLEXIVE] and se is [+ REFLEXIVE](needless to say, a similar ternary opposition for other morphological featureswould lead to inventories much larger than those actually attested). In the presentproposal, however, the different binding properties of Participant, Other andbare CL clitics are seen instead as direct consequences of their morphologicalstructure, and need not be stipulated.

In addition to being more constrained in its use of privative monovalentfeatures and hierarchical implication relationships, a Feature Geometry approachhas the advantage of being able to express limits on the numbers of markedfeatures which can appear in a representation. Whatever the form whichComplexity Constraints eventually take, they clearly express an importantgeneralization about markedness: that morphological units (like phonologicalsegments) cannot exhibit all the contrasts which are provided by their featuralmakeup, and that more specification in one area implies less in another.

The present analysis has however only scratched the surface of the sociolectaland geolectal variation which is attested in Spanish object clitic systems. Thereare at least three different types of leísmo with different plural systemsdocumented by Fernández-Ordóñez (1994), and these are just the attested variantsof the referential paradigm within Castilian. Such extensive variation is a sourceof further data which should allow us to refine our hypotheses about the featuregeometric representation of morphological paradigms.

In this theory, variation has a source in structural properties. Languages may makedifferent demands on whether structure is required, and, if some is, just how much; incases of equipollent structure, languages make a choice as to which is more marked.(Rice forthcoming:36)

66 DAVID HEAP

Ultimately, formal analyses of variable data allow us to gain a clearer view of theunderlying properties of morphological structures, and of the nature ofmarkedness.

REFERENCES

Béjar, Susana. 1999. �Structural Markedness and Minimalist Checking Theory�.Proceedings of the West Coast Conference on Formal Linguistics 18 ed. bySonya F. Bird, Andrew Carnie, Jason D. Haugen & Peter Norquest, 27-42.Somerville, Mass.: Cascadilla Press.

---------. 2000. �Structural Markedness and Formal Features�. Revue québécoisede linguistique 28.1.47-72. Special thematic issue on �Traits et interfaces� ed.by David Heap & Juvénal Ndayiragije

---------. & Daniel Currie Hall. 2000. �Marking Markedness: The UnderlyingOrder of Diagonal Syncretisms�. Ms., University of Toronto.

Benveniste, Émile. 1956. �La nature des pronoms�. For Roman Jakobson. LaHaye: Mouton. Reprinted in Problèmes de Linguistique Générale, 1966, vol 1,251-257. Paris: Gallimard.

Bloomfield, Leonard. 1938. Language. New York: Holt, Rinehart & Winston.Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in Romance

Ph.D. Dissertation, MIT.--------- 1995. �The Feature Structure of Romance Clitics�. Natural Language and

Linguistic Theory 13.607-617.Fernández-Ordóñez, Inés. 1993. �Leísmo, laísmo y loísmo: estado de la cuestión�.

Los pronombres átonos ed. by Olga Fernández Soriano, 63-96. Madrid: Taurus.---------. 1994. �Isoglosas internas del castellano. El sistema referencial del

pronombre átono de la tercera persona�. Revista de filología española 74.71-125.

Grimshaw, Jane. 1997. �The Best Clitic: Constraint Conflict in Morphosyntax�.Elements of Grammar ed. by Liliane Haegeman, 169-196. Dordrecht: Kluwer.

---------. 1999. �Optimal Clitic Position and the Lexicon in Romance CliticSystems�. OT Syntax ed. by Géraldine Legendre, Jane Grimshaw & StenVikner. Cambridge, Mass.: MIT Press.

Hanson, Rebecca, Heidi Harley & Elizabeth Ritter. 2000. �Underspecification anduniversal defaults for person and number features�. Paper presented at theAnnual Meeting of the Canadian Linguistics Association, held at the Universityof Alberta, Edmonton, May 2000.


Harley, Heidi. 1994. �Hug a Tree: Deriving the Morphosyntactic FeatureHierarchy�. Papers on Phonology and Morphology (MITWPL 21) ed. byAndrew Carnie & Heidi Harley, 289-320. Cambridge, Mass.: MITWPL.

--------- & Elizabeth Ritter. 1998. �Meaning in Morphology: Motivating a feature-geometric analysis of person and number�. Ms., University of Calgary &University of Pennsylvania.

Heap, David. 1996. �Optimalizing Romance Clitic Sequences�. Paper presentedat the Linguistic Symposium on Romance Languages XXVI, held at theUniversidad Metropolitana, Mexico City, April 1996.

---------. 1998. �Optimalizing Iberian Clitic Sequences�. Theoretical Advances inRomance Linguistics ed. by José Lema & Esthela Treviño (=Current Issues inLinguistic Theory 157), 227-248. Amsterdam & Philadelphia: John Benjamins.

---------. Forthcoming. �Constraining Optimality: Clitic sequences and FeatureGeometry�. Submitted to Perspectives on Clitic and Agreement AffixCombinations (Series: Linguistik Aktuel) ed. by Lorie Heggie & FranciscoOrdóñez. Amsterdam & Philadelphia: John Benjamins.

Kayne, Richard. 1998. �Person morphemes and reflexives�. Ms., New YorkUniversity.

Klein-Andreu, Flora. 1981. �Distintos sistemas de empleo de le, la, lo. Perspectivasincrónica, diacrónica y sociolingüística�. Thesaurus 36.284-304. Reprinted inLos pronombres átonos ed. by Olga Fernández Soriano. 1993.337-353. Madrid:Taurus.

Rice, Keren. Forthcoming. �Featural markedness in phonology: variation�. Glot.--------- & Peter Avery. 1991. �On the relationship between laterality and

coronality�. Phonetics and phonology 2. The special status of coronals. Internaland external evidence ed. by Carole Paradis & Jean-François Prunet, 101-124.San Diego, Calif.: Academic Press.

--------- & Peter Avery. 1993. �Segmental complexity and the structure ofinventories�. Toronto Working Papers in Linguistics ed. by Carrie Dyck12.2.191-202. Toronto: Toronto Working Papers in Linguistics.

--------- & Peter Avery. 1995. �Variability in a Deterministic Model of LanguageAcquisition: A Theory of Segmental Elaboration�. Phonological acquisitionand phonological theory ed. by John Archibald, 23-62. Hillsdale, N.J.: L.Erlbaum Associates.

Ritter, Elizabeth. 1997. �Agreement in the Arabic Prefix Conjugation: Evidencefor a non-linear approach to person, number and gender features�. CanadianLinguistics Association Annual Conference Proceedings ed. by Leslie Blair,Christine Burns & Lorna Roswell, 191-202. Calgary: University of Calgary,Department of Linguistics.

68 DAVID HEAP

Rooryck, Johan. 1994. �On 0- and α- underspecification in syntax andphonology�. The Morphology-Syntax Connection (MITWPL 22), ed. by HeidiHarley & Colin Phillips, 197-216. Cambridge, Mass.: MITWPL.

CATALAN PHONOLOGYCLUSTER SIMPLIFICATION AND NASAL PLACE ASSIMILATION*

DYLAN HERRICKUniversity of California, Santa Cruz

0. IntroductionThis paper examines the complex interaction of nasal place assimilation

(hereafter NPA) and cluster simplification (hereafter CS) from the perspectiveof a parallel non-serial version of Optimality Theory (OT, Prince & Smolensky1993). Within OT, the solution makes use of Correspondence Theory(McCarthy & Prince 1995) and output-output constraints (Benua 1995) toillustrate that by straightforwardly combining an analysis of CS with ananalysis of NPA we can account for the apparent opacity exhibited in (1)without the need for additional theoretic machinery such as cyclicity, multiplelevels, or underspecification.

(1) /tin+k bint bota+s/[ti@N bi@m bo@t´s]I have twenty wineskins“I have twenty wineskins”

While the surface forms for the first two words show the effects of CS, theeffects of NPA are only partially obvious; [ti@N] is surface-opaque since [N]does not share place with the [b] of the following word, but [bi@m] is surface-transparent since [m] shares place with the [b] of the following word.

* I would like to thank the following people for comments on earlier versions of this paper:Laura Downing, Rodrigo Gutierrez, Ryuji Harada, Junko Ito, Kazutaka Kurisu, Bill Ladusaw,Armin Mester, Jaye Padgett, Jason Riggle, Nathan Sanders, Phillip Spaelti, Adam Ussishkin,Andy Wedel, Caroline Wiltshire, the members of the UCSC Research Seminar 1999, themembers of the UCSC Phonology Interest Group in Spring 1999, the audience at LSRL30, andtwo anonymous reviewers for this volume. None of these people should be taken asnecessarily agreeing with the analysis, and all errors are my sole responsibility.

DYLAN HERRICK70

Mascaró (1976) first presented this data as crucial evidence in favor of thephonological cycle. He showed that while no single (non-cyclic) ordering ofCS and NPA can account for the attested surface forms, a cyclically orderedset of (rewrite) rules can. For Kiparsky (1985), on the other hand, the samedata was used to illustrate the benefits of splitting phonology into lexical andpostlexical modules. He improved on Mascaró’s analysis by, among otherthings, showing that the interleaving of phonological rules with morphologicaloperations accounts for the restriction of cyclic rules to derived environments.

In contrast, the analysis presented here accounts for the data without theadded theoretical machinery of a cycle, levels, or underspecification. Thesimultaneous surface transparent and surface-opaque outputs of (1) areaccounted for without adding to the analyses which are independentlynecessary to account for CS and NPA. The remainder of the paper is organizedas follows; section 1 covers the analysis of CS, section 2 provides an accountfor NPA, section 3 illustrates how the two analyses automatically account forthe complex forms such as (1) without any additional assumptions, and finally,section 4 presents the conclusions.

1. Cluster simplificationIn Catalan CS, obstruents fail to surface word finally when preceded by a

homorganic consonant; e.g. /nt/ surfaces as [n], /mp/ surfaces as [m], and soon. The relevant data is given in (2) below. The first column gives words inisolation, and the second column provides evidence for the makeup of theunderlying consonant cluster. All data is from Mascaró (1976:53-55).

(2) Cluster simplification data ([Et]= diminutive)word final word internal gloss

a. [ka@m] [k´mpE@t] “field”b. [pu@n] [puntE@t] “point”c. [ba@N] [b´NkE@t] “bank”d. [m´la@l] [m´l´ltE@t] “ill, sick”e. [ç¤r] [urtE@t] “orchard”f. [gu@s] [gustE@t] “taste”g. [se@rp] [s´rpE@t] “snake”h. [di@sk] [diskE@t] “disk”i. [mç@…k] n/a “I grind”

The data in (2a-f) exhibit cluster simplification, while (2g-i) do not. Thecrucial distinction is that the data in (2a-f) have underlying homorganic

CATALAN PHONOLOGY CS & NPA 71

consonant clusters (ending in obstruents), while in (2g-i) the underlyingconsonant cluster is not homorganic.

1.1 An OT analysis of cluster simplificationThe intuition behind the OT analysis of Catalan CS presented here is that

complex codas simplify only if the loss of a segment does not result in the lossof consonantal place specification. In other words, CS arises because amarkedness constraint which bans complex codas outranks a faithfulnessconstraint which preserves segments. The cases that show no CS effects areexplained by ranking a faithfulness constraint which preserves consonantalplace specification over the markedness constraint on complex codas.Whenever the loss of a segment entails the loss of place, CS will not beobserved (2g-i); however, wherever a complex coda can simplify without lossof place (2a-f) it will. The specific constraints are given below.

(3) MAX(PLACE) (Lombardi 1998)Every input place feature has an output correspondent.

(4) MAX SEGMENT INPUT-OUTPUT (McCarthy & Prince 1995)Every segment of the input has a correspondent in the output.“No phonological deletion.” (abbreviated as MAXIO)

(5) *COMPLEXCODA (Prince & Smolensky 1993)Syllables must not have complex codas. (abbreviated as *COMP)

By ranking *COMP over MAXIO, we predict a language with no complexcodas. This works well with (2a-f) but not (2g-i). To account for (2g-i),MAX(PLACE) must outrank *COMP, thereby ensuring that the loss of codasegments will not entail the loss of place. In other words, complex codas willsimplify only if they are homorganic. This is illustrated by Tableaux 1 and 2.

The input in Tableau 1 contains a word final homorganic consonant clusterand shows the crucial ranking of *COMP >> MAXIO. The attested output,candidate (a), satisfies *COMP at the expense of MAXIO, while candidate (b)incurs a fatal violation of *COMP. Candidate (a) does not violate MAX(PLACE)because the output coda consonant is specified for a labial feature.

In Tableau 2, the input contains a consonant cluster but, crucially, it is not ahomorganic cluster. This allows us to establish the ranking of MAX(PLACE) >>*COMP. Since the word final consonant cluster contains a coronal followed bya labial, candidate (a), which fails to realize the labial segment, is eliminated

DYLAN HERRICK72

by a high ranking MAX(PLACE) violation. Candidate (b), which violates*COMP, satisfies MAX(PLACE) and surfaces as the winner.1

/ k a m p/

[lab]

MAX(PLACE) *COMP MAXIO

☞ a. k a m |[lab]

*

b. k a m p

[lab]

*!

Tableau 1: *COMP >> MAXIO

/ s e r p/ | |[cor][lab]

MAX(PLACE) *COMP MAXIO

a. s e r |[cor]

*! *

☞ b. s e r p | |[cor][lab]

*

Tableau 2: MAX(PLACE) >> *COMP

At this point, we have the core analysis for CS, but there is one morecomplication – Catalan cluster simplification shows identity effects betweenwords and phrases. When the diminutive morpheme [E@t] combines with thenoun [ka@m], the resulting form is [kampE@t]. This provides evidence that theunderlying form of [ka@m] is /kamp/. However, when [ka@m] is combined withthe copula [e@s], the attested output does not contain a labial stop; it is [ka@me@s]and not *[ka@mpe@s]. Since the copula and the diminutive have the samephonological shape, [V@C], the resulting generalization is that the word finalobstruent of an underlying homorganic cluster can only surface within amorphologically complex word.

Within a parallel non-serial version of OT, this generalization can beaccounted for with an output-output constraint (Benua 1995).2 Output-output

1 I do not consider candidates such as [kap] or [sep]. I assume that high ranking LINEARITYand CONTIGUITY constraints (McCarthy & Prince 1995) eliminate such candidates.2 Another possibility would be to assume a multi-level system in OT where the output of onelevel serves as the input to the next. A full discussion would lead us too far astray, but the


constraints enforce an identity relation between a base word and the base wordplus affixal material. The proposal here is that we allow the identity relation tobe calculated between a base word and the base word in a phrase. In a sense,this requires that a word in a phrase be identical to its form in isolation. Morespecifically, in Catalan, an output-output DEP constraint will demand that allthe output segments of a phrase correspond to output segments of a word asdefined in (6) below.3

(6) DEPOUTPUTWORD-OUTPUTPHRASE

All output segments of a word in a phrase must correspond to outputsegments of that word (when evaluated) in isolation. (abbreviated asDEPOOPHRASE)

With DEPOOPHRASE in place, we can now explain the distinction between[k´mpE@t] and *[ka@mpe@s]. Tableau 3 shows the evaluation for a single word,and since this is a single word, DEPOOPHRASE is not active. Tableau 4, on theother hand, contains a multi-word input, and therefore the output words areevaluated by DEPOOPHRASE against their single word correspondents.

/kamp/“field”

MAX(PLACE)

*COMP DEP-OOPHRASE

MAXIO

a. ka@mp *!

☞ b. ka@m *

Tableau 3: DEPOO inactive for a single word

In Tableau 3, we are evaluating a word in isolation, and therefore,DEPOOPHRASE can be considered inactive. Thus, [ka@m] emerges as the winner.

Tableau 4 illustrates a case in which the input is a phrase; the single wordoutputs are evaluated in parallel, but for ease of reference they are includedunderneath the input. Looking only at single words, the output for the copulais [e@s], and the output for “the field” is [´lka@m]. An output such as candidate(a) violates DEPOOPHRASE because there is no word level correspondent for the

main arguments against such a system can be found in Benua (1995). For counterargumentssee Kiparsky (1999).3 This constraint needs to have access to the output forms of the individual words in isolation.For the purpose of this paper, I assume that el camp es gran ([´lka@m e@s gra@n]) “the field isbig” constitutes a phrase, and, phonologically speaking, [´lka@m], [e@s], and [gra@n] constitutewords which are evaluated in parallel to the phrase.

DYLAN HERRICK74

labial stop [p]. The winning candidate is more faithful to the phonologicalshape of the individual output words than it is to the entire input. The twocandidates considered below differ only in the presence or absence of [p].

/el kamp es gran/ [´lka@m] [e@s] [gra@n] “the field is big”

MAX(PLACE)

*COMP DEP-OOPHRASE

MAXIO

a. ´lka@mp e@s gra@n *!(p)

☞ b. ´lka@m e@s gra@n *

Tableau 4: DEPOO active at phrasal level

1.2 Summary of OT constraint ranking for CS in CatalanThe overall constraint ranking for Catalan CS is shown in Figure 1 below.

CS is viewed as a drive to simplify complex codas, with *COMPLEX outrankingMAXIO. However, since CS only occurs when the word final consonantcluster is homorganic, the constraint MAX(PLACE) is ranked above the ban oncomplex codas. This prevents CS in exactly those cases where loss of asegment entails loss of place – non-homorganic consonant clusters. Finally, anoutput-output constraint, DEPOOPHRASE, ensures that the word final obstruent ofan underlying homorganic cluster surfaces only within a morphologicallycomplex word.

MAX(PLACE)

*COMPLEX DEPOOPHRASE

MAXIOFigure 1: Final Constraint Ranking for Catalan CS

2. Nasal place assimilationNPA is a well attested cross-linguistic phenomenon in which nasal

consonants share the same place specifications as adjacent consonants. InCatalan, nasals in coda position assimilate the place features of immediatelyadjacent obstruents as shown in (7) below. There are three variants for thecoronal nasal, and these are written as; [n] for the [+anterior, +distributed]coronal, [n] for the [+anterior, −distributed] coronal, and [n,] for the [−anterior,+distributed] coronal. All data is from Mascaró (1976).


(7) NPA in Coronal Nasals: [so@n] “they are”a. so[n] [a]mics “they are friends”b. so[m] [p]ocs “they are few”c. so[M] [f]eliCos “they are happy”d. so[n] [d]os “there are two”e. so[n] [s]incers “they are sincere”f. so[n,] [z&]ermans “they are brothers”g. so[N] [g]rans “they are big”

The data in (7) exhibit full NPA since the adjacent segments share all placefeatures; however, the actual situation in Catalan is not so straightforward.The data in (8)-(10) describe the numerous exceptions to full NPA in Catalan.The first exception is that coronal nasals do not assimilate fully to palatalsegments as shown in (8) below.

(8) Incomplete-Assimilation to Palatals:so[n,] [¥]iures “they are free”

Full assimilation would predict the palatal nasal [¯] in (8); however, we findinstead the pre-palatal nasal, [n,].

The second major exception to full NPA is that underlying labial nasalsonly assimilate to adjacent labial consonants. The labial nasal remains labial inall cases; the only case where assimilation can be seen is in (9c) which showsthe assimilation of [+distributed] (assimilation in (9b) is vacuous).

(9) NPA in Labial Nasals: [so@m] “we are”a. so[m] [a]mics “we are friends”b. so[m] [p]ocs “we are few”c. so[M] [f]eliCos “we are happy” (assimilation of [+distr.])d. so[m] [d]os “we are two”e. so[m] [¥]iures “we are free”f. so[m] [g]rans “we are great”

The final exception to Catalan NPA, is that neither palatals nor dorsalsexhibit any assimilation effects at all. An adjacent labial consonant will fail toaffect the place features of a palatal nasal (10a) or a dorsal nasal (10b). Sincethese segments never show NPA, only two forms are given.

DYLAN HERRICK76

(10) NO NPA in Velar and Palatal Nasals:a. a[¯] [f]eliC “a happy year”b. ti[N] [p]a “I have bread”

The remainder of section 2 proposes and illustrates an OT analysis ofCatalan NPA.

2.1 OT analysisThe core OT analysis for NPA is essentially that of Padgett (1995). The

intuitive idea is that adjacent consonants are forced to share place features,even at the expense of being unfaithful to the input. The constraint whichdemands place assimilation is called SPREAD(PLACE) and belongs to the largerclass of spreading constraints defined in (11). Since NPA is directional, aspecific form of MAX(PLACE) which is sensitive to the feature [+release]preserves the place features of consonants which precede more sonoroussegments (and obstruents in word final position).4 This constraint is calledMAXRELEASE(PLACE) and is defined in (12).

(11) SPREAD(X): ∀∀∀ ∀ x,y x(y)(in some domain) (Padgett 1995)Every feature is linked to every segment (in some domain), with xranging over features and y over segments. “Assess one violationmark for each feature which fails to link to each segment.”(abbreviated as SPREAD(PLACE))

(12) MAXRELEASE(PLACE) (Padgett 1995)Let S be a [+release] output segment. Then every place feature in theinput correspondent of S has an output correspondent in S.(abbreviated as MAXREL(PLACE))

In addition to (11) and (12), the constraint (3), MAX(PLACE), is alsonecessary. Since NPA requires violations of faithfulness, SPREAD(PLACE) mustbe ranked above MAX(PLACE). The ranking of MAXREL(PLACE), however,cannot be established since any violation of MAXREL(PLACE) will entail aviolation of MAX(PLACE). For the remainder of this paper, it will be placed tothe left of SPREAD(PLACE). Tableau 5 depicts how this ranking works.

4 Consonants which precede sonorant segments are released (Steriade 1997), and, in Catalan,word final obstruents have a small burst (Recasens 1991, 1993) which, for the purpose of thispaper, is equated with release.


/son poks/‘they are few’

MAXREL(PLACE) SPREAD(PLACE) MAX(PLACE)

a. so@n po@ks *!(np)☞ b. so@m po@ks *(m)

c. so@n to@ks *!(t) *(t)Tableau 5: SPREAD >> MAX

Candidate (a) is totally faithful to the input, but this results in a fatalviolation of SPREAD(PLACE). Candidates (b) and (c), on the other hand, fullysatisfy SPREAD(PLACE) at the expense of MAX(PLACE). This leavesMAXREL(PLACE) to decide the outcome in favor of candidate (b) which remainsfully faithful to the [+release] segment [p].5

The ranking given in tableau 5 provides an account of full NPA, but it doesnot account for the numerous exceptions to NPA in Catalan. In order toaccount for the non-assimilation of non-coronal nasals, the faithfulnessconstraint to place must be decomposed into its more specific instantiations asshown in (13) below.

(13) Decompose MAX(PLACE) into feature specific constraints.MAX(PLACE) → MAX(LAB), MAX(COR), MAX(PAL), MAX(DOR)

What (13) affords us is the possibility of blocking spreading effects to non-coronal segments by giving a relatively high ranking to MAX(LAB), MAX(PAL),and MAX(DOR). I simplify and refer to these instantiations of MAX(PLACE) asMAX(NONCOR). This is shown in (14).

(14) MAX(NONCOR) = MAX(LAB), MAX(PAL), MAX(DOR)

Whether or not there is a universally fixed ranking between MAX(COR) andMAX(NONCOR) remains to be seen, but the universal low ranking of MAX(COR)would provide a partial explanation of the inherent weakness of coronals(Paradis & Prunet 1991 and references therein). Additionally, the low rankingof MAX(COR) is crucial to this analysis. Since spreading is most evident forinput coronals, MAX(COR) must be ranked below SPREAD(PLACE). 5 The MAXREL(PLACE) analysis is superior to a purely prosodic account provided byIDENTONSET(PLACE). The reasoning is as follows: in Catalan lateral place assimilation, word-final obstruents surface (e.g. /mol+k/ surfaces as [mo@…k] “I grind”; see (2i) above).IDENTONSET(PLACE) fails to make a prediction for the directionality of spreading in word finalcoda clusters; since only onsets are special, IDENTONSET(PLACE) makes no prediction aboutcodas. MAXREL(PLACE), on the other hand, predicts that the place of the word final obstruentwill be preserved since word final obstruents are [+release].

DYLAN HERRICK78

MAX(NONCOR), on the other hand, must outrank the spreading constraint sinceit is precisely the non-coronals which resist NPA. The ranking is provided in(15).

(15) MAX(NONCOR) >> SPREAD >> MAX(COR)

Tableaux 6 through 11 illustrate how the ranking in (15) can allow NPA incoronals while blocking it for labials and palatals. In addition, the partialassimilation seen in labials gains an explanation, as shown in Tableau 9.Tableau 6, immediately below, shows assimilation in coronals.

/son grans/“they are big”

MAXREL

(PLACE)Max

(NONCOR)SPREAD(PLACE)

MAX(COR)

a. so@n gra@ns *!(ng)☞ b. so@N gra@ns *(n)

c. so@n dra@ns *!(g) *(g) *Tableau 6: Assimilation for coronals

Candidate (a) of Tableau 6 is fully faithful and therefore eliminated byspreading violations. Candidate (c) satisfies the spreading constraint, but itviolates all faithfulness constraints by choosing to assimilate to the coronalcoda consonant rather than the dorsal. Candidate (b) wins because itminimally violates faithfulness while completely satisfying the spreadingconstraint.6 Tableaux 7 and 8 show how MAX(NONCOR) effectively blocksNPA in non-coronals.

/som grans/“we are big”

MAXREL

(PLACE)Max


MAX(COR)

☞ a. so@m gra@ns *(mg)b. so@N gra@ns *!(m) *(n)c. so@m pra@ns *!(g) *(g) *(n)

Tableau 7: Non-assimilation for input labials

/a¯ felis/“a happy year”

MAXREL

(PLACE)Max


MAX(COR)

☞ a. a@̄ f´li@s *(¯f )b. a@M f´li@s *!(¯) *(n)c. a@̄ c´li@s *!(f) *(f)

Tableau 8: Non-assimilation of input palatals 6 The [g] is [+release] because it is followed by a more sonorous segment.


In both Tableau 7 and Tableau 8, the winning candidates violate thespreading constraint. However, any candidate which satisfies the spreadingconstraint will incur a violation of faithfulness, in particular, a fatal violation ofthe high-ranking MAX(NONCOR) constraint. However, the high ranking ofMAX(NONCOR) does not block the partial assimilation of the distributed featureseen in labial segments as shown in Tableau 9.

/som felisus/“we are happy”

MAXREL

(PLACE)Max


MAX(COR)

a. so@m f´l i@ sus *!(mf)☞ b. so@M f´l i@ sus

Tableau 9: Partial assimilation for input labials

In Tableau 9, the totally faithful candidate is eliminated because it incurs aspreading violation. However, the candidate which perfectly satisfies thespreading constraint does not violate either MAX(COR) or MAX(NONCOR)(although it would violate general faithfulness to input features; IDENT[F]).

This leaves us with only one more puzzle for Catalan NPA – the partialassimilation of coronals to palatal segments. At present, the ranking predictsthat input coronals will assimilate fully to a following palatal segment. Thefact that this does not occur can be explained by assuming that a markednessconstraint against palatal segments eliminates the candidate which fullysatisfies the spreading constraint. Such a constraint is given in (16) below.

(16) *PALATALSEGMENTPalatal segments are marked. Assess one violation mark for eachpalatal segment. (Abbreviated as *PAL)

The markedness constraint *PAL is a constraint against segments, notfeatures. Thus, its violation marks are assigned for each palatal segment in acandidate, not each palatal feature.

/son ¥iures/“they are free”

MAXREL

(PLACE)MAX

(NONCOR)*PAL SPREAD

(PLACE)MAX(COR)

a. so@n ¥iu@r´s *(¥) ***!b. so@¯ ¥iu@r´s **!(¯¥) *(n)

☞ c. so@n, ¥iu@r´s *(¥) ** *(n)Tableau 10: Partial Assimilation to Palatals *PAL >> SPREAD

DYLAN HERRICK80

Tableau 10 shows that by ranking *PAL above SPREAD(PLACE), the fullassimilation candidate will lose because it contains an additional palatalsegment. A non-assimilation candidate or a partial assimilation candidate willavoid this extra violation of *PAL, and therefore be preferable to the fullassimilation candidate. At this point, it becomes important to remember thatthe spreading constraint is gradiently violable, and therefore the candidate withthe segment which comes closest to being palatal without actually becomingpalatal will win.7 For this reason, candidate (c) wins out over candidate (a);candidate (c) incurs fewer violations of SPREAD(PLACE).8

/ a¯ felis /“a happy year”

MAXREL

(PLACE)MAX

(NONCOR)*PAL SPREAD

(PLACE)MAX(COR)

☞ a. a@̄ f´li@s *(¯) *(¯f )b. a@M f´li@s *!(¯)c. a@n f´li@s *!(¯) *(nf )a. a@̄ c´li@s *!(f ) *(f ) **(¯c)

Tableau 11: Input palatals still surface; MAX(NONCOR) >> *PAL

There is an additional ranking for *PAL. It must be ranked belowMAX(NONCOR) in order for input palatal segments to surface in the output. InTableau 11, candidate (a) preserves the input palatal segment and violates*PAL. However, it is able to win because the candidates which satisfy *PALincur fatal violations of the higher ranked MAX(NONCOR) constraint.

2.2 Summary of NPA in OTSection two presented an OT analysis of Catalan NPA, and the overall

constraint ranking is shown in Figure 2 below. The basic intuition behind thisanalysis is that a markedness constraint which demands that adjacent segmentsshare all place features interacts with faithfulness to place. In particular,sharing place is more important than being faithful to input coronals segments(while the opposite is true for non-coronals).

7 An analysis which made use of the categorical spreading constraint AGREE (Beckman 1997)would not be able to distinguish between the non-assimilation candidate and the partial-assimilation candidate.8 Padgett (1995) provides a detailed discussion of how to calculate the exact number ofspreading violations for partial assimilation. Such a discussion has been omitted here forreasons of space. For the purpose of this paper, it is only necessary to acknowledge that theprepalatal nasal [n,] is more similar to a palatal [¯] in featural specifications than [n] is.


MAX(NONCOR)

*PAL MAXREL(PLACE)

SPREAD(PLACE)

MAX(COR)Fig. 2: OT Grammar for Catalan NPA

3. Combining the analyses CS and NPAThis section combines the analysis of CS with that of NPA to provide a

straightforward analysis of the simultaneous surface opacity and surfacetransparency of data such as that exhibited in (1) above. Where the ranking forCS had MAX(PLACE), we can simply insert the entire ranking for NPA. Thisessentially adds *COMP, DEPOO and MAXIO beneath MAX(COR) in Figure 2above.

Tableau 12 evaluates the input /tin+k/ “I have” – the first word of theexample (1). Candidate (a) wins despite its violations of MAX(COR) andMAXIO. Its closest competitor, candidate (b), is eliminated by *COMP.

/tin+k/“I have”

MAXREL

(PLACE)MAX


MAX(COR)

*COMP DEPOOPHRASE

MAXIO

☞ a. ti@N *(n)

*(k)

b. ti@Nk *(n)

*!(Nk)

c. ti@nk *!(nk)

*(nk)

d. ti@n *!(k)

*(k)

e. ti@nt *!(k)

*(k)

*( nt)

Tableau 12: NPA and CS in a single word

Tableau 13 evaluates the entire phrase presented in (1): /tin+k bint bot´+s/“I have 20 wineskins”.

DYLAN HERRICK82

/tin+k bint bot´+s/[ ti@N bi@m bo@t´s ]“I have 20 wineskins”

MAXNONCOR

SPREAD(PLACE)

MAX(COR)

*COMP DEPOOPHRASE

MAXIO

a. ti@nk bi@nt bo@t́ s **!*(nk, kb, tb)

**(nk, nt)

**(k, t)

b. ti@Nk bi@nt bo@t́ s **!(kb, tb)

*(n)

**(Nk, nt)

**(k, t)

c. ti@N bi@n bo@t́ s **!(Nb, nb)

*(n)

**(k, t)

d. ti@m bi@m bo@t́ s *!(k)

**(n, n)

**(k, t)

e. ti@Nk bi@m bo@t´s *(kb)

**(n, n)

*! * *

☞ f. ti@N bi@m bo@t́ s *(Nb)

**(n, n)

**

Tableau 13: NPA and CS in a phrasal input

In Tableau 13, candidate (d), representing the candidate with full CS andtotally transparent NPA is eliminated by MAX(NONCOR). Candidate (a), thetotally faithful candidate, fatally violates SPREAD(PLACE).9 Candidates (b) and(c), which show assimilation to input adjacent segments, fatally violateSPREAD(PLACE) on the surface. Candidate (e), which poses the greatest threatto the winner, shows only partial CS, and is eliminated due to the extraviolation of *COMPLEX. This leaves the simultaneously surface transparentand surface opaque candidate (f) as the winner.

4. ConclusionTableau 13 shows that a parallel, non-serial theory of phonology such as

OT is capable of accounting for simultaneous surface transparency and surfaceopacity. Most important, though, is that the complexities of the data can behandled without recourse to underspecification, multiple levels, or the cycle.In Catalan, this result is achieved by the high ranking of MAX(NONCOR) whichpreserves input non-coronals (labials, palatals, and dorsals) in the output at theexpense of SPREAD(PLACE). This ranking allows for surface opacity to arisewhen there are input noncoronals. The low ranking of MAX(COR), on the otherhand, ensures that SPREAD(PLACE) will apply fully to segments correspondingto input coronals. In these cases, the surface forms are fully transparent.

9 Actually, since the only established ranking for DEPOOPHRASE is that it ranks above MAXIO, itwould be possible for DEPOOPHRASE to rank above SPREAD in which case, candidates (a) and (b)would be crucially eliminated by DEPOOPHRASE and not SPREAD.


REFERENCES

Beckman, Jill. 1997. Positional Faithfulness. Ph.D. Dissertation, University ofMassachusetts, Amherst.

Benua, Laura. 1995. Transderivational Identity: Phonological RelationsBetween Words. Ph.D. Dissertation, University of Massachusetts, Amherst.

Kiparsky, Paul. 1985. “Some Consequences of Lexical Phonology”.Phonology Yearbook 2. 85-138.

---------. 1999. “Paradigm Effects and Opacity”. Ms., Stanford University.Lombardi, Linda. 1998. “Evidence for MaxFeature Constraints from

Japanese”. Ms., University of Maryland, College Park. [ROA-247-0298]Mascaró, Joan. 1976. Catalan Phonology and the Phonological Cycle. Ph.D.

Dissertation, MIT. [Bloomington, Ind.: Indiana University Linguistics Club,1978.]

McCarthy, John & Alan Prince. 1995. “Faithfulness and ReduplicativeIdentity”. University of Massachusetts Occasional Papers 18: Papers inOptimality Theory ed. by Jill Beckman, Laura Walsh Dickey & SuzanneUrbanczyk, 249-384. Amherst, Mass.: Graduate Linguistic StudentAssociation.

Padgett, Jaye. 1995. “Partial Class Behavior and Nasal Place Assimilation”.Arizona Phonology Conference: Proceedings of the South WesternOptimality Theory Workshop, Volume 5 Coyote Papers ed. by KeiichiroSuzuki & Dirk Elzinga, 145-183. University of Arizona, Tucson.

Paradis, Carole & Jean-François Prunet. 1991. “The Special Status ofCoronals: Internal and External Evidence”. Phonetics and Phonology; v.2.San Diego, Calif.: Academic Press.

Prince, Alan & Paul Smolensky. 1993. “Optimality Theory: Constraintinteraction in generative grammar”. Ms., Rutgers University & University ofColorado, Boulder.

Recasens, Daniel. 1991. Fonètica Descriptiva del Català. Barcelona: InstitutD’Estudis Catalans, Biblioteca Filològica XXI.

---------. 1993. Fonètica I Fonologia. Barcelona: Enciclopèdia Catalana.Steriade, Donca. 1997. Phonetics in Phonology: The Case of Laryngeal

Neutralization. Ms., University of California, Los Angeles.

THE ARTICULATOR GROUP AND LIQUID GEOMETRYIMPLICATIONS FOR SPANISH PHONOLOGY PRESENT AND PAST*

D. ERIC HOLTUniversity of South Carolina

0. IntroductionThe present work reconsiders several aspects of Spanish phonology in light

of recent theoretical advances regarding the internal organization of thesegment. In building on and synthesizing these insights, I propose a novelapproach to the understanding of the ambivalent status of the feature[±continuant] of /l/, whose value is not universally accepted. Indeed, considerthe following statement regarding continuancy from the foundational work ingenerative phonology, The Sound Pattern of English:

In the production of continuant sounds, the primary constriction in the vowel tract isnot narrowed to the point where the airflow past the constriction is blocked; in stopsthe air flow through the mouth is effectively blocked��The characterization of the liquid [l] in terms of the continuant-noncontinuant scaleis even more complicated. If the defining characteristic of the stop is taken (as above)as total blockage of air flow, then [l] must be viewed as a continuant and must bedistinguished from [r] by the feature �laterality�. If, on the other hand, the definingcharacteristic of stops is taken to be blockage of airflow past the primary stricture,then [l] must be included among the stops. (Chomsky & Halle 1968:317-318)[emphasis added; DEH] 1

* I would like to thank the participants at the 30th Linguistic Symposium on RomanceLanguages (University of Florida, 24-27 February 2000), particularly Jim Harris, DylanHerrick, Haike Jacobs, John Lipski, Rafael Núñez Cedeño, Carlos-Eduardo Piñeros and LoriRepetti, as well as two anonymous referees, for helpful observations, data, discussion andreferences. I also thank Laura Walsh Dickey for on-going discussion of liquid geometry andthe implications of it. Of course, none are to blame for any shortcomings of the presentanalysis.1 /l/ has been analyzed as [+cont] in several languages: Walsh Dickey (1997) cites Musey(Shryock 1994), Zoque (Wonderly 1951), and Chipewyan (Li 1946, Chomsky & Halle 1968);Gussenhoven & Jacobs (1998) cite Frisian. Others who have argued that [l] is [+cont] areHarms (1968), Anderson (1974) and Ladefoged (1982) (all cited in Tatò 1981).

D. ERIC HOLT86

The paper is organized as follows: in §1, I present the data of the first partof the paper, the spirantization facts of Spanish, and lay out my theoreticalassumptions regarding the articulator group hypothesis and liquid geometry; in§2, I show how applying the proposed liquid geometry to Modern Spanishallows us to resolve the issue of spirantization, and I review previous accounts;in §3, I discuss further implications for the treatment of several historicalchanges, including delateralization of [λ], formation of ch, intrusive stopformation, simplification of -ns- and other clusters, and vocalization of LateLatin /-l/ along with /-k, -g/. In §4, I offer concluding remarks.

1. Primary data and other theoretical preliminariesThe initial focus of this paper will be spirantization in Spanish. Data is

presented immediately below, and the remainder of this section presents thearticulator group hypothesis and its relevance to a revised liquid geometry.

Restricting ourselves to the distribution of the voiced obstruents instandard dialects of Spanish in Table 1, the generalization to be made is thatstops occur after pause and after homorganic sonorants, while spirants obtainafter a continuant. Thus, [b d g] and [β D γ] are in complementary distribution(and so may be represented as /B D G/, unspecified for [±cont], in the spirit ofLozano 1978).

a. b. c. d. e. f.|| ___ N ___ l ___ V ___ s ___ r ___bote umbote elββββote reβote rezβala, lozβotes arβoldaño un5daño el 55 55daño reDaño dezDe, lozDaños arDegato uNgato elγγγγato aγatas dezγrana, lozγatos erγido

Table 1: Distribution of [b d g] and [β D γ] (standard dialects)(Adapted from Hualde 1989:25)

However, [l] is ambiguous with regard to its value for [±cont], as seen inthe bolded examples above in (c), where /lD/ yields [l5d], but /lB, lG/ yield [lβ,lγ]. Crucially, then, [l] appears before both stops and fricatives.

This matter has occupied the attention of theoretical phonologists for morethan 30 years, going back to Harris (1969). Previous accounts have differed inhow to treat the above data and in their assumptions about the [±cont] status of

/l/ has been analyzed as [-cont] for Basque (Hualde 1989), English (McCawley 1979;

cited in Walsh Dickey 1997), Scots English (Chomsky & Halle 1968), Gallo-Romance (Jacobs1991, following Clements 1987), Korean and Luganda (mentioned in Walsh Dickey 1997; noreferences given). See also Kaisse (1998) for a review of evidence from a wider range oflanguages. (Kaisse arrives at the conclusion that laterals are [-cont].)

THE ARTICULATOR GROUP AND LIQUID GEOMETRY 87

/l/. The most innovative and unorthodox of the approaches is that of Mascaró(1991), who views /l/ as phonologically [+cont] but phonetically [-cont] before/d/. Although this analysis has been sharply criticized for mixing phonetics andphonology and for circularity of argumentation, I vindicate the essentialinsight of it here.

Specifically, I consider the articulator group of Padgett (1992, 1994, 1995),Selkirk (1990) and others, and extend its definitional tenet that the place ofarticulation and stricture of a segment are intimately related and are executedas a single gesture (in the spirit of Browman & Goldstein 1989), with stricturestructurally dependent on place.

Here I reassess the feature geometry proposed by Walsh Dickey (1997) forliquid consonants and apply the articulator group structure in (a) to the liquidgeometry in (b). We thus arrive at the revised structure of /l/ in (c), where bothvalues of [±cont] are phonological and underlyingly present (all structuressimplified for present purposes):

a. (e.g., Padgett 1995) b. (Walsh Dickey 1997) c. (current proposal)

Place /l/ ([liquid, son]) /l/ ([liquid, son]) wgo ! !

LAB COR DOR C-Place C-Place ! ! ! 2 2[±cont] [±cont] [±cont] COR V-Place COR V-Place

1 ! ! !-dist.+ant. DOR [-cont] DOR

![+cont]

Fig. 1: The articulator group and its application to liquid geometry

Ample evidence for the dual articulation of laterals is given by WalshDickey (1994, 1997). Relevant data supporting the separate specification of[±cont] for each articulator are well known facts about syllable-final /l/:2

2 Padgett (1995:ch. 3, �Complex Segments�) makes a similar argument for Kabardian�harmonic clusters� (obstruents), which he argues require independent underlying [±cont]specifications for their multiple articulators. Though he treats the Spanish spirantization facts,he does not appeal to dual specification for /l/, as his work predates Walsh Dickey�s. However,see fn. 7 for a similar insight he pursues.

D. ERIC HOLT88

(1) Changes in syllable-final position [l]:a. by loss of primary COR features (including [-cont]; vocalic DOR

remains, so resulting segment is [+cont]):l > u Belear Catalan, Old Frenchl ~ w Mehri (South Arabian Semitic), Brazilian Portugueseł > o Serbo-Croatianl > ƒ English (no variety specified by Walsh Dickey 1997)(Also l > w in Polish in all positions)

b. by loss of secondary DOR features (including [+cont]; primaryCOR remains, so resulting segment is a [-cont] consonant):l > r Florentine Italian, Modern Greek, Caipira Portuguese,

Andalusian Spanish (presumes that tap [r] is [-cont])l > t Gascon, Aranés (masc. sg. definite article et < Lat. ILLE;

for Gascon, see Montreuil 2000)

2. Application of proposed liquid geometry to Modern Spanish dataHaving motivated this approach to the geometry of [l], I now discuss the

implications of it for a variety of phenomena, occurring both in the Spanish ofthe present day and in the evolution of Spanish through history.

2.1 Spirantization in Standard SpanishGiven the nasal/lateral + obstruent assimilation imperative observed in

Table 1 (b,c), and the rightward directionality of assimilation in continuancy(contrast (a) with (d,e,f); also a[βD]icar, uste[D], e.g.), Spirantization may beformulated maximally simply, as the rightward spreading of [+cont]:

(2) Spirantization in Standard Spanish: (SPIR)Spread [+continuant] (L > R)(Implemented straightforwardly within any theoretical framework.)

a. b. c./s, r, etc./ /B, D, G/ /l/ /B, G/ /l/ /D/

! ! ! ! ! !C-Place C-Place C-Place C-Place C-Place C-Place

! ! rh ! fp|COR LAB, COR, DOR COR V-Place LAB, DOR V-Place COR

! ! ! ! | ![+cont] [øcont] DOR [øcont] [+cont] [-cont]

> [+cont] g > [+cont] [+cont]

= [s, r, etc. +β, D, γ] = [lβ, lγ] = [l5d]


Given the modified articulator group of Figure 1, when place assimilationoccurs, [-cont] necessarily follows the place of articulation (PA) of /N/ andprimary (C-Place) coronal of /l/. Thus, /lD/ > [l5d] (shown in (c)), but /lB, lG/> [lβ, lγ] (where the secondary or V-Place [+cont] of /l/ spreads).3

Stops [b, d, g] result after pause and wherever SPIR has not applied, as[-cont] is universally unmarked for obstruents:4

(3) Default assignment for [±continuant]:[-sonorant] > [-continuant](Kenstowicz 1994:64)/|| Bote, Daño, Gato/ > [|| bote, daño, gato] (from Table 1 (a))

Spirantization (SPIR) is now phonologically motivated and internallyconsistent, as the alternations in Table 1 are understood as due to assimilationin continuancy, and all instances of [+cont] participate, with universal defaultrules filling in missing values when SPIR does not apply.

2.2 Comparison with previous approachesIn this section I review the analyses and assumptions of previous

approaches and show that we need not appeal to a Linking Condition to eitherblock Spirantization (Harris 1984, Padgett 1995) or to enable it (Hualde 1989),nor to rule ordering of place assimilation and continuant spreading (as in allnon constraint-based accounts) to account for the data; these are inextricablyintertwined phenomena, and the former implies the latter, which falls outentirely from the geometry proposed here, seen above.

The current approach shares aspects of many previous studies, but isclosest in spirit to those of Harris (1984) and Padgett (1995), though it differsfrom these in important ways. Let me first present a synthesis of previousresearchers� assumptions regarding various theoretical issues.

3 I assume that when the C-Place COR nodes are merged, the dental features of [d] retain the[-cont] value of [l]. Alternatively, [-cont] is assigned by a modified version of default (3). See(6b) below.4 This also explains why when obstruents disagree in place of articulation, the continuants aredisallowed (*i[nβ]ierno, *la[mD]a, *albu[mγ]rueso, etc., Harris 1985): the stops obtain viauniversal default. I thank Jim Harris for alerting me to this additional data. Further, for thosedialects where [lb, lg, rd, rb, rg, sd] occur (e.g., those of Mexico City and Bogotá; see Harris1984, Amastae 1995, Widdison 1997 and Carreira 1998 for discussion), the stops likewiseresult from (3). Finally, the difference among most Argentine Spanish speakers noted byHarris and Kaisse (1999) regarding the alternation between [d�] and [�] (appearing afternoncontinuants and continuants, respectively) appears to follow from (3) as well.

D. ERIC HOLT90

First, let us consider the underlying status of the voiced obstruents andtheir ultimate realization as either stops or fricatives. Most researchers (Lozano1978, Goldsmith 1981, Harris 1984, Hualde 1989, Branstine 1991) haveassumed that these are underspecified for [±continuant] and are representedunderlyingly as /B, D, G/.5 I follow this approach here, as it allows for amaximally simple analysis of the stop-continuant alternations of Table 1 as afeature-filling operation, complemented by the universal default rule of [-son]> [-cont] in (3), assuming the modified liquid geometry of Figure 1. WhileHarris (1984) also assumes the latter default rule to account for those stops thatare not place-linked, Hualde (1989) must stipulate a separate rule of post-pausal stop formation that complements the Spanish-specific default rule of[+cont] assignment. Likewise, Carreira (1998) seems to adopt this into heroptimality-theoretic account via the constraint VOICED-CONTINUANT (�if[+voice] then continuant�). However, given that she is silent on the matter ofthe underlying representation of the voiced obstruents, it is unclear howCarreira can account for the post-pausal stops. Padgett (1995) avoids the issueof default assignment by assuming (without argument) /b, d, g/, andGnanadesikan (1997) does so as well, though there seems to be no clearstatement in this regard. For the latter researchers, Spirantization consequentlymust be feature-changing.

Regarding the underlying [±cont] status of Spanish /l/, there has likewisebeen much debate. /l/ is assumed to be [-cont] by Goldsmith (1981), Tatò(1981), Hualde (1989), Branstine (1991) and Carreira (1998), but [+cont] byHarris (1984) and Kenstowicz (1994:36), who notes that �the inconsistentbehavior of the lateral leaves the [±continuant] status of [l] unresolved.� Amixed or ambiguous position in addition to that of the oft-criticized Mascaró(1984, 1991) is that of Padgett (1995), who claims that /l/ is �basically�[+cont], but is compatible with either value; that is, otherwise [+cont] [l] maytake [-cont] by virtue of coronal linking to following [-cont] [d].6, 7

Researchers have also offered differing characterizations of the best way totreat the stop-continuant alternations, specifically, whether [β, δ, γ] result froma genuine process of spirantization. For Harris (1984), Padgett (1995) and 5 Under Lexicon Optimization in Optimality Theory (Prince & Smolensky 1993, Inkelas1994), /B, D, G/ are favored, as Spirantization would be feature-changing otherwise, thusincurring gratuitous violations of FAITHFULNESS constraints.6 Gnanadesikan (1997) essentially agrees, though for her the feature in question is really theternary-valued Consonantal Stricture (CS).7 Interestingly, Padgett (1995: 151 fn.12) suggests for [l] that [approximant] may be borne by aseparate articulator, DOR, argued to be part of laterals. He does not directly connect this tospirantization, though. Nonetheless, this is very much in the spirit of the present analysis.


Gnanadesikan (1997), the answer is �yes�. These researchers believe that thesealternations are due to a progressive assimilation in stricture (airflow); this is anatural assumption, and one made here as well. However, for Hualde (1989),Branstine (1991) and Carreira (1998), on the other hand, the fricatives resultfrom default assignment of [+cont], a position at odds with assumptionsregarding universal markedness.

As for the realizations [b, d, g], Harris (1984) argues that in these cases toothere is progressive assimilation in stricture, and [-cont] propagates rightwardfrom the nasal and lateral to yield the stops, and for Padgett (1995), this fallsout naturally under the articulator group (also assumed here). For both authors,place-linked [l5d] escapes Spirantization due to the Linking Condition, underwhich association lines in structural descriptions are interpreted as exhaustive.8For Hualde (1989), a rule of stop formation via spreading of [-cont] affectsonly place-linked segments, with the nasal and lateral spreading their [-cont]value to the following underspecified voiced obstruent; as mentioned above,his special rule of post-pausal stop formation yields the other instances of [b,d, g], and further, [lβ, lγ] result from his Spanish-specific default rule of[+cont] insertion.9

2.3 Summary of the present workThere is a phenomenon of �spirantization�, and it is characterized as the

rightward spreading (understood nonderivationally, if one pleases) of thefeature [+cont] (from vowels, glides, fricatives, and the V-Place of laterals).Nasal/Lateral Assimilation interacts with SPIR in that /B, D, G/ are providedthe specification [-cont] because they share a Place node with the precedingconsonant, which entails that they will share stricture features. Nasals andlaterals are alike at their primary articulation, being [-cont].

Like Harris, I argue that /l/ is [+cont], but at its secondary articulator node,and because I adopt Walsh Dickey�s feature geometry and apply Padgett�s 8 This definition of the Linking Condition, also called the Linking Constraint or UniformityCondition (Kenstowicz 1994:413), comes from Hayes (1986), cited in Padgett (1995:148).However, because Harris� (1984) rule does not specify the Place node, Spirantization shouldapply even when there are place-linked structures, and we should expect *[l 55 55D]. Padgettremedies this by including Place in his rule of Spirantization, but notes (1995:149), however,that the matter is �patched up, but [the] question [is] still unexplained�. For Harris, [d] after [l]arises via universal default; for Padgett, [d] retains its underlying stop value.9 Goldsmith (1981) argues for a �Minimal Distance Principle� that yields results similar toHualde�s, in that a place-linked nasal or lateral spreads its [-cont] value, while when there is noassimilation, the closest vowel spreads its [+cont] value (following, [al.βa], as regressiveassimilation; or preceding, [us.teD], as progressive assimilation). [|| b, d, g] obtain via aseparate rule, though it is unclear why [|| β, D, γ] do not obtain via [+cont] spreading from thefollowing vowel.

D. ERIC HOLT92

own articulator group hypothesis to it, I avoid the need to invoke the dubiousand stipulative Linking Condition. To invert the phrasing of Padgett, we mightthen say that [l] is �basically� [-cont] (as airflow is blocked at its primaryconstriction), but that its secondary/vocalic [+cont] has the same effect as thevocalic place of vowels and glides. That is, Mascaró was on the right track: [l]is both [+cont] and [-cont]; however, both are phonological. That is, [l5d, n5d,mb, Ng] obtain via assimilation in place,10 where assimilation in manner is anautomatic consequence of the articulator group. Additionally, [|| b d g] surfaceby universal default, and thus there is no need for [+cont] as a Spanish-specificdefault rule. Finally, [lβ, lγ] are due to spirantization (SPIR); markednessconsiderations rule out place assimilation, and there is assimilation in mannerfrom the preceding [+cont] segment, here from the V-Place of [l].11

3. Implications for several historical changesThe proposal also accounts for several historical changes.

3.1 �Delateralization�In early Old Spanish [λ] (< [lj]) became [d�] (ALIUM �garlic� > [aλo] >

[ad�o] (> [a�o] > MSp. [axo] ajo)). Under the assumptions made here, thismay be seen as simplification of the complex palatal via loss of [liquid] (andconcomitantly of [+sonorant]); the result is a multiply-articulated voicedpalatal obstruent that is both [- and + cont], i.e., affricate [d�].

10 Perhaps because Place features are not licensed in the coda, as argued by Carreira (1998).11 However, as Harris (1985) notes, the specific class of triggering segments appears to varyaccording to dialect. That is, in dialects where stops occur after /r/ or where the stops [b, g]occur after /l/, it appears that we must restrict the process of Spirantization to the spreading of[+cont] from vowels only; the universal default rule (3) inserts [-cont]. Other dialectdifferences, including variation in the Argentine data of fn. 4, would presumably be handled inan analogous fashion. I leave this matter open here. I thank Jim Harris for making me aware ofthe Argentine data and for discussion of them.


(4) �Delateralization�:12, 13

λ = Root (simplified)[cons, son, liquid]

goC-Place [+voice]2

COR V-Place! !

[-cont] DORg

[+cont]

3.2 Development of Latin Cl (voiceless C plus l) clusters to ch [tš]Following the approach given here, ch may result from a stage of *Cλ 88 88 (<

/p, t, k + l/, where �delateralization� and assimilation in voicelessness yield [t�],e.g., MACULA �stain� > *ma(n)kλa > mancha.

(5) The creation of [t�]:C λ = Root (simplified)

[cons, son, liquid]fi

[-voice] [+voice] C-Place2

COR V-Place! !

[-cont] DOR!

[+cont]

Here, by assimilation of [-voice], *Cλ yields [Cλ 88 88], analyzed in turn by thelistener as [t�], and lexicalized as /t�/. (See Holt 1997, 1998 for details.)

3.3 Intrusive Stop FormationWhat happens when the sonorant does try to assimilate to a following

[+cont] segment? Although the sonorant generally respects the place spreadingimperative, marking conditions will disfavor success here, as according to the

12 �Delateralization� is actually a misnomer if Walsh Dickey (1997) is right in eliminating[lateral] from the feature inventory; instead, �lateral� = doubly-articulated nonnasal sonorant.See Keating 1988 and Lipski 1989 for corono-dorsal structure of palatals.13 This also accounts for those modern dialects that have eliminated [λ] for affricate [d�] orfricative [�, �], or for those where these sounds alternate freely. In other dialects [λ] may havebecome [j] first, with later (or stylistic) fortition to [d�, �, �]. I thank John Lipski for a questionregarding this point that has obliged me to clarify my views.

D. ERIC HOLT94

articulator group this would result in a continuant nasal or lateral (at its mainarticulation).

(6) Marking conditions against sonorant continuants:a. *[nasal, +consonantal, +continuant]

�Continuant nasal consonants are disfavored�(Branstine 1991:10; Padgett 1995:147; Carreira 1998:147)

b. *[lateral, +consonantal, +continuant]14

�Continuant laterals are disfavored�(Branstine 1991:10; Carreira 1998:147)

Instead of violating these conditions, other consequences result. One suchcase is the well-known epenthesis of a stop consonant before [+cont] trill [r#]:

(7) Sp. future sal(i)r + á > saldrá; conditional sal(i)r + ía > saldría(Also with nasals: pon(e)r + á > pondrá, and pon(e)r + ía > pondría;similar cases are cited in Padgett 1995, e.g., Kikuyu �hardening�.)

/l/ (d)15 / r #/ ! !

C-Place C-Place 2 !

V-Place COR COR! ! !

DOR [-cont] [+cont]!

[+cont]

There now appears to be no need to invoke �bad syllable contact� as atheoretical prime to motivate the intrusive stop formation (Wetzels 1985,Clements 1987, Martínez-Gil 1991). Instead, epenthesis falls out from otherconstraints on the grammar, those of spreading and marking conditions. 14 This actually must be understood as applying to the C-Place COR node of laterals, since, as Ihave argued here, /l/ is indeed [+cont] at its V-Place DOR node. Further, if Walsh Dickey(1997) is right regarding the invalidity of the feature [lateral], (6b) must be reformulated suchthat it applies only to doubly-articulated nonnasal sonorants.15 Given that the affricate-like structure that results is singly place-articulated, the transitionalelement is not �lateral� and so is not interpreted as [+sonorant] (see fns. 12 and 14); that is, it isinterpreted as [d] (later lexicalized as /d/). The intrusive [d] after a nasal (pon(e)r + á >pondrá) presumably arises in a similar fashion. Contrast this with the linking of /lD/ in (2c), aswell as /n + B, D, G/; given that /B, D, G/ are underspecified for [±cont], there is no affricate-like structure created and no transitional element results.


3.4 Simplification of clusters of nasal/lateral plus [+continuant]Another possible resolution of the interaction between the spreading

imperative and the marking conditions is the deletion of the potentiallyoffending segment, thus vacuously satisfying the marking conditions of (6):

(8) a. /-ns-, -nf-/ > /-s-, -f-/ (MENSA > mesa, INFANTE > OSp. iffante)(Parallel cases cited in Padgett 1995, e.g., Lithuanian, Zoque,Malayalam.)

b. Ptg. future and conditional forms sal(i)r + á > sairá; sal(i)r +ía > sairia; etc. (Also with nasals: pon(e)r + á > porá; pon(e)r+ ía > poria; terei, teria; etc.)

/n/ > ø /s, f/ ! !

C-Place C-Place b |

COR COR, LAB! !

[-cont] [+cont]

This loss of /n/ in (a) is due to the interaction of the nasal placeassimilation imperative with the marking condition against continuant nasals,which would result under the articulator group. To avoid violation of thisconstraint, /n/ is lost, presumably passing through a stage where nasality wasretained on the preceding vowel, subsequently lost.

For the Portuguese cases in (b), we need to invoke the modified version ofmarking condition (6b). Here, the �lateral� place assimilation imperative urgesassimilation of the C-Place of /l/ to the following [+cont] segment, but the highranking of the marking condition favors the loss of /l/. Data of this type thusprovide further evidence in favor of the [-cont] status of /l/.16

3.5 Vocalization of /-l/ along with /-k, -g/Finally, the approach taken here also allows us to understand why it is that

/l/ patterned with the syllable-final velars [-k, -g] in Late Latin, whichvocalized to [j] (or [w], depending on the preceding vowel):17

16 I leave open why it should be that both Spanish and Portuguese lose the nasal before theobstruents /s, f/ but differ in their resolution of the marking condition when a [+cont] nasalwould arise before sonorant r.17 I thank Rafael Núñez Cedeño for raising the issue of how /-l/ is able to become either [-j] or[-w]. Upon considering these and other Romance data, it is clear that the surrounding contextplays a decisive role: when the preceding vowel is [a], [-w] results; in other cases, the

D. ERIC HOLT96

(9) a. /-kt-/ OCTO �eight� > Hispano-Romance and ModernPortuguese oito; NOCTE �night� > noite/-gr-/ INTEGRU �whole� > enteiro

b. /-ulC-/ MULTU �much� > muito; VULTURE �vulture� > buitre;CULTELLU �knife� > OSp. cuchiello

c. /-alC-/ ALTERU �other� > outro; ALTARIU �hill� > otero

It has been unsatisfactorily explained so far why the more sonorous /-l/should have been affected like /-k, -g/ (which became [-x, -γ], then [-j]).Previous approaches (e.g., Penny 1991, Holt 1997, 1999) had to stipulate that/-l/ was velarized, and so underwent vocalization as did the pure velars.However, if we understand the motivation for vocalization to be theelimination of [-cont] features from coda position as a reaction to a constraintrequiring increasingly sonorous moraic elements, then the change from /l/ to[j] may be seen as the suppression of primary C-Place COR due to itsdominating the feature [-cont], which contributes less sonority than [+cont].This is thus a more principled account of this change.

4. Summary and conclusionsIn appealing to recent insights on the internal structure of the segment,

namely the articulator group hypothesis and liquid geometry, this work shedsnew light on several long-standing recalcitrant issues in Spanish phonologyspecifically, including spirantization and a host of historical developments, andphonological theory more generally, and promises much for analyses of otherlanguages where /l/�s behavior with regard to [±continuant] has been seen asambivalent or contradictory in the past.18 These are welcome results.

following consonant, usually [t] or [s], seems to require another COR articulation, here thepalatal [-j]. (That is, featural agreement/assimilation appears to play a role.) I must assume thatwhen C-Place COR is affected/suppressed, V-Place DOR may survive (as in (1a) and (9c)), orthat the COR articulation may be preserved in the V-Place, but with the latter�s [+cont]specification (as in (9a,b)).18 It would also be interesting to revisit the analyses of [±continuant] status of [l] in otherlanguages (see also Kaisse 1998); Havana Liquid Assimilation (Harris 1985), Catalan StopAssimilation, Cibaeño liquid gliding (Guitart 1976, Núñez Cedeño 1997), /-l, -r/ neutralizationin Andalusian and Caribbean Spanish; the historical change /ld/ > /ll/ in Aragonese(Menéndez-Pidal 1950:294-6), South Central Italian, inner Apulia and Modern Roman (callofor caldo, sollato for soldado; Rohlfs 1949:400) (all cited in Tatò 1981; for Aragonese, seealso Lapesa 1981 and Zamora Vicente 1989). This may perhaps be explained as assimilationof the lateral to the obstruent of the entire Place, not just the C-Place COR. Recall that underWalsh Dickey (1997), �lateral� really means �complex corono-dorsal�.


REFERENCES

Amastae, John. 1995. �Variable spirantization: Constraint weighting in threedialects�. Hispanic Linguistics 6/7. 265-285.

Branstine, Zoann. 1991. �Stop/spirant alternations: On the representation ofcontrast�. Studies in the Linguistic Sciences 21.1.1-24.

Browman, Catherine & Louis Goldstein. 1989. �Articulatory gestures asphonological units�. Phonology 6.2.201-251.

Carreira, Maria M. 1998. �A constraint-based approach to Spanishspirantization�. Selected Papers from the XXVI Linguistic Symposium onRomance Languages ed. by José Kena & Esthela Treviño, 143-157.Amsterdam & Philadelphia: John Benjamins.

Chomsky, Noam & Morris Halle. 1968. The Sound Pattern of English. NewYork: Harper & Row.

Clements, G. Nick. 1987. �Phonological feature representation and thedescription of intrusive stops�. Proceedings of the 23rd Chicago LinguisticSociety 23 ed by A. Bosch, B. Need & E. Schiller, 29-50. Chicago, Ill.:Chicago Linguistic Society.

Gnanadesikan, Amalia E. 1997. Phonology with Ternary Scales. Ph.D.Dissertation, University of Massachusetts, Amherst.

Goldsmith, John. 1981. �Subsegmentals in Spanish phonology: anautosegmental approach�. Selected Papers from the 9th LinguisticSymposium on Romance Languages ed. by William W. Cressey & DonnaJo Napoli, 1-16. Washington, D.C.: Georgetown University Press.

Guitart, Jorge. 1976. Markedness and a Cuban Dialect of Spanish.Washington, D.C.: Georgetown University Press.

Gussenhoven, Carlos & Haike Jacobs. 1998. Understanding Phonology. NewYork: Oxford University Press.

Harms, Robert T. 1968. Introduction to Phonological Theory. EnglewoodCliffs, New Jersey: Prentice Hall.

Harris, James W. 1969. Spanish Phonology. Cambridge, Mass.: MIT Press.---------. 1984. �La espirantización en castellano y la represenatción fonológica

autosegmental�. Estudis Gramaticals 1.149-167.---------. 1985. �Autosegmental phonology and liquid assimilation in Havana

Spanish�. In Selected Papers from the XIIIth Linguistic Symposium onRomance Languages ed. by Larry D. King & Catherine A. Maley, 127-148.Amsterdam & Philadelphia: John Benjamins.

Another open question is the fuller consideration of the argument that [l] is like an

affricate in having both a stop and a continuant articulation and, more broadly, the differencebetween (obstruent) affricates and laterals. I leave these matters for further research.

D. ERIC HOLT98

Harris, James W. & Ellen M. Kaisse. 1999. �Palatal vowels, glides andobstruents in Argentinian Spanish�. Phonology 16.117-190.

Holt, D. Eric. 1997. The Role of the Listener in the Historical Phonology ofSpanish and Portuguese: An Optimality-Theoretic Account. Ph.D.Dissertation, Georgetown University. [ROA #278-0898]

---------. 1998. �The role of comprehension, reinterpretation and theUniformity Condition in historical change: The case of the development ofCl clusters from Latin to Hispano-Romance�. Proceedings of the WesternConference on Linguistics (WECOL) 1996 ed. by Vida Samiian, 133-148.Fresno, Calif.: Department of Linguistics, California State University,Fresno.

---------. 1999. �The moraic status of consonants from Latin to Hispano-Romance: The case of obstruents�. Advances in Hispanic Linguistics:Papers from the Second Hispanic Linguistics Symposium ed. by JavierGutiérrez-Rexach & Fernando Martínez-Gil, 166-181. Somerville, Mass.:Cascadilla Press.

Hualde, José I. 1989. �Procesos consonánticos y estructuras geométricas enespañol�. Asociación de Lingüística y Filología de América Latina 1.7-44.

Inkelas, Sharon. 1995. �The consequences of optimization for under-specification�. North Eastern Linguistic Society 25 ed. by J. Beckman,287-302. Amherst, Mass.: GLSA. [ROA #40-1294].

Jacobs, Haike. 1991. �A nonlinear analysis of the evolution of consonant + yodsequences in Gallo-Romance�. Canadian Journal of Linguistics/La Revuecanadienne de Linguistique 36.27-64.

Kaisse, Ellen M. 1998. �Laterals are [-cont]�. Paper presented at EncuentroLingüístico del Noroeste held at Hermosillo, Mexico, November 1998.

Keating, Patricia A. 1988. �Palatals as complex segments: X-ray evidence�.UCLA Working Papers in Phonetics 69. 77-91.

Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Cambridge,Mass.: Blackwell.

Lapesa, Rafael. 1981. Historia de la lengua española. Madrid: Gredos.Lipski, John M. 1989. �Spanish yeísmo and the palatal resonants: Toward a

unified analysis�. Probus 1.211-223.Lozano, Maria del Carmen. 1978. Stop and spirant alternations: fortition and

spirantization processes in Spanish phonology. Bloomington, Ind.: IndianaUniversity Linguistics Club.

Martínez-Gil, Fernando. 1991. �The insert/delete parameter, redundancy rules,and neutralization processes in Spanish�. Current Studies in SpanishLinguistics ed. by Héctor Campos & Fernando Martínez-Gil, 495- 571.Washington, D.C.: Georgetown University Press.


Mascaró, Joan. 1984. �Continuant spreading in Basque, Catalan and Spanish�.Language Sound Structure ed. by Mark Aronoff & R.T. Oehrle, 287-298.Cambridge, Mass.: MIT Press.

---------. 1991. �Iberian spirantization and continuant spreading�. CatalanWorking Papers in Linguistics. 167-179. Barcelona: Universitat Autònomade Barcelona.

Montreuil, Jean-Pierre. 2000. �Optimal allomorphs of the Gascon definitearticle�. Paper presented at the 30th Linguistic Symposium on RomanceLanguages, held at the University of Florida, Gainesville, February 2000.

Núñez Cedeño, Rafael. 1997. �Liquid gliding in Cibaeño and feature geometrytheories�. Hispanic Linguistics 9:1.143-164.

Padgett, Jaye. 1992. Stricture and nasal place assimilation. LRC-92-08. SantaCruz, Calif.: Linguistics Research Center.

---------. 1994. �Stricture and nasal place assimilation�. Natural Language andLinguistic Theory 12.465-513.

---------. 1995. Stricture in Feature Geometry. Stanford, Calif.: CSLI.Penny, Ralph J. 1991. A History of the Spanish Language. Cambridge:

Cambridge University Press.Prince, Alan and Paul Smolensky. 1993. Constraint Interaction in Generative

Grammar. Ms., Rutgers University & University of Colorado, Boulder.Selkirk, Elizabeth O. 1990. �A two-root theory of length�. Papers in

Phonology: University of Massachusetts Occasional Papers in Linguistics14 ed. by E. Dunlap & J. Padgett, 123-71. Amherst, Mass.: GLSA.

Tatò, Pasquale. 1981. �Romance phonological evidence for the noncontinuantstatus of /l/�. Selected Papers from the 9th Linguistic Symposium onRomance Languages ed. by William W. Cressey & Donna Jo Napoli, 69-82. Washington, D.C.: Georgetown University Press.

Walsh Dickey, Laura. 1994. �Representing laterals�. Proceedings of the 25th

Meeting of the North Eastern Linguistic Society ed. by J.N. Beckman, 535-550. Amherst, Mass.: GLSA.

---------. 1997. The Phonology of Liquids. Ph.D. Dissertation, University ofMassachusetts, Amherst.

Wetzels, W. Leo. 1985. �The historical phonology of intrusive stops: A non-linear analysis�. Canadian Journal of Linguistics/La Revue canadienne deLinguistique 1.3-54.

Widdison, Kirk A. 1997. �Physical parameters behind the stop-spirantalternation in Spanish�. Southwest Journal of Linguistics 16.73-84.

Zamora Vicente, Alonso. 1989. Dialectología española. Madrid: Gredos.

INTONATION IN SPANISHAND THE OTHER IBERO-ROMANCE LANGUAGES

OVERVIEW AND STATUS QUAESTIONIS

JOSÉ IGNACIO HUALDEUniversity of Illinois at Urbana-Champaign

0. IntroductionFor several decades since the publication of the outstanding pioneering work

of Navarro Tomás (1944), progress in our understanding of Spanish intonationwas relatively slow, as is evident in the overview offered in Kvavic & Olsen(1974) (cf. also Kvavic 1978). In the last decade, however, great strides have beenmade in the study of Spanish intonation. An important reason for this change hasbeen the increasing availability of sophisticated computer-based programs for thestudy of intonation. Regarding the analytical framework, most recent work onSpanish intonation takes as its theoretical foundation the Autosegmental-Metrical(AM) model (Pierrehumbert 1980, Pierrehumbert & Beckman 1988, Beckman &Pierrehumbert 1986, Ladd 1996). Work which applies this model to Spanishintonation includes Sosa (1991, 1999), Prieto (1998b), Prieto et al. (1995, 1996),Nibert (1999, 2000), Face (2000), among others. Outside of this paradigm wefind, among others, Chela-Flores (1994) and Alcoba & Murillo (1998). The AMframework has also been applied with interesting results to Portuguese (Frota1997, 1998) and to Catalan (Prieto 1995, 1997, 1998a). Thanks to the work ofthese and other researchers, it is now possible to examine differences andsimilarities across Ibero-Romance languages and dialects in detail and also topoint out similarities and differences with respect to other languages.

In this paper, rather than chronologically reviewing work on intonation inIbero-Romance in the last two or three decades, we will examine some specificissues in the intonation of Spanish and its close relatives that have arisen in recentwork and are currently controversial. In particular we will focus on three issues:the phonological analysis of rising pitch-accents, the nature and phonologicalcharacterization of final declarative contours, and intonational phrasing. These are

JOSÉ IGNACIO HUALDE102

fundamental issues in the sense that they concern the analysis of the most basicutterances.

1. Phonological analysis of rising accents in SpanishBy intonation we understand the overall melody of an utterance, as reflected

primarily by its tonal or F0 contour. To account for the linguistic productivity ofintonation, it must be the case that intonational contours can be decomposed intosmaller units or primitives, in the same ways as the segmental component ofutterances. In the AM model of Pierrehumbert (1980) (and more referencesabove), which is currently the most influential theoretical framework ofintonational analysis, the claim is made that in languages like English or Spanishonly certain points in the utterance are phonologically specified for tone, the restof the contour being filled in by phonetic interpolation between tonally-specifiedpoints. In particular, in languages like Spanish or English, tonal events areassociated with either stressed syllables or phrasal-boundaries at the phonologicallevel.

Tones associated with stressed syllables are referred to as pitch-accents.These can be monotonal, T*, or bitonal, T*+T or T+T*, where T ranges overH(igh) and L(ow) and the asterisk indicates which of the two tones is associatedwith the stressed syllable (the other, unstarred, tone closely precedes, in the caseof T+T*, or follows, in the case of T*+T; e.g.: H*+L is a high tone associatedwith the stressed syllable followed by a low tone; H+L* is a low tone on thestressed syllable preceded by a high tone, etc.). Boundary tones are indicated asT% (that is, H% or L%). A special type of boundary tones are phrasal tones, T-

(H- or L-), which may occur at the end of an intermediate phrase and also betweenthe last pitch-accent of the intonational phrase and a final boundary tone (toproduce a greater number of finalities than the two boundary tones would allow).More will be said about phrasal tones below.1

It is now abundantly clear that although stressed syllables function asanchoring points for intonational pitch-accents in Spanish, these pitch-accents donot have an invariant shape regardless of context, say a tonal peak on the stressedsyllables; nor, for that matter, do all stressed syllables receive a pitch-accent.Giving all stressed syllables a high tone, as one could do in a very simplisticspeech synthesis model, produces patterns that sound highly unnatural in Spanish.Consider, on the other hand, the contour in Figure 1, which represents a simple 1 Sosa (1991, 1999) analyzes Spanish intonation without making use of phrasal tones. Nibert(2000), on the other hand, argues that these tones must be postulated for an adequate analysis ofseveral intonational contrasts in Spanish.

INTONATION IN SPANISH 103

declarative sentence in Spanish (Syllables with lexical stress are underlined. Thetext of the example is taken from Sosa 1999):

Fig.1: le dieron el número de vuelo �they gave him the number of the flight�

Something that contributes to the naturalness of this example is the fact thatthe non-utterance-final stressed syllables (the prenuclear accents) do not have anF0 maximum. Rather, there is an abrupt rise over the stressed syllable, but thepeak is actually realized on the posttonic. The frequent displacement of non-utterance-final accentual peaks, which was already noticed by Navarro Tomás(1974[1944]:49), can now be taken as an established fact of Spanish intonation(Fant 1984, Garrido et al. 1993, Mota 1995, 1997, Llisterri et al. 1995, Prieto etal. 1995, Sosa 1999, Face 2000, Nibert 2000). There is substantial agreement thatthe displacement of the accentual peak to the posttonic syllable represents theunmarked case for accented words in nonfinal position in declarative sentences.Where there is less agreement is in the phonological analysis of these facts. Thequestion is whether displaced and non-displaced accentual peaks should be giventhe same phonological representation.

Sosa (1999) and Face (2000) use the label L*+H for the configuration with arise from a valley at the very beginning of the stressed syllable to a peak on theposttonic. The L tone in this bitonal pitch-accent bears the asterisk because it isthe beginning of the rising contour, the valley, that is realized within the stressedsyllable, not the peak. Face (2000) further proposes the existence of aphonological contrast between this L*+H pitch-accent, which he takes ascharacteristic of non-final or prenuclear accents in Castilian Spanish, and adifferent pitch-accent, L+H*, with the peak realized within the stressed syllable,which would signal the nuclear accent of the utterance (either the final accent or anon-final constituent with narrow focus). Thus, in the analysis proposed in Face


(2000), in the example in Figure 1, the words dieron and número would bear aL*+H pitch-accent, whereas the nuclear accent on vuelo would be L+H*. Thenuclear accent is followed by a final boundary L%.2

Face (2000) argues that this is a categorical phonological distinction. Anargument for this view is that placing a L+H* accent (with the peak on thestressed syllable) on a non-utterance-final word would result in making this anuclear accent, the interpretation being narrow focus on this word.

Prieto (forthcoming) and Nibert (2000), on the other hand, defend the viewthat whether the peak is displaced to the posttonic or not, phonologically we havesimply a H* accent. In this analysis, all three accented words in Figure 1 wouldbear a H* pitch-accent:

(1) Competing analysesle dieron el número de vuelohim they-gave the number of flight�they gave him the number of the flight�

Analysis A, le dieron el número de vuelo L*+H vs. L+H*: | | |

L*+H L*+H L+H* L%

Analysis B, le dieron el número de vuelo all rising accents are H*: | | |

H* H* H* L%

In Analysis B, the possible displacement of the peak towards the right is to beconsidered not the result of a binary choice between two distinct and contrastivepitch-accents, but a gradual phenomenon conditioned by such factors as theproximity of word- and phrasal boundaries. That is, the accentual H* will movetowards the right within the syllable or across syllable boundaries �if there isenough room� before a prosodic boundary or another tonal specification.

Prieto (p.c.) points out that there is a very strong tendency for accentual peaksto remain within word boundaries (cf. Llisterri et al. 1995) and that, consequently,in the case of phrase-medial oxytones, the peak is generally realized within thestressed syllable. For instance, in an example such as le darán el número de vuelo�they will give him/her the number of the flight� the accentual peak of darán willtend to occur within the stressed syllable, instead of being displaced to the 2 For the record, this is also the analysis I used in Hualde (1999), although now I have somedoubts about its correctness, is explained in the text. Sosa (1999) uses the labels H* and L* forfinal accents, as will be discussed in section 2.


posttonic as with paroxytonic dieron in the same context in Figure 1. If thisfinding is confirmed by further research, it would appear to present a seriousproblem for Analysis A. At the very least, it would perhaps require aneutralization between the two proposed pitch-accents in oxytones (to keep thepitch-accent within a word boundary).

Face�s (2000) and Sosa�s (1999) use of the label L*+H also seems to besomewhat at variance with the definition of this contour assumed by otherresearchers: �L*+H is an accent contour that is low for a good portion of theaccented syllable and then rises sharply, often into the following unstressedsyllable if there is one� (Ladd 1996:83). This definition would seem to be directlyapplicable to a contour like the so-called �rising accent� of Serbo-Croatian(Godjevac 2000, Smiljanić & Hualde 2000), but much less so to the Spanishexamples that we are considering.

If we adopt Ladd�s distinction between association and alignment, the fact thatthe peak is not realized within the stressed syllable is not a decisive reason todiscard the label H* for accents with displaced peaks:

The fact of association entails no specific predictions about alignment: if a H tone isassociated with a given prominent syllable, we may expect to find a peak of F0somewhere in the general vicinity of the syllable, but the peak may be early in thesyllable or late, and indeed it may be outside the temporal limits of the syllablealtogether. For example, it is particularly common in accented syllables at thebeginning of an utterance to see the high F0 peak aligned in time with the followingunstressed syllable (Ladd 1996:55).

Let us return now to the fact mentioned above that having a peak within thestressed syllable with fall on the posttonic in a non-final word, as in Figure 2 ledieron el número, results in a different interpretation, namely narrow focus (�theyGAVE him the number�). Isn�t this evidence for a phonological contrast betweenL*+H and L+H*, as argued by Face (2000)?


Fig. 2: le dieron el número �they GAVE him the number�Nibert (2000) postulates that in this case there is an intermediate-phrase L-

tone immediately following the H* of the first accent (Ladd 1996:96-98 offers asimilar analysis for the very similar narrow focus contour in English). Theabsence of displacement here would thus be due to the presence of a phrasal-toneafter narrowly-focused words. Notice that narrow focus on a non-final word is notsimply a matter of peak-alignment with respect to the stressed syllable. Anotherimportant effect is the compression of the range or deletion of any accents furtherto the right, as can be seen in Figure 2.

I tend to agree with Prieto (forthcoming) and Nibert (2000) in thinking that wedo not have a phonological contrast between two pitch-accents L*+H and L+H*in examples like the ones we have examined so far. On the other hand, I think thatrepresenting all these accents as H* is not adequate either. Clearly what typicallycharacterizes accented syllables in Spanish declarative sentences is an abrupt risein pitch across the stressed syllable. The peak may or may not be realized withinthe syllable. In fact, the impression of increasing pitch may be better conveyed bydelaying the peak when there is �enough room�. On the other hand, the presenceof a tonal valley (a L tone) at the very beginning of the stressed syllable appearsto be a quite consistent fact (Prieto et al. 1995). It seems to me that it is importantto incorporate this syllable-initial L tone to the phonological definition of theaccent (cf. Arvaniti & Ladd 1995, Arvaniti et al. 1998 for prenuclear risingaccents in Greek). The pitch-accent may be defined as (L+H)* with both tonesassociated (in Ladd�s sense) with the stressed syllable. The preferred alignment ofvalley and peak within this rising contour would be the following:

(2) Phonological representation of rising accents in Spanish *

[.....(σ)....] word | L+H

In (2), the L tone is linked to the beginning of the stressed syllable, whichallows the maximization of the rising contour throughout this syllable, and the Htone is more loosely associated before the end of the word domain.

Postulating a simpler monotonal H* accent, as Prieto (forthcoming) andNibert (2000) do, would fail to capture the important point that generally there isa valley at the very beginning of the stressed syllable. Let us consider the examplein Figure 3, Maria-Elena me la dio �M-E gave it to me�, where the first accentoccurs further to the right than in previous examples:


Fig. 3: Maria-Elena me la dio �Maria-Elena gave it to me�

Characterizing the accent on the stressed syllable of Maria-Elena as simplyH* would fail to account for the fact that the tone remains low up to the onset ofthe stressed syllable (which is a systematic feature, at least in Castilian Spanish).This F0 contour is better characterized by a (L+H)* specification. A simple H*tone would imply a gradual rise from the beginning of the utterance to theaccented syllable.

I would thus propose the label (L+H)* for rising accents in Spanishdeclarative utterances, both in nuclear and in prenuclear position, without makingthe phonological distinction claimed by Face (2000). The different alignmentbetween prenuclear and nuclear accents seems to follow from the fact that in thecase of nuclear accents there is another low tone that must be realized within theword. This low boundary tone would prevent the displacement of the accentualpeak towards the right.

The presence of an intermediate-phrase L- would also prevent the peak frommoving to the posttonic in cases of narrow focus, as Nibert (2000) argues. For thecontour in Figure 2, where the word dieron has narrow focus, the tonal markingwould be as indicated:

(3) Narrow focus (contradictory)Context: ¿le pidieron el número? �Did they ask him for the number?�

le dieron el número �They GAVE him the number� |

(LH)* L- L%


The presence of an additional L- right after the (L+H)* pitch-accentsufficiently explains the difference in alignment between neutral and narrow focuswithout the need to postulate a phonological contrast between two types of pitch-accent.

2. Accents on final words in declarativesAs Roca (1986) remarks, the last stressed syllable in a neutral declarative

utterance is usually perceived as bearing the highest level of prominence. Henotices that in an example such as Juan vino en el tren de las tres �Juan came onthe three o�clock train� main prominence falls on the last word, tres. Evidently,speakers� intuitions somehow compensate for the fact that, in the absence ofspecial emphasis, accentual peaks decrease progressively from the beginning ofthe utterance (durational factors may also contribute to the perception of relativeprominence). Even more surprising is the fact that, in spite of its perceivedprominence, the last stressed syllable often bears no distinctive tonal peak.Comparing a number of renditions of the sentence le dieron el número de vueloproduced by different speakers, Sosa (1999) makes a categorical distinctionbetween two types of final accents. He uses the label H* for examples where thereis a clear rise on the last stressed syllable (vue-) (cf. Sosa 1999:Figs. 3-2, 3-53-8)and the label L* for examples where the last stressed syllable shows a fallingcontour (cf. Sosa 1999:Figs. 3-1, 3-4, 3-6, 3-7, 3-9, 3-10). Sosa (1999:195)suggests that the final contour H* L% (with a rise on the last stressed syllable)conveys a greater degree of emphasis or focalization on the last stressed wordthan the final contour L* L% (with a falling contour on the last stressed syllable).He finds this contrast both in Latin American and in Peninsular Spanish.3

A question is whether we are indeed dealing with a choice between twodistinct phonological categories for the final accent in declarative sentences, asSosa (1999) maintains. The fact is that even the very contours that Sosa (1999)includes as illustrations can be interpreted as providing evidence for a differentview: namely, that what we have is a continuum of realizations from a clear peakon the last stressed syllable to no peak at all, with tokens in-between showingdifferent degrees of reduction. This is basically the position that Nibert (2000)adopts.

Frota (1997, 1998), finds a very similar distinction in the realization of finalaccents in European Portuguese declaratives to that claimed by Sosa (1999) forSpanish, although she uses different labels for the accents. Frota (1997, 1998)uses the label H*+L for a final accent with a peak over the stressed syllable (= 3 Grice (1995:147) reports very similar facts for Palermo Italian.


Sosa�s H*) and the label H+L* for a fall over the final stressed syllable (= Sosa�sL*). Frota (1997, 1998) states that the falling H+L* accent is the one that is usedin neutral declaratives in European Portuguese, whereas the accent H*+L is usedto convey narrow focus on the last word. This is essentially the same pragmaticcharacterization Sosa (1999) gives, although expressed in stricter terms.Nevertheless, it appears that the evidence for postulating a phonological contrastbetween two types of final accents in declaratives is stronger in Portuguese thanin Spanish. Frota (1997, 1998) shows that in Portuguese we may have a surfacecontrast of the type indicated even in one-word utterances and that the twocontours would be used in different pragmatic contexts. This is illustrated with theidealized contours in (4), which are based on F0 tracings provided in Frota(1997:377):

(4) European Portuguese (Frota 1997)

casaram casaram �they got married�neutral intonation narrow focus

Frota (1997) indicates that the first contour, where the utterance casaram�they got married� is pronounced with a fall from the pretonic to the tonic,conveys new information and would be a natural answer to a question such as E oRoberto e a Maria? �what about Roberto and Maria?�, whereas the secondcontour, with a peak on the tonic syllable, would only be acceptable in a contextsuch as Eles separaram-se? �did they split up?�, where the word casaram as ananswer would express contrastive focus. Sosa (1999) has not shown that a clear-cut contrast of this type obtains in Spanish or in some subset of its dialects. Werethis evidence forthcoming, the separate issue concerning the correct labeling ofthe accents would then have to be raised.

In the absence of evidence for Spanish of the sort adduced by Frota forPortuguese, I believe it is more prudent to conclude that the examples analyzed bySosa (1999) as containing a nuclear L* accent simply represent cases of accentreduction. We should note that accent reduction and deaccentuation is by nomeans limited to the final position in the utterance. Non-final accents can also bereduced or eliminated depending on the emphasis that the speaker wishes to giveto particular words. That is, not every lexically stressed word necessarily bears apitch-accent, even outside of contexts of narrow focus. Consider for instance,


Figure 4, where the word número has been deaccented, or at least its accent hasbeen considerably reduced:4

Fig. 4: le dieron el número de vuelo �they gave him the number of the flight�

3. Prosodic phrasing: intonational signaling of old and new informationExamples such as the one in Figure 1, uttered as they would be pronounced in

a �neutral� context, typically show a pattern where each rising accent is lower thanthe preceding accent. That is, we have a series of progressively lower peakscreating a uniformly descending pattern from beginning to end of the utterance.

In sentences with a subject-predicate (or predicate-subject) structure, on theother hand, we often have an uptrend followed by a downtrend, as in the two F0tracings for the sentence Mariano me dio la moneda de oro �Mariano gave me thegold coin� in Figure 5 and Figure 6:

Fig. 5: Mariano me dio, la moneda de oro �M. gave me the gold coin�

4 See Willis (2000) for other examples of deaccentuation of nonfinal words.


Fig. 6: Mariano, me dio la moneda de oro �M. gave me the gold coin�

The very high value at the end of dio in Figure 5 and at the end of Mariano inFigure 6 can be interpreted as signaling the presence of a phrasal high tone H-.Although this phrasal high tone may coincide with the boundary marking themain syntactic division of the sentence, as we see in Figure 6, this is notnecessarily or even typically the case in Spanish. In this respect Spanish contrastswith English. Whereas in English a sentence like John ate the apple wouldnormally be prosodically divided as John, ate the apple, in Spanish we wouldperhaps more commonly have Juan comió, la manzana with the main prosodicbreak right after the verb (that is Juan comióH- la manzana).5 This is the structureof the sentence Mariano me dio, la moneda de oro in Figure 5.

The key for understanding the variable position of the phrasal H- in examplessuch as these is a consideration of the pragmatic information conveyed by each ofthe utterances. Clearly the contours are not pragmatically equivalent. Whereas thefirst intonational contour in Figure 5 would be appropriate in a context such as¿qué te dió Mariano? �what did Mariano give you?�, the contour in Figure 6could be used as answer to the question ¿qué hizo Mariano? �what did Marianodo?�. That is, the phrasal H- is positioned to coincide with the end of theconstituent conveying old information.

These considerations explain certain cases of prosodic disambiguation andnon-disambiguation of homophonous texts in Catalan studied in Prieto (1997)(and comparable cases in Spanish studied in Nibert 2000). Prieto selected anumber of ambiguous sentences in Catalan such as la vella llança l’amenaça �theold spear threatens her� or lit. �the old woman throws out the threat�, and asked 5 This is not to say that this kind of prosodic grouping is impossible in English (see Steedman1991), but the relative frequency and markedness of the groupings seem to be different in Spanishand English.


speakers to read them in a context where the intended meaning was unambiguous.She then performed a perceptual test on the utterances thus obtained to see towhat extent listeners were able to identify the intended meaning. The result wasthat some contours were consistently interpreted as noun-verb-noun (e.g. �the oldwoman throws out the threat�), but contours where the speaker had intended astructure adjective-noun-verb (e.g. �the old spear threatens her�) were in factambiguous. Although Prieto (1997) expresses some surprise at these findings(since they do not agree with previous descriptions of the facts, which claimed theexistence of a direct one-to-one mapping between syntactic and prosodicstructures), the interpretations are totally consistent with the possible pragmaticstructures of the sentences and consequent placement of H- at the end of the oldinformation:

(5) Syntactic and prosodic structurea. Syntax: [la vella llança] NP [l’amenaça] VP

the old spear threatens-her�the old spear threatens her�

Prosody: la vella llança H- l’amenaça

b. Syntax: [la vella] NP [llança l’amenaça] VPthe old-woman throws-out the threat�the old woman throws out the threat�

Prosody: 1. la vella H- llança l’amenaçacontext: �what does the old woman do?�

2. la vella llança H- l’amenaçacontext: �what does the old woman throw out?�

The partial ambiguity in interpretation that we find in these cases is aconsequence of the fact that the placement of the phrasal H- tone is determined bythe informational structure of the sentence: that is, by the division of utteranceinto theme and rheme or old and new information, and not directly by syntacticstructure.

It seems clear that an understanding of the pragmatic context of the utteranceis essential for the analysis of intonational contours. Hopefully in the near futurewe will see an increasing focus on the study of free conversation and other formsof natural speech.


4. ConclusionAt the current state of research on the intonation of the Ibero-Romance

languages we can ask very specific questions regarding the system of contrastsand the phonological representation of the contrasting elements. In this paper,after reviewing the evidence that has been put forward for different analyses, Ihave made specific proposals regarding the representation of raising contours,final falling contours and intonational phrasing.

REFERENCES

Alcoba, Santiago & Julio Murillo. 1998. �Intonation in Spanish�. IntonationSystems: a survey of twenty languages ed. by Daniel Hirst & Albert Di Cristo,152-166. Cambridge: Cambridge University Press.

Arvaniti, Amalia & D. Robert Ladd. 1995. �Tonal alignment and therepresentation of accentual targets�. Proceedings of the 13th InternationalCongress of Phonetic Sciences ed. by K. Elenius & R. Branderad, vol. 4.220-223. Stockholm.

---------, D. Robert Ladd & Ineke Mennen. 1998. �Stability of tonal alignment: thecase of Greek prenuclear accents�. Journal of Phonetics 26. 3-25.

Beckman, Mary & Janet Pierrehumbert. 1986. �Intonational structure in Japaneseand English�. Phonology Yearbook 3.255-310.

Chela-Flores, Bertha. 1994. �Entonación dialectal del enunciado declarativo deuna región de Venezuela�. Lexis 18.55-68.

Face, Timothy. 2000. �A phonological analysis of rising pitch-accents in CastilianSpanish�. Paper presented at the 30th Linguistic Symposium on RomanceLanguages held at the University of Florida, Gainesville, February, 2000.

Fant, Lars. 1984. Estructura informativa en español: estudio sintáctico yentonativo (Acta Universitatis Upsaliensis, Studia Romanica Upsaliensia 34).Stockholm: Almqvist & Wiksell.

Frota, Sónia. 1997. �On the prosody and intonation of focus in EuropeanPortuguese�. Issues in the Phonology and Morphology of the Major IberianLanguages ed. by Fernando Martínez-Gil & Alfonso Morales-Front, 359-392.Washington, DC: Georgetown Univ. Press.

---------. 1998. Prosody and Focus in European Portuguese. Ph.D. Dissertation,Universidade de Lisboa.


Garrido, Juan M., Joaquim Llisterri, Carme de la Mota & Antonio Ríos. 1993.�Prosodic differences in reading style: isolated vs. contextualized sentences�.Eurospeech 93.573-576.

Godjevac, Svetlana. 2000. �An autosegmental-metrical analysis of Serbo-Croatianintonation�. Ohio State University Working Papers in Linguistics 53.79-142.

Grice, Martine. 1995. The Intonation of Interrogation in Palermo Italian:implications for intonation theory. Tübingen: Niemeyer.

Hualde, José I. 1999. �Basic intonational contours in Spanish�. Paper presented atthe First Sp-ToBI Workshop held at the Ohio State University, Columbus,October 1999.

Kvavic, Karen. 1978. �Directions in recent Spanish intonation analysis�.Corrientes actuales en la dialectología del Caribe ed. by Humberto LópezMorales, 181-197. Río Piedras: Editorial Universitaria.

--------- & C. L. Olsen. 1974. �Theories and methods in Spanish intonationalstudies�. Phonetica 30.65-100.

Ladd, Robert. 1996. Intonational Phonology. Cambridge: Cambridge UniversityPress.

Llisterri Joaquim, Rafael Marín, Carme de la Mota & Antonio Ríos. 1995.�Factors affecting F0 peak displacement in Spanish�. ESCA, Eurospeech’95. 4th

Conference on Speech Communication and Technology, 2061-2064.Mota, Carme de la. 1995. La representación gramatical de la información nueva

en el discurso. Ph.D. Dissertation, Univ. Autònoma de Barcelona.---------. 1997. �Prosody of sentences with contrastive new information in

Spanish�. Intonation: theory, models and applications, Proceedings of an ESCAworkshop ed. by Antonis Botinis, 75-78. Grenoble: European SpeechCommunication Association.

Navarro Tomás, Tomás. 1944. Manual de entonación española. New York:Hispanic Institute of the United States. (4th ed., 1974, Madrid: Guadarrama).

Nibert, Holly. 1999. �A perception study of intermediate phrasing in Spanishintonation�. Advances in Hispanic Linguistics ed. by Javier Gutiérrez-Rexach &Fernando Martínez-Gil, 231-247. Somerville, Mass.: Cascadilla Press.

---------. 2000. Phonetic and Phonological Evidence for Intermediate Phrasing inSpanish Intonation. Ph.D. Dissertation, University of Illinois at Urbana-Champaign.

Pierrehumbert, Janet. 1980. The Phonology and Phonetics of English Intonation.Ph.D. Dissertation, MIT.

--------- & Mary Beckman. 1988. Japanese Tone Structure. Cambridge, Mass.:MIT Press.


Prieto, Pilar. 1995. �Aproximació als contorns entonatius del català central�.Caplletra 19.161-186.

---------. 1997. �Prosodic manifestation of syntactic structure in Catalan�. Issuesin the Phonology and Morphology of the Major Iberian Languages ed. byFernando Martínez-Gil & Alfonso Morales-Front, 173-194. Washington, DC:Georgetown University Press.

---------. 1998a. �Entonació�. Gramàtica del català contemporani ed. by JoanSolà. Forthcoming.

---------. 1998b. �The scaling of the L values in Spanish dowstepping contours�.Journal of Phonetics 26.261-282.

---------. Forthcoming. Review of Sosa 1999. Linguistics.---------, Jan van Santen & Julia Hirschberg. 1995. �Tonal alignment patterns in

Spanish�. Journal of Phonetics, 23.429-451.---------, Chilin Shih & Holly Nibert. 1996. �Pitch downtrend in Spanish�. Journal

of Phonetics 24.445-473.Roca, I.M. 1986. �Secondary stress and metrical rhythm�. Phonology Yearbook

3.341-370.Smiljanić, Rajka & José I. Hualde. 2000. �Lexical and pragmatic functions of

tonal in two Serbo-Croatian dialects�. Chicago Linguistic Society 36, vol. 1: TheMain Session ed. by Arika Okrent & John Boyle, 469-482. Chicago, Ill.:Chicago Linguistic Society.

Sosa, Juan Manuel. 1991. Fonética y fonología de la entonación del españolhispanoamericano. Ph.D. Dissertation, University of Massachusetts, Amherst.

---------. 1999. La entonación del español. Madrid: Cátedra.Steedman, Mark. 1991. �Structure and intonation�. Language 67.260-296.Willis, Erik. 2000. �Acoustic evidence for the Spanish imperative intonation:

Pitch-accent range and alignment�. Paper presented at the 30th LinguisticSymposium on Romance Languages held at the University of Florida,Gainesville, February, 2000.

‘PARTIAL’ SPANISHSTRATEGIES OF PIDGINIZATION AND SIMPLIFICATION

(FROM LINGUA FRANCA TO ‘GRINGO LINGO’)

JOHN M. LIPSKIThe Pennsylvania State University

0. IntroductionA recent comic strip, resuscitating racial stereotypes which had purportedly

disappeared at least a century ago, depicted a dialogue between a Spanish priestand an outrageous parody of an African ‘native’. The latter begins byaddressing the priest as follows:

Yo estar muy enojado. Yo haber tenido 10 hijos, “todos de color”! Ahora el 11vo.nacer blanco! Ud. ser el único hombre blanco en 200 km. Ud. deber“EXPLICARME.”[I am very angry. I’ve had ten children, “all colored!” Now the 11th one has beenborn white! You are the only white man within 200 km. You must “explain to me.”]

This stereotype is confirmed by independent observations. Ferguson (1971,1975) says that

a speaker of Spanish who wishes to communicate with a foreigner who has little orno Spanish will typically use the infinitive of the verb or the third singular ratherthan the usual inflected forms, and he will use mi “my” for yo “I” and omit thedefinite and indefinite articles: mi ver soldado “me [to-] see soldier” for yo veo alsoldado “I see the soldier.” Such Spanish is felt by native speakers of the languageto be the way foreigners talk, and it can most readily be elicited from Spanish-speaking informants by asking them how foreigners speak. (Ferguson 1971:143-4).

Thompson (1991) presented native speakers of Spanish with options as to howto address a newly-hired employee who spoke little Spanish. A large numberof respondents preferred sentences with bare infinitives, such as ¿cómo estarfamilia? “How is [the] family?” and ¿Jugar niños afuera? “[are the] childrenplaying outside?” When asked to ‘speak like Tarzan,’ the same respondents

JOHN M. LIPSKI118

came up with sentences using uninflected infinitives and null subjects: estarcomiendo “[we] are eating,” ¿estar plátano bajo tierra? “is [the] banana down[on the] ground?”, ¿quién ser hombre? “who is [the] man?” Hinnenkamp(1984) similarly lists the bare infinitive as a typical strategem of foreigner talk,while Corder (1975) believes that foreigner talk, baby talk, interlanguages, andpidgins share the general lack of copulas and functional categories, a singlepronominal paradigm, and little or no verbal inflection.

It is unlikely that any modern reader or speaker has heard Spanish spokenin this fashion, by natives of Africa or elsewhere. Those who struggle toemploy Spanish as a weak second language do not combine the bare infinitivewith correct vocabulary, compound verb formation, and NP-internalagreement. And yet this model of Spanish ‘foreigner talk’ has been inexistence for at least 500 years and probably much longer. Nor are blackAfricans the only group to be branded with this type of language; at one time oranother, similar reductions of Spanish have been attributed to speakers ofArabic, Berber, Chinese languages, Tagalog, Basque, French, German, English,and a variety of Native American languages. In Renaissance Spain, this wasthe language of the moro or Moor; today, it is the hands-down winner fortypecasting the gringo or ‘ugly American.’ Moreover, when unsuspectingSpanish speakers throughout the world are asked to imagine how Tarzan orsome other ‘ape-man’ might talk, the spontaneous responses are suspiciouslysimilar to the above-mentioned literary parodies. Lest it be thought that such‘broken Spanish’ is nothing but a fanciful invention, the product of bigotry,previously documented or currently surviving offshoots of Spanish-,Portuguese-, Italian-, and French-based foreigner talk reveal that this type ofpidginized language has existed in real speech communities, although today noknown second-language learners of Spanish speak in this fashion. What, then,is the relationship between imagined and real ‘foreigner’ Spanish, and how hasa reasonably cohesive model of such ‘almost-Spanish’ remained in the Spanishcollective unconscious for so long? In order to answer these questions, it isnecessary to examine real examples of reduced Romance, seeking paths ofhistorical evolution, crossovers among languages, and recurring patterns.Particular attention will be paid to the choice of the infinitive as default verband mí as subject pronoun.

1. Afro-Iberian languageReduced forms of Spanish, used both by native speakers and by second-

language learners, have coexisted with the full language since its origins, butaccurate documentation of L2 varieties of Ibero-Romance does not emerge untilthe end of the medieval period. Beginning towards the middle of the 15th

‘PARTIAL’ SPANISH 119

century with Portuguese explorations in West Africa, various forms of reducedPortuguese and then Spanish arose between Europeans and sub-SaharanAfricans. Trade, then slavery, provided the momentum for Afro-Europeancontact languages, some of which were immortalized in the form of creoles, inWest Africa and the Americas. Portuguese trade and colonization in southernand southeastern Asia also resulted in the formation of numerous creoledialects, in which a strong African component was also present. Finally, anumber of Spanish-based creoles developed in the Philippines, according tosome (e.g. Whinnom 1956) relexified from earlier Asian Portuguese creoles,and therefore possibly bearing an African imprint.

For more than three centuries beginning in the 1450’s, Portuguese andSpanish authors, in Europe and later in the Americas, would embellish theirpoems and plays with the fala de preto/habla de negro ‘black speech’. In Spainand Portugal, these literary imitations persisted long after African-born blacksceased to be a commonplace in the Iberian Peninsula (Lipski 1995). Despitethe high degree of racist stereotyping, many of the linguistic traits present inthe first round of Afro-Iberian literary texts appear in established creoles, thuslending more credibility to the remaining features.

The first couple of Afro-Portuguese texts, found in the Cancioneiro geral ofGarcia de Resende published in 1516, contain the bare infinitive and use ofmi(m) as subject pronoun, contain non-agreeing null subjects, and lack articlesand some copulas:

(1) FERNAM DA SILVEIRA [1455]: A min rrey de negro estar Serra Lyoa,lonje muyto terra onde viver nos, andar carabela, tubno de Lixboa“I am [a] king from Sierra Leone, from from the land where we live, [I]travelled by caravelle/shark to Lisbon”

Bare infinitives were soon replaced by some form of conjugated verb in theearly Afro-Portuguese texts; by the time of the first major writer to use suchlanguage (Gil Vicente 1912, writing in the early 16th century), ‘Africanized’Portuguese uses a combination of correctly conjugated verbs and incorrectforms, some of which are inappropriate members of the same paradigm, whileothers are wild inventions. In Spain, Afro-Hispanic pidgin language was firstrepresented in the “Coplas a los negros y negras” by Rodrigo de Reinosa,written at the turn of the 16th century. Most verbs are left in the infinitive;occasional defective attempts at conjugation (e.g. sabo) also occur. There aresome null subjects and articles are frequently missing:

JOHN M. LIPSKI120

(2) A mí llamar Comba de terra Guinea, y en la mi tierra comer buencangrejo“My name is Comba from the land of Guinea, and in my land [we] dinewell on crabs”

Uninflected infinitives were soon replaced by some semblance of conjugatedverbs. Only a decade or two after Reinosa’s poems, Lope de Rueda, widelyacknowledged as one of the most accurate imitators of early Afro-Hispaniclanguage, alternated bare infinitives and conjugated verbs in the speech of hisAfrican characters. Conjugated verbs--some bizarre, others only slightlydeviant--formed the basis for Afro-Hispanic language for several centuriesfollowing these early Golden Age writings. The subject pronoun mí wasreplaced by yo by the middle of the 16th century, and invariant verbs gravitatetoward the 3 s., with some use of the 1 p. in -mos. Beginning in the late 18thcentury a new group of texts appears in Latin America, representing newlyarrived Africans in the three regions in which the highest concentrations ofbozales or African-born L2 speakers of Spanish were to be found: Cuba,coastal Peru, and Buenos Aires/Montevideo. Particularly in the first tworegions, conjugated verbs (often in the invariant 3 s. form) did alternate withbare infinitives, suggesting the rapid acquisition of Spanish by freshly arrivedAfricans who could not always tap into a previously established Afro-Hispaniccommunity language. Spanish may have briefly creolized in the slave barracksof 19th century Cuban sugar plantations, but such cases were exceptional.Another source of creole-like structures in Cuban bozal Spanish is the influx ofcane-cutters from other Caribbean islands, most of whom spoke Afro-European creole languages with a high degree of structural similarity (Lipski1996, 1998c, 1999a). Descriptions of actually occurring Afro-Cuban Spanishwere given in the 19th century (Pichardo 1848, Bachiller y Morales 1883), andin the 1960’s (Barnet 1966). Imitations of bozal Spanish also form part of therituals of the negros congos, Afro-Hispanic communities along Panama’sCaribbean coast, who during the annual Carnival season speak a deliberatelymodified language which combines fanciful wordplay with what they claim tobe remnants of earlier Afro-Hispanic pidgin (Lipski 1989). Ortiz López (1998)traveled to Cuba and interviewed elderly Afro-Cubans, who recalled the timewhen bozal Spanish was still to be heard, and who produced examples similarto those found in literary texts. A number of examples from 19th centuryBuenos Aires and Montevideo (Fontanella de Weinberg 1987, Lipskiforthcoming) confirm the tendencies noted for Afro-Cuban Spanish. All ofthese actual observations--and most of the literary imitations--portray the lastcentury of Afro-Hispanic pidgin as a series of idiolects which converged to a


greater or lesser degree with vernacular Caribbean Spanish (cf. Lipski 1986a,1986c, 1998b).

In the modern world, the only stable interface between Spanish and sub-Saharan African languages occurs in Equatorial Guinea, formerly SpanishGuinea, which still maintains Spanish as the official national language.Grammatically, Guinean Spanish has no systematic differences fromPeninsular Spanish, but is rather characterized by a considerable instabilitywith regard to proper verb conjugation, syntactic formation, prepositionalusage, sequence of tenses and nominal agreement (Lipski 1985, Quilis &Casado-Fresnillo 1995). In contemporary Angola, Portuguese is spoken as asecond language in the musseques or working class neighborhoods of Luandaand other cities; this L2 Portuguese shares many of the characteristics ofEquatorial Guinean Spanish, for similar reasons. Although most Angolans usefinite verb forms (gravitating toward the 3 s. as unmarked form), there isdocumentation of the use of the Portuguese infinitive in the most pidginizedforms of Angolan Portuguese, for example:

(3) Senhor, já ter estado eschola aqui, agora já nno estar mais aquieschola, já eschola n'outra parte“Sir, there was a school, now there’s no school, the school issomewhere else” (Vidal 1916:426)

2. ‘Moorish’ SpanishThe use of Ibero-Romance as a second language by speakers of Arabic and

Berber began as early as the Moorish invasions of the 8th century, but writtendocumentation of such speech does not come until the expulsion of the last ofthe Moors from Spain at the end of the 15th century. Appearing nearlysimultaneously with Afro-Hispanic imitations, the speech of the moro ormorisco became a literary stock in trade for Spanish writers during the 16thand 17th centuries (Sloman 1949). In the 20th century a similar stereotype wasrevived in parts of Latin America in the speech of the (usually Arabic-speaking) turco “Turk,” in reality natives of Lebanon, Syria, and Palestine(Biondi Assali 1992). Grammatically, the morisco verb is almost always in theinfinitive, and some 16th century texts exhibit use of mí as subject pronoun.Non-agreeing null subjects are frequent, articles are often eliminated, and theverb estar, which in 16th century Spanish had fewer purely copular functionsthan in modern Spanish, became the default copula, used even with predicatenominatives, for example:

JOHN M. LIPSKI122

(4) GIL VICENTE, Cortes de Júpiter (ca. 1520): Mi no xaber que exto extar,mi no xaber que exto xer. Mi no saber onde andar. “I don’t know whatthis is, I don’t know where [I] am going”

Contemporary Latin American literary imitations of the Arabic-speakingMiddle Eastern immigrant rely predominantly on phonetic distortions, butoccasional grammatical modifications are also found. Only very occasionallyare the bare infinitive or generic copula estar used, and mí as subject pronounis not present:

(5) JOSÉ ROMÁN (1978), Cosmapa [Nicaragua]: Usted sabe, badrón, yoestar en este bueblo de Chinandega, borque combro granosbaradísimos “You know, boss, I am here in this town of Chinandegabecause I buy grain cheaply”

3. Anglo-Spanish pidginA common element in much 20th century Latin American literature is the

representation of pidginized Spanish as used by speakers of English. Thetypical pidgin speaker hails from the United States, and is often portrayed asthe domineering gringo or yanqui (expatriate travelers, entrepreneurs, andmilitary personnel), but similar linguistic traits have been attributed to nativesof England, and to West Indians in Central America. Most of the examples arederisive and macaronic, but occasionally the English speakers are cast in asympathetic light. Although there is considerable variation among texts, use ofthe uninflected infinitive is a common denominator, as is use of mí as subjectpronoun. The infinitive sometimes alternates with finite verbs, usually in thethird person singular. Many texts exhibit no articles and few prepositions,suggesting a phrase structure composed entirely of lexical projections.

(6) a. BENITO LYNCH (1926), El inglés de los güesos [Argentina]: mítrabaca ... mí busca huesas antiguos, viecas.“I work, I look for old bones,”

b. JOAQUÍN GUTIÉRREZ (1977), Puerto Limón [Costa Rica]: No, míno pueda llevar. Mí llevar y después joden a Tom. Míster, yosabe bien“No, I can’t take [you]. I take you, and then Tom gets screwed. Mister, I know very well”

c. JOAQUÍN BELEÑO (1963), Curundú [Panama]: ¿quién mandaraquí? si tú no saber, ¿quién sabe?“Who’s in charge here? If you don’t know, who knows?”


These examples contrast with more realistic literary representations of English-influenced Spanish, as well as by actual field observations. In the DominicanRepublic, the creole-English speaking contract laborers from the AnglophoneCaribbean frequently appear in regionalist literature. Their L2 variety ofSpanish bears great resemblance to the Spanish of Anglophone students in theUnited States and stands in contrast to the previously mentioned literarystereotypes. Another source of second-language Spanish data comes in thespeech of descendents of black Americans, in the Samaná Peninsula of theDominican Republic. Samaná by native speakers of English (Ferreras 1982):

(7) Yo se sabe lo que tú se quiere decir, pero para que tú se consigue esecosa que tú se dice, yo se va a dar un buen consejo.“I know what you’re trying to say, but in order for you to get what youwant, I’m going to give you some good advice”

Examples of the L2 Spanish of West Indians actually recorded by myself inPuerto Rico include:

(8) a. yo conoce Trinidad, yo fuite de vacacione“I know Trinidad, I went on vacation” [Jamaica]

b. yo vengo pa cá y yo aprende“I came here and I learned” [St. Kitts]

Literary imitations of Haitian canecutters in the Dominican Republic and Cuba,who speak a pidginized Spanish similar to that of English-speaking WestIndians, coincides substantially with actual observations of Haitians’ Spanish,although early literary attempts were only crude parodies. In general, finiteverb forms are used to instantiate Spanish finite verbs, with a noteworthypreference for the third person singular. Some Haitians occasionally use theSpanish infinitive instead of a conjugated verb, perhaps reflecting thewidespread homophony between Haitian Creole verbs and French infinitives,but such examples are not common. Actual examples of Haitians’ L2 Spanishrecorded in Cuba (Ortiz López 1999) are:

(9) a. Yo prende hablá catellano con cubano“I learned to speak Spanish among the Cubans”

b. yo hacel mucho trabajal; coltal caña balato; recogel café a sei kilo

“I worked hard; I cut sugar cane for little money; I pickedcoffee for six cents”

JOHN M. LIPSKI124

Vestigial Spanish as spoken by transitional bilinguals or semi-speakers, e.g. inthe United States, shows the same characteristics as pidginized L2 Spanish,including overuse of the 3s. verb form, and unstable agreement systems (Lipski1986b).

4. Philippine pidgin SpanishThe Spanish language was present in Philippines for more than 350 years,

although only a very small proportion of the Philippine population ever spoke(non-creolized) Spanish either natively or as a strong second language (Lipski1987, 1988). Philippine Creole Spanish has survived as a viable first andstrong second language in several cities; in addition to creole and quasi-nativePhilippine Spanish, several Spanish-based pidgins evolved, particularly asspoken by Chinese residents, but also used by non-fluent Filipinos whenaddressing Spaniards. This was a rough pidgin, and contained few if any of theconsistent grammatical structures which characterize Creoles; pidgin Spanishas spoken by Chinese immigrants in the Philippines is typified by:

(10) sigulo, señolía ... como no tiene ahola talabajo; como no tiene capé,y ha de ganalo la vida, sigulo tiene que hace tabaco“of course, sir; since {I} do not have a job now, and since {I} don’thave any coffee, and {I} have to earn a living, of course {I} have tomake cigars” (López 1893:58)

There are many examples of Philippine pidgin Spanish as used by nativeFilipinos, with some creoloid characteristics but still representing animperfectly acquired second language, e.g. Bueno, señor, aquí comer “Well,sir, here [you can] eat” (Feced 1888:24).

5. Chinese pidgin SpanishIn the second half of the 19th century, Cuba received at least 150,000

Chinese laborers, while more than 90,000 Chinese workers were imported intocoastal Peru. Smaller numbers arrived in Panama, Venezuela, and CentralAmerica. The Chinese worked in the sugar plantations and mills as virtualslaves, side by side with Africans and--in Cuba--workers from other Caribbeanislands. The linguistic conditions surrounding the lives of Chinese laborersclosely parallels that of African bozales, and Chinese workers’ acquisition ofSpanish followed similar paths (Lipski 1998a, 1999c). Verbs--which almostnever agreed with subjects--were sometimes based on the third person singular,sometimes on the infinitive:


(11) a. ANTONIO ORTEGA, “China olvidado” (Bueno 1959:54-73) [Cuba]:Yo no sabel. Chino olvilalo, chino no tenel palientes ... no tenelamigos ... chino estal solo ...“I don’t know. Chinese man forgets, Chinese man has norelatives, has no friends ... Chinese man is alone”

b. TRAZEGNIES GRANDA (1994:238) [Peru]: tu cleel que sólo neglohacel velso. Pelo pala chino sel palte de su elucació ...“you think that only blacks can make up verses. But for [us]Chinese, it’s part of [our] upbringing”

c. CARLOS LUIS FALLAS, Mamita Yunai (1975) [Costa Rica]: Yolevanta templano pelo quela mucho lato convelsando coLamilo.“I got up early but I stayed a long time talking to Ramiro”

Nowadays there is little authentic Chinese pidgin Spanish remaining.Individual speakers of Chinese learning Spanish exhibit significantly differentcharacteristics, particularly the use of articles and copulas, as well asconjugated verbs instead of bare infinitives; an actually occurring examplefrom one Chinese speaker’s L2 Spanish (Clements 1999) is: yo dise tíoconmigo hablando, puede fuela China “I said, uncle was telling me I couldleave China”.

6. ‘Basque’ SpanishIn the Basque Country of northern Spain, Spanish was a recessive second

language for many centuries, and, even today, Basque-dominant and evenSpanish-dominant bilinguals exhibit linguistic traits which set them apart frommonolingual speakers in other regions. In earlier centuries the literarystereotype of the vizcaíno or Viscayan took its place alongside the other well-developed parodies of second-language Spanish speakers: Moors, Gypsies,black Africans, Galicians and Portuguese, French, Italians, and speakers ofnon-prestige regional dialects of Ibero-Romance. Major Spanish writers suchas Cervantes and Lope de Vega, as well as numerous writers of skits, poems,and longer plays, incorporated the vizcaíno, consolidating a stereotype whicheventually needed no preamble to clue the audience as to the characters’identity (Legarda 1953, Herrero García 1966:chap. IX). The linguistic featuresof the Basque-Spanish stereotype center around incorrect subject-verbagreement, with an overwhelming preference for the second person singular (-s) as invariant verb form. Bizarre word-order alteration frequently occurred, asdid unstable gender and number agreement. These humorous lapses were oftencombined with an extensive command of Spanish vocabulary and syntactic

JOHN M. LIPSKI126

structures, not a likely combination in actually occurring bilingual speech. Theimpression of vizcaíno Spanish is more chaotic and even demented than itsclosest relative, Afro-Hispanic pidgin, due to the startling juxtaposition ofsophisticated vocabulary and improbable syntactic transpositions:

(12) MELCHOR DE SANTA CRUZ (1996:321-9), Floresta española {1574}:Juras a Dios, andas por arte del diablo“[I] swear to God, [the mill-wheel] is turning by the Devil’s artifice”

Although it is impossible to completely rule out the possibility that the vizcaínoimitations may once have been accurate imitations of Spanish-Basquebilinguals, contemporary Basque-influenced Spanish shows strikingly differentcharacteristics (Urrutia Cárdenas 1995). There is no evidence of the 2s. orother verb form being overextended in an invariant verb paradigm.

7. Amerindian pidgin SpanishThroughout the Americas, reduced forms of Spanish are spoken by isolated

indigenous populations, with characteristics different from the more fluentSpanish-based interlanguage used, e.g. in the Andean region, Paraguay, andcentral Mexico. At times, the pidgins are used only by indigenous residentswhen speaking to native Spanish speakers or members of other languagegroups, while in other cases native Spanish speakers consciously adopt thepidgin when speaking to members of an indigenous community. In theAmazonian basin (Colombia, Peru, Ecuador), Spanish is used as a recessivesecond language by speakers of various indigenous languages. The latter’s L2Spanish exhibits overgeneralization of the 3s. finite form, but no systematicuse of the infinitive (Rodríguez de Montes 1981). In the Venezuelan Orinocobasin, members of the Panare group have used a stable Spanish-derived pidginfor interchange with fluent Spanish speakers as well as indigenous groupsspeaking other languages (Riley 1952). In this pidgin, the verb is representedin the gerund form, few articles or prepositions are used, and Spanish subjectpronouns are used in all cases: yo no sabyendo ke tu disyendo “I don’t knowwhat you are talking about.” Another Spanish-based pidgin spoken in theVenezuelan Orinoco region is used by the Marquitare and appears in the novelCanaima by Rómulo Gallegos (1991). Like the Panare pidgin, the gerund isused to represent verbs: Yéndote con Marcos, que no siendo maluco “Go withMarcos, who isn’t sick.” Another Orinoco group, the Guaraúnos, also speakswith the jerga de gerundios ‘gerund-jargon’: Yo dándote chinchorro, túdándome sal “[if] I give you a fishing net, you will give me salt.” The novelJuyungo by the Ecuadoran writer Adalberto Ortiz (1976) provides examples of


the pidginized Spanish of the Cayapa of northwestern Ecuador. The Cayapaalso prefer the gerund as invariant verb form: Tú, compadre, chiquito,gustándome. Tú, sabiendo números, ¿no? Yo necesitándote aquí “I like you,fellow. You know about numbers, don’t you?” At another point, Ortizimitates the pidginized Spanish of the Colorados, another indigenous groupfrom northwestern Ecuador (1976:93): Eso estando bueno, entren en mi casa,yo llamando otra gente “That’s good, come into my house, I will call someother people.” Even the less fluent Spanish as used by speakers of Quechuaand Aymara in the Andean region contains an overuse of the gerund as well asother anomalies of word order and verb morphology. As with the earlierexamples, the bare infinitive is never used instead of a finite verb (Laprade1981, Stratford 1989): ¿Qué diciendo nomás te has venido? “Just why haveyou come?” The L2 Spanish of Otomí speakers in Mexico, which is notimitated by native Spanish speakers, contains many non-agreeing verbs, withthe 3 s. being the preferred form. The gerund occasionally appears in lieu offinite verbs (Lastra 1995): Tú siguiendo trabajando “You keep on working.”

8. The evidence from Spanish child languageGiven the popular equation PIDGIN LANGUAGE = BABY TALK, it is useful to

gather data from Spanish early child language to judge the likelihood that adultimitation of child Spanish lies at the root of some or all Spanish-based pidgins.Careful examination of Spanish child language reveals that, whereas articlesmay be missing in the earliest stages, the other recurring features of Spanishpidgins are not common in child language. In particular, the choice of theinfinitive and mi as subject pronoun are not typical of Spanish, Italian, andPortuguese early child language (e.g. López Ornat et al. 1994). In Spanish, thefinite-non finite distinction is morphologically no more complex than thedifference between other members of the verbal paradigm; the infinitivemorpheme, consisting of the theme vowel + /r/ takes the place of anothersuffix. There is no sense in which the infinitive is morphologically simpler ormore ‘basic’ than finite forms. Thus it is not accurate to speak of the Spanishinfinitive as ‘uninflected’ as opposed to ‘conjugated’ verbs, since all Spanishverb forms consist of at least a stem, a theme vowel, and some other indicationof tense, mood, person, and number. The infinitive appears as the dictionaryrepresentation of a Spanish verb and is used as a citation form by nativespeakers, when discussing verbs in an abstract sense. In terms of frequency ofusage, Spanish infinitives are considerably less common than many othermembers of the verbal paradigm, particularly the present indicative forms. Thesituation in Italian child language is similar (Pizzuto & Caselli 1992, Guasti1993/4). The same holds for other Ibero-Romance languages, including

JOHN M. LIPSKI128

Catalan and Galician. French child language is exceptional in this regard, inexhibiting a significant use of the bare infinitive in lieu of finite verbs (Clark1985, Pierce 1992, Ferdinand 1996). Moi is also common as subject in childFrench (Ferdinand 1996:201-2). There is also evidence that in French, as inEnglish, objective case is the default case when a non-finite verb fails to assignnominative case to a subject (cf. Ferdinand 1996:142-4; Haegeman 1996:286-7; Friedemann 1994 suggests that the case filter is not yet operative in thisstage of child language). L2 French may also show some use of moi and toi assubject, with both finite verbs and root infinitives (White 1996). German,Dutch, and Scandinavian child language also frequently exhibits rootinfinitives instead of finite verbs (Harris & Wexler 1996, Wexler 1994, 1998,Rizzi 1993/4, Hoekstra & Hyams 1998, Haegeman 1996). Wexler (1994,1998) affirms that the optional infinitive (OI) stage is not found in early childspeech of languages in which INFL licenses null subjects: Italian and Ibero-Romance. OI does occur in languages where INFL does not fully license nullsubjects: English and other Germanic languages. It is not coincidental that inthe latter languages the infinitive is morphologically indistinct from somemembers of the finite paradigm; in the terminology of Jaeggli & Safir (1989),only languages with morphologically uniform verb paradigms permit nullsubjects.

9. Sources of the uninflected infinitiveThe repertoire of simplified or reduced forms of Spanish contains some

striking dichotomies as regards morphosyntactic patterns, in particular phrasestructure, use of articles, and verbal morphology. One group of reducedlanguages, typified by ‘Moorish’ speech, some contemporary Native Americancontact varieties, and pidgin Spanish as attributed to speakers of English, ischaracterized by the use of the bare infinitive, use of mí as subject pronoun,minimal use of articles and prepositions, and in general a phrase structureconsisting only of lexical categories. The second group, represented by Afro-Iberian pidgins, Philippine pidgin Spanish, and to a certain extent Chinesepidgin Spanish, has a verb stem based largely on the third person singular(present indicative), some use of articles and prepositions, and generalevidence of functional projections and limited use of subordinate clauses.Objectively, when contemporary L2 varieties of Spanish are brought into theequation, they align more closely with the second group. True Spanish childlanguage, often implicated in ‘baby talk’ theories of pidgin formation, is notidentical to either group of reduced languages; early child Spanish shares withthe first group the limited use of functional projections. It shares with thesecond group a verbal morphology based on finite forms and proper use of


subject pronouns. Due largely to differences in the morphology of verbparadigms, the use of bare infinitives is more common in French-derivedpidgins and Creoles, in the Caribbean, Africa, and the South Pacific (Goodman1964:105; Holm 1988:16; Hollyman 1964, Reinecke 1971, Stageberg 1956,Nguyen 1977, Schuchardt 1888, Göbl-Gáldi 1934:271; Chaudenson 1978,Poyen-Bellisle 1894:43; Niedzielski 1989, Duponchel 1979, Makouta-Mboukou 1975, Kokora 1983, Véronique 1994).

10. Use of mi as subject pronounPerhaps even more than the uninflected infinitive, the choice of mi as

subject pronoun defines the most xenoglossic forms of Spanish foreigner-talk.Spanish speakers throughout the world associate the subject pronoun mi with‘Tarzan talk’ and infantile pidgin, almost as much as do English speakers,despite the lack of ready models in actually occurring colloquial or childspeech (Pensalfini 1995). Mi as subject is not a mindless creation out ofignorance; Portuguese-based creoles in Africa and Spanish/Portuguese-basedcreoles in the Americas (e.g. Papiamentu and Afro-Colombian Palenquero) usepronouns derived from mi. Unlike French, Ibero-Romance subject pronounshave not evolved to phonological clitics. In French, the disjunctive objectpronouns moi, toi, etc. are used as free-standing pronouns or for emphasis.This is reflected in all French-based creoles. Colloquial English also uses meas an alternative to I in answer to questions or (together with the subjectpronoun) for emphasis. English-based creoles have adopted mi as the genericfirst person singular pronoun, but in some areas this may also have beeninfluenced by English speakers’ stereotypes of foreigner-talk or early pidginEnglish. Spanish and Portuguese subject pronouns may stand alone, withcontrastive stress, although due to their redundant status in the face of a richverbal inflection, they do not ordinarily receive strong tonic stress. In anysituation where emphasis is required, or when a free-standing subject pronounis called for (e.g. in response to a question), it is invariably the subject pronounwhich is used. By the same token, overt subject pronouns are by definitionemphatic, and are never replaced by disjunctive object pronouns for emphasis,to answer questions, etc.

11. In search of sources: Mediterranean Lingua FrancaThe search for the origins of contrived Spanish moves one step closer to a

solution upon consideration of a series of Romance-based contact vernacularswhich antedate the reduced Spanish examples discussed above. For more thanhalf a century, ‘monogenetic’ theories of creole formation and their lessexplicit precursors have implicated another language, which apparently

JOHN M. LIPSKI130

stretched at least from the early medieval period through the turn of the 19thcentury: the Mediterranean Lingua Franca. Much has been written about thiselusively unwritten contact language, but tangible and trustworthy attestationsare as scarce as hens’ teeth (Bonaparte 1877, Cifoletti 1978, 1989, Coates1971, Collier 1976, Cortelazzo 1965, 1972, 1977, Coutelle 1977, Fronzaroli1955, Grion 1891, Hadel 1969, Harvey & Whinnom 1967, Kahane & Kahane1976, Lang 1992, 2000, Schuchardt 1909, Vianello 1955, Whinnom 1977,Wood 1971). Only a handful of texts or descriptions of Lingua Franca antedatethe 19th century, making reconstruction speculative and venturesome.Surviving texts are suspect as true specimens of a pan-Mediterranean LinguaFranca, rather than local attempts at mimicking broken Romance spoken byforeigners or derogatory stereotypes of ‘infidel’ Arabs and Turks. All knownLingua Franca texts employ the infinitive as invariant verb stem to instantiatethe entire verb paradigm, although occasional conjugated verbs crop up insome texts. The subject pronouns mi and ti predominate except in the veryearliest Lingua Franca examples, in which null subjects occur:

(13) a. ANON., ITALY [ca. 1353]: come ti voler parlare? “how do youwant to speak?”

b. GIGIO GIANCARLI, La cingana [ca. 1550]: mi no saber certa “Iam not sure”

c. DIEGO DE HAEDO, Topografía e historia general de Argel [ca.1612]: mirar como mi estar barbero bono y saber curar, si estarmalato y ahora correr bono “Look what a good doctor I am andhow I know how to cure [him], if [he] is sick, and now [he] runswell”

d. MOLIÈRE, Le bourgeois gentilhomme [ca. 1671]: Mi star Mufti; tiqui sar qui? “I am Mufti; who are you?”

e. [ALGIERS, 1884] (Faidherbe 1884): Moi meskine, toi donner sordi“I am poor; you [will] give me money”

The choice of the infinitive in Lingua Franca is not a natural consequence ofthe imperfect acquisition of Italian by speakers of Eastern Mediterraneanlanguages, but reflects an originally conscious choice by speakers of Italian andother Romance languages to simplify their verbal system when speaking toforeigners deemed incapable or unworthy of learning a full version of theselanguages. The use of mi and ti as subject pronouns in Lingua Franca has amore straightforward explanation (Lipski 1991). These pronouns are linked toregional dialects of medieval Italy, particularly Venetian and Genoese, whichshortly before had begun to employ mi and ti as subject pronouns (Vanelli1984, 1987).


12. Lingua Franca as a model for ‘foreigner’ SpanishLingua Franca was in the right place at the right time to serve as a model

for Moorish pidgin Spanish, as it appears in Golden Age literature. ManySpaniards and Portuguese had traveled to the eastern Mediterranean.Continuing skirmishes with the Barbary Coast put Spaniards and Portuguese incontact with the westernmost varieties of Lingua Franca, which according toHaedo (1927) already contained more Spanish than Italian. Lingua Francastrategies became implanted as a powerful national image of ‘foreign’ Spanishand gave native Spanish speakers a template upon which to base their ownforeigner-talk, when such was necessary. So indelible was the imprint ofLingua Franca that it has remained in the collective consciousness of theSpanish-speaking world as the most ‘savage’ form of foreigner-talk, to be usedwhen more ‘civilized’ approximations to Spanish are felt to be inappropriate.

The same basic Lingua Franca patterns found in Moorish imitations formedthe basis for the earliest Afro-Iberian pidgin, from 1455 to the early 1500’s.The use of (a)mí as subject in early Afro-Iberian language was reinforced bythe fortuitous similarity among the first person singular pronouns in a widevariety of West African languages, all of which center around the formsmi/emi/ami (Lipski 1991). French colonists continued to use Lingua Francamodels (known as petit mauresque), combining French and Italian words, inNorth Africa until the final decades of the 19th century. Continuing well intothe 20th century, reduced forms of Italian have been spoken in North Africaand the former Italian East Africa, particularly Libya, Ethiopia (Eritrea) andSomalia (Migliorini 1963:696; Marcos 1976). The bare infinitive is used fornon-past reference, while the Italian past participle forms the basis for past-tense forms. Lingua Franca was reborn not far from its original birthplace; inLibya, it is possible that the Lingua Franca of the Barbary Coast corsairsevolved seamlessly into colonial Afro-Italian.

13. Additional sources for Spanish foreigner talkThe preceding discussion has demonstrated that reduced Spanish,

Portuguese, and Italian could not come directly from L2 learners’ spontaneousacquisition of these languages, nor from early child language, although somesimilarities with the latter can be observed. In Italy and Spain, contact withGerman (Mühlhäusler 1984) and French dialects, respectively, in which rootinfinitives occur in child speech and foreigner talk since at least the MiddleAges, may well have spurred the use of root infinitives and disjunctive objectpronouns as subjects in contrived foreigner talk. Given the condescending

JOHN M. LIPSKI132

nature of much Romance-based foreigner talk and the negative attitudesextended to the intended recipients, another source of inspiration is likely: thespeech of adults with language disorders. Developmental dysphasia, parti-cularly the cluster of phenomena known as specific language impairment (SLI),is characterized by the prolongation into late childhood and even adulthood ofmorphological and syntactic mismatches characteristic of early child language.In particular, extended optional infinitives are frequent in impaired English,German, and French (Clahsen 1989, 1991, Wexler 1996, Rice & Wexler 1996,Leonard 1998). Adult agrammatism, a form of aphasia, is typically caused bystrokes and other brain lesions, and like other forms of aphasia covers a widegamut of speech impairments, many of which are familiar to unsophisticatedmembers of Romance speech communities. Some of the traits of aphasicspeech coincide with child language and vestigial speech (Menn 1989, Menn &Obler 1990:1372-3). Since the Romance infinitive, particularly in Spanish,Portuguese, and Italian, is a morphologically rather marked form, it does notcommonly appear in aphasic speech in substitution of finite forms. Miceli &Mazzuchi (1990) do describe some root infinitives in impaired Italian. Thestereotype of the simple or bobo ‘simpleton’ was a frequent concomitant ofSpanish Golden Age literature, although such characters usually madeinopportune or accidentally perspicacious remarks rather than producing SLISpanish. However, awareness of language disorders, although not referred toas such, is as old as the human species, and the abundant models of childlanguage, delayed language development/SLI, and adult agrammatismconverge with the more grotesque foreigner-talk stereotypes, to an extentwhich suggests more than coincidence.

14. Was Spanish deliberately simplified?The data from reduced Spanish define a fundamental typology, based on

attitudes, power and influence, and urgency of communication. Deliberatelyreduced foreigner talk--replete with infinitives, null subjects, default copulas,and lack of functional categories--is used when communication with anothergroup is felt to be essential to native speakers, while the L2 either have littleopportunity to learn Spanish or are felt to be incapable of or unwilling to do so.This occurs regardless of the power differential: Spaniards to Moors, LatinAmericans to ‘gringo’ expatriates, Turks to German tourists (Hinnenkamp1984), and possibly Portuguese sailors and traders to West Africans (Naro1978). Lingua Franca was used by speakers of Romance to a variety of EasternMediterranean speakers, regardless of relative social status, based on the need(commercial, military, religious) to communicate and the unlikelihood that thenon-Romance speakers would ever acquire a full variety of Italian, French, or


Spanish. Similarly, Spaniards may have used morisco language when speakingto the despised Moors or at least expected the latter to be incapable of risingabove such language. At once unwanted and essential to economic prosperity,post-Reconquest Moors were reluctantly assigned their own foreigner-talkregister. Reduced Spanish is not used when speaking to groups who willultimately have to acquire fluency in Spanish and/or when communication canbe mediated through a bilectal group (overseers, slavedrivers, teamsters,traders): Spanish speakers to African slaves and Chinese indentured laborers,Dominicans to Haitian canecutters, Spanish-speaking Central Americans toWest Indians. This comes despite the observation of Naro (1978:324-6) thatattitudes of superiority were irrelevant in the decision to use (Portuguese)foreigner-talk, facilitating comprehension always being foremost. Only thecase of ‘Basque’ Spanish apparently breaks the rule; Spaniards had no essentialneed to communicate with Basque speakers in a reduced language, and thewildly improbable nature of vizcaíno parodies makes it quite unlikely that anysuch reduced language ever existed.

Lingua Franca and colonial Italian were not directly based on Italian childlanguage, in which the infinitive is much less frequent than, e.g. in Spanish orFrench. Rather, the choice of the infinitive as verb stem was an earlyforeigner-talk strategy employed by fluent speakers of Italian and otherRomance languages as being the verb form most likely to speed alongcommunication with the many linguistic groups of the eastern Mediterranean.Once inaugurated via the Lingua Franca as the vehicle for verbal expression,the infinitive acquired a life of its own, becoming a juggernaut which propelledsubsequent reduced versions of Italian, French, and Ibero-Romance wheneveraccommodation of non-Romance speakers was required. The Lingua Francaparadigm of the bare infinitive was put back into play, having served admirablyto discredit potential speakers of Spanish in earlier centuries. Afro-Hispanicpidgin emerged at a time when Lingua Franca was fresh in the minds ofsouthern Europeans, and these pidginized Romance dialects coexisted for atleast two centuries. Unlike Lingua Franca, Afro-Hispanic pidgin--except forthe very earliest examples--shows little evidence of deliberate modification byEuropeans, an exception being the 15th-16th century training of Africaninterpreters in Portugal (Naro 1978). Awareness of the basic features of Afro-Hispanic pidgin was high among Spaniards and Portuguese living in cities withlarge African bozal populations, but nothing suggests that Europeans ever usedbozal pidgin themselves or otherwise modified their speech when speakingwith Africans in their midst. Substantially the same was true of Chineseworkers taken to Latin America. The Chinese laborers were imported toreplace African slaves in the most demanding and degrading plantation work;

JOHN M. LIPSKI134

when they first arrived, they were despised by blacks and whites alike; theywere forced to learn Spanish from fellow plantation workers, many of whomwere African-born bozales, creole-speaking canecutters from other Caribbeanislands, or marginalized free blacks who may have retained some bozal traits intheir spoken Spanish. Thus not only did Chinese workers acquire a makeshiftpidgin in the absence of sustained contact with native speakers of Spanish, butthey built their pidgin upon foundations already laid by speakers of African andAfro-American languages. The result was a reduced form of Spanish bearingmore than coincidental resemblance to bozal and Afro-Caribbean Spanish, butwith some uniquely Chinese traits.

Consistently pidginized Spanish has developed in isolated communitieswhen trade with the dominant Spanish-speaking population was expanded. Theenticement of trade or missionary activity provided the impetus for fluentSpanish speakers to deliberately adopt a reduced form of the language. In areaswhere no particular attempt was made to accommodate the language ofindigenous speakers (e.g. the Andean region, Paraguay, Guatemala), aspontaneous interlanguage with significantly different characteristics is theresult.

15. ConclusionsThe overview of reduced Spanish varieties yields the unmistakable

conclusion that a collective stereotype of imperfect Spanish has existed forseveral centuries, independently of the existence of actual specimens of suchlanguage. This remarkable survival stems from the convergence of severalfactors, including a nearly unbroken--if unflattering--literary tradition, apoignant if inaccurate awareness of early child language, and a consciousdecision to homogenize Spanish grammar to a degree not found in spontaneoussecond language acquisition. It is the latter phenomenon which sets Spanishforeigner talk apart from the majority of L2 Spanish varieties, in whoseevolution native speaker simplification has played an insignificant role. At theroot of the decision by native speakers to use a contrived language is theperceived status of the pidginizing group, the likelihood that the L2 speakerscan and will acquire and understand complete Spanish, and the necessity forsustained communication between native speakers and the L2-speaking group.Non-interference with second language speakers’ acquisition of Spanish is theunmarked case, and results in varieties such as Afro-Hispanic bozal languageand the actually occurring L2 Spanish of Anglophone Central Americans,Chinese speakers, and North Africans (Casado-Fresnillo 1995, Tarkki 1995).The original Lingua Franca, a comprehensive Romance-based contact languageformed with the active collaboration of speakers of languages cognate with


Spanish, never fully disappeared beneath the horizon as the Spanish languageevolved and spread to other continents. In circumstances in whichcommunication with groups felt--if only grudgingly--to be essential tradepartners, Spanish speakers have always been able to slip back into a form ofdiscourse which owes little to coincidental similarity or the meanderings ofuniversal grammar. The ready availability of this artificially generated butculturally potent linguistic template has at times deflected attention away fromcontact languages formed through natural selection, and in which substratumand universal features predominate over intentional simplification.

REFERENCES

Bachiller y Morales, Antonio. 1883. “Desfiguración a que está expuesto elidioma castellano al contacto y mezcla de razas”. Revista de Cuba 14.97-104.

Barnet, Miguel. 1966. Biografía de un cimarrón. Havana: Instituto deEtnología y Folklore.

Beleño, Joaquín. 1963. Curundú. Panama: Ministerio de Educación.Biondi Assali, Estela. 1992. “‘Beine ... beineta’: el uso de (p) en el habla

española de los inmigrantes de origen árabe en la Argentina.” HispanicLinguistics 5.143-168.

Bonaparte, L. L. 1877. Lingua Franca. The Athenaeum. May 19, 1877, p.640; June 2, 1877, p. 703.

Bueno, Salvador, ed. 1959. Los mejores cuentos cubanos, tomo II. Havana:Segundo Festival del Libro Cubano.

Casado-Fresnillo, Celia. 1995. “Resultados del contacto del español con elárabe y con las lenguas autóctonas de Guinea Ecuatorial”. Silva-Corvalán1995.281-292.

Chaudenson, Robert. 1978. “Créole et langage enfantin: phylogenèse etontogenèse”. Language Française 37.76-90.

Cifoletti, Guido. 1978. “Lingua Franca e Sabir: considerazioni storiche eterminologiche”. Incontri Linguistici 4.205-212.

---------. 1989. La lingua franca mediterranea. Padova: Unipress.Clahsen, Harald. 1989. “The grammatical characterization of developmental

dysphasia”. Linguistics 27.897-920.---------. 1991. Child language and developmental aphasia. Amsterdam &

Philadelphia: John Benjamins.Clark, Eve. 1985. “The acquisition of Romance, with special reference to

JOHN M. LIPSKI136

French”. The crosslinguistic study of language acquisition, vol 1., ed. byDan Slobin, 687-782. Hillsdale, NJ: Lawrence Erlbaum.

Clements, J. Clancy. 1999. “Solo pidginization: the TMA system in anaturally acquired L2 with minimal input”. Speaker series lecture,Department of Linguistics, Cornell University, November 19, 1999.

Coates, William. 1970. “The German pidgin-Italian of the 16th-centuryLanzichenecchi”. Papers from the fourth annual Kansas LinguisticsConference ed. by Herbert Harris, 66-74. Lawrence: University of Kansas,Linguistics Department.

---------. 1971. “The Lingua Franca”. Papers from the fifth annual KansasLinguistics Conference ed. by Frances Ingemann, 25-34. Lawrence:University of Kansas, Linguistics Department.

Collier, Barbara. 1976. “On the origins of Lingua Franca”. Journal of CreoleStudies 1.281-298.

Cortelazzo, Manlio. 1965. “Che cosa s’intendesse per ‘lingua franca’”.Lingua Nostra 26.108-110.

---------. 1972. “Nuovi contributi alla conoscenza del grechesco”. L'ItaliaDialettale 35.

---------. 1977. “Il contributo del veneziano e del greco alla lingua franca”.Venezia, centro di mediazione tra Oriente e Occidente (secoli XV-XVI),volume II, ed. by H.-G. Beck, M. Manoussacas & A. Pertusi, 523-535.Florence: Leo S. Olschki Editore.

Coutelle, Louis. 1977. “Grec, greghesco, lingua franca”. Venezia, centro dimediazione tra Oriente e Occidente (secoli XV-XVI), volume II, ed. by H.-G.Beck, M. Manoussacas & A. Pertusi, 537-544. Florence: Leo S. OlschkiEditore.

Duponchel, Laurent. 1979. “Le français en Côte d’Ivoire, au Dahomey et auTogo”. Le français hors de France ed. by Albert Valdman, 403-417. Paris:Honoré Champion.

Estrada y Zenea, Ildefonso. 1980. El quitrín. Havana: Editorial LetrasCubanas.

Faidherbe, Général. 1884. “L’Alliance français pour la propagation de lalangue française dans les colonies et les pays étrangers”. Revue SciéntifiqueJanuary 1, 1884, 104-109.

Fallas, Carlos Luis. 1975. Mamita Yunai. San José: Lehmann.Feced, Pablo [Quioquiap]. 1888. Filipinas, esbozos y pinceladas. Manila:

Estab. Tipog. de Ramírez y Compañía.Ferdinand, Astrid. 1996. The development of functional categories: the

acquisition of the subject in French. Ph. D. Dissertation, University of


Leiden. The Hague: Holland Academic Graphics (HIL dissertations, 23).Ferguson, Charles. 1971. “Absence of copula and the notion of simplicity: a

study of normal speech, baby-talk, foreigner talk and pidgins”. Hymes1971.141-151.

---------. 1975. “Towards a characterization of English foreigner talk”.Anthropological Linguistics 17.1-14.

Ferreras, Ramón Alberto. 1982. Negros (media isla: 4). Santo Domingo:Editorial del Nordeste.

Fontanella de Weinberg, María Beatriz. 1987. “Variedades lingüísticas usadaspor la población negra rioplatense”. Anuario de Lingüística Hispánica(Valladolid) 3.55-66.

Friedemann, Marc-Ariel. 1994. “The underlying position of externalarguments in French: a study in adult and child grammar”. LanguageAcquisition 3.209-255.

Fronzaroli, Pelio. 1955. “Nota sulla formazione della lingua franca”. Atti eMemorie dell'Academia Toscana de Scienze e Lettere "La Colombaria"20.211-252.

Gallegos, Rómulo. 1991. Canaima. Critical edition by Charles Minguet.Madrid: CEP de la Biblioteca Nacional.

Giancarli, Gigio. 1991. Commedie. Critical edition by Lucia Lazzerini.Padova: Editrice Antenore.

Göbl-Gáldi, L. 1934. “Esquisse de la structure grammaticale des patoisfrançais-créoles”. Zeitschrift für französische Sprache und Literatur 58 (5-6).257-295.

Goodman, Morris. 1964. A comparative study of creole French dialects. TheHague: Mouton.

Grion, G. 1891. “Farmacopea e lingua franca del dugento”. ArchivioGlottologico Italiano 12.181-186.

Guasti, Maria Teresa. 1993/4. “Verb syntax in Italian child grammar: finiteand nonfinite verbs”. Language Acquisition 3.1-40.

Gutiérrez, Joaquín. 1977. Puerto Limón. Havana: Casa de las Américas.Haedo, Diego de. 1927. Topografía e historia general de Argel. Madrid:

Sociedad de Bibliófilos Españoles.Hadel, Richard. 1969. “Modern creoles and Sabir”. Folklore Annual of the

University Folklore Association 1.35-43.Haegeman, Liliane. 1996. “Root infinitives, clitics and truncated structures”.

Generative perspectives on language acquisition ed. by Harald Clahsen, 271-307. Amsterdam & Philadelphia: John Benjamins.

Harris, Tony & Ken Wexler. 1996. “The optional-infinitive stage in child

JOHN M. LIPSKI138

English: evidence from negation”. Generative perspectives on languageacquisition ed. by Harald Clahsen, 1-42. Amsterdam & Philadelphia: JohnBenjamins.

Harvey, L. P., R. O. Jones & Keith Whinnom. 1967. “Lingua Franca in avillancico by Encina”. Revue de Littérature Comparée 41.572-579.

Herrero García, Miguel. 1966. Ideas de los españoles del siglo XVII. Madrid:Gredos.

Hinnenkamp, Volker. 1984. “Eye-witnessing pidginization? Structural andsociolinguistic aspects of German and Turkish foreigner-talk”. Papers fromthe York Creole Conference. York Papers in Linguistics ed. by Mark Sebba& Loreto Todd, v. 11, 153-166. York: University of York, Department ofLanguage.

Hoekstra, Teun & Nina Hyams. 1998. “Aspects of root infinitives”. Lingua106.81-112.

Hollyman, K. J. 1964. “L’ancien pidgin français parlé en Nouvelle-Calédonie”. Journal de la Société de Océanistes 20.57-64.

Holm, John. 1988. Pidgins and creoles, volume I: theory and structure.Cambridge: Cambridge University Press.

Hymes, Dell, ed. 1971. Pidginization and creolization of languages.Cambridge: Cambridge University Press.

Jaeggli, Osvaldo & Kenneth Safir. 1989. “The null subject parameter andsyntactic theory”. The null subject parameter, ed. by Osvaldo Jaeggli &Kenneth Safir, 1-44. Dordrecht: Kluwer.

Kahane, Henry & Renée Kahane. 1958. The Lingua Franca in the Levant;Turkish nautical terms of Italian and Greek origin. Urbana, Ill.: Universityof Illinois Press.

---------. 1976. “Lingua franca: the story of a term”. Romance Philology30.25-41.

Kokora, Pascal. 1983. “Language contact: the case of Ivory Coast”. CahiersIvoiriens de Recherche Linguistique 13.148-172.

Lang, George. 1992. “The literary settings of lingua franca (1300-1830)”.Neophilologus 76.64-76.

---------. 2000. Entwisted tongues: comparative creole literatures. Amsterdamand Atlanta: Rodopi.

Laprade, Richard. 1981. “Some cases of Aymara influence on La PazSpanish”. The Aymara language in its social and cultural context ed. byMartha J. Hardman, 207-227. Gainesville, Fla.: University Presses ofFlorida.

Lastra, Yolanda. 1995. “Is there an Indian Spanish?” Contemporary research


in Romance linguistics ed. by Jon Amastae, Grant Goodall, MarioMontalbetti & Marianne Phinney, 123-133. Amsterdam & Philadelphia:John Benjamins.

Legarda, Anselmo de. 1953. Lo vizcaíno en la literatura castellana. SanSebastián: Biblioteca Vascongada de los Amigos del País.

Leonard, Laurence. 1998. Children with specific language impairment.Cambridge, Mass.: MIT Press.

Lipski, John. 1985. The Spanish of Equatorial Guinea. Tübingen: MaxNiemeyer.

---------. 1986a. “Golden Age ‘black Spanish’: existence and coexistence”.Afro-Hispanic Review 5(1-2).7-12.

---------. 1986b. “Creole Spanish and vestigial Spanish: evolutionaryparallels”. Linguistics 23.963-84.

---------. 1986c. “Convergence and divergence in bozal Spanish”. Journal ofPidgin and Creole Languages 1.171-203.

---------. 1987. “Contemporary Philippine Spanish: comments on vestigialusage”. Philippine Journal of Linguistics 18.37-48.

---------. 1988. “Philippine creole Spanish: reassessing the Portugueseelement”. Zeitschrift für romanische Philologie 104.25-45.

---------. 1989. The speech of the negros congos of Panama. Amsterdam andPhiladelphia: John Benjamins.

---------. 1991. “On the emergence of (a)mí as subject in Afro-Iberian pidginsand creoles”. Linguistic studies in medieval Spanish ed. by Ray Harris-Northall & Thomas Cravens, 39-61. Madison: Hispanic Seminary ofMedieval Studies.

---------. 1995. “Literary ‘Africanized’ Spanish as a research tool: datingconsonant reduction”. Romance Philology 49.130-167.

---------. 1996. “Contactos de criollos en el Caribe hispánico: contribucionesal español bozal”. América Negra 11.31-60.

---------. 1998a. “El español de los braceros chinos y la problemática dellenguaje bozal”. Montalbán 31.101-139.

---------. 1998b. “El español bozal”. América negra: panorámica actual delos estudios lingüísticos sobre variedades criollas y afrohispanas ed. byMatthias Perl & Armin Schwegler, 293-327. Frankfurt: Vervuert.

---------. 1998c. “Latin American Spanish: creolization and the Africanconnection”. Publications of the Afro-Latin American Research Association(PALARA) 2.54-78.

---------. 1999a. “Creole-to-creole contacts in the Spanish Caribbean: thegenesis of Afro Hispanic language”. Publications of the Afro-Latin American

JOHN M. LIPSKI140

Research Association (PALARA) 3.5-46.---------. 1999b. “Evolución de los verbos copulativos en el español bozal”.

Lenguas criollos de base lexical española y portuguesa ed. by KlausZimmermann, 145-176. Frankfurt: Vervuert.

---------. 1999c. “Chinese-Cuban pidgin Spanish: implications for the Afro-creole debate”. Creole genesis, attitudes and discourse ed. by John Rickford& Suzanne Romaine, 215-233. Amsterdam & Philadelphia: JohnBenjamins.

---------. Forthcoming. “Panorama del lenguaje afrorrioplatense: vías deevolución fonética”. Anuario de Lingüística Hispánica (Valladolid).

López, Ventura. 1893. El filibustero. Madrid: Imprenta de la Viuda de M.Minuesa de los Ríos.

López Ornat, Susana, Almudena Fernández, Pilar Gallo & Sonia Mariscal.1994. La adquisición de la lengua española. Madrid: Siglo Veintiuno.

Lynch, Benito. 1926. El inglés de los güesos. Madrid: Espasa-Calpe.Makouta-Mboukou, Jean-Pierre. 1975. “Pourquoi un dialecte créole n’est-il

pas né sur le continent négro-africain d'expression française?”. Le françaishors de France 1973 ed. by Alain Guillermou, 102-109. Dakar & Abidjan:Les Nouvelles Editions Africaines.

Marcos, Habte-Mariam. 1976. Italian. Language in Ethiopia ed. by M.Bender, J. Bowen, R. Cooper & C. Ferguson, 170-180. London: OxfordUniversity Press.

Menn, Lise. 1989. “Some people who don’t talk right: universal andparticular in child language, aphasia, and language obsolescence”.Investigating obsolescence: studies in language contraction and death ed. byNancy Dorian, 335-345. Cambridge: Cambridge University Press.

--------- & Loraine Obler. 1990. “Cross-language data and theories ofagrammatism”. Agrammatic aphasia: a cross-language narrativesourcebook ed. by Lise Menn, Loraine Obler & Helen Goodglass, 1368-1389. Philadelphia & Amsterdam: John Benjamins.

Miceli, Gabriele & Anna Mizzuchi. 1990. “Agrammatism in Italian”.Agrammatic aphasia: a cross-language narrative sourcebook ed. by LiseMenn, Loraine Obler & Helen Goodglass, 717-816. Philadelphia andAmsterdam: John Benjamins.

Migliorini, Bruno. 1963. Storia della lingua italiana. Florence: Sansoni.Mühlhäusler, Peter. 1984. “Tracing the roots of pidgin German”. Language

and Communication 4.27-58.Naro, Anthony. 1978. “A study on the origins of pidginization”. Language

45.314-347.


Niedzielski, Henry. 1989. “A French based pidgin in Burundi”. Wheelswithin wheels ed. by Martin Pütz & René Dirven, 81-107. Frankfurt amMain: Peter Lang.

Nguyen, Dang Liem. 1977. “Cases and verbs in pidgin French (Tay Boi) inVietnam”. Journal of Creole Studies 1977.127-156.

Ortiz, Adalberto. 1976. Juyungo. Barcelona: Seix Barral.Ortiz López, Luis. 1998. Huellas etno-sociolingüísticas bozales y

afrocubanas. Frankfurt: Vervuert.---------. 1999. “El español haitiano en Cuba y su relación con el habla bozal”.

Lenguas criollas de base lexical española y portuguesa ed. by KlausZimmermann, 177-203. Frankfurt: Vervuert.

Pensalfini, Rob. 1995. “Pronoun case errors, both syntactic andmorphological”. Papers on language processing and acquisition (MITWorking Papers in Linguistics vol. 26) ed. by Carson Schütze, JenniferGanger & Kevin Broihier, 305-324.

Pichardo, Esteban. 1848. Diccionario provincial casi razaonado de vocescubanas. La Habana: Imprenta de M. Soler, 2nd ed.

Pierce, Amy. 1992. Language acquisition and syntactic theory. Dordrecht:Kluwer.

Pizzuto, Elena & Maria Cristina Caselli. 1992. “The acquisition of Italianmorphology: implications for models of language development”. Journal ofChild Language 19.491-557.

Poyen-Bellisle, René de. 1894. Les sons et les formes du créole dans lesAntilles. Baltimore: John Murphy & Co.

Quilis, Antonio & Celia Casado-Fresnillo. 1995. La lengua española enGuinea Ecuatorial. Madrid: Universidad Nacional de Educación aDistancia.

Reinecke, John. 1971. “Tây Bô'i: notes on the pidgin French spoken inVietnam”. In Hymes (1971).

Rice, Mabel & Kenneth Wexler. 1996. “A phenotype of specific languageimpairment: extended optional infinitives”. Toward a genetics of languageed. by Mabel Rice, 215-237. Mahwah, NJ: Lawrence Erlbaum.

Riley, Carroll. 1952. “Trade Spanish of the Piñaguero Panare”. Studies inLinguistics 10(1).6-11.

Rizzi, Luigi. 1993/4. “Some notes on linguistic theory and languagedevelopment: the case of root infinitives”. Language acquisition 3.371-393.

Rodríguez de Montes, María Luisa. 1981. Muestra de literatura oral enLeticia, Amazonas. Bogotá: Instituto Caro y Cuervo.

Román, José. 1978. Cosmapa. Managua: Distribuidora e Impresora de

JOHN M. LIPSKI142

Libros especializados.Santa Cruz, Melchor. 1996. Floresta española. Critical edition by

Maximiliano Cabañas. Madrid: Ediciones Cátedra.Schuchardt, Hugo. 1888. “Beiträge zur Kenntnis des kreolischen Romanisch

I: Allgemeineres über das Negerportugiesische”. Zeitschrift für romanischePhilologie 12.242-254. Translation in Schuchardt 1979.59-72.

---------. 1909. “Die Lingua Franca”. Zeitschrift für romanische Philologie33.441-461. Translation in Schuchardt 1979.26-47.

---------. 1979. The ethnography of variation: selected writings on pidginsand creoles ed. and translated by T. L. Markey. Ann Arbor: Karoma.

Silva-Corvalán, Carmen ed. 1995. Spanish in four continents: studies inlanguage contact and bilingualism. Washington, DC: GeorgetownUniversity Press.

Sloman, Albert. 1949. “The phonology of Moorish jargon in the works ofearly Spanish dramatists and Lope de Vega”. Modern Language Review44.207-217.

Stageberg, Norman. 1956. “Pidgin French grammar: a sketch”. ModernLanguage Journal 40.167-169.

Stratford, Billie Dale. 1989. Structure and use of Altiplano Spanish. Ph.D.Dissertation, University of Florida, Gainesville.

Tarkki, Pekka. 1995. El español en los campamentos de refugiados de laRepública Arabe Saharaui Democrática. Helsinki: Universidad de Helsinki,Instituto Iberoamericano.

Thompson, Roger. 1991. “Copula deletion in Spanish foreigner-talk: usingquestionnaires as a research tool”. Journal of Pidgin and Creole Languages6.89-106.

Trazegnies Granda, Fernando de. 1994. En el país de las colinas de arena:reflexiones sobre la inmigración china en el Perú del S. XIX desde laperspective del Derecho, tomo I. Lima: Pontificia Universidad Católica delPerú.

Urrutia Cárdenas, Hernán. 1995. “Morphosyntactic features in the Spanish ofthe Basque Country”. Silva-Corvalán 1995.243-259.

Vanelli, Laura. 1984. “Pronomi e fenomeni di prostesi vocalica nei dialettiitaliani settentrionali”. Revue de Linguistique Romane 48.281-295.

---------. 1987. “I pronomi soggetto nei dialetti italiani settentrionali dalMedio Evo a oggi”. Medioevo Romanzo 12.173-211.

Véronique, Daniel. 1994. “Naturalistic adult acquisition of French as L2 andFrench-based creole genesis compared: insights into creolization andlanguage change”. Creolization and language change ed. by Dany Adone &


Ingo Plag, 117-137. Tübingen: Niemeyer.Vianello, N. 1955. “ ‘Lingua franca’ di Barberia e ‘lingua franca’ di

Dalmazia”. Lingua Nostra 15/67-68.Vicente, Gil. 1912. Obras de Gil Vicente. Coimbra: França Amado.Vidal, João Evangelista de Lima. 1916. Por terras d’Angola. Coimbra: F.

França Amado.Wexler, Kenneth. 1994. “Optional infinitives, head movement and the

economy of derivations”. Verb movement ed. by David Lightfoot & NorbertHornstein, 305-350. Cambridge: Cambridge University Press.

---------. 1996. “The development of inflection in a biologically based theoryof language acquisition”. Toward a genetics of language ed. by Mabel Rice,113-144. Mahwah, NJ: Lawrence Erlbaum.

---------. 1998. “Very early parameter setting and the unique checkingconstraint: a new explanation of the optional infinitive stage”. Lingua106.23-79.

Whinnom, Keith. 1956. Spanish contact vernaculars in the Philippines.Hong Kong: Hong Kong University.

---------. 1977. “The context and origins of Lingua Franca”. Langues encontact--pidgins-creoles--languages in contact ed. by Jürgen Meisel, 3-18.Tübingen: Gunter Narr.

White, Lydia. 1996. “Clitics in L2 French”. Generative perspectives onlanguage acquisition ed. by Harald Clahsen, 335-368. Amsterdam &Philadelphia: John Benjamins.

Wood, Richard. 1971. “The Lingua Franca in Molière's Le bourgeoisgentilhomme”. USF Language Quarterly 10.2-6.

THE DEATH OF FRENCH IN MEDIEVAL ENGLAND*

D. GARY MILLERUniversity of Florida

0. IntroductionIt is widely agreed that Anglo-French (AF) was dead by ca.1400. Several

new pieces of evidence in support of that thesis are adduced here. In therecords of the London Grocers’ Company (LGrC), one can literally watch AFdie between 1400 and 1430. Not fortuitously, this correlates with the decline ofFrench calques in English and the increase of French suffixes in Englishhybrids, neither of which is characteristic of code-switching bilinguals.

Section 1 overviews the expansion of French in England. Section 2 reviewshow English survived the onslaught. In section 3, we switch to languages incontact and the signs of language death. Section 4 reviews some characteristicsof late Anglo-Norman (AN) literature, and section 5 targets the death of AF inthe LGrC records.

1. The Proliferation of French in EnglandAccording to the Domesday Survey [1086] of land, ownership, values for

taxation, etc., England had 190 lay barons, only five or six of which wereEnglish. Some 20 of the lay fiefs, plus a dozen of the larger monastic andepiscopal estates, held approximately 40% of the total surveyed land ofEngland (Berndt 1965:151-152). On the other hand, the figures for Frenchsettlers in rural areas were minuscule. Even urban figures were low: 160 inNorwich (E Midl), 43 in Shrewsbury (W Midl), 65 in Southampton (deepsouth), etc., and in no case greater than the number of English inhabitants(Berndt 1965:148). While French immigrants had thus penetrated much ofEngland, the greater density in the east implies that they landed in the SE andmigrated north to London and East Anglia.

Because of the absence of vernacular texts, it is difficult to ascertain thedegree of French assimilation during the early period. It appears that after 1100 * I wish to thank the organizers of LSRL XXX for inviting me to present this ‘outreach paper’,which benefited from discussion at the time and two subsequent anonymous reviewers.

D. GARY MILLER146

the use of Latin began to decline while Norman French became the official andprestige language to the extent that, in 1154, the English monks who wrote theAnglo-Saxon Chronicle discontinued their work — a sign of their pessimismregarding the future of English rule (SOE 58). In 1166 the highest court ofappeals became French (but the first legal documents and treatises in Frenchdid not appear until 1272). By the end of c12, even 70% of the Englishpeasants had abandoned their Anglo-Saxon names “in favor of thosefashionable with the Norman settlers” (Kibbee 1991:24).

Baugh & Cable (1993:174) estimate that some 10,000 French wordsentered English between 1150 and 1400 and that roughly 75% remain. Themore technical study by Dekeyser & Pauwels (1990) shows that massiveborrowing from OF started ca.1200, and that between 1175 and 1225, OEvocabulary dropped from 71.3% to 26.6%, while the OF component increasedfrom 6.7% to 12.6%, reaching a high of 30.2% in 1300, after which it droppedoff in favor of an increased Latin component, as shown in Table 1.1

OE OF ML L-F OE OF ML L-F1100 85.2 .0 1.0 5.9 1300 8.0 30.2 3.3 6.21125 78.7 4.9 .0 3.3 1325 5.3 28.2 2.8 7.41150 88.6 2.0 4.0 .0 1350 5.5 27.6 3.1 5.51175 71.3 6.7 1.2 .6 1375 2.3 15.6 16.7 12.41200 52.1 14.8 .9 2.1 1400 1.8 15.2 21.0 6.11225 26.6 12.6 1.5 1.8 1425 1.6 15.5 15.5 4.31250 17.8 14.8 2.2 7.4 1450 1.2 13.9 16.4 3.21275 13.0 23.0 4.4 7.5 1475 1.3 11.4 24.6 3.8

Table 1: Sources of English Borrowings 1100-1475

By the figures in Table 1, the increase of French loans during the period1375–1400 is less dramatic than the technical transfers from Medieval Latinand learnèd Latin-French constructs.

Evidence for even greater French infusion into English comes frommorphological transfer, which is rare (Weinreich 1966:31ff). Approximately130 hybrid derivatives (native base plus French suffix) are attested prior to1500 (see Miller 1997). The 15 top-frequency suffixes alone attest no fewerthan 100 hybrids prior to 1450 (38 -able, 18 -ess, 13 -ery, 10 -age, 9 -ment, 7-ard, 5 -ity). Of these, at least 64 antedate 1400 (21 -able, 18 -ess, 9 -ery, 3-age, 5 -ment, 4 -ard, 4 -ity). Given that most of the texts are literary, and that

1 OE = Old English; OF = Old French; ML = Medieval Latin; L-F = Latin-French. Forexpository simplicity, borrowings from other sources are not included in this count.

THE DEATH OF FRENCH IN MEDIEVAL ENGLAND 147

hybrids are most prevalent in monolingual authors, these relatively highnumbers are indicative of a major shift in language dominance.

2. The Resurgence of EnglishFollowing are the factors traditionally adduced for the survival of English

(cf. Kibbee 1991, LOB 218-227, Baugh & Cable 1993:ch.6):1) Because of fusion with Old Norse, English was too robust to be

obliterated (SOE 59), and French was a literary and cultural prestige languageof an elitist minority, especially feudal lords, bureaucrats, ecclesiastics,attorneys, educators, and writers.

2) A century after the Conquest, according to the Dialogus de Scaccario(“Dialogue of the Exchequer”) [1176-77], the judicial distinction betweenNormans and English could not be maintained because the Normans hadbecome totally assimilated through intermarriage. In c12 Anglo-NormanFrench was unknown to the vast English majority (Kibbee 1991:19-20).

3) In 1204, King John lost Normandy to the French. A decree by LouisIX in 1244 made it obligatory for nobility with estates on both sides of theChannel to give up one or the other as a declaration of allegiance (SOE 59f).

4) Even literary Anglo-Norman by c13 (144 works), and especially c14(only 14 works), was in serious decline.2 Rampant errors, treatises on Frenchorthography and grammar, French courses at Oxford, interlinear glosses ofFrench texts, and the wide circulation of Walter of Bibbesworth’s textbook3

[ca.1250] (Koch 1934) to teach French as a foreign language, reveal thatEngland had very few native French speakers (Käsmann 1961:14; Kibbee1991:57). This is also implied by injunctions against speaking English in theBenedictine abbeys after about 1250 (Berndt 1976:140).

5) In strong reaction against the Francophile Henry III, Edward I(crowned in 1272) in a royal letter of 1295 appealed to the patriotism of hissubjects to support his war against the French king who was bent on wiping theEnglish language off the earth (cf. Berndt 1976:146f, Kibbee 1991:34).

6) The plague of 1349 reduced the population by a third, and another in1361 ravaged the upper classes, the last stronghold of French. That, togetherwith the Hundred Years’ War [1337-1453], a trade war between England and 2 Vising (1923) lists a total of 281 AN works in the 13th century vs. only 48 in the 14th.3 “That it is intended for the gentry is proved by its subject-matter. It progresses from parts ofthe body to common animals and then household affairs necessary for a householder to knowin order to supervise servants, then to agriculture and hunting, and finally to the menu for agreat banquet. It was evidently very popular, since fourteen manuscripts [...] survive [...]. In allthe copies of the text which survive, English words are often written above the French, butthese so-called glosses increase in the later copies” (Legge 1980:112f).

D. GARY MILLER148

France, provided the English middle and lower classes with the wealth toacquire power. To replenish educators and professionals, new colleges openedup at Oxford and Cambridge which catered to merit rather than aristocracy.

7) By an edict of 1356, proceedings in the sheriff’s courts of Londonand Middlesex were to be in English (LOB 218).

8) The Statute of Pleading enacted by Parliament [1362] required thatall governmental and legal affairs be conducted in English rather than French“q’est trop desconue” (“which is too unknown”). The statute was aimed at thechicanery enabled by the incomprehensibility of legalistic French rather than atFrench per se, and pleadings continued to be in French (until c18), whileEnglish became the language of argument (LOB 218).

9) By 1362, when the chancellor opened Parliament in English for thefirst time, there was a renewed sense of nationalism prompted by the rise of thenew middle class. This led to a dislike of foreign things, including foreignlanguages (Kibbee 1991:60).

10) The 1370s witness the end of Anglo-Norman literature (Vising1923:73; Kibbee 1991:70-71). The last major work was John Gower’s Mirourde l’omme “Mirror of Man”, a moralizing poem of over 30,000 lines.

11) By the end of c14, English was the literary language of England. Thepopularity of Chaucer [c14e] gave particular impetus to a central Midland-London dialect.

12) The Bible translation and related works effected by John Wyclif inthe 1380s mark “the first appearance in literature of a widespread spokenkoiné” (Poussa 1982:79). Wyclif’s expressed purpose was to render the Bibleaccessible to the people in the language they knew. He singled out Latin andFrench as languages not used in speaking.

13) From the time of the Norman Conquest, all (known) wills werewritten in French until 1383. After 1400, English wills become frequent.

14) By 1385, according to John of Trevisa (quoted in Baugh & Cable1993:147), at least partly because of the results of the plagues, Englishpredominated as the medium of instruction in schools. The end of c14 andbeginning of c15 witness a proliferation of dialogues to teach French.

15) The 1430s were pivotal. Chancery, which handled all administrativework of the government, switched from Latin and French to English. The lastFrench petitions are recorded in 1447. Local town ordinances begin to betranslated into English ca.1430, and after 1450 all transactions are generally inEnglish.

16) Loss of the trade war with France in 1453 destroyed “the lastremaining motivations for maintaining French in English administration”(Kibbee 1991:68).


17) By the time of Malory’s Le Morte Darthur [a1470], knowledge ofFrench in England had become so limited even among writers that the Frenchphrases interspersed through his work are fraught with errors, “and presumablyhis contemporary readers were not sufficiently familiar with French to havebeen disturbed by some of his more surprising phrases” (Muir & Field1971:498).

3. Competing Languages, Morphological Transfer, and Language DeathThe most usual type of word-formation among code-switching bilinguals is

the calque, or loantranslation. Modern French has many phrasal and compoundcalques on English (Picone 1994a:263ff; 1996a:32-252; 1996b:76-86).Similarly, Middle English had numerous French calques. Prins (1952)documents some 370 loantranslated French phrases in the texts in his samplefrom 1056 to 1450 (cf. the roughly 500 phrasal calques and idioms of Frenchorigin studied by Orr 1962). From Prins’ table (1952:32) and list of eachcalque by date of first occurrence (1952:291-306), the (minimum) figures inTable 2 are available for genuine (G) and possible (P) French calques in ME.

G P TOTAL1100–1150 1 1 21150–1200 2 2 41200–1225 55 10 65 (52G, 9P in Ancrene Riwle)1225–1250 15 0 151250–1275 4 0 41275–1300 50 6 56 (14G, 5P in Cursor Mundi)1300–1350 52 3 551350–1400 85 6 91 (35G, 3P in Chaucer)1400–1450 26 3 29

Table 2: Genuine and Possible French calques in ME

The peak between 1200 and 1225, followed by an abrupt drop, correlates withthe end of the Norman Period. The subsequent surge correlates with the influxof the literarily prestigious Central/Parisian French. The sharp drop between1400 and 1450 not fortuitously correlates with the death of Anglo-French (seebelow) and the initial productivity of the loansuffixes in English.

While Modern French has many English loans with -ing, -er, -man (Picone1996a:334f), application of these to purely French bases is rare in ordinaryspeech, despite transfers such as forcing, on which the base is French as wellas English.

The highest rates of morphological transfer correlate with imminentlanguage death (see Thomason & Kaufman 1988:214-342), even though the

D. GARY MILLER150

extensive transfer on all levels can be a successful effort at self-preservation.4Consider the relevant features of language death (Dressler 1988):

1) There are “massive lexical loans from the dominant into the recessivelanguage” (Dressler 1988:185).

2) Native lexical items are replaced by substituted transfers.3) “[T]he borrowing of morphological suffixes is a symptom of decay if

the synonymous indigenous suffixes become unproductive at the same time”(ibid.:185).

4) There is a shift from indigenous neologisms (e.g. calquing) to “syste-matic borrowing of all neologisms from the dominant language” (ibid.: 186).

4. Late Anglo-NormanThis section reviews a few well-known features of late Anglo-Norman that

indicate convergence with English, but are not signs of language death.Examples are culled from the prose romance [c14] of Fouke le Fitz Waryn(FFW), based on a lost verse romance of late c13 (ed. Hathaway et al. 1975).

In (1), each entry contains an adjective without plural marking, for whichthe absence is indicated by a vowel in boldface.

Another typical feature, the recomposition of fused contractions into theirconstituent parts, is illustrated in (2).

(3) illustrates qe as a generic complementizer. Especially relevant are theexamples in which, from the standard French viewpoint, a special nominativeform (qui) would be expected, e.g. as in (3c) or the second qe in (3a).

4 Convergence was a successful strategy for Malinche Mexicano (Picone 1994a:267) andMiddle English (Miller 1997). Neither of these languages gave up their own word-formationprocesses in the face of language competition. Language death occurs when convergencebecomes “so total that a substrate language ceases to have its own identity” (Mike Picone,p.c.). According to Dressler (1988:184), “[l]anguage death occurs in unstable bilingual ormultilingual speech communities as a result of language shift from a regressive minoritylanguage to a dominant majority language. Language shift typically involves a gradualtransition from unstable bilingualism to monolingualism, that is the loss or ‘death’ of therecessive language.”


(1) Inconsistent concord on prenominal modifiers (cf. Vising 1916:11f;Pope 1952:463)

a. des douce chauntz5 (FFW 3.3f)of.the.pl sweet.sg song.pl“with sweet songs”

b. ces franke tenauntz (FFW 24.31)these frank.sg tenant.pl“these frank (free-born) tenants”

c. totes bone gentz (FFW 41.4)all.pl good.sg folk.pl“all good folks”

(2) Frequent recomposition of au(x) to à le(s), du/des to de le(s), etc.(cf. Pope 1952:465)

a. les portes de le chastel (FFW 18.37)“the gates of the castle”

b. quant vindrent al somet de le mont (FFW 46.30)“when they came to the summit of the mountain”

c. e autres dartz launcerent e gitterent a les gentz (FFW 35.28f)“and other darts they launched and threw at the people”

(3) qe as complementizer in all relative and verb-complement structures(cf. Pope 1952:467)

a. je vous pri qe mon frere qe la gist, quant il est mors, qeI you pray that my brother that there lies since he is dead thatvous facez enterrer son cors, qe bestes savages ne le devourentyou make bury.INF his body that beasts savage not him devour

(FFW 52.6f)“I pray you that my brother who lies there, since he is dead, thatyou have his body buried, that savage beasts not devour it”

b. molt estes folhardys qe vous osez entrer la mer6 (FFW 42.7f)much you.are foolhardy that you dare enter the sea“you are very foolhardy that you dare enter the sea”

c. la dame e l’enfant, qe febles erent, furent aporteez (FFW 39.12f)the dame and the child that weak were were transported“the woman and the child, who were weak, were transported”

5 The spelling aun for an is a typical AN feature dating from c13b (Vising 1916:53); similarly,oun for on occurs since c13m (ibid. 56f).6 Entrer “to enter” with accusative direct object is a calque on English (Burghardt 1906:80f).

D. GARY MILLER152

While q(u)e as generic complementizer is frequent in other dialects, AN qehas the same distribution as older English þe/þæt and MnE that: they are(relativizing) complementizers, not pronouns, and cannot receive case. With acase-assigning preposition, qy must be used, as shown in (4).

(4) qy as relative pronoun P-complementa. Dieu, en qy vous creez (FFW 44.26)

“God, in whom you believe”b. ce fust Fouke, son frere, ou qy il avoit combatu (FFW 55.38f)

“it was Fouke, his brother, with whom he had fought”

For the parallel between AN qe and E that, note that, just as qe is apparentlyungrammatical in (4), that cannot be used here either: *his brother with that hehad fought.

Syntactic calques on English are especially noticeable when they involveprepositions. One type involves preposition-noun combinations, as in (5),where outre bord is a calque on OE ofor bord [Ælfric+] (uncompounded)rather than on the Old Icelandic compound útanborðs “overboard”.

(5) Preposition-Noun calquesil chay outre bord enmy la mer (FFW 53.28f)“he fell overboard in the middle of the sea”

Another type of syntactic calque involves verb-preposition combinations,as in (6).

(6) Verb-Preposition calques (cf. Price 1985:224)a. le counte ly dona sur Asshesdoune (FFW 58.9)

the count to.him gave up Ashdown“the count gave up to him Ashdown”

b. chescun7 corust sur autre (FFW 10.28f)each ran over other“each overran (attacked) the other”

Especially interesting is (6b), which also occurs with the prepositionstranded when a clitic is extracted, as in (7a/b).

7 Chescun is the preferred form in Anglo-Norman (Vising 1916:58).


(7) P-stranding with clitic extractiona. Fouke e sa compagnie les corurent sur (FFW 45.2f)

Fouke and his company them ran over“Fouke and his comrades attacked them”

b. les chevalers e les esquiers de Lacy les corurent sur (FFW 17.36f)the knights and the squires of Lacy them ran over“the knights and the squires of Lacy attacked them”

The same type of P-stranding by clitic movement is typical of Old andMiddle English (ME), cf. (8a/b).

(8) P-stranding with clitic movement in early ME: LaZamon (L) [?a1200]a. Arður him faht wið (L 10,890)

Arthur him fought with“Arthur fought with him”

b. ofte he hire lokede on (L 9251)often he her looked on“often he looked on (at) her”

P-stranding of a slightly different type is found in some other dialects ofOld and Middle French, e.g. the Picardian Escoufle [1200–1225]; cf. (9).

(9) P-stranding in EscoufleL’empereïs ... Li vait encontre (Escoufle 1412f)the empress him.DAT goes toward“the empress goes toward him”

Examples of type (7) do not occur in continental Norman literature, e.g. theChanson de Roland, but those in (9), with dative complements, do (cf. Roland926), suggesting that they are in reality adverbs rather than Ps. Moreover, sincetype (7) is extremely rare in FFW, except in the very expressions that can beindependently demonstrated to be English calques, it is highly probable thatthe examples in (7) are calques on English syntax.

To conclude this section, Anglo-Norman literature shows convergence withEnglish to a certain degree, but the language remains distinctively recognizableas a variety of French.

D. GARY MILLER154

5. The Death of Anglo-FrenchAnglo-Norman “survived perhaps longest in the ports, witness the words in

The Oak Book of Southampton and The Register of Daniel Rough, bothfourteenth century” (Legge 1980:116).

Striking as the late features in §4 are, they are nothing compared to thechanges in the records of the London Grocers’ Company (LGrC) between 1400and 1430. The LGrC records are in Anglo-French (Rothwell 1992), or (rarely)Latin, until ca.1400. After the 1390s, French and English are distinguishedprimarily by function words (de/of(f), pur/for, le/the, etc.). Following are sometypical examples, cited from Kingdon (1883-1884) by date and/or volumeplus page (with modern numbers substituted throughout).

In (10) and (11), from the earlier period, there are no major surprises fromthe point of view of late Anglo-French. The generalization of qe in (10) tolocationals is perhaps not unexpected though apparently unparalleled in earlyliterature; cf. example (3) above.

(10) Item pur eskyper j tonel eill ou vyn...sour la keye qe il gist [1379] (1.55)

“Item: for shipping 1 ton of oil or wine on the quay where it lies”

(11) pur le8 netture et garbellage de poiure gingerue canellaet autres merchandises de Grocerie garbellables [1394] (1.73)

“for the cleaning and garbelling of pepper, ginger, cinnamon,and other wares of grocery capable of being garbelled”9

Some of the earliest uses of E –ing on French bases occur in (12). A calque onthe for-ing gerundial (12c) replaces the standard pur “for” + inf. purposive in(10).

(12) a. paie a le Gyldehalle pur enrollynge et escriuyngde diuerez dedez [1418] (1.126)

“paid to the Guildhall for enrolling and writingof diverse deeds”

b. pur inrollenge dun dede “for enrolling of a deed” [1420] (1.133)

c. Also payed for wrytyng and Enrollyng off owr dedys [1431] (2.207)

“also paid for writing and enrolling of our deeds”

8 On the numerous gender derailments in AN, see the ample documentation in Burghardt(1906:1-23); one even finds le nature and le Scripture.9 Rothwell (1992:35) translates garbellables as “requiring sifting”.


In (13) and (14), another convergence with English is seen in theadaptation of F en to E in, but the syntax is paralleled in AN literature; cf. en lesgardyns (FFW 16.16) “in the gardens”. And of course in + gerundial (14) is nodifferent syntactically from French en + -ant (etc.), but note that the noun basefaute is treated as a verb here, evidently under English influence.

(13) Item paie a Richard Burton pur Rentde nostre measoun in le Barge [1420] (1.133)

“Item: paid to Richard Burton for rent of our house in the barge”(14) Item ressu in fautynge de processions [1420] (1.133)

“Item: received in (de)fault(ing) of processions”

In (15a) and (16b), we see the typical inconsistent concord on prenominalmodifiers, as in (1). We also have in (15a) the purely English form shearingsubstituted for F tonsure “shearing”, which occurs only rarely in the laterentries, e.g. (15b). For a similar entry in English, cf. (15c).

(15) a. Item paye pur scheryng de Tout cest drapz et dyenge etGrowneyng de le memes drapz [1420] (1.137)

“Item: paid for shearing of all these cloths and dyeing andgraining of the same cloths”

b. paie pur tonsure de nostre drape [1418] (1.126)

“paid for the shearing of our cloth”c. payyd for our clothe greynyng (cf. Ross 1947/48:95) [1427] (1.153)

“paid for graining of our cloth”(16) a. Item paye a Bridde Browderer

pur browdynge de 94 gownes [1420] (1.138)

“Item: paid to Bridde, embroiderer,for embroidering of 94 gowns”

b. ... pur browdyng de lour gownys (1.136)

“for embroidering of their gowns”

By the time of entries (17) through (19), there is not much left that one cancall French. By the same token, from the OE point of view, there is not muchleft that one can call native Germanic-English either.

(17) Item paie pur takyn downe off tyle off an olde housz [1432] (2.211)

“Item: paid for taking down of tile off an old house”

D. GARY MILLER156

(18) a. Item paie a laborersz pur serchyng off grondys [1432] (2.211)

“Item: paid to laborers for searching of grounds”b. Item paies ... pur laborers pur les wers [1421] (1.143)

“Item: paid for laborers for the wares”(19) Item paie a ij laborersz pur dygyg vp off stone [1432] (2.213)

“Item: paid to 2 laborers for digging up of stone”[4 entries later: pur dyggyng vp off stone]

The -er agentives in (18) and (19) constitute an especially sticky area. Thefrequent contamination between E -er and L/F -o(u)r in ME is amplydocumented (Dellit 1906:78f; Adolphi 1910:34ff). French itself, of course,has a number of agentives in -er, but in late AN literature, there is an increasein -er forms. FFW has many -our agentives, e.g. conquerour (36.17), veneours(48.33) “hunters”, robbeours (45.4, 5) “robbers”, traytour (37.11, 14) “traitor”,jogelour (32.22) “juggler, minstrel”. Nevertheless, FFW attests not only formslike porter (F portier) “porter” (29.7), but also -er alternants to the -our forms,cf. jogelers sunt mensungers (33.18) “minstrels are liars”, and also the veryEnglish-looking mariner (41.34) (vs. F marin “sailor”).

Given the consistent -our spellings in earlier LGrC entries, e.g. toundour“shearer” [1345] (1.21), garbelour “garbeller” [1394], etc. (Rothwell 1992:34),the -er words, such as bocher “butcher” [1418] (1.125), carpenter [1432](2.211), laborer (18/19), garbeler (1.73), etc., if not outright hybrids (despitestandard French boucher, charpentier), at the very least exhibit strongconvergence between French and English.

6. ConclusionThe LGrC entries constitute strong evidence of a marked deterioration in the

knowledge of French in England after 1400 (cf. Miller 1997). The sharpdecline ca.1430 is doubtless due to the fact that AF was already dead by 1400and barely hanging on at the hands of a few diehards.

The same English formatives abound in LGrC records ca.1400 for which theFrench today have a predilection (§3; Picone 1996a), namely -er (laborer (18),browderer (16), etc.), -man (pur lez bargeman “for the bargemen” (1.149),etc.), and especially -ing. Frequent are scheryng, as in (15a), dyenge/dyyng(15a), enrollynge (12), browdyng (16), serchyng (18a), etc.; cf. also pavynge[1408], wynnynge [1418], dyggyng vp (19) – most in the pur gerundialconstruction, which in earlier entries was correctly construed with a Frenchinfinitive; cf. (10). However, despite the hundreds of -ing borrowings intoModern French, there are no genuine hybrids in speech (Picone 1996a). In ANliterature, English suffixes occur only on loanwords; very rare hybrids in -ness


occur without standard French variant readings (Burghardt 1906: 88ff). In itsdying phase, AF had many hybrids: escriuyng “writing” (12a), fautynge“default(ing)” (14), browdyng (16), browderer (16), etc.

In conclusion, the AF hybrids provide testimony of a typical language deathsituation, in which the dying language employs extreme measures ofconvergence as an attempted survival strategy. The death of AF ca.1400correlates with the decline (by two-thirds) of French calques in English (§3)and the increase (by double) of French suffixes in English hybrids (§1). Sincetrue code-switching bilinguals produce calques but not hybrids, both of thesefacts point to a monolingual (English) society ca.1400.

REFERENCES

Adolphi, Paul. 1910. Doppelsuffixbildung und Suffixwechsel im Englischen,mit besonderer Rücksicht auf das lateinisch-romanische Element. Inau-gural-Dissertation, Universität Marburg. Tübingen: Franz Fues.

Baugh, Albert C. & Thomas Cable. 1993. A History of the English Language.4th ed. Englewood Cliffs, N.J.: Prentice-Hall.

Berndt, Rolf. 1965. “The Linguistic Situation in England from the NormanConquest to the Loss of Normandy”. Philologica Pragensia 8.145-163.

---------. 1976. “French and English in Thirteenth-Century England: Aninvestigation into the linguistic situation after the loss of the Duchy ofNormandy and other Continental Dominions”. Aspekte der anglistischenForschung in der DDR: Martin Lehnert zum 65. Geburtstag (= Sitzungs-berichte der Akademie der Wissenschaften der DDR: Gesellschafts-wissenschaften, Jahrgang 1976, Nr.1), 129-150. Berlin: Akademie-Verlag.

Burghardt, Ernst. 1906. Über den Einfluss des Englischen auf dasAnglonormannische. Halle: Max Niemeyer.

Dekeyser, Xavier & Luc Pauwels. 1990. “The Demise of the Old EnglishHeritage and Lexical Innovation in Middle English: Two intertwineddevelopments”. Leuvense Bijdragen 79.1-23.

Dellit, Otto. 1906. Über lateinische Elemente im Mittelenglischen: Beiträgezur Geschichte des englischen Wortschatzes. Marburg: Elwert.

Dressler, Wolfgang U. 1988. “Language Death”. Linguistics: The Cambridgesurvey ed. by Frederick J. Newmeyer, Vol.IV: Language: The Socio-cultural context, 184-192. 4 vols. Cambridge: Cambridge University Press.

Escoufle: See Sweetser (1974).FFW: Fouke [Foulques] le Fitz Waryn: See Hathaway et al. (1975).

D. GARY MILLER158

Hathaway, E. J., P. T. Ricketts, C. A. Robson & A. D. Wilshere, eds. 1975.Fouke le Fitz Waryn. Oxford (Anglo-Norman Text Society): BasilBlackwell. [= Vising 1923: #214.]

Käsmann, Hans. 1958. “Zur Rezeption französischer Lehnwörter im Mittel-englischen”. Anglia 76.285-298.

---------. 1961. Studien zum kirchlichen Wortschatz des Mittelenglischen 1100–1350: Ein Beitrag zum Problem der Sprachmischung. Tübingen: MaxNiemeyer.

Kibbee, Douglas A. 1991. For to Speke Frenche Trewely: The Frenchlanguage in England, 1000–1600; its status, description and instruction.Amsterdam & Philadelphia: John Benjamins.

Kingdon, John Abernethy, ed. 1883–84. Company of Grocers of the City ofLondon, A.D. 1345–1463. Facsimile, transcription, and translation [Frenchtext only]. 2 vols. London: Richard Clay & Sons (1886).

Koch, John. 1934. “Der anglonormannische Traktat des Walter von Bibbes-worth in seiner Bedeutung für die Anglistik”. Anglia 58.30-77.

Legge, M. Dominica. 1963. Anglo-Norman Literature and its Background.Oxford: Clarendon Press.

---------. 1980. “Anglo-Norman as a Spoken Language”. Proceedings of theBattle Conference on Anglo-Norman Studies II. 1979 ed. by R. AllenBrown, 108-117. Woodbridge, Suffolk: The Boydell Press.

LGrC = London Grocers’ Company: See Kingdon (1883–84); cf. Ross(1947/48).

LOB = Price (1985).McCrum, Robert, William Cran & Robert MacNeil. 1993. The Story of

English. 2nd ed. New York: Penguin Books.MED = H. Kurath, S. M. Kuhn, J. Reidy, R. E. Lewis et al., eds. 1952-. Middle

English Dictionary. Ann Arbor: Univ. of Michigan Press.Miller, D. Gary. 1997. “The Morphological Legacy of French: Borrowed

suffixes on native bases in Middle English”. Diachronica 14.233-264.Muir, Margaret A. & P. J. C. Field. 1971. “French Words and Phrases in Sir

Thomas Malory’s Le Morte Darthur”. Neuphilologische Mitteilungen 72.483-500.

Orr, John. 1948. The Impact of French Upon English: The Taylorian lecture1948. Oxford: Clarendon.

---------. 1962. Old French and Modern English Idiom. Oxford: Basil Black-well.

Picone, Michael D. 1994a. “Lexicogenesis and Language Vitality”. Word45.261-285.


---------. 1994b. “Code-intermediate Phenomena in Louisiana French”. PapersFrom the 30th Regional Meeting of the Chicago Linguistic Society ed. byKatherine Beals et al., 1.320-334. Chicago, Ill.: Chicago Linguistic Society.

---------. 1996a. Anglicisms, Neologisms and Dynamic French. Amsterdam &Philadelphia: John Benjamins.

---------. 1996b. “Stratégies lexicogéniques franco-louisianaises”. Pluri-linguismes 11.63-99.

Pope, Mildred K. 1952. From Latin to Modern French with Especial Consi-deration of Anglo-Norman. 2nd ed. Manchester: Manchester UniversityPress.

Poussa, Patricia. 1982. “The Evolution of Early Standard English: The creol-ization hypothesis”. Studia Anglia Posnaniensia 14.69-85.

Price, Glanville. 1985. The Languages of Britain. [Chap.17: “Anglo-Norman”(pp.217-231).] London & Baltimore: Edward Arnold.

Prins, A. A. 1952. French Influence in English Phrasing. Leiden: Universi-taire Pers Leiden.

Purg: See Vising (1916).Roland: See Short (1990).Ross, Alan S. C. 1947/48. “The Vocabulary of the Records of the Grocers’

Company”. English & Germanic Studies 1.91-100.Rothwell, William. 1992. “The French Vocabulary in the archive of the

London Grocers’ Company”. Zeitschrift für französische Sprache undLiteratur 102.23-41.

Short, Ian. 1990. La Chanson de Roland: Édition critique et traduction. Paris:Librairie Générale Française.

SOE = McCrum et al. (1993).Sweetser, Franklin, ed. 1974. Jean Renart: L’Escoufle. Roman d’aventure.

Nouvelle édition d’après le manuscript 6565 de la Bibliothèque del’Arsenal. (= Textes Littéraires Français 211). Paris & Geneva: Droz.

Thomason, Sarah Grey & Terrence Kaufman. 1988. Language Contact,Creolization, and Genetic Linguistics. Berkeley & Los Angeles: Universityof California Press.

Vising, Johan, ed. 1916. Le purgatoire de Saint Patrice: Des manuscritsharléien 273 et fonds français 2198. Göteborg. (Repr., Geneva: Slatkine,1974.) [= Vising 1923: #104.]

---------. 1923. Anglo-Norman Language & Literature. Oxford & London:Oxford Univ. Press.

Weinreich, Uriel. 1966. Languages in Contact: Findings and problems. TheHague: Mouton.

DISCOURSE CONTEXT AND POLYSEMYSPANISH CASI

SCOTT A. SCHWENTERThe Ohio State University

0. IntroductionThis paper presents a study of the Spanish adverb casi “almost”, with

implications both for Hispanic linguistics and for the field of pragmatics moregenerally. First, for Hispanic linguistics, it provides a novel analysis of thecanonical meaning of casi, and a description/analysis of an innovative use ofthis adverb in the spoken Spanish of one region in Spain. For pragmatics (andalso semantics, given the diffuse boundary between the two), it offers up a casestudy which illustrates how to distinguish between contextually-determinedinterpretations of a lexical item, on the one hand, and conventionalizedsenses—polysemies—of the same lexical item, on the other.

The structure of the rest of the paper is as follows. In Section 1, themeaning of canonical casi is shown to consist of two separate components ofunequal status. Section 2 illustrates the innovative, ‘inverted’, use of casifound mainly in Spain’s Valencian Community. Section 3 provides argumentsfor treating this innovation as a separate polysemy, not as a contextual variantof canonical casi. Section 4 discusses other extensions of the meaning of casiand their relation to the analysis of canonical casi in Section 1. Section 5 offersconcluding remarks.

1. The meaning of canonical casiIn the field of pragmatics, there exists a longstanding debate over the

meaning of adverbs expressing approximation, e.g. Spanish casi or Englishalmost. The crucial question which this debate revolves around can beformulated as follows: Is the negative meaning of casi (or almost) part of theword’s conventional semantic content (a truth-conditional entailment, or a non-truth-conditional conventional implicature), or is it merely a conversationalimplicature (cf. Sadock 1981) and therefore felicitously cancelable in anappropriate discourse context? In other words, does casi p, where p = some

SCOTT A. SCHWENTER162

proposition, entail the negation of that proposition (~p), or does it merelyconversationally implicate ~p?

To answer this question, I will assume that the meaning of casi consists oftwo separate (but interacting) components, which can be termed the proximalcomponent and the polar component (cf. Hitzeman 1992, Sevi 1998). Considerthe contribution of these two components to the interpretation of the sentencesin (1) and (2):

(1) María casi terminó el trozo de pizza.“Maria almost finished the piece of pizza.”Proximal component = She approximated finishing it.Polar component = She did not finish it.

(2) María casi no terminó el trozo de pizza.“Maria almost didn’t finish the piece of pizza.”Proximal component = She approximated NOT finishing it.Polar component = She finished it.

It is important to note that any attempt to cancel the polar component in (1)leads to a contradictory interpretation (unless the first conjunct is interpreted asechoing an interlocutor). The uncancelability of (3) thus provides evidenceagainst an implicature analysis, and in favor of an analysis of the polarcomponent as part of the conventional semantic, specifically truth-conditional,content of casi (# = pragmatic infelicity):1

(3) #María casi terminó el trozo de pizza y (de hecho) lo terminó (deltodo).“Maria almost finished the piece of pizza and (in fact) she finishedit (all).”

On the other hand, the polar component of the meaning of casi can beexplicitly reinforced, as in (4) below. The felicity of this reinforcementpotentially supports an implicature analysis, since conversational implicatures,but not entailments, are typically reinforceable (Sadock 1981). Nevertheless,the examples show that reinforcement with casi is only possible when arhetorical contrast between the two conjuncts is highlighted, thereby requiring 1 Another possibility is that the polar component of casi is presupposed (Ducrot 1973).However, it fails standard presupposition tests in questions and conditionals. Thus, in ¿Casiterminó el trozo de pizza María? (“Did Maria almost finish the piece of pizza?”), or Si Maríacasi terminó el trozo de pizza, estará contenta (“If Maria almost finished the piece of pizza,she’ll be happy”) it is not necessarily presupposed that Maria did not finish the piece of pizza.

DISCOURSE CONTEXT AND SPANISH CASI 163

pero “but” instead of y “and” conjunction. As shown by Horn (1991), this isthe only context where entailed material can be redundantly affirmed in afelicitous fashion:

(4) a. María casi terminó el trozo de pizza {pero/#y} no lo terminódel todo.“Maria almost finished the piece of pizza but/#and she didn’tfinish it all.”

b. María no terminó el trozo de pizza {pero/#y} casi lo terminó.“Maria didn’t finish the piece of pizza but/#and she almost did.”

The possibility of redundant affirmation as in (4) therefore does not provideevidence in favor of an implicature-based account, since under conditions ofrhetorical contrast entailed material can be reinforced in just this way (Horn1991). The reinforceability of the polar component of casi thus remainscompatible with an entailment-based analysis (cf. Aranovich 1995).

What the examples in (4) also suggest is that the rhetorical orargumentative value of casi is positive, i.e., oriented towards p, despite thepresence of the polar component (entailing ~p). This orientation is wellillustrated by the pragmatic asymmetry between the constructed mini-discourses in (5) and (6):

(5) a. Vamos a sentarnos (ya), está casi lista la cena.“Let’s sit down (now), dinner’s almost ready.”

b. #No vamos a sentarnos (todavía), está casi lista la cena.“Let’s not sit down (yet), dinner’s almost ready.”

(6) a. #Vamos a sentarnos (ya), no está lista la cena.“Let’s sit down (now), dinner’s not ready.”

b. No vamos a sentarnos (todavía), no está lista la cena.“Let’s not sit down (yet), dinner’s not ready.”

Despite the fact that in neither (5a) nor (5b) is dinner ready yet, only (5a) is afelicitous utterance, while (5b) borders on contradiction. As (6a,b) without casishow, the felicity of (5a) is due precisely to the presence of this adverb. Thus,while casi p does indeed entail ~p, and therefore shares this entailmentproperty with the negative particle no, casi p is not argumentatively equivalentto no, as shown most clearly by the difference in acceptability between (5b)and (6b).

The argumentative orientation of casi can be seen even more clearly in thecontrast between casi and fellow adverb apenas “barely”, which both


semantically and argumentatively is the mirror image of casi (cf. Horn 1996),as illustrated by the contrast between (7a) and (7b):

(7) a. ¡Qué bien! Casi funciona el teléfono.“Great! The phone almost works.”

b. #¡Qué bien! Apenas funciona el teléfono.“Great! The phone barely works.”

It is crucial to note that this contrast between the two adverbs obtains despitethe fact that the phone actually works only in the second example (7b). That is,casi conveys a positive evaluation of the state of affairs in (7a) and contrastswith the negative assessment contributed to (7b) by apenas. More explicitly,casi p entails ~p but is argumentatively oriented toward p, while apenas pentails p but is argumentatively oriented toward ~p.2

Returning to the two components of meaning introduced above, we findthat there also exists a clear asymmetry between the proximal and the polarcomponents with respect to whether they can be the target of a response to aquestion (cf. Sevi 1998). This asymmetry is illustrated by the possible andimpossible interpretations of B’s responses in (8a) and (8b), given inparentheses:

(8) a. A: ¿Está casi lista la cena?“Is dinner almost ready?”

B: Sí.“Yes.”(= It is at least “almost ready” but no more than that)(≠ It is completely ready)

b. A: ¿Está casi lista la cena?“Is dinner almost ready?”

B: No.“No.”(= It is not at least “almost ready”)(≠ It is completely ready)

Indeed, these examples show that a bare sí/no answer to A’s question can onlybe interpreted as affirming or denying the PROXIMAL, not the polar, componentof the meaning of casi. The polar component of meaning is transparent, or‘inert’ (Horn 2000), to affirmation and denial. 2 Note that (7b) could be appropriate as an ironic utterance.


The ‘inertness’ of the proximal component has further empiricalconsequences. For instance, if casi p actually asserted a proposition of theform ~p (and, likewise, if apenas p actually asserted p), this assertive propertyought to be reflected in the patterning of other phenomena, such as thedistribution of negative polarity items (NPIs). That is, one would expect casi tobe a negative trigger for NPIs (and apenas not to be one). However, examples(9) and (10) show that we actually find the opposite result: it is apenas, notcasi, that serves as a trigger for NPIs (the presence of two triggers—apenasand no—in [10] also leads to ungrammaticality):

(9) Juan *casi/apenas movió un dedo por ellos.“Juan barely lifted a finger for them.”

(10) Juan casi/*apenas no movió un dedo por ellos.“Juan almost didn’t lift a finger for them.”

In line with Horn’s (2000) analysis of English almost, the evidenceprovided above shows that the polar component of the meaning of casi is infact entailed but ‘assertorically inert’. Hence it is transparent toaffirmation/denial and also to NPIs, which require a negative licensor. Bycontrast, the proximal component of meaning is both entailed and asserted.This asymmetry between the two components is summarized in Table 1:

Entailed AssertedProximal Component + +Polar Component + –

Table 1: Semantic/pragmatic status of casi’s meaning components

2. The innovation: ‘inverted’ casiBeyond the canonical use of casi, there exists an innovative use of this

adverb which can be heard frequently in (at least) Spain’s ValencianCommunity.3 This innovation, which I term ‘inverted’ casi for reasons whichwill become clear shortly, is exemplified by the naturally-occurring examples(documented in the cities of Valencia and Alicante) shown in (11) to (14):

3 Although speakers from other regions in Spain have also reported hearing this use, or reportusing it themselves, widespread consensus on this use has so far only been found amongspeakers from the Valencian Community. The use does not exist in Valencian dialects ofCatalan, and the speakers who produced (11)-(14) were not Valencian speakers, thus anexplanation in terms of language contact does not seem plausible.


(11) [Context: Speaker is trying to get out of his car, parked on a narrowstreet, but many cars are passing by and he has to wait. When he isfinally able to get out, he says]¡Casi salgo!almost I:exit“I almost did not get out!”

(12) [Context: Speaker is waiting for a friend at the door of atheater. The friend arrives just one minute before the session starts]¡Casi llegas!almost you:arrive“You almost did not arrive!”

(13) [Context: A conversation between two graduate students]A: Por cierto, mañana tenemos el examen.

“By the way, tomorrow we have the exam.”B: ¡Casi me lo dices!

almost me it you:tell“You almost did not tell me!” (OR “Now you tell me!”)

(14) [Context: Speaker opens a bag of cookies two days after havingbought them and sees that there are just a few left. Looking at herspouse, she says]¡Casi quedan!almost they:remain“There are almost none left!”

The utterances in (11), (12), and (13), though differing in terms of theirgrammatical subject (first person singular in [11], second person singular in[12] and [13]), are nevertheless quite parallel in interpretation. All of theseemphasize the ‘lateness’ of the event depicted in the casi clause and bringabout the interpretation that the speaker expected the event to occur earlier.And, while (14) seems rather different on the surface (it contains a third personsingular subject and a stative verb), it likewise evokes a temporal expectation,in this case related to the quantity of cookies left in the bag at utterance time.

In each of the preceding examples, casi p clearly cannot be understood asentailing the truth of ~p, since it occurs in a discourse context where the truthof the proposition p is obvious. Indeed, casi p in the above examples is beingused where speakers of many other dialects of Spanish would use the negatedversion casi NO p, in other words, where they would use canonical casi + thenegative no, as in (11’) to (14’) (read the symbol ! as “entails”):


(11’) ¡Casi no salgo!(! I have gotten out of my car)

(12’) ¡Casi no llegas!(! You have arrived)

(13’) ¡Casi no me lo dices!(! You have told me)

(14’) ¡Casi no quedan!(! There are some cookies left)

The ‘inverted’ use of casi as in (11)-(14) is heavily constrained, bothcontextually and grammatically. The principal contextual restriction is that itcan only occur when the event/situation described by the verb is patentlyobvious in the immediate discourse context. As for grammatical restrictions,the ‘inverted’ interpretation can only occur with verbs; an example like (15)where casi modifies an adjective can only receive the canonical interpretationthat the car is not blue, but nonetheless ‘approximates’ blue. That is, (15)cannot have the putative ‘inverted’ interpretation in which the car is blue, but itis only blue ‘by a little’.

(15) Ese coche es casi azul.“That car is almost blue.”(= The car is not blue, but it is ‘almost’ blue)(≠ The car is blue, but it is only blue ‘by a little’)

In addition, ‘inverted’ casi can only co-occur with verbs in the simplepresent tense, as illustrated by the infelicity of the preterite, perfect, andconditional verb forms in (16) (based on [12] above):

(16) Casi #llegaste/#has llegado/#llegarías.“You almost arrived/have arrived/would arrive.”

3. ‘Inverted’ casi: irony or polysemy?A question immediately arising from the interpretation of (11) through (14)

above is whether this use of casi represents a case of ironic usage, i.e., a casewhere casi p is being used to convey casi no p. The irony involved in such ause would hinge on exploitation of the polar component of meaning associatedwith the adverb’s canonical usage (see Section 1 above). Specifically, anutterance normally conveying ~p is produced in a discourse context where thetruth of p is obvious, thereby making the ~p interpretation absurd in thatcontext. Nevertheless, while an explanation of these examples in terms of irony


appears plausible a priori, it is interesting to note that they are certainly not‘typical’ cases of ironic usage with casi, like the following example in (17) is:

(17) [Context: I give my son a big plate of spaghetti for dinner. I goaway for a few minutes and when I return I see that he hasdevoured it all. I say,]¡Casi ha comido el niño!“The boy has almost eaten!”(interpretation: The boy has eaten A LOT)

In (17), an ironic interpretation arises from the conflict between myutterance, which entails that the boy has not eaten, and what is evident fromthe situational context: the boy has devoured all the spaghetti. This kind ofironic usage of casi is often found (in Peninsular Spanish, at least) inexpressions like casi nada, as in (18), where what B intends to convey, in spiteof what she says literally, is that Juan knows a lot about cars.4

(18) A: Juan no sabe de coches.“Juan doesn’t know about cars.”

B: Sí, ¡casi nada!“Yeah, almost nothing!”

(interpretation: he knows A LOT about cars)

The cases seen in (17) and (18) depend on the exploitation of the proximalcomponent (proximal-interpreted-as-distant) for their correct interpretation asironic utterances. In stark contrast, examples (11)-(14), if considered ironicusage, would depend on the exploitation of the polar component (negative-interpreted-as-positive). This difference is notable, but is not necessarilyevidence in favor of an analysis which treats (11)-(14) as something other thanironic usage. However, there do exist two stronger arguments against ananalysis of ‘inverted’ casi in terms of ironic usage and in favor of its status as aseparate polysemy, distinct from canonical casi.

The first argument comes from the clear non-equivalence between‘inverted’ casi and similar expressions like por poco “almost, nearly”. Withrespect to their canonical meanings, these two adverbs overlap to a greatextent. This overlap even extends to ironic uses of por poco, as shown by thepossibility of either utterance in the context of the following example:

4 Though not necessary for the argument being made here, intonation also plays an importantrole in examples like (18).


(19) [Context: The TV weather forecast calls for rain all day. A looksoutside and sees clear, sunny skies. She says the followingto B]Casi aciertan.“They’re almost correct [about the weather].”OR:Por poco aciertan.

Both of the possible utterances here are easily interpretable as being ironic:they convey, in the context, that the forecasters were ‘not even close’ to beingright. The parallelism between the utterances in (19) demonstrates that bothcasi and por poco can be used to convey similar ironic meanings. Thisparallelism is expected, since irony, like other kinds of (particularized)conversational implicature, is not necessarily linked to specific linguistic forms(Levinson 1983, Reyes 1994, Sperber & Wilson 1995).5

What is most striking, however, is that adverbial por poco (which is nearlysynonymous with canonical casi in many contexts) cannot be used in contextswhere the ‘inverted’ use of casi is possible, as shown by (20) (based on [13]above). If this use of casi were a simple case of irony, one would expect porpoco to be just as possible an option in such a context:

(20) A: Por cierto, mañana tenemos el examen.“By the way, tomorrow we have the exam.”

B1: ¡Casi me lo dices!“You almost tell me!” (OR “Now you tell me!”)

B2: Casi no me lo dices.“You almost don’t tell me.”

B3: #Por poco me lo dices.“You almost tell me.”

B4: Por poco no me lo dices.“You almost don’t tell me.”

As (20) illustrates, all of B’s possible responses are acceptable with the glaringexception of B3, a putative ‘inverted’ use of por poco. The impossibility ofresponse B3 thus offers evidence that: (i) the use of casi as in B1 is not strictlyironic, since near-synonyms should also allow similar ironic uses; and, as a

5 Indeed, according to Levinson (1983:117), an “ironic reading will be shared by all thedifferent ways of expressing the proposition that gives rise to it”. Such a view does notpreclude the possibility of preferred ways of expressing irony becoming conventionalized.


consequence, (ii) the use in B1 should be considered an innovative extensionof the adverb casi, i.e., the emergence of a new polysemy, here called‘inverted’ casi.

The second strong argument against analyzing ‘inverted’ casi as ironicusage—and in favor of analyzing it as a distinct polysemy—is that ‘inverted’casi always expresses contravention of some temporal expectation. As (21a)below shows, in some cases Casi no p is ambiguous between a temporal andanother kind of interpretation, e.g. a quantitative interpretation regarding thequantity of food eaten in (21a).

(21) a. ¡Casi no como!“I almost didn’t eat!”(Possible interpretation 1: I almost ran out of time to eat)(Possible interpretation 2: I ate a small quantity of food)

b. ¡Casi como!“I almost eat!”(= I almost ran out of time to eat and I expected to eat earlier)(≠ I ate a small quantity of food)

By contrast, as (21b) illustrates, ‘inverted’ casi p can only receive the temporalinterpretation; this utterance could not be used in a context where what thespeaker is trying to convey is that she ate a small quantity of food.

As the interpretation of (21b) implies, an added nuance is the contraventionof a temporal expectation: the speaker expected to eat earlier than she actuallydid. The importance of this nuance is shown by the fact that ‘inverted’ casi isin fact not a felicitous option in contexts where there is no contravention of atemporal expectation. Consider again example (14) above: here what thespeaker meant to express was that there ought to have been more cookies leftin the bag after only two days in the house (i.e., an expectation about thequantity of cookies remaining at a specific point in time). In this samediscourse context, the speaker could also have produced the utterance in (22):

(22) ¡Casi no quedan!“There are hardly any left!”

Crucially, however, the speaker in (14) could not have employed ‘inverted’casi in the constructed situation presented in (23) below. This is because theexpectation that a certain quantity of cookies should remain in (23) is muchweaker—if one exists at all—given the greater amount of time that has elapsedsince the cookies were bought:


(23) [Context: Speaker opens the bag of cookies two months afterhaving bought them and sees that there are just a few cookies left.Looking at her spouse, she says]:¡Casi no quedan!#¡Casi quedan!

In sum, the impossibility of the non-temporal reading in (21b), and theinfelicity of ‘inverted’ casi in the discourse context of (23), together providecorroboration of the claim that, in order to use the adverb with this meaning, itmust express contravention of specifically temporal expectations.

This semantics of expectation contravention is strong evidence for callingthe ‘inverted’ use a separate polysemy of the adverb casi. More evidencecomes from the fact that, with ‘inverted’ casi, speakers are drawing on thecanonical interpretation of casi p as “approximation to p”, but in a context inwhich the truth of p is plainly evident. As a result, the “not p” interpretationstrongly associated with canonical casi (i.e., the polar component of meaning)does not even enter into the interpretative picture. Rather, this negativemeaning has been transformed in such a way that the event described in p ispresented as having been on the verge of non-realization. The overall meaningof ‘inverted’ casi p can be schematically represented as in Figure 1:

Fig. 1: Schematic representation of the meaning of ‘inverted’ casi

Figure 1 represents in schematic fashion the meaning of examples (11)-(13).6 In each case, there is an underlying temporal expectation regarding theaction described in p. The shaded area of the figure represents the temporalrange in which p can possibly occur. The expected versus observed outcomesare shown using two distinct timelines, and the location of p in each is marked

6 Example (14), which contains an interrelated temporal and quantitative expectation (thequantity of cookies left at time t), requires a more complex representation which for spacereasons will not be presented here. Nonetheless, it also clearly involves the juxtaposition of anexpected quantity at time t and the actual observed quantity at time t.

p @ time t

p @ time t+x

Expected

Observed


by the vertical line within the shaded areas. The specific effect of ‘inverted’casi, then, is to invoke the expected timeline, and situate p at the end of thetemporal range in which this proposition can possibly take place on theobserved timeline.

4. Other extensions of the meaning of casiThere are other uses of casi which, like ‘inverted’ casi, disregard the polar

component of the canonical meaning in order to extend the proximalcomponent to other kinds of interpretation.

One type of contextually-bound extension found very commonly in Spain(and presumably in other Spanish-speaking regions too) is the use of casi as anillocutionary ‘hedge’, often for purposes of politeness, even in cases where thetruth of p is easily understood in the discourse context:7

(24) [Context: In a store in Spain]Clerk: ¿Se lo pongo en una bolsa?

“Should I put it in a bag for you?”Customer: Pues, casi mejor.

well almost better“Well, that would be better.”

The customer’s reply in (24) counts as an affirmative answer to the clerk’squestion. Her response cannot be cooperatively understood as directing theclerk not to put the merchandise in a bag, i.e., it cannot be understood asmeaning literally “that would be almost better”. Hence, this is a case where thepolar component of canonical casi is again left out of the interpretation, but theproximal component is pragmatically extended for purposes of attenuating theassertion. In (24), this attenuation strategy serves specifically to lessen thedegree to which the customer is imposing on the clerk.

The acceptable and unacceptable continuations of the customer’s responsein (24), illustrated in (25a) and (25b), respectively, show more clearly that thisresponse is compatible with an affirmative interpretation, but not with anegative one:

7 My thanks to Salvador Pons for drawing this use of casi to my attention and for pointing outits similarities to ‘inverted’ casi.


(25) a. Pues, casi mejor, si no le importa.well almost better if not you matter“Well, almost better, if you don’t mind.”

b. #Pues, casi mejor, así que no me lo ponga en una bolsa.well almost better so that not me it you:put in a bag“Well, almost better, so don’t put it in a bag for me.”

The continuation si no le importa “if you don’t mind” in (25a) is understood asreferring to the clerk’s putting the merchandise in a bag and is perfectlyfelicitous in the context of the customer’s response. However, the negativedirective in (25b) is infelicitous, since casi mejor, despite its literal (canonical)meaning, could not be interpreted as a negative response to the clerk’squestion.

A different, and apparently more semantically conventionalized, extensionof the meaning of casi has been discussed by García (1991). García presentsdata from the Mexican/Mexican-American speech community of ElPaso/Juárez which illustrate an extension of the adverb’s meaning to a distinctadverbial meaning of “generally, usually, principally”, as illustrated in (26)below (García 1991:18). Importantly, this extension too is highly compatiblewith the form’s canonical meaning, presented above in Section 1.

(26) Aquí, casi comemos en nuestra casa. En restorán, no.here almost we:eat in our house in restaurant not“Here, we generally eat at home. Not in a restaurant.”

As the English translation of (26) suggests, casi in this use limits the extent towhich a particular proposition maps onto real-world situations, but it does notexpress the negation of the proposition. The sense of approximation in thismeaning is transferred over to a different, aspectual, realm, specifically that ofhabituality, and conveys a degree reading which is slightly less than totalhabituality. Thus, once again, it is the proximal component of meaning whichhas advanced at the expense of the polar component of meaning, as one wouldpredict given the analysis above of this polar value as ‘non-asserted’.

5. ConclusionThere are two major conclusions to be drawn from this paper. The first of

these is that the polar component of the canonical meaning of casi, i.e., ~p, isindeed an entailment of this adverb, but it is one which is transparent or‘assertorically inert’ (Horn 2000). The ‘inert’ status of the polar component hasimportant consequences for possible extensions of the meaning of canonical


casi. In particular, it restricts semantic/pragmatic innovations to uses whichextend the proximal component of meaning in some way; indeed, in meaningextensions like the ‘inverted’ use and other non-canonical uses seen above, thepolar component is discarded entirely.

Second, the ‘inverted’ use of casi found in Spain’s Valencian Communityrepresents a distinct and innovative polysemy of this lexeme. Specifically, itillustrates the extension of the adverb’s canonical meaning to the realm oftemporal expectation, and to the contravention of this expectation. Thoughsimilar in some respects to a case of ironic usage, this meaning cannot beconsidered to belong to this class of contextually-dependent implicatures, asshown above in Section 3. The role of the discourse context is crucial for theextension of the meaning of casi to its ‘inverted’ sense: it can only arise incontexts in which the truth of p is evident/obvious, and where there is clearcontravention of an underlying temporal expectation.

REFERENCES

Aranovich, Raúl. 1995. “Spanish casi as a scalar operator”. BerkeleyLinguistics Society 21 ed. byJocelyn Ahlers, Leela Blimes, Joshua S.Guenter, Barbara A. Kaiser & Ju Namkung.12-23. Berkeley, Calif.:Berkeley Linguistics Society.

Ducrot, Oswald. 1973. La preuve et le dire. Paris: Maison Mame.García, MaryEllen. 1991. “Casi se usa así, casi. Reanálisis semántico en una

comunidad bilingüe”. Letras de Deusto 21.17-33.Hitzeman, Janet. 1992. “The selectional properties and entailments of

‘almost’”. Papers from the 28th Regional Meeting of the Chicago LinguisticSociety ed. by J. Denton, G. Chan & C. Canakis, 1.225-38. Chicago, Ill.:Chicago Linguistic Society.

Horn, Laurence R. 1991. “Given as new: When redundant affirmation isn’t”.Journal of Pragmatics 15.313-36.

---------. 1996. “Exclusive company: Only and the dynamics of verticalinference”. Journal of Semantics 13.1-40.

---------. 2000. “Assertoric inertia”. Paper presented at Linguistic Society ofAmerica, held in Chicago, January 2000.

Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge UniversityPress.

Reyes, Graciela. 1994. Los procedimientos de cita: citas encubiertas y ecos.Madrid: Arco/Libros.


Sadock, Jerrold M. 1981. “Almost”. Radical pragmatics ed. by Peter Cole,257-71. New York: Academic Press.

Sevi, Aldo. 1998. A semantics for almost and barely. Unpublished M.A. thesis,Tel-Aviv University.

Sperber, Dan & Deirdre Wilson. 1995. Relevance: communication andcognition. 2nd ed. Oxford: Blackwell.

NEW INSIGHTS INTO FRENCH REDUPLICATION*

MARY ELLEN SCULLENUniversity of Maryland, College Park

0. IntroductionMorin (1972) argues that spoken French contains an autonomous and

productive class of reduplicated words. Yet a unified analysis of this process hasbeen elusive due to problems of predictability, variation, and locus ofreduplication. In fact, the attested data have been so problematic that somescholars have privately suggested that reduplication in French is neitherproductive nor worthy of study.1

In this paper, I hope to show that both of these claims are patently false. Inparticular, I will present new data on ‘invented’ reduplications in the domain ofFrench baby-talk which strongly suggest that French reduplication is indeedproductive and that it can be analyzed straightforwardly within a constraint basedapproach such as Optimality Theory (Prince & Smolensky 1993, McCarthy &Prince 1993). Furthermore, from a theoretical standpoint, this analysis of Frenchreduplication is especially interesting in that several instances of the emergenceof the unmarked or TETU (McCarthy & Prince 1994) arise from careful analysisof the new data.

The paper is organized as follows. I first present the apparently problematicdata from French and in section 2 briefly discuss the challenges they pose for aunified analysis. Section 3 describes the new data on French reduplication, and anOT analysis is sketched out in section 4. Finally, section 5 discusses the TETUeffects which emerge in the domain of French reduplication.

* I would like to acknowledge my colleagues and students at the University of Maryland forparticipating in the data collection, as well as the participants of the LSRL XXX for insightfulcomments on the oral version of this paper. All errors remain my responsibility.1 See Scullen (1997) for a comprehensive list of the attested reduplicated forms in French culledfrom Kjellman (1920), Boudard & Etienne (1970), Morin (1972), Paradis (1979), George (1980),Plénat (1982), Borrell (1986), and Thiele (1987).

MARY ELLEN SCULLEN178

1. DataReduplicated forms in French are always consonant-initial disyllables with an

initial light syllable.2 Examples of reduplication from a monosyllabic input areprovided in (1). When the input is a light syllable, the final syllable of thereduplication is also light (1a), and when the input is a heavy syllable, the finalsyllable of the reduplicated form is also heavy (1b).

(1) a. foufou [fufu] “foolish” < fou [fu] cucul [kyky] “silly” < cul [ky]

b. baballe [babal] “ball” < balle [bal]bébête [bebet] “silly” < bête [bet]

If the input is polysyllabic, reduplication apparently involves truncation, as in (2).

(2) jojo [¥o¥o] “pretty” < joli [¥]li]Bubul [bybyl] “Bulgarian” < Bulgare [bylgar]tuture [tytyr] “car” < voiture [vwatyr]

In the special case of vowel-initial inputs, it appears that the first syllable to theright of the initial onsetless syllable is reduplicated, as in (3).

(3) fanfan [fãfã] “child” < enfant [ãfã]toto [toto] “car” < auto [oto]titi [titi] “sharp shooter” < artilleur [artijœr]

In consonant-initial polysyllabic inputs, either the left-edge or the right-edge of theinput can reduplicate, as shown in (4) and (5), respectively.

(4) gaga [gaga] “foolish” < gâteux [gatø]coco [koko] “communist” < communiste [k]mynist]pipi [pipi] “to urinate” < pisser [pise]

2 For the sake of completeness, there are a handful of reduplicated forms with an initial heavysyllable. These cases of total reduplication are either cases of morphological derivation (i) oronomatopoetic forms (ii). These forms will not be considered further here.

(i) cache-cache [kaïkaï] “hide and seek” < cache “hides3sg”pousse-pousse [puspus] “rickshaw” < pousse “pushes3sg”

(ii) miamiam [mjamjam] “yum yum”tam-tam [tamtam] “drum beat”

NEW INSIGHTS INTO FRENCH REDUPLICATION 179

(5) popeau [popo] “hat” < chapeau [ïapo]pépère [peper] “grand-father” < grand-père [grãper]

To summarize, reduplicated forms in French are always consonant-initialdisyllables with an initial light syllable and a variable second syllable.

2. Apparent problems in an OT account of French reduplicationGiven this description of French reduplication, a constraint-based account

within OT would seem rather straightforward. The constraint ONSET could accountfor the lack of vowel-initial reduplicated forms and NOCODA for the first lightsyllable. Assuming further Correspondence Theory (McCarthy & Prince 1995),which governs the relationship between the input and the output and that betweenthe base and the reduplicant, the constraints MAXIO and MAXBR would certainlyalso play a role. The ranking: MAXIO >> NOCODA >> MAXBR is established inTableau (1) with the reduplicated bébête “silly” from bête.

/ RED + b e t/Input MAXIO NOCODA MAXBR

a. b e t . b e t tt !b. L b e . b e t t t

c. b e . b e t !

Tableau 1: bébête < bête

The most faithful candidate (a) does not win because NOCODA dominates MAXBR,which would require identity between the base and the reduplicant. Note however,that MAXIO in turn dominates NOCODA and accounts for the final coda consonantin the attested output in (b).

Somewhat unexpectedly in the attested forms in (6), the final coda consonantfrom the input is not present in the output. Accounting for these forms would leadto an apparent ordering paradox as NOCODA would have to dominate MAXIO,contrary to the constraint ranking established above.

(6) sœsœ [sœsœ] “sister” < sœur [sœr]fifi [fifi] “little girl” < fille [fij]

The new data on ‘invented’ reduplications in French, presented in section 3.2, willelucidate this apparent ordering paradox.

The truncation accompanying the reduplication of polysyllabic inputs(illustrated in (4)-(5) above) poses a further complication. In fact, given only theconstraint ranking established thus far, the high ranking MAXIO would favor


retaining all segments of the input in the output. Thus, a reduplicated polysyllabicinput would never yield a disyllabic output. Tableau (2) illustrates how theincorrect winner, the unattested *jojoli, would emerge as the optimal candidatesince it best satisfies the constraint MAXIO.

/ RED + ¥ o l i / MAXIO NOCODA MAXBR

a. ¥ o . ¥ o t t !b. ¥ o . ¥ o l t ! t t

c. * L ¥ o . ¥ o . l i t t

Tableau 2: Incorrect winner for joli

A final problem in analyzing French reduplication is identifying the locus ofreduplication. In other words, does the right-edge or the left-edge of the inputreduplicate? The preponderance of left-edged forms like those in (4) as well asforms like titi from artilleur “sharp shooter” would seem to argue thatreduplication in French operates from the left-edge of words and hence for aconstraint like ANCHOR-L (viz, the left edge of the reduplicant must correspond tothe left edge of the base). On the other hand, it is difficult to see how to accountfor forms like those in (5) above, in which the right-edge of the input isreduplicated.

How then to offer an insightful account of French reduplication with theseseemingly important stumbling blocks? To wit, it does not appear to be clear fromthe data whether reduplication in French is right-edged or left-edged. Further, thepresence of a final coda consonant in forms like bébête “silly” and baballe “ball”versus its absence in fifi “little girl”and sœsœ “sister” is also troubling. Finally, howshould one account for the apparent truncation which accompanies the process ofreduplication? In the next section, I will show that new data on the productiveprocess of reduplication in the domain of ‘baby-talk’ elucidates these problemsand allows for an elegant account of reduplication within OT.

3. A new look at French reduplication: the import of new dataFaced with the problematic data described above and doubts about the

productive nature of French reduplication, I devised a survey to elicit new data onpotential reduplications in French. The survey itself is discussed in section 3.1 andthe interesting results in 3.2.

3.1 The survey instrument and subjectsThe goal of the study was to encourage native speakers to produce new

reduplicated forms that were not necessarily attested in the literature cited above.


The survey contained a list of 95 words that were read to the subjects, and theywere asked to provide a reduplicated form for each word in turn.

Given the productivity of reduplication in ‘baby-talk’(i.e., words used toimitate children and talk to small children, pets, and occasionally loved ones), Iincluded words from semantic fields like toys, animals, playground equipment,clothing, and food. A representative sample of these words is provided in (7).

(7) jupe [¥yp] “skirt” image [ima¥] “picture”chemise [ïcmiz] “shirt” éléphant [elefã] “elephant”salopette [salopet] “overalls” jardin [¥ardE‚] “yard”

I also included a number of forms which were attested in the literature on Frenchreduplication, to test their validity in current spoken French. A number of theseforms are listed in (8) with the attested forms provided on the right.

(8) Attested Formssœur [sœr] “sister ” sœsœ [sœsœ]Bulgare [bylgar] “Bulgarian” Bubul [bybyl]artilleur [artijœr] “sharp shooter” titi [titi]communiste [komynist] “communiste” coco [koko]

The survey was administered to five subjects who were all native speakers ofMetropolitan French and ranged in age from 22 to over 50. Nearly all wereuniversity educated. Although some individual variation was observed, thespeakers were remarkably consistent in producing ‘invented’ reduplications of theshapes discussed below and in refusing reduplications that do not fit thesepatterns, even those reduplications which were attested in the literature.3 The nextsection provides an overview of the findings from this survey.

3.2 Results and findings from survey on ‘invented’ reduplicationsAccording to the results of this survey, the most productive process of

reduplication in current spoken French is when the input is a single heavy syllableas shown in (9).4

3 Variation among speakers was revealing. For example, one conservative speaker refused toprovide a reduplicated form for any word which did not have a pre-existing reduplicated form inFrench (preferably in the dictionary!). A few other speakers were happy to provide reduplicatedforms for nearly every word. However, there were some words which all speakers uniformlyrefused to reduplicate. Where relevant, these cases are discussed below. 4 I am defining productivity here not only in terms of how many individual words are reduplicatedbut by how many speakers. In other words, not only were a large number of heavy monosyllabic


(9) jujupe [¥y¥yp] “skirt” < jupe [¥yp]tatasse [tatas] “cup” < tasse [tas]véverre [vever] “a glass” < verre [ver]popomme [pop]m] “apple” < pomme [p]m]gogomme [gog]m]5 “eraser” < gomme [g]m]

All of the reduplicated forms in (9) correspond to the preferred shapesdescribed above, namely a consonant-initial disyllable with an initial light syllable(cf. bébête “silly”). Crucially, the final syllable of these reduplicated forms isheavy just like the final syllable of the input.

Reduplication from inputs consisting of a single light syllable was lessproductive but still rather robust. Examples are given in (10) and are similar to theforms in (1a) like foufou [fufu] “foolish.”

(10) chatchat [ïaïa] “cat” < chat [ïa]chienchien [ïjE‚ïjE‚] “dog” < chien [ïjE‚]

In the case of vowel-initial inputs, all of the speakers uniformly respected theconstraint that reduplicated forms in French be consonant-initial and declined toprovide vowel-initial reduplications. Some of the consonant-initial reduplicationsthey did give are listed in (11).

(11) phantphant6 [fãfã] “elephant” < éléphant [elefã]gnéegnée [njenje] “spider” < araignée [arenje]siesiette [sjesjet] “plate” < assiette [asjet]

Although these forms conform to the expected shape for French reduplication,they are surprising in another way. Contrary to the vowel-initial inputs in (3) liketiti from artilleur “sharp shooter,” these reduplications are not formed from thefirst consonant-initial syllable of the input but rather from the right-edge of theword. This would seem to provide a tentative answer to the question of the locus

inputs in the survey reduplicated, but they were reduplicated by a high percentage of the speakersas well.5 All of my informants seemed to agree that the reduplicated forms popomme and gogomme werepronounced with a closed o, [o], in the initial syllable. However, this may be simply due toregional influences. Several native French speakers at the LSRL meeting were unanimous inagreeing that in their dialects of French, these words would be pronounced with an open o, []], inthe first syllable, hence [p]p]m] and [g]g]m].6 I have simply invented the spellings (based on the input’s orthographic form), as the informantswere not asked to write the forms down but only to pronounce them aloud.


of reduplication in French; these forms suggest that French reduplication is right-edged.

This hypothesis is confirmed by the productive process of reduplication withpolysyllabic inputs as shown in (12).

(12) toto [toto] “boat” < bateau [bato]toto [toto] “knife” < couteau [kuto]mimise [mimiz] “shirt” < chemise [ïcmiz]sésette [seset] “socks” < chaussette [ïoset]nanane [nanan] “banana” < banane [banan]pépette [pepet] “overalls” < salopette [salopet]

In virtually every case of a polysyllabic input, the ‘invented’ reduplicated formsupplied by the informants was formed on the right-edge of the word. Combinedwith the data above on vowel-initial inputs, it seems clear that Frenchreduplication is indeed right-edged. This interpretation finds additionalcorroboration in the stress facts of French, since stress falls on the final syllableof a word or word group and hence can be said to be right-edged as well. Theproblem of the attested left-edged reduplications (e.g., the forms in (4) and titifrom artilleur “sharp shooter”) will be addressed in section 4.4 below.

With regard to attested reduplications, the survey data is also quite revealing.For example, my informants did not spontaneously produce the problematicreduplicated forms sœsœ “sister” and fifi “little girl” when asked for reduplicatedforms for sœur and fille. Instead they produced the forms in (13).

(13) sœsœur [søsœr] “sister” < sœur [sœr]fifille [fifij] “little girl” < fille [fij]

As discussed above, the attested forms sœsœ and fifi (Kjellman 1920) raised anapparent ordering paradox due to the absence of the input coda consonant in theoutput. It appears, however, that although these forms existed at one point, theyare no longer acceptable reduplications in spoken French. In fact, when I asked theinformants if I could say fifi or sœsœ in addition to the forms they had suggested,they explicitly rejected these forms.7

7 The speakers were also quasi-unanimous in refusing to produce a reduplicated form for eitherBulgare or artilleur, even when told that such a form did exist and had been reported for French.In fact, the forms Bubul and titi were documented by George (1980), who claims they were partof the creation of military slang during the First World War.


A final intriguing point raised by the new data on reduplicated forms was therefusal by several speakers to produce a reduplicated form for the inputs frère [frer] “brother” and fils [fis] “son.” This is somewhat surprising given therobustness of the phonologically and semantically similar reduplicated forms sœsœr“sister” and fifille “little girl/daughter.” Several speakers speculated thatreduplicated forms like *fréfrère [frefrer] and *fifils [fifis] were possible but wereblocked by the existence of another slang term derived from the same input (e.g.frérot [frero] “brother” and fiston [fistõ] “son”8). It is not clear at this time how toprovide a theoretical account for this interesting observation, but it is certainlyworth noting.

In the next section, we turn to a re-analysis of French reduplication based onthe insights provided by the new data on ‘invented’ reduplications.

4. (Re-)analysis of French reduplicationAs mentioned above, insights into the process of reduplication in French

gleaned from the new data on ‘invented’ reduplications have answered many ofthe questions raised in section 2 and allow for an elegant account of Frenchreduplication within OT as presented in the following sections.

4.1 A prosodic constraint [óó]In order to account for the invariant disyllabic output shape of French

reduplicated forms as well as the apparent truncation which accompanies thereduplication of polysyllabic inputs, some sort of size constraint is required. Isuggest that one way to tackle this problem is to posit a prosodic template forFrench reduplication in the form of a high-ranking constraint, [óó], which wouldprohibit reduplicated forms longer than two syllables.9

4.2 ANCHOR-RAs was made clear by the new data in section 3.2 above, French reduplication

is clearly right-edged. To account for this, the constraint ANCHOR-R certainly plays 8It is not simply the presence of another slang term for the word which appears to blockreduplication, as the slang term frangine “sister” does not prevent speakers from producing sœsœr.Nor is it the presence of another slang reduplicated form, since the common toutou “dog” did notprevent the formation of chienchien by several speakers. A reviewer suggests that this blockingcould have something to do with the phonological similarity between the potential forms and thealready existing forms as a type of ‘gradient attraction’ (Burzio 1999).9 Positing a prosodic template such as [óó] is somewhat problematic in the current view ofProsodic Morphology. For example, McCarthy & Prince (1999) state clearly that such templatesdo not exist and argue that the ‘template’ itself can be seen as a result of TETU. It is not clear thatthe French reduplication data can be explained in this way, and I leave this as a question for furtherresearch.


a role.10 The interaction of [óó] and ANCHOR-R is shown in Tableau (3). It shouldbe noted that the only ranking which can be clearly established is the basic rankingdiscussed above: MAXIO >> NOCODA >> MAXBR.

/ RED + ï c m i z/ [ ó ó ] ANCHOR-R MAXIO NOCODA MAXBR

a. ï c m i z ï c m i z t ! tt

b. ï c ï c m i z t ! t ttt

c. ï c ï c t ! ttt

d. L m i m i z tt t t

e. m i m i ttt !

Tableau 3: The ‘invented’ mimise < chemise

4.3 The role of ONSETNow that French reduplication has been characterized as right-edged, it

appears there is no need to say anything special about vowel-initial forms. If thereduplicated form is created from the right-edge of the word, the presence orabsence of an onset consonant on the left-edge of the word is irrelevant. Thus, itappears that the constraint ONSET plays no role in French reduplication. There isat least one vowel-initial form, however, which shows the crucial interactionbetween the constraints ONSET and ANCHOR-R and establishes the constraintranking ONSET >> ANCHOR-R. This is illustrated in Tableau (4) for thereduplicated form baba from ébahi [ebai] “astonished.”

/ RED + e b a i / ONSET ANCHOR-R

a. i i t t !b. L b a b a t

c. e e t t ! t t

Tableau 4: Interaction of ONSET and ANCHOR-R

In this case, the candidate satisfying ANCHOR-R (candidate a) is not the optimalcandidate. Rather, candidate (b), which satisfies the higher ranked ONSETconstraint, is the winner.

10This account runs counter to Nelson (1999) who argues, in part based on data from Frenchhypocoristics, that there is no need for the constraint ANCHOR-R in the grammar. She argues insteadfor the interaction of ANCHOR-L with other constraints such as ANCHOR- ó�. Given that theprosodic head in French is the right-edge of a word, it is possible that one could simply useANCHOR-ó�instead of ANCHOR-R to account for French reduplication. I leave this as an openquestion for future research.


4.4 Left-edge reduplicationAlthough the importance of the constraint ANCHOR-R in accounting for French

reduplication has been demonstrated, there are still a fair number of left-edgedreduplications in French, including a few from the corpus of ‘invented’reduplications, which must be explained. Attested left-edge reduplications aregiven in (14) and ‘invented’ forms in (15).

(14) jojo [¥o¥o] “pretty” < joli [¥oli]cracca [krakra] “dirty” < crasseux [krasø]coco [koko] “communist” < communiste [komynist]Bubul [bybyl] “Bulgarian” < Bulgare [bylgar]

(15) soso [soso] “socialist” < socialiste [sosjalist]gégé [¥e¥e] “General” < général [¥eneral]

In virtually every case, left-edge reduplication occurs when the right-edge containsa morphological suffix (e.g., -eux, -iste, -al).11 So descriptively, left edgereduplication in French occurs when the right edge of the word contains suffixalmaterial. I propose to account for this with a constraint specifying that thereduplicative morpheme contain lexical material or RED . ROOT. This isillustrated by Tableau (5) for cracra “dirty” from crasseux.

/ RED + k r a s ø / RED . ROOT ANCHOR-R

a. s ø s ø t !b. L k r a k r a t

Tableau 5: Cracra < crasseux

The unusual reduplication titi for artilleur “sharp shooter” can be explained withreference to both ONSET and RED . ROOT. Anecdotal confirmation of the interplayof these constraints can be found in the words of one of my informants whoexplained why the only possible reduplication for that word would be titi. Looselytranslating from French, she said, “It can=t be arar [reflecting implicit knowledgeof ONSET]. It can=t be llelleur, [jœjœr], [knowledge of RED . ROOT]. So you=re leftwith titi.” A perfect prose explanation for constraint interaction at work!

11 The case of bibi from biberon “bottle” is a lexical exception.


5. TETU effects in French reduplicationFrom a theoretical standpoint, this analysis of French reduplication is also

interesting as it reveals several instances of ‘The Emergence of the Unmarked’ orTETU (McCarthy & Prince 1994). TETU refers to the observation that a markedstructure generally allowed in a language, (e.g. in French, syllables without anonset as well as syllables with a coda) can be banned in a special domain, likereduplication, where an unmarked structure emerges (e.g. the ban on onsetlessreduplicated forms and the invariant initial light syllable in French reduplication).TETU results from the differential ranking of two sets of Faithfulness Constraints(e.g. MAX IO and MAX BR) with regard to a Markedness constraint (e.g. ONSET orNOCODA).

5.1No coda in reduplicantOne manifestation of TETU in French arises from the constraint ranking MAXIO

>> NOCODA >> MAXBR, which effectively prohibits codas in the reduplicant (theinitial syllable of the reduplicated form) which do show up in the final syllable.Examples include attested reduplications like bébête “silly” and fofolle “foolish”and ‘invented’ forms like mimise “shirt” and nanard “duck”. Compare thesereduplicated forms with regular words like armoire “wardrobe,” bifteck “steak,”and personne “person,” which all contain word-internal codas.

5.2No onset-less reduplicantA further example of TETU in French reduplication is the absence of vowel-

initial reduplicated forms, although onsetless syllables occur word-initially andword-medially in French. Recall that vowel-initial inputs such as ébahi“astonished,” auto “auto,” and araignée “spider” yield consonant-initialreduplications such as baba, toto, and gnéegnée.

5.3Mid-vowel qualityAlthough this example is less clear-cut than the two cases mentioned above,

a further TETU effect to emerge in French reduplication arises with the quality ofthe mid-vowel in the reduplicant. From inputs with a lax (or open) vowel such asbête [bet] “stupid” and verre [ver] “glass,” the preferred output contains a tense(or closed) vowel in the initial open syllable (e.g. bébête [bebet] “silly” andvéverre [vever] “glass”). The closing of the mid-vowel in an open syllable appearsto correlate with a tendency in French, among other languages, to prefer tensevowels in open syllables and lax vowels in closed syllables and might beaccounted for by some type of constraint which prohibits lax vowels in open


syllables.12 Yet, this case of TETU in French reduplication is problematic in thatthere appears to be much variation in the vowel quality of the reduplicant. Forexample, some speakers pronounce an open vowel in the initial open syllable (e.g.[bebet]). I have also only mentioned the case of the front unrounded mid-vowelpair e/e. As mentioned in footnote 6, the case of o/] is much less clear, with somespeakers preferring forms like fofolle [f]f]l] and others [fof]l]. This is clearly aquestion for more empirical data as well as further research.

6. ConclusionLike much else about the phonology of French, the process of reduplication is

far from straightforward and not without exceptions. Yet it is clear from the‘invented’ reduplication data discussed above that reduplication in spoken Frenchis a productive process which can be insightfully analyzed within a constraint-based framework such as OT. Further, it is interesting to note the emergence ofseveral unmarked structures in the domain of reduplication, such as the absenceof onsetless syllables, the absence of a coda in the initial syllable of thereduplicant, and the quality of the mid-vowel pair e/e. That much work needs tobe done to solve all the mysteries of French phonology scarcely bears repeating.

REFERENCES

Borrell, André. 1986. “Le vocabulaire ‘jeune’, le parler ‘branché’ : création et/ourécréation lexicale?”. Cahiers De Lexicologie 47.69-87.

Boudard, Alfonse & Luc Etienne. 1970. La méthode à Mimil. Paris: La JeuneParque.

Burzio, Luigi. 1999. “Surface-to-Surface Morphology: when your Representationsturn into Constraints”. Paper presented at the Maryland Mayfest, held in CollegePark, Maryland, May 1999.

George, Kenneth E. M. 1980. “L=apocope et l=aphérèse en français familier,populaire et argotique”. Le Français Moderne 48.16-37.

12 In some dialects of French (such as Meridional French), this is the absolute rule for all mid-vowels (e/e, ø/œ and o/]) in stressed final position (cf. la loi de position).

Kjellman, Hilding. 1920. Mots abrégés et tendances d=abréviation en français.Uppsala: Akademiska Bokhandeln.


McCarthy, John & Alan Prince. 1993. “Prosodic Morphology I: ConstraintInteraction and Satisfaction”. Ms., University of Massachusetts, Amherst &Rutgers University

---------.1994. “The Emergence of the Unmarked: Optimality in ProsodicMorphology”. Proceedings of the 24th Meeting of the North EasternLinguistics Society ed. by M. Gonzalez, 333-79. Amherst, Mass.: GLSA.

---------.1995. “Faithfulness and Reduplicative Identity”. University ofMassachusetts Occasional Papers in Lingusitics: Papers in Optimality Theoryed. by J. Beckman, L Walsh-Dickey & S Urbanczyk, 249-384. Amherst,Mass.: GLSA.

---------. 1999. “Faithfulness and Identity in Prosodic Morphology”. The Prosody-Morphology Interface ed. by R. Kager, H. van der Hulst & W. Zonneveld,Cambridge: Cambridge University Press.

Morin, Yves-Charles. 1972. “The Phonology of Echo-Words in French”.Language 48.97-105.

Nelson, Nicole. 1999. “Right Anchor, Aweigh”. Ms., Rutgers University.Paradis, Michel. 1979. “Baby talk in French and Québécois”. The Fifth LACUS

Forum ed. by Wolfgang Wölck, 355-66. Columbia: Hornbeam Press.Plénat, Marc. 1982. “Quatres notes sur la morphologie des hypocoristiques à

redoublement”. Cahiers De Grammaire 5.80-143.Prince, Alan & Paul Smolensky. 1993. “Optimality Theory: Constraint Interaction

in Generative Grammar”. Ms., Rutgers University & University of Colorado,Boulder.

Scullen, Mary Ellen. 1997. French Prosodic Morphology: A Unified Account.Bloomington: Indiana University Linguistic Club Publications.

Thiele, Johannes. 1987. Wortbildung der französischen Gegenwartssprache.Translated by André Clas., La formation des mots en français moderne.Montréal: Presses de l’Université de Montréal.

LOCAL CONJUNCTION IN ITALIAN AND FRENCH PHONOLOGY∗∗∗∗

BERNARD TRANEL & FRANCESCA DEL GOBBOUniversity of California, Irvine

0. IntroductionThis paper presents evidence from Dutch, Italian, and French motivating

the existence of the a priori reasonable, yet so far typologically-challenged,conjoined constraint {ONSET&NOCODA}. This constraint makes sense of theacquisition pattern of Dutch syllable structure (Section 2) and is shown to playa key role in explaining the suppletive distribution of the masculine pluraldefinite article in Italian (Section 3.1) and the exceptional behavior of h-aspiréwords with respect to optional schwa deletion in French (Section 4.1).Interestingly, the local domains of relevance for {ONSET&NOCODA} are foundto differ in the Italian/Dutch vs. French cases, thereby challenging thehypothesis that conjoined constraints absolutely self-define their domains.Domain-indexation, independently proposed for self-conjoined constraints,may thus be more generally necessary in Local Conjunction.

{ONSET&NOCODA} does not go unviolated in either Italian or French. Theallomorphic distribution of the masculine singular definite article in Italian(Section 3.2) and ‘liaison sans enchaînement’ in French (Section 4.2) are casesin point and are analyzed in some detail.

Besides Local Conjunction, other general issues touched upon in this paperinclude the cross-theoretical evidence provided by suppletion in favor ofOptimality Theory (OT), the nature of mental representations in relation tosurface allomorphs, and the formalization of variation within OT.

∗ Some of this material was also presented to the University of Southern California PhonologyGroup (April 24, 2000), at the Fifth Southwest Workshop on Optimality Theory held at theUniversity of California, San Diego (May 15, 1999), and in seminars and workshops at theUniversity of California, Irvine. We thank these audiences, as well as the LSRL-30participants, for their interest and feedback, in particular Jake Anderson, Barbara Bullock,Dani Byrd, Jaeshil Kim, Frida Morelli, Lori Repetti, Mario Saltarelli, Rachel Walker, and DiWu.

BERNARD TRANEL & FRANCESCA DEL GOBBO192

1. Background on Local ConjunctionA basic tenet of OT is ‘strictness of strict domination’ (Prince & Smolensky

1993), illustrated in Tableau 1.

/input/ C1 C2 C3☞ candidate (a) * *

candidate (b) *!Tableau 1: Strictness of strict domination

Even though candidate (b) behaves more harmonically than candidate (a)on two out of three constraints, it loses because of its violation of dominant C1.The empirical need to circumvent this tenet has proved pervasive, leading tothe use of Local Conjunction, illustrated in Tableau 2.1

/input/ {C2&C3} C1 C2 C3candidate (a) *! * *

☞ candidate (b) *Tableau 2: Local Conjunction

Three main properties characterize conjoined constraints. First, a conjoinedconstraint is violated if and only if both conjuncts are violated. Second, aconjoined constraint is universally ranked above its simplex constraints. Third,a conjoined constraint operates in a local domain. In Tableau 2, the conjoiningof C2 and C3 and its ranking above C1 allows candidate (b) to emerge as thewinner. The ranking of {C2&C3} just below C1 would yield the reverse result(the same as in Tableau 1). Local Conjunction therefore increases expectationsabout typological diversity.

1 First proposed by Smolensky (1993), Local Conjunction has since been used to explain (andin many cases unify) a variety of phenomena, e.g. assimilation, dissimilation, and OCP effects(Alderete 1997, Fukazawa 1999, Itô & Mester 1998, Ohno 1995, Suzuki 1995, 1998), chainshifts (Kirchner 1996), derived-environment effects (Lubowicz 1998), harmony (Bakovic1999, 2000, McCarthy 1997, Smolensky 1995, 1997), opacity (Itô & Mester 1999), positionalmarkedness (Itô & Mester 1998, Smolensky 1995, Zoll 1998) and subject choice (Aissen1999). Many important questions regarding Local Conjunction remain to be settled, includingthe extent to which constraints can freely conjoin, the universal availability of conjoinedconstraints, and the definition of ‘Local Domain’. See preceding references for somediscussions and proposals (in particular Bakovic 1999, Itô & Mester 1998, 1999, Lubowicz1998).

LOCAL CONJUNCTION IN ITALIAN AND FRENCH PHONOLOGY 193

2. DutchDutch allows syllables with complex onsets and codas. In acquisition,

various stages can be observed toward this potential maximal syllable (Levelt& Vijver 1998). Table 1 shows the first four stages.

Stages I II III IVAllowed syllables CV CV, CVC CV, CVC, V CV, CVC, V, VC

Table 1: Stages I-IV in the acquisition of Dutch syllable structure

Stage III is noteworthy, because the addition of [V] syllables to theestablished inventory of [CV] and [CVC] syllables does not entail the inclusionof [VC] syllables. The Stage-III grammar requires the conjoined constraint{ONSET&NOCODA}, pace Kager (1999:400), and the constraint hierarchy in (1)to account for the exclusive ban on [VC] syllables.

(1) *COMPLEX, {ONSET&NOCODA} » FAITH » ONSET, NOCODA

Levelt & Vijver also note a striking difference between attested grammarsacross languages and Dutch developmental grammars. As illustrated in Table2, Type-III languages are apparently missing.

Syllable types Languages Grammars (Constraint Hierarchies)I CV Hua *COMPLEX, ONSET, NOCODA » FAITHII CV, CVC Thargari *COMPLEX, ONSET » FAITH » NOCODAIII CV, CVC, VIV CV, CVC, V, VC Mokilese *COMPLEX » FAITH » ONSET, NOCODA

Table 2: Cross-linguistic vs. developmental syllable types

If real, this cross-linguistic gap may prove problematic for linguistictheory, because it implies that children can construct grammars, or perhapseven use constraints, which are otherwise not possible. In the following twosections, we seek to fill the typological gap by arguing that a Type-III syllableinventory is attested in Italian and that {ONSET&NOCODA} is functional in bothItalian and French, albeit in somewhat hidden corners of the phonology andmorphology of these two languages.2

2 A recent syllable-structure sketch of Sentani identifies this Papuan language as a Type-IIIlanguage (Elenbaas 1999:50). Such languages may thus be rare rather than unattested.


3. Italian3.1 The distribution of the masculine plural definite article

Standard Italian has two allomorphs for the masculine plural definitearticle, i [i] and gli [¥i] (sometimes pronounced [ji]). As illustrated below, ioccurs before words beginning in ‘licit onsets’ (2a), while gli occurs beforewords beginning in ‘illicit onsets’ or vowels (2b).

(2) a. i ponti “the bridges”i flagelli “the whips”

b. gli specchi “the mirrors”gli amici “the friends”

The distinction between ‘licit’ and ‘illicit’ onsets is sonority-based (Davis1990a,b).3 The general idea is that in Italian a sufficiently steep rising sonorityslope is required in order for consonant clusters to be ideally parsed intocomplex syllable onsets (see Table 3).

Licit onsets(adequate sonority slope)

Illicit onsets(inadequate sonority slope)

single consonants obstruent + obstruent (e.g. [sp, ps])oral stops, [f] + liquid (e.g. [pr, fl]) geminates ([S, ts, dz, ¯ ])

Boundary case: voiceless stop + nasal stop (e.g. [pn])Table 3: Examples of ‘licit’ vs. ‘illicit’ onsets in Italian

The distribution of the masculine singular definite article (il [il] vs. lo [lo])is a standard diagnostic test for the distinction. As illustrated below, wordsbeginning with licit onsets take il (3a), while words beginning with illicitonsets take lo (3b). Idiolectal variations occur in the boundary case (3c).

(3) a. il ponte “the bridge”il flagello “the whip”

b. lo specchio “the mirror”lo scialle ([loSSalle]) “the shawl”

c. il pneumatico ~ lo pneumatico “the tire”

The basic explanation is that lo is used when it allows the consonants in theillicit onsets to be distributed in coda and onset positions, as shown in (4). 3 See also Chierchia (1986), Kaye (1991/92), and Wiltshire & Maranzana (1999). For a criticalview of this approach, see Marotta (1993) and Morelli (1999).


(4) lo specchio [los.pekkjo]lo scialle [loS.Salle]

These phonetic structures offer the advantage of fully parsing consonantsand consonantal length while avoiding dispreferred sonority slopes in onsets.4

With these preliminaries in place, we are in a position to address twoessential questions raised by the allomorphic distribution of the masculineplural definite article: (i) why the difference across consonant-initial words((5a) vs. (5b))? and (ii) why the strange bed-fellows (illicit clusters (5b) andvowels (5c) both take gli)?

(5) a. i ponti *gli pontib. gli specchi *i specchic. gli amici *i amici

We show here that an OT approach that includes Local Conjunction is ableto make sense of these data. The basic idea to be implemented is that both i andgli are freely available lexical allomorphs (thereby doubling the number offaithful candidates) and that the standard mechanism of the theory (constraintranking) accounts for their distribution.5

The first puzzle concerns why i ponti is preferred over *gli ponti (5a).Given ONSET, the reverse should hold. We propose that the effect of ONSET isthwarted by *STRUCTURE (see Tableau 3). This constraint dominates ONSETand is thus responsible for the preference for the shorter allomorph (see Hargus1997 for a similar appeal to BREVITY in allomorph selection). *STRUCTURE isactually a large family of constraints, often used in relation to syllables (e.g.Takeda 1998, Tranel 1999, Zoll 1996). We assume that it is relativized toconsonants here (*STRUC-C).6

4 When words with illicit onsets are pronounced in isolation, clusters survive intact (specchio[spekkjo]) and geminates are reduced to simple consonants (scialle [Salle]). A formal accountis provided in Section 3.2 below.5 By contrast, a rule-based framework seems unable to provide an insightful account. If lexicalsuppletion is assumed, the allomorphic distribution must be recorded in special statements thatare purely descriptive and outside the basic formalism of the theory. If a single underlyingrepresentation is assumed (/¥i/), a similarly unexplanatory rule is required deleting /¥/ before/i/ when the vowel is in syllable-final position and followed by a consonant (/¥/ —› Ø / i . C).6 A possible alternative is to exploit the markedness of the sequence [¥i] and to view thepreference for i over gli as an OCP effect analogous to the one which in a number of languagesbans sequences like [ji] and [wu].


/i ~ gli + ponti/ *STRUC-C ONSET☞ a. i ponti *

b. gli ponti *! ([¥])Tableau 3: i ponti > *gli ponti

This account of (5a) creates a problem for the two cases where gli is usedinstead of i (5b-c). A ranking paradox arises, since ONSET appears to prevailover *STRUC-C in both instances. We resolve the paradox by proposing that ineach case, another constraint is dominant and thwarts the effect of *STRUC-C.

We thus come to the second puzzle: why is gli specchi better than *ispecchi (5b)? The situation here involves illicit onsets, for which the opensyllable of the article provides a free coda position. We therefore need tocompare the two structures in (6).

(6) a. glis.pecchib. is.pecchi

An explanation is now readily available for the preference for (6a). Theinitial syllable in (6a) violates NOCODA, but satisfies ONSET. By contrast, theinitial syllable in (6b) violates both ONSET and NOCODA. As shown in Tableau4, Local Conjunction enables us to capture the difference: the conjoinedconstraint {ONSET&NOCODA} provides the correct decision in favor of (6a), aslong as it dominates *STRUC-C.

/i ~ gli + specchi/ {ONSET&NOCODA} *STRUC-C ONSET☞ a. glis.pecchi *

b. is.pecchi *! *Tableau 4: gli specchi > *i specchi

The Constraint Hierarchy established in Tableau 4 does not resolve thethird puzzle: why is gli preferred to i before vowel-initial words (5c)? This isshown in Tableau 5, where i is incorrectly selected by the grammar.

/i ~ gli + amici/ {ONSET&NOCODA} *STRUC-C ONSET! a. gli amici *! *☛ b. i amici **

Tableau 5: Problem for gli amici > *i amici


A reasonable explanation for ruling out candidate (b) is that it violatesONSET in contiguous syllables. As shown in Tableau 6, Local Conjunction, inthis case Self-Conjunction (Alderete 1997, Itô & Mester 1998, Suzuki 1995),again provides a resolution.7

/i ~ gli + amici/ {ONSET}2 *STRUC-C ONSET☞ a. gli amici * *

b. i amici *! **Tableau 6: gli amici > *i amici

In sum, our treatment of the allomorphic distribution of the masculineplural definite article in Italian crucially relies on two conjoined constraints,{ONSET&NOCODA} and {ONSET}2, dominantly ranked as shown in (7).8

(7) {ONSET&NOCODA}, {ONSET}2 » *STRUC-C » ONSET

3.2 The case of the masculine singular definite articleWhile {ONSET&NOCODA} is unviolated in Dutch at Stage III, this is not the

case in Italian. In particular, as shown below, the constraint is violated by themasculine singular definite article (8a), even though the potential exists foravoiding such a violation by using a different allomorph (8b). The goal of thissection is to explain this striking puzzle.

(8) a. il ponteb. *lo ponte (cf. lo specchio)

As illustrated in (9), the masculine singular definite article has threesurface allomorphs.

7 One might alternatively view this restriction on contiguous onsetless syllables as a type ofOCP effect and capture it by using Suzuki’s (1998) Generalized OCP formalism: *[Ø~Ø]ONSET.8 The constraint hierarchy in (7) claims that the uses of gli before illicit onsets and beforevowels are independent. The potential re-rankings of the conjoined constraints with respect to*STRUC-C predict the possibility of dialects of the types {i ponti, i specchi, gli amici} (with{ONSET}2 » STRUC-C » {ONSET&NOCODA}) and {i ponti, gli specchi, i amici} (with{ONSET&NOCODA} » *STRUC-C » {ONSET}2). Whether such dialects are attestedremains to be determined. With both conjoined constraints ranked below *STRUC-C, wepredict the use of i throughout (i ponti, i specchi, i amici), i.e., the elimination of gli as a viableallomorph. The Roman dialect exhibits this pattern (Mario Saltarelli, personalcommunication). And both Giovanna Marotta and Laura Vanelli have independently reportedto us a geographically broader range of occurrence for this pattern.


(9) a. l’amico *il amico, *lo amico “the friend”b. lo specchio *il specchio, *l specchio “the mirror”c. il ponte *lo ponte, *l ponte “the bridge”

The allomorph [l] is used before vowel-initial words (9a), [lo] is usedbefore words beginning in illicit onsets (9b), and [il] is used before wordsbeginning in licit onsets (9c).

The puzzle caused by the fact that il ponte is preferred over *lo ponte isactually independent of the existence of {ONSET&NOCODA}, since given thesimplex constraints ONSET and NOCODA alone, a [CV] syllable will always winover a [VC] syllable. Tableau 7 formalizes the problem.

/il ~ lo + ponte/ {ONSET&NOCODA} ONSET NOCODA! a. il ponte *! * *☛ b. lo ponte

Tableau 7: Problem with il ponte > *lo ponte

Our answer to this puzzle is to attribute a special status to the vowel of thelo allomorph, such that its surface occurrence entails a DEP violation. Theimplementation of this proposal can take two different routes, which we onlydescribe briefly here. The first (partial epenthesis) is to assume that the vowelin /lo/ is not a full-fledged lexical segment, but rather a floating segment(floating segments are underlyingly defective at some structural level, lackingeither an inherent position or a root node which is inserted when the segment isphonetically realized).9 The second possibility is to assume that the [o] in [lo]is a full epenthetic vowel whose quality is morphologically determined ([o] isthe vowel normally used to mark masculine singular words in Italian).10 Undereither implementation, the crucial point is that the use of [lo] leads to someDEP violation, henceforth noted as DEP(o) (we also note as /l(o)/ the lexicalrepresentation of the allomorph lo).

As in the case of the masculine plural definite article, we assume herelexical allomorphy (the two forms /il/ and /l(o)/ are equally available in

9 See for instance Tranel (1996b) and Zoll (1996) for different, but here equivalent, proposalsregarding the representation of floating segments.10 The full epenthesis approach raises issues with respect to Consistency of Exponence(McCarthy & Prince 1993), since it involves an inserted segment which is morphologicallyaffiliated on the surface (it is clearly part of the article).


candidates requiring the masculine singular definite article), and we let theconstraint hierarchy govern the distribution.11

The selection of [il] rather than [lo] before words beginning in licit onsetsis attributable to the ranking of DEP(o) above {ONSET&NOCODA} (Tableau 8).

/il ~ l(o) + ponte/ DEP(o) {ONSET&NOCODA}☞ a. il ponte *

b. lo ponte *!Tableau 8: il ponte > *lo ponte

This case shows that the surface allomorph [lo] cannot correspond to a full-fledged lexical /lo/, because if it did, there would be no way for il ponte to winover *lo ponte (cf. Tableau 7 above). Not all surface representations cantherefore be lexical representations. This observation argues against‘Generalized Suppletion’, the proposal seeking to eliminate traditionalunderlying representations in favor of lexical representations that are merelysurface representations (e.g. Burzio 1996, 1997, Flemming 1995, 1997).12

Words beginning in illicit onsets select [lo] rather than [il]. The reason isthat the vowel [o] of [lo] provides a coda position for the initial consonant ofthe offending clusters. We are therefore dealing here with the part of theconstraint hierarchy where DEP(o) is relatively low-ranked. Tableau (9)illustrates the case of consonant clusters like [sp] (lo specchio), while Tableau(10) illustrates the case of initial geminates (lo scialle).13

/il ~ l(o) + specchio/ *COMPLEX(CODA) MAX-C SON-SLOPE DEP(o)a. il.specchio *!b. ils.pecchio *!c. il.pecchio, is.pecchio *!

☞ d. los.pecchio *Tableau 9: lo specchio > *il specchio

11 As suggested to us by Lori Repetti, the [i] in [il] may be viewed as an epenthetic vowel aswell, but of a purely phonological nature. An analysis worth considering for the masculinesingular definite article could thus consist of a lexical representation /l/ and a constrainthierarchy selecting epenthetic [o] and [i] in the appropriate contexts.12 See Tranel (1998a,b, 1999, 2000) for similar arguments drawn from French against‘Generalized Suppletion’.13 To save space, candidates with just [l] as the article (*l specchio, *l scialle) are not enteredin these tableaux. They are out for obvious syllable-structure reasons.


In Tableau 9, candidate (a) is out because it violates the sonority-sloperequirement for onsets. Candidate (b) is out because of a strong restriction oncomplex codas. The candidates in (c) are out because of the unlawful deletionof a consonant. Candidate (d) wins because it satisfies these requirements, atthe expense of lower-ranked DEP(o). MAX-C is ranked above SON-SLOPE toaccount for the fact that in isolation, specchio is pronounced [spekkjo], with itsinitial consonant cluster intact, but in violation of SON-SLOPE.

/il~l(o)+Salle/ *COMPLEX(CODA) MAX-C SON-SLOPE GEM-S DEP(o)a. il.SSalle *!b. ilS.Salle *!c. iS.Salle *!d. il.Salle *!

☞ e. loS.Salle *Tableau 10: lo scialle > *il scialle

The word scialle begins in a consonant ([S]) which, along with [ts, dz, ¯],has traditionally been posited as underlyingly geminate in Italian (e.g.Chierchia 1986). Given Richness of the Base, this is not a possiblerepresentation in OT. We therefore assume a constraint (mnemonically notedGEM-S here) which demands the gemination of this class of consonants.14 InTableau 10, the candidates in (a), (b), and (c) fatally violate SON-SLOPE,*COMPLEX(CODA), and MAX-C, respectively. Candidate (d) is also out, becauseof its failure to geminate [S]. Candidate (e) wins because it satisfies theserequirements, at the expense of lower-ranked DEP(o). SON-SLOPE is rankedabove GEM-S in order to account for the fact that in isolation, scialle ispronounced with a simple rather than geminate onset consonant ([Salle]).

Finally, before vowel-initial words, [l] occurs, rather than [il] or [lo],because, as shown in Tableau 11, it yields the optimal syllable structure.15

14 These consonants are the cross-linguistically marked counterparts of [s, tS, dZ, n]. One mighttherefore view the gemination requirement as an instantiation of the general principle ofdispersion, which tends to create maximal perceptual distance between contrasting sounds(Flemming 1995).15 The separate existence and relative rankings of *VV (no hiatus) and ONSET are justifiedelsewhere (Del Gobbo 2001, Del Gobbo & Tranel 2000). On the independence of *VV andONSET (pace McCarthy & Prince 1993), see also Tranel (1998a).


/il1 ~ l2(o) + amico/ MAX-V *VV {ONSET&NOCODA} ONSETa. il1.a.mi.co *! **b. i.l1a.mi.co *!c. l2o.a.mi.co *! *d. l1a.mi.co *!

☞ e. l2a.mi.coTableau 11: l’amico

It is worth observing that l’amico cannot correspond to /lo+amiko/ (with/lo/ containing a full-fledged lexical /o/ getting deleted in hiatus). The reason isthat MAX-V is dominant in Italian, as shown in (10) by the preservation of thefull-fledged lexical vowels in the plural articles gli (masc.) and le (fem.) beforevowel-initial words.16

(10) gli amici, le amiche “the friends” (masc., fem.)

We conclude this section by underscoring the central role played by{ONSET&NOCODA} in the analysis.

(11) … DEP(o) » {ONSET&NOCODA} » *STRUC-C …

The conjoined constraint is active in determining the allomorphicdistribution of the masculine plural definite article: it overrides an otherwiseoperative constraint economizing structure (hence gli specchi > *i specchi,despite i ponti > *gli ponti). But when it comes to the masculine singulardefinite article, its effect is thwarted by a higher-ranked faithfulness constraint(hence il ponte > *lo ponte, despite lo specchio).17

3.3 Conclusion on ItalianIn a small corner of its morphology (the masculine plural definite article),

Italian turns out to be a Type-III language (see Table 4). 16 The optional deletion of [i] in gli before [i]-initial words (e.g. Italiani “Italians”) is aseparately explainable phenomenon.17 Our analysis accounts for the dialect with the coexisting patterns {il ponte, lo specchio,l’amico} and {i ponti, gli specchi, gli amici}. But given constraint re-ranking, these are not theonly predicted patterns, since the constraint rankings for the distribution of the two sets ofallomorphs are relatively independent of each other. For example, nothing at this pointprevents the possibility of {il ponte, lo specchio, l’amico} coexisting with {i ponti, i specchi,gli amici} in some speakers. Further investigations of which patterns are actually attested arewarranted. See also note 8 above.


Syllable typesallowed disallowed

Examples

CV gli (amici)CVC glis (pecchi)

V i (ponti)VC *is (pecchi)

Table 4: Masculine plural definite article

Italian is therefore comparable to Dutch at Stage III (at least in kind if notin scope). Our analysis thus provides some typological grounding for assumingthat Dutch children do not behave in aberrant fashion at Stage III and that theconjoined constraint {ONSET&NOCODA} is legitimate.18

However, a major difference between Dutch at Stage III and Italian is that{ONSET&NOCODA} rules in a narrow sub-domain of Italian phonology andmorphology, whereas it governs the whole phonology of Dutch at Stage III.Still in need of an explanation therefore is the apparent non-occurrence (orextreme rarity) of languages exhibiting a Type-III syllable inventorythroughout their phonologies. At least, the issue has been narrowed down. Thequestion may no longer be whether {ONSET&NOCODA} is a legitimateconstraint, but rather why it seems irrevocably (or preferentially) demotedbelow FAITH when ONSET and NOCODA are themselves ranked below FAITH.

Our analysis of the Italian data crucially relies on two conjoined constraints({ONSET&NOCODA} and {ONSET}2) and as such provides support for LocalConjunction, including Self-Conjunction. It also speaks to the issue of ‘LocalDomain’ for conjoined constraints. Regarding {ONSET&NOCODA}, the syllableitself is the relevant domain in Italian, an uncontroversial possible localdomain given the two simplex constraints involved. By contrast, the existence

18 Japanese loanword phonology lends credence to the existence of {ONSET&NOCODA} aswell. Word-final /d/ is among the consonants geminating in borrowings (Katayama 1998:76):Fred —› [∏u.red.do], *[∏u.re.do]. But Ed yields [e.do], *[ed.do], in compliance with{ONSET&NOCODA}. We note that other factors modulate the scope of the generalization.Word-final /t/ geminates even if {ONSET&NOCODA} is violated: hit —› [hit.to], *[hi.to]; at—› [at.to], *[a.to]. The ban on *[ed.do] is thus due not only to {ONSET&NOCODA}, but alsoto the relative markedness of voiced geminates. Finally, egg is borrowed as [eg.gu], *[e.gu], anindication that spelling may play a role too (for other cases where orthography influencesJapanese phonology, see Itô, Kitagawa & Mester 1996 and Katayama 1998:31-33, 79). Thanksto Eric Bakovic for pointing out to us the Fred/Ed contrast (an observation he attributes toYoko Futagi and Takeo Kurafuji), and to Shigeyuki Fujimoto, Maki Irie, Hironobu Kasai,Takaomi Kato, Andrew Martin, and Hajime Ono for help with the data.


of {ONSET}2 as a self-conjoined constraint can be questioned on the basis of astrict definition of ‘Local Domain’. Thus, following Lubowicz (1998), Bakovic(1999) defines a local domain as “the smallest domain evaluable by theconjuncts” and concludes that Self-Conjunction is vacuous, because thesmallest domain evaluable by a self-conjoined constraint reduces to a singleposition. We advocate instead an interpretation requiring each token in a self-conjoined constraint to identify its own separate position of relevance. In thecase at hand, the two instances of ONSET in {ONSET}2 require considering twodifferent syllables, and locality forces these two syllables to be adjacent.

Finally, we underscore two broad cross-theoretical implications connectedto the relation between mental and surface representations. The first concernssuppletive allomorphy, which is involved in both i~gli and il~lo alternations inItalian (the lexical entry of a given article -- plural or singular -- must beassumed to include separate allomorphs, instead of containing a uniquephonological representation). This, however, does not mean that the surfacedistributions of these lexical allomorphs are haphazard; on the contrary, as wehave seen, they follow well-motivated regular patterns. Yet, a rule-basedframework can only accommodate these facts through arbitrary distributionalstatements and/or equally unenlightening parochial rules. By contrast, the basictool of OT (constraint ranking) is able to capture the distributions in aprincipled way. Suppletive allomorphy thus constitutes an important focalpoint of comparison between the two theories.19 Second, we wish to emphasizethe problem which the pattern il ponte > *lo ponte poses for any approachbroadly claiming that surface representations are systematically stored aslexical representations (‘Generalized Suppletion’). The Italian il~lo caseclearly demonstrates that although [lo] occurs as a surface allomorph (lospecchio), it must not correspond to an identical lexical representation /lo/,because if it did, the correct form il ponte could never surface.20

4. French4.1 H-aspiré words and optional schwa deletion

The argument for the relevance of {ONSET&NOCODA} in French involves aregular process of vowel deletion (optional schwa deletion), which is blockedin front of a special class of words (h-aspiré words). As we shall see, an 19 For other applications of this argument, see for instance Dolbey (1997), Hargus (1997),Kager (1996), Mascaró (1996a,b), McCarthy & Prince (1993), Mester (1994), Tranel(1996a,b).20 Morelli (1999) circumvents this problem by positing that a constraint *lo dominates anotherconstraint *il. However, this formalism resorts to highly language-specific constraints andstipulates that il is the default allomorph instead of explaining why this is the case.


interesting twist in this instance is that the required domain for the conjoinedconstraint is different from the Italian and Dutch cases.

We begin with separate nutshell characterizations of optional schwadeletion and h-aspiré words. Optional schwa deletion refers to the variabledeletion of the weak vowel ‘schwa’ illustrated in (12) vs. (13) (whenpronounced, this vowel is generally realized as [œ]).

(12) fais le plein a. [fE.lœ.plE‚] b. [fEl.plE‚] “fill it up” (2nd sg)(13) faites le plein a. [fEt.lœ.plE‚] b. *[fEtl.plE‚] “fill it up” (2nd pl)

The deletion depends on various factors such as rate of speech, style, andsyllable structure. Typically, it may occur only if a single consonant precedesthe schwa (Dell 1985, Morin 1974). Thus, the “e” in le can optionally delete in(12), but not in (13). The process is attributable to SYLLABLE ECONOMY (SE), avariably ranked *STRUCTURE constraint relativized to syllables (Tranel 1999).When SE is low-ranked (e.g. in careful elocution), no deletion takes place, as in(12a). SE is inactive: NOCODA prevails (NOCODA » SE). In less formal styles, SEis promoted in the constraint hierarchy, making deletion possible, as in (12b).SE is active: it now prevails over NOCODA (SE » NOCODA). SE’s upward reachwithin the constraint hierarchy is restricted by markedness constraints. Forexample, as shown in (13), if the coda position before le is already occupied,the deletion may not occur. *COMPLEX(CODA) is an upper bound to thepromotion of SE (*COMPLEX(CODA) » SE).

H-aspiré words are exceptional vowel-initial words (Tranel 1981). Asillustrated in (14), elision and liaison, which can be viewed as anti-hiatusrepairs (Tranel 2000), normally occur before vowel-initial words. They resultin ONSET satisfaction, at the expense of left-edge misalignments between themorphology and the phonology (i.e., ALIGN-L, the constraint demanding thatthe left edge of words coincide with the left edge of a syllable, is violated. Insubsequent examples, “{” denotes relevant morphological left edges).

(14) a. Elision l’étau [.l{eto] *[lœ{.eto] “the vise”b. Liaison les étaux [lE.z{eto] *[lE{.eto] “the vises”(cf. le bateau [lœbato] “the boat”, les bateaux [lEbato] “the boats”)

By contrast, as shown in (15) with héros [ero], a small set of vowel-initialwords (traditionally called “h-aspiré words” because they often include aninitial orthographic h) are exceptional in that they bar elision and liaison; theirinitial vowel idiosyncratically prevents a consonant from intruding in onsetposition in violation of ALIGN-L.


(15) a. No elision le héros *[.l{ero] [lœ{.ero] “the hero”b. No liaison les héros *[lE.z{ero] [lE{.ero] “the heroes”

The data can be accounted for by assuming that ONSET dominates generalALIGN-L, but is dominated by word-specific ALIGN-L (Russell 1999; see alsoTranel 1996a): ALIGN-L (héros, σ) » ONSET » ALIGN-L (word, σ).

Two independent observations emerge from these characterizations ofoptional schwa deletion and h-aspiré words (they are important because theircombination leads to the French puzzle ultimately solvable by{ONSET&NOCODA}). First, schwa deletion is allowed to create NOCODAviolations. Second, in the case of h-aspiré words, the force of ONSET isneutralized by the dominant word-specific ALIGN-L.

We now turn to the interaction of optional schwa deletion with h-aspiréwords, which is illustrated in (16).

(16) fais le héros a. [fE.lœ{.ero] b. *[fEl{.ero] “be the hero”

The puzzle is the ungrammaticality of (16b), made vivid in Tableau 12.

/fE l´ ero/ ALIGN-L(héros, σ) ONSET SE NOCODA! a. fE.lø{.e.ro * ****!☛ b. fEl{.e.ro * *** *

Tableau 12: Problem with *[fEl{.e.ro]

Tableau 12 reflects the problematic situation which arises when SE is active(SE » NOCODA). The two candidates tie on ALIGN-L(héros, σ) and ONSET,which incorrectly leads to the selection of candidate (b), because of its betterperformance on SE.21 Observing that candidate (b) violates both ONSET andNOCODA, we propose that {ONSET&NOCODA} is responsible for its ill-formedness, by constituting another upper bound to SE’s potential climb in theconstraint hierarchy. This analysis is illustrated in Tableau 13.22

21 Candidate (a) is unproblematically generated when SE is inactive (NOCODA » SE). Theissue here is that it must also be generated when SE is active (SE » NOCODA).22 Given constraint re-ranking, the (plausible) potential exists for idiolects where SE woulddominate {ONSET&NOCODA} and thus allow candidate (b) to surface as a possible form.


/fE l´ ero/ ALIGN-L(héros, σ)

{ONSET&NOCODA}

ONSET SE NOCODA

☞ a. fE.lø{.e.ro * ****b. fEl{.e.ro *! * *** *

Tableau 13: [fEløero] > *[fElero]

The twist here is that the conjunction of ONSET and NOCODA requiresdifferent domains in the French vs. Italian and Dutch cases. Whereas for Italianand Dutch the relevant domain was established as the syllable (17a), theconstraint must operate across syllables in French (17b).

(17) Relevant domains for {ONSET&NOCODA}a. Italian/Dutch the syllable [VC]σb. French across syllables [… C]σ [V…]σ

While an obvious local domain for {ONSET&NOCODA} is no doubt thesyllable (as noted in Section 3.3 above), the domain required in the French caseis a plausible local domain as well, since it deals with the interface betweenadjacent structural positions in adjacent syllables.

A strict interpretation of ‘Local Domain’ as “the smallest domain evaluableby the conjuncts” (Bakovic 1999, Lubowicz 1998) is however problematic for(17), since a corollary of this definition is that “two local conjunctions cannotdiffer solely by their domain of application” (Bakovic 1999). On the otherhand, although many conjoined constraints appropriately self-define theirdomain as the smallest domain evaluable by their conjuncts,23 domain-indexation has also been proposed in the literature on Local Conjunction, atleast for self-conjoined constraints. Thus, according to Itô & Mester’s (1998)analysis of Japanese Rendaku, {VOP}2, the self-conjoined constraint banningrepeated voiced obstruents, must be domain-indexed for ‘stem’ and for ‘word’in order to explain the difference between Old and Modern Japanese. The caseof {ONSET&NOCODA} might therefore be viewed as an instance of a non-self-conjoined conjoined constraint where domain-indexation is warranted(adjacent syllables vs. the syllable).24

23 A well-known example is {NOCODA&VOP}, the conjoined constraint banning voicedobstruents in coda position (‘VOP’ stands for “Voiced Obstruent Prohibition”).24 A conceivable alternative is to assume that conjoined constraints may differ in the order oftheir conjuncts, leading in cases such as the combination of ONSET and NOCODA to theconstruction of different local domains. Thus, {ONSET&NOCODA} (listed in this order)would demand that the onset position domain-precede the coda position, hence the syllable asits self-defined smallest domain, whereas {NOCODA&ONSET} (listed in this order) would


It is important to underscore the continued importance of locality in theoperation of {ONSET&NOCODA} across syllables. As shown in (18), when twoadjacent syllables constitute the presumed relevant domain, there are inprinciple two configurations which could be banned: (18a), which contains aviolation of ONSET in the first syllable and a violation of NOCODA in thesecond syllable, and (18b), which contains a violation of NOCODA in the firstsyllable and a violation of ONSET in the second syllable (the absence of anonset in (18) is denoted by Ø, and the presence of a coda by C).

(18) a. *[ØV]σ [CVC]σb. *[CVC]σ [ØV]σ

Long-distance restrictions like (18a) are not expected cross-linguistically,while (18b) is the needed banned configuration for French. The difference isthat (18b) is local, since it concerns two adjacent positions, whereas (18a) isnon-local, referring to non-adjacent positions. Locality correctly determineswhich restriction can be operative.25

In sum, the conjunction of ONSET and NOCODA appears to be active inFrench. The conjoined constraint explains why two independently allowedphenomena (the creation of codas through optional schwa deletion and therequired absence of initial onsets with h-aspiré words) cannot occur incombination. The relevant local domain for {ONSET&NOCODA} is different inFrench than in Italian or Dutch at Stage III (adjacent syllables vs. the syllable),which suggests a need for domain-indexation in Local Conjunction extendingbeyond self-conjoined constraints.

4.2 Liaison sans enchaînementAs in Italian, {ONSET&NOCODA} does not go unviolated in French. We

focus here on a well-known and rather intriguing case usually described as‘liaison sans enchaînement’ (LSE) and illustrated in (19c).

(19) j’avais un rêve a. [Za.vE.E‚‚.rEv] no liaison“I had a dream” b. [Za.vE.zE‚‚.rEv] ‘standard’ liaison

c. [Za.vEz.E‚‚.rEv] liaison sans enchaînement

demand that the coda position domain-precede the onset position, hence adjacent syllables asits self-defined smallest domain. See also the discussion about (18) in the text.25 Given any domain for {ONSET&NOCODA} greater than the syllable, locality will in factsystematically limit the operation of the constraint to adjacent positions in adjacent syllables.


LSE may variably occur in the speech of some speakers in contexts whereliaison is optional (Encrevé 1988). The phrase in (19) offers such a contextbetween the verb avais and the article un. (19a) shows that liaison is notobligatory here. (19b) illustrates ‘standard’ liaison: the linking consonant [z] ispronounced (liaison) and it occurs in onset position, the normal position forlinking consonants (enchaînement). (19c) exhibits LSE: the linking consonantis in coda rather than onset position, in violation of {ONSET&NOCODA}. Thegoal of this section is to explain why such a violation is possible.

We begin with the familiar descriptive distinction between Latent vs. FixedFinal Consonants (LFC vs. FFC). The label LFC refers to linking consonantssuch as [z] in (19), which may occur or not phonetically, depending on context.When pronounced, LFC do not normally lead to violations of{ONSET&NOCODA}. Thus, as already mentioned, (19b) reflects ‘standard’liaison, and, as shown in (20), LFC must occupy an onset position in obligatoryliaison contexts (e.g. between an adjective and a noun).

(20) petit anneau a. [pœ.ti.ta.no] b. *[pœ.tit.a.no] “small ring”(cf. petit [pœti], *[pœtit]; petit pot [pœtipo], *[pœtitpo] “small pot”)

The label FFC refers to consonants that are always present phonetically,regardless of context. The final /t/’s in net “clear” and petite “small” (fem) areexamples of FFC: they are invariably pronounced ([nEt], [pœtit]).

Another characteristic distinguishing FFC from LFC is illustrated in (21).

(21) net avantage a. [nE.ta.vA‚.taZ] b. [nEt.a.vA‚.taZ] “clear advantage”petite astuce a. [pœ.ti.tas.tys] b. [pœ.tit.as.tys] “small trick”

(21) shows that FFC can occur in onset or coda position (Morin 1998,Tranel 1996b). Our interpretation is that alignment considerations mayoptionally suspend onset formation with FFC, thus leading to violations of{ONSET&NOCODA}, as in (21b). The two options can be accounted for byvariably ranking {ONSET&NOCODA} and ALIGN-FFC-R (a constraint forcingalignment of morphological and phonological right edges).26

26 H-aspiré words behave as expected here, given (a) dominant ALIGN-L(héros, σ), (b) theobservation that LFC must normally go into onset position if pronounced, and (c) the fact thatFFC must be pronounced, but don’t have to occupy an onset position.

(i) LFC petit héros *[pœti.te.ro] [pœ.ti.e.ro] *[pœ.tit.e.ro] “small hero”(ii) FFC sept héros *[sE.te.ro] *[sE.e.ro] [sEt.e.ro] “seven heroes”


(22) a. {ONSET&NOCODA} » ALIGN-FFC-R —› [nE.ta.vA‚. taZ], [pœ.ti.tas.tys]b. ALIGN-FFC-R » {ONSET&NOCODA} —› [nEt. a.vA‚.taZ], [pœ.tit.as.tys]

Our next step is to explain why LFC, as opposed to FFC, are normallyforced into onset position. As (22) indicates, we take the syllabification of finalconsonants in onset or coda position to be a struggle between phonological andmorphological forces.

(23) Input /… VC # V …/Candidates a. […V] [CV…]

b. [… VC] [V …]

Given the input in (23), the phonological force P ({ONSET&NOCODA}) issatisfied by candidate (a), but violated by candidate (b). By contrast, themorphological force M (ALIGN-R) is satisfied by candidate (b), but violated bycandidate (a). As anticipated in (22), we propose that ALIGN-R be relativized toFFC and LFC (24b).

(24) a. Phonological force P: {ONSET&NOCODA}b. Morphological force M1: ALIGN-FFC-R

M2: ALIGN-LFC-R

The split of M into M1 and M2 allows us to capture the different relativestrengths of the morphological affiliations of final consonants. The intuitiveidea is that M1 is stronger than M2 (i.e., M1 inherently dominates M2). Thereason for the difference is that FFC are integral members of words (‘intrinsic’consonants), whereas LFC are only partially so (although LFC are lexicallydependent on the preceding word for some of their properties, they otherwisehave all the characteristics of epenthetic segments).27 Given the constraints in(24), the rankings in (25) allow FFC to occur in onset or coda position, butdemand that LFC be in onset position.

(25) M1 ~ P » M2 (“~” indicates variable ranking between M1 and P)

With FFC, only M1 and P are relevant. The variable ranking between thetwo constraints yields the two output types noted in (21)-(22) above. With LFC,

27 Formally, LFC have thus been variously analyzed as ‘floating’ or ‘borrowed’ or ‘inserted’consonants (see Tranel 2000 for a survey and references).


M2 is relevant, but its dominated status relative to P makes it inactive andforces LFC into onset position.

We turn finally to an explanation for ‘liaison sans enchaînement’ (LSE).Two important reminders are warranted at this point. First, LSE is possible onlyfor some speakers.28 Two different, albeit related, grammars must therefore beconstructed, one for ‘standard speakers’ and one for ‘LSE speakers’. Thegrammar for ‘standard speakers’ has already been established above as (25).Second, LSE is usually observable in cases of optional, as opposed toobligatory, liaison. This fact, as we shall see, is not coincidental and deservesan explanation.

For the distinction between obligatory and optional liaison, we assume thetraditional intuitive notion that the connection between two words entering intoliaison is less tight in optional cases than in obligatory cases.29 In general, therelative strengths of phonological constraints like P and morphologicalconstraints like M can be affected by the degree of tightness between twoconstituents.30 In particular, with two words in a comparatively looseconnection (such as optional liaison contexts), the phonological requirement Pcan be expected to carry less weight than the morphological requirement M2.This is in our view the basis for LSE. Optional liaison occurs between wordsthat are loosely connected, and M2 may as a result variably take over anddominate P, hence the possible violation of {ONSET&NOCODA}.

The LSE grammar just suggested presumes that P and M2 can be variablyranked, thus yielding the two options in (19b-c) above for LSE speakers. Thishypothesis, however, requires some fine-tuning. If M2 comes to dominate P in‘LSE mode’, the prediction is that LSE will occur in obligatory liaison contextsas well. But this result is incorrect. As noted earlier, LSE is not typically foundin obligatory liaison contexts. The remedy to this problem is to integrate theidea that P is domain-dependent, specifically that its relative strength is afunction of the degree of cohesion between words (see note 30). As shown in(26), we indicate that {ONSET&NOCODA} is stronger in obligatory than inoptional liaison contexts by splitting P into two intrinsically ranked constraints.

28 LSE can be recognized and is often derided by non-users. In Labov’s 1972 sociolinguisticclassification, it is thus more a ‘stereotype’ than a ‘marker’ or an ‘indicator’.29 There is a fairly abundant literature seeking to formally define the appropriate domains forobligatory vs. optional liaison (e.g. De Jong 1990, Kaisse 1985, Selkirk 1974).30 Across languages, [CV] resyllabification effects are stronger within morphemes than acrossmorphemes than across words than across phrases, etc., as considerations of morphologicaland syntactic alignment with phonological structures correspondingly weigh in more and moreagainst purely phonological preferences.


(26) a. P1: {ONSET&NOCODA} in obligatory liaison contextsb. P2: {ONSET&NOCODA} in optional liaison contextsc. P1 » P2

The final versions of our two grammars are given in (27), where we recruitthe concept of ‘constraint range’ proposed by Boersma & Hayes (2001) toaccount for acquisition and variation. The lines at the bottom of each grammarrepresent the continuous scale along which constraints are ranked from high-ranking (toward the left) to low-ranking (toward the right). Each boxrepresents the range of a given constraint. The center of a constraint’s range isthe most likely choice for its ranking value. The odds of picking a given valuewithin the range diminish as we move away from the center point. Allconstraints are assumed to have the same range, hence the same size box foreach constraint. If boxes overlap, the constraints they represent are rankedvariably. If boxes do not overlap, the relevant constraints are strictly ranked.

(27) a. Standard grammar

b. LSE grammar

The ‘standard’ grammar (27a) is similar to our earlier (25), where P can beviewed as encapsulating P1 » P2 (26c). The range overlaps between P1 and M1and between P2 and M1 capture the fact that FFC can generally occur either inonset or coda position. The lack of overlap between P1/P2 and M2, togetherwith M2’s lower ranking, captures the fact that LFC must occur in onsetposition everywhere. Because P1 partially dominates M1, which itself partiallydominates P2, this model (plausibly) predicts that the syllabification of FFC inonset position should statistically occur more often in obligatory than inoptional liaison contexts.

P1 P2

M1 M2

P1 P2

M1 M2


The LSE grammar (27b) differs from the standard grammar (27a) only inpresenting some overlap between P2 and M2. (27b) thus contains (27a) in thesense that it allows all the outputs generated by (27a). But the range overlapbetween P2 and M2 produces additional possible outputs, namely the desiredLSE cases. With P2 partially dominating M2, this model actually predicts therelative rarity of LSE (in Encrevé’s 1988 corpus, LSE is on average observed inonly about 11% of the realized optional liaisons).

Because each of the intrinsically ranked P and M hierarchies (P = P1 » P2; M= M1 » M2) is presumably independent of the other, sliding the two sets ofconstraints along the continuous scale of ranking values can yield additionalpotential grammars besides (27a) and (27b). We do not investigate the variouspossibilities and restrictions here, but focus briefly on one intriguing caseraised by the following extremely rare, but apparently attested, instance of LSEin an obligatory liaison context (Morin & Kaye 1982:300).

(28) un grand ethnologue [E‚.grA‚t.Et.nç.lçg] “a great ethnologist”

The grammar producing such a form must allow a range overlap betweenP1 and M2. Such an overlap seems impossible if the following four assumptionsare all true.

(29) a. P1 » P2b. M1 » M2c. M1 must overlap with both P1 and P2 (see (27) above)d. All constraints have the same range

Interestingly, both the existence and the extreme rarity of (28) can beexplained by Boersma & Hayes’s actual hypothesis that “the values of thenormal distribution [assumed for each constraint] never actually reach zero”. Inprinciple, all constraints therefore overlap, but the odds of a variable rankingare “vanishingly low” between constraints whose center points are relativelyfar apart from each other. The example in (28) may be the freak realization ofsuch a low probability.

4.3 Conclusion on FrenchThe local conjunction of ONSET and NOCODA resolves an interesting puzzle

which arises in French out of the interaction of optional schwa deletion and h-aspiré words. However, adjacent syllables constitute the relevant domain forthis conjoined constraint in French, whereas the syllable itself is the requireddomain in Dutch at Stage III and in Italian. These results are in conflict with


the idea that the domain of a conjoined constraint is uniquely self-defined byits constituent simplex constraints. Domain-indexation for Local Conjunctionmay thus be needed independently of its already established motivation in Self-Conjunction.

‘Liaison sans enchaînement’ violates {ONSET&NOCODA}. Our explanationfor this phenomenon and related syllabification effects across words relies onthe combination of two important concepts, intrinsically ranked hierarchies andconstraint range. The intrinsically ranked hierarchies P and M capture theessence of the two major competing forces at work in the syllabification ofword-final consonants, based on context type for the phonological hierarchy(obligatory vs. optional liaison contexts) and on the nature of final consonantsfor the morphological hierarchy (FFC vs. LFC). The notion of constraint rangeinsightfully captures the variable interactions between the members of the twohierarchies and makes plausible predictions about the relative frequency of thesyllabification effects.

5. General ConclusionWe have argued in this paper for the existence of {ONSET&NOCODA} as a

legitimate conjoined constraint, drawing evidence from three separate strandsof research: language acquisition (Dutch), suppletion (the distribution of themasculine definite plural article in Italian), and exceptions (the interaction ofoptional schwa deletion and h-aspiré words in French).

Several interesting by-products have emerged in the course of making thisargument. From a language-specific perspective, we have naturally been led toconsider cases in Italian and French where {ONSET&NOCODA} is violated, andwe have developed analyses that address puzzling chestnuts in both languages,namely the distribution of the masculine singular definite article in Italian and‘liaison sans enchaînement’ in French.

From a broader perspective, our analyses first provide supplementaryevidence for Local Conjunction. They also speak to the issue of ‘LocalDomain’ for conjoined constraints, as they militate in favor of domain-indexation and Self-Conjunction, without denying the basic principle oflocality. In addition, our treatment of the Italian definite articles exemplifies animportant general argument in support of OT. Suppletive allomorphy is inprinciple out of reach for rule-based analyses, even when the distribution of thesuppletive allomorphs follows regular patterns linked to solid linguisticgeneralizations. By contrast, OT can generate such distributions in a non-arbitrary fashion with its single customary tool, universal constraints and theirlanguage-specific rankings. At the other end of the spectrum, we have alsoargued against ‘Generalized Suppletion’, the view that all surface allomorphs


can be mental representations and that traditional underlying representationscan be completely dispensed with.

The French case of ‘liaison sans enchaînement’ and the variability of theinteractions between phonological and morphological forces acting onsyllabification across words support an interesting novel conception ofconstraint ranking: constraints have a value range on a continuous rankingscale, rather than a set point. The constraint-range approach not only allows theformal encoding of the variable syllabification effects observed in differentcontexts by different types of final consonants before vowel-initial words; italso makes plausible and verifiable frequency predictions about these effects.

REFERENCES31

Aissen, Judith. 1999. “Markedness and Subject Choice in Optimality Theory”.Natural Language and Linguistic Theory 17.673-711.

Alderete, John. 1997. “Dissimilation as Local Conjunction”. Proceedings ofthe Twenty-Seventh Annual Meeting of the North East Linguistic Society, ed.by Kiyomi Kusumoto, 17-31. Amherst, Mass.: GSLA.

Bakovic, Eric. 1999. “Assimilation to the Unmarked”. Ms., The PennsylvaniaState University. [ROA-340]

---------. 2000. Harmony, Dominance and Control. Ph.D. Dissertation, RutgersUniversity. [ROA-360]

Boersma, Paul & Bruce Hayes. 2001. “Empirical Tests of the GradualLearning Algorithm”. Linguistic Inquiry 32.45-86. [ROA-348]

Burzio, Luigi. 1996. “Surface Constraints versus Underlying Representation”.Current Trends in Phonology: Models and Methods ed. by Jacques Durand& Bernard Laks, 123-141. University of Salford, Manchester: EuropeanStudies Research Institute.

---------. 1997. “Strength in Numbers”. University of Maryland WorkingPapers in Linguistics, Volume 5 ed. by Viola Miglio & Bruce Morén, 27-52.College Park, Maryland: Department of Linguistics, University of Maryland.

Chierchia, Gennaro. 1986. “Length, Syllabification and the Phonological Cyclein Italian”. Journal of Italian Linguistics 8.5-34.

Davis, Stuart. 1990a. “Italian Onset Structure and the Distribution of il and lo”.Linguistics 28.43-55.

31 Items listed as “ROA-xxx” are available from the Rutgers Optimality Archive(http://roa.rutgers.edu).


---------. 1990b. “The Onset as a Constituent of the Syllable: Evidence fromItalian”. Papers from the 26th Regional Meeting of the Chicago LinguisticSociety (Volume 2: The Parasession - On the Syllable in Phonetics andPhonology) ed. by Michael Ziolkowski, Manuela Noske & Karen Deaton,71-79. Chicago, Ill.: Chicago Linguistic Society.

De Jong, Daan. 1990. “The Syntax-Phonology Interface and French Liaison”.Linguistics 28.57-88.

Del Gobbo, Francesca. 2001. “An OT Account of the Distribution of Articlesin Italian”. Proceedings of the Fifth South Western Optimality TheoryWorkshop. ed. by Wind Cowles, Linda Godson, Andy Hickl, Alicia Munoz,Paola Nieddu, Todd O’Bryan & Heidi Tonomura. Linguistic Notes from LaJolla, 1-18. La Jolla, Calif.: Department of Linguistics, University ofCalifornia, San Diego.

--------- & Bernard Tranel. 2000. “On the Allomorphic Distribution of ItalianDefinite Determiners: Theoretical Implications”. Handout of ColloquiumTalk, University of Southern California, April 24, 2000.

Dell, François. 1985. Les règles et les sons. Paris: Hermann.Dolbey, Andrew. 1997. “Output Optimization and Cyclic Allomorph

Selection”. Proceedings of the Fifteenth West Coast Conference on FormalLinguistics ed. by Brian Agbayani & Sze-Wing Tang, 97-112. Stanford,Calif.: CSLI Publications.

Elenbaas, Nine. 1999. A Unified Account of Binary and Ternary Stress. Ph.D.Dissertation, Utrecht University. [ROA-397]

Encrevé, Pierre. 1988. La liaison avec et sans enchaînement. Paris: Seuil.Flemming, Edward. 1995. Auditory Representations in Phonology. Ph.D.

Dissertation, University of California, Los Angeles.---------. 1997. “Some Arguments against Underlying Representations”.

Colloquium Talk, University of California, Irvine, October 24, 1997.Fukazawa, Haruka. 1999. Theoretical Implications of OCP Effects on Features

in Optimality Theory. Ph.D. Dissertation, University of Maryland at CollegePark. [ROA-307]

Hargus, Sharon. 1997. “Prosodically Conditioned Allomorphy in OptimalityTheory: Brevity in Witsuwit'en”. Ms., University of Washington, Seattle.

Itô, Junko, Yoshihisa Kitagawa & Armin Mester. 1996. “Prosodic Faithfulnessand Correspondence: Evidence from a Japanese argot”. Journal of EastAsian Linguistics 5.217-294.

Itô, Junko & Armin Mester. 1998. “Markedness and Word Structure: OCPeffects in Japanese”. Ms., University of California, Santa Cruz. [ROA-255]

Itô, Junko & Armin Mester. 1999. “On the Sources of Opacity in OT: Codaprocesses in German”. Ms., University of California, Santa Cruz. [ROA-347]


Kager, René. 1996. “On Affix Allomorphy and Syllable Counting”. Ms.,University of Utrecht. [ROA-88]

---------. 1999. Optimality Theory. Cambridge: Cambridge University Press.Kaisse, Ellen. 1985. Connected Speech: The interaction of syntax and

phonology. Orlando, Florida: Academic Press.Katayama, Motoko. 1998. Optimality Theory and Japanese Loanword

Phonology. Ph.D. Dissertation, University of California, Santa Cruz.Kaye, Jonathan. 1991/92. “Do you Believe in Magic? The story of s+C

sequence”. SOAS Working Papers in Linguistics and Phonetics 2.293-313.Kirchner, Robert. 1996. “Synchronic Chain Shifts in Optimality Theory”.

Linguistic Inquiry 27.341-350.Labov, William. 1972. “The Social Setting of Linguistic Change”.

Sociolinguistic Patterns ed. by William Labov, 260-325. Philadelphia:University of Pennsylvania Press.

Levelt, Claartje & Ruben Van de Vijver. 1998. “Syllable Types in Cross-Linguistic and Developmental Grammars”. Ms., Free University ofAmsterdam & University of Tübingen. [ROA-265]

Lubowicz, Anna. 1998. “Derived Environment Effects in OT”. Ms., Universityof Massachusetts, Amherst. [ROA-239]

Marotta, Giovanna. 1993. “Selezione dell'articolo e sillaba in italiano: uninterazione totale?” Studi di Grammatica Italiana 15.255-296.

Mascaró, Joan. 1996a. “External Allomorphy and Contraction in Romance”.Probus 8.181-205.

---------. 1996b. “External Allomorphy as Emergence of the Unmarked”.Current Trends in Phonology: Models and Methods ed. by Jacques Durand& Bernard Laks, 473-483. University of Salford, Manchester: EuropeanStudies Research Institute.

McCarthy, John. 1997. “Process-Specific Constraints in Optimality Theory”.Linguistic Inquiry 28.231-251.

--------- & Alan Prince. 1993. Prosodic Morphology I: Constraint Interactionand Satisfaction. Ms., University of Massachusetts, Amherst & RutgersUniversity.

Mester, Armin. 1994. “The Quantitative Trochee in Latin”. Natural Languageand Linguistic Theory 12.1-61.

Morelli, Frida. 1999. The Phonotactics and Phonology of Obstruent Clustersin Optimality Theory. Ph.D. Dissertation, University of Maryland at CollegePark.

Morin, Yves-Charles. 1974. “Règles phonologiques à domaine indéterminé:chute du cheva en français”. Cahiers de Linguistique 4.69-88. Montréal:Presses de l'Université du Québec.


---------. 1998. “Remarks on Prenominal Liaison Consonants in French”. Ms.,Université de Montréal.

--------- & Jonathan Kaye. 1982. “The Syntactic Bases for French Liaison”.Journal of Linguistics 18.291-330.

Ohno, Sachiko. 1995. “Synchronically Unified Ranking and Distribution ofVoice in Japanese”. Proceedings of the First South Western OptimalityTheory Workshop ed. by Keiichiro Suzuki & Dirk Elzinga, 126-144. Tucson,Arizona: Department of Linguistics, University of Arizona.

Prince, Alan & Paul Smolensky. 1993. Optimality Theory: ConstraintInteraction in Generative Grammar. Ms., Rutgers University and Universityof Colorado, Boulder.

Russell, Kevin, 1999. “MOT: Sketch of an OT Approach to Morphology”.Ms., University of Manitoba. [ROA-352]

Selkirk, Elisabeth. 1974. “French Liaison and the X’ notation”. LinguisticInquiry 5.573-590.

Smolensky, Paul. 1993. “Harmony, Markedness, and Phonological Activity”.Handout for Rutgers Optimality Workshop 1, October 23, 1993. [ROA-87]

---------. 1995. “On the Internal Structure of the Constraint Component Con ofUG”. Handout for UCLA Colloquium Talk, April 7, 1995. [ROA-86]

---------. 1997. “Constraint Interaction in Generative Grammar II: LocalConjunction”. Handout for Hopkins Optimality Theory Workshop /University of Maryland Mayfest 1997, May 12, 1997.

Suzuki, Keiichiro. 1995. “Double-Sided Effects in OT: Sequential groundingand Local Conjunction”. Proceedings of the First South Western OptimalityTheory Workshop ed. by Keiichiro Suzuki & Dirk Elzinga, 209-224. Tucson,Arizona: Department of Linguistics, University of Arizona.

---------. 1998. A Typological Investigation of Dissimilation. Ph.D.Dissertation, University of Arizona, Tucson. [ROA-281]

Takeda, Kazue. 1998. “Single Consonant Reduplication and CausativeFormation in Kammu”. UCI Working Papers in Linguistics, Volume 4 ed. byHidehito Hoshi, 163-198. Irvine, Calif.: Department of Linguistics,University of California, Irvine.

Tranel, Bernard. 1981. Concreteness in Generative Phonology: Evidence fromFrench. Berkeley, CA: University of California Press.

---------. 1990. “On Suppletion and French Liaison”. Probus 2.169-208.---------. 1996a. “French Liaison and Elision Revisited: A unified account

within Optimality Theory”. Aspects of Romance Linguistics ed. by ClaudiaParodi, Carlos Quicoli, Mario Saltarelli & María-Luisa Zubizarreta, 433-455.Washington, D.C.: Georgetown University Press. [ROA-15]


---------. 1996b. “Exceptionality in Optimality Theory and Final Consonants inFrench”. Grammatical Theory and Romance Languages ed. by KarenZagona, 275-291. Amsterdam & Philadelphia: John Benjamins. [ROA-60]

---------. 1998a. “Questioning Generalized Suppletion”. Paper presented at theFourth South Western Optimality Theory Workshop, held at the University ofArizona, Tucson, April 4, 1998.

---------. 1998b. “Suppletion and OT: On the issue of the syntax/phonologyinteraction”. Proceedings of the Sixteenth West Coast Conference on FormalLinguistics ed. by Emily Curtis, James Lyle & Gabriel Webster, 415-429.Stanford, Calif.: CSLI Publications.

---------. 1999. “Optional Schwa Deletion: On syllable economy in French”.Formal Perspectives in Romance Linguistics ed. by J.-Marc Authier, BarbaraE. Bullock & Lisa A. Reed, 271-288. Amsterdam & Philadelphia: JohnBenjamins.

---------. 2000. “Aspects de la phonologie du français et la théorie del’optimalité”. Langue Française 126.39-72.

Wiltshire, Caroline & Elisa Maranzana. 1999. “Geminates and Clusters inItalian and in Piedmontese: A case for OT ranking”. Formal Perspectives onRomance Linguistics ed. by J.-Marc Authier, Barbara E. Bullock & Lisa A.Reed, 289-303. Amsterdam & Philadelphia: John Benjamins.

Zoll, Cheryl. 1996. Parsing Below the Segment in a Constraint BasedFramework. Ph.D. Dissertation, University of California, Berkeley.

---------. 1998. “Positional Asymmetries and Licensing. Ms., MIT. [ROA-282]

ON THE RELATION BETWEEN QUANTITY-SENSITIVE STRESSAND DISTINCTIVE VOWEL LENGTH

THE HISTORY OF A PRINCIPLE AND ITS RELEVANCE FOR ROMANCE

W. LEO WETZELSFree University of Amsterdam/Holland Institute of Generative Linguistics

0. IntroductionThe relevance of syllable weight for word stress assignment in Spanish,

Italian, and Portuguese remains controversial. Whereas for Spanish and Italianopinions are divided, there seems to be a tacit consensus among Lusitanists thatsyllable weight is irrelevant as a parameter for stress distribution. Most analysesof Portuguese stress are based on the observation that, at least in non-verbs, stressfalls on the last vowel of the stem, where �stem� is defined as the lexical word,i.e., without the �desinence�, which in most cases is the thematic vowel. Thus, thethematic noun cabeça �head� has prefinal stress, whereas in the athematic nariz�nose� stress is on the word-final syllable. This stem-based analysis successfullypredicts main stress in the great majority of the Portuguese non-verbs. However,both the fact that stress could never fall on the antepenultimate vowel in a wordlike aberto �open� (with a prefinal heavy syllable) and the fact that athematicnon-verbs usually end in a heavy syllable remain, from the point of view of therule that assigns stress, purely accidental.

It would be at least equally successful to posit a mechanism that takessyllable weight to be a conditioning factor for main-stress placement. A quantity-based account could provide an explanation for the systematic absence ofproparoxitonic stress in words with a prefinal heavy syllable and also explainwhy, in Portuguese, stress placement in newly created words is systematicallygoverned by the closed vs. open syllable distinction.

Elsewhere, I have submitted a number of observations which stronglysuggest that mora count has a major role to play in an analysis of main stress innon-verbs in Brazilian Portuguese (henceforth BP), as well as in other parts ofthe BP phonological grammar. Some of the arguments are well-known,especially the ones that also apply to Spanish and which have been identified by

W. LEO WETZELS220

James Harris in his work on Spanish stress. I repeat the most important ones in(1):

(1) Main arguments for a weight-sensitive stress rule in BrazilianPortuguese

a. Prefinal heavy syllables can never have stress to their left, whereasprefinal light syllables can.

b. Final heavy syllables are usually (86%) stressed in the existingvocabulary.1

c. Newly created words, especially acronyms, are almost exceptionallyadapted to fit the (right-to-left) moraic trochee pattern.

d. Portuguese has two contexts for productive stressed mid vowellowering which can be reduced to a single mora-countinggeneralization: Trimoraic Lowering (d[ç¤]lar �dollar�, m[ç¤]vel�movable�, proj[E¤]til �projectile�, el[E¤]tron �electron�, m[E¤]dico�physician�, esquel[E¤]tico �skeletal� (compare esquel[é]to�skeleton�), ab[ç¤]bora �pumpkin�, aer[ç¤]dromo �aerodrome�). A midvowel in a stressed syllable is lowered when followed by a heavysyllable or two light syllables, i.e., when followed by two moras.

e. �Impossible� codas (plosives and /f/) are more easily tolerated instressed syllables than in unstressed syllables mógno ~ móg[i]no�mahogany�, but ag[i]nóstico ~ *agnóstico �agnostic�, where [i] isepenthetic (see Collichon 2000).

I have also addressed elsewhere the arguments put forward against a weight-sensitive stress rule and shown that they do not apply to Portuguese (Wetzels1996, in preparation).

1. Quantity-sensitive stress and distinctive vowel length: PreliminariesHere I will concentrate on one interesting argument that is frequently

mentioned to dismiss the relevance of a moraic approach to Romance stress,which is the one that makes contrastive vowel length a prerequisite to weight-sensitive stress. The interest of the argument not only derives from its persistencein the literature (see, for instance, Kuryłowicz 1973 [1948], Newman 1972,Greenberg & Kaschube 1976, Hyman 1977, 1985, Ohsiek 1978, Roca 1990), butalso from the fact that Trubetzkoy�s Grundzüge der Phonologie (1939) is usuallyreferred to as the source of the principle from which the argument is derived.Indeed, Trubetzkoy has expressed himself in his Grundzüge, translated in 1 Statistics taken from Bisol (without date).

STRESS AND VOWEL LENGTH 221

English as Principles of Phonology, to which we will henceforth refer, in a waythat could be interpreted along the lines suggested by the critics of a moraicapproach to Spanish and Portuguese stress. I provide Trubetzkoy�s full statementbelow:

(2) The interpretation of long nuclei as geminated, or in terms of multinumberconstituency in general, may be regarded as an �arithmetic conception of quantity�.Languages in which this conception finds expression are �mora-counting� languagessince in these languages the smallest prosodic unit does not always coincide withthe syllable (1939/69:177).

I will assume that when contemporary phonologists claim that languageswithout a distinctive vocalic quantity opposition cannot have a weight sensitivestress rule, what they mean is that in such languages the location of stress cannotbe based on a mora count. We will see below that, according to Trubetzkoy, alength contrast on vowels can be expressed in various ways. One is by ageminate vs. single vowel opposition. In Trubetzkoy�s view, one prerequisite fora language to be a mora-counting language is to have not only an oppositionbetween short and long vowels, but also to instantiate the long vowel as ageminate nucleus. As the statement in (2) stands, Trubetzkoy�s words can beinterpreted in at least two ways. One is that all languages in which long vowelsare treated as consisting of a sequence of two identical units are mora-countinglanguages. The second interpretation is that all and only the languages with ageminate interpretation of long vowels are mora counting languages. Clearly,under the first interpretation, Trubetzkoy does not exclude the possibility thatlanguages which do not have contrastive vowel length can be of the mora-counting type.

In this paper I intend to support the following hypotheses:1. Trubetzkoy�s intention is reflected by the strong �all and only�

interpretation of (2).2. The distinction Trubetzkoy makes between syllable-counting languages

and mora-counting languages is the result of a search for an adequaterepresentation of distinctive vowel length. His research strategy wasnever directed toward the question of formulating stress rules or otherphenomena involving mora count. Therefore, the generalization ofTrubetzkoy�s conclusions to areas of phonology that are not strictlyrelated to the question of contrastive vowel length is necessarily based onshaky empirical foundations.

3. The strong interpretation has consequences for mora theory that probablyno modern phonologist would wish to accept, including those linguists

W. LEO WETZELS222

who refer to Trubetzkoy to dismiss a weight-sensitive stress rule for therelevant Romance languages.

4. Whereas the focus of structural phonology was on the notion of contrast(except for some 20 pages, Principles of Phonology is aboutphonological oppositions, correlations and neutralization of oppositions),in generative phonology it has shifted to the notion of rule. In particular,the class of what we now call lexical rules or generalizations would havebeen, in general, irrelevant to the discovery of phonological contrast andtherefore was never the object of systematic study by structuralistphonologists. Neither were the present so-called post-lexical rulesconsidered relevant, at least to the extent that they did not neutralizephonological oppositions. With the changing theory, the class oftheoretically relevant generalizations has drastically changed, and in factincreased, and with it our understanding of what conditioning factors canbe at the basis of phonological generalizations. Therefore, no claimregarding the class of possible phonological generalizations can, withoutgreat caution, be based on structuralist claims.

2. The history of a principle2.1 Trubetzkoy’s view of the mora

It is well-known that the Grundzüge is not only Trubetzkoy�s opus magnum,but also the endpoint of a lifetime�s reflection on phonological ideas and ofintensive discussions with fellow linguists, most noticeably his friend andcollaborator Roman Jakobson. The phonological interpretation of vowel quantitywas one of Trubetzkoy�s favorite topics, as well as a thorny problem, as we canjudge from the amount of energy he spent on disentangling it. The viewexpressed in the citation (2) above represents the culmination of a discussion thathad started almost twenty years earlier.

Trubetzkoy first addressed the question of the relation between stress andvowel quantity in a paper that deals with the disintegration of the commonRussian linguistic unity, entitled Einiges über die russische Lautentwicklung unddie Auflösung der gemeinrussischen Spracheinheit, which he published in 1925.In this study Trubetzkoy explains the disappearance of the Old-Slavic(�Urslavisch�) pitch accent as a consequence of the loss of quantitative voweloppositions. According to Trubetzkoy, this supposed causal relation followsfrom a universal implicational law, which establishes that �languages withpitch accent but without quantitative oppositions, as well as languages withfree quantity and free dynamic stress � do not occur anywhere in the world.�(1925:303-4) [My translation; LW].


In this citation, �free quantity� must be interpreted as �vowel lengthcontrast� and �free dynamic stress� as �(surface) contrastive expiratory accent�.Let us, for clarity�s sake, rephrase the universal implications formulated byTrubetzkoy as in (3):

(3) Universal implicational relations between vowel quantity and stress(Trubetzkoy 1925)a. If L has contrastive pitch accent, then L has a vocalic length contrast.b. If L has a vocalic length contrast, then L cannot also have a

contrastive dynamic (stress) accent, which implies:c. If L has a contrastive dynamic (stress) accent, then L cannot have

also contrastive vowel length.

Very probably, what Trubetzkoy meant by contrastive pitch accent in his1925 study was the opposition between at least one nucleus-internal pitchmovement (contour) and one or more other tones, either level or contour, as theexpression of word prominence. Such a contrast, he hypothesizes, can onlyoccur in a language that has phonological vowel length.

The implicational laws (3b,c) suggest the activity of some universalfunctional principle that militates against the multiple distinctive use of asingle phonetic parameter, in this case duration: vocalic length and dynamicaccent cannot both be contrastive in one and the same language.2 In a paper

2 The implicational laws in (3) embody the suggestion of a functional division of labor betweenpitch on the one hand and duration/intensity on the other: languages that use pitch as an across-the-board contrast cannot also use it for purposes of word prominence, and, similarly,languages that use duration as a general vocalic opposition cannot use it also to mark wordprominence. The suggestion that the phonetic parameters used to express word prominence arerelated to the structure of the phonological system reminds us of the conclusions of more recentdiscussions on the phonetic correlates of stress. Bolinger (1958) establishes a hierarchy ofperceptual cues for word prominence, arguing that the most important one is a change in pitch,followed by increase in duration and greater intensity in this order. According to Hyman(1977), cited in Berinstein (1979:1), this hierarchy is universal. Berinstein (1979) shows thatthe order within the hierarchy established by Bolinger is indeed dependent upon structuralproperties of the phonology of a given language. In a tone language, fundamental frequencycannot be a cue for stress because it is already used to signal the distinctive tone contrast.Inversely, in languages with a length contrast on vowels, fundamental frequency and intensityare the most important cues for the perception of word stress. The major differences betweenthe more recent proposals and Trubetzkoy�s laws is, first, the discovery that there is a universalhierarchy that languages in which neither tone nor length is contrastive respect, second, that theFunctional Load Hypothesis (Berinstein 1979) appears to hold also in languages with fixed,i.e., non-contrastive, stress.

W. LEO WETZELS224

entitled Zur allgemeinen Theorie der phonologischen Vokalsysteme publishedin 1929, Trubetzkoy returns to these laws in somewhat greater detail. He pointsout that the implication given above in (3b,c) had already been proposed byJakobson in 1923, in a study on Czech metrics, called O češskom stiche.3Trubetzkoy then uses the implication as an illustration in a lesson onphonological analysis that aims to show the fundamental difference betweenlinguistic (phonological) and non-linguistic facts. Although Trubetzkoy citesEnglish and German as counterexamples to Jakobson�s universal, he takes it asan important cross-linguistic generalization for which he tries to find aprincipled explanation. According to his view, the reason why the implicationsin (3b,c) hold is because the basic (universal) phonological opposition involvedin both contrastive vowel quantity and contrastive dynamic accent is Intensity.This means that contrastive vowel quantity as well as contrastive dynamicaccent are language-specific implementations of a phonological oppositionbetween minimal intensity and maximal intensity, as schematized in (4):

(4) Intensity (minimal~maximal)

Free Quantity Free Expiratory Accent(Unrestricted number (One contrastive �strong� vowel perof contrastive �weak� and �strong� word, e.g. Spanish, Italian,

vowels per word, e.g. Latin) Portuguese)

Languages do not need to implement the Intensity opposition, in which caseneither word stress nor vowel quantity are used contrastively, as in Polish4 orArmenian.

According to my understanding of Trubetzkoy�s lesson, Classical Latinshould be a language where the feature of Intensity is implemented as anacross-the-board vowel length opposition. It follows that main stress inClassical Latin, which according to Trubetzkoy is dynamic5, cannot becontrastive. It must be a non-phonological or �mechanical� fact, in more 3 Thanks to Ben Hermans for his translation and discussing the relevant passages with me.4 In Polish, a language without contrastive length or tone, stress is indicated as predicted byBolinger: most importantly by fundamental frequency, secondarily by duration and least of allby intensity (cf. Jassem, Morton & Steffen Batóg 1968).5 Notice that, if Bolinger and Hyman are right, Classical Latin stress should have been signalledprimarily by fundamental frequency and only secondarily by intensity. As said before, it is notmade clear by Trubetzkoy what exactly he meant by �dynamic� or �expiratory� stress. We mustprobably understand these terms as meaning �non-musical (or non-pitch) accent� as definedunder (3), i.e., either duration, intensity, or both.


modern terms, a surface-true generalization, which it is. On the other hand, in alanguage like Portuguese, where Intensity is implemented as a (surface)contrastive expiratory accent, only one syllable per word may be contrastivelystrong, as in sábia �wise-FEM�, sabía �know-IMPF.SG�, sabiá �(type of) bird�.The notions of quantity and expiratory accent are, according to Trubetzkoy,phonetic, not phonological. What matters from a phonological point of view isthe opposition between minimal and maximal intensity. Languages may usethat opposition to create main word stress, or to create quantitative voweloppositions across the board.

In the same paper, Trubetzkoy briefly returns to law (3a), which establishesthat languages with a musical accent must have a vocalic length contrast, butnothing essential is added to what he had said before.

The importance that Trubetzkoy attaches to the issue of vocalic quantity,and probably also his intellectual struggle to get a grip on the problem, isshown by the fact that he dedicated two more papers exclusively to this subject.One is relatively short, called Die Quantität als phonologisches Problem,presented at the Fourth International Congress of Linguists held in Copenhagenin 1936. Another one is much more elaborate, entitled Die phonologischenGrundlagen der sogenannten “Quantität” in den verschiedenen Sprachen,written in 1934 and published in 1938. Since the two papers express essentiallythe same ideas, although sometimes worded in a different way, I will focus onthe longer 1938 study (Trubetzkoy 1938b). Compared to his earlier writings onthe subject, this paper shows two new insights. One explains why English andGerman are not exceptions; the other makes precise what is meant by musicalaccent and how it relates to the representation of vocalic quantity.

The fact that Trubetzkoy was not entirely satisfied with his earlier (1929)explanation of Jakobson�s implicational laws was almost predictable, since hehimself had noticed that there were at least two languages, German andEnglish, that seemed to oppose the claim that distinctive vowel length andcontrastive expiratory stress are incompatible entities. Obviously, if theselanguages were really counterexamples, his �universalist� explanation could notbe true. Indeed, the major change that he introduces in the new study is clearlyinspired by his desire to understand the seemingly exceptional status of Englishand German, to which he now adds Dutch.

According to Trubetzkoy, we should conceive of the short and longelements in a quantitative opposition as a point opposed to a line. While theline can be stretched out, the point cannot. He subsequently asks the questionin what different ways can the long element be long. In the view he nowdevelops, the languages of the world are classified into three different types

W. LEO WETZELS226

with regard to the way in which they implement the correlation of phonologicalquantity, as is represented in (5):

(5) Possible implementations of Phonological Quantity

Opposition of Intensity Analytic Quantity Opposition of(1 vs. 2 �points�) Close Contact

non-culminative culminativeLatin Spanish pitch-accent languages Dutch

Italian EnglishPortuguese German

The oppositions of Intensity and Analytic Quantity appear to be mutuallyexclusive, and this is why they are represented in (5) as two alternative ways ofimplementing the more basic opposition of Phonological Quantity. On theother hand, the opposition of Close Contact is independently available and cancombine with either Intensity or Analytic Quantity. Dutch, English and Germanare languages that realize the former combination, i.e., Close Contact and(Culminative) Intensity, whereas Siamese and Hopi, which Trubetzkoydiscusses in his Principles (1969:179), are claimed to combine the correlationof Close Contact with Analytic Quantity. The expression �Close Contact� isused in the English translation of Grundzüge as the equivalent of whatTrubetzkoy termed the �Silbenschnittkorrelation�, literally the �syllable cutcorrelation�. The opposition refers to the way in which consonants may relateto the preceding vowel. In languages using this opposition, long vowels are notimpeded in their articulation by the following consonant, whereas short vowelssound short because of the fact that the articulation of the following consonantcuts off its normal flow, as in German: satt ~ Saat, but when no consonantfollows, only the long vowel appears: sah [za:] �saw 1p�~ *sa; or in Dutch lat~ laat (�slat�~ �leave�), but without a following consonant only la �drawer� isa possible word, with long [a:], not *l[a].

So, given the existence of the close contact opposition, Dutch, English andGerman are no longer exceptions to the claim that distinctive quantity isincompatible with distinctive durational word-stress.

Contrary to these Germanic languages, in languages which use distinctivevowel quantity in a non-culminative function, such as Latin, phonemic lengthis the expression of Intensity. The reason why this must be so is now clear:these languages never use Intensity for the purpose of contrastive word stress.Rather, in such languages the locus of expiratory accent is always non-


distinctive, because mechanically assigned, either through reference to theword-boundary (e.g. initial stress in Finnish, final stress in Persian, Yakut, etc.)or to the quantity of the boundary syllables, counting either from the beginningor the end of the word, like in Latin. Consequently, in these languages theposition of stress serves not the differentiation of meaning but the delimitationof words. On the other hand, as we already have seen above, languages that useintensity for the purpose of differentiating meaning by emphasizing a singlesyllable in the word cannot also use it for other distinctive purposes. In theselanguages stressed syllables are always long, and unstressed syllables are short,as in the Romance languages.

Another conception of quantity is one in which the long vowel isconsidered to be divisible in two parts, a beginning and an end. Length can becreated here because the transition from the first part to the second can take anyamount of time. Usually this �analytic� quantity is expressed phonetically as anopposition between contour tones that are realized on the long nuclei but noton the short ones. However, Trubetzkoy now adds that it does not really matterwhat exactly causes the segment-internal movement, whether it be tonal ordynamic (whatever he means by the latter, in phonetic terms). The importantphonological fact is the language�s potential to foreground only one part of along nucleus, which presupposes its composite nature. Here Trubetzkoy clearlyoffers a refinement of the first part of his law, the one that requires pitch-accentlanguages to have distinctive vowel length. It seems that the implication hasnow become one between distinctive moraic stress and distinctive analyticquantity, although the exact statement of the revised law is left to the reader�sinterpretation. We propose that the formulation in (6) comes close to whatTrubetzkoy had in mind:

(6) If L has distinctive moraic stress than L must have (across-the-boarddistinctive) analytic quantity.

The part between parentheses seems crucial, because if the opposition betweengeminate and non-geminate vowels only occurred in stressed syllables, theimplication would be tautological. We must consequently assume that if alanguage is of the mora type, all long vowels are bimoraic, even if there is noaudible proof for it (i.e., in unstressed long vowels; see also Trubetzkoy1969:173). A way of understanding the functionality behind the new definitionof the law is, perhaps, to consider that if a language has both non-culminativeand culminative distinctive quantity, then the quantity must be analytic for the

W. LEO WETZELS228

stress to create a kind of distinctive prominence (contour) that is different fromthe contrastive prominence of long vowels (durational) in unstressed position.

Trubetzkoy uses the term mora for the first time in his 1938 study. In hisview, moras are subsyllabic elements to which the two points of a long vowelindividually belong (cf. 1938b:164). In deciding which languages belong to themora type, i.e., the analytic type, Trubetzkoy clearly proceeds on a �what yousee is what you get� basis, probably because he considers the analyticalquantity opposition as exceptional, and something you need clear empiricalevidence for.6 The kind of evidence for analytic quantity that he wishes toconsider is the one that relates to specific types of vocalic distinctiveproperties, in particular those that use intra-vocalic positions as points ofcontrast, i.e., the beginning of a vowel as opposed to its end. As such,Trubetzkoy gets trapped in circular reasoning, which is particularly visiblewhen he proposes that real tone languages (he refers to African languages,1938b:164,fn1) are also mora-counting, despite the fact that in many languagesof this type the realization of a sequence of two or even three tones on a singlevowel does not coincide with a length contrast, very often not even a phoneticone. In Trubetzkoy�s view as expressed in his 1938b study, the set of moracounting languages coincides more or less with the class of languages thatshow contrastive vocalic contour properties, either as a pitch-accent or as anacross-the-board tonal contrast.

From what precedes it is clear that Trubetzkoy�s preoccupation is with theexplanation of distinctive vowel length and how to represent it, not with thenotion of weight as it is conceived of in modern theories of phonological(prosodic) representations. There is more proof for this conclusion, as well asfor the circularity of his reasoning. Trubetzkoy discusses Italian as an exampleof a language that opposes single consonants to geminate consonants.However, the fact that a language uses the analytical quantity correlation inconsonants is in itself not enough for it to be a mora type language. SinceItalian belongs to the class of languages that implement vocalic quantity in aculminative distinctive function and because it is not a pitch accent languagenor a tone language, it cannot be a mora language. Consequently, in Italian,geminates are not moraic, although they consists of two �points�. It musttherefore be the case that one point of the geminate belongs to the left-hand

6 This is very clear from the following statement (1938b:159): �Die einfachste Auffassung derQuantität liegt in den Sprachen mit nichtgipfelbildender Intensitätskorrelation vor� (italics asin the original text). In translation: �The simplest conception of quantity occurs in thelanguages that have a non-culminative intensity correlation.� Recall that in Trubetzkoy�s viewthese are languages like Latin, which are not considered to belong to the moraic type.


syllable, as opposed to the mora as a subsyllabic constituent, the other point tothe right-hand syllable.7 Trubetzkoy�s analysis of Italian quantity is particularlyrevealing. We must now conclude that, since the mora is a subsyllabic prosodicelement that only occurs in languages with an analytical quantity opposition, itcannot possibly be the case that Portuguese or Spanish, like Italian, is a mora-counting language. However, it also means from Trubetzkoy�s perspective thatthese languages have no moras at all. Therefore, one clear and probablyunacceptable consequence of Trubetzkoy�s position is that there can be nolanguages without distinctive vowel length that have compensatorylengthening, which in modern phonology is seen as mora stability underdeletion of a moraic segment, or any other generalization based on a moracount. This implication is demonstrably wrong, also for Portuguese (cf. 1dabove).

Illuminating for other reasons is the fact that Trubetzkoy considerslanguages like Latin as using a monolithic length opposition, not as a moralanguage. Latin counts syllables, and yet distinguishes, also according toTrubetzkoy himself, between heavy and light syllables (see also 1938b:165,fn.1). How the Latin type of �heaviness� is defined is not made clear, which isof course disturbing, because one wishes to understand why long vowelspattern with diphthongs and short vowels in closed syllables to make a Latinsyllable heavy8. Leaving this aside, the critical fact here is that the light vs.heavy syllable distinction, which is crucial for the definition of the Latin stressrule, does not rest upon the presence of moras. Indeed, Trubetzkoy is veryexplicit in distinguishing between languages that oppose light syllables toheavy syllables and languages that oppose monomoraic syllables to bimoraicsyllables:

(7) Whereas in the languages with an �energetic� conception of quantity onlymonolithic syllables exist, of which one part is heavy, the other light, inlanguages with an analytic conception of quantity not syllables but moras areconsidered units. Here, not heavy and light syllables are opposed to each other,but bimoraic and monomoraic syllables (Trubetzkoy 1938b:164; italics areoriginal). [My translation; LW].

According to Trubetzkoy, there are languages that count moras andlanguages that distinguish heavy syllables from light syllables. Obviously, thisdistinction only makes sense if we understand that Trubetzkoy�s idea of the 7 For further discussion of Italian, see also Principles (1939/69:201).8 And, of course, why Latin could nevertheless have compensatory lengthening, as in n[i:]dus(< *nisdos) �nest�, c[a:]nus ( < *casnos) �grey�, c[o:]mis (< cosmis) �cheerful�, etc.

W. LEO WETZELS230

mora is the (provisional) conclusion of his search for the proper representationof distinctive vowel length, combined perhaps with some intuitive (andpossibly wrong) notion of the unmarked (i.e., non-analytic) phoneticimplementation of this opposition.

2.2 Jakobson’s reactionIf we now reread Trubetzkoy�s statement in (2) above, we would probably

not wish to refer to it in order to dismiss a Portuguese or Spanish stress rulebased on a mora count. We would probably feel that the statement is wronganyway, because it is based on too narrow a conception of the mora. Unless, ofcourse, something happened in the short period from 1934 when he wrote his1938b study and the years 1935 to 1938, when he was writing his Principles,something which persuaded him to revise his view on the mora dramatically.As a matter of fact something of this kind did happen. While he considered theuse of analytical quantity the exceptional case in his 1938 study, Trubetzkoystates in his Principles:

(8) It may be noted that the (non-culminative) correlation of intensity is acomparatively rare phenomenon. In any event, the correlation of prosodicgemination occurs much more frequently (1969:184).

So what happened that made him change his mind so drastically? In a paperpublished in 1937, but written in 1936, Jakobson investigates the functions thatword stress can have in different languages and how these functions relate tovocalic quantity. In this paper, he states that he will adopt what he considersthe important distinction proposed by Trubetzkoy between languages wherelength is the expression of the �stretchability� of the phoneme, on the one hand,and, on the other hand, languages that use the opposition of close contact.Surprisingly he does not mention Trubetzkoy�s further division of quantitylanguages into systems of monolithic and analytic quantity. Could it be thatJakobson rejects the relevance of this distinction? The idea that he is indeedpreparing the ground for a severe criticism of Trubetzkoy�s conception ofquantity is confirmed by the fact that he relativizes Trubetzkoy�s generalization(which is actually his own), stated as (3b,c) above, according to whichdistinctive vowel length cannot be combined with distinctive dynamic stress.He calls it a tautology, which fact he explains in (9) below:

(9) If in a language with distinctive quantity the accent is mobile [i.e., surfacecontrastive: LW], it can choose between two short contiguous syllables as wellas between the beginning and the end of a long vowel, and would thus


necessarily become a polytonic accent. [Jakobson 1937/1971:258; Mytranslation; LW]

The fact that Jakobson calls his own implicational law a tautology isinteresting, because implicitly it means that, in his view, vowel length isalways of a composite nature, as is very clear from the explanation he providesin (9).9 Distinctive vowel quantity is always of the analytic kind and languagesthat combine this with distinctive main accent must realize it as a pitch accent.If stress is non-distinctive, it does not really matter how it is realized, because itis phonologically irrelevant. It looks as if Jakobson wants to break out of thecircle in which Trubetzkoy had placed himself. Indeed, he goes on to claim thatthe analytic quantity opposition is much more widespread than Trubetzkoysuggests. In particular, it must at least be assumed for any language in which atleast one of the following conditions holds:

(10) Languages with analytic vowel length (according to Jakobson 1937):a. Languages where a morpheme boundary can fall �in the middle� of

a long vowel.b. Languages where it can be shown that diphthongs and long vowels

function together in phonotactic restrictions.c. Languages where long vowels count as two units for the

localization of word stress (Latin now becomes a mora countinglanguage. Trubetzkoy adopts Jakobson�s formulation of the Latinstress rule which is that �stress is on the penultimate mora beforethe last syllable�).

d. Pitch-accent languages.e. Languages that have the contrast �with stød/without stød� where the

stød appears contrastively only in long vowels, diphthongs andvowel + sonorant consonant sequences.

f. Languages that have diphthongs.

Jakobson obviously succeeded in convincing Trubetzkoy to change his mindabout the relative markedness of the language-specific implementation of thelength opposition. In his Principles he adopts all of Jakobson�s criteria, exceptfor one, which is the one stated in (f). 9 At least it means that long vowels can at any time be analyzed as consisting of two parts,which amounts to the same. Observe that in Trubetzkoy�s interpretation of the law it was nottautological, since his view of the unmarked vocalic length was long vs. short. In such a view,stress cannot automatically distinguish between two parts of a long vowel. It does raise thequestion of how monolithic stress can become analytic in the history of a language.

W. LEO WETZELS232

According to Jakobson we can consider diphthongs (phonetic orphonological) as long vowels with a segment-internal shift in their articulation,much like long vowels that carry a tone movement. In languages that have bothlong vowels and diphthongs, the diphthongs show the composite nature of the�flat� long vowels. Jakobson�s reasoning is particularly revealing, since itimplies that diphthongs must be considered as intrinsically bimoraic. It followsthat languages that have diphthongs without having long vowels are potentiallyalso mora-counting. Hence, we can now refer to Jakobson instead of toTrubetzkoy to defend the idea that Portuguese, which has diphthongs, is amora-counting language. Obviously Trubetzkoy could not accept the presenceof diphthongs as a generally valid criterion for the identification of vowelquantity as analytic. It would almost imply that he had to give up completelyhis original distinction between monolithic and analytic length. He was notready to do so � understandably in a way. Instead, he maintained his originalclassification, but reversed the exceptional vs. normal relation betweenmonolithic and analytic length in favor of the latter. For Portuguese this stillmeans that it has no moras, since there is no distinctive vowel length. Theawkward implications of this radical statement are made explicit above.

3. ConclusionIt is now clear that Trubetzkoy�s principle is the conclusion of a long

debate about the representation of vowel length, and all the empirical evidenceconsidered, even by Jakobson, was selected in function of that uniqueobjective. We now know that there are languages which have no contrastivevowel length but which do have word-level, non-morphologically conditionedstress rules based on mora count, some of which are mentioned in Ohsieck(1978). Of course we cannot blame Trubetzkoy for not knowing all thelanguages of the world. We can certainly not blame him for not havingobserved that Spanish and Portuguese are examples of languages that countmoras despite the fact that they have no long vowels. Stress in these languagesis used distinctively and is therefore non-automatic. This is, in the eyes of aEuropean structuralist, in contradiction with the idea that it can be rulegoverned. We had to wait for a completely different conception of what aphonological generalization may look like, in order to state the Romance stressfacts as being rule governed. To conclude, it does not make much sense to referto Trubetzkoy to dismiss the possibility of a weight-sensitive stress rule forRomance.


REFERENCES

Berinstein, Ava E. 1979. A Cross-linguistic Study on the Perception andProduction of Stress. Los Angeles, Calif.: UCLA Working Papers inPhonetics 47.

Bisol, Leda. Without date. Sobre o Acento em Português. Ms., PontifíciaUniversidade Católica do Rio Grande do Sul, Porto Alegre, Brazil.

Bolinger, D. 1958. �A Theory of Pitch Accent�. Word 14:109-149.Collichon, Gisela. 2000. �A Epêntese Vocálica no Português do Sul do Brasil:

Análise Variacionista e Tratamento pela Teoria da Otimalidade�. Letras deHoje 119. 285-318.

Greenberg Joseph and D. Kaschube. 1976. �Word Prosodic Systems: aPreliminary Report�. WPLU 20.1-18.

Hyman, Larry. 1977. �On the nature of linguistic Stress”. Studies in Stress andAccent. (=Southern California Working Papers in Linguistics, 4) ed byLarry Hyman, 37-82. Los Angeles, Calif.: University of SouthernCalifornia.

---------. 1985. A Theory of Phonological Weight. Dordrecht, ForisPublications.

Jakobson, Roman. 1923. O češskom stiche. Berlin.---------. 1937. “Über die Beschaffenheit der Prosodischen Gegesätze”.

Mélanges de linguistique et de philologie offerts à J. van Ginneken. Paris.(Citations refer to the same paper published in Roman Jakobson. 1971.Selected Writings I, second edition. 254-261. The Hague: Mouton.)

Jassem, Morton & Steffen Batóg. 1968. �The Perception of Stress in SyntheticSpeech-like Stimuli by Polish Listeners�. Speech Analysis and Synthesis 1.289-308.

Kuryłowicz, Jerzy. 1973 [1948]. �Contribution à la théorie de la syllabe�. Esquis-ses linguistiques I.193-220. Munich: Wilhelm Fink Verlag.

Newman, Paul. 1972. �Syllable Weight as a Phonological Variable�. Studies inAfrican Linguistics 3.301-323.

Ohsiek, Deborah. 1978. �Heavy Syllables and Stress�. Syllables and Segmentsed. by A. Bell & J.B. Hooper, 35-43. Amsterdam: North-Holland.

Roca, Iggy. 1990. �Diachrony and Synchrony in Word Stress�. Journal ofLinguistics 26.133-164.

W. LEO WETZELS234

Trubetzkoy, Nikolaj. 1925. �Einiges über die russische Lautentwicklung unddie Auflösung der gemeinrussischen Spracheinheit�. Zeitschrift fürslavische Philologie I.287-319.

---------. 1929. �Zur allgemeinen Theorie der phonologischen Vokalsysteme�.Travaux du Cercle Linguistique de Prague 1.39-67.

---------. 1938a. �Die Quantität als phonologisches Problem�. Actes du IVeCongrès des linguistes tenu à Copenhague du 27-8 au 1-9-1936. 117-122.Copenhagen: Einar Munksgaard.

---------. 1938b. �Die phonologischen Grundlagen der sogenannten �Quantität�in den verschiedenen Sprachen�. Scritti in Onore de Alfredo Trombetti.155-174. Milan: Ulrico Hoepli Editore.

---------. 1939/1969. Principles of Phonology (trans. by Christiane A.M. Bal-taxe). Berkeley & Los Angeles: University of California Press.

Wetzels, W. Leo. 1996. �Syllable Structure and Stress in BrazilianPortuguese�. Paper presented at the 26th Linguistic Symposium on RomanceLanguages, held in Mexico City, April 1996.

---------. In preparation. The Sound Pattern of Brazilian Portuguese.

INDEX OF TERMS & CONCEPTS

AAccent, 5, 42-44, 53, 103-110, 222-

226, 230-233pitch accent, 5, 42, 222-223, 228,

231Acquisition, 3, 7-8, 14, 64, 67, 120,

124, 130-131, 134, 136-138, 141,143, 191, 193, 211, 213

Alignment, 105-108, 113, 115, 208,210

Allomorphy, 1, 99, 191, 194-195, 197-199, 201, 203, 213suppletion, 3, 191, 195, 203, 213,

217-218syncretism, 57-58

Alternations, 89-90, 97-98, 203Aphasia, 132, 135, 140Approximation, 161, 171, 173Argumentative value, 163Article

definite, 3, 88, 99, 191, 194-195,197-199, 201-202, 213

Articulator group, 5, 85-87, 89, 91- 92,94-96

Assimilationnasal place, 1, 3, 69, 95, 99place, 3, 74, 76-77, 89, 92, 95

Attenuation, 172

BBilinguals, 7, 8, 27, 29-30, 124- 126,

145, 149-150, 157

CCalque, 8, 145, 149, 151-154, 157Child Language, 2, 7, 11, 127-128,

131, 133-134, 140

Clitic, 6, 59-61, 63-65, 152-153Codas, 14, 18, 71, 74, 77-78, 92, 96,

179-180, 183, 187-188, 193- 194,196, 199-200, 204, 206- 209, 211,220

Codeswitching, 6-9, 27-28, 30, 33- 38,145, 149, 157

Complement, 27, 29, 34, 36-37, 90,153

Complementizer, 7, 31-35, 150-152Comprehension, 7, 29-30, 34, 98, 133Consonants

cluster simplification, 3, 69-72Constraints

complexity, 6, 63-65conjunction, 3, 20-21,23 191-192,

197, 202, 206, 212-213discourse, 45, 49

linking, 91range, 3, 211, 213ties, 49

Contact, 5-7, 38, 94, 119, 128-129,131, 134, 138, 142-143, 145, 165,226, 230

Continuant, 5, 85-86, 88-90, 94-97, 99Contrast, 5, 46, 55, 70, 88, 97, 103,

105-106, 108-109, 123, 162-165,168, 170, 195-196, 202-204, 209,213, 221-223, 225, 228, 231

Convergence, 30, 134, 150, 153, 155-157

Coronal, 11, 18-19, 72, 75, 78, 89- 90Correspondence, 1-3, 17, 19-20, 23

Output-Output, 2-3, 69, 73-74; seealso Faithfulness

COUNT, 55, 58Creoles, 7, 119-120, 124, 129, 137-

139, 142-143

INDEX OF TERMS & CONCEPTS236

Cyclicity, 3, 69-70

DDeclaratives, 5, 101, 103, 106-108Delateralization, 5, 86, 93Deletion, 3, 71, 95, 106, 142, 200-

201, 203-204, 229Dialects, 1-2, 5, 11, 16-17, 24, 49, 55,

86, 89, 92-93, 97, 101, 109, 113,115, 119, 125, 130-131, 133, 137,152-153, 165-166, 182, 188, 197

Dieresis, 11-13, 17, 20, 22, 24Diphthongs, 2, 11, 13, 16, 17, 19, 21,

22-24, 229, 231Discourse, 34, 45-47, 49, 52, 135, 140,

161, 166-167, 170-172, 174constraints, 45, 49

Domainslocal, 191-192, 202, 206-207

EElision, 13, 15, 204-205Epenthesis, 94, 198Etymological, 17, 55, 57-58, 60, 62Exceptions, 75, 77, 133, 169, 186, 188,

213, 225-226Expectation contravention, 171

FFaithfulness, 11, 18-23, 71, 76-79, 81,

201; see also CorrespondenceFeature Geometry, 5-6, 59, 87, 92, 99

liquid, 85-88, 90, 96Features

binary, 55, 58-59, 63monovalent, 59-60, 65privative, 65

Floating segments, 198Focus, 4, 34, 39-40, 42-47, 49, 52- 53,

86, 101, 103-109, 112-113, 207,212, 222, 225

Foreigner, 7, 117-118, 131-132, 134,137

Fricative, 86, 90-91, 93Functional elements, 7, 27-30, 32, 35,

37

GGames, 14

javanais, 14-15, 25largonji, 14-15, 25

Geminates, 194-195, 199-200, 202,228,230

Generalized Suppletion, 199, 203, 213,218; see also Allomorphy

Glides, 2, 11-15, 17-21, 23-24, 91-92,98

HHabituality, 173H-aspiré, 3, 191, 203-205, 207, 212-

213Hybrids, 8, 58, 145-146, 156, 157

IImplicatures

conventional, 161conversational, 161-162, 169

Infinitive, 117-119, 121-124, 126-133, 143, 156

Input, 1-3, 19, 23, 40, 45-46, 64, 71-82, 136, 178-184, 187, 192, 209

Intensity, 223-225, 227-228, 230Intonation, 4-5, 9, 42, 44, 52-53, 101-

103, 109, 112-115, 168Intrusive Stop Formation, 86, 94Irony, 167, 169

LLaísmo, 55-58, 66Language

death, 6, 8, 145, 149-150, 157disorders, 132games, 2, 11, 14, 17

Leísmo, 55-58, 65

INDEX OF TERMS & CONCEPTS 237

Liason sans enchâinement, 207-208,210-212

Linking Condition, 89, 91-92Loanwords, 146, 149-150, 156Locality, 203, 207, 213Loísmo, 55-58, 66L-vocalization, 5, 86, 96

MMarkedness, 1-3, 6, 8-9, 11, 14, 17-24,

41, 60-61, 63-67, 71, 79, 81, 91-92,97, 111, 132, 156, 171, 187, 192,195, 200, 202, 204, 214-215, 217,231

MASS, 55Mental representations, 191, 203, 214Monolinguals, 35, 125, 146, 157Mora, 4, 219, 221-222, 227-232

NNegative Polarity Item (NPI), 165Nuclear Stress Rule (NSR), 4, 41- 43,

45-47, 51-52Nucleus, 2, 11-15, 17, 23, 221, 227

OOnsets, 2-3, 11-14, 16-19, 21-23, 77,

107, 185, 187, 193-200, 204, 206-211, 214, 215

Opacity, 1, 3, 69, 81-82, 192Opposition, 65, 221, 223-231Optimality Theory, 1-4, 9, 25, 39, 42,

45, 48, 50, 52, 61, 66, 69, 71, 73-74,76, 81-83, 90, 177, 179-180, 184,188-189, 191-192, 195, 200, 203,213-218

Output, 3, 12, 16-21, 23, 48-50, 52,70-74, 76, 80, 82, 179, 180, 183-184, 187, 209, 212, 215

PPalatal, 75-76, 79-80, 92, 96, 98

Paradigms, 6, 55, 57-60, 62-63, 65,101, 118-119, 126-127, 129-130,133

Parallel, 3, 7, 69, 73-74, 82, 152, 166Peak, 15, 17, 21, 102-109, 114, 149Phonetics, 5, 11, 17, 87, 102, 122, 195,

223, 225, 227-228, 230-231Phrase, 3, 5, 31, 33, 36, 40-42, 44, 73-

74, 82, 92, 102, 110, 113- 114, 122,128, 208

Pidgins, 7, 118-120, 122, 124-129,131, 133, 137-142

Place of Articulation, 87, 89Polar, 6, 162-165, 167-168, 171- 173Polysemy, 1, 6, 161, 167-168, 170-

171, 174Pragmatics, 109, 111-112, 115, 161-

163, 165, 174-175Pronouns, 6, 55, 60-61, 63, 118- 122,

126-131, 152Prosodic, 4-5, 25, 34, 77, 104, 111-

112, 184-185, 221, 228, 230Proximal, 6, 162, 164-165, 168, 172-

174

RRanking, 2, 4, 19-21, 42, 45-48, 51-

52, 71-72, 74, 76-82, 95, 179, 185,187, 192, 195-196, 199, 203, 208,211-212, 214, 218variable, 209, 212

Reading, 7, 29, 34, 63, 113, 169, 171,173

Reduction, 1, 5, 108-109, 139Reduplication, 2, 177-188Referential, 55-60, 62, 65REFLEXIVE, 61, 65Representations, 1, 5-6, 25, 34, 45, 59,

62-65, 90, 97, 103, 106, 112-113,122-123, 127, 171, 195, 198-200,203, 214, 221, 225, 228-229, 232

Resyllabification, 210

INDEX OF TERMS & CONCEPTS238

SSchwa Deletion, 17, 204-205

optional, 3, 191, 203-205, 207, 212-213

Scrambling, 41, 49Second Language, 7, 118, 120, 121,

123-126, 128, 131-132, 134, 136,143

Sonority, 96, 194-195Spirantization, 5, 86-87, 90-92, 96- 99Stereotypes, 7, 117, 121, 123, 125,

129, 130, 132, 134, 210Stress, 4, 24, 41-42, 44, 46-47, 51,

103, 115, 129, 183, 219-232dynamic, 222-223, 230sentential, 4, 40-47, 51-52

Structuralism, 1, 3-4, 222, 232SUBJECT, 40, 45-49, 51-52Subject Inversion, 39, 40, 50-52Substratum, 135Suffixes, 2, 14, 127, 146, 186Syllable, 3-5, 11-12, 14, 17-19, 22, 25,

42, 45, 94, 102-109, 178- 179, 181-183, 187-188, 191, 193- 194, 196,198, 200, 202, 204, 206-207, 212,218-221, 225-229, 231economy, 218weight, 219

Syneresis, 11-13, 18, 20-21Syntax, 4, 30, 35-37, 39, 42-45, 47-

48, 66, 68, 111-112, 114, 121, 125,132, 137-138, 141, 152- 153, 155,210, 216, 218

TTemplate, 12, 131, 135, 184Tones, 5, 102-103, 105-107, 111- 112,

223-224, 227-228, 232boundary, 102, 107contour, 5, 102-103, 105-109, 111,

223, 227-228Transfer

morphological, 146, 149Truncation, 178-180, 184Typology, 1, 3-4, 38, 65, 132

UUnmarked, 22, 60-61, 63, 65, 89, 103,

121, 134, 187-188, 230-231emergence of the, 177, 184, 187-

188

VVariation, 1-6, 8, 13, 15, 17-18, 20,

22-24, 55, 58, 61, 65, 67, 92, 122,142, 177, 181, 188, 191, 211, 214dialectal, 2, 5, 11, 16word-order, 4, 39, 41

Velar, 18Vowels, 2, 4, 11-13, 15, 17, 20, 23-

24, 91-92, 98, 187, 194-195, 197,201, 221, 223-224, 226-227, 229,231-232length, 4, 220-221, 223-225, 227,

229, 230-232

WWeight, 4, 210, 219, 221, 228

CURRENT ISSUES IN LINGUISTIC THEORY

E. F. K. KoernerZentrum für Allgemeine Sprachwissenschaft, Typologie

und Universalienforschung, Berlin

The Current Issues in Linguistic Theory (CILT) series is a theory-oriented series which welcomes contributions from scholars who have significant proposals to make towards the advancement of our understanding of language, its structure, functioning and development. CILT has been established in order to provide a forum for the presentation and discussion of linguistic opinions of scholars who do not necessarily accept the prevailing mode of thought in linguistic science. It offers an alternative outlet for meaningful contributions to the current linguistic debate, and furnishes the diversity of opinion which a healthy discipline must have. A complete list of titles in this series can be found on the publish-ers website, www.benjamins.com/jbp

��

209 KING, Ruth: The Lexical Basis of Grammatical Borrowing. A Prince Edward Island French case study. 2000.

210 DWORKIN, Steven N. and Dieter WANNER (eds.): New Approaches to Old Problems. Issues in Romance historical linguistics. 2000. xiv, 235 pp.��

211 ELŠÍK, Viktor and Yaron MATRAS (eds.): Grammatical Relations in Romani. The Noun Phrase. with a Foreword by Frans Plank (Universität Konstanz). 2000. x, 244 pp.��

212 REPETTI, Lori (ed.): Phonological Theory and the Dialects of Italy. 2000. x, 301 pp.��

213 SORNICOLA, Rosanna, Erich POPPE and Ariel SHISHA-HALEVY (eds.): Stability, Variation and Change of Word-Order Patterns over Time. with the assistance of Paola Como. 2000. xxxii, 323 pp.��

214 WEIGAND, Edda and Marcelo DASCAL (eds.): Negotiation and Power in Dialogic Interaction. 2001. viii, 303 pp.��

215 BRINTON, Laurel J. (ed.): Historical Linguistics 1999. Selected papers from the 14th International Conference on Historical Linguistics, Vancouver, 9–13 August 1999. 2001. xii, 398 pp.��

216 CAMPS, Joaquim and Caroline R. WILTSHIRE (eds.): Romance Syntax, Semantics and L2 Acquisition. Selected papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000. 2001. xii, 246 pp.��

217 WILTSHIRE, Caroline R. and Joaquim CAMPS (eds.): Romance Phonology and Variation. Selected papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000. 2002. xii, 238 pp.��

218 BENDJABALLAH, Sabrina, Wolfgang U. DRESSLER, Oskar E. PFEIFFER and Maria D. VOEIKOVA (eds.): Morphology 2000. Selected papers from the 9th Morphology Meeting, Vienna, 24–28 February 2000. 2002. viii, 317 pp.��

219 ANDERSEN, Henning (ed.): Actualization. Linguistic Change in Progress. Papers from a workshop held at the 14th International Conference on Historical Linguistics, Vancouver, B.C., 14 August 1999. 2001. vii, 250 pp.��

220 SATTERFIELD, Teresa, Christina M. TORTORA and Diana CRESTI (eds.): Current Issues in Romance Languages. Selected papers from the 29th Linguistic Symposium on Romance Languages (LSRL), Ann Arbor, 8–11 April 1999. 2002. viii, 412 pp.��

221 D’HULST, Yves, Johan ROORYCK and Jan SCHROTEN (eds.): Romance Languages and Linguistic Theory 1999. Selected papers from ‘Going Romance’ 1999, Leiden, 9–11 December 1999. 2001. viii, 406 pp.��

222 HERSCHENSOHN, Julia, Enrique MALLÉN and Karen ZAGONA (eds.): Features and Interfaces in Romance. Essays in honor of Heles Contreras. 2001. xiv, 302 pp.��

223 FANEGO, Teresa, Javier PÉREZ-GUERRA and María José LÓPEZ-COUZO (eds.): English Historical Syntax and Morphology. Selected papers from 11 ICEHL, Santiago de Compostela, 7–11 September 2000. Volume 1. 2002. x, 306 pp.��

224 FANEGO, Teresa, Belén MÉNDEZ-NAYA and Elena SEOANE (eds.): Sounds, Words, Texts and Change. Selected papers from 11 ICEHL, Santiago de Compostela, 7–11 September 2000. Volume 2. 2002. x, 310 pp.��

225 SHAHIN, Kimary N.: Postvelar Harmony. 2003. viii, 344 pp.��

226 LEVIN, Saul: Semitic and Indo-European. Volume II: Comparative morphology, syntax and phonetics. 2002. xviii, 592 pp.��

227 FAVA, Elisabetta (ed.): Clinical Linguistics. Theory and applications in speech pathology and therapy. 2002. xxiv, 353 pp.��

cilt.serieslist.indd 1 22/03/04 16:02:41

228 NEVIN, Bruce E. (ed.): The Legacy of Zellig Harris. Language and information into the 21st century. Volume 1: Philosophy of science, syntax and semantics. 2002. xxxvi, 323 pp.��

229 NEVIN, Bruce E. and Stephen B. JOHNSON (eds.): The Legacy of Zellig Harris. Language and information into the 21st century. Volume 2: Mathematics and computability of language. 2002. xx, 312 pp.��

230 PARKINSON, Dilworth B. and Elabbas BENMAMOUN (eds.): Perspectives on Arabic Linguistics. Papers from the Annual Symposium on Arabic Linguistics. Volume XIII-XIV: Stanford, 1999 and Berkeley, California 2000. 2002. xiv, 250 pp.��

231 CRAVENS, Thomas D.: Comparative Historical Dialectology. Italo-Romance clues to Ibero-Romance sound change. 2002. xii, 163 pp.��

232 BEYSSADE, Claire, Reineke BOK-BENNEMA, Frank DRIJKONINGEN and Paola MONACHESI (eds.): Romance Languages and Linguistic Theory 2000. Selected papers from ‘Going Romance’ 2000, Utrecht, 30 November - 2 December. 2002. viii, 354 pp.��

233 WEIJER, Jeroen van de, Vincent J. van HEUVEN and Harry van der HULST (eds.): The Phonological Spectrum. Volume I: Segmental structure. 2003. x, 308 pp.��

234 WEIJER, Jeroen van de, Vincent J. van HEUVEN and Harry van der HULST (eds.): The Phonological Spectrum. Volume II: Suprasegmental structure. 2003. x, 264 pp.��

235 LINN, Andrew R. and Nicola McLELLAND (eds.): Standardization. Studies from the Germanic languages. 2002. xii, 258 pp.��

236 SIMON-VANDENBERGEN, Anne-Marie, Miriam TAVERNIERS and Louise J. RAVELLI (eds.): Grammatical Metaphor. Views from systemic functional linguistics. 2003. vi, 453 pp.��

237 BLAKE, Barry J. and Kate BURRIDGE (eds.): Historical Linguistics 2001. Selected papers from the 15th International Conference on Historical Linguistics, Melbourne, 13–17 August 2001. Editorial Assistant: Jo Taylor. 2003. x, 444 pp.��

238 NÚÑEZ-CEDEÑO, Rafael, Luis LÓPEZ and Richard CAMERON (eds.): A Romance Perspective on Language Knowledge and Use. Selected papers from the 31st Linguistic Symposium on Romance Languages (LSRL), Chicago, 19–22 April 2001. 2003. xvi, 386 pp.��

239 ANDERSEN, Henning (ed.): Language Contacts in Prehistory. Studies in Stratigraphy. Papers from the Workshop on Linguistic Stratigraphy and Prehistory at the Fifteenth International Conference on Historical Linguistics, Melbourne, 17 August 2001. 2003. viii, 292 pp.��

240 JANSE, Mark and Sijmen TOL (eds.): Language Death and Language Maintenance. Theoretical, practical and descriptive approaches. With the assistance of Vincent Hendriks. 2003. xviii, 244 pp.��

241 LECARME, Jacqueline (ed.): Research in Afroasiatic Grammar II. Selected papers from the Fifth Conference on Afroasiatic Languages, Paris, 2000. 2003. viii, 550 pp.��

242 SEUREN, Pieter A.M. and Gerard KEMPEN (eds.): Verb Constructions in German and Dutch. 2003. vi, 316 pp.��

243 CUYCKENS, Hubert, Thomas BERG, René DIRVEN and Klaus-Uwe PANTHER (eds.): Motivation in Language. Studies in honor of Günter Radden. 2003. xxvi, 403 pp.��

244 PÉREZ-LEROUX, Ana Teresa and Yves ROBERGE (eds.): Romance Linguistics. Theory and acquisition. Selected papers from the 32nd Linguistic Symposium on Romance Languages (LSRL), Toronto, April 2002. 2003. viii, 388 pp.��

245 QUER, Josep, Jan SCHROTEN, Mauro SCORRETTI, Petra SLEEMAN and Els VERHEUGD (eds.): Romance Languages and Linguistic Theory 2001. Selected papers from ‘Going Romance’, Amsterdam, 6–8 December 2001. 2003. viii, 355 pp.��

246 HOLISKY, Dee Ann and Kevin TUITE (eds.): Current Trends in Caucasian, East European and Inner Asian Linguistics. Papers in honor of Howard I. Aronson. 2003. xxviii, 426 pp.��

247 PARKINSON, Dilworth B. and Samira FARWANEH (eds.): Perspectives on Arabic Linguistics XV. Papers from the Fifteenth Annual Symposium on Arabic Linguistics, Salt Lake City 2001. 2003. x, 214 pp.��

248 WEIGAND, Edda (ed.): Emotion in Dialogic Interaction. Advances in the complex. vi, 280 pp. + index. Expected Summer 2004��

249 BOWERN, Claire and Harold KOCH (eds.): Australian Languages. Classification and the comparative method. xii, 377 pp. (incl. CD-Rom) Expected Spring 2004��

251 KAY, Christian J., Simon HOROBIN and Jeremy SMITH (eds.): New Perspectives on English Historical Linguistics. Selected papers from 12 ICEHL, Glasgow, 21-26 August 2002. Volume I: Syntax and Morphology. Expected Fall 2004��

252 KAY, Christian J., Carole HOUGH and Irené WOTHERSPOON (eds.): New Perspectives on English Historical Linguistics. Selected papers from 12 ICEHL, Glasgow, 21-26 August 2002. Volume II: Lexis and Transmission. xii, 265 pp. + index. Expected Fall 2004 ��

cilt.serieslist.indd 2 22/03/04 16:02:42

caroline r. wiltshire, joaquim camps romance phonology and variation selected papers from the 30th...

Documents

current issues

john benjamins b

thlinguistic symposium

usaacknowledgmentsthe

floridaromance phonology

selected papersfrom

series iv

romancelanguages lsrl