PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
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 212
Lori Repetti (ed.) Phonological Theory and the Dialects of Italy
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
Edited by
LORI REPETTI State University of New York, Stony Brook
JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA
8
TM
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.
Library of Congress Cataloging-in-Publication Data Phonological theory and the dialects of Italy / edited by Lori Repetti. p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 212) Includes bibliographical references and index. 1. Italian langauge--Dialects--Phonology. 2. Italian language--Phonology. I. Repetti, Lori. II. Series. PC1726.P46 2000 457--dc21 00-064163 ISBN 90 272 3719 0 (Eur.) / 1 58811 026 5 (US) © 2000 – 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 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
ACKNOWLEDGMENTS This project began exactly five years ago, in May 1995 when I met with Pier Marco Bertinetto to discuss the possibility of publishing a collection of essays on Phonological Theory and the Dialects of Italy. The book was originally intended to be a companion volume to Adriana Belletti's excellent reader on Syntactic Theory and the Dialects of Italy (Torino: Rosenberg & Sellier, 1993). I was encouraged to pursue the project by Pier Marco Bertinetto, to whom I am particularly grateful for the support and encouragement he has shown me. I contacted the contributors soon thereafter, and the project was under way. I would like to extend my most heartfelt thanks to the contributors who have always been extremely patient and cooperative despite the many delays that we have experienced. Each article was peer reviewed, and I would like to thank the scholars who generously helped in the review process, including Mark Aronoff, Christina Bethin, Ellen Broselow, Luigi Burzio, Thomas Cravens, Robert Hoberman, José Hualde, Marie Huffman, Christine Kamprath, Michael Kenstowicz, Elinor Payne, Keren Rice, Bert Vaux, and Draga Zee. Thanks also go to Konrad Koerner and Anke de Looper for their precious assistance in publishing this volume. Finally, many many people have helped me during the preparation of this collection in various ways that are not always easy to define, and I would especially like to thank Mark Aronoff, John Bailyn, Anna Cardinaletti, Laura Minervini, Christina Tortora, my parents, my sisters, my daughter, and most of all Peter Ludlow, to whom 1 dedicate this book. Lori Repetti Port Jefferson, NY 26 May 2000
CONTENTS
Acknowledgments
v
List of Contributors
ix
Phonological Theory and the Dialects of Italy LoriRepetti
1
The Relationship Between Vowel Length and Consonantal Voicing in Friulian Marco Baroni & Laura Vanelli
13
Consonant Gemination in Neapolitan Barbara E. Bullock
45
The Feature [Advanced Tongue Root] and Vowel Fronting in Romance Andrea Calabrese
59
Vowel Alternation, Vowel/Consonant Assimilation and OCP Effects in a Barese Dialect Francesco Dlntrono & Rosemary Weston
89
How Many Moras? Overlength and Maximal Moraicity in Italy John Hajek
111
Stress Stability Under Cliticization and the Prosodie Status of Romance
137
Clitics
Michele Loporcaro Phonological Dissimilation and Clitic Morphology in Italo-Romance Martin Maiden
169
VIII
CONTENTS
Oxytone Infinitives in the Dialect of Pisa Giovanna Marotta
191
Sonority and Derived Clusters in Raeto-Romance and Gallo-Italic Jean-Pierre Montreuil
211
Stress and Schwa in Faetar Naomi Nagy
239
Vowel Lengthening in Milanese Pilar Prieto i Wives
255
Uneven or Moraic Trochees? Evidence from Emilian and Romagnol Dialects Lori Repetti
273
Index of Names
289
Index of Terms & Concepts
297
CONTRIBUTORS Marco Baroni Department of Linguistics UCLA Los Angeles, CA 90024 USA
[email protected]
John Hajek Dept. of French and Italian Studies University of Melbourne Parkville Vic 3052 Melbourne Australia
[email protected]
Barbara E. Bullock Department of French & Linguistics Penn State University University Park, PA 16802 USA
[email protected]
Michele Loporcaro Romanisches Seminar University of Zurich Plattenstrasse 32 8028 Zurich Switzerland loporcar @ rom. unizh. ch
Andrea Calabrese Department of Linguistics University of Connecticut Storrs,CT 06269 USA calabres @ uconnvm. uconn.edu
Martin Maiden Trinity College University of Oxford Oxford OX1 3BH UK
[email protected]
Frajncesco D'Introno Department of Spanish and Portuguese University of Massachusetts Amherst, MA 01003 USA
[email protected]
Giovanna Marotta Dipartimento di Linguistica Universita di Pisa Via S.Maria 36 56126 Pisa Italy
[email protected]
X
CONTRIBUTORS
Jean-Pierre Montreuil Department of French and Italian University of Texas Austin, TX 78712 USA
[email protected]
Lori Repetti Department of Linguistics SUNY Stony Brook, NY 11794-4376 USA lrepetti ©notes, cc. suny sb. edu
Naomi Nagy Department of English University of New Hampshire Durham, NH 03824-3574 USA ngn @ hopper, unh.edu
Laura Vanelli Dipartimento di discipline linguistiche, comunicative, e dello spettacolo Universitadegli Studi viaB.Pellegrino 1 35137 Padova Italy
[email protected]
Pilar Prieto i Vives Departament de Filologia Catalana Universitat Autonoma de Barcelona Facultat de Lletres, Edifici B 08193 Bellaterra (Barcelona) Spain
[email protected]
Rosemary Weston Department of Spanish Amherst College Box 2255 Amherst, MA 01002-5000 USA rwestongil @amherst.edu
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY LORI REPETTI State University ofNew York, Stony Brook 0.
Introduction The linguistic situation in Italy is quite complex. In addition to the standard language, scores of minor languages or 'dialects' are spoken. The Italian 'dialects' are not dialects of Italian in the usual sense of the term. They are daughter languages of Latin and sister languages of each other, of standard Italian, and of other Romance languages, and they may be as different from each other and from standard Italian as French is from Portuguese. Standard Italian is just one of the many languages spoken in Italy: it is derived from a Florentine variety of Tuscan which achieved literary and cultural prestige in the Fourteenth Century, but only in the Twentieth Century did it become a widespread spoken language. This book is part of an ongoing effort to present to a wider (English speaking) group of linguists a taste of the linguistic wealth that Italy has to offer, and the essays contained in this collection examine various aspects of the phonology of the minor languages of Italy. 1.
The Minor Languages of Italy Italy is unique within Europe in that there is so much linguistic variation concentrated in so small a geographic area. Why has so much linguistic variety survived in Italy? The answer has to do with Italy's political history: Italy was politically united only in the late Nineteenth Century. At the time of unification in 1861 it is estimated that only 2.5-10% of the citizens of the newly formed nation spoke Italian (Castellani 1982, De Mauro 1970). That means that the vast majority of Italians were monolingual (or polylingual) speakers of a 'dialect' or 'dialects', and for them 'Italian' was, for all intents and purposes, a foreign language. What happened in Italy over the next century is unprecedented in modern Europe: a language that had been foreign for about 90% of the Italians became a living and
2
LORI REPETTI
national language. Today nearly all Italians understand the national language, the vast majority can speak some variety of Italian, and many use it regularly both in and out of the home. According to a survey conducted in 1988, 34.4% of the population speaks Italian within the family (an additional 26% uses both Italian and a 'dialect'), and 47% usually speaks Italian outside of the home (an additional 19.5% uses both Italian and a 'dialect') (DOXA 1988). The study of the minor languages of Italy has now entered a period of unprecedented urgency. The socio-economic forces in effect in Italy have led many 'dialect' speakers to emigrate from small villages to larger towns and cities. There, the diverse local languages are replaced by local varieties of Italian and then by the standard language. It is very possible that within a few generations most of the 'dialects' will be so Italianized that very little will be left of their unique grammatical structure. With their loss, we are deprived of an invaluable and irreplaceable font of information on Romance languages and language itself. There are many reasons that linguists in general and phonologists in particular should be interested in the minor languages of Italy. Italy provides a remarkable laboratory for the study of synchronic and diachronic variation, and there are empirical and theoretical reasons that these languages are particularly significant. First, by examining minor variation in closely related languages, we can better understand the ways in which languages work. The in-depth study of the grammar of a particular language may help us to understand certain patterns attested in that language, but can not improve our understanding of other patterns. By comparing languages which look very different from one another we are not able to control the variables and thereby understand what the variation means. However, by examining minor variation among closely related languages we can keep the variables to a minimum and thereby better understand the variation itself. Since the goal of linguistic theory is not just to describe but to explain the fundamental properties of natural languages, it is critical that we understand the way variation is encoded into the system. Empirically, we have a unique situation in Italy since we find massive amounts of data that scientists have been collecting over the past century and a half. In addition, we have a very detailed knowledge of the history of these languages since most of the work done in Italian dialectology has been in the area of historical phonology. The 'dialects' of Italy offer a unique case study of what happens when a single language (Latin, in this case) evolves unchecked by a national norm for millennia. The many minor Romance languages spoken on the Italian peninsula can provide insight into how languages change and evolve across time and space.
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
3
2.
Italian Dialectology For decades 'Italian Linguistics' has been closely tied to the study of the minor languages of Italy, and 'Italian Dialectology' has been synonymous with historical phonology. This is due in large part to the history of linguistic studies in Italy. Italian (and Romance) Dialectology as a field of scientific investigation was established in the Nineteenth Century, unofficially with the publication of Ascoli's seminal paper, "Saggi ladini," in the first issue of his Archivio Glottologico Italiano (1873).1 Crucially, Ascoli's work, as well as that of his predecessors, was written within the framework of comparative-historical phonology, and their work established the tradition of studying the minor languages from the perspective of historical phonology, a tradition which continues uninterrupted to this day. In their recent manual on Fondamenti di Dialettologia Italiana, Grassi, Sobrero, & Telmon (1997:91-126) dedicate §3.3 to "I dialetti in Italia: I principali fenomeni linguistici" with four subsections divided among "I sistemi vocalici," "Altri fenomeni vocalici," "II consonantismo," "Qualche fenomeno morfologico." The heavy emphasis on phonology, specifically historical phonology, with only a brief mention of morphology and morphosyntax, is typical of the approach to the field even today. The study of historical phonology encompasses many topics. Most frequently, scholars have examined the evolution of Latin vowels and consonants in different grammatical contexts, but their work has also included the study of the influence of the substratum on modern varieties (for example, whether the spirantization of intervocalic voiceless stops in Tuscan can be attributed to the influence of Etruscan), toponomastics (especially those place names that are of non-Latin origin), philology (the connection between the written, usually literary, form and grammatical structure), and lexicon (of note in this context is the proliferation of linguistic atlases in the Twentieth Century: Atlante italo-svizzero, Atlante linguistico italiano, Atlante storico linguistico etnograficofriulano,Atlante linguistico etnografico italiano della Corsica, Atlante linguistico dei laghi d'ltalia, etc.). These scholars have also worked on the classification of the 'dialects', categorizing them usually on the basis of historical phonology (for example, the northern Gallo-Romance varieties are distinguished from the other Italo-Romance varieties on the basis of degemination vs retention of Latin geminate consonants) and the lexicon (for example, one of the isoglosses distinguishing the northern 'dialects' from the central and southern 'dialects' is the one which divides the areas using a form of HINC HODIE from the areas using a form of HODIE for 'today'). 1
For a introduction to the history of dialectology in Italy, see Benincà (1996), Cortelazzo (1980), Grassi, Sobrero, & Telmon (1997: 33-69), and references therein.
4
LORI REPETTI
The traditional approach to dialectology as a field of comparative-historical phonology has had some competition in the recent past. Structuralism represented a new approach to the study of language in that it shifted the focus of investigation to grammar as a synchronic, autonomous system. Perhaps the earliest application of Structuralism to an Italian 'dialect' is Soffietti (1949), but the first truly influential structuralist study of an Italian 'dialect' is Heilmann (1955). Structuralism was adopted by many dialectologists in the 1960s and into the 1970s, when another model was introduced: Sociolinguistics. Sociolinguistics first received widespread attention in Italy with the publication of a special issue of the Rassegna italiana di sociologia (1968) dedicated to this topic, and has been perhaps the most fruitful approach to Italian dialectology in the past few decades. In viewing language as a social phenomenon, it represents an alternative model to the study of language, and it has broadened the field of investigation to include different social dimensions (such as class) and geographic areas (such as urban environments). Studying the phonetic structure of the minor languages has also recently gained favor in many research centers in Italy including Bologna, Padua, Pisa, and Naples, and the results have been significant and promising. The development of Generative Grammar in the second half of the Twentieth Century has revolutionized the way we think about language, as well as about human cognition in general. In Italy, the introduction of Generative Grammar meant that, for the first time, syntax was the focus of intensive study, and the Italian 'dialects' have proved to be a particularly fertile field for these syntactic investigations. The study of the phonology of Italian 'dialects' has played a more subdued role in Generative Linguistics. Beginning with Saltarelli's pioneering work on Italian (Saltarelli 1970) many phonologists have studied Italian phonological processes within the framework of Generative Grammar; however, those working on the phonology of Italian 'dialects' have been, until very recently, few. Why have generative phonologists not ventured into the realm of Italian 'dialect' studies? There are a number of reasons. Phonologists (not specifically trained in Italian or Romance linguistics) may not be aware of the great linguistic variety found in Italy. And until quite recently no introductory manual was available in English. Fortunately, that lacuna is beginning to be filled (Maiden & Parry 1997, etc.). For those phonologists trained in Italian, the traditional approach to the study of the phonology and specifically the comparative-historical phonology of the minor languages of Italy has been so rich and so successful, that most students of Italian phonology are readily drawn to this area of investigation. However, Generative Phonology (and especially recent theories of Generative Phonology, such as Optimality Theory) can be particularly helpful in better
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
5
understanding precisely those areas that the traditional method has focused on, including language change, language variation, and language typology. 3.
Summary of Articles The goal of this book is both theoretical and empirical. First, we aim to provide information to non-Italianists on the phonological structures of the minor languages of Italy. Since most work on Italian dialectology has been written in Italian (and, to a lesser extent, German), most linguists have not even had the opportunity to investigate the riches that Italian 'dialects' hold for them. Second, the reader will find detailed analyses of phonological phenomena that provide important contributions to phonological theory. The richness of the empirical material and the sophistication of the theoretical analyses make this collection a particularly significant addition to the field of phonology. The languages represented in this collection are nearly equally divided between northern and southern Italy, with five papers focusing almost exclusively on northern Gallo-Romance varieties (Baroni & Vanelli, Hajek, Montreuil, Prieto, Repetti), and three on southern varieties (Bullock, Calabrese, D'Introno & Weston), with one paper on a Francoprovençal variety spoken in southern Italy (Nagy), one paper on Tuscan (Marotta), and two papers in which data are gleaned from among many 'dialects' (Loporcaro, Maiden). Conspicuously absent is any detailed discussion of Sicilian and Sardinian (except for the Catalan dialects spoken on the island of Sardinia discussed in Loporcaro). Over half of the articles contain data collected by the authors, and most of the data have not been available in English language publications. Some of the articles included in this volume deal with closely related topics. Syllable weight, and specifically vowel length, continues to occupy the attention of many phonologists, and Hajek, Baroni & Vanelli, and Prieto investigate this aspect of the northern Italian 'dialects'. In his paper "How Many Moras? Overlength and Maximal Moraicity in Italy", Hajek questions the traditional wisdom that syllables are maximally bimoraic, and, therefore, may not be, for example, trimoraic. Early work on moraic structure, including Hayes (1989), noted that many languages (Old English, Farsi, German and Danish dialects, Estonian, etc.) are best described as having trimoraic syllables. By studying phonological patterns and phonetic evidence, Hajek argues that trimoraic syllables can also be found in standard Italian, Friulian, Lombard, and Bolognese. The second paper dealing with syllable structure and vowel length is Baroni & Vanelli's "The Relationship Between Vowel Length and Consonantal Voicing in Friulian". The authors present the results of an acoustic study of Friulian
6
LORI REPETTI
vowels and consonants which show that the relationship between vowel length and consonant voice is more complex than previously assumed (Hualde 1990; Repetti 1992,1994; Vanelli 1979, 1986). They show that the burden of 'contrast' is not due to a single feature (consonant voice, vowel length), but to a group of features, (various degrees of consonant and vowel length, vowel quality, and tonal shape). Prieto ("Vowel Lengthening in Milanese") reexamines Milanese data in which long vowels surface only in word-final stressed syllables, but not in (ante)penultimate stressed syllables or in unstressed syllables. She notes that previous (derivational) accounts of these facts do not explain this particular distribution, and she offers a principled motivation for the fact that vowels are lengthened only in this context. She couches her analysis in Optimality Theory and uses three well-attested constraints to account for the patterns: Foot Binarity (feet should be analyzable as binary), High-Sonority-Nucleus (favoring nuclei of higher sonority over those of lower sonority), and Fill (requiring input structure be respected). Two other papers deal with metrical structure: Repetti's "Uneven or Moraic Trochees? Evidence from Emilian and Romagnol Dialects", and Marotta's "Oxytone Infinitives in the Dialect of Pisa". Repetti examines the status of (HL) trochaic feet. According to one model of trochaic feet (the 'moraic trochee' model) such a foot is not permitted, while according to the 'uneven trochee' model it is. Repetti finds evidence of (HL) feet in Emilian and Romagnol dialects. First, in some dialects phonotactic constraints apply to heterosyllabic consonants in the same foot, i. e., consonants found in the following context: (CVC.CV). A foot with this structure, namely (HL), is an uneven trochee. Second, some words have a metrical template which is part of the input. That template is a disyllabic trochaic foot -- ('σσ) — which may be realized with an (HL) structure. Marotta discusses the aberrant rhythmical pattern of the oxytone infinitives in the Tuscan dialects and their interaction in various phonological processes. She notes that an analysis of these structures and processes is problematic within a theory of metrical structure that includes the Principle of Uniformity (languages necessarily have only one type of foot stucture). She suggests that Pisan can best be analyzed as having left-headed and right-headed feet, binary and ternary feet. By adopting this analysis, Extrametricality and Degenerate Feet are no longer necessary components of the theory, and we can provide a more natural analysis of words with final or antepenultimate stress, and of the metrical structure of the various prosodic levels. The prosodic hierarchy is the topic of investigation of Loporcaro's "Stress Stability under Cliticization and the Prosodie Status of Romance Clitics". He
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
7
begins by noting the three different Romance stress patterns of verb + enclitic pronoun structures: (i) stress is never reassigned when enclitic pronouns are added to the verb, (ii) stress is reassigned only when two or more enclitics follow the verb, (iii) stress is reassigned whenever an enclitic pronoun is added to the verb. Previous accounts of the data are based on the assumption that clitics subcategorize for different prosodic categories (Prosodic Word, Phonological Phrase, or Clitic Group), but Loporcaro argues that there is no conclusive evidence of this. He instead proposes that Romance clitics are invariably adjoined to the Prosodie Word. The author focuses on varieties with stress stability under enclisis, type (i), and discusses segmental rules in these varieties that only apply within the domain of the Prosodie Word: /epenthesis in Algarés, vowel raising in Romanesco and Friulian, vowel insertion and vowel harmony in two northern Italian 'dialects'. By showing that these rules do in fact apply to the verb + enclitic structures, he concludes that the clitics must be part of the Prosodie Word. Therefore, the three different types of stress patterns are not due to the different prosodie status of the clitics, but to a parametric option allowing or forbidding stress to be reassigned postlexically. Montreuil's paper, "Sonority and Derived Clusters in Raeto-Romance and Gallo-Italie", accounts for the complex word-initial onsets found in five 'dialects' of Northern Italy and Eastern Switzerland. Montreuil shows that the properties of derived clusters is different from those of underlying clusters, and he argues that the sonority differential (and not absolute sonority) plays a central role in determining acceptability in clusters and in understanding dialect variation. Three papers deal with the vowel quality of languages spoken in the Apulia region: Nagy ("Stress and Schwa in Faetar"), D'Introno & Weston ("Vowel Alternation, Vowel/Consonant Assimilation and OCP Effects in a Barese Dialect"), and Calabrese ("The Feature [Advanced Tongue Root] and Vowel Fronting in Romance"). In her paper on Faetar (a Francoprovençal dialect spoken in Faeto, in the province of Foggia), Nagy questions the phonemic status of schwa: is it predictable, or is it lexical? She considers various analyses according to which schwa is lexical and rejects them. She also considers the hypothesis that word stress is lexical: if word stress is marked in the lexicon, then the presence of schwa is predictable according to the following rules: stressed vowels are never schwa, pre- and post-tonic vowels may reduce to schwa. Evidence in support of this analysis (and against the other hypotheses) comes from the vowel alternations in verbal paradigms, Raddoppiamento Sintattico (gemination of a word-initial consonant following a word-final stressed vowel), the retention of historically motivated stress assignment, the alternation of vowel and schwa in pretonic
8
LORI REPETTI
position, the absence of non-schwa vowels in post-tonic position, and the restriction of stress to one of the three final syllables. D'Introno & Weston discuss some interesting data from the Barese dialect of Corato in which unstressed vowels may be preserved, reduced to schwa, or changed in other ways: the non-low vowels generally reduce to [a], while the low vowel /a/ is reduced to [a] only when it is phrase final. The labial vowels present further complications: labial vowels adjacent to a labial or a velar consonant are retained, but /o/ is raised to [u] when adjacent to a velar consonant; and labial vowels reduce to [ә] when adjacent to a labial consonant and followed by a stressed back vowel. The authors claim that the behavior of unstressed vowels is determined by the place of articulation of the unstressed vowel, the place of articulation of the consonants adjacent to the unstressed vowel, the place of articulation of the stressed vowel that follows the unstressed vowel, and the position of the unstressed vowel within the phrase, and they provide a feature geometric account of the data. Calabrese argues that the feature [ATR1 plays a role in the vowel systems of Romance languages, and shows how this feature allows us to account for vowel fronting in the Barese dialect of Altamura. The vowels /o/ and /u/ (both [+ATR]) are fronted when in stressed syllables which are closed or in antepenultimate position. We do notfindfronting in the following situations: the vowel / / is never fronted (because it is [-ATR]); no fronting takes place in open penultimate syllables (because the vowel is diphthongized); /o/ and /u/ do not front after a labial consonant, and /u/ does not front following a velar consonant (because of a secondary labiovelar articulation of the consonant). Calabrese analyzes the motivation for these processes within the constraint-and-repair model of phonology (Calabrese 1995, forthcoming). It is interesting that the data analyzed in D'Introno & Weston and Calabrese are perfectly complementary. For example, in Corato unstressed /o/ and /u/ adjacent to a labial consonant do not reduce to schwa (as expected), but are retained. In Altamura stressed /o/ and /u/ adjacent to a labial consonant do not front (as expected), but are retained. While the theoretical models adopted are different, the two analyses of these patterns are similar in that they both rely on the notion (i) that a particular feature can be doubly linked to two positions, and (ii) that a doubly linked structure is resistant to certain changes. Bullock ("Consonantal Gemination in Neapolitan") deals with a metrical phenomenon in Neapolitan which occurs at the intersection of syntax, morphology, and phonology: Raddoppiamento Sintattico. While standard Italian Raddoppiamento Sintattico is, in some cases, phonologically predictable (word initial consonants are geminated after words with final stress), Neapolitan
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
9
Raddoppiamento Sintattico is lexicaiiy determined: it is triggered by a class of function words that display a unique 'double' prosodic structure. Initial gemination is the compromise between two competing constraints: (i) that certain syllables be bimoraic, and (ii) that final vowels remain short. Bullock's analysis provides evidence of the notion of 'double prosody', and supports Selkirk's (1995) hypothesis that function words can have unique phonological properties. In Neapolitan, for example, lexical words and function words parsed as monomoraic syllables can produce different phonological effects. Interestingly, both Bullock and Repetti independently come to the same conclusion: that certain aspects of a language's prosodic structure (syllable weight and foot structure, respectively) can be represented as abstract phonological properties. Finally, Maiden analyzes the complex and often ignored topic of phonological dissimilation in his article, "Phonological Dissimilation and Clitic Morphology in Italo-Romance". He shows that, diachronically, the existence of morphological and lexical variants is eliminated when speakers select the phonologically 'better' variant, such as the one which does not have successive identical segments. Crucially, the motivation for the variants themselves is not phonological. For example, the sequence of indirect object clitic /le/ + direct object clitic /lo/ results not in the expected **[le lo], but in the following forms in various Italian 'dialects': [λe lo], [t e lo], [ne lo], [ve lo], [se lo]. The clitic variants [λe], [t e], [ne], [ve], [se] are existing pronouns which are selected over the nondissimilated form [le]. This analysis suggests a close tie between morphological variation and a general phonological preference for dissimilated structures. 4
Conclusion The minor Romance languages spoken in Italy provide a wealth of resources for the linguist in general and the phonologist in particular. They are languages whose history is well documented and well studied, which exhibit a vast degree of variation concentrated in a restricted geographical area, and whose resources (native speakers) are still largely untapped. The articles contained in this collection describe complex and poorly understood and under-studied phonological phenomena attested in these languages, they provide a sophisticated analysis of the phenomena, and in doing so they help to refine our understanding of Romance phonology as well as the phonological component of grammar. I am hopeful that these papers will inspire further discussion and study of Italy's most precious resource and best kept secret: its linguistic heritage.
10
LORI REPETTI
REFERENCES Ascoli, Graziadio I. 1873. "Saggi ladini". Archivio glottologico italiano 1.1-556. Belletti, Adriana. 1993. "Syntactic Theory and the Dialects of Italy". Syntactic Theory and the Dialects of Italy ed. by Adriana Belletti, 7-23. Torino: Rosenberg & Sellier. Benincà, Paola. 1996. Piccola storia ragionata della dialettologia italiana. Padova: Unipress. Calabrese, Andrea. 1995. "A Constraint-Based Theory of Phonological Markedness and Simplification Procedures". Linguistic Inquiry 26.373-463. . forthcoming. "Markedness and Economy in a Derivational Model of Phonology". Ms., University of Connecticut. Castellani, Arrigo. 1982. "Quanti erano gl'italofoni nel 1861?". Studi linguistici italiani 8.3-26. Cortelazzo, Manlio. 1980. Avviamento critico allo studio della dialettologia italiana: I: Problemi e metodi. Pisa: Pacini. De Mauro, Tullio. 1970. Soria linguistica delVItalia unita. Bari: Laterza. DOXA. 1988. "Pariare in dialetto". Bollettino della DOXA. 42.55-62. Grassi, Corrado, Alberto A. Sobrero, & Tullio Telmon. 1997. Fondarnenti di dialettologia italiana. Roma & Bari: Laterza Hayes, Bruce. 1989. "Compensatory Lengthening in Moraic Phonology". Linguistic Inquiry 20.253-306. Heilmann, Luigi. 1955. La parlata di Moena nei suoi rapporti con Fiemme e con Fassa. Bologna: Zanichelli. Hualde, José I. 1990. "Compensatory Lengthening in Friulian". Probus 2.31-46. Maiden, Martin & Mair Parry, eds. 1997. The Dialects of Italy. London & New York: Routledge. Repetti, Lori. 1992. "Vowel Length in Northern Italian Dialects". Probus 4.155182. . 1994. "Degenerate Syllables in Friulian". Linguistic Inquiry 25.186-193. Saltarelli, Mario. 1970. A Phonology of Italian in a Generative Grammar. The Hague: Mouton. Selkirk, Elisabeth. 1995. "The Prosodic Structure of Function Words". University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory ed. by Jill Beckman, Laura Walsh Dickey, & Suzanne Urbanczyk, 439-471. Amherst, Mass.: Graduate Linguistic Student Association.. Soffietti, James P. 1949. Phonemic Analysis of the Word in Turinese. New York.
PHONOLOGICAL THEORY AND THE DIALECTS OF ITALY
11
Vanelli, Laura. 1979. "L'allungamento delle vocali in friuiano". Ce fastu? 55.6676. ——. 1986. "La fonologia dei prestiti in friuiano". Raetia antiqua et moderna. ed. by Gunter Holtus & K. Ringger, 355-376. Tübingen: Niemeyer.
THE RELATIONSHIP BETWEEN VOWEL LENGTH AND CONSONANTAL VOICING IN FRIULIAN*
MARCO BARONI University of California, Los Angeles LAURA VANELLI University of Padua
0.
Introduction In Friulian (a Romance variety spoken in northeastern Italy), there is a systematic relationship between vowel length and consonant voicing: stressed vowels are lengthened before phonemically voiced obstruents that undergo word-final devoicing. The aim of this paper is to review the theoretical approaches that have been proposed in order to account for Friulian vowel lengthening, and to present new empirical evidence that should be taken into account in future research on this phenomenon. In §1, we present the basic data on the distribution of long vowels and on the relationship between vowel length and consonantal voicing. In §2, we present and discuss the accounts of the Friulian patterns proposed by Vanelli (1979), (1986), Hualde (1990) and Repetti (1992), (1994). In §3, we present the new evidence provided by Vanelli (1986), (1989), Yamamoto (1993) and Bais (1997), which shows how the relationship between vowel length and consonant voicing is a synchronic feature of Friulian grammar. In §4, we present a preliminary acoustic investigation of vowels and consonants in the relevant contexts, and we discuss the implications of the results of this investigation for the theoretical analysis of the Friulian patterns. We conclude with some suggestions for possible directions for future research.
Although this paper is the result of a joint project, §0-§3 were written by Laura Vanelli, and §4 by Marco Baroni. An earlier, shorter version of this paper appeared as Baroni & Vanelli (1999).
MARCO BARONI & LAURA VANELLI
14
1. 1.1
The data The Distribution of Long Vowels in Friulian In Friulian 1 , phonologically long vowels can only occur in stressed wordfinal syllables closed by a single consonant. (1)
[Vi:f] [can'ta: t] [go'lo:s] ['fre:t] ['ri:t] [tat]
'alive (m.)' 'sung (m.)' 'greedy (m.)' 'cold (m.)' '(he/she) laughs' 'row (m.)'
Stressed vowels are always phonologically short when they are not in the last syllable of a word (2a), 2 when they occur in the final open syllable (2b), 3 and when they are in the final syllable closed by a consonant cluster (2c). (2)
a
[Vive] [can'tade] [go'loze] [Trede] ['ridi] ['krude] ['triste] ['calde] *['vi:ve]
'alive (f.)' 'sung (f.)' 'greedy (f.)' 'cold (f.)' 'to laugh' 'row (f.)' 'bad (f.)' 'warm (f.)'
1 All the data presented in this paper are from Central Friulian, the most widespread variety. For a survey of Friulian, see Vanelh (1997). 2 The acoustic analysis presented in §4 shows that vowels are slightly lengthened before voiced obstruents in any position. Furthermore, in some varieties long word-internal vowels are attested before sonorants in words such as ['ne:re] 'black (f.)', ['pa:ri] 'father' and [Vo:li] 'eye'. In these cases, the vowel was followed, in an earlier stage of the language, by a consonantal cluster. Thus, the long vowels are probably due to some form of compensatory lengthening. 3 In some varieties, there are some forms with long word- final vowels. Mainly, these are verbal infinitives, such as [can'ta:] 'to sing', [ta'ze:] 'to be silent', [dur'mi:] 'to sleep'. As in the exceptional cases discussed in the previous footnote, this is probably a case of compensatory lengthening, since in all these forms the vowel was followed by a coda [r] during an earlier stage of the language.
VOWEL LENGTH AND CONSONANTAL VOICE b.
c.
[ku'si] [du'la]
'so' 'where'
[a'mi] [das'po] *[ku'si:] ['gust] ['bosk] ['calt] *['gu:st]
'friend (m.)' 'after'
15
'taste' 'forest' 'warm (m.)'
Furthermore, vowels are always short when followed by nasals 4 (3a) or affricates (3b). (3)
a
b.
['ma )]
'hand'
['bo ]
'good(m.)'
['prig] ['fur}] ['a ] *['ma: ]
'first (m.)' 'smoke' 'year'
['mjet ] [brat ] ['pots] *['mje:t ]
'half 'arm' 'well'
Finally, stressed vowels are always long when followed by [r]. (4)
['ca:r] ['spc:rk] [Vs:rt]
'dear (m.), cart, meat' 'dirty (m.)' 'green (m.)'
['la:rk] *[car]
'large (m.)'
1.2
The Relationship Between Vowel Length and Consonantal Voicing Vowel length is distinctive in word-final syllables closed by single obstruents or by [1]. In this context, both stressed long (5a) and short (5b) vowels can occur.
4
In final position only velar nasals are allowed: [ ] alternates with labial, alveolar or palatal nasals in word-internal position: [mani'a] 'to handle', [fu'ma] 'to smoke', [le' o:s] 'wooden'.
16
MARCO BARONI & LAURA VANELLI
(5)
a
b.
['la:t] ['bru:t]
'gone (m.)' 'broth, mother-in-law'
[fi'ni:t] [a'ze:t] ['fu:k]
'finished (m.)' 'vinegar' 'fire'
[lu:s] [Vo:s]
light' 'voice'
['pe:s] ['pa:s] ['ne:f] ['vi:f]
'weight' 'peace' 'snow' 'alive (m.)'
['lo:f]
'wolf
[Va:l]
'(it) is worth'
['mi:l] 'honey' [lat] 'milk' [brut] 'ugly' ['frit] 'fried (m.)' [met] '(he/she) puts' [tok] 'piece' ['ssk] 'skinny (m.)' ['trop] 'how much?' ['mus] 'donkey' fros] 'red (m.)' [lus] luxury' ['pes] 'fish' ['pas] 'step' [pa'taf] 'slap' [zgar'duf] 'tuft' [Val] 'valley' ['mil] 'one thousand'
Notice the minimal pairs: ['la:t] vs. ['lat], ['bru:t] vs. ['brut], ['lu:s] vs.[lus], ['pa:s] vs. ['pas], [Va:l] vs. [Val], ['mi:1] vs. ['mil]. Vowel length in word-final syllables before obstruents is predictable on the basis of the following generalizations. (i) If a morpheme displays a vowel followed by a voiced obstruent when it occurs word-internally, the same morpheme will display a long vowel followed by a voiceless obstruent when it occurs in a word-final stressed syllable (6a). (Word-final obstruents are always voiceless in Friulian.) (ii) If a morpheme displays a vowel followed by a
VOWEL LENGTH AND CONSONANTAL VOICE
17
voiceless obstruent when it occurs word-internally, the same morpheme will display a short vowel followed by a voiceless obstruent when it occurs wordfinally (6b). (6)
a
Part] / ['lade] [fi'ni:t] / [fi'nide] ['pe:s] / [pe'za] ['ne:f] / [neve'a] [Vi:f] / [Vive] ['lo:f] / ['love] ['lu:s] / [lu'zo:r] ['fu:k] / [fogo'la:r] flat] / [la'ta] ['mat] / ['mate] ['brut] / ['brute] ['pas] / [pa'sa] ['met] / ['msti] ['ssk] / ['ssce] ['ros] / ['rose] [pa'taf] / [pata'fa] [tck] / [tu'kut]
'gone (m.)/(f)' 'finished (m.) / (f.)' 'weight / to weight' 'snow / to snow' 'alive (m.) / (f.)' 'wolf (m.) /(f)' 'light/diffuse light' 'fire / fire place' 'milk / to breast-feed' 'crazy (m.) / (f.) 'ugly (m.) /(f)' 'pass / to pass' '(he/she) puts / to put' 'skinny (m.) / (f.)' 'red (m.) / (f.)' 'slap / to slap' 'piece / little piece'
Following previous authors who have worked on Friulian vowel length, we will concentrate on the systematic pattern presented in (6). 2.
Approaches to Friulian vowel length Three authors have proposed synchronic accounts of the distribution of long and short vowels in Friulian. For Vanelli (1979), (1986) and Hualde (1990), the long vowels in (6) are the result of a synchronic vowel lengthening rule that lengthens vowels before word-final devoiced obstruents; while for Repetti (1992), (1994) long and short vowels are lexically represented as such, and, synchronically, the relationship between vowel length and consonantal voicing is not a straightforward one. 2.1
Vanelli (1979), (1986) and Hualde (1990) Since vowel length before obstruents is entirely predictable from morphophonological alternations, as shown in (6), Vanelli hypothesizes that all vowels are underlyingly short. Consider, for example, the following paradigms. (Final Id is the feminine marker, whereas masculine forms are bare roots.)
18
MARCO BARONI & LAURA VANELLI
(7)
['la:t] / [lade] ['mat] / ['mate]
'gone (m.) / (f.)' 'crazy (m.) / (f.)'
Vanelli argues that the underlying representation of the first root is /'lad/, with a short vowel and a voiced consonant. The underlying representation of the second root is identical to its surface form: /'mat/. When the final consonant of /'lad/ occurs word-finally, it is devoiced. However, the loss of the obstruent voicing contrast in word-final position is compensated by the creation, on the surface, of a vowel length contrast. To achieve this, Vanelli postulates a vowel lengthening rule, triggered by word-final voiced consonants. The context of the rule is subsequently canceled by the final devoicing rule. (In Vanelli 1979 these rules are stated in an informal way; for a formal account in the framework of 'classical' generative phonology, see §3.4.) Following an observation in Vanelli (1979), who claimed that vowel lengthening is a sort of compensation for the loss of voicing of the consonant, Hualde (1990) proposes to treat the Friulian vowel lengthening pattern as a form of compensatory lengthening. Hualde, who works within the framework of 'moraic phonology' (Hayes 1989), proposes that in contemporary Friulian only voiced segments in a rhyme can be mora-bearing units. Thus, when a word-final coda obstruent is devoiced, it cannot bear its mora anymore, and the mora is associated to the preceding vowel. In (8), we show how the /'lad/ -> ['la:t] derivation is represented within Hualde's formalism. (8)
1
a
d
| JA
l |
JA
a
t
I JA
l a t /\
JA
JA JA
Of course, compensatory lengthening does not concern the /'mat/ > ['mat] derivation, in which the voiceless obstruent is not moraic. Notice that Hualde's solution, like the one suggested by Vanelli, assumes rule ordering. In particular, the mora assigned to the word-final voiced obstruent cannot be lexically specified, since, in the framework of moraic phonology, underlying moras associated to consonants mark phonemically long (geminate) consonants (Hayes 1989), and the word-final consonants of Friulian are not phonemically long. The mora associated to word-final voiced obstruents must be assigned by some syllabification algorithm. Subsequently, word-final devoicing applies, and the syllabification algorithm applies again, this time linking the mora of the devoiced final obstruent to the preceding vowel. Thus, Hualde must assume an intermediate level of representation, since the
VOWEL LENGTH AND CONSONANTAL VOICE
19
assignment of a mora to word-final voiced obstruents cannot take place in the underlying representation (because these consonants are not phonologically long), and it cannot take place on the surface, since the word-final obstruents are not voiced on the surface. Hualde's solution is elegant and empirically sound. However, since there is no independent evidence that only voiced codas are moraic in Friulian, this assumption is completely ad hoc. Furthermore, if moras are timing units, as usually assumed (Kenstowicz 1994:428), then Hualde's treatment looks particularly problematic since voiced obstruents are typically shorter than voiceless obstruents, and, as we will see in §4, Friulian voiced obstruents are not an exception to this general trend. Thus, it seems that voiceless obstruents should be represented as moraic. Both Vanelli and Hualde treat vowel length before obstruents as a surface feature depending on the voicing of the following consonant. Thus, both accounts must explain why there is no vowel lengthening in paradigms such as the one in (9). (As we stated in §1, long vowels are never allowed before affricates.) (9)
['mjet ], *['mje:t ] / ['mjed3e] 'half (m.) / (f.)'
Since affricates are complex segments, they tend to be longer 'by nature' Thus, the fact that lengthening does not take place before affricates can be explained in terms of a non-ad hoc constraint prohibiting extra-long syllables/rhymes, and does not pose a major challenge to Vanelli's and Hualde's accounts. Vanelli and Hualde also have to assume that vowel length before wordfinal obstruents and vowel length before word-final /1/ are different phenomena, since the former is an allophonic property, whereas the latter is a phonemic, lexically-specified feature. As the following pairs of paradigms shows, vowel length before /1/, unlike vowel length before obstruents, is not predictable and is present in the underlying representation. (10)
/Va:l/'(it) is worth' /'mi:l/ 'honey' /Val/ 'valley' /'mil/ 'one thousand'
2.2 Repetti (1992), (1994) For Repetti (1992), all long vowels are not derived by rule, but specified in the lexicon. Thus, for Repetti the underlying representations of ['la:t] and ['mat] are /'la:d/ and /'mat/, respectively. The fact that underlying long vowels
20
MARCO BARONI & LAURA VANELLI
surface as short when they occur word-internally (['lade], *['la:de]) is due to a metrical vowel shortening rule, which only allows long vowels in final stressed syllables. Since Repetti does not postulate vowel lengthening, she does not need to state a special constraint to handle the absence of long vowels before affricates: ['mjet ], *['mje:t ] / ['mjed3e]. Furthermore, in Repetti's treatment long vowels before obstruents and long vowels before /1/ have the same status, i. e., they are both lexically specified. However, if vowel length is a phonemic property, then it is not clear why long vowels only occur before obstruents that are underlyingly voiced, i. e. why an underlying form such as */la:t/ is not allowed (*[Ta:t] / [late]). Repetti (1994) proposes a solution for this problem, based on some assumptions on the possible syllabic and moraic structures of Friulian. Repetti argues that Friulian only allows mono- and bimoraic syllables. Word-final CVCC and CVVC structures are treated as bimoraic syllables followed by a 'degenerate' syllable (following the analysis of the Arabic dialects presented by McCarthy & Prince 1990 and Broselow 1992). The degenerate syllable dominates a consonant and a mora. Thus, CVCC and CVVC syllables have the following structures.
In order to account for the ban against underlying forms like */'la:t/, Repetti postulates that only sonorants and voiced obstruents can project degenerate syllables. Thus, */la:t/ is not a legitimate structure because Friulian does not tolerate trimoraic syllables (hence, the syllabification in (12a) is impossible), and a voiceless consonant cannot project a degenerate syllable (hence, the syllabification in (12b) is impossible).
1
a
t
VOWEL LENGTH AND CONSONANTAL VOICE
21
Repetti has to explain why word-final CVCC sequences with an (underlyingly) voiceless obstruent as the last segment are well-formed, as in the following examples. (The feminine forms show that the last segment of these roots is underlyingly voiceless.) (13)
['trist] ['fwart] ['alt]
'sad (m.)' 'strong (m.)! 'tall(m.)'
['triste] ['fwarte] ['alte]
'sad (f.)' 'strong (f.)' 'tall (f.)'
Given that trimoraic syllables are not acceptable, and that voiceless obstruents cannot project a degenerate syllable, a form like /'trist/ should not be grammatical. Repetti explains the grammaticality of the forms in (13) by assuming that, in these cases, the final obstruent is adjoined to the second mora of the bimoraic syllable early in the derivation, i.e., to the mora of the preceding consonant. (14)
a
/\ μμ
! A a 1 t
Given this assumption, one may wonder why a structure like */la:t/ cannot be parsed with the final consonant sharing the second mora of the vowel. Repetti rules out this possibility by assuming that a single mora cannot dominate both a vowel and a consonant. (15) *o
1
a
t
Notice that Repetti's condition against voiceless segments in degenerate syllables does not hold at the surface level, since forms like /'la:d/ can be realized as ['la:t]. Thus, Repetti's solution is also derivational. Repetti must postulate an intermediate level, after syllabification but before devoicing, at which the condition against degenerate syllables projected by voiceless segments is checked.
22
MARCO BARONI & LAURA VANELLI
While Repetti's solution works from an empirical point of view (but see §3), it is based on the following set of ad hoc assumptions which are not independently motivated by the analysis of other aspects of Friulian and/or by general phonetic properties of language: (i) some word-final consonants are not parsed as codas of the last syllable of a word, but as 'degenerate' syllables; (ii) only voiced obstruents and sonorants can form degenerate syllables; (iii) wordfinal voiceless obstruents can share the mora of the preceding consonant; (iv) mora-sharing is only possible between consonants. 2.3
Conclusions on the Reviewed Approaches to Vowel Length Vanelli's traditional approach, which is not based on an autosegmental representation of length, appears to be the only one that derives the Friulian pattern from a phonetically grounded and cross-linguistically common rule, namely vowel lengthening before voiced consonants (all three authors assume word-final devoicing, another phonetically grounded and cross-linguistically common rule). Differences in duration between vowels before voiced vs. voiceless obstruents are probably related to the fact that voiced obstruents are shorter than voiceless obstruents - thus, for some isochrony principle similar to open syllable lengthening / closed syllable shortening, vowels before the short voiced segments tend to be longer than before the longer voiceless segments (Kluender, Diehl, Wright 1988). This pattern, while not universal, is crosslinguistically very common (Keating 1985). On the other hand, Hualde's and Repetti's accounts are based on assumptions that are not grounded in typology nor in phonetics, and which do not find support in other areas of Friulian phonology. Probably, in this case the autosegmental approaches are less successful than a more traditional approach because in an autosegmental framework, such as the one of moraic phonology, length is represented as a feature independent from segmental features such as voicing. Thus, autosegmental approaches to length are well-suited to represent processes in which phonological length behaves as an independent suprasegmental dimension (such as standard cases of compensatory lengthening), but not processes, like lengthening before voiced obstruents, in which length interacts with segmental features, such as voicing. In the next section, we will show that, against Repetti's claim that vowel length and voicing are independent properties of synchronic Friulian grammar, these two features are still perceived as related by contemporary speakers.
23
VOWEL LENGTH AND CONSONANTAL VOICE
3.
The Relationship Between Vowel Lengthening and Voicing in Synchronic Friulian: Empirical Evidence 3.1 Loanwords from Italian The first two arguments in favor of a synchronic correlation between vowel length and consonant voicing come from the analysis of loanwords from Italian (Vanelli 1986, 1989). In Italian, unlike in Friulian, consonantal length is distinctive, and vowel length is allophonic. Following a common crosslinguistic pattern (see Maddieson 1985), stressed vowels are longer in open syllables and shorter in closed syllables, and, consequently, before geminates (see the data in Farnetani & Kori 1986). Thus, we have pairs such as the ones in (16).5 (16)
['pa la] vs. ['fa to] vs.
['pal:a] ['fat:o]
'shovel' vs. 'ball' 'fate'vs.'fact'
When Friulian borrows Italian words, the consonantal length contrast is lost. However, the shorter vowels that occur before geminates in Italian are treated as short Friulian vowels; the longer (stressed) vowels that occur before single consonants in Italian are treated as long Friulian vowels, if they occur in word-final position: [impje'ga:t], from Italian [impje'gato]. Thus, the allophonic vowel length contrast of Italian is restructured as a phonological vowel length contrast. Interestingly, when the new Friulian long vowels created in this way occur in word-internal position, they become short, but the obstruent following them is voiced (17a) (in word-final position, all obstruents are voiceless, because of the systematic and productive process of word-final devoicing). Instead, when borrowed short vowels occur in word-internal position, the obstruents following them stay voiceless (17b). (17)
a
b.
[impje'ga:t] / [impje'gade] [deli'ka:t] / [deli'kade] [bandi:t] / [bandi'dut] [ste'ka:t] / [steka'dut] [a'fit] / [afi'tut] [kon'trat] / [kontra'tut] [stan'tufl / [stantu'fut]
'clerk (m.)/(f.)' 'delicate (m.)/(f.)' 'outlaw / little outlaw' 'fence / little fence' 'rent / little rent' 'contract / little contract' 'piston / little piston'
It. [impje'ga-to] It. [deli'ka-to] It. [ban'di-to] It. [ste'k:aTo] It. [a'f:it:o] It. [kon'tratro] It. [stan'tuf:o]
The symbol "•" after a vowel or consonant stands for the IPA 'semi-long' diacritic.
24
MARCO BARONI & LAURA VANELLI
Notice that in the cases in (17a) the Italian forms display voiceless stops. Hence, voicing is not a property inherited from the Italian forms. Furthermore, notice that intervocalic voicing is no longer a productive process of Friulian. Indeed, as the following examples show, Italian intervocalic obstruents that are not preceded by long vowels do not undergo voicing when imported into Friulian. (18)
[kapi'ta:l] 'capital' [gra'matike] 'grammar'
It. [kapi'ta-le] It. [gra'm:a-tika]
Vanelli (1986) interprets the voicing in (17a) as a case of 'rule inversion'. Since vowel length is not an underlying property of Friulian, the surface long vowel + word-final voiceless obstruent sequences borrowed from Italian are reinterpreted by Friulian speakers as underlying short vowel + voiced obstruent sequences (and, consequently, these are the structures that surface in wordinternal position). Independently from the theory proposed to account for the relevant patterns, the treatment of loanwords illustrated in (17a) constitutes evidence that contemporary speakers are aware of the relationship between vowel length and voicing. 3.2
More Evidence from Loanwords: Verbs with Voiced Affricates Yamamoto (1993) presents fieldwork data on the borrowing of Italian verbs with an infinitival form ending in -(g)gere [-dj(:)ere] in several Friulian dialects. (This ending becomes -gi [dji] in Friulian). Yamamoto is interested in what happens to the stressed vowel before /dj/ in the forms with zero suffixation, in which the voiced affricate occurs word-finally and, consequently, is devoiced. Remember from §1 that in native forms long vowels can never occur before affricates. Yamamoto found that, while in some varieties (the Enemonzo and Tolmezzo dialects) the loanwords, like the native words with affricates, are not subject to lengthening, while in many other varieties (the Qualso, Buia, Tarcento, Artegna and Taboga (Gemona) dialects) the borrowed verbs are subject to vowel lengthening in the forms with zero suffixation. (19)
[ko'redji] / [ko're:tJ] [dis'tru dji] / [dis'tru:t
'to correct / (s)he corrects' 'to destroy / (s)he destroys'
It. [ko'r:edj:ere] It. [dis'trudj:ere]
VOWEL LENGTH AND CONSONANTAL VOICE
25
Yamamoto interprets his data in terms of the conflict between two characteristics of Friulian: the phonological well-formedness constraint which bans long vowels before affricates, and the morphophonological pattern that requires lengthening before word-final devoiced obstruents. The latter tendency wins in the borrowing of verbal forms in the dialects with the paradigms of (19). Notice that, in this case, the Italian forms have short stressed vowels (these vowels are short because they occur in closed syllables). Thus, the long vowels in the examples in (19) must be the product of the Friulian rule of vowel lengthening before a voiced consonant. If there was no synchronic relationship between vowel length and final devoicing, similar cases should not exist. 3.3
Reading of Non-Words The third argument in favor of a synchronic correlation between vowel lengthening and voicing comes from a non-word reading test conducted by Bais (1997). Bais asked three Friulian native speakers to read two sets of non-words. The first set (150 items) had sequences such as the ones in (20a), with a stressed vowel followed by a word-final voiceless obstruent, corresponding to possible, although non-existing, Friulian words. The second set (153 items) had sequences such as the ones in (20b), with a stressed vowel followed by a wordfinal voiced obstruent — an ill-formed structure in Friulian, because it violates final devoicing. The forms were presented in standard Italian orthography, and thus vowel length was not marked in the written forms. (20)
a b.
klat, ta'lut, praf, ri'luk kuv, li'nud, tu'ped, prod
Notice that in Italian words with final obstruents are extremely rare, and, with the exception of some words ending in /s/, they are recent loanwords. Thus, it was reasonable to expect that, even if Italian is the typical language of written communication for Friulian speakers, the subjects would have treated the nonwords as Friulian forms, rather than as Italian forms. For the first set (words such as the ones in (20a)), since both long and short vowels can occur in Friulian before word-final (surface) voiceless obstruents, and since vowel length was not marked on the orthographic vowels, the expectation was that speakers would oscillate between producing them with long and short vowels. The results confirm this expectation. The subject R.M.G. produced the words of the first set with a long vowel 54% of the times and with a short vowel 45% of the times; F.M. produced them with a long vowel 65% of the times and with a short vowel 35% of the times; J.F. produced them with a
26
MARCO BARONI & LAURA VANELLI
long vowel 73% of the times and with a short vowel 27% of the times. Considering the responses of all the subjects together, the non-words in the first set were produced with a long vowel 61% of the times, with a short vowel 39% of the times. 6 Let us now consider the case of sequences ending in a voiced consonant. 23% of the stimuli were produced in the way in which they were orthographically presented, i. e., with a short vowel and a voiced consonant. This is not surprising since subjects were simply asked to read what was presented. 17% of the stimuli were produced with a voiced final consonant, but with a long vowel, i. e., the subjects applied the spontaneous lengthening process which occurs before voiced consonants. However, the most interesting results are found when subjects devoiced the final consonant. This happened with 60% of the stimuli. Once devoicing takes place, one would expect the distribution of vowel length to be similar to the one we saw for the words with a voiceless consonant: a long vowel in 61% of the cases, and a short vowel in 39% of the cases. Instead, the results were entirely different. The forms with a long vowel were by far the most common, with an average of 94%, whereas in only 6% of the cases was the vowel short. These data provide further support for the hypothesis that there is a significant correlation between final consonant devoicing and the occurrence of a long vowel. 3.4
Conclusions on the Empirical Evidence about the Synchronic Relationship Between Vowel Length and Consonant Voice The data presented in this section show that lengthening before devoiced final consonants is a synchronically active property of Friulian. The data are problematic for Repetti's account. In Repetti's framework, a short vowel before a word-final devoiced obstruent is a well-formed structure. Thus, it is not clear why Friulian speakers should lengthen short Italian vowels in borrowed forms such as the ones in (19), when these vowels occur before word-final devoiced obstruents. Similarly, it is not clear why Friulian speakers reading non-words such as the ones in (20b) would lengthen the vowel whenever they devoice the following obstruent.7 6
It is interesting that all the subjects show a preference for long vowels, and that this preference appears to be inversely correlated with the degree of education of the readers. J.F., the less educated reader, is the one who produced the highest number of long vowel responses. 'The data on voicing of consonants after long borrowed vowels discussed in §3.1 are not problematic for Repetti, because she argues that, when Friulian speakers produce word-final V:C sequences, they must postulate that the final obstruent is underlyingly voiced, since a final voiceless consonant cannot be parsed as a degenerate syllable, and it cannot form a trimoraic syllable with the preceding long vowel.
VOWEL LENGTH AND CONSONANTAL VOICE
27
3.5
Formalizing the Vowel Lengthening Rule In the preceding paragraphs we showed that: (1) the presence of long and short vowels is predictable when they are followed by an obstruent, and (2) the relevant context for a long vowel is the presence of a underlying voiced obstruent, which is devoiced in word-final position. We can account for the occurrence of long vowels assuming the interaction of two rules, vowel lengthening and final devoicing (as suggested in Vanelli 1979), which can be formalized as follows within the framework of 'classical' generative phonology. (21)
a. Rules: Vowel lengthening
Final devoicing
V
C [-snt]
V:/
C ## [+voi] [-snt] [voi] / ##
b. Examples : Underlying Representations :
/'lad/
/'lad+e/
Vowel lengthening:
/'la:d/
n/a
Fnal devoicing: Output:
/'la:t/ [la:t]
n/a
Underlying Representations:
/'mat/ n/a
/'mat+e/ n/a n/a
Vowel lengthening: Final devoicing: Output:
n/a ['mat]
['lade]
['mate]
Notice the necessity of rule ordering: final devoicing makes the final consonants identical; hence vowel lengthening, a rule sensitive to the voiced/voiceless distinction, must apply before final devoicing. Since the devoiced and voiceless obstruents are identical on the surface, vowel lengthening cannot be stated as a surface rule/constraint. Adopting the terms of traditional generative phonology, we can state that vowel lengthening and final devoicing stand in a counterbleeding relation and that vowel lengthening is an opaque rule, since its triggering context is canceled by the final devoicing rule (Kiparsky 1968, 1971; Kisseberth 1973). The two postulated processes, i. e., vowel lengthening before voiced consonants and word-final devoicing are phonetically grounded and cross-linguistically common (see Keating 1985 and Westbury & Keating 1986, respectively).
28
MARCO BARONI & LAURA VANELLI
However, we realize that the formalization proposed above presents an aspect which is not completely satisfactory. The structural description of the lengthening rule requires only a final voiced obstruent as the context which triggers the lengthening process. The fact that this consonant is voiceless on the surface is derived independently by the application of another rule, i. e., final devoicing, which is not formally related to the lengthening rule. On the contrary, it would be better to relate both processes, to account for the fact that long vowels are found not simply before phonologically voiced consonants (voiceless on the surface form). What is relevant is that consonants are devoiced, i. e., phonetically voiceless and phonologically voiced. But it seems that, on the surface form, voiceless and devoiced obstruents are identical and completely neutralized. This means that the lengthening rule must be 'projected' onto an intermediate, phonological level, where we can still distinguish between /lad/ and /'lat/. But in this case the lengthening rule is opaque, as we said before, and is not sensitive to its phonetic context, as we would like. For these reasons, since the previous work on Friulian vowel lengthening has been based on impressionistic data, we have been induced to conduct an acoustic investigation of vowels and consonants in the relevant contexts. In particular, we conducted an acoustic analysis of the stressed vowel + obstruent sequences in word-internal and word-final position. We wanted to obtain a nonimpressionistic description of the phonetic features that characterize the contrast between paradigms such as ['la:t]/['lade] vs. ['mat]/['mate]. In particular, we wanted to assess, to the extent possible in an acoustic analysis, the validity of the impressionistic observation that voiceless and devoiced obstruents in wordfinal position are completely neutralized, i. e., identical. If this was not the case, it would not be necessary to assume that vowel lengthening and final devoicing are ordered rules. If voiceless and devoiced obstruents are not identical, then vowel lengthening can be stated as a surface rule/constraint. 4, 4.1
Acoustic Analysis of Long Vowel + Obstruent Sequences in Friulian The Materials We prepared a list of nonsense forms in which each Friulian stressed vowel (/i e a o u/) occurred before an alveolar stop (It d/). The nonsense forms were all possible Friulian adjectives from a morphophonological point of view, and they were coupled in pairs of 'masculine' and 'feminine' forms. Each vowel occurred in two pairs: in one pair, the vowel was followed by /d/ plus the suffix /-e/ in the feminine, and by word-final /D/ in the masculine, which corresponds to the bare form of the adjective (ID/ = word-final devoiced /d/); in the other
VOWEL LENGTH AND CONSONANTAL VOICE
29
pair, the vowel was followed by /t/ plus the suffix /-e/ in the feminine, and by word-final /t/ in the masculine. In (22) we present an example consisting of the forms with the stressed vowel /a/. (In parenthesis we present the nonsense forms transcribed in Friulian orthography, as they were presented to the speaker). (Cf. lade / lât = 'gone (f) / (m)'; mate / mat = 'crazy (f) / (m)'.) (22)
feminine a'tade (atade)
masculine a'ta:D (atât)
a'tate (atate)
a'tat (atat)
Notice that in Friulian orthography both /D/ and /t/ are transcribed as
; thus, the fact that the Friulian speaker, as we will see, did not produce a complete neutralization between /D/ and /t/ cannot be attributed to an attempt to preserve an orthographic contrast. The length of the vowel before /D/ is marked with a circumflex accent (^). All the vowels were inserted in the same "at T(e)" skeleton presented in (22) and each 'pseudo-adjective' was inserted in one of the carrier sentences in (23): the feminine forms in the carrier sentence (23a), the masculine forms in the carrier sentence in (23b). (23)
4.2
a
A
b.
to her DAT-cl (I) tell 'I tell her ___ today as well.' A lui i dîs to him DAT-cl (I) tell 'I tell him __ today as well.'
je
i
dîs
ancje ué. also
today.
ancje ue. also today.
Data Collection and Analysis Our subject was a male Friulian native speaker of middle age. We interviewed him before the data collection session to verify that he spoke a standard central variety of Friulian. Before the recording session, the subject had some time to train himself with a list of sentences containing true adjectives with the same structure as the pseudo-adjectives, and with the list of the sentences containing the pseudo-adjectives used as stimuli. After these preliminaries, the subject was recorded while reading the list of sentence pairs
30
MARCO BARONI & LAURA VANELLI
for six times, in random order. The recording took place in the sound booth of the Padua CNR Phonetics Laboratory. The data were digitized at a 20 kHz sampling rate and analyzed using the Kay CSL package at the UCLA Phonetics Laboratory. Six tokens of each category were analyzed, and the following properties were measured. • •
•
•
Duration of the stressed vowel: measured on the waveform, from the onset of the first complex periodic wave to the offset of the last one; First and second formant frequency of the stressed vowel: measured on an LPC analysis of the central portion of the vowel (all the formant values based on the LPC analysis were double-checked on a wide-band spectrogram and on a FFT plot of the same portion of the wave, in order to assess their plausibility); Position of the F0 peak in the stressed vowel: the relative position of the F0 peak of the vowel, i.e., the ratio of the time lag from the onset of the vowel to the peak to the overall duration of the vowel (both the onset-peak and the overall duration were measured on the pitch track, ignoring the initial and final portions of the vowel, where the pitch was perturbed by the neighboring consonants); Closure duration of the consonant following the stressed vowel: measured on the waveform and on a synchronized wide-band spectrogram, from the point corresponding to a sudden weakening of the formants to the release burst or, when this cue was absent, to the onset of the first complex periodic wave. (In Friulian, stops are unaspirated.)
4.3
Results The tables in the appendix report mean, standard deviation, minimum and maximum values of each category with respect to each measured property. All the statistical results presented below come from a series of One Way ANOVA with Scheffé's post hoc test (significance level at a = .05). 4.3.1 Some Non-Quantified Observations on the Characteristics of the Obstruents The acoustic waveform and spectrographic displays were visually inspected in the sections corresponding to /d/ and /D/ in order to detect the presence of periodic vibration and/or of the voicing bar. While both cues of voicing were present in all the tokens of /d/, they appeared to be completely absent from all the instances of /D/. This suggests that word-final /D/ is indeed
VOWEL LENGTH AND CONSONANTAL VOICE
31
voiceless, but we realize that the presence of vocal fold vibration should be assessed on the basis of articulatory data. We also noticed that in the 'voiced' stops (Id/ and /D/) the burst is often weak and hard to identify, whereas it is always clear and easy to identify in /t/. Again, this cue of partial word-final neutralization should be confirmed by articulatory data. 4.3.2 Duration of the Stressed Vowel The vowels before word-final /D/, i.e., the vowels traditionally labeled as long, are indeed extremely long ~ often, more than twice as long as vowels before word-final /t/. We will refer to the vowels before word-final /D/ as the 'extra-long' vowels. However, the vowels before word-internal /d/ (traditionally considered short) are also consistently longer than the vowels before /t/. In the extreme cases of /a o/, the vowels before word-internal /d/ are on average one fourth longer than the vowels before word-internal /t/, but the contrast is always systematic. Vowels before word-internal /t/ are only slightly longer than vowels before word-final /t/, and the contrast is not systematic. These observations are largely supported by the statistical analysis of the results. • •
• •
As expected, the difference between any vowel before /D/ and the same vowel in any other context is always statistically significant. In the cases of /a e o/, the difference between the duration of the vowel before word-internal /d/ and the duration of the vowel before the two // classes is statistically significant. (Notice, however, that in the case of Id the assumption of homogeneity of variances required by the ANOVA is violated, according to the results of the Levene test (p < .01).) In the case of /u/, the vowel before word-internal /d/ is only significantly longer than the vowel before word-final til. In the case of /i/, the difference between the vowels before word-internal /d/ and the vowels before lil is not significant. Notice, however, that even in the cases of /i/ and /u/ the difference between the vowels before word-internal /d/ and word-final/internal /t/ is quite consistent. In both cases, the longest token of the vowel before word-final lil is still shorter than the shortest instance of the vowel before word-internal Id/. For /u/, this is also true for the difference between the vowel before word-internal /d/ and /t/, whereas in the case of /i/ the overlap between these categories is minimal. (HI before word-internal /d/ ranges from 115 to 126 msec, whereas /i/ before wordinternal HI ranges from 122 to 157 msec.) The lack of statistical significance
32
•
MARCO BARONI & LAURA VANELLI
in the case of the high vowels is probably due to the fact that, overall, high vowels tend to be shorter 'by nature' (Lehiste 1970), and thus the duration contrast is less prominent. The difference between the vowels before word-internal vs. word-final /t/ is statistically significant only in the case of /o/.
Thus, our results confirm the existence of a marked length contrast between the vowels before word-final /D/ vs. /t/, but they also reveal the existence of a consistent vowel length distinction before word-internal /d/ vs. /t/. The latter cannot be considered a 'physiological' fact, since Keating (1985) has shown that the vowel length differences before voiced and voiceless consonant is a cross-linguistically common, but not universal, phenomenon: Hence, it cannot be physiological, and speakers must be phonologically aware of the lengthening. 4.33 First and Second Formant Frequency of the Stressed Vowel In the discussion that follows, we treat Fl and F2 as cues of vowel quality, since Fl is inversely correlated with height and F2 is directly correlated with anteriority (inversely correlated with posteriority). The general trends emerging from the data are that long vowels tend to be peripheral and short vowels tend to be central. (These are cross-linguistically common and phonetically grounded patterns; see, among others, Lindau 1978, Labov 1994.) However, the realization of these tendencies varies from vowel to vowel. In the cases of /a o/, it is the longest vowel (the one before word-final /D/) that is more peripheral (more back in the case of /a/; higher and more back in the case of /o/). In the case of /e/, we find a contrast between the two longest categories, /e/ before word-internal /d/ and word-final /D/, that are higher and more front, and the two shortest ones, /e/ before word-internal and word-final /t/, that are lower and less front.8 In the cases of /i u/, it is the shortest vowel (the one before word-final /t/) that is centralized (in both cases, the vowel before word-final /t/ is virtually identical to the most peripheral instances of the corresponding mid vowels /e o/). The statistical analysis gives the following results.
8 Impressionistic analysis of fieldwork data collected by Vanelli and others suggests that, for the mid vowels, /e o/, there is wide variation in the distribution of the tense and lax allophones even within the Central Friulian varieties.
VOWEL LENGTH AND CONSONANTAL VOICE • •
•
• •
33
The difference in F2 between /a/ before word-final /D/ and /a/ before wordinternal and word-final ill is statistically significant. The difference in Fl between /o/ before word-final /D/ and any other category is statistically significant; the difference in F2 between /o/ before word-final /D/ and /o/ before word-internal /t/ is statistically significant. The differences in Fl and F2 between /e/ before word-internal /d/ and wordfinal /D/, on one hand, and /e/ before word-internal and word-final /t/, on the other, are statistically significant.9 The difference in Fl between /i u/ before word-final /t/ and the other categories is statistically significant. For /i/, the difference in F2 between the vowel before word-final /t/ and the other categories is also statistically significant. In the case of /u/, F2 of the vowel before word-final /t/ differs significantly from the vowels before word-internal /t/ and /d/.
The schema in (24) summarizes the general correlation between length and peripherality. As the schema illustrates, before word-final consonants (in the contexts t# and D#) there is always a vowel quality contrast accompaning the vowel quantity contrast. (24)
shorter t# a
»
longer D#
dV#
tV# a
a
A O E U
0
0
0
ε u
ε u
e u
a o e u
I
i
i
i
4.3.4 Position of the FO Peak in the Stressed Vowel In stressed vowels before word-final /D/ (the extra-long vowels) the fundamental frequency peak occurs in the first half of the vowel. This supports the impressionistic observation that these vowels are characterized by a descending High-Low tonal contour. In all the other contexts, the peak occurs towards the end of the vowel (with the exception of the peak in /u/ before wordinternal /d/). From a statistical point of view, we found the following results:
9 However, for Fl the assumption of homogeneity of variances is violated (according to the Levene test, p < .01).
34
•
• •
MARCO BARONI & LAURA VANELLI
The difference between the relative location of the peak of the vowel before word-final /D/ and in the other contexts is statistically significant for /a e i o/; whereas for /u/ the vowel before word-final /D/ significantly differs from the vowel before word-internal and word-final /t/, but not from the vowel before word-internal /d/ (where the peak occurs rather early). In the case of /e/, the peak in the vowel before word-internal /d/ occurs significantly earlier than the peaks in the vowels before /t/. In the case of /u/, the peaks in the vowels before word-internal /d/ and /t/ occur significantly earlier than the peak in the vowel before word-final /t/.
Here, we will simply make the obvious remark that a contour tone is more likely to arise and survive on longer vowels, and we will not make any claim about the exact nature and correct representation of the contour High-Low tone that characterizes the extra-long vowels. The difference in tonal shape, together with the vowel quantity and quality distinctions, is another factor contributing to the contrast between word-final stressed vowels before /D/ and /t/. 43.5 Closure Duration of the Consonant Following the Stressed Vowel As expected, word-internal /d/ is shorter than /t/. (See Lisker 1978, among others, on the relationship between consonantal voicing and duration.) However, it is also the case that word-final /D/ (the devoiced consonant) is consistently shorter than /t/. In general, word-internal /d/ and word-final /D/ have very similar duration. Finally, word-final /t/ tends to be the longest category. These observations are confirmed by the statistical analysis. • • •
•
The difference between word-internal /d/ and word-internal /t/ after /e i/ is statistically significant. Word-final /D/ differs significantly from word-final ill after /a e i u/. The tables in the appendix show that even in the non-significant cases the voiceless categories are consistently longer than the 'voiced' ones: the longest token of word-internal /d/ after /a/ is still shorter than the shortest token of word-internal /t/; the overlap is minimal in the case of wordinternal /d/ vs. /t/ after /o/, and only slightly larger in the cases of wordinternal /d/ vs. /t/ after /u/ and word-final /D/ vs. /t/ after /o/. The difference between word-internal /d/ and word-final /D/ is never statistically significant.
VOWEL LENGTH AND CONSONANTAL VOICE
•
•
35
Word-final ill is significantly longer than any other category after /e/ and /i/; it is significantly longer than the two 'voiced' categories after /a/ and /u/, and it is significantly longer than word-internal /d/ (the shortest category) after /o/. Word-internal /t/ is significantly longer than word-final /D/ after /a e i/.
We can conclude that word-final neutralization is not complete: while word-final devoiced /D/ is not characterized by vocal fold vibration (the phonetic feature that is disfavored in word-final position), the length distinction between voiced and voiceless stops is maintained. This distinction must also have been, historically, the contextual factor causing the rise of the vowel length contrast, as a compensatory lengthening/shortening phenomenon. (See Kingston & Diehl's 1994 notion of the V/C duration ratio as a cue to the voicing distinction.) The impressionistic claim that word-final devoiced consonants become identical to word-final voiceless consonants is not correct. 4.4 Conclusions on the Acoustic Data The features that, according to the results of our investigation, distinguish each VC pair in each context are summarized by the tentative transcriptions in the following schema. (We use the symbol " " to represent the High-Low tone on the final extra-long vowels). (25) A O E U I
Word-Internal
ut
od ed ud
Vt# atotst ut-
u':t
it-
id
it-
i:t
Vt atotεt
Vd ad
Word-Final VD# a':t o':t e':t
In this section we will try to draw some tentative conclusion based on our preliminary results. We realize that the results of our acoustic analysis should be confirmed and complemented by articulatory data, by the instrumental investigation of data from other speakers, and by the analysis of real words and othcr consonants. Furthermore, in order to compare Friulian partial neutralization to other well-known cases (such as German; Port & Crawford 1989), other characteristics of the phenomenon should be analyzed: for example, to what extent partial neutralization depends on pragmatic or
36
MARCO BARONI & LAURA. VANELLI
grammatical factors, and whether Friulian speakers can perceive the difference between partially neutralized stops. The idea, first suggested by Vanelli (1979), that lengthening before wordfinal devoiced obstruents represents the phonologization of the general tendency towards lengthening of vowels before voiced consonants, is confirmed: vowels before word-internal voiced obstruents are longer than vowels before wordinternal voiceless obstruents, but the duration contrast is emphasized in wordfinal position, where the vowels before the devoiced obstruents are more than twice as long as the vowels before voiceless obstruents. However, the traditional assumption that word-final devoiced obstruents become identical to word-final voiceless obstruents was not confirmed by our analysis: word-final voiced obstruents are devoiced, in the sense that they are not characterized by vocal fold vibration, but they are as short as word-internal voiced stops, and thus they are not identical to word-final voiceless obstruents. Thus, it is possible to formulate an analysis that does not assume rule ordering, in which vowel lengthening follows from a rule/constraint on surface forms, requiring wordfinal vowels before (extra-)short consonants to be extra-long. This, or a similar rule/constraint, will be able to distinguish between vowels before word-final devoiced (= shortest) vs. voiceless (= longer) consonants. However, we feel that an analysis along these lines has an unsatisfactory aspect: while a rule/constraint that lengthens vowels before short consonants is empirically, phonetically and typologically sound, this formulation misses an important fact about Friulian vowel lengthening, namely, that it is crucially connected with the devoicing of the following consonant. We would like to finish this paper suggesting some speculations on a possible 'functionalist' approach to this problem. Intuitively, vowel lengthening before word-final devoiced obstruents is not simply a phenomenon of duration compensation within a VC sequence. The latter is probably the case in word-internal position, where vowels before voiced (hence, shorter) obstruents are slightly longer than vowels before voiceless (hence, longer) obstruents. The lengthening of vowels before word-final voiceless obstruents is clearly more extreme, well beyond a similar compensation pattern. It is plausible that vowel lengthening is emphasized in word-final position as a form of compensation for the loss of the most salient feature ensuring the contrast in word-internal position, namely voicing. Notice that this is not a form of compensatory lengthening (such as the one proposed by Hualde): there is no compensation in terms of duration, but rather a compensation in terms of contrastivity. Since the voicing distinction is lost in word-final position, due to final devoicing, the burden of contrast is left
VOWEL LENGTH AND CONSONANTAL VOICE
37
to the vowel length and to the consonantal length distinctions. However, the distinction between slightly longer and slightly shorter vowels attested in wordinternal position is not very robust, and thus it is emphasized in word-final position, where it acquires the role of the main contrast-carrying feature. Probably, vowel length is preferred as the phonologized feature because even a sharp consonantal contrast in word-final position would be hard to perceive. We will not address here the issues of how such an approach could be formalized, and whether similar considerations on contrast preservation should be formulated as components of a synchronic grammar. More in general, the data suggest a notion of contrast less categorical, more flexible than the traditional one: in word-internal position, as well as in word-final position, the burden of contrast is not entirely on one single feature, but on a set of phonetically correlated features (voicing, different degrees of consonantal and vowel lengthening, vowel quality, tonal shape). While it is probably the case that voicing is the most salient feature distinguishing the word-internal VC sequences, and vowel length is the most salient feature distinguishing the word-final VC sequences, in both contexts other features contribute to the contrast. APPENDIX Duration of the stressed vowel: The tables from 1 to 5 report the mean, standard deviation and minimum and maximum values of duration of each vowel in each context (duration values are expressed in milliseconds). |CATEGORY| [MEAN S.D. MIN 9 167 ad 178 132 7 124 at aD# 261 18 238 at# 124 9 11 Table 1. duration of lal MIN CATEGORY [MEAN S.D. ed j166 146 13 et 133 9 122 eD# 259 29 220 et# 124 121 Table 2. duration of /e/
MAX 192 145 293 135
MAX 180 | 146 301 129
38
MARCO BARONI & LAURA VANELLI
MIN CATEGORY | [MEAN S.D. id 135 12 122 120 5 115 it iD# 236 23 196 it# 102 112 112 Table 3. duration of lil CATEGORY | [MEAN S.D. MIN 7 169 od 181 5 125 ot 135 279 oD# 296 11 6 109 !ot# 120 Table 4. duration of /o/ CATEGORY | MEAN^ S.D. MIN 130 ud 143 18 4 116 121 ut uD# 221 16 201 ut# 115 8 102 Table 5. duration of /u/
MAX 157 126 257 120
MAX 189 142 304 128
MAX ! 177 127 239 122
First and second formant frequency of the stressed vowel: The tables from 6 to 15 report the mean, standard deviation and minimum and maximum values of the first and second formant frequency of each stressed vowel in each context (frequency values are expressed in Hertz). CATEGORY ad at aD# at#
MEAN | 703 711 695 733 Table
MIN S.D. 23 688 38 688 18 688 41 688 6. F1 of /a/
MAX 733 779 733 779
VOWEL LENGTH AND CONSONANTAL VOICE
CATEGORY| |MEAN S.D. MIN 65 1513 1422 ad 56 1467 1574 at 47 1376 !l437 aD# 1544 37 1513 at# Table 7. F2 of /a/ CATEGORY | [MEAN S.D. MIN 24 ed 381 366 0 et 550 550 24 381 eD# 366 19 |et# 550 557 Table 8. Fl of /e/ CATEGORY ed et eD#
|et#
MEAN S.D. 2117 53 1865 37 2132 48 1834 29 Table 9. F2 of /e/
MIN 2018 1834 2064 1788
CATEGORY| |MEAN id 275 it 275 267 iD# it# 404 Table
S.D. MIN 0 275 0 275 19 229 19 366 10. Fl of /i/
CATEGORY id it iD# it#
S.D. 24 37 45 80 11. F2 of
MEAN 2355 2355 2393 2178 Table
MIN 2339 2293 2339 2064 /i/
MAX 1605 1605 1513 1605
MAX ! 412 550 412 596
MAX
2155 1926 2201 1880
MAX 275
275 275 412
MAX 2388 2385 2477
2247
40
MARCO BARONI & LAURA VANELLI
CATEGORY \ MEAN S.D. MIN 614 25 596 od 596 604 19 ot 403 366 9 oD# 16 550 ot# 619 Table 12. Fl of /o/ CATEGORY od ot oD# ot#
MEAN S.D. MIN 1018 1009 20 1054 58 963 935 101 779 1047 45 1009 Table 13. F2 of /o/
CATEGORY
MIN MEAN S.D. 34 275 328 19 275 313 328 18 321 366 |427 37 Table 14. Fl of /u/
ud ut uD#
|ut# CATEGORY ud ut uD#
|ut#
1
MEAN S.D. MIN 79 863 733 817 19 779 902 63 825 1032 150 917 Table 15. F2 of /u/
MAX 642 642 412 642
MAX 1055 1100 1055 1100
MAX ! 366 321 366 458
MAX ! 963 825 1009 1330
Position of the F0 peak in the stressed vowel: The tables from 16 to 20 report the mean, standard deviation and minimum and maximum values of the ratio of the time lag from the vowel onset to the F0 peak to the overall duration of the stressed vowel (these proportional values range from 0 to 1).
VOWEL LENGTH AND CONSONANTAL VOICE
CATEGORY ad at aD# |at# Table
MIN MAX MEAN S.D. .668 1 .129 .865 .503 .848 .199 1 .417 .546 .475 .049 .833 1 .967 .074 16. position of F0 Peak in lal
MIN MAX CATEGORY| [MEAN S.D. .635 .814 ed .714 .070 .788 et .932 .105 1 .334 eD# .394 .041 .455 .804 |et# .967 .080 1 Table 17. position of F0 Peak in lei CATEGORY id it iD# it# Table
MEAN S.D. MIN .767 .602 .182 .854 .079 .800 .463 .103 .307 .822 .157 .599 18. position of F Peak in /i/
MAX
1 1 .583 1
MIN CATEGORY MEAN S.D. MAX od .779 .934 .099 .667 Ot | .917 .139 .267 oD# .430 .319 ot# .000 1 |l Table 19. position of F0 Peak in /o/ CATEGORY MEAN ud i .488 .651 ut uD# .228 ut# .966 Table 20. position
S.D. MIN .212 .126 .169 .428 .154 .000 .082 .799 of F0 Peak in /u/
MAX | .667 .800 .456
1
41
42
MARCO BARONI & LAURA VANELLI
Closure duration of the consonant following the stressed vowel: The tables from 21 to 25 report the mean, standard deviation and minimum and maximum values of the closure duration of the alveolar stop following each vowel in each context (duration values are expressed in milliseconds). MIN MAX CATEGORY MEAN S.D. 64 56 ad 6 72 at 86 77 96 9 56 39 64 11 aD# 101 |at# 25 52 122 Table 21. duration of the stop after lal MIN MAX CATEGORY MEAN S.D. 50 10 33 65 ed et [91 12 71 108 51 19 14 64 eD# et# 101 132 119 12 Table 22. duration of the stop after /e/ MIN MAX CATEGORY | [MEAN S.D. 64 71 id 4 59 86 it 8 77 96 55 iD# 42 69 11 125 16 113 151 it# Table 23. duration of the stop after /i/ CATEGORY od ot oD#
MEAN S.D. MIN MAX 61 10 50 70 79 9 69 90 76 11 62 89 91 22 57 123 Table 24. duration of the stop after /o/
VOWEL LENGTH AND CONSONANTAL VOICE
43
MIN MAX CATEGORY 1 MEAN S.D. 67 74 ud 7 58 77 13 93 ut 61 12 52 82 uD# 67 ut# 93 12 77 110 Table 25. duration of the stop after /u/ REFERENCES Bais, Maurizio. 1997. "La lunghezza vocalica nella lettura di parole friulane e di non parole". Ce fastu? 73.7-29. Baroni, Marco & Laura Vanelli. 1991. "Il contrasto di lunghezza vocalica in friulano" Fonologia e morfologia dell'italiano e dei dialetti d'ltalia ed. by Paola Benincà, Alberto Mioni, & Laura Vanelli, 291-317. Roma: Bulzoni. Broselow, Ellen. 1992. "Parametric Variation in Arabic Dialect Phonology". Perspectives on Arabic Linguistics IV (Papers from the Fourth Annual Symposium on Arabic Linguistics ed. by Ellen Broselow, Mushira Eid & John McCarthy, 7-45. Amsterdam & Philadelphia: John Benjamins. Farnetani, Edda & Shiro Kori. 1986. "Effects of Syllable and Word Structure on Segmental Duration in Spoken Italian". Speech Communication 5.17-34. Hayes, Bruce. 1989. "Compensatory Lengthening in Moraic Phonology". Linguistic Inquiry 20.253-306. Hualde, José I. 1990. "Compensatory Lengthening in Friulian". Probus 2.31-46. Keating, Patricia A. 1985. "Universal Phonetics and the Organization of Grammars". Phonetic Linguistics: Essays in Honor of Peter Ladefoged ed. by Victoria A. Fromkin, 115-132. Orlando, Florida.: Academic Press. Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Oxford: Blackwell. Kingston, John & Randy L. Diehl. 1994. "Phonetic Knowledge". Language 70.419-455. Kiparsky, Paul. 1968. "Linguistic Universals and Language Change". Universals in Linguistic Theory ed. by Emmon Bach & Robert Harms, 191 212. New York: Holt, Rinehart & Winston. . 1971. "Historical Linguistics". A Survey of Linguistic Science ed. by W. Dingwall, 576-649. College Park: Linguistic Program, University of Maryland. Kisseberth, Charles. 1973. "Is Rule Ordering Necessary in Phonology?". Issues in Linguistics: Papers in Honor of Henry and Renée Kahane ed. by B. Kachru, 418-441. Urbana, Illinois: University of Illinois Press.
44
MARCO BARONI & LAURA VANELLI
Kluender, Keith R., Randy L. Diehl & Beverly A. Wright. 1988. "Vowel Length Differences before Voiced and Voiceless Consonants: An Auditory Explanation". Journal of Phonetics 16.153-169. Labov, William. 1994. Principles of Linguistic Change: Internal Factors. Cambridge, Mass.: Blackwell. Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, Mass.: MIT Press. Lindau, Mona. 1978. "Vowel Features". Language 54.541-563. Lisker, L. 1978. "Rapid vs. Rabid: A Catalogue of acoustic features that may cue the distinction". Status Report on Speech Research 54.127-132. McCarthy, John & Alan Prince. 1990. "Prosodic Morphology and Templatic Morphology". Perspectives on Arabic linguistics IV (Papers from the Fourth Annual Symposium on Arabic Linguistics ed. by Ellen Broselow, M. Eid & John McCarthy, 1-54. Amsterdam & Philadelphia: John Benjamins. Maddieson, Ian. 1985. "Phonetic Cues to Syllabification". Phonetic Linguistics: Essays in Honor of Peter Ladefoged ed. by Victoria A. Fromkin, 203-221. Orlando, Florida: Academic Press. Port, Robert & Penny Crawford. 1989. "Incomplete Neutralization and Pragmatics in German". Journal of Phonetics 17.257-282. Repetti, Lori. 1992. "Vowel Length in Northern Italian Dialects". Probus 4.155182. . 1994. "Degenerate Syllables in Friulian". Linguistic Inquiry 25.186-193. Vanelli, Laura. 1979. "L'allungamento delle vocali in friulano". Ce fastu? 55.66-76. . 1986. "La fonologia dei prestiti in friulano". Raetia antiqua et moderna ed. by Gunter Holtus & Kurt Ringger, 355-376. Tübingen: Niemeyer. . 1989. "Friulano e italiano: aspetti contrastivi". Scuola, Lingue e Culture Locali ed. by Nereo Perini, 227-240. Codroipo: Comune di Codroipo. . 1997. "Friuli". The Dialects of Italy ed. by Mair Parry & Martin Maiden, 279-285. London & New York: Routledge. Westbury, John. & Patricia Keating. 1986. "On the Naturalness of Stop Consonant Voicing". Journal of Linguistics 22.145-166. Yamamoto, S. 1993. "Alcuni ampliamenti dei casi dell'allungamento vocalico nel friulano". Per Giovan Battista Pellegrini. Scritti degli allievi padovani ed. by Laura Vanelli and Alberto Zamboni, 645-655. Padova: Unipress.
CONSONANT GEMINATION IN NEAPOLITAN
BARBARA E. BULLOCK The Pennsylvania State University
0.
Introduction Neapolitan, a dialect of the center-south of Italy, has a robust population of over seven million speakers. In both its lexicon and its phonetic evolution from Vulgar Latin, it differs dramatically from standard Italian which is based on northern-central (Tuscan) dialects and it is generally not comprehensible to speakers of standard Italian. There is a large body of literature, particularly poetry, written in Neapolitan and the dialect is internationally known through its popular folk songs. Some of the distinctive properties of the phonology of Neapolitan, many of which are shared by other southern dialects, are: reduced vowels, especially final atonic vowels which are often reduced to schwa [9], apocopated verbal infinitives that are oxytonic ([jc'ka] 'to play' ~ standard Italian giocare), imperatives that are often partially reduplicated ([jcke 'jo] 'play!'), and geminate consonants in lexical items where Italian cognates show a single consonant ([ad'dcre] 'adore', [am'mcre] 'love' ~ standard Italian a[d]ora, a[m]ore), and also where Italian has a cluster (Neapolitan [kckke] ~ standard Italian qualque ['kwalke] 'several'). Another distinguishing property of Neapolitan is the gemination of word initial consonants in particular syntactic environments, a phonomenon known as Raddoppiamento Sintattico (RS). In phonological analyses of standard Italian, RS has received considerable theoretical attention. In standard Italian RS, word initial consonants are geminated after words with final stress. Vogel (1982:68-72) views the process as one in which a consonantal slot is inserted word finally and the following onset is geminated and subsequently resyllabified to close the final stressed syllable (di.ven.to # ság.gio di.ven.tóCsag.gio diven.tó [s.s]âggio 'I will become wise'). (See Fanciullo 1986:82 for discussion.) While initial consonant gemination in Italian is argued to be phonologically predictable on the
46
BARBARA E. BULLOCK
basis of final stress (Vogel 1982), this is not the case in Neapolitan. Here, RS is controlled by a variety of factors. First, it is triggered only by certain lexical items; final stress is neither a necessary nor sufficient condition. Secondly, it can serve a specific morphological function in the dialect, distinguishing gender morphology both in the singular and plural forms. Finally, it causes systematic alternations in the quality of word-initial voiced obstruents. In essence, while both standard Italian and Neapolitan illustrate the phenomenon of RS, it is clearly not a unitary process and analyses of gemination in standard Italian will not extend to Neapolitan. In what follows, I detail the facts of gemination in Neapolitan. I will argue that RS is triggered in Neapolitan by a class of function words that display a unique 'double' prosodic structure. Following the phonological analyses of Repetti (1991) and Bullock (1991) for Italian RS, I maintain that gemination in Neapolitan falls out of the competing requirement that certain syllables be bimoraic but that final vowels remain short. That is, RS occurs to best satisfy the structural requirements on prosodic word size. A strong implication of this analysis is that historically dominant aspects of a language's prosody (specifically syllable weight) can become encoded lexically as abstract phonological properties. 1.
The Basic Data and the Contextfor RS In Neapolitan, select grammatical items automatically produce RS. 1 These are the conjuction é 'and' (la), the prepositions a 'at, to' (lb), pé 'by, for' (lc), cu [ku] 'with' (1d), the number tre 'three' and its derivatives (except trirece 'thirteen'), the interrogatives che 'what' (le) and pecché 'why' (1f), the indefinite qualifiers quacche 'some' and ogni 'every', the adverbs cà [ka] 'here' (lg), là 'there', (lh), the verb esse re 'to be' in its present forms é, sì, só (1i), the verb forms stó 'existential be', po 'can' (1j), and the neuter singular definite article o (1k), the feminine plural definite article é or i (11), and the intensifiers cchiú 'more' (1m) and accussi 'so' (1n),and the negative nu (Bichelli 1974). (Glosses in Standard Italian are included for comparison.)
1
Unless otherwise specified, the data within this analysis come from fieldwork I conducted in 1987 with a male native speaker of Neapolitan in the Italian community of Wilmington, Delaware. The participant was a 48 year old cobbler who had lived in the U.S. for 10 years and who still spoke the dialect regularly with his family and in his community. While some of the information is from spontaneous conversation, some was also collected based on the participant's retelling of Neapolitan folkales. These tales were collected, transcribed and published by Valerio Petrarcha. However, I am only in possession of a mimeograph extract of this work and have been unable to track down a complete reference.
CONSONANT GEMINATION
47
No other items in the language cause gemination although there are forms with lexicalized geminates in initial position: [bb]uona 'good', [bb]elle 'pretty', [kk]iù 'more', [rr]e 'king' always occur with geminate onsets. (1) Examples of RS in Neapolitan a. b. c.
[e ssautavenette] [a kkja
әrә]
[a t t e a t r ] [pe b b i ] [pe nne ffune parts]
d e.
[ku mme] [ke bbwc fa]
'e si voltavano' 'a piangere 'a teatro'
'con me' 'che vuoi fare'
f.
[pek'ke rri punne]
'perche rispondi'
[kadderete]
'qua dietro'
[laffora]
'là fuori' 'tu sei scemo'
j k.
[tu si eme] [so bbjekkje] [po ttrasi] [o bbo ke] [o mmele]
1. m.
[e pporte] [kkju fforta]
n. o.
[kkju u ave] [akkussi kkumbinate] [nu mme fa eme]
'to the theater'
'per via' 'by way of' 'per nessuna parte' 'no where'
g. h. i
'and they jumped' '(to) to cry'
'sono vecchio'
'with me' 'what do you want to do' 'why reply?' (Bichelli 1974:60) 'here behind' (Bichelli 1974:61) 'there outside'(Bichelli 1974:61) 'you are silly' 'I'm old' 'you can enter'
'puo entrare' 'il bosco'
'woods'
'il miele' 'le porte'
'honey' 'the doors'
'più forte'
'more strong'
'più soffïava' 'cosi conciati'
'the more it blew' 'so transformed'
'non mi fai scemo' 'don't make me stupid'
Gemination serves to mark a gender morphology distinction that it cannot mark in Italian. Neapolitan has a neuter definite article and direct object pronoun. Although the pronoun itself is indistinguishable from the masculine form, the neuter, which is used for mass nouns, nominalized adjectives, and abstract or generic objects, causes RS while the masculine does not (Andalò & Bafile 1991, Bichelli 1974, Fanciullo 1996, Iannucci 1948-49, Rohlfs 1969). This retention of neutral gender distinct from the masculine is shared by many Meridional dialects although the difference between markers is often manifested through a vocalic distinction as well as RS. (For example, in the Marches masculine is lu/ru, and neuter is lolro; Rohlfs 1966.)
BARBARA E. BULLOCK
48
(2) Neuter -vs- Masculine Neuter Article: [o n'niru] [od'dctt e] [o ll'atte] [offjerru] [okkaffe] Masculine Article:
Neuter Pronoun:
Masculine Pronoun:
[o 'kans] [o kaffe] [o Vskkju] ['porti: fjerra] [ob'bids] [o rn'majijie] [od'dika] [o Vids]
'il nero' 'il dolce' 'il latte' 'il ferro' 'il caffè' 'il cane' 'il caffè' 'il vecchio' 'porte di ferro'
'black1 'the sweet (one)' 'milk' 'iron' (Fanciullo 1986) 'coffee' (Fanciullo 1986) 'dog' 'a cup of coffee' 'the old one' 'iron doors' 'I see it'
'lo vedi' 'lo mangio' 'I eat it' 'I tell it' 'lo dico' 'lo vedi (il cane)' 'I see it, (the dog)'
Notice that the generic use of a term (o ffiero 'iron') is opposed to its specific or qualificatory use also by gemination (porti fierro 'doors of iron'). The feminine plural definite article, e or i, and direct object pronoun function in a similar fashion, distinguishing themselves from the masculine plural forms by provoking gemination. (3) Gemination in Feminine Plural Feminine Sg: [a 'femmene]
Feminine P1:
Direct Object:
[a'kasa] [a preggjere] [e f'femmene] [e k'kase] [i ppreg'gjsre] [s i mman'navene]
'la donna' 'la casa' 'la preghiere' 'le donne'
'the woman' 'the house' 'the prayer' 'the women'
'le case' 'the houses' 'le preghiere' 'the prayers' 'se le mangiavano' 'they atethem(f.)'
In a refreshing twist on gender morphology polemics, it is the feminine plural form with gemination that is the default for plural direct objects, rather than the masculine (Iannucci 1948-49). (4) Default Gender in Plural Direct Objects: Masculine and/or Feminine Plural: [ev'vedu] 'I see them (m./f.)' [e t'tengu] [e mman'navene]
'I have them (m./f.)' 'they ate them (m./f.)'
CONSONANT GEMINATION
49
These distinctions carry over into all determiners that are marked by gender morphology: ati kkose ('altre cose') 'other things (f.)' (Fanciullo 1986), certe [bb]ipere ('certe vipere') 'certain snakes' (Rohifs 1969:237). Finally, RS occurs on the word following these elements and those listed above in ( 1) even if that word is itself a qualifier or a specifier (5). (5) Gemination of Subsequent Element [e t'tenteε'msssa](f.pi) 'le cento messe' [nati d'djεtt/ 'anni] 'altri died anni ' [o m'mεyyo 'panә] (n) 'il miglior pane' [tre b'brave fiyyjuclә] 'tre brave figliuole'
'100 masses' 'another 10 years' 'the best bread' (Fanciullo 1986) 'three fine girls'
However, as observed by Bichelli (1974), Rohlfs (1966) and Fanciullo (1986), there are syntactic conditions on RS in the Meridional dialects that are much more rigid than those found in standard Italian. For instance, the RS triggering words will only provoke gemination when they specify the phrase that immediately follows, thus there is a difference between a phrase such as (1h), [la ffora] 'there outside' and [la | fcra: pcrtә] 'there, outside the door' where a phrasal break intervenes between the adverbs. There is further evidence of the strict syntactic nature of RS in Neapolitan in that the forms che 'what' and pecche 'why' trigger the process in their interrogative forms only (che bbuò [kebbwc] 'what does he/she want?'), but the complementizer forms of these words resist gemination (che lu a fatto [ke lus fatte] '[timt]comp he did'). According to Fanciullo (1986:88), RS is also blocked in Neapolitan whenever the triggering word in not 'connected rather rigidly to the terms to which they refer'. Thus, he states that it is possible in Neapolitan to form the phrase [o m ' m e o temεndә pe ffabrә'ka] 'the best cement for building' with the neuter [o] triggering RS, but not *[o | ppe ffabrә'ka | 'meyyo t emεndә] where the prepositional complement 'for building' is dislocated to the position immediately follow the RS triggering determiner in surface structure. 2.
The Prosodic Status of RS Triggers: Fnc/ Lex From the above discussion, we see that Neapolitan RS is clearly a morphosyntactically conditioned process. Only a certain class of items serve as triggers and only under rigid syntactic conditions. A principle concern for this analysis is the types of words involved and their internal structure. Selkirk (1995) proposes an alignment based theory which accounts for the lexical (lex) versus function (fnc) word contrast cross-linguistically and its interaction with the phonological component of the grammar. In general, function words are distinct from lexical
50
BARBARA E. BULLOCK
words in that they each have unique phonological properties. For instance, in English, fnc may display a stressless and stressed form, while lex can only be stressed. Within Selkirk's theory, function words may be prosodized either as PWd or they may be prosodized as clitics. The series [fnc+lex] or [lex+fnc] is, in turn, parsed into a hierarchically dominating Phonological Phrase (PPh). Following Selkirk's theory, I propose that there are two types of fnc words in Neapolitan that are distinguished by their prosodic properties. The first type includes those words in (1) which provoke RS; these are fnc words whose final syllables are parsed into a foot (cp); that is, they are minimally bimoraic. The second type of fnc are those, like the masculine and feminine singular definite articles, that do not equal a foot but are instead parsed directly into a PWd as a syllable (a). These configurations are given in (6a) and (6b), respectively. (6) The Hierarchical Structure offnc (following Selkirk 1995:450) a. fnc as qp (+RS) b. fnc as σ (-RS) PPh PPh
/\
/\ |
PWd PWd
PWd
| | || qp
|
|
| fnc
lex
a
| I I fnc lex
The configuration in (6a) in which a. fnc is parsed as a full cp is the one which gives rise to RS in Neapolitan. Outside of this specific PPh domain, RS does not occur. That is, a phonological phrase that may consist of two lex does not contain the morphological constituency to invoke RS. Similarly, the structure in (6b) does not contain the phonological constituent necessary to create the environment for RS since, here, a. fnc is parsed only as a syllable. There are historical reasons to believe that (6a) is the correct prosodic structure for the words that induce RS since the majority of these forms were once consonant final and bimoraic. For instance, the difference between the masculine singular article illu(m) in Latin and the neuter singular illud resides in the fact that the nasal consonant deleted, leaving a final open and short syllable, while the consonant final neuter retained its final closed syllable. Although such phonological constrasts have been lost in the surface forms of these function words, I maintain that a morphological contrast with phonological repercussions remains in the language's grammar. The different structures in (6) would seem to indicate a critical prosodie difference since a foot (6a) may be expected to bear
CONSONANT GEMINATION
51
stress while a syllable (6b) would not. While this difference is not manifested directly in Neapolitan, many other Meridional dialects (the Marches) show such a contrast with a reduced vowel /u/ in the masculine and a non-reduced /o/ in the neuter. Despite the lack of sychronic surface contrast, the etymological evidence of final consonants in the RS triggers in Neapolitan is overwhelming. With the exceptions of two forms, ogni < omnem and quacche < qual(e) che (sia), where no etymological consonant can be reconstructed in final position, all other forms are attested with final consonants. The non verbal RS triggers with their Latin etymons appear in (7). (7) Latin Etymons for RS Provoking Words a. e < ad b. c. d e. f. g. h. i. j. k.
pé < per cu < cum tre < tre: s che < quid ca < (ec) (cum) ha: c la < iliac o < illud (neuter, sg) e,i < illa(e)s, illaec cchiu < plus accussi < eccu(m) sic
Such etymological evidence sheds some light on the differences encoded in gender mophology. That is, the masculine singular o < ilium contrasts with neuter singular o < illud and the feminine plural e/i < Mas or illaec contrasts with the homophonous masculine e/i
52
BARBARA E BULLOCK
language. However, the presence of a diachronic initial consonant is still felt in French since, like all other initial consonants, the silent h-aspiré blocks vowel elision. Thus, le hibou 'the owl' is pronounced [lœibu], rather than *[libu]. A parallel case to French h-aspiré is attested in Salentine where an etymological initial consonant also blocks elision [la urpe], rather than * [lurpe] 'the fox' (volpe) (Fanciullo 1986:97). 3.
The Phonology of RS The kind of abstract phonological property that I associate with RS provoking function words in Neapolitan is actually well attested across Romance languages.2 It is a phenomenon which I call 'double prosody' meaning that there is a mismatch between the structure of a syllable and its weight scansion (Bullock ms). In one type of 'double prosody', the deletion of a coda element often leaves behind its inherent timing so that while a syllable may appear structurally light due to its internal segmental constituency, it nonetheless can often scan as a heavy, bimoraic one. In many cases, the missing coda element is eventually supplanted by a lengthening of the vocalic nucleus, as in compensatory lengthening. However, examples of double prosody appear throughout the history of Latin phonology and metrics which demonstrate the opposite effect, moraic reduction. For instance, in the Preclassical Latin metrical rule of brevis brevians, a long vowel following a short one is scanned as light although it remains quantitatively long (Mester 1994). Double prosody occurs at some cost to the grammar. That is, I perceive of the mismatch between internal syllable structure and the syllable's prosodic weight as a violation of the fundamental properties of mapping between the segmental and prosodic tiers. The double prosody constraint that is violated in Neapolitan is illustrated below. (8) Double Prosody Constraint: *CV - μμ A light syllable parsed as heavy violates well-formedness. 2 I assume that Neapolitan, like Italian is a quantity sensitive language with a trochaic (leftheaded) rhythmic pattern. Historical Latin vowel length has largely been supplanted in Neapolitan by word internal consonant gemination so that syllable quantity in stressed syllables has been maintained. There is evidence from the movement of stress in enclitic forms that points to a binary, moraic foot. In Neapolitan, the addition of a single enclitic does not displace the primary stress from the stem (magna+ti = mágnati 'eat (you)'). However, the addition of two enclitics causes the stress to be moved forward onto the first of the two clitics (magna+te+lo mangatéllo (you) 'eat it') (Bichelli 1974, Kenstowicz 1991). The conclusion I draw is that a single enclitic fails to constitute a binary foot; thus, the stem is not reparsed. Two enclitics do form a trochaic foot, and stress is recalculated to strike that foot.
CONSONANT GEMINATION
53
Within a constraint-based theory of phonology, violations of wellformedness do occur but all constraint violation is minimal and the impact of a constraint violation on output can only be determined relatively. That is, constraints must be ranked with respect to one another and it is this ranking that determines a language's grammar. I have proposed above that the RS triggers each constitute a PWd and, necessarily, a bimoraic foot. The most harmonic form of a foot in a quantity sensitive language like Neapolitan is the bimoraic foot [ΜΜ]. This is formalized by a constraint on the prosodic shape of the foot, FrBlN, (Prince & Smolensky 1993: Chapter 4) that analyzes all feet as binary at the moraic level. In citation form, the monosyllabic fnc words under consideration here seem to fail the requirement on foot binarity because they surface with short final vowels {cu [ku] not *[ku:] 'with', etc.). This is similar to standard Italian where even though stressed word internal open syllables manifest lengthened vowels, final stressed open syllables never do. However, there is a conflict between these constraints and other structural constraints at the segmental level since Neapolitan prohibits vowel lengthening. Long vowels in Neapolitan only occur when vowels coalesce at the the juncture between morphemes (Bichelli 1974). I assume that such a prohibition falls out from a constraint prohibiting the lengthening or shortening of vowels, proposed by Urbanczyk (1994), and formulated in (9). (9) Weight Identity (WTlDENT): *V:
Let us consider an input in the RS environment: an input like cu me 'with me' has several possible output forms which can be informally evaluated as in (10). The optimal candidate (10e) violates the structural constraint prohibiting codas, but it satisfies the others. It is clear from the evaluation of candidate (10a) that adhering to input/output FAITHFULNESS is not enough if a violation of prosodic structure occurs. Thus, FTB0IN outrank faithfulness. However, (10c) demonstrates that leaving a second mora unattached to the segmental tier leads to a fatal violation of Double Prosody. Consonant insertion in (10d) also leads to a fatal violation. The final ranking of constraints is then: FTBIN » WTlDENT : DOUBLE PROSODY : FAITHFULNESS » NOCODA. At this point, the three constraints in the middle cannot be ranked with respect to one another since neither one picks out the better candidate.
54
BARBARA E BULLOCK
(10) Possible Candidate Outputs for Input cu me 'with me' a. b.
[ku 'me] Violates FTBIN ['ku: 'me]
c.
Satisfies FTBIN Violates WTIDENT [ku[μ]'me]
d
Satisfies FTBIN Violates DOUBLEPROSODY ['kut'me] (with an epenthesized consonant)
e.
Satisfies FTBIN Violates FAITHFULNESS ([t] is not contained in the input) Violates NOCODA(a universal constraint against codas) ['kum'me] Satisfies FTBIN Satifies FAITHFULNESS ([m] is contained within input) Violates NOCODA
If we were to consider one of the RS fnc words in isolation, its best parse satisfies FTBIN since its is a PWd. But its output form contains no long vowel or epenthesized consonant. Thus, the constraints against vowel lengthening and epenthesis must both outrank the constraint against DOUBLE PROSODY. Briefly, this means that it is better to allow a final floating mora (outside the proper RS environment) unattached to segmental material than it is to augment the base through epenthesis or vowel lengthening. The constraint ranking is shown in (11). (11) Constraint Tableau for RS input [cu(PWd) me (pwd)]phrase FAITH DOUBLE FTBIN WTIDENT
NOCODA
PROSODY
a. r(ku) mé] b. [(kü:) mé] c. r(kut) mé] d. [(kú[μ])mé] e. fkùm mé] |
*! *! *!
* *! *
In (11), parentheses mark foot boundaries and an asterisk denotes a constraint violation. Violations which take a constraint out of the running are marked with an exclamation point, and all cells that are inconsequential for a candidate's evaluation are shaded. The best candidate in (11e) still incurs a
CONSONANT GEMINATION
55
constraint violation since it has a coda consonant, but due to the order of these constraints, violation here is low ranked and minimal. It succeeds as the 'optimal' candidate although it is still not perfect. The advantage to this kind of constraint based approach over rule based approaches is that it obviates the need for derivations in phonology and presents consonant gemination in Neapolitan as the best output for a given input rather than as the only possible output. As I view it, RS in Neapolitan arises because its triggers are endowed with a bimoraic prosodic structure. This essentially provides these function words with an additional timing slot that is imperceptible in citation forms and that ultimately surfaces in the proper syntactic configuration. Another phonological phenomenon that arises in the context of RS in Neapolitan is a predictable alternation involving the voiced obstruents that might be directly related to the availability of an additional timing unit. In non RS environments, initial velar and dental voiced obstruents may be deleted. All voiced initial obstruents show an alternation between a spirantized (or sometimes deleted) word initial consonant and a geminated obstruent in a RS environment (12).3 (12) Initial Voiced C Mutation a. e [bb]atténe 'e vattene' [v]âttene Vattene' b. ti [dd]ongo 'te lo do' tu me [r]aje 'tu mi dai' c. a [y ]atta/a atta 'la gatta' e[g g]atte 'le gatte'
'and go + you + away 'go + you + away' 'I give you it (n.)' 'you give me' 'the cat' (f.) 'the cats' (f.pl)
Outside of the RS environment, a single voiced obstruent surfaces with a spirantized articulation, giving us the distribution in (13). (13)
Voiced Obstruent Distribution RS Elsewhere bb V dd r gg
3
Y
In a non-RS word initial position, the velar voiced obstruent tends to delete before [a] and [u]: a umma (sg.) -vs- e [gg]umme (pl.) 'la gomma'/'the eraser(s)'. In my corpus, the voiced dental can also delete before [a]: ammenne 'dammeneV'give me some' -vs- ti [ddjongo 'te lo doVT give you it (n.)'.
56
BARBARA E BULLOCK
The distribution in (13) is categorical, posing the problem of which voiced obstruent, stop or fricative, is underlying. It must be pointed out that /v/ and /r/ are phonemes in Neapolitan, each which geminate without any featural change in initial position under RS: se mette a [rr]irere 'si mette a ridere' 's/he begins to laugh'. Fanciullo (1986) argues for the underlying stop in word initial position. An underlying stop would imply that the voiced obstruent series undergoes lenition intervocalically and in inital position outside of the RS environment. A fricative representation would mean that RS is a process of fortition. This is the contention of Andalò & Bafile (1991) who argue that the lenited series constitutes the underlying forms because younger Neapolitans are beginning to generalize the fricative articulations and geminate them in RS: e [vv]enuto 'he came' along wih e [bb]enuto. Categorical alternations such as these provide the critical data that make the notion of underspecification (radical, redundant, or contrastive) indispensible to phonological theory (Inkelas 1994). In fact, the distribution of the voiced stops in Neapolitan is entirely predictable; they only appear in consonant clusters or when geminated. No singleton voiced stops surface intervocalically nor do they appear in word initial position unless following a consonant in the previous word or when they occur in an RS environment. While an in-depth analysis of the voiced obstruent series in Neapolitan is beyond the scope of this paper, I would like to advance the notion that this is an area where a phonetic, gestural approach to phonology along the lines of Kirchner (1997) might prove to be fruitful. That is, it appears that the production of a voiced stop, as opposed to a voiced fricative, in Neapolitan depends upon a sustained consonantal gesture. Lacking the ability (i. e., timing) to prolong such a gesture, the resulting occlusion is incomplete to the point that in fast speech, there may be no closure at all and the consonant deletes. A gestural approach to the problem of voiced consonant gemination/lenition in Neapolitan is not at odds with the analysis of RS presented here. In fact, this analysis sees the process of RS as providing the extra timing slot (mora) that would allow for word initial voiced stops. 4.
Conclusion This analysis of Neapolitan has demonstrated that the process of RS in this dialect occurs at the intersection of syntax, morphology, and phonology. This morpho-syntactic approach is based on Selkirk's insight that function words display unique phonological behaviors because they can be prosodically distinct in the grammar. This prediction is borne out in Neapolitan by the fact that only certain, lexically determined function words are prosodized with final bimoraic syllables. These, in turn, produce consonant gemination. Lexical words and
CONSONANT GEMINATION
57
function words parsed only as monomoraic syllables fail to produce the same effect in the grammar. A strong implication of this analysis is that historically dominant aspects of a language's prosody can become encoded lexically as abstract phonological properties. In Neapolitan, the function words that provoke RS can almost uniformly be traced back to Latin etymons that were consonant final. The etymological trace of the final bimoraic syllable in these function words remains, giving these words a kind of double prosodic structure that is manifested, much like the h-aspiré in French, only in a syntactically determined sandhi environment.
REFERENCES Andalò, Adriana & Laura Bafile. 1991. "On some morphophonological alternations in Neapolitan dialect". Bertinetto, et al. 1991. 247-257. Bertinetto, Pier Marco, Michael Kenstowicz & Michele Loporcaro, eds. 1991. Certamen Phonologicum II. Torino: Rosenberg & Sellier. Bichelli, Piro. 1974. Grammatica del dialetto napoletano. Bari: Pegaso. Beckman, Jill, Laura Walsh Dickey & Suzanne Urbanczyk, eds. 1995 University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory. Amherst, Mass.: Graduate Linguistic Student Association. Bullock, Barbara E. 1991. "Mora-bearing consonants in coda position and related quantity effects". Bertinetto, et al. 1991. 105-20. . ms. "Double prosody and stress shift in Proto-Romance". The Pennsylvania State University. Fanciullo, Franco. 1986. "Syntactic reduplication and the Italian dialects of the Centre-South". Journal of Italian Linguistics 8.67-104. Iannucci, James E. 1948-49. "Gemination of initial consonants and its semantic function in Neapolitan". Romance Philology 2.237-239. Inkelas, Sharon. 1994. "Consequences of optimization for underspecification". ROA-40-1294 (Rutgers Optimality Archive http://ruccs.rutgers.edu/roa.html). Kenstowicz, Michael. 1991. "Enclitic accent: Latin, Macedonian, Italian, Polish". Bertinetto, et al. 1991.173-186. Kirchner, Robert. 1997. "Geminate inalterability and lenition". ROA-249-1398 (Rutgers Optimality Archive http://ruccs.rutgers.edu/roa.html). Mester, Armin. 1994. "The Quantitative Trochee in Latin". Natural Language and Linguistic Theory 12.1-61. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in a Generative Grammar. Ms., Rutgers University and University of Colorado, Boulder.
58
BARBARA E. BULLOCK
Repetti, Lori. 1991. "A Moraic analysis of Raddoppiamento Fonosintattieo". Rivista di Linguistica 3.307-330. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialettiI:Fonetica. Torino: Einaudi. . 1969. Grammatica storica della lingua italiana e dei suoi dialetti II: Morfologia. Torino: Einaudi. Selkirk, Elisabeth. 1995. "The Prosodic Structure of Function Words". Beckman, et al. 1995.439-471. Urbanczyk, Suzanne. 1995. "Double Reduplication in Parallel". Beekman, et al. 1995. 499-532. Vogel, Irene. 1982. La sillaba come unità fonologica. Bologna: Zanichelli.
THE FEATURE [ADVANCED TONGUE ROOT] AND VOWEL FRONTING IN ROMANCE*
Andrea Calabrese University of Connecticut, Storrs
0.
Introduction As its name implies, the feature [Advanced Tongue Root] controls the advancement of the tongue root. The contrast between sounds with and without ATR is used in the vowel harmony of a great many West African languages. (See Stewart (1967), Halle & Stevens (1969), Perkell (1971), Ladefoged et al. (1972), Lindau (1978, 1979), Ladefoged & Maddieson (1990), Archangeli & Pulleyblank (1994), Vaux (1996a, 1996b).) This feature has also been used to describe the contrast between tense and lax vowels in Romance languages (Hualde 1989, 1992; Polgardi 1998). In Calabrese's (1988, 1999) analysis of metaphony in southern Italian dialects, this feature plays a fundamental role in characterizing the high vowels that result from the raising of [-ATR] mid vowels as marked and therefore the targets of simplification processes. High [-ATR] vowels are indeed marked in African and other languages, as shown by their phonological behavior and by the fact that they tend to be eliminated in historical changes differently than other vowels, such as [+ATR] /i, u/ (Archangeli & Pulleyblank (1994), Calabrese (1995), Vaux (1992, 1996a, 1996b)). In this paper I argue that if we assume that this feature does indeed play a role in the vowel system of the Romance languages, we can account for the peculiar sound change in (1) which occurred in several Romance varieties, the most well-known of which is French. Observe that the sound change in (1) is context-free, in particular it is not conditioned by adjacent front vowels or coronals. Previous explanations of this change, based on substratum theory I am thankful to Klaus Abels, Morris Halle, Lori Repetti and an anonymous reviewer for comments and suggestions. The usual disclaimers apply.
60
ANDREA CALABRESE
(Ascoli 1881) and structural pressure due to independent sound changes, besides being outdated, are quite unsatisfactory.1 There is no generative study of this change, and it usually is considered to be an 'unnatural' sound change that defies explanation (cf. Dressler (1974)). (1)
/u/
[y]
This paper deals with the southern Italian dialect of Altamura, which displays a sound change similar to (1). This dialect is studied in Loporcaro (1988), from which all of the data presented here are taken. An account of the change in ( 1) will be proposed that crucially relies on the feature [+ATR]. Section 1.1 contains a detailed discussion of the Altamura dialect facts. As we will see, fronting of rounded vowels occurs only in stressed syllables which are either closed or in antepenultimate position. Stressed open syllables undergo diphthongization in this dialect. Fronting fails to occur in diphthongs. Fronting also fails to occur after consonants such as labials (for [o] and [u]) and velars (only for [u]). In both cases, an independent process applies which assigns a secondary labio-velar articulation to the consonants. Section 1.2 discusses phonological alternations which illustrate some of the processes which are relevant in the analysis of the Altamura dialect facts. These alternations also provide evidence for assuming that there is a process which fronts underlying back vowels. Section 2 outlines the derivational constraint-and-repair approach of Calabrese (1995, forthcoming) which includes a traditional rule system and a UG constraint module. Section 2.1 introduces the notion of marking statements, the constraints contained in UG. Section 2.2 deals with the workings of the repair component. The idea is that repairs are implemented freely, but then checked by an evaluation mechanism based on an overarching principle of economy.
1 The most plausible analysis is in Liidtke (1957) where fronting in Old French is considered. According to this explanation the structural pressure due to the creation of the diphthong [ow] from the velarization of P] in coda position lead to the monophthongization of the original diphthong [ow] to [ø]. This last change according to Liidtke then lead to the fronting of [u] to [y]. Liidtke does not provide any clear reasons why the monophthongization of [ow] should lead to the fronting of [u]; there is simply no motivation for it. There are other obvious problems with Lüdtke's explanation. First, it is unclear why the introduction of the new diphthong [ow] should lead to the change of [ow] to [ø] and not just to the merger of the two diphthongs. Secondly, the diphthong [ow] is found only in open stressed syllables in penultimate position. The question then arises of why [u] is fronted in all positions in Old French. On accounts based on the substratum theory see Rohlfs (1966), Tekavcic (1972). See §4 for a brief discussion of fronting in Old French.
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
61
Section 3 develops an analysis of the fronting process. The central claim is that this process can be explained if the vowels [u] and [o] which, unlike [c], undergo fronting, are [+ATR]. Advancement of the tongue root involves an automatic fronting of the tongue body which must be suppressed in back vowels. The consequence of this, I propose, is the marking statement *[+ATR, +back] characterizing back [+ATR] vowels as phonologically complex. In the dialect of Altamura this constraint is active, and triggers a repair that fronts back vowels. Crucially this constraint interacts with the independently needed constraint * [-back, +round] disallowing front rounded vowels. As we will see, the constraint * [-back, +roundl is deactivated only in the nucleus of stressed syllables, but not in other positions. The strictly positional deactivation of this constraint, in conjunction with the notion of geminate inalterability, will account for the distributional properties of the fronting process in this dialect. Section 4 concludes the article with a brief discussion of fronting in Old French. 1. 1.1
The Dialect of Altamura The Vowel System The surface vowel inventory of the Pugliese dialect of Altamura, shown in (2), displays both front and back rounded vowels,2 but these two sets of vowels are in complementary distribution. The high round back /u/ and the mid [+ATR] round back /o/ never occur in a stressed syllable which is either closed or in antepenultimate position. In contrast their front counterparts /y, ø/ occur only in these positions. This distributional asymmetry can be easily accounted for by assuming a process of fronting affecting underlying /u, o/ in these positions. This process produces the derivations in (3). However, the lower mid vowel /c/ does not undergo fronting (4). (2)
i e e
y ø
u o c a
2
The high vowels are phonetically realized as lax[I,U](Loporcaro 1988:29). I assume that this is due to a late phonological process (See Calabrese (1999) for discussion of this process.)
62
ANDREA CALABRESE
/u/: /'ursә/ --> ['yrsә] /'unna/ --> [ ynnә] /'tunnә/ ~> ['tynne] /dunne/ -> [ d j y n ә] /'luggJә/ -> ['lyggJә] /'frutta/ --> ['fryttә] /'bruttә/ --> ['brytta] /'suttsә/ - > ['syttsә] /'ruddjәnә/ --> ['ryddjәnә] /dә' unәnә/--> [dә' ynәnә] /an'nu әnә/ —> [an'ny әnә] /'sutәnә/ --> ['sytanә] /'nuvәlә/ --> ['nyvәlә] /'stu әnә/ - > ['sty әnә] /o/ /'tostә/ --> ['tøstә] /'omnia/ --> ['ømmә] /'loddjә/ - > [løddjә] /'lonnә/--> ['lønnә] /'nottә/ - > ['nøttә] /'nostә/ --> ['nøstә] /'torta/ --> f'tørtә] /'donna/ ~ > f'dønnә] /ra'kotta/ --> [rә'køttә] /'koma/ - > ['kørnә] /ak'koggJә/ ~> [ak'køggJә] /'kokkjә/ -> ['køkkiә] /'ssoltә/ --> ['ssøltә] /'komәtә / ~> ['kømәtә/ /'stodәkv/ --> ['stødvkә] /'okkJәrә/ --> ['økkJәrә] /'ossәrә/ —> ['øssәrә] /arrә'kordәnә/ -> [arrә'kørdәnә] /'stommәkә/ --> ['stømmekә] /ja'roffәlә/ --> [ja'røffәlә]
'bear' 'rush' 'round.ms.' 'June' 'July' 'fruit' 'ugly' 'dirty' 'rust' 'fast.3pl' 'bring.3pl' 'sweat' 'cloud' 'smoke' 'hardfem.' 'man' 'verandah' 'long.fem.' 'night' 'our.fem.' 'crookedfem.' 'Honorific.fem. (Donna)' 'Ricotta cheese' 'hom' ' gather, lsg.' 'couple' 'dissolved' 'comfortable' 'stupid fern.' 'eyes' 'bones' 'remember.3pl.' ' stomach, sg.' 'carnation'
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING (4) /c/: /'stoppa/ —> ['stoppa] /'dolt ә/ --> ['dolt ә] /'sordә/ - > ['sordә] /'tcttә/ --> ['totta] /ka'no ә/ - > [ka'no ә] / ә'nokkJara/ - > [ ә'nokkJәrә] /ka'no әnә/ --> [ka'no әnә]
*'stœppә *'dœlt ә *'sœrdә *'tœttә *ka'nœ ә *Ja'nœkkJәrә *ka'nœ әnә
63
'tow' ' sweet' 'deaf 'all .fem.' 'know' 'knees' 'know.3pl.'
In several consonantal environments fronting does not occur. First of all, there is no fronting of /u, o/ after labial consonants (5). (5) /'puttsә/ - > ['pwuttsә] /'fumana/ —> f' w umәnә] /fufәnә/ —> ['fwufәnә] /'portә/ --> ['pwortә] /'wostә/ —> ['wostә] /'foggJә/ - > [TwoggJә] /'spoggJә/ —> [*spwoggJә] /'morta/ ~> ['mwortә] /'moddә/ —> ['mwoddә]
'well' 'smoke.3pl.' 'run away.3pl.' 'door 'your.fem.' 'greens' 'undress. lsg.' 'dead.fem.' 'wet.fem.'
In addition, there is no fronting of [u] after a velar consonant (6). The [+ATR] mid [o], however, does front in this position: [rә'køttә], [ak'køggjә]. (6) /'kuddә/--> ['kwuddә] /'ngutanә/ —> ['ngwutәnә]
'that.mas.' 'smoke.3pl.'
In the examples in (3)-(5) a coronal is adjacent to the fronted vowels; one might therefore propose that fronting is due to the coronal consonant. Vowel fronting due to adjacent coronal consonants is quite common (Clements (1991), Clements & Hume (1995), Halle, Vaux & Wolfe (2000)). This is not the case for the Altamura dialect. Although there are no cases showing [y] in a context where it is not preceded or followed by a coronal, 3 there are many cases where [o] is fronted in this context.
3
Latin forms such as UMIDU, ULTIMU, where, etymologically, we find [u] not preceded by a coronal, developed into this dialect as ['jymәtә], ['jyltәmә] with an epenthetic glide before the word-initial vowel.
64
ANDREA CALABRESE ['kømәtә] ['køppәlә] ['øbbrә] (in [fe 1 'øbbra]) ['ømmә] ['skøkkә] [vәrmә'køkkә]
'comfortable' 'beret' 'make noise' 'man' 'bark' 'apricot'
A secondary labio-velar articulation of the preceding consonant is found in non-fronting environments such as those in (5) and (6). Observe, however, that this secondary articulation is found also in labials before the f-ATR] back [c], which never fronts. (8) /t ә'pcddә / --> [tfә'pwoddә] /'mokka/ --> ['mwckkә]
'onion' 'mouth'
As implied earlier, there is no fronting of rounded vowels in unstressed syllables. As the last two examples in (9) show, rounded vowels in pretonic position become high. (In posttonic position all vowels except /a/ are reduced to schwa.) (9) /sul'dεjtә/ - > [sul'dεjta] / ur'nεjtә/ - > [ ur'nεjtә] /tu'towrә/ --> [tu'towra] /lun'dεjn/ —> pun'dεjn] /assu'kws/ —> [assu'kwε] /fu'sijma/ ~> [fu'sijmә] / b'kwεjtә / --> [ u'kwεjtә] /tscm'bwε/ —> [tsum'bwε]
'soldier' 'day 'tutor' 'far' 'dry.inf *spill. 1pl.' 'play.2pl.' 'jump.inf
In stressed open penultimate syllable, rounded vowels are diphthongized, as are all other vowels. No fronting of vowels or secondary rounding of the preceding consonant occurs in this case, as shown in (10). Loporcaro's discussion of the diphthongization process is somewhat unclear. He claims that diphthongization occurs only in stressed open penultimate syllable at the end of a phonological phrase (before a pause, in his terminology). He provides the strings in (11) in support of his claim. (Prepausally schwas are deleted).
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING (10) /'sors/ —> ['sowr] /'rota/ --> ['rowtә] /'kors/ - > ['kowrә] /'bona/ --> ['bownә] /'kofә/ ~> ['kowfә] /' oka/ - > [' bwkә] /'fumә/ ~> ['fuwmә] /'mute/ - > ['muwlә] /'lunә/ ~> [luwnә] /'lu ә/ --> ['luw ә] /dә' unә/ --> [dә' uwnә] /an'nukә/ ~> [an'nuwkә]
'sister' 'wheel' 'heart' 'good.fem.' 'cook.3sg' 'play/3sg.' 'smoke' 'mule' 'moon' Tight' 'fast' 'bring. 1sg'
(11) 'kjinә 'kjijn## 'sulә 'suwl## 'bbona 'bown## 'neve 'nejv## 'kjεnә 'kjejn##
elative form of... 'full.ms.sg. (Ml full)' 'only.mas.sg. (only only)' 'good.fem.sg. (good good)' 'new.mas.sg. (new new)' 'slow.mas.sg. (slow slow)
65
Diphthongization in most other Romance varieties has a close relationship with vowel lengthening. Vowels lengthen in an open stressed syllable in penultimate position at the word level regardless of their position in the phrase. Long vowels may then diphthongize. This is the case of lax mid vowels in Italian, for example, which diphthongize in this position, whereas all other vowels lengthen: [Vεngo]/[Vjεne] T come/he comes' vs. ['kre:do], ['di:ko], [Tu:mo]. This is also the case in the other northern Italian dialects which are characterized by generalized vowel diphthongization (cf. Rohlfs 1966). The fact is that in the Altamura dialect, as in other Romance varieties, lengthening does occur in an open stressed syllable in penultimate position regardless of phrase position (cf. Loporcaro 1988:181). We should then also expect to find diphthongization in this position. The alternation in (12) is evidence for the expected diphthongization process; Loporcaro (1988:182) states that the alternation in (12), as opposed to that in (11), occurs in all types of phrase positions. (12)
ne'pawtә/ne'pwctәmә
' nephew/ my nephew'
The best account for this alternation is the following. The underlying vowel in such alternations is [c]; it lengthens because of the process descriptively stated in (13).
66
ANDREA CALABRESE
Once long, the vowel diphthongizes by changing the second component of the long vowel into a glide (14). (See below for more discussion.)
A further process that lowers [-ATR] vowels when followed by a coda glide, as in (14), produces [aw|. Once we assume the preceding analysis we can account for the otherwise mysterious alternation ['kjajn3/'kjijnә] 'full.fem./mas.'. The vowel underlying this alternation is [e|. In the mas.sg., it undergoes metaphony (cf. §1.2) which produces [i]. Diphthongization then gives [ij]. In the fem.sg., it does not undergo metaphony, only diphthongization. The vowel lowering rule mentioned above then gives [aj]. I propose that lengthening and diphthongization apply to all vowels in open penultimate stressed syllables at the word level. Loporcaro's spectrograms, although difficult to read, seem to indicate that there is some diphthongal movement in the formant structure of stressed vowels of open penultimate syllable in non-phrase final position (cf. in particular #6 on p. 160). The issue is what happens in phrase-final position. One could propose that phrase-final position is the locus for a further process of extra-lengthening. This process could be connected to the deletion of the final schwa which also occurrs in phrase final position (cf. Hayes 1989 on Old English /talә/ —> [ta:l], etc.). The alternations in (11) should then be reinterpreted as those in (15). (15) 'kjijnә 'kjij:n## 'suwlә 'suw:l## 'bbowna 'bbow:n## 'nejve 'nej:v## 'kjεjnә 'kjej:n##
elativc form of...
'full.mas.sg. (full full)' 'only.mas.sg. (only only)' 'good.fem.sg. (good good)' 'new.mas.sg. (new new)' 'slow.mas.sg. (slow slow)'
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
67
(My own experience with speakers of areas dialectally close to Altamura seems to confirm this analysis, but more careful phonetic studies should be done.) Phrase-final diphthongs are then characterized by extra-lengthening. Ï propose that in the extra-long diphthongs, the extra-lengthening leads to a perceptive enhancement of the glide component of the diphthong. This enhancement leads to the misperception of the contrasts in (15) as those in (11): the shortness of the glide component of the non phrase final diphthong in opposition to its length when it occurs phrase finally may lead to its misperception as non-existent. More research is needed to check this point. In the light of this reinterpretation of Loporcaro's data, I assume that diphthongization is a word-level phenomenon. The forms in (10) then display diphthongization at this level. 1.2
Some Phonological Alternations of the Altamura Dialect The alternations in (16) will allow us to clarify the different processes relevant to our analysis of the Altamura dialect. The diphthongs are the outcome of lengthening (19) in conjunction with the diphthongization process in (20-21) and of the morphophonemic rule of metaphony described in (18). The high back vowel [u] is the outcome of the rule of raising in unstressed syllables. In addition, the process of fronting is also active. (16) a. underlying lol: i. 'sowrә/'sørdә 'sister/your sister' ii. present of u'ka ' p l a y ' 4 1s. 2s 3s 'jowkә ' wejk ' owkә iii. present of rnu'ri ' die' 1s 2s 3s 'mowrә 'mwejrә iv. 'nøstә/'nwestә * our (fem/masc)' v. present of arrәkur'da ' remember' 1s 2S 3S arrә'kørdә arrә'kwerdә b. underlying /u/: i. present of an'nu ә 'bring' 1s 2s 3s an'nuwkә an'nuw ә an'nuw ә 4
1p u'kwejmә
2p u'kwejtә
3p ' økәnә
2p 'mowrә
3p 'mørәnә
2p arrә'kørdә
3p arrә'kørdәnә
lp
lp
lp 2p 3p annu' ejmә annu' e j t ә an'ny әnә
In the forms [fu'kwejmә], [fukwejtә] the velar stop acquires a secondary labial articulation due to an independent assimilation process triggered by ihe preceding high round vowel.
68
ANDREA CALABRESE
Let us consider the different processes involved in these alternations, beginning with metaphony. The structural changes characterizing this process in the Altamura dialect are provided in (17). They occur in morphological environments such as the masculine singular and plural of the adjectives of class II, the plural of class III adjectives, the second singular of the indicative present, etc.5 The metaphonic process in (17) can be expressed by the rules in (18) (feature tree simplified). (See footnote 5 for more discussion.) (17) a.
e --> i:
b.
s -> i
c.
o —> we
d
o --> u
(18) a.
X |
6
in certain morphological contexts
-->
XX \/
(cf. 17a)
[-high, -low, -back, +ATR]
[+high]
b.
X
X
[+high, +back]
[high,-low,-back]
X
-->
(cf. 17c)
||| [-high, -low, +back, +ATR] 5
Metaphony is a common sound change in southern Italian dialects. It can be reconstructed as involving height assimilation of stressed mid vowels before high vowels, and in fact high vowels historically underlie all the morphological categories triggering the metaphonic changes in (i), in conjunction with a process diphthongizing the [+high, -ATR] vowels obtained when mid [ A T R ] vowels are targeted, as shown in (i). (See Calabrese 1999 for an analysis. Alternative views are expressed by Kaze 1989, Maiden 1991, Cole 1999.) (i) e --> i / i, u ε--> I-->iε o --> u c --> u—>uc The synchronic alternations we observe in (i) are the outcomes of several historical changes which modified the original metaphonic alternations, such as reduction of final vowels and monophthongization of [is] in closed syllables. But the major change disturbing the original alternations was an exchange of [ATR] values in the mid vowels so that etymological [+ATR] [e] and [o] became [ATR] [ε] and [c] and etymological [ATR] [s] and [c] became [+ATR] [e] and [o], as in (ii). (ii) [αATR] --> [α ATR] / [ , -high, -low] By (ii) the natural basis of the metaphonic process in (i) was obscured and the alternations can only be expressed by the stipulatory rules in (18). 6 This long vowel has a special behavior: in open syllables it may shorten, triggering gemination of the following consonant, whereas in closed syllables and in antepenultimate position it always shortens without causing any change in the consonantal environment. (See discussion in Loporcaro 1988:37.)
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING c.
X || [high, -low, -ATR]
-->
X
69
(cf. 17b, 17d)
[+high, +ATR]
We can now turn to the process of diphthongization. As mentioned earlier diphthongization is related to lengthening. Lengthening is a complex phenomenon which requires a detailed discussion of the stress system of Romance which cannot be done here. For the sake of simplicity, I will assume the descriptive rule in ( 13), repeated as in ( 19). (19) N N I /\ X --> X X / I \/ +stress V V
The rule in (19) lengthens vowels in a penultimate open syllable, but not in closed or antepenultimate syllables. Diphthongization itself may be accounted for by assuming a constraint forbidding long vowels (20).7 This constraint triggers the repair in (21). By the repair in (21) the second component of the long vowel is assigned to coda position where it automatically becomes [+high]. Therefore, phonetically we obtain a sequence of vowel plus homorganic glide. For the sake of the exposition I will not go into details concerning this repair which involves discussion of various aspects of syllable structure. (Feature trees are simplified.) (20)
7
N /\ * X X \/ [-consonantal]
As discussed in Calabrese (forthcoming) and below, restrictions on the internal structure of a constituent should be expressed with constraints, whereas rules should be used to account for restrictions on the distribution of elements across constituents.
70
ANDREA CALABRESE
Finally we can assume the rule in (22) raising vowels in unstressed syllables. (22)
[-consonantal]
--> [+high] /
Let see how these different processes account for the forms in (16). First of all, observe that the alternations in (16) can be accounted for only if we assume that the segment underlying them is a back vowel. Consider the form in (16aii), where we observe an alternation between [ow/wej/u/ø]. If we assume an underlying /ø/, it would be difficult to account for the shape of the diphthongal [ow] or of the high vowel [u]. There is no such problem once we assume an underlying /o/: it becomes the diphthong [ow] by application of lengthening and diphthongization, and [u] by rule (22). The relevant derivations are shown in (23) and (24). (23)
' okә --> (19) --> ' o:kә --> (21) --> ' owke
(24) Jb'kemә -> (19+21) -> o'kejmә -> (22) -> u'kejmә -> (other rules) -> u'kwejmә
Crucially a three-step derivation is required in the case of the form [ wejk9]. As we will see in more detail later, a derivation involves the serial application of rules and repairs. The alternant [wej] is obtained by the application of the rule of metaphony in (18) followed by the application of lengthening and diphthongization as shown in (25). (25)
' okә --> (18) - > ' wekә - > (19) - > ' we:kә --> (21) --> ' wejkә
The change [o] -> [ø] is due to a repair triggered by the constraint *[+ATR, +backl discussed in §3 (see (33)). Crucially, as discussed in more detail
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
71
there, we have to order the application of this repair after the application of all the other rules discussed up to this point. 2.
Theoretical Background of Analysis The analysis of the Altamura dialect facts proposed in this paper is couched in the derivational constraint-and-repair model developed in Calabrese (1988, 1995,2000) and independently in Paradis (1988). The general architecture of this model is provided in (26). One of the fundamental properties of this model is that it is derivational: surface phonological representations are constructed step by step from an input underlying representation. As discussed in Calabrese (forthcoming), by assuming the notion of phonological derivation, more efficient, simpler and more elegant analyses can be proposed. Furthermore, this model assumes a markedness module containing universal constraints and repairs which adjust violations of these constraints. In addition to this module, there is a Phonological/Morphophonological component which contains a more traditional system of rules which can be ordered relative to each other and also with respect to the repairs in the markedness module. The idea is that not all modifications of a phonological string are due to constraint-induced repairs. They are instead implemented by rules, i. e., deterministic instructions to perform a specific structural change on a given structure.8 Crucially, these rules can be ordered among each other and also with respect to the repair of the markedness module. One of the fundamental concerns underlying the analyses proposed here is to build a theory of the phonological component that meets what Chomsky (2000) calls the minimal design specifications, i. e., a theory that tries to achieve explanatory adequacy by minimizing the theoretical apparatus. In trying to reach this goal of theoretical minimization, I will propose in §2.3 that an overarching principle of economy governs the phonological component controlling repair 8
As discussed in Calabrese (forthcoming), rules are the best means to account for processes characterized by changes where only a single output is conceivably possible, such as directional assimilation processes. Morphophonemic changes typically belong to this category as well. If a constraint is used to account for this kind of process, other outputs in addition to the correct one must be produced by the repair component. These other outputs must then be excluded by introducing other constraints or devices. In the case of these processes, this is an unwarranted complication that violates the basic principles of economy governing phonology. (See Calabrese (forthcoming) for more discussion.) Calabrese (forthcoming) observes that this type of process typically involves distributional restrictions on elements across constituents. Constraints appear to be instead the best means to account for the internal structure of constituents where there are typically different conceivable ways to achieve well-formedness. This is, for example, the case of restrictions on feature cooccurrence or on syllable structure. (See Calabrese (forthcoming) for more discussion.)
72
ANDREA CALABRESE
operations and determining their outcomes. (See Calabrese (forthcoming) for more discussion.) (26) Morphology Underlying Representations
Phonology/ Morpho-Phonology Rule Component
Markedness Module Constraints System: Repairs
Surface Representation
2.1
Marking Statements and the Phoneme 'Alphabet' Languages differ as to the phoneme 'alphabet', i. e., as to the feature complexes that may figure in their vocabulary items. Following Calabrese (1995) we can say that every speaker's innate linguistic competence — i. e., the uniquely human capacity that allows her/him to acquire a natural language -- includes a set of filters or marking statements, which assign a cost to specific configurations of features. An example of these marking statements is given in (27). (27)
*r-back,+roundl
(27) characterizes front round vowels as phonologically complex by assigning a cost to the use of the feature [+round] (underlined in (27)) in the context of the feature [-back]. In other words, (27) characterizes the use of the feature [+round] as 'marked' in front vowels. Simple and optimal feature combinations are those which have the opposite value for the marked feature. Segments characterized by a feature combination mentioned in a marking statement may occur in a language if and only if the relevant marking statement is
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
73
deactivated.9 If the marking statement is active, the relevant segment is ruled out. For example, in standard Italian, where front rounded vowels are absent, (27) is active. Languages differ from one another in deactivating certain marking statements but not others.10 For example languages with front rounded vowels deactivate the marking statement (27). This is the case of German, for example. (For further details, see Calabrese 1995.) Under this view, acquiring a language involves learning which marking statements are deactivated in this language.11 As proposed in Calabrese (1988,1995), marking statements are ranked in a universally fixed hierarchy indicating their different status in terms of markedness/phonological complexity. Thus, for example, the marking statement *[-high, -low] governing the presence of mid vowels in phonological systems is ranked much lower than the marking statement in (27). This indicates that the mid vowels, although phonologically more complex than the unmarked vowels [i, u, a], are much less complex than front rounded vowels, and therefore much more common across languages. Note that a marking statement does not need to be deactivated in all environments in a language. It is in fact well-known that many languages allow an increase in phonological contrast in certain linguistically prominent positions such as stressed or word/root-initial position, allowing the most marked members of the inventory in those positions (Steriade 1993, 1995, Beckman 1995). In my 9 In addition to marking statements, there are prohibitions. Prohibitions always exclude cooccurrence of the relevant feature pairs, expressing the acoustic/articulatory impossibility of a given feature combination. Obviously prohibitions can never be deactivated. An example of a prohibition is given in (i). (i) *[+high, +low] Segments characterized by a feature combination mentioned in a prohibition are obviously impossible and never occur in a phonological system. 10 What is active or not is an idiosyncratic aspect of a given language. There is no way of predicting the pattern found in a given language except in the cases predicted by the implicational relations between marking statements as discussed in Calabrese (1988, 1995). The pattern of deactivation of marking statements found in a given language is due to its historical idiosyncracies. 11 The marking statements also provide the basis for the definition of contrastive features. In particular, when a marking statement is deactivated, the marked feature and its opposite ate contrastive. Contrastive feature values distinguish minimally distinct pairs of segments and play an important role in the phonology of languages. In particular, phonological rules behave differently with regard to which features are visible to them. For some rules, only marked features are visible; for others the larger class of contrastive features is visible; for still another class of rules all types of features are visible. By making these distinctions among rules, it is possible to account for facts that are usually explained by relying on Underspecification Theory. (See Calabrese (1995) for more discussion.)
74
ANDREA CALABRESE
framework, such a situation can be accounted for by assuming that the deactivation of marking statements can be restricted to these positions. Under this proposal, a marking statement would be deactivated only in a given position, say in stressed syllables, but not in other positions, i. e., in unstressed syllables. Thus the marked segment which is blocked by that marking statement will appear in the former position, but not in the latter. For example, the African language Uwie displays a nine-vowel system /i, τ, e, s, a, c, o, u, u/ in roots, but not in suffixes where it is reduced to /i, e, e, a, c, o, u/. We can say that in this language the marking statement *[+high, -ATR1 is deactivated only in roots, but not in affixes.12 As we will see a crucial property of the dialect of Altamura is that the constraint in (27) is deactivated only in stressed syllables. Observe that the underlying inventory of segments may be different from that characterizing the output of the markedness module. To account for this difference, I assume, following Prince & Smolensky (1993), the so-called Richness of the Base Hypothesis, according to which underlying representations are not governed by constrains. In terms of the diagram in (26), they are outside of the control of the markedness module; thus there is no arrow leading from this module into the base component. The only goal of underlying representations is to allow a simple and general account of the lexical alternations characterizing a given language, and there is no need to have them checked by the markedness module. In the dialect of Altamura there are no front rounded vowels underlyingly; their distribution is in fact predictable. This does not mean however that the marking statement in (27) must be active underlyingly to be (positionally) deactivated later. What is present/absent underlying does not matter for the establishment of the patterns of marking statements deactivation characterizing a language. What matters are the output lexical forms of the language. 2.2
Repair Strategies Configurations which are marked as ill-formed by a given active constraint must be repaired. I propose that repairs are implemented freely by a single operation that I call REPAIR. Obviously, given a structure containing an illformed configuration, only the ill-formed configuration should be affected by REPAIR, nothing else. This property follows from a general economy principle that tries to minimize cost in all the different aspects of the phonology. In this case, assuming that phonological manipulations have a cost attached to them, this economy principle requires minimal application of phonological manipulations. Therefore we have the following corollary of this economy principle. 12
The positions allowing this local deactivation may be characterized by properties that support the appearance of marked features (Steriade 1995, Beckman 1995).
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING (28)
75
Last Resort: Do not implement any modification of the phonological string unless it is required.
(28) governs the free manipulations implemented by REPAIR and characterizes them as last resort operations applying only to ill-formed configurations. Obviously (28) does not affect the rules of the phonological component because by definition they involve deterministic instructions to modify a given string. REPAIR consists of the free application of basic operations such as insertion and deletion. In the case of nonlinear segmental representations, we thus have the four basic situations shown in (29) and (30). (29) a. 0 --> F
b. F --> 0
(30) a. F --> F
b. F
G
G
G
REPAIR freely constructs all possible derivational combinations of these basic operations. Most of these derivations will be excluded by Last Resort. A subset of the derivations that survive Last Resort involves a single application of a basic operation such as line addition or removal. Other derivations will involve a sequential application of the basic operations. Sequential application of basic operations is allowed by Last Resort only if it involves a further REPAIR of a disallowed configuration. Each REPAIR then involves several parallel derivations. As mentioned earlier, most will be excluded by Last Resort. A majority of the surviving ones will produce ill-formed configurations, i. e., configurations disallowed by active marking statements. We can say that they 'crash'. A few will produce wellformed configurations. I will call these 'convergent derivations'. The outputs of convergent derivations and the derivations themselves are then evaluated by the Evaluation component: the best output according to the Evaluation component is chosen. The Economy principle evaluates candidate repairs with respect to the number of phonological modifications they implement: it requires that repairs must be minimal, and it governs how minimal they can be. Candidates in the Evaluation component are evaluated according to the Economy Principle (31). The cost of a given configuration is determined by markedness, by its derivational history, or by other means discussed below.
76
ANDREA CALABRESE (31)
Economy Principle: Given two candidate forms A and B, choose the form which is less costly.
As discussed in Calabrese (forthcoming), markedness and derivational history are not able to choose the correct candidate by themselves, and some other device is needed. Consider the case in which both the output repairs [aB, -bG], [-aB, bG] of a disallowed configuration *[aB, bG] are convergent and equivalent in terms of markedness. We can choose between them by assigning a certain value to one of the features of the input configuration. I call this feature the prized feature. The prized feature needs to be preserved in the input, otherwise a cost is assigned to its modification. Suppose that [aB] is the prized feature in the repair of the disallowed configuration *[aB, bG], therefore the output [aB, -bG] is chosen in the Evaluation component because it is less costly than output [-aB, bG] where the prized feature [aB] is modified.13 Remember that at every stage in the phonological derivation, the string is checked for well-formedness by the markedness module. If there is a violation of an active marking statement, repair procedures are invoked that remove the violation. However, as mentioned before, these repairs do not apply automatically, but can be ordered with respect to rules or other repairs. This is precisely what occurs in the dialect of Altamura where the repair fixing [+ATR, +back] vowels must be ordered after the rules discussed in § 1.2. 3.
Fronting in the Dialect of Altamura We can now turn to the process which fronts [u, o]. First we need a feature which is common to [u] and [o], and distinguishes them from [c] (remember that lower mid [c] is not fronted). This feature must be an appropriate trigger of vowel fronting. If we hypothesize that the distinction between higher and lower mid vowels involves the feature [ATR], we will have the following feature assignment for vowels.
13 We expect cases involving complex interactions among considerations of markedness, derivational economy, structural complexity and feature pricing. (See Calabrese (forthcoming) for discussion.) Observe that the function of feature 'pricing' is similar to that of faithfulness constraints in Optimality Theory in assigning a cost to given modifications of the input (Kager 1999:55). In the model presented here cost is assigned directly to the lexical inputs and the effect of the faithfulness constraints of disallowing these modifications by ranking is derived from the general principle of economy as discussed in the text above. We can thus dispose of faithfulness constraints and restrict the constraints used by grammatical apparatus only to the markedness constraints, thus achieving a substantial simplification of the theory. (See Calabrese (forthcoming) for more discussion.)
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING (32) high
i +
low back round ATR
e
£
a
0
0
-
-
-
+ +
+
+
-
+
+
-
-
-
-
+ +
+ +
+ +
+ +
-
-
u +
0
77
y +
+ +
The feature [ATR] is shared by the high /u/ and the higher mid /o/, as well as /ø, y/. This is the feature triggering the fronting. Specifically, I would like to propose that this fronting is a repair due to the marking statement in (33), which characterizes the use of the feature [+back] in the context of the feature [+ATR] as marked. (33)
*[+ATR, +backl
There is a natural motivation for this constraint involving the mechanics of tongue root movement. Lindau (1975:30) first observed that advancing the tongue root tends to push the tongue body up and forward (cf. also Archangeli & Pulleyblank 1994), as schematized in (34) (from Vaux 1992). (34)
[+ATRJ --> fronting and raising
tongue root advanced
tongue body pushed up and forward
Therefore in pronouncing [+ATR] back vowels speakers need to suppress the natural tendency to front them. The need for this suppression makes the configuration [+ATR, +back] articulatorily complex and therefore phonologically marked. Evidence for the marking statement in (33) is provided by the historical development in which [+ATR] back vowels in ATR systems are fronted so that from a protosystem with [ATR] oppositions as in (35a) we get the system in (35b) where the [+ATR] vowels are also fronted (Vaux 1992,1996a, 1996b). (/A/ refers to a [+low, +ATR] vowel.)
78
ANDREA CALABRESE (35)
a,
b.
original ATR system i e A o I s a c
u u
derived ATR/back system i e æ ø ï e a o
y u
Consider for example the Somali vowel system. Somali has two series of vowels contrasting in terms of the feature [ATR]. Interestingly the [+ATR] vowels are also fronted, as shown by the [+ATR] [u], [o] and [a] which surface as [y], [ø] and [æ], respectively (Anteil, et al. 1974:38). (36)
[+ATR] i e
[ATR] I s
ø y æ
o u a
Recent phonetic work has shown that the Tungusic vowel systems are based on ATR oppositions and employ ATR harmony, rather than [back] harmony (Svantesson 1985, Ard 1980). Rialland & Djamouri (1984) and Svantesson (1985) have provided similar evidence that various Mongolian languages also employ ATR harmony. Crucially [+ATR] vowels are also relatively front in both groups of languages. The fronting we observe in these languages can be explained by assuming that the marking statement in (33) is active and that a configuration violating it is repaired as in (37). (The feature [+ATR] is the prized one.) (37)
Input: *[+ATR, +back] (prized feature: [+ATR]) Repairs allowed by Last Resort: a. [+ATR, -back] (=[y]) b. [-ATR,+back](=[u])
The evaluation component chooses the output in (37a) because it is less costly in so far as the prized feature [+ATR] has not been changed. By the repair in (37), the vowels /u, o/ are fronted into [y, ø].
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
79
Further evidence for a correlation between the feature [ATR] and the feature [back] is provided by Adjarian's Law in Armenian (Vaux 1992, 1998). Vaux shows that in several Armenian dialects, for example in the dialect of Van, round back vowels are fronted after voiced stops. Classical Armenian 'spade' bah 'knife' danak 'sheep' gam
Van päx tänäk kjär
buk-h durs go*
' snowstorm' 'out' 'thief
pykh tys kjøB
pat tasn kartf
'wall' 'ten' ' short'
pat tas kartf
Vaux accounts for this change by assuming that voiced stops are [+ATR]. 14 In the case of Adjarian's Law, we can assume that the feature [+ATR] of the obstruents was spread to the following vowel, as in (39). (39) [+cons]
[cons]
[+ATR]
Fronting is then due to the active marking statement in (33) and the repair in (37). Obviously the constraint in (33) cannot be assigned a high degree of complexity and must be ranked very low among the UG Marking Statements. In fact it can be easily deactivated, like the marking statements governing the presence of mid vowels or of mid [+ATR] vowels. Thus back [+ATR] vowels are common across languages. It can however be active in some languages, as in Somali and the 14 There is strong phonetic evidence supporting the notion that voiced obstruents are [+ATR]. Phoneticians have long known that an expansion of the pharyngeal cavity, implemented primarily by lowering of the larynx and advancing of the tongue root, which lowers the supraglottal pressure sufficiently, is required in the case of stops to allow vocal cord vibration. (For more discussion of this issue, see Vaux 1996b.)
80
ANDREA CALABRESE
other languages discussed above. This is also the case of the dialect of Altamura. Crucially in this dialect, as argued in §1.2, [+ATR, +back] vowels are present underlyingly, as allowed by the Richness of the Base Hypothesis mentioned in §2.1. The constraint in (33) is active only in the phonological component. Crucially the repair it triggers is ordered after lengthening, diphthongization, and application of the rules in (43) and (44). The disallowed configuration [+ATR, +back] is then repaired as in (37) which fronts the vowels /u, o/ to [y, ø]. The crucial feature of this aspect of the phonology of this dialect is then the status of the marking statement in (27) forbidding front rounded vowels. There is in fact an interaction between the constraint in (27) and the constraint in (33). Constraint interactions in the model used here are governed by the principle in (40) which determines when a given constraint must be inactive by inspecting the lexical input to see what features are prized and by inspecting the grammar to see which other constraints are active. (40)
A constraint [aF, -bG] cannot be active if there is an active constraint [bG, gZ], and if the features aF, gZ are prized.
The effect of principle (40) is that of preserving underlying or previously created configurations by banning costly derivations which are either vacuous or involve changes affecting prized features. In this case, deactivation, or blocking the activity of a constraint, is the preferred choice. (See Calabrese (forthcoming) for more discussion of this principle, called there the Structure Preservation Principle, and for an extension of (40) to constraints dealing with syllable structure.) In the case of the dialect of Altamura we are dealing with the two constraints *[+ATR, +back] and *[-back, +roundl. We know that the feature [+ATR] is prized because of the repair in (37). Let us suppose that the feature [+round] is also prized. This constraint interaction therefore satisfy the conditions of principle (40). Constraint *[+ATR, +back] cannot be active unless *[-back, +round] is deactivated. This is what occurs in the dialect of Altamura, but with one restriction: the marking statement in (27) is deactivated only in the nucleus of stressed syllables. Therefore, in stressed syllables, the marking statement in (33) can be active, and trigger the repair that produces a front rounded vowel. In all other positions, however, the constraint against front rounded vowels remains active. Therefore, the constraint in (33) must be deactivated and thus [+ATR, +back] vowels are allowed. This immediately explains why [+ATR] back vowels are not fronted in stressless syllables. Since (27) is active in that position, the constraint in (33) must be deactivated. Therefore, the fronting inducing repair in (37) does not occur.
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
81
Let us now consider the other cases where fronting does not occur, i. e., cases in which the back vowel is either diphthongized or preceded by a secondary articulation. Consider first the diphthongs. Diphthongization produces structures such as those in (41).
In the structure in (41), the feature bundle of [u] is simultaneously linked to the nucleus and coda positions. This situation is identical to the one we encounter in cases of geminate inalterability where a configuration which is the target of a rule or of a constraint appears in a linked configuration. Following Schein & Steriade (1986), we can account for this type of situations by assuming the principle in (42) (a modified version of what they call the Uniform Applicability Condition).15 (42)
Given a node n and a set S consisting of all nodes directly dominating n, and given a rule or a constraint T, if T refers to n and any member of S, it must refer to all members of S to be active.
It follows that for a constraint to be active in a given position, it must be exhaustively satisfied in all of the positions linked to that position as required by the Uniform Applicability Condition. In (41) the feature bundle of [u] is linked to a position, the coda, where (33) must be deactivated.16 Therefore given the Uniform Applicability Condition, the constraint in (33) cannot be active in the feature bundle of the nuclear [u]. This explains the special behavior of diphthongs. We can account for the absence of fronting after the secondary articulation in the same way. The two rules of assimilation informally given in (43) and (44)
15 I refer the reader to Schein & Steriade (1986) for a criticism of Hayes' (1986) Linking constraint which the Uniform Applicability Condition replaces. 16 As discussed above, according to (40), (33) can be active only in the nucleus of stressed syllables where (27) is deactivated.
82
ANDREA CALABRESE
(see footnote 17 for a correct formulation of the rules)17 create structures where the feature bundles of [u] and [o] are simultaneously linked to nucleus and onset position. As before, the constraint in (33) cannot be active in their feature bundles 17 Assuming Halle, Vaux & Wolfe's (forthcoming) model of feature geometry and feature spreading, where only terminal feature spread, the two rules in (43) and (44) would be formally expressedas follows. (See Halle, Vaux & Wolfe (forthcoming) for the formal underpinning of the operations in (τ)-(ττ)).
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
83
because of the Uniform Applicability Condition since it does not hold for al! the position to which these feature bundles are linked.
We thus have an explanation for the various properties of fronting in the dialect of Altamura.18
18 We also have to account for the reason the labiovelar secondary articulations appear only in closed syllables and in antepenultimate position, i. e., whenever there is no diphthongization. I propose that the rules apply before stressed vowels in all positions. However, the constraint in (i) triggers the repair in (ii) by which the first link of a feature already associated to two different timing positions is delinked.
84
ANDREA CALABRESE
4.
Fronting in Old French Before ending this paper, I would like to consider briefly the difference between the Altamura dialect and Old French in which context-free fronting occurred as a historical sound change. In contrast to the dialect of Altamura, fronting in Old French, involved only [u], not [o], and applied in both stressed and unstressed positions (45). (45) 'duru > dyr 'fuste > fyst 'bruma > brym du'rare>dyrer pur'gare > pyrdgier
The higher mid [o] was not fronted. (46)
'tottu > tot (> later tut) tor'mentu > tormant (> later turmant) dobi'tare > doter (> later duter) or'tica > ortie (> later urtie)
First let us assume that, as in the dialect of Altamura, the process of fronting is due to the activation of the constraint in (33). Crucially we have to assume that Proto-French fu] is characterized by the feature [+ATR]. Now, we can account for the difference between the Altamura dialect and French by assuming that these two languages resort to different patterns of positional deactivation. In particular, whereas in the dialect of Altamura the constraint *[-back, +round] is deactivated only in stressed syllables, in Old French it was deactivated in all types of syllables but only in the case of high vowels, i. e., in the environment [ +high]. 19 This immediately accounts for the Old French situation. 5.
Conclusion We have seen that an account for the mysterious sound change in ( 1) can be found once the role of the feature [+ATR] in high and higher mid back vowels is considered. As we have seen, the marking statement *[+ATR, +back]
19 The higher mid [o] in open syllables eventually became fronted to [ø] through a diphthongal stage [ou]. We can assume that there followed stages in which the diphthongal [u] was fronted and then monophthongization occurred, as shown in (i). (i) o: -> ou --> oy --> 0
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
85
characterizes back f+ATRJ vowels as phonologically complex. If this constraint is active, these vowels are repaired by fronting. Interestingly, in the dialect of Altamura fronting does not apply across-theboard to [+ATR] back vowels in all positions. It occurs only in the nucleus of stressed closed syllables or open syllables in antepenultimate position; in both cases fronting occurs only when the back vowel is not preceded by labials (for /o/ and /u/) or velars (only for /u/). This complex distribution of the fronting process was accounted for by assuming that in this dialect the constraint against front rounded vowels (27) is deactivated only in the nucleus of stressed syllables. Given principle (40), wherever this constraint is active, the marking statement [+ATR, +backj must be deactivated, and therefore there is no fronting. This immediately accounts for the failure of fronting in unstressed syllables. The independently needed rules of diphthongization (20) and of secondary articulation assignment (43)-(44) create linked structures between the back vowels and syllabic margins where (27) is active, and therefore (33) must be deactivated. Given the Uniform Applicability Condition in (42), if (33) cannot be active in one part of a linked structure, it cannot be active in the other part of this structure. This accounts for the failure of fronting in the other environments. REFERENCES Anteil, Stephen, et al. 1974. "Nilo-Saharan Vowel Harmony from the Vantage Point of Kalenjin". Research Notes from the Department of Linguistics and Nigerian Languages 6.1 -58. Archangeli, Diane & Douglas Pulleyblank. 1994. Grounded Phonology. Cambridge, Mass.: MIT Press. Ard, Joshua. 1980. "A Sketch of Vowel Harmony in the Tungusic languages". Studies in the Languages of the Soviet Union ed. by Bernard Comrie, 23-43. Edmonton: Linguistic Research, Inc. Ascoli, Graziadio Isaia. 1881. Una lettera glottologica pubblicata nelVoccasione che raccoglievasi in Berlino il Quinto Congresso Internationale degli Orientalisti. Torino: Loescher. Beckman, Jill. 1995. "Positional Faithfulness, Positional Neutralization, and Shona Vowel Harmony". Phonology 14.1-46. Calabrese, Andrea. 1988. Towards a Theory of Phonological Alphabets. Ph.D. dissertation, MIT. . 1995. "A Constraint-Based Theory of Phonological Markedness and Simplification Procedures". Linguistic Inquiry 26.373-463. . 1999. "Metaphony Revisited" Rivista di Linguistica 10.7-68.
86
ANDREA CALABRESE
. 2000. "Gemination, Glide Formation and Sievers' Law in Vedic Sanskrit". Ms., University of Connecticut. . forthcoming. "Markedness and Economy in a Derivational Model of Phonology". Ms., University of Connecticut. Chomsky, Noam. 2000. "Minimal Inquiries". Studies in Honor of Howard Lasnik ed. by D. Michaels. Cambridge, Mass.: MIT Press. Clements, George. 1991. "Place of Articulation in Consonants and Vowels: A Unified Theory". Working Papers of the Cornell Phonetics Laboratory 7.3576. Clements, George. & Elizabeth Hume. 1995. "The Internal Organization of Speech Sounds". The Handbook of Phonological Theory ed. by Jonathan Goldsmith, 245-307. Cambridge, Mass.: Blackwell. Cole, Jennifer. 1999. "Deconstructing Metaphony". Rivista di Linguistica 10.6998. Dressier, Wolfgang. 1974. "Diachronic Puzzles for Natural Phonology". Natural Phonology ed. by Anthony Bruck, Robert A. Fox, & Michael W. La Galy, 95-102. Chicago: Chicago Linguistic Society. Halle, Morris & Kenneth Stevens. 1969. "On the Feature [Advanced Tongue Root]". Quarterly Progress Report (Research Laboratory of Electronics, MIT) 94.209-215. Halle, Morris, Bert Vaux & Andrew Wolfe. forthcoming. "Feature Spreading and Vocalic Place". Linguistic Inquiry 31.2. Hayes, Bruce. 1986. "Inalterability in CV Phonology". Language 62.321-51. . 1989. "Compensatory Lengthening in Moraic Phonology". Linguistic Inquiry 20.253-306. Hualde, José I. 1989. "Autosegmental and Metrical Spreading in the Vowel Harmony Systems of Northwestern Spain". Linguistics 27.773-805. . 1992. "Metaphony and Count/Mass Morphology in Asturian and Cantabrian Dialects". Theoretical Analyses in Romance Linguistics ed. by Christiane Laeufer & Terrel A. Morgan, 99-114. Amsterdam & Philadelphia: John Benjamins. Kager, René. 1999. Optimality Theory. Cambridge: Cambridge University Press. Kaze, Jeffrey. 1989. Metaphony in Italian and Spanish Dialects Revisited. Ph.D Dissertation, University of Illinois. Ladefoged, Peter, et al. 1972. "An Auditory Motor Theory of Speech Production". UCLA Working Papers in Phonetics 22.48-75. Ladefoged, Peter & Ian Maddieson. 1990. "Vowels of the World's Languages". Journal of Phonetics 18.93-122.
THE FEATURE [ADVANCED TONGUE ROOT] AND FRONTING
87
Lindau, Mona. 1975. "Features for Vowels". UCLA Working Papers in Phonetics 30. . 1978. "Vowel Features". Language 54.541-63. -. 1979. "The Feature Expanded". Journal of Phonetics 1 A6i-76. Loporcaro, Michele. 1988. Grammatica storica del dialetto di Altamura. Pisa: Giardini. Liidtke, Helmut. 1956. Die strukturelle Entwicklung des romanischen Vokalismus, Bonn: Romanisches Seminar an der Universität. Maiden, Martin. 1991. Interactive Morphonology: Metaphony in Italy. London: Routledge. Paradis, Carole. 1988. "On Constraints and Repair Strategies". Linguistic Review 6.71-97. Perkell, Joseph S. 1971. "Physiology of Speech Production: A Preliminary Study of Two Suggested Revisions of the Features Specifying Vowels". Quarterly Progress Report (Research Laboratory of Electronics, MIT) 102.123-39. Polgardi, Krisztina. 1998. Vowel Harmony: An Account in Terms of Government and Optimality. Ph.D. dissertation, Leiden University. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in a Generative Grammar. Ms., Rutgers University and University of Colorado, Boulder. Rialland Anne & R. Djamouri. 1984. "Harmonie vocalique, consonantique et structures de dependance dans le mot en Monglo Khalka". Bulletin de la Societe de Linguistique de Paris 79.333-383. Rohlfs, Gerhard. 1966. Grammatica storica delia lingua italiana e dei suoi dialetti:Fontica. Torino: Einaudi. Schein Barry & Donca Steriade. 1986. "On Geminates". Linguistic Inquiry 17.691-744. Steriade, Donca. 1993. "Positional Neutralization". Paper presented at NELS 24, University of Massachusetts, Amherst. — — . 1995. "Underspecification and Markedness". Handbook of Phonology ed. by Jonathan Goldsmith, 114-174. London: Blackwell. Stewart, John M. 1967. "Tongue Root Position in Akan Vowel Harmony". Phonetica 16.185-204. Svantesson, Jan-Olof. 1985. "Vowel Harmony Shift in Mongolian". Lingua 67.283 329. Tekavcic, Pavao. 1972. Grammatica storica dell'Italiano: Fonematica. Bologna: II Mulino.
88
ANDREA CALABRESE
Vaux, Bert. 1992. "Adjarian's Law and Consonantal ATR in Armenian". Proceedings of the Fourth International Conference on Armenian Linguistics ed. by John Greppin, 271-293. New York: Caravan Books. . 1996a. "ATR Harmony in the Altaic Languages". Linguistic Studies in the Non-Slavic Languages of the Commonwealth of Independent States and the Baltic Republics ed. by Howard Aronson, 389-405. Chicago: Chicago Linguistic Society. . 1996b. "The Status of ATR in Feature Geometry". Linguistic Inquiry 27.175-182. . 1998. The Phonology of Armenian. Oxford: Oxford University Press.
VOWEL ALTERNATION, VOWEL/CONSONANT ASSIMILATION AND OCP EFFECTS IN A BARESE DIALECT*
FRANCESCO D'INTRONO University of Massachusetts, Amherst ROSEMARY WESTON Amherst College
0.
Introduction In Coratino, a Northern Barese dialect of Apulia spoken in the town of Corato, stressed vowels undergo a series of processes such as metaphony, diphthongization, lengthening, opening and laxing. Unstressed vowels are reduced to [e], but in some cases they are preserved or changed. Here we will deal only with unstressed vowels, and we will try to explain why they are reduced, preserved or changed. (For a historical overview of Coratino and other Barese dialects, see Melillo 1955.) There are four major factors determining the reduction, preservation or change of the unstressed vowel. i. the Place of Articulation (PA) of the unstressed vowel ii. the PA of the consonant adjacent to the unstressed vowel iii. the PA of the stressed vowel that follows the unstressed vowel iv. the position of the unstressed vowel within the phrase.
We would like to thank Giuseppe Tandoi and Anna Zaza, who were our main source of information during our data collection, and John McCarthy for his suggestions on a previous version of this paper, which was partially supported by a 1992-1993 University of Massachusetts Faculty Grant. Errors are our own. An earlier version of this paper was presented at the 1994 meeting of the Società internazionale di linguistica e filologia italiana, and was published in the Proceedings (D'Introno & Weston 1997).
90
FRANCESCO D'INTRONO & ROSEMARY WESTON
Examples in (1) illustrate the relevance of the first factor, that is, the quality of the unstressed vowel. Notice that in these examples the consonant adjacent to the vowel under consideration is a coronal consonant, i. e., a non-labial, non-dorsal consonant. (Dorsal is synonymous with back, and labial with round.) (1) Vowels adjacent to a coronal consonant PA of V stressed position 'ritstsә 'curl' a. Front High b. Front Mid 'seddә'chair' c. Low 'sals 'salt' d Labial Mid 'rots 'wheel' e. Labial High 'surd 'deaf
unstressed position rәts'tsutә 'curly' sәd 'darә'chair maker' sa'latә 'salty' rә'teddә 'small wheel' ndzәr'di 'to become deaf'
We deduce from (1) that non-low vowels in contact with a coronal consonant are reduced to schwa in an unstressed position, while the low vowel /a/ is preserved (1c). The relevance of the second factor, that is, the type of consonant adjacent to the unstressed vowel, is illustrated by the examples in (1), where, as we have seen, the vowel is adjacent to a coronal consonant, in (2), where the vowel is adjacent to a labial consonant, and in (3), where the vowel is adjacent to a dorsal consonant.1 (2) Vowels adjacent to a labial consonant unstressed position PA of V stressed position 2.1. Preceded by a labial consonant a. Front High 'mirrә 'mile' mәjjarә 'about a thousand' b. Front Mid 'mestra 'master' mәst'ruddә 'small master' c. Low 'baps 'bath' bәna'rolә 'bathtub' d Labial Mid bor'dzsttә 'pocketbook' 'bordzә 'bag' e. Labial High bus'tina 'small envelope' 'busts 'envelope' 2.2. Followed by alabial consonant a. Front High 'lima 'file' lә'matә 'filing' b. Front Mid 'Jems 'stupid' fәmu'nitә 'became stupid' c. Low 'lamә 'blade' lә'mettә 'razor blade' 'omәnә 'man' omә'niaә 'small man' d. Labial Mid e. Labial High 'lumә 'oil lamp' lu'minә 'wick'
[c]/|j] represent palatal stops, [c] is sometimes represented as /k1/.
VOWEL ALTERNATION
91
(2. lc) and (2.2c) show that in unstressed position the low vowel is preserved when adjacent to a labial consonant, while (2.1a, b) and (2.2a, b) show that front vowels are reduced to [9] in this context. The examples in (2. 1d, e) and (2.2d, e) show that unstressed /o/ and /u/ are preserved when adjacent to a labial consonant.2 (3) Vowels adjacent to a dorsal consonant stressed position unstressed position PA of V 3.1. Preceded by a dorsal consonant a. Front High 'kilә 'kilo' kә'latә 'about a kilo' b. Front Mid (mә)'kelә 'Michael' (mә)kә'linә 'little Michael' c. Low 'kasә 'house' ka'ssddә 'hut' d Labial Mid 'kondzә 'tanning fluid' kundza'torә 'tanner' e. Labial High 'kundә 'account' kun'dandә 'cash' 3.2. Followed by a dorsal consonant a. Front High 'rikkә 'rich' arәik'kitә 'enriched' b. Front Mid 'sekkә'dry' assәk'katә'dried' c. Low 'sakkә 'sack' sakku'tεdtә 'small sack' d Labial Mid 'sokә 'rope' su'kwiddә 'thin rope' e. Labial High 'sukә 'juice' ndzu'kwatә 'wet, a rinse' (3.1c) and (3.2c) show that, as in the previous cases, unstressed /a/ is preserved, while front vowels are reduced to [9] ((3.1a, b) and (3.2a, b)). In the case of labial vowels the pattern is that /u/ is preserved, as in example (3. le) and (3.2e), while /o/ becomes [u], as in examples (3. 1d) and (3.2d). Palatalized velars (usually geminates in the second syllable) adjacent to vowels, however, do not provide a context for the retention of labial vowels, as shown in (4). Notice that the low vowel is preserved in this case too.
2
Back vowels are also preserved when the labial consonant following the vowel is in the same syllable as the vowel: ['stumbә] 'gimp', [stumba'riәdiә] 'little gimp', etc. But in the context of the voiceless labials, [p] and [f], and of the voiced fricative [v], the retention of the back vowels, /o/ and /u/, is not categorical. We find ['lupә] 'wolf' and [lu'pinә] little wolf', but also [lә'pinә]; ['furnә] 'oven' and [fur'nstta] 'small oven', but also [fәr'nεttә]; ['voskә] 'woods' and [voska'riaddә] 'little woods', but also [vәs'kwiddә] 'small woods'; [Vovә] 'bull' and [vә'tiәddә] 'veal', but not *[vo'tiәddә]. These phenomena are part of a study we are conducting on other aspects of Coratino.
92
FRANCESCO D'INTRONO & ROSEMARY WESTON (4) Vowels adjacent to a palatalized dorsal consonant PA of V a. Front High b. Front Mid c. Low d Labial Mid e. Labial High
The third factor mentioned above for the retention or change of the unstressed vowel is the type of stressed vowel that follows it.3 This factor is irrelevant for the front vowels, but it is determinant in the case of labial vowels. In fact, the examples in (5.1) show that a labial vowel in contact with a labial consonant is not retained as expected (see (2)) if the following stressed vowel is a back vowel. (5) Unstressed vowels before a stressed back vowel PA of V stressed position unstressed position 5.1. Adjacent to a labial consonant a. Labial Mid b. Labial High 5.2. Adjacent to a dorsal consonant a. Labial Mid b. Labial High
In examples (3.1. d, e), repeated here as (5.2), however, the labial vowel is retained, although it is followed by a stressed back vowel. But in this case the unstressed vowel is adjacent to a dorsal consonant and not to a labial consonant, as in (5.1). This is, as we will show, the reason why the unstressed labial vowel is not reduced. The fourth factor considered, the position of the vowel within the phrase, is relevant in the case of the low vowel /a/, which, as we have seen, does not reduce. Examples in (6), however, indicate that /a/ reduces to schwa if it is phrase final and unstressed.
3
We are assuming that [a] is back in Coratino as well as in other Barese dialects (in Ruvese, spoken in Ruvo, at 5 miles from Corato, stressed [a] is realized in many words as [c]) as well as in some other Romance dialects. For example, in some Asturian dialects of Northern Spain stressed [a] alternates with [o] by metaphonic raising: [*bonku] 'bank' for ['banku], which shows that /a/ is back in these dialects; while in other dialects it alternates with [e]: [benku], which shows that in these dialects it is front.
VOWEL ALTERNATION
93
(6) Unstressed phrase final /a/ a.daughter' b . ' m a j 9 'my daughter' c. maja 'beltø 'my beautiful daughter'
A summary of the data is presented below. (7) unstressed non-low vowels reduce to [ә]: (1) (8) • unstressed labial vowels adjacent to a labial or a velar consonant are not reduced: (2d),(2e),(3e) • /o/ becomes [u] when adjacent to a velar consonant: (3d) • unstressed labial vowels are reduced to[әJ if they are adjacent to a labial consonant and are followed by a stressed back vowel: (5.1) (9) the low vowel /a/ is reduced to [a] only when it is unstressed and phrase final: (6)
1.
Coratino Vowel System According to the analysis that Stehl (1980, 1988) proposes for Northern Barese dialects, Coratino should have a five vowel system. Stehl also argues that the 'Neapolitan' vocalic system of Northern Barese, which includes Coratino, once had seven phonemic vowels, which was then reduced to five, with the exception of the dialect spoken in the city of Bari. In this dialect, [Venә] 'he/she comes' and ['vena] 'vein' are a minimal pair (Stehl 1988:703). This is also a minimal pair in Coratino, as is ['sokә] 'I am' and ['sckә] 'rope'. On the basis of these observations we will take Coratino to have a seven vowel system, not five, a fact that has no major bearing on the issues discussed in this article. (The feature that distinguishes closed from open mid vowels {Id vs. /e/, and /o/ vs. /o/) is ATR, which does not play a major role in the hypothesis sustained in this article.) The stressed vowel system is exemplified in (10). (10)
'sikkjә 'bucket' 'kussә 'this' 'venә 'he/she comes' 'sokә 'I am' 'vεnә 'vein' 'sckә 'rope' 'tsappe 'hoe'
The feature definition of the Coratino vowels in (11) is compatible with the Underspecification Theory of Archangeli (1988),4 and follows in part the Feature 4
In this paper we are not strictly subscribing to Archangeli's Radical Underspecification Theory (Archangeli 1988). Rather, we are going to develop a hypothesis that is an overlap between hers
94
FRANCESCO D'INTRONO & ROSEMARY WESTON
Dependency Hypothesis of Mester (1986),5 in that features are hierarchical. More specifically, we are assuming that low, dorsal, labial, high and ATR are feature tiers organized in such a way that ATR is dependent on high, high on labial, labial on dorsal, and dorsal on low. In each tier a feature will be either specified, for example H in the high tier for /i/, or not specified, as is the case for H for mid and low vowels. Specified features are positive; unspecified features are negative. Assuming that all features have to be spelled out at the surface level, they will be introduced by redundancy rules. Accordingly, the underlying representation of lil and /a/, for example, will be as in ( 12). (11) i
e
s
low dorsal labial high ATR (12)
L
a
0
o
u
(D)
D (Lab)
D
D
(Lab)
(Lab) H
H
ATR
/i/ || [low
ATR /a/
]
[low
I
L] I
[dorsal
]
[dorsal (D)]
i [labial
]
| [labial
]
[high
]
[ATR
]
|| [high H] ||] [ATR ]
A few observations are in order. In the first place, as can be seen in ( 11) and (12), all segments contain the same tiers although those with an unspecified feature are empty. In second place, a specified feature F a that is redundant with respect to another feature F b that immediately dominates F a , as is the case of labial and Steriade's (1987) and Clements' (1987) Contrastive Specification Theory, which we find to be the most adequate empirically and theoretically for the data presented here. 5 Our hypothesis is similar to that of Mester (1986), who sustains, for example, that in Ainu, high depends on back. We think that the order is parameterized and that in other languages it can be reversed, as seems to be the case in Ngbaka.
VOWEL ALTERNATION
95
with respect to dorsal, is represented in parentheses. More precisely, Feature F a is redundant with respect to feature F b if F b directly dominates F a and F a is predictable and recoverable from Fb. Finally, some features are exclusive, in the sense that they are incompatible with another feature. More precisely, Feature F a is exclusive with respect to feature F b if the presence of F a in a segment S excludes the presence of F b in the same segment S, that is, F b cannot be dominated by F a in S. For example, low is exclusive with respect to high and labial: a low segment cannot also be high or, at least in Coratino, labial. 2.
Vowel Reduction and Retention As we have seen, Coratino unstressed non-low vowels adjacent to coronal consonants are reduced, that is, their features are delinked and surface as [9]. This can be seen in examples (1) repeated here as (13). (13) a. b. c. d e.
Front High 'ritstsә 'curl' Front Mid 'seddә'chair' Low 'sala 'salt' Labial Mid 'rotә 'wheel' Labial High 'surdә'deaf
rәts'tsutә 'curly' sәd d arә 'chair maker' sa'lata 'salty' ra'tεddә 'small wheel' ndzәr'di'to become deaf
The low vowel /a/ does not reduce to [9], as shown in (13c), except in word final position (6). 6 We will discuss this reduction in §6. How can we account for the reduction of non-low vowels in Coratino? We will assume that there are two processes that affect features in Coratino, Delinking and Linking, which can be defined as in (14). (14)
a. Delinking: Tier T of an unstressed vowel V is severed b. Linking: Segment S a and segment S b link at feature F iff: (i) S a and S b areadjacent (ii) F is present in S a and S b
We will also assume that Delinking applies to the dorsal tier, erasing the dorsal tier, and along with it any other tier below it. If Delinking applies to /i/, for example, represented in (12), it will erase all the tiers except the low tier, which can not be filled in since it is not a redundant feature. The result is a featureless vowel root node realized phonetically as [9]. 6
The status of the word final suffix for feminine nouns will be clearly established later on. As for the status of the masculine suffix, it will be left open, since it has no bearing on the problems discussed in this paper.
96
FRANCESCO D'INTRONO & ROSEMARY WESTON
By applying Delinking, as stated in (15), to the unstressed vowels in (13) we get [9] in all cases, except when the vowel is /a/. Here Delinking of the dorsal has no effect because dorsal is redundant in the case of /a/, thus it will be reintroduced. (15) Delinking: Delink the dorsal tier of an unstressed vowel
In ( 16) we see the effect of Delinking on unstressed /u/ which becomes [ә]. (16)
/u/ || D --> | (Lab) | H
[9] 0
To ensure that Delinking and Linking apply not just to a particular tier but also to all the tiers dependent from it, we will add the following conventions to Delinking and Linking.7 (17) a. Delinking Convention: If tier T a of segment S is delinked, tier T b of S dominated by T a will delete b. Linking Convention: If segment S a and segment Sb are linked by feature F a , feature Fb of S a dominated by F a will be assigned to S b , iff Fb is not exclusive for S b
The Delinking Convention in (17a) stipulates that if Delinking applies to a tier that dominates other tiers, these tiers delete, independently of the type of feature they contain. For example, they can contain a redundant feature or they can be unspecified. Thus, if Delinking applies to /u/ so as to sever the dorsal tier which contains the feature D, it severs tier Lab which contains the redundant feature Lab, tier High which contains the feature H, and tier ATR which is empty for /u/. This is shown in (16), where /u/ reduces to [9]. Because Delinking applies to all tiers, even if they are empty, it will apply to front vowels as well, resulting in [9]. 8
7
The Delinking Convention, which is a consequence of the Dependency Theory adopted here, is similar to the Shared Feature Convention of Steriade (1982). 8 One could ask what the difference between a front mid vowel and schwa is. Within our framework, the difference is that a front mid vowel has empty tiers that can be filled in by
VOWEL ALTERNATION
97
Examples in (2) and (3) show, however, that reduction of non-low vowels in Coratino sometimes does not take place. In fact, as illustrated in the data in (2), and as can be deduced from examples (18b) and (19b), when an unstressed labial vowel is adjacent to a labial consonant, the vowel retains its features. Examples (20b) and (21b) show that the same is true of a labial vowel adjacent to a velar consonant. (18) (19) (20) (21)
a. 'mollә 'spring' a. 'lumә 'oil lamp' a. 'kundә 'account' a. 'dzukkәrә 'sugar'
b. mollette 'clothes pin' b. lu'min9 'wick' b. kun'dandә 'cash' b. dzukka'rinә 'sweet'
Our hypothesis for the retention of vowels in (18b)-(21b) is as follows. Under the assumption that vowels and consonants are defined by the same set of features, labial and velar consonants have a feature in common with back, rounded vowels, i. e., the major articulatory features labial and dorsal, respectively. Given this, a labial vowel and an adjacent labial or dorsal consonant will have in common a feature, through which they link. Because of that, they can not be reduced, and they are preserved.9 Let us make this hypothesis more explicit, and assume that the point of articulation of consonants is defined with the same set of hierarchical features as the vowels, although, as suggested by Clements (1985,1987) some features might not be relevant for consonants, and vice-versa. For example, low is not applicable to consonants because there are no low consonants. This is also true for ATR. Moreover, following Clements, we could distinguish two types of features for consonants: the major articulatory features, which are coronal, labial and dorsal, and minor articulatory features, such as high. We are also assuming that coronal consonants are unspecified in terms of point of articulation, as is the case of mid front vowels. 10 Finally, we will assume that the Obligatory Contour Principle (OCP), as defined by McCarthy (1986) which prohibits a sequence of two identical elements, is the triggering condition for the retention of the vowels because it is also operative at the level of feature tiers. Here it checks for identity redundancy rules with non-low, non-dorsal, etc. Schwa is the phonetic realization of an empty V slot. This might not be true for other languages. 9 [monu'msnda] 'monument' shows that underlyingly the stem can contain more than one labial vowel and that they can be preserved if they link. 10 This idea also appears, for example, in Hume (1996), who describes vowel to consonant assimilation. In her analysis the process affects front vowels and coronal consonants, which in our analysis do not link because they are transparent, as seems to be the case in many other languages.
98
FRANCESCO D'INTRONO & ROSEMARY WESTON
of features, and if a vowel and an adjacent consonant share a feature, an OCP violation occurs. The ensuing repair strategy consists of linking the two segments, after which Delinking can not apply because of the Inalterability Convention which stipulates that linked structures are not affected by phonological rules (Steriade 1982, Hayes 1986, and Schein & Steriade 1986).11 (22) Obligatory Contour Principle (OCP): At the melodic level, adjacent identical elements are prohibited
Let us see how this hypothesis can account for the data at hand. In (18) and (19), for example, we have a labial consonant adjacent to a labial vowel. Both segments are specified for labial. This results in an OCP violation, which is repaired by Linking of the labial feature as shown in (23) for the sequence /mo/. 12 In (19) and (20), on the other hand, we have a velar adjacent to /u/, both dorsal, so they link at the dorsal tier as shown in (24) for the sequence /ku/. (23) m o | || [] D => | | Lab (Lab) (24) k u | | D D => ||| [] [Lab] ||| H H
m o []
D \/ Lab
k
u \/ D
[Lab] H
Notice that Linking of dorsal in (24), for example, requires Linking of other dependent features of the two segments that are compatible with dorsal, i. e., labial,
11 Linking, however, seems to be able to link more than a segment. For example, in [sfun'ga] 'to give milk', there are three segments that are linked, [ung]. However, this is not surprising. A similar case of linking of three segments can be seen in Spanish in the spreading of [cont] in [adßer'tir] 'to caution', where this feature is spread from [a] to the two following consonants. 12 In Casali (1995) roundness harmony between vowels in Nawuri is blocked by an intervening labial consonant adjacent to the triggering word final vowel. In our hypothesis this process would be blocked by linking the round word final vowel and the adjacent consonant.
99
VOWEL ALTERNATION
or are identical in the two segments, i. e., high. Notice also that we are assuming that this last feature is a minor articulatory feature for velar consonants. This hypothesis offers a straightforward explanation for the fact that unstressed front mid vowels are not retained: they do not share any feature with consonants, so they cannot link, and Delinking erases their tiers. The end result is [9]. (ATR is not a feature present in consonants.) What about the high front vowel /i/? The only feature present in /i/ is high. The same feature is present in velar consonants, as we have just pointed out, thus an unstressed /i/ in contact with a velar consonant should be retained. This happens in cases like (25b) and (26b) where [I] is a vowel slightly more closed and shorter than [i], but it can also reduce to [9] as in (25c) and (26c), a fact that suggests that Linking of high, a minor articulatory feature for consonants and the lowest one in the feature hierarchy, does not allow for the preservation of other features. 13 (25) a. 'kite 'kilo' (26) a. 'rikkә 'rich'
b.kr'latә'about a kilo' b. arrik'kitә 'enriched'
c. kә'latә c. arrәk'kitә
3.
OCP and Tier Scanning The analysis we have proposed does not account, however, for the alternations presented in (3), where /o/ becomes [u] when it is adjacent to a velar consonant. (27) a.'kordә'cord' (28) a.'skopә 'broom'
b. kur'dεddә 'string' b. sku'pinә 'toilet brush'
Our hypothesis is that when Linking of dorsal between the vowel and the consonant takes place, the feature high of the consonant is shared by the vowel, and /o/ becomes [u]. 14 This is also true for the labial feature of the vowel which is shared by the velar consonant, which is allophonically realized as labialized. The claim here is that /o/ becomes high and that velars become allophonically labial.
13 It is not easy to find examples with stressed [i] in contact with a velar because most of them come from an underlying /e/ which raises by metaphony, therefore when unstressed [i] reduces to [a] the question is if it comes from underlying /i/ or Id, something that is not always easy to establish. 14 There are a few cases where an unstressed /o/ adjacent to a velar does not change into [u], for example, [koper'tonә] 'big blanket; (surname)'. /o/ does not change into [u] in words like [kon'drorә], 'afternoon; siesta', because they are compounds. See §5.
100
FRANCESCO D'INTRONO & ROSEMARY WESTON (29) k o | D D ||| | (Lab)
k |
u \/ D
=>
(Lab)
|| H
H
The analysis developed so far, however, leaves unexplained the change of unstressed /o/ and lui to [9] in cases like those in (30b-33b), because we would expect /o/ and lui to link to the adjacent labial and thus be preserved. Notice that in the examples (30b-33b), the stressed vowel that appears after /o/ and lui is a back vowel, and not a front one as in ( 18), [mo'llεttә], and in ( 19), [lu'minә]. Let us suppose then that the OCP is in effect in (30b)-(33b) and, in particular, that in Coratino there is an OCP effect between the unstressed vowel and the following stressed vowel. (30) (31) (32) (33)
a. a. a. a.
lә'monә lemon' 'noma 'name' 'botta 'a blow' 'Jumma 'hump'
b.lәmә'natә'lemonade' b. nәmә'natә 'big name, fame' b. abbәt'turә 'injury from a blow' b. Jәmma'rutә 'hunchback'
This being the case, /o/ and lui must be delinked, as shown in (34), because both the unstressed and stressed vowel are dorsal. They cannot link because stressed vowels do not link, so the unstressed vowel is reduced. (34) o
a
'V
|||| D D | (Lab)
=>
'V
Ø D
This hypothesis gives the right results, but leaves unexplained why Linking of labial between the vowel and the adjacent consonant does not occur prior to Delinking. The answer could be that Delinking precedes Linking. The following examples, however, show that this is not the case.
VOWEL ALTERNATION (35) (36) (37) (38)
a. vrә'gonә 'shame' a. 'kondzә 'tanning fluid' a. 'skortsә 'tree bark; peel' a. 'skolә 'drainage'
101
b. vrәgu'nusә feels ashamed' b. kun'dzarә/kundza'tors 'tanner' b.skur'tsutә 'with thick bark or peel' b. skula'turә'sediment of drainage'
In (35b)-(38b), the unstressed back vowel /o/ is adjacent to a velar consonant and is followed by a stressed back vowel. There are two OCP violations: one between the unstressed vowel and the adjacent consonant, and another between the unstressed vowel and the stressed one. To satisfy the OCP, either the unstressed /o/ and the adjacent consonant can undergo Linking, or the unstressed vowel can undergo Delinking. In order to account for the fact that the /o/ raises, we could say that Linking precedes Delinking, but this is exactly the opposite of what we suggested to account for (30b)-(33b). The alternative is to leave Linking and Delinking unordered. We must ensure, however, on one hand, that the OCP, Linking and Delinking operate at the Dorsal tier prior to the Labial tier, otherwise we could not explain (30b)-(33b). Furthermore, we must ensure that they first affect segments at the beginning of the word before affecting segments at end of the word, otherwise we could not explain cases like (35b)-(38b). Therefore, we propose the following convention, which can be parameterized differently in other languages. (See also Archangeli & Pulleyblank 1987.) (39)
Feature Tier Scanning: Phonological processes scan feature tiers top to bottom, and left to right.
(39) requires that the OCP, Linking and Delinking apply from top to bottom, and from left to right. Following this strategy, as soon as the conditions for a phonological process P are met, P applies. For instance, in (35b)-(38b), scanning of the dorsal tier from left to right reaches first the velar consonant and /o/, both specified for dorsal. Since this is an OCP violation and the two segments are adjacent, they undergo Linking. Continuing with the scanning from left to right, when the dorsal tier of the stressed vowel is reached, unstressed /o/ is already linked and cannot be reduced. So, although there is an OCP violation between the two vowels, it is overcome by the fact that the unstressed vowel is linked to the consonant, and it is therefore preserved. This hypothesis is supported by examples (30b)-(33b). For instance, in (30b) analyzed here as (40), scanning applies first to the dorsal tier, the OCP is violated between the unstressed and the stressed vowel, Linking between them cannot apply (sinced stressed vowels can not link), so the dorsal tier of unstressed
102
FRANCESCO D'INTRONO & ROSEMARY WESTON
/o/ is deleted. This happens before the labial tier is scanned, so /o/ does not link to /m/. The result is loss of /o/. (40) 1 a m o n a t |||||| [] D (D) ||| Lab (Lab)
a
l a m =>
[]
a 0
n
a
t
a
(D)
Lab
Two questions can now be asked. First, is it the case that for the OCP to be effective between the two vowels, the intervening segments must be coronal, as is the case in (40)? Second, would a dorsal segment other than the stressed vowel trigger an OCP violation with the unstressed vowel? The answer to these questions is no, as shown by the following examples. (41) a. 'stomәkә 'stomach' (42) a. 'monәkә 'monk'
b. stoma'kiccә 'little belly' b. mona'kiddә 'alter boy'
In (41b) and (42b) there is a back vowel, /a/, and a dorsal consonant, /k/, between /o/ and the stressed vowel. If these segments created the appropriate environment for an OCP violation,/o/would have to reduce to [9]. Since /o/ does not reduce, the examples prove that the relevant segment for the OCP effect is the stressed vowel of the suffix, which in this case is a front one and does not create an OCP violation with /o/. Let us look at other cases with /a/. The first [9] in ['stomәkә] and ['monәkә] of (41a) and (42a) is derived from /a/, which surfaces in [stoma'kicca] (41b) and [mona'kiddә] (42b). The examples in (6), ['fijja 'majә], however, show that the correct environment for the reduction of /a/ is phrase final. We will come back to this issue below. Examples (43)-(46) show that non-low vowels in the suffix reduce to [9] when unstressed. In examples (43b)-(45b), we find a stressed non-low vowel in the penultimate suffix - [s], [i] and [o] - which reduces to [9] when it is unstressed (43c)-(45c). There are no similar examples with /u/. However, in (46b) the stressed low vowel in the penultimate suffix - [a] - does not reduce to [9] when it is unstressed, but not phrase final (46c). Instead, it is preserved. In conclusion, unstressed non-low vowels reduce independently of their position, while unstressed /a/ reduces only when phrase final.
VOWEL ALTERNATION (43) a.
'karr-ә 'cart' (44) a. 'tarand-ә Taranto' (45) a. 'tsapp-9 'hoe' (46) a. 'skarp-ә 'shoe'
103
b.
karr-'εtt-ә c. karr-әtt-'iәr-ә 'light cart' 'cart driver' b. tarand-rin-ә c. tarand-әn-!iәdd-ә 'person from Tarante' 'little person from Taranto' b. tsapp-at'or-ә c. tsapp-at-әr-'iәdd-ә 'peasant' 'small peasant' b. skarp-'ar-9 c. skarp-ar-'iәdd-ә 'shoemaker' 'lousy shoemaker'
Notice that in (45c) /o/ is not in contact with a consonant to which it could link. Will a labial vowel in the suffix link to an adjacent velar or labial consonant? We do not have examples of this type where the consonant is part of the suffix, but (47c) and (48c) show that the first vowel of the suffix can link to the last consonant of the stem. In fact, [u] in (47c)-(48c) derives from [o] in (47b)-(48b), proving that a labial vowel of a suffix can link to an adjacent consonant. 15 (47) a.
'sakk-ә 'sack' (48) a. 'pakk-ә 'pack'
b.
sakk'ott-ә 'big sack' b. pakk-'ott-ә 'package'
c. sakk-ut-'εdde 'drawstring sack' c. pakk-ut-'εdde 'small package'
4.
Labial Vowels in the Stem Our hypothesis accounts for the examples in (49a) but leaves unexplained (49b), from which the forms in (49a) are derived. 16 (49) a. |kukumә'riәdlә]'cucumber' [kukuts'tsarә] 'zucchini seller'
b. [k9'komәrә] 'watermelon' [kә'kotstsә] 'zucchini1
Since the first vowel in (49a) is [u], we would expect the first vowel in (49b) to be [o] or [u], but not [9]. How can we account for this? We will offer the following explanation. The second vowel of the stem is underlyingly /o/, which surfaces as [u] in (49b) by linking to the velar. Assuming that the first vowel is 15
Notice that although this does not provide conclusive proof for Linking in this case, there seems to be no other explanation for the appearance of [u] in (47c)-(48c). There are no examples of a stem ending in a labial consonant, followed by an underlying unstressed labial vowel, so we cannot present positive proof in this case. But if our analysis is correct, it would predict that the labial vowel would be preserved because of Linking. 16 Similar examples include [skukurjja] 'to research', [skukuts'tsa] 'to cut the top', [kukuva'Ju9l9] 'a type of herb', [kuku'tsεttә] 'a type of hat', [kukuVaJә] 'owl', [akkuku'vat9] 'crouched'.
104
FRANCESCO D'INTRONO & ROSEMARY WESTON
underlyingly /o/, the underlying representations of (49a) and (49b) are (50a) and (50b), respectively. (50) a. /kokomәr/ /kokotsts/
'iәdiә 'arә
b. /ko'komәr/ ә /ko'kotsts/ ә
The difference between these forms is that the second /o/ is unstressed in (50a) while it is stressed in (50b). We could then say that the first /o/ in (50b) reduces because it is followed by a stressed /o/. Let us try to incorporate this idea into our hypothesis. Suppose that the OCP as well as Linking and Delinking apply between vowels before applying between vowels and consonants. 17 Suppose also that the OCP is operative among vowels if they share all the specified features. Linking will then associate vowels if they are identical and unstressed, otherwise they will reduce unless stressed. This predicts that an OCP violation will result if two labial vowels have the same features, in which case they will link - if neither one is stressed - and will surface unchanged, otherwise the unstressed ones will surface as [a]. In (50a) both labial vowels are /o and unstressed, therefore they link and are preserved, as shown in (49a). If one of them is stressed, as in (50b), the unstressed one loses its features because it can not associate with the stressed vowel (remember that stressed vowels do not link) as proven by (49b). We will also assume that in the case of feature sharing between identical unstressed vowels, the features of the second one are spread to the first one. This process can be summarized informally the following way, assuming once again that processes apply from left to right. (51) If two vowels, V1 and V 2 , agree in all specified features and are unstressed, the features of V 2 are assigned to the root node of V 1 .
According to (51), given two unstressed labial vowels in the stem with the same height, as in (49a) and (50a), the first one will lose its features, and the features of the second one will spread to its root node (52). On the other hand, given an unstressed vowel and a stressed one, both with the same roundness and height, as in (49b) and (50b), the first one will lose its features but will not receive the features of the second one. As a result, it will surface as [a] (53). (We have placed vowels and consonants in different levels for ease of exposition.) 17
We claim, following a hypothesis of McCarthy (1981) on Semitic languages, that vowels are on a separate tier from consonants, but as a reviewer points out this is justified on independent grounds for Semitic languages, but not for Romance languages.
VOWEL ALTERNATION
105
After the process described in (51) takes place, the OCP, Linking and Delinking will reapply to vowels and consonants from left to right. When this occurs the second /o/ of (52) will link to the velar consonant, surfacing as [u]. Since the set of features for this [u] have spread to the first vowel, this also will surface as [u]: [kukumәr-'i9ddә]. In (53), on the other hand, when vowels and consonants are scanned together, the first vowel, [a], will not link to the velar consonant. Neither, will the second vowel, because it is stressed, and the result will be [kә'komәrә]. The examples in (54) provide evidence in support of our hypothesis. 18 (54)
a. /tfoko'latә/ > [tfuku'latә] 'chocolate' b. /kolo'nεddә/ > [kolon'εddә] 'little column; night table' c. /ko'lonnә/ > [kә'lonnә] 'column' d. /kum'bostә/ > [kum'bostә] 'pepper preserve'
In (54a) the first vowel is associated with the second since both are unstressed, and they surface as [u] because of the Linking of the second vowel with the velar consonant. 19 In (54b) the features of the second /o/ will spread to the root node of the first /o/. Since the first vowel obtains its features from the second one, it will also surface as [o] even though it is adjacent to a velar consonant. In (54c) the first /o/ does not link to the second, which is stressed, so it 18
Examples similar to (54b) include [bombo'lonә] 'big male doll; (surname)' ; examples similar to (54c) include [k9'lummә] 'fig', [kә'tunә] 'quince'; examples similar to (54d) include [kum'blcttә] 'conspiracy'. 19 In our analysis the only requirement is that there are two identical vowels. Even if the underlying form were with /u/ the derivation of this word would support our analysis.
106
FRANCESCO D'INTRONO & ROSEMARY WESTON
loses its features. The featureless vowel root node will not link to the adjacent velar consonant, and it surfaces as [9]. In (54d) the two labial vowels of the stem are not identical so the first is not delinked at the vowel tier. When scanning takes place between vowels and consonants the first vowel links to the velar consonant and is preserved. Our analysis can not account for the vowels in [mor'rojdә] 'hemorrhoid' and [ro'soljә] ' alcoholic rose water' (we would expect the first /o/ to reduce or delink at the vowel tier); [kulә'nεltә] 'small column' and [kutә'niәddә] 'small quince' (Bucci 1982) (we would expect the first two vowels to remain identical); [dot'torә] 'doctor' and [o'norә] 'honor' (we would expect the unstressed [0] to delink because it cannot link to an adjacent consonant, as happens in [mo'tor9] ' motor'). However, it might be that these words do not follow the expected pattern because they are loan words from Italian and are used in more formal speech. 5.
Stem and Suffix The following seem to be counterexamples to the hypothesis just presented because (55b)-(57b) contain two labial vowels with the same height, the first one being unstressed. (55) (56) (57)
a.ku'radjdjә'courage' a. 'kusә 'sew' a. 'kundzә 'tanning fluid'
b.[kuradjdj]'usә'brave' b. [kus] ә'turә 'couture' b. [kundz]a'turә'leather tanning'
We would expect the first /u/ of (55b)-(57b) to lose its features when vowels are scanned (because of an OCP violation with the following stressed /u/ with which it can not link) and surface as [9]. But it does not. Why is the first /u/ maintained? The answer relies on the observation that in the previous examples the two labial vowels were in the stem, while in (55b)-(57b) the first /u/ is in the stem and the second /u/ is in the suffix. This indicates that the OCP, Linking and Delinking operate first on the vowels of the stem (or separately in each morpheme), and then on all the segments at a lexical level. Accordingly, the first /u/ of (55b)-(57b) will not be reduced when vowels in the stem are scanned (since there is no OCP effect at that point). When the OCP, Linking and Delinking apply to all the segments of the word, the unstressed /u/ will link to the velar consonant and will be preserved before the stressed /u/ in the suffix is reached. This analysis implies that the first vowel in the examples in (58) should reduce for two reasons: both because it is followed by an identical stressed vowel, and because it can not link to an adjacent consonant. It is, however, preserved. This
VOWEL ALTERNATION
107
is because the word is a compound: the first vowel does not delink because it has a secondary stress. (58)
a. stuta'tubbә 'tube-extinguisher, stupid, (surname)' < [[[.stut]stemalword [['tubb]stemә]word]word b. sottaskwa'dronә 'big under-slap' < [[[.sott]stema]WQrd[[skwadr]stem'onә]word]word
This analysis can explain why the unstressed labial vowel in (59) can surface as [u]. (The pronunciation with [9] is also possible.) (59)
a. mun'danә 'mountain' b. mus'tarda 'mustard; garbage'
Our hypothesis predicts that in cases like (30b) and (31b) where an unstressed labial vowel is followed by a stressed /a/, the labial vowel is reduced because of an OCP effect between the two vowels. So, /u/ in (59) should reduce. Why is it preserved? What sets (59) apart from the other examples is that the stressed /a/ in (59) is not in the suffix but in the stem. For reduction to take place, there must be two identical labial vowels in the stem. Since this is not the case in (59), /u/ is not reduced. We must add, however, that when all segments are scanned, the OCP takes into account the unstressed vowel and its adjacent consonants, or the unstressed vowel and the following stressed vowel in the suffix, if there is one. In this way the /u/ of (59) will only create an OCP violation with the labial consonant, /m/, because there is no stressed vowel in the suffix. Thus /u/ will link to /m/ and will be preserved. 6.
Low Vowel Reduction As we have previously pointed out, unstressed /a/ is reduced and surfaces as [9] in phrase final position, due to the erasure of the feature low (see (6) above). This is not, however, completely correct. (60a), whose underlying representation is (60b) with two unstressed /a/s after stressed /o/, shows that /a/ reduces in a poststress position although it is not necessarily word final. But it is also clear from (61)-(63) that /a/ reduces only after the final stress of a phrase, making low Delinking a phrase level process. In fact, it does not apply to the final /a/ of the first word of a compound, as can be seen in the examples in (58), nor does it apply to the final /a/ of the first feminine adjective in reduplication constructions such as (61a) (compare with 61b), or the first /a/ of an adverbial compound (62). Example (63) lends support to this analysis (compare with 64).
108
FRANCESCO D'INTRONO & ROSEMARY WESTON (60) (61) (62) (63)
(64)
a. 'monәkә 'nun' b. /'monaka/ a. Tina 'finә Very fine (f.)' b. finә finә 'very fine (m.)' 'sotta 'sawpә 'upside down' a. 'finә 'daughter' b. 'fina 'rnajә 'my daughter' c. 'fina 'maja 'bellә 'my beautiful daughter' 'fnә 'mi 'bellә 'my beautiful son'
It is important to add that final /a/ reduces within a phrase of more than one word if there is no pause between words, otherwise the process applies to each individual word. It does not apply to monomorphemic monosyllabic elements either, as the article Ja ' the (f.)', the preposition a ' to', and the complementizer ça ' that', probably because these are a type of clitic and appear at the beginning of the phrase of which they are a part (i. e., DP, PP, and CP, respectively). Finally, /a/ reduction does not correspond to an OCP violation. In fact, there are no OCP effects on /a/. This is a consequence of the feature organization we have given in (11) and (12), and the Delinking hypothesis we have proposed. Since the only specified non redundant feature of /a/ is low, which is not shared by consonants, there cannot be an OCP effect between /a/ and an adjacent consonant. Thus, the only possible OCP effect could be between an unstressed /a/ and a following stressed /a/ (at the low tier) as in /[skarp]'arә/ 'shoemaker'. In this case the unstressed /a/ does not reduce because Delinking in a pre-stress environment applies only to the dorsal tier. Therefore, /a/ is preserved: [skar'parә]. 7.
Conclusion We have shown that the four major processes affecting unstressed vowels in Coratino are those in (65). In (66) we list the hypotheses we have proposed to account for the data. (65)
Major Processes Affecting Unstressed Vowels a. non-low vowel reduction b. labial vowel retention c. labial vowel raising d. low vowel reduction
VOWEL ALTERNATION (66)
109
a.
Features are organized in tiers dependent on each other, the highest tier being low and the lowest ATR. b. Positive features are specified, so that front mid vowels are unspecified, and back vowels are specified. c. Features are scanned for OCP violations, the repair strategy being either vowel Linking (to an identical unstressed vowel at the stem level, or to an adjacent consonant with the same feature at the lexical level) or vowel reduction by Delinking the dorsal tier. d Scanning for OCP violations, Linking and Delinking of features take place from top to bottom, and left to right. e. /a/ is reduced after the last stress of a phrase.
REFERENCES Archangeli, Diana. 1988. "Aspects of Underspecification Theory". Phonology 5.183-207. Archangeli, Diana & Douglas Pulleyblank. 1987. "Maximal and Minimal Rules: Effects of Tier Scansion". NELS 17.16-35. Bucci, Cataldo. 1982. Dizionario Etimologico Coratino. Cassano Murge, Bari: Tipografica Meridionale. Casali, Roderic. 1995. "Labial Opacity". Natural Language and Linguistic Theory 13.649-663. Clements, George. 1985. 'The Geometry of Phonological Features". Phonology 2.225-252. . 1987. 'Towards a Substantive Theory of Feature Specification". Paper presented at the UCLA Symposium on Segment Structure, October 1987. D'Introno, Francesco & Rosemary Weston. 1997. "Preservazione, cambio e riduzione vocalica in coratino: effetti e strategie dell'OCP". Atti del terzo convegno delia Società internazionale di linguistica e filologia ialiana ed. by Luciano Agostiniani, et a l , 287-302. Napoli: Edizioni scientifiche italiane. Hayes, Bruce. 1986. "Inalterability in CV Phonology". Language 62.321-351. Hume, Elizabeth. 1996. "Coronal Consonant, Front Vowel Parallels in Maltese". Natural Language and Linguistic Theory 14.163-203. McCarthy, John. 1981. "A Prosodic Theory of Nonconcatenative Morphology". Linguistic Inquiry 12.373-418. . 1986. "OCP Effects: Gemination and Antigemination". Linguistic Inquiry 17.207-263. Melillo, Michele. 1955. Atlante fonetico pugliese. Roma: Arti Grafiche S. Marcello.
110
FRANCESCO D'INTRONO & ROSEMARY WESTON
Mester, Armin. 1986. Studies in Tier Structure. Ph. D. dissertation, University of Massachusetts, Amherst. Schein, Barry & Donca Steriade, 1986. "On Geminates". Linguistic Inquiry 17.691-744. Stehl, Thomas. 1980. Die Mundarten Apuliens Historische und Structurelle Beiträge. Münster Westfalen: Aschendorff. . 1988. "Italienisch: Arealinguistik XI: Apulien und Salerno". Sprachen und Sprachgebiete 6.696-716. Steriade, Donca. 1982. Greek Prosodies and the Nature of Syllabification. Ph. D. dissertation, MIT. . 1987. "Redundant Values". Papers from the 23rd annual regional meeting of the Chicago Linguistic Society: Part 2: Parasession on Autosegmental and Metrical Phonology ed. by Anna Bosch, Barbara Need & Eric Schiller. 339-362. Chicago: Chicago Linguistic Society.
HOW MANY MORAS? OVERLENGTH AND MAXIMAL MORAICITY IN ITALY
JOHN HAJEK University of Melbourne 0.
Introduction In discussion of vowel length in Standard Italian and Italian dialects, consensus is that where a length difference of any kind is noted, the contrast is maximally binary, i. e., between long and short vowels only. Descriptions of Standard Italian, for instance, generally report vowel length differences to be important but entirely predictable: stressed vowels are always short in closed syllables and in word-final open syllables ([fatto] 'fact', ['tanto] 'so', [fi'ni] 'he finished'), but always long in word-medial open syllables (['fa:to] 'fate', [setti'ma:na] 'week'). As the examples also indicate, unstressed vowels are always short - in both open and closed syllables. The predictable distribution of vowel length in Standard Italian has important consequences for the permissible syllable structure of the language: stressed syllables, at least word-medially, are always heavy, i. e., VV$ and VC$. The phenomenon of even syllable weight in Standard Italian and many other Italian dialects is well-known and often referred to as the Bimoraic Norm: heavy syllables must always be equivalent to two moras in weight (Hajek 1997, Repetti 1989,1992, 1997, Young 1993). In moraic phonology, each weight-bearing segment is assigned one mora. Long vowels, and falling diphthongs (VG) - both equivalent to two short vowels - are each also equivalent in weight to two moras. (See Hayes 1989 for details.) Light syllables (V), often found in unstressed position, are assigned only one mora but are not permitted in stressed position in Italian. With rcspect to consonants, slightly different conditions apply in moraic theory: in onset position they are weightless. In coda position, singleton consonants have no underlying weight but may be assigned weight by rule while long consonants are underlyingly already attached to one mora.
112
JOHN HAJEK
If the Bimoraic Norm does indeed apply in Italian, then so-called superheavy syllables with three moras, VVV, VVC or VCC, are impermissible in this language. (See Absalom & Hajek 1997 for details.) Examples of the bimoraic representation of stressed syllables in word-medial position in Italian are given in Figure 1.
f a t o f a t o 'fate' 'fact' Figure 1. Bimoraic Structures in Stressed Position in Standard Italian
The notion of bimoraic maximality in Italian is consistent with claims made about maximal moraic weight in much of the literature on moraic phonology (Bickmore 1995, McCarthy & Prince 1993). The assumption that syllables may have no more than two moras is easily explained: since in discussion of syllable weight, languages most frequently contrast light and heavy syllables, this binary contrast in syllable weight is transposed into moraic analysis. There is, however, no a priori theoretical reason that should block the possibility of trimoraicity or three moras per syllable. Indeed, Hayes (1989) in his seminal paper on moraic phonology, has argued strongly for the existence of trimoraic syllables in a number of different languages, including Old English, Farsi, German and Danish dialects, and Estonian. To this list of trimoraic languages we can also add languages such as Finnish which permit sequences of bimoraic long vowel and monomoraic geminate consonant, as in [rr:p:Y:] 'drink' represented diagrammatically below.
r
Y:
p
'drink' Figure 2. Example of a Trimoraic Structure in Finnish
According to Hayes, the existence of trimoraic syllables can be established indirectly by means such as (1) patterns of compensatory lengthening, and (2) quantitative metrics, and directly by (3) the existence of three-way or ternary length contrasts. More recently, Féry (1997) has also argued for trimoraic
HOW MANY MORAS?
113
syllables in German on the basis of patterns of stress placement and syllable structure. Phonological patterning is not the only diagnostic for determining mora count. The use of phonetic evidence is a recognized technique in testing and, where necessary, justifying mora count (Broselow, et al. 1997, Hayes 1989, Hubbard 1995, Maddieson 1993). These studies make it clear - with very good experimental data - that the relationship between mora and segment(s) assigned to it is not a purely abstract phonological one but one which also correlates consistently at a phonetic level in terms of relative duration - even at levels of fine detail. Ternary vowel length contrasts appear to provide the clearest evidence in support of trimoraicity, since in moraic phonology each degree of length is normally distinguished by mora count. In Dithmarschen German, as in Estonian, a three-way vowel length contrast in stressed position leads to the following distribution of moras: stressed vowels (and syllables) may be short (monomoraic), long (bimoraic) or overlong (trimoraic): Dithmarschen /sprekst/ '(you) speak' vs. /spre:kn/ 'we speak' vs. /spre::k/ 'I speak'. Although three-way contrasts are considered to be exceptionally rare amongst the world's languages, the number of identified examples continues to climb slowly. Ladefoged & Maddieson (1996) refer to Mixe and Yavapai, while Laver (1994:441) adds Scottish Gaelic, Lappish, Hopi, and Nenets to the list. Hayes (1995) refers to surface three-way vowel length contrasts - derivable by rule - in Choctaw, Chickasaw, Malecite-Passamaquoddy, and St. Lawrence Island Yupik. 1.
Evidence of Overlength in Romance Languages The possibility of overlength and of three way length contrasts in the Romance-speaking world has not, to my knowledge, been previously surveyed. Outside of Italy, I am aware of only one clear example of a three-way vowel length contrast occurring in otherwise identical circumstances. Morin (1994) in a different context provides the following set of minimal pairs found in Burgundy regional French taken from Taverdet ( 1989). (1) [kry] 'raw' (m. sg.) [kry:] 'believed' (m. sg.) [kry::] 'raw' (f. sg.); 'believed' (f. sg.)
While it is accepted in the more general literature on Italian phonology that binary length contrasts (short vs. long vowels and/or short vs. long consonants)
114
JOHN HAJEK
are widespread throughout the Italian peninsula, the possibility of any kind of three-way contrast has generally missed the attention of Italian phonologists and dialectologists. Yet close inspection of the data and of the dialect literature shows considerable evidence in favor of three-way contrast in segment length in a range of Italian dialects, and, as will be argued below, in Standard Italian as well. 1.1
Overlength in Calabria and Lucania Trumper (1980) in a detailed study of related Calabrian and Lucanian dialects of Southern Italy insists that a three-way distinction is indispensable for an understanding of the vowel systems in these dialects. The distribution of vowel length is predictable, and occurs as follows. (2)
a b. c.
short ['V] long['V:] overlong ['V::]
/ / /
__CCV#, __CCVCV# __CVCV# __CV#, _ V #
Foot and syllable structure combine to condition vowel length. Vowels are always short in closed syllables, always long in antepenultimate open syllables, and always overlong in penultimate open syllables. The three degrees of length also have important secondary effects on vowel quality. Short vowels are normally lax and open, [frunna] 'frond'. Overlong vowels surface as tense monophthongs but also undergo variable diphthongization. The resultant diphthongs are clearly trimoraic: [i::] ~ [I:j], [e::] ~ [e:I] in the dialect of Amendolara. Long vowels, unlike their overlong equivalents, surface as lax monophthongs and are never diphthongized: [I:], [ε:]. 1.2
Duration and Overlength in Standard Italian Following Trumper's analysis, there is evidence to suggest a similar threeway length distinction exists in Standard Italian. The traditional view of vowel length distribution in Italian stressed syllables has already been briefly described, but is summarized here as follows. (3) a. b.
short ['V] / _CCV(CV)#, _ C # , _ # long['V:] / _CV(CV)#, __V#
With the apparent exception of open final syllables, the surface result is always considered to be a heavy/bimoraic syllable in stressed position. Even open final syllables may surface as heavy as a result of initial consonant gemination of following words, or optional vowel lengthening: [fi'ni b'bε:ne] and [fi'ni: 'bε:ne] 'it
HOW MANY MORAS?
115
finished well' (Absalom & Hajek 1997). Unstressed vowels regardless of syllable structure are of course always short. Historical developments evident in Standard Italian provide additional evidence in support of a long vs. short distinction in stressed syllables. Only long vowels in word-medial position may be diphthongized: BONU > ['bwc:no] 'good', *MELE > ['mjε:le] 'honey'. Short vowels are never diphthongized: DENTE > ['dente] 'tooth'. Expected diphthongization of long' vowels in antepenultimate position is much less regular, with a strong historical tendency towards failure to diphthongize or to secondary elimination of earlier diphthongized forms in this context, ['pε:kora] 'sheep', and ['pc:polo] for earlier ['pwc:polo] 'people'. These data are strongly suggestive that the traditional view of vowel length distribution in Standard Italian needs revision. Elsewhere (Hajek 1997) I have referred to the unusual behaviour of so-called long' vowels in antepenultimate position, and noted that compared to penultimate long vowels, they are perceptually usually much shorter, being at best half-long as confirmed by available experimental data discussed below. This same vowel shortening also accounts for absence of vowel diphthongization that, as is noted above, would otherwise be expected in long open syllables. Results of experimental studies appear to confirm antepenultimate halflength in Italian and provide important phonetic confirmation of a possible threeway length and mora count distinction. (See also Hajek 1997.) Sources 'V# 'VCVCV# 'VCV# Vayra (1984)1 219 268 183 Vayra(1984)2 180 234 159 Bertinetto(1981) 174 139 157 Romito (1994) (RC) 75 91.5 121 Romito (1994) (NA) 167 237.5 144 (RC= Reggio Calabria) (NA = Naples) Table 1. Experimental data showing stressed vowel duration in three stress conditions in Standard Italian (average of multiple tokens, no standard deviations available)
The results are consistent with no evidence of overlap across categories.1 Internal comparison of results indicates that for all speakers stressed open vowels 1 I have included in the table results only for speakers who shorten final stressed vowels in open syllables as per the traditional description of vowel length distribution in Standard Italian. The traditional description is consistent with normal patterns in Tuscany (the source of Standard Italian) and the rest of Central Italy as well as most of Southern Italy. Excluded are speakers
116
JOHN HAJEK
in penult position have by far the longest duration. Antepenultimate vowels are intermediate in duration and final stressed syllables are always the shortest. All experimental studies on Standard Italian also confirm the traditional view that vowels in closed stressed syllables, and in unstressed syllables are phonetically short. (See Hajek 1997 for details.) Vogel (1980), for instance, gives the following average figures in milliseconds based on Fava & Magno Caldognetto's (1976) large sample with three speakers.
(4)
'vccv#
'v#
*vcv
107.7
113
103 113 (*V - unstressed vowel)
*v#
'vcv# 207
Duration of stressed antepenults was not examined in this particular study, but the distribution is otherwise clear and consistent: penults are long in open syllables but short in closed syllables. Stressed vowels are short when final and unstressed vowels are always short. Short vowels across all contexts and regardless of stress condition are approximately the same in duration. The phonetic results presented above allow for the following three-way length parameter to be formulated for Standard Italian. (5)
minimum duration 'V# » 'VCCV# *V(C) (* = unstressed syllables)
'VCVCV#
»
maximum duration 'VCV#
This tripartite distribution of duration suggests that stressed vowels in penultimate position are in fact overlong and should be assigned three moras. Short vowels are monomoraic but may appear in bimoraic closed syllables, while stressed vowels in antepenultimate position are half-long and should be assigned two moras.
from Northern Italy who show a very different pattern: they never lengthen word-initial consonants and regularly lengthen the stressed vowel in open final syllables in all contexts, e.g. finì bene > [fi'ni: bε:ne] and never [fi'ni bbs:ne] which is instead common in Central and Southern Italy. Lengthening is confirmed by experimental results for Northern speakers. But note that antepenults are always clearly distinguishable by duration: they remain shorter than penults (and finals).
HOW MANY MORAS?
117
'so' 'sheep' 'bread' Figure 3. Examples showing suggested three-waymoralcdistribution
At this point, an alternative bimoraic analysis, based on recent work by Broselow, et al. (1997), Hubbard (1995), and Maddieson (1993), needs also to be considered. Following the work of these researchers, it is possible to argue that three-way durational differences of the kind reported for Standard Italian can be accounted for within a maximally bimoraic analysis by the presence and absence of so-called mora-splitting: vowels that share a mora with a following consonant are phonetically shorter, as a result of temporal overlap, than vowels which are uniquely linked to moras. If this analysis is correct, all the examples in Figure 3 would appear with maximally bimoraic stressed syllables as in Figure 4.
'so' 'sheep' 'bread' Figure 4. Examples of an alternative bimoraic distribution with mora-splitting
While such an account appears to be justified in most of the languages these authors have examined, mora-splitting is difficult to justify in Standard Italian, given the specific syllabification requirements of the language. All intervocalic segments that are in any way attached to the prosodic structure of preceding syllables as well as to that of following syllables must be long in Italian, as seen previously in the moraic representation of [fa:to] and [fatto] in Figure 1 above. Yet there is no evidence of prosodically conditioned consonant lengthening after antepenults as predicted by mora-splitting, i. e., we find ['pekora] 'sheep' and not predicted ['pek:ora] (or perhaps ['pekora] with only half-lengthened post-vocalic consonant). Moreover, all syllables in Italian, regardless of word position, which have some kind of consonantal coda attached to a mora are always closed as a result, [fatto] and never [fa.tto]. Lengthened consonants can never be assigned in Italian exclusively to syllable onset position. There is no dispute that the stressed antepenultimate syllable in words like ['pekora] is always considered to be fully
118
JOHN HAJEK
open by native speakers. An understanding of syllabification issues in Italian is critical to the analysis here, since it allows us to exclude the evidence presented by Broselow, et al. (1997), Hubbard (1995) and Maddieson (1993) in favor of morasplitting in other languages. Broselow, et al. ( 1997) make a particularly convincing case in favor of mora-splitting in Malayalam and Levantine Arabic, but in both cases mora-splitting only occurs within the same syllable (VC$) and never across syllables (V$CV). Additional evidence in support of trimoraicity in Standard Italian comes from the patterning of diphthongs. In an experimental phonetic study Salza, Marotta & Ricca (1987) measured the duration of the Italian diphthongs /au/ and /ai/ in unstressed and stressed positions. Results, presented here, are strongly suggestive that unstressed diphthongs are bimoraic while their stress-bearing equivalents are fully trimoraic. UNSTRESSED mss
STRESSED SD
mss
SD
suggested transcription
arcaicità 171 9 dàino 218 17 ['da:jno] 'archaicness' 'buck' /au/ cautèla 145 15 fàuci 242 24 ['fa:wtfí] 'caution' 'jaws' Table 2. Average duration in milliseconds (with standard deviations) of Italian diphthongs /ail/and /au/ (Salza, Marotta & Ricca 1987) /ai/
Notice first the clear separation of duration results - supported by relatively low standard deviations - between the stressed and unstressed diphthongs. Moreover, in unstressed position, the duration of the diphthongs (145-171 mss.) is much greater than that reported for monomoraic simple vowels in unstressed position (103-113 mss noted above). They are in the same duration range as sequences of the same two vowels in unstressed hiatus, e. g., ba$ullno 'small trunk' and a$izzatòre 'inciter' (hiatus marked with $), as Salza, et al. (1987) note. The duration of stressed diphthongs (218-242 mss) is very great, and is consistent with the trimoraic analysis of stressed penults in Italian. Further evidence in support of the trimoraic nature of diphthongs in stressed position is provided in the description and transcription of them as superheavy syllables - a fact missed in the literature on the bimoraic norm in Italian - either as glide + long vowel (['bwc:no] 'good') or as long vowel + glide (['ka: wza] 'cause') by linguists such as Chapallaz (1979), Lepschy & Lepschy (1988), and Maiden (1995).
HOW MANY MORAS?
119
On the basis of the experimental and descriptive evidence presented here, the suggested moraic representation of unstressed and stressed diphthongs is therefore the following.
'caution' 'jaws' Figure 5. Suggested moraic representation of Italian unstressed and stressed /au/
1.3
Overlength in Northern Italy As the discussion below is intended to make clear, Trumper (1980) is neither the first nor the only linguist to note the existence of three degrees of vowel length in varieties of Romance spoken within the confines of Italy. Although Trumper (1980) referred to three degrees of vowel length in a Southern Italian dialect, evidence of overlength can also be found in Northern Italy and is presented in the sections that follow. 1.3.1 Lombard Most descriptions of Milanese Lombard (Sanga 1988) refer to the existence of a binary vowel length contrast in this dialect, /'kru:s/ 'cross' vs. /'tus/ 'cough'. The same sources also suggest that evidence of contrast is restricted to stressed word-final syllables only. In all other positions, vowels are normally described as short: /'lana/ 'wool', /'gabja/ 'cage'. However, as early as 1928, Pavia provided a very explicit account of three degrees of vowel length in the same dialect: short, long and overlong.2 Examples of a clear three-way contrast are very difficult to find because of the particular distribution patterns of each degree of length. In general, overlong vowels are normally found in word-final syllables only: /'ki::/ 'who'. Word-medial examples are very unusual and appear only in hiatus: /'mi::a/ 'not' and 'mine'. Short vowels are found in all syllable positions, except when in hiatus: /fa/ 's/he does' and /'kamer/ 'latrine'. The middle grade of so-called long vowels is normally only found in penultimate syllables, in both open and closed position, such as /'ka:mis/ 'shirt', /'ma:rtor/ 'fool' and /'stro:leg/ 'astrologer'. Long vowels are also found, albeit rarely, in hiatus and in one or two word-final open syllables,
2
Sanga's (1988) description of length in Milanese has in recent times been taken by other linguists interested in Milanese phonology to be the primary source. Sanga refers to Pavia's (1928) description of Milanese as the best available, but does not follow his analysis.
120
JOHN HAJEK
/'mi:a/ 'my' and /'a:/ 'has'. Evidence of contrast between two degrees of length in the form of minimal or near minimal pairs can be found relatively easily. (6)
short /'kamer/ 'latrine' /'plata/ 'know-all' /'roma/ 'Rome' /'strypi/ 'crippled'
long /'ka:mola/ 'tarma' /te'a:ter/ 'theatre' /'o:men/ 'men' /'ky:pola/ 'cupola' /'mi:a/ 'my' /'tu:a/ 'your'
/ki/ 'here' /gat/ 'cat'
overlong
/'mi::a/ 'mine', 'not' /'tu::a/ 'yours' /'ku::a/ 'tail' /ki::/ 'who' /na::s/ 'nose'
The best example of a three-way set of minimal pairs that I have been able to identify to this point are the following. (7) /a/'at' /a:/ 'has; ah' /a::/ 'aah; yes, yes, it's OK'
It appears possible that confirmation of Pavia's description of three-way vowel length in Milanese Lombard may be found in the results of a recent experimental phonetic study by Bosoni (1995) of vowel duration and length contrasts in the dialects of Northern Lombardy, all closely related to Milanese. At this point, it should be noted that Bosoni did not, himself, consider the possibility of a three-way length contrast. His intention was to investigate the extent to which three Northern Lombard dialects agree or differ in terms of a simple binary vowel contrast, i.e. short vs. long. Not surprisingly, given the traditional views regarding possible length contrasts in the Italian peninsula, no attention is given to the possibility of a three-way contrast. However, closer inspection of Bosoni's data and results are suggestive of a short vs. long vs. overlong distribution. For reasons of consistency and clarity, focus here is given to the dialect of Lecco, as described by Bosoni (1995). Although the data provided by Bosoni is extremely restricted, they follow the general patterns of length distribution noted for Milanese: in the limited Lecchese data overlong vowels occur only in word-final position in open and closed syllables. Short vowels may occur finally or medially, while medium grade long vowels are only found word medially.
121
HOW MANY MORAS?
SHORT
ms
LONG
ms
OVERLONG
ms
['ba:d a] 'strap'
198
['ba::tf] 'straps'
295
['pa::k] 'pays'
272
['mat)] 'May'
121
['mat/a] 'bunch'
122
['sak] 'sack'
106
[Va:ka] 'cow'
173 201 176
['ga::mp]'legs'
237
174
['fa: :f] 'beans'
250
['ra::f] 'turnips'
258
[Tam] 'hunger'
103
['pa:ga] 'pay' ['ga:mba] 'leg'
['baf] 'steam'
130
['ba:va]'slaver'
['raf| 'by hook'
140
['ra:fa] 'snatches' 160
Table 3. Examples of vowel length in Lecchese Lombard with average duration in milliseconds (3 tokens, no SDs)
In Table 3. examples of each degree of length, with suggested length transcriptions and duration measurements in milliseconds, are provided. Each degree of length has its own approximate duration range: short vowels are 90-140 milliseconds long, long vowels are approximately 150-200 milliseconds in duration, while overlong vowels range from at least 200 to 323 milliseconds. Overlong vowels at the extreme end of this range are substantially longer than the longest vowels in Standard Italian (max. 268 mss. reported above in Table 1). Bosoni's experimental results need to be considered with some care, in particular with reference to possible confounding phonetic effects. That is, to what extent are these Lombard results influenced by or predictably dependent on the structure of the word and/or the nature of the following consonant? The possibility that the results and the suggested analysis in favor of overlength in fact be indirect artifacts of either or both of these factors needs to be eliminated. With respect to word structure and distribution of vowel length, there is no doubt that there is some strong interaction in Lecchese Lombard as in Milanese Lombard: long and overlong vowels occur typically in medial and final syllables respectively. But Milanese examples of overlong vowels can also be found in medial position; ['ku::a] 'tail'. Most importantly contrast with long vowels does occur, as noted in (6) and (7). The fact too that short vowels in light syllables can contrast word finally with overlong vowels in superheavy syllables in Lecchese (/raf/ vs. /ra::f/) and in Milanese, also suggests that word-final stress is in itself not sufficient to explain the presence of overlength. Finally, the existence of a threeway minimal pair set in Milanese (7) is also confirmation that interaction between observed patterns of length distribution and word structure may be very strong but is not completely overriding. With respect to long vowels in medial position, there appears to be frequent correlation between the middle degree of length and voicing of the following obstruent. Some lengthening before voiced consonants is to be expected as a low-
122
JOHN HAJEK
level effect (Ghen 1970, Hyman 1975), and is a sufficient basis, therefore, to account for slightly longer vowels in ['ba:va] and ['pa:ga] when compared to ['ra:fa] and [Va:ka] respectively. Note that the contextually determined lengthening does not impact on the clear category differences between suggested short and long vowels. Moreover, the correlation is not in any case a consistent one as the examples in Table 4 indicate. Vowels are apparently always long before intervocalic [g] and [b], but short before [d]. They tend to be much shorter before intervocalic [p], but longer before [k] and [t]. LONG (bimoraic) ['go :ba]'hunchback' 177 ['pa:ga] 'pay' 201 ['tapa]'halt' 146 ['va:kal'cow' 173 ['tsapa]'hoes' 152 ['ga:ta]'cat'(f) 201 Table 4. Examples of vowel duration before voiceless and voiced obstruents (average of three tokens, no SDs)
SHORT (monomoraic) ['seda]'silk' 138
Interpretation of Bosoni's (1995) experimental is of course somewhat tentative here. Clearly, a larger more detailed phonetic study of the same dialects is needed before firm conclusions can be drawn. However, it is worth noting at this stage that the same contextual correlation/distribution in Lombard is closely paralleled in Friulian (discussed in detail below). My own fieldwork on Friulian indicates that the perceptually shortest vowels (perhaps better referred to as 'overshort') are characteristically also found before [p], ['kcpe] 'nape', ['klap] 'rock' and before intervocalic [d], ['lade] 'gone' (f.sg.), [kla'made] 'called' (f.sg.). While phonetic vowel shortening before [p] is not unexpected (as reported earlier, vowels tend to be shorter before bilabials), the consistent shortening before [d] and not before [b] is more difficult to explain. Moreover the overshort quality of vowels before [p] in Friulian is striking to an English ear, and appears to be much more reduced in duration than might be expected if merely a low-level contextual effect. This phenomenon of very specific shortening (in both Friulian and Lombard) should be treated as phonological in a manner consistent with Hyman's ( 1975) and Hajek's ( 1997) view that where language-specific differences are much less or greater than the expected (low-level) effect, then this variation should be taken as evidence of phonologization of the phenomenon in the language under investigation.
HOW MANY MORAS?
123
1.3.2 Friulian According to the most widely cited descriptions of Friulian (Frau 1984, Rizzolatti 1981, Salzmann, Francescato & Bender 1952), a two-way short vs. long vowel contrast of high frequency has developed in word-final position in Friulian: flat] 'milk', [la:t] 'gone' (m.sg.). Word-medially all vowels are described as generally predictably short: [lane] 'wool*, [lade] 'gone' (f.sg.). Long vowels are described as appearing very rarely in word-medial position, but since they are assumed to derive secondarily from other phenomena such as simplification of earlier diphthongs (PAUCA > ['po:ce] 'few' (f.sg.)), they are generally disregarded in any discussion of vowel length in Friulian. The phenomenon of word-final contrast in Friulian has long attracted the attention of a number of phonologists (Hualde 1990, Morin 1992, Repetti 1989, 1992, Vanelli 1979, 1986) on two counts: (1) Friulian long vowels have a reportedly quite different distribution from long vowels in Standard Italian, and (2) there is an apparent correlation between vowel lengthening and underlying obstruent voicing and word-final devoicing: /lat/ > ['lat] 'milk', but /lad/ > [la:t] 'gone' (m.sg.) and /lade/ > [lade] 'gone' (f.sg.). With respect to the second point, the matter will not be referred to further. (Repetti 1992, citing comparative dialect data, has rightly pointed out that lengthening is independent of obstruent voicing.) Recently, I have undertaken a full re-evaluation of the Friulian vowel length question (Hajek ms). All primary data sources on Friulian, including linguistic atlas materials, and other descriptive material (such as Frau 1984) have been checked. I have also carried out new fieldwork with a native speaker of central Friulian - the variety usually described in grammars of Friulian. All evidence suggests that the standard description of the vowel length contrast and the distribution of long vowels in Friulian (such as Rizzolatti 1981) is a highly idealized one and does not in fact reflect the real situation in Friulian. Instead, all the evidence points to a pattern of three phonological degrees of vowel length in Friulian similar to the one found in Lombard: short, long and overlong. From a perceptual perspective, each degree is subject to contextual phonetic variation, so that at least four, if not more, phonetic degrees of length are discernible: short vowels may be further sub-categorized into short and overshort, long vowels vary from half-long to long, while overlong vowels are normally perceptually extremely long, although often subject to some phonetic shortening in some phrasal contexts. (See Greenbcrg's 1996 discussion of four phonetic dcgrees of length in a Slovene dialect.) Overlong vowels are generally restricted to word-final position where they contrast with short vowels: [la::t] 'gone' (m.sg.) vs. [lat] 'milk', but can also be found word-medially in some examples: ['ca::ze] 'house', [pa'tfje::tj(je] 'patience'. Short vowels can also be found word-medially: ['sute] 'dry', ['fate] 'cat' (f.sg.),
124
JOHN HAJEK
[Vene] Vein', while long vowels appear to be restricted to word-medial position: [ga'ba:ne] 'overcoat', [ba'le:no] 'do we dance?'. Striking confirmation of this new analysis of vowel length in Friulian is given in the description of vowel length by Castellani (1980), a published work which is difficult to locate and is consequently ignored in recent discussion regarding Friulian phonology. Castellani whose focus is on western Friulian, is quite explicit about three degrees of length. He notes that "si hanno tre distinte quantità, o gradi di durata, delle vocali semplici" (p. 30) in many Friulian dialects. He refers to the second degree of vowel length as Vocali medie' and note they have "un suono allungato e uniforme, teoricamente sdoppiabile, per quanto riguarda la durata, in due brevi" (p. 31). The overlong vowels are called Vocali lunghe' which have "un valore che può essere rappresentato dal prolungamento di una media con una successiva breve. " Castellani's attention to detail includes a description of the particular tonal properties of overlong vowels, a fact which was later independently confirmed by Trumper, et al. (1991) in a small-scale phonetic study of word-final length in Friulian. Word-final diphthongization in many western Friulian varieties provides further evidence in support of overlength: they are trimoraic in nature with long nucleus and short vocalic off-glide: Maniago ['me:jl], Clauzetto Friulian ['mi:el], alongside earlier ['mi::1] 'honey' still found in other Western and Central Friulian dialects. Castellani also notes that the three-way length contrast is found only in some Friulian dialects. In others the contrast is reduced to a binary short vs. long distinction, while in still others all evidence of length contrast is lost and all vowels are short. He gives the precise geographical distribution of each of these systems in addition to providing explicit cross-dialectal material of the kind given in (8). (8) Carnico Colza Spilimbergo Azzano X
overlong mu::r 'wall' pa::ri'father'
long
mu:r pa:ri
short cjar'clear' spera 's/he hopes' cjar spera mur cjar pari spera
Castellani's description of the phonotactic distribution of each grade of vowel length corresponds to the one I have already described for Friulian and to the one Pavia (1928) reports for Milanese Lombard: short vowels can be found in all positions, overlong vowels most frequently in word-final position, and the intermediate grade of long vowels word-medially. But he does note that relatively
125
HOW MANY MORAS?
rare examples of non-final overlong and final long vowels do occur. As a result it is in fact possible to identify three-way sets of close to minimal pairs. (9)
2.
short ['mus] 'donkey' ['sut] 'dry' [Viva]'alive' ['rara] 'rare' (f.sg.)
long ['bu:s] 'hole' ['ru: c] 'creek' ['bi:za] 'beige' (f.sg.) ['rno:ra] 'dark' (f.sg.)
overlong ['lu:: ] 'light' ['kra: : t] 'raw! (m.sg.) ['gri::za] 'grey'(f.sg.) [Vo::ra] 'work'
Problems and Issues in Identifying Overlength and Three Degrees of Length in Italy If, on the basis of the experimental data and descriptive material referred to, we accept that a three-way length contrast exists in Lombard, Friulian, and elsewhere, and indeed that it has been there all along, it must be asked why the phenomenon has missed the attention of practically all linguists working on the description of Italian dialects over the last one hundred years? One might argue that the differences between so-called long and short vowels are too slight for listeners to hear them. But experimental perceptual studies show that the separation of results of the type reported in this study is sufficient enough to be perceptually distinct (Hajek 1995, 1997, 1998). Moreover, it needs to be made very clear that some linguists had in fact quite clearly identified three degrees of length, Pavia (1928) and Castellani (1980) bear testimony to this fact. Their observations appear simply to have gone unobserved by other Italian dialectologists and linguists. There also other more straightforward reasons behind the failure. In the first instance, Italian linguists are trained to expect no more than a binary long short contrast in vowel and/or consonant length in any Italian dialect. Furthermore, the tremendous difficulty, if not impossibility of finding a set of three minimal pairs differing only in length has further clouded any sensitivity to more than a binary difference in length in any Romance variety spoken in the peninsula. It is not surprising that only non-native speakers of Standard Italian (Trumper for Calabrian) and native speakers of dialects with three degrees of length (Pavia for Milanese, Castellani for Western Friulian) have not felt constrained by traditional assumptions about length in the Italian peninsula. In the second instance, failure to identify three degrees of length can be explained by the distribution patterns of phonetic duration. In Figure 6 I present a simple overview of the relative duration of each degree of length (based on the statistics cited throughout the text above). In the first line Lombard is taken to be typical of all Northern Italian dialects which appear to have overlength. In the
126
JOHN HAJEK
second line I present the traditional binary interpretation of length in Standard Italian. The third line represents the new interpretation of vowel length in Standard Italian based on the data presented in this study. short Lombard
Italian 1
long | short |
overlong |
|
|
long |
|
short long overlong Italian 2 | | | | Figure 6. Phonetic duration of degrees of length in Lombard (and other Northern Italian dialects), Standard Italian 1 (according to the traditional model), and the revised distribution of Standard Italian 2 (based on the presentation of data in this study)
Comparison of the first and second lines (Lombard and the traditional view of Standard Italian) suggest that Northern Italian long vowels overlaps with both short and long degrees of length in Standard Italian. However, most of the overlap falls mainly on the side of the category of short vowels in Standard Italian. It is not surprising then that Italian linguists on hearing the middle degree of long vowels in Lombard or any other Italian dialect have generally perceived them as short - since this is in fact the most consistent categorization in a simple binary short vs. long system. My own fieldwork with a Friulian speaker confirms this same general patterning: as previously reported, Friulian short vowels are short to extremely short. The middle degree of long vowels often sound only half-long to long, while the overlong vowels are in most cases instantly identifiable as extremely long. 3.
Overlength and Trimoraicity in Bolognese: A Special Case There is evidence of an unusual pattern of segmental and syllabic overlength in Bolognese. This dialect appears to have generalized overlength and trimoraicity more than most other Italian dialects. The details of the phenomenon are different enough for the dialect to be treated quite separately from other Italian dialects dealt with in previous sections. Traditionally, Bolognese is described as having a binary vowel length contrast, and predictable consonant lengthening after short stressed vowels, for which /V:C/ > [V:C] and /VC/ > [VC:], /'ga:ta/ > ['ga:ta] »cat' (f.) but/Vela/ > [Vetta] 'life'. Post-tonic consonant lengthening is a relatively recent phenomenon, having
HOW MANY MORAS?
127
developed only in the 18th and 19th centuries. The distribution in Bolognese of long consonants is also unusual: they also occur in word-final position ([pas:] 'fish') — something which is not permitted in Standard Italian and most other Italian dialects. Two recent studies (Hajek 1995, 1998) examined phonetic aspects of the distribution and duration of segmental quantity in Bolognese. It should be noted that like Bosoni's (1995) previously reported study of Lombard, experimental work by Hajek (1995, 1998) was designed to test the validity of a simple shortlong distinction, and was not specifically intended to uncover evidence of possible three-way length. For the purposes of this section, I will restrict discussion to urban Bolognese only, since it is the results for this variety which are of most interest here: duration levels were found to be extreme in all contexts when compared to the rural variety. Selected data for one speaker of urban Bolognese are presented in Table 5 below. As the table shows, the tokens considered are similar in structure, identical in terms of vowel and post-tonic consonant, differing only in terms of expected vowel and/or consonant length. Four tokens of each item were collected in isolation (to facilitate elicitation since Bolognese is not usually written). a(:)
SD
k(:)
SD
suggested transcription sa:ka'large sack' 343 64 101 20.4 [sa::ka] sak:a'dry'(f) 114 12 314 64.5 [sak::a] sa.k'sack' 380 13.5 176 8.8 [sa::k:] sak: 'dry' (m) 144 4.5 230 18 [sa:k::] Table 5. Average duration in milliseconds of stressed vowel & post-tonic consonant in di- and monosyllables in Bolognese (4 tokens), and suggested alternative transcription
Statistical analysis (simple non-parametric) found highly significant differences between three degrees of segment duration in both V and C categories, as seen in (10). Where items are listed together in (10) under the same heading, no significant differences were found. Despite the tripartite division, I will maintain for the moment the traditional transcription, marking binary length only, of the tokens under examination, and place them between II to indicate what appears to be their more phonological than phonetic nature. However, in the rightmost column of Table 5 I have included phonetic transcriptions that more clearly indicate the apparently three-way differences in vowel and consonant durations. These transcriptions will be referred to at a later point.
128
JOHN HAJEK
(10)
a.
b.
V /sak:a/
p >.005
C /sa:ka/
p >.003
»
»
vv
/sak:/ CC /sa:k:/
» p > .0001 p > .005
» p > .001 p > .0001
VVV /sa:k/ /sa:ka/ CCC /sak:/ /sak: a/
The items analysed have been carefully chosen to avoid as much as possible the influence of possible confounding factors such as duration effects triggered by different types of vowels and consonants. In any case, analysis of the larger sample originally presented elsewhere (Hajek 1998) shows the same statistically significant distribution as in (10). One potential influencing factor that needs to be specifically considered here is possible phonetic lengthening of final segments and/or syllables in tokens uttered in isolation. If this did occur to a major degree, monosyllables and final segments should be significantly longer than their nonfinal counterparts. The results in (10) suggest that this phenomenon was only marginal in influence, with no statistically significant difference in vowel duration in the mono-/disyllable pair /sa:k/ and /sa:ka/ nor in consonant duration in the mono-/disyllable pair /sak:/ and /sak:a/. I will return to this point below. Returning to the results in Table 5, it is worth noting the extreme length of overlong vowels in Bolognese - especially when compared to figures previously given in Table 1 for Standard Italian. In Italian the reported duration range for Tong' stressed vowels in open penults is 167-268 mss. These same Italian vowels are analysed as trimoraic in this study. In urban Bolognese the same category of 'long' vowels has an average duration of 343-380 mss. Existence of vowel overlength in urban Bolognese has also been confirmed by subsequent additional fieldwork with other speakers of urban Bolognese: the same overlong vowels are immediately noticeable and may even alternate with sequences of long vowels plus breathy voiced approximant: ['pa::na] and ['pa:hna] 'cream'. In addition, diphthongs are also superheavy: ['na:jv] 'snow', [Va:wlp] 'wolf'. The phonetic evidence and additional data cited here support the notion that 'long' vowels in Bolognese are better described as overlong and trimoraic. At the other extreme, the vowel in /sak:a/ is short and monomoraic (V). In between, the vowel in /sak:/ is medium in length. It is suggested here that it be treated as bimoraic (VV) and joined with bimoraic middle length vowels found in other contexts: ['pa:sta] 'pasta'(CV:CCV). An unexpected twist in the Bolognese data - not reported to date anywhere else in Italy - is evidence in support of three degrees of consonant duration. 'Short' consonants are indeed short (101 ms) in intervocalic position after long
HOW MANY MORAS?
129
vowels, but are lengthened (175 mss.) after the same overlong vowels when wordfinal The apparently third degree of overlong duration (230-314 mss.) is found after short vowels in both medial and final positions. While each degree of consonant length appears to be contextually restricted, the statistical separation between the three degrees, as the cited figures indicate, is clear and highly significant. The first issue is to grapple with to what extent is the consonant duration effect in the Bolognese data simply a phonetic phenomenon rather than a fully contrastive phonological one. I have been careful to avoid the use of the term 'contrast' in discussion of consonant duration in Bolognese since it is possible that in this case we are indeed seeing a word-final phonetic lengthening effect in this specific context. On the other hand, three degrees of consonant length are not impossible in the world's languages: it is found in at least two languages known to this author: (1) Estonian (alongside a ternary vowel length contrast); and (2) in Sami (alongside a binary vowel length contrast). Moreover in both of these languages, consonant length is not directly inverse to vowel length: Estonian [o::p::] 'hit' and Sami ['mana:] 'you walk', ['ma:n:a:] 'the child', and ['man::a:] 'he starts to walk' (Engstrand 1987). An alternative mora-splitting analysis which allows for the combination of elements of both phonetic and phonologically contrastive analyses appears to be the most suitable in the case of Bolognese. While syllabification conditions in Standard Italian were found not to be compatible with mora-splitting (§1.2), the same difficulties do not apply to the Bolognese data. In this way, the apparently three-way duration difference need not necessarily be reflected in a direct threeway difference in mora count (ie μ vs. μμ vs. μμμ), but instead in the specific pattern of mora assignment and sharing, as Broselow, et al. (1997) and others have argued for other languages. In the examples in Figure 7, we see that in 3 out of 4 contexts, the stressed syllable is analysed as trimoraic in Bolognese. Only medial VC: sequences, e. g., [sak:a], are bimoraic. Closed syllables in word-final position on the other hand are trimoraic: [sa:k:] (V:C:,μμμ). By way of contrast, the same type of syllable is always maximally bimoraic in Standard Italian: [bar] (VC, μμ) 'bar'. In Bolognese overlong stressed vowels in medial open position ([sa::ka] 'large sack') are trimoraic, as in Standard Italian. The short duration of intervocalic /k/ corresponds to its lack of moraic weight Unmoraified, what limited duration it has comes from its syllable onset position. The increased phonetic duration of its final counterpart in /sa:k/ 'sack' (transcribed phonetically as [sa::k:]) - reported above to be the middle degree of consonant duration - can now be accounted for as a mora-splitting phenomenon, seen in Figure 7d. Its extra duration is the result
130
JOHN HAJEK
of shared moraification. The overlong degree of consonant duration is treated as fully moraic, as can be seen in Figure 7a and 7c.
a. [sak:a] 'dry (f)'
b. [sa::ka] 'large sack"
c. |sa:k:| 'dry (m)' d. [sa::k:J 'sack' Figure 7. Suggested Moriac Structure of BologneseWords
The suggested moraiç distribution of vowels and consonants are summarized in (11). Vowels can be variously linked to one, two or three moras. In the case of consonants, mora-splitting and weightless onset syllabification help to reduce the momification of three degrees of duration to a maximum of one mora. Phonotactic restrictions operating in Bolognese then conspire to block the combination of maximally trimoraic vowels with fully moraic consonants. As a result in Bolognese the maximum count per stressed syllable is three moras. (11) V duration:
Μ. >> ΜΜ >> ΜΜ + shared μ >>> μμμ V .>> V: >> V:+ C: V:: [sak::a] [sa:k::j [sa::k:] [sa::ka]
Cduration: no mora C [sa:: ka]
4.
>>
shared μ >> Μ C: C:: [sa::k:] [sa:k::] [sak::a]
Conclusion A strong traditional tendency toward favoring binarity as the upper limit in phonological analysis and description has discouraged Italian and moraic phonologists in general from looking for evidence of three-way length contrasts as well as the possibility of three moras per syllable. However, by means of both indirect and direct analysis, supported in many cases by explicit descriptions made
HOW MANY MORAS?
131
by other linguists as well as experimental evidence, this study aims to show that three-way contrasts in length and trimoraic syllables are not uncommon phenomena in Italy. In the case of Standard Italian, Friulian and Lombard, it has been argued that traditional binary accounts of segment length are inadequate and need revision. With respect to Bolognese, the available evidence points to a rather different and unusual distribution of length. In this dialect, lengthened consonants appear to be possible after overlong vowels, and trimoraic syllables occur in a wider range of contexts. The question of overlength in Italy has not yet been directly examined from an experimental phonetic perspective. However, available experimental data can be interpreted as providing indirect support in favor of trimoraicity and overiength in Italian dialects. Conclusions using such data are of course tentative, but are nevertheless intended to encourage more detailed experimental investigation. Phonetic data has been used effectively elsewhere to help establish moraic relationships down to a fine phonetic level. There is no reason, therefore, to believe that similar phonetic data cannot be used to establish moraicity, and overiength in Italy. Finally, since the varieties of Italo-Romance discussed here represent merely a first sample rather than an exhaustive survey, it is likely that continued investigation will uncover other examples of three-way length and trimoraicity in Italy.
REFERENCES Absalom, Matthew & John Hajek. 1997. "Raddoppiamento sintattico: What Happens when the Theory is on Too Tight?". Bertinetto, Gaeta, Jetchev & Michaels 1997. 159-179. Albano Leoni, Federico, ed. I dialetti e le lingue delle minoranze d il fronte all'italiano. Rome: Bulzoni. Bender, Byron, Giuseppe Francescato & Zdenek Salzmann. 1952. "Friulian Phonology". Word 8.216-223. Bertinetto, Pier Marco. 1981. Strutture prosodiche dell'italiano. Firenze: L'Accademia della Crusca. Bertinetto, Pier Marco, Michael Kenstowicz, & Michele Loporcaro, eds. 1991. Certamen Phonologicum. Torino: Rosenberg & Sellier. Bertinetto, Pier Marco, Livio Gaeta, Georgi Jetchev & David Michaels, eds. 1997. Certamen Phonologicum HL Torino: Rosenberg & Sellier. Bickmore, Lee. 1995. "Accounting for Compensatory Lengthening in the CV and Moraic Frameworks". Durand & Katamba 1995. 119-179.
132
JOHN HAJEK
Bosoni, Giorgio. 1995. "Dialettologia lombarda: un esempio di approccio strumentale allo studio delle opposizioni di quantità vocalica in sillaba tonica". Studi Italiani di Linguistica Teorica e Applicata 24.345-364. Broselow, Ellen, Su-I Chen & Marie Huffman. 1997. "Syllable weight: convergence of phonology and phonetics". Phonology 14.47-82. Castellani, Riccardo. 1980. Il friulano occidentale. Udine: De Bianco Editore. Chapallaz, Marguerite. 1979. The Pronunciation of Italian. London: Bell & Hyman. Chen, Matthew. 1970. "Vowel Length Variation as a Function of the Voicing of the Consonant Environment". Phonetica 22.129-159. Connell, Bruce & Amalia Arvaniti. 1995. Phonology and Phonetic Evidence: Papers Laboratory Phonology IV. Cambridge: Cambridge University Press. Dressier, Wolfgang, Martin Prinzhorn & John Rennison, eds. 1994. Phonologica 1992. Torino: Rosenberg & Sellier. Durand, Jacques & Francis Katamba, eds. 1995. Frontiers of Phonology: Atoms, Structures and Derivations. London: Longman. Engstrand, Olle. 1987. Salient Features of Lule Sami Pronunciation. Samisk Ljudlära. Department of Phonetics, University of Umea. Fava, Elisabetta & Emanuela Magno Caldognetto. 1976. "Studio sperimentale delle caratteristiche elettroacustiche delle vocali toniche e atone in bisillabi italiani" Simone, et al. 1976. 189-235. Féry, Caroline. 1997. 'The Mora as a Measure of Weight and a Syllabic Constituent". Bertinetto, Gaeta, Jetchev, & Michaels 1997. 91-110. Francescato, Giuseppe. 1989. "Friaulisch: Interne Sprachgeschichte I. Grammatik". Holtus, Metzeltin & Schmitt 1989. 563-610. Frau, Giovanni. 1984. Friuli (Profilo dei dialetti italiani). Pisa: Pacini. Giannelli, Luciano, Nicoletta Maraschio, Teresa Poggi Salani, & Massimo Vedovelli, eds. 1991. Tra rinascimento e strutture attuali. Torino: Rosenberg & Sellier. Greenberg, Marc L. 1996. 'The Vowel System of the Sredisce Dialect in Prlekija, Slovenia, based on the Descriptions of Karol Ozvald". Studies in South Slavic and General Linguistics 23.87-99. Languages. Haiman, John & Paola Benincà. 1992. The Rhaeto-Romance London: Routledge. Hajek, John. 1995. "A First Acoustic Study of the Interaction between Vowel and Consonant Duration in Bolognese". Rivista italiana di acustica 19.1-9. . 1997. Universals of Sound Change in Nasalization. Oxford: Blackwell. . 1998. "Analisi acustica delle quantità segmentali in area bolognese". Rivista italiana di dialettologia 21.133-147.
HOW MANY MORAS?
133
. ms. "Length, overlength and moraicity in Friulian". University of Melbourne. Hayes, Bruce. 1989. "Compensatory Lengthening in Moraic Phonology". Linguistic Inquiry 20.253-306. Holtus, Günther & Kurt Ringger, eds. 1986. Raetia antiqua et moderna. Tübingen: Max Niemeyer. Holtus, Günther, Michael Metzeltin & Christian Schmitt, eds. 1989. Lexikon der Romanistischen Linguistik (vol. 3). Tübingen: Max Niemeyer. Hualde, José. 1990. "Compensatory Lengthening Friulian". Probus 2.31-46. Hubbard, Kathleen. 1995. "Towards a Theory of Phonological and Phonetic Timing: Evidence from Bantu". Connell & Arvaniti 1995. 168-187. Hyman, Larry. 1975. Phonology, Theory and Analysis. New York: Holt, Rinehart & Winston. Ladefoged, Peter & Ian Maddieson. 1996. The Sounds of the World's Languages. Oxford: Blackwell. Laver, John. 1994. Principles of Phonetics. Cambridge: Cambridge University Press. Lehiste, Ilse. 1970. Suprasegmentals Cambridge: MIT Press. Lepschy, Anna Laura & Giulio Lepschy. 1988. The Italian Language Today. London: Hutchinson. Maddieson, Ian. 1993. "Splitting the Mora". UCLA Working Papers in Phonetics 83.9-18. Maiden, Martin & Mair Parry, eds. 1997. The Dialects of Italy. London: Routledge. Maiden, Martin. 1995. A Linguistic History of Italian. London: Longman. Marchetti, Giuseppe. 1952. Lineamenti di grammatica friulana. Udine: Società Filologica Friulana. Mazzola, Michael L., ed. 1993. Issues and Theory Romance Linguistics. Washington, D. C : Georgetown University Press. McCarthy, John & Alan Prince. 1993. Prosodic Morphology I. Technical Reports of the Rutgers University Center for Cognitive Science 3. Morin, Yves-Charles. 1994. "Phonological Interpretations of Historical Lengthening". Dressier, Prinzhorn & Rennison 1994. 135-155. Nicoli, Franco. 1983. Grammatica milanese. Busto Arsizio: Bramante Editrice. Pavia, Luigi. 1928. Sulla parlata milanese e suoi connessi. Nuovi studî fonicografici, filologici, storici, comparativi. Bergamo. Prieto, Pilar. 1993. "Historical Vowel Lenghtening in Romance: the Role of Sonority and Foot Structure". Mazzola 1993. 87-107.
134
JOHN HAJEK
Repetti, Lori. 1989. The Bimoraic Norm of Tonic Syllables in Italo-Romance. Ph.D. dissertation, UCLA. . 1992. "Vowel Length in Northern Italian Dialects". Probus 4.155-182. . 1994. "Degenerate Syllables in Friulian". Linguistic Inquiry 25.186-193. . 1997. "The Syllable". Maiden & Parry 1997. 52-57. Rietveld, A. C. M. & U. N. Frauenfelder. 1987. "The Effect of Syllable Structure on Vowel Duration". Proceedings of the International Congress of Phonetic Sciences 11.28-31. Rizzolatti, Piera. 1981. Elementi di linguistica friulana. Udine: Società Filologica Friulana. Romito, Luigi. 1994. "Cenni sui correlati elettro acustici dell'accento in alcune varietà d'italiano". Salza 1994. 107-120. Salza, Pier Luigi, Giovanna Marotta & Davide Ricca. 1987. "Duration and Format Frequencies of Italian Bivocalic Sequences". Proceedings of the International Congress of Phonetic Sciences 11.113-116. Salza, Pier Luigi, ed. 1994. Gli aspetti prosodici delVitaliano. Rome: Tipografia Esografica. Sanga, Glauco. 1988. "La lunghezza vocalica nel milanese e la coscienza fonologica dei parlanti". Romance Philology 61.290-297. Simone, Raffaele, et al., eds. 1976. Studi di fonetica e fonologia. Roma: Bulzoni. Taverdet, Gérard. 1989. Le français régional parlé en Bourgogne: Étude phonologique. Dijon: CRDP. Tranel, Bernard. 1995. "On Phonetic Evidence for the Phonological Mora: Comments on Hubbard". Connell & Arvaniti 1995. 188-201. Trumper, John. 1980. "La zona Lausberg ed il problema della frammentazione linguistica". Albano Leoni 1980. 267-303. Trumper, John, Luigi Romito, & Marta Maddalon. 1991. "Double Consonants, Isochrony and Raddoppiamento Sintattico: Some Reflections". Bertinetto, Kenstowicz & Loporcaro 1991. 329-360. Vanelli, Laura. 1979. "L'allungamento delle vocali friulano". Ce fastu? 40.66-76. . 1986. "La fonologia dei prestiti friulano". Holtus & Ringger 1986. 355376. Vayra, Mario. 1991. "Appunti su un fenomeno di 'centralizzazione' nel vocalismo dell'italiano standard". Giannelli, Maraschio, Poggi Salani & Vedovelli 1991.195-212. Vayra, Mario, Cinzia Avesani & Carol Fowler. 1984. "Patterns of Temporal Compression in Spoken Italian". Proceedings of the Tenth International Congress of Phonetic Sciences ed. by M. P. R. van den Broeke & A. Cohen, 541-546. Foris: Dordrecht.
HOW MANY MORAS?
135
Vogel, Irene, 1980. La sillaba comeumitàfonoiogica. Bologna: Il Mulino. Young, Diane. 1993. "Compensatory Processes and the Bimoraic Canon'5. Romance Philology 46.231-250.
STRESS STABILITY UNDER CLITICIZATION AND THE PROSODIC STATUS OF ROMANCE CLITICS* MICHELE LOPORCARO University of Zurich
0.
Introduction Among central and southern Italo-Romance varieties, three different patterns of enclitic stress are observed, which are exemplified in (1) with data from Standard Italian and two dialects of Campania.1
In (li), the Standard Italian pattern, stress remains on the lexically stressed syllable of the verb form, regardless of the number of enclitics hosted by the verb. At the other extreme, in Stabiese ((liii)), as in many dialects of southern Campania, Lucania (Liidtke 1979:32), and northern Calabria (Formentin 1994:222, Pace 1993-94:41), stress is reassigned whenever one or more clitics are attached to the verb. In the intermediate pattern, exemplified by Neapolitan in (1 ii), I am grateful to Lori Repetti and one anonymous referee for many helpful suggestions. I am of course solely responsible for any inaccuracies and errors which may have persisted. The following abbreviations will be used throughout the paper: ANT = antipassive, CG = clitic group, DO = direct object, F = foot, IO = indirect object, ISV = intervocalic /s/-voicing, MPR = morphonological rule, MVT = mid-vowel tensing, NSI = northern standard Italian, N&V = Nespor & Vogel (1986), PPh = phonological phrase, PR = phonological rule, PW = phonological (or prosodic) word, SI = Standard Italian, TNA = total nasal assimilation, SVL = stressed vowel lengthening. 1 For the data on Stabiese (spoken in Castellamare di Stabia, south-east of Naples) I am indebted to Livio Gaeta. All other data, whenever unreferenced, are drawn from my own field notes.
138
MICHELE LOPORCARO
stress is reassigned only when two enclitics follow the host. This is the most widespread pattern of stress reassignment among the dialects of the Center-South, from Lazio and Abruzzi to Apulia (Loporcaro 1988:244), Calabria (Scerbo 1886:47) and Sicily, and it is also found in Sardinia.2 The functional motivation underlying the patterns of stress reassignment in (1ii-iii) resides in a constraint on Italo-Romance stress: as in many Romance varieties, stress usually falls on one of the three last syllables of the word. Stress shift in (lii-iii), thus, prevents the creation of strings stressed on the fourth to last syllable. Conversely, as shown in (1i), such strings do arise in Standard Italian. Current explanations of the contrast between the Standard Italian (henceforth, SI) stress pattern and those found in the dialects (Bafile 1993, 1994, Monachesi 1995, Peperkamp 1995a, 1997) are based on the following assumptions (Berendsen 1986, Inkelas 1989). (2)
a. b.
clitics are prosodically dependent elements clitics can subcategorize for different prosodic categories
Assumption (2a), which is uncontroversial, follows from the fact that clitics are endowed with defective prosodic structure: by definition, they do not bear stress by themselves and in order to be prosodified and phonetically realized they have to adjoin to an adjacent prosodie category. It is further argued in (2b) that this category can be the PW or some higher prosodie constituent (the PPh, or the CG, for those who include the latter in the prosodie hierarchy). Assumption (2b), in Bafile's or Peperkamp's accounts, is motivated by the empirical evidence in ( liiiii). If clitics force stress to be reassigned, it is suggested, this implies that they form part of the domain for stress assignment, i. e., the PW. Conversely, if clitics fail to induce stress reassignment, as is the case in SI ((1i)), it is inferred that they are not adjoined to the PW but instead are directly incorporated into the PPh. This paper questions the validity of the claim in (2b). I will argue that there is no conclusive evidence that Romance clitics can subcategorize for different prosodie domains. In particular, I will show that the pattern of SI columnar stress in (li) is not a valid argument to infer that clitic incorporation into prosodie structure takes place directly in the PPh domain, skipping the PW. Rather, I will propose that Romance clitics are invariably adjoined to PW, as shown in (3). 3 2 For example, Logudorese Sardinian ['ma:niya] 'eat', ['ma:niyalu] /[maniya'öi:lu] 'eat it'. A different pattern of stress shift is found in the dialects of Campidano (Bolognesi 1998:452). 3 Similar analyses have been put forward, within different theoretical frameworks, by Dressier (1985:34), Booij (1996). See also Zec (1993:367), who replaces the notion 'clitic group' by 'postlexical phonological word'.
STRESS STABILITY
139
(Prosodic constituency, in (3) and henceforth, is marked by tagged, round brackets.) (3)
a. b. c.
('piy:a)PW (('piy:a)PW lo)PW (('piy:a)PW telo)PW
The adjunction of one or more clitics to the lexical PW (the host) results in the creation of an outer (postlexical) PW including the host plus the clitic(s). This is, I propose, the only prosodic representation required to account for all the phonological properties of Romance clitics. This has the following consequence for the argument to be developed in what follows: unlike recent dialectological literature in the field which takes the PW-external status of SI clitics for granted and basically focuses on the formalization of the two patterns of stress shift in (1ii-iii), I will instead concentrate on SI and on other dialects displaying stress stability under cliticization. The organization of this paper is as follows. In § 1 I show that, given the representations in (3), the contrast between SI ((1i)) and southern dialects ((liiiii)) simply follows from a parametric option: the former, unlike the latter, never allows stress to be reassigned postlexically. If this is correct, then lack of stress shift is no longer a relevant diagnostic to conclude that clitics in SI are PWexternal. In §2, I discuss segmental rules of SI which are commonly held to be PW-sensitive, and could consequently bear on the issue of the prosodification of clitics. I conclude that none are actually PW-bound. In §3, the empirical scope of the discussion is extended by considering genuinely PW-bound rules applying in other Romance varieties of Italy which display the stress pattern in (li): /epenthesis in Algarés (§3.1), and two rules of vowel raising applying in Romanesco (§3.2) and Friulian (§3.3). I then briefly review a process of vowel insertion and a process of vowel harmony observed in some northern Italian dialects (§3.4). These data, it is shown, are straightforwardly accounted for under (3), whereas they would be impossible to deal with under the assumption that stress stability implies adjunction of clitics to some prosodie category higher than the PW. Based on the evidence discussed, in §4 I conclude that (3) should be acknowledged as the only answer to the question of how Romance clitics are prosodified. Finally, one remark concerning the style of the arguments is in order here. This paper crucially relies on notions such as 'phonological word' and on the distinction between a lexical and a postlexical level. These notions are, in my opinion, indispensable theoretical tools for any phonologist. Thus, making use of
140
MICHELE LOPORCARO
them and discussing their status does not necessarily imply commitment to any of the specific theoretical frameworks which have elaborated on these notions. The lexical/postlexical distinction, while first termed that way within Lexical Phonology (Kiparsky 1982), became the subject of theoretical discussion outside that framework (see Dressier 1985:115 for Natural Phonology, Lombardi 1996:32 for non-derivational frameworks such as Optimality Theory). As to the PW, I believe that any model of phonology needs to distinguish between at least a notion 'word' defined in morpho-lexical terms and one defined in phonological terms. This is the only implication carried by the use of the label PW in this context. Put in other words, this is a paper about phonology. It deals with empirical data from the dialects of Italy and establishes some empirical results, which bear on the general theoretical (i.e., non-theory-internal) issue of the prosodic status of clitics. 1.
Stress Stability in Standard Italian The patterns of variation in enclitic stress in southern Italian dialects ((1iiiii)) has received much attention recently (Bafile 1993, 1994, Monachesi 1995, Peperkamp 1995a, 1997).4 This type of stress shift is contrasted with SI stress stability, which is in turn explained with the assumption (2b): clitics do not form part of the same PW with their host; therefore, stress assignment is never affected by cliticization. To illustrate this approach, consider Peperkamp's (1995a) prosodic analyses of SI vs. Neapolitan strings consisting of verb plus two clitics.5 4
An important piece of philological work on this topic is Formentin (1994:218), who shows among other things that in Old Neapolitan (14th century) the attraction of stress by strings of two clitics was not yet categorical if the host was an oxytone. Variation is attested as late as one century ago by Capozzoli (1889:21): ['dim:әlә]/[di'm:el:ә] 'tell me (it)', ['dam:әla]/[da'm:el:a] 'give her to me', ['fam:әnә ]/[fa'm:en:ә] 'make some for me'. It can be noted in passing that this fact seems to pose a problem for Peperkamp's (1995a) Optimality Theoretic account of the Neapolitan stress pattern. Peperkamp's constraint ranking derives categorical stress shift in both [da'm.el'.a] (oxytonic host) and [piy:a'tel:9 j (paroxytonic host). Thus, there is no obvious way in which the difference between the two could be derived. In depth discussion of this topic, however, exceeds the scope of the present paper. 5 In what follows, when providing prosodie analyses of the materials discussed, I will neglect the foot. I agree with Peperkamp that double clitics form a foot in Neapolitan where they receive main stress postlexically, and that this cannot be the case in Standard Italian where they remain stressless. Peperkamp further assumes, following a widely held view, that feet are maximally binary. However, for both SI and the dialects exemplified in (1ii-iii), there is solid evidence that ternary feet must also be acknowledged. In other words, there is no firm empirical ground to support the claim that the last syllable of proparoxytones is extrametrical. (See Bafile 1994, 1996 for a discussion of the Italian and Neapolitan facts, leading to this conclusion.) In my opinion, the foot has to be defined, very traditionally, as the sequence of a stressed syllable plus (if left-headed) all the following unstressed syllables. This implies that in SI pígliamelo the
141
STRESS STABILITY (4)
a. b.
(((('piy:a)F)PW('tel:ә)F)PW)PPh ((('piy:a)F)PW telo)PPh
Neapolitan Standard Italian
However, an alternative account of the Italian stress facts readily comes to mind. Assume that in both varieties clitics are adjoined to PW, as shown in (3). 6 The contrast between SI and Neapolitan with respect to main stress assignment can then be accounted for by means of a simple parametric choice ((5b)). (5) a. b.
stress assigned lexically (= in the inner PW) stress assigned postlexically (= in the outer PW)
(5a) must be set positively for all languages which have phonologically nonpredictable stress. These in turn divide into those, like Neapolitan, which choose the positive option for (5b), and those, like Italian, in which main stress assignment is strictly lexical (hence the minus in (5b)). This statement mirrors in a straightforward way the observation that the stress patterns in (lii-iii), but not the one in (li), involve postlexical re-assignment of stress. This is a piece of factual evidence which has to be acknowledged independently. And once this is acknowledged, as in (5), there is no need to assume that the prosodic status of clitics varies across Romance languages. It is easy to conceive of crucial empirical evidence to discriminate between the two alternative types of approaches: the one which assumes the contrast in prosodic constituency in (4), vs. the one advocated for here which assumes that the conditions under which clitics enter prosodie structure are constant across Romance varieties (as in (3)). The two approaches make diverging empirical predictions: under the one upheld here, it is predicted that strings of host+clitic(s) (one, two or three) always behave like PWs with respect to phonological rules of the language which make reference to the PW-domain in their structural description. Under the alternative view, conversely, which assumes PW-recursion for Neapolitan only and PPh-incorporation for SI, it is predicted that strings lexical foot of the host ('σ σ) is restructured postlexically to include all of the unstressed syllables provided by the clitics ('orσσσσ).While discussion of foot structure is crucial for those dialects in which stress reassignment takes place, it is less so for varieties like Standard Italian, to which the rest of this paper is devoted. 6 This proposal, which has been put forward by Booij (1996), is rejected by Peperkamp (1997:204).
142
MICHELE LOPORCARO
composed of host plus two clitics will behave like PWs in the former, but not in the latter. We consequently have to survey varieties of the SI type, in order to find out (a) whether they display phonological rales related to the PW, and, if they do, (b) how clitics behave with respect to these rules. We will begin our investigation, in §2, with SI. 2.
(Allegedly) PW-Sensittve Rules in Standard Italian In the relevant literature, some segmental rules applying in SI are commonly said to be sensitive to the PW-domain. In Nespor & Vogel (1986:124-34) (henceforth, N&V), four such rules are discussed. All of them are PW-span rules: intervocalic /s/-voicing (ISV), total nasal assimilation (TNA), stressed vowel lengthening (SVL), and mid-vowel tensing (MVT). Among these processes, SVL is not relevant to our present concern, since we are looking for diagnostics of PWhood to be applied to clitics within varieties in which these never carry stress. The remaining three processes are discussed in the following sections (§2.1-23). 2.1
Total Nasal Assimilation (TNA) The final nasal of a prefix obligatorily assimilates in place to the following stem-initial consonant. If this is a sonorant, assimilation also affects nasality. As shown in (6a), this fact is reflected in Standard Italian orthography: (6)
a.
irregolare/*inr~ immaturo/Inmcollaterale/*conl-
'irregular' 'immature' 'collateral'
b.
in rima 'in rhyme' con molli 'with many con loro 'with them'
This assimilation does not take place, N&V argue, when the nasal occurs as the final segment of a preposition (see (6b)). Note that alternatives such as *ir rima, *com molti, etc. would be nowadays unacceptable. Based on the data in (6ab), the authors propose the rule in (7) (with minor simplifications). (7)
[+nasal] -->
α nasal ß place
/ (...
[+sonorant [a nasal [ß place
..)PW
Let us assume this is correct and elaborate on this conclusion. We seem to have here a test for PW-hood which could in principle be applied to clitics. If TNA takes place at the boundary between host and clitic (or vice versa), this implies that the analysis proposed here (in (3)) is correct, and that alternative
STRESS STABILITY
143
analyses which treat Italian clitics as PW-extemal are to be rejected. The latter include Bafile's, Peperkamp's and N&V's analyses, in which it is claimed that clitics form a PW separate from their host. The above prediction is not tested in the Prosodic Phonology literature on Italian for a quite obvious reason: this literature has only dealt with contemporary SI, and no clitic in SI ends in a nasal. The same is true of verb forms (imperative, gerund and infinitive) to which clitics can encliticize. So there is no possible SI input to feed TNA. However, this has not always been the case in the history of the language. In earlier stages of Romance, clitics could be enclitics of finite verb forms. Among these, 1st and 3rd person plural forms (/-Vmo/, /-Vno/) could end in [m] and [n], respectively, if they underwent troncamento (i.e., word-final post-sonorant vowel deletion). For Old Italian, then, we are in a position to carry out the test. In (8) some relevant examples are extracted from the 13th century Florentine texts edited by Castellani (1952). (The page number is included in parentheses after each example.) It is appearent that TNA takes place before clitics beginning with a sonorant. The conclusion which follows from combining (7) and (8) is straightforward: in Old Italian, clitics were adjoined to the PW, not directly incorporated into the PPh/CG. (8)
a.
b.
c.
d
avenue (
Unfortunately, things are not so simple. The description of TNA mirrored in (6a-b) captures a genuine contrast which is, however, best expressed in the following terms: TNA is compulsory in (6a), while it is optional in (6b). In fact, although this is not apparent in (modern) orthography, all of the phrases in (6b) can be realized - and indeed they normally are in connected speech at least by speakers from Tuscany and central Italy - as [ko 'l:o:ro], [ko 'm:olti], [i 'r:i:ma].
144
MICHELE LOPORCARO
This has always been the case in the history of Italian, and earlier, in Latin, at least since the 1st century B.C., as proved by the fact that such assimilations were graphically recorded before the orthographic standard reflected in (6b) was established. Assimilation of final nasals is massively attested in the written sources of Old Italian (see the 13th century examples in (9a)) and, even earlier, in late Latin inscriptions ((9b)), as well as in metalinguistic judgements by Latin authors dating back to Cicero ((9c)). (See Loporcaro 1997a:43 for a discussion of these and other Latin sources recording sandhi TNA.) (9)
a.
co llo [r] rio portare 'with their bad behavior' (Contini 1960, I:415), i-llei 'in her' (II:500), co-llei 'with her' (II:561), be-llo saccio 1 know very well' (<— /'ben lo/, I:491), be-llo dëon sapere 'they have to know' (<— /'ben lo/, II:253); u rromeo 'a pilgrim' (Schiaffini 1926:6), co-mmaravigliosa forza 'with incredible strength' (Schiaffini 1926:175).
b.
plus amor imment est (- immentest) 'in mente est' 'is on (his) mind' CIL VI 30125; feminas im matrimonio suo iu [n]xerint '(they) married:SUBJ women', im mu[ro]pos[t templum divi] Aug. ad M[inervam] 'on the wall behind the temple of divus Augustus' (Dessau 1954-55 2 ,I:407, n°2010)
c.
«Libenter etiam copulando verba iungebant ut sodes pro si audes, sis pro si vis [...] Quid, illud non olet unde sit, quod dicitur cum illis, cum autem nobis non dicitur, sed nobiscum? quia si ita diceretur, obscaenius concurrerent litterae, ut etiam modo, nisi autem interposuissem, concurrissent» (Cic. Orat. 154) '[our ancestors] used to connect words when they put them together, as in sodes instead of si audes, sis instead of si vis [... ] How could one fail to recognize the reason why we say cum illis, not however cum nobis, but rather nobiscum? Because if we did say so, the letters would knock into each other giving a rather obscene result, as they would have right now, had I not interposed autem'
We can now combine the data discussed so far ((6)-(9)) with a fact concerning Latin, and that is that TNA could apply optionally at a prefix boundary as well. In that context, this optionality was reflected in standard orthography: irritus/inritus 'vane, useless', collido/conlido 'I hit', commisceo/conmisceo 'I mix together'.7 This historical evidence, thus, allows us to reconstruct the history ofTNA as shown in(10). 7
Actually, there is consensus among Indo-European scholars and Latin philologists that orthographic optionality does not represent true phonetic variation in this case. Loporcaro (1999:131-133) provides arguments against this traditional view.
145
STRESS STABILITY
word boundary
(10) a. b.
TNA: stage 1: LATIN TNA: stage 2: ITALIAN
prefix boundary
optional
optional
optional
categorical
(=postlexical) (=lexical)
If, following standard assumptions, we take optionality as a diagnostic for postlexical rule application, this forces us to admit that TNA was originally born as a postlexical rule, applying regardless of morphological structure. It was later extended upwards through the grammar of Proto-Romance. The end product of this change is that TNA applies lexically in Italian, ever since the earliest records of the language, when it is fed by prefixation. This corresponds nicely to a recurrent pattern of phonological change, often described in the literature, according to which "rules are added postlexically and gradually move up through the levels" (Hargus & Kaisse 1993b: 14; see also Dresher 1993, Kaisse 1993, Zec 1993). As Kaisse (1993:345) puts it, "as a rule ages, it extends its domain upward to earlier strata within the lexicon, perhaps curtailing its postlexical application as it moves". In the case of TNA, such a curtailment has not taken place. On the one hand, TNA has undergone a change in that its application has become categorical in some morphologically defined contexts; on the other hand, however, TNA has never ceased to apply optionally at the postlexical level. As a consequence, two instances of TNA application have to be distinguished for today's SI which we can call TNA-1 and TNA-2, as shown in (11). (11) a. TNA-1
type
applies
input
level
restrictions
PR
optionally
any N
post-lexical
none
b. TNA-2
MPR
categorically prefix-final N
lexical
derived
before [+son] before f+son]
environment
|
All current approaches to Italian TNA fail to recognize this duplicity. N&V take only TNA-2 into account, but they treat it as though it were postlexical. This is why a prosodic domain has to be mentioned in its context. (Recall that in N&V's model the only rules for which no prosodie domain needs to be specified are those of lexical phonology; see footnote 14 below.) However, once it is recognized that TNA-1 is also at work in Italian (and that it is postlexical since it applies optionally and across word boundaries), it follows that TNA-2 must be lexical. Otherwise, no reasonable account could be given of the diverging properties listed in ( 1 la) and ( 1 lb). Peperkamp (1995b, 1997:84-6), on the other hand, suggests that the data in (6) are best accounted for by assuming that TNA is a lexical rule. The empirical
146
MICHELE LOPORCARO
argument provided to substantiate this claim is that "the proper names Enrico and Manlio contain consonant sequences of nasal plus sonorant that are not subject to assimilation" (1995b:233). The same also goes, she observes, for lexicalized compounds such as manrovescio 'backhanded blow' (from mono 'hand' + rovescio 'back'), monregalese 'inhabitant of Mondovì', as well as for prefixed words such as circonlocuzione 'circumlocution', panromanzo 'common to all Romance languages', etc. This is, however, only half of the story. It is true that the lexical representation - and the orthography - of some borrowed (Enrico) or learned (Manlio) names, as well as words like manrovescio, circonlocuzione, etc. contain such sequences. But it is surely not true that their phonetic realizations are immune to assimilation. All of these words, in fact, can be pronounced with assimilation: [e'r:i:ko], ['mal:jo], [mar:o'vεf:o]. Indeed, for speakers from Tuscany and central Italy this is the most natural realization. Non-application of TNA is possible for these speakers (including myself) only in slow and hyperarticulated speech. If we now turn to dialect variation, clear evidence for the postlexical application of TNA-1 is offered by Romanesco. This variety has a postlexical rule degeminating /rr/ —> [r], both word-internally and at word boundaries. In (12a) degemination affects morpheme-internal geminates, while in (12b) the geminate arises from the final /r/ of the definite article /er/ plus the initial /r/ of the following word. ((12c) is added for comparison.) (12)
a. b. c.
(13)
a.
b. c.
d
/'kar: o/ /'ter:a/ /er rega't:sino/ /er ro'mano/ /er 'kane/
—> —> —> —> —>
['ka.ro] ['t:ra] [e rega't:si:no] [e ro'ma:no] [er'ka:ne]
'carriage' 'earth' 'the little boy' 'the Roman' 'the dog'
/en'riko/ /man'riko/
—> —>
[e'ri:go] [ma'ri:go]
'Enrico' 'Manrico'
/manro'vε :o/ /in + r / /in + r / /un re'galo/ /un ro'mano/ /in 'rima/ /un 'kane/ /in'kasa/
—> [.maro'vε :o] —> [irego'la:re] —> [irespon'tsab.ile] —> [u re'ga:lo] —> [uro'ma:no] —> [i'ri:ma] —> [un 'ka:ne] —> [in'ka:sa]
'backhanded blow' 'irregular' 'irresponsible' 'a present' 'a Roman' 'in rhyme' 'a dog' 'at home'
STRESS STABILITY
147
A degeminated [r] is the final output of /n/ + /r/ strings as well, both wordinternally (in derived as well as non-derived environments) and at a wordboundary. This can be seen from the data in (13), paralleling those in (12). Obviously, for /nr/ to take part in [r:]-degemination, it is necessary that it first be converted into the appropriate input [r:]. This follows from the application of TNA-1. The Romanesco pattern would remain unexplained if TNA were assumed to be either a lexical rule, as proposed for SI by Peperkamp, or a PWspan rule, as N&V suggest. In sum, strings consisting of nasal plus sonorant never ceased to represent a difficulty for speakers of Italian throughout the history of the language. Any account of both the synchronic status and the diachrony of TNA has to recognize that this mirrors a genuine constraint on pronounceability for speakers of the language (Stampe 1979). The evidence witnessing to this persistent status is overwhelming. Historically, morpheme-internal nasal plus sonorant sequences, occurring exclusively within loanwords, underwent assimilation as testified by the fact that, for example, Germanic names such as Konrad or Heimrich were adapted into Tuscan as Corrado, Arrigo (the latter co-occurring with Enrico, also normally pronounced, although not spelled, with a geminate [r:]).8 The same goes for Gallo-Romance loans: Old Provençal onrat 'honored' was borrowed into Old Tuscan as orrato, Old French denrée was borrowed as derrata 'victuals', etc. An alternative strategy to solve the difficulty is epenthesis: this shows up in lexicalized forms such as Endrigo (family name, also from Heimrich, via Enrico) as well as in language acquisition. My daughter, Laura (at age 2;6), has free variation between TNA-1 and epenthesis, as in [i r:o'vε :o]/[in dro'vε :o], literally in rovescio, for SI a rovescio 'upside down'. Peperkamp (1995b:234) is correct in observing that the occurrence of the lexical strings /nl/, /nr/ in the words mentioned above militates against N&V's account: if TNA were a categorical PW-span rule, as in (7), then none of these strings would ever be allowed to surface. But Peperkamp's approach fails to predict that Manlio, Enrico, manrovescio and the like can optionally undergo TNA. 8
In Medieval manuscripts, however, the spelling Errico also occurs (for example, in the Ottimo Commento della Commedia, Florence 1334: Errico di Brettagna). Variability is also observed in the spelling of the other proper name mentioned by Peperkamp as proof of the nonapplication of TNA morpheme-internally. This is easily ascertained through a search in the database of the TLIO, available online at the site www.csovi.fi.cnr.it, which contains all published Italian texts up to 1375. Along with the Latinate spelling Manlio, which occurs 190 times, there are 68 occurrences of the spelling Mallio, recording the application of TNA, in texts from Tuscany, Urnbria, and Rome. (I am indebted to Retro Beltrami for allowing me to consult the database.)
148
MICHELE LOPORCARO
The approach sketched in (11), on the other hand, accounts for all of the above facts. TNA-2 must apply at the lexical level, since (a) it is structurepreserving and neutralizing; (b) it applies categorically only when its input arises via a morphological operation, such as prefixation; and (c) it suffers lexical exceptions (/en'riko/ vs. /irre'ale/ 'unreal'). These exceptions, indeed, do not have to be listed, contrary to Peperkamp's claim. Rather, they follow straightforwardly from one of the basic properties of lexical rules: TNA-2 is blocked in non-derived environments. That proper names fit into this category is obvious; but this is also true of manrovescio which - as Peperkamp rightly argues - is lexicalized, i. e., was historically restructured through deletion of the internal morpheme boundary originally separating the two stems. From blocking of TNA-2 in non-derived environments, it follows that the phonemic strings /nl/, /nr/ are allowed to occur within lexical morphemes. However, that these can optionally be input to assimilation is evidence for the existence of TNA-1, which must be postlexical because (a) it is optional, and (b) it applies in sentence phonetics across word-boundary, as well as (c) in non-derived environments, within lexical morphemes (/en'riko/ —> [e'r:i:ko]/[en'ri:ko]). Needless to say, once the existence of TNA-1 is acknowledged, the pattern of assimilation affecting clitics in Old Italian (see (8)) is also explained nicely as the product of postlexical application of TNA-1. However, the final consequence of our scrutiny of TNA for our present discussion of the prosodic status of clitics is that we have lost a potential test. In fact, as will have become clear by now, neither TNA-1 nor TNA-2 can be legitimately regarded as PW-bound: the former, because it applies across words; and the latter, because the statement in (7) predicts, contrary to fact, that the only possible pronunciation of Enrico and Manlio is the one displaying assimilation. The MPR deriving irreale from in+reale, whose application is obligatory, must refer to the morphological categories involved (prefix/stem boundary), rather than to prosodic constituency. It applies, after all, only under prefixation. Thus, we are entitled to expunge TNA from the list of PW-related phonological processes of Italian. 2.2
Mid-Vowel Tensing (MVT) MVT applies to the low mid-vowels /ε/ and /c/ in forms such as those in (14b-c), as opposed to (14a, d). (14)
a. b. c. d
p['ε]rdo I lose' p[e]rdiâmo 'we lose' p[e]rdénte loser' p['ε]rditémpo 'time-waster'
t[o]sto t[o]stiamo t[o]statúra t['c]stapáne
'I toast' 'we toast' 'toasting' 'bread toaster'
STRESS STABILITY
149
As apparent from (14), MVT affects /s/ and /o whenever they become unstressed under inflection or derivation. The fact that tensing does not apply to /s/ and /o/ occurring in the first stem of compounds leads N&V to conclude that MVT has the PW as its domain of application, as seen in (15).9 (15)
V
-->
[+tense]/( ...
Again, reference to the prosodic domain is no more compelling for the statement of this rule than it was for TNA. Tensing takes place in unstressed position only. Consequently, there is no reason to expect that it should affect the stressed vowel in the first stem of a compound. Reference to stress, which is independently needed (see (15)), is both necessary and sufficient to derive the correct result. That reference to [-stress] is, and always was, a necessary and sufficient condition for the application of MVT becomes even clearer if one takes into account the behavior of three further classes of morphemes. (16)
a. b. c.
clitics: le 3fplDO < *lε < ILLAE; ne PART < *nε < INDE monosyllabic prefixes: pre- < prε < PRAE- 'pre-' unstressed monosyllabic conjunctions: e < e < ET 'and'; o < o < AUT 'or'
Let us begin with clitics. As shown in (16a), the 3fplDO clitic [le] and the partitive clitic [ne] should have surfaced with [ε] if they were stressed, since the diachronic sources of SI /e/ and /o/ are Latin E, AE and O, AU, respectively. This fact could provide evidence that they are part of the same PW with their hosts if MVT were a reliable test for PW-hood. But this account runs into problems with (16b) and (16c). Monosyllabic prefixes ending in a vowel, such as pre-, are analyzed as separate PWs by N&V. (Vogel & Scalise 1982 even claim that they carry stress underlyingly.) This analysis is at odds with the fact that MVT has applied to their vowels. No problem arises, conversely, if we assume that: (17)
a. b. c.
monosyllabic prefixes are unstressed they form one single PW with their hosts MVT applies to unstressed vowels, regardless of prosodic constituency
9 In the original formulation, instead of [stress] a specification [DTE] (Designated Terminal Element) occurs. The difference is irrelevant here, since a DTE is one which bears "primary stress in the domain indicated" (N&V 1986:130).
150
MICHELE LOPORCARO
Claim (17a) needs no demonstration; rather, it reduces to a factual observation given the application of MVT in ( 16b). Claim ( 17b) is at odds with the claim that ISV is a PW-span rule; in fact, ISV does not apply in, for example, pre[s]entire 'to have a presentiment'. This point will be taken up again in §2.3, where it is shown that ÎSV is actually not a PW-span rule. Claim (17c), finally, is further confirmed by the data in (16c). The monosyllabic conjunctions /e/ and /o/ underwent MVT. Obviously, they do not form one single PW with the lexical material to which they are adjacent in the speech chain. From this it follows that: (18)
a. b.
/e/ 'and' and loi 'or' are unstressed (contrary to what is claimed by Basbøll (1989:24), Sluyters (1990:73))10 for MVT to apply, [stress] is both a necessary and sufficient condition
Thus we have to conclude that MVT is not PW-bound. Consequently, its potential value as a diagnostic for the prosodic status of clitics evaporates. 2.3
Intervocalic /s/- Voicing (ISV) The last potentially relevant rule to test the prosodic status of clitics is ISV. This is a much-debated phenomenon whose basic facts (for Northern Standard Italian = NSI) are summarized in (19).11 N&V (1986:129) propose the statement of intervocalic/s/-voicing in (20), which they call 'prosodie' voicing. (See also Vogel 1991:93, Nespor 1993:173.)
10 Basbøll (1989:24) and Sluyters (1990:73) make this claim to account for the fact that these monosyllables trigger Raddoppiamento Fonosintattico (= RF), e [t:]u 'and you', o [t:]u 'or you'. As argued in Loporcaro (1997a:3), however, this cannot be true; e and o trigger RF because they carry an idiosyncratic rule feature [+RF] in the lexicon. Note that this must be assumed anyway for the few penultimate stressed polysyllables which also trigger RF (come [t:]e 'like you'): final -e in ['ko:me], after which RF takes place, cannot possibly be underlyingly stressed. 11 NSI rather than Tuscan SI (= TSI) is usually considered when discussing ISV because NSI has no phonemic voicing contrast for /s/. ISV, thus, is free to apply across the board (although with a restriction to be discussed below). In TSI, on the contrary, /s/ and /z/ do contrast, as shown below (Muljacic 1972:47, Camilli & Fiorelli 1965:45-8). a. ['fu:zo] 'melted' b. [Tu: so] 'spindle' ['kjε:ze] 'churches' ['kjε:se] 's/he asked' [in'tfi:za] 'carved:fsg' [in't i:sa] 'Incisa'(place name) Within lexical morphemes, ISV is not allowed to apply: this leaves room for voice contrast in TSI (a-b)) vs. NSÏ (19a~b). For all the remaining cases in (19c-h), however, TSI shows the same distribution of [s] and [z] as NSI. From this we can conclude that ISV must be active in TSI too. However, in this variety it has to be constrained to apply only in derived environments.
151
STRESS STABILITY (19)
a. b. c. d
kre[z]ima 'confirmation' ro[z]+ellina 'little rose' di[z]+educare 'to miseducate' bi[z]+avolo 'great-grandfather'
(20)
Prosodie s-voicing: s —> z/ (... V
e. f. g. h.
a+[s]ociale 'asocial' ri+[s]alire 'to go up again' anti+[s]ociale 'antisocial' para+ [s]ole 'sun-shade'
V ... )PW
The authors assume that /s/-voicing, as stated in (20), has the phonological word as its domain of application. On this assumption, the distribution exemplified in (19) naturally leads to the following definition of the phonological word in Italian. 12 (21)
Italian PW (N&V 1986:129, Vogel 1991:92, Nespor 1993:173) a. stem plus all suffixes (derivational, evaluative, inflectional) (= (19a-b)) b. stem plus monosyllabic prefix ending in a consonant (= ( 19c-d)) c. other prefixes (monosyllables ending in a vowel, bisyllables) (= (19e-h))
/s/-voicing, thus, becomes a diagnostic for the upper limit of the prosodic domain PW. Whenever /s/ between vowels does not voice, this implies that it occurs on the edge, rather than in the middle, of the prosodie constituent. Based on this premise, ISV is used as evidence for the prosodie status of clitics. As shown in (22), /s/-voicing does not take place in Italian at the boundary between a clitic and its host. (22)
a. b. c.
lo [s]enti vedendo[s]i recato(vi)[s]i
'you feel it' 'seeing oneself' 'having gone (there)'
Peperkamp (1995a:245, 1995b:230-2, 1997:74) adopts N&V's account of ISV as a PW-span rule (although she has a definition of PW which differs from (21) as to the status of prefixes). Consequently, she also uses it as an empirical argument to substantiate her analysis of Italian clitics, which coincides with N&V's in considering clitics as PW-external, although adjoined directly to PPh, without the intermediate prosodie category CG.
12 The distinction made between vowel- vs. consonant-final monosyllabic prefixes in (21b-c) is highly controversial. The issue cannot be dealt with here. See, however, the comments referring to (16b) above for a discussion of some incorrect predictions which follow from the assumption that monosyllabic prefixes such as a-, ri-, pre- ((19e-f), (16b)) are PWs separate from their hosts.
152
MICHELE LOPORCARO
However, several alternative accounts of ISV have been proposed recently which do not at all make reference to the PW-domain. Kenstowicz (1996:371-3) puts forward an Optimality Theory account based on Correspondence Theory, according to which voicing of stem-initial /s/ after prefixes is blocked whenever the stem independently occurs as an autonomous word in the language. Thus, voicing in (19e) is blocked given the existence of sociale 'social'. Loporcaro (1999), on the other hand, after scrutiny of comparative evidence from other Romance varieties, concludes that a statement of ISV such as (20) fails to account for cross-linguistic variation in ISV. Crucially, many western Romance varieties display the same patterns as NSI (i.e., (19)), but in addition they show voicing of final /s/ when this becomes intervocalic in sentence phonetics. A few examples from Algarés (the Catalan dialect spoken in Alghero; see §3.1 below) illustrate this fact. (The underlying forms of the words concerned are added for comparison.) (23)
/'fas/ /Vas/ /'altrus/ /'pres/ /'kutjus/
'you do like this' ['faza' i] 'you go there' ['vazana'ya] ['altruz a'mits] 'other friends' [I 'e 'prez a'vuj] 'I have taken it today' [j 'a 'masa 'kutfuz anag'ki] 'there are too many dogs here
Clearly, it is impossible to claim that final /s/ is PW-internal here. It follows that (20), as it stands, cannot be the correct description of ISV neither in this variety nor in Standard Catalan, Portuguese or Rumantsch, which display the same voicing pattern (i.e., (19) plus (23)). On the contrary, all cross-linguistic ISV-data can be handled nicely once it is recognized that ISV, across Romance, is not constrained prosodically, but it is blocked by a morphological boundary preceding /S/.13
As to our issue concerning the prosodic status of SI clitics, this implies that ISV loses its value as a diagnostic for PW-hood. Thus, we have to conclude this section by recognizing that our search for segmental processes providing reliable tests to assess the prosodic status of clitics has been frustrated so far. Actually, upon closer inspection of the rules of SI which are usually claimed to be PW-
13 Actually, inflectional boundaries do not block voicing, as shown by the data in (23), where plural and 2nd person -/s/ does undergo the process. This can be explained by appealing to morphological structure: inflectional boundaries are the weakest morphological boundaries (Loporcaro 1999:141-143, Bertinetto 1999). Therefore, it is natural that precisely these boundaries fail to block the application of a phonological rule.
153
STRESS STABILITY
sensitive, it turns out that these must indeed be described without any reference to PW (or to other prosodic domains). 14 Even though the TNA and the MVT facts discussed in §2.1-2.2 would have corroborated the prosodic analysis of clitics proposed here, we have to abandon these arguments and extend our empirical database beyond standard Italian in search of relevant evidence. 3.
PW-Sensitive Rules in Some Romance Dialects of Italy The next step is to look for other varieties, related to SI, in which cliticization does not affect stress-placement ((1i)), and in which some genuinely PW-related rule is at work. Three such cases are discussed in §3.1-3.3.
3.1
Algarés The Catalan dialect spoken in the linguistic enclave of Alghero (northwestern Sardinia) parallels Standard Italian in that clitic placement never results in stress shift. In (24), a list of Algarés DO and IO clitics is provided. (24) DO IO
singular 1st 2nd ma ta ma ta
3rd m/f lu/la (l)i
plural 1st 2nd mus vus mus vus
3rd m/f lus/las (l)i
refl. sa sa
As in SI, clitics always follow the verb when this is in the imperative or in the gerund, whereas they are proclitic to infinitives. (See Loporcaro 1998:117-124 for more details on the syntax of clitics in this variety.) The examples in (25) show that no stress shift occurs. 15
14
The mention of a prosodie domain for all of the rules discussed above is deductively, rather than empirically, motivated. It follows from a basic starting hypothesis of N&V's model of Prosodie Phonology according to which "the domain of application must be stated explicitly for all rules" (N&V 1986:15). This applies, in their model, to truly phonological rules, to the exclusion of rules of lexical phonology. More recently, stronger claims have been made in the same line of research, a line which Kisseberth (1994:133) calls "[t]he domain-based view of phonology," aiming at a unification of lexical and prosodie phonology, as is the case for Inkelas' (1989) Prosodie Lexical Phonology. According to Inkelas (1993:79), "[w]ithin the lexicon, just as outside it, phonological rules never apply to morphological strings directly. Rather, they always apply within domains described by p-constituents" (p-constituent = phonological rule domain). 15 In feminine third person clitics, /l/-flapping followed by deletion may optionally take place: /1/ —> [r] --> 0.
MICHELE LOPORCARO
154 (25)
a. b. c.
['pclta] 'bring' ['pcltama] 'bring:1sg REFL' ['pcltamaxa] 'bring:1sgREFL 3fsgDO'
['fes] 'make' [Testa] 'make:2sg REFL' [festara] 'make:2sg REFL 3fsgDO'
[pu'zaw] 'put:2p [pu'zavus] 'put:2pl REFL' [pu'zavuras] 'put:2p! REFL 3fsgDO'
If stress stability under cliticization entails the analysis of clitics as PWexternal (see (4b)), it follows that the strings in (25b-c) should not behave as PWs. Conversely, if PW-recursion is assumed, as in our proposal in (3b-c), it is expected that the strings in (25b-c) will exhibit PW-like behavior with respect to relevant tests. The next step is to find evidence other than stress assignment in order to check which prediction is correct. As argued in Loporcaro (1997b), Algarés indeed displays a PW-sensitive rule, which can be used as a test for our purposes. As shown in (26a), an epenthetic [i] is inserted at word boundary ( P W 1 _ P W 2 ) , whenever PW 2 begins with a consonant and PW1 ends in a consonant cluster, in a non-continuant obstruent, or in /f/. (The underlying representation of PW1 is added for comparison in (26b), and corresponds to the prevocalic/prepausal phonetic realization.) (26)
a.
[as'tiki Tent] ['sinki d al'mants] [l 'e feti pru'ra] [l'e 'visti pru'rant] [as'krifi 'be]
'I am doing' 'five brothers' 'I made him cry' 'I saw him crying' 'I write well'
b.
/as'tik/ /'sink/ /'fet/ /Vist/ /as'krif/
The process is clearly syllable-driven: the word final consonants (or consonant clusters) in (26) do not correspond to admissible word-internal codas. Epenthesis allows their resyllabification as onsets of the syllable built postlexically around the epenthetic vowel When however PW1 ends in a consonant which does correspond to a possible word-internal coda, epenthesis fails to apply, as illustrated in (27). (27)
a.
[se ) fent] ['trez djal'mants] [lu Tas pxu'ra] [a'kel 'djova]
'we are doing' b. 'three brothers' 'you make him cry' 'that young man'
/'sem/ /'tres/ /Tas/ /a'kel/
STRESS STABILITY
155
A simplified statement of the epenthesis process is given in (28). 16 (28)
0
-->
[i] /
*C) P W __PW(C [where *C cannot be syllabified]
Given (28), epenthesis qualifies as a domain juncture rule. That the proper context for its application is the PW~boundary is further shown by the following facts. On the one hand, as shown in (29a), [i]-epenthesis does not apply within phonological words, even when the appropriate segmental conditions arise via morpheme concatenation (such as in plural formation). (29)
a. b.
[Vent] 'wind' [pa'rent] 'relative' [lu 'venti s 'es kal'mat] pu pa'renti 'sap]
plural: [Vent+s]/*['ventis] plural: [pa'rent+s]/*[pa'rentis] 'the wind abated' 'the relative knows'
As shown by comparison of (29a) and (29b), the very same clusters are epenthesized across PW-boundary. On the other hand, epenthesis regularly applies between the two members of compounds, whenever the syllabic/segmental condition in (28) is satisfied. (30)
['kap]'head'+ [msstra]'master, craftsman' —> [,kapi'mεstra] 'masterbuilder' ['sent] 'one hundred' + ['kambas] legs' --> [.senti'kambas] 'centipede'
Application of epenthesis in (30) precludes the possibility that the relevant prosodic domain is the PPh or any other prosodic category higher than the PW. Thus, this process seems to be a reliable diagnostic for PW-hood: given (28), any string at whose boundaries [i]-epenthesis takes place will qualify as a PW. Let us now have a look at what happens at the boundary between host and clitic. Given the assumptions expounded in §0 above, according to which clitics are not phonological words themselves (see (2a)), we expect epenthesis not to take place between a clitic (proclitic or enclitic) and its host, even when the segmental/syllabic condition for its application is satisfied. A potential site for epenthesis is created when a clitic encliticizes to a gerund, since gerunds end in a 16
Actually, the main point in Loporcaro (1997b) consists in showing that the epenthesis rule cannot be reduced to syllabification conditions. However, the simplified statement in (28) assumed as a provisional generalization in Loporcaro (1997b:210) - is sufficient for our present purposes, since what is crucial here is the prosodie domain mentioned in the epenthesis rule, not the exact nature of the rale as such.
MICHELE LOPORCARO
156
consonant cluster (/nt/: [panti'nant] 'combing', [pal'dent] 'losing', [pal'tint] 'leaving'). As shown in (31), epenthesis takes place between gerund and clitic when the latter is proclitic to a following finite verb. (Small capitals stand for contrastive focusing.) (31)
1sg 2sg 3sg 1pl 2pl 3pl
[panti'nanti [panti'nanti [panti'nanti [panti'nanti [panti'nanti [panti'nanti
ma'so] ta 'ses] s 'es] mus 'sem] vus 'sew] sa 'son]
'I am COMBING myself 'you are COMBING yourself 's/he is COMBING him/herself 'we are COMBING ourselves' 'you are COMBING yourselves' 'they are COMBING themselves'
However, when the clitic is hosted by the gerund, epenthesis never takes place, and cluster simplification applies instead. 17 (32) [panti'nama]/*[pantinantima] [panti'nanta]/*[panti'nantita] [panti'namus]/* [panti'nantimus] [panti'na vus]/*[panti'nanti vus] [panti'nantsa]/*[panti'nantisaj
The same happens when two clitics follow the verb. (33)
a.
b.
([la 'ma]) [as'tiki f an'tamara/*ran'tantimafa] '(my hand:fsg) I am washing itrfsg' /ran'tant+mala/ (pus 'psws]) p 'e 'visti fan'tantsalts/*fan'tantisalts] '(his feetimpl) I have seen him washing them:mpl' /ran'tant+salus/
From the absence of epenthesis in (32)-(33), thus, it is possible to conclude that the starting assumption in (2a) is indeed correct: clitics are prosodically 17 The converging strategies for cluster simplification at work in (32) arc reminiscent of the mainland Catalan situation (Mascaró 1978:52f, 1989). Algarés used to be like Barcelona Catalan until the last century, as cluster simplification applied in all contexts including in sentence phonetics. As shown in Loporcaro (1997b), epenthesis is a relatively recent rule which is only sporadically attested at the end of the 19th century (Guarnerio 1886). It had become optional by the time of Kuen's (1934) description. Nowadays, it is a categorical postlexical PR, as described in (28).
157
STRESS STABILITY
defective units, not separate PWs, as assumed, for instance, in N&V. 18 (If they were separate PWs, the starred forms in (33a-b) would be predicted to be grammatical.) On the other hand, the occurrence of epenthesis in (31) shows that it would be difficult to extend to Algarés the analysis proposed for Italian clitics by Vogel (1994). In her proposal, the PW in Italian consists of the stem plus any suffixes, whereas "items that are not PWs (i.e. prefixes) [...] join as sisters of PWs into some larger unit" (1994:113). This larger unit is taken to be the clitic group, as shown by the prosodic analysis proposed for prefixed words, reported in (34). (34)
(anti(sociale)PW)CG 'antisocial' 'unequal' (dis(uguale)PW)CG ((ripre(lava)PW)CG ((subito)PW)CG)PPh Xs/he) prewashes again immediately'
Although the prosody of clitics is not discussed in Vogel (1994), it follows from her definition of PW that proclitics, like prefixes, are not part of the same PW as their hosts, although they are no longer treated as separate PWs as they were in N&V. If extended to Algarés, this approach would yield the representations in (35a), which can be compared with the one following from the proposal advocated for here, given in (35b). (35)
a. b.
(((panti'nant)PW)CG (ma ((panti'nant)PW (ma
('so)PW)CG)PPh ('so)PW)PW)PPh
(= Vogel 1994) (= (3b))
As can be seen, analysis (35a) would not account for the occurrence of epenthesis in (31) unless the statement of epenthesis in (28) is expanded, so that not only PW-boundary but also CG-boundary qualifies as the appropriate prosodiC context. Conversely, analysis (35b) predicts the occurrence of epenthesis without any ad hoc adjustment being required.
18 Note that the data in (31)-(32) also witness to the correctness of the statement in (28), where PW-boundary is represented as in (a) rather than as in (b): a. ...)PW — PW(... b. ...)pw or PW(... Had the rule mentioned only the right edge of the preceding PW, we would have expected epenthesis to be triggered also before enclitics, which is not the case. Since the relevant prosodie category is here the PW, even if clitics are involved, we seem to have here one more argument to expunge the category 'clitic group' from the prosodie hierarchy (ZeC 1993:367).
158
MICHELE LOPORCARO
We now have to check the behavior with respect to epenthesis of sequences consisting of a host plus two clitics. Consider the imperative forms of the verb 'to give', listed in (36). (36)
['dona] 'give:2sg' /1/ '3msgDO' /la/ 3fsgDO' /lus/ '3mplDO' /las/ 3fplDO'
['do+ma] 'give.me' ['do+ma+1] ['do+ma+ra] ['do+ma+lts] ['do+ma+xas]
['dona+mus] 'give.us' ['dona+mu+1] ['dona+mu+Ra] ['dona+mu+1ts] ['dona+mu+xas]
The only forms which provide us with a relevant test are those in which the second clitic is the 3mplDO form /lus/. This clitic undergoes vowel deletion and consequently ends in a consonant cluster: [V+1ts]. All the remaining clitics, on the other hand, end in a vowel or in [s], an admissible coda consonant, and consequently do not satisfy the description of the epenthesis rule. If a verb form imperative or gerund - is followed by two clitics the latter of which is 3mplDO [1ts], epenthesis takes place between this string and the following word, as shown in (37). (37)
a. b.
['donamultsi 'legu] 'give us them immediately' ([lus 'pews])[1'e 'visti Ran'tantsaltsi 'sentsa sa'bo] '(his feet) I have seen him washing them without soap'
Here again, the PW-recursion analysis accounts for the occurrence of epenthesis, while alternative analyses would not. Following Peperkamp's (1995a) proposal, stress stability under cliticization is evidence that sequences of two clitics which do not attract stress (as in SI, but not in Neapolitan) are directly incorporated into PPh skipping the PW. If extended to Algarés, this would yield the analyses in (38a). (In (38b), the analyses following from the approach developed here are added for comparison.) (38)
a. b.
(('dona)PWmults)PPh (('dona)PWmults)PW
((ran'tant)PWsalts)PPh ((ran'tant)PWsalts)PW
(cf. (4b)) (cf. (3b))
The analyses in (38a), however, would fail to account for the occurrence of epenthesis in (37), given the statement of the rule in (28). It follows that either the rule (28) or the analysis (38a) must be incorrect. Assume for a moment that the former is the case. We could then restate the rule substituting PPh-boundary for PW-boundary: this would account for epenthesis after(('dona)PWmults)PPh, if
STRESS STABILITY
159
analyzed as in (38a). That this cannot be true, however, is easily understood as soon as we consider (39) and (40). (39)
a. ((as'tiki)PW b. (('vwiti)PW
(ran'tant)PW)PPh ('filts)PW)PPh
'I am washing' eight sons'
(40)
a./'set +'sent+s/ b./'vwit + 'mil/ c. /'solt + 'mut/
—> —> —>
'sevenhundred' 'eight thousand' 'deaf-mute'
[.seti'sents] [.vwiti'mil] [.solti'mut]
As shown in (39), epenthesis applies between the (aspectual) auxiliary and the main verb in the progressive periphrasis (39a), as well as between the determiner and the noun in (39b). Under current assumptions about prosodic constituency (Ghini 1993), the strings in (39a-b) form one single PPh. Even if we were willing to reconsider such basic assumptions of prosodic constituency, this would not suffice to rescue the analysis (38a). Epenthesis, in fact, applies within compounds, as shown in (30) and here again in (40). Obviously, compounds cannot be assumed to span over two distinct PPh's. The only possibility left, under the prosodie analysis in (38a), is to complicate rule (28) by introducing a disjunction into the application context ('either at PW or at PPh boundary'). This is of course undesirable and unnecessary, as (28) in its present form perfectly accounts for all of the relevant data. Conclusion: if (28) is correct, then this demonstrates the superiority of the prosodie analysis in (38b) over the alternative in (38a). We consequently have empirical evidence to conclude that clitics, in Algarés, are added to their host forming one single postlexical PW. Stress stability, in Algarés as in SI, is straightforwardly accounted for as shown in (5): in neither variety is stress ever reassigned postlexically. This is all we need to say to explain the behavior of clitics with respect to stress. 3.2
Romane sco Algarés is not at all an isolated case. On the contrary, many other Romance varieties display PW-related segmental processes in which clitics behave as PWinternal. Romanesco offers evidence which is relevant for our present concerns. In this dialect, as in SI, stress is never shifted in verb plus clitic clusters. (41)
a. ['ko:ri] 'run' b. ['ko:rife] 'run there' c. [ko:rimefe] 'run there for me'
[Ven:i] 'sell' ['ven:ila] 'sell it:fsg' [Ven:imela] 'sell me it:fsg'
160
MICHELE LOPORCARO
In Romanesco a rule applies in verb inflection turning /a/ (occurring as the 1st conjugation thematic vowel, as well as in the imperfect morpheme -va) into /e/ when unstressed and non-final. This is illustrated by the verb forms (all in the indicative, present or imperfect) in (42). (42) a.
' to steal' [ru'b:a:mo]
'to eat'
' to beat'
[ma' :a:mo]
[kor'ka:mo]
pres. 1p1 imperf.3sg
pres.3sg
b.
['rub:a]
['ma :a]
c.
[ru'b:a:va]
[ma' :a:va]
['korka] [kor'ka:va]
d
['rub:eno]
['ma :eno]
['korkeno]
pres.3pl
e.
[ru'b:a:veno]
[ma' : a: veno]
[kor'ka:veno]
imperf.3pl
The vowel surfaces as fa] when stressed ((42a, c)) or word-final ((42b, c)), whereas it turns to [e] when it is followed by other phonetic material before the word-boundary ((42d-e)). Note that the environment cannot be specified in purely phonological terms. An /a/ in the penultimate syllable of a proparoxytonic word remains unaffected if it occurs within the root, no matter whether this corresponds to a verb (as in (43a)) or to other lexical categories ((43b-f)).19 (43)
a.
[se'skr :fana]
*[se 'skr :fena]
' s/he gorges'
b. c.
[na'b:u:fala] [er'tsirr.ago]
*[na'b:u:fela] *[er'tsin:ego]
' a fake' ' the mayor'
d e.
[na'k :fana] ['sab: ado]
*[na 'k fena] *['sab:edo]
' a big pot, a lot' ' Saturday'
f.
['s :malo]
*['sc:melo]
* Somali'
The phonological contexts in (42d-e) and (43) are identical; the only contrast is in the difference in morphological structure. Given the above, the rule can be stated as shown in (44). (44)
X)PW
Crucial for our discussion is the fact that (44) affects the final -a of 2sg imperatives (which are phonetically identical to the 3rd person indicative forms in (42b)) when one or more clitics are attached to them. This is illustrated in (45). 19 The morphological constraint illustrated in (43) may not have been at work in past stages of the language, as witnessed by some lexicalized remnants of a-raising in nouns: ['mc:niga] 'nun', ['stcm:igo] 'stomach' (cf. SI monaca, stomaco).
161
STRESS STABILITY (45)
a. b. c.
[korka]/* ['korke] ['korkelo] 'beat him' ['korkemelo]
['rubra]/* ['rubre] ['rub.elo] 'steal it' ['rub:emelo] 'steal it for me'
'eat.ANT with it
'beat him for me'
The fact that (44) applies in both (42d-e) and (45b-c) follows straightforwardly under the PW-internal analysis of Romance clitics advocated for here: since the clitics in (45) enter the PW postlexically, they create the appropriate environment for verb-final /a/ to change to /e/. On the other hand, on the assumption that clitics are PW-external in all varieties in which they are invisible to stress, there is no obvious way to explain jointly the application of (44) in both (42d-e) and (45b-c). It has been claimed that in both SI and Italian dialects the 3rd person ending -no, as in (42d-e), is PW-external in order to account for the fact that only in 3rd person plural forms can fourth to last stress occur. (See Bafile 1994:15 who proposes extrametricality of the -no ending for Neapolitan, etc.) This is shown in (46) for Romanesco. (46)
a. b.
[te'lε:fona] 's/he calls up' [te'lε:foneno] 'they call up'
['pεt:ina] 's/he combs' ['pεt:ineno| 'they comb'
If then both clitics and the 3rd person ending -no are PW-external, it could be assumed that they join the host (or the verb stem) at a higher node (again, the PPh or the CG). Under this assumption, (44) could be restated as applying in PPh-internal position, rather than PW-internally. But this is incorrect, as shown by the fact that stem-final -a in compounds whose first member is a verb stem remains unchanged.20 (47)
potato-peeler' brown nose' 'junk yard'
Under the account developed here, non-application of (44) in (47) follows automatically given the current prosodic analysis of Romance compounds: stem20
Here, as noted by one referee, specification of the domain in (44) as CG rather than PW would still derive the observed facts. However, that the clitic group has to be dispensed with on independent grounds was already argued above (footnote 18). Note further that reference to PPh in (44) would not work either: within compounds (47), the first stem bears stress in Romanesco, as in Standard Italian.
162
MICHELE LOPORCARO
final /a/ is in PW-final position; therefore, (44) cannot apply. Conversely, if a — > e were PPh-bound, its non-application in (47) would remain unexplained, since a would occur after stress but not in domain-final position. 33
Friulian Friulian displays a rule directly comparable with the one just described for Romanesco. 21 In most dialects, as well as in the standardized koinè from which the following data are drawn, Latin word-final -A was raised to ~e. This -e occurs in the forms in (48i), which exemplify 1st conjugation 3sg present indicative ((48ia)), along with 1st class feminine nouns ((48i-b)) and adjectives ((48i-c). 22 (48) a b. c.
i. puarte 'bring:3sg' cjale 'look:3sg' c jase 'house' orele 'ear' straiche 'tired:fsg' gnove 'new:fsg'
ii. puartis 'bring:2sg* cjalis 'look:2sg' c jasis 'houses7 orelis 'ears' strachis 'tired:fpl' gnovis 'new:fpl'
puartin 'bring:3pl' cjalin look:3pl
As illustrated by the (48ii) examples, -e is raised to i when it is not in wordfinal position. The relevant domain for raising to take place is the PW. This is shown by non-application of raising in compounds, where -e at the end of the first stem (be it nominal, as in (49a), or verbal, as in (49b)) remains unchanged. (49)
a.
b.
jarberave 'beetroot' -jarbe 'grass' + rave 'turnip' aghedivite 'aqua-vitae' = aghe 'water' + di vite 'of life' tristelenghe 'slanderer' = triste 'wicked' + lenghe 'tongue' uzzefnàrfis 'knife-grinder' = uzze- 'grind' +fuarfis 'scissors' puzzefadie 'idler' -puzze- 'stink' + fadìe 'labor' gjavestropui 'corkscrew' = gjave- 'remove' + stropui 'corks'
This -e is not word-final, morphologically, but it does occur in PW-final position phonologically. First conjugation 2sg imperative also has final -e (< Latin -A); as in many Romance varieties, it is homophonous with the pure stem (see
21
I am indebted to Laura Vanelli for pointing out to me the relevance of the Friulian data in this connection. 22 All the examples in (48)-(50), from Marchetti (1977), are given in Friulian orthography (cj =
[c], gj =[I],gh =[g],gn = [n]
STRESS STABILITY
163
(49b)) and the 3sg indicative present (see (48a)). Crucially, -e is raised to -i when the imperative hosts one or more enclitics. (50)
a. b.
cjale look' cjalimi look at me' cjaliti look at yourself ' cjalinus look at us'
puàrte 'bring' puàrtimal 'bring it to me' puàrtime 'bring her to me' puàrtimes 'bring them: f to me'
Note that, as apparent from the (50b) examples, Friulian belongs to the (1i) type varieties: encliticization does not result in stress shift. 3.4
Vowel Epenthesis and Vowel Harmony in the Dialects of Coli and Piverone Much other evidence could be discussed here to show that Romance clitics are (postlexically) PW-internal. In this section, I briefly mention two further phonological processes observed in some northern Italian dialects. It will be shown that with respect to these processes host plus clitic clusters pattern like lexical PWs. Consider first the vowel insertion process discussed in Repetti's (1995:284) study of the dialect of Coli (prov. Piacenza). In this dialect, vowel insertion applies to allow resyllabification of word-final consonant clusters. The epenthetic vowel is inserted between coda consonants of increasing sonority ((51a)), whereas it is inserted word-finally after clusters of equal or decreasing sonority ((51b)). (51)
a.
[Ve:dәv] 'widower' [re:dәn] 'reins'
b.
['fumnә] 'females' ['ms:rlә] 'blackbird'
Crucially, the same pattern is observed with enclitics, as shown by the contrast between (52a) and (52b), paralleling (51a) and (51b), respectively. (52)
a.
/'di m 1/ ---> 'tell:sg me it'
['dimәl]
b.
/'dε n n/ --> ['dεnnә] 'give:pl us some'
Another relevant phenomenon is observed in the Piedmontese dialect of Piverone (prov. Vercelli), which displays a systematic vowel harmony process first described by Flechia (1898). (Sec also Loporcaro 1997c: 17 for discussion.). In Piveronese, final non-low vowels are harmonized to the stressed vowel, surfacing as [e]/[o] if preceded by a low or low mid stressed vowel ((53a)), and as [i]/[u] if they follow a high or high mid stressed vowel ((53b)).
164
MICHELE LOPORCARO (53)
a
[kcöe] ['pεre] [kanto] ['maro]
'cooked:fpl' 'stones' 'sing:3pl' 'sour'
b.
[bryti] ['fridji] ['styju] ['tfitu]
'ugly:fpl' 'cold fpl' 'cleanse:3pl' 'silent'
As shown in (54a-b), the final vowels of enclitics follow the same harmony pattern. (54)
a.
['dame]'give:sg me' b. ['matte] 'put.REFL on' ['dam-ne] 'give:sg mes ome' ['mat-lo] 'put:sg it'
['mus-mil 'show:sg me' [das'vidj-ti] 'wake.REFL up' [an'dum-s-ni] 'let's go away' ['pij-lu] 'take: sg it'
The domain within which vowel harmony applies in this variety, following a cross-linguistically common pattern (van der Hulst & van de Weijer 1995), is the PW. Thus, clitics must be PW-internal. 4.
Conclusion In the foregoing discussion, I aimed to demonstrate that clitics throughout Romance are always adjoined to the (lexical) PW and form a postlexical PW with their host. To show this, I had to refute alternative accounts according to which clitics, in some varieties, may be directly prosodified into domains such as the CG or the PPh, higher up in the prosodic hierarchy. The main argument on behalf of these alternative accounts is based on stress stability under cliticization in varieties such as SI (see (li)). However, I have shown that stress-stability effects do not provide compelling proof, since they follow independently from the distinction between lexical vs. postlexical levels (§1). I have further argued that several SI segmental rules which are usually considered PW-bound and invoked as evidence for the PW-external status of SI clitics are better analyzed without reference to the PW. Finally, I have shown that whenever varieties other than SI display genuinely PW-bound segmental rules, these are best described if citics are analyzed as PWinternal at the postlexical level. This holds true, as shown in §3, for dialects in which, parallel to SI, enclitics do not attract stress. Thus, the study of dialect variation and in-depth comparison of related varieties has proven crucial for the present argument: it has provided us with a rich empirical testing ground and has helped us gain a better understanding of the theoretical question at issue.
STRESS STABILITY
165
REFERENCES Bafile, Laura. 1993. Fonologia prosodica e teoria metrical accento, cliticizzazione e innalzamento vocalico nel napoletano. Ph.D. dissertation, University of Florence. . 1994. "La riassegnazione postlessicale dell'accento nel napoletano". Quaderni del dipartimento di linguistica delVUniversità di Firenze 5.1-23. —. 1996. "Sulla rappresentazione delle strutture metriche ternarie". Quaderni del dipartimento di linguistica delVUniversità di Firenze 7.2-24. Basbøll, Hans. 1989. "Phonological weight and Italian raddoppiamento fonosintattico". Rivista di Linguistica 1.5-31. Berendsen, Egon. 1986. The Phonology of Cliticization. Dordrecht: Foris. Bertinetto, Pier Marco. 1999. "Boundary Strength and Linguistic Ecology (Mostly exemplified on intervocalic /s/-voicing in Italian)". Folia linguistica 33.267-286. Bolognesi, Roberto. 1998. The Phonology of Campidanian Sardinian. A Unitary Account of a Self-Organizing Structure. The Hague: HAG. Booij, Geert. 1996. "Cliticization as Prosodic Integration: The Case of Dutch". The Linguistic Review 13.219-242. Camilli, Amerindo & Piero Fiorelli. 1965. Pronuncia e grafia dell'italiano. Florence: Sansoni. Capozzoli, Raffaele. 1889. Grammatica del dialetto napoletano. Naples: Chiurazzi. Castellani, Arrigo. 1952. Nuovi testi fiorentini del Dugento. (2 vols.). Florence: Sansoni. Contini, Gianfranco. 1960. Poeti del Duecento. (2 vols.). Milan & Naples: Ricciardi Dessau, H. 1954-55. Inscriptiones latinae selectae. (3 vols.). Berlin: Weidmann. Dresher, B. Elan. 1993. "The Chronology and Status of Anglian Smoothing". Hargus & Kaisse 1993a. 325-341. Dressier, Wolfgang U. 1985. Morphonology: the dynamics of derivation. Ann Arbor, Mich.: Karoma. Flechia, Giovanni. 1898. "Atone finali determinate dalla tonica nel dialetto pi Veronese". Archivio glottologico italiano 14.111-120. Formentin, Vittorio. 1994. "Dei continuatori del latino ille in antico napoletano". Studi linguistici italiani 20.40-93,196-233. Ghini, Mirco. 1993. " -formation in Italian: a new proposal". Toronto Working Papers in Linguistics 12.41-78.
166
MICHELE LOPORCARO
Guarnerio, Pier Enea. 1886. "Il dialetto catalano d'Alghero". Archivio glotîologico italiano 9.201-364. Hargus, Sharon & Ellen M. Kaisse, eds. 1993a. Studies in Lexical Phonology, San Diego: Academic Press. Hargus, Sharon & Ellen M. Kaisse. 1993b. "Introduction". Hargus & Kaisse 1993a. 1-19. van der Hulst, Harry & Jeroen van de Weijer. 1995. "Vowel Harmony". The Handbook of Phonological Theory ed. by John A. Goldsmith, 495-534. Cambridge, Mass. & Oxford: Blackwell. Inkelas, Sharon. 1989. Prosodic Constituency in the Lexicon. Ph.D. dissertation, Stanford University. — — . 1993. "Deriving Cyclicity". Hargus & Kaisse 1993a. 75-110. Kaisse, Ellen M. 1993. "Rule Reordering and Rule Generalization in Lexical Phonology: A Reconsideration". Hargus & Kaisse 1993a. 343-363. Kenstowicz, Michael. 1996. "Base-Identity and Uniform Exponence: Alternatives to Cyclicity". Current Trends in Phonology: Models and Methods ed. by Jacques Durand & Bernard Laks, vol. 1,363-393. Manchester: ESRI. Kiparsky, Paul. 1982. "From Cyclic Phonology to Lexical Phonology". The structure of phonological representations ed. by Harry van der Hulst & N. Smith, vol. 1,131-175. Dordrecht: Foris. Kisseberth, Charles. 1994. "On Domains". Perspectives in Phonology ed. by Jennifer Cole & Charles Kisseberth, 133-166. Stanford: Center for the Study of Language and Information. Kuen, Heinrich. 1934. El dialecto de Aiguer y su posición en la historia de la lengua catalana. Barcelona: Biblioteca Balmes, Duran i Bas. Lombardi, Linda. 1996. "Postlexical rules and the status of privative features". Phonology 13.1-38. Loporcaro, Michele. 1988. Grammatica storica del dialetto di Altamura. Pisa: Giardini. . 1997a. L'origine del raddoppiamento fonosintattico. Basel & Tübingen: Francke. 1997b. "On vowel epenthesis in Aiguer Catalan". Certamen phonologicum III, Papers from the Third Cortona Phonology Meeting, April 1996 ed. by Pier Marco Bertinetto, Livio Gaeta, Georgi Jetchev & David Michaels, 205-227. Torino: Rosenberg & Sellier. —. 1997c. "L'importanza del piemontese per gli studi di tipologia linguistica". Piemonte: Mille anni di lingua, di teatro e di poesia, Atti del convegno, Vercelli, 11-12 ottobre 1997,11-29. Vercelli: Vercelliviva.
STRESS STABILITY
167
. 1998. Sintassi comparata dell'accordo participiale romanzo. Torino: Rosenberg & Sellier. . 1999. "Teoria fonologica e ricerca empirica sull'italiano e i suoi dialetti". Fonologia e morfologia dell'italiano e dei dialetti d'Italia: Atti del XXXI Congresso delia Società di Linguistica Italiana ed. by Paola Benincà, Alberto Mioni & Laura Vanelli, 117-151. Rome: Bulzoni. Liidtke, Helmut. 1979. Lucania (Profilo dei dialetti italiani, vol. 17). Pisa: Pacini. Marchetti, Giuseppe. 1977. Lineamenti di grammatica friulana. Udine: Società Filologica Friulana. cycle. Mascaró, Joan. 1978. Catalan phonology and the phonological Bloomington, Ind.: Indiana University Linguistic Club. . 1989. "On the form of segment deletion and insertion rules". Prohus 1.31-61. Monachesi, Paola. 1995. A Grammar of Italian clitics. Tilburg: ITK Dissertation Series. Muljacic, Zarko. 1972. Fonologia della lingua italiana. Bologna: II Mulino. Nespor, Marina. 1993. Fonologia. Bologna: II Mulino. Nespor, Marina & Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris. Pace, A. 1993-94. Ricerche di morfosintassi sui dialetti di Trebisacce e Castrovillari. Tesi di Laurea, University of Calabria. Peperkamp, Sharon. 1995a. "Enclitic Stress in Romance". CLS 31: Papers from the 31st Regional Meeting of the Chicago Linguistic Society. Vol. 2: The Parasession on Clitics. 234-249. Chicago: Chicago Linguistic Society. . 1995b. "Prosodic Constraints in the Derivational Morphology of Italian". Yearbook of Morphology 1994 ed. by Geert Booij & Jaap van Marie, 207244. Dordrecht: Kluwer. . 1997. Prosodie Words. The Hague: HAG. Repetti, Lori. 1995. "Constraints on Prosodie Structure. A Study of the Dialect of Coli (PC)". Studi Italiani di Linguistica Teorica e Applicata 24.279-288. Scerbo, Francesco. 1886. Sul dialetto calabro. Florence: Loescher & Seeber. Schiaffini, Alfredo. 1926. Testi fiorentini del Dugento e dei primi del Trecento. Florence: Sansoni. Sluyters, Willebrord. 1990. "Length and stress revisited: a metrical account of diphthongization, vowel lengthening, consonant gemination and word-final vowel epenthesis in modern Italian". Probus 2.65-102. Stampe, David. 1979. A Dissertation on Natural Phonology. Bloomington, Ind.: IULC. Vogel, Irene. 1991. "Level Ordering in Italian Lexical Phonology?". Certamen phonologicum II, Papers from the 1990 Cortona Phonology Meeting ed. by
168
MICHELE LOPORCARO
Pier Marco Bertinetto, Michael Kenstowicz & Michele Loporcaro, 81-101. Torino: Rosenberg & Sellier. —. 1994. "Phonological interfaces in Italian". Issues and Theory in Romance Linguistics: Selected Papers from the 23rd Linguistic Symposium on Romance Languages (LSRL XXIII), April 1-4y 1993 ed. by Michael L. Mazzola, 109-126. Washington, DC: Georgetown University Press. Vogel, Irene & Sergio Scalise. 1982. "Secondary Stress in Italian". Lingua 58.213-242. Zec, Draga. 1993. "Rule Domains in Phonological Change". Hargus & Kaisse 1993a. 365-405.
PHONOLOGICAL DISSIMILATION AND CLITIC MORPHOLOGY IN ITALO-ROMANCE
MARTIN MAIDEN Trinity College, Oxford University
0.
Introduction Most languages, and certainly all Romance languages, have undergone phonological dissimilation at some point in their history. Yet dissimilation has always been something of a Cinderella in historical phonology, for it is rarely 'lautgesetzlich' and its characteristic sporadicity and unpredictability have actually led some scholars to doubt its very existence as a phonological process. Such neglect is undeserved: I shall argue in this study that dissimilation represents a singularly important variety of sound change. Italo-Romance provides very little evidence of dissimilatory changes which have occurred specifically in order to obviate sequences of identical segments: rather, the fortuitous fact that successive consonants are not identical confers an advantage wherever variation happens to arise between forms containing successive identical segments and those that do not. In most of the examples I shall examine, the existence of rival sets of phonologically distinct but functionally non-distinct clitic pronouns has tended diachronically to lead to the grammaticalization of one of the variant clitics (to the exclusion of the other), just where the other variant would have involved a sequence of phonologically identical segments. What we observe suggests an interestingly 'symbiotic' relationship between morphological variations which have fortuitously dissimilatory effects, and a general phonological preference for dissimilated structures which can only operate where there is paradigmatic variation between 'dissimilated' and 'non-dissimilated' morphological forms, and which ultimately tends to eliminate such morphological variation in favor of 'dissimilated' forms.
170
MARTIN MAIDEN
1.
The Nature of Dissimilation Three characteristics of dissimilation seem to be generally accepted (Brugmann 1909:157, Grammont 1933:270, Posner 1961:16, Hoenigswald 1965:211, Kiparsky 1995:658, Ohala 1993:255). i.
Dissimilation is negatively motivated: its essence is that sequences of successive identical sounds be avoided. ii. Dissimilation is often lexically sporadic (this is certainly the case for all varieties of consonantal dissimilation attested in Italo-Romance). iii. Dissimilation always results in a segment which is already part of the segment inventory of the language; 'new' segments never arise under dissimilation. In fact, two distinct varieties of dissimilatory motivation are proposed in the phonological literature. We encounter a widespread notion that there is something fundamentally difficult for speakers in the production of sequences of identical sounds. Brugmann (1909:175) ascribed dissimilation to a rather nebulously defined 'horror aequi', while Palmer (1936:32) related dissimilation to the fact that "it is difficult for human muscles to execute a rapid succession of identical movements". Many modern phonologists (McCarthy 1988, Yip 1988, Cohn 1992, Clements & Hume 1995) hold that dissimilation is a stratagem for eliminating violations of the 'Obligatory Contour Principle', which prohibits adjacent identical segments (the presumed 'adjacency' of segments involved in dissimilation would occur at the level of the 'tier' on which the relevant articulator features are arrayed). A different approach views dissimilation as, in effect, an attempt to 'undo' what is erroneously perceived to be the result of assimilation (Donegan & Stampe 1976:143, Ohala 1986:392-395, 1993:249-257). For Ohala (1993) segments (among them notably laterals) characterized by the long time interval needed for the perception of their features tend to 'overlap' with nearby segments. Sequences of identical segments may be erroneously attributed to the effects of such overlap, and 'corrected', so that the supposedly 'overlapping' feature is removed from one of the segments. It is perfectly possible that both types of dissimilatory motivation (avoidance of sucesive identical segments and perceptual miscue) coexist,1 but appeal to perceptual miscue provides a particularly satisfactory account of the apparently paradoxical fact that 1 In principle, the 'miscue effect' could 'feed' avoidance of successive identical segments. The suppression of some feature (or even, in cases of haplology, of whole syllables) is a hypercorrection whose diachronic survival is favored by the fact that it dispenses with an otherwise cumbersome repetition.
PHONOLOGICAL DISSIMILATION
171
dissimilations sometimes co-occur with countervailing assimilations. (See also Posner 1961:36). What is of principal concern to this study is that in dissimilation there is a significant hiatus between motivation and realization. To say that some sequence of identical segments undergoes dissimilation does not tell us which of a multitude of possible alternatives (in fact, the entire inventory of segments and sequences of segments otherwise admissible in the grammar) should be selected. On the other hand, that the factors motivating dissimilation give no indication as to its implementation helps to explain why the resolution of dissimilation always seems to be a 'default', a sound already extant in the phonological inventory of the language. In OCP-oriented accounts of dissimilation, it is generally assumed that in the case of two tier-adjacent identical feature specifications, one of them is delinked, and the missing value is subsequently filled in by a default rule (which necessarily yields an existing segment). This provides theoretical formulation of an elegant and intuitively satisfying account of many types of dissimilation, but the fundamental fact remains, as Cohn (1992:199) notes, that "[d]issimilation [...] consists of two independent processes, delinking and default fill-in" [my emphasis]. One reason for the relative neglect of dissimilation in the history of phonological theory is probably the widespread perception that it does not behave like a 'normal' sound change, principally because of its lexical sporadicity. Perhaps the most trenchant voicing of this view is to be found in Togeby (1964),2 for whom the arbitrary and sporadic nature of many alleged dissimilations was such that virtually no phenomena generally hailed as dissimilatory actually qualified as 'phonetic laws', and any attempts at generalization were to be abandoned in favor of the dictum that 'each word has its own history'. In particular, Togeby rejected many alleged 'dissimilations' on the grounds that they were apparently not the product of sound change but involved lexical or morphological substitutions, sometimes through folk etymology, hypercorrection or onomatopoeia: e. g., Latin suffix -ILIS replacing -INUS in FEMININUS 'feminine' > Italian femminiie; MONUMENTUM 'monument' + MORI 'die' > Logudorese murimentu 'tomb'; DENTE 'tooth' + GINGIUA 'gum' > Old Neapolitan dengiva 'gum'; RETORTA 'wound' + LIGARE 'bind' > Ligurian litorta 'osier', Cosentine lituorta 'piece of cloth wound round the head'. Togeby's approach presupposes a 'Neogrammarian' perspective on sound change: all sound changes are exclusively and wholly phonetically determined, 2 See Maiden (1997) for a more detailed analysis of Togeby's approach and of his review of existing theories of dissimilation.
172
MARTIN MAIDEN
being 'blind 5 to morphological or lexical factors; any change which makes reference to such factors lies outside 'phonetic laws'. In a recent discussion of the apparent incompatibility of 'Neogrammarian'-style exceptionless sound change on the one hand, and minor and typically sporadic sound changes, on the other, Kiparsky (1995) argues that sound change, even of the Neogrammarian type, is not, in fact, 'blind'. If it were, one could not account for the fact that all languages seem to maintain ordered phonological systems, consistent with certain universal phonological principles. Kiparsky proposes to account for the apparent sensitivity of sound change to phonological systems in terms of 'variation and selection' (1995:659). [T]he neogrammarians [...] were wrong [...] in believing that sound change per se, as a mechanism of change, is structure-blind and random. The process also involves the integration of speech variants into the grammar, at which point system-conforming speech variants have a selective advantage which causes them to be preferentially adopted. In this way, the language's internal structure can channel its own evolution, giving rise to long-term tendencies of sound change. The crucial element in Kiparsky's account is the notion of the languagespecific phonological system acting as a 'filter' which tends to screen out articulatory or perceptual innovations inconsistent with it. The facts of ItaloRomance dissimilation suggest that the role played by 'filters' in sound change is rather broader. The filter may favor not only 'system-conforming' variants, but also variants which conform to the apparently universal tendency to avoid sequences of identical segments. And the type of 'variation' to which the filter is exposed can include not only low-level phonetic variation of the sort considered by Kiparsky, but also a type of morphological variation in which an identical set of morphosyntactic and semantic features comes to have (for whatever reason, e.g., dialect borrowing) more than one possible phonological realization. Posner (1961:30,204, 206) presents the view that 'dissimilatory influence' can exercise a filtering influence on sociolinguistic variation and perhaps on 'latent' sound changes geographically diffused from other dialects. Thus Posner (1961:206) states that in East France (the first [r] is regularly dissimilated in [r] + consonant + [r]).
PHONOLOGICAL DISSIMILATION
173
[A]ll pre-consonantal [r]s tend to fall — although [r] is regularly retained before a labial It is easy to see the examples of the dissimilatory change as remnants of a once regular change that conservative or other influence has caused to 'retreat'. In other cases, the examples of dissimilatory change appear to be precursors of a regular change, that, for some reason, was never fully carried out: the frequent dissimilatory change of intervocalic [1] to [r] in France and Italy, for example. There is apparently no systematic consonantal dissimilation in any ItaloRomance variety, and there is no easy way of substantiating Posner's claim that Italo-Romance dissimilatory rhotacism of intervocalic [1] is a 'precursor' of more widespread change. The scattered and sporadic examples of dissimilation gleaned ( 1980:145) indeed involve the sonorants [1], by Rohlfs ( 1966:460) and fr] and [n], and to the small number of cases of dissimilation of laterals that these authors cite we may add the extremely widespread type [skarpεllu] for **[skalpεllu] 'chisel' (AIS map 265) and some cases of dissimilation of successive palatal laterals brought about by the following mechanism:3 LILIUM > 'lily' (widespread, with many local > variants in Italo-Romance; see AIS map 639); LOLIUM > 'couch grass' (attested in most Tuscan and central and southern varieties). The pan-Italian type 'July' (AIS map 322), for expected (which should have given * * , seems to be a hypercorrect reaction to this development. But if there is a faintly detectable general Italo-Romance tendency to dissimilate successive laterals, there is one context in which such dissimilation appears to happen quite systematically, namely in sequences of clitics. 2.
The Absence of **[l- l-] in Italo-Romance Clitic Sequences In most Italo-Romance varieties, the third person singular indirect object pronoun derives from a Proto-Romance form containing *[1]: singular *[li] (< Latin dativeILLI)and plural *[lis] (< Latin dative ILLIS); some varieties have a distinct feminine indirect object form in the singular, derived from *[le]. The source of the third person direct object clitics is *[lo] (MSg), *[la] (FSg), *[li] 3
I am assuming (with Rohlfs 1966:463) that there has been prior assimilation of the first [1], yielding before . But Rohlfs fails to explain the presence of initial [d ] in and , and [j] or [0 in many southern dialects in Lazio, in Salento); now, [%], [j] and [ ] are all regular reflexes of earlier *[j], but not of (See Rohlfs 1966:212-215, 396398). The presence of [dj] (etc.) is plausibly explained if we assume dissimilation of the lateral feature in the first consonant to initial *[j], later giving [dj], [j], or [ ] (cf. IUNIU(M) >
174
MARTIN MAIDEN
(MPI), *[las] (FP1) (< ILLU(M), ILLA(M), EU, ELLAS). Concatenations of third person indirect and direct object clitics might therefore be expected to yield two syllables each beginning with [1-]. But in Italo-Romance as, indeed, in the Romance languages generally (Maiden 1997),4 the expected [1-1-] is conspicuous by its near-total absence. True, the predicted [1-1-] occurs (Bottiglioni 1933-42; map 221) in most dialects of central and southern Corsica (e.g., from ALEIC: II Mugale[dillali], S. Pietro di Tenda [dillila], La Volpaiola [dillilu], Luri [dillila], Bastia [dillila] 'say it to him'), but in Italy proper there are (to judge from the linguistic atlases) just four localities in which the two pronouns in sequence each begin with [1]. Thus AIS map 1659 diglielo 'say it to him': Siena [dillelo], Montemarciano [dillalu] (both in Tuscany), Verbicaro (Calabria) [dit eddidd] (where [dd] < [11]), and Vernole (Puglia) [dillilu]. In certain areas this state of affairs simply reflects the fact that the indirect object clitic in [1-] happens to have been supplanted by another form bearing some other initial segment. Thus in most northern Italian dialects (apart from Emilian-Romagnol and Piedmontese) *[li] has been replaced by an originally locative [ge] or [ga] (possibly from *HIC HABET; Rohlfs 1969:253). Elsewhere we find that one of the expected clitics in [1-] has been substituted by some other, phonologically distinct, form, solely or principally where a [1-1-] sequence would otherwise be expected. 3. Typology of Italo-Romance Clitic Dissimilations 3.1. Dissimilatory Conservation of Positional Variants The sequence (for e x a m p l e , r a t h e r than **[dillelo] 'say it to him') is extremely widespread in southern Italy. On the face of it, there may be an explanation for such cases of rather than **[1-1- ] which has nothing to do with dissimilation. The palatal is the regular reflex of a prevocalic * [li], the Italo-Romance singular and plural indirect object (Latin FILIAM > 'daughter'). Since certain Italo-Romance varieties originally preserved an initial vowel in the clitic reflexes of direct object ILLU(M), etc., it is at least plausible that a palatalized variant of *[li] should appear before the clitic direct object: *[li] + * [ello] > But what is possibly regular from a historical phonetic point of view is strikingly irregular from a historical morphological perspective. For in many dialects the original phonologically conditioned variants [li] and prevocalic have come to coexist in free variation, except where the result would be a sequence of clitics each beginning with [1-]. The history of Tuscan provides an especially clear illustration of this: both li and the originally prevocalic gli coexist without differentiation of function as late as the nineteenth century in the 4
The present study represents a modification of some of the positions taken in Maiden (1997).
175
PHONOLOGICAL DISSIMILATION
case of literary Italian, and to this day in many parts of Tuscany. The AIS gives both [li] and for the Tuscan dialects of Incisa (534), Fauglia (541), Montecatini (542), Arezzo (544), Castagneto Carducci (550), Scansano (581). (See also Malagoli 1937:707 for Pisa.) Sequences such as lile, lili were admitted in Old Tuscan (Lombard 1934:42, 44), but already by the fourteenth century the variant had become predominant in the first clitic, with [1] in the second, so that sequences of identical laterals were systematically avoided: glili, glieli.gliele, etc., never **gligli, * *gliegli, **lili", **lile, etc. According to Lombard (1934:56 n2) some Renaissance Italian grammarians recommended the use of gli rather than li before any following liquid5 (e.g., gli leggo rather than li leggo 'I read them'). In many central and southern dialects there is general replacement of the type by [1-], except where an [1-1-] concatenation would otherwise arise; thus, for example, the AIS maps 1659 diglielo 'say it to him', 150 gli pose (le mani sulle spalle) literally 'he put to him the hands on the shoulders' = 'he put his hands on his shoulders', 158 gli fa male literally 'it does ill to him' - 'it hurts him', show retention of the palatal clitic (or a reflex thereof) before [1-] (the enclitic or proclitic position of the clitics is irrelevant to questions regarding the consonantal structure of the clitics); (1) Tuscany and Lazio Campori(AIS 511)
dijjalo
li mese li fa mmale
Prunetta (AIS 513) Pisa (AIS 530)
1 a messo fa mmale
Fauglia (AIS 541)
li pose
Montecatini (AIS 542)
li c pposatu
li or
Castagneto Carducci (AIS 550)
li mese
li fa mmale or
li mise
li ssnte
| Pomonte (AIS 570) Marsciano (AIS 574)
dijjalo
le pcso
Scansano (AIS 581)
sent
| sεnte
|
je fa mmale li fa mmale
Pitigliano (AIS 582)
dijjalu
li mese
li fa male
Ronciglione (AIS 632)
dijjale
1 o messo
le fa male
|
In addition, in Old Tuscan the forms [li], ' and also [le] were all available as feminine indirect object clitics; later, [le] ousts the alternative forms, except before another object clitic: hence modern Italian le do un libro 'I give to her a book', but glielo do 'I give it to her'. The Old Spanish replacement of potential 5
But among * liquids' was included [m]: gli mando was apparently preferred over li mando ' I send them'.
176
MARTIN MAIDEN
10-DO sequences **le (sg.) lo and **les (pl) lo with the original prevocalic singular form ge (ge lo) offers a suggestive parallel development (Menéndez Pidal 1944:253). The fact that Tuscan glie is also the sole form of the third person dative clitic possible before the genitive-partitive ne (gliene do ' I give her some' ) might be felt to weaken any appeal to a strictly dissimilatory motivation. But it should be borne in mind that in Old Tuscan, as in modern Aretine and Florentine, the sequence gliele6 also appeared in the form gliene (Lombard 1934:35, 51, 54). 7 The origins of the final -ne in this form are unclear; Rohlfs (1968:168) suggests that this development is itself dissimilatory, the second lateral apparently being further dissimilated from the first by substitution with [ne] (gnele is also attested). In any case, the existence of a form gliene, itself a variant of gliele, probably guaranteed the survival of gliene in its modern meaning 'to her of it'. It is interesting to note that in some western Tuscan and Lazio 8 dialects, [ne], [ni] or [ni], tend to establish themselves as third person IO pronoun in all contexts. According to Rohlfs (1968:155) these originate as combinatory variants of [li] or [yi] after [n] (e.g., after non): in the modern Corsican and coastal Tuscan dialects (Bottiglioni 193342; map 1590 non prestargli) of Stazzema (ALEIC 55), Pisa (Putignano) (ALEIC 53), Marciano (ALEIC 52), Tempio Pausania (ALEIC 51), Sartene (ALEIC 45), Petreto-Bichisano (ALEIC 40), Coti-Chiavari (ALEIC 41), the sequence *[non li] is indeed realized as [un ni]. In these examples, the existence of a 'dissimilated' sequence of clitics may be viewed as exerting a negative influence on morphological change. A possible morphological change is blocked (at the cost of introducing systematic allomorphy into the pronominal paradigm) where the result would otherwise have led to a sequence of identical laterals. 3.2
Dissimilation by Replacement Other patterns of dissimilation in clitic sequences all involve replacement of one of the clitics by an etymologically unrelated form; for example, in many regions of southern Italy we find, instead of expected **[dilelo] 'say it to him', 6
This gliele (and gliene) form could stand for any possible combination of number and gender in the two pronouns ('them to her', 'him to them', 'them to them', etc.). 7 A further old Tuscan form is gnene, a type preserved in certain modern Tuscan dialects such as Incisa (AIS 534), Florence (AIS 523), Stia (AIS 526), Caprese Michelangelo (AIS 535), Arezzo (AIS 544), Cortona (AIS 554). Here initial [n] seems to be due to an assimilation of nasality (cf. also Ronjat 1937:63 for an analogous development in the Occitan dialect of Vinzelles), which nonetheless maintains the difference in palatality. 8 Arsoli and Canterano have [pi]. See Pulcini (1972:22); Merlo (1930:76).
177
PHONOLOGICAL DISSIMILATION
the type [dit elo], with the basically locative clitic [tfe] functioning as indirect object. It is helpful to view such dissimilations in the light of wider diachronic tendencies in Italo-Romance generally. In addition to those clitics in [1-] derived from accusative and dative case forms of Latin ILLE, and excluding the first and second person singular forms, there have emerged in Italo-Romance a range of other personal and locative clitics derived as follows. (2) 3rd reflexive 1pl. 2pl. Locative
si O.Tuscan no
< <
acc. refl. SE(or dat. SIBI) lpl. acc. NOS
O.Tuscan vo
< <
2pl. acc. UOS HINCE 9 'hence'
Locative
ci vi
Locative, partitive / genitive
ne
<
IBI 'thither'
<
INDE 'thence'
The synchronie and diachronic relationship between form and meaning in the Italo-Romance system is, however, very far from biunique; syncretism and suppletion abound. Some of the asymmetries of form and meaning observable in the clitic systems of central and southern Italy are represented schematically in the following table, where the rows show the various grammatical and semantic functions of the clitics, and the columns the Latin etyma of the modern forms. (3)
IILU
ILLA
ILLI
ILLÆ
ILLI
ILLAS
ILLIS
3sgDO 3sgIO 3plDO 3plIO 3rdRef lpl. 2pl. Loc.
9 It is likely that Italian ci and the [t i ] , [t e], etc. of many central and southern dialects derive from HINCE, although *ECCEHIC (lit. To, here') is also a possible etymon. At any rate, all other forms derived from ECCE(or *ECCU) are stressed demonstratives, e.g., questo (< *[ekku este]), etc.
178
MARTIN MAIDEN
SE
NOS
UOS
HINCE
IBI
INDE
3sgDO 3sgIO 3plDO 3plIO 3rdRef lpl. 2pl. Loc.
The boxes in the above tables identify the widespread tendency for original locatives to supplant, or coexist with, third person (and sometimes first and second person plural) clitics, and an apparently reciprocal tendency for the third person IO form sometimes to assume locative functions and thence, like the locatives, to percolate (albeit rarely) into first and second person plural object forms. 10 Reflexes of HINCE and IBI have also encroached on each other, there remaining no semantic distinction of proximity between them in dialects (such as Tuscan) where both survive. More detailed exemplification of the developments diagrammed above is available in Maiden ( 1997). In brief, reflexes of HINCE (Rohlfs 1968:155) assume third person IO function in parts of Tuscany, and extensively in Lazio (e.g., Castro dei Volsci (Vignoli (1925:165)) [t ә 'dornә nә 'I give him a book'), Campania; Calabria and Sicily (see also AIS, maps 65 and 150 for Trasacco (646), S. Francesca (664), Sonnino (682), Scanno (656), Lucera (707), Serrastretta (771), and all Sicily). In much of the Italo-Romance domain (see Rohlfs 1968:159, 160, and AIS maps 1111, 1654), a reflex of HINCE replaces the object forms of the Ipl. pronoun: e.g., Italian ci alziamo, northern Lazio [t i le'vamo]; Bari (AIS 719) [ndj al'tsame] lit. 'we raise us' = 'we get up9. In parts of Lazio (e.g., Paliano, Sassa (AIS 625)) we find [t e] as a second person plural indirect object clitic. In Abruzzo and parts of northern Puglia HINCE ([t i], [tfe]) replaces the third person reflexive (and impersonal) clitic; [t i] occurs in the same function in parts of Tuscany (e.g., Isola del Giglio: [t i parte] - Italian si parte 'one leaves'). In Salento, Lucania and Calabria we sometimes encounter 10 Thus in Assisan one hears [je 'mando la 'karta] ' I send the paper there', and popular Tuscan non gli è carne for nonc'e came 'there's no meat'. In Lazio at AIS Montefiascone (612), Rieti (624), Palombara (643), clitics from ILLI function as 2pl. datives: e.g., Rieti 'enne]; Montefiascone [jele Venne] T sell (subjunctive) them to you'. At Rome (AIS 652) a continuant of ILLI serves as a lpl. clitic [je 1 a 'ddato] 'he gave it to us' (map 1111 of the AIS); cf. also, in Abruzzo, Crecchio (AIS 639) and Fara S. Martino (AIS 648).
PHONOLOGICAL DISSIMILATION
179
continuants of INDE encroaching on the IO forms; e.g., Oriolo [nn 'dowәdә] 'it hurts (to) him', Nardò [di 'skrive] 'he writes to him' (see Rohlfs 1968:156, Lausberg 1939:142). INDE, perhaps because it shares the nasal consonant generally characteristic of lpl. forms, assumes lpl. value in parts of Basilicata, Salento, and Calabria: e.g., Galiipoii [nde Vite] 'he sees us' (see Rohlfs (1968:158)). Instances of cliticized IBI assuming indirect object functions, commonly attested in Gallo-Romance (e.g., popular French J'y ai dit ça for Je lui ai dit ça 'I told him that'; also Alibert (1976:64) for Occitan and Remacle (1952:252) for Wallon), are rare in Italo-Romance proper, although the old Logudorese Condaghe di San Pietro di Silki, ui < IBI tends to replace the third person dative pronoun; further examples from modern Sardinian will be discussed later. In addition to a widely observed tendency for the clitics from SE (or from the dative SEBI) to become universal reflexive forms in all persons and numbers (Wunderli 1989), SE / SIBI extends its domain to non-reflexive forms (Cennamo 1991; 1993): e. g., southern Sardinian fsi mus'siaða] 'he bites us', fsi yum'prendu] 'I understand you (pl.)' (Blasco Ferrer 1984a:256). The reflexive clitic also does duty as a lpl. object clitic in eastern Tuscany and many northern Italian dialects (Rohlfs 1968:159, Blasco Ferrer 1984b:21, Cennamo 1991:17, 1993). The reflexive is probably also the source of third person non-reflexive IO [si] found in a small enclave around Benestare in southern Calabria. What is remarkable is that in dialects where such syncretistic developments among clitics are a mere tendency, or are not systematically attested, they none the less occur obligatorily in sequences of clitics where [1-1-] would otherwise arise. For example, in Calvello (Basilicata) (Gioscio 1985:64), the IO clitic may be either [ndjә] or [lu], [la], [lә], (identical to the accusative clitics); but [ndjә] is obligatory before a third person direct object in [l-]: [lu 'rannә nu ka'vaddә] or [ndjә 'rannә nu kaVaddә] 'they give him a horse', but only [ndjә lu 'rannә] 'they give it to him', never **[lu lu 'rannә]. From the Italo-Romance texts gathered by Battisti (1921) for Arpino (Lazio), Pescosolido, Vallerotonda (Campania) and Andria (Puglia), it is clear that a reflex of HINCE appears as IO clitic precisely where it is followed by a clitic in [1-], whilst a reflex of n u is used outside this context. Thus Pescosolido oissә la 'mamma]11 'mother said to him', 'òisә 'kella 'fada] 'that fairy said to her', ['kwannә 'jettә la 'mamma, tfә lә pi'di] 'when mother went, she asked her for it'; Andria [le 'dess9] 'he said to him', [tjә 'a 'dœtt9] 'he gave it to him'. The AIS maps 150 (gli pose (le mani)), 158 (gli fa male), 1659 (diglielo)
11
The palatalization of p] apparent in these examples is a secondary effect conditioned by original following [i] or [u].
180
MARTIN MAIDEN
suggest that [la] or [li] must give way to a reflex of HINCE ([t i], [t ә], [nt e], [ndje], etc.), when followed by a [1-] clitic. (4)
Porto Santo Stefano (AIS 590) Morrone del Sannio (AIS 668) Lucera(AIS 707) Sangiovanni Rotondo (AIS 708) Vico del Gargano (AIS 709) Gallo (AIS 712) Colle Sannita (AIS 714) Ascoli Satriano (AIS 716) Canosa di Puglia (AIS 717) Ottaviano (AIS 722) Ripacandida(AIS 726) Carovigno (AIS 729) Matera (AIS 736) Palagiano (AIS 737)
li mese i mәtεttә i mәttijә li messә rә mәttεttә rә misә lә mәtti lә metti i mәttivә li mittivu la la mәtti
li fa mmale i dolә i fa malә li fa mala li fa mmalә ro fa mmalә li fa malә li fet ә i fa mmalә 1 a puostә
ditt alo dit ala dit illә dit alә dcjtt әlu dit ala ditt ello dәndjillә dindjlla rint ello dәndjillә dint ulu dәndjyllә dint i l l ә |
At Benestare (AIS 794) we find [dintfillu] but [si misi], [si doli]. Presumably in this case [si] (< SIBI) replaces an earlier form in [1-], as is found in most surrounding dialects. Lausberg (1939:142) indicates that in Nova Siri (and elsewhere in southern Basilicata), the reflex of INDE, otherwise an optional form of the IO, is obligatory in combination with an accusative clitic of the third person ([ji] is a reflex of earlier [li]): [je ji nu libbr] or [je nnә nu libbr] 'I give him a book', but [dan'nillә] 'give it to him', 'let's say it to him'. Continuants of INDE ([ne], [n9], [nde], [ndi], etc.) replace those of n u ([li], [lә], etc.) before DO forms in [1-] in a number of southern Italian localities: Camaiore (AIS 520), ['dinnelo], [li 'posa], Acquafredda (AIS 742), [dit in'nilu], [li mit'tivu]. The data from Bottiglioni (1933-42; map 221) suggest an analogous situation in northern Corsica at Rogliano, and on the Tuscan coast at Stazzema and Marciano. A reflex of IBI replaces IO ([li]/[lis]) before clitics in [1-] in Logudorese and northern Sardinian varieties (AIS, Bottiglioni 1933:42, Blasco Ferrer 1984a:255).
181
PHONOLOGICAL DISSIMILATION
(5)
Ploaghe (AIS 923) Sassari (AIS 922/ALEIC 50) Tempio (AIS 916/ALEIC 51)
li pon'dzeidi 'he put on him' le b'bongu 'I put on him' 1 a d'datu 'he gave to him'
'narahilu 'say it to him' 'diviru a iddu 12 'say it to him' 'divvila 'say it to him'
|
At Tagliacozzo in central Italy (AIS 645) it is again a reflex of IBI which provides the IO pronoun where [1-] follows: (['ibbilo] 'say it to him'). Finally, some Sardinian varieties replace a [1-] clitic with a reflex of SIBI when another [1-] follows (Blasco Ferrer 1984a:256,1984b:28): Dorgali (AIS 949) [nara'silu] 'say it to him', [1 a 'postu] literally, 'he put to him', and Desulo (AIS 957) [nara'siddu], [dd a 'postu]. There is an interesting parallel development in Castilian se lo doy ' I give it to him' for *le lo doy (Schmidely 1979). In the above cases, the dissimilation has occurred, once again, simply because there happened to become available an alternative sequence of clitics not involving iteration of [1]; correspondingly, the phonological system appears to have exercised selective pressure in favor of the 'dissimilated' sequences, leading to the grammaticalization of those sequences. 3.3
Other Types of Clitic Dissimilation and the Domain of Clitic Dissimilation In addition to the glielo type, standard Italian evinces a number of others in which sequences of clitics containing identical segments are dispreferred where a 'dissimilated' alternative is available. Potential sequences of the clitic si (reflexive and 'impersonal' clitic) are debarred, ci substituting the first clitic: not **si si lava but ci si lava 'one washes oneself', not **si se lo dà but ci se lo dà 'one gives it to oneself'. Locative ci and vi may not be combined with their personal homophones ci and vi: not **ci ci porta nor **vi vi porta but vi ci porta 'he takes you/us there'. There is also a preference for vi before host words in initial [tf], and for ci before hosts with initial [v] (Wanner 1977:121): thus vi cercano diamanti 'they look for diamonds there' is preferable to ci cercano diamanti', and ci vedono diamanti 'they see diamonds there' is preferable to vi vedono diamanti. A similar dissimilatory preference is claimed to hold for the synonymous and functionally identical prepositions fra and tra 'among, between': fra treni 'among trains' rather than tra treni, but tra fratelli 'among brothers' rather than fra fratelli. It may seem at first puzzling that clitic dissimilation should occur in, at 12
In this case [ru] < [lu]. The stressed form is here copresent with the clitic.
182
MARTIN MAIDEN
best, a desultory way across the boundary between clitic and host, yet systematically across the boundary between concatenated clitics. But in fact Romance dissimilation normally occurs only word-internally, and there is distributional and diachronic evidence to suggest that the boundary between adjacent clitics is akin to that between morphemes within words. The range of lexical host verbs with which a clitic can be combined is an open set, whereas sequences of concatenated clitics form a closed set whose members — for example the sequence combining third person direct and indirect object clitics — may attain very high token frequency. Such frequent cooccurrence may confer on these chains the status of independent lexical entries: the fact that at certain stages in the history of Italian the third person sequence gliele (preferred over its undissimilated alternatives lile and lele) was the phonological realization of any possible permutation of number (singular or plural), gender (masculine or feminine) and case (direct + indirect object or vice versa) indicates that the chain came to be analysed as a structurally opaque 'word'. 13 3.4
Suppression of a Consonant Finally, I have encountered one type of clitic dissimilation in which the source of variation is apparently phonetic variation rather than the coexistence of morphological alternatives (and which is rather closer than most of my examples to the type of selection of phonetic variants described by Kiparsky). This occurs in the Sardinian dialect of Nuoro (Pittau (1972:83)), where the second [1] is regularly suppressed in sequences of unstressed IO [li] + DO Msg. [lu], Fsg. 0a], Mpl. [los], Fpl. [las]. Thus we have [li u], [li a], [li os], [li as], not **[li lu], **[li la], **[li los], **[li las]. These deletions of [1] are, on the available evidence, limited solely to clitic sequences containing two [l]s (otherwise, we have [mi lu], [ti lu], etc.). Pittau writes (1972:22-29) that there is a tendency to lenite many intervocalic consonants in Nuorese and we may surmise that lenition sporadically resulted in deletion, an outcome which appears ultimately to have been retained only in the clitic system, where the result is dissimilatory. It is also worthy of note that in sequences of [nos] (lpl. clitic) or [bos] (2pl. clitic) + DO, the expected result, allowing for regular assimilation of the preceding [s], should be **[nol lu] etc., and **[bol lu] etc. ; but what actually occurs is [no lu], [bo lu] etc.
13 It is noteworthy in this connection that in some southern Italian dialects vowel harmony rules operating among clitics are blocked by the boundary between a clitic and a following host. (See Maiden 1995:119.)
PHONOLOGICAL DISSIMILATION
183
4.
Other 'Morphological' Dissimilations in Italo-Romance Morphologically specified dissimilation is not limited to clitics. A common noun suffix in Italo-Romance derives from the originally diminutive ending * [olu] (M.), *[ola] (F.). Reflexes of Latin GRANDINE(M) 'hail' (AIS 373) and of *['lendine] 'nit' (AIS 476) display a reflex of *[ola], with dissimilatory effect, in a number of dialects where intervocalic [n], in *[ine] and elsewhere, is otherwise retained: 520 (Camaiore) ['grajiola], 559 (S. Elpidio a Mare) ['grannola], 569 (Grottammare) ['jenala], 590 (Porto S. Stefano) ['lendili], 612 (Montefiascone) ['nennale], 625 (Sassa) ['ranola], 648 (Fara S. Martino) [linәlә], 656 (Scanno) [linnәlә], 668 (Morrone del Sannio) ['grannәlә], ['linnәlә], 682 (Sonnino) ['rannola], [ninnulә], 707 (Lucera) ['granәtә], ['linn9te]; the type ['grandile] found in most of Sardinia may reflect the influence of another suffix, *[ile]. Perhaps the development ARBORE(M) > ['arbulu] 'tree', sporadically found in southern Italy and much of Sicily (AIS map 533) has a similar origin (cf. also the sporadic ROBORE(M) > ['roguia], ['rovulu] 'oak tree' in northern Italy and Sicily, AIS map 598). The Fpl. suffix *['ane] may have provided the basis for the extremely widespread, and dissimilatory, southern Italian type [pa'tane], revealed by AIS map 1387 for le patate', 'potatoes'. 14 There are sporadic examples in Tuscan of dissimilation of [r- r-] to [r- d-]: RARUM > rado 'scarce', PRORAM > proda 'prow', ARMARIUM > armadio 'cupboard', CONTRARIUM > old Tuscan contradio, PORPHYRUM > old Tuscan porfido 'porphyry'). In certain infinitives we find that [r- r-] has also undergone dissimilation, except that in this case [d] (in the verb root) precedes [r] (in the infinitive ending): QUÆRERE > chiedere (alongside old Tuscan cherere) 'ask', INTERERE > intridere 'knead', FERIRE > old Tuscan fedire / ferire 'wound', CONQUIRERE> old Tuscan conquidere 'conquer'. Substitution of [r] by [d] is not restricted to the infinitive, since it is observed throughout the paradigm of the relevant verbs (cf. the third person singular present forms chiede, intride, fiede, conquide). The history of this development in the verb would need a more detailed analysis than is possible here, but it seems likely that the presence of [d] is partly explicable by a proportional analogy: the verbs in question (except ferire) form their preterite in [s] (chiesi, intrisi, etc.), a characteristic which they share with a large number of verbs otherwise characterized by stem-final [d] (divisi/dividere 'divide',persi/perdere 'lose', rasilradere 'shave', etc.), so that on the model of divisildividere, etc., there developed chiesi/chiedere, etc. On the other hand, these are the only Italian verbs thus analogically affected, a fact which suggests either
14
For an account of a dissimilatory deployment of suffixes in French, see also Malkiel (1966:327).
184
MARTIN MAIDEN
that the analogical changes took root in these verbs precisely because they gave rise to a phonologically desirable dissimilated infinitive, or, alternatively, that the effects of a phonological dissimilation survived in these verbs precisely because they fitted an existing pattern of alternation.15 5.
The Conditioning of Clitic Dissimilation I have shown that certain substitutions in the clitic pronoun system and elsewhere occur principally or solely in such a way that their effect is one of phonological dissimilation. The linguistic atlases and other dialectological studies indicate that such systematic substitutions are often in the vanguard of 'latent' changes in the clitic system, where by 'latent' I mean that the substitution also occurs optionally in other contexts, or at any rate is likely to be known to speakers of the relevant dialects because it systematically occurs in neighbouring ones. In short, the obligatory dissimilatory clitic sequences seem to make use of clitic forms which are in variation with an alternative clitic. But are the dissimilatory effects of these substitutions really any more than an accident, from a phonological point of view? Blasco Ferrer (1984b:28) certainly has no doubt that clitic dissimilation in Sardinian is due to a tendency to avoid two successive liquids, and indeed sequences of liquids are recurrent loci of dissimilation in Romance and well beyond (Grammont 1933:276, 284, 291, 307). It might be suggested, however, that we are dealing with an exclusively morphosyntactic phenomenon, which has emerged as follows. First, there is a tendency for one of the clitics to be absent, motivated perhaps by the fact that the reference of that clitic is very easily recoverable from context and from the argument structure of verbs such as 'give' and 'say'. Second, the resulting 'gap' might sometimes be filled by using any available alternative clitic. According to the AIS 1 6 there is deletion of the IO clitic at Chiavaretto (AIS 545) ([dillo] 'say it to him'), Ancona (AIS 539) ([dilo]), Zuoz (AIS 28) ([dil]), Rex Platta (AIS 47) ([dil]). At Ascrea (Fanti (1939:133)), the DO clitic [lu] is omitted in such cases: [li itso], but in reality there are very few examples of deletion in the Italo-Romance corpus which I have consulted. Arguably, this is a consequence of using a corpus which relies heavily on questionnaires couched in 15 Malkiel (1973:214-225) hypothesizes what is, to my mind, a rather less plausible scenario in which first verbs like chiedere are analogically remodelled on the basis of their preterites and then, subsequently, the resultant dissimilated infinitives furnish a model for dissimilation of [r- r ] sequences elsewhere in the lexicon. 16 Comparison of AIS maps 150 and 1627 suggests that some Sardinian varieties (Tempio (916), Nuoro (937), Macomer (943), Milis (941), and Picerno (732) in Basilicata), tend to avoid DO-IO sequences, expressing the IO with a stressed pronoun: Tempio [dillu a 'iddu].
PHONOLOGICAL DISSIMILATION
185
the standard language, and might have induced informants simply to translate into their dialect the syntactic structure of the original question. Precisely because one of the clitics is normally absent, the informant, obliged to produce a clitic, might cast around for any adequate substitute, whence the introduction of novel forms. Now it is quite possible that the inherent recoverability of third person clitics plays some role in their suppression, but this does not adequately account for the dialect facts, for it predicts that suppression should occur regardless of the phonological context. In reality, the dissimilatory changes occur predominantly in the environment of another clitic in [1]. The very artificiality of the questionnaire, paradoxically, guarantees the authenticity of the response: if informants normally suppress a clitic, but find themselves forced to furnish one in imitation of standard Italian, we should expect that the restored clitic should be that closest to Italian gli(e), not a reflex of HINGE, INDE, etc. But the hypothesis of a phonological dissimilation of successive laterals is also supported by the instances, in standard Italian and in many dialects, of differential conservation of a clitic with dissimilatory effect (e.g., glielo do), despite a general tendency to eliminate pronominal allomorphy. Dissimilations of the kind surveyed in this study do not appear to be 'phonologically driven' in the sense that they might be the product of phonological rules transmuting an underlying consonant into a 'dissimulated' variant. There are in the Italo-Romance dialects no dissimilatory rules having [1-1-] (or [n-n-] or [r-r-]) as their input: all of these sequences are, after all, perfectly admissible and any notion that, for example, an 'output' [ne] could be 'derived' by some dissimilatory rule from an 'underlying' form [lo] would require postulation of an implausibly powerful rule arbitrarily limited to the environment of another following clitic. Indeed most of the clitic dissimilations we have seen, like the suffixal dissimilations involving reflexes of *[olu], and the apparent dissimilations involving morphological 'blends' (e.g., DENTE 'tooth' + GINGIUA 'gum' > Old Neapolitan dengiva 'gum'), involve not modifications of segments but substitutions of whole morphemes. All these facts suggest that the phonological mechanism involved in the dissimilation of Italo-Romance clitics is a kind of filter, which has as its basis not a language-specific dissimilatory rule acting on 'undissimilated' underlying forms, but a universal phonological preference exercising selective pressure on a synonymous pair of surface alternatives. The coexistence of such alternatives need be no more than the fortuitous result of independent lexical or morphological changes, one of which offers a phonologically preferable alternative to its undissimilated counterpart. This scenario implies a synchronic stage between the introduction of the alternative clitic form and its eventual grammaticalization in combination with
186
MARTIN MAIDEN
another clitic in which the phonological filter acts to reject the 'undissimilated' clitic sequence in favor of the 'dissimilated' one; but such 'rejection' can only occur if the phonological filter has access to paradigmatic information. It must 'know', for example, that the combination of semantic and syntactic features for the third person singular direct object pronoun and third person indirect object pronoun may be phonologically realized either as [li lo] or [ne lo] ; that the root of the Old Italian verb chiedere might be either [kjsr] or [kjεd]; that the Neapolitan word for 'gum' might be either gengiva or dengiva, and so forth. Given such information, the filter will give preference to the phonologically 'optimal' realization; without such information, nothing happens. The hypothesis sketched here, that certain kinds of phonological change presuppose the existence of synonymous variants, carries with it the implication that once a 'dissimilated' clitic sequence becomes the norm it can no longer be viewed as being phonologically determined: for this reason there is no necessary contradiction between the position I adopt and that taken by Bonet ( 1991) who, in her analysis of a partially similar set of clitic phenomena in the speech of Barcelona, emphatically denies (1991:154 nl7) the possibility of a synchronic phonological motivation, and offers a cogent account of the synchronic data in terms of feature delinking and default realizations (see further my remarks in Maiden 1997). My focus, rather, is diachronic, but entails the following claim which can be formulated, in a very strong form, as follows: in a system where a given set of syntactic and semantic features (e.g., those associated with third person, number, and direct or indirect object) has more than one possible phonological realization then, all other things being equal, speakers will tend to prefer that realization which is phonologically 'better'. What counts as 'better' may be determined in some cases by the language specific phonological system (the kind of filtering Kiparsky invokes for the output of sound change), but it may also be measured in terms of more widespread cross-linguistic tendencies such as the avoidance of successive identical consonantal segments (dissimilation) or, for example, a preference for consonant-initial syllable structures of the kind discussed in Dixon (1982). A richer development of this hypothesis demands a more extensive diachronic investigation of the relevant phenomena than has ever been undertaken to date. Within Romance, for example, a major task ahead will involve attempting to answer Togeby's objection that dissimilated forms of the dengiva type are really a matter of folk etymology, whose effect is only accidentally dissimilatory. If my outline hypothesis is correct, and given a
PHONOLOGICAL DISSIMILATION
187
sufficiently large comparative and historical corpus of popular creations, 17 it may become apparent that such by-forms are favored diachronically precisely when they happen to conform to some general phonological preference such as dissimilation, and to be rejected when they do not. REFERENCES AIS = Jaberg & Jud 1928-1940 ALEIC = Bottiglioni 1933-1942 Alibert, Lois. 1976. Gramàtica occitana. Montpellier: Centre d'Estudis Occitans. Battisti, Carlo. 1921. Testi dialettali in trascrizione fonetica. Parte IIa: Italia centrale e meridionale. (Zeitschrift für romanische Philologie. Beiheft 56). Halle: Niemeyer. Blasco Ferrer, Eduardo. 1984a. Storia linguistica delta Sardegna. Tübingen: Niemeyer. . 1984b. "L'estensione del pronome riflessivo SE in sardo e nelle lingue romanze". Revue de linguistique romane 43.12-35. Bonet, Eulàlia. 1991. Morphology after Syntax : Pronominal Clitics in Romance. Cambridge, Mass.: MIT Working Papers in Linguistics. Bottiglioni, Gino. 1933-1942. Atlante linguistico etnografico italiano delta Corsica. Pisa: L'Italia dialettale. Brugmann, Karl. 1909. "Das Wesen der lautlichen Dissimilationen". Abhandlungen der K. S. Gesellschaft der Wissenschaften. phil.-hist. Klasse 27.141-178. Cennamo, Michela. 1991. "Se, sibi, suus nelle Inscriptiones Latinae Christianae Veteres ed i successivi sviluppi romanzi". Medioevo romanzo 16.3-20. , 1993. "L'estensione del dominio referenziale del riflessivo in testi italiani antichi". Archivio glottologico italiano 78.53-62. Clements, George & Elizabeth Hume. 1995. "Internal organization of speech sounds". The Handbook of Phonological Theory ed. by John A. Goldsmith, 244-306. Cambridge, Mass. & Oxford: Blackwell. Cohn, Abigail. 1992. "The consequences of dissimilation in Sundanese". Phonology 9.199-220.
17
It is, of course, often observed that dissimilation occurs in the more recondite and unfamiliar sectors of the lexicon, such as technical and botanical names, foreign words and certain proper names (Togeby 1964:652, Posner 1961:84), where popular creation of more 'transparent' forms might have been particularly rife.
188
MARTIN MAIDEN
Dixon, Robert. 1982. "Olgolo syllable structure and what they are doing about it". Where Have All the Adjectives Gone? ed. by Robert Dixon, 207-210. Berlin, New York, Amsterdam: Mouton. Donegan, Patricia & David Stampe. 1979. "The study of natural phonology". Current Approaches to Phonological Theory ed. by Daniel Dinnsen, 126173. Bloomington, Ind.: Indiana University Press. Fanti, Renata. 1939. "Note fonetiche, morfologiche e lessicali sul dialetto di Ascrea". Vltalia dialettale 15.101-135. Gioscio, Joseph. 1985. Il dialetto lucano di Calvello. Stuttgart: Steiner. Grammont, Maurice. 1933. Traité de phonétique. Paris: Delagrave. Hoenigswald, Henry. 1964. "Graduality, sporadicity, and the minor sound change processes". Phonetica 11.202-215. Jaberg, Karl & Jacob Jud. 1928-1940. Sprach- und Sachatlas Italiens und der Südschweiz. Zofingen: Ringier. Kiparsky, Paul. 1995. "The phonological basis of sound change". The Handbook of Phonological Theory ed. by John A. Goldsmith, 640-670. Cambridge, Mass. & Oxford: Blackwell. Lausberg, Heinrich. 1939. Die Mundarten Südlukaniens. Halle: Niemeyer. Lombard, Alf. 1934. "Le groupement des pronoms personnels atones en italien". Studier i modem Språkvetenskap 12.19-76. Maiden, Martin. 1995. "Evidence from the Italian dialects for the internal structure of prosodic domains". Linguistic Theory and the Romance Languages ed. by John Charles Smith & Martin Maiden, 115-131. Amsterdam & Philadelphia: John Benjamins. . 1997. "La dissimilation à la lumière des pronoms clitiques en roman". Zeitschrift fur romanische Philologie 113.531-562. Malagoli, Giuseppe. 1937. "La parlata pisana nell'AIS". Vltalia dialettale 13.5373. Malkiel, Yakov. 1973. "Etiological studies in Romance diachronic phonology". Acta Linguistica Hafniensia 14.201-242. . 1966. "Genetic analysis of word formation". Theoretical Foundations: Current Trends in Linguistics vol. 3 ed. by Thomas Sebeok, 305-364. Paris & The Hague: Mouton. Menéndez Pidal, Ramón. 1944. Manual de gramática historica espanola. Madrid: Espasa-Calpe. Merlo, Clemente. 1930. La Novella I,9 del 'Decameron' tradotta nei parlari del Lazio: I, Valle dell'Aniene. Roma: Società Filologica Romana.
PHONOLOGICAL DISSIMILATION
189
Ohala, John. 1986. "Phonological evidence for top-down processing in speech perception". Invariance and Variability in Speech Processes ed. by Joseph Perkell & Dennis Klatt, 386-401. New Jersey: Earlbaum. . 1993. "The phonetics of sound change". Historical Linguistics. Problems and Perspectives ed. by Charles Jones, 237-278. London: Longman. Pittau, Massimo. 1972. Grammatica del sardo-nuorese. Bologna: Pàtron. Posner, Rebecca. 1961. Consonantal Dissimilation in the Romance Languages (Publications of the Philological Society, vol. XIX). Oxford: Blackwell. Pulcini, Walter. 1972. Il dialetto di Arsoli. Tivoli: Tipografia Tivoli. Rohlfs, Gerhard. 1966. Grammatica storica delta lingua italiana e dei suoi dialetti. I Fonetica. Turin: Einaudi. ——-. 1968. Grammatica storica delta lingua italiana e dei suoi dialetti. II Morfologia. Turin: Einaudi. . 1969. Grammatica storica delta lingua italiana e dei suoi dialetti. III Sintassi e formazione delle parole. Turin: Einaudi. Ronjat, Jules. 1937. Grammaire istorique des parlers provençaux modernes, Vol. III. Montpellier: Société des langues romanes. Schmidely, Jack. 1979. "De ge lo à se lo". Cahiers de linguistique hispanique médiévale 4.63-70. Tekavcic, Pavao. 1980. Grammatica storica delVitaliano. Bologna: Mulino. Togeby, Knud. 1964. "Qu'est-ce que la dissimilation?". Romance Philology 17.642-667. Vignoli, Carlo. 1925. Il vernacolo di Veroli. Rome: Società Filologica Romana. Wanner, Dieter. 1977. "On the order of clitics in Italian". Lingua 43.101-128. Wunderli, Peter. 1989. "L'explication de «se» omnipersonnel dans les langues romanes". Revue de linguistique romane 43.25-34. Yip, Moira. 1988. 'The OCP and phonological rules: a loss of identity". Linguistic Inquiry 19.65-100.
OXYTONE INFINITIVES IN THE DIALECT OF PISA*
GIOVANNA MAROTTA University of Pisa 1.
Oxytone Forms of the Infinitive In the dialects of Tuscany,1 there are many phonological phenomena involving word final position, and special note must be made of the oxytone infinitives of verbs, i. e., verbs ending in a stressed stem vowel, without the unstressed final syllable /-re/ of the standard form. During informal conversation, it is not unusual to hear Tuscan speakers saying, for instance, cantá, vedé, senti, instead of cantore 'to sing', vedére 'to see', sentire 'to hear'.2 The interest of such forms is twofold: on one hand, they interact with Raddoppiamento
Ï would like to thank Michael Kenstowicz, Laura Bafile, and an anonymous reviewer for their useful comments. 1 Tuscan dialects are traditionally called vernacoli, in order to keep Tuscany distinct from other regional areas of Italy. As is well known, the Tuscan varieties are more similar to so-called standard Italian than those spoken elsewhere. While there are historical reasons for such closeness, recent developments have resulted in a shifting of the standard towards Northern models, while Tuscan varieties, including Florentine, are becoming less and less prestigious in the sociolinguistic consciousness of speakers. On the other hand, the similarity between the standard language and the dialects spoken in Tuscany makes it sometimes difficult to single out the structural peculiarities of the vernacolo from the characteristics belonging to the national language. From this point of view, the study of other regional areas appears to be easier, since the greater the distance between dialect and Italian, the simpler the distinction between the two varieties. As frequently happens, similarity may then become a handicap instead of an advantage. On general problems connected with the analysis of Tuscan speech, see Agostiniani & Giannelli (1990). 2 As is well known, in Italian there are four verb conjugations which end in -áre, -ére, -ere, -íre. The truncation process does not lead to oxytony in the case of the third conjugation, where stress falls on the root vowel and not on the stem vowel: prèndere/prènde; conóscere/conósce, etc. See §5 for a discussion of the reflexes of the different metrical patterns on phrasal prosody.
192
GIOVANNA MAROTTA
Fonosintattico (henceforth, RF),3 on the other, they give rise to a degenerate foot, i. e., to a rhythmical pattern which is not consistent with the normal metrical structure of the language where feet are assumed to be binary, bounded, and leftheaded. Since in Italian, especially in poetic Italian, infinitives show truncated allomorphs which end with the liquid sonorant after loss of the final vowel (cantâr, vedér, sentír), the oxytone forms may also be considered truncated, if truncation is defined as the loss not only of the final vowel after a sonorant (amâre > amár 'to love', like mare > mar 'sea'), but of the entire final syllable (amâre > amár > amâ).4 In the following pages, the terms 'oxytone' and 'truncated' will be used synonymously. Indeed, we will aim to show that Pisan forms like cantâ, vedé are lexicalized for speakers and, therefore, at least synchronically, they are no longer related to the truncation process active in nominal and verbal inflection. Although the truncated forms of infinitives are attested over all Tuscany,5 some peculiarities can be noted with regard to their frequency and modalities of use. This is the reason why we decided to concentrate our attention on the variety spoken in Pisa, where oxytone infinitives not only are widely used by the majority of speakers, but also systematically trigger RF, showing interesting relations with other phonological processes at work in this dialect.6 3 Raddoppiamento Fonosintattico is a phonological process of Italian and of many Italian dialects of the Center-South occurring at word boundaries and producing the lengthening of the initial consonant of the second word after a final stressed vowel and after some morphemes. In recent years the topic has been dealt with by many scholars; however, for the basic aspects, we refer the reader to Agostiniani (1992), Chierchia (1983-86), Fanciullo (1983-86, 1997), Loporcaro (1988,1997). 4 Troncamento is normally defined as the deletion of a word final unstressed vowel after a sonorant: sono andato > son andato '(I) have gone', signore > signor 'sir'. Those truncations showing an optional character have to be considered distinct from those which are obligatory: nessun bicchieref*nessuno bicchiere 'no glass', andiam via/andiamo via '(we) go away'. Although the process of truncation has not yet been studied in all its morphophonological aspects, the basic contributions recently published on the topic are Vogel, et al. (1983), Agostiniani (1989), Nespor (1990,1993). 5 Tuscany shares this reduction of the infinitives with a wide area of the Center-South of Italy. (See Rohlfs 1968: 359-360, Savoia 1989.) However, in some of these dialectal varieties a final unstressed syllable may be adjoned to the oxytone form, with the consequence of producing a paroxytone pattern: Castrovillari (Calabria) [kan'ta:di] instead of [kan'ta] 'to sing' (Rohlfs 1968). The syllable adjunction, or epithesis, indicates that in these systems the oxytone pattern is not allowed by the prosodic system. The situation for Tuscan, at least in the case of the truncated infinitives, is different, as we will see. 6 See Pieri (1890: 180), Malagoli (1916: 325), and Giannelli (1976: 54, 1988: 599). Even in the vernacular literature, the doubling of the initial consonant following an oxytone infinitive is normally indicated in the orthography; for instance, Guidi (1991): mètte ggallo (p. 214), sapé
OXYTONE INFINITIVES
193
As a first remark, we have to observe that in Pisa the oxytone infinitives belong not only to the dialect, but also to the local pronunciation of Italian.7 In spite of the fact that these forms are a clear cue of a rather low sociolinguistic level, it is frequent to hear them produced even by speakers who do not normally use the dialectal subcode and who have an advanced education. For instance, a school teacher could use oxytone infinitives during his/her lessons as well as while speaking with colleagues, although his/her linguistic performance would be judged to be not in line with the profession he has. In other words, censure is at work, but these forms are still alive and do not show any sign of decay in the real communication process, expecially in contexts not marked by a high degree of formality. The relevance of pragmatic factors in the manifestation of the phenomenon is confirmed by the lack of truncated forms in the case of verbs belonging to registers marked as inherently formal. For example, we believe that it is very difficult for a Tuscan speaker to say esaudi instead of esaudire 'to satisfy' or auspicá, conferí instead of auspicdre 'to augur', conferíre 'to confer'. In fact, these verbs do not occur in everyday speech. 2.
Method of Investigation Our first field of investigation has been the spontaneous speech of Pisans, especially Pisan retailers selling their products in open-air markets. After this stage, where the oxytone infinitives was produced frequently and spontaneously in different phonological and morphosyntactic contexts, we administered a test where the specific phenomena we are interested in were included. We interviewed speakers born and living in Pisa, 8 who differed in age (24-50 years), sex and education. During the interview, every speaker was first asked to speak for a while about himself/herself. After this phase of relatively spontaneous speech, the subject was given a test in which s/he was asked to repeat in the dialect a series of phrases and sentences produced by the author of this article. From the analysis of the results, we find that age and gender are not significant factors in the production of oxytone infinitives, while education is relevant: speakers who have a university ffá (p. 286), sortì ffori (p. 307); Guidi (1994: 20): devi fá ccosì e ccosè; a ffá ccóme voleva lui; with clitics, sfotticchiâlla, attaccàcci, fdssi. These spellings witness the speaker's consciousness of the vitality of the phonological process. 7 It is well known that so-called Standard Italian is an aspiration more than a reality, since any speaker of Italian, even one who does not know a dialect, shows specific traits of his/her regional background. This is especially true of the phonology. 8 In one case only we interviewed a woman born and living not in Pisa, but in Zambra, a village close to Pisa. Her data do not differ in any respect from the data relative to the speakers of the city.
194
GIOVANNA MAROTTA
degree used the truncated forms in spontaneous speech, but seldom during the test. A few words have to be said about the test. There is no doubt that we are tackling a thorny matter, inasmuch as the speaker's metalinguistic competence is involved. For instance, when we ask somebody 'how do you say in your dialect se vuoi prenderlo?' (= 'if you want to take it'), we are asking for a translation, which is always a difficult task, and even more so in the case of a Tuscan speaker who believes s/he is not using a dialect, but the standard language. 9 It is obvious that every speaker of every language or dialect has intuitions about his/her linguistic competence, but the elicitation of the data by a test can be influenced by many factors, including the attitudes of the investigator. This issue is a very thorny one, and we do not have enough space here to discuss it properly. In short, we believe that tests should be used together with linguistic data spontaneously produced by speakers, although we recognize the difficulty in studying spontaneous speech. 10 3,
Truncated Infinitives and 'Rafforzamento Fonosintattico' From the data we collected, it turns out that the most frequent truncated infinitives are undoubtedly those belonging to the first conjugation. The reasons for such a morphological dominance are numerous. a. In the Italian lexicon, the number of verbs belonging to the first conjugation is higher than that of the other conjugations. Not only have the Latin verbs in /-are/ been normally preserved in Italian (amare 'to love', turbare 'to disturb'), but many new verbs have taken the endings of the first conjugation (abbassare 'to lower' from basso 'low', imbiancare 'to whiten' from bianco 'white'). b. In the first conjugation, we find many verbs of common usage: mangiare 'to eat', camminare 'to walk', andare 'to go', lavare 'to wash'. c. The first conjugation is the most regular, and therefore, if it is possible, it is normally preferred, as child language clearly shows. d. The morphosyntactic processes deriving verbs from native as well as from loan words show the adoption by default of the first conjugation: compattare 'to compact' from compatto, like filmare 'to film' from film, stoppare 'to stop' from stop. 9
The traditional method followed by dialectologists is based on a list of written questions usually having to do with individual words. 10 A critical discussion of the traditional methodology used in dialect interviews can be found in Carpitelli & Iannáccaro (1995), where a pragmatic perspective of analysis is proposed as a possible tool for reaching "una ricerca dialettologica dall'interno" (p. 113, fn. 32).
OXYTONE INFINITIVES
195
e. In the case of synonymous pairs, Italian speakers tend to prefer verbs in /-are/. We observe that Tuscans tend to say pigliarelpigliá ' to take', pigiareipigid 'to press', cascarelcascá 'to fall', levaretlevá 'to take from', garbare/garbd 'to like' instead of the synonymous prèndere, prèmere, cadére, tdgliere, placére. As far as sandhi phenomena are concerned, we have to note that Pisan oxytone infinitives feed RF: 1 1 the initial consonant of the word following the infinitive form is indeed long: [vcy:o an,ta d:a 's.o.la] '(I) want to sing alone', 1 2 [LO'VCV:O ve,de 'b:ε:ne] '(I) want to see it well', [um 'vc:le dor.mi 'm:ai] '(s/he) never wants to sleep', [lo 'dε:vi sen,ti 'f:orte] '(you) have to hear it loudly'. In principle, the gemination we find after the oxytone forms may be interpreted in two opposite ways, depending on which interpretation of RF is given. While there is no doubt of the assimilatory nature of the process in diachrony, in synchrony RF can be considered as the product of a consonantal assimilation at a word boundary (see Loporcaro 1988) or as an effect of the Strong Rhyme Constraint (SRC) holding in Italian and in Tuscan on the stressed syllable of a word (see Chierchia 1983-1986, Agostiniani 1992, Marotta 1995). Following a rule-based approach, in the case of a surface phrase like mangiá [t:]anto 'to eat much' we would have two different rule orderings. (1)
underlying form Truncation RF (= assimilation)
mangiâre tanto mangiâr tanto mangiá [t:]anto
11 RF after truncated infinitives may occur in other dialects such as Roman and Sardinian (Loporcaro 1988), while it is normally lacking in the Center-South (Fanciullo 1983-1986: 98, footnote 45). 12 In this utterance as well as in the following ones, we indicate primary and secondary stresses according to a general principle of alternation, taking into account that, in Italian as well as in Tuscan, the metrical domain is right-headed at all levels of the prosodic hierarchy except the foot (Nespor & Vogel 1986). Another aspect about the phonetic transcription which has to be pointed out concerns the velar voiceless stop. As is well known, in Tuscany this segment undergoes the spirantization process taking place after a vocalic nucleus (the so-called gorgia toscana), and giving rise to [h] instead of [k]. (See Giannelli & Savoia 1978, 1979-80.) In the dialect of Pisa (as in Livorno) the glottal fricative can also be deleted, whereby a hiatus is produced in this context. For Italian forms like [a'mi:ko] 'friend', [la 'kat:fa] 'the hunting', we find in Florentine [a'mi:ho], [la 'hat: a], but in Pisan we find both [a'mi:ho], [la 'hat: a], and [a'mi:o], [la 'at: a], the latter even more frequently (Giannelli 1976: 48). As for the other voiceless stops, while in Florence and the central area of Tuscany they normally undergo spirantization, in Pisa the process is not systematic. Therefore, in the phonetic transcription, we decided to maintain the symbols for the stops instead of the corresponding fricatives.
196
GIOVANNA MAROTTA (2)
underlying form RF (via SRC)
mangiâ tanio mangiâ [t:]anto
As we can see, the basic point of difference between the two hypotheses lies on the status of the oxytone infinitive: on one hand, mangiâ is the point of arrival of a synchronic phonological process, on the other, it is the point of departure for the application of RF. We prefer the ordering in (2), since in our opinion truncated infinitives in Pisa dialect are lexicalized forms, given the lack of any alternation between forms like mangiáre/mangiâr, on the one hand, and the form mangiâ, on the other. We are going now to present the evidence for this claim. 4.
Restructuring The first context we consider is the one relative to the final position of a sentence. If truncated infinitives were the result of an assimilatory process taking place synchronically, we would expect to find the full forms in absence of the triggering context, i. e., when the infinitive is not followed by any other word in the utterance. But in the Pisan dialect we always find the reduced forms in prepausal position. None of the speakers we interviewed produced a full form of the infinitivein this context; therefore, we have, ['se l:o ,vci to'k:a] 'if you want to touch it', ['t c ,d:et:o d u' :i] '(I) told you to go out', ['un t ï ,vc:le an'da] '(s/he) does not want to go there', ['um:elo ,fa v:e'de] 'do not show it to me'. Even quotation forms of the infinitive surface without a final /-re/ syllable, for instance, as the answer to the question 'What did you say? Could you repeat the last word?'. On the other hand, it has already been noted that phonological processes involving weakening and reduction of segments or syllables find a preferential field of application in verbal morphology (Giannelli & Savoia 1991, Agostiniani, et al. 1997): truncated infinitives in Tuscan would represent another instance of such a trend. The second context which is relevant is the prevocalic one: here we find only the oxytone variant: [do'vresti ve,de 'an:a] ' you should see Anna', [bi'son:a an,da a 'p:rεnde del: artri 'scrdi] 1 3 'we have to go to get more money'. The same happens when there is a sequence of two infinitives, for instance, [do.ve an'da] ' to have to go', [po.te ar:i'va] ' to be able to come', [vo'r:εi po.te u :i] '(I) would like to be able to go out'. If the underlying form of the infinitive were not the oxytone, the rhotic consonant should surface in the prevocalic context, because it would have the function of avoiding the hiatus. But it does not. A hiatus 13 The rhotacization of the lateral before a consonant ([.artri 'scrdi] instead of Italian [altri 'soldi]) occurs not only in Pisa, but in a wide area of Tuscany as well (Giannelli 1976, 1988).
OXYTONE INFINITIVES
197
is in fact produced not only in the case of two different nuclei (dové andá, fá entra), but even in the case of the same vowel (dové entrá, fá andá) where the vowel is articulated twice (a real hiatus) and not pronounced as a long vowel. Furthermore, in Pisan speech the hiatus also occurs in the context of the so-called gorgia toscana, i. e., the process of spirantization of a voiceless velar stop associated with a syllable onset following a vowel: fa'mi:a] 'friend (fem.)', [la 'a:sa] 'the house', instead of [a'mi.ha], [la 'ha:sa], which are the outputs in other areas of Tuscany. Another piece of evidence against the assimilatory hypothesis can be found in the phonological process of /s/ affrication occurring after a sonorant segment in Tuscan and in many other central and southern varieties of Italian: ['bortsa], ['pεntso], ['saltsa] instead of ['borsa] 'bag', ['pεnso] '(I) think', ['salsa] 'sauce'. Cardinaletti (1993:107) has shown that in the variety of Ancona (in the CenterEast), the process is active not only within the word domain, but also in noun phrases and prepositional phrases: il [ts]ole 'the sun', del [is]ale 'some salt', con [ls]andro 'with Sandro', per [ts]andro 'for Sandro'. The same is valid in Tuscan. Cardinaletti (1993:108-109) claims that in verb phrases, and in particular after a truncated infinitive, affrication can also occur, although these cases are marginal; ?credo di aver [ts]entito la voce di qualcuno ' (I) think that (I) have heard the voice of someone', lesser [ts]inceri è una qualità di pochi 'to be sincere is a quality of few'. These examples would not be grammatical in the Pisan dialect, where /s/ affrication does not occur after the infinitive. If the underlying form were the one with the final sonorant, in the context of /-r/ + /s-/, we would expect to find the phonetic sequence [-r ts-]. However, s gemination occurs here, and there is no trace of the sonorant: [a,ve s:en'ti:to] ' to have heard', [vo.le 's:ε:mpre a,ve r:a'jo:ne] ' to want to always be right'. Even in the case of the reflexive clitic /si/, we find consonant lengthening, but not affrication: [diver'ti s:i] 'to have fun' (*[diver'tir tsi]), [ve'de s:i] 'to see each other', [pet:i'na s:i] 'to comb one's hair'. 1 4 On the other hand, /s/ affrication may occur after other truncated verbal forms, where the sonorant is preserved in word final position: [a'b:jan tsen,ti:to to'na] '(we) have heard it thunder', [son 'tsordi] ' (they) are deaf', ['sjan .tsεntsa 'scrdi] ' (we) are without money'.
14
It is worthwhile observing that /s/ affrication may also occur in Pisa at a high sociolinguistic level. In formal speech, closer to the Italian norm, where oxytone infinitives are sanctioned and then avoided, /s/ affrication is accepted, probably because truncation is a morphophonological phenomenon, while affrication is phonological. Therefore, a sort of implicational scale based on sociolinguistic judgements could be postulated, where the affrication of the sibilant would occupy a different position with respect to the truncation of infinitives.
198
GIOVANNA MAROTTA
An argument supporting the hypothesis of assimilation and against restructuring could be derived by the future forms of Pisan as well as Italian verbs, where/-r-/ occurs: finirò 'I will finish', mangerò 'I will eat', leggerò 'I will read'. If we assume that the two verbal forms (infinitive and future) are strictly related, there could be evidence that an underlying /-r-/ is still present in the infinitive. However, the infinitive form is involved only in the historical formation of the future; the fact that leggerò comes from legere habeo does not seem sufficient for claiming that for speakers of contemporary Italian or Tuscan there is a paradigmatic relation between future and infinitive. In our opinion, the hypothesis of lexicalized oxytone infinitives is not in contrast with the formation of the future tense, because future tense is an inflectional stem related to the infinitive only diachronically, but not synchronically. 5.
RF is Prosodically Constrained Taking into account the arguments presented so far, we conclude that the situation in Pisa is different from that of other Tuscan dialects, where the surfacing of /-r/ in prevocalic context has been used as an argument in favor of underlying representations for the infinitives (Savoia 1974-75:289, Loporcaro 1988:349). In other words, if the loss of /-r/ is not yet complete in Florence, in Pisa it is. As a consequence, if in Florentine "RF after infinitives has to be considered a sandhi assimilation of phonological /r/" (Loporcaro 1988:350), the same does not hold for Pisan, where RF is prosodically constrained. Since /-r/ (or /-re/) does not surface in any context where an infinitive form is used, in the dialect of Pisa there is no alternation for infinitives. Therefore, we are not confronted with a synchronic phonological process producing truncated forms. Truncation as a synchronic rule is not active, since the historical alternation between long and short infinitive forms, as well as the assimilation of /-r/ to the following consonant, are no longer recoverable for speakers. However, there is one crucial piece of evidence which goes against our hypothesis of truncated infinitives as triggering RF because of prosodic constraints: the truncated forms of verbs belonging to the third conjugation may trigger RF although even though they are not oxytones: [voile a't: εnde ,t:ut:e le an'de:le] '(s/he) wants to light all the candles', [lo 'pos:o ,prende 's:ubito] ' (I) can take it immediately', [do'vresti o,no :e 'p:aola] 'you should know Paola'. The speakers we interviewed did not all present RF in this context, although the majority does. The variability depended on the speaker and the prosodic and segmental context. However, in the typical expression of every day speech, [a ,εs:e sin't ε:ri] ' to be sincere; to tell you the truth', only a few of our speakers produced /s~/ gemination. This seems to indicate that after an infinitive of the third conjugation RF is not systematic since it is not phonologically governed.
OXYTONE INFINITIVES
199
For speakers who produce RF after truncated infinitives regardless of stress, we could resort to analogy as an explanation: after prènde, conósce, métte, the following word-initial consonant lengthens just as it does after andd, vedé, senti, where prosodic RF is active. 6.
Articles Beginning with /l-/ The special behavior of articles beginning in lateral consonant {la, lo, le) has to be noted. As has already been discussed (for example, Agostiniani 1992), in Tuscany (Florentine area excluded) the allomorphs la, le, lo normally do not undergo RF: [man.djc la 'torta] '(s/he) ate the cake', [pu,li la hu' i:na] '(s/he) cleaned the kitchen'. In the same way, after an oxytone infinitive these morphemes do not undergo RF: [,vc:le man,dja la 'torta] '(s/he) wants to eat the cake', ['dε.ve pu,li la hu' i:na] ' (s/he) has to clean the kitchen'. A possible reason for this peculiar behavior could be found in the rhythmic structure of the phrase. It has already been suggested (Nespor & Vogel 1979, Marotta 1983-86) that RF is sensitive to the interstress interval, that is, it is realized at the maximum degree in the case of a stress clash: caffè fòrte, tè fréddo. In a sequence verb + article + noun, there is no such prosodic configuration, and therefore RF can be suspended, or at least weakened. Agostiniani (1992:18) reports for Tuscan some interesting instances of such a phonetic weakening with a, a strong monosyllabic preposition/conjunction which triggers RF according to the following stress pattern: a [l]ivórno 'in Livorno' vs. a [l:]ucca 'in Lucca', s'andrá a [v]edere '(we) will go to see' vs. s'andrá a [1:]èggere '(we) will go to read'. However, the lack of RF in this context does not feed the spirantization process taking place in Tuscan in postvocalic position (the gorgia toscana; see footnote 11): a [k]ari[h]are 'to load' (not *a [h]ari[h]are), a [t ]ercâre 'to search' (not *a [ ]ercáre). In our data, RF is not systematically found when there is no stress clash. For instance, we find both la voglio mangid [d:]ománi '(I) want to eat it (fem.) tomorrow', la devi pulí [p:]er bène '(you) must clean it (fem.) very well', gli garba omprá [v:]estíti bòni 'he likes to buy good clothes', and mangid [d]omdni, pulí [p]er bène, omprá [v]estíti. Here, the interstress interval is the same as in the phrases with the articles quoted above; the initial segment may or may not be lengthened, but it is never spirantized. Moreover, the alternation between [+RF] and [-RF] may depend on the syntactic and metrical structure. Consider, for instance, the following utterances with RF: [vcy:o man'dja ,d:u pa'ni:ni] '(I) want to eat two sandwiches' ; [um:elo 'fa n:e,m:en ve'de] 'do not show it to me', [si.po ve'de,t:ut:a la 'rc:ba] 'we can see all the things'; some speakers produced them without RF: [man'dja ,du], [Ta
200
GIOVANNA MAROTTA
ne,m:en], [ve'de.tut:a]. In these cases, as with the articles, the element which does not undergo RF is not the syntactic head of the phrase. The same alternation between [+RF] and [-RF] occurs with the articles. Most speakers do not lengthen the initial lateral consonant, but some do: ['dε:vo la,va l:e a'mi: e] '(I) have to wash the shirts', ['pci spo.sta l:a 'sε:dja] '(you) can move the seat', ['vcy:o pu,li l:a 'stantsa]' (I) want to clean the room'. Therefore, we might conclude that RF is sensitive to the rhythmical structure of the phrase. 7.
Clitics The paradigmatic pressure which leads to the general application of RF after all truncated forms does not show the same strength in all the contexts. We have seen that RF is normally produced after the truncated infinitives of the third conjugation, although they are not oxytone: prènde [t:]ùtto, mètte [s:]ùbito). On the other hand, in the sequence infinitive + clitic, we observe a difference between paroxytone and oxytone forms, since only the latter triggers RF. Consider first the pronominal clitics. With oxytone infinitives we find RF: [se'det:i] 'to sit down', [,ab:ra't :al:a] 'to embrace her', [man'djal:o] 'to eat it', [pu'li:o] ' to clean it' ; with non-oxytone infinitives we do not find RF: ['prendela] 'to take her', [Vendela] 'to sell it (fem.)', ['prεndeti] 'to take it for you', [ko'no ela] 'to know her', ['kjudelo] 'to close it'. However, for Pisan speakers the sequence infinitive + clitic is disfavored. When the infinitive is governed by an auxiliary verb, speakers tend to place the clitic before the finite verb. For instance, instead of ['pc:so sen,til:o do'ma:ni] '(I) can hear it tomorrow', which was produced during the test, speakers would spontaneously say [lo 'pcs:o sen,ti d:o'ma:ni]. The same raising process is found in many other cases. 15 Utterances with raised clitics like [lo 'pcs:o ve,de 'd:o:po] '(I) can see him afterwards', [ti ,vci 'prende ,t:ut:o 'te] '(you) like to take everything for youself', [ti ,vcyo hom'pra ,k:wer ves'ti:to] '(I) want to buy that dress for you', are judged more natural than the corresponding sentences in which the clitic follows the infinitive: ['pcs:o ve,del:o 'do:po], [vci 'prendeti ,tut:o 'te], [vcy:o hom'prat:i ,ver ves'ti:to]. Even with the second person negative imperative, which is expressed with the infinitive form, we have noticed a trend toward the raising of pronominal clitics: ['urn:i ,fa 'm:a:le] 'do not hurt me' ; ['uy:i ,di 'n:ul:a] 'do not say anything to him', [un ti 'mc:ve] 'do not move yourself. (The corresponding forms in Italian would be non mi fare male, non gli dire niente, non ti muovere, or, without 'clitic climbing', non farmi male, non dirgli niente, non muoverti.) In Pisan, the 15 For clitic raising, see Calabrese (1988: 572) for Italian, and Kayne (1991; 1994: 79) for Romance languages in general. Within generative syntax, structures with preverbal clitics are considered to be derived from an order where the clitic follows the verb.
OXYTONE INFINITIVES
201
preference for the raising of the clitic pronoun is not surprising: it is well-known that with a negative imperative the enclisis of a pronoun is typical of the educated language, while in the normal speech, and even more so in the dialects, the preferred construction is with proclisis. 16 In the case of the clitic pronouns ci and si, respectively, locative clitic and reflexive clitic, we again find the doubling of the initial consonant after an oxytone infinitive: ,d:et:o d an'dat: i] ' (I) told you to go there', [,pe d:or'mit: i 'bε:ne I t i | :a ,vc:le r 'lεt.o] 'to sleep well there, you need the bed', [pe v:e,det: î 'hja:ro ,mεt:esi I: o'k:ja:lil 'to see well, you have to put on the glasses', [t i ,met:e tre 'o:re a ,p:et:i'nas:i] ' (s/he) takes three hours to comb him/herself, :le ves,tis:i da 's:e] '(s/he) want to get dressed by him/herself'. If the truncated infinitive is not oxytone, there is no lengthening: [se 'mεt:i i r 'tsa:le I ,ε 'l:i], 'if you want to put on the salt, it is there'. However, as before, we observe the preference for preverbal position: [si ves,ti d:a 's:e], ['se t: i, :1e dor'mi], [se t: î .mεt: e r 'tea:le]. These last utterances were preferred by the native speakers to those quoted previously where the clitic was in postverbal position. 8.
Degenerate Feet From a theoretical point of view, it is interesting to note the metrical structure of oxytone infinitives. Italian and Italian dialects, Tuscan dialects included, are normally considered systems where the foot is bounded and left-headed (Nespor 1993, Helsloot 1996). 17 Within standard metrical theory, the final stressed syllable in forms like andá, vedé, senti constitutes a so-called 'degenerate foot' (henceforth DF) since the foot consists of only one syllable (Goldsmith 1990:172, Kenstowicz 1994:558). In the metrical representation of an oxytone infinitive like andd, two syllables are present: the first is linked to the word node directly, whereas the second projects a monosyllabic foot.
16
Battaglia & Pernicone (1951: 248) already observed that "nell'uso più vivo, ed è spesso un tratto dialettale, la particelia pronominale si suole anche anticipare: Non affaticarti e Non ti affaticare". 17 The question of Quantity Sensitivity is quite complex. According to some scholars, stress on a heavy penultimate syllable (finès tra 'window', arròsto 'roast', formaggio 'cheese') is a clear cue of sensitivity to syllable quantity (Chierchia 1983-86, Sluyters 1990), while for others an unstressed heavy penultimate syllable in borrowings like Ámsterdam, Ròtterdam, Istanbul, as well as in a few native words like pòlizza 'policy', mándorla 'almond', or in place names like Tárante, Lèpanto, indicate that Italian is not quantity-sensitive (Bafile 1996, Marotta 1999).
202
GIOVANNA MAROTTA (3)
an d á \/ \/
σw σs |
||
| F \ / W However, this representation appears to be problematic for a theory which includes the recognition of relative prominence: stress is a prosodic phenomenon which has contrastive but not absolute nature. A syllable is identified as stressed only because it can be compared to another one which is unstressed, i. e., which has a different value in perceptive prominence. There are two problems with the representation in (3): first, the two syllables of the word do not belong to the same metrical constituent; second, only one foot consisting of a strong syllable. There is no doubt that words with final stress in systems assumed to be trochaic have always been a problem for metrical theory. This is why this metrical pattern is normally dealt with only briefly and with special rules. In traditional metrical theory, at least up to Hayes (1995), the theoretical aspects relative to DF may be summarized as follows (Kager 1995:399): (i) in many languages, there are prosodic constraints requiring that minimal words contain two syllables or two moras; therefore, DF cannot be considered as proper foot templates; (ii) in weak position, DF show the trend to lose their status, normally via destressing; (iii) repair strategies often occur in order to restore binary parsing. According to this model, in the Pisan infinitive andâ, we can identify two degenerate feet since both are daughters of non-branching nodes; furthermore, the first foot has no real metrical status, being attached directly to the word node. Despite the difficulties surrounding DF, the theory has tried to motivate them basically on the grounds of Exhaustivity Principle, requiring that all syllables have to be parsed as part of a foot. This principle is strictly connected with the more general principle of Prosodie Mapping: only the phonological material which has been interpreted prosodically, by syllable as well as metrical structure, can receive phonetic interpretation. Hayes (1995:86) has recently challenged this traditional view by proposing that Exhaustivity be considered a 'soft' constraint which can be overriden by 'stronger' constraints. According to this new approach, only stressed syllables (but not unstressed ones) may form DF. Italian could be considered a language with this parameter set at the 'weak prohibition', so a stressed syllable can form a DF. However, within this model, DF are still present with the same term marking the lack of well-formedness.
OXYTONE INFINITIVES (4)
203
Degenerate Foot Parameter: Foot parsing may form degenerate feet • strong prohibition: absolutely disallowed • weak prohibition: allowed only in strong position
9.
Catalexis Since words like città 'city', virtu 'virtue', caffè 'coffee', porto 's/he brought', mangid 's/he ate', clearly belong to the Italian language, the theory must represent them not as monstra (remember the term 'degenerate'), but as structures showing a metrical pattern normally produced and recognized by Italian speakers. That oxytones are not anomalous items is demonstrated by their stable occurrence in speech; there is no resort to special repair strategies in order to override their 'degenerate' nature. No processes of destressing oxytones are normally at work in the language. Moreover, in the Pisan dialect, as well as in many other varieties of the Center-South of Italy, the oxytone pattern is not residual, but productive, as it is spreading to new domains such as infinitives.18 On the other hand, it is significant that oxytone nouns and adjectives fail to pluralize: un caffé 'one coffee', due caffé 'two coffees'; una città 'one city', molte città 'many cities'. The lack of a plural morpheme reveals that something else, beside stress, is unusual at the right edge of these structures. An interesting way to representfinalstress without resorting to DF could be through catalexis (Kiparsky 1991; Jacobs 1994). Catalexis is complementary to extrametricality in many respects (Burzio 1994): both are assumed to occur at word edges; extrametrical syllables receive phonetic interpretation but not full prosodic structure, while catalectic syllables are visible to metrical patterning but do not have segmental material. According to this model, in the representation of oxytone words such as caffè or andá, syllable catalexis would be involved, and a trochaic foot would be recognized. For instance, in a form like andá, there would be a binary, left-headed foot, with the stressed syllable as head and a following dependent syllable lacking segmental material (5). (5)
an d á @ \/ \/ | σw σs σw | \/ \ F \/ W
18 Another dialect where the iambic pattern emerges clearly is Neapolitan, in the hypocoristics Totd < Antonio, Fefè < Raffaele (Bafile 1996).
204
GIOVANNA MAROTTA
This model consisting of a final syllable which is not phonetically interpreted is consistent with the historical development of oxytones. In Latin no words (except monosyllables and a handful of words which underwent final vowel deletion: illìc, istìc, illùc, Arpinás, etc.) were stressed on the final syllable. In Italian, oxytones come from the deletion of the final syllable. The catalexis hypothesis would suggest that the same syllabic skeleton is maintained through time, so that phonological change occurred only at the segmental tier. However, even if we assume catalexis for the metrical representation of Italian words with final stress, it must be lexically marked. Then, as Jacobs (1994:59) writes, "one kind of lexical marking (that is, final foot) has been replaced by another one (that is, final mora - or syllable, G.M. - catalexis)". Furthermore, the epenthesis process applying in some Tuscan varieties (such as Florentine) after word-final stressed vowels (fu =fùe '(s/he) was', perd = perde 'however'; Sluyters 1990:74) does not constitute real evidence for catalexis, at least for mora catalexis, since in this case we would expect vowel lengthening instead of epenthesis (Jacobs 1994:61). In conclusion, we do not believe that catalexis, though interesting, could represent an exhaustive solution for the interpretation of final stress in Italian, because it does not show independent motivation outside the scope of the rhythmic pattern we are dealing with. 10.
Iambic Feet? In our opinion, a theory should represent the outputs of the language in the most natural way is possible. In the case of bisyllabic oxytones, an analysis which does not group the two syllables together goes against naturalness and speakers' intuitions. An alternative way of building feet in these words is to admit an iambic structure, i. e., a binary right-headed foot. (6)
an dá V V σw σs
\/ F | W
But such a representation is incompatible with the setting of the parameter relative to location of the head at the foot level, because in Italian, as in Pisan, the foot is assumed to be left-headed.
OXYTONE INFINITIVES
205
The Uniformity of Algorithm Principle (UAP) is a commonly accepted notion of metrical theory (Roca 1988). According to this principle, in any language there exists only one metrical algorithm which is responsible for word stress. The possible exceptions which are not captured by the algorithm are interpreted through specific rules or conditions. However, as Roca (1988:418) correctly points out, the UAP is useful in languages which have iterative patterns. But in languages where there is not such a regular stress pattern, the UAP cannot be effective without the addition of specific constraints and the relaxation of general principles.19 The preservation of the UAP becomes even more difficult for languages like Italian, where stress is lexical, and therefore cannot be derived from a prosodic algorithm. In this case, the language might not undergo a straightforward analysis because theoretical assumptions like binary branching and uniform headedness only apply through resort to ad hoc machinery. The power of an algorithm predicting the stress position in an Italian word cannot be high, given the variability of the stress patterns. Even the weight of the penultimate syllable, which is the only residue of the Latin stress rule and which has been interpreted as a clear cue of quantity-sensitivity, turns out to play a reduced role in contemporary Italian, as clearly shown in place names like Amsterdam, Ròtterdam, abbreviations like Fíninvest, Italmec, as well as words like internet, èternit20 Italian and Italian dialects have lexical stress and allow for different stress patterns, the basic ones being trochees, dactyls and iambs. Although the relative frequency of these patterns is not the same, all three are present and well-attested in the language. A theory which aims to have explanatory power, has to deal with all the stress patterns and to find a plausible interpretation for them all, although such a goal might mean that some previous theoretical assumptions are not valid. In our opinion, it is time to challenge the UAP. The recognition of one metrical pattern as dominant, because it is more frequent than the others, does not imply the automatic elimination of all the other patterns occurring in the language. A theoretical assumption strictly connected with the UAP is the obligatory binarity of bounded feet (Halle & Vergnaud 1987, Kager 1995). Standard metrical theory recognizes only one size for foot constituents, with the exclusion of ternary feet. But in Italian and Italian dialects, to assume that feet are always uniformly binary does not reflect the outputs of the language. Such a view implies the 19 For instance, see Harris (1983), where Spanish stress is accounted for by a metrical theory based on UAP and ad hoc machinery. See the criticism expressed by Roca (1988). 20 Indeed, initial word stress is becoming more and more dominant in Italian, probably under the influence of the increasing number of English loans. This pattern is found even in the case of a penultimate heavy syllable, which should be stressed if Italian were quantity-sensitive: Fìninvest, not *Finïnvest (Marotta 1999).
206
GIOVANNA MAROTTA
introduction of special diacritics like Extrametricality into the algorithm. Recent proposals (Haraguchi 1991, Nespor 1993, Burzio 1994, Bafile 1996) aim to provide alternatives to the bmarity constraint: a ternary constituent is recognized in the foot inventory, although with a greater degree of markedness. The introduction of ternary feet has two relevant consequences: (i) extrametricality is no longer necessary; (ii) the constraint on adjacency is relaxed. Strict bmarity, as applied to Italian and Italian dialects, appears to be forced by the theory rather than being a constraint on naturalness. No Italian speaker would say that the final syllable of a word like távolo has a different prosodic status from that of the penultimate syllable. In the case of the oxytone words, the rejection of the UAP might have the consequence of introducing an iambic foot. A more relaxed notion of the foot parameter should therefore be adopted, admitting right-headed feet along with left-headed ones, which are, nonetheleess, recognized as dominant. There are some advantages in adopting this new view. First of all, we no longer need DF, which have always been a problematic aspect of traditional metrical theory. Second, a greater degree of naturalness would be reached in the representation of bisyllabic words like città or andà. Third, the iambic foot would be in line with the rhythmical structure at the higher prosodic levels. It is wellknown that the Phonological Word as well as the Phonological Phrase are rightheaded. The asymmetry between the position of the head of the foot, on the one hand, and higher metrical domains, on the other, would be reduced if the iamb were inserted into the foot inventory of the language. Moreover, with respect to Latin, the Italian iambic pattern represents the only true innovation in the position of word stress. The Latin stress system was quantity-sensitive, while Italian is not. However, in Italian (and in Italian dialects) the stress position has been normally preserved from Latin in the overwhelming majority of the words. The novelty in metrical patterning comes exclusively from the introduction of final stress: 21 the current analysis ought to accept the new pattern instead of confining it into a sort of peripheral limbo, not belonging to the core of grammar. Once Uniformity (as a constraint on binarity and headedness) has been rejected, the inventory of possible feet in Italian as well as in the Pisan dialect will include trochees, dactyls, and iambs. Such an increase of the foot inventory may involve a better fit with the data, but the degree of generality in the representation is lower. 21 The only relevant exceptions are the changes in words containing the muta cum liquida clusters (Lat. integrum > It. intéro, Sp. entéro, Fr. entier) or the vocalic sequences /'ie, /'io/, /éo/ (Lat. muUerem > South. It. moglièra, Sp. mujér, Old Fr. moille). In both cases, the stress is shifted to the penultimate syllable of the word (Lausberg 1969: §149).
OXYTONE INFINITIVES
207
Even if we enlarge the foot inventory, it is still difficult to permanently ban DF from the representation. Consider first the case of initial unstressed syllables. For instance, in words like patáte ' potatoes' or rivedé ' to see again', how do we represent the first syllable in a metrical tree? It could be a DF or a light unparsed syllable. In this particular context, the fact that an unstressed syllable is not part of a foot is only a minor problem since the unparsed element is at the left edge of the word, and Italian is right-dominant at the levels of Phonological Word and Phonological Phrase. Therefore, we could associate this unparsed syllable directly to the word node. Another possibility would be to admit amphibraching feet (Halle & Vergnaud 1987, Hayes 1995). The most problematic aspect which survives in the sketch of metrical structure we have outlined so far concerns monosyllables. Though in Italian there are very few monosyllabic stressed words (re ' king', tu 'you', tre ' three', tè 'tea'), they do exist, and the theory has to deal with this pattern as well. For this specific case, we do not have any solution at hand which could be better than the re-proposal of a unary foot. Such a foot would be assigned the highest degree of markedness, since it is composed only of the head. At least for the moment, it seems that the metrical representation of oxytone words and, therefore, of the truncated infinitives, is still awaiting an exhaustive solution.
REFERENCES Agostiniani, Luciano. 1989. "Fenomenologia dell'elisione nel parlato in Toscana". Rivista italiana di dialettologia 13.7-46. Agostiniani, Luciano. 1992. "Su alcuni aspetti del ' rafforzamento sintattico' in Toscana e sulla loro importanza per la qualificazione del fenomeno in generale". Quaderni del Dipartimento di Linguistica dell'Università di Firenze 3.1-28. Agostiniani, Luciano & Luciano Giannelli. 1990. "Considerazioni per un'analisi del parlato toscano". Vitaliano regionale: Atti del XVIII Congresso della Società di Linguistica Italiana ed. by Michele A. Cortelazzo & Alberto M. Mioni, 219-237. Roma: Bulzoni. Agostiniani, Luciano, Paola Bonucci & Margherita Castelli. 1997. "Osservazioni su un fenomeno di sandhi nel parlato perugino". Atti del Terzo Convegno delta Società Internazionale di Linguistica e Filologia Italiana (volume I) ed. by Luciano Agostiniani, et al., 3-29. Napoli: Edizioni Scientifiche Italiane. Bafile, Laura. 1996. "Sulla rappresentazione delle strutture metriche ternarie". Quaderni del Dipartimento di Linguistica delVUniversità di Firenze 7.45-67. Battaglia, Salvatore & Vincenzo Pernicone. 1951. La grammatica italiana. Torino: Chiantore.
208
GIOVANNA MAROTTA
Burzio, Luigi. 1994. Principles of English Stress. Cambridge: Cambridge University Press. Calabrese, Andrea. 1988. "I pronomi clitici". Grande grammatica italiana di consultazione (volume 1) ed. by Lorenzo Renzi, 549-592. Bologna: Il Mulino. Cardinaletti, Anna. 1993. "The affrication of [s] in a Central Italian variety: Syntactic conditions on a 'sandhi' rule". Syntactic Theory and the Dialects of Italy ed. by Adriana Belletti, 99-129. Torino: Rosenberg & Sellier. Carpitelli, Elisabetta & Gabriele Iannáccaro. 1995. "Dall'impressione al metodo: per una ridefinizione del momento escussivo". Dialetti e lingue nazionali: Atti del XXVII Congresso della Società di Linguistica Italiana ed. by Maria Teresa Romanello & Immacolata Tempesta, 99-120. Roma: Bulzoni. Chierchia, Gennaro. 1983-1986. "Length, syllabification and the phonological cycle in Italian". Journal of Italian Linguistics 8.5-34. Fanciuilo, Franco. 1983-86. "Syntactic reduplication and the Italian dialects of the Centre-South". Journal of Italian Linguistics 8.67-104. . 1997. Raddoppiamento sintattico e ricostruzione linguistica nel Sud italiano. Pisa: ETS. Giannelli, Luciano. 1976. Toscana. Pisa: Pacini. . 1988. "Toscana". Lexicon der Romanistischen Linguistik (volume 4) ed. by Gunter Holtus, Michael Metzeltin & Christian Schmitt, 594-606. Tübingen: Niemeyer. Giannelli, Luciano & Leonardo M. Savoia. 1978. "L'indebolimento consonantico in Toscana: 1". Rivista italiana di dialettologia 2.23-58. . 1979-80. "L'indebolimento consonantico in Toscana: 2", Rivista italiana di dialettologia 3/4.38-101. — . 1991. "Restrizioni sull'esito [h] da t in fiorentino e nelle altre varietà toscane". Studi italiani di linguistica teorica e applicata 20.3-57. Goldsmith, John A. 1990. Auto segmental and metrical phonology. Oxford: Blackwell. Guidi, Guido. 1991. A Ppisa si dice 'osì. Pisa: il Portone. . 1994. "Mi baciasti, mi baciasti". Er tramme. Trimestrale di vernacolo pisano e delle tradizioni popolari 10.20-21. Halle, Morris & Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge, Mass.: MIT Press. Haraguchi, Shosuke. 1991. A Theory of Stress and Accent Dordrecht: Foris. Harris, James. 1983. Syllable Structure and Stress in Spanish. Cambridge, Mass.: MIT Press. Hayes, Bruce. 1995. Metrical Stress Theory. Principles and Case Studies. Chicago: University of Chicago Press.
OXYTONE INFINITIVES
209
Helsloot, Karijn. 1996. Metrical Prosody. A Template-and-Constraint Approach to Phonological Phrasing in Italian. The Hague: HAG. Jacobs, Haike. 1994. "Catalexis and stress in Romance". Issues and Theory in Romance Linguistics. Selected Papers from the Linguistic Symposium on Romance Languages XXIII ed. by Michael L. Mazzola, 49-65. Washington, D.C.: Georgetown University Press. Kager, René. 1995. "The metrical theory of word stress". The Handbook of Phonological Theory ed. by John Goldsmith, 367-402. Oxford: Blackwell. Kayne, Richard S. 1991. "Romance Clitics, Verb Movement, and PRO". Linguistic Inquiry 22.647-686. . 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, Mass.: Blackwell. Kiparsky, Paul. 1991. "Catalexis". Unpublished ms, Stanford University. Lausberg, Heinrich. 1969. Romanische Sprachwissenschaft: I. Einleitung und Vokalismus; IL Konsonantismus. Berlin: de Gruyter. Loporcaro, Michele. 1988. "History and geography of raddoppiamento fonosintattico: Remarks on the evolution of a phonological rule". Certamen Phonologicum ed. by Pier Marco Bertinetto & Michele Loporcaro, 341-387. Torino: Rosenberg & Sellier. . 1997. Vorigine del raddoppiamento fonosintattico. Saggio di fonologia diacronica romanza. Basel & Tübingen: Francke. Malagoli, Giuseppe. 1916. La letteratura vernacola pisana posteriore al Eucini, con note linguistiche e glossario. Pisa: Bemporad. Marotta, Giovanna. 1983-86. "Rhythmical constraints on syntactic doubling". Journal of Italian Linguistics 8.35-52. . 1995. "Apocope nel parlato di Toscana". Studi italiani di linguistica teorica e applicata 24.297-322. . 1999. " Degenerate Eeet nella fonologia metrica dell'italiano". Fonologia e morfologia dell'italiano e dei dialetti d'Italia: Atti del XXXI Congresso della Società di Linguistica Italiana ed. by Paola Benincà, Alberto Mioni & Laura Vanelli, 97-116. Roma: Bulzoni. Nespor, Marina. 1990. "Vowel deletion in Italian: The organization of the phonological component". The Linguistic Review 7.375-398. . 1993. Fonologia. Bologna: II Mulino. Nespor, Marina & Irene Vogel. 1979. "Clash avoidance in Italian". Linguistic Inquiry 10.467-482. . 1986. Prosodic phonology. Dordrecht: Foris. Pieri, Silvio. 1890. "Appunti morfologici concernenti il dialetto lucchese e il pisano". Archivio glottologico italiano 12.161-180.
210
GIOVANNA MAROTTA
Roca,Iggy. 1988. "Theoretical implications of Spanish word stress". Linguistic Inquiry 19393-423. Rohlfs, Gerhard. 1968. Grammatica storica della lingua e dei suoi dialetti (volume 2). Torino: Einaudi. Savoia, Leonardo M. 1974-75. "Condizioni fonetiche del fiorentino comune e alcune proposte per una teoria fonologica concreta". Studi di grammatica italiana 4.209-330. —. 1989. "Proprietà vocaliche e struttura metrico-sillabica in alcune varietà abruzzesi". Studi in memoria di Ernesto Giammarco, 331-380. Pisa: Giardini. Sluyters, Willebrord. 1990. "Length and stress revisited: A metrical account of diphthongization, vowel lengthening, consonant gemination and word-final vowel epenthesis in Modern Italian". Probus 2.65-101. Vogel Irene, Marina Drigo, Alessandro Moser Sc Irene Zannier. 1983. "La cancellazione di vocale in italiano". Studi di Grammatica italiana 12.191-230.
SONORITY AND DERIVED CLUSTERS IN RAETO-ROMANCE AND GALLO-ITALIC
JEAN-PIERRE MONTREUIL University of Texas at Austin
0.
Introduction Romance languages are not known for the complexity of their consonant clusters. More often than not, they offer no particular challenge to traditional sonority-based analyses. However, some Raeto-Romance and varieties do display unusual sequences of consonants in word-initial position, as shown in (1). 1 Gallo-Italic (1)
Bravuogn [djle:r] Comelicano [dnei]
Lombardo Piemontese
[vdel] [bzu'ne]
In this paper, I examine these clusters from a point of view which incorporates both synchronic and historical considerations. However, I will limit both the coverage of the data and the scope of the discussion in the following way. Data. I will consider only word-initial clusters which are derived rather than underlying. This automatically excludes most of the clusters which occur freely in many Romance languages, whether they obey sonority ([fr]) or not ([sp]). The exact meaning of 'derived' and the justification for this choice will be discussed in §1.1. I will also limit my survey to representative varieties of five dialects from Northern Italy and Eastern Switzerland. They have in common the fact that, at 1
Due to the abundance of illustrations this article contains, glosses have not been provided. The reader is referred to the sources cited in each section. General references for all of include Battisti (1988), Bruni (1992), Cortelazzo & Marcato (1992). For each dialect, the transcription is given using IPA conventions, but in the illustration tables, words are given as they are found in the original sources. Gallo-Italic
212
JEAN-PIERRE MONTREUIL
some point in their history, they all were deeply influenced by Germanic, in such a way that unstressed vowel deletion, and hence the emergence of clusters, was heavily favored. Two dialects are Raeto-Romance: (1) Surmiran, a variety of central Romansch, spoken around Bravuogn (Bergün) in the Swiss Graubunden; (2) Comelicano, a variety of Eastern Dolomitic, spoken in Comelico, north of Ampezzo (province of Belluno). The other three dialects are Gallo-Italic: (3) Lombardo, especially in its peripheral varieties; (4) Piemontese, as heard in Oleggio, on the Ticino river; and (5) Romagnolo, spoken in Romagna, east of Emilia. Analysis. Our understanding of the behavior of consonant clusters in natural languages has benefited recently from the results of two trends of research: (1) defining the true nature of sonority, and its role in the determination of syllable structure, and (2) recognizing the segmental constraints on adjacency that rule out many otherwise sonority-acceptable clusters and allow some otherwise nonacceptable clusters. This feature-based approach is particularly promising and has yielded numerous insights in the analysis of English clusters (Yip 1989, Padgett 1991, Steriade 1994, Truckenbrodt 1994, among others, and most recently Morelli 1998). However, I will ignore here the latter approach and focus on sonority and in particular on the question of sonority differential. I will show that RaetoRomance and Gallo-Italic clusters pose a number of problems in terms of structure, and that they lend themselves to several competing analyses among which it is not always possible to decide. By contrast, and this is most clearly apparent in §3, the notion of sonority differential proves useful in understanding dialectal variation. 1. 1.1
Preliminaries Sonority and Derived Clusters To account for the phonotactics of consonants, many different sonority scales have been proposed, some of which go deeply into the detail of segmental features. It will be sufficient to adopt here a 'major category' sonority scale of the kind proposed by Selkirk (1982, 1984) and others. When restricted to consonants, this scale consists in a four-degree hierarchy which encodes the decreasing sonority that obtains between the major categories of Liquids, Nasals, Fricatives and Stops. We will refer to the Principle + Hierarchy package shown in (2) as SON.2
2
Sonority is a phonetically grounded notion, and, as such, functions well as a primitive. Several other phonological notions can be derived from it, for instance markedness (Lekach 1979, Eckman & Iverson 1993). At the same time, there is no reason to think that sonority
213
SONORITY AND DERIVED CLUSTERS
(2) Sonority (= SON) Sequencing Principle: Sonority must increase from syllable margins to syllable peaks Hierarchy: L(iquid) > N(asal) > F(ricative) > S (top) 4 3 2 1 Sonority distance diagram: Son = -3 Son = -2 Son =-1 Son = 0 Son = +1 Son = +2 Son = +3
Table 1 shows the Romance canvas of our data tables (Tables 3-7) for twoconsonant clusters (C1C2). Columns marked -3, -2, -1, 0, +1, +2, +3, as per the diagram in (2), indicate the sonority distance (C2 sonority minusC1sonority). For instance, a cluster [fn], with respective sonorities of 2 and 3, as per SON, can be found under +1; the cluster [lv], with respective sonorities 4 and 2, can be found under -2 (compare with Baertsch 1998:3). Accordingly, I will refer to clusters whose sonority decreases from right to left (-3, -2, -1) as negative sonority clusters, to plateaus as equal sonority clusters, and to those whose sonority increases (+1, +2, +3) as positive sonority clusters. The Romance background. In input forms, and again if we exclude from consideration [sC] clusters which violate sonority,3 Romance languages typically
itself cannot be derived (Carnie 1994, Clements 1990, Rice 1992, 1993). In this paper, however, we will leave this issue aside and simply assume that sonority is a primitive. 3 The existence of sC clusters in Indo-European languages and in others, sometimes even the coexistence of sCi sequences with Cis sequences, has been discussed extensively. Although they violate SON, such clusters do not in and of themselves invalidate it; in fact, SON becomes all the more useful as a tool as it allows us to express generalizations concerning the degree of violation incurred. It is not our purpose here to review or discuss the abundant literature on the subject. Italian sC clusters have been especially well studied since they trigger occasional epenthesis, force differences in allomorph selection (il vs. lo), and interact with larger issues such as syllabification and raddoppiamento. We will simply note that, in confronting such violations, there are four fundamental approaches which still preserve the hierarchy: i. making SON non-inclusive, in such a way that /s/, /ƒ/, and possibly all coronal fricatives, are not ranked within it. This is done by stipulation. ii. relaxing SON in such a way that no violation occurs; for instance, Jespersen's (1904) hierarchy classified in the same group /s/ and voiceless stops (however, /z/ and voiced stops belong to different groups).
214
JEAN-PIERRE MONTREUIL
accept all +3 clusters except for [tl] and [dl], and rule out all other clusters except +2 clusters of labial fricatives + liquid. SON-3
SON-2
SON-1
SON 0
SON + 1
SON+2
out
out
out
out
out
out, except
SON+3 OK, except *tl, dl
Table 1. Two-consonant clusters in Romance
Although this is a simplification, we will define as ROMANCE SONORITY violations any cluster which contradicts Table 1. For instance, French has, in addition to all unmarked clusters, a large number of clusters which violate ROMANCE SONORITY, as shown in the representative list given in (3). (3)
a tøf], fips], [pt], [p ], [kt], [ks], [gz], [t ], [dz], [ts], [dj] as in pfennig, psy, ptose, pschitt!, cténaire, czardas, xylophone, choke-bore, tsar, tsé-tsé, djinn b. [pn], [km], [kn], [tm] as in pneu, khmer, knout, tmèse
The consonants of the first group (3a) are often analyzed as affricates or 'near-affricates', and to the extent that they associate to only one metrical unit, they do not constitute a problem for any theory that builds the syllable on intermediate levels like skeletal slots. They can also be analyzed as partial geminates, sharing a voice node. Either analysis accounts for the fact that epenthesis can never occur in the middle of such clusters. The second group (3b) involves nasals, with a potential disagreement in voice, although the nasal itself is often at least partially devoiced. The clusters are the site of occasional epenthesis, as in [pønø] for pneu, and accordingly are usually analyzed as presenting a greater challenge for syllabification, since they associate to two skeletal slots. Thus, the words bleu, psy and pneu represent three types of clusters which can be represented as in (4).
iii. modifying the structure of sC clusters in such a way as to remove the violation; for instance, by viewing them as near-affricates, thus associated to a single metrical unit (Selkirk 1982) iv. letting SON apply, so that /s/ will be stray, and will readjoin as an appendix. Unlike option 3, this makes /s/ less bound to the following consonant, and allows for variations in epenthesis sites such as SCOLA 'school' > Old French escole vs. Walloon sicole.
SONORITY AND DERIVED CLUSTERS
215
• The +3 cluster in bleu respects ROMANCE SONORITY and accordingly is syllabified into the onset of the core syllable (4a). • The +1 cluster in psy, which violates ROMANCE SONORITY, is not a cluster but a single metrical unit as in (4b). • The +2 cluster in pneu also violates ROMANCE SONORITY. The presence of epenthesis can then be related to the fact that [p] readjoins to the syllable node (4c). Additionally, we need to distinguish between underlying clusters and derived clusters. Derived clusters result from a contextual rule of vowel deletion, or from the aggregation of two onsets. Again, a clear illustration can be taken from French, which has clusters derived from vowel-deletion, as in déjà [de3a] 'already' pronounced [dja] in fast speech, and clusters derived from the aggregation of two onsets, as in genoux 'knees', pronounced [jønu] after a consonant, as in sept genoux 'seven knees', but [gnu] after a vowel, as in trois genoux 'three knees', given that the underlying representation of genou contains an empty segment (= schwa), as in (5). (See also Lauefer 1991). (5)
Underlying representation of genou (with constituents added)
Derived clusters pose the problem of preservation vs. restructuration. It is clear that historically, most C-C clusters originate in CVC sequences where V was unstressed and underwent deletion. This raises a number of questions, two of which are basic.
216
JEAN-PIERRE MOMTREUIL
1) Are derived clusters in any way structurally different from other CC clusters in the language; i.e., is [ n] in genou to be represented any differently from the initial cluster in bleu, psi or pneu? 2) More specifically, must two identical surface clusters be analyzed as structurally different if one of them is derived and the other underlying; in other words, what is the representation of CC clusters from CVC which obey ROMANCE SONORITY?
Table 2 illustrates this contrast. Latin etyma have been added to show the historical deletion in column A. A. DERIVED CLUSTERS
B. UNDERLYING CLUSTERS
French
ple
pler
< PLACERE
Brav. Sunn.
fle
flamә
< FLAMMA
Brav. Surm.
Brav. Surm.
d εmber dzε:әt
d εndәr dzi:
< GENERU
Brav. Surm.
fregdә
fre:r
Comelico
dretu
drapu
< *DRAPPU
Comelico
kri
< *QUAERIRE
kride
< *CRIDARE
Comelico
pliz
< PELLICE
pleura
< *PLEUR-
Table 2. Derived vs. underlying clusters: no sonority
violation
This touches on the very broad issue of structural preservation in historical linguistics: after deletion of a metrical unit, is the new, shorter sequence of segments rearranged through the application of a metrical template or is reassociation done without upsetting the pre-existing stracture? The viability of these two options is well documented in studies on compensatory lengthening. We call the first option RESTRUCTURATION, whereby the new cluster (as long as it obeys sonority restrictions) is indistinguishable from pre-existing clusters. We call the second option PRESERVATION, whereby the new cluster differs structurally from pre-existing clusters (for instance, through appendix-formation). This is illustrated in (6) and (7) by comparing a hypothetical form [pla] < PLA to a newly created form [pla] < PALA. (6) Restructuration σσσσ /\
/\
pa l a
/\
/\
like
\ pla
/\
/\\ p 1 a < PLA
SONORITY AND DERIVED CLUSTERS
217
(7) Preservation /\ /\0 pa l a
/|\ p1a
σσσσ unlike /\ /\ \ p l a
For the clusters under consideration here, the data will show that preservation is the correct option, even when derived clusters obey sonority restrictions. This is motivated essentially by the numerous cases where 'new' CCclusters co-exist with a CVC- variant, either stylistically or in free variation. A few examples from Comelicano will suffice to illustrate what is indeed a pervasive situation. In Comelicano, underlying clusters never change shape, whereas the consonants of derived clusters are 'separable', either through zero/vowel alternations, as in (8), or through metathesis, as in (9). (Note that the entry in the de Lorenzo Tobolo (1977) dictionary is systematically the vowelless one.) (8) Comelicano: 0/V alternations [dbotu] / [dubotu] [dgurenti] / [dugrenti] / [ dugurenti] [dman]/[duman] [ksi] / [kusi]
[kri]/[kari] [tváia] I [tuváia] [vlimentu] / [vilimentu] [vnuda]/[venuda] / [vinuda]
(9) Comelicano: metathesis [rdondi] / [drondi]
[flizada]/[filzada]
Plangg (1973) makes a similar point for Badia Dolomitic, and distinguishes between normal styles, which elicit forms without vowels, and high styles, where a vowel can be heard. The phenomenon is similar to zero / schwa alternations in codas (colloquial [ry tl] vs. formal [ry tәl]), with one difference: in onsets, different vowels may appear in the long form. (10) Badia Dolomitic stylistic alternations Colloquial Formal [mpe] [impe] [ndc:] [indo:] [nkø:] [inkø:] [lbo] [әlbo] [rdi] [ardi]
218
JEAN-PIERRE MONTREUIL
Comelicano in particular displays regular morphological patterns based on this kind of alternation, as shown in (11). Their regularity is only blocked by OCP effects, as in [zu'ze] (*[zze]), [zu'zu] (*[zzu]), [zu'zo] (*[zzo])4. Again, such alternations motivate the choice of (7) over (6), since they could never occur in underlying clusters. (11)
Infinitive
1 Sing. Pres. Indie.
Past participle
[pda]
[peau]
[pdo]
[θsa]
[θeesu]
[rve]
[revu]
[6so] [rvo]
[spra]
['speru]
[spro]
Accordingly, the data presentation includes some forms of non-coronal stop + liquid clusters, even though most are excluded for obeying ROMANCE SONORITY in the first place: the ones we consider all constitute derived clusters. Similarly, sC clusters also divide into derived and non-derived categories. Again, it is particularly crucial to distinguish between underlying clusters, which originate from Latin and are characteristic of the phonology of standard Italian, from derived sC clusters, which are specific to dialects. For instance, Fortina, et al (1992) list 719 sC forms from Oleggio, but very few are derived: [smenza] 'It. semente', [sne] 'It. segnare', [srœn] 'It. sereno'. We find near-minimal pairs of the type [smaie] 'It. smagliare' vs. [smai] 'It. assomigliare'. The short list in (12) provides examples from Piacentino. In our Tables (3-7), only derived clusters are included. (12)
A. DERIVED CLUSTERS
B. UNDERLYING CLUSTERS
srein sdass
sbaleinc
It. sbilenco
spëss
It. spesso
It. sereno It. setaccio
snä
It. cenare
snell
It. snello
slär smell
It. sellaio It. gemello
slitta
It. slitta
smort
It. smorto
steimbar
It. settembre
stemma
It. stemma
svëüa
It. civetta
svani
It. svanire
\
Phonetic articulation occasionally indicates whether the cluster is derived or not. For instance, Bravuogn Surmiran has forms like [stçikr] from SECURU, which contains a derived cluster. All [s]'s from underlying initial sC clusters have 4
This remark pertains to Comelicano only. OCP violations are not unheard of. For instance, the Piacentino cognate of Italian 'sesino' is [zzeai] (Gorra 1890:142).
SONORITY AND DERIVED CLUSTERS
219
palatalized to [f]; but the derived form is [stçikr] and not *[ tçikr]. Such phenomena provide additional motivation for our structural distinction. A quick survey of the five dialects in question will now allow us to display the data tables (Tables 3 - 7 ) over which generalizations will be drawn. These tables, which present the Raeto-Romance and Gallo-Italic data, are organized like Table 1 with one additional column: the leftmost column (C1) indicates the first consonant of the cluster. Most fricative + liquid clusters are not included in +2, nor are most stop + liquid clusters in +3, except for [tl], [dl], with exceptions to be discussed. It is understood that clusters that do not violate ROMANCE SONORITY abound in all dialects, as expected. Numerical impossibilities are indicated by darkened boxes: for instance, [r] and [1] show black boxes under +1, +2, +3, since they are located at the most sonorous edge of the scale. 1.2
Bravuogn Surmiran The 25-phoneme consonant inventory of Bravuogn Surmiran, as documented in Lutta (1923), is given in (13). Its relative complexity stems from the coexistence of a series of medio-palatals [ ], [3], [if] and [d3] with a series of palatals: [ç], [j], [tç] and [dj]. Affricates are to be analyzed as monophonemic; for sonority purposes, we classify them under 1 in SON, together with stops. (13)
p/b f/v
t/d s/z
m
n 1 r
k/g ƒ/ 3 t/d3 n
ç/j tç/dj n y
All consonants are attested as word-initial single onsets, except [r]. Palatals [ç] [j]' [n] and [y] cannot be the first member of a branching onset. One might ascribe that property to their complex underlying structure. However, a subset of affricates, namely [ts], [tj] and [d3] can be the first members of a branching onset; see [tslen], [t nar], [d3le:r]. Thus, they are included in the table. Liquids [1] and [r] are not attested as the first member of complex onsets, but are included here for comparison with other languages.
220
JEAN-PIERRE MONTREUIL
C l ! -3
2
-1
0
+1
+2 pne:tә
p
ptço
pse:r
b
bke
bzents
+3 pyer
hyer
t
tsiәvә
tmεgvә
d
dzukr
dmjets
d3εmbәr knaps
k
g tslen
ts tfnar
rf
d3le:r ftsegre
fne:ftrә
vdzevә, vduәtçә
V
vzevә, v3әn]
1 r Table 3. Two-consonant onset clusters: Bravuogn Surmiran
1.3 Comelicano The consonant inventory of Comelicano (Tagliavini 1926, de Lorenzo Tobolo 1977) consists of 19 phonemes, all of which can be single onsets and all of which, except the affricates and [h], can be the first member of an initial cluster. (Note the interdental [8]; its voiced counterpart [ö] is only an allophone.)
C1
-3
-2
! -1
0 \ pta,pdana
p
+1
+2
+3
p8e
pnel
pliz
b t
tvaia
tnaia
tlei
dfora, d v a n i
dman, dnei
dlibre,dretu
k
ksi
kneilu
g
gnocco
d
dbotu, dgurenti
6
6sa
fni
f
v
vnuda
vdel
N
mdaia
1
Iva
r..
8rumi
rdondi
mna
vlimentu
mlestar
rve Table 4. Two-consonant onset clusters: Comelico
221
SONORITY AND DERIVED CLUSTERS
(14)
p/b
t/d θ
f/v
k/g
s/z
h
tf/dj m
n
n 1 r
1.4 Lombardo The inventory of Lombardo also consists of 19 phonemes. (In Table 5, the asterisked forms are from Piacentino, a transition dialect spoken at the very edge of the Lombardo domain, but most aptly characterized as Western Emiliano (Bearesi 1982, Bruni 1992, Gorra 1890).5) (15)
p/b
t/d
f/v
s/z
m
n
k/g /3 t /d3 n
1 r
C1 -3
-2
-1
+2
0
+1
p
pka, ptass*,
pfilar, psiga*, pnell*
pdagna*
pzä*
b
bdjer, bcon*
bsognäva
blëtt*
tsen
t
tnir,
dzegr,dséva,
d
+3 plus*
taäia*
där*, trüss*
dmànda
dla,dritt
kmencia
clazion*
dvün csa*,cverc'*
k
g f
fteina*
flipp* frameint*
vtär, vdel
V
lgnus*
1 rfœstat
Table 5. Two-consonant onset clusters: 5
vritä*, vlüd
mner
N
r
vgnüd vjinz, vcin*, vgilia*, vzola*
Lombardo
In addition, Lombardo displays very unusual Glide + C initial clusters: [jla], [j e], [jdei], [wliia], not analyzed here.
223
SONORITY AND DERIVED CLUSTERS
1.6 Romagnolo The consonant inventory of Romagnolo (Mattioli 1986) consists of 20 phonemes. (17)
p/b
t/d
f/v
s/z
k/g
m
n
n
1 r
y
/5
ts/dz
C1
-3
-2
-1
0
p b
+2 pnoda
blëzza
bsogna tsi
tni
Üarana
dbend, dgivi
dvinte, dsen
dmanda,dnin5
dla
kminzipié, knoss
k gverna
rf V
+3
bton, bcunê
t d
+1 pfe, psion
fness vden
vnend
vle
1 r Table 7. Two-consonant onset clusters:
Romagnolo
2. Structural Analyses 2.1 Analyses of Onset Clusters Given the large number of proposals concerning syllable structure in general (see most recently van der Hulst & Ritter 1998), the array of theoretical possibilities for analysis of unusual clusters is considerable. We will only sketch out four specific proposals. 2.1.1 No Sonority Violation Consider the Surmiran word [bke] (Table 3). Under this proposal, the initial cluster in [bke] has exactly the same structure as that in [ble]. This possibility emerges from the concept of MSD (= Minimal Sonority Distance), which can be set at zero. It is always possible to relax the sonority requirements sufficiently so that many clusters are viewed as normal and thus constitute no violation at all. This is commonly done in the analysis of Slavic languages, where such clusters abound. In Slovak, Rubach & Booij (1992) find that SON must be relaxed in two ways: (1) fricatives and stops cluster together, and (2) there is no MSD in
224
JEAN-PIERRE MONTREUIL
obstruents. As a result, any two obstruents in any order are allowed: [sk], [ k], [zb], [ps], [p ] , [kt], [tk], [db], [bd], [v ]. In this fashion, potential sonority problems that arise from the co-occurrence of clusters such as [sp] and [ps], such as in French spore vs. psaume, automatically disappear. 2.1.2 Re-Adjunction This is a favored mechanism in derivational models, where processes apply in a sequence, and the level at which they apply is a crucial element of the phonology. Under this view, any violation of SON creates a situation in which it is not possible to syllabify an initial consonant. For instance, if MSD is set at 2 in SON, then the [f] in [fne tra] or the [b] in [bke] are left stray. As a result, they are exposed to possible deletion unless some process comes to license them prosodically. While deletion is not infrequent at the right edge of a prosodic domain, it is extremely rare at left edges. 6 This rarity correlates with two observations: (1) the initial consonant of roots generally shows remarkable resistance to deletion in Romance, and (2) the prefixai stray material is protected from deletion whenever deletion would result in the obliteration of the entire morpheme; this is known as an 'exhaustivity effect'. It is thus imperative that some repair strategy apply. Epenthesis (N-creation or u-creation) is common, especially with sonorants. In the absence of epenthesis, readjunction reassociates strays to an existing node. The controversial aspects of this process involve mainly (i) the determination of the level at which strays reassociate, and whether or not they can remain unassociated for a long stretch; in some models, the derivational distance between syllable assignment and their eventual incorporation into the prosody is quite important; (ii) the determination of the node at which they readjoin, whether it is at the constituent level (here, onset), at the syllable node, or whether all readjunction involves creation of an upper node above sigma. Re-adjunction in French has been argued to correlate to phonetic factors. For instance, Rialland (1994) finds that in C1C2V- sequences co-articulation effects that occur between C1 and the vowel are much reduced when C1 is extrasyllabic: thus co-articulation effects between [k] and [i] in [kri] are important, while the same effects between [k] and [u] in [knut] are negligible. Even though this fact is not incompatible with re-adjunction to the syllable node, Rialland posits a structure in which the extrasyllabic consonant is associated to the Prosodic Word node. The clusters in [kri] and [knut] will thus be structurally different (branching onset vs. readjunction), but Rialland's model makes the claim that no structural differences exist between underlying and derived clusters which defy 6
It does occur sporadically in some high-frequency words, for instance forms like [ni] < [vni] < VENIRE in Romantsch and in Piacentino (Gorra 1890:143).
225
SONORITY AND DERIVED CLUSTERS
sonority, as in pneu [pnø] vs. recommencer [rkomäse]. The difficulties involved in buttressing structural differences by phonetic corroboration cannot be underestimated. Differences in co-articulation effects, where the nature of C2 varies, as in the [kri] and [knut] examples, are not directly relevant; perfect minimal pairs should be studied. However intuitively satisfying it may be to suppose that harmony between two consonants in the onset could be closer than that between two daughters of a higher node, no convincing phonetic evidence can be adduced. The two possibilities mentioned so far (relaxation of the scale and readjunction) can easily be combined. In Slovak, for instance, it is not the case that even with a relaxed scale, no violation occurs. Slovak has negative sonority clusters, as in [rmut], [rdest], [lkatt']. In such cases, Rubach & Booij (1992) posit readjunction. Whether or not readjunction is made to the syllable node (or) or to the prosodic word node (PW) is a matter of debate. Rubach & Booij (1990) argue that in Polish, strays readjoin to PW. Rubach (1993:250) further argues that Initial Adjunction targets σ, while Postcyclic Adjunction targets PW. Note that, even though Slovak is a language that favors syllabic resonants, initial strays, even when they are liquids, do not become syllabic in words like [rdest], [lkatt']. This opens a third possibility: that the strays may remain unsyllabified. 2.1.3 Non-Exhaustive Syllabification Rubach & Booij (1992) assume that if a segment is left unsyllabified, it will be deleted by Stray Erasure. Their framework, in which the very purpose of adjunction is to 'rescue' a segment from deletion, assumes exhaustive syllabification. Other options have been suggested: for instance Bagemihl (1991) analyzes Bella Coola as having non-exhaustive parsing (see also Downing 1993). In Bella Coola, a language with no vowelless syllables, segments may be left unsyllabified, as are [p] and [s] in (19). In fact, entire words can consist of unsyllabified segments (18). This is because segments which are unsyllabified are moraic, hence licensed. Bagemihl argues that evidence for such a mechanism is motivated by reduplication, which looks for monomoraic syllables and ignores unsyllabified segments. (18)
(19)
cr
cr
/A JA
JA
| q
JA
JA
| p
JA
| s
t
JA
| t
X
|
|
/ i
I JA JA
JA
JA
|
Ml
|
|
|
? a t p s
JA
| t u
226
JEAN-PIERRE MONTREUIL
This leaves open the theoretical possibility that the initial C in 'new' CC clusters be represented as moraic but unsyllabified. 2.1.4 Vowelless Syllables Shaw (1993) analyzes the so-called minor syllables of Mon-Khmer, which can consist of one or two consonants without a vowel. Mon-Khmer displays sesquisyllabic words, i.e., words consisting of a minor syllable followed by a major syllable, as in [cp-ci:p]. Shaw's analysis assumes exhaustive parsing, but makes a formal distinction between nuclear and non-nuclear moras (in Kamme, for instance, only the latter exist). σ
(20)
σ
(21)
/ |I
/|
||
|
||
||
I
μ
|μ
||
|| C
M
C
C
V
Shaw argues that phonological syllables can be preferentially vowelless, and that consonantal syllables are maximally binary: onset + non-nuclear mora. Three consequences arise for general theory. Shaw proposes that, universally, (1) all moras have a resonant nucleus; consequently, (2) an obstruent never functions as the nuclear mora of a heavy syllable, and (3) all obstruent-only syllables are light. Thus, the initial C in 'new' CC clusters could be represented as forming its own syllable, even in the absence of a buttressing epenthetic vowel 2.2
Analyses of Syllable Structure These four basic possibilities, combined with theoretical choices concerning levels of representation (X, μ, N, etc.) allow at least for the following options in representing a word like the Surmiran [bke], cited in Table 3. i. no sonority violation, relaxation of MSD ii. adjunction to onset iii. adjunction to syllable These three options are illustrated in (22-24), using a constituent framework with Onset and cr as the relevant nodes.
SONORITY AND DERIVED CLUSTERS
227
We have already cited evidence that points toward (24) as the preferred form. While (23) differs from (22) only in a purely derivational sense, (24) directly relates phonogical differences discussed in § 1.1. to a structural difference, i.e., the level of reattachment of the stray consonant. The general correlation is what motivates any configurational analysis: there exists greater cohesion between two elements represented as daughters of a lower node. To the extent that [bk] is viewed as more separable a cluster than [bl] in Surmiran, the distinction between (24) and (22)-(23) is well-motivated. Moraic Theory posits no intervening node between onset segments and syllable nodes, and, as a result, all three structures are represented as in (25). The claim that Moraic Theory makes is that there is no empirical evidence related to such structural differences. (25)
σ / / | i ! JA !
|
|
b k e
To represent different degrees of separability, Moraic Theory uses different contrasts: appendix status (26), non-exhaustive syllabification (27), vowelless syllables (28). (26)
PW
(27)
(28)
228
JEAN-PIERRE MONTREUIL
It appears that (26) represents the appropriate structure for most derived clusters, that (27) should be reserved for negative clusters beginning with a sonorant (see §23), and that (28) should not be retained within the context of Romance. 2.3
Sonorant-Initial Clusters The data presented thus far can be summarized as in Table 8, which shows that all types of positive clusters are attested in all languages examined. The fact that there are gaps in negative and zero clusters turns out to be deceiving for two reasons. Sometimes, epenthesis comes in to bleed the occurrence of some negative clusters. For instance, a form like [rve], appearing in Table 4 under -2, would not have been considered as having an initial cluster if epenthesis had applied, giving [arve]. Yet the role of epenthesis as a low-level therapy frequently applied to allow for the syllabification of strays is well-documented. To that extent, if some sonorants are considered as strays prior to epenthesis, and the initial cluster to which they belong is included in the Tables, then all boxes would be filled. -3 Bravuogn Comelico Lombardia Piemonte Romagna
V
-2
V V V
-1
0
+1
+2
+3
V
V V V V V
V V V V V
V V V V V
V V V V
V V V
Table 8. Summary of Tables 3-7
v
The therapeutic character of epenthesis correlates to its status as a phrasal rule: it does not need to apply if the preceding word (in the proper prosodic domain) ends in a vowel. Dictionaries, as they list words in isolation, are not always consistent in choosing whether the main entry of a form should be its bare form or its epenthetic form, e.g. rve or arve. The occurrence of epenthesis is subject to much variability. In Lombardia, for example, stray liquids surface without a vowel, but in Piacentino, [a]-epenthesis is frequent: [alka] 'Ital. leccare', [arlikja] 'Ital. reliquia'. It also occurs, but less frcquently, before stray nasals: [anta] 'Ital. nettare'. In addition, epenthesis is not always restricted to sonorants: in Romagnolo, for instance, it occurs before other stray consonants; thus, along with armor 'Ital. rumore', arvena 'Ital. rovina', amstir 'Ital. mestiere', one encounters forms like [avtur] 'Ital. vetture', [avsti] 'Ital. vestire'. This epenthetic vowel, transcribed here simply as [a], is often described as a 'dark [a]', suggesting schwa-like coloring.
229
SONORITY AND DERIVED CLUSTERS
Yet, it remains true that the occurrence of epenthesis correlates statistically with the high sonority of C1 in a cluster. In absolute terms, it is more frequent before liquids than before any other segment. In relative terms, it is more frequent in -3 clusters than in any other cluster. This can be represented functionally as a strategy to reduce markedness and structurally as the syllabification of stray moras. To the extent that sonorants tend to behave differently from obstruents in their ability to remain strays, they should be analyzed as moralc. If it is correct to retain (27) as the proper representation of sonorant + C clusters, epenthesis can be interpreted as the tendency to resolve (27), a marked structure, into an unmarked bisyllabic form. stray nasals
epenthesis
Bravuogn ComeHco Gardena Lombardia Piemonte Romagna
epenthesis
stray liquids
9
V
V V V V V
a
V
a a
Table 9. Epenthesis in sonorant + C clusters
3.
The OT Perspective Optimality Theory (OT) evaluates constraints on output candidates to select the optimal candidate. Lack of phonological activity is viewed as input-output (IO) faithfulness, and any modification affecting the input incurs a violation of MAX (= No deletion) or, crucially in our case, DEP (= No epenthesis). Another constraint, *COMPLEX, favors single consonants over clusters (or non-branching over branching). To our concerns, *COMPLEX means: Onsets may not branch. 3.1
The Indeterminacy of the Input Consider a hypothetical input /pla/. In a language in which it corresponds to output [pela], *COMPLEX would outrank DEP. But Raeto-Romance and GalloItalic languages prefer the opposite order. In all the languages we have referred to here, it is more important to be faithful than to avoid clusters; thus the order is DEP » *COMPLEX, as shown in (29). (29) input /pla/ [pela]
[pla]
DEP
*COMPLEX
*! *
230
JEAN-PIERRE MONTREUIL
What if the input were /pela/? From input /pela/ to output [pla], both faithfulness (here, MAX) and *COMPLEX are violated, and [pla] could not be retained as the optimal candidate. This clearly shows that another, more powerful constraint is at play. We will assume here that the constraint that makes [pla] preferable to [pela] is not exclusively concerned with syllable economy, but also with prosody, in the sense that it selects one specific way of making words shorter by targeting exclusively pre-stress vowels. It can best be viewed as a constraint that limits the occurrence of some prosodically unlicensed elements. This approach is reminiscent of the analysis proposed by Hammond (1997) to account for syncope in English. Hammond shows that syncope of stressless vowels in English can be accounted for most accurately through the interaction of faithfulness with a constraint *FOOTLESS which rules out unstressed syllables. In a variety of English which prefers [prejd] to [pәrejd] (for parade, at least in certain styles), it can be said, in Hammond's terms, that "the pressure to foot syllables exceeds the pressure to pronounce some of them". In our case, not all unstressed syllables, but only initial unstressed syllables are affected by syncope. To keep matters simple,7 we will call this constraint SYNCOPE and formulate it informally as: SYNCOPE: No initial unstressed syllables. A language where syncope occurs to turn /pela/ into [pla] shows the higher ranking of SYNCOPE with respect to the other constraints. Raeto-Romance and Gallo-Italic languages follow this general constraint ordering. When the grammar is organized in this fashion, with the order SYNCOPE » DEP, MAX » *COMPLEX, the same output is generated from two different inputs, as shown in (30). The input could remain indeterminate.
(30)
input /pela/ [pela]
[pla] 7
SYNCOPE
MAX
DEP
*COMPLEX
|
*! *
*
This is a simplification: indeed, there are cases where an unstressed vowel is left before the main stress even after syncope has occurred. These are all words whose input can be represented as showing two unstressed vowels before the main stress. For instance: Comelicano: [du'grenti] / [dgu'renti], [vli'mentu]; Lombardo (and Piacentino): [bso'nava], [kla'zion], [fra'meint], [pfi'lar], [rræs'tat], [vri'tæ]; Piemontese: [bzu'ne], [dli'ka], [djun'or], [dven'te], [fna'rø], [kman'ze], [tka'li], [tvu'grai]; Romagnolo: [tla'rana], [kminzi'pie]. In the first example, the form *[dgrenti], which does not seem to occur, needs to be blocked out as a possible output. Due to space limitations, this avenue of research is not explored here, but it is clear that a more complete account should account for variability in deletion patterns: [dugrenti] / [dgurenti], but not *[dgrenti]; as well as [vlimentu], but not *[vilmentu].
SONORITY AND DERIVED CLUSTERS
input /pla/
MAX
SYNCOPE
j
*!
[pela]
231
PEP
*COMPLEX
j
* *
[pla]
3.2
The Parameterization of Sonority The complexity of the situation described earlier stems from the fact that some clusters are tolerated while other are not, which means that *COMPLEX has to be divided into smaller constraints. This division follows sonority lines, implicationally and transitively. Accordingly, constraints must be formulated in such a way as to respect the degrees of sonority given in (2). This is shown in (31). (31)
*SON+l means: "No clusters whose sonority index is +1 or less" *SON 0 means: "No clusters whose sonority index is 0 or less" *SON-1 means: "No clusters whose sonority index is -1 or less" *SON-2 means: "No clusters whose sonority index is -2 or less" *SON-3 means: "No clusters whose sonority index is -3 or less"
In this fashion, a hypothetical language U which obeys ROMANCE SONORITY would represent an ordering where *SON+1 dominates SYNCOPE, since +3 clusters are acceptable and +1 clusters are out. (32) Language U: outputs [pla], but [pesa] input /pela/
._*SON +
SYNCOPE
MAX
DEP
*
* *SON+l
SYNCOPE
MAX
[pela]
DEP
* [*SON+l
input /psa/
MAX
DEP
*COMPLEX
DEP
*COMPLEX
*
*! *SON+l
SYNCOPE
*
[pesa]
[psa]___
SYNCOPE
*
[pesa] [psa]
*COMPLEX
*
*!
[pla] input /pesa/
•COMPLEX
*!
[pela] |[pla] input /pla/
|
*!
MAX
* * *
1
232
JEAN-PIERRE MONTREUIL
In hypothetical language V, where +1 clusters are in, but plateaus are out, SYNCOPE outranks *SON+l, but not *SON 0. It obeys the constraint ordering shown in (33). (33) Language V: outputs [pla] and [psa] but [petal input /pesa/
[ *SON0
SYNCOPE
[pesa]
[psa] hinput /psa/
*SON0
MAX [ DEP
SYNCOPE
*
*
*COMPLEX
*SON+l
MAX
* i DEP
*
[psa] input /peta/
|
*SÖN0
SYNCOPE
*COMPLEX
*
*!
[pesa]
*
*SQN+1
MAX
*
*
*SON+l
MAX
DEP
»COMPLEX
DEP
_.*ÇOMPLEX
*
[petal [pta]
|
*!
input /pta/
|
*SON0
SYNCOPE
*
*
[petal
i
*SON+l
*\
[pta]
* *
*!
*
Language W, in which SYNCOPE outranks *SON 0, but not *SON-l, is shown in (34). (34) Language W: outputs [pla], [psa] and [pta], but [veta] input /peta/
L *SQN-1
SYNCOPE
[pta] input /pta/
| *SON-l
SYNCOPE
*
*SON0
MAX
•COMPLEX
•
DEP
| [ *SON-I
SYNCOPE
*COMPLEX
* *
[pta]
•SONO
* MAX \
DEP
*COMPLEX j
*
[veta]
*
*!
[vta] j
•SON-1
SYNCOPE
•SONO
*
[veta]
[vta]
*
DEP
*!
[peta]
input /vta/
MAX
*!
[peta]
nput/veta/
*SON0
*!
*
•
*
i
MAX i
! |
DEP
*COMPLEX
* *
SONORITY AND DERIVED CLUSTERS
233
In language X, SYNCOPE outranks * s O N - l , but not *sON-2. (35) Language X : outputs [p]a], [psa], [pta], [vta], but resa] input /veta/
*SON-2
SYNCOPE
[vta] input /vta/
|
*SON-2
SYNCOPE
*
*
*SON-l
MAX
DEP
•COMPLEX
* DEP
•COMPLEX
* *
[[vta] ]
*SON-2
SYNCOPE
*
*SON-1
MAX
*
*
•SON-1
MAX
*
[resa]
*!
Ursa] input/rsa/
MAX
*!
[veta]
input /resa/
•SON-1
*!
[veta]
|
•SQN-2
SYNCOPE
PEP
| * DEP
*
[resa]
[rsa]
*COMPLEX
* *
*!
*COMPLEX
*
j
In language Y, SYNCOPE outranks * SON-2, but not *SON~3. (36) Language Y : outputs [pl a], [psa], [pta |, [vta] and [rsal, but [reta] input /resa/
1 *SON-3
[resa]
SYNCOPE
L*SONj3_
SYNCOPE
[rsa] input /reta/
*SON-3_
*
*
• SON-2
MAX DEP *
*COMPLEX
*
j
•
*!
[resa]
SYNCOPE __*SON^_j
MAX
DEP_ •COMPLEX
*
[reta]
•!
|
*SON-3_
SYNCOPE
"
*
•
•SON-2
MAX
•
. D E P _ •COMPLEX
*
[reta]
1|
*;
•COMPLEX]
| *
input /rsa/
[rta]
MAX j PEP
•!
[rsa]
input /rta/
•SON-2
* *
j
Finally, in language Z, SYNCOPE outranks even *SON-3.
•
234
JEAN-PIERRE MONTREUIL
(37) Language Z: outputs tola], [psa], [pta] [vta] [rsa] and [rta] 1 SYNCOPE
input /reta/
J [rta] input /rta/ [reta] [rta]
*SON-3
MAX j
DEP
*CQMPLEX
DEP
*COMPLEX
*!
[reta] \
SYNCOPE
*
*
*SON-3
MAX
*
*!
* *
|
*
On the basis of the limited data summarized in Table 8, which again did not include some NC» clusters or sC~ clusters, we arrive at a typology based on ordering, within which our Raeto-Romance and Gallo-Italic languages take their place. • Language U typifies Italian or Spanish. It does not correspond to any of the Raeto-Romance and Gallo-Italic languages, since all of them allow at least clusters like [ps]. • Neither do Language V and Language W, since they all allow -1 clusters.8 • Language X typifies Bravuogn Romansch and Romagnolo. • Language Y typifies Lombardo and Piemontese. • Language Z typifies Comelicano. The fact that Raeto-Romance and Gallo-Italic dialects are accurately characterized in this fashion shows that the notion of sonority differential can be an accurate predictor of cluster well-formedness. There is no need, in this particular case, to appeal to featural content or even to absolute sonority. Only the differential, i. e., the relative sonority of two segments, matters. 4.
Conclusion The consonant clusters of Standard Italian have been well documented and analyzed, and it is a measure of the conservative character of Italian phonotactics that initial clusters do not diverge very far from their Latin ancestors. In contrast, the five Romance dialects studied in this limited survey display singular phonotactic patterns, which invite more comparisons with Gallo-Romance than with Italo-Romance. 8 However, Language V would have been instantiated had our database extended to other, more conservative varieties of Raeto-Romance, like Gardenese, for instance (Bammesberger 1974). If, on the other hand, we had included in our data NC clusters derived from VNC, which abound in the language, then even Gardenese would pattern after Language Y.
SONORITY AND DERIVED CLUSTERS
235
The properties of derived clusters have been shown to differ widely from those of underlying clusters. In analyzing these properties, two concerns have been addressed. (Î) The first concern has to do with structure, and specifically with showing the motivation behind a structural representation favoring readjunction. Several theoretical approaches were discussed, and even though in each case arguments were given as to why a particular structural representation should be preferred over another, the fact remains that such arguments are more dependent upon theoretical preassumptions than dictated by phonetic realizations. (2) The second concern has to do with the concept of sonority differential and leads to the following question: when structural differences are put aside and only surface forms are being considered, is it necessary to appeal to various aspects of melodic information (including absolute sonority), or is it sufficient to compare the relative sonorities of the two members of a cluster? The answer that emerges does not rule out the possibility that additional featural information may be required, but it does suggest that the concept of sonority differential can elicit useful classifications and accurate characterization, indicating that, in derived clusters, sonority is very much - and possibly foremost - a matter of contrast REFERENCES Baertsch, Karen. 1998. "Onset Sonority Distance Constraints Through Local Conjunction". Chicago Linguistic Society 34.1-15. Bagemihl, Bruce. 1991. "Syllable Structure in Bella Coola". Linguistic Inquiry 22.589-646. Bammesberger, Alfred. 1974. "Le parier ladin dolomitique du Val Gardena". Bulletin des jeunes romanistes 20.5-74. Battisti, Carlo. 1988 [1914-1921 ]. Testi dialettali italiani in trascrizione fonetica. Bologna: Arnaldo Forai. Bearesi, Luigi. 1982. Piccolo dizionario del dialetto piacentino. Piacenza: Libreria editrice Berti. Bruni, Francesco, ed. 1992. Vitaliano nelle regioni. Torino: UTET. Carnie, Andrew. 1994. "Whence Sonority? Evidence from Epenthesis in Modern Irish". MIT Working Papers in Linguistics 21.81-108. Clements, George. 1990. "The Role of the Sonority Cycle in Core Syllabification". Papers in Laboratory Phonology 1.283-334. New York: Cambridge University Press. Cortelazzo, Manlio & Carla Marcato. 1992. Dizionario etimologico dei dialetti italiani. Torino: UTET.
236
JEAN-PIERRE MONTREUIL
de Lorenzo Tobolo, Elia. 1977. Dizionario de dialetto ladino di Comelico Superiore Bologna: Tamari. Downing, Laura J. 1993. "Unsyllabified Vowels in Aranda". Chicago Linguistic Society vol. 29 ed. by Katharine Beals, et al., 171-187. Chicago: Chicago Linguistic Society. Eckman, Fred & Gregory K. Iverson. 1993. "Sonority and Markedness among Onset Clusters in the Interlanguage of ESL Learners". Second Language Research 9.234-252. Fortina, Peppo, et al. 1992. Vocabolario del dialetto di Oleggio. Torino: Edizione dell'Orso. Gorra, Egidio. 1890. "Fonetica del dialetto di Piacenza". Zeitschrift für Romanische Philologie 14.113-159. Hammond, Michael. 1997. "Optimality Theory and Prosody". Optimality Theory: An overview ed. by Diana Archangeli & Terence Langedoen, 33-59. Oxford: Blackwell. van der Hulst, Harry & Nancy Ritter, eds. 1998. The Syllable: Views and Facts. Berlin: Mouton De Gruyter. Jespersen, Otto. 1904. Lehrbuch der Phonetik. Leipzig & Berlin. Lauefer, Christiane. 1991. "Syllabification and Resyllabification in French". New Analyses in Romance Linguistics ed. by Dieter Wanner & Douglas Kibbee, 19-36. Amsterdam & Philadelphia: John Benjamins. Lekach, A. Farmer. 1979. "Phonological Markedness and the Sonority Hierarchy". Papers in Syllable Structure, Metrical Structure and Harmony Processes: MIT Working Papers in Linguistics 1. Lutta, C. Martin. 1923. "Der Dialekt von Bergün und seine Stellung innerhalb der Rätoromanischen Mundarten Graubündens". Beihefte zur Zeitschrift für Romanische Philologie 71.1 -356. Mattioli, Antonio. 1986 [1879]. Vocabolario romagnolo-italiano. Bologna: Arnaldo Forni. Morelli, Frida. 1998. "Markedness Relations and Implicational Universals in the Typology of Onset Obstruent Clusters". ROA-251 (Rutgers Optimality Archive http://ruccs.rutgers.edu/roa/html). Padgett, Jaye. 1991. Stricture in Feature Geometry. Ph.D. dissertation, University of Massachusetts, Amherst. Plangg, Guntram. 1973. Sprachgestalt als Folge und Fügung. Zur Phonologie des Dolomitenladinischen (Badiot) und seiner Nachbarn. Tübingen: Max Niemeyer. Rialland, Annie. 1994. "The Phonology and Phonetics of Extrasyllabicity in French". Phonological Structure and Phonetic Form, Papers in Laboratory
SONORITY AND DERIVED CLUSTERS
237
Phonology (vol. 3) ed. by Patricia Keating, 136-160. Cambridge: Cambridge University Press. Rice, Keren. 1992. "On Deriving Sonority: A structural account of sonority relationships". Phonology 9.61-99. —. 1993. "A Re-examination of the Feature [sonorant]: the status of sonorant obstruents". Language 69.308-344. Rubach, Jerzy. 1993. The Lexical Phonology of Slovak. Oxford: Oxford University Press. Rubach, Jerzy & Geert Booij. 1992. "Resolutions of Extrasyllabicity in Slovak". Linguistics 30.699-729. Selkirk, Elizabeth. 1982. "The Syllable". The Structure of Phonological Representations (vol. 2) ed. by Harry van der Hulst & Norval Smith, 337-383. Dordrecht: Foris. . 1984. "On the Major Class Features and Syllable Theory". Language Sound Structure ed. by Mark Aronoff & Richard Oehrle, 107-137. Cambridge, Mass.: The MIT Press. Shaw, Patricia. 1993. "The Prosodic Constituency of Minor Syllables". West Coast Conference on Formal Linguistics 12.117-132. Steriade, Donca. 1994. "Complex Onsets as Single Segments: the Mazateco pattern". Perspectives in Phonology ed. by Jennifer Cole & Charles Kisseberth, 203-293. Stanford: CSU. Tagliavini, Carlo. 1926. "Il dialetto di Comèlico". Archivum Romanicum 10.1200. Truckenbrodt, Hubert. 1994. "The OCP, Underspecification and Articulatory Faithfulness". MIT Working Papers in Linguistics 21.227-275. Yip, Moira. 1989. "Feature Geometry and Co-Occurrence Restrictions". Phonology 6.349-374.
STRESS AND SCHWA IN FAETAR
NAOMI NAGY University of New Hampshire
0.
Introduction Faetar is a Francoprovençal dialect which is spoken in two small southern Italian villages, Faeto and Celle, located in the mountains on the western border of Apulia in the province of Foggia. They were settled approximately 600 years ago by emigrants from what is now southeastern France. Due to contact with Italian, Faetar has adapted many properties of Italian (Nagy 1996). However, one way in which it remains distinct from standard Italian is that the vowel schwa frequently appears in unstressed syllables. The phonemic status of schwa structure is unclear. This paper explores three possible accounts of the of Faetar which are intertwined with the status of schwa. The first, the Marked Stress Hypothesis (§3.1), is that primary word stress prosodic is marked in the lexicon. In that case, vowel quality (full vs. reduced vowel) is predictable: tonic vowels are never schwa, post-tonic vowels categorically reduce to schwa, and pretonic vowels (variably) reduce to schwa. Under this analysis, schwa is not phonemic. The second possibility is the Marked Quantity Hypothesis (§3.2): vowels are lexically marked as long or short (bi- or monomoraic). Under this hypothesis, we can observe that stress appears on the rightmost long (bimoraic) vowel, and short (monomoraic) vowels may reduce to schwa. The lack of a moraic difference between vowel and schwa in Faetar will be shown by the existence of Raddoppiamento Sintattico in Faetar. The Marked Quality Hypothesis (§3.3), that vowel quality is lexically marked (schwa as a mid-central vowel vs. other vowels), is one other possible alternative. In that case, the location of word stress is fully predictable: it always appears on the rightmost non-schwa vowel Schwa, then would be phonemically distinct from other vowels: the mid-central vowel would be a phoneme. The
240
NAOMI NAGY
word-position distribution of vowels undermines this hypothesis: if schwa were a phoneme, why would it be the only vowel with the privilege of appearing post-tonically? Like many nonstandard languages and dialects, Faetar has no systematic orthography, long-term historical record, or native speakers trained as linguists. This eliminates many of the tools often used by linguists to determine underlying structure. Therefore, other methods are required to choose among these three hypotheses (§2). I present several arguments to support the first hypothesis, and contradict the latter two (§3). Before presenting these arguments, I provide necessary background information about Faetar and the languages with which it is in contact (§ 1). 1. 1.1
Background Phonological Information Faetar Consonants and Vowels The vowels and consonants shown in (la) and (2a) exist in Faetar ((lb) and (2b) are discussed below). The shading indicates distributional differences between Faetar and standard Italian: phones which are shaded exist in Faetar but are not Italian phonemes, and empty shaded boxes indicate phonemes which exist in Italian but not Faetar (though the sounds may exist as allophones of other phonemes). Faetar has five tense vowel phonemes, shown between slashes. The four lax vowels of Faetar, shown in square brackets, are not phonemes, but allophones of the corresponding tense vowels. They appear optionally in closed syllables, in free variation with the tense forms, but they never appear in open syllables (Hoffman 1968:13). As the phonemic status of schwa is at issue, it appears here in parentheses. Schwa is distinct from the four lax vowels in that it appears in both open and closed syllables, but never in stressed syllables. a. Faetar vowels (Nagy 1996:1190 Front Mid High /i/ [I] Mid Id [ε] (e) Low /a/
b. Apulian vowels (Valente 1975:12)1 Front Mid Back
(1)
Back [ü] /u/ [c] /o/
High Mid Low
i(j) e s
d
c
u(w) o
a
* Although Valente (1978) provides a non-phonemic source for schwa, he lists it as a phoneme in Apulian. It is unclear which phones are phonemic in his analysis, thus I have not distinguished between phonemes and allophones in the Apulian charts.
STRESS AND SCHWA IN FAETAR
241
The phonemic inventory of Faetar consonants is shown in (2a) (Kattenbusch 1982:151-168). Additionally, Faetar has phonemic geminate consonants word-internally (Nagy 1994). There are contrastive geminates for all the consonants except /n/, /y/, /v/ and the affricates. (2)
Stop Fricative Affricate Nasal Liquid Trill
a. Faetar consonants (Nagy 1996:116) Labial Labio- Dental Palatal Velar dental p b t d k g f v s ƒ ts t d3 m
n
n
1
y
r
b. Apulian consonants2 (Valente 1975:12) Labial Labio- Dental Palatal Velardental t d c f k g P b f V s z ƒ ts dz n m 1 r
1.2
Faetar Stress Stress always appears on the rightmost non-schwa vowel of a word. The stressed syllable is always one of the last three syllables in the word. However, stress is not on a fixed syllable, counting from the left or from the right, nor is it consistently on either an odd or an even syllable. Furthermore, stress may be on either an open or closed syllable, as shown in (3). (3) Possible stressed syllable positionsWord From end From start [bro.kala] antepenult 1st penult 1st [Tra.ra] [am.'inap.pafe] antepenult 2nd [kut.'tejs] penult 2nd penult [am.bu.'lan.za] 3rd penult [am.mattuf.'fli.jg] 4th [am.ma.tün/ta] final 4th [am.mun.tu.nam.'men] final 5th
Italian forchetta fratello caspita cotello ambulanza accumulare ammaccare ammasso
'fork' 'brother' 'interjection' 'knife' ' ambulance' 'to accumulate' 'to bruise' 'accumulation'
2 Valente (1975) does not include a palatal liquid in the inventory but uses it in a number of words. 3 Transcriptions throughout are broadly phonetic and make use of certain allophones which are not part of the phonemic inventory. For example, /s/-->[z]/[+vd, +cons] in Faetar.
242
NAOMI NAGY
In Faetar, any vowel except schwa may be stressed, and any vowel may appear in an unstressed syllable, as shown in (4). (4) Possible stressed and unstressed vowels
i
Stressed Word [a'vi]
I
[djә 'linnә]
oggi gall(in)o
Unstressed Word
Italian 'today'
[tfin'nl] [rәmir'ri]
'rooster'
Italian sfacciato
'saucy'
guardato
'looked'
u
[a'lunnә]
alunno
'student'
[kut'tejә]
cotello
'knife'
u
[mokkә]
mucca
'cow'
[ammatün'ta]
ammaccare
'to bruise'
e
[kut'tejә]
cotello
'knife'
[vekkar'dun]
vecchio
'old man'
s
[bu'fεttә]
tavola
'table'
[allεr'dji]
allergia
'allergy'
o
[hrokәlә] [bal'dcrjә]
forchetta gaiezza
'fork'
[kontftfettә]
Concetta
'gaiety'
[cr'majә]
ormai
'Concetta' 'henceforward'
a
[frarә]
fratello
'brother'
[a'vi]
9
—
oggi gall(in)o
'rooster'
0
[dg3 linnә]
'today'
These facts can be accounted for in a straightforward manner. One binary foot, a generalized trochee (Kager 1993), is built at the right edge of the word. 4 Some words have one extrametrical syllable at their right edge, but no more than one. Thus the possible structures for the right edge of Faetar words are shown in (5). (5) a. Antepenult stress F
b. Penult stress F
/\
| σ
cr cr
|||
/\
||
μw |||
V
F
/\
σ σ <σ> μs
c. Final stress
e
μ 9
μs
μw
|| V e
μs | V
μw
(5) shows only the assignment of vowels to syllable nuclei. The strong (left) branch of any foot may have a coda consonant, the weak branch may not. The limitations of one extrametrical syllable and a binary-branching foot restrict stress to one of the rightmost three syllables of the word.
4 The absence of other feet in the word is indicated by the lack of syllable-by-syllable alternation between full vowels and schwa in the pretonic syllables.
STRESS AND SCHWA IN FAETAR
243
Nagy & Reynolds (1997:44-5) show that the variable process of wordfinal deletion (of segments and syllables) is more likely to occur in Faetar words with antepenultimate stress than in words with penultimate or final stress. This can now be explained as deletion of unfooted segments/syllables, a less costly deletion than for footed segments/syllables, supporting the analysis that extrametrical syllables exist in words with antepenultimate stress. Such extrametrical vowels reduce to schwa as an intermediary step in the deletion process. Similarly, pretonic vowels, all unfooted, may reduce to schwa. An unassociated mora appears in the structure for (all?) final stressed words, which satisfies the requirement that a trochee be binary-branching. The source of the unassociated mora may be from analogy with Italian, or it may exist for historical reasons. Faetar exhibits much variable word-final deletion, a synchronic symptom of the diachronic process of increasing apocope mentioned above. Final unassociated moras may be remnants of word-final segments present at an earlier stage. Apulian Phonology While distinctions between Faetar and standard Italian have already been highlighted, the relevant comparison is to the local Apulian dialect, as this is the variety with which Faetar speakers have regular contact. The principal source used for this description of Apulian is Valente (1975), and his findings are confirmed by recordings of speakers from towns neighboring Faeto (Nagy 1996). Valente (1975:64) writes, "Apulia represents a unified and continuous linguistic area that has its center of gravity and of dissemination in the region of Bari [my translation]." The most distinctive trait of the dialect is that Latin A in both stressed and final syllables is intact, that is, not reduced (Valente 1975:54). All other unstressed vowels reduce to [9] (Valente 1975:36). In Apulian, stress is always on one of the last three syllables (Valente 1975:29). From the data available, it seems to have the same stress system as standard Italian. The phonemic inventories of Apulian, are given above in (lb) and (2b). Shading in these two figures indicates differences between the inventories of Apulian and Faetar. 1.3
1.4 Raddoppiamento Sintattico Like many southern Italian dialects, Faetar exhibits a process of wordinitial consonant lengthening following a word-final stressed vowel, known as Raddoppiamento Sintattico (RS). I adopt Repetti's (1991:307) analysis of RS: "raddoppiamento is caused by an empty mora which is present in the underlying representation (not added by a rule) ... [the mora] is now filled by spreading
244
NAOMÏNAGY
from the initial consonant of the following word." If there is an empty mora that needs to be associated to an element in the following word, then the vowel in that syllable can be linked to only one mora to begin with. That is, the structure of a word-final stressed open syllable is that shown in (5c). As in Italian, this provides the necessary structure for RS to apply. In §2.3, acoustic evidence that RS does occur in Faetar is presented. I make use of the presence of RS to illustrate that the Marked Stress Hypothesis is viable and the Marked Quantity and Marked Quality Hypotheses are not. 2.
Empirical Evidence ofRS in Faetar In this section, I first illustrate the existence of historical sources for RS in Faetar, an important point since Gallic Francoprovençal is not documented to exhibit RS. I then present data showing that Faetar speakers do exhibit RS. This fact supports the prosodic analysis presented above. 2.1 Historical Source for RS Jaberg & Jud's (1928-40) transcriptions provide evidence that lexically triggered RS occurs in the Apulian dialects surrounding Faeto. (6) shows forms elicited by fieldworkers for several phrases which contain words that trigger RS (Bertinetto 1985:612, Dulcibella 1934, Grandgent 1927). These forms show a strong tendency to be transcribed with a double initial consonant in most southern Italian data points. I have chosen three representative points for comparison: the two towns closest to Faeto and the closest city. The data from Napoli, the next closest city, show RS applying categorically in these words. (6) RS in the neighboring towns and closest cities (Jaberg & Jud 1928-40) Entry Italian Colle Sannita Ascoli Satriano Bari 519 va a cacccia [va a kkat ] [vat e a kkat ] [va:j a kkat ] 647 a dormire [addor'mi] [a ddcrme] [a kkclkese] 1584 a scegliere [jt e kka'pa] [a kka'pa] [a kka'pa] 646 è andata [s ε ggju:t] [e ju:t] [e u:t] 668 èvestito [s εvv estu:te] [e vәstu:tә] [s a vvәsturte] 1340 èbuono [e bbo:nә] [je bbu:nә] [je bbwe:nә] 1602 è cattivo [e mmalә'mεntә] [je kkat'ti:vә] [je ffәεnt]
'go hunting' 'to sleep' 'to choose' 'went' 'was dressed' 'is good' 'is bad'
The phrases in (7) give more recent evidence of lexically triggered RS in Apulian. They are extracted from narratives recorded in three southern locales and transcribed in Valente (1975:66-69). The list includes phrases which have
STRESS AND SCHWA IN FAETAR
245
contexts for RS, but includes some instances where the lengthening did not occur, or at least was not transcribed. (7) RS patterns in near-by dialects (Valente 1975:66-69) San Marco
MS? Yes
Foggia [e rәsortә]
'(he) came out'
[e kki]
'and that' 'and who'
Yes
[e kuan]
'and when'
No
[e llsvә]
' and lift'
Yes
'and said'
Yes
[e lu] [ke fannә]
No 'and the' 'what (he) does' No
[e ffatgfvә ] Bari
[pә ffa]
'(for) to do'
Yes
[a ffa] [a ppezzә]
' (for) to do' 'to pieces'
Yes Yes
[e kkuddә]
RS? No
[a da]
'to give'
No
[a pprimә] [e ffasulә]
'for first'
Yes
'and beans'
[e pәsiәddә]
'and peas'
Yes No
[e 11 amm]
'and to them give'
Yes
2.2
Design of Experiment The experiment described in this section forms part of a corpus of material collected by the author during three summers of fieldwork in Faeto, during which interviews, direct elicitation, and quantitative analysis techniques were developed and used. The full corpus consists of approximately 50 hours of recorded speech from more than 100 speakers, recorded between 1992 and 1994 The relevant portion of the corpus, produced in order to have a directly comparable quantity of speech data from many different speakers, was gathered by recording 60 speakers, ranging in age from six to 87 years. Speakers were asked to name objects pictured in a children's book (Amery & Cartwright 1987) and then describe the scenes pictured on ten pages. This allowed for the elicitation of a careful pronunciation of each word in isolation and then, ideally, several repetitions of the word in sentence context.5 5 In conducting this experiment, it was necessary to choose between recording naturallyoccurring speech and eliciting carefully-selected isolated sentences. There are advantages to both methods, but I believe the importance of analyzing language the way people actually speak it outweighs the advantages to be gained by a strictly controlled lab experiment. The variety of words and contexts used here is meant to cancel out any biasing effects of particular independent variables. Furthermore, in a non-written language it is difficult to design a traditional cued experiment as oral cues or a written representation of the language could influence the subjects' speech. Transcription may also confuse the issue due to an ongoing conflict in Faeto: proponents of developing a Faetar orthography are at odds about whether the orthography should represent the true phonetic status of the language or show its etymological relations. Finally, there are Faetar speakers who do not read at all and who do not speak any language except Faetar, making a sentence elicitation task with natural-sounding data well nigh possible.
246
NAOMI NAGY
For this experiment, the durations of word-initial consonants in various contexts were examined in the speech of one representative speaker: a 29 year old woman recorded in 1994. 87 segments were measured: all of the clearly recorded instances of word-initial /m/ and If/ in the storybook recording. After transcribing the speech in full, transcriptions were searched for all appearances of word-initial /m/ and /f/.6 A list of the phrases examined and the relevant consonant duration for each appear in the appendix. Recordings were digitized and the duration of the initial segments measured from on-screen spectrograms using Xwaves on a Unix Sparcstation in Mark Liberman's lab. The measurements were grouped according to the class of the preceding segment: consonant (C), full vowel (V), or schwa (9). 2.3
Results of Experiment There is a significant effect of preceding segment. Table 1 and Figure 1 show that word-initial consonants are longer following V than following [ә] and C. (The two graphs show durations for the two target segments examined.) Statistical significance was determined by Kruskal-Wallis One-Way Anova, a non-parametric test which assumes neither normal distributions nor equivalent variances for the samples being compared, and which can operate on small samples, necessary assumptions for this analysis. The statistical analysis was carried out in Systat 6 with the excellent assistance of Suzanne Mitchell. These data illustrate two points. The first is that RS occurs in Faetar. This is shown by the difference between length of word-initial consonants which follow RS-triggers (V) and those following non-triggers (C, [9]). The second is that V triggers RS and [9] does not, indicating a difference in prosodic structure between syllables with schwa and those with other vowels. Target Trigger segment segment C /f/
3
V C /m/
3
V
N 5
Average duration
17 20 6 8
Standard deviation
Significance of difference from V
0.07
0.03
0.07
0.03
p = 0.089 p= 0.003 (sig.)
0.11
0.07
Total N for /f/ =42
0.09 0.06 0.12
0.12 0.06 0.16
p = 0.087 p= 0.006 (sig.)
Total N for/m/ = 45 31 Table 1. Durations according to environment (in seconds) 6
/m/ and /f/ were chosen because they are easy to spot in spectrograms and were plentiful in the transcriptions.
STRESS AND SCHWA IN FAETAR
247
Preceding segment Figure 1. Average duration according to different preceding segments
2.4
Discussion of Experiment The difference in average duration between V and [ә] is significant (at the p = 0.05 level) for both target consonants examined, indicating that V triggers RS and [9] does not. The difference in average duration between V and C is significant only at the p = 0.1 level, due to the small number of consonant-final words in the corpus. However, even in these cases, the trend is in the necessary direction to support the hypotheses. These data indicate that RS occurs in Faetar following stressed vowels but not following schwa. This is explained as the initial consonant linking to an empty mora in the final stressed syllable of the preceding word (Repetti 1991). As syllables can contain no more than two moras, only one mora may underlyingly be linked to a stressed vowel, so there cannot be a difference in association of moras to stressed vowels and schwa (as shown in (5)). Since RS does not apply when there is a word-final schwa, the difference in stressed and unstressed syllable and foot structure shown in (5) is motivated. Therefore,
248
NAOMI NAGY
stress is marked in the lexicon, and vowel quality/quantity need not be marked in order to account for Faetar's patterns. 3.
The Problem Unlike standard Italian, Faetar has many words in which [ә] appears word-internally (and word-finally). The question arises as to whether the segment has phonemic status or is the reduced allophone of some other vowel(s) in unstressed position. This issue is crucially linked to the analysis of stress. There are three ways to think about the interaction of stress and schwa in Faetar. 3.1
The Marked Stress Hypothesis The first hypothesis proposes that stress is lexically marked and vowel length is not. Specifically, certain words have an extrametrical final syllable marked in the lexicon. A binary-branching trochee is built at the right edge of each word, as shown in (5). All vowels in the weak branch of the foot surface as schwa, accounting for the fact that immediately post-tonic vowels are schwa. All vowels in the strong branch surface as a full vowel (not schwa). Pretonic vowel quality is unpredictable: a given speaker may produce a given word at one time with a full vowel and at another time with a reduced vowel. I have heard, for example, [tawә'linә] and [tawo'linә] 'table' (dimin.), and [piә'rann] and [pia'rann] 'grandfather'. If we assume that only one foot is built on each word (at the right edge), this accounts for the allophonic alternations between schwa and non-schwa vowels in pretonic position: these vowels are not footed, so there is no requirement that they be either schwa or not schwa.7 Support of lexically marked stress is found in the marking of particular morphological classes, such as past participles and many infinitives, by wordfinal stress. Examples are given in (8). Such forms cannot be analyzed as words ending in underlying long vowels because RS applies to these forms when they appear in the appropriate context, and, as discussed in §2.4, RS would not apply if the vowel were long. Rather, these words must have a bimoraic final syllable, where one mora is not underlyingly associated to a segment.
7 There is no evidence that any pretonic vowel is either categorically unreduced or categorically reduced. One way to determine this would be to conduct an acceptability judgment test with native speakers.
STRESS AND SCHWA IN FAETAR (8)
Past participles and Infinitives 'grew/to grow' 'went/to go' [krә'fi] ' stayed/to stay' [a'la] [arrama'ne] 'married/to marry' [ma'ri] 'married/to marry' [fpu'za]
249
Past participles [a'ßi] 'had' [u'le] 'wanted' [tә'ni] 'held' [mun'd3a] 'ate'
Historical evidence also supports lexically marked stress: Francoprovençal, the source language of Faetar, has lexically marked stress. Although Francoprovençal requires that stress be on one of the last two syllables, there are minimal pairs like ['8ata] 'sing, 2nd sg. imperative' vs. [8a'ta] 'sing, 2nd pl. imperative' (Martin 1990:673). There is no way that stress could be predicted to appear on different syllables of the two members of this pair, as they are otherwise identical. For a number of words, comparison with the Italian source form of borrowings into Faetar indicates that a final segment or syllable has been deleted since the Italian word was borrowed into Faetar, without affecting the stress pattern. Apparently, stress appears on the last syllable of the word in the Faetar word because that is the syllable marked for stress in Italian. Such diachronic evidence is not available to a speaker's grammar, so stress must be lexically marked for these words. (9) Italian borrowings with final stress in Faetar (transcriptions from Minichelli 1994) Faetar Possible Italian source allergia /aller'djia/ 'allergy' /allεrd3i/ /alie'ri/ 'happiness' aliegria /alle'gria/ /anә'ma/ /ani'male/ animale 'animal' /ar'ka/ arcata /ar'kata/ 'arcade' /au'ta/ altare /al'tare/ altar' /bar'ri/ harile barrel' /ba'rile/ /bәkkjә'ra/ hicchierata /bikkje'rata/ glassful' /kande'da/ /kandi'dato/ candidate' candidate /kapat'ta/ capitale /kapi'tale/ capital' /kartfe'ri/ carceriere /kartfe'rjsre/ prison warden'
3.2
The Marked Quantity Hypothesis A second potential analysis is the Marked Quantity Hypothesis: that vowels are marked as short or long in the lexicon (distinguishing short schwa from other, long, vowels), and then stress is predictably assigned to the
250
NAOMI NAGY
rightmost long vowel. The problem with this analysis is that the difference between schwa and other vowels in Faetar cannot be accounted for moraically, and cannot be lexically marked. A phonemic distinction in vowel length would have to be represented by different associations between segments and moras for schwa vs. non-schwa vowels. The prosodic structures presented in (5) do not allow for any leeway: both long (i.e. non-schwa) vowels and schwa must be associated to exactly one mora. Furthermore, this hypothesis is contradicted by the fact that only schwa appears in post-tonic syllables. If schwa were a phoneme, there is no reason why only it would appear in this position. Similarly, the alternations between schwa and other vowels in pretonic position could not be accounted for. Finally, if the difference between schwa and other vowels were that other vowels were bimoraic (lexically marked as long) and schwa were not, then RS would not be expected to apply: there would be no unassociated mora in wordfinal full vowels. §2 shows that RS does apply when there is a full vowel, and not when there is a schwa, supporting the Marked Stress Hypothesis. 3.3
The Marked Quality Hypothesis One alternative approach is that Faetar vowel quality is lexically marked. The vowel phonemes are then /i, u, e, o, a and 9/. By rule or constraint, stress will be placed on the last syllable with a non-schwa vowel. While apparently descriptively adequate, such an approach cannot account for alternation in verb paradigms. Many paradigms contain pairs as listed in (10): some verb forms have a stressed full vowel in the last syllable of the root and other forms have an unstressed schwa in the same position (Giuliani 1995:147). Presumably, the underlying root forms of various elements of a paradigm are identical. (10) [ijә i Vint] 'he comes' [djid3ә'leva] ' I lift'
~ ~
[nus na vә'nunnә] 'we come' [nus na lәVunnә] ' we lift'
Second, if we assume that vowel quality is marked in the lexicon, there is no way to correctly assign stress by rule or constraint in Faetar, given the diversity of word types shown in (3). For example, if we take a common Faetar word, ['brokәtә] 'fork', there is no way to predict that it is pronounced ['brok9l9] rather than [brәkә'le] or [br9'kel9]. A regular right-to-left trocheeforming grammar would produce the latter form, rather than the appropriate one, if schwa were a phoneme.
STRESS AND SCHWA IN FAETAR
4.
251
Conclusion Several arguments have shown that Faetar is a language in which stress is lexically marked and schwa is not a phoneme, but rather a reflex of some other vowel(s) in unstressed position. Support for the Marked Stress Hypothesis (and against the other hypotheses) includes: the vowel alternations in many verb paradigms, the existence of RS in Faetar, the retention of historically-motivated stress assignment, the alternation of vowel and schwa in pretonic position, the categorical absence of non-schwa vowels in post-tonic position, and the restriction of stress to the three final syllables. APPENDIX Word list for RS experiment /f/ initial words Consonant Duration 0.036 0.045 0.070 0.086 0.113 0.026 0.035 0.041 0.041 0.044 0.045 0.049 0.052 0.055 0.065 0.070 0.078 0.089 0.096 0.108 0.110 0.110
Preceding Segment C C C
c c ә ә ә ә ә ә 9 9 ә ә 9 ә ә 9 9 ә ә
Phrase (Preceding & Target words) anfan forsә sun fiaws in fttә kristian forsә in fatti atә fij kattә fiann ands fεrma attә fij dә favolεt kworrә fors stunds fors ss firmatә firmats firmats kontә firma uns famiAs stεssә f amiyә atә fiaw atә fen ands fa atә fen ats fen
English Gloss child maybe his son in fact people maybe in fact other daughter four children have stopped other daughter of story runs maybe (they) are maybe (he) stops stops stops sure thing a family same family other son other woman did other woman other woman
252 0.063 0.065 0.067 0.069 0.078 0.084 0.085 0.089 0.092 0.092 0.092 0.093 0.095 0.099 0.123 0.124 0.146 0.151 0.231 0.336
NAOMI NAGY
V V V V V V V V V V V V V V V V V V V V
/m/ initial words Consonant Preceding Segment Duration 0.028 0.031 0.049 0.052 0.057 0.330 0.025 0.033 0.034 0.036 0.037 0.042 0.046 0.192
e ferma u fonds la fermats sta famiÅ sta fiÄ sta famiyә sta famiÅ sta famiy n a fenns n a fets
trefijaw kifa i fa la fermats
is closed at the bottom the stop this family this girl this family this family this family a woman a party a party a family a party the girl this girl drinks maybe three children that do he does the stop
Phrase (Preceding & Target words)
Gloss
n a fets na famiÅs n a fets la fiyjs sta M s baij forsә
English
C C
anfan m e papanon mamanon
c c c c
sun mari un mufil anfan m e anfann m e
more children grandpa grandma her husband a kitten more children more children
9
esta mәtann pә mεt lә m a n lә m a n n bambols makәnotsv kakә maijә kakә maijә bit fiktεttә mototfikεttә
this morning (in order) to put the hand the hand dolls, cars some more some more bicycle- scooter
9 9 ә 9 9 9 9
STRESS AND SCHWA IN FAETAR
0.018 0.041 0.044 0.044 0.051 0.052 0.063 0.067 0.072 0.077 0.080 0.081 0.082 0.082 0.082 0.084 0.087 0.087 0.089 0.090 0.095 0.097 0.098 0.103 0.121 0.122 0.180 0.233 0.401 0.889
V V V V V V V V V V V V V V V V V V V V V V V V V V V V V V
a mi/ du mart/ a piεd u mwen du martf a pied u mit/ bu me mê me dapoi mәnd3a bume sa mar sa mar sa maÄija la msnnzota sa mar i rrrindg la makkana sa mar sa mar na motqfikktt sa mar e marra u mit/s sa marra sa msAijs errøAi su mar per o mank 9 umar u mwenn domwen
253
in the middle on the sidewalk the man on the sidewalk in the middle bit more more more after eathing bit more the mother the mother his wife the shelf the mother he eats the ear the mother the mother a scooter the mother and mother in the middle the mother his wife and wife his mother but (is ) missing the mother the man some men
REFERENCES Amery, Heather & Stephen Cartwright. 1987. The first hundred words. Tulsa: Educational Developmental Corporation. Bertinetto, Pier Marco. 1985. "A proposito di alcuni recenti contributi alla prosodia dell'italiano". Annali della Scuola Normale Superiore. Classe di Lettere e Filosofia (Serie UI) 25.581-643.
254
NAOMI NAGY
Dulcibella, J. W. 1934. The Phonology of Sicilian Dialects. New York: AMS Press. Fiorelli, Piero. 1958. "Del raddoppiamento da parola a parola". Lingua Nostra 19.122-127. Giuliani, Francesca. 1995. Ricerche sul francoprovenzale di Faeto e Celle San Vito (in provincia di Foggia). Ph.D. dissertation, University of Florence. Grandgent, Charles. 1927. From Latin to Italian: An Historical Outline of the Phonology and Morphology of the Italian Language. Cambridge, Mass.: Harvard University Press. Hoffman, Robert Joseph. 1968. The Franco-Provençal Dialect of Faeto. Ph.D. dissertation, Ohio State University. Jaberg, Karl & Jakob Jud, eds. 1928-1940. Sprach- und Sachatlas Italiens und der südschweiz. Zofingen: Ringier. Kager, René. 1993. "Shapes of the Generalized Trochee". Proceedings of the 11th West Coast Conference on Formal Linguistics ed. by Jonathan Mead, 298-312. Palo Alto: CSLL Kattenbusch, Dieter. 1982. Das Frankoprovenzalische in Süditalien: Studien zur synchronise hen und diachronischen Dialektologie. Tübingen: Gunter Narr. Martin, Jean-Baptiste. 1990. "Französisch: Frankoprovenzalisch". Lexikon der Romanistischen Linguistik (vol. 2). ed. by Gunter Holtus, Michael Metzeltin, & Christian Schmitt, 671-685. Tübingen: Max Niemeyer. Minichelli, Vincenzo. 1994. Dizionario francoprovenzale. Celle di San Vito e Faeto. Alessandria: Edizione dell'Orso. Nagy, Naomi. 1994. "Language Contact: Italian (?) Geminates in Faetar". Belgian Journal of Linguistics 9.111-428. —. 1996. Language Contact and Language Change in the Factor Speech Community. Ph.D. dissertation, University of Pennsylvania. Nagy, Naomi & Bill Reynolds. 1997. "Optimality theory and variable wordfinal deletion in Faetar". Language Variation and Change 9.37-56. Repetti, Lori 1991. "A Moraic Analysis of Raddoppiamento Fonosintattico". Rivista di Linguistica 3.307-330. Saltarelli, Mario. 1970. A Phonology of Italian in a Generative Grammar. The Hague: Mouton. Valente, Vincenzo. 1975. Puglia (Profilo dei dialetti italiani, vol 15). Pisa: Pacini.
VOWEL LENGTHENING IN MILANESE* PILAR PRIETO I VIVES Universitat Autonoma de Barcelona 0.
Introduction Modern Milanese displays a very interesting set of vowel length contrasts, with long vowels usually surfacing in word-final stressed syllables before underlyingly voiced consonants: ['nœ:v]~['nœ:f] 'new.ms' vs. ['nœva] 'new.fs'; [Vi:v]~['vi:f] 'alive.ms' vs. [Viva] 'alive.fs'. Recent analyses of Milanese synchronic vowel length alternations have relied on a combination of a Bimoraic Enforcement condition (every stressed syllable has to be heavy) together with a moralc distinction between voiced and voiceless consonants. However, these analysis leave unexplained the fact that vowel lengthening occurs exclusively in the final syllable of the word and not in syllables in penultimate and antepenultimate positions. The present article offers a different interpretation of these alternations and claims that the distribution of short and long vowels in this dialect can be successfully accounted for by resorting to the following optimization constraints (Prince 1990, McCarthy & Prince 1993, Mester 1994). (1)
a. FT-BIN: Feet should be analyzable as binary b. HNUC: A higher sonority nucleus is more harmonic than one of lower sonority c. FILL: Syllable positions are filled with segmental material
The first principle, FT-BlN, is one of the essential constraints at play in this proposal and motivates the fact that monomoraic feet are prone to be repaired. In the trochaic system under consideration, monomoraic feet are found in word-final The article has benefited from my participation in the following research groups: 'Laboratori de Fonètica-IEC' (project 1997SGR0Q290) and 'La estructura de las unidades prosodicas' (project PB-96-1199-C04-04). I would like to thank Lori Repetti and an anonymous reviewer of this article for very helpful comments on an earlier version.
256
PILAR PRIETO I VIVES
stressed syllables, precisely the syllables that tend to be lengthened. The second constraint, HNUC, evaluates which segment should rescue a given moralc unit, deciding the cases in which the vowel will be lengthened. Thus, it will be shown that an optimality approach can straightforwardly account for a phenomenon that was not satisfactorily handled within derivational models. Also, the present analysis provides further evidence of the crosslinguistic applicability of the three constraints employed. The article is organized as follows. The first section presents a detailed description of the distribution of short and long vowels in Milanese. Section 2 revises previous analyses of the Milanese facts, and section 3 presents an optimality analysis of the data. 1.
Vowel Length Distribution in Milanese Milanese is a Gallo-Italian dialect spoken in the northern Italian province of Milan (Lombardy region) which belongs to the Western Lombard dialectal area1. As in other northern Italian dialects, vowel length is distinctive in Milanese. The set of vowels that can appear in stressed positions is shown in (2). 2 (2)
i
y (e)
u s
i: c
œ a
y: e:
u: (s:) œ:
(o:) a:
In (3) we find minimal (and quasi-minimal) pairs illustrating vowel length contrasts in stressed word-final position. The examples in (3b) show a case of morphologically conditioned lengthening: while the infinitive form of a given verb ends in a final short vowel ([por'ta] 'to bring', [fi'ni] 'to finish'), the masculine past participle form ends in a final long vowel ([por'ta:] 'brought.ms', [fi'ni:] 'finished.ms'). (For presentation of the data, I follow IPA conventions throughout the article.) 1 According to Merlo (1960: 2) and Sanga (1984: 8), Western Lombard dialects comprise the following areas: the provinces of Milan, Como, Varese, Sondrio and the northern part of the Pavia province (plus the Canton Ticino in Switzerland and the eastern part of the Novara province). Linguistic features characterizing Western Lombard dialects as opposed to Eastern dialects are: 1. Proto-Romance [u] became [y]; 2. Word-final [r, 1] were deleted after tonic vowels; 3. Vowel length is distinctive in word-final stressed positions (Sanga 1984:60-61). 2 The examples given throughout the article come from the following sources: Cherubini (1839) = Ch; Gökcen (1991) = G; Lepschy (1978) = L; Montreuii (1990) = M; Nicoli (1983) = N; Pavia (1928) = P; Salvioni (1975) = Sal-75; Salvioni (1884) = Sal; Sanga (1984) = S-84; Sanga (1988)= S.
VOWEL LENGTHENING IN MILANESE (3) a.
['si]
'boy' (N:46) 'to take' (S:293) 'sun' (N.41) 'mine.ms' (S:293) 'foot' (N:39) 'yes' (S:293)
[fjœ:] [toe:] ['su:] ['me:] [pε:] ['si:]
'girl' (N:46) 'your.p' (S:293) 'I am'(N:41) 'mine.mp' (S:293) 'feet' (P:207; N:39) 'seven* (S:293)
[an'da] [por'ta] [gana'sa] [fi'ni] [ves'ti] [ri'dy]
'to go' (S:292) 'to bring' (P.207) 'to eat eagerly' (N:37) 'to finish'(S:293) 'to dress' (N:39) 'to reduce'(N:39)
[an'da:] [por'ta:] [gana'sa:] [fi'ni:] [ves'ti:] [ri'dy:]
'gone.ms' (S:292) 'brought.ms' (P:207) 'devoured.ms' (N:37) 'finished.ms' (S:293) 'dress* (N:39) 'reduced.ms' (N:39)
[fjoe] ['toe] ['su] ['me] ft*]
257
In unstressed positions, vowels are always short: (4)
The data in (5) illustrate the process of vowel reduction in Milanese. When in unstressed position, /ε, c, œ/ become [e, u, u], respectively, and the rest of the vowels (/a, e, i, u, y/) maintain their vocalic quality (Pavia 1928:16, Nicoli 1983:46). (5) [ga'nasa] ['asen] [ba'las] ['tern] ['pes] Tbel] Tfjœ] Tvin] ['tus] ['mo sa]
'jaw' 'donkey' 'artful' 'earth' 'fish' 'beautiful' 'boy' 'wine' 'cough' 'movement'
[gana'sa] [as'nin] [balu'set] [te'rε] [pe'sit] [be'lsdza] [fju'let] [vi'net] [tu'si] [mu'seta]
'to eat eagerly' (N:37) 'donkey.dim'(N:110) 'artful dim' (N:47) 'terrain' (N:50) 'fish.dim' (N:38) 'beauty' (N:38) 'boy.dim' (N.46) 'wine.dim' (N:46) 'to cough' (S:296) 'movement.dim' (N:50)
Vowels are always short in unstressed positions, as the derived words in (6) show.
PILAR PRIETO I VIVES
258 (6) [tfi'ta:] [mala:] ['pe:s] [mes'te:] [ves'ti:] [par'ti:]
[tfita'dina] [mala'din] [pe'za] [meste'rin] [vesti'din] [parti'don]
'city' 'sick' 'weight' 'skill' 'dress' 'party'
'city.dim'(N:109) 'sick.dim'(N:109) 'to weigh' (S:39) 'skill.dim'(N:109) 'dress.dim' (N:39) 'party.augm' (N: 109)
Apart from these underlying contrasts, Milanese displays a productive alternation between long and short vowels. Vowels surface as long when in wordfinal stressed syllables before underlyingly voiced obstruents (7a) and as short when in stressed penultimate positions (7b). The examples in (7b) also show the optionality of the rule of devoicing of word-final obstruents (Nicoli 1983:52-53; Montreuil 1990:45). (7) a.
['nœ:v] ~ ['næ:f] [Vi:v] ~ [Vi:f] ['tu:z] ~ ['tu:s]
'new.ms' 'alive.ms' 'boy'
b.
[nœva] [Viva] ['tuza]
'new.fs' (S:296) 'alive.fs' (S:296) 'girl' (N:63)
(8) exhibits the difference in behavior between vowels preceding word-final voiced obstruents ([z, g, d, g]) and vowels preceding word-final voiceless obstruents ([f, s, t, k]). While vowels preceding voiced obstruents are long (8a), vowels preceding voiceless obstruents are short (8b). The examples listed below are from Nicoli (1983:45,50), Sanga (1988) and Montreuil (1990:38).3 (8) a. ['na:z] ~ ['na:s] [la:g] ~ [la:k] ['fi:g] ~ ['fi:k] ['re:d]~['re:t] ['se:t]
3
'nose' lake' 'fig' 'net' 'thirst'
[na'za] [lage'tin] [fi'ge:] [re'dina] [se'dafa]
'to smell' 'lake.dim' 'fig tree' 'net. dim' 'thirst.augm'
Similar vowel length alternations obtain in Friulian. Vowels are long before underlyingly voiced consonants and short before voiceless consonants. One of the major differences between Milanese and Friulian is the obligatoriness of the final obstruent devoicing rule in Friulian. (All data are taken from Hualde 1990.) [la:k] lake' [la'gune] 'lagoon' 'ship' [navi'ga:] 'to navigate' Pna:f] [fa'mo:s] 'famous.ms' [fa'moze] 'famous.fs' ['la:t] 'gone.ms' ['lade] 'gone.fs' ['lu:f] 'wolf [lu'vin] 'wolf. dim' But: ['mat] 'mad, crazy, ms' ['mate] 'mad, crazy.fs'
VOWEL LENGTHENING IN MILANESE
[myf] [pεs] ['spεs] ['pas] ['tas] ['gat] ['set] ['sak]
'mouldy.ms' 'fish' 'dense, ms' 'step' 'taxes' 'cat' 'seven' 'sack'
['myfa] [pe'sam] ['spεsa] [pa'sa:] [ta'sa:] [ga'tc] ['sεte'mεz] [sa'ket]
259
mouldy.fs' quantity of fish' dense.fs' to pass' to tax' big cat' ancient coin' sack, dim'
As we have already mentioned, vowels are always short in penultimate position (9a) and in antepenultimate position (9b). ['pever] [fever] [noeva] [viva] [tuza] ['rnosa] ['tern] Ipa'dsla] ['strolega] ['unika] [pegura] ['nivula]
'pepper'(Sal-75:4) 'fever' (Sal-75:4) 'new.fs' (S:296) 'alive.fs' (S:296) 'girl' (N:63) 'movement' (N:50) •earth'(N:50) Trying pan' (N:38) 'gypsy' (N:47) 'unique.fs' (N:47) 'sheep' (N:38) 'cloud'(N:83)
Even though the general trend for word-final voiced consonants is to trigger lengthening of preceding vowels, underlying /l/ and /n/ behave in a special way. Before these consonants, vowels can be either long or short. (10a) and (10b) show instances of long and short vowels, respectively, before word-final [1].4 (10)
a.
[pe:l] ['fe:l] ['mœ:l] ['ku:l] ['fi:l] [ka:l] [ba:l]
'hair' (M:39; S:291) 'faithful.ms' (S:292) 'mass' S:295) 'strainer' (S:295) 'son' (G:25) 'drop, fall'(M:39) 'bales'(L:208)
b.
[pεl] [fel] ['mol] ['kcl] ['stil] [kal] [bal]
'skin'(M:39;L: 208) 'do i t 3s' (S:292) 'soft.ms' (S:295) 'neck' (S:295) 'style' (S:292) 'corn' (M:39) 'dance'(L:208)
The different outcome is due to the etymological origin of the consonant. As Sanga (1988:291) points out, "avanti liquida semplice originaria abbiamo sempre vocale lunga"
260
PILAR PRIETO I VIVES
The examples in (11) illustrate the fact that vowels can be long or short before word-final underlying in/. Like in the case of final /1/ (10), final /n/ can be preceded by a short vowel (11a) or by a long nasalized vowel, in which case the nasal consonant is also deleted ( 1 lb). 5 (11) a.
['kan]'reed.p' (cf. [ka'nin] 'reed.dim' (Ch:206))
b. ['kä:]*dog' (cf. ['kani] 'dog.p' (Ch:200))
['pan]'cloth' (cf. [pa'ncs] 'cloth.dim' (Ch: 1068))
['pa]'bread' (cf. [pana'tεl] 'breaddim' (Ch: 1064))
[dcn]'woman' (cf. [do'nat] 'effeminate' (Ch:436))
['ma:]'hand' (cf. [ma'nina] 'handdim' (N:58)) [se'rε:] 'serene.ms' (cf. [sere'nisim] 'serene, super' (N: 128))
In contrast to the behavior of vowels preceding final ill and /n/, vowels preceding word-final /r/ are always long (12). Vowel lengthening before [r] might have been due to a generalized phonetic effect caused by syllable-final rhotics, a common process reported in many northern Italian dialects (Prieto 1994). (12)
['pe.r] ['ka:r] [to:r] ['u:r]
'pear'(S:293) [fer] 'cart; dear.ms' (G:255; S:292) ['sc:r] 'tower' (G:246) [do:r] 'hour'(S:292) [du'lu:r]
'iron' (S:37, G:255; Sal:37) 'soft.ms* (S:293) '(it) hurts' (S:291) 'pain*(N:54)
5 The difference between long and short vowels in this environment can be traced back to the historical difference between geminate and simple nasals: short vowels show up before Latin geminate nasals and long vowels before single nasals: CANNAS > ['kan] 'reed.pl' (S:292), CANE> ['ka:] 'dog' (S:292). Milanese also displays a productive morphophonemic alternation between masculine and feminine forms ending in underlying /n/. While in the masculine form the final nasal deletes, in the feminine form it does not (examples are from Nicoli 1983:57-8 and Lepschy 1978:208): ['bu:] 'good.ms' I ['bon] 'good.fs', [upur'ty:] 'opportune.ms' / [upur'tyn] 'opportune.fs', [kwaj'dy] 'someone.ms' / [kwaj'dyn] 'someone.fs'. Syllable-final [n] is deleted in word-internal stressed positions (['sidik] 'mayor', [zgü:fi] 'swollen.ms', [Vy:des] 'eleven'), but is never deleted in unstressed syllables (['azen] 'donkey', ['kardin] 'hinge', [Verdjin] 'virgin', [dzjuvin] 'young.ms', ['orgen] 'organ', [ba'ritun] 'baritone', ['urdin] 'order', [tran'vaj] 'train', [ande'ge:] 'antiquarium', [ven'dyda] 'sold.fs'). According to Tuttle (1991), this pattern of nasal deletion is very common across northern Italian dialects. Other nasal consonants (bilabials and palatals) never delete or nasalize preceding vowels (['lym] 'lamp', flam] 'flame.p', f'om] 'man', ['soji] 'sleep', f'lsji] 'wood', ['pyji] 'fist', ['psji] 'ledge'. Similarly, syllable-final [ij does not delete(fdugka] 'therefore', faigka] 'also', [Vagga] 'spade', [luqk] 'long.ms', ['tjîigk] 'five', ['bjarjt] 'white.ms', ['sagg] 'blood'). (The abovementioned examples are from Nicoli 1983:58-59.)
261
VOWEL LENGTHENING IN MILANESE
Vowels before word-final consonant clusters can also be long or short, depending on the voicing properties of the cluster. In general, vowels are short before voiceless clusters [sk, st, stj] (13a), and vowels tend to be long before clusters consisting of a liquid plus an obstruent (13b). In the latter case, we find some degree of variation (['kyrt] ~ ['ky:rt] 'short.ms'; ['kyrv] ~ ['ky:rv] 'curved.ms'; see Sanga 1988:192, 293) 6 . Again, vowels in this environment are short when in penultimate syllables: ['a:1t] (S-84:66) 'tall.ms' / ['alta] 'tall.fs', ['d3a:lt] 'yellow.ms' / [djalda] 'yellow.fs'. (13)
6
a.
voiceless cluster [Tjask] 'bottle' (M:40) ['risk] [rost] 'roastedms' ['past] [Tyst] 'trunk' (M:40) ['pest] ['ristj] 'risk' b. voiced cluster i. long vowel ['se:rk] [pc:rk] 'pig'(S:293; P:35) ['kc:rp] 'body' (P:35) ['spa:rg] ['mc:rt] [a'vε:rt] 'open.ms'(P.33) [To:ftf] 'scythe' (S-84:63; P:35) ['ka:dd] ['ne:rv]~['ne:rf] 'nerve' (S:293; P:33) [la:rg] ~ [la:rk] long.ms' (N:53; S:292) [Ve:rd] ~ [Ve:rt] 'green.ms' (B:251; S:293; G:250) ii. short vowel ~ long vowel ['kyrt] ~ ['ky:rt] 'short.ms' (S-84:62) ['kyrv] ~ [ky:rv] 'curved.ms' (S-84:62) iii. short vowel [sfils] 'strings' (S-84:62) ['puls] 'wrist' (S-84:62) [Vulp] 'fox' (S-84:62) ['crp] 'blind.ms' (N:51; P:35) [dis'tyrp] 'trouble' (S:293) ['durd]~['durt] 'thrush' (S:293)
'curl' (S:293) 'grazeland' 'poundedms'
'circle'(G:255) 'asparagus' (P:32) 'dead.ms'(P:35) 'hot.ms' (S-84:62)
According to Sanga (1984), there is a contrast between the behavior of high and non-high vowels before word-final liquid plus consonant clusters. While high vowels [i, y, u] are short, non-high vowels [s, c, e, a, œ] are long. Although this might reflect a general tendency in Milanese, we have to admit that this is the only study that describes this behavior. Other descriptions of the phenomenon point out that vowels preceding such clusters are generally long. I only found the following counterexamples to Sanga's observations: ['pert] 'opened.ms' (G:255), ['perd] 'to loose' (Sal:58; G:255), [Tals] 'false.ms' (N:62).
262
PILAR PRIETO I VIVES
2.
Previous Analyses of Milanese Vowel Lengthening: The Role of Weight and Sonority Authors such as Salvioni (1884), Nicoli (1983) and Sanga (1988:294) have already acknowledged the role that voicing properties play in the synchronic vowel length alternations found in Milanese. (See Gokçen 1990 for a review.) Within the autosegmental framework, Montreuil (1990) has worked out the details of an analysis that relies on both the role of sonority and syllable structure.7 (See also Prieto 1993.) In order to account for the Milanese distribution of long and short vowels, Montreuil assumes that word-final voiced and voiceless consonants have different underlying weight properties. He argues that Milanese words have two possible lexical moraic representations, depending on whether long vowels show up in the phonetic representation. (14)
Moraic analysis of Milanese vowel length (Montreuil 1990:42) 1. Syllables with short vowels are underlyingly bimoraic (15a) 2. Syllables with long vowels have the underlying structure given in (15b)
(15) Milanese Underlying Coda Contrasts (Montreuil 1990:43) a.
σ
σ
A μ
μ
| kas 'case' b.
σ
A μ
|
μ
H f ys 'moated.ms'
A
JA JA
|
JA JA
| p an 'cloth'
σ
σ
σ
A
σ
|I p εs 'fish' σ
|||| JA
\\ kaz 'house'
JA
\\ fyz 'melted.ms'
μ
l\ pan bread'
μ
l\ pez 'weight'
Montreuil claims that a process of Bimoraic Enforcement is active in Milanese, as Saltarelli (1983) and Chierchia (1986) had proposed for Standard
7
Montreuil frames his analysis in a moraic view of the syllable structure. Even though he uses the moraic tier for syllabic representations, he points out that this assumption should not interpreted as an argument against other possible intrasyllabic constituents.
263
VOWEL LENGTHENING IN MILANESE
Italian.8 By virtue of this constraint, every stressed syllable becomes bimoraic and, in a later process, the empty moraic unit is rescued by the vowel, thus becoming long. (16a) illustrates the underlying contrast between words with word-final nonmoraic (voiced) and moraic (voiceless) consonants (/kaz/ vs. /kas/). (16b) shows the application of the Bimoraic Norm: indeed, the rule is only applicable in cases where final consonants are non-moraic (/kaz/ and /pan/). It cannot apply in the case of /kas/ because the bimoraic constraint is already fulfilled. Finally, (16c) shows the filling procedure whereby vowels link to the empty mora added through Bimoraic Enforcement. (16) Montreuil's Analysis of Vowel Lengthening in Milanese a. Underlying representations σ cr cr | A ! JA
JA JA
JA
l\ || w p an k az kas b. Bimoraic Stress Norm and Parasitic Delinking cr σ cr A A A JA JA
JA JA
| || kas k az c. Filling Operation from the Vowel cr A
8
JA JA
|| p an σ A
JA JA
JA JA
| p an
II p an
σ
A JA JA
JA JA
|/ k az d. Output ['ka:z] 'house
σ A
n. a. ['kas] 'case'
|/ p an ['pa:n] 'bread'
n. a. ['pan] 'cloth'
Two phonological processes present in Standard Italian, open-syllable lengthening and Raddoppiamento Sintattico, have been generally interpreted as processes triggered by a rule of Bimoraic Enforcement on stressed syllables. The two processes represent two different repair strategies to reach an ideal bimoraic syllable, namely, lengthening of the stressed vowel or lengthening of the following consonant.
264
PILAR PRIETO I VIVES
To account for the nasalization process in the word 'bread' (pronounced ['pä:] and not *['pa:n]), Montreuil introduces the rule of nasal spreading in (17), a process that applies to underlying representations and spreads nasality only 'inside' the mora. (17)
Nasal spreading: Nasalize inside the mora (Montreuil 1990:45)
The rule in (17) ensures that nasalization can only spread when the nasal consonant shares a moraic unit with the preceding vowel (see Montreuil's representation of monomoraic /pan/ 'bread' in (15b)); however, the rule cannot apply to bimoraic /pan/ 'cloth', as the two segments in the rime license two separate moras. Finally, Montreuil's analysis would also need a later rule of nasal deletion. Even though the Bimoraic Enforcement approach correctly handles the basic facts of Milanese vowel length alternations, we should point out some of its weaknesses. On the one hand, the process of Bimoraic Enforcement predicts that all stressed vowels in open syllables should be long: instead, we find a clear contrast between vowels in word-final syllables (which are long) and vowels in penultimate and antepenultimate positions (which remain short). How, then, would words such as [Viva] 'alive.fs' or ['nœva] 'new.fs' ever be able to avoid the Bimoraic Enforcement rule? We will argue that we are not dealing with a process of optimization at the syllable level (through Bimoraic Enforcement), but at a higher level, namely, at the level of the foot. Montreuil's analysis has another drawback. By lexically marking voiceless segments as moraic, such a proposal is in direct contradiction with the crosslinguistic tendency to represent sonorant and voiced consonants as having more prosodic weight than less sonorous segments. For example, in Classical Greek Osthoff's Law had the effect of shortening long vowels before rimes containing sonorants: gno:-nt-es — > gnontes 'knowing.nom.pl' (cf., gi-gno:sko:); e-phan-e:-nt —> ephanent —>ephanen 'appear.aorist.pl' (cf., e-phan-e:-n). Both Zec (1988) and Steriade (1990) interpret the Greek example as an argument in favor of distinguishing sonorants from non-sonorant consonants with respect to weight licensing. As sonorants are mora-bearing in Greek, an application of the Bimoraic Enforcement principle forces vowels to shorten. Other instances of this type of behavior have been reported in languages such as Kwakiutl and Nootka, which treat CVS syllables (S=Sonorant) as heavy and CVO syllables (0=Obstruent) as light for stress assignment purposes. For more arguments dealing with stress attraction and stress assignment and shortening processes, see Zec (1988) and Steriade (1990).
VOWEL LENGTHENING IN MILANESE
265
Repetti 's (1992) analysis for Milanese and Friulian offers a solution to the 'weight adequacy' problem by positing that word-final voiced consonants (voiced obstruents and sonorants) can constitute degenerate syllables, in contrast with voiceless consonants, which cannot. She considers long vowels to be the underlying units (/bandiid/ [ban'di:t] 'banditms 5 , /viiv/ ['vi:f] 'alive-ms'), and postulates a rule of vowel shortening in non-final syllables in order to account for Friulian words such as [ban'dide] 'bandit.fs', ['vive] 'alive.fs'. "In derived words, the consonant remains voiced (from the underlying form) and the vowel surfaces as a short vowel because of the metrical rule in Friulian that allows long vowels in final stressed syllables only" (Repetti 1992:170).9 This proposal recognizes the moraic properties of voiced and voiceless consonants and nicely accounts for cases where short vowels appear before word-final underlyingly voiced consonants: ['mjεtf] 'half.ms', ['mjεdje] 'half.fs'. Similarly, the proposal explains recent loanwords that never reflect a CVVC structure with a voiceless final C in the underlying representation (generally, the final consonant becomes voiced if it is preceded by a long vowel). Yet, the analysis is also stipulating a shortening process in open penultimate and antepenultimate syllables precisely where we would expect a lengthening effect. Such a rule runs against the well-known crosslinguistic tendency for vowels to lengthen in open stressed syllables. (For a more detailed analysis of the 'degenerate syllable' proposal applied to the Friulian data, see Repetti 1994). Finally, Hualde's (1990) analysis of the synchronic vowel length alternations in Friulian relies on the Final Devoicing rule in this language. This phonological process delinks the moraic unit of the voiced consonants and the result crucially feeds a rule of compensatory lengthening: the empty mora is filled by the preceding vowel, making it bimoraic. However, this proposal completely loses its motivation when applied to Milanese, where the rule of Final Devoicing is only optional. 3.
Vowel Lengthening in Milanese: An Optimality Approach The analysis presented here is framed in the Optimality Theory (OT) defended by Prince & Smolensky (1993). As Archangeli & Pulleyblank (1994) put it, "the optimization hypothesis is that grammars prefer relations between representations that are defined in formal and grounded terms as optimal: synchronic and diachronic processes should gravitate towards optimal relations". Following Prince & Smolensky (1993) and McCarthy & Prince (1993), I assume 7
As Repetti points out, the vowel shortening rule has some exceptions in Friulian, namely, a group of words where the long vowel comes from the deletion of the first consonant of an onset obstruent cluster: NIGRU> ['ne:ri] 'black.ms'.
266
PILAR PRIETO I VIVES
that Universal Grammar is defined by a set of constrains on structural wellformedness, and that these universal constraints are not inviolable. Each grammar resolves conflicts between universal constraints in a language-particular way: certain constraints can be prioritized over others in a strict dominance hierarchy in a way that a given constraint has absolute priority over all the others lower in the hierarchy. The present analysis claims that vowel lengthening in Milanese is the result of a composite action of the following three constraints on representational wellformedness. (18)
a. FT-BIN: Feet should be analyzable as binary b. HNUC: A higher sonority nucleus is more harmonic than one of lower sonority c. FILL: Syllable positions are filled with segmental material
FT-BIN (18a) expresses the crosslinguistic tendency to avoid monomoraic feet. If this constraint is highly ranked, and requires that the smallest foot should be bimoraic. It has been repeatedly shown that monomoraic feet are very highly marked and tend to be repaired or reanalyzed in various ways. In Latin, for example, monosyllabic words consisting of only one short vowel are completely ungrammatical (cf. fe 'thing', spe 'faith', *re, *spe), even though they are stable in polysyllables {lege 'he/she read.imp', anima 'soul'; Mester 1994). Mester (1994) has suggested the following prosodic well-formedness scale of foot types: "bimoraic trochees ([σμμ] or [σσ]) have the highest degree of harmony, followed by trimoraic trochees [σμμσ], and with monomoraic trochees [σ] ranking lowest on the harmony scale, i. e.: [μμ] > [ΜΜ Μ] > [Μ]" In the OT framework, the ban against monomoraic feet has been expressed with the so-called binarity principle (FT-BIN) formalized by Prince (1990) and McCarthy & Prince (1993). (19)
Foot Binarity (Prince 1990:6): FT-BIN: Feet should be analyzable as binary
As with most of the Romance languages, Milanese has a trochaic stress system. Stress falls in one of the last three syllables of the word, and typically antepenultimate stress is only possible when the penultimate syllable is light. This analysis assumes that word-final syllables in proparoxytones have to be marked as extrametrical at the lexical level. (20)
/noev/ 'new.ms'
/nœ va/ 'new.fs'
/uni/ 'unique.fs'
267
VOWEL LENGTHENING IN MILANESE
In (21), two possible output candidates are evaluated against the input forms /nœva/ and /unika/. Being bimoraic, the candidates [('nœva)] and [('uni)ka] do not violate FT-BIN, while [('nœ:va)] and [('u:ni)ka] do violate the constraint. Thus, the output forms that best satisfy this constraint are [('nœva)] and [('uni)ka]. (21)
FT-BIN
/nœva/
*!
a. [('nœ: va)] b. [('nœva)]
/unika/
|
FT-BIN
a. [('u:ni)ka]
*•'
b. [(u:ni)kal
In the case of oxytones, the difference between the behavior of vowels before voiced and voiceless obstruents cannot be obtained by resorting exclusively to the FT-BIN requirement. Neither ['(næ:)v] nor ['(næv)] violate FT-BIN, as the two moras can be filled, in one case by a long vowel (['(nœ:)v]), and in the other by a short vowel and a consonant (['(næv)]). The same is true of pairs such as ['(my:)f] and ['(myf)], as we can see in the following graphic representation. (22)
JA JA
\! n œ: v
JA JA
|
JA JA
| n œv
\/ m y: f
JA JA
|
| m uf
The difference between /nœv/ and /myf/ can be explained if we consider the following. First, let us assume that certain consonants at the end of the word contribute an underlying mora to the structure. Only voiced consonants will be able to license a moraic unit, with the exception of special cases like final /n/ or/1/), in line with the crosslinguistic observation that sonorous consonants tend to have more prosodic weight (Zec 1988, Steriade 1990). (23)
JA JA
μ
/n œv/
/m y f/
Second, the mora unit contributed by voiced consonants will have to be parsed into the structure. We argue that the principle called Nuclear Harmony
268
PILAR PRIETO I VIVES
Constraint (HNUC) (proposed by Prince & Smolensky 1993:16) will determine, through a process of harmonic evaluation, which segments constitute the best candidate for rescuing the empty mora. (24) HNUC: A higher sonority nucleus is more harmonic than one of lower sonority
In general, vowel elements will be preferred over any consonantal elements, as they are the most sonorous segments and will most likely represent an overall increase in the sonority of the nucleus. In the constraint tableau, the candidate that best satisfies HNUC is the output [(nœ:)v], (as opposed to ['(nœv)]), as it constitutes the most harmonic nucleus. Another possible candidate, ['(noe)v], is ruled out because it violates the FT-BIN constraint (25) μ |_ /nœv/ a. [(nœ:v)] b. [(nœv)] [ c. [(næ)v]
FT-BIN
HNUC
|
| œ-œ oe - v *!
œ
_
Finally, one of the basic constraints from the faithfulness family, FILL, will play a crucial role in selecting the optimal output candidate for forms such as /myf/. PARSE and FILL are both representatives of the set of constraints demanding a tight relation between input and output Both principles ensure that the initial input structure will be fully parsed and filled. The FILL constraint expresses the idea that every node in the input structure must be filled properly (Prince & Smolensky 1993:250). We will interpret this constraint in a general way so that it is able to ban any situation in which the input structure is either unfilled or overfilled. (26) FILL: Syllable positions are filled with segmental material
Tableau (27) evaluates three possible candidates for the underlying form /myf/. The candidate ['(my:)fj crucially violates FILL, as the segment ['y:] is adding segmental and prosodic material in the output that is not present in the input structure. Instead, the winning candidate, ['my:f], does not add any unnecessary material to the structure.
VOWEL LENGTHENING ÎN MILANESE
(27) /myf/
|
*!
a. [(my:)f] b. [('my)f] c.
FT-BIN
FILL
269
HNUC
j
y-y *!
[('myf)]
y
y-f
Similarly, word-final vowels in oxytones do not lengthen (cf. vesti [ves'ti] ' to dress') because, unlike vowels before word-final voiced obstruents, no moraic unit from the following consonant is available in the input structure. This set of examples represent an argument for the relative ordering of FT-BIN and FILL, showing that FILL has to be ranked higher than FT-BIN. In this case, the winning candidate [ves'ti] is chosen over [ves'ti:] because the lengthening of the final long vowel represents a violation of the higher-ranked principle FILL.10 (28)/vesti/ a. [ves'(ti)l Lb. [ves'(ti:)]
|
FILL
FT-BIN
*
*
As we know, the output [ves'ti:] 'dress' is also a well-formed word in Milanese. In this case (and in all cases of oxytones with final long vowels) we assume that the input forms have an underlying long vowel (cf. /vesti:/). In sum, we claim that the analysis presented here successfully accounts for the productive vowel length alternations in Milanese. Residual phenomena such as the set of cases where word-final in/ or /l/ are preceded by short vowels can just be obtained through lexical marking (i. e., /n/ or ill will not be mora-bearing). Thus, the vowel length difference between ['pa:] 'bread' and ['pan] 'cloth', both underlyingly /pan/, will come from the fact that the former has an underlying mora in the lexical representation, while the latter does not (cf. [pa'ncos] 'cloth.dim' and [pana'tεl] 'bread.dim' (Ch: 1064)). Similarly, the unusual presence of short vowels before word-final voiced obstruents (['fred] 'cold', ['gœb] 'hunchback', ['mez] 'half) will be treated in a similar way: such consonants will be specifically marked not to license a moraic unit. 4.
Conclusion This article has claimed that the productive alternation of long and short vowels in Milanese can be successfully accounted for by resorting to three basic constraints on representational well-formedness. A universal binarity constraint on 10 We do not deal with the possible candidate [Vesti] (an ouput which satisfies both the FT-BIN and FILLrequirements) because it would involve stress assignment constraints.
270
PILAR PRIETO I VIVES
foot structure (FT-BlN) and two other constraints: HNUC (the tendency for nuclei to increase their sonority level) and FILL (the demand that the input structure is respected). These three constraints, which have been shown to be active in different languages, are the key factors in deciding the optimal outputs and in successfully predicting vowel length distribution in this dialect. In contrast with previous proposals based on a process of Bimoraic Enforcement (Montreuil 1990), I believe that the present analysis has the advantage of accounting for the vowel length distribution in a way that is crosslinguistically motivated. The analysis offers a principled motivation for the fact that vowels are only lengthened in final syllables. In particular, the FT-BIN constraint explains why vowels are lengthened in this context and stay short when in penultimate or antepenultimate positions. The difference between the behavior of vowels in penultimate vs. word-final syllable positions has been reported in other northern Italian dialects (Prieto 1994). For example, in the Lombard dialect spoken in Valle Spluga stressed vowels are long before word-final /r/ plus consonant clusters (cf. ['tc:rt] 'twisted.ms', ['kc:rp] 'body', ['ta:rt] Mate', [inVε:rrn] 'winter') and stay short in penultimate syllables (cf. ['porta] 'door', ['barba] 'beard', ['kwerta] 'covered.fs'; data are taken from Zahner 1989). Similarly, in the Canton Ticino dialect of Sonogno there is also a tendency to lengthen vowels before nasal consonants only in word-final syllables (['bja:nk] 'white.ms', [ve:nt] 'wind', ['gra:nt] iarge.ms', ['sa:nt] 'saint.ms' vs. [ma'renda] 'snack', ['djanda] 'acorn', ['santa] 'saint.fs' (data come from Salvioni 1886, Saunders 1975, 1976, Jaberg & Jud point 42). In the cases above, the foot binarity constraint, FT-BIN, is ranked higher than the syllable binarity constraint (i. e., lengthen every stressed syllable), that is, optimization at the foot level has priority over the optimization at the syllable level. Optimization at one level in some cases applies at the expense of another level: the evolution of French syllable structure, for example, represents an instance of syllable structure becoming more complex due to the pressure to optimize different foot types (Jacobs 1992). In the case of Milanese we have also witnessed the competition and interplay between two different optimization requirements: on the one hand, the pressure to optimize foot structure (FT-BIN) and, on the other, the pressure to optimize the prosodic representation of segments, as more sonorous segments tend to increase their prosodic weight. REFERENCES Archangeli, Diana & Douglas Pulleyblank. 1994. Grounded Phonology. Cambridge, Mass.: MIT Press.
VOWEL LENGTHENING IN MILANESE
271
Cherubini, Francesco. 1839. Vocabolario Milanese-Italiano. Milano: Regia Stamperia. (Ristampa Anastatica Rusconi Libri). Chierchia, Gennaro. 1986. "Length, Syllabification and the Phonological Cycle in Italian". Journal of Italian Linguistics 8.5-34. Gökcen, A. 1990. "Delia quantità nel vocalismo milanese". Quaderni d'italianistica 2.238-264. Hualde, José Ignacio. 1990. "Compensatory Lengthening in Friuiian". Probus 2.31-46. Jacobs, Haike. 1992. "The Interactions between Syllable Structure and Foot Structure in the Evolution from Classical Latin to Old French". Theoretical Analyses in Romance Linguistics ed. by Christiane Laeufer & Terrell A. Morgan, 55-79. Amsterdam & Philadelphia: John Benjamins. Jaberg, Karl & Jakob Jud. 1928-40. Sprach- und Sachatlas Italiens und der Südschweiz. Zofingen: Ringier. Lepschy, Giulio C. 1978. Una fonologia milanese del 1606: Il Prissian da Milan della parnonzia Milanesa, Bologna: II Mulino. McCarthy, John & Alan Prince. 1993. Prosodic Morphology L Technical Reports of the Rutgers University Center for Cognitive Science 3. Merlo, Clemente. 1960. "I dialetti lombardi". L'Italia Dialettale 24.1-12. Mester, Armin. 1994. "The Quantitative Trochee in Latin". Natural Language and Linguistic Theory 12.1-62. Montreuil, Jean-Pierre Y. 1990. "Length in Milanese". New Analyses in Romance Linguistics ed. by Dieter Wanner & Douglas A. Kibbee, 36-47. Amsterdam & Philadelphia: John Benjamins. Nicoli, Franco. 1983. Grammatica milanese. Busto Arsizio: Bramante Editrice. Pavia, Luigi. 1928. Sulla parlata milanese e suoi connessi. Nuovi studi fonicografici, filologici, storici, comparativi. Bergamo: Presso L Autore. Prieto, Pilar, 1993. "Historical Vowel Lengthening in Romance: The Role of Sonority and Foot Structure". Issues and Theory in Romance Linguistics ed. by Michael Mazzola, 87-108. Washington D.C.: Georgetown University Press. . 1994. Vowel Lengthening in Romance: A Case for Segmental and Prosodic Optimization. Ph.D. dissertation, University of Illinois at UrbanaChampaign. Prince, Alan. 1990. "Quantitative Consequences of Rhythmic Organization". CLS. Papers from the Parasession on the Syllable in Phonetics and Phonology ed. by Karen Deaton, Manuela Noske & Michael Ziolkowski, 134. Chicago: Chicago Linguistic Society.
272
PILAR PRIETO ï VIVES
Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in a Generative Grammar. Ms., Rutgers University and University of Colorado, Boulder. Repetti, Lori. 1992. "Vowel Length in Northern Italian Dialects". Probus 4.155182. . 1994. "Degenerate Syllables in Friulian". Linguistic Inquiry 25.186-193. Saltarelli, Mario. 1983. "The Mora Unit in Italian Phonology". Folia Linguistica 17.7-24. Salvioni, Carlo. 1884. Fonetica del dialetto moderno delia cìttà di Milano. Roma: Ermano Loescher. . 1886. "Saggio intorno ai dialetti di alcune vallate all'estremità settentionale del Lago Maggiore". Archivio Glottologico Italiano 9.188-260. . 1975. "Fonetica e morfologia del dialetto milanese (ed. by Dante Isella)". L'Italia Dialettale 38.1-47. Sanga, Glauco. 1984. Dialettologia lombarda: lingua e culture popolari. Pavia: Dipartimento di Scienza delia Letteratura dell'Università. . 1988. "La lunghezza vocalica nel milanese e la coscienza fonologica dei parlanti". Romance Philology 3.290-7. Saunders, Gladys E. 1975. A Comparative Study of the Gallo-Italian Dialects. Ph.D. dissertation, University of Michigan, Ann Arbor. . 1976. "Evolution of Vowel Length in Gallo-Italian". General Linguistics 18.14-27. Steriade, Donca. 1990. "Moras and Other Slots". Proceedings of the First Meeting of the Formal Linguistics Society of Midamerica ed. by Denise Meyer, Satoshi Tomioka & Leyla Zidani-Eroglu, 254-280. Madison, Wisc: University of Wisconsin, Madison. Tuttle, Edward F. 1991. "Nasalization in Northern Italy: Syllabic Constraints and Strength Scales as Developmental Parameters". Rivista di Linguistica 3.2392. Zahner, Gérard. 1989. Il dialetto delia Val San Giacomo (Valle Spluga). Pubblicazioni delia Universite Cattolica del Sacro Cuore. Zec, Draga. 1988. Sonority Constraints on Prosodie Structure. Ph.D. dissertation, Stanford University.
UNEVEN OR MORAIC TROCHEES? EVIDENCE FROM EMILIAN AND ROMAGNOL DIALECTS*
LORI REPETTI State University of New York, Stony Brook
Left foot. Right foot. Feet. Feet. Feet. Oh, how many feet you meet! The Foot Book by Dr. Seuss
0.
Introduction There has been much discussion in the recent literature as to the exact nature of the metrical foot in quantity-sensitive trochaic systems. Some argue that this type of foot consists of two moras ('moraic trochee') which may be part of a single heavy syllable (H) or a sequence of two light syllables (LL). In a different approach, this type of foot defines three groupings of syllables ('uneven trochee'): a single heavy syllable (H), or two syllables, the second one being light: (LL) or (HL). The two types of trochaic feet characterize the sets of syllables in
(1)
a. b.
moraic trochee: (H), (LL) uneven trochee: (H), (LL), (HL)
I would like to thank Mark Aronoff, Christina Bethin, Ellen Broselow, John Hajek, and Peter di Ludlow, as well as audiences at the Incontro di Grammatica Generativa XXIV, the Firenze, and the Università di Padova for helpful comments and suggestions. I would also like to thank my informants for their time and invaluable assistance. An earlier versions of this work is published in Repetti (1998). 1 Most recent studies acknowledge the existence of all foot structures, although some are Università considered 'less harmonic'. For example, Prince (1992) sets up the following scale: (H), (LL) » (HL) » (L).
274
LORI REPETTI
Both models predict the same footing of a heavy syllable, (H), and of a sequence of two light syllables, (LL). The controversy involves a heavy-light sequence of syllables. According to the moraic trochee analysis, these syllables are parsed as a monosyllabic (H) foot followed by a single light syllable. According to the uneven trochee analysis, these syllables together form a single foot: (HL). To see which model (the moraic trochee or the uneven trochee) can best account for the metrical processes in languages with quantity-sensitive trochaic feet, linguists have examined stress patterns and other metrical processes such as shortening, syncope, etc. After examining various metrical patterns, some have concluded that the moraic trochee is the best way to represent the metrical foot in quantity sensitive trochaic systems (Hayes 1995, Kager 1993, Mester 1994, Prince 1992, Prince & Smolensky 1993, and references therein), while others argue for the uneven trochaic foot (Bullock 1995, 1998, Jacobs 1990, 1997, 2000, Kenstowicz 1994, Lahiri, et al. 1999, Mazzola 1997, Thornton 1996, and references therein). In this article, I provide information on metrical patterns observed in the dialects of northern Italy, and specifically those of the Emilia-Romagna region (henceforth, E-R), to see if the uneven trochaic foot or the moraic trochaic foot can best account for these processes.2 Proof that these dialects have a quantity sensitive trochaic foot comes from stress patterns. Stress assignment in the E-R dialects is only partly predictable. Stress can fall either on the final or penultimate syllable. (Antepenultimate stress is extremely rare.) If a word ends in a long vowel or a superheavy syllable (CVCC or CV:C), stress invariably falls on the final syllable. If the word ends in a CV or a CVC syllable, stress falls either on the final or penultimate syllable. We can account for the patterns in (2a) with either the moraic trochee or the uneven trochee model: a trochaic foot is constructed at the right edge of the word, and stress is assigned to therightmostfoot. The patterns in (2b) consist of words ending in a CVC syllable. Whichever model of the trochaic foot we adopt (moraic or uneven), words with this configuration should have final stress. Therefore, words with penultimate stress ending in a CVC syllable must be marked in the lexicon for irregular stress assignment. Stress assignment is not predictable in words ending in a light syllable (2c): both the moraic trochee model and the uneven trochee model predict penultimate stress in these cases, so we can conclude that words ending in a light syllable with final stress must be marked in the lexicon for stress assignment.
2 See Hajek (1997) for a description of the main characteristics of these dialects, and Repetti (to appear a) for a study of their prosodic structures.
UNEVEN OR MORAIC TROCHEES?
(2) a.
b.
final stress attested σJ.CV:# σCV:C# predicted σ.CVCC# attested σ.CVC# predicted attested σ.CV# not predicted (exceptional)
275
penultimate stress not attested notpredicted attested (only in anaptyxis dialects) not predicted (exceptional) attested predicted |
The information in (2) shows that stress patterns in the E-R dialects can best be accounted for if we assume that these dialects have a quantity sensitive trochaic foot. However, we cannot decide between the uneven trochee model and the moraic trochee model because both make the same predictions about stress assignment and account for the patterns in (2) equally well. In the following paragraphs I show that in some dialects we find constraints that apply to heterosyllabic adjacent consonants within the same foot; a foot with a (CVC.C V) shape is, by definition, an uneven trochee (§ 1). I also show that some words are characterized by the presence of a metrical template which is part of the lexical representation of the word, and the template is an uneven trochee (§2). 1. 1.1
Phonological Processes and the Uneven Trochee Uneven Trochees in the Dialect of Coli In the dialect of Coli word-internal onset clusters and coda clusters are prohibited, and only a restricted set of clusters is permitted word-initially and word-finally (for example, [grps] 'big' and [u:rs] 'bear'). 3 An unacceptable word-final cluster is eliminated through the insertion of an epenthetic vowel wordfinally ('paragoge') or in penultimate position ( 'anaptyxis'). (3) Coli paragoge ['fe:rmә] 'still' ['ne:rvә] 'nerve' [ka.rgә] loaded' [sa:rzә] 'willow' ['te:vdә] 'tepid' [tœzgә] 'poisonous'
anaptyxis ['ma:gar] 'thin' ['kwa:tәir] 'four' [pe.gәr] 'sheep(pi)' [veday] 'widower' ['kwe:daz] fifteen' F3u:vsn] 'young'
3 Coli is in the province of Piacenza. Data come from Jaberg & Jud (1928-1940) and field research.
276
LORI REPETTI
The position of the epenthetic vowel (in final or penultimate position) is determined by the sonority contour of the word-final consonant cluster. If the cluster has falling sonority, the two consonants may be adjacent, and the epenthetic vowel is inserted at the end of the word. If the cluster has rising sonority, the two consonants may not be adjacent, and the epenthetic vowel is inserted between them. To account for the data, we must posit the following constraints. (4) SonCon - Sonority Contour - a coda must be more sonorous than the following onset. This constraint on the sonority contour of adjacent consonants falls out from general principles of syllabification: an intervocalic consonant cluster of rising sonority will form an onset cluster, while an intervocalic consonant cluster of falling sonority will be heterosyllabic. (See Vennemann 1988.) (5) Align-RTFt, PrWd) - the right edge of the prosodic word must be aligned with the right edge of the main foot. (6) Align-R(C, PrWd) - the right edge of the prosodic word must be aligned with a consonant. (This is a constraint of the Anchor family requiring the prosodie word to end in a consonant. )
The constraints above allow us to account for the patterns exemplified in (3). (7) coda cluster of falling sonority /ferm/ Align('Ft) SonCon a. ('fe:r.ma)
*! *!
b. ('fe:r)m9 c. ('fe:rә)m
|
Align(C)
!
* *
(8) coda cluster of rising sonority /magr/
,
b. ('ma:g)ra c. ('ma:.gә)r
SonCon
*! *!
a. ('ma:g.rә) |
Align('Ft)
* *
Align(C)
* *
The SonCon constraint does not apply to any of the candidates in (7) since in (7a) and (7b) coda /r/ is more sonorous than the following onset /m/. SonCon
277
UNEVEN OR MORAIC TROCHEES?
does apply to candidates (8a) and (8b) since coda /g/ is less sonorous than the following onset /r/.4 Now consider the following data with an unacceptable word-initial consonant cluster. (9)
a. b.
/dman/ /dman matεn/
> >
[әd'man] [dema'tεn)5
'tomorrow' 'tomorrow morning'
The question arises as to why the SonCon constraint does not apply to (9a), [ad.'man], eliminating it from the competition (since coda /d/ is less sonorous than onset /m/), and resulting instead in the output * [da.'mal)]. In the tableau in (10), we see that this is precisely what the constraints, as formulated above, would predict. [(l0)/dman/ a. (әd)('man)
SonCon |
Align-R('Ft, PrWd)
Align-R(CPrWd)
*!
b. (dә)('man) ©
We need to refine the SonCon constraint so that it applies to forms like * ['ma:g.rә], eliminating them from the competition, but not to foms like [ad.'mag]. We can do this by limiting the domain of application of this constraint to heterosyllabic consonants within the same trochaic foot. In this way, the SonCon constraint does not apply if the two consonants are in different feet. (11)
SonCon-2 - a coda must be more sonorous than the following onset in the same foot.
This new formulation of the SonCon constraint (SonCon-2) does not apply to either of the candidates in (10): in (a) the two consonants are in different feet, and in (b) the two consonants are not adjacent. Therefore, the tableau in (10) does not give us a result: candidate (10b) must be eliminated from the competition. We can do this by introducing the low ranking Foot Binarity (FtBin) constraint which bars feet that are not binary at some level of representation (bimoraic or disyllabic). Given these constraints, the correct output is selected.
4
A candidate such as ['ma.grs] is ruled out because of a ban on word-internal onset clusters. There is also a full version of this phrase: [әdman) ma.'tεnl]. I will assume that the reduced form, [dema'tsi)], is derived by the deletion of the first of two (nearly) identical syllables. 5
278
LORI
REPETTI SonCon-2
(12)/dman/ a. (әd)('man) d.(dә)('man)
FtBin
|
*!
There is further evidence that the domain of application of SonCon-2 is the trochaic foot supporting the formulation of the constraint as given in (11) rather than (4). We have seen that [ad.'mag] does not violate SonCon-2; however, in (9b) we find [de.ma'tεn] and not *[әd.ma.ftsg] because the latter violates SonCon-2 and is eliminated from the competition.6 (13)/dman matεn/
SonCon-2
a. (әdma)('tεn) b. (әd)(ma)('tεn)
FtBin
*! *
c. (dә ma)('tεn) d. (dә)(ma)('tεn)
**
The data in (14) provide additional support of the formulation of SonCon as given in (11). 7 (14)
a. b.
/511: vn/ > /3u:vn + ct/ >
[3u:vәn] [3uv'nct]
'young (mas)' 'young (dim)'
In (14a) we find [('3u:.vi)n] and not *[05u:v.n9)] because of SonCon. In (14b) we find [(3uv)(nct)] which violates the SonCon constraint as originally formulated (without reference to foot structure), but it does not violate the SonCon-2 constraint which makes reference to foot structure.8 These data clearly show that SonCon must be formulated in such a way as to apply to forms like *[('3u:v.ng)] and *[(edma)('fë]g)] (eliminating them from the competition), but not to forms such as [(3uv)('not)] and [(ed)('mai))]. We can
6
We cannot attribute the choice of [dә.ma.'tsï)] over *(әd.ma.'teij] to a preference for an [(LL)('H)] configuration over an [(HL)('H)] configuration since we also find forms such as [әn. vu:d]/[dn.mld&ijL| 'nephew'/ '(dim.)'. The form |әn.vu.'dsεn] has an [(HL)('H)] configuration (and crucially does not violate SonCon), and it is selected over *[nә.vu.'dεn)jj with an [(LL)(!H)] structure. 7 The feminine form, [('ju:v.na)], violates SonCon. We do not find *[('3u:.va)na] because of the ban on words with antepenultimate stress. 8 The candidate *[(5u.v3)('not)] is eliminated because of a low ranking DEP violation.
UNEVEN OR MORAIC TROCHEES?
279
best do this by restricting the domain of application of SonCon to the metrical foot. (15) a. b. c. d
*[('3u:v.nә)] *[(edma)('t£i))] [(әdX'man))] [(5uv)('nct)]
SonCon (4) (no reference to foot structure) * * * *
SonCon-2 (11) (applies within domain of foot) * *
What we get for free from this analysis is that the revised SonCon constraint forces a particular configuration of trochaic feet: they must be uneven (and not moraic) since the first syllable of the foot ends in a consonant, i. e., it is heavy. The uneven trochaic model permits (HL) feet, while the moraic trochaic model does not. If, instead, we assume that trochaic feet are moraic trochees, we could no longer formulate the SonCon constraint to apply within a particular phonological domain. For example, there would be no explanation for the presence of wordinitial epenthesis in [әd.'man)], but not in *[3d.ma.'tsg]. 1.2
Uneven Trochees in the Dialect of Vediceto Vediceto is a town approximately 10 km from Coli.9 The dialects of these two towns are, for all intents and purposes, identical, specifically with regard to the patterns illustrated in (3). However, there is one significant difference: while nonhomorganic nasal clusters are permitted in the dialect of Coli, such clusters are not allowed in the dialect of Vediceto (Repetti to appear b). (16)
Coli ['gcrmdә] ['mangә] ['stomgә]
Vediceto ['gumәdl ['manag] ['stumәg]
'elbow' 'handle' 'stomach'
These data suggest that there is a constraint banning non-homorganic clusters in Vediceto. Given an input /gumd/, the Nasal Homorganicity constraint rules out a candidate *[('gum.d9)], and the form [('gu.m9)d] is selected. (17)
9
NasHom - Nasal Homorganicity - non-homorganic nasal clusters are not permitted
Vediceto is in the province of Piacenza. Data come from Zörner (1989) and field research.
280
LORI REPETTÎ
However, we do find non-homorganic nasal clusters pre-tonically: [әm.'te] 'you (pl) put', [әm.'dy.ra] 'thrashing9. Why doesn't the Nasal Homorganicity constraint rule out these forms? As with the SonCon constraint, it appears that the Nasal Homorganicity constraint applies within the domain of the foot, the uneven trochee, (18)
NasHom-2 - non-homorganic nasal clusters are not permitted in the same foot
The revised Nasal Homorganicity constraint applies to forms like *[('gum.dә)], eliminating them from the competition, and resulting in the output [(gu.mә)d], but it does not apply to forms like [(9m)('te)] and [(9m)('dy.ra)] since the non-homorganic nasal clusters, /mt/ and /md/, are contained in different feet. The revised Nasal Homorganicity constraint makes sense only if we admit uneven trochaic feet in the foot inventory. If, instead, we assume that trochaic feet are necessarily moraic trochees, i. e., if (HL) feet are not permitted, we could no longer formulate the Nasal Homorganicity constraint to apply within a particular phonological domain so that it applies to *[('gum.da)], but not to [(9m)('te)]. 2. 2.1
Epenthesis and the Metrical Template Epenthesis Patterns The E-R dialects divide into two groups based on the position of the epenthetic vowel needed to syllabify word-final consonant clusters. 'Paragoge dialects' have epenthetic vowels that are paragogic (in final position), and 'anaptyxis dialects' have epenthetic vowels that are anaptyctic (in penultimate position).10 (I will not discuss here the 'mixed dialects' like Coli and Vediceto in which the position of the epenthetic vowel varies depending on phonotactic constraints. See§I.) (19)11 a. b.
paragoge dialects ['fe:rmә] ['ma-grә]
anaptyxis dialects ['fe:rәm] ['ma:gәr]
'still' 'thin'
Since the presence, position, and quality of the post-tonic vowel in (19) are predictable, we can assume it is epenthetic. Its presence is required because the words end in an unacceptable consonant cluster (determined independently for each dialect; see Repetti 1995, 1996), its quality is that of the epenthetic vowel 10 The post-tonic vowel in (19) is represented as schwa, but its quality varies from dialect to dialect. 11 The feminine forms for all dialects are the same: ['fe:rma], ['ma:gra].
281
UNEVEN OR MORAIC TROCHEES?
found in other contexts, and its position is predictable based on the constraints given below. Therefore, we can posit an input form which does not contain a posttonic vowel, and a series of constraints which determine its presence, position, and quality. The constraints that allow us to derive the correct outputs are the alignment constraints given in (5) and (6), plus a high ranking constraint banning deletion of material from the input: MAX. The key difference between the paragoge dialects and the anaptyxis dialects is the relative ranking of the two alignment constraints. In the paragoge dialects Align-R('Ft, PrWd) is ranked higher than Align-R(C, PrWd), while in the anaptyxis dialects Align-R(C, PrWd) is ranked higher than Align-R('Ft,PrWd). (20) paragoge dialects /ferm/ | a. ('fe.r.m) ә b. ('fe:r)mә c. ('fe:.rә)m d.('fe:r) (21) anaptyxis dialects /ferm/ | a. ('fe:r.nә) b. ('fe:r)m3 c. ('fe:.rә)m d ('fe:r)
Align-R('Ft, PrWd)
MAX
*! *!
*! MAX
Align-R(C, PrWd)
* *
! Align-R(C,PrWd)
*! *!
1 Align-R('Ft, PrWd)_
* *
*!
The grammars in (20) and (21) allow us to account for all the data with an unacceptable word-final cluster, as well as the absence of certain types of forms, such as the following. (22)
2.2
paragoge dialects *[fe:rәm] * ['ma: gәr]
anaptyxis dialects *[fe:rma] * ['ma: grә]
The Trochaic Template There is a group of words that is exceptional in both the paragoge dialects and the anaptyxis dialects. The forms in (23a) are exceptional in the paragoge dialects because of the presence of a final vowel. In these dialects, unstressed final
282
LORI REPETTI
vowels are strictly prohibited unless they are morphemes or epenthetic vowels (see (19)). Why do we find ['ka:rgә] instead of *[ka:rg] (since [la:rg] is an acceptable form)? The forms in (23a) are exceptional in the anaptyxis dialects because of the presence of an unstressed vowel: in these dialects, syncope is productive. Why do we find ['ka:rәg] instead of *[ka:rg] (since [la:rg] is an acceptable form)? Furthermore, these words exhibit the irregular stress pattern described in (2b). Why do we find ['ka:rәg] instead of *[ka'rәg]? (Note that lexical schwa can be stressed: ['pәs]/['pәska] 'fish'/'s/he fishes'.) (23) a. b. (24)
paragoge dialects 'ka:rgә la:rg
anaptyxis dialects 'ka:rәg la:rg
'loaded' 'wide'
exceptional features of words in (23a): ['ka:rgә J (paragoge dialects) • no apocope • no syncope ('ka:rәg] (anaptyxis dialects) • irregular stress assignment
We can account for the forms in (23a) either by assuming that the vowel in the final syllable is lexical or by assuming it is epenthetic. If we assume that the vowel in the final syllable is lexical, then these words must be marked so as to block apocope in the paragoge dialects and syncope in the anaptyxis dialects, and they must be marked for irregular stress assignment in the anaptyxis dialects. If we assume that the vowel is epenthetic, these irregularities disappear. We do not find apocope in the paragoge dialects because the final vowel is epenthetic, and we do not find syncope in the anaptyxis dialects because the unstressed vowel is epenthetic. In addition, we do not need to mark these words for exceptional stress assignment in the anaptyxis dialects because the vowel in the final (heavy) syllable is epenthetic. A high-ranking Head-DEP constraint bans stress assignment on epenthetic vowels (see Alderete 1999): the final vowel in these words is epenthetic, so stress predictably falls on the rightmost lexical vowel. According to this analysis, stress assignment is now predictable for words in category (2b). Further evidence in support of the analysis of the vowel in the final syllable in (23a) as epenthetic comes from that fact that both the quality and the position of that vowel is consistent with the quality and position of an epenthetic vowel (compare (19) and (23a)). If we assume that the vowel in the final syllable in (23a) is epenthetic, we need to account for the reason that an epenthetic vowel is needed in that position.
UNEVEN OR MORAIC TROCHEES?
283
In other words, why not say *[ka:rg]? I would like to suggest that the input consists of the melody /karg/ plus a disyllabic metrical template ( σ σ ) 1 2 The monosyllabic input must become disyllabic in the output in order to satisfy the requirements of the template, so an epenthetic vowel is added. 13 The quality and position of that vowel are determined by the constraint hierarchy described above in §2.1. (25) paragoge dialects /karg/, (σσ) | MAX Align-RfR, PrWd) a. ('ka:r)g b. ('ka:r,gә) c. ('ka:r)gә d. ('ka:.ra)g
Align-R(C, PrWd)
*
[*!(σσ)
* *|
*
*!
|
(26) anaptyxis dialects /karg/, ( ) i 1 MAX \ Align-R(C, PrWd) Align-R('Ft, PrWd) a. ('ka:r)g | *! (σσ) b. ('ka:r.ga) c. (fka:r)ga ! d. ('ka:.P3)g
j
*
*! *!
* *
Evidence that the template is a disyllabic foot (i, e., a trochee) and not simply a disyllabic minimum requirement comes from words with a disyllabic input like /bazilg/ 'basil' plus a template. If the requirement were simply that the output be disyllabic (with no reference to foot structure) (27a), we would expect an output *[ba'zi:lg]. This word is disyllabic, has stress (as predicted) on the superheavy final syllable, and ends in an acceptable coda cluster. However, this is not the form found in most E-R dialects. It is found only in those dialects with forms like [ka:rg] in which the template is no longer part of the input (see footnote 13). The
12 The presence of templates in Romance languages is well attested: Davis & Napoli (1994), Lipski (1991), Pharies (1986), Thornton (1996), and Benincà & Vanelli (1975) who do not discuss 'templates' per se, but analyze certain patterns using the notion of templates. See Inkelas (1994) for an OT analysis of exceptional stress patterns in Turkish. Inkelas argues that a template based account of the data is superior to one based on output constraints, and concludes that there is no principled reason not to posit metrical templates in the input. 13 It should be noted that in some E-R dialects the output is [ka.rg]. In these dialects, the disyllabic template has been eliminated from the input. The input form, /karg/, does not have a template associated with it, so the output form is monosyllabic jka:rg].
284
LORI REPETTl
actual output forms are [ba'zi:lәg]/[ba'zi:lgә] proving that the output must contain a disyllabic foot and not simply be a disyllabic word (27b). 14 (27) a. /bazilg/, crcr (template consists of disyllabic minimum requirement) *[ba'zi:lg] found only in dialects with no template (see fn. 13) b. /bazilg/, ('σσ*) (template consists of disyllabic trochaic foot) [ba'zi:lgә]/[ba'zii:lәg] output forms
If the metrical template is a disyllabic trochee, and not just a minimum disyllabic requirement, is there any evidence whether the trochaic template is a moraic trochee or an uneven trochee? The disyllabic template must be an uneven trochee and not a moraic trochee as shown by the tableau in (25). The output form in (25b) contains an uneven trochee, while the most similar form with a moraic trochee in (25c) was not selected. In the next section we will see why some words are characterized by the presence of a metrical template — /karg/, ('crcr) — while others are not — /larg/. The metrical template is a means of encoding information contained in the original Latin form. Even though the melody has changed, the metrical structure has not. 2.3
Latin Proparoxytones Words like those in (23a) have something else in common: they all derive from Latin proparoxytones (words with antepenultimate stress). 15 E-R words deriving from Latin proparoxytones, regardless of whether or not the final cluster would otherwise be acceptable, have penultimate stress (28).
14
We find a form like *['ba:zilg] because penultimate stress is not permitted in words ending in a superheavy syllable, nor do we find *['ba:zilgai]/*[,ba:zilag] because antepenultimate stress is not permitted. 15 In many Romance varieties, words deriving from Latin proparoxytones have a different metrical structure from words deriving from Latin paroxytones. See Rohlfs (1966) for data from other northern Italian dialects (Piedmontese, Ligurian, Lombard), and Cremona (1955) for an argument that the post-tonic vowel in Gascon words deriving from Latin proparoxytones is etymological (and not epenthetic).
UNEVEN OR MORAIC TROCHEES? (28) Latin CARICU SALIGE TEPIDU QUATTUORDECI QULNDECI TO[ks]ICU NU:BULU
anaptyxis dialects (Gazzoli16) [kar.әg] ['sa:lәzj [te:vәd] [kwa'torrctez] [kwëxtez] ['tœsәk] ['ny:v9r]
285
paragoge dialects (Busana17) [karge]
fsalje] [tevde] [kwa'tor^e] ['kwin%e] ['tcjge] [nyvle]
loaded' 'willow' 'tepid' 'fourteen' 'fifteen' 'poisonous' 'cloud'
We know that in these dialects all Latin words underwent apocope, so at some point in their history all E-R dialects had the following forms (after intervocalic voicing of obstruents and reduction of unstressed, non-low vowels to [a]). (See Schurr 1974) (29)
LARGU > 'larg CARICU > 'karәg
In order to retain stress on the penultimate syllable in ['karәg], metrical information had to be included in the lexical representation of the word. So the template originated as a means of indicating an irregularly stressed syllable. In the paragoge dialects, a form like ['karәg] was restructured as ['karg9], possibly because of the effects of the high ranking constraint Align-R('Ft, PrWd) which selects [('kargә)] over [('karә)g]. At that point the E-R dialects divided into the two familiar groups: paragoge dialects with ['ka:rg9] and anaptyxis dialects with ['kairøg]. 3.
Conclusions The data from the E-R dialects provide support for the hypothesis that the trochaic foot is uneven and permits the structures in (lb). Evidence comes from many aspects of the grammars of these dialects. First we saw that in some dialects (Coli and Vediceto) constraints apply to heterosyllabic consonants in the same foot. The best way to describe this domain is as a trochaic foot which, by definition, is uneven since the first syllable contains a coda consonant. We also saw that some words are characterized by the presence of a metrical template as part of the lexical input. It was shown that the template is a disyllabic foot (and not 16
Gazzoli is in the province of Piacenza. Data come from field research. Busana is in the province of Reggio Emilia. Data come from the unpublished linguistic archives of the Atlante Linguistico Italiano housed at the University of Torino, Italy. 17
286
LORI REPETTI
a disyllabic minimum requirement), and the foot is an uneven (not. moraic) trochee. The template is satisfied by epenthesizing a vowel in post-tonic position, and the quality and position of that vowel is determined by the constraint hierarchy. This analysis also shows that the input form of some words can contain information not only about the string of phonemes but about metrical structure as well.
REFERENCES Alderete, John. 1999. "Head Dependence in Stress-Epenthesis Interaction". The Derivational Residue in Phonological Optimality Theory ed. by Ben Hermans & Marc van Oostendorp, 29-50. Amsterdam & Philadelphia: John Benjamins. Benincà, Paola & Laura Vanelli. 1975. "Morfologia del verbo friulano: il presente indicativo". Lingua e contesto 1.1-62. Bullock, Barbara E. 1995. "The Uneven Trochee in French". Rivista di linguistica 7.273-292. 1998. "The myth of equivalence: Where two lights do not make a long". Theoretical Analyses on Romance Languages: Selected Papers from the 26th Linguistic Symposium on Romance Languages (LSRL XXVI) ed. by José Lema & Esthela Trevino, 53-70. Amsterdam & Philadelphia: John Benjamins. Cremona, J. 1955. "Le problème des voyelles d'appui dans la réduction des proparoxytons en Gascon". VII congreso internacional de lingüistica românica (tomo II) ed. by A. Badía, A. Griera, & F. Udina, 673-678. Barcelona. Davis, Stuart & Donna Jo Napoli. 1994. A Prosodic Template in Historical Change: The Passage of the Latin Second Conjugation into Romance, Torino: Rosenberg & Sellier. Hajek, John. 1997. "Emilia-Romagna". The Dialects of Italy ed. by Martin Maiden & Mair Parry, 271-278. London: Routiedge. Hayes, Bruce. 1995. Metrical Stress Theory. Chicago: University of Chicago Press. Inkelas, Sharon. 1994. "Exceptional Stress-Attracting Suffixes in Turkish: Representations vs. the Grammar". Paper presented at the Workshop on Prosodic Morphology, Utrecht University, 1994. Jaberg, Karl & Jakob Jud. 1928-1940. Sprach- und Sachatlas Italiens und der Südschweiz (8 vols.). Zofingen: Ringier. Jacobs, Haike. 1990. "On Markedness and Bounded Stress Systems". The Linguistic Review 7.81-119. . 1997. "Latin Enclitic Stress Revisited". Linguistic Inquiry 28.648-661.
UNEVEN OR MORAIC TROCHEES?
287
. 2000. "The Revenge of the Uneven Trochee: Latin Main Stress, Metrical Constituency, Stress-Related Phenomena and OT". Analogy, Levelling and Markedness: Principles of Change in Phonology and Morphology ed. by Aditi Lahiri. Berlin: Mouton de Gruyter. Kager, René. 1993. "Alternatives to the Iambic-Trochaic Law". Natural Language and Linguistic Theory 11.381-432. Kenstowicz, Michael. 1994. "On Metrical Constituents: Unbalanced Trochees and Degenerate Feet". Perspectives in Phonology ed. by Jennifer Cole & Charles Kisseberth, 113-131. Stanford: CSLI. Lahiri, Aditi, Thomas Riad, & Haike Jacobs. 1999. "Diachrony". Word Prosodic Systems in the Languages of Europe ed. by Harry van der Hulst, 335-422. Berlin: Mouton de Gruyter. Lipski, John M. 1991. "Metathesis as a Template-Matching: A Case Study from Spanish". Folia Linguistica Historica 12.127-145. Mazzola, Michael L. 1997. "From Stress in Latin and Romance to Lexical Stress in Italian". Certamen Phonologicum III: Papers from the Third Cortona Phonology Meeting, April 1996 ed. by Pier Marco Bertinetto, et al., 229-246. Torino: Rosenberg & Sellier. Mester, R. Armin. 1994. "The Quantitative Trochee in Latin". Natural Language and Linguistic Theory 12.1-61. Pharies, David A. 1986. Structure and Analogy in the Playful Lexicon of Spanish. Tübingen: Max Niemeyer. Prince, Alan. 1992. "Quantitative Consequences of Rhythmic Organization". CLS 26: Papers from the 26th Regional Meeting of the Chicago Linguistic Society: The Parasession on the Syllable in Phonetics and Phonology (vol. 2) ed. by Michael Ziolkowski, Manuela Noske, & Karen Deaton, 355-398. Chicago: Chicago Linguistic Society. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in a Generative Grammar. Ms., Rutgers University and University of Colorado, Boulder. Repetti, Lori. 1995. "Variazione nella sillabificazione: il caso dei dialetti emiliani e romagnoli". Rivistaitaliana di dialettologia 19.41-56. . 1996. "Syllabification and Unsyllabified Consonants in Emilian and Romagnol Dialects". Aspects of Romance Linguistics: Selected Papers from the Linguistic Symposium on Romance Languages XXIV, March 10-13, 1994 ed. by Claudia Parodi, et al., 373-382. Washington, D. C : Georgetown University Press. . 1998. "Uneven Trochees in Latin: Evidence from Romance Dialects". University of Venice Working Papers in Linguistics 8.95-119.
288
LOW REPETTI
. to appear a. The Prosodic Structures of Northern Italian Dialects Amsterdam: Kluwer. . to appear b. "Non-Homorganic Nasal Clusters in Northern Italian Dialects" Current Issues in Linguistic Theory: Selected Papers from the Linguistic Symposium on Romance Languages XXIX, Ann Arbor, April 8-11, 1999 ed. by Diana Cresti, et al. Amsterdam & Philadelphia: John Benjamins. Rohlfs, Gerhard. 1966. Grammatica storica delta lingua italiana e dei suoi dialetti: Fonetica. Torino: Einaudi. Schiirr, Friedrich. 1974. La voce della Romagna. Ravenna: Edizioni del Girasole. Thornton, Anna M. 1996. "On Some Phenomena of Prosodic Morphology in Italian: Accorciamenti, Hypocoristics and Prosodie Delimitation" Probus 8.81-112. Vennemann, Theo. 1988. Preference Laws for Syllable Structure and the Explanation of Sound Change. Berlin: Mouton. Zörner, Lotte. 1989. Die Dialekte von Travo und Groppallo. Wien: Der Österreichischen Akademie der Wissenschaften.
INDEX OF NAMES
A. Absalom, Matthew: 112, 114, 131 Agostiniani, Luciano: 192, 195, 196, 199,207 Albano Leoni, Federico: 131 Alderete, John: 282, 286 Alibert, Lois: 179, 187 Amery, Heather: 245, 253 Andalo, Adriana: 47, 56, 57 Anteil, Stephen: 78, 85 Archangeli, Diane: 59, 77, 85, 93, 101, 109,265,270 Ard, Joshua: 78, 85 Arvaniti, Amalia: 132 Ascoli, Graziadio Isaia: 3, 10, 60, 85 Avesani, Cinzia: 134 B, Baertsch, Karen: 213,235 Bafile, Laura: 47, 56, 57, 138, 140, 143, 161, 165, 201, 203, 206, 207 Bagemihl, Bruce: 225,235 Bais, Maurizio: 13,25,43 Bammesberger, Alfred: 234,235 Baroni, Marco: 5, 6, 13, 43 Basbøll, Hans: 150, 165 Battaglia, Salvatore: 201, 207
Battisti, Carlo: 179, 187, 211, 222, 235 Bearesi, Luigi: 221,235 Beckman, Jill: 57, 73, 74, 85 Belletti, Adriana: 10 Bender, Byron: 123,131 Benincà, Berendsen, Egon: 138, 165 Bertinetto, Pier Marco: 57, 115, 286 131, 152, 165,244,253 Bichelli, 57 Paola: 3, 10, 132, 283, Bickmore,Piro: Lee: 112, 131 Blasco Ferrer, Eduardo: 179, 180, 181, 184, 187 46, 47, 49,165 52, 53, Bolognesi, Roberto: 138, Bonet, 186, 187 Bonucci, Paola: 207 Booij, Eulàlia: 225, 237 Bosoni, Giorgio: 120, 121, 122, Geert: 127, 132 Bottiglioni, Gino: 174, 176, 180, 138, 141, 165, 223, 187 Broselow, Ellen: 20, 43, 44, 113, 117, 118, 129, 132 Brugmann, Karl: 170, 187 Bruni,
Francesco: 211, 221, 235
290 Bucci, Cataldo: 106, 109 Bullock, Barbara: 45, 46, 52, 57, 274, 286 Burzio, Luigi: 203, 206, 208 C. Calabrese, Andrea: 5, 7, 8, 10, 59, 50,61,68,69,71,72, 73, 76, 80, 85-86, 200, 208 Camilli, Amerindo: 150, 165 Capozzoli, Raffaele: 140, 165 Cardinaletti, Anna: 197, 208 Garnie, Andrew: 213, 235 Carpitelli, Elisabetta: 194, 208 Cartwright, Stephen: 245, 253 Casali, Roderic: 98, 109 Castellani, Arrigo: 1, 10, 143, 165 Castellani, Riccardo: 124, 125, 132 Castelli, Margherita: 207 Cennamo, Michela: 179, 187 Chapallaz, Marguerite: 118,132 Chen, Matthew: 121, 132 Chen, Su-I: 132 Cherubini, Francesco: 256, 271 Chierchia, Gennaro: 192, 201, 195, 208, 262, 271 Chomsky, Noam: 71, 86 Clements, George: 63, 86, 94, 97, 109, 170, 187, 213, 235 Cohn, Abigail: 170, 171, 187 Cole, Jennifer: 68, 86 Connell, Bruce: 132 Contini, Gianfranco: 144, 165 Cortelazzo, Manlio: 3, 10, 211, 235 Crawford, Penny: 35,44 Cremona, J.: 284, 286
INDEX OF NAMES
Davis, Stuart: 283, 286 de Lorenzo Tobolo, Elia: 217, 220, 236 De Mauro, Tullio: 1, 10 Dessau, H.: 144, 165 Diehl, Randy: 22, 35, 43, 44 D'Introno, Francesco: 89, 109 Dixon, Robert: 186, 188 Djamouri, R.: 78, 87 Donegan, Patricia: 170, 188 Downing, Laura: 225, 236 Dresher, B. Elan: 145, 165 Dressier, Wolfgang: 60, 86, 132, 138, 140, 165 Drigo, Marina: 210 Dulcibella, J. W.: 244, 254 Durand, Jacques: 132 E. Eckman, Fred: 212,236 Engstrand, Olle: 129,132 F. Fanciullo, Franco: 45, 47, 48, 49, 52, 56, 57, 192, 195, 208 Fanti, Renata: 184, 188 Farnetani, Edda: 23,43 Fava, Elisabetta: 116, 132 Féry, Caroline: 112, 132 Fiorelli, Piero: 150, 165, 254 Flechia, Giovanni: 163, 165 Formentin, Vittorio: 137, 140, 165 Fortina, Peppo: 218, 222, 236 Fowler, Carol: 134 Francescato, Giuseppe: 123, 131, 132 Frau, Giovanni: 123, 132 Frauenfelder, U. N.: 134
291
INDEX OF NAMES
G. Gaeta, Livio: 131 Ghini, Mirco: 159, 165 Giannelli, Luciano: 132, 192, 195, 196, 207, 208 Gioscio, Joseph: 179, 188 Giuliani, Francesca: 250, 254 Gokçen, A.: 256, 271 Goldsmith, John: 201, 208 Gorra, Egidio: 218, 221, 224, 236 Grammont, Maurice: 170, 184, 188 Grandgent, Charles: 244, 254 Grassi, Corrado: 3, 10 Greenberg, Marc: 123, 132 Guarnerio, Pier Enea: 156, 166 Guidi, Guido: 192, 193,208 H. Haiman, John: 132 Hajek, John: 111, 112, 114, 116, 122, 123, 125, 127, 128, 131, 132, 274, 286 Halle, Morris: 59, 63, 82, 86, 205, 207, 208 Hammond, Michael: 230, 236 Haraguchi, Shosuke: 206, 208 Hargus, Sharon: 145, 165 Harris, James: 205, 208 Hayes, Bruce: 5, 10, 18, 43, 66, 81, 86, 98, 109, 111, 112, 113, 133, 202, 207, 208, 274, 286 Heilmann, Luigi: 4, 10 Helsloot, Karijn: 201, 209 Hoenigswald, Henry: 170,188 Hoffman, Robert Joseph: 240, 254 Holtus, Günther: 133
Hualde, José: 6, 10, 13, 17, 18, 19, 22,43,59,86, 123, 133, 258, 265, 271 Hubbard, Kathleen: 113, 117, 118, 133 Huffman, Marie: 132 van der Hulst, Harry: 164, 166, 223, 236 Hume, Elizabeth: 63, 86, 97, 109, 170, 187 Hyman, Larry: 122, 133 I. Iannáccaro, Gabriele: 194,208 Iannucci, James: 47, 48, 57 Inkelas, Sharon: 56, 57, 138, 153, 166, 283, 286 Iverson, Gregory: 212,236
J. Jaberg, Karl: 188, 244, 254, 270, 271,275,286 Jacobs, Haike: 203, 204, 209, 270, 271,274,286-287 Jespersen, Otto: 213,236 Jetchev, Georgi: 131 Jud, Jakob: 188, 244, 254, 270, 271,275,286 K. Kager, René: 76, 86, 202, 205, 209, 242, 254, 274, 287 Kaisse, Ellen: 145, 166 Katamba, Francis: 132 Kattenbusch, Dieter: 241, 254 Kayne, Richard: 200,209 Kaze, Jeffrey: 68, 86
292 Keating, Patricia: 22, 27, 32, 43, 44 Kenstowicz, Michael: 19, 43, 52, 57, 131, 152, 166, 201, 209, 274, 287 Kingston, John: 35, 43 Kiparsky, Paul: 27, 43, 140, 166, 170, 172, 182, 186, 188, 203, 209 Kirchner, Robert: 56,57 Kisseberth, Charles: 27, 43, 153, 166 Kluender, Keith: 22,44 Kori, Shiro: 23, 43 Kuen, Heinrich: 156, 166 L. Labov, William: 32, 44 Ladefoged, Peter: 59,86, 113, 133 Lahiri, Aditi: 274, 287 Lauefer, Christiane: 215, 236 Lausberg, Heinrich: 179, 180, 188, 206, 209 Laver, John: 113, 133 Lehiste, Ilse: 32,44, 133 Lekach, A. Farmer: 212, 236 Lepschy, Anna Laura: 118,133 Lepschy, Giulio: 118, 133, 256, 260, 271 Lindau, Mona: 32, 44, 59, 77, 87 Lipski, John: 283, 287 Lisker, L.: 34, 44 Lombard, Alf: 175, 176, 188 Lombardi, Linda: 140,166 Loporcaro, Michele: 57, 60, 61, 64, 65, 66, 67, 68, 87, 131, 137, 138, 144, 150, 152, 153, 154, 155, 156, 163, 166-167, 192, 195, 198, 209
INDEX OF NAMES
Lüdtke, Helmut: 60, 87, 137, 167 Lutta, C. Martin: 219,236 M. Maddalon, Marta: 134 Maddieson, Ian: 23, 44, 59, 86, 113, 117, 118, 133 Magno Caldognetto, Emanuela: 116, 132 Maiden, Martin: 4, 5, 9, 10, 68, 87, 118, 133, 169, 171, 174, 178, 182, 186, 188 Malagoli, Giuseppe: 175, 188, 192, 209 Malkiel, Yakov: 183, 184, 188 Maraschio, Nicoletta: 132 Marcato, Carla: 211,235 Marchetti, Giuseppe: 133,162, 167 Marotta, Giovanna: 118, 134, 191, 195, 199,201,205,209 Martin, Jean-Baptiste: 249, 254 Mascaró, Joan: 156, 167 Mattioli, Antonio: 223, 236 Mazzola, Michael: 133, 274, 287 McCarthy, John: 20, 44, 97, 104, 109, 112, 133, 255, 265, 266, 271 Melillo, Michele: 89, 109 Menéndez Pidal, Ramón: 176,188 Merlo, Clemente: 176, 188, 256, 271 Mester, Armin: 52, 57, 94, 110, 255, 266, 271, 274, 287 Metzeltin, Michael: 133 Michacls, David: 131 Minichelli, Vincenzo: 249, 254 Monachesi, Paola: 138, 140, 167
INDEX OF NAMES
Montreuil, Jean-Pierre: 211, 256, 258, 262, 263, 264, 270, 271 Morelli, Frida: 212, 236 Morin, Yves-Charles: 113, 123, 133 Moser, Alessandro: 210 Muljacid, Zarko: 150, 167 N. Nagy, Naomi: 239, 240, 241, 243, 254 Napoli, Donna Jo: 283, 286 Nespor, Marina: 142, 143, 147, 149, 150, 151, 153, 157, 167, 192, 195, 199,201,206,209 Nicoli, Franco: 133, 256, 257, 258, 260, 262, 271 O. Ohala,John: 170, 189 P. Pace, A.: 137, 167 Padgett, Jaye: 212, 236 Paradis, Carole: 71, 87 Parry, Mair: 4, 10, 133 Pavia, Luigi: 119, 120, 124, 125, 133,256,257,271 Peperkamp, Sharon: 138, 140, 141, 143, 145, 147, 148, 151, 158, 167 Perkell, Joseph: 59, 87 Pernicone, Vincenzo: 201, 207 Pharies, David: 283, 287 Pieri, Silvio: 192, 209 Pittau, Massimo: 182, 189 Plangg, Guntram: 217, 236 Poggi Salani, Teresa: 132
293
Polgardi, Krisztina: 59, 87 Port, Robert: 35, 44 Posner, Rebecca: 170, 171, 172, 173, 187, 189 Prieto i Vives, Pilar: 133, 255, 260, 262, 270, 271 Prince, Alan: 20, 44, 53, 57, 74, 87, 112, 133, 255, 265, 266, 268,271,272,273,274,287 Prinzhorn, Martin: 132 Pulcini, Walter: 176, 189 Pulleyblank, Douglas: 59, 77, 85, 101, 109,265,270 R. Rennison, John: 132 Repetti, Lori: 1, 5, 6, 10, 13, 17, 19,20,21,22,26,44, 46, 58, 111, 123, 134, 163, 167, 243, 247, 254, 265, 272, 273, 274, 279, 280, 287-288 Reynolds, Bill: 243, 254 Riad, Thomas: 287 Rialland, Anne: 78, 87, 236, 224 Ricca, Davide: 118, 134 Rice, Keren: 213, 237 Rietveld, A. C. M.: 134 Ringger, Kurt: 133 Ritter, Nancy: 223, 236 Rizzolatti, Piera: 123, 134 Roca, Iggy: 205, 210 Rohlfs, Gerhard: 47, 49, 58, 60, 65, 87, 173, 174, 176, 178, 179, 189, 192, 210, 284, 288 Romito, Luigi: 115, 134 Ronjat, Jules: 176, 189 Rubach, Jerzy: 223, 225, 237
294
S. Saltarelli, Mario: 4, 10, 254, 262, 272 Salvioni, Carlo: 256, 262, 270, 272 Salza, Pier Luigi: 118, 134 Salzmann, Zdenek: 123,131 Sanga, Glauco: 119, 134, 256, 258, 259,261,262,272 Saunders, Gladys: 270,272 Savoia, Leonardo: 192, 195, 196, 198,208,210 Scalise, Sergio: 149, 168 Scerbo, Francesco: 138, 167 Schein Barry: 81, 87, 98, 110 Schiaffini, Alfredo: 144, 167 Schmidely, Jack: 181,189 Schmitt, Christian: 133 Schurr, Friedrich: 285, 288 Selkirk, Elisabeth: 9, 10, 49, 50, 56, 58, 212, 214, 237 Shaw, Patricia: 226, 237 Simone, Raffaele: 134 Sluyters, Willebrord: 150, 167, 201,204,210 Smolensky, Paul: 53, 57, 74, 87, 265, 268, 272, 274, 287 Sobrero, Alberto: 3, 10 Soffietti, James: 4, 11 Stampe, David: 147, 167, 170, 188 Stehl, Thomas: 93, 110 Steriade, Donca: 73, 74, 81, 87, 94, 96, 98, 110, 212, 237, 264, 267, 272 Stevens, Kenneth: 59, 86 Stewart, John: 59, 87 Svantesson, Jan-Olof: 78, 87
ÏÏNDEX OF NAMES
T. Tagliavini, Carlo: 220, 237 Taverdet, Gérard: 113,134 Tekavcic, Pavao: 60, 87, 173, 189 Telmon, Tullio: 3, 10 Thornton, Anna: 274, 283, 288 Togeby, Knud: 171, 186, 187, 189 Tranel, Bernard: 134 Truckenbrodt, Hubert: 212,237 Trumper, John: 114, 119, 124, 125, 134 Tuttle, Edward: 260, 272 U. Urbanczyk, Suzanne: 53, 57, 58 V. Valente, Vincenzo: 240, 241, 243, 244, 245, 254 Vanelli, Laura: 5, 6, 11, 13, 14, 17, 18, 19, 22, 23, 24, 27, 32, 36,44, 123, 134,283,286 Vaux, Bert: 59, 63, 77, 78, 79, 82, 86,88 Vayra, Mario: 115, 134 Vedovelli, Massimo: 132 Vennemann, Theo: 276, 288 Vergnaud, Jean-Roger: 205, 207, 208 Vignoli, Carlo: 178, 189 Vogel, Irene: 45, 46, 58, 116, 135, 142, 143, 145, 147, 149, 150, 151, 153, 157, 167, 168, 192, 195, 199, 209, 210
295
INDEX OF NAMES
w. Walsh Dickey, Laura: 57 Wanner, Dieter; 181, 189 van de Weijer, Jeroen: 164, 166 Westbury, John: 27, 44 Weston, Rosemary: 89, 109 Wolfe, Andrew: 63, 82, 86 Wright, Beverly: 22,44 Wunderli, Peter: 179, 189
Y. Yamamoto, S.: 13, 24, 25, 44 Yip, Moira: 170, 189,212,237 Young, Diane: 111, 135 Z. Zahner, Gérard: 270,272 Zannier, Irene: 210 Zee, Draga: 138, 145, 157, 168, 264, 267, 272 Zörner, Lotte: 279, 288
INDEX OF TERMS & CONCEPTS
A. Affricates: 15, 19, 20, 24, 25, 214, 219, 220, 222, 241 Assimilation: 81, 137, 142, 144, 146, 147, 148, 170, 171, 173, 176, 182, 195, 196, 197, 198 Vowel/Consonant Assim.: 7, 89,97 C. Catalexis: 203, 204 Clitics: 7, 9, 50, 52, 108, 137-143, 148-161, 163, 164, 169, 173186, 192, 197, 200, 201 Clitic Group: 7, 137, 138, 143, 151, 157, 161, 164 Cliticization: 7, 137, 139, 140, 153, 154, 157, 158, 164 Enclisis/Enclitic: 7, 137, 138, 140, 143, 155, 157, 163, 164, 175, 201 Clusters: 7, 14, 45, 56, 154, 155, 156, 158, 159, 163, 211-229, 231, 232, 234, 235, 260, 261, 265, 270, 275, 276, 277, 279, 280,281,283,284 Compensatory Lengthening: 14, 18, 22, 35, 36, 52, 112, 216, 265
Constraint-and-Repair Model: 8,60, 71 Constraints: 5, 6, 9, 19, 20, 25, 27, 28, 36, 52, 53, 54, 55, 60, 61,69,70,71,74,76, 77, 7985, 138, 147, 160, 195, 198, 202, 205, 206, 212, 229-232, 250, 255, 256, 263, 266-270, 275-283, 285, 286 D. Degemination: 3, 146, 147 Diphthong: 60, 67, 70, 81, 111, 114, 118, 119, 123, 128 Diphthongization: 8, 60, 6470, 80, 81, 83, 85, 89, 114, 115, 124 Dissimilation: 9, 169-174, 176, 177, 181-187 Double Prosody: 9, 46, 52, 53, 54, 57 E. Epenthesis: 7, 54, 139, 147, 154159, 163, 204, 213, 214, 215, 224, 226, 228, 229, 279, 280 Anaptyxis: 275, 280, 281, 282, 283, 285 Paragoge: 275, 280, 281, 282, 283, 285
298
INDEX OF TERMS & CONCEPTS
Extrametricality: 7, 161, 203, 206, 207
Unary: 207 see also Metrical Structure
F. Features: Advanced Tongue Root ([ATR]): 7, 8, 59, 60, 61, 63, 64, 66, 68, 69, 70, 74, 76-80, 8 4 , 8 5 , 9 3 , 9 4 , 9 6 , 9 7 , 9 9 , 109 Back: 8, 67, 68, 90, 91, 92, 93, 94, 97, 100, 101, 102, 109, 240 Dorsal: 90, 91,92, 94, 95, 96, 97, 98, 99, 100, 101, 102, 108, 109 High: 32, 59, 61, 64, 67, 68, 69,70,73,74,77,84, 90, 91, 92, 94, 95, 96, 97, 99, 109, 163, 240 Labial: 8, 60, 63, 64, 85, 90104, 106, 107, 108, 173, 241 Low: 8, 68, 69, 73, 77, 90-97, 102, 107, 108, 109, 240 Round: 60, 61, 64, 67, 72, 73, 74, 77, 79, 80, 84, 85, 90, 97, 98, 104 Feet: Binary: 6, 52, 53, 192, 202206, 242, 243, 248, 255, 266, 269, 270, 277 Degenerate:7, 192, 201, 202, 203 Iamb/Iambic Foot: 203-206 Ternary: 6, 205, 206 Trochee/Trochaic Foot: 6, 52, 203, 205, 206, 242, 243, 248, 250, 255, 266, 273, 274, 275, 277-281,283-286
G. Gemination: 8, 9, 45-49, 51, 52, 55,56,68, 114 Geminate Consonants: 3, 18, 23, 47, 45, 55, 91, 112, 146, 147, 241 Geminate Inalterability: 61, 81 L. Lexical Phonology: 140, 145, 153 Lexicon: 3, 7, 19, 45, 145, 194, 239, 248, 249, 250, 274 M. Marked(ness): 7, 59, 71-77, 193, 204, 206, 207, 214, 229, 239, 244,248-251,266 Marking Statements: 60, 61, 72-80, 84, 85 Metaphony: 59, 66, 67, 68, 70, 89, 92,99 Metrical Structure: 6, 7, 192, 199, 201,202,207,284,286 see also Feet Mora: 5, 18, 19, 20, 21, 22, 53, 54, 56, 111, 112, 113, 115, 116, 117, 129, 130, 202, 204, 225, 226, 229, 243, 244, 247, 248, 250, 267, 268, 269, 273 Bimoraic: 5, 9, 20, 21, 46, 50, 51, 52, 53, 55, 56, 57, 112, 113, 114, 116, 117, 118, 122, 128, 129, 239, 248, 250, 262267, 270, 277
INDEX OF TERMS & CONCEPTS
Bimoraic Norm / Bimoraic Enforcement: 111, 112, 255, 262, 263, 264, 270 Monomoraic: 9, 57, 112, 113, 116, 118, 122, 128, 225, 239, 255, 264, 266 Mora-Splitting: 117, 118, 129, 130
Moraic Theory/Phonology: 18, 22, 111, 112, 113 Trimoraic: 5, 6, 21, 26, 112, 113, 114, 118, 124, 126, 128131,266 Morphology: 3, 9, 46-51, 56, 169, 171, 172, 174, 176, 182, 183, 185, 194, 196 Morpho-Syntax: 3, 49, 56, 193, 194 O. Obligatory Contour Principle (OCP): 89, 97-102, 104-109, 170 Optimality Theory (OT): 5, 6, 76, 140, 152,229,265,266 Orthography: 25, 26, 29, 142, 143, 144, 146, 162, 240, 245 P. Prosody: 46, 57, 157, 191, 224, 230 Phonological Phrase: 7, 137, 138, 141, 143, 151, 155, 157, 158, 159, 161, 162, 164, 206, 207 Prosodic Domain: 7, 138, 145, 149, 151, 153, 155,224,228
299
Prosodic Hierarchy: 7, 138, 157, 164, 195 Prosodie (Phonological) Word: 7, 46, 50, 64, 137-143, 147155, 157-164, 206, 207, 224, 225, 276 Prosodie Phonology: 143, 153 Prosodie Structure: 9, 50, 53, 55, 117, 138, 141, 203, 239, 246, 250 see also Clitic Group, Double Prosody R. Raddoppiamento (Fono)Sintattico /Rafforzamento Fonosintattico (RS)/(RF): 8, 9, 45, 46, 47, 49-57, 150, 191, 192, 195, 196, 198, 199, 200, 239, 243248,250,251 Rules: 7, 8, 17, 18, 19, 20, 22, 25, 27,28, 36, 52, 55, 60, 66-71, 73, 75, 76, 80-83, 85, 94, 96, 98, 111, 113, 137, 139, 141, 143, 145-160, 162, 164, 171, 182, 185, 195, 198, 202, 205, 215, 228, 243, 250, 258, 263, 264, 265, 268 Rule Ordering: 18, 24, 27, 28, 36, 195 S. Sonority: 6, 7, 163, 211-219, 223226,229, 231, 234, 235, 255, 262, 266, 268, 270, 276
300
INDEX OF TERMS & CONCEPTS
Stress: Antepenultimate (Proparoxytonic) Stress: 6, 7, 8, 60, 61, 85, 116, 117, 140, 160,241, 242, 243, 266, 274, 278, 284 Final (Oxytonic) Stress: 6, 7, 8, 9, 13, 14, 16, 17, 20, 45, 46, 53, 107, 116, 119, 121, 140, 191, 192, 193, 195, 196201, 203, 204, 206, 207, 241244, 247, 248, 249, 255, 256, 258, 265, 266, 267, 269, 274, 275, 283 Penultimate (Paroxy tonic) Stress: 8, 64, 65, 66, 69, 140, 191,200, 241, 242, 243, 258, 274, 275, 284, 285 Secondary Stress: 107, 195 Stress Shift: 138, 139, 140, 153, 159, 163 Stress Stability: 7, 137, 139, 140, 154, 158, 159, 164 see also Syllable Position Substratum: 3, 59, 60 Syllable Position: Antepenultimate: 6, 8, 60, 61, 68, 69, 83, 85, 241, 255, 259, 264, 265, 270 Final: 6, 13, 15, 16, 17, 45, 50, 51, 53, 56, 57, 66, 111, 114-117, 119, 121, 241, 243, 244, 247, 248, 249, 250, 255256, 258, 260, 264, 265, 266, 270, 274, 282, 283
Penultimate: 6, 8, 60, 64, 65, 66, 69, 114, 115, 116, 118, 119, 128, 160, 205, 206, 241, 255,258, 259, 261, 264, 265, 266, 270, 274, 285 Posttonic: 8, 64, 250 Pretonic: 8, 64, 242 Stressed/Tonic: 6, 8, 14, 15, 16,20,45,53,60,61, 64, 65, 6 6 , 6 9 , 7 3 , 7 4 , 8 0 , 8 1 , 84, 85, 90-93, 111-119, 129, 130, 137, 140, 150, 240, 241, 243, 244, 247, 255, 256, 258, 263, 265, 270, 274, 275, 283, 285 Unstressed/Atonic: 6, 8, 45, 64, 67, 70, 74, 80, 84, 85, 9093, 111, 116, 118, 140, 149, 230, 239, 242, 247 see also Stress Syllable Structure: Closed Syllable: 8, 14, 15, 22, 23,25,50,60,61,68, 69, 83, 85, 111, 114, 116, 117, 119, 120, 129,240,241 Coda: 14, 15, 18, 19, 22, 5255,66,69,81, 111, 117, 154, 158, 163, 217, 242, 244, 262, 275, 276, 277, 283, 285 Degenerate Syllable: 20, 21, 22, 26, 265 Nucleus: 6, 52, 61,80, 81, 82, 85, 124, 197, 226, 242, 255, 266, 268, 270 Onset: 7, 30, 40, 45, 47, 82, 111, 117, 129, 130, 154, 197, 215, 217, 219-227, 229, 275, 276, 277
INDEX OF TERMS & CONCEPTS
Open Syllable: 8, 13, 14, 22, 23, 50, 53, 60, 64, 65, 66, 68, 69,84,85, 111, 114-120, 128, 129, 240, 241, 244, 263, 264, 265 Superheavy Syllable: 112, 118, 121, 128,274,283,284 see also Mora Syntax: 4, 9, 56, 153, 199, 200 T. Template: 6, 216, 202, 275, 280, 281,283,284,285,286 Toponomastics: 3 U. Underspecification Theory: 93 Universal Grammar (UG): 5, 60, 79, 266
301
V. Vowels: Vowel Alternation: 7, 8, 89, 217,251 Vowel Fronting: 7, 8, 59, 60, 61,63, 64, 67, 76-81, 83, 84, 85 Vowel Harmony: 7, 59, 139, 163, 164 Vowel Length(ening): 5, 6, 1337, 89, 53, 54, 65, 111-116, 119-131, 142, 204, 248, 250, 255, 256, 258, 260, 262-266, 269, 270 Vowel Raising: 7, 59, 67, 70, 77, 108, 139 Vowel Reduction: 8, 89-93, 95, 96, 97, 99-102, 104, 106, 107, 108, 109, 257 Reduced Vowel: 45, 51, 239, 248