Approaches to Metaphony in the Languages of Italy 3110369567, 9783110369564

Table of contents :
Contents
List of contributors
Introduction
Harmonic processes and metaphony in some Italian varieties
Metaphony and diphthongization in Southern Italy: reconstructive implications for sound change in early Romance
On the morphophonology of metaphonic alternations in Altamurano
Metaphony in the Ticino Canton and phonological features
Progressive metaphony in the Abruzzese dialect of Teramo
Metaphony and Raddoppiamento Fonosintattico in plural nouns in the dialect of Airola
On integrating different methodologies in phonological research: acoustic, articulatory, behavioral and neurophysiological evidence in the study of a metaphony system
Which syntactic information gets interpreted by phonosyntactic processes? Evidence from /u/ propagation in Southern Italian dialects
On the typology of metaphony/stress dependent harmony
Metaphonic chain shifts, vowel height and markedness
Positional prominence and consonantal interactions in metaphony and post-tonic harmony
Opaque interactions between vowel merger and metaphony
Abruzzese metaphony and the A eater
Index

Citation preview

Francesc Torres-Tamarit, Kathrin Linke and Marc van Oostendorp (Eds.) Approaches to Metaphony in the Languages of Italy

Phonology and Phonetics

Editor Aditi Lahiri

Volume 20

Approaches to Metaphony in the Languages of Italy Edited by Francesc Torres-Tamarit, Kathrin Linke and Marc van Oostendorp

ISBN 978-3-11-036956-4 e-ISBN (PDF) 978-3-11-036631-0 e-ISBN (EPUB) 978-3-11-039327-9 ISSN 1861-4191 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2016 Walter de Gruyter GmbH, Berlin/Boston Typesetting: Compuscript Ltd., Shannon, Ireland Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany

www.degruyter.com

Contents List of contributors

vii

Kathrin Linke, Marc van Oostendorp and Francesc Torres-Tamarit Introduction 1 Leonardo M. Savoia Harmonic processes and metaphony in some Italian varieties

9

Michele Loporcaro Metaphony and diphthongization in Southern Italy: reconstructive implications for sound change in early Romance 55 Andrea Calabrese On the morphophonology of metaphonic alternations in Altamurano 89 Stefano Canalis Metaphony in the Ticino Canton and phonological features

127

Diana Passino Progressive metaphony in the Abruzzese dialect of Teramo

147

Mirella De Sisto Metaphony and Raddoppiamento Fonosintattico in plural nouns in the dialect of Airola 179 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese On integrating different methodologies in phonological research: acoustic, articulatory, behavioral and neurophysiological evidence in the study of a metaphony system 195 M. Rita Manzini and Leonardo M. Savoia Which syntactic information gets interpreted by phonosyntactic processes? Evidence from /u/ propagation in Southern Italian dialects 221

vi 

 Contents

Joan Mascaró On the typology of metaphony/stress dependent harmony Martin Krämer Metaphonic chain shifts, vowel height and markedness

259

277

Rachel Walker Positional prominence and consonantal interactions in metaphony and post-tonic harmony 301 Francesc Torres-Tamarit and Kathrin Linke Opaque interactions between vowel merger and metaphony Roberta D’Alessandro and Marc van Oostendorp Abruzzese metaphony and the |A| eater 349 Index

369

333

List of contributors Andrea Calabrese, University of Connecticut Stefano Canalis, University of Padua Roberta D’Alessandro, Leiden University Mirella De Sisto, Meertens Institute Mirko Grimaldi, University of Salento Martin Krämer, UiT – The Arctic University of Tromsø Kathrin Linke, University of Vienna Michele Loporcaro, University of Zurich M. Rita Manzini, University of Florence Joan Mascaró, Universitat Autònoma de Barcelona Sandra Miglietta, University of Salento Marc van Oostendorp, Meertens Institute and Leiden University Diana Passino, University Nice Sophia-Antipolis Leonardo M. Savoia, University of Florence Francesco Sigona, University of Salento Francesc Torres-Tamarit, UMR 7023 CNRS/Paris 8 Rachel Walker, University of Southern California

Kathrin Linke, Marc van Oostendorp and Francesc Torres-Tamarit

Introduction

The term metaphony refers to a type of vowel assimilation in Romance, by which a post-tonic inflectional high vowel, or its diachronic remnant in the form of a schwa or even a zero suffix, causes raising of—at least—a stressed vowel b ­ elonging to the root. Even this short description may remind the reader that a proper theoretical description of the phenomenon involves the understanding of several important areas of linguistic theory: we have to understand at least what the internal ­structure of vowels is, how vocalic properties can ‘spread’, what the potential influence (and representation) of stress is, how morphological properties—being part of a root or a suffix—can codetermine phonological behavior, and how historically complex phenomena can get an (opaque) representation in current grammar. Furthermore, since the phenomenon shows up in different ways in ­different dialects, understanding metaphony potentially enhances studies on language variation and typology. The phenomenon is currently best attested in the Romance languages and dialects of Italy. Although still operative in Northwestern Romance varieties in Spain (Cantabrian Spanish, Asturian, Galician), Iberian metaphony is recessive (see Penny 2009 for a descriptive overview). In Italy, however, metaphony is widespread and pervasive across most varieties. Italo-Romance is thus a good empirical domain for studies on metaphony in general. Although several articles present extremely interesting and some of them even new data, the emphasis of the present volume is on theoretical issues, and our aim is mostly to contribute to theoretical understanding of the phenomena at hand. Several of these phenomena are addressed in the different papers of this book; we distinguish between the morphosyntax-phonology interface, the ontology and representation of phonological primes, phonological opacity, the relation between diachrony and synchrony and between grammar and the lexicon. Each of these will be briefly discussed in turn in this volume.

1 Morphology-phonology interface One important issue is how morphology affects the process at hand. Metaphony typically (or even always) means that an affix influences the phonological shape

2 

 Kathrin Linke, Marc van Oostendorp and Francesc Torres-Tamarit

of the stem. In this respect it is different from so-called root-controlled vowel harmony in languages like Turkish, in which a phonological feature spreads from the stem to the affixes. Furthermore, given the fact that the process only involves some affixes and not others, one has to decide how this kind of specific behavior of individual affixes can be represented in the grammar: should it be done by assigning specific phonological properties (e.g. floating features) to some affixes, or rather by formulating the process itself as being sensitive to morphology? Or should the whole process be called morphological, with some kind of heavy stem suppletion that is sensitive to certain morphemes? Possibly, there is no theory-neutral way to decide on these questions; whether or not we find any answer to any of these questions plausible will depend at least to some extent on our other views on the organization of the grammar and the division of labor among its various components. The phenomenon can therefore work as a testing ground for those different sets of assumptions. Different approaches to these and related questions are found among the papers in this volume. Calabrese, in his chapter On the morphophonology of metaphonic alternations in Altamuro, argues against analyses based on abstract representations such as floating features to represent the phonological exponent of specific morphemes, and against suppletion to account for opaque metaphonic alternations due to historical processes of reduction. His proposal is couched in the framework of Distributed Morphology and resorts to phonological rules that operate on specific morpheme exponents. Mascaró, in his chapter On the typology of metaphony/stress dependent harmony, argues instead in favor of floating features to account for morpho­logized metaphonic systems. According to the structure of the metaphonic trigger, Mascaró distinguishes between morphemes that consist of a single floating feature or a set of features (which he calls featural affixation), morphemes that consist of both segmental and floating material (segmental and featural ­affixation), or just a phonological element encoding no relevant morphological information (phonological harmony). While Calabrese and Mascaró thus each argue for an approach to metaphony that is at least partially morphologically informed, D’Alessandro and Van Oostendorp, in Abruzzese metaphony and the |A| eater, argue for considering metaphony to be a purely phonological phenomenon, which is blind to all morphological and lexical information. They develop a representational solution to metaphony in Ariellese in which the |A| element belonging to the stressed vowel is detached from its original position and linked to a final empty mora that is the phonological exponent of [plural] and of the feature bundle [masculine, plural]. The morphology thus only influences phonology indirectly, by the fact that some morphemes take a ‘funny’ phonological shape.

Introduction 

 3

Manzini and Savoia, however, suggest in Which syntactic ­information gets interpreted by phonosyntactic processes? Evidence from /u/ propagation in Southern Italian dialects that the externalization component of a ­minimalist grammar needs to read syntactico-semantic representations in the light of harmony processes triggered by /u/ in Southern Italian dialects. Based on the metaphony process in the dialects under their discussion, they thus suggest a reinterpretation of what kind of morphosyntactic information the phonology can see in general. These different approaches demonstrate the difficulty in establishing a clear distinction between purely phonological processes and morphophonological processes in metaphonic systems. At the same time, they each in their own way succeed in bringing the discussion about the boundaries between the different modules to a higher level.

2 The nature of phonological primitives A second important theoretical issue addressed in this volume is the ­appropriateness of feature-based versus element-based theories in accounting for metaphony. Features are the traditional primitives of phonological representation since the seminal work of Roman Jakobson in the 1940s, but in particular since the 1980s, Element Theory has been quite popular as an alternative to this, in particular within Europe (see Backley 2011 for an overview). Element Theory has the advantage that it can describe some typological universals with respect to vowel systems quite well (e.g. that mid and high front vowels behave as a natural class, as do mid and high round vowels) and that it is also theoretically very elegant and quite restrictive. When confronted with ­metaphony, however, Element Theory suffers from several problems, such as the fact that it has no uniform representation of vowel height (i.e. no correspondent to the feature [+high]). Some papers explicitly compare the two approaches and contrast their empirical predictions. Canalis, in Metaphony in the Ticino Canton and phonological ­features, argues that representing vowel height as a multi-valued feature, or by using privative phonological primes like elements, accounts for the facts of the Romansh dialect of Ticino Canton in a more unified way than standard binary features. Passino, in Progressive metaphony in the Abruzzese dialect of Teramo, offers a complete formal account of progressive, and also regressive, metaphony in the dialect of Teramo, and compares an element-based analysis of progressive metaphony in terms of |A| demotion and a feature-based analysis in terms of height assimilation. Savoia, in Harmonic processes and metaphony in some Italian varieties,

4 

 Kathrin Linke, Marc van Oostendorp and Francesc Torres-Tamarit

offers an analysis of harmony and metaphony in Central Sardinian and Central Sicilian based on elements and also compares it with an alternative analysis based on features. Finally, De Sisto, in Metaphony and Raddoppiamento Fonosintattico in plural nouns in the dialect of Airola, develops an analysis based on Element Theory and CVCV. Her analysis is again based on the idea of element demotion, by which the |A| element is detached from the stressed vowel due to specific licensing conditions. These chapters all conclude that the element account gives a more unitary account of the phenomena, as does the paper by D’Alessandro and Van Oostendorp, and therefore prove that the element approach to metaphony is gaining ground. This does not mean that the binary feature approach is no longer used in current analyses of metaphony, as shown in the chapters by Calabrese, Krämer, Mascaró, and Walker. The debate between feature-based and element-based approaches to metaphony is interesting as it sheds light on another related and central question: which kind of phonological operation is responsible for metaphony? Two basic answers to this question can be found in the volume: firstly, the more traditional answer that interprets metaphony as a case of assimilation of the feature [+high]. Within this interpretation, Walker’s paper Positional prominence and consonantal interactions in metaphony and post-tonic harmony stands as the one arguing in favor of a view of metaphony as driven by positional licensing in the light of consonant blocking in post-tonic harmony. Positional licensing is also assumed, but with modifications, in Torres-Tamarit and Linke’s paper. Another approach, like the one by Krämer, abandons the idea of metaphony as licensing of [+high] in favor of an interpretation based on correspondence relations between the weak position and the strong position within the same output form. Secondly, a newer interpretation of metaphony within the framework of Element Theory is presented that characterizes this process as a case of |A| demotion, or deletion or movement of lowness.

3 Phonological opacity and diachrony The relation between metaphony and so-called phonological opacity has r­ eceived quite some deserved attention, both in previous literature and in the ­chapters of this volume. The issue is that many dialects have gone through two h ­ istorical steps: first one of metaphony proper, i.e. assimilation of the stressed vowel to a high vowel in a suffix, and subsequently reduction of the suffixed vowel to schwa or even deletion of the vowel. The puzzle for synchronic phonology is how those two steps can be represented in the grammar: do they correspond to two

Introduction 

 5

­extrinsically ordered rules? Do we need some other specific device for opacity in a constraint-based view of the grammar? Or can the synchronic grammar be organized ­differently, without mirroring the history of the process? A similar question arises for those dialects which show chain shifts, e.g. raising the high mid /e/ to [i] and the low mid /ɛ/ to [e]. The diachronic ­analysis could be e.g. that the high mid vowel raised first, and the low mid vowel ­followed later when it was already too late to further raise to [i] (this would be a drag chain). The question is how a synchronic analysis should look like in a language in which both processes are arguably productive. Krämer, in his chapter Metaphonic chain shifts, vowel height and markedness, considers opaque metaphonic systems in which raising proceeds stepwise. His account is based on two ingredients. First, he makes use of the local conjunction of faithfulness constraints, as in the first Optimality Theory approach to synchronic chain shifts by Kirchner (1996). Second, he resorts to syntagmatic correspondence, by which metaphony is not understood as licensing of height, but as an identity correspondence between the two positions within the same output form, as mentioned before. The opacity of the process thus does not mean that there is some ordering of the operations, but that the low mid vowels are restricted to how divergent they are allowed to become from their underlying representation Torres-Tamarit and Linke, in Opaque interactions between vowel merger and metaphony, develop a parallel OT analysis of metaphony in Arpinate and ­Salentino, which show opacity of the underapplication and overapplication type due to reduction processes in the trigger. Their analysis makes use of privative features and Turbidity Theory, a version of Optimality Theory that assumes Containment. Also in this case, the synchronic analysis does not mirror the historical developments; those developments have simply led to a system that can be represented with the tools that, according to the authors, are independently necessary, without giving up the idea that phonology has no ordering of processes. The historical origin of metaphony in Italo-Romance itself is the topic of Loporcaro’s chapter Metaphony and diphthongization in Southern Italy: reconstructive implications for sound change in early Romance. This chapter suggests that diphthongization and metaphony in Romance have a different historical source, as only metaphony, but not diphthongization, was sensitive to the quality of the final unstressed vowel. The historical development of diphthongization and metaphony in Romance appears in this chapter as a relevant case study of the different effects that phonetics and (morpho)phonology exert on sound change. Both phonological opacity and the question whether phonological patterns require a synchronic or a diachronic explanation have been, and will continue to be, the topic of discussion for phonology for quite some time. The various

6 

 Kathrin Linke, Marc van Oostendorp and Francesc Torres-Tamarit

c­ hapters in our volume show that metaphony can work as a good testing ground for all those theories.

5 Methodology Finally, the volume includes one paper that presents the results of a set of ­experimental studies on metaphony. In On integrating different methodologies in phonological research: acoustic, articulatory, behavioral and neurophysiological evidence in the study of a metaphony system, Grimaldi, Miglietta, Sigona, and Calabrese make integrated use of different methodological techniques in the study of metaphony in Southern Salento. Ultrasound images of the tongue corroborate that two different features spread depending on the trigger: [+ATR] spreads from /i/ triggers, and [+high] spreads from /u/ triggers. Behavioral and electrophysiological methods are also used to explore the allophonic alternation between [ɛ] and [e] and the phonemic contrast between /e/ and /i/. No difference was found between the two pairs, suggesting that allophones and phonemes are computed in the same way during auditory processing.

6 Conclusion In this brief introduction, we have posited every chapter with respect to some theoretical debate. That does not do complete justice to those chapters, since they all actually contribute to all of those debates. This is one of the things that makes theoretical discussions often complicated: one’s view on e.g. phonological primitives has repercussions also on one’s view of opacity. This is one reason why it is important to sometimes discuss specific phenomena—such as metaphony in Italo-Romance—rather than just theoretical notions: to see in detail how various theoretical assumptions interact. It is obviously not the purpose of this volume to argue in favor of one specific theoretical approach over the other, although that has been the purpose of most, if not all, of the individual papers. The main objective of this volume is instead to bring together pioneering work on metaphony from different ­perspectives and provide a state-of-the-art update of this classical topic in Romance ­linguistics that still informs linguistic theory to this day. We are confident that this volume offers a ­ etaphony poses diverse and thorough discussion of some of the challenges that m for different linguistic theories, and becomes a reference and inspiring book for those linguists interested in this phenomenon. We know that we will have many interesting discussions ahead of us.

Introduction 

 7

References Backley, Philip. 2011. An introduction to Element Theory. Edinburgh: Edinburgh University Press. Jakobson, Roman. 1940/1992. Kindersprache, Aphasie, und allgemeine Lautgesetze. Frankfurt am Main: Suhrkamp. Kirchner, Robert. 1996. Synchronic chain shifts in Optimality Theory. Linguistic Inquiry 27: 341−350. Penny, Ralph. 2009. Vowel harmony and metaphony in Iberia: a revised analysis. Estudos de Lingüística Galega 1: 113−124.

Leonardo M. Savoia

Harmonic processes and metaphony in some Italian varietiesi

Università di Firenze

Abstract: This article examines restrictions on the distribution of the phonological content of nuclei within the prosodic domain of the stressed nucleus. Besides metaphony, two types of harmonizing mechanisms are considered which govern the distribution of vocalic quality, specifically aperture properties, in the string: restrictions on the mid post-tonic vowels in Central Sardinian varieties and the distribution of [−ATR] vowels in Central Sicilian dialects. Metaphony and harmonic phenomena offer an interesting test for notions like contrastivity and markedness, supporting the hypothesis that the spreading of phonological cues in the string is connected to an improved ‘perceptibility’ of specifications relevant for the interpretation (Kaun 1995). The comparison with a treatment in terms of elements clarifies of the intuitive link that connects metaphony to other harmonic phenomena.

1 Introduction The classical generative framework assigns phonology a special status, inherently different from that of syntax. Chomsky (2000: 118) restates this difference in terms of Inclusiveness Condition. Whereas syntax obeys Inclusiveness in the sense that ‘No new features are introduced by CHL (Computational procedure for Human Language)’ in lexical items, phonology escapes this condition, given that ‘there are true phonological features that are visible only to the phonological component and form a separate subsystem of FL (Faculty of Language), with its own special properties’. In particular, ‘the [phonological] subsystem reflects special properties of the sensory-motor systems, which are in a certain sense «extraneous» to language…’. In this perspective, Berwick & Chomsky (2011: 37) argue that

i I wish to thank my informants, who collaborated with intelligence and generosity, and in particular Musu (Allai), Sabina Lorusso (Andria), Benedetta Mazzei (Iacurso), Paola Melosu (Làconi), Giuseppina De Tommaso (Mascioni), Emanuela Serrao (Milena), Antonietta Sanna (Paulilàtino), Marziale Mascia (Settimo S. Pietro), Serafina Landro (Sutera), Maria Fiandaca (Villapriolo), and Raimondo Usai (Sèneghe).

10 

 Leonardo M. Savoia

phonology, together with morphology, is an interpretive component working in the externalization procedure of internal mental objects made up of the faculty of language. This characterization of phonology confirms the original distinction between phonology and syntax repeated in Bromberger & Halle (1989). Actually, this disparity seems to be too drastic. Naturally, we can expect that (morpho-) phonology interacts with the sensory-motor system, at least in the sense that phonology incorporates phonetic information and phonetic changes or imperfections. However, when phonetic data are included in the grammar, they cannot have a cognitive status substantially different from syntax objects. Traditional approaches are based on the separation between underlying and surface representations, in the sense that the former register distinctive properties which phonological rules can make opaque at the level of the latter. In this view, phonology would have the function of manipulating and possibly obscuring the phonetic content specified at the level of lexical ­representations. Recent phonological models imply a revision of this pattern. In particular autosegmental requirements reduce the possible types of rules; analogously, fully specified underlying representations (Halle et al. 2000; Calabrese 1998; Nevins 2010) bring about a more adequate phonological analysis. Along this line, Government Phonology (Kaye 1986/87, 1990; Kaye, Lowenstamm & Vergnaud 1990; Charette 1990, 1991; Harris 1994a) has a crucial role in con­ straining the format of phonological representations and the nature of possible processes in terms of phonetic interpretability. In particular, if according to Harris and Lindsey (1995), phonology operates on mental objects, both lexical underlying phonological representations and surface representations must be equally phonetically interpretable. As a consequence, there is no abstract phonetically uninterpretable level, as assumed in models of underspecification. Processes admitted include mechanisms of linking/spreading of properties which are included in the representations or delinking of features, thus satisfying the Projection Principle (Kaye 1986/87, 1990), which prevents arbitrary changes in prosodic relations at all levels of representation. As for the nature of phonological content, the recourse to privative monovalent elements (Kaye, Lowenstamm & Vergnaud 1985; Harris & Lindsey 1990, 1995; Backley 2011) concurs to answer many descriptive problems. In what follows we will apply autosegmental representations, syllabic and prosodic structure, government relations between syllabic positions, and licensing by prominent segments in the prosodic organization of the string. Metaphony and other phenomena will be analyzed in terms of features; section 4.1 will be devoted to an approach based on elements. Phonological specifications and processes depend on well-formedness constraints which fix the distribution of vocalic/consonantal properties on the basis of prosodic requirements and licensing relations in the relevant domain. Strict phonological mechanisms of



Harmonic processes and metaphony in some Italian varieties 

 11

metaphony and other harmonic phenomena associate with their ability to ‘externalize’ morpho-syntactic information, in the sense of Berwick and Chomsky (2011: 37) which identify phonology and morphology with ‘the linguistic processes that convert internal syntactic objects to the entities accessible to the sensory-motor system’. In this perspective, the traditional view which connects metaphony with morphological categories like number/gender/verbal person1, may imply a valid intuition. The fact that the association with the stressed nucleus can enhance the perceptibility of phonological features fits in with other aspects of the process. In this sense, the stressed nucleus is in many cases the place for phonological contrasts externalizing distinct inflectional categories.

2 Some remarks on metaphony In Italo-Romance metaphony the stressed nucleus partially or completely agrees with the following high vowel, [i u], in the aperture degree (cf. Rohlfs 1966 [1949]). Two fundamental types of Italo-Romance metaphony are recognized, i.e. metaphony of high-mid stressed vowels /e o/ and metaphony of low-mid stressed vowels /ɛ ɔ/. The first process causes the raising of high-mid vowels of lexical bases to [i u], whereas metaphony of low-mid vowels brings about different results, in particular raising to [e o], diphthongs [iə/iɐ uə/uɐ], raising to [i u], according to different varieties. In any case, a system of alternations between lexical level stressed nuclei and metaphonized ones normally shows up. Metaphonic realizations normally preserve cavity properties [+/−back, +/−round] of the underlying stressed vowel, possibly in the former element of the diphthong. Finally, in many varieties post-tonic vowels, including the final ones externalizing nominal inflections, have been affected by different types of neutralization. Thus, in many Southern dialects we find a schwa, [ə]; raising of original *-e *-o to [i u] occurs in Calabrian, Sicilian, and Southern Sardinian dialects.

1 Here I adopt the model proposed in Manzini & Savoia 2005, 2007, where the nominal inflection includes nominal class/gender and number; verbal inflection refers to verbal (person, tense) categories. This model depends on the analysis generally applied in generative grammar (cf. Chomsky & Halle 1968; Harris 1991), where inflectional morphemes are sub-word units endowed with a specialized semantic content. Traditionally all of these specifications are associated with terminal nodes in syntactic derivation and externalized by morphological exponents (as for instance assumed in a Distribute Morphology framework, cf. Halle & Marantz 1994). Different approaches assume that words are the true units, and the morphological segmentation is a secondary abstraction (cf. Blevins 2006). This conclusion is based on the alleged unpredictable character of the relation between inflection and lexical base.

12 

 Leonardo M. Savoia

Metaphony of high-mid and low-mid vowels have a common root insofar as they can be conceived as assimilatory mechanisms affecting the distribution of height vocalic properties in the word domain. Nevertheless, the two types are neither entirely unifiable from a phonetic point of view nor in terms of a theoretical treatment. Empirical data seem to confirm this conclusion. In fact, their chronology and areal distribution do not coincide (Rohlfs 1966 [1949]; Maiden 1985, 1987, 1991; Savoia & Maiden 1997). In particular, metaphony of high-mid vowels has a wider areal distribution; this possibly supports the hypothesis that it is older than metaphony of low-mid vowels (Maiden 1985, 1987). The two metaphonic processes correspond to different harmonizing properties, which work with independent modalities according to the original vocalic system and, what is more, to the morphological procedures of externalization. Moreover, metaphony cannot be isolated from the set of the other harmonic processes that coexist and interact with it. Some authors have attempted a reconstruction of a historical and phonetic link between the two types of metaphony of low-mid vowels. A traditional idea supported for example by Castellani (1973) is that high-mid outcomes [e o] could derive from the monophthongization of original metaphonic diphthongs [ie uo]. However, other hypotheses are discussed in the literature. A proposal, discussed in Barbato (2008) and Loporcaro (this volume), assumes that metaphonic diphthongs would be a secondary outcome from older high-mid [e o] outcomes deriving from original low-mid vowels. In this perspective, a crucial role is assigned to a conjectured/reconstructed intermediate level whereby simple raising would precede the current diphthongizing metaphony in Center-Southern varieties. Indeed, the evidence appears to be uncertain, if we consider that the observable situation shows a strong areal overlapping of the two types of metaphony. What is more, if diphthongization was a secondary process affecting /e o/ deriving from metaphony, we would expect that it could affect all /e o/ nuclei, including the etymological ones as well. Barbato (2008) notes this problem suggesting that diphthongization could be determined by an impulse towards the morphologization of metaphony. As regards the typology and distribution of metaphonic outputs of low-mid vowels, we see that a large variability between diphthongized outcomes, typically [iə uə] (not [je wo]), and monophthongs [i u] shows up. The outcomes [i u] from low-mid vowels seem to have a clear relation with original diphthongs [iə uə], while there is no evidence for a link between [e o] and [iə uə]. If we take the existing phonetics as a serious clue for interpreting the history of sounds, these outcomes suggest different lines of evolution. Anyway, metaphony goes back to a distant past and another solution could be plausible, as, for example, the hypothesis that the outcomes [e o] are due to a process of monophthongization. As for the relation between diphthongization and metaphony, we know that in many Central and Southern Italian dialects (Abruzzi, Apulia, Lucania)



Harmonic processes and metaphony in some Italian varieties 

 13

diphthongization has affected tonic nuclei according to the metric-syllabic contexts, possibly also including metaphonic outputs. However, metaphony of low-mid vowels has normally different outcomes from ‘spontaneous’ diphthongization/aperture processes, which on the contrary systematically include the outcomes of metaphony of high-mid vowels, but only partially those of low-mid vowels. An interesting suggestion can come from the Sardinian, Calabrian, and Sicilian varieties characterized by a three height degree vocalic system lacking high-mid vowels. In Calabrian varieties, raising and diphthongizing metaphony of low-mid vowels coexist, whereas other types of diphthongization are absent and sensitivity to syllabic structure does not emerge. Similarly, in Sicilian dialects metaphonic diphthongization of low-mid vowels is present while other types of diphthongization are absent. This distribution weakens both the hypothesis that metaphony of high-mid vowels precedes the metaphony of low-mid vowels, and the assumption that metaphonic diphthongization is secondary on the raising of low-mid vowels to [e o]. In other words, there is no evidence to argue for the older character of simple raising to [e o]2. In fact, reconstruction is an interesting exercise, but on the basis of the available data the most reasonable conclusion we can arrive at is that the two types of metaphony of low-mid vowels—raising vs. diphthongization—correspond to two different harmonizing processes applying to the etymological low-mid vowels in tonic position (Rohlfs 1966 [1949]: 128, 154). For that matter, no phonetic or cognitive ­constraint would seem to exclude this possibility. It is of note that there are authors who reject the assimilatory nature of metaphony. So, Sánchez Miret (1998) and Russo (2007) propose that metaphonic outcomes are nothing but spontaneous evolutions in contexts where the final vowel is different from /a/. In particular, Russo (2007) holds that raising and diphthongization correspond to the introduction of ‘default’ specifications in the stressed vowels of the system; analogously, the preservation of stressed nuclei would implement a requirement of vocalic aperture triggered by the morphological context -a. This entails an overturning of the traditional perspective, requiring explanation of why -a would block the default diphthongization or raising of high-mid vowels. The most evident inadequacy is the recourse to an unjustified set of default rules to account for not only ‘metaphonic’ outcomes but also the basic vocalic system.

2 More in general, in dealing with history of metaphony we must allow for the fact that diph­ thongization and other vocalic processes have affected stressed nuclei in Northern varieties as well, with separate results from metaphony. For example, metaphony of low-mid vowels emerges in Romansh varieties, at least in a subset of adjectives. In these varieties, however the non-metaphonic outputs also remain distinct from the metaphonic ones: ɔ/iə, etc.

14 

 Leonardo M. Savoia

The idea I start from is that metaphony is not intrinsically different from other types of harmonies. All these phonological mechanisms enhance perceptibility of certain phonological properties within a prosodic domain by increasing their durational extension (Kaun 2004). In the recent phonological debate, harmonic processes and metaphony have been connected with the need for making certain vowel properties recognizable. As for harmonies, according to Kaun, they increase the exposition of the listener to (morpho-)phonological elements characterized by vulnerable phonological properties (Kaun 1995; Zoll 1998; Walker 2005). Distinctiveness can play a crucial role: ‘harmony serves to extend the duration of phonetic information which is phonologically important (i.e. distinctive), but which is transmitted by means of relatively subtle acoustic cues.’ (Kaun 1995: vii). In particular, Kaun (1995) applies these ideas to [round] harmony, noticing that rounding contrasts are ‘perceptually subtle’ in front and non-high vowels. Walker (2005), referring back to Kaun (1995), characterizes harmonies and metaphonies as processes in which the ‘perceptual weakness’ of some features is circumvented by associating it with the prominent vowel in the string: ‘trigger conditions on spreading constraints may single out perceptually marked properties…’ (cf. Zoll 1998; Walker 2001). This picture provides harmonies and metaphonies with the same interpretive basis, in the sense that harmonies satisfy the durational extension of a phonological structure, whereas metaphony satisfies both durational extension and phonological strengthening of a feature by associating it with a strong position. In particular, associating a feature with the stressed position concurs to preserve its contrastive force (Kaun 2004; Nevins 2010). Returning to metaphony, the crucial question concerns the relation between raising metaphony of high-mid vowels and that of low-mid ones. Although traditionally unified, they do imply two different types of assimilation, given that in the first case the metaphonizing feature is [+high] ([high] metaphony), while in metaphony of low-mid vowels [+ATR] is involved ([ATR] metaphony). In the literature both descriptive and theoretical approaches assume that the same feature [+high] is involved. Naturally, in the classic feature system worked out by Chomsky & Halle (1968), the problem arises that [e o], or [i u], cannot be treated as the result of a rule changing [−high] to [+high] starting from [−high, +low] /ɛ  ɔ/ underlying vowels. Some readjustments are required which manipulate the combination of the two features, yielding [−high, −low], i.e. [e o], or [+high, −low], i.e. [i u], because of the impossibility of [+high, +low] (cf. Kaze 1991). Anyway, metaphony of low-mid vowels requires a treatment distinct from that of high-mid nuclei. In frameworks which have recourse to the feature [ATR] for dealing with vocalic tenseness/laxness phenomena (Vaux 1995), metaphony of the high-mid vowels combines [+high] with [+ATR], yielding the basic high vowels [+high, +ATR], [i u]. In the contexts where the underlying vowel affected by metaphony is low-mid, [−high,



Harmonic processes and metaphony in some Italian varieties 

 15

−ATR], spreading of [+high], as maintained in Calabrese (1998, 2009), creates an outcome [+high, −ATR], [ɪ ʊ] that corresponds to a possible articulation. However, the usual outcome is [−high, +ATR], i.e. [e o], or [+high, +ATR], i.e. [i u]. Calabrese (1998, 2009) explains this by resorting to repair mechanisms which modify a marked vocalic specification. In particular, the combination [+high, −ATR], derived from spreading [+high] to low-mid underlying vowels, is normally reinterpreted as [−high, +ATR], i.e. [e o], by the repair operation of ‘negation’, [+high, +ATR], i.e. [i u], by ‘delinking’, or, finally, as a diphthong, [ie uo], by ‘fission’. Frigeni (2003, 2004) deals with metaphony in Sardinian varieties as spreading of the feature [ATR]. Optimalist approaches (Walker 2011; Mascarò 2010) relate metaphony of low-mid vowels [−high, −ATR] to the prohibition of *[+high, −ATR] as a marked specification. In particular they assume specialized constraints which block the assimilation to [+high] and admit only the assimilation to [+ATR], bringing about the outcomes [e o]. In most systems, stressed [a] does not metaphonize, except for some Central Italian and Lombard Alpine varieties. The simplest explanation of the behavior of [a] is that there are no possible phonological configurations [+low] being able to license the combination [+high]/[+ATR] in post-tonic/final position. This seems to be the case, although actually many varieties have central vowels, so that nothing would prevent an outcome [ə ɨ] from occurring in raising contexts. However, metaphony of [a] generally shows the outcome [ɛ]. This suggests that metaphony is connected with the distribution of mid vowels in stressed position, thus recalling many of the harmonies studied in the literature (Archangeli & Pulleyblank 1989, 1994). More precisely, [i a u] in stressed positions occur without restrictions, whereas mid vowels depend on the height of the following vowel. We can express this point by concluding that [+high]/[+ATR] on the post-tonic vowel has to be licensed by the same specification in the stressed vowel. If the metaphony corresponds to such a mechanism, /a/ is excluded. The fact that in the majority of varieties stressed [a] is not subject to metaphony, can be therefore related to its isolated position in the vocalic system. In other words, the application of metaphony seems to involve the contrastive capability of the relevant properties. In the terms of Calabrese (1995, 1998) and Nevins (2005, 2010), contrastivity is defined with respect to the existence of two segments differentiated by opposite values for a feature, whereby a segment is contrastive for a feature [α F] in a given position if and only if a segment [−α F] may occur in this position. Contrastivity has a role in determining the processes working on phonological properties of segments. Along this line, we can think that metaphonies generally exclude the only vowel, i.e. [a], which lacks a corresponding segment contrasting with it in the features [high] and [ATR]. In other words, metaphonies apply to [high]/[ATR] contrastive properties. In short, harmonies and metaphony can be interpreted as phonological devices which associate some properties with a strong position which licenses

16 

 Leonardo M. Savoia

(authorizes) them. As discussed in Harris (1994a,b, 1997), Harris & Lindsey (1995, 2000), Zoll (2004, 1998), Walker (2001, 2011), metaphony and harmony correspond to the requirement whereby some properties in a domain are authorized if they appear in the prosodic head of the domain. As a consequence licensing can manifest itself in the form of the phonological agreement between the strong position, the licenser, and the licensed position, as in (1) (Harris 1990b). (1)  A property x in a governed position in the prosodic domain is licensed if it is associated with the head of the domain. As we will see, this model can be generally applied to the harmonic phenomena we will examine.

2.1 Metaphonies Let us now take a look at the basic properties of metaphony examined in some concrete cases beginning with metaphony of the dialect of Mascioni (Abruzzo). This grammar includes the set of vocalic contrasts in (2a) distinguishing high-mid vowels from low-mid ones. Metaphony implies the raising of lexical /e o/ to [i u], in (2b), and the raising of lexical /ɛ ɔ/ to [e o], in (2c). (2)  Mascioni a. i e

u o

[+high] [−high, −low]

[+ATR]

[−high, −low] [−ATR] ɔ a [+low] b. Metaphony of high-mid vowels [ˈveta] ‘fingers’ [ˈvitu] ‘finger’ [ˈroʃ ʃa] ‘red-f.sg’ [ˈruʃ ʃu] ‘red-m.sg’ [ˈroʃ ʃe] ‘red-f.pl’ [ˈruʃ ʃi] ‘red-m.pl’ c. Metaphony of low-mid vowels [ˈpɛrdo] / [ˈpɛrde] ‘I lose / he loses’ [ˈperdi] / [ˈperdu] ‘you / they lose’ [ˈnɔa] ‘new-f.sg’ [ˈnou] ‘new-m.sg’ [ˈnɔe] ‘new-f.pl’ [ˈnoi] ‘new-m.pl’ ɛ





Unlike the system of Mascioni, Sardinian systems, like that of Paulilàtino (Central Sardinia) in (3a), have three height degrees, excluding stressed high-mid vowels, except for metaphonic contexts. In the system of Paulilàtino stressed high-mid [e o] occur only when followed by vowels [+high], in (3b); in the other contexts stressed low-mid [ɛ ɔ] occur, in (3c).



Harmonic processes and metaphony in some Italian varieties 

(3)  Paulilàtino a. i e/ɛ

u o/ɔ

a b. [ˈbɛttsa] / [ˈbɛttsɔzɔ] [ˈbɛssɔ] [ˈɔɣɔzɔ] [ˈdrɔmmɔ] c. [ˈtɛndzɔ] / [ˈtɛnɛzɛ] [ˈkɔttsɔ] / [ˈkɔɣɛzɛ]

[+high, +ATR] [−high, −low, +/−ATR] [+low, −ATR] ‘old-f.sg / m.pl’ [ˈbettsu] ‘I go out’ [ˈbessizi] ‘eyes’ [ˈoɣu] ‘I sleep’ [ˈdrommiði] ‘I keep / you keep’ ‘I cook / you cook’

 17

‘old-m.sg’ ‘you go out’ ‘eye’ ‘he sleeps’

Let us consider first metaphony in the Mascioni dialect in (2b, c) and test the possibility that it is sensitive to the contrastive nature of features [high] and [ATR]. We see that in systems like that of Mascioni all different vowels can co-occur occupying identical vocalic contexts, as illustrated in (4), except for sequences including mid vowels, just like (2b, c), where a low-mid stressed nucleus occurs only when followed by a [−high] vowel. [+high] post-tonic vowels require a stressed [+ATR] nucleus in compliance with metaphony. [high] and [ATR] values associated with [i a u] can freely combine in the domain, admitting sequences like i...a, a…i, i…i, etc. Thus, [high] is contrastive in [+ATR] nuclei, and [ATR] is contrastive in [−high] vowels. Metaphonic contexts select the values [+high]/[+ATR]. (4)  Mascioni [ˈsete] ‘thirst’ [ˈit ʃ e] ‘he says’ [ˈkalle] ‘hot-f.pl’ [ˈbejo] ‘I drink’ [ˈiko] ‘I say’ [ˈsallo] ‘I jump’ [ˈveta] ‘fingers’ [orˈtika] ‘nettle’ [ˈkalla] ‘hot-f.sg’ [ˈko ʃ e] / [ˈko ʃ o] ‘he sews / I sew’ [ˈkrue] / [ˈkrua] ‘raw-f.pl / f.sg’ [ˈmɛle] ‘honey’ [ˈkɔre] ‘heart’ [ˈlɛo] / [ˈlɛa] ‘I take / he takes away’ [ˈjɔko] / [ˈjɔka] ‘I play / he plays’ In the grammar of Paulilàtino, [i u] [+high] and [ɛ a ɔ] [−high] can occur in the same vocalic contexts, where they precede [a] or low-mid vowels, as in (5). The contrast between low-mid and high-mid vowels is limited to the stressed position, where it is governed by metaphony, as in (3b). (5)  Paulilàtino [ˈlimba]  ‘tongue’ [ˈuŋgra]  ‘nail’ [ˈbɛttsa] ‘old-f.sg’ [ˈnɔa]  ‘new-f.sg’ [ˈsana]   ‘sound’

[ˈkrikkɔ]     ‘I look for’ [isˈkurtɔ]  ‘I listen’ [ˈbɛttsɔzɔ] ‘old-m.pl’ [ˈtɛndzɔ] / [ˈtɛnɛzɛ] ‘I / you keep’ [ˈsɔllɛ]  ‘sun’ [ˈkɔttsɔ] / [ˈkɔɣɛzɛ]  ‘I / you cook’ [ˈfrakkɛ]     ‘scythe’

18 

 Leonardo M. Savoia

We can account for metaphonic processes in (2b, c) and (3b) by assuming that metaphony reflects contrastive specifications, as expressed in the constraint in (6a) for high-mid vowels and in (6b) for low-mid ones. The non-contrasting value [ATR] of [i a u] in stressed position is not involved. In the Paulilàtino system in (3b) high-mid and low-mid vowels are in complementary distribution according to the value of [ATR] in the following vowel; we conclude that only (6b) is at work. In this variety [−high, +ATR] vowels interpret metaphonic contexts, where they anticipate and license [+ATR] in the following vowel. Naturally, raising metaphony of low-mid vowels in the dialect of Paulilàtino and that one in the dialect of Mascioni are not different. (6) a. Raising metaphony (of high-mid vowels) Contrastive [+high] ([i u] vs. [e o]) in the stressed nucleus licenses [+high] in the following vowel. b. ATR metaphony of low-mid vowels Contrastive [+ATR] ([e o] vs. [ɛ ɔ]) in the stressed nucleus licenses [+ATR] in the following vowel. In the systems preserving non-neutralized weak vowels, as in (2) and (3), (6) is satisfied when the stressed nucleus realizes relevant phonological properties thus licensing their occurrence in governed positions. Licensing implies delinking of incompatible phonological features, as in (7). The mechanism in (7) s­ ubsumes metaphony of low-mid vowels in Paulilàtino variety as well. (7)   Mascioni N

N

N

>

O

N

O N

x x

x

x x

/p r

ε

t

N

i /

[−ATR] [+ATR] [+high]

→

>

O

N

O N

x x

x

x x

p

e

t

r

i

[−ATR] [+ATR] [+high]

A crucial dimension of variation concerns the metaphonic status of intermediate weak vowels in trisyllables. In many systems the intermediate nucleus is opaque and triggers metaphony on the stressed nucleus, as in the Paulilàtino variety in (8a) and in general in Sardinian dialects. In other varieties, the in-between vowel is transparent and only the final nucleus is involved, as in the Mascioni variety in (8b).

Harmonic processes and metaphony in some Italian varieties 



(8) a. Paulilàtino [ˈfemmina] ‘woman’ [ˈomminɛ] ‘man’  b.  Mascioni [ˈolepe] ‘fox’ [ˈpɛtteno] ‘I comb’ [ˈɔme] ‘man’

 19

[ˈpessiɣɛ] ‘peach’ [ˈpoɖɖiɣɛ] ‘finger’ [ˈulepi] ‘foxes’ [ˈpetteni] ‘you comb’ [ˈommeni] ‘men’

Varieties like that of Paulilàtino in (8a) show the result we expect according to the phonological context; we can simply think that (6b) applies to sequences in (8a) as well. In the dialect of Mascioni the intermediate vowel of proparoxytones is in any case a high-mid vowel; in other words, in this position only high-mid vowels or [a] are admitted. So, the intermediate vowel is devoid of contrastive height properties. We can conclude that in these positions the intermediate vowel is not seen by metaphony in that its [−high] value is not contrastive, as suggested in (9). (9)   Mascioni >

word N foot N

>

N

O

N

O

N

x

x

x

x

x

u l e p i [+high]contr [−high] [+high]contr Let us consider finally the variety of Iacurso (Calabria) where metaphony of low-mid vowels shows up as a diphthong. Its vowel pattern coincides with that of Sardinian varieties in (3) insofar as it includes only three aperture degrees, high, low-mid and low vowels. A post-tonic [+high, +ATR] vowel rejects a [−high, −ATR] stressed nucleus. In this context, the stressed nucleus must anticipate these specifications and the diphthong [iɐ]/[uɐ] is realized in alternation with the lexical low-mid vowel, as in (10a). Again, the intermediate post-tonic high vowel of proparoxytones regularly triggers metaphony as in (10b). (10)   Iacurso a. [ˈpɛðɛ] [ˈvɛcca] [ˈpɛrdɛ] [ˈnɔva]

‘foot’ ‘old-f.sg’ ‘s/he loses’ ‘new-f.sg’

[ˈpiɐði] [ˈviɐccu] / [ˈviɐcci] [ˈpiɐrdu] / [ˈpiɐrdi] [ˈnuɐvu] / [ˈnuɐvi]

‘feet’ ‘old-m.sg / m.pl’ ‘I lose / you lose’ ‘new-m.sg / m.pl’

20 

 Leonardo M. Savoia

[ˈkɔrɛ] ‘heart’ [ˈkuɐri] ‘hearts’ [ˈdɔrmɛ] ‘s/he sleeps’ [ˈduɐrmu] / [ˈduɐrmi] ‘I sleep / you sleep’  b. [ˈpriɐvitɛ] / [ˈpriɐviti] ‘priest / priests’ [ˈpiɐkura] ‘sheep’ (11a) expresses the requirement whereby the formation of a diphthong in stressed position licenses both [+ATR], as in Sardinian dialects, and [+high], as in the variety of Mascioni in (2). The first position in the diphthong, i.e. its head, introduces the phonological content which needs to be licensed, i.e. [+high, +ATR]. (11)  Iacurso a. ATR metaphony [+high] and [+ATR] in the stressed nucleus license [+high, +ATR] in the following vowel.  b. N N >

O

N

x

x

/ p

ε

O

N

x

x

x

ε

ð + i /

[−high] [−ATR]

→

[+high] [+ATR]

N

>

O

N

O

N

x

x

x

x

x

p

i

a

N

ð

i

[+high] [−high] [+high] [+ATR] [−ATR] [+ATR]

Phonological information of the lexical level, namely [−high, −ATR], is preserved in the weak segment of the diphthong, in (11b), where the lexical base combines with the -i formative of m.pl.

2.2 Morphologized metaphonies Let us now pass on to the relation between morphological properties and metaphony. In the South Sardinian variety, original mid vowels in post-tonic position neutralize in [i u]; the result is that, except [a], only [+high] post-tonic vowels occur, as in (12) for Settimo S. Pietro. So, metaphonic contexts are opaque given that only a sub-set of post-tonic [i u], in (12b), deriving from etymological [i u], trigger metaphony (Contini 1987; Frigeni 2003, 2004). (12a) exemplifies contexts in which high vowels deriving from original mid vowels are inactive.



Harmonic processes and metaphony in some Italian varieties 

 21

(12)   Settimo S. Pietro  a. [aˈβɛrru] ‘I open’ b. [aˈβerrizi] / [aˈβerriði] ‘you open / [ˈfɛrruzu] ‘irons’    (s)he opens’ [ˈbrɛmmi] / [ˈbrɛmmizi] ‘worm/s’ [aˈβerrinti] ‘they open’ [ˈɔɣuzu] ‘eyes’ [ˈferru] ‘iron’ [ˈdrɔmmu] ‘I sleep’ [ˈoɣu] ‘eye’ [ˈdrommizi] / [ˈdrommiði] ‘you sleep / (s)he sleeps’ [ˈdromminti] ‘they sleep’ The comparison between the data of Paulilàtino in (5) and those in (12) points out two different distributions in the post-tonic position. In Northern Sardinian varieties the post-tonic position admits low-mid [ɛ ɔ], high [i u] and low [a] vowels, as provided for by the constraint in (13a). In this position, Settimo S. Pietro variety licenses only [i a u]. The exclusion of mid vowels from the post-tonic position can be expressed by a constraint like (13b), which authorizes only [+high] or [+low]. (13)  a.  Post-tonic vowels (Paulilàtino)  Post-tonic position in the word domain licenses [+ATR] only if it combines with [+high].  b.  Post-tonic vowels (Settimo S. Pietro)  Post-tonic position in the word domain licenses [+high] or [+low]. The distribution of metaphonic outcomes in (12) recalls the one expressed in (6b), with the complication that only a sub-set of [i u] is allowed for by the stressed nucleus, as in (12b). Other [i u], in (12a), do not trigger metaphony. We can single out the metaphonic sub-set of [i u] in terms of the morphological information (gender, number, person) they externalize. To be more precise, we must say that metaphony is selected by the masculine singular inflection and by the 2nd sg and 3rd person of the III verbal class present. We can discover no phonological property that separates metaphonizing [i u] from inert [i u], unless different abstract underlying qualities are involved. Morphologized metaphonies like that of Settimo S. Pietro in (12) demand a partial revision in formalizing metaphony. The non-arbitrary link between high-mid nuclei and spreading of [+ATR] is still valid; however we must be able to link metaphony to the morphological context. We conclude that the phonological content of a sub-set of inflectional exponents -u/-i is not working, although it is phonetically interpreted. If the phonological content of [i u] was involved, we would expect all [+high] vowels to trigger metaphony. In other words, in these varieties metaphony does not depend on phonetic properties of post-tonic vowels anymore; it is now a morphologized process triggered by morphological information (cf. Calabrese 1998).

22 

 Leonardo M. Savoia

In the case here examined, the nominal inflectional property of II class masculine (m.sg) and verbal inflectional features of 2nd sg / 3rd person of III class, are externalized by metaphonic properties realized on the stressed nucleus, as in (14b). In this sense, the metaphonic outcome of the stressed nucleus is the true exponent of these morpho-syntactic categories. On the other hand, contrastive values of [ATR] are licensed by the stressed nucleus, as in (14a). Finally, (14c) schematizes the process, where [+ATR] is introduced in correspondence to the specification m.sg as a property subsumed by the stressed head, that licenses it. (14)  a.  Morphologized [+ATR] metaphony (Settimo S. Pietro) [ATR] contrastive is authorized on the stressed nucleus.  b. Inflections m.sg, 2nd sg / 3rd person of III class verbs require [+ATR] in the stressed nucleus [−high, −low]. c. N N N

> O

N

x

R N x

x

x

/ f

ε

O

x

[−high] [−ATR]

r

+ um.sg / [+ATR]

N

> O

N

x

R N x

x

x

f

e

r

u

O →

x

[−high] [+ATR]m.sg

[−ATR]   The classical type of morphologized metaphony in Southern Italian varieties is due to the neutralization to [ə] of post-tonic vowels, as in (15) for Andria (Apulia). Metaphony of low-mid vowels is illustrated in (15a) and that of highmid vowels in (15b). The outcomes of metaphony of original high-mid vowels follow the phonological development of the original high vowels /i u/, which diphthongize in open syllables and centralize in closed syllables. Thus [oi eu] in open syllables and [ɪ ə] in closed ones correspond to the metaphonic outcomes. However, for the sake of simplicity, in the discussion we concentrate on low-mid metaphony in (15a). (15)   Andria a. [ˈdɛndə] ‘tooth’  [ˈdində] ‘teeth’ [ˈpɛrdə] / [ˈpɛrdə] / [ˈpɛrdənə] ‘I lose / he loses / they lose’   [ˈpirdə] ‘you lose’ [ˈɔccərə] ‘eyes’   [ˈuccə] ‘eye’ [ˈʃɔukə] / [ˈʃɔukə] / [ˈʃɔkənə] ‘I play / he plays / they play’         [ˈʃu:kə] ‘you play’

Harmonic processes and metaphony in some Italian varieties 



 b. [ˈmaisə] ‘month’ [ˈkrɛdənə] ‘they believe’ [ˈnautʃə] ‘walnut’ [ˈkɔrtə] ‘short-f’

 23

[ˈmoisə] ‘months’ [ˈkrɪdə-mə] ‘believe me!’ [ˈneutʃə] ‘walnuts’ [ˈkərtə] ‘short-m(.pl)’

In (15) metaphonic contexts are opaque given that only a sub-set of post-tonic [ə], deriving from etymological [i u], trigger metaphony. Nevertheless, metaphony appears to be a regular phonological mechanism that realizes the specification [+high, +ATR] in alternation with low-mid vowels. According to the preceding discussion, we relate metaphony to inflectional categories, as in (16a, b). (16) a. [+high, +ATR] are licensed in the stressed nucleus. b. Inflectional categories m.sg, m.pl, III noun class pl, 2nd sg / 3rd pl of the present associate [+high, +ATR] with the stressed nucleus. c.  Andria N N

>

O

N

O

N

x

x

x

x

∫

u [−high]

x

k + ə 2sg [+high] [+ATR]

Interestingly, the result of metaphony of low-mid lexical nuclei in this variety is a [+high] outcome, i.e. [i u]. This implies that lexical low-mid content is replaced by the metaphonic specification that needs to be licensed by the stressed nucleus, as in (16c).

3 H  armony and metaphony in Central Sardinian varieties Metaphony combines with restrictions on the distribution of unstressed vowels in Central Sardinian varieties. The domain of this harmonization is the same as for metaphony, i.e. the strong foot or the sequence including the stressed nucleus and week nuclei, both the following and preceding ones. The ­harmonic

24 

 Leonardo M. Savoia

­ istribution of aperture properties in the stressed nucleus and the weak fold lowing nucleus characterizes some Central Sardinian varieties, like those of Àllai, Làconi (cf. Wagner 1941; Contini 1987; Loporcaro 2002−03, 20053; Savoia 2005). In particular the constraints on the vocalic sequences include metaphony as well, that, as we saw in (6b), affects the [ATR] content of stressed mid vowels followed by post-tonic [i]/[u]. (11a) is in turn operating in limiting the occurrence of [+ATR] in final position. Moreover, in the contexts including an intermediate onset, the occurrence of a low-mid vowel [ɛ]/[ɔ] in post-tonic position implies a low-mid vowel in stressed position, as illustrated in (17), but the contrary is not true. Stressed [ɛ]/[ɔ] can co-occur with [a] in post-tonic position, whereas posttonic [ɛ]/[ɔ] are excluded from the contexts in which the stressed vowel is [a], as in (18). (17)  Àllai [ˈkɛrdzɔ] / [ˈkɛrɛzɛ] ‘I want / you want’ [ˈbɛttʃa] / [ˈbɛttʃɔzɔ] ‘old-f.sg / old-m.pl’ [ˈmɔvvɔ] / [ˈmɔvvɛzɛ] ‘I move / you move’ [nɛˈßɔðɛ] / [nɛˈßɔðɛzɛ] ‘nephew / nephews’ Làconi [ˈbɔʒɔ] / [ˈbɔllɛzɛ] ‘I want / you want’ [ˈɔɣɔzɔ] ‘eyes’ [ˈpɛrdɔ] / [ˈpɛrdɛzɛ] ‘I lose / you lose’ [ˈbɛttʃa] / [ˈbɛttʃɔzɔ] / [ˈbɛttʃaza] ‘old-f.sg / old-m.pl / old-f.pl’ Vowels [i a u] freely co-occur in the positions here considered, as in (18). So, stressed [a] co-occurs with post-tonic [i a u], and correspondingly stressed [i]/[u] co-occur with post-tonic [i a u]. (18)   Àllai [ˈsannu] / [ˈsanna] / [ˈsannuzu] / [ˈsannaza] ‘healthy-m.sg / f.sg / m.pl / f.pl’ [ˈpatti] ‘part, portion’

3 The correlation between final vowels and stressed vowels is noticed for the dialects of Làconi and Baunei in Wagner (1941: 39): “In Làconi hab ich beobachtet, dass -e in Ausgang nach vorhergehenden -e oder -o bevorzugt wird, sonst meistens -i; ebenso im Pl. -os mit Vorliebe nach e und o, sonst -us …”. The harmony of Àllai and Sèneghe was first highlighted in Loporcaro (2002–03, 2003). In Contini (1987: 443–444) the presence of both mid and high final vowels in the dialects of Central Sardinia is observed; however, Contini does not consider the relation between the mid and high final vowels and the quality of stressed vowels. He notes that the infinitival inflection -re generally resists raising.



Harmonic processes and metaphony in some Italian varieties 

 25

[ˈliddʒu] / [ˈliddʒizi] ‘I read / you read’ [ˈkrikku] / [ˈkrikkaza] ‘I look for / you look for’ [ˈsuðru] / [ˈsuðra] / [ˈsuðruzu] / [ˈsuðraza] ‘deaf-m.sg / f.sg / m.pl / f.pl’ Làconi [ˈfraði] ‘brother’ [ˈsannu] / [ˈsanna] / ['sannuzu] / [ˈsannaza] ‘healthy-m.sg / f.sg / m.pl / f.pl’ [ˈliddʒu] / [ˈliddʒizi] ‘I read / you read’ [ˈitta] ‘what’ [ˈsurdu] / [ˈsurda] / [ˈsurduzu] / [ˈsurdaza] ‘deaf-m.sg / f.sg / m.pl / f.pl’ High-mid vowels [e]/[o] occur only in stressed position when followed by ­post-tonic [i u], as in (19), agreeing with the requirements of metaphony analyzed in (6b). (19)   Àllai Làconi [apˈperizi] ‘you open’ [aˈβerrizi] ‘you open’ [ˈbettʃu] ‘old-m.sg’ [ˈbettʃu] ‘old-m.sg’ [ˈmorizi] ‘you die’ [ˈdrommizi] ‘you sleep’ [ˈoɣu] ‘eye’ A different distribution appears in the contexts in which the stressed nucleus and the following unstressed nucleus are adjacent, i.e. in strings where the intermediate onset is absent. In the dialect of Àllai, in these contexts, exemplified in (20a)–(20b), stressed [ɛ]/[ɔ] co-occur with post-tonic [ɔ]/[ɛ] and post-tonic [i]/[u] trigger metaphony combining with stressed high-mid vowels, as in [ˈmeu] ‘my’. When the stressed nucleus is [+high], parametric variation shows up. In particular, high stressed nuclei can co-occur with low-mid vowels, in (20a), conflicting with (17). (20) Àllai a. [ˈniɛ] ‘snow’ [ˈbiu] / [ˈbiɛzɛ] ‘I see / you see’ [ˈmia] / [ˈmiaza] ‘my-f.sg / f.pl’ b. [ˈdʒɛɔ] ‘I’ [ˈpɛi] / [ˈpɛzi] ‘foot / feet’ [ˈkrɛi] ‘to believe’ [ˈmeu] / [ˈmɛɔzɔ] ‘my-m.sg / m.pl’

[ˈtui] ‘thou’ [ˈtua] / [ˈtuaza] ‘your-f.sg / f.pl’ [ˈtuɔ] / [ˈtuɔzɔ] ‘your-m.sg / m.pl’ [ˈɔɛ] / [ˈɔɛzɛ] ‘ox / oxen’

[ˈnou] ‘new-m.sg’

In the Làconi system the only low vowel that can follow a high stressed nucleus is [a], as in (21a); thus, a form like [ˈkrua] ‘raw’, encompasses both masculine and feminine reference. Data concerning mid stressed vowels are presented in (21b).

26 

 Leonardo M. Savoia

(21)   Làconi a. [ˈnia] ‘snow’ [ˈmia] / [ˈmiaza] ‘my-f.sg / f.pl’ [ˈbiu] / [ˈbiaza] ‘I see / you see’ b. [ˈmeu] / [ˈmɛɔzɔ] ‘my-m.sg / m.pl’ [ˈpɛi] / [ˈpɛizi] ‘foot / feet’ [ˈðɛɔ] ‘I’

[ˈtua] / [ˈtuaza] ‘your-sg / pl’ [ˈkrua] / [ˈkruaza] ‘raw-sg / pl’ [ˈnou] / [ˈnɔa] ‘new-m.sg / f.sg’ [ˈnɔɔzɔ] / [ˈnɔaza] ‘new-m.pl / f.pl’ [ˈprɔɛðɛ] ‘it rains’

As we saw in (3), in Sardinian varieties the distribution of mid stressed vowels takes into account the height of the post-tonic vowel. In the vocalic systems of Àllai and Làconi in (22) distributional constraints that relate the height degree of post-tonic vowels to the height degree of stressed ones let [−high, −ATR] combine both with final low-mid vowels and [a], preserving its contrastive status. In parallel, metaphonic outcome [−high, +ATR] is contrastive with respect to stressed [i u], since the latter require in turn final vowels [+high, +ATR]. As to mid vowels, [−low, −ATR] contrasts with [−low, +ATR]. (22)   Àllai, Làconi i u ɛ ɔ e o a

[+high, −low, +ATR] [−high, −low, −ATR] [−high, −low, +ATR] [−high, +low, −ATR]

The distribution in (17)–(21) suggests that height properties of post-tonic vowels are licensed by the stressed vowel, i.e. the head of the prosodic domain. In this perspective, metaphony represents only one among the different harmonic devices which govern the possible combinations of vowels in the domain of the stressed nucleus, i.e. the foot or the phonological word. A problem stands out for the classical generative framework, i.e. the fact that [a] and the high vowels [i u] form the class of the vowels that exclude post-tonic mid vowels. On the contrary, the traditional feature composition proposed in Chomsky & Halle (1968) gathers [a] together with low-mid by means of the specification [−high, +low], whereas it cannot capture a natural class including [a] and [+high] vowels, [i u]. This difficulty remains in a model with [ATR], as (22), where [a] [+low, −ATR] cannot be put together with [+high, +ATR] vowels [i u]. As noted by Loporcaro (2002–03, 2005) and Savoia (2005) the distribution in (17)–(18) recalls conditions discussed in the literature as regards height harmony in Chicheŵa (Harris 1990a, 1994b), and in other Bantu languages (Kula & Marten 2000), where the nucleus of the verbal stem fixes the height degree of the vowel in the suffix. In Chicheŵa, a on a par with i u selects [+high] vowels in suffixes; Archangeli & Pulleyblank (1989, 1994) and Scullen (1992) connect the harmonic



Harmonic processes and metaphony in some Italian varieties 

 27

process with spreading of [−high] only from mid vowels. In these analyses [+high] outcomes are derived from the application of default rules which assign [+high] to the underspecified vowel in the suffix. In other words, the coincidence of the phonological outcomes in the context of [a]/[i u] is nothing but the surface product of the non-application of harmony. Harris (1990a, 1994b) has recoursed to an ‘element’ approach in which it is the element [A] which spreads, whereas the high outcomes [i u] simply realize the lexical content specified by the elements [I]/[U]. We can suppose that also in the harmonies in Sardinian dialects the relation between [i u] and [a] is only a surface effect of different conspiring processes. Actually, according to Contini (1987), these harmonizing systems represent the passage between the dialects of Logudoro, preserving low-mid final vowels and Campidanian ones that neutralize final mid vowels in high vowels. Loporcaro (2002–03, 2003) provides an explanation involving mechanisms internal to the system. In particular, according to Loporcaro (2003: 41) the existence of Central Sardinian dialects with the harmonic distribution of mid and high final vowels would confirm that Campidanian raising of final mid vowels worked through intermediate stages partially reflected by the areal distribution. The change of mid final vowels to high vowels would have been driven by lexical, morphological and phonological factors, in particular the quality of the stressed vowels. The insight is that harmony corresponds to an incomplete expansion of the neutralization in [+high] vowels expressed in (13b). In other words, the distribution attested by the dialects of Àllai and Làconi would depend on a more limited application of neutralization. In particular neutralization in (13b) interacts with metaphony, since in harmonizing dialects metaphonic contexts with [+high] post-tonic vowels are kept separated from post-tonic mid vowels which are preserved in the context of mid stressed vowels. Neutralization applies to the contexts where stressed nuclei occur which are not involved in metaphony, i.e. [i a u]. We will assume that the distribution of unstressed vowels is governed by a constraint which connects the properties of the post-tonic mid vowel with the ones of the stressed vowel, while metaphony obeys the constraint proposed in (6b) for the variety of Paulilàtino. The distribution of mid post-tonic vowels depends on a requirement that relates stressed low-mid vowels to post-tonic low-mid ones, as in (17), and stressed [i a u] to following [i a u], as in (18). Posttonic [ɛ ɔ] are legitimated by the presence of the same specification [−low, −ATR] in the stressed vowel, as in (23). As matters stand, all other combinations include high vowels in post-tonic position, and in the case of a stressed mid vowel, the application of (6b) generates the metaphonic outcome [e o]. (23) Harmony of low-mid vowels  The contrastive specification [−low, −ATR] in the stressed nucleus licenses [−low, −ATR] in a following vowel.

28 

 Leonardo M. Savoia

The condition in (23) accounts for sequences like (24a), excluding [a], that, actually, is able to follow every stressed nucleus and to precede post-tonic [i u], as in (24b). (24)  Làconi N  a.

x

x

m

O

N

O

x

x

x

x

x x

x

s

a

n

u

>

x

>

N R N

c

O

N R N

N

b.

v

[+low]

c

[−low] [−ATR]



 

O

N

[+ATR] [+high]

Hiatus contexts in the Àllai variety in (20) block or relax the requirement in (23), since stressed [i u] co-occur both with [a] and post-tonic [ɛ ɔ]. This can be accounted for by the prosodic structure of the hiatus sequences, where the unstressed nucleus does not license a preceding onset. As a consequence, the stressed nucleus licenses the phonological content of the following nucleus without having to authorize the post-tonic nucleus as a licenser. In this position, then, an unstressed mid vowel can be licensed by a stressed high nucleus as well, as in (25b) for the string i … ɛ. The only remaining restriction obliges the stressed and post-tonic nuclei to share the same cavity properties: stressed [i] excludes the combination with [+back] and [u] with [−back]. (25)  Àllai a.  [+high, αback] in the stressed nucleus licenses an adjacent unstressed contrastive specification [−high, αback] inside the foot. N  b. N

>

O

N

O

x

x

x

n

i

ε

[+high]

[−back]

N



Harmonic processes and metaphony in some Italian varieties 

 29

Morpho-lexical contexts can influence the distribution of the process (Loporcaro 2005). Consider in this connection the data of the variety spoken in Sèneghe (also examined by Loporcaro 2005). In this system harmony applies to the same phonological contexts as in Àllai and Làconi, i.e. in the position following stressed [i a u]. (26a) exemplifies the co-occurrence of low-mid vowels ɛ/ɔ - ɛ/ɔ and ɛ/ɔ-a. (26b) illustrates harmony in the sequences verbal stem-inflection combining i/u-a. (26c) illustrates metaphony, which reproduces the distribution observed in Sardinian dialects like those of Paulilàtino, Làconi and Àllai. (26) Sèneghe a. [ˈkɛrdzɔ] / [ˈkɛrɛzɛ] ‘I want / you want’ [ˈbɛttsa] / [ˈbɛttsɔzɔ] / [ˈbɛttsaza] ‘old-f.sg / m.pl / f.pl’ [ˈmɔvvɔ] / [ˈmɔvvɛzɛ] ‘I move / you move’ a’. [ˈpɛa] [ˈpɛaza] ‘foot / feet’ [ˈkɔa] / [ˈkɔaza] ‘tail / tails’ b. [ˈsamunu] / [ˈsamunaza] ‘I wash / you wash’ [ˈliu] / [ˈliaza] ‘I take off / you take off’ [ˈbuffu] / [ˈbuffaza] ‘I blow / you blow’ [ˈtua] / [ˈtui] / [ˈtuaza] ‘your-m.sg / f.sg / m.pl / f.pl’ c. [apˈpɛrdzɔ] / [apˈperizi] ‘I open / you open’ [ˈlettu] / [ˈlɛttɔzɔ] ‘bed / beds’ [ˈoɣu] / [ˈɔɣɔzɔ] ‘eye / eyes’ [ˈdrɔmmɔ] / [ˈdrommizi] ‘I sleep / you sleep’ However, in Sèneghe some morphological limitations are in force. The harmony does not apply to the masculine plural inflection -ɔs, as in (27a); moreover, it excludes other lexical sub-sets, like the form [ˈbattɔrɔ] ‘four’. Furthermore, [ɛ] is allowed to follow stressed [a], as in (27b), while it is banned in the contexts ­following stressed [i u], where harmonic outcome [i] occurs, as in (27b’). In the contexts of hiatus in (27c, c’), the sequences a-ɛ and i-ɛ are allowed. (27) Sèneghe   a. [ˈsãũ] / [ˈsana] / [ˈsãɔzɔ] / [ˈsanaza] ‘healty-m.sg / f.sg / m.pl / f.pl’ [fraˈittsu] / [fraˈittsɔzɔ] ‘fox / foxes’ [kru] / [ˈkrua] / [ˈkruɔzɔ] / [ˈkruaza] ‘raw-m.sg / f.sg / m.pl / f.pl’   b. [ˈfardɛ] ‘brother’, [ˈfaɛðɛ] ‘to make’   b’. [ˈnuɣi] ‘walnut’, [ˈpiski] ‘fish’   c. [ˈnia] ‘snow’   c’. [ˈbiu] / [ˈbiɛzɛ] / [ˈbiɛðɛ] ‘I see / you see / he sees’ As we saw in the case of metaphony, we can expect that morpho-phonological procedures make syntactic information show up. In this case Sèneghe’s speakers

30 

 Leonardo M. Savoia

apply the constraint (23) only to the verbs, i.e. syntactic objects of the form [v X], as in (28). (28)  In a phonological string associated with a terminal node v, the contrastive specification [−low, −ATR] in the stressed nucleus licenses [−low, −ATR] in a following vowel. Differently from the noun, in the variety of Sèneghe verb inflections -a/-i/-u are subject to harmonic distribution. As for other combinations, the only context which is left out by (23) is the sequence a … ɛ. This suggests that (23) must be formulated in a more adequate way, as in (29a). In (29a) the relevant requirement is that [−ATR] in post-tonic position is authorized by [−ATR] in stressed position. The restriction on the contrastivity of [ATR] singling out only mid vowels is no longer working, as shown in (29b), where sharing of the feature [−ATR] is independent of [+/−low] nature of the stressed vowel. Naturally, (29a) operates over the vowel inventory of this variety. (29) a.  Harmony of low-mid vowels in Sèneghe [−ATR] in the stressed nucleus licenses [−ATR] in post-tonic position.  b.

c. 

N

O

N

O

x

x x

x

x

x

c

v

c

m

>

N R N

f

a

N R O N x

N

[−ATR]



> O

N

x

x

x

r

d

ε

[−ATR]

Loporcaro (2003) aims at explaining the current harmonic distribution of the vocalic properties in Làconi, Àllai, and Sèneghe varieties, connecting it with the generalized raising of Southern Sardinian (Campidanian) varieties, like that of Settimo S. Pietro in (12). His idea is that the harmonic constraints which govern the distribution of the height degree of final vowels in the dialects here examined preserve the original conditions which would have initially triggered the raising of final vowels in Campidanian varieties. More precisely, according to Loporcaro (2003), raising of final vowels in Campidanian dialects would derive from an initial mechanism of coarticulation whereby stressed vowels have influenced the raising of post-tonic final vowels. These conditions would be preserved just by the harmonic distribution in dialects like Làconi and Àllai. In other words, the intermediate area including these dialects would show that the original phonetic cause for raising of final vowels would be a



Harmonic processes and metaphony in some Italian varieties 

 31

type of assimilation at a distance rather than a simple process of optimization of the vowel space. Thus, perceptive and articulatory mechanisms like maximization of the articulatory space, coarticulation, inertia, economy would be at work, possibly driven by morphological and lexical constraints. In particular, we saw that in the variety of Sèneghe the plural inflection -ɔs escapes raising in harmonic contexts. In accordance with Loporcaro (2002–03, 2003) we can think that external functional factors are involved which prevent neutralization. A reasonable hypothesis is that the plural morpheme -ɔs has preserved its low-mid vowel in virtue of the fact that it contrasts with other -us inflections occurring in nouns or verbs. Indeed, the inflection -us characterizes the singular of nouns like [ˈtempuzu] ‘time’, [ˈkorpuzu] ‘body’, alternating with plural forms [ˈtɛmpɔzɔ] ‘times’, [ˈkɔrpɔzɔ] ‘bodies’, and generally contrasting with the occurrence of plural -os. Moreover, -us occurs as 1st plural inflection in verbs, cf. [buf ˈfauzu] ‘we.drink’. The need for preserving the morphological difference between -ɔs and -us can explain why harmonic raising did not extend to -ɔs inflection. In the case of -ɔrɔ in [ˈbattɔrɔ] ‘four’ of Sèneghe, Loporcaro (2002–03) assumes that it is preserved on the analogy of -ɔs, given that in many Sardinian dialects in sandhi final -r has the same phonetic outcomes as the final -s. This is possible; nevertheless I note that the phonetic identity in the realization of -r and -s is not systematically attested in the Sardinian varieties. In particular, speakers of Sèneghe realize the form [ˈbattɔrɔ] with final vowel in sandhi too; moreover in this dialect pre-consonant -r- is maintained. An evolutionary perspective adds interesting elements to the comprehension of this process, specifying the relation between external and internal factors. Indeed, we can think that the external factors that Loporcaro indicates were involved in causing and driving the harmonic distribution in the varieties here studied. More precisely, factors like coarticulation, maximization of phonetic space and generalization, etc. can be conceived as cognitive or perceptive mechanisms working along the boundary between sensory-motor and thinking systems. They would concur to fix the morpho-phonological units which externalize the computational system, in the sense discussed by Berwick & Chomsky (2011)4. When the phonetic processes

4 Berwick & Chomsky (2011: 37) note that “Externalization … has to relate two quite distinct systems: one is a sensorimotor system that appears to have been basically intact for hundreds of thousands of years; the second is a newly emerged computational system for thought, which is perfect, insofar as the strong minimalist thesis is correct. We would expect, then, that morphology and phonology—the linguistic processes that convert internal syntactic objects to the entities accessible to the sensorimotor system—might turn out to be quite intricate, varied, and subject to accidental historical events. Parameterization and diversity, then, would be mostly—possibly entirely—restricted to externalization”.

32 

 Leonardo M. Savoia

or morphological/lexical constraints driving harmony have been acquired by children, they become a part of their internal language system.

4 Harmony in Sicilian varieties Some varieties of Central Southern Sicily are characterized by a harmony process involving the distribution of [+ATR]/[−ATR] in the vowels inside the domain of word5. This phenomenon has been described in Rohlfs (1966 [1949]) and ­Piccitto (1959) and analyzed in detail in Mocciaro (1980) and Cruschina (2007). By virtue of this process [(+high), −low] vowels in the word domain assimilate on the value [−ATR] except in metaphonic outcomes [iə uə]/[i u] of mid vowels; crucially, the post-tonic/final vowel [+low] blocks the assimilation. Consider the data of varieties spoken in some villages in the province of Caltanissetta, precisely Milena, Sutera, Villapriolo. Given that the data are very similar in all the varieties, we will use in particular the data of Milena for simplicity’s sake (a complete presentation is in Savoia 2015). Realizations [+high, +ATR] occur in stressed position in contexts where the following vowel is [a], as in (30a, 30’a, b), in the post-tonic span of a stressed head [+ATR], as in (30’c), and, possibly in the pretonic domain of stressed [a], as in (30’b). Non-final post-tonic [a] is able to select [+ATR] in the stressed [+high], as in [pırˈdivatʊ] ‘you lost’ in (30’c). In oxytones in (30’d) the stressed nucleus is [−ATR] exactly as in the strings with post-tonic [+high] vowels in (30b) and with post-tonic [+high] vowels in (30’a, b, c). Stressed [a] in (30b) and metaphonic diphthongs [iə uə] in (30d) combine with final [ı ʊ], [−ATR]. (30)  Milena     Paroxytone strings V-v a. i-a [ˈfiʎʎa] ‘daughter’ [ˈtira] ‘pull-3sg’ o-a [ˈrota] ‘wheel’ [ˈkoʃʃa] ‘thigh’ [ˈgrossa] ‘big-f.sg’

u-a [ˈkruda] ‘raw-f.sg’ [ˈfuma] ‘smoke-3sg’

e-a [ˈbeɖɖa] ‘nice-f.sg’ [ˈleva] ‘take.off-3sg’

a-a [ˈamma] ‘leg’

5 Really, the pronunciation of high vowels of the type we characterize as [−ATR] is attested in many Sicilian varieties, as in (i) for S. Marco d’Alunzio, though it is variable and it does not show clear distributional constraints (cf. the discussion in Cruschina 2006). (i) [ˈpjɛrІ] ‘foot / feet’, [ˈkwɔrІ] ‘heart / hearts’, [ˈjirІtʊ] ‘finger’, [kaˈnʊʃʃʊ] ‘I know’, [ˈfrІjʊ] ‘I fry’



Harmonic processes and metaphony in some Italian varieties 

 33

 b. ı-ı ı-ʊ ʊ-ı ʊ-ʊ [ˈfɪʎʎɪ] ‘sons’ [ˈfɪʎʎʊ] ‘son’ [ˈkrʊdɪ] ‘raw-pl’ [ˈkrʊdʊ] ‘raw-m.sg’ [ˈtɪrɪ] ‘pull-2sg’ [ˈtɪrʊ] ‘pull-1sg’ [ˈfʊmɪ] ‘smoke-2sg’ [ˈfʊmʊ] ‘smoke-1sg’ a-ı a-ʊ [ˈammɪ] ‘legs’ [ˈvrattsʊ] ‘arm’  c.  ɛ-ı ɔ-ı ɔ-ʊ [ˈpɛðɪ] ‘foot’ [ˈrɔtɪ] ‘wheels’ [ˈɔttʊ] ‘eight’ [ˈnɛʃʃɪ] ‘go out-3sg’ [ˈkɔʃʃɪ] ‘thighs’ [ˈɔmʊ] ‘man’ [ˈdɔrmɪ] ‘sleep-3sg’  d. iə-ı iə-ʊ uə-ı [ˈpiəðɪ] ‘feet’ [ˈviəccʊ] ‘old-m.sg’ [ˈuəccɪ] ‘eyes’ [ˈbiəɖɖɪ] ‘nice-pl’ [ˈniəʃʃʊ] ‘go out-1sg’ [ˈgruəssɪ] ‘big-pl’ [ˈniəʃʃɪ] ‘go out-2sg’ [ˈduərmɪ] ‘sleep-2sg’ uə-ʊ [ˈuəccʊ] ‘eye’ [ˈgruəssʊ] ‘big-m.sg’ [ˈduərmʊ] ‘sleep-1sg’ (30’) Milena Paroxytone strings v-V-v  a. a-i/u-a a-e/o-a [katˈtiva] ‘widow’ [asˈpetta] ‘wait-3sg’ [kaˈrusa] ‘girl’



a-a-ı/ʊ [caˈmarɪ] ‘to call’

a-V-ı/ʊ [katˈtɪvɪ] ‘widows’ [aʃˈtʊtʊ] ‘put out-1sg’ [asˈpiəttʊ] ‘wait-1sg’  b. [+high]-[+ATR] [+high]-[−high] [+high]-a [mɪ / iˈluna] ‘melons’ [kʊ / uˈteɖɖa] ‘knives’ [li / ɪˈvatɪ] ‘take off-2pl’ [krɪ / iˈðiva] ‘believe-3sg’ [pu / ʊrˈtammʊ] ‘bring-1pl’

[+high]-[−ATR] [+high]-iə-/-uə [nɪʃˈʃɪvʊ] ‘went out-1sg’ [kʊˈtiəɖɖʊ] ‘knife’ [vɪˈvɪtɪ] ‘drink-2pl’ [nɪˈʃiəmmʊ] ‘go out-1pl’ [mɪˈlʊnɪ] ‘melon’ [dʊrˈmiəmmʊ] ‘sleep-1pl’  c.  Proparoxytones V-[+high]-a V-[+high]-[−ATR] V-a-[+high] [fʊrˈmikula] ‘ant’ [fʊrˈmɪkʊlɪ] ‘ants’ [ˈjennarʊ] ‘son-in-law’ [ˈnivika] ‘snow-3sg’ [ˈfɪmmɪnɪ] ‘women’ [fʊˈmavamʊ] ‘smoked-1pl’ [ˈfimmina] ‘woman’ [ˈrʊppɪrʊ] ‘broke-3pl’ [ˈʎommarʊ] ‘ball’ [ˈniura] ‘black-f.sg’ [ˈnɪʊrʊ] ‘black-m.sg’ [pɪrˈdivatʊ] ‘lost-2sg’

34 

 Leonardo M. Savoia

ɛ/ɔ-v-v [ˈkrɛðʊnʊ] ‘believe-3pl’ iə-/uə-v-v [ˈpiəkura] ‘sheep’ [ˈpiərdɪnʊ] ‘lose-3pl’ [ˈduərmɪnʊ] ‘sleep-3pl’ [ˈuəmɪnɪ] ‘men’  d.  Oxytones [−ATR] [nɪʃˈʃɪ] ‘went out-3sg’ [fɪˈnɪ] ‘finished-3sg’ [tʊ] ‘thou’

[−high] [jɛ] ‘be-2sg’ [sɛ] ‘be-2sg’ [pɔ] ‘can-3sg’

a [kka] ‘here’ [sa] ‘know-2sg’

Mid vowels do not occur in unstressed positions; in these contexts [a] and [+high] vowels are admitted in correspondence to [−low] underlying vowels, as illustrated in (31). (31)  Milena [ˈlava] / [laˈvamʊ] ‘he washes / we wash’ [arˈrıdʊ] / [arrıˈdiəmmʊ] ‘I laugh / we laugh’ [ˈpɛrdɪ] / [pɪrˈdiəmmʊ] ‘he loses / we lose’ [ˈporta] / [pʊrˈtammʊ] ‘he bring / we bring’ Villapriolo [ˈvɛnɪ] / [vɪˈnɪmʊ] ‘he comes / we come’ [ˈjɔka] / [jʊˈkamʊ] ‘he plays / we play’ The vocalic system of these varieties can be characterized as in (32). (32)

 Milena i u ı ʊ e o ɛ ɔ a

[+high, −low, +ATR] [+high, −low, −ATR] [−high, −low, +ATR] [−high, −low, −ATR] [−high, +low, −ATR]

The scatterplots in (33)–(34) show the distribution of the vocalic types [+high], i.e [i u] [+ATR] and [ı ʊ] [−ATR], [−high], i.e. [e o] vs. [ɛ ɔ], and of [a] in the acoustic space. Average values are calculated on the basis of a corpus including 250 realizations obtained by means of field research with native informants. In particular, the values represented in graphs in (33)–(34) concern productions of a female native speaker of the dialect. The realizations recorded have been subjected to a spectrographic analysis using the current software Praat. As we

Harmonic processes and metaphony in some Italian varieties 



 35

can see, average differences in frequency between F1 and F2 in particular in [+high] vowels, distinguish two different vocalic types, sensitive to the context. The F1 / F2 values can be related respectively to [i u] and to lax [ı ʊ] on the basis of the vowel prototypes of the corpus UPSID (UCLA Phonological Segment Inventory Database). In the context Xa, we find average frequencies of 348 Hz and 2818 Hz for F1 and F2 respectively in [i], and average F1 of 366 Hz and F2 of 885 Hz for [u]; in the harmonizing context, we find [i] with an average F1 of 537 Hz and an average F2 of 2464 Hz, [ʊ] with average values of 547 Hz for F1 and of 1046 Hz for F26. (33)7



6 A complete statistical analysis of the data is presented in Savoia (2015). 7 The mean values for Milena in the scatterplot in (33) are detailed in (a) and (b) a. i(ə) F1 318,62 Hz, F2 2818,18 Hz u(ə) F1 366,28 Hz, F2 885 Hz iXa F1 348,21 Hz, F2 2826,28 Hz uXa F1 366,28 Hz, F2 1046,7 Hz ɪXɪ / ʊ F1 537,52 Hz, F2 2464,56 Hz ʊXɪ / ʊ F1 547,6 Hz, F2 1046,7 Hz eXa F1 479,16 Hz, F2 2469 Hz oXa F1 596,25 Hz, F2 1232,5 Hz ɛXɪ / ʊ F1 620,44 Hz, F2 2241,88 Hz ɔXɪ / ʊ F1 609,25 Hz, F2 1143 Hz a F1 826,17 Hz, F2 1658,35 Hz b. ɪ# F1 533,3 Hz, F2 2462,3 Hz ʊ# F1 565,8 Hz, F2 1047,12 Hz

36 

 Leonardo M. Savoia

(34)8



As regards the significance level of differences between the means carried in the graphs, we note that critical values of the Student’s t distribution are satisfied for P < 0,01 in the case of the most important contrasts. Thus, the contrast between vowels [+ATR], with higher mean values for F1 and lower mean values for F2, corresponds to significant differences. The difference is stronger in the [−back] series, which therefore realizes in a more recognizable and perceptible way the contrast between [+ATR] and [−ATR]. In the case of unstressed realizations, we

8 The values of F1 and F2 for Milena in (34) are reported in a–b. a. post-tonic context ˈVXiXa F1 332 Hz, F2 2738,5 Hz ˈVXuXa F1 399,5 Hz, F2 1213 Hz ˈVXɪ(Xɪ / ʊ) F1 483,31 Hz, F2 2400,29 Hz ˈVXʊ(Xɪ / ʊ) F1 512,61 Hz, F2 1266,31 Hz a F1 705,68 Hz, F2 1708 Hz b. pretonic context ɪXˈi / u F1 426,18 Hz, F2 2476,09 Hz uXˈi / u F1 447,6 Hz, F2 1173,4 Hz ɪXˈɪ F1 516,5 Hz, F2 2307,62 Hz ʊXˈɪ / ʊ F1 546 Hz, F2 1227,33 Hz ɪXˈa F1 524,5 Hz, F2 2228 Hz ʊXˈa F1 492,4 Hz, F2 1302,2 Hz a F1 680,93 Hz, F2 1767,81 Hz

Harmonic processes and metaphony in some Italian varieties 



 37

note that they are generally less clearly differentiated in the different contexts and that they tend to more centralized and laxer realizations (see also Cruschina 2007). In particular in the pretonic context Xˈa a realization occurs which is more centralized than stressed alternants before post-tonic [a]. This is the general case for the [+ATR] realizations. The mean values of F1 and F2 in pretonic [i] and stressed [i] are significantly different for P < 0,01; in fact, pretonic [i] is on average more centralized than stressed [i]. Again, the differences for [u] do not gain the significance level for Student’s t. The table in (35) schematizes this distribution, where [+high, +ATR] realizations occur only in stressed position, in the context of final [a], as in [ˈfiʎ ʎa] ‘daughter’ and in the domain of a [+ATR] stressed vowel, in (30’c), as [ˈfimmina] ‘woman’, i.e. when they are followed by [a]. In the pretonic domain [+high] vowels are normally more centralized; [+ATR] realizations are preferred before a [+ATR] stressed vowel, as in [miˈluna] ‘melons’ in (30’b). In other stressed contexts, including stressed [a], outcomes are basically [−ATR], as in [lɪˈvatı] ‘you.pl take off’. (35)  [+high] stressed vowels [+high] post-tonic vowels [+ATR] / [−ATR] / [−ATR]#

Xa [−ATR]

[+/−ATR] / [−ATR]

Xa

[+high] pretonic vowels [+/−ATR] / [−ATR]

Xˈi / ˈu / ˈa

My informants of Milena favour [ı ʊ] in pretonic contexts independently of the aperture degree of the stressed vowel. Data from Sutera, although generally analogous to those of Milena, show a stronger variation. We can think that acoustic properties of unstressed realizations are at least partially governed by prosodic devices, whereby the vowels in weak position tend to neutralize. The clearest generalization is that [+ATR] stressed nuclei are able to control the realization [+ATR] in pretonic contexts, whereas stressed [a] seems to lack or limit this ability. In other words, the fact that tonic [a] and tonic [+ATR] nuclei lack the capability of licensing [+ATR] suggests that the harmonic domain includes the stressed nucleus and the weak vowels on its right. The occurrence of [+ATR]/[−ATR] is licensed in the immediate domain of the stressed nucleus, i.e. its foot (paroxytones/oxytones) or an enlarged span including its foot (proparoxytones). Let us now consider the phonological mechanism which governs harmony in these dialects. Evidence in favour of the hypothesis that the [−ATR] pronunciation of [+high] vowels is basic, is provided by oxytone contexts in (30’d) where the final [+high] stressed vowel is realized as [−ATR]. In this context there is no following vowel which could spread the [−ATR] property to the stressed nucleus. Moreover this suggests that harmony is not a spreading

38 

 Leonardo M. Savoia

process right-to-left and that stressed nuclei introduce autonomous aperture and cavity specifications. As a first step, the restriction in (36) accounts for the fact that these systems authorize only a sub-set of vowels in unstressed positions, excluding mid vowels, as pointed out in (31). (36) Vowels in unstressed position A weak (unstressed) vowel in the word domain licenses [−low] if it is [+high] As we saw in (30)–(30’) in final position vowels [−ATR], [ı a ʊ], occur. This would seem to suggest that a process of [−ATR] spreading is at work. In the literature, spreading of [−ATR] in centralization / laxing processes is a traditional solution, applied, for example, in Hualde (1989) in the case of harmony with centralization of all vowels in the word domain in the Montañes of Tudanca (Cantabria). A similar solution is adopted by Cruschina (2007), which assumes that [+high] final vowels become [−ATR] in word final position and [−ATR] spreads on all [+high] vowels in the domain with the exception of [a]. The data in (30)−(30’) show that occurrence of [−ATR] is authorized in the domain including the stressed nucleus and the following vowels, while pretonic nuclei have variable behaviour. In contexts in (30’b) where the stressed nucleus is [−ATR], as in [nıʃ ˈʃıvʊ] ‘went out-1sg’, a [+high] pretonic vowel is generally [−ATR]. In contexts where a [+ATR] tonic nucleus is realized, as in [fʊrˈmikula] ant, a [−ATR] pretonic vowel is favoured. If [−ATR] was spread from right to left, we would not expect [−ATR] in pretonic vowel preceding a [+ATR] stressed vowel. Summing up, the behaviour of pretonic nuclei and that of stressed [−low] vowels in oxytones suggest that the realization of [−ATR] is fixed not by spreading but in virtue of a different mechanism. The different explanation that we propose is that the value of [ATR] is determined by the stressed nucleus, as in the constraint in (37a). The comparison between (33) and (34) points out that in weak positions a general reduction of contrasts shows up; on the contrary in stressed nuclei the difference between [+ATR] and [−ATR] is fully recognizable. This suggests that harmony takes into account the contrastive nature of [ATR] specifications in [−low] vowels, in particular opposing [i]/[u] to [ɪ]/[ʊ]. In conclusion, [−ATR] is selected only in vowels in which it is contrastive, i.e. high and possibly (for some speakers) mid vowels, as in (37a). [−ATR] stressed nuclei precede post-tonic [−ATR] nuclei, as in (37b), while [+ATR] stressed nuclei precede final [a], as in (37c). (37)  a. [ −ATR] contrastive in the stressed nucleus licenses [−ATR] in the following vowel.

Harmonic processes and metaphony in some Italian varieties 



 39

 b.  Milena N

O

  c.

N

>

N

O

N

x

x

x

x

x

k

r ʊ ð i [−ATR] [−ATR] >

N

 

O x f

N

N

>

R N x

O N

O

N

O

N

x

x

x

x

x

x

ʊ r [−ATR]

x

m i k u l a [+ATR] [+ATR] [+ATR]

(37a) obliges the stressed nucleus and the following weak nucleus to agree on contrastive [−ATR], as in (37b). A consequence of (37a) is that non-contrastive [−ATR] of [a] is left out by the harmonic process and cannot be licenced by [−ATR] in the stressed nucleus. Since the requirement is fixed on the contrastive properties of the stressed nucleus, the stressed nucleus realizes as [+ATR]. Naturally (37a) accounts for not only the distribution of high vowels but also that of mid vowels. As we have seen in (30)−(30’), at least for some speakers, mid vowels have a distribution recalling that of high vowels, with low-mid outcomes in contexts preceding [ı ʊ] and high-mid outcomes when [a] follows. Other informants do not apply (37a) on low-mid vowels, realizing for example [ˈrɔta] and not [ˈrota] ‘wheel’. Thus, these varieties at least optionally lack outcomes [e o], and, consequently, (37a) does not allow for them. Finally, in the case of pretonic vocalism, (37a) is only optionally working, given that many speakers systematically introduce [−ATR] or intermediate outcomes. The proposal of Mocciaro (1980), relating lowering of high vowels to metaphony, is based on the fact that in many Central Sicilian varieties the metaphonic outcome is [i u]. Mocciaro assumes that etymological i, u changed to [ı ʊ] in order

40 

 Leonardo M. Savoia

to be distinguishable from the outcomes [i u] introduced by metaphony. As a consequence, the original contrasts of the type *muəru ‘I die’ vs. *muru ‘wall’ would be re-created like /muru/ ‘I die’ vs. /mʊru/ ‘wall’. Indeed this link does not seem to be so crucial, in that in many varieties, including that of Milena, metaphony triggers a diphthongized outcome (Ruffino 1984), as in (30d). However, Mocciaro (1980) captures at least the other face of metaphonic distribution, i.e. the fact that in the context a, where metaphony does not apply, [+ATR] vowels can freely occur, merging with the monophthongized outcomes of metaphony. In Sicilian varieties metaphony affects only stressed low-mid vowels [ɛ ɔ], since high-mid vowels are absent in these systems. The possible phonetic outcomes alternate between a diphthong [iə uə], as in Milena in (30d), (38), and a simple nucleus [i u], as in Villapriolo in (39). Both these outcomes escape harmony, occurring before following post-tonic [ɪ ʊ]. As for unstressed vowels, original mid vowels changed to [+high]; so, in post-tonic position we find only [+high] vowels. This has obscured the original distribution of metaphony, deleting the difference between original post-tonic mid vowels and high ones. As a consequence, in Sicilian systems metaphony is totally or partially morphologized, as shown by the data in (38)−(39). What we see is that the alternation between low-mid and diphthongized outcomes is not due to phonetically distinct properties, given that, except for [a], in post-tonic position only [+high] vowels occur and both metaphonic outcomes and low-mid stressed nuclei alternate in the same contexts with post-tonic [ɪ ʊ]. In varieties like that of Villapriolo both metaphonic outcomes [i u], in (39b), and stressed harmonized nuclei, [ɪ ʊ], in (39a), can occur before post-tonic [ɪ ʊ]. (38) Milena [ˈpiəðɪ] ‘feet’ [kʊˈtiəɖɖʊ] ‘knife’ [ˈniəʃʃʊ] ‘I go out’ / [ˈniəʃʃɪ] ‘you go out’ [ˈpiəkura] / [ˈpiəkʊrɪ] ‘ewe / ewes’ [ˈduərmʊ] ‘I sleep’ / [ˈduərmɪ] ‘you sleep’ [ˈduərmɪnʊ] ‘they sleep’ [ˈtuəkkʊ] ‘I touch’ / [ˈtuəkkɪ] ‘you touch’

[ˈpɛðɪ] ‘foot’ [kʊˈtɛɖɖa] ‘knives’ [ˈnɛʃʃɪ] ‘s/he goes out’ [ˈdɔrmɪ] ‘s/he sleeps’ [ˈtokkwa] ‘s/he touches’

(39) Villapriolo a. harmonized outcomes b. metaphonic outcomes [ˈjɪtʊ] / [ˈjita] ‘finger / fingers’ [ˈpɛðɪ] / [ˈpiðɪ] ‘foot / feet’ [ˈfɪʎ ʎʊ] / [ˈfiʎ ʎa] ‘son / daughter’ [ˈdɛndɪ] / [ˈdindɪ] ‘tooth / teeth’ [fʊrˈmɪkʊlɪ] / [fʊrˈmikula] ‘ants / ant’ [kuˈtiɖɖʊ] / [kuˈtɛɖɖa] ‘knife / knives’ [ˈpɪʎ ʎʊ] / [ˈpiʎ ʎa] ‘I take / s/he takes’ [ˈpirdʊ] / [ˈpirdɪ] / [ˈpɛrdɪ] ‘I / you lose / s/he loses’ [ˈpirdɪnʊ] ‘they lose’

Harmonic processes and metaphony in some Italian varieties 





 41

[aʃ ˈtʊtʊ] / [aʃ ˈtuta] [ˈdurmʊ] / [ˈdurmɪ] / [ˈdɔrmɪ] ‘I put out / s/he put out’ ‘I / you sleep / s/he sleeps’ [ˈmʊrʊ] ‘wall’ [ˈmurʊ] / [ˈmurɪ] / [ˈmɔrɪ] ‘I die / you die / s/he dies’ [ˈfʊrnʊ] / [ˈfurnara] ‘oven / ovens’ [ˈnuvʊ] / [ˈnɔva] ‘new-m.sg / f.sg’ [mʊˈlʊnɪ] / [muˈluna] ‘melon / melons’ [ˈɔmʊ] / [ˈummɪnɪ] ‘man / men’ b’. [ˈpikura] / [ˈpikʊrɪ] ‘sheep-sg / pl’

Hence, metaphony is no longer connected with the phonetic content [+high] in the post-tonic vowel, but is governed by morphological information, as suggested in (40a) where metaphonizing properties are associated with the morphological content. As in the case of Settimo S. Pietro in (12)–(14), inflectional categories, specifically II class masculine (m.sg), 1sg, 2sg and III class 3pl, assign metaphonizing properties to the stressed nucleus, which externalizes them. Let us consider the different outcomes of metaphony, i.e. [iə uə] in (38) and [i u] in (39). As in the other types of metaphony we have investigated, the stressed nucleus subsumes and licenses [+high, +ATR]. The specification [+high, +ATR] is realized as the first part of the diphthong, in (38) for Milena; in the recessive second-position a low-mid element [−high, −ATR] is realized. Otherwise the simple vowel [i u] is realized, independently of the [−ATR] nature of post-tonic vowels, as in (39) for Villapriolo9. This metaphonic outcome implements a complete reversal from [−high, −ATR] to [+high, +ATR], which cannot be obtained through the simple propagation of [+ATR] or [+high]. We assume that metaphony in these varieties requires licensing of [+high] and [+ATR] by the stressed nucleus in morphologically fixed contexts, as in (40). By virtue of (40) relevant inflectional categories introduce [+high, +ATR] as a phonological specification of the inflectional morpheme. The prominent nucleus in the string licenses it by subsuming. In the case of the diphthong, we can expect that it is its head position which realizes the phonological content [+high, +ATR] to be licensed. Metaphonic outcomes in proparoxytones, like (39b’), can be dealt with as lexicalized forms. Differently, 3pl proparoxytones in the 2nd and 3rd verbal class trigger morphologized metaphony. (40) Metaphony Contrastive [+high] and [+ATR] are licensed in the stressed nucleus associated with II class masculine (m.sg / m.pl), 1sg, 2sg and III verbal class 3pl of the present.

9 Metaphonic [i u] deriving from original low-mid vowels ɛ, ɔ, characterize many Southern Italian varieties.

42 

 Leonardo M. Savoia

Unlike the diphthong in (41a), the outcome [i u] in correspondence to underlying low-mid vowels is derived through delinking of the lexical content of the stressed nucleus. More precisely, the stressed nucleus is required to realize the specification introduced by the metaphony, which replaces its incompatible lexical content, as in (41b). (41) a. Milena N

N

N

> O

N

x

R N x

x

x

/ n

ε

∫ + i 2sg /

[−high] [−ATR]

[+high] [+ATR]

x

b. Villapriolo

x

R N x

n

i

O →

/ d

c

x

[−high] [−ATR]

N

x

x

∫

i

N >

N R N x

x

O

[+high]2sg [−high] [+ATR]2sg [−ATR]

N

O

x e

O

>

N

O

N

x

x

x

r

m + i2sg / [+high] [+ATR]

N

> O

N

x

R N x

x

x

x

d

u

r

m

i

O →

[−high] [+high]2sg [−ATR] [+ATR]2sg

Metaphony interacts with harmony introducing outcomes which cannot be interpreted by the harmonizing process. The fact that metaphonic [i u] do not undergo harmony confirms the fact that the stressed nature of the nucleus is involved in both harmony and metaphony. As we saw, in the discussion in section 2, these processes improve and increase the perceptibility of particular properties within a phonological domain. In the case of harmony [−ATR] is extended along the entire span governed by the stressed nucleus, which licenses it. Metaphony associates the



Harmonic processes and metaphony in some Italian varieties 

 43

contrastive [+high, +ATR] specification with the strong nucleus in the domain. We note that metaphony does not work in the contexts where the underlying stressed vowel is [+high], like in (42) for Villapriolo. (42) Villapriolo [ˈpɪʎ ʎʊ] ‘I take’, cf. *[ˈpiʎ ʎʊ] vs. [aʃ  ˈtʊtʊ] ‘I put out’, cf. *[aʃ  ˈtutʊ]

[ˈpirdʊ] ‘I lose’ [ˈdurmʊ] ‘I sleep’

metaphony metaphony

We must conclude that in these varieties metaphony is selected by a sub-set of lexical entries which can be characterized in phonological terms, i.e. as [−high, −low] stressed nuclei. More precisely, metaphony takes into account the phonological content of the stressed nucleus in the lexical representation. This prevents metaphony from affecting underlying nuclei [+high], excluding them as potential inputs. Underlying contexts [+high] license [−ATR] independently of metaphony. The fact that the stressed nucleus can license only once in its domain seems to be parametrical, in the sense that we could expect grammars where this is not true. Nevertheless, generally, metaphony overcomes other possibly competing processes blocking them, as in some one among the propagations examined by Manzini and Savoia in this volume. Similarly, in many Southern Italian varieties laxing of the [+high] stressed nuclei is prevented when [+high] realizes metaphony (Savoia 2015). We can think that the relation between metaphony and inflectional categories it externalizes overcomes phonetic requirements, like harmony in Sicilian varieties. So, the externalizing procedure could represent the more specific process that precedes the application of the harmonic rule. More precisely, the strings that undergo metaphony are a specialized sub-set of those that undergo the harmonic rule in the terms of the ‘elsewhere’ condition proposed in Kiparsky (1973) (see also Harris 1974 for an application). Nevertheless, the child has to learn that metaphony excludes other processes.

5 M  etaphony and harmonic processes in the element framework A model based on elements (Kaye, Lowenstamm & Vergnaud 1985; Harris 1994b; Harris & Lindsey 1990, 1995, 2000; Backley 2011) involves a different ­formalization of processes. In particular, the notion of contrastivity is subsumed by the very nature of the elements as monovalent privative properties. Thus, coming to metaphonic processes examined in section 2, they can be interpreted in terms of alignment inducing [I]/[U] heads in the post-tonic nucleus to co-occur with [I]/[U] in the stressed

44 

 Leonardo M. Savoia

nucleus. The stressed nucleus licenses the whole autosegmental content of the vowel position in its foot, as in the constraint in (43a). More precisely, (43a) assumes a complete alignment whereby only [I]/[U] are admitted in the stressed position and in unstressed ones. This analysis is available for all metaphonies in­troducing the outcomes [i u], independently from the aperture degree of the lexical mid vowel. Thus, the constraint in (43a) is operating not only in varieties like that of Mascioni in (43b) where /e o/ alternate with [i u], but also in metaphonies which raise /ɛ ɔ/, [I/U, A], to [i u], like in many Southern Italian varieties, as Villapriolo in (43d), or introduce a diphthong [iɐ/ə uɐ/ə], as in Iacurso, (43c), and Milena, (43e). The aperture element [A] is delinked, as in (43b, d) or preserved and interpreted as the governed part that diphthongs realize in the strong nucleus, as in (43d). (43) a.  Metaphony  The stressed position licenses [I/U] in the prosodic (foot) domain. b.  Mascioni c.  Iacurso N N N

>

O

N

O

N

x

x

x

v

i I

N

>

O

N

O

N

x

x

x

x

x

x

t + u U

p

i I

A

a

ð + i I

A d.  Villapriolo e.  Milena N

N >

O

N

O

N

x

x

x

x

i I

ð + im.pl I

p

A

N

>

O

N

O

N

x

x

x

x

x

p

i I

e

N

A

ð +

im.pl I

Harmonic processes and metaphony in some Italian varieties 



 45

Varieties which raise low-mid vowels /ɛ ɔ/, [I/U, A], to high-mid [e o], [I/U, A], in the context of post-tonic [i u], realize a partial alignment. So, the head element [I]/[U] in post-tonic position aligns with the head [I]/[U] in tonic position, i.e. with stressed high-mid vowels, as in (44b). These dialects then adopt a weaker version of (43a), requiring simply head alignment, as in (44a). In the Sardinian system this is the only restriction which accounts for metaphony, as in (44b). In the Central dialects, like that of Mascioni in (44c), the constraint in (44a) affects only lexical low-mid vowels, while metaphony of lexical high-mid ones is interpreted by (43a). (44) a. Low-mid metaphony The [I]/[U] head in the governing nucleus in the word domain licenses [I]/[U] head in the post-tonic position. b. Paulilàtino c.  Mascioni N

N >

O

N R N

O

N

O

N R N

x

x

x x

x

x

x

e I A

ts

u U

p

e I A

b



> O

N

x

x

x

r

d

i I

Processes analyzed in (43)–(44) concur to preserve the contrastive nature of the phonological content of the stressed vowel. Thus, in alternation with lexical high-mid vowels [I/U, A], [I] and [U] are selected in the content of the stressed nucleus. Analogously, the outcome [I/U, A] is preserved in many dialects, which thereby exploit its ability to contrast with [I/U] and [I/U, A] vowels. An evident result of the element treatment of metaphony is that all the metaphonic processes which bring about [i u] outcomes can be unified without allowing for the lexical starting point. By contrast, the high-mid outcomes require a more restricted application of the metaphonic constraint. Moreover, element theory points out the similarity between metaphony and other harmonic processes. In fact, (43) and (44) analyze metaphony like many harmonic processes as the result of a total or partial alignment of the phonological content involving the stressed nucleus and the post-tonic ones. Differently from other

46 

 Leonardo M. Savoia

traditional approaches there is neither feature-changing nor spreading of phonological material.

5.1 Harmonies Sicilian harmony examined in section 4 involves the distribution of [−ATR] / [+ATR] specifications in the domain of the stressed nucleus. In the element framework [ATR] can be interpreted in terms of headedness (Backley 2011). The expressions including a head element correspond to [+ATR] vowels, like [I], [U] for [i], [u], and [I, A], [U, A] for high-mid [e] and [o]. [−ATR] vowels lack the head, like [I], [U] for [ɪ] and [ʊ], [I, A], [U, A] for [ɛ ɔ]. [a] is in turn characterized by a headed expression [A], whereas the non-headed [A] corresponds to a centralized lax vowel, like [ɐ ə]. We can analyze the harmonic process of Sicilian varieties as a constraint on the occurrence of the headedness in vowels, both in the stressed nucleus and in intermediate vowels in proparoxytones, as in [ˈnivika] ‘snow-3sg’ in (30’c). As seen in section 4, in these varieties in unstressed position, including the posttonic position, high vowels or [a] occur; this constraint is interpreted by (45a). Nevertheless, while [i u], i.e. [I]/[U], alternate with [ɪ ʊ], i.e [I]/[U], [a] has no nonheaded alternant like [A]. In other words the only low vowel in the vocalic system is the headed [A] expression, as in (45b). (45) a. Unstressed positions authorize vocalic expressions including only one element. b. [a] is headed, [A]. In harmonizing varieties of Milena and Sutera the behaviour of [a] directly follows if we assume that the headed / or non-headed nature of the unstressed (post-tonic) vowels must align to headedness of the stressed nucleus, as in (46). Thus, a post-tonic [a] can be licensed only by a headed stressed vowel, namely [I]/[U], [A], or the metaphonic diphthong [iə uə], where the prominent vowel of the diphthong is a headed expression. This imposes alignment on the head role of cavity elements. For some informants, mid vowels are realized as high-mid in the context of a following [a] (in 30a) as well, as we saw. In the other contexts only non-headed vowels [I]/[U] occur, as suggested in the representations in (47a, b). (46) Alignment in the internal structure of vowels (harmony) An unstressed headed vowel, [I]/[U], [A], is licensed by a stressed headed vowel;

Harmonic processes and metaphony in some Italian varieties 



(47) Milena a.

O




N

N

>

N

O

N

x

x

x

x

x

O

k

r

ʊ U

ð

i I      

x f

 47

N

N

>

R N x

O N

O

N

O

N

ʊ U

x x

x

x

x

x

x

r m

i I

k

u U

l

a A

The gist of this analysis is that alignment of [a] with [I/U] headed nuclei realizes the fundamental vocalic cavity properties, [i a u], namely the segments endowed with the unmarked intrinsic content in Chomsky & Halle (1968); Kaye, Lowen­ stamm & Vergnaud (1985, 1990). In particular, [a] is able to subsume the resonance properties within the domain of the stressed nucleus, including the foot and first metrical projection on its right. In other words, [a] can control the harmonizing effect in the domain of the stressed nucleus. The positions preceding the stressed nucleus are basically insensitive to harmony, and, as Milena’s data show, we normally find [ı ʊ], even if it is not excluded that [ATR] harmony can extend to the string preceding a harmonizing stressed nucleus, as (36’b) shows. In other words, the domain of the stressed nucleus includes the entire word in the grammar of some speakers. As a consequence, the headed outcomes occur on its left as well. This is true not only in sequences in which [a] is in post-tonic position, as in (36a), but also systematically in contexts where [a] is the stressed nucleus preceded in its turn by a high vowel, as in [liˈvatı] ‘you take off’ in (36’b) and (37’b). In these grammars, (46) also includes the sequence on the left of stressed [a]. In metaphonic contexts [I]/[U] occur in the stressed nucleus independently of the headed / non-headed nature of the post-tonic vowels in the string, as we saw in (43). In particular, [I]/[U] contrast with mid vowels, including [A], and, crucially, with lax [I]/[U] stressed nuclei. In other words metaphonic stressed nuclei manifest a specialized content that separates them from other possible stressed nuclei, including harmonizing ones. As suggested in section 3, we observe that metaphony affects a sub-set of the strings that otherwise could undergo harmony. As a consequence, metaphony excludes [−ATR] harmony. Like the preceding metaphonies and harmonies, Central Sardinian harmony discussed in section 3 for the varieties of Àllai and Làconi in (17)–(18) can also

48 

 Leonardo M. Savoia

be dealt with as a type of head alignment. These varieties exclude a [I/U, A] expression from the post-tonic position, but, differently from the Campidanian varieties, they lack the neutralization of the post-tonic original mid vowels to [I/U], as in (47a). A low-mid vowel in final position is admitted in the context of a low-mid stressed vowel, as in (47b). Finally, it is a [I]/[U] stressed vowel which selects a simple headed expression [I/U] in the post-tonic position, as in (48). (47) a. A post-tonic position legitimates simple expressions [I]/[U]/[A] or nonheaded complex expressions. b. [A] in a complex expression in post-tonic position is legitimated by [A] in a complex expression in stressed position. (48) The [I/U] head in the stressed nucleus legitimates the [I/U] head in a ­post-tonic position. Again, the identical behavior of [i u] and [a] is the surface result of the distributio­ nal constraints that limit the occurrence of the possible final vowels. More precisely, (47a) authorizes in post-tonic contexts only [i a u] or [ɛ ɔ]; (47b) authorizes post-tonic [ɛ ɔ] only if the stressed vowel is [ɛ] / [ɔ] in its turn. Finally, (48) fixes the specific distribution of [I/U], whereby stressed [i u] and the metaphonic outcomes [e o] legitimate final / post-tonic [I]/[U]. [a] simply escapes these constraints. (47) and (48) satisfy a complexity principle generally involved in prosodic systems requiring that the segments in a weak / governed position are not more complex than their governing segments (Harris 1990a, 1997, 2005). Actually, this requirement seems to implement a substantial property whereby licensing capacity manifests itself as the intrinsic phonetic properties of the prosodic head. In virtue of (47a, b), a vowel in a governed position cannot be more complex than the stressed vowel that licenses it, as in (49a–d). The distribution determined by (47)–(48) undergoes metaphony, in (43a). (49) Làconi a. N

b.  >

O

N R N

O

N

x

x

x x

x

c

v

U A

c

m

N

U A

>

O

N R N

O

N

x

x

x x

x

s

a A

n

u U

Harmonic processes and metaphony in some Italian varieties 





c.

d. 

N N R O N

N

> O

N

O

N R N

x

x

x x

x

x

x

b

e I A

ts

u U

b

ε I A



 49

>

x

O

N

x

x

ts

a A

On the basis of this analysis, the fact that [i u] and [a] appear to have identical behaviour (cf. section 3) in selecting post-tonic [i u], is the surface result of the distributional constraints in (47), limiting the occurrence of the possible final vowels.

6 Concluding remarks The analysis of metaphony is a real testing ground for phonological theory. For Italo-Romance kinds of metaphony, the crucial point is the difference between raising metaphony of high-mid vowels and [ATR] metaphony of low-mid vowels. Current studies try to unify them both from a descriptive (Calabrese 1998) and historical point of view (Barbato 2008; Loporcaro this volume). Contrary to these approaches, this article, as a first step, proposes that the two types of metaphony can be adequately interpreted, allowing for their specific phonetic implementation separating [+high] metaphony from [+ATR] metaphony and involving contrastivity (in the sense of Nevins 2010). This solution can work without having recourse to expensive formal devices like markedness requirements and re-adjustment rules. In the analysis I present here, the prosodic and autosegmental organization of the string concur to point out the role of the stressed nucleus in metaphony. A second issue discussed in this article is the relation between metaphony and harmonic processes. Metaphony, height harmony in Central Sardinian varieties and [ATR] harmony in Central Sicilian can be seen as a licensing procedure where the stressed nucleus authorizes height degree of vowels in the string. It is no accident that these phenomena coexist in the same systems and interact in influencing the overall distribution of the vocalic aperture properties inside the domain of the stressed nucleus. The comparison between the feature-based analysis and the treatment in the element theory highlights interesting differences in the explanations they give rise to. A feature-based model allows a characterization of the metaphonic and

50 

 Leonardo M. Savoia

harmonic processes in a descriptively adequate way. An explanatory difficulty shows up when the link between metaphony and harmony must be expressed. In particular, a traditional feature treatment separates the two types of metaphony, whereby raising metaphony is based on [+high] spreading while [ATR] metaphony depends on [+ATR] spreading. Moreover, this approach is not able to connect harmony and metaphony which work within the same language. Thus, harmony in Central Sicilian varieties in section 4 is explained in terms of [−ATR] spreading, while metaphony spreads [+high]. Analogously, in Central Sardinian varieties in section 2, harmony is explained in terms of [+high] while metaphony involves [+ATR]. This result seems to derive from the splitting of the height properties in different features. On the contrary, the element analysis concurs to throw light on the substantial affinity between harmony and metaphony explaining them both as alignment on the properties [I/U] of the stressed nucleus. The interaction between metaphony and harmony is especially evident in the fact that height harmony in Central Sardinian varieties and [ATR] harmony in Central Sicilian ones are blocked (bled) by metaphony. This bleeding relation seems to depend on the fact that metaphony and harmony answer complementary requirements implied in the externalization procedure that connects morpho-syntactic information to Sensory-Motor (SM) interpretation, in the sense of Berwick & Chomsky (2011). In fact, metaphony introduces a specialized phonological characterization in the stressed nucleus in the context of the inflectional exponents; harmony concurs in its turn to make certain vocalic qualities (aperture degree) associated to the phonological organization of the word more recognizable. In other words, these phonological mechanisms increase perceptibility of phonological properties which have a crucial interpretive role within a prosodic domain (Kaun 2004; Walker 2005, 2001, 2011). Thus, both the harmonic and metaphonic outcomes are required to show up in the strings in correspondence of morpho-phonological properties which must be legible to SM.

References Archangeli, Diana and Douglas Pulleyblank. 1989. Yoruba vowel harmony. Linguistic Inquiry 20 (2): 173–217. Archangeli, Diana and Douglas Pulleyblank. 1994. Grounded Phonology. Cambridge: MIT Press. Backley, Phillip. 2011. An Introduction to Element Theory. Edinburgh: Edinburgh University Press. Barbato, Marcello. 2008. Metafonia napoletana e metafonia sabina. In Alessandro De Angelis (ed.), I dialetti italiani meridionali tra arcaismo e interferenza, 275–289. Palermo: CSFLS.



Harmonic processes and metaphony in some Italian varieties 

 51

Berwick, Robert and Noam Chomsky. 2011. The biolinguistic program: the current state of its evolution and development. In Di Anna Maria Sciullo and Cedric Boeckx (eds.), The Biolinguistic Enterprise, 19–41. Oxford: Oxford University Press. Blevins, James P. 2006. Word-based morphology. Journal of Linguistics 42: 531–573. Bromberger, Silvain and Morris, Halle. 1989. Why phonology is different. Linguistic Inquiry 20 (1): 51–70. Calabrese, Andrea. 1995. A constraint based theory of phonologicl markedness and ­simplification procedures. Linguistic Theory 26 (3): 373–463. Calabrese, Andrea. 1998. Metaphony revisited. Rivista di linguistica, 10: 7–68. Calabrese, Andrea. 2009. Metaphony, ms. Online University of Connecticut. Castellani, Arrigo. 1973. I più antichi testi italiani. Edizione e commento. Bologna: Pàtron. Charette, Monik. 1990. License to govern. Phonology 7: 233–253. Charette, Monik. 1991. Conditions on Phonological Government. Cambridge: Cambridge University Press. Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Roger Martin, David Michaels and Juan Uriagereka (eds.), Step by Step, 89–155. Cambridge: The MIT Press. Chomsky, Noam and Halle, Morris. 1968. The Sound Pattern of English. New York: Harper and Row. Contini, Michel. 1987. Étude de géographie phonétique et de phonétique instrumentale du sarde, 2 vv: Texte, Atlas et album phonétique, Alessandria, Edizioni Dell’Orso. Cruschina, Silvio. 2007. Il vocalismo della Sicilia centrale: il tratto [ATR], metafonesi e armonia vocalica. Rivista Italiana di Dialettologia 30: 75–101. Frigeni, Chiara. 2003. Metaphony in Campidanian Sardinian: a domain-based analysis. Toronto Working Papers in Linguistics 20: 63–91. Frigeni, Chiara. 2004. Deriving metaphony in Southern Sardinian with all the ingredients it takes, ms., Colloquium McGill Linguistics. Halle, Morris, Vaux, Bert and Wolfe, Andrew. 2000. On feature spreading and the ­representation of place of articulation. Linguistic Inquiry 31 (3): 387–444. Halle, Morris and Alec Marantz. 1994. Some key features of distributed morphology. In Andrew Carnie and Heidi Harley (eds.),  MITWPL 21: 275–288. Harris, James W. 1974. Evidence from Portuguese for the “Elsewhere Condition” in phonology. Linguistic Inquiry 5 (1): 61–80. Harris, James. 1991. The exponence of gender in Spanish. Linguistic Inquiry 22: 27–62. Harris, John. 1990a. Reduction harmony, GLOW Phonology Workshop, University of London. Harris, John. 1990b. Segmental complexity and phonological government. Phonology 7: 255–300. Harris, John. 1994a. English Sound Structure. Oxford: Blackwell. Harris, John. 1994b. Monovalency and opacity: Chicheŵa height harmony. UCL Working Papers in Linguistics 6: 509–547. Harris, John. 1997. Licensing inheritance: an integrated theory of neutralisation. Phonology 14 (3): 315–370. Harris, John. 2005. Vowel reduction as information loss. In Philip Carr, Jacques Durand and Colin J. Ewen (eds.), Headhood, elements, specification and contrastivity, 119–132. ­Amsterdam: Benjamins. Harris, John and Geoff Lindsey. 1990. Phonetic interpretation in generative grammar. UCL Working Papers in Linguistics 2: 355–369. Harris, John and Geoff Lindsey. 1995. The elements of phonological representation. In Jacques Durand and Francis Katamba (eds.), Frontiers of phonology, 34–79. London: Longman.

52 

 Leonardo M. Savoia

Harris, John and Geoff Lindsey. 2000. Vowel patterns in mind and sound. In Noel BurtonRoberts, Philip Carr and Gerard Docherty (eds.), Phonological Knowledge. Conceptual and Empirical Issues, 185–205. Oxford: Oxford University Press. Hualde, José Ignacio. 1989. Autosegmental and metrical spreading in the vowel-harmony systems of northwestern Spain. Linguistics 27: 773–805. Kaun, Abigail. 1995. The Typology of Rounding Harmony. An Optimality Theoretic Approach. PhD dissertation, UCLA. Kaun, Abigail. 2004. The typology of rounding harmony. In Bruce Hayes, Robert Kirchner and Donca Steriade (eds.), Phonetically based phonology, 87–116. Cambridge: Cambridge University Press. Kaye, Jonathan. 1986/87. Government in phonology. The case of Moroccan Arabic. The Linguistic Review 6: 131–135. Kaye, Jonathan. 1990. ‘Coda’ licensing. Phonology 7: 301–330. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. 1985. The internal structure of phonological elements: A theory of charm and government. Phonology Yearbook 2: 305–328. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7: 293–231. Kaze, Jeffery W. 1991. Metaphony and two models for the description of vowel systems. Phonology, 8 (1): 163–170. Kiparsky, Paul. 1973. ‘Elsewhere’ in phonology. In S. R. Anderson and P. Kiparsky (eds.), Festschrift for Morris Halle, 93–106. New York: Holt, Rinehart and Winston. Kula, Nancy Chongo and Lutz Marten. 2000. Constraints and processes: Evidence from Bemba, Herero and Swahili. SOAS Working Papers in Linguistics and Phonetics 10: 91–102. Loporcaro, Michele. 2002–03. Di una presunta reintroduzione preromanza di -US di accusativo plurale in Sardegna. Studi e Saggi Linguistici 40–41: 187–205. Loporcaro, Michele. 2003. Coarticolazione e regolarità del mutamento: l’innalzamento delle vocali medie finali in sardo campidanese. In G. Marotta and N. Nocchi (eds.), La coarticolazione. Atti delle XIII giornate di studio del Gruppo di Fonetica Sperimentale (AIA), Università di Pisa, 28–30 settembre 2002. Pisa: ETS, 23–44. Loporcaro, Michele. 2005. Typological remarks on Sardinian: 1. Vowel harmony 2. Sardinian in a correlative typology of the Romance languages. Sprachtypologie und Universalienforschung, 58 (2–3): 210–227. Loporcaro, Michele. (this volume). Metaphony and diphthongization in southern Italy: reconstructive implications for sound change in early Romance. Maiden, Martin. 1985. ‘Displaced’ metaphony and the morphologisation of metaphony. Romance Philology XXXIX, 1: 22–34. Maiden, Martin. 1987. New perspectives on the genesis of Italian metaphony. Transactions of Philological Society 85 (1): 38–73. Maiden, Martin. 1991. Interactive Morphonology: Metaphony in Italy. New York: Routledge. Manzini, M. Rita and Leonardo M. Savoia. 2005. I dialetti italiani e romanci. Morfosintassi generativa, 3 vols. Alessandria: Edizioni dell’Orso. Manzini, M. Rita and Leonardo M. Savoia. 2007. A unification of morphology and syntax. London: Routledge. Manzini, Maria Rita and Leonardo M. Savoia. (this volume). Which syntactic information gets interpreted by phonosyntactic processes? Evidence from /u/ propagation in Southern Italian dialects.



Harmonic processes and metaphony in some Italian varieties 

 53

Mascarò, Joan. 2010. An analysis of stress-dependent harmony in Servigliano, communication presented at X Incontro di dialettologia italiana -10th Italian Dialectology Meeting, University of Bristol, 22–24 September 2010. Mocciaro, Antonia. 1980. Per una interpretazione fonologica del vocalismo delle parlate della Sicilia centrale, in Bollettino del Centro di Studi filologici e linguistici siciliani XIV, 3–14. Nevins, Andrew. 2005. Microvariations in harmony and value-relativized parametrization. LIVY Yearbook 5: 187–232. Nevins, Andrew. 2010. Locality in Vowel Harmony. Cambridge: The MIT Press. Piccitto, Giorgio. 1959. Il siciliano dialetto italiano, in Orbis VIII, 183–199. Rohlfs, Gerhard. 1966 [1949]. Grammatica storica della lingua italiana e dei suoi dialetti. Fonetica [Historische Grammatik der Italienischen Sprache und ihrer Mundarten. I – Lautlehre], Torino, Einaudi. Ruffino, Giovanni. 1984. Isoglosse siciliane. In Adriana Moreschini Quattordio (ed.), Tre millenni di storia linguistica della Sicilia, 161–224. Pisa: Giardini. Russo, Michela. 2007. La metafonia napoletana: evoluzione e funzionamento sincronico. Bern: Peter Lang Sánchez Miret, Fernando. 1998. La diptongación en las lenguas románicas. München – Newcastle: LINCOM Europa. Savoia, Leonardo M. 1987. Teoria generativa, modelli fonologici e dialettologia. La ­propagazione di u in una varietà lucana. Rivista Italiana di Dialettologia, XI: 185–263. Savoia, Leonardo M. 2005. Armonizzazioni vocaliche in alcune varietà romanze. In Biffi, Marco, Calabrese, Omar and L. Salibra (eds.), Italia linguistica: discorsi di scritto e di parlato. Nuovi studi di linguistica italiana in onore di Giovanni Nencioni. Siena: Protagon, 217–233. Savoia, Leonardo. 2015. I dialetti italiani. Sistemi e processi fonologici nelle varietà italiane e romance. Pisa: Pacini. Savoia, Leonardo M. and Martin Maiden. 1997. Metaphony. In Martin Maiden and Mair Parry (eds.), The dialects of Italy, 15–25. London: Routledge. Scullen, Mary Ellen. 1992. Chichewa vowel harmony and underspecification theory. Linguistic Analysis 3/4: 218–245. Vaux, Bert. 1995. The status of ATR in feature geometry. Linguistic Inquiry 26: 175–182. Wagner, Max Leopold. 1941. Historische Lautlehre des Sardischen, Max Niemeyer, Halle [Beihefte 93 zur ZRPh] Walker, Rachel. 2001. Positional markedness in vowel harmony. In Caroline Féry, Antony D. Green and Ruben van de Vijver (eds.), Linguistics in Potsdam, 12, Proceedings of HILP, v. 5, 212–232. University of Potsdam. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural language and linguistic theory 23: 917–989. Walker, Rachel. 2011. Vowel patterns in language. Cambridge: Cambridge University Press. Zoll, Cheryl. 1998. Positional asymmetries and Licensing, ms. Zoll, Cheryl. 2004. Positional Asymmetries and Licensing. In John J. McCarthy (ed.), Optimality Theory in Phonology: A Reader, 375–378. Oxford: Blackwell.

Michele Loporcaro

Metaphony and diphthongization in Southern Italy: reconstructive implications for sound change in early Romance*

University of Zurich

Abstract: Both diphthongization and metaphony per se, and the interaction between the two, are among the most investigated topics in Romance historical linguistics. In this paper, I suggest that the two do not originally belong together, in the sense that a) the so-called Romance diphthongization, which affected PRom /ɛ ɔ/ in most Romance languages, was not originally sensitive to the quality of the final unstressed vowel; and b) the different types of metaphony observed across (­ Italo-­ )Romance, often involving diphthongization, are most economically explained as changes which all first arose as assimilatory raising, not as diphthongization. These data provide an interesting case study in what one might call the division of labor of phonetics and (morpho)phonology for the explanation of sound change. In this connection, experimental phonetic evidence from Southern Italian dialects will prove crucial.

1 Introduction Metaphony occurs widely throughout the Romance language family, from Portuguese to Romanian, and most Romance languages have undergone several kinds of diphthongization as well. Especially for the most widespread of those diphthongization processes, the one affecting ­Proto-­Romance /E, O/ (e.g. Italian viene ‘come-3sg’, nuovo ‘new-m.sg’, French (il) vient, noeuf ‘new-m.sg’ < OFr nuef, Spanish viene, nuevo)—sometimes called “the Romance diphthongization” par excellence—a causal relation with metaphony has been repeatedly suggested (most influentially by Schürr 1936, 1970). In most Romance varieties in which

* Previous versions of this paper were presented at the Universities of Munich (15th LIPPSymposium, “Language change”, July 2011) and Venice (December 2011). The following abbreviations are used throughout: AMR = allomorphic morphological rule, CL = Classical Latin, MPR = morphophonological rule, OF = Old French, OHG = Old High German, PR = ­phonological rule, PRom  = Proto-Romance. Thanks to Marcello Barbato, Domenico Meo, Fernando Sánchez Miret and an anonymous reviewer for comments on a previous draft. Usual disclaimers apply.

56 

 Michele Loporcaro

diphthongization occurs, such a relation is not (anymore) transparent, as exemplified in (1) (from Loporcaro 2011b: 120): (1) / ɛ ɔ/-diphthongization a. Florentine b. Old Aretino c. Neapolitan d. Castilian sensitive to i. s yllable structure

+

+

–

–

ii.  quality of the final vowel

–

+

+

–

As seen here, in its ­present-­day distribution, diphthongization may be sensitive to metaphony, as in Neapolitan and many other ­Italo-­Romance dialects of the ­Center-­South (1c), or to syllable structure (cf. the ­open-­syllable diphthongization of ­Florentine-­based standard Italian, (1a), or French). There are (or have been) also varieties in which diphthongization is sensitive to both conditions (Old Aretino, (1b), cf. Castellani 2000: 368) as well as languages in which it is sensitive to neither (Castilian, (1d)). In addition, there are also Romance languages and dialects (­Galician-­Portuguese, Sardinian, and several varieties of Central Italy) in which metaphony brings about raising, rather than diphthongization (e.g. ­Logudorese Sardinian boːnu ‘good-m.sg’ vs. bɔːna ‘good-f.sg’). Given this picture, and the fact that changes which led to the four types in (­1a–­d) took place before the earliest extant documents of the Romance languages, it is clear that any diachronic account of this c­ ross-­linguistic differentiation must rely on reconstruction. According to those who claim that diphthongization was originally tied to the quality of the final vowel (most recently, ­Sánchez-­Miret 1998, 2007, Maiden, to appear), the distribution in (1c) (and, possibly, that in (1b)) must be original, whereas (1a) and (1d) are regarded as the products of later change, obscuring the allegedly original metaphonic ­conditioning. In this paper, I shall provide arguments for the disjunction of metaphony and /ɛ ɔ/-diphthongization, which are best regarded in my opinion as two distinct processes. To prove this point, it is essential to demonstrate that the Neapolitan type (1c) is indeed a later development and that, conversely, the kind of metaphony today observed in Logudorese Sardinian has to be regarded as the most conservative one. In this demonstration, reconstructive arguments based on experimental phonetic evidence, drawn partly from two Southern Italian dialects (Southern Salentino and Agnonese), will play a crucial role. The paper is organized as follows. In section 2, I provide an overview of the different kinds of diphthongization and metaphony observed in (­Italo-­)Romance. Section 3 reports the contrasting hypotheses on the rise of metaphony and Romance

Metaphony and diphthongization in Southern Italy 



 57

diphthongization, while section 4 addresses the issue of phonetic explanation for sound change, focusing on the recruitment of reconstructive arguments from experimental phonetics. Section 5 moves on to consider the rise of metaphonic diphthongization in Old Neapolitan, defending the hypothesis that metaphony first arose as an assimilatory process, but disjoined from diphthongization, which only later modified the output of the metaphony rule. Section 6 reviews experimental phonetic studies on other modern Southern Italian dialects which provide evidence in favor of the ­non-­primary character of metaphonic diphthongization. Section 7 finally shows that the reconstructive scenario outlined here can be successfully extended to the whole of Romance, and, in particular, accounts straightforwardly for the intricate development of metaphony and diphthongization in Romanian.

2 D  iphthongization and metaphony in Romance: the basic facts Classical Latin (2a) had five contrasting stressed vowel qualities plus a quantity contrast, which was given up with different outcomes, exemplified in (­2b–­d) for stressed vowels: (2) The vowel system from Latin to Romance a.  CL

iː

b.

i

   Spa c.   Rom d.    Sar

i

eː

e

e

hilo

ɛ

peras crezco hierro

i

e

fir

pere i

cresc ɛ

a

aː

o

oː

a

ɔ

vaca

ocho

ɛ

a

fier

vacă a

u

uː

o ɔ

sol

u

gola muro u

opt soare gură ɔ

mur

u

filu piras cresco ferru bacca otto sole bula muru ‘thread’ ‘pears’ ‘grow-1sg’ ‘iron’ ‘cow’ ‘eight’ ‘sun’ ‘throat’ ‘wall’

In the majority of them, as illustrated with Spanish examples in (2b), there was a merger of ĭ ē and of ŭ ō, whereas the outcomes of Latin short ĕ and ŏ remained distinct (following common practice in Romance historical linguistics, Latin phonemes are given in small caps). This system, usually labelled the “Common Romance” vowel system, is shared by all of ­Ibero-­and ­Gallo-­Romance as well as by ­central-­northern ­Italo-­Romance. A different development took place in Romanian and Sardinian. As seen in (2d) (examples are from Logudorese), the Sardinian vowel system shows a plain merger of (Classical Latin) long and short

58 

 Michele Loporcaro

­counterparts into one phoneme, yielding a f­ ive-­vowel system. Romanian, on the other hand, has the asymmetrical vowel system (2c), which is like the common Romance system for front vowels (where the outcomes of Latin ĭ and ē have merged) but like Sardinian for back vowels since the outcomes of ŭ and ō have not merged.1 The label “Romance diphthongization” is commonly used to designate the diphthongization of the outcomes of Latin short ĕ and ŏ. Those vowels turned into rising diphthongs in many Romance varieties, under different conditions, as schematically exemplified in (3) by comparing the major Romance languages to the exclusion of Romanian, which also displays diphthongs, as shown in (2c), but, given its distinct vowel system, poses special problems, to be addressed in section 7. (3)

nŏ.vum

vĕ.nit

cŏl.lum

pŏr.tam

Spa

nuevo

viene

cuello

puerta

Ita

nuovo

viene

collo

porta

Fre

neuf < OF nuef

vient

cou

porte

Por

novo

vem

colo

porta

In Spanish, PRom /ɛ ɔ/ have diphthongized, irrespective of syllable structure, whereas in Italian and French diphthongization occurred only when the syllable was originally (and still is, in Italian) open, a distribution ­preserved to this day in standard Italian. In Portuguese, however, no diphthongization took place. One reason to call this the “Romance diphthongization” is that it is shared by many Romance branches: Spanish and Italian, for instance, diphthongized no other PRom vowels. French underwent radical reshaping in the vowel system, including other instances of diphthongization (affecting the outcomes of PRom /e o/ too), whereas diphthongs arisen from ­low-­mid /ɛ ɔ/ are a shared trait between French, Spanish and Italian. In principle, features shared among sister languages are likely to be older than ­language-­specific innovations, and indeed there is evidence from both absolute and relative chronology that the Italian (i.e. Tuscan) diphthongs arose by the mid 7th century (Castellani 1961: 95), whereas /ɛ ɔ/ > /jɛ wɔ/

1 This asymmetry has been explained as a consequence of a later date of the ŭ ō merger with respect to that of ĭ ē, as documented e.g. in Herman’s (1971: 139–­143), (1985: 75–­76) studies of deviant spellings in epigraphic Latin from different parts of the Western Empire. Some harbingers of the common Romance development are to be found in Romanian too (e.g. scoate < excŭtere ‘to shake’, roş < rŭsseum ‘red’, with ŭ > /o/ lowering), as observed e.g. by Sánchez Miret (2001: 377). Yet, the diachronic correspondences in (2c) account for the overwhelming majority of the Romanian lexicon.

Metaphony and diphthongization in Southern Italy 



 59

diphthongization appears to be somewhat earlier in French (6th century, according to Bourciez 1937: 94).2 The examples in (3) do not exhaust the possible conditions/environments for diphthongization of PRom l­ow-­mid vowels. As shown in (4), several western Romance languages display diphthongization conditioned by a following palatal consonant (cf. Sánchez Miret 1998: 213; Lausberg 1976: 232). (4)

French

Provençal

Catalan

Sursilvan

medium

mi

mieg

mig

miez

vet(u)lum

vieil

vielh

vell

‘half-m.sg’ ‘old-m.sg’

We will not be dwelling on diphthongization before palatal consonants here (for which Barbato 2013 convincingly proposes a joint analysis with ­vowel-­induced metaphony), but focus on what is usually called metaphonic diphthongization, as exemplified with Neapolitan data in (­5a–­b) (I shall comment on (­5c–­d) directly), and on its relationship with the other types of diphthongization exemplified above in (3). (5) Neapolitan type (Naples; cf. Vignuzzi & Avolio 1994: 644–­645) Before -a -e -o

Before -ī -ŭ

open syllable

checked syllable open syllable

a. ŏ

[ˈbːɔːnə] ‘good-f’

[ˈmɔrtə] ‘dead-f’ [ˈbːwoːnə] ‘good-m’ [ˈmwortə] ‘dead-m’

b. ĕ

[ˈpɛːrə] ‘foot’

[ˈvɛrmə] ‘worm’ [ˈpjeːrə] ‘feet’

[ˈvjermə] ‘worms’

[ˈsordə] ‘deaf-f’ [nəˈpuːtə] ‘nephews’

[ˈsurdə] ‘deaf-m’

c. ō / ŭ [nəˈpoːtə] ‘nephew’

d. ē / ĭ [ˈmeːsə] ‘month’ [ˈsekːə] ‘dry-f’

[ˈmiːsə] ‘months’

checked syllable

[ˈsikːə] ‘dry-m’

In this variety, PRom ­low-­mid vowels diphthongized, irrespective of syllable structure, when they were originally followed by a final high vowel which was subsequently neutralized to schwa. Conversely, when the final vowel was ­non-­high, no diphthongization is observed. As for the trigger of metaphony, this may differ across Romance: while in Neapolitan and Sardinian (cf. (7) below) both high vowels bring about metaphony (and

2 Much earlier dates (­3th–­­4th century AD) are assumed by Straka (1953: 268), Richter (1934: 138, ­155–­156), Wright (1982: 59–­60), Zink (19996: 53), etc. However, they are ­ill-­founded, resting on faulty readings of passages by Latin grammarians and Latin epigraphic evidence (cf. Herman 1970: 30; Loporcaro 2011b: 692 n. 20).

60 

 Michele Loporcaro

the same is the case for Portuguese—cf. n. 6 below—and Romansh), this is triggered only by (the outcome of Latin) final -ī in many Northern ­Italo-­Romance dialects and Abruzzian (in the upper part of Southern Italy). The implementation of metaphony may differ across Romance not only with respect to its trigger—an aspect we will not be focusing on here—but also to its target and the structural change involved.3 Note that the two terms diphthongization and metaphony are specular: diphthongization labels the structural change, whose different subtypes, as seen in (3)–(5), can then be distinguished referring to the context in which the change applies. Specularly, the term metaphony refers in itself to the environment of the change, while different subtypes of the process are classified according to the input of the rule and the structural change this input undergoes.4 Thus, in Neapolitan, one finds diphthongization of ­low-­mid vowels and, at the same time, raising of ­high-­mid vowels. In fact, as shown in (­5c–­d), before etymological high vowels (which later merged), the original h ­ igh-­mid vowels /e o/ were raised, a change shared by all ­Central-­Southern Italian dialects displaying metaphony.5 As for ­low-­mid vowels, two subtypes must be distinguished, since alongside the Neapolitan type, with diphthongization of ­low-­mid vowels, the so-called Sabino type occurs, documented widely across central Italy (as well as less extensively in Southern Italy), in which PRom /ɛ ɔ/ were raised instead. This is shown in (6) with examples drawn from one dialect of the Marche, that of Servigliano (in the province of Ascoli Piceno). (6) Sabino type (Servigliano; cf. Camilli 1929: 224–­231) Before -a -e -o open syllable

Before -ī -ŭ

checked syllable open syllable

checked syllable

a.

ŏ

[ˈmɔːre] ‘die-3sg’ [ˈmɔrta] ‘dead-f’ [ˈmoːri] ‘die-2sg’ [ˈmortu] ‘dead-m’

b.

ĕ

[ˈpɛːde] ‘feet’

c.

ō / ŭ [ˈloːpa] ‘­she-­wolf’ [ˈkorsa] ‘run-f’

d.

ē / ĭ [ˈveːde] ‘see-3sg’ [ˈmetːe] ‘put-3sg’ [ˈviːdi] ‘see-2sg’

[aˈpɛrta] ‘open-f’ [ˈpeːdi] ‘foot’ [ˈluːpu] ‘wolf’

[aˈpertu] ‘open-m’ [ˈkursu] ‘run-m’ [ˈmitːi] ‘put-2sg’

3 As to the input, recall that the scope of the present discussion is limited to diphthongization of ­Proto-­Romance /E, O/, from Lat. ĕ ŏ. 4 This is somewhat simplified, since also metaphony can be triggered by different (sets of) final vowels, as said above. Anyway, the term metaphony itself abstracts away from those differences, on the one hand, and on the other hand, contrary to the term diphthongization, it does not say anything as to the nature of the structural change that is going on. 5 An isolated exception is Old Romanesco, which had metaphonic diphthongization of PRom /ɛ ɔ/ but not metaphonic raising of PRom /e o/ (cf. Ernst 1970: ­53–­58 and, more recently, Formentin 2012, 2013, who first discovered instances of an incipient metaphonic raising of PRom /e o/ in 14th century Romanesco, restricted to nouns denoting humans, which never evolved into a regular process and eventually disappeared).

Metaphony and diphthongization in Southern Italy 



 61

In Logudorese Sardinian too, the implementation of metaphony is by raising. (7) Logudorese Sardinian (cf. Loporcaro 2003) Before -a -e -o open syllable

Before -i -u

checked syllable open syllable

checked syllable

[ˈmɔrta] ‘dead-f.sg’

[ˈmortu] ‘dead-m.sg’

a.

ŏ ō [ˈbɔːna] ‘good-f.sg’

b.

ĕ ē [ˈbɛːnɛ] ‘well’ [ˈbɛlːa] ‘­beautiful-f’

[ˈboːnu] ‘good-m.sg’

[ˈbeːni] [ˈbelːu] ‘beautiful-m.sg’ ‘come-imp.2sg’

Metaphony is fed by the outcomes of all Latin mid vowels due to the merger of Lat. ŏ ō and ĕ ē in the Sardinian vowel system (2d). Note that syllable structure is irrelevant for the operation of metaphony in all the data sets in (5)–(7). This is also true for Portuguese, which, like Logudorese Sardinian, does not show diphthongization but rather displays raising metaphony, though unlike in Logudorese, this is fully morphologized. (8) Portuguese (cf. Williams 1962: 129–­130; Mateus & Andrade 2000: 17) Before -u

Elsewhere

a.

ŏ n[o]vo ‘new-m.sg’ n[ɔ]va, -as, -os ‘new-f.sg / f.pl / m.pl’ gr[o]sso ‘big-m.sg’ gr[ɔ]ssa, -as, -os ‘big-f.sg / f.pl / m.pl’

b.

ō form[o]so ‘beautiful-m.sg’

c.

ō t[o]do, -a, -as, -os ‘all-m.sg / f.sg / f.pl / m.pl’ < tōtum

form[ɔ]sa, -as, -os (still form[o]sos in the 16th century) ‘beautiful-f.sg / f.pl / m.pl’ < formōnsum

PRom ­low-­mid /ɔ/ was raised before an etymological final -ŭ. While PRom ­ igh-­mid /o/ was preserved in principle (for instance, in (8c)), in several cases h metaphonic alternations were extended by analogy to some adjectives with original stressed /o/, as shown for form[o]so/form[ɔ]sa in (8b).6 This analogical leveling shows that metaphony has been morphologized: as its context was opacized phonologically (note that its alternants occur elsewhere as distinct phonemes: e.g. b[ɔ]la ‘a kind of cake’ vs. b[o]la ‘ball’), it became linked to morphological specifications only.

6 An alternative explanation was proposed by Lüdtke (1956: 175–­185), who interpreted application of metaphony in form[o]so/form[ɔ]sa as a remnant of a previous diachronic layer in which Portuguese may have had a vowel system of the Sardinian type ((2d)). While the trigger of metaphony in the examples in (8) is final -ŭ, also -ī caused metaphony in Portuguese (e.g. fiz ‘do-prt.1sg’ < fēcī).

62 

 Michele Loporcaro

3 On the rise of metaphony and diphthongization The diachronic relationship between metaphonic diphthongization, as observed in Neapolitan, and the other types of diphthongization, which occurred in different environments, e.g. in Florentine and Castilian ((1a), (1d)), and do not correlate with the quality of the final vowel, has been widely discussed. The opinions which have been defended may be classified as seen in (9): (9)   Does /ɛ ɔ/-diphthongization (in Tuscan, French, etc.) depend on the nature of the -V? a. no (two independent processes): Bourciez (1937: 94), Wartburg (1950: 82, ­122–­123, 141), Castellani (1962, ­1970a–­c), Lüdtke (1956: 82–­83), Loporcaro (2011b: 130–­135); b. yes: Schürr (1936, 1970), Lausberg (1976: 207, 228 on Tuscan, 230 on French), Ferguson (1976), Maiden (1995: 54–­55), ­Sánchez-­Miret (1998, 2007, 2010). For proponents of (9a), metaphony is an assimilatory process, while ­non-­metaphonic diphthongization of the French and Italian type is rooted in ­Proto-­Romance open syllable lengthening, and Castilian diphthongization is an implementation of a lengthening process which is not constrained by syllable structure. On the other hand, proponents of (9b) all agree that there is a relationship between diphthongization and the quality of the final vowel, but part ways as to the exact nature of this relationship. Most authors listed in (9b) share Friedrich Schürr’s view, according to whom Neapolitan metaphonic diphthongization is an assimilatory process, by which the degree of closure of the subsequent high vowel is anticipated on the onset of the stressed vowel (Schürr 1936: 279, 1970, etc.): “L’effet inflexionnant ou métaphonique des -ī, -ŭ […] suivants est causé par la particularité de leur articulation, à savoir l’élévation de la langue contre le palais ou le voile […] et il consiste à faire anticiper cette élévation ou fermeture suivantes dans la tension de la voyelle tonique.” (Schürr 1970: 14) [The inflecting or metaphonic effect of following -ī, -ŭ […] is caused by the peculiarity of their articulation, viz the raising of the tongue towards the palate or the velum […] and it consists in anticipating this following raising or closure in the tensing of the stressed vowel]

Under this view, all Romance languages, including those that display ­non-­metaphonic diphthongization (French, Tuscan, Castilian, etc.), and those ­on-­ diphthongizing metaphony (Sardinian (7), Portuguese (8), that display n



Metaphony and diphthongization in Southern Italy 

 63

­Sabino-­type dialects (6), etc.), must have gone through a preliterary stage of metaphonic diphthongization of the Neapolitan type: “Les résultats issus de la métaphonie de , ǫ́ s’expliquent tout naturellement par une monophtongaison même très ancienne des phases intermédiaires iẹ́, uọ́, ce qui est le cas du ­galicien-­portugais […], castillan primitif […], sarde […] non moins que des dialectes italiens ­ci-­dessus mentionnés” (Schürr 1970: 24–­25). [The results of the metaphony of [ˈe], [ˈo] are naturally explained by an even very old monophthongization of intermediate [ˈje], [ˈwo], as is the case for ­Galego-­Portuguese […], the primitive Castilian […], Sardinian […] as well as for the above mentioned Italian dialects]

In a series of more recent contributions, Fernando Sánchez Miret (henceforth FSM) proposes a version of (3a) in which, like in Schürr’s, the quality of the final vowel had an influence on diphthongization. However, he claims that this influence was not properly understood by Schürr and others: “Lo evidente del condicionamiento de [-i, -u] ha hecho que muchos lingüistas consideren superfluo un paso que a mi modo de ver es necesario en la explicación de un cambio fonético: la búsqueda de sus causas fonéticas” (Sánchez Miret 2007: 488). [The obviousness of the conditioning by [-i, -u] led many linguists to regard as superfluous a step that, in my opinion, is necessary for the explanation of a sound change: the search for its phonetic causes]

He thus takes—so to speak—an Ohalian stance, from which he raises the following objection to Schürr’s explanation: “Schürr was aware of a weak point in this explanation: why should anticipatory tongue raising manifest itself at the onset of the stressed vowel?” (Sánchez Miret 2010: 84)

To demonstrate the n ­ on-­plausibility of Schürr’s assumption, he mentions experimental work on ­coarticulation by D. Recasens and others, whose results “do not confirm the hypothesis that ­V-­­to-­V anticipatory effects in V1CV2 sequences can be traced at V1 onset only” (Sánchez Miret 2010: 85). His counterproposal is the following: Romance diphthongization is a spontaneous lengthening process, rather than an assimilatory one, which in languages like Spanish affected all stressed vowels, whereas in other varieties it was impeded, to different degrees, by several factors, one of which being the intrinsic duration of the vowel occurring in the following unstressed syllable. Since high vowels are intrinsically shorter than low vowels, diphthongization in Neapolitan was favored by final [-i, -u]: “word final high vowels cooccur with longer stressed vowels to a larger extent than word final low vowels. It is argued that this relationship could be at the origin of Southern Italian metaphonic diphthongization” (Sánchez Miret 2010: 83).

64 

 Michele Loporcaro

Thus, final [-i, -u] did not cause diphthongization through an assimilatory process. Rather, they allowed it due to their intrinsically shorter duration.

4 M  etaphony, diphthongization and the explanation of sound change The phonetic (im)plausibility argument that FSM raises, to dismiss Schürr’s account, is a powerful one. There is a wide consensus on the fact that recon­ structive arguments based on experimental phonetic evidence are of paramount importance for explanation in diachronic phonology. This has been stressed by many, perhaps most prominently over the past few decades by John Ohala, in his seminal work in the research line that he christened laboratory phonology (cf. e.g. Ohala 1989, 2003, 2005 etc.). However, application of such phonetic plausibility arguments is not in itself a guarantee of sound explanation. For it to be so, one must make sure that such arguments are applied to the appropriate objects, that means, to the reconstruction of sound change which is truly and purely phonetically motivated. Or, to put it in other words, to sound change in its incipient stage, as Ohala (1989: 191) is keen to stress, in observing that his methodology, with its “mechanistic purity”, elucidates the perceptual and articulatory preconditions of sound change, which can favor its first rise in individual speech. Sound change does not reduce to this, though, as was emphasized by an equally prominent research tradition, that concentrated on what one might want to call the grammatical implications of sound change. These often arise in a further step, as it were, with respect to the initial one focused on in explanations à la Ohala. In this vein, Kiparsky (1995: 657–658) discusses the ‘life cycle of phonological rules’, in which “[s]ound change can be assumed to originate through synchronic variation in the production, perception and acquisition of language, from where it is internalized by language learners as part of their phonological system. The changes enter the system as l­ anguage-­specific phonetic implementation rules” which “may in turn become reinterpreted as phonological rules, either ­post-­lexical or lexical”. Finally, “[i]n the phonologized stages of their life cycle, rules tend to rise in the hierarchy of levels”. Along similar lines, Dressler (1980, 1985) addresses the “deiconization path” which phonological rules follow over time: they start out as phonetically motivated to then progressively lose phonetic transparency, while at the same time acquiring the function of encoding morphological categories/processes. When it acquires morphological function while retaining phonetic motivation, a phonological rule



Metaphony and diphthongization in Southern Italy 

 65

(= PR) becomes a morphophonological rule (= MPR), to eventually turn into an allomorphic morphological rule (= AMR) as phonetic motivation is completely lost: (10) P  R > MPR > AMR deiconization of PRs (Dressler 1980, 1985) decreasing phonetic transparency/motivation ­non-­encoding (PRs) vs. encoding (MPRs, AMRs) of morphological meaning In Loporcaro (2011a) I discuss the advantages of such an approach to ­Italo-­Romance metaphony over most current work on the topic within Generative Phonology, from the ­SPE-­era to OT (e.g. Calabrese ­1984–­85, 1998; Walker 2005 etc.), which model all instances of I­talo-­Romance metaphonic alternations as though they were (still) synchronically phonological, rather than morpho(no)logical (a discussion that cannot be resumed here due to space limitations: on the morphologization of metaphony in Southern Italian dialects cf. Tuttle 1985; Maiden 1989; 1991; Fanciullo 1994). Note that the traditional label “sound change”, translating the Neogrammarian term Lautwandel, is vague as to the stage along the “deiconization” path in (10) in which (the rise of) a given process/rule is to be located. This ­vagueness must be resolved, whenever an explanation of a specific instance of sound change is proposed: without pinning this down, in fact, one cannot be sure that one is applying the appropriate explanatory tools to the right kind of object. The discussion on the rise of Romance diphthongization exemplifies this nicely: in fact, FSM’s phonetically based objection to Schürr’s explanation holds only if metaphonic diphthongization arose as a ­low-­level phonetic change, or, in ­Bermúdez-­Otero’s (2007: 506) terms, as an instance of stabilization, i.e. “a process of regular categorical change that creates a new phonological counterpart for an existing phonetic rule”. Should it turn out that this was not the case, then FSM’s argument would lose its force.

5 R  econstructing the diachronic relationship between metaphony and diphthongization We have seen in section 2 that there are different types of metaphony, classified by the nature of the input and the structural change: on the one hand, raising metaphony affected ­high-­mid vowels all over ­Central-­Southern Italy ((­5c–­d), (­6c–­d)), and ­low-­mid vowels in the Sabino and the Sardinian types ((­6a–­b) and (7)). On the other, diphthongizing metaphony is found on l­ow-­mid vowels in the Neapolitan type, (­5a–­b). If one wants to capitalize on this geographic variation for diachronic reconstruction, the question arises whether (and, if so, how)

66 

 Michele Loporcaro

metaphonic diphthongization and raising are related. The three logically conceivable answers are listed in (11): (11)  a. no (two unrelated processes): Mengel (1936), Rohlfs (­1966–­69: 1.­126–­8, ­153–­5); b. yes (different and successive steps in one and the same development): i. diphthongization > raising: Parodi (1892), Castellani (1973: 83), Avolio (1996: 321), Lausberg (1976: 228); ii. raising > diphthongization: Lausberg (1947), Lüdtke (1956), Barbato (2008). On the one hand, some scholars have disjoined the two types of metaphony and assumed that they have arisen independently from each other (11a). Obviously, a joint explanation would be more economical, though. Such is Schürr’s (1936), (1970) strategy (11b.i): metaphony, in his view, must be at the origin of all instances of Romance diphthongization, and all different types of metaphony of PRom ­low-­mid vowels nowadays observed must have started out as diphthongization, even in Sardinian, Portuguese or other varieties that provide no evidence of diphthongization whatsoever. Should (11b.ii) emerge as the right answer, this would crucially disconfirm Schürr’s scenario, and would also have consequences for FSM’s account: in fact, his phonetic argument from coarticulation, meant to deny that diphthongizing metaphony is caused by a final high vowel, would fall, because raising before a high vowel is a perfectly natural assimilatory change and does not pose any problems in terms of phonetic plausibility. The question has been dealt with by Barbato (2008), who provides three arguments in favor of (11b.ii): (12) a. philological: medieval texts from several areas that today display diphthongizing metaphony had raising metaphony, not diphthongization; b. geographical: the areal distribution of raising vs. diphthongizing metaphony; c. structural: the relationship of raising vs. diphthongizing metaphony with preservation vs. loss of contrasts in final vowels. The first argument—(12a)—comes from the philological record. Metaphony of  the Sabino type used to reach much further south and east during the Middle Ages: “Da tempo è noto che l’area cassinese, m ­ olisano-­abruzzese e marchigiana meridionale— oggi di tipo napoletano—producono testi di veste schiettamente mediana, sia per l’assenza di dittongazione metafonetica che per la conservazione di -u […] Più recentemente si è avanzata l’ipotesi che rientrassero nell’area mediana anche la Campania settentrionale



Metaphony and diphthongization in Southern Italy 

 67

(in senso antico, quindi inclusa Gaeta) e forse persino la Capitanata” (Barbato 2008: 280) [It has been known for a while that the Cassino, ­Molisano-­Abruzzese and Southern Marchigiano areas—today of the Neapolitan type—yield [Medieval, M.L.] texts with a clearly mediano outfit, both for the lack of metaphonic diphthongization and for the preservation of -u […] More recently, the hypothesis has been put forward that also northern Campania (in the old sense, i.e. including Gaeta) and perhaps even the Capitanata belonged to the area mediana]7

The Sabino type has shrunk over time, yielding to the expansion of the Neapolitan vowel system, as Naples, which was for centuries the capital of (mainland) Southern Italy, spread linguistic innovations. This is reflected—that is the second argument (12b)—in the areal distribution of the two types of metaphony in ­Central-­Southern Italy, as seen on Barbato’s (2008: 289) map, reproduced in Fig. 1, which elaborates on the AIS charts. The map shows that the Sabino type (red) today occurs in marginal areas: also Vignuzzi & Avolio (1994: 686 n. 20) observe that the Sabino type is today confined to Apenninic (i.e. montaineous) territories, less accessible—and less readily reached by innovations—than the plains in which diphthongizing metaphony prevails. As shown by the arrows on the map, the diphthongizing type spread not only from Naples but also, more to the north, from Rome, whose dialect had metaphonic diphthongs until the 15th century (see fn. 5), and later lost them as its dialect was Tuscanized. Barbato’s map brings us to the third argument (12c), the structural one. A black horizontal stroke across the yellow or red dot means that the dialect preserves the distinction of final unstressed -ŭ vs. -o, which is relevant because -ŭ was a trigger of metaphony whereas original -o was not. As one sees at first glance, there are more red dots with a black stroke, than yellow. In other words, preservation of the -ŭ vs. -o contrast ((13a)), which singles out area mediana with respect to the other dialect areas of (­Central-­Southern) Italy ((­13b–­d)), tends to ­co-­occur with raising rather than diphthongization. There are some exceptions, admittedly, but this is the overall tendency.

7 Area mediana (cf. Loporcaro & Paciaroni, to appear) is the traditional label designating the major dialect subdivision interposed between Tuscany (to the north) and the Upper South (the area centring on Naples, to the south). This subdivision includes mostly dialects featuring metaphony of the Sabino type, which is also sometimes labeled metafonesi ciociaresca (Sabina and Ciociaria beeing parts of area mediana, lying n ­ orth-­­north-­east and ­south-­east of Rome, respectively). On the broader extension of “mediano”-type features into today’s Upper South during the Middle Ages cf. Vignuzzi (1994: 332–­339), Barbato (2000: 111–­113), (2002: 39).

68 

 Michele Loporcaro

Rome

Naples Fig. 1. (Barbato 2008: 298 on AIS data). Darker-gray dots: raising (Sabino type) metaphony. Lighter-gray dots: diphthongizing (Neapolitan type) metaphony. Barred dots: -/u/ vs. -/o/ phonetically distinct.

(13) Final unstressed vowels from Latin to ­Central-­Southern ­Italo-­Romance a. Area mediana (Sabino type) -i -e -o -u -a b. Old Neapolitan -ə -o -a c. Peripheral Campanian, Molisano etc. -ə -a d. Modern Neapolitan -ə



Metaphony and diphthongization in Southern Italy 

 69

Barbato capitalizes on this observation in order to propose the stepwise development synthesized in (14) (exemplified only for the front mid vowel, for ease of presentation). (14) a. Sardinian type: metaphony still an allophonic PR (there is no *[e] ← /e/) /ɛ/ → [e] / - {/i/, /u/} b. Sabino type: metaphony a MPR (there is a distinct [e] ← /e/ in the language) /ɛ/ → /e/ / - {/i/, /u/} c. Pre-­Neapolitan (reconstructed): metaphony an AMR /ɛ/ → /e/ / -/o/ [m.sg] (only some -[o]’s [those < -ŭ] trigger metaphony) d. Old Neapolitan: metaphony an AMR /ɛ/ → /je/ / -/o/ [m.sg] (the rule’s output becomes more differentiated) e. Modern Neapolitan: metaphony an AMR /ɛ/ → /je/ / [morphologically specified context] In the first stage, still observed in Logudorese (14a), metaphony is still an allophonic PR. The next step towards opacization of the metaphony rule (14b) is observed in the ­Sabino-­type dialects, where raising is still transparently triggered by a following unstressed high vowel, except that the phonological system also features an independent phoneme /e/, whose surface realization (→ [e]) merges with the output of metaphonic raising /ɛ/ → [e]. Metaphony, at this stage, is an MPR, that retains phonetic plausibility but at the same time takes on a morphological function.8 (14b) is (a fragment of) the kind of system exemplified in (6), which we have compared with Neapolitan in (5). Contemporary Neapolitan, whose metaphony AMR is schematized in (14e), was preceded by at least two previous distinct stages: (14d) is attested for Old Neapolitan (­13th–­16th centuries), which already had diphthongizing metaphony but still contrasted three unstressed final vowels, as shown in (13b). This is the result of the standard analysis of the Old Neapolitan vowel system by Formentin (1998: 100–­116, ­187–­188, 350, 711), who analyzes 15th century data such as accuoncio/accuoncie ‘in agreement-m.sg/m.pl’, moro/muore ‘die-1sg/3sg’ and shows that the final -ī in the masculine plural and the second singular endings was spelled —attesting to merger with original -e > [ə]—whereas was regularly spelled as such, but never , in the masculine singular ending 8 Note that while metaphonic alternations may signal morphological category values (for gender, number, or person), it would be wrong to claim—as is sometimes done—that metaphony arose in order to compensate for the loss of distinction caused by the merger of final vowels. This is trivially proven by the fact that dialects of area mediana like Serviglianese (in (6) above), as well as several dialects of Calabria and Salento, display metaphonic alternations and yet preserve (three to four) contrasting unstressed final vowels.

70 

 Michele Loporcaro

(< -ŭ), attesting to its merger with original ‑o (as found in 1sg moro)—except that the latter did not trigger metaphony—but not with [ə] < -e = -ī. (14c), on the other hand, must be reconstructed for a still earlier ­pre-­documentary stage of Old Neapolitan, given the evidence in (12), which demonstrates the gradual expansion of metaphonic diphthongization and the retreat of raising metaphony. We have positive documental evidence that this replacement took place in Molise, Abruzzo, Southern Marche and today’s South Lazio (Cassino, Gaeta): this suggests that the same substitution may have taken place, before the earliest documents of the dialect, in Naples as well.9 The crucial difference between reconstructed (14c) and attested (14d) is that in the former, metaphony must still have been of the Sabino (i.e. raising) type. The structural reason why metaphonic raising may have yielded to diphthongization, in Naples like elsewhere, is that the rule had been opacized via the merger of final unstressed -/u/ and -/o/. In fact, as seen on the map in Fig. 1, the -/u/ vs. -/o/ contrast ­co-­occurs more often with raising than with diphthongizing metaphony. This squares well, as Barbato (2008: 283) points out, with the fact that MPRs and AMRs “generally involve more radical change” (Dressler 1985: 63) between input and output than PRs, i.e. tend to be “hypercharacterized”. In sum, this reconstruction of the rise of Neapolitan diphthongizing metaphony leads Barbato to dismiss FSM’s objection to the assimilatory nature of metaphony: “È evidente che, se si considera il dittongo metafonetico come un effetto della perdita della motivazione fonologica della metafonia (e quindi del suo carattere assimilativo), l’apparente contraddizione tra metafonia e dittongazione scompare.” (Barbato 2008: 283) [It is evident that, if one considers the metaphonic diphthong as an effect of the loss of the phonological motivation of metaphony (and hence, of its assimilatory character), the apparent contradiction between metaphony and diphthongization disappears]

This refutation is in my view impeccable: FSM’s objection is devoid of empirical content since in the reconstructive scenario in (14) the only primary change in this domain of the phonology of Southern Italian dialects was the rise of

9 The change did not affect all instances of surface [ˈe], as shown by the fact that stressed [ˈe] from PRom /e/ remained unaffected (e.g. [ˈveːrə] ‘see-1sg’ < vĭdet). Rather, what was modified through the rise of metaphonic diphthongization was only the surface realization of /ˈɛ/ (< PRom /ɛ/) in all environments where it was subject to the application of the metaphony rule. An anonymous reviewer objects that “if diphthongization was a secondary process affecting /e o/ (deriving from metaphony) we would expect that it could affect all /e o/ nuclei, taking on the etymological ones as well”. This is not a necessary expectation, though, under the view of phonology adhered to here (cf. section 8), given the distinction between surface and underlying representations.



Metaphony and diphthongization in Southern Italy 

 71

metaphonic raising of the Sabino type, while diphthongizing metaphony was a later development of its output, implemented as the rule had lost phonetic motivation. Not dissimilarly, early modern German sünge ‘sing-sbjv.pret.1sg’, which had arisen from OHG sungi via metaphony, changed to modern German sänge (cf. Nübling 1999) centuries after that the phonological metaphony rule had ceased to operate, becoming an AMR (as shown by Dressler 1985: 127). As argued in Loporcaro (2011b: 132–­134), (2013: 125–­126), crucial evidence in support of the reconstruction in (14) comes from recent experimental studies on some Southern Italian dialects, to which we now turn.

6  Experimental evidence for the primacy of metaphonic raising over diphthongization 6.1   ­Co-­existence of raising and diphthongizing metaphony in some ­Upper-­South dialects For several dialects scattered throughout the northern part of the U ­ pper-­South subdivision, coexistence of metaphonic and raising diphthongization of /ɛ ɔ/ has been described. For instance in Minturno (province of Latina, Southern Lazio) one finds on the one hand [ʎu ˈpɛːrə] ‘the foot’, [ʎi ˈpeːrə] ‘the feet’, but on the other also [n də ˈsjendə ˈbːwoːnə] ‘don’t you feel alright?’. A similar vacillation is reported e.g. for nearby Isola Liri, province of Frosinone, by Vignuzzi (2005: 86), for the dialects of Sannio Beneventano, in northern Campania, by Maturi (2002: 47–­50), or for Greccio, in the province of Rieti, northern Lazio—e.g. [ˈwortu] ‘vegetable garden’/[ˈordzu] ‘barley’ vs. [ˈpɔrta] ‘door’—by Paiella (1973: 425, 427). These instances of vacillation have been interpreted in either of the two different ways synthetized in (11b.­i–­ii). I shall now consider in some detail one such dialect, the one spoken in Agnone (province of Isernia, Molise), drawing on an experimental phonetic study by Loporcaro et al. (2007), in order to show that (11b.ii) must be preferred. In this dialect, one finds, at word level, diphthongizing metaphony of the Neapolitan kind. (15) Dialect of Agnone (Molise) PRom /ɛ/ a. Metaphonic alternant (masc): b. ­Non-­metaphonic alternant (fem):

PRom /ɔ/

[ˈvie̯cːə] [ˈvɛcːɐ]

[ˈgruo̯sːə] [ˈgrɔsːɐ]

‘old’

‘big’

72 

 Michele Loporcaro

In addition, the dialect of Agnone also shows many other diphthongs conditioned by syllable structure (cf. Ziccardi 1910: 417), as seen in (16), where glosses are in Italian, in order to show approximately how the corresponding words reconstruct in ­Proto-­Romance. (16) PRom a. Open syllable

b. Closed syllable

/i/

[vəˈtʃoi ̯nə] ‘vicino’

[ˈviɲːɐ] ‘vigna’

/e/

[ˈpɔi ̯pə] ‘pepe’

[ˈʃtelːɐ] ‘stella’

/ɛ/

[ˈpɔi ̯də] ‘piede’

[ˈvɛcːɐ] ‘vecchia’

/a/

[ˈsɛæ̯nə] ‘intero’

[ˈkakːə] ‘qualche’

/ɔ/

[ˈreu̯sɐ] ‘rosa’

[ˈʃtɔtːsə]

/o/

[kuˈlau̯rə] ‘colore’

[ˈkortɐ] ‘corta’

/u/

[ˈliu̯nɐ] ‘luna’

[ˈrudːzə] ‘rozzo’

‘tozzo (di pane)’

The open syllable diphthongs in (16a) are not stable: rather, they occur only variably. The diphthong is normally realized prepausally, whereas in ­non-­utterance final position it shows a strong tendency to monophthongize, as shown in (17) (the boxes highlight ­non-­diphthongized ­utterance-­internal stressed vowels). (17) a. Prepausal

b. ­Non-­prepausal etyma

[ˈji lə ˈva i  ̯də  ]

[lə ˈvedə ˈjojːə]

‘I see him’

< vĭd(e)o

[ˈsanə ˈsɛæ̯nə  ]

[ ˈsanə ˈsɛæ̯nə]

‘totally untouched’

< sanum

The ­non-­prepausal realization clearly preserves the older stage here, as can be seen from comparison with the PRom vowels listed in the first column in (16), ­ uscan-­based standard Italian (see the glosses). A comwhich stayed as such in T parable variation between prepausal diphthongs and ­utterance-­internal monophthongs is found in other Southern Italian dialects too (cf. Loporcaro 2011c: 76–­77 for previous references, most notably Rohlfs 1938, and the experimental study on the dialect of Pozzuoli, province of Naples, by Abete 2006). What has never been described for any other Southern Italian dialect but Agnonese is that here also metaphonic diphthongs participate in this s­ entence-­phonetic alternation. (18)

a. Prepausal

b. ­Non-­prepausal

Agnone

[ɲːa tə ˈsie̯ndə  ] ‘how do you feel?’ [n də ˈsendə ˈvuo̯nə] ‘aren’t you feeling good?’ [ˈvoːnə ˈvuo̯nə  ] ‘very good’

 [ ˈvoːnə ˈvuo̯nə]



Metaphony and diphthongization in Southern Italy 

 73

Note that, when not diphthongized in utterance internal position, the vowel surfaces as a ­high-­mid, not a ­low-­mid, vowel: in other words, it does not retain its etymological quality but is raised, as exemplified with the metaphonic diphthong [ie̯] in Fig. 2.

Fig. 2. The metaphonic diphthong [ie̯] in Agnonese, compared with the corresponding monophthongs (one male speaker).

In Agnonese, the stressed metaphonic diphthong [ie̯] (as in [ˈsie̯ndə] ‘feel-2sg’, (18a)) displays considerable variation in its offglide, as shown by the two arrows in Fig. 2, which indicate the trajectories in the vowel space of the two main variants of this diphthong. In more conservative speech, this diphthong has an [e] offglide, marked as an x on the vowel chart. Now, whenever the same stressed vowel occurs as a monophthong ­utterance-­internally ([ˈsendə] ‘feel-2sg’, (18b)), this vowel— marked as a “+” sign—almost overlaps with the abovementioned offglide, and is clearly distinct from the ­non-­metaphonic outcome of the same etymological vowel (as e.g. in [ˈsɛndə] ‘feel-1/3sg’), marked as a circle in Fig. 2. Thus, synchronic variation in (18a–b) shows that even in a dialect like this, in which diphthongization has established itself as a ­word-­level implementation of metaphony, raising metaphony of the Sabino type (/ɛ/ → [e]) still occurs in ­utterance-­internal position, i.e. in a position that preserves a more conservative realization of the stressed syllable nucleus, as demonstrated by the independent evidence from ­non-­metaphonic diphthongs in (17). This is in turn s­ ystem-­internal evidence for the chronological and structural priority of metaphony by raising over metaphony by diphthongization. The

74 

 Michele Loporcaro

latter, in Agnonese, seems to be dependent on a w ­ ell-­known phonetic universal: prepausal lengthening (cf. Maddieson 1997: ­631–­632; ­Pompino-­Marschall 2010: ­413–­414). This suggests a complement to the reconstruction of the rise of metaphonic diphthongization presented in section 5: prepausal lengthening, on the phonetic side, may have converged with the opacization of the trigger, on the phonology side, to favor the change from raising to diphthongization (14c) > (14d). An anonymous reviewer suggests the following alternative: “We could see the alternations discussed in (18) as a result of the application of these spontaneous phenomena of diphthongization/aperture [i.e. the n ­ on-­metaphonic diphthongization processes affecting stressed wowels “in many southern dialects (Abruzzi, Apulia, Lucania)”, M.L.].” Two objections can be raised to this alternative account. Firstly, the contextual variation in (18) is not subject to any ­metric-­syllabic constraints, unlike Abruzzian, Apulian and eastern Lucanian ­non-­metaphonic diphthongizations. This lack of sensitivity characterizes metaphony all over Central-­Southern Italy (cf. (1c)). Secondly, as for the sounds involved, it is a fact that the two variants in (­18a–­b) correspond to the outputs of the metaphony of PRom ­low-­mid /ɛ ɔ/ all over ­Central-­Southern Italy, and there is abundant evidence that several dialects which now show metaphonic diphthongues used to display raising metaphony of the Sabino type at earlier stages. Thus, there is evidence for the occurrence of S ­ abino-­type metaphony in medieval texts from areas where today metaphonic diphthongization occurs (cf. (12a) above and Barbato 2008: 280), and the same replacement has been observed much more recently: it is still ­on-­going e.g. in Minturno (province of Latina, Southern Lazio), while it has been completed recently in nearby Formia (cf. Del Puente & Fanciullo 2004: 163–­164). Under the reviewer’s counteranalysis, disjoining (­18a–­b) from metaphony, the identity in both the affected segments and the application context of the rule should be regarded as merely coincidental, whereas they are naturally explained under the view defended here: Agnonese still preserves the initial raising metaphony only in ­non-­prepausal position (18b), whereas it has generalized the metaphonic diphthongue prepausally (18a).

6.2 “Unexpected” raising metaphony in ­South-­eastern Salentino Most of the ­ Italo-­Romance dialects of the subdivision traditionally labelled “extreme South” (Pellegrini 1977: 31) do not provide any decisive ­system-­internal evidence, of the kind discussed in sections 5 and 6.1, for the chronological priority of raising over diphthongizing metaphony (11b.i). This is stressed by an anonymous reviewer: “In Calabrian, raising and diphthongizing metaphony coexist, whereas other types of diphthongization are lacking—nor does sensitivity to

Metaphony and diphthongization in Southern Italy 



 75

syllabic structure emerge. There is no evidence to argue for the older character of the simple raising. Similarly, in Sicilian dialects metaphonic diphthongization is present while other types of diphthongization are absent.” Note however that the claim that the evidence from Calabrian—which shows metaphony north of the Vibo ­Valentia-­Punta Stilo line, cf. Loporcaro (2013: 159)—is neutral, as to the choice between (11a), (11b.i), and (11b.ii), overlooks the (cogent, in my opinion) geolinguistic argument by Barbato (2008: 279), according to whom the area showing ­Sabino-­type metaphony in central Calabrian (e.g. in Belsito, Mangone, Rogliano in the province of Cosenza, or in Casabona, in the province of Crotone; cf. Rohlfs 1­ 966–­69: 1.128, Tucci 1­ 996–­97: 7–­11) “configura un caso da manuale di area isolata, facendo sorgere il dubbio che la metafonia ‘originale’ sia quella sabina” [represents a textbook case of an isolated area [ergo conservative, in terms of Bartoli’s (1943) first areal norm, M.L.], motivating the suspect that the ‘original’ metaphony is the Sabino one]. Moreover, the dialects of the extreme South hold compelling evidence in store, bearing on the controversy over the origins and relationship of metaphony and Romance diphthongization. In fact, further evidence in favor of hypothesis (11b.ii) is provided by the results of the experimental study of the dialects of Southern Salento by Grimaldi (2003). Prior to his study, the traditional view on Salentino has it that metaphony is realized exclusively as diphthongization, and occurs in Northern and Central Salento, as exemplified with Leccese in (19a). (19) a. Central Salentino Before -a -e -o

Before -i -u

ŏ [ˈsɔ̝ːna] < sonat ‘play / sound-3sg’ [ˈseːni] (< [ˈsweːni]) ‘play / sound-2sg’ ĕ

[ˈtɛ̝ːne] < tenet ‘keep-3sg’

[ˈtjeːni] ‘keep-2sg’

b. Southern Salentino (standard account: e.g. Parlangeli 1953, 1960) Before -a -e -o

Before -i -u

ŏ [ˈsɔ̝ːna] < sonat ‘play / sound-3sg’

[ˈsɔ̝ːni] ‘play / sound-2sg’

ĕ [ˈtɛ̝ːne] < tenet ‘keep-3sg’

[ˈtɛ̝ːni] ‘keep-2sg’

This metaphonic diphthongization did not spread so far south as to cover the whole Salentine peninsula, as shown on the map in Fig. 3 (from Loporcaro 2013: 158). The relevant isoglosses are 5 and 6 (plus 7, to be discussed directly): ŏ > [we] stops at a line N ­ ardò-­San ­Cesareo-­Vèrnole (isogloss 5), whereas ĕ > [je] (isogloss 6) reaches further south (Gallipoli). However, Southern Salentino (19b) lacks metaphonic diphthongs altogether.

76 

 Michele Loporcaro

Fig. 3. Metaphony in Salentino (after Loporcaro 2013: 158) Isogloss 5: southern limit of diphthongization of Latin stressed Ŏ > [we] Isogloss 6: southern limit of diphthongization of Latin stressed Ĕ > [je] Isogloss 7: northern and western limit of raising metaphony affecting the outcomes of Latin stressed Ĕ and/or Ŏ.

For historical and geographical reasons, it is clear that diphthongizing metaphony spread southwards during the early Middle Ages, possibly as one of the innovations favored by the Langobardic conquest, as argued by Parlangeli (1953, 1960): the territories which resisted the innovation were the extreme strongholds of the area which remained under Byzantine rule at that time.1⁰ Lacking metaphonic diphthongization, those southernmost dialects (near the

10 This kind of external factors (spread of sound change, possibly favored by the sociopolitical unraveling caused by conquest by new rulers) adds to the phonological and phonetic factors involved in the rise and/or spread of metaphonic diphthongization discussed in sections 5 and 6.1.

Fig. 29(a): Melissano Metaphony and diphthongization in Southern Italy 

 77

200 i

[E] per -/i/

u

400

[E] per -/u/ F1

[E] per -/e, a/

600

a

[O] per -/i/ [O] per -/u/

800

[O] per -/e, a/ 1000 3000

2500

2000

F2

1500

1000

500

Fig. 4. The vowel system of the dialect of Melissano (province of Lecce, after Grimaldi 2003: 220).

Capo di Leuca) have been commonly claimed not to display metaphony at all. Grimaldi’s study, however, shows that this is indeed true only of the dialects of the South-western corner of Salento, from Gallipoli down to Ugento, west of isogloss 7, as exemplified with the dialect of Melissano in the vowel chart in Fig. 4: ellipses for stressed mid vowels divided by final vowel coincide. Here, no significant coarticulatory effect of final high vowels on stressed [ɛ̝ ɔ̝] is observed: this confirms that the dialect of Melissano is a genuine instance of type (19b). However, the dialects spoken east of isogloss 7 do show a significant raising of stressed mid vowels either (according to the specific dialects) just before -/i/, or only before -/u/, or before both high vowels. Also, parallel to the asymmetry in diphthongizing metaphony in the dialects spoken a couple of miles to the north, in South-eastern Salentino as well, raising affects [ɛ̝] in more dialects, while it affects also [ɔ̝] in fewer: (20)

Before -/i/

Before -/u/

Before -/a/, -/e/

[ˈbːɛ̝ɖːʐu] [ˈdɛ̝ntʰɛ̝] ‘tooth’ ‘handsome-m’ /ˈɔ/ [ˈʃɔːki] ‘play-2sg’ [ˈʃɔːku] ‘play-1sg’ [ˈkɔːrɛ̝] ‘heart’

gray = raising

a. /ˈɛ/ [ˈdentʰi] ‘teeth’

S. Maria di Leuca

b. /ˈɛ/ [ˈdentʰi] ‘teeth’

Patù, Tiggiano

[ˈbːɛɖːʐu] [ˈdɛntʰɛ] ‘tooth’ ‘handsome-m’ /ˈɔ/ [ˈʃoːki] ‘play-2sg’ [ˈʃɔ̝ːku] ‘play-1sg’ [ˈkɔ̝ːrɛ] ‘heart’

78 

 Michele Loporcaro

c. /ˈɛ/ [ˈdentʰi] ‘teeth’

[ˈbːɛɖːʐu] [ˈdɛntʰɛ] ‘tooth’ ‘handsome-m’ /ˈɔ/ [ˈʃoːki] ‘play-2sg’ [ˈʃoːku] ‘play-1sg’ [ˈkɔːrɛ] ‘heart’

Alessano

For instance in S. Maria di Leuca, a dialect of type (20a), [ˈpeːdi] ‘feet’ vs. [ˈpɛ̝ːdɛ̝] ‘foot’, [ˈdenthi] ‘teeth’ vs. [ˈdɛ̝nthɛ̝] ‘tooth’ show metaphony, while no allophonic variation for stressed /ɔ/ is observed. In dialects of type (20b) like those of Patù or Tiggiano, on the other hand, also /ɔ/ undergoes metaphony: [ˈʃoːki] ‘play-2sg’ vs. [ˈʃɔ̝ːka] ‘play-3sg’, [ˈkoːri] ‘hearts’ vs. [ˈkɔ̝ːrɛ̝] ‘heart’ (Grimaldi 2003: 60–65). An example of type (20c), where both /ɛ/ and /ɔ/ are raised before /i/ whereas only /ɔ/ is raised to [o] also before final -/u/, is shown in Fig. 5 for the South-eastern Salentino dialect of Fig. 7(a): Alessano Alessano.

200 i

F1

[e] per -/i/

u

400

[E] per -/u/ [E] per -/e, a/

600

[o] per -/i/

a

[o] per -/u/

800

1000 3000

[O] per -/e,

2500

2000

F2

1500

1000

500

Fig. 5. The vowel system of the dialect of Alessano (province of Lecce, after Grimaldi 2003: 192).

These data call for a straightforward interpretation, and shed crucial light on the issue of the diachronic relationship between diphthongization and metaphony. Spoken at the south- eastern corner of the whole Romania, the dialects of South- eastern Salento were never attained by diphthongizing metaphony, which in Salento extends only as far south as the line Nardò-San Cesareo-Vèrnole (isogloss 5 on the map in Fig. 3), for ŏ > [we], and as far south as Gallipoli (isogloss 6) for ĕ > [je]. However, South- eastern Salentino does show raising metaphony, just like Logudorese Sardinian ((7)). This confirms that Logudorese metaphony has to be taken at face value, pace Schürr (1970: 26–27), Ferguson (1976: 126–127), and others: metaphony in statu nascendi was (and

Metaphony and diphthongization in Southern Italy 

 79

still is, in Logudorese as well as in South- eastern Salentino) a process of raising, not of diphthongization.11 In the many Romance varieties in which metaphony (induced by original -/i/ and, possibly, -/u/) occurs in the form of diphthongization (e.g. Neapolitan, (5)), this mirrors a further development ([ɛ] > [e] > [je], [ɔ] > [o] > [wo]), as still arguable synchronically from data such as those from Agnonese considered above in section 6.1.

7 Diphthongization and metaphony in the prehistory of Romanian Among the Romance languages outside of Italy, Romanian stands out in that its stressed mid vowels show the effects of both metaphony and diphthongization. It thus provides a crucial testing ground for the reconstruction just outlined. In this language, the diphthongization facts are particularly intricate and have been interpreted in contrasting ways. The relevant data, as for the modern language, can be summarized as follows. (21) a. b. c.

Before -a ŏ / ō poartă ‘brings’, coadă ‘tail’ ĕ iarbă ‘grass’, iapă ‘mare’, piatră ‘stone’ ē / ĭ seară ‘evening’, neagră ‘black-f.sg’

Before -e soare ‘sun’, noapte ‘night’ iepe ‘mares’, pietre ‘stones’ vede ‘see-3sg’, lege ‘law’

Before -i -u -o Romanian portu ‘bring-1sg’, nopţi ‘nights’ fier ‘iron’, piept ‘breast’, ieri ‘yesterday’ leg ‘tie-1sg’, negri ‘black-m.pl’, cred ‘believe-1sg’

Firstly, syllable structure is not relevant here to determine the different outcomes. Secondly, due to the asymmetric mergers seen in (2c), Lat. ō was also affected by the same diphthongization process as ŏ, whereas on the palatal side diphthongization under the same contextual conditions—but with different outcomes— affected not only ĕ but also ē / ĭ. Thirdly, these contextual conditions clearly involve the quality of the final vowel. Note that the seeming difference in the influence of final -e on the stressed vowels in (21a) vs. (21b–c) is a mirage, due to a process of harmonization of the stressed diphthongs from ĕ and ē / ĭ to the unstressed final vowel. Before its

11 Note that, in both Sardinia (cf. Loporcaro 2003: 86) and South-eastern Salento, raising metaphony carries over to the local pronunciation of Italian: e.g. (in Salento) [paˈreːri] ‘advices’ vs. [paˈrɛ̝ːrɛ̝] ‘advice’ (Grimaldi 2003: 64). This proves that metaphony is still a synchronically active phonological rule.

80 

 Michele Loporcaro

application, in Old Romanian (as seen in (22)) the same outcomes were observed before -e as before -a, just like it is still the case for the outcomes of ŏ ō in (21a) (identical to Old Romanian (22a)). (22) a. b.

c.

Before -a ŏ / ō poartă ‘brings’, coadă ‘tail’ ĕ iarbă ‘grass’, iapă ‘mare’, piatră ‘stone’ ē / ĭ seară ‘evening’, neagră ‘black-f.sg’

Before -e soare ‘sun’, noapte ‘night’ iape ‘mares’, piatre ‘stones’

Before -i -u -o Old ­Romanian portu ‘bring-1sg’, nopți ‘nights’ fier ‘iron’, piept ‘breast’, ieri ‘yesterday’

veade ‘see-3sg’, leage ‘law’

leg ‘tie-1sg’, negri ‘black-m.pl’, cred ‘believe-1sg’

ORom veade ‘see-3sg’, leage ‘law’ vs. piatre ‘stones’ then changed to vede, lege vs. pietre (cf. Lambrior 1878: 85–­86 fn. 4, Lausberg 1976: 227–­228) in ­Daco-­Romanian, whereas the older stage ([ˈle̯adʒɛ] ‘law’, [ˈse̯atɛ] ‘thirst’) is still preserved in ­Macedo-­Romanian. The (Old) D ­ aco-­Romanian (and ­Macedo-­Romanian) facts, thus, show that in a first step the same outcomes were found before -a -e, on the one hand, and before -i -u -o, on the other. Since in Romanian -o merged with -u early on, let us call the two contexts, simplifying a bit, “before ­non-­high” vs. “before high vowel” respectively. Now, this is precisely the contrast found in the environment of metaphony, as exemplified above in (5)–(7), except that ŏ ō and ē ĭ diphthongized in Romanian before ­non-­high vowels, whereas metaphonic diphthongization of the Neapolitan type (in (5)) occurs before high vowels, not before ­non-­high ones. This led many scholars to regard ŏ ō > [o̯a] and ē ĭ > [e̯a] as conditioned diphthongization, calling it metaphonic (cf. e.g. recently Avram 2005: 22; and cf. Sánchez Miret 1998: 193, who cites a long list of previous acoounts along those lines, starting with Mussafia 1868). However, it is easily objected that the same diphthongization takes place where no conditioning environment can be possibly spotted: e.g. dea < dēt ‘give-3sg.subj’, stea < stēt ‘stay-3sg.subj’.12 These data are incompatible with statements such as the following: “Vocalele e și o au fost diftongate în limba română, când silaba următoare a cuvântului conținea un a (ă) sau un e” (Rosetti 1986: 634) [The vowels e and o have been diphthonziged in Romanian, when the word’s following syllable contained a (ă) or e]. Rather, as argued by Nandriş (1963: 212–­213) and Sala (1976), (2004: 272–­274), the Romanian facts demonstrate that this diphthongization is not conditioned. 12 There are several such examples, which did not prevent supporters of the conditioned nature of diphthongization from proposing ad hoc analogical explanations for each one of them (cf. e.g. Schürr 1964: 359 and the discussion by Avram 2005: 20).

Metaphony and diphthongization in Southern Italy 



 81

Consequently, since [o̯a] and [e̯a] developed by default, then the complementary context must have been positively specified in the phonological rule which must have accounted for the distribution of the allophones prior to diphthongization: this is in fact what has been argued by the many who reconstruct for P ­ roto-­­Balcan-­Romance (23) and ­pre-­literary Romanian (24) a stage in which a final high vowel induced an allophonic realization [e o] of stressed mid vowels, in complementary distribution with stressed [ɛ ɔ] (later > [e̯a o̯a]) occurring elsewhere (cf. Vasiliu 1968: 40–­41). (23)

Before -a

Before -e

Before -i -u -o

a. ŏ / ō

ɔ

o

b. ĕ

jɛ

je

c.

ɛ

e

ē / ĭ

(24)

Before -a a. ŏ / ō b. ĕ c. ē / ĭ

Before -e

Before -i -u -o

o̯a

o

*je̯a

je

e̯a

e

­Proto-­­Balcan-­Romance

­pre-­literary Romanian

Sala (2004: 272) parallels the reconstructed allophonic distribution in (23) with the working of metaphony in ­Galego-­Portuguese. Note that, at the stages in (23)–(24), ĕ diphthongization to [je] (21b) must have already been applied. This has to be assumed on two grounds. First, had this not been the case, the outcome of ĕ would have merged with either *[ɛ] or *[e] from ē / ĭ, contrary to fact. Second, the output of ĕ-­diphthongization came to be subject to the same metaphonic variation, giving e.g. *[ˈfjeru] ‘iron’ (before high vowels) vs. *[ˈjɛrba] ‘grass’ (elsewhere). The nucleus of the latter was then seized by the later diphthongization process affecting *[ɛ] from ē / ĭ (Lambrior 1878: 86–­87). This yielded an intermediate *[je̯a] (24b), later reduced to [ja]. The above reconstruction has an inescapable consequence, as to the chronology of diphthongization and metaphony: in Romanian, the former must have preceded, because the latter applied to both its output and to previous /E/ (from Lat. ē / ĭ) unaffected by (Romance) diphthongization. (25) schematizes this earlier stage, in which diphthongization has already applied, but metaphony has not yet (upper case /E O/ stand for mid vowels unspecified for tenseness/ATR). (25)

Before -a Before -e Before -i -u -o ­pre-­­Proto-­­Balcan-­Romance a. ŏ / ō

O

b. ĕ

jE

c. ē / ĭ

E

82 

 Michele Loporcaro

Summing up, the D ­ aco-­ Romance facts provide evidence for a process of Ĕ-­diphthongization prior to and independent from the application of metaphony. This conclusion confirms that diphthongization of PRom lower mid vowels is old,13 a result independently arrived at from inspection of the G ­ allo-­and ­Italo-­Romance facts, and that it is not amenable to an original metaphonic conditioning.

8  Conclusion Ohala (1986: 22) concludes his “Consumer’s guide to evidence in phonology” with the observation that “although phonologists may be able to extract useful information from all sources of evidence […] [e]xperimental evidence is most efficient in allowing us to weed out ­non-­competitive hypotheses”. This is obviously true— as shown in section 6 with the evidence from Southern Italian dialects for the chronological and structural primacy of metaphonic raising over diphthongization—though when the method is applied to sound change, one crucial proviso must be added: as shown in section 5, one has to make sure that arguments from experimental evidence are applied in the right place, i.e. to the reconstruction of sound change in its incipient stage along the cline in (10). This decision cannot be made automatically, but is the linguist’s responsibility: as Coseriu (1973: 181 n. 8) humorously put it, [talking about (the rise of) allophonic distinctions], “en este sentido, no podemos esperar nada de los laboratorios, que no suelen resolver problemas racionales” [in this sense, we can hardly expect anything from the labs, which do not usually solve rational problems]. More generally, the results of the present study support a classical modular feedforward model of phonology (cf. ­Bermúdez-­Otero 2007: 503), and seem hard to reconcile with reductionist views of phonology as ­non-­distinct from phonetics. From the empirical point of view, in this paper I have answered the two questions posed at the outset: firstly, all changes rubricated under the heading “metaphony” can be reduced to one primary sound change (phonetic rule, in ­Bermúdez-­Otero’s 2007: 506 terms), viz. assimilatory raising triggered by a following unstressed high vowel; secondly, and relatedly, if metaphony did not

13 The absolute chronology assumed in Avram (2005: 21), after Rosetti (1986: 93), for the diphthongization of the outcome of Lat. ĕ is “after the 5th century”, i.e. early enough for this to be considered as part of the L ­ atin-­Romance transition in the Balcans as well (cf. section 2 for the absolute chronology of Romance diphthongization in ­Gallo-­and ­Italo-­Romance). The asymmetric vocalism of Romanian caused a mismatch here with respect to the outcome of ŏ: despite this, ­pre-­literary ŏ diphthongization is sometimes assumed for Romanian too (e.g. Schürr 1970: 65; Lausberg 1976: 227): see the discussion of the counterarguments by Avram (2005: 19–­20).



Metaphony and diphthongization in Southern Italy 

 83

originally start out as diphthongization in the first place, then metaphony, pace Schürr, cannot be the original kernel of Romance diphthongization, and, more generally, claims of a causal link between diphthongization of PRom /ɛ ɔ/ and the quality of the following unstressed vowel do not stand a closer inspection of the available evidence. This evidence shows that metaphony and diphthongization of PRom /ɛ ɔ/ are best conceived as two independent changes: in Romanian, as seen in section 7, diphthongization of stressed /ɛ/ must have been applied prior to the metaphonic raising of mid vowels before unstressed high vowels, whereas the Southern ­Italo-­Romance facts analyzed in sections 5 and 6, conversely, are most economically accounted for by assuming that metaphonic raising occurred first, to then yield to metaphonic diphthongization in a substantial subset of the dialects of Southern Italy, exemplified by (Old) Neapolitan in (5)/(14).

References Abete, Giovanni. 2006. Sulla questione della sillaba superpesante: i dittonghi discendenti in sillaba chiusa nel dialetto di Pozzuoli. In Renata Savy & Claudia Crocco (eds.), Analisi prosodica. Teorie, modelli, sistemi di annotazione (Atti del 2° Convegno Nazionale AISV), ­379–­398. Torriana: EDK Editore. AIS: Karl Jaberg & Jakob Jud. ­1928–­1940. ­Sprach-­und Sachatlas Italiens und der Südschweiz. 8 vols. Zofingen: Ringier. Avolio, Francesco. 1996. Il ‘neutro di materia’ nei dialetti ­centro-­meridionali: fonti, dati recenti, problemi aperti. Contributi di Filologia dell’Italia Mediana 10: ­291–­337. Avram, Andrei. 2005. Metafonie și fenomenele conexe în limba română. București: Editura Academiei Române. Barbato, Marcello. 2000. Ancora sui frammenti volgari di Agnello di Gaeta. Medioevo romanzo 24: ­110–­113. Barbato, Marcello. 2002. La formazione dello spazio linguistico campano. Bollettino linguistico campano 2: ­29–­64. Barbato, Marcello. 2008. Metafonia napoletana e metafonia sabina. In Alessandro De Angelis (ed.), I dialetti italiani meridionali tra arcaismo e interferenza, ­275–­289. Palermo: CSFLS. Barbato, Marcello. 2013. La métaphonie romane occidentale. Revue de linguistique romane 77: ­321–­341. Bartoli, Matteo. 1943. Lineamenti di linguistica spaziale. Milan: Le Lingue Estere. Bermúdez-Otero, Ricardo. 2007. Diachronic phonology. In Paul de Lacy (ed.), The Cambridge Handbook of Phonology, 497–517. Cambridge: Cambridge University Press. Bourciez, Édouard. 1937. Précis de phonétique française. 8th edn. Paris: Klincksieck. Calabrese, Andrea. ­1984–­85. Metaphony in Salentino. Rivista di Grammatica Generativa ­9–­10: ­3–­140. Calabrese, Andrea. 1998. Metaphony revisited. Rivista di Linguistica 10: ­7–­68. Camilli, Amerindo. 1929. Il dialetto di Servigliano. Archivum Romanicum 13: ­220–­271.

84 

 Michele Loporcaro

Castellani, Arrigo. 1961. Sulla formazione del tipo fonetico italiano. Fenomeni vocalici. Studi linguistici italiani 2: ­24–­45 [repr. in Castellani (1980,I: 73–­95)]. Castellani, Arrigo. 1962. Quelques remarques à propos de la diphtongaison toscane. Réponse à M. Schürr. ZRPh 78: ­494–­502 [repr. in Castellani (1980,I: 139–­145)]. Castellani, Arrigo. 1970a. Dittongamento senese e dittongamento aretino nei dialetti dell’Italia mediana (in epoca antica). In I dialetti dell’Italia mediana con particolare riguardo alla regione umbra. Atti del V Convegno di studi umbri (Gubbio, 28 maggio – 1 giugno 1967). Gubbio: Centro di Studi Umbri presso la Casa di Sant’Ubaldo – Perugia: Ed. Fac. di Lettere e filosofia, ­311–­380 [repr. in Castellani (1980,I: 358–­422)]. Castellani, Arrigo. 1970b. Note sul dittongamento toscano. In Mille: i dibattiti del Circolo ­linguistico fiorentino 1945–1970. Firenze: Olschki, ­41–­53 [repr. in Castellani (1980,I: 146–­155)]. Castellani, Arrigo. 1970c. Ancora sul dittongamento italiano e romanzo. Seconda risposta a Friedrich Schürr. CN 30: ­117–­130 [repr. in Castellani (1980,I: 156–­176)]. Castellani, Arrigo. 1973. I più antichi testi italiani. Edizione e commento. Bologna: Pàtron. Castellani, Arrigo. 1980. Saggi di linguistica e filologia italiana e romanza (­1946–­1976). 3 voll. Rome: Salerno. Castellani, Arrigo. 2000. Grammatica storica della lingua italiana. I. Introduzione. Bologna: Il Mulino. Coseriu, Eugenio. 1973. Sincronía, diacronía e historia. El problema del cambio lingüístico. Segunda edición, revisada y corregida. Madrid: Gredos [first edn. Montevideo: Universidad de la República 1958]. Del Puente, Patrizia & Franco Fanciullo. 2004. Per una “Campania dialettale”. In Franco Fanciullo. Dialetti e non solo, ­149–­175. Alessandria: Edizioni dell’Orso. Dressler, Wolfgang U. 1980. A semiotic model of diachronic process phonology. In Winfred P. Lehmann & Yakov Malkiel (eds.), Perspectives on historical linguistics, ­93–­131. Amsterdam – Philadelphia: Benjamins. Dressler, Wolfgang U. 1985. Morphonology: the dynamics of derivation. Ann Arbor: Karoma. Ernst, Gerhard. 1970. Die Toskanisierung des römischen Dialekts im 15. und 16. Jahrhundert, Tübingen: Niemeyer (Beiheft 121 ZRPh). See Richter. Fanciullo, Franco. 1994. ­Morfo-­metafonia. In Palmira Cipriano, Paolo Di Giovine & Marco Mancini (eds.), Miscellanea di studi linguistici in onore di Walter Belardi, ­571–­592. Rome: Il Calamo. Ferguson, Thaddeus. 1976. A history of the Romance vowel systems through paradigmatic reconstruction. The Hague: Mouton. Formentin, Vittorio, ed. 1998. Loise de Rosa, Ricordi. 2 vols. Rome: Salerno. Formentin, Vittorio. 2012. Un nuovo testo per la storia del romanesco medievale. In Michele Loporcaro, Vincenzo Faraoni & Piero Adolfo Di Pretoro (eds.), ­29–­78. Vicende storiche della lingua di Roma, Alessandria: Edizioni dell’Orso. Formentin, Vittorio. 2013. A proposito di romanesco antico: la metafonia nel registro di Giovanni Cenci. Lingua e Stile 48: ­299–­316. Grimaldi, Mirko. 2003. Nuove ricerche sul vocalismo tonico del Salento meridionale. Analisi acustica e trattamento fonologico dei dati. Alessandria: Edizioni dell’Orso. Herman, József. 1970. Les particularités de l’évolution du latin provincial. In Actele celui ­de-­al ­XII-­lea Congres Internaţional de Lingvistică şi Filologie Romanică. Bucureşti: Editura Academiei, vol. 1, ­125–­130 [repr. in Herman (1990: 29–­34)].



Metaphony and diphthongization in Southern Italy 

 85

Herman, József. 1971. Essai sur la latinité du littoral adriatique à l’époque de l’Empire. In Sprache und Geschichte. Festschrift zum 65. Geburtstag von Harri Maier. Munich: Fink, ­199–­226 [repr. in Herman (1990: 121–­146)]. Herman, József. 1985. La différenciation territoriale du latin et la formation des langues romanes. In Actes du XVIIe Congrès International de Linguistique et Philologie Romanes. ­Aix-­­en-­Provence, 29 ­août-­3 septembre 1983. ­Aix-­­en-­Provence: Université de Provence. Vol. 2, ­15–­62 [repr. in Herman (1990: 62–­92)]. Herman, József. 1990. Du latin aux langues romanes. Etudes de linguistique historique. Tübingen: Max Niemeyer. Joseph, Brian D. & Richard D. Janda (eds.), 2003. The Handbook of Historical Linguistics. Malden, ­MA-­Oxford, UK: Blackwell. Kiparsky, Paul. 1995. The phonological basis of sound change. In John Goldsmith (ed.), The Handbook of Phonological Theory, ­640–­670 [also in Joseph & Janda (2003: 313–­342)]. Oxford: Blackwell. Lambrior, Alexandru. 1878. L’e bref latin en roumain. Romania 7: ­85–­93. Lausberg, Heinrich. 1947. Zum romanischen Vokalismus. Romanische Forschungen 60: ­295–­307. Lausberg, Heinrich. 1976. Linguistica romanza. I. Fonetica. 2nd edn. Milan: Feltrinelli. Loporcaro, Michele. 2003. Dialettologia, linguistica storica e riflessione grammaticale nella romanistica del Duemila. Con esempi dal sardo. In Fernando Sánchez Miret (ed.), Actas del XXIII CILFR. Salamanca, ­24–­30 September 2001. Tübingen: Niemeyer, vol. 1, ­83–­111. Loporcaro, Michele. 2011a. Morphological complexity without abstractness: ­Italo-­Romance metaphony. Poznań studies in contemporary linguistics 47.2: ­177–­202. Loporcaro, Michele. 2011b. Phonological processes. In Maiden et al. (2011: 109–­154, ­689–­698). Loporcaro, Michele. 2011c. Syllable, segment and prosody. In Maiden et al. (2011: 50–­108, ­684–­689). Loporcaro, Michele. 2013. Profilo linguistico dei dialetti italiani. 2nd edn. ­Rome-­Bari: Laterza. Loporcaro, Michele, Nadia Nocchi, Tania Paciaroni & Michael Schwarzenbach. 2007. Dittongazione e metafonia nel dialetto di Agnone (IS). Paper read at the 4th Conference of the Associazione Italiana di Scienze della Voce, Cosenza, December ­3–­5, 2007. Lüdtke, Helmut. 1956. Die strukturelle Entwicklung des romanischen Vokalismus. Bonn: Romanisches Seminar an der Universität. Maddieson, Ian. 1997. Phonetic universals. In William J. Hardcastle & John Laver (eds.), The handbook of phonetic sciences, ­619–­639. Oxford: Blackwell. Maiden, Martin. 1989. Sulla morfologizzazione della metafonesi nei dialetti italiani meridionali. ZRPh 105: ­178–­192. Maiden, Martin. 1991. Interactive morphonology: Metaphony in Italy. London, New York: Routledge. Maiden, Martin. 1995. A Linguistic History of Italian. New York: Longman [Italian translation Storia linguistica dell’italiano. Bologna: Il Mulino 1998]. Maiden, Martin. To appear. Italo-Romance metaphony and the Tuscan diphthongs. To appear in the Transactions of the Philological Society. Maiden, Martin, John Charles Smith & Adam Ledgeway (eds.), 2011. The Cambridge History of the Romance Languages. Vol. 1. Structures. Cambridge: Cambridge University Press. Mateus, M. Helena & Ernesto d’Andrade. 2000. The Phonology of Portuguese. Oxford: OUP.

86 

 Michele Loporcaro

Maturi, Pietro. 2002. Dialetti e substandardizzazione nel Sannio beneventano. Frankfurt am Main: Lang. Mengel, Erich. 1936. Umlaut und Diphthongierung in der Dialekten des Picenums. PhD dissertation, Köln. Mussafia, Adolfo. 1868. Zur rumänischen Vokalisation. Sitzungsberichte der Wiener Akademie der Wissenschaften 58: ­125–­154. Nandriş, Octave. 1963. Phonétique historique du roumain. Paris: Klincksieck. Nübling, Damaris. 1999. Wie die Alten sungen … Zur Rolle von Frequenz und Allomorphie beim präteritalen Numerusausgleich im Frühneuhochdeutschen. Zeitschrift für Sprachwissenschaft 17.2: ­185–­203. Ohala, John J. 1986. Consumer’s guide to evidence in phonology. Phonology Yearbook 3: ­3–­26. Ohala, John J. 1989. Sound change is drawn from a pool of synchronic variation. In L. Egil Breivik & Jahr, Ernst Håkon (eds.), Language change. Contributions to the study of its causes, ­173–­198. Berlin – New York: Mouton de Gruyter. Ohala, John J. 2003. Phonetics and Historical Phonology, in Joseph & Janda (2003: 669–­686). Ohala, John J. 2005. Phonetic explanations for sound patterns. Implications for grammars of competence. In William J. Hardcastle & Janet Mackenzie Beck (eds.), A figure of speech. A festschrift for John Laver, ­23–­38. London: Erlbaum. Paiella, Anna Rita. 1973. Le parlate del Basso Velino. Rieti 1: ­413–­443. Parlangeli, Oronzo. 1953. Sui dialetti romanzi e romaici del Salento. Memorie dell’Istituto lombardo di scienze e lettere. Classe di Lettere, Scienze morali e storiche ­25–­26 (3rd Series): ­93–­200. Parlangeli, Oronzo. 1960. Storia linguistica e storia politica nell’Italia meridionale. Firenze: Le Monnier. Parodi, Ernesto G. 1892. Il dialetto d’Arpino. AGI 13: ­299–­308. Pellegrini, Giovan Battista. 1977. Carta dei dialetti d’Italia. Pisa: Pacini. ­Pompino-­Marschall, Bernd. 2010. Einführung in die Phonetik. In Ludger Hoffmann (ed.), Sprachwissenschaft. Ein Reader, ­405–­430. Berlin: De Gruyter. Richter, Elise. 1934. Beiträge zur Geschichte der Romanismen, I. Chronologische Phonetik des Französischen bis zum Ende des 8. Jahrhunderts. Halle a.S.: Max Niemeyer (Beiheft 82 ZRPh). Rohlfs, Gerhard. 1938. Der Einfluß des Satzakzentes auf den Lautwandel. ASNS 174: ­54–­56. Rohlfs, Gerhard. ­1966–­69. Grammatica storica della lingua italiana e dei suoi dialetti. 3 vols. Turin: Einaudi. Rosetti, Alexandru. 1986. Istoria limbii române. I. De la Origini pâna la începutul secolului al ­XVII-­lea. Ediție definitivă. București: Editura științifică și enciclopedică. Sala, Marius. 1976. Contributions à la phonétique historique du roumain. Paris: Klincksieck. Sala, Marius. 2004. Dal latino al romeno. Alessandria: Edizioni dell’Orso. Sánchez Miret, Fernando. 1998. La diptongación en las lenguas románicas. München – Newcastle: LINCOM Europa. Sánchez Miret, Fernando. 2007. El papel de la fonética en la explicación de los cambios fonológicos dentro de la gramática histórica de las lenguas románicas. In Alexandra Cuniță, Coman Lupu & Liliane Tasmowski (eds.), ­484–­493. Studii de Lingvistică şi Filologie Romanică. Hommages offerts à Sandra Reinheimer Rîpeanu, Bucureşti: Editura Universității. Sánchez Miret, Fernando. 2010. The effect of word final unstressed high vowels on stressed vowel duration and its consequences for metaphonic diphthongization in Southern Italian. In Daniel Recasens, Fernando Sánchez Miret & Kenneth J. Wireback (eds.), 83–­97. ­Experimental phonetics and sound change. München: Lincom Europa. ­



Metaphony and diphthongization in Southern Italy 

 87

Schürr, Friedrich. 1936. Umlaut und Diphthongierung in der Romania. Romanische Forschungen 50: ­275–­316. Schürr, Friedrich. 1964. Rumänische und romanische Diphthongierung. Acta philologica (Societas Academica Dacoromana) 3: ­351–­360. Schürr, Friedrich. 1970. La diphthongaison romane. Tübingen: Narr. Straka, Georges. 1953. Observations sur la chronologie et les dates de quelques modifications phonétiques en roman et en français prélittéraire. Revue des Langues Romanes 71: ­247–­307. Tucci, Maria. ­1996–­97. Il dialetto di Casabona. Studio ­fonetico-­fonologico con un’analisi sperimentale sul vocalismo tonico. Tesi di laurea, Università della Calabria. Tuttle, Edward F. 1985. Morphologization as redundancy in central Italian dialects. RPh 39: ­35–­43. Vasiliu, Emanuel. 1968. Fonologia istorică a dialectilor dacoromâne. Bucureşti: Editura Academiei. Vignuzzi, Ugo. 1994. Il volgare nell’Italia mediana. In Luca Serianni & Pietro Trifone (eds.), Storia della lingua italiana. Vol. 3. Le altre lingue, ­329–­372. Turin: Einaudi. Vignuzzi, Ugo. 2005. La lingua e i dialetti. In Lazio. Guida d’Italia del Touring Club Italiano. Milano: Touring Club Italiano, ­83–­89. Vignuzzi, Ugo & Francesco Avolio. 1994. Per un profilo di storia linguistica “interna” dei dialetti del Mezzogiorno d’Italia. In Giuseppe Galasso & Rosario Romeo (eds.), Storia del Mezzogiorno. Vol. 9. Aspetti e problemi del Medioevo e dell’età moderna, ­631–­699. Rome: Editalia. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language and Linguistic Theory 23: ­917–­989. Wartburg, Walter von. 1950. Die Ausgliederung der romanischen Sprachräume. Bern: Francke. Williams, Edwin B. 1962. From Latin to Portuguese: historical phonology and morphology of the Portuguese language. 2nd edn. Philadelphia: University of Pennsylvania Press. Wright, Roger. 1982. Late Latin and Early Romance in Spain and Carolingian France. Liverpool: Francis Cairns. Ziccardi, Giovanni. 1910. Il dialetto di Agnone. La fonetica e la flessione. Zeitschrift für romanische Philologie 34: ­405–­436. Zink, Gaston. 19996. Phonétique historique du français. Paris: PUF.

Andrea Calabrese

On the morphophonology of metaphonic alternations in Altamurano*

Università degli Studi Firenze

Abstract: The focus of this paper is morphophonological metaphony and deals with one of the dialects displaying it, Altamurano. I will argue that this process is simply and adequately accounted for by morphophonological rules operating on single basic (and abstract) forms of morpheme exponents, as in Chomsky and Halle, and more recently in Distributed Morphology, and show that alternative analyses involving abstract representational elements, such as floating features, or memorized (suppletive) stem alternants are unable to do so.

1 Introduction Many Italian dialects are characterized, at least historically, by a process in which stressed stem vowels are raised, diphthongized, or changed in other ways, before final high vowels. This process is traditionally referred to as metaphony. In a large subgroup of these dialects, the final vowels, including the high vowels that triggered this phenomenon, were reduced to schwas or deleted. Final high vowels were the exponents of morphological categories such as 2sg forms of verbs, the masculine of nouns/adjectives of the II class, the plural of nouns/adjectives of the III class, etc. Once they were lost or reduced, the metaphonic changes became associated with the morphological categories they used to represent. Metaphony thus became a morphophonological (MP) process, i.e. a morphologically conditioned phonological process. The focus of this paper is MP metaphony and deals with one of the dialects displaying it, Altamurano, the Apulian dialect of the town of Altamura, which is carefully described and discussed in Loporcaro (1988).1

* I would like to thank Paola Beninca, Jonathan Bobaljik, Eugene Buckley, David Embick, Rolf Noyer, Laura Vanelli, an anonymous reviewer, and audiences at the University of Connecticut, at the International Ferrara Phonology Meeting, at the University of Pennsylvania, and at the University of Padova, where different versions of this paper were presented, for their comments and suggestions. 1 More or less similar data can be found in any other Apulian dialects (Stehl 1980; Valente 1975), and also in most upper southern dialects of the Adriatic coast such as Abruzzese and Molisano (Gianmarco 1979). Altamurano, and all of these dialects are in the same way characterized by metaphony, reduction or deletion of final vowels, and widespread changes in the stressed vowel system.

90 

 Andrea Calabrese

In recent times, there has been a resurgence of interest in MP processes (see Bermúdez-Otero 2012, 2013; Bonet & Harbour 2012; Bye 2007; Bye & Svenonius 2012; Paster 2006; Mascaró 2007, a.o.). Leaving aside Cophonology Theory (Inkelas & Zoll 2007) and morpheme-indexed constraints (Pater 2000, 2010) (see Bermúdez-Otero 2012 for a critical assessment), MP processes have been accounted for by resorting to abstract representational elements, such as floating features (Lieber 1992; Akinlabi 1996; Zoll 1998) or partially specified segments (de Lacy 2006), or to memorized stem alternants (suppletion) (see Bye & Svenonius 2012 and Bermúdez-Otero 2012 for recent examples of such accounts). In this paper, I will argue that abstract representational elements2 and memorized (suppletive) stem alternants are unable to deal with MP metaphony in Altamurano, and that, instead, this process is simply and adequately accounted for by MP rules operating on a single basic (and abstract) form of the morpheme exponent, as in Chomsky & Halle (1968) and more recently in Distributed ­Morphology (DM) (Halle & Marantz 1993). The paper is organized as follows. In section 2, after describing the main properties of the Altamurano vowel system, I discuss its metaphony alternations focusing on the verbal system. In section 3, I will provide an account for them in a DM framework by using MP operations. In section 4, I will consider possible alternative analyses not using such operations: firstly, one based on floating feature (section 4.1) and then one based on suppletive stem forms (4.2). I will argue that both are untenable. Section 5 discusses metaphony in nouns and adjectives. A brief conclusion ends the paper.

2 Metaphony in Altamurano: facts I will begin by focusing on the relatively simple metaphonic alternations found in the Altamurano verbal system; the more complex metaphonic alternations of the nominal and adjectival systems are addressed in section 5. In the verbal system, stressed mid vowels are regularly affected by metaphonic changes in the 2sg singular. This occurs systematically in all tenses3 and with all verbs. The trigger of

2 I will be arguing against floating features in particular, but my arguments also hold for all other types of abstract representational elements. 3 Altamurano has a reduced tense morphology—I assume that the single tense morpheme, as in Italian, includes Aspect and Mood features, (see Calabrese 2015a). It has a present indicative, an imperfect indicative and subjunctive, and a perfect. Metaphonic alternations are found in the present (see 1), in the imperfect indicative (mandƷay / mandƷiːv ‘eat-ipf.ind-3sg / 2sg’), and in the imperfect subjunctive (mandƷɛss / mandƷiss ‘eat-ipf.sbjv-3sg / 2sg’). In the perfect, stress falls on the thematic vowel, which is underlyingly high or low in this tense, and, therefore, does not undergo metaphony.



On the morphophonology of metaphonic alternations in Altamurano 

 91

the process is purely morphological; however, the target is phonological, a mid stressed vowel. In (1), I give sample present forms (Loporcaro 1988).4,5 (1)

a)    underlying /i/    1sg 2sg 3sg 1pl 2pl 3pl Gloss   ˈdiyk ˈdiy ʃ ˈdiy ʃ ˈdikǝnǝ ‘say’ b)   underlying /e/:   ˈley ʃ ˈliː ʃ ˈley ʃ ˈle ʃǝnǝ ‘read’   ˈveŋŋ ˈvinn6 ˈveyn ˈvenǝnǝ ‘come’   ˈspetts ˈspitts7 ˈspetts ˈspettsǝnǝ ‘break’ c)    underlying /ɛ/:   ˈkrɛyt ˈkriyt ˈkrɛyt ˈkrɛtǝnǝ ‘believe’   ˈvɛyf ˈviyf ˈvɛyf vǝˈviym vǝˈviyt ˈvɛvǝnǝ ‘drink’   ˈmɛtt ˈmitt ˈmɛtt ˈmɛttǝnǝ ‘put’ d)   underlying /a/:   ˈkyɛym ˈkyɛym ˈkyɛym ˈkyamǝnǝ ‘call’ e)    underlying /ɔ/:   aˈdɔwr aˈduwr aˈdɔwr aˈdɔrǝnǝ ‘smell’   ˈɲɲɔtt ˈɲɲütt ˈɲɲɔtt ˈɲɲɔttǝnǝ ‘swallow’   asˈkɔnn asˈkwunn asˈkɔnn asˈkɔnnǝnǝ ‘hide’ f)     underlying /o/:8   ˈ ʃowk ˈ ʃeyk9 ˈʃowk  ʃuˈkwɛym  ʃuˈkwɛyt ˈʃøkǝnǝ ‘play’   ˈmowr ˈmweyr ˈmowr ˈmørǝnǝ ‘die’   arrǝˈkørd arrǝˈkwerd arrǝˈkørd arrǝˈkørdǝnǝ ‘play’

4 These are citation forms as given in Loporcaro (1988). As discussed later, citation forms, insofar as they occur prepausally, undergo post-tonic vowel deletion in forms with penultimate stress and vowel diphthongization in open penultimate syllables. There is also fronting of [+ATR] back vowels in closed or antepenultimate stressed syllable (see below). 5 Loporcaro (1988) provides 1pl and 2pl present sample forms only for a few verbs. 6 The form to be expected here is [viːn] with a long [iː]. This long vowel, however, has a special behavior: when it is in an underlying open syllable and is followed by a sonorant consonant, it shortens, triggering gemination of the following sonorant. Thus, we have the following derivation: underlying /ven-ǝ/ → metaphony → [ˈviːnǝ] → gemination → [ˈvinnǝ] → posttonic deletion → [ˈvinn]. Furthermore, in underlying closed syllables and in antepenultimate position, this long vowel always shortens without causing any change in the consonantal environment. Hence, underlying /spetts-ǝ/ → metaphony → [ˈspiːttsǝ] → shortening → [ˈspittsǝ] → posttonic deletion → [ˈspitts]. See discussion in Loporcaro (1988: 37). 7 See note 6. 8 In the forms [ ʃuˈkwɛym], [ ʃuˈkwɛyt the velar stop acquires a secondary labial articulation due to an independent assimilation process triggered by the preceding high round vowel. 9 The labiovelar glide [w] is deleted after coronal consonants and sometimes in word-initial position. In the case of surface [ ʃeyk], we have the following derivation: /ʃok-ǝ/ → metaphony → [ˈʃwekǝ] → diphthongization → [ˈʃweykǝ] w-deletion → [ˈʃeykǝ] → posttonic deletion → [ˈʃeyk].

92 

 Andrea Calabrese

g)    underlying /u/:   anˈnuwk anˈnuw ʃ ˈannuw ʃ annuˈ ʃɛym

anˈnüʃǝnǝ ‘bring’

To understand the alternations in (1), as already mentioned in footnote 4, we need to consider some of the many phonological processes affecting vowels in Altamurano. First of all, Loporcaro shows that there is a process deleting post-tonic vowels, which can only be schwas,10 at the end of the phonological phrase, i.e. prepausally (see 2). For example, this process deletes the post-tonic word-final vowel in the second component of the elative constructions in (3) and in the noun [kɛsǝ] when at the end of the sentence in (4b). This process may be blocked in words with antepenultimate stress (see 5) although the conditions for this blocking are not clear. In many cases, deletion would create an illicit consonantal cluster, but not always, e.g. the licit cluster rk that would result from schwa deletion in ˈsɔrǝkǝ (see discussion of this point in Loporcaro 1988: 166–68). (2) ǝ (3)

 →

[* Ø/ σ

[* = +stress

##

ˈkjinǝ ˈkjijn## ‘elative form of ˈsulǝ ˈsuwl## ˈnevǝ ˈnejv## ˈkjenǝ ˈkjejn##

(4) a. je la ˈkɛsǝ de ˈfratt## is the house of your brother b. mǝ nǝ ˈwɔk a ˈkeys## me from-here go to house (5) ˈdƷüvǝtǝ## ˈfɛmǝnǝ## ˈʃɔʃǝlǝ## ˈsɔrǝkǝ##

full-m.sg (full full)’ ‘only-m.sg (alone alone)’ ‘new-m.sg (new new)’ ‘slow-m.sg (slow slow)’

‘it is the house of your brother’ ‘I am going home’

‘elbow’ ‘female’ ‘type of fruit’ ‘mouse’

All examples, from now on, with the exception of section 5, will include the final schwa.11 Another process is diphthongization, which affects stressed vowels—see the underlying vowel system in (6)—in word penultimate position when at the end of the phonological phrase. Sample forms are given in (7):12

10 See Loporcaro (1988: 63–66) for some exceptions. 11 So they will be different in surface shape from those found in Loporcaro (1988) who uses mostly citation forms, which being prepausal undergo vowel deletion in (2). 12 Short stressed high vowels, i.e. stressed high vowels in closed or antepenultimate syllable, are [−ATR] in surface representations because of late phonetic process (see Loporcaro 1988 and Calabrese 1998 for discussion). I will neglect the effects of this phonetic process in the representations used in this paper.

On the morphophonology of metaphonic alternations in Altamurano 



(6) i e ɛ

a

(7) i: e: ɛ: a: ɔ: o: u:

ˈdiyʃǝ / ˈdikǝnǝ ˈleyʃǝ / ˈleʃǝnǝ ˈkrɛytǝ / ˈkrɛtǝnǝ ˈkyɛymǝ / ˈkyamǝnǝ aˈdowrǝ / aˈdɔrǝnǝ ˈʃowkǝ / ˈʃØkǝnǝ anˈnuwʃǝ / anˈnüʃǝnǝ

 93

u o ɔ ‘say-3sg/pl’ ‘read-3sg/pl’ ‘believe-3sg/pl’ ‘call-3sg/pl’ ‘smell-3sg/pl’ ‘play-3sg/pl’ ‘smell-3sg/pl’

In Calabrese (2000), I proposed that behind the process of diphthongization in Altamurano, there is the rule in (8), which lengthens stressed vowels in the penultimate syllable of the phonological phrase. This process is followed by diphthongization process as in (9).13,14 (8)

N X

[*

N → X

[–cons]

X

/

s ##

[–cons]

(9) R N X

X

[−cons] [+high]

13 This lengthening in an open penultimate syllable is due to constraints on foot structure holding in most Italo-Romance varieties (see Krämer 2009 for further discussion). The nature of this lengthening is irrelevant here, and I will not discuss this issue further. 14 In some words, [ɛ, ɔ] diphtongize into [ay, aw], respectively, instead of [ɛy, ɔw]. Loporcaro argues that these outcomes are representative of an older stage of the language in which only these vowels underwent diphthongization (into [ay, aw]). According to Loporcaro, these older diphthongs are undergoing a process of monophthongization (into ɛ, ɔ, respectively), which however has not ­applied to all words in the lexicon. Thus, there are still forms such as kraytǝ / krɛtǝnǝ ‛see 3sg/ pl’, paysǝ / pɛsǝnǝ ‘weigh-3sg/pl. The diphthongs [ay, aw] behave differently from the other diphthongs insofar as they can appear in non-prepausal position. This can be accounted for by assuming that, in the relevant words, the lengthening and diphthongization rules in (8) and (9) can apply to [ɛ, ɔ] at the phonological word level. A subsequent process lowers their nucleus at this level.

94 

 Andrea Calabrese

Citation forms always show the effect of final vowel deletion and diphthongization insofar as they appear prepausally. Vowel /a/ is fronted in stressed open penultimate syllable at the phonological word level. The outcome of this process, then, undergoes diphthongization (a → ɛ → ɛy): (10) ˈkyɛymǝ / ˈkyamǝnǝ ‘call-3sg/pl’ ˈkɛysǝ / ˈkasǝrǝ ‘house-sg/pl’ ˈfrɛytǝ / ˈfratǝmǝ ‘brother / brother+1sgPoss. also: ˈrannǝ ‘big’, ˈkyattsǝ ‘square’, ˈpaggyǝ ‘straw’ Stressed [+ATR] back vowels are fronted (cf. 11a vs. 11b) except when they are diphthongized (uw)—namely, when they are in closed or antepenultimate syllables—or when they are preceded by a homorganic secondary articulation (kwu)15. See Calabrese (2000, 2015b) for further discussion of this process: (11) a. [+ATR] back vowels ˈmowrǝ / ˈmØrǝnǝ ‘die-3sg/pl’ ˈʃowkǝ / ˈʃØkǝnǝ ‘play-3sg/pl’ anˈnuwʃǝ / anˈnüʃǝnǝ ‘bring-3sg/pl’ arrǝˈkwerdǝ / arrǝˈkØrdǝ ‘remember-3sg/pl’ see also: ˈtünnǝ ‘round’, ˈfrüttǝ ‘fruit’, ˈØmmǝ ‘man’, ˈlØɲɲǝ ‘long’ b. [−ATR] back vowels aˈdɔwrǝ / aˈdɔrǝnǝ ‘smell-3sg/pl’ ˈɲɲɔttǝ / ˈɲɲɔttǝnǝ ‘swallow-3sg/pl’ (12) [* N X [+ATR, +back]   →   [−back]/    Finally, vowels are raised or reduced to schwas in unstressed syllables. (13) ˈʃowkǝ / ʃuˈkwɛymǝ ˈvɛyfǝ / vǝˈviymǝ

‘play-1sg/pl’ ‘drink-1sg/pl’

We can see the activity of the just discussed phonological forms in the case of the verbal paradigm in (14): (14) 1sg 2sg 3sg 1pl 2pl 3pl Gloss ˈʃowkǝ ˈʃ weykǝ ˈʃowkǝ ʃuˈkwɛymǝ ʃuˈkwɛytǝ ˈʃØkǝnǝ   ‘play’

15 Calabrese (2000) accounts for these conditions on fronting by saying that this process does not apply to vowels in a linked configuration.



On the morphophonology of metaphonic alternations in Altamurano 

 95

The forms in (14) can be derived as follows: (15) a. ˈʃokǝ → Lengthening (L) → ˈʃoːkǝ → Diphthongization (D) → ˈʃowkǝ b. ʃoˈkɛmǝ → L→ ʃoˈkɛːmǝ → D → ʃoˈkɛymǝ → Reduction → ʃuˈkɛymǝ → Other rules → ʃuˈkwɛymǝ c. ˈʃokǝ → Metaphony (see below) → ˈʃwekǝ → L → ˈʃweːkǝ → D → ˈʃweykǝ d. ˈʃokɛnǝ → vowel fronting → ˈʃØkɛnǝ → Other rules → ˈʃØkǝnǝ   As discussed above, diphthongization and fronting, as well as post-tonic schwa deletion, are regular phonological processes in the dialect of Altamura. If we parse away their effect, we can closely examine the metaphonic alternations in this language: (16) a. General allomorph b. 2sg underlying /e/: ˈleʃǝ ˈliːʃǝ ˈvenǝ ˈviːnǝ ˈspettʃǝ ˈspiːttsǝ underlying /ɛ/: ˈkrɛtǝ ˈkritǝ ˈvɛvǝ ˈvifǝ ˈmɛttǝ ˈmittǝ underlying /ɔ/: aˈdɔrǝ aˈdurǝ ˈɲɲɔttǝ ˈɲɲuttǝ asˈkɔnnǝ asˈkunnǝ underlying /o/: ˈʃokǝ ˈʃwekǝ ˈmorǝ ˈmwerǝ arrǝˈkordǝ arrǝˈkwerdǝ

‘read’ ‘come’ ‘break’ ‘believe’ ‘drink’ ‘put’ ‘smell’ ‘swallow’ ‘hide’ ‘play’ ‘die’ ‘play’

We have two basic stem allomorphs: one appearing in the 2sg and the other for the other persons. Following standard phonological practice, we can assume that the basic allomorph of the verb is the one with the widest distribution, therefore the one in (16a). The allomorph with the more restricted distribution can be derived by the changes in (17) with respect to the basic one: (17) Stressed [+ATR] [−ATR] [+ATR] [−ATR]

General allomorph e ɛ o ɔ

These changes are due to metaphony.

2sg iː i we u

96 

 Andrea Calabrese

Before going on, let us consider the diachronic development of the meta­ phonic alternations in Altamurano. Comparative evidence from other dialects (cf. Loporcaro 1988; Maiden 1991; Calabrese 2011) shows that these alternations developed from a traditional “Neapolitan” type system of metaphony. In such a system, stressed mid vowels undergo the changes in (18) before high vowels (in (18), only vowel and not consonant shape is reconstructed): (18) Historical metaphony 3sg 2sg +ATR mid vowels ˈkrete ˈkriti ‘believe’ aˈdore aˈduri ‘smell’ −ATR mid vowels ˈvɛne ˈvyeni ‘come’ ˈʃɔke ˈʃwoki ‘play’ As one can see in (18), [+ATR] mid vowels are raised to [+ATR] high vowels, and [−ATR] mid vowels are diphthongized. An analysis of these alternations is provided in Calabrese (1985). The main concern of this paper was analyzing them in terms of maximal simplicity and generality. Proposing two different metaphony processes, one of raising applying to stressed [+ATR] mid vowels and one of diphthongization applying to stressed [−ATR] mid vowels, meant failing to recognize that these two processes applied in exactly the same environment, and therefore overlapped strikingly in their structural description. To get out of this theoretical redundancy, I proposed that there was a single, unitary process behind them: specifically, the same rule in (19) raised all stressed mid vowels regardless of their [ATR] specification: (19)

[* N X

X

−cons

−cons

[−high]

[−low] [+high]

This rule directly raises mid [+ATR] vowels into high [+ATR] ones and, thus, accounts for the behavior of these vowels. However, in the case of the mid [−ATR] vowels, this rule creates high [−ATR] vowels. Allowing the creation of this outcome was central for my analysis. Although these segments can be found as surface variants of the high

On the morphophonology of metaphonic alternations in Altamurano 



 97

vowels in some dialects (see Loporcaro 1991 and Calabrese 2011), they are consistently not phonemic in all Italian dialects. Adopting a variant of Kiparsky’s (1985) structure preservation, I assumed that they were disallowed by the constraint *[+high, −ATR]16 and proposed that this constraint, instead of blocking the generation of these illicit segments as in the original version of Kiparsky’s structure preservation, triggered a repair adjusting them. I hypothesized that the repair that occurred in this case was similar to the one that often adjusts the German or French front rounded vowel ü in the non-native pronunciation of these languages by changing it into the diphthong [iu] with the sequencing of fronting and rounding (cf. Trubetzkoy 1969: 135). This repair was eventually called fission in Calabrese (1988) (see Calabrese 1995, 2005 for further discussion of fission and for more examples). The idea was that feature constraints hold only among feature specifications dominated by the same root node, i.e. in the same feature bundle. Fission takes a feature bundle containing an illicit combination of features and breaks it into two different feature bundles, each containing one of them. Thus the violation of the constraint is repaired. Application of the metaphony rule in (19) creates a feature bundle containing the feature specifications [+high, −ATR] in violation of the active constraint *[+high, −ATR] (see 20a). Application of fission to this feature bundle creates the structure in 20b with two feature bundles, one containing the feature [+high] and the other the feature [−ATR] which are then matched with compatible features.17 This accounts for the diphthongization we observe in the case of metaphony of mid [−ATR] vowels. Further changes discussed in Calabrese (1985, 1998) account for the conversion of the high vowel into a glide and for other dialectal outcomes. (20) a.

(=[ɪ, ʊ] (the output of → b.

X [−cons]

[high]

the application of

[−cons]

metaphony to [ɛ, ɔ]))

[–ATR]

X



(=[iɛ, wɔ]18) [−cons]

[+high] [−high] [+ATR] [−ATR]

16 Acoustically, vowels characterized by this configuration are not optimal in so far as tongue root retraction does not enhance the acoustic effects produced by the raised tongue position (see S ­ tevens (1986); Keyser & Kawasaki 1986). High [−ATR] vowels tend to be historically eliminated from vowel systems. At the same time, the presence of [+high, −ATRl vowels in a vowel system implies the presence of [+high, +ATR], [−high, −ATR] and [−high, −ATR] vowels (see Calabrese 1988, 1995). 17 See Calabrese (1995, 2005) for discussion of the linear order of the fissioned bundles. 18 A further change account for the fronting of vowel [o] ([wo] → [we]). This change occurs only in some southern varieties, Altamurano being one of them.

98 

 Andrea Calabrese

Under this analysis, the metaphony rule is unitary. Metaphony in itself does not display any variation. It is a process spreading the high feature of high vowels onto a preceding mid stressed vowel. Diphthongization, instead, results from an independently motivated process. (See Kaze 1989; Maiden 1991; Cole 1998; Walker 2005 for alternative views) (21) Stressed  [−high]  Metaphony   [+high]  Fission (20) [+ATR] e → i [−ATR] ɛ → ɪ → ie [+ATR] o → u [−ATR] ɔ → ʊ → wo (→ [we]19) After discussing the metaphony process that historically underlies the metaphonic alternations in Altamurano, let us consider the other diachronic changes that affected this dialect. A diachronic change that is crucial to the development of the current metaphony system of this dialect is the reduction of post tonic vowels to schwas. Observe that there is no synchronic evidence—no synchronic alternations or other type of evidence—in Altamurano showing that this reduction process is still active synchronically. So there is no way of postulating abstract underlying final vowels synchronically. This historical change simply eliminated morphological contrasts in inflectional endings. In particular, it eliminated evidence for inflectional high vowels, the vowels that triggered metaphony. For example, the exponent of the 2sg subject agreement morpheme was /-i/. This final vowel triggered metaphony of the stressed mid vowels. Once this vowel was neutralized with the other final vowels (for example, with the /-o/ of the 1sg or the /-a/-e/ of the 3sg), the metaphonic change was reanalyzed as being associated with the 2sg. It is important to observe, though, that metaphonic changes have not simply become the morphological index of the 2sg. Metaphony still remains a phonological process targeting stressed mid vowels. Thus, only when the stem stressed vowel is mid, do we have metaphonic alternations. If the stem stressed vowel is low or high, there are no alternations, and therefore there is simply neutralization of morphological distinction: the verbal form [kyamǝ]—after fronting and diphthongization, surface [kyɛymǝ]—is at the same time 1, 2, and 3sg. Further historical changes modified the original metaphonic alternations and generated the synchronic alternations we observe in (16). First of all, there was monophthongization of [ye], the metaphonic outcome of [ɛ], to [iː]. This change involved first a resyllabification of the glide [y] to nuclear status, which was then followed by stress retraction ([ yé] → [íe]). This is a common change across

19 See note 18.



On the morphophonology of metaphonic alternations in Altamurano 

 99

southern Italian dialects. The mid vowel was then changed to schwa ([ie] → [iǝ]) and furthermore assimilated to the preceding vowel ([iǝ] → [iː]). All of these different steps are synchronically found in other Apulian dialects (see Loporcaro 1988). However, 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] [ɛ] and [ɔ] and etymological [−ATR] [ɛ] and [ɔ] became [+ATR] [e] and [o], as in (ii).20 (22) [aATR] → [−aATR] / [ 

, −high, −low]

These historical changes affected Altamurano as shown below: (23) a. Reduction of final vowels
 3sg 2sg +ATR mid vowels ˈkretǝ ˈkritǝ aˈdorǝ aˈdurǝ −ATR mid vowels ˈvɛnǝ ˈvyenǝ ˈʃɔkǝ ˈʃwokǝ

‘believe’ ‘smell’ ‘come’ ‘play’

b. Monophthongization of [ye] to [iː] (and fronting of [o] in [wo]) 
 +ATR mid vowels ˈkretǝ ˈkritǝ ‘believe’ aˈdorǝ aˈdurǝ ‘smell’ −ATR mid vowels ˈvɛnǝ ˈviːnǝ ‘come’ ˈʃɔkǝ ˈʃwekǝ ‘play’

c. [aATR] → [−aATR] / [  , −high, −low]
 −ATR mid vowels ˈkrɛtǝ ˈkritǝ ‘believe’ aˈdɔrǝ aˈdurǝ ‘smell’ +ATR mid vowels ˈvenǝ ˈviːnǝ ‘come’ ˈʃokǝ ˈʃwekǝ ‘play’

It is important to observe at this point that there is no synchronic evidence— no synchronic alternations or other type of evidence—in Altamurano showing that the processes in (23), and specifically (23c), are still active synchronically. Altamurano does not provide any evidence for postulating an abstract phonolo-

20 See Loporcaro (1988) for some speculations on the phonetic nature of this sound change.

100 

 Andrea Calabrese

gical analysis where the different changes we see in (23) apply in a synchronic derivation. These changes cannot be recovered synchronically but only reconstructed historically. Thus, by the changes in (22), the natural bases of the original metaphonic alternations in (18), as analyzed in Calabrese (1985) (see above discussion), were totally obscured synchronically in Altamurano.

 The consequence is that in Altamurano there is no synchronic unitary generalization that allows capturing all the different alternations in terms of a single process: no single operation—even combined with subsequent ones— can account for the changes we observe in metaphony. This will become clear when I discuss the floating feature analysis below in section 4.1. The historical changes in (23) disrupted the original metaphony system in such a way that the metaphonic alternations can be accounted for only in terms of different generalizations. Through these changes, a single phonologically based unitary process as in (19) was transformed into different phonologically unmotivated, morphologically based—the trigger is in fact a purely morphological category— processes. Still, despite all of the disruption brought about by the historical changes, if we consider the effects of metaphony in the verbal morphology, we observe that metaphonic changes are a systematic property of the 2sg of all present tense forms if their stressed vowel is mid. If the verb has a stressed mid vowel, then it is systematically affected by metaphony in the 2sg of the present. Metaphony is fundamentally regular: metaphonic changes apply without exceptions under the relevant conditions. Furthermore, although the metaphonic changes are not phonologically motivated, as mentioned above, they do involve plausible, “natural”, phonological operations. Their structural change is fully phonological: a specific phonological segment is targeted—a stressed mid vowel, and the structural change involve simple phonological operations: lengthening, raising, tensing, diphthongization. The systematic and regular phonological properties of metaphony ­combined with its purely morphological conditioning point to a straightforward MP account.

3 Metaphony in Altamurano: analysis The analysis proposed in this paper is set in the DM framework (Halle & Marantz 1993; Embick & Marantz 2008; Embick 2010; Bobaljik 2012). The theory of DM proposes a piece-based view of word formation, in which the syntax/morphology interface is as transparent as possible.



On the morphophonology of metaphonic alternations in Altamurano 

 101

(24) The grammar (Syntactic Derivation) Morphology PF

LF

DM crucially incorporates hierarchical structure into morphology; essentially, it assumes the input to morphology to be syntactic structure. Features (or feature bundles) are distributed over nodes forming morphemes, which in turn are subject to Vocabulary Insertion (rules that add phonological material (exponents) to these morphemes). Competition between Vocabulary Items (VI)—typically, with ordering by specificity determining the winner—yields one winner and “blocks” lessspecified VI. The remaining part of the information about the phonological form of morphemes is provided by a set of morpho-syntactically conditioned phonological rules (called Readjustment Rules in DM but simply referred to as MP rules here), and by a set of standard phonological rules. It follows that contextual allomorphy in DM is accounted for by the application of: 1) vocabulary insertion rules that add phonological material (exponents) to morphemes formed by morphosyntactic feature bundles; 2) MP rules;21 and 3) phonological rules. I assume that the principle that governs the selection of these analytic options is the one adopted by classical generative analyses since its early days (Kenstowicz & Kisseberth 1979: chapter 6). MP analysis always starts at the lowest level of abstraction. All alternations are, therefore, at first considered to be instances of suppletion. The analysis proceeds forward towards higher levels of abstraction only when gains in linguistically significant generalizations can be achieved. The natural analytic sequence in terms of abstraction is the following: alternations are first analyzed as involving suppletion, i.e. in terms of different VIs. If there is evidence that they can be accounted for in terms of processes, MP rules are postulated unless there is also evidence for a pure phonological analysis, in which case phonological rules are postulated.22

21 This distinction between VI and MP rules parallels the traditional distinction between ­suppletion and conditioned allomorphy on the one hand and morphophonemic alternations (i.e. allomorphs related by a set of morphologically conditioned phonological rules) on the other. 22 Furthermore, although it is not relevant for this paper, I assume that filters, i.e. negative constraints, can be postulated only under further examination of the processes accounting for the alternations, specifically, when there is evidence of a conspiracy among processes, or when a process can be analyze as a repair triggered by a UG marking statement (a UG filter) (see Calabrese (forthcoming) for more discussion of these issues).

102 

 Andrea Calabrese

It is assumed that contextual allomorphy can occur only under certain locality conditions. The locality condition that will be relevant in this paper enforces “inside out” cyclicity (e.g. Halle & Marantz 1993; Bobaljik 2000). For further discussion of locality conditions in DM, see Embick (2010); Bobaljik (2012) and also Calabrese (2015a): (25) Insertion proceeds from the inside out. We can again turn to Metaphony in Altamurano at this point. Altamurano metaphony is a phonological process changing the quality of the stressed mid vowels in certain morphological contexts. In DM terms, it is a MP rule. Assuming the structure for verbal forms in (26), in a model such as DM, we postulate the VI in (28) and (29) for the agreement suffixes and the thematic vowels. The roots have the underlying representations in (27) which contain the vowel found in the general allomorph in the metaphonic alternations. Rule (29a), which deletes the verb thematic vowel before a vowel-initial suffix, is an important MP rule of Altamurano, as of all other Italo-Romance varieties. It applies as in (29b), and its application must precede the application of metaphony. (26) Structure of Altamurano verbal forms (where TV is the thematic vowel, T is tense and AGR the subject agreement morpheme) T AGR

T V  Root

(27) Verbal roots: /leʃ-/ /krɛt-/ /ʃok-/ /adɔr-/ etc.

(T)23 TV ‘read’ ‘believe’ ‘play’ ‘smell’

(28) VI for subject agreeement suffixes: [+participant, +author, +plural] ↔ /-mǝ/ / [+present] [+participant, −author, +plural] ↔ /-tǝ/ [−participant, +plural] ↔ /-nǝ/ [ ] ↔ /-ǝ/ (elsewhere exponent in singular)

23 As proposed below, T is pruned or missing in the present.

On the morphophonology of metaphonic alternations in Altamurano 



(29) TV24

↔

/-i-/ /-ɛ-/

(30) a. V b. ʃ ˈok-ɛ-ǝ

→ →

]TV V Ø/ ʃ ˈoːkǝ ‘play-TV-sg’

 103

/rooti

The task now is to account for the metaphonic alternations in terms of MP rules. As discussed earlier, there is no unitary generalization that allows capturing all the different alternations in terms of a single process, and specifically, in terms of a single operation. As a matter of facts, the historical changes in (23) disrupted the original metaphony system in such a way the metaphonic alternations can be accounted for only by assuming different MP rules. This is what I do in (31) where I propose three MP rules, differentiated by the target vowel, but sharing the same morphological context (/ [AGR, +part, −auth, −plur]). These rules take the processes we observe in the alternations at their face value as involving different phonological processes of lengthening, raising, and diphthongization occurring in a specific morphological context. Notice that this is the least abstract solution for the learner to achieve: it essentially states formally what is occurring in the observed processes. (31a) involves two operations: i. morphologically conditioned one—a MP rule—splits the nucleus into two skeletal positions; and ii. a phonological one the raises the long [eː] which results from the nucleus splitting in i. (31) a. Metaphony Rule I i. [* s R N

/

[AGR, +participant, −author, −plural]

X

X

(followed by automatic spreading of vowel root node (e → eː)

[−cons] [−low] [−back] [+ATR]

[−high]

24 Various idiosyncratic changes affect the thematic /ɛ/ of non-present tenses, for example it diphthongizes as [ay] in the imperfect indicative. A detailed analysis of all the morpho-syntax and morpho-phonology of Altamurano verbal forms is not possible here

104 

 Andrea Calabrese

ii.

[* s R N X

X [−cons]

(eː → iː)

[−low] [−high] [+high] (31b) also involves two operations: i. a morphologically conditioned one—a MP rule—inserts an empty onset position, which is then filled in by the following nucleus; and ii. a subsequent phonological one inserts the feature [−back] in the nuclear part of the diphthong)  b. Metaphony Rule II

i.

[* s R N X

X

/

(followed by automatic spreading of vowel root node (o → wo)

[−cons] [−high]



[−low]

[+back]

[+ATR]

ii.

X

X

[–cons]

[–cons]

[+back]

[AGR, +participant, −author, −plural]

[–back]

[+back]

s R N

(wo → we)



On the morphophonology of metaphonic alternations in Altamurano 

 105

(31c) is the simplest MP rule. It just inserts the feature [+high] (with automatic delinking of [−high]). c. Metaphony Rule III [*

/

[AGR, +participant, −author, −plural]

R N X [−cons] [−high] [+high] [−low] [−ATR]

(followed by delinking of [−ATR] due to the constraint *[+high, −ATR]25, see Calabrese 2011 for discussion) (i.e. e → ɪ → i, O → ʊ → u)

As discussed earlier, these operations are not phonetically motivated, however they are phonologically natural in the sense that they involve the basic operations of insertion and deletion applying on well grounded phonological representations. They are fully constrained in terms of what is a possible phonological rule. The rules in (31) require reference to stress information. A detailed discussion of stress in verbal forms in Altamurano is not possible here. What is important to observe, however, is that the position of stress is crucially dependent on morphosyntactic information: stress falls on the thematic vowel preceding the tense morpheme—with the exception of the present where one can assume that this morpheme is missing or pruned away (cf. Embick & Halle 1995), otherwise on the root. In the present indicative first and second plural, however, stress always falls on the thematic vowel preceding the subject agreement marker (for the sake of simplicity here I use intermediate representations without the surface processes of diphthongization, vowel fronting, unstressed vowel raising and final vowel deletion).

25 This constraint is independently required to account for the phonemic system of ­Altamurano where no contrast in ATR values is found in high vowels. Note that surface high vowels are [−ATR] because of a late phonetic process (see note 12).

106 

 Andrea Calabrese

(32) 1sg Present ˈʃokǝ

2sg

3sg

1pl

2pl

3pl

Gloss

ˈʃwekǝ

ˈʃokǝ

ʃoˈkwɛmǝ

ʃoˈkw ɛtǝ

ˈʃokǝnǝ

‘play’

manˈdƷɛmǝ fiˈnɛmǝ

manˈdƷiːvǝ manˈdƷɛnǝ ‘eat’ fiˈniːvǝ fiˈnɛnǝ ‘end’

Imperfect indicative:26 manˈdƷɛ manˈdƷiːvǝ manˈdƷɛ fiˈnɛ fiˈniːvǝ fiˈnɛ Imperfect subjunctive kanˈdɛssǝ kanˈdissǝ

kanˈdɛssǝ kanˈdɛssǝmǝ kanˈdissǝvǝ kanˈdɛssǝnǝ‘sing’

Perfect: kanˈdibbǝ

kanˈdau

kanˈdɛmmǝ kanˈdɛstǝ

(33) Generalizations a. In non-present T: stress always on TV T

AGR T

ess

TH ε ε ε ε

V Root

AGR TH ε i

e e eu



ImpI. ImpS. Perf. 1/2PLPr

T

eu

Root

‘sing’

b.  In present T: Stress on root except in 1 and 2pl T

V

kanˈdɛrnǝ

eu

kanˈdɛstǝ

rn m  

Therefore, the computation of stress requires reference to tense and inflectional endings, in particular subject agreement endings (cf. the first and second plural in the case of the present). It follows that stress must be must computed at the level of the morphological word (the m-word) (see Embick & Noyer 2001) where the outer inflectional endings can be accessed. It also follows that the metaphony rules must apply at the level of the morphological word, after stress is computed. A complete treatment of metaphony in Altamurano verbal system requires a brief discussion of “hypermetaphony” (Maiden 1991). In hypermetaphony, the metaphonic process applies differently in a certain determined morphological

26 The thematic /ɛ/ of the imperfect indicative diphthongizes as [ay] and not as [ɛy] (see fn. 14)



On the morphophonology of metaphonic alternations in Altamurano 

 107

environment. For example, in many Italian dialects, where we observe the usual metaphonic alternations in the nominal system, we find—or at least we can reconstruct—a different metaphony pattern in the verbal system. Thus, in Arpinate, metaphony in the verbal system, but not in the nominal one, also affects the low vowel /a/, e.g. ˈparlǝ vs. ˈpyerlǝ ‘speak-1sg/2sg’, in addition to the usual mid vowels. Accounting for “hypermetaphony” requires an additional mp rule. Thus, in the case of Arpinate we need to add (34) to the usual metaphony rules affecting mid vowels, where AGR indicates that the rule applies only in a verbal context. Fission of the illicit combination *[+low, +high], other vocalic readjustments and resyllabification account for the change /a/ → [ye] (see Calabrese 1985, 2011 on /a/-metaphony): (34)

[* X [–consonantal]

/

[AGR, –part, –auth, –plur]

[+high]

[+low]

In Altamurano, hypermetaphony is found in the 2nd singular and plural of the imperfect indicative, where the thematic vowel /ɛ/ is not changed into [iy] by metaphony, as expected by the rule in (31c), but into [i:]. Therefore we need to add the rule in (36) to those in (31): (35) manˈdƷɛ fiˈnɛ

manˈdƷiːvǝ manˈdƷɛ manˈdƷɛmǝ manˈdƷiːvǝ manˈdƷɛnǝ fiˈniːvǝ fiˈnɛ fiˈnɛmǝ fiˈniːvǝ fiˈnɛnǝ

(36) Hypermetaphony: [* N X [−cons]

/ X

[T−AGR, +Imperfect, +part, −auth]

followed by the automatic spreading of the root node (ε → εː], by application of (31aii)(εː → Iː] and by delinking of [−ATR] as in (31c)(Iː → iː]

[−low] [−back] [−ATR] [−high] We thus have an account of the metaphonic alternations in the Altamurano verbal system (see section 3 on nouns and adjectives).

108 

 Andrea Calabrese

4 Discussion of alternative analyses The key aspect of the analysis developed in the preceding section is the possibility of postulating MP operations such as those in (31) and (36). However, a few linguists have recently argued against the mixing of phonological operations and morphological features characteristic of MP operations. For example, Bermúdez-Otero (2012) (following Baudouin de Courtenay (1895=1972), Martinet (1965) a.o on the autonomy of morphology) proposes the principles in (37): (37) a.   Morphological operations do not alter the syntactic or phonological contents of morphs. (his 41, p. 50) b.  A phonological constraint may not refer to syntactic, morphological, or lexical information […] (his 71, p. 77)) According to (37), morphological processes can only change morphological configurations. In the same way, phonological processes can only change phonological configurations. No mixing of phonological and morphological features is possible. If we assume (37), we have two analytic options in the case of a metaphonic system like that of Altamurano. We can account for the metaphonic alternations: 1. in purely morphological terms by proposing that the alternations in (65) are due to suppletion; or 2. in purely phonological terms by assuming abstract phonological representations such as floating features or other floating prosodic devices. I will argue that both options are untenable. Therefore, principle (37) needs to be rejected as too restrictive. MP operations such as those in (31) and (36) must be allowed in the grammar.

4.1 Floating features Let us first try to account for Altamurano metaphony by using abstract phonological representations such as floating features. To illustrate this solution, consider a Pseudo-Italian dialect with a simplified metaphony system. (38)  i u  ɛ ɔ a (39)  ˈvɛdǝ ˈvidǝ ˈvɛdǝ  ˈʃɔkǝ ˈʃokǝ ˈʃɔkǝ  1sg 2sg 3sg

‘see-pres’ ‘play-pres’

In this case, the generalization is that stressed vowels are raised to high in the second singular.



On the morphophonology of metaphonic alternations in Altamurano 

 109

In a purely phonological process, in addition to a phonological structural change, both trigger and target of the process refer only to phonological information. In contrast, in MP processes, either the trigger or the target, or both, refer to morphological information. In analyses using floating features, a MP process is decomposed into two parts: i) a VI that inserts an abstract phonological element, in particular a “floating” phonological specification in the context of some morphological information; ii) a purely phonological operation triggered by the floating specification. This decomposition allows an account of a MP process in terms of a simple phonological rule. The decomposition needed by a floating feature analysis of the alternations in (39) requires the vocabulary items in (40), where (40a) inserts a floating feature [+high] in the case of the exponent of the 2sg. (40) 
 [+high] a. [+participant, −author, −plural] ↔  /− ǝ/ b. [−plural] ↔  /− ǝ/ The phonological rule in (41), which spreads the floating feature [+high], would then account for the metaphonic alternations in (39): (41) [ * X [−cons] [+high] [−high] This rule can apply to all stressed vowels, vacuously in case of high [i] and [u]. The constraint *[+high, +low] would block its application in the case of the low vowel [a]. This analysis requires a VI with a more abstract UR such as that in (40a); however it not only satisfies the requirements of the Autonomy of Morphology principle but also allows a phonological rule simpler than the MP rule in (42). (42) [ * X [−cons]

/

[+part., −auth., −plur.]

[+high] [−high] In a simple system like the Pseudo-Italian dialect, this floating feature solution can be easily maintained. However, this is not the case for the dialect of Altamura. I will first show that no single feature, set of features, or complex

110 

 Andrea Calabrese

­ honological configuration can account for all of the changes we see in (65). We p need to resort to different floating feature configurations. Then I will discuss the problems caused by using different floating features. Let us first consider if we can capture the different Altamurano metaphonic changes in terms of a single abstract representational element, a floating feature or some other abstract device that could be encoded in the exponent of a single VI. This requires accounting for the different changes in terms of a single generalization: a unitary analysis of the changes. The Altamurano metaphonic alternations involve raising, lengthening, and diphthongizing mid vowels. Let us first consider the possibility that Altamurano metaphony involves a single process raising mid vowels to high ones as expected historically. Under this possibility, we can assume the vocabulary items in (40) followed by the rule in (41). This solution obviously gives us the wrong outcomes as we can see in (43): (43) [  +ATR] e [  −ATR] ɛ [  +ATR] o [  −ATR] ɔ

Metaphony (expected) → *i iː → *ɪ i → *u we → *ʊ u

Assuming fission and subsequent synchronic reduction of the diphthong [ye] to [iː] would not save this analysis as shown in (44): (44) [  +ATR] e [  −ATR] ɛ [  +ATR] o [  −ATR] ɔ

Metaphony (expected) → *i iː → ɪ →ye →ie→*iː i → *u we → ʊ →wo→*we u

There is no plausible phonological process that could lengthen [+ATR] [i] and diphthongize [+ATR] [o]. And even if such a process is postulated, it would lead to the wrong results insofar as it would also apply to underlying [i] and [u] that do not change in metaphonic contexts.27 The only option I see in this case is to assume an abstract analysis where the rule exchanging [ATR] values is a synchronic rule of Altamurano. This rule, repeated in (45), would have to apply after metaphony in (41). Under this analysis,

27 I will not explore analyses employing underspecification of the target vowels or other phonological tricks since I do not see how they could adequately account for the Altamurano alternations or improve the analyses I discuss here.



On the morphophonology of metaphonic alternations in Altamurano 

 111

which essentially reproduce history, the underlying values for the feature [ATR] in the case of the mid vowels is reversed so that we have intermediate metaphonic alternations in (46). These alternations can be analyzed as I did for the original historical alternations in (21) (with addition of synchronic reduction for the diphthong [ye] to [iː] and fronting in the case of [wo]→[we]. (45) [aATR] → [−aATR] / [ 

, −high, −low]



Metaphony Fission Further changes (expected) (46) [−ATR] ɛ → ɪ → ye → ie → iː iː [+ATR] e → i i [−ATR] ɔ → ʊ → wo → we we [+ATR] o → u u The across-the-board application of (45) after metaphony—a case of absolute neutralization—would account for the surface alternations (e/iː, ɛ/i, o/we, ɔ/u). The point is that there is absolutely no independent synchronic motivation for this abstract analysis. In particular, there are no synchronic alternations, or other possible phonological cues, that could provide evidence for (45). It is then unclear how the learner could ever postulate the underlying specifications for [ATR] in (46) other than from the metaphony process itself. And even if this were done, it is unclear why the difference in length [iː] vs. [i], or the difference in diphthongization [o] vs. [we], should ever lead to the postulation of an underlying difference in [ATR] values, i.e. [ɛ] vs. [e], [ɔ] vs. [o]. The analysis assuming that there is a generalized process of raising behind metaphony in Altamurano can, therefore, be reasonably rejected. Another possibility is to assume that behind this phenomenon, there is a single process of lengthening as seen in the alternation [ɛ/iː]. We could assume that the relevant vocabulary item contains a floating mora, which is then somehow attached to the stressed vowel (I will not try to formalize this process here). Under such an analysis all stressed vowels would be lengthened in a metaphonic context as below: Stressed (47)  [+ATR] [  −ATR]  [+ATR]  [−ATR]

e ɛ o ɔ a i u

2sg (expected) eː iː ɛː i oː we ɔː u aː a iː i uː u

112 

 Andrea Calabrese

Obviously other rules are needed to obtain the expected outcomes in the metaphonic context. One could propose that long mid [+ATR] vowels are dipthongized as in (48). This is a plausible operation, although usually in Romance it is the [−ATR] [ɛ] and [ɔ] that diphthongize as [ye] and [wo], (see Loporcaro 2010) and not the [+ATR] [e] and [o], which instead diphthongize as [ey] and [ow]. (48) eː   →   ye     (ye → ie → iː) oː   →   wo The long [−ATR] vowels in (47) must be raised (and tensed) by the rules in (50). (49) ɛː   →    ɪː     →    iː ɔː   →  ʊː   →    uː (50) a. X

X

[−ATR]

  [+high]  

b. X

X

[+high]

[+ATR]

Then they must be shortened. This could be done by the rule in (51) which is also needed to shorten the long high vowels that were lengthened in a metaphonic context (see 47). Also another rule, (52), is need to shorten the long low vowel in (47). Obviously, these operations of shortening must apply before other regular processes of Altamurano lengthen (and dipthongize) vowels in open stressed syllables: (51) X

X

→

[+high] (52) X

X

[+low]

X [+high]

→

X [+low]

(51) and (52) are very stipulative rules with no independent motivation in the phonology of Altamurano. Furthermore, (51) must be prevented from applying to the long [iː] that is the outcome of metaphony of [e]. This could be done by ordering the application of (51) before the application of relevant rules in (48). It follows that if we assume that metaphony involves generalized lengthening as in (47), we would be also forced to assume an analysis that is fundamentally unmotivated and unsatisfactory with ad hoc rules and rule orderings. I do not see any other plausible solution in which Altamurano metaphony can be accounted for by means of an operation involving a single feature, set of features, or phonological configuration. In conclusion, assuming a single VI inserting an abstract feature



On the morphophonology of metaphonic alternations in Altamurano 

 113

such as a floating [+high] or mora fails to provide an adequate account for this phenomenon. Note also that we still need to account for hypermetaphony. In the case of hypermetaphony in Altamurano, thematic vowel /ɛ/ of the imperfect indicative is not changed into [i] by metaphony, as expected, but into [iː]. Both of the analyses proposed above would fail to account for what happens in this case. If we postulate that thematic vowel [ɛ] is actually underlying [e], as expected according to (45), assuming the metaphony rule in (41) would lead to expect a surface [i] (see 46), contrary to the facts. If we postulate that metaphony involves a lengthening process as in (47), the thematic vowel [ɛ] must undergo the changes in (50) and (51), and again should become surface [i]. I do no see any way out of it except postulating that (51) is blocked in the imperfect indicative, an obvious violation of Autonomy of Morphology. Therefore, the only possibility that I see in the case of Altamurano hypermetaphony is to assume a different VI, and, for example, postulate that the VI of the 2sg in the imperfect indicative contains both a mora and the feature [+high]. Given that an independent VI for the 2sg suffix must be assumed for the imperfect indicative anyway, let us then consider the possibility that Altamurano metaphonic alternations indeed always involve different VIs for the 2sg, each inserting a different floating feature configuration. Crucially, in this case, the choice of the floating feature configuration in the VI must depend on the nature of the target stressed vowel in the relevant metaphonic alternation. Thus, we are in a situation in which an inflectional VI must refer to the phonological nature of the stressed vowel in the stem, where, as discussed in section 3, stress is determined at word level. For example, the VI triggering the [+high] raising in (13c) should be that in (23): 
       [+high] (53) [+participant, −author, −plural]  ↔  /− ǝ /

[* / [[…ɛ.. ]

]W

Putting aside the issue of having a VI including predictable phonological information in its statement—an obvious violation of autonomy of morphology, such a VI is simply impossible since it must be inserted before the determination of stress at the word level. Therefore it cannot refer to it. We have a simple case of an ordering paradox. Assuming different VIs to account for Altamurano metaphonic alternation is therefore not a possible solution. Again the simplest thing to do is to say that the metaphonic changes are due to rules that apply differently to the different vowels, as it was done in section 3. Using MP rules still appears the best way to account for Altamurano metaphonic alternations.

114 

 Andrea Calabrese

4.2 Suppletion Another possible way to account for the type of stem allomorphy we see in Altamurano metaphony in a way that satisfy the autonomy of morphology principle is to resort to stem suppletion. Kiparsky (1996) proposes the following criteria for suppletive vs. MP analysis of alternations. (54) a) Properties of alternations due to suppletion/morpho-lexical allomorphy: i. they are idiosyncratic (item-specific) ii. they may involve more than one segment iii. they obey morphological locality condition (cyclic insertion, morpheme adjacency) iv. they are ordered prior to MP and phonological processes b) Properties of alternations due to MP processes: i. they are general (not item-specific) ii. they involve a single segment iii. they observe phonological locality conditions, not morphological ones iv. they follow suppletion/morpholexical processes (i.e. vocabulary insertion) According to these criteria, Altamurano metaphonic alternations are clearly MP: i) They are general. ii) They involve a single segment (stressed vowels). iii) They do not obey morphological locality condition (cyclic insertion, morpheme adjacency). For example, metaphony, which is triggered by the AGR morpheme, applies across the T morpheme—of the indicative imperfect and of the past participle—onto the stem in (56) (see morphosyntactic structure in 55ː) (55)

T T v Root

AGR T

TH

(56) kanˈd-ɛ-ss-ǝ    kanˈd-i-ss-ǝ   ‘sing-Imperfect Subjunctive--1sg/2sg aper-t-ǝ apir-t-ǝ ‘open-PastParticiple-fsg/msg Furthermore, iv) metaphonic alternations also require previous application of phonological and MP processes, not only of stress, as discussed ealier, but also of thematic vowel deletion. Adequate selection of the metaphonic alternant, in



On the morphophonology of metaphonic alternations in Altamurano 

 115

fact, requires previous deletion of the thematic vowel, which otherwise would be expected to appear in its metaphonic shape: (36) *ʃok-i-ǝ vs. ʃwek-ǝ

‘play-Present2sg’ (UR / ʃok / ʃwek-ɛ / i-ǝ/)

Thus, the criteria in (54) tell us that Altamurano metaphony should be analyzed in terms of a MP process. But let us suppose that these criteria are not valid and assume that suppletive stems must be postulated anyway because of (20). The suppletion solution requires two partially distinct stem allomorphs sharing all segments but the alternating vowel where the allomorphs restricted to the 2sg would have to be fully prosodified to prevent its appearance in unstressed positions. A constraint blocking multiple stresses per word would also be needed to account for its correct distribution:28 (57) a. Verbs:

{ { { {

} } } }



     ˈiː   /l   ʃ-/     e

‘read’



      ˈi   /kr       t-/     ɛ

‘believe’



    ˈwe   /ʃ     k-/     o

‘play’

    ˈu   /ad    r-/     ɔ b. Thematic vowel:

{}

  -ˈi  -         -ɛ-

‘smell’

‘TV-I Conj’.

28 The thematic vowel should also be included in the stems in (57), i.e. liˈːʃ-ǝ / leʃ-ǝ, etc. I will not do it here for the sake of simplicity, since this issue is not relevant.

116 

 Andrea Calabrese

An informal selection rule (a lexical redundancy rule (Jackendoff 1975) or a morpholexical rule) (Lieber 1981, 1992) is given below: (58) Select the vowel { iˈ  ː, ˈi, ˈwe, ˈu} in the allomorphs in (37) if the allomorph is in the context [V, +part., −auth., −plur.], otherwise select {e, o, ɛ, ɔ}, respectively. The selection mechanism requires access to stress. As discussed earlier, stress can be computed only after inflectional exponents are inserted (=selected). If vocabulary insertion is cyclic as assumed earlier, we are in a paradoxical situation in which insertion/selection in an inner node must be done only after insertion/selection in an outer node is done. (59)

T T v Root

AGR T

TH

This solution requires assuming that vocabulary insertion is not cyclic and that stem competition and selection is done at word level where both stress and morphological information—PHI features that are in the outer position in the m-word—can be accessed at the same time. This can be done by assuming phonologically driven allomorph selection by output optimization (e.g. Tranel 1996, 1998; Kager 2009: 420ff; Mascaró 1996, 2007; Rubach & Booij 2001; Bermúdez-Otero 2012). In this account, the lexicon supplies two listed allomorphs for alternating items. The morphology inserts both allomorphs at the appropriate place in the underlying representation, with an instruction for the phonology to resolve the disjunction. If all things are equal, phonological markedness considerations cannot account for the selection of the metaphonic alternant insofar as they contain phonological configurations (long vowels, diphthongs, stressed high vowels) that are more marked than the pure mid-vowels of the non-metaphonic alternant. Some other device is needed to ensure that the metaphonic alternant is chosen. Mascaró (2007) and Bonet, Lloret & Mascaró (2007) propose that there are “precedence” relations, x > y, between allomorphs, which mark certain allomorphs as being dispreferred, essentially the selection rule in (58) restated in OT terms. They also propose the faithfulness constraint priority, which penalizes selection of the dispreferred allomorph. In this framework, Altamurano metaphonic alternations could be analyzed as follows. The suppletive alternants can be those in (37). Crucially, the precedence relations [iː > e, i > ɛ, we > o, u > ɔ] must be restricted to the context 2sg

On the morphophonology of metaphonic alternations in Altamurano 



 117

(i.e. ( iː > e, i > ɛ, we > o, u > ɔ] / 2sg). The tableaux below illustrate how the combination of Priority, a markedness constraint against diphthongs (*we—one of the marked configurations appearing in a metaphonic context) and a constraint on foot structure (Foot) could account for the distribution of the stem alternants ʃwek- / ʃok- requiring the positioning of the metaphonic (inherently stressed) stem allomorph in the last foot of the word and excluding multiple stresses per word (further constraints must account for the pretonic vowel reduction (surface [ʃǝkiːv])):29 (60) input: { we > o /

2sg}, [ [ ]

2sg

Foot

Priority

*we *

a.  ˈʃwek b.    ˈʃok (61) input: { we > o / a.

*! 2sg} [ [ ]

3sg

Foot

Priority

*we *!

 ˈʃwek

b.  ˈʃok (62) input: { we > o / a.   ˈʃ weˈkiːv

2sg} [ [ ]

Imp 2sg

Foot

Priority

*we

*!

* *

b.  ʃoˈkiːv

(63) input: {we > o / _ 2sg},{iː > ɛ / _I2sg}, [ [ ] _I2sg Foot Priority a.   ˈʃweˈkɛvǝ

*!

b.   ˈʃweˈkiːvǝ

*!

*

*we

*i:

* *

* *

c.   ʃoˈkɛvǝ

**!

d.  ʃoˈkiːvǝ

*

*

e.   ˈʃwekɛvǝ

*!

*

This account is successful. However, the problem remains of why we should encode regular phonological changes in suppletive stem forms. By definition, suppletion is used when the relations between allomorphic alternants is idiosyncratic, arbitrary, and item specific, therefore when no phonological

29 The thematic vowels should also be included as part of the stem. The issue is, however, irrelevant there.

118 

 Andrea Calabrese

generalizations can correlate the alternants. This is simply not true in the case of Altamurano metaphony. Metaphonic alternations are general, regular and predictable. They involve representations that can be related phonologically in quite simple ways, very differently than in true cases of suppletions: e.g. went / go in English or vad / and in Italian. They are not restricted to some special irregular verb or noun but they occur in hundreds of otherwise regular verbs and nouns (see next section). Metaphony is a general process of the language. It is unclear why the learner should be unable to extract the relevant information when exposed to such massive evidence and to construct the relevant generalization that certain vowels are being changed in a certain morphological context, and why s/he should resort to memorized stem alternants. The shape of the different stems in Altamurano is fully predictable. As discussed at length, stem alternation, and therefore suppletion in this framework, occurs only when the basic, non-metaphonic stem alternant contains a stressed mid vowel. If the stem has the vowels [e, ɛ, o ɔ] as their stressed syllable, they will have [i, iː, we, u] in the relevant morphological contexts. This fact needs to be captured by the grammar instead of stipulating it in the shape of the suppletive stem. If we resort to suppletion to account for robust generalizations such as those governing the metaphonic alternation of Altamurano, it is unclear what remains to be accounted for in terms of synchronic processes. It, for example, becomes unclear why one should not account for generalizations on the surface shape of stems like those involving vowel fronting and vowel diphthongization discussed in section 1 by encoding them into suppletive forms of the stems, thus essentially transforming them into stipulations. Therefore, the suppletive solution leads to unsatisfactory results. It follows that both a suppletive analysis and a floating feature analysis do not provide adequate accounts of metaphony in Altamurano. I submit then that the most adequate way of accounting for the metaphonic alternation of Altamurano is to assume that MP rules such as those in (31) and (36) are possible. An adequate grammar must include MP operations like the MP rules of DM. These are the “powerful readjustment rules” which Bermúdez-Otero (2012: 80) claims “utterly destroy the empirical content of morphological and phonological hypotheses”. It is actually unclear what the empirical grounds of Bermúdez-Otero’s claim are; the only motivation for it appears to be ideological: the assumption of the autonomy of morphology. From the discussion in this section, however, it appears that this principle leads to extremely abstract phonological analyses with very little empirical content, to logical paradoxes, or to analyses in which robust MP generalizations are simply stipulated into idiosyncratic



On the morphophonology of metaphonic alternations in Altamurano 

 119

s­ uppletive stem forms. It can, thus, be claimed, pace Bermúdez-Otero, that MP rules—his scary “powerful readjustment rules”30—are actually the most concrete, least abstract and most adequate way to capture MP processes and generalizations.

5 M  etaphony in the nominal and adjectival systems of Altamurano I will turn now to the alternations found in the nominal and adjectival systems of Altamurano. Given the absence of final class markers, different inflectional classes in nouns must be identified by the presence/absence of metaphonic changes. Such changes are obviously dependent on the phonological make up of the stem in so far as only underlying mid vowels may undergo metaphony. We thus have the following stem classes. Here I will give nouns and adjectives in their citation forms as provided by Loporcaro (1988) so that reader can have an idea of the complexities of the surface phonology of this language. As discussed in section 2, identification of the metaphonic alternation requires parsing away surface diphthongization in (intermediate) open stressed penultimate syllables, and vowel fronting of [+ATR] vowels in stressed syllable that are either closed or in antepenultimate position. Remember that fronting is blocked in linked structures. Also remember that final schwas are deleted in citation forms. (64) a.  Nominal stems with high and low stressed vowels do not undergo metaphony. ˈfiyk / ˈfiyk ‘fig-sg/pl, ˈtsiyt / ˈtsiyt ‘boyfriend / girlfriend’, ˈluwʃ / ˈluwʃ ‘light-sg/pl’ ˈkɛyn / ˈkɛyn ‘dog-sg/pl’ ˈkavadd / ˈkavadd ‘horse-sg/pl’

30 Bermúdez-Otero points the finger at the “excessive power” of DM MP rules, but he fails to notice that the same criticism could be leveled against the OT formalism he adopts, in particular against faithfulness constraints. In fact, whereas markedness constraints are constrained by the universal substance of language, no such restriction governs faithfulness constraints. They are fundamentally unconstrained, in the same way as GEN that they control is unconstrained. So the same criticism as the one leveled against SPE rules by OT proponents can be leveled against faithfulness constraints: they are completely unrestricted, arbitrary, and thus characterized by an excessive richness in expressive power. There are basic limits on the operations of rules: they can only insert or delete nodes in well-grounded phonological representations. It is not so for faithfulness constraints: their only limits are in the imaginative capacity of the OT practitioner (see Calabrese 2005, forthcoming for more discussion of faithfulness constraints).

120 

 Andrea Calabrese



b. Nominal stems with mid stressed vowels and metaphonic changes in the masculine (singular/plural) but not in the feminine (singular/plural) Masculine: paˈtruwn / paˈtruwn ‘master-sg/pl’, ˈʃ ing / ˈʃ ing ‘male pack animal / donkeysg/pl’, ˈkweyk / ˈkweyk ‘chef-sg/pl’, ˈpwerk / ˈpwertʃ  ‘pig-sg/pl’ Feminine: ˈʃɛng / ˈʃɛng ‘female pack animal / mare-sg/pl’, paˈtrawn / paˈtrawn ‘mistress-sg/pl’, ˈrɛkky / ˈrɛkky ‘ear-sg/pl.’ Alternation masculine vs. feminine: paˈtruwn / paˈtrawn “master/mistress”31, ˈʃing / ˈʃɛng ‘male / female pack animal’, sǝɲˈɲuwr / sǝɲˈɲawr ‘sir / lady’ d. Masculine nominal stems with metaphonic changes in singular but not plural sruˈvidzy / srǝˈvɛdzy ‘service-sg/pl’, raˈtʃepp32 / raˈtʃØpp ‘bunch of grapessg/pl’, ˈkwern / ˈkØrn ‘horn-sg/pl’ Some of them have the suffix- /-ǝrǝ/ in the plural ˈess33 / ˈØssǝrǝ ‘bone-sg/pl’, ˈert / ˈØrtǝrǝ ‘vegetable garden-sg/pl’, ʃaˈnükky / ʃaˈnɔkkyǝrǝ ‘knee-sg/pl’, ˈtʃiggy / ˈtʃɛggyǝrǝ ‘eye-lash-sg/pl’, ˈdiʃt / ˈdɛʃtǝrǝ ‘finger-sg/pl’, ˈpwutts / ˈpwɔttsǝrǝ ‘well-sg/pl’ e. Nominal stems (both masculine and feminine) with metaphonic changes in plural but not singular ˈvɛrm / ˈvirm ‘worm-sg/pl’, ˈpeyt / ˈpiːt ‘foot-sg/pl’, ˈsɔrǝkǝ / ˈsürǝkǝ ‘mouse-sg/pl’, ˈwowf / ˈweyf ‘ox-sg/pl’, ˈnawʃ / ˈnuwʃ ‘nut-sg/pl’, paˈyɔwn / paˈyuwn ‘peacock-sg/pl’, ˈsɛddƷ / ˈsiddƷ ‘chair-sg/pl’, ˈwosk / ˈwesk ‘wood-sg/pl’ In the case of adjectives, we have the following situation: (65) a. Adjectival stems with high and low stressed vowels do not undergo metaphony: ˈrann ‘big-m/f.’, ˈsɛyn ‘sane-m/f.’, ˈtʃitt ‘silent-m/f.’, ˈfiyn ‘thin-m/f.’, ˈbrütt ‘ugly-m/f.’

31 Remember that in some words, underlying [ɔ] and [ɛ] may diphthongize as [aw] and [ay], respectively, instead of [ɛy, ɔw]. This is the case of the words for ‘master’ and ‘sir/lady’. So, ­before the application of diphthongization (and final schwa deletion), the alternations [ paˈtruwn / paˈtrawn], [sǝɲˈɲuwr / sǝɲˈɲawr] are [ paˈtrunǝ / paˈtrᴐnǝ], [sǝɲˈɲuwrǝ / sǝɲˈɲawrǝ]. 32 Remember that [w] of the diphthong [we] is deleted after coronal consonants and sometimes in word initial position, so [raˈtʃɛpp] derives from intermediate [raˈtʃwɛpp], see also [ˈess] ‘bonesg’ below from intermediate [ˈwess]. 33 See note 26.



On the morphophonology of metaphonic alternations in Altamurano 



 121

b. Adjectival stems with mid stressed vowels and metaphonic changes in the masculine and not in feminine: ˈneyf / ˈnowf ‘new-m/f.’, ˈbbweyn / ˈbbown ‘good-m/f.’, ˈtirn / ˈtern ‘tender-m/f.’, ˈtünn / ˈtɔnn ‘round-m/f.’, ˈritt / ˈrɛtt ‘straight-m/f.’,ˈfriʃk / ˈfrɛʃk ‘fresh-m/f.’ c. In the case of these stems, metaphonic changes appear to be conditioned by the morpho-syntactic context: If the adjective modifies a noun referring to masculine human being, metaphonic changes occur in the singular and plural. If it does not, the metaphonic changes occur only in the masculine singular: u ˈbbwenǝ waɲˈɲawn i ˈbbwenǝ waɲˈɲuwn ‘the good boy’ ‘the good boys’ la ˈbbona waɲˈɲedd / ˈkaws i ˈbbonǝ waɲˈɲedd / ˈkaws ‘the good girl / thing’ ‘the good girls / things u ˈpwɛtǝ34 / kwaˈpiddǝ ˈkwurt / ˈleŋŋ / ˈɲɲuwr i ˈpiːtǝ / kaˈpiddǝ ˈkØrt / ˈlØŋŋ / ˈɲɲawr ‘the foot / hair short / long / black’ ‘the feet / hairs short / long / black’ u mwaˈrandƷǝ ˈtest / ˈmwedd i maˈrandƷǝ ˈtØst / ˈmwodd ‘the orange hard / soft’ ‘the oranges hard / soft’ l ˈarvǝ ˈtert l ˈarvǝ ˈtØrt ‘the tree crooked’ ‘the trees crooked’ u kwaˈvaddǝ ˈrüss i kaˈvaddǝ ˈrɔss ‘the horse red’ ‘the horses red’ l anǝˈmɛlǝ ˈbbweyn / ˈlürd l anǝˈmɛlǝ ˈbbown / ˈlɔrd ‘the animal good / dirty’ ‘the animals good / dirty’ u ɲɲummuˈriddǝ s e ˈkkwettǝ ˈbbweyn i ɲummuˈridːǝ s ˈawɔnnǝ ˈkːØttǝ ˈbbown ‘the meat bundle is cooked well’ ‘the meat bundles are cooked well’ u ˈʃkwatǝlǝ / l ˈarvǝ s e ˈrrütt u ˈʃkwatǝlǝ / l ˈarvǝ s ˈawɔnnǝ ˈrɔtt ‘the box / tree is broken’ ‘the box / tree are broken’ d. Adjectival stems with mid stressed vowels and metaphonic changes only in the masculine pl ˈvekky / ˈvikky ‘old-m.sg;f.sg;f.pl / m.pl’,ˈddƷɔvǝnǝ / ˈddƷüvǝnǝ ‘-m.sg; f.sg;f.pl / m.pl’

34 Remember that non-coronal consonants are rounded after [u].

122 

 Andrea Calabrese

The following analysis can be proposed. Nouns with high and low stressed vowels do not undergo metaphony. Otherwise, we have two classes of nouns. The first class of nouns is divided into two subgroups: feminine and masculine nouns. Feminine nouns do not undergo metaphony. In the case of the masculine nouns there are two further subgroups. In a group of nouns there is metaphony both in the singular and the plural, in the second group only in the singular. I refer to the second group as the b group, and account for the behavior of this group as in (66) (b is a diacritic on the root; if present, metaphonic changes are restricted to a [−feminine, −plural] context, N indicates that the rule applies when the root is in a nominal context). In (66), I am adding the context relevant for the Class I of nouns to the metaphony rule as in (31):

}

{

[AGR, +part, −auth, −plur] [RootI , N, [−fem], ]

{

[AGR, +part, −auth, −plur] [RootI, , N, [−fem], ] [RootII, N, [+plur]

Class I: (66) Metaphonic changes (=31) /

In the second class of nouns, there is metaphony only in the plural. Thus we can add another morpho-syntactic context to those in (66), as in (67): Class II:   /    (67) Metaphonic changes (=31)

}

Adjectives can belong to the same classes. However, if they belong to Class I, the following rule holds: (68) [−human]

→

b / [Adjective, RootI

]

A complete account of the metaphonic alternation in Altamurano can thus be achieved. Observe that the analysis of nominal and adjectival alternations crucially requires the use of diacritics, and of rules manipulating diacritics such as that in (68), something that is natural in DM (see Calabrese 2013 for discussion and arguments in support of the use of diacritics). Although nominal and adjectival alternations do not pose further major problems to the floating feature-based model and the suppletion-based model other than those already mentioned in the preceding section, these analyses require the use of diacritics to treat them adequately. Consider the pairs in (69) as representative of their own classes. (69) a. sruˈvidzy / srǝˈvɛdzy ‘service-sg/pl       b. ˈsɛddƷ / ˈsiddƷ ‘chair-sg/pl’ Metaphony applies in the singular in (69a) and in the plural in (69b). There is no phonological cue that could distinguish these pairs. Therefore, the



On the morphophonology of metaphonic alternations in Altamurano 

 123

different VI required by the floating feature analysis would be forced to refer to a lexical property of the root, hence to a diacritic. Under the suppletive analysis, the selection rule in (58) must also crucially refer to different classes of stems, and, therefore, to some idiosyncratic lexical property—again a diacritic— identifying them. The issue is whether or not the use of diacritics can be naturally assumed in these models. It is definitely natural in DM, and therefore this fact can be considered as further evidence for this model and for the MP rules that it assumes.

3 Conclusion In this paper I have provided evidence that the best analysis of the metaphonic alternations in Altamurano includes MP operations changing phonological features in morphological contexts. The best way of modeling these operations is provided by the MP rules of DM. Notice that although these rules may be quite idiosyncratic and not phonetically motivated, they are always naturally restricted by the nature of the phonological operation and of phonological representations. They will involve the basic phonological operation of insertion and deletion operating on well-grounded phonological ­representations. As observed in Calabrese (2005), the synchronic phonology of a language is both natural and conventional. It is natural because sound changes are motivated by markedness considerations. An analysis based on considerations of this type has an advantage in terms of economy and efficiency and will tend to be preserved in the language. But phonology is also conventional because of the way in which language is transmitted in a social group. Because of its conventional nature, the effects of idiosyncratic accumulation of historical changes and events can be encoded in the surface shape of the language. These effects introduce arbitrariness in the synchronic phonology of the language: we therefore have exceptions, diacritics, idiosyncratic rules and extrinsic ordering of processes in the derivation, all of which may obscure the natural side of phonology. As discussed in this paper, MP processes also arise in this way. Phonological theory must have an architecture such that processes involving universal markedness considerations and purely language specific processes interact with each other smoothly and efficiently (see Calabrese 2005, 2009, forthcoming). MP rules must be part of this architecture.

124 

 Andrea Calabrese

References Akinlabi, Akinbiyi. 1996. Featural affixation. Journal of Linguistics 32: 239–289. Baudouin de Courtenay, Jan N. 1972. An attempt at a theory of phonetic alternations. In Edward Stankiewicz (ed.), A Baudouin de Courtenay anthology: the beginnings of structural linguistics (Indiana University Studies in the History and Theory of Linguistics), 144–212. Bloomington: Indiana University Press. First published (1895), Versuch einer Theorie ­phonetischer Alternationen: ein Kapitel aus der Psychophonetik, Straßburg: Trübner. Bermúdez-Otero, Ricardo. 2012. The architecture of grammar and the division of labour in exponence. In Jochen Trommer (ed.), The morphology and phonology of exponence (Oxford Studies in Theoretical Linguistics 41), 8–83. Oxford: Oxford University Press. Bermúdez-Otero, Ricardo. 2013. The Spanish lexicon stores stems with theme vowels, not roots with inflectional class features. Probus 25 (1): 3–103. Bobaljik, Jonathan D. 2000. The ins and outs of contextual allomorphy. In Kleanthes K. Grohmann and C. Struijke (eds.), Proceedings of the Maryland Mayfest on Morphology 1999, 35–71. University of Maryland Working Papers in Linguistics 10. College Park: ­University of Maryland. Bobaljik, Jonathan D. 2012. Universals in comparative morphology: suppletion, superlatives and the structure of words. Cambridge, MA: The MIT Press. Bonet Eulalia and Daniel Harbour DA. 2012. Contextual Allomorphy. In Jochen Trommer (ed.), The morphology and phonology of exponence (Oxford Studies in Theoretical Linguistics 41), 195–236. Oxford: Oxford University Press. Bonet, Eulalia, Maria Rosa Lloret, and Joan Mascaró. 2007. Allomorph selection and lexical preferences: two case studies’ Lingua 117 (6): 903–27. Bye, Patrik. 2007. Allomorphy – selection, not optimization. In Sylvia Blaho, Patrik Bye, and Martin Krämer (eds.), Freedom of analysis (Studies in generative grammar 95), 63–91. Berlin: Mouton de Gruyter. Bye, Patrik and Peter Svenonius. 2012. Non-concatenative morphology as epiphenomenon. In Jochen Trommer (ed.), The morphology and phonology of exponence (Oxford studies in theoretical linguistics 41), 427–495. Oxford: Oxford University Press. Calabrese, Andrea. 1985. Metaphony in Salentino. Rivista di grammatica generative. IX−X: 3–140. Calabrese, Andrea. 1988. Towards a theory of phonological alphabets. PhD dissertation, MIT. Calabrese, Andrea. 1995. A constraint-based theory of phonological markedness and simplification procedures. Linguistic Inquiry 26 (2): 373–463. Calabrese, Andrea. 1998. Metaphony revisited. Rivista di Linguistica 10 (1): 7–68. Calabrese, Andrea. 2000. The fronting of ü and the status of the feature ATR in Romance. In Lori Repetti (ed.) Phonological theory and the dialects of Italy, 59–88. Amsterdam: John Benjamins. Calabrese, Andrea. 2005. Markedness and economy in a derivational model of phonology. Berlin: Mouton de Gruyter. Calabrese, Andrea. 2009. Markedness theory vs. phonological idiosyncracies in a realistic model of language. In Charles Cairns and Eric Raimy (eds.), Contemporary views on architecture and representations in phonological theory, 261–304. Cambridge: MIT Press. Calabrese, Andrea. 2011. Metaphony in Romance. In C. Ewen, M. van Oostendorp, B. Hume (eds.), The Blackwell companion to phonology, Vol.V, 2631–2762. London: WileyBlackwell.



On the morphophonology of metaphonic alternations in Altamurano 

 125

Calabrese, Andrea. 2015a. Locality effects in Italian verbal morphology. In Di Domenico, E., Hamann, C. and Matteini, S. (eds.), Structures, Strategies and Beyond: Studies in honor of Adriana Belletti, 97–132. Benjamins. Calabrese, Andrea. 2015b. Markedness Effects in the Gallo-Romance Vowel System. To appear in Diachroniques. Calabrese, Andrea. Forthcoming. Essays on Morpho-Phonology. Ms. University of Connecticut. Chomsky, Noam and Morris Halle. 1968. Sound patterns of English. New York: Harper & Rows. Cole, Jennifer. 1998. ‘Deconstructing Metaphony. Rivista di Linguistica 10: 69–98. De Lacy, Paul. 2006. Markedness: Reduction and preservation in phonology (Cambridge Studies in Linguistics). Cambridge: Cambridge University Press. Embick, David. 2010. Localism versus globalism in morphology and phonology. Cambridge, MA: MIT Press. Embick, David and Morris Halle. 2005. On the status of stems in morphological theory. In Twan Geerts, Ivo van Ginneken, and Haike Jacobs (eds.), Romance languages and linguistic theory 2003: selected papers from ‘Going Romance’ 2003, Nijmegen, 20–22 November (Current Issues in Linguistic Theory 270), 37–62. Amsterdam: John Benjamins. Embick, David and Alec Marantz. 2008. Architecture and blocking. Linguistic Inquiry 39: 1–53. Embick, David, and Rolf Noyer. 2001. Movement operations after syntax. Linguistic Inquiry 32: 555–595. Gianmarco, Ernesto. 1979. Abruzzo. Pacini Editore, 1979 Halle, Morris and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In Ken Hale and Samuel J. Keyser (eds.), The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger, 111–176. Cambridge, MA: MIT Press. Inkelas, Sharon and Cheryl Zoll. 2007. Is grammar dependence real? A comparison between cophonological and indexed constraint approaches to morphologically conditioned phonology. Linguistics 45 (1): 133–71. Jackendoff, Ray. 1975. Morphological and semantic regularities in the lexicon. Language 51 (3): 639–71. Kager, René. 2009. Lexical irregularity and the typology of contrast. In Kristin Hanson and Sharon Inkelas (eds.), The nature of the word: studies in honor of Paul Kiparsky, 397–432. Cambridge, MA: The MIT Press. Kaze, Jeffery. 1989. Metaphony in Spanish and Italian dialects revisited, PhD dissertation, University of Illinois, Urbana-Champaign. Kenstowicz, Michael and Charles Kisseberth. 1979. Generative phonology: description and theory. New York: Academic Press. Kiparsky, Paul. 1985. Some consequences of Lexical Phonology. Phonology Yearbook 2: 83–138. Kiparsky, Paul. 1996. Allomorphy or morphophonology? In Rajendra Singh (ed.), Trubetzkoy’s Orphan: Proceedings of the Montréal Roundtable “Morphophonology: Contemporary Responses”, 1996. Amsterdam: Benjamins. Krämer, Martin. 2009. The phonology of Italian, Oxford: Oxford University Press. Lieber, Rochelle. 1981. On the organization of the lexicon. PhD Dissertation, MIT. Lieber, Rochelle. 1992. Deconstructing morphology: word formation in syntactic theory. Chicago: University of Chicago Press. Loporcaro, Michele. 1988. Grammatica storica del dialetto di Altamura. Pisa: Giardini. Loporcaro, Michele. 1991. The natural phonological process [V, +high] → [+Tense] and the vowel system of some Southern Italian dialects. Folia Linguistica XXV: 3–4.

126 

 Andrea Calabrese

Loporcaro, Michele. 2010. Phonological processes. In Martin Maiden, John Charles Smith and Adam Ledgeway (eds.), The Cambridge history of Romance languages, Vol. 1. Structures, p. 109–154. Cambridge: CUP. Maiden, Martin. 1991. Interactive morphonology: metaphony in Italy. New York: Routledge. Martinet, Andrea. 1965. De la morphonologie. La Linguistique 1: 15–30. Mascaró, Joan. 1996. External allomorphy as emergence of the unmarked. In Jacques Durand and Bernard Laks (eds.), Current trends in phonology: models and methods, 473–483. Salford, Manchester: European Studies Research Institute, University of Salford. Mascaró, Joan. 2007. External allomorphy and lexical representation. Linguistic Inquiry 38 (4): 715–735. Paster, Mary E. 2006. Phonological conditions on affixation. PhD dissertation, University of California, Berkeley. Pater, Joe. 2000. Nonuniformity in English secondary stress: the role of ranked and lexically specific constraints. Phonology 17: 237–74. Pater, Joe. 2010. Morpheme-specific phonology: constraint indexation and inconsistency resolution. In Steve Parker (ed.), Phonological argumentation: essays on evidence and motivation (Advances in Optimality Theory). London: Equinox Publishing. Rubach, Jerzy and Geert Booij. 2001. Allomorphy in optimality theory: Polish iotation. Language 77: 26–60. Stehl, Thomas 1980. Die Mundarten Apuliens. Historische and strukturelle Beiträge. Münster: Ascendorf. Stevens, Kenneth, Samuel J. Keyser and H. Kawasaki. 1986. Toward a phonetic and phonological theory of redundant features. In Perkell, J. and D. Klatt (eds.), Invariance and variability in speech processes, 426–449. Hillsdale, N.J.: Lawrence Erlbaum. Tranel, Bernard. 1996. French liaison and elision revisited: A unified account within Optimality Theory. In Claudia Parodi, Carlos Quicoli, Mario Saltarelli, and María Luisa Zubizarreta (eds.), Aspects of Romance linguistics: selected papers from the Linguistic Symposium on Romance Languages XXIV, March 10–13, 1994, 433–455. Washington, DC: Georgetown University Press. Tranel, Bernard. 1998. Suppletion and OT: on the issue of the syntax/phonology interaction. In Emily Curtis, James Lyle, and Gabriel Webster (eds.), WCCFL16: The proceedings of the Sixteenth West Coast Conference on Formal Linguistics, 415–429. Stanford, CA: CSLI Publications. Trubetzkoy, Nikolai S. 1969. Principles of Phonology. University of California Press, B ­ erkeley. English Translation of Grundzge der Phonologie, Travaux du Cercle Linguistique de Prague, 7, 1939). Valente, Vincenzo. 1975. Puglia. Pisa: Pacini. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language and Linguistic Theory 23: 917–989. Zoll, Cheryl. 1998. Parsing below the level of the segment in a constraint-based framework. Stanford, CA: CSLI.

Stefano Canalis

Metaphony in the Ticino Canton and phonological features*

Università di Bologna

Abstract: The goal of this paper is to describe a (morpho)phonological process attested in the Romance dialect of the Ticino Canton (Switzerland) and to discuss its relevance to theories of phonological features. In this dialect all stressed front vowels are raised stepwise, and stressed back vowels are fronted. The actual nature of the phonological feature(s) involved is not obvious and provides relevant material to compare the predictions of different feature theories. Whereas standard binary features represent Ticino metaphony as the accidental combination of various unrelated processes, feature systems representing vowel height as a multivalued feature, as well as the phonological primes of Dependency/ Particle/Government Phonology, can describe its alternations as a uniform process.

1 Introduction Metaphony in I­talo-­Romance dialects is a w ­ ell-­known and widely studied phenomenon, yet some of its aspects still remain to be fully investigated and understood. One of the reasons is that ‘metaphony’ is an umbrella term covering a family of similar and historically related processes rather than a single phonological process; given the significant amount of variation across I­talo-­Romance dialects, metaphonic processes also display a fairly high level of ­inter-­dialectal variation, and some of them are less studied than others. In most dialects metaphony is a process raising stressed ­mid-­high vowels to high (and often also affecting ­mid-­low vowels, which are raised to ­mid-­high or diphthongized) if the w ­ ord-­final unstressed vowel is high (/i/ or /u/). However, many varieties in ­north-­western Italy display the more or less productive remnants of a partially different process. The main differences are as follows: first, only w ­ ord-­final /i/ caused metaphony; second, stressed vowels alternate without a w ­ ord-­final unstressed vowel (still) being present. Third, most if not all vowels undergo meta­ etaphonic dialects of phony (including /a/, which is seldom raised in the other m

* I would like to thank the audience at the Meertens Institute Workshop on Metaphony— especially Ben Hermans, Joan Mascaró, Marc van Oostendorp, and Markus Pöchtrager—as well as an anonymous reviewer for their valuable comments and suggestions.

128 

 Stefano Canalis

Italy). Fourth, these a ­ lternations not only involve the raising of ­non-­high vowels, but also fronting of back ones; finally, metaphony also affects front rounded vowels, which are unattested in the vowel inventories of the other ­Italo-­Romance dialects. Since this metaphonic process creates a large and complex set of vowel alternations, it provides an excellent empirical basis to test theories of phonological features; hence, the main subject of this paper will be the phonological nature of the metaphonic alternations, with the goal of understanding which phonological features are involved. Three different (families of) theories of phonological features will be compared, and their relative merits discussed.

2 Properties of Ticinese metaphony Before discussing phonological features in detail, in this section I will provide a brief description of some general aspects of this metaphonic process, and succinctly present some theoretical assumptions whose detailed motivation would lie outside the main focus of this paper.

2.1 Diachronic origin and geographical distribution As mentioned above, this process is attested in some varieties of the dialects of ­north-­western Italy—the so called ‘­Gallo-­Italic’ dialects (Piedmontese, Lombard, Ligurian, Emilian, Romagnolo). Diachronically, stressed vowels became falling diphthongs with a front glide if the ­word-­final vowel was an unstressed /i/ (which usually was a verbal or nominal inflectional morpheme). Later on, w ­ ord-­final unstressed /i/ was lost (as many ­word-­final unstressed vowels in ­Gallo-­Italic), and still later the diphthongs coalesced into a simple vowel, usually more raised and fronted than the nucleus of the previous diphthong. The first stage can be observed in Medieval texts (e.g. laron ‘thief’ vs. laroin ‘thieves’, can ‘dog’ vs. cayn ‘dogs’ in Old Ligurian [Rohlfs 1966: 103]). Some varieties still show an intermediate stage between diphthongization and coalescence, with both diphthongs and coalesced vowels attested; for example a variety spoken in the Canavese area (Piedmont) has [ʧiˈtɔt] sg vs. [ʧiˈtɔi ̯t] pl (gloss not reported, but the word probably means ‘little/small/child’), [kulp] ‘blow-sg’ vs. [kui ̯p] ‘pl’ and [tyt] ‘all/ entirely-sg’ vs. [tyi t̯ ] ‘pl’, but [to] ‘your-sg’ vs. [tø] ‘pl’, [gat] ‘cat-sg’ vs. [gɛt] ‘pl’, [au̯t] ‘tall/high-sg’ vs. [ɛu̯t] ‘pl’, etc. (Rossebastiano 1995). However, the most common pattern is diachronic coalescence of all (or nearly all) diphthongs. Over time, in most dialects metaphonic alternations have been levelled, generally in

Metaphony in the Ticino Canton and phonological features 



 129

favor of the ­ non-­ diphthongized/­ non-­ metaphonized alternant, so that many modern dialects present few if any traces of metaphony: “[t]he number of lexical exceptions to metaphony in northern Italy seems everywhere to expand through history” (Maiden 1991: 198). One of the areas in which metaphony is best preserved is actually not in Italy but in Switzerland, in the Ticino Canton (in which a ­Gallo-­Italic dialect similar to Lombard is spoken). Even in this area, several contemporary varieties have very few traces of metaphony, and it is difficult to find a fully regular process. However, the fact that metaphonic alternants can be found in virtually all words in one dialect or another suggests that, at least in the past, metaphony was a (nearly) systematic process; consequently, the empirical basis of this paper will not be a contemporary variety, but the detailed description of metaphony in the dialects north of the Maggiore lake provided by Salvioni (1886); more specifically, I will mainly focus on the dialect of the small town of Cevio (now having a population of about 1200), in the Maggia valley, in which the process seems to be (or have been) most productive and about which most data are available. In any case, it must be pointed out that even in Salvioni’s data sporadic lexical exceptions exist, and a recent investigation on metaphony in the same area (Cecconello 2009) shows that it has receded further, although metaphonic alternations are still numerous.

2.2 Data and examples The vowel system of the Cevio dialect is presented in (1). (1)  i

y

u e

ø ɛ

ɔ a

It must be pointed out that this is the system of stressed vowels; in unstressed position the contrast between /ɛ/ and /e/ is neutralized. Salvioni also states (p. 208) that /ɔ/ (or /ɔ/ and /o/, in the varieties having both) is reduced to [u] in unstressed vowels (e.g. [ˈmɔvi] ‘I move’ vs. [muˈvɛva] ‘­move-­­pst.­­ipfv.­3sg’) “in most of the region” (“per gran parte della regione”)—although, somehow inconsistently, in several of his examples he uses [o] to represent unstressed vowels. In the Maggia valley unstressed /ɛ e i/ turn to [a] with an ‘almost ­law-­like tendency’ (“tendenza [...] prossima a diventare una legge” ibid: 206;

130 

 Stefano Canalis

e.g. [ˈbɛvi] ‘I drink’ vs. [baˈvy] ‘­drink-­­pst.­p tcp’). Stressed [o] is attested in the Cevio dialect, but a ­ pparently just as an allophone of /ɔ/, occurring only before nasal consonants in coda position (e.g. [from] ‘strawberry’, [bom] ‘good’, ibid.: 201). In other areas however /o/ is an independent phoneme along with /ɔ/. Metaphony affects stressed vowels, for example modifying them in the plural forms of nouns and adjectives; in (2) all the attested alternations in nominal inflection are provided, together with some examples. In general, front vowels undergo scalar raising (although /ɛ/ ‘skips’ one step, as it alternates with /i/) and back vowels are fronted. As already mentioned in § 1, diachronically metaphony was triggered by a w ­ ord-­final unstressed /i/, but an overtly present /i/ no longer is the trigger of the process. Actually, this vowel is still sometimes present in some metaphonic alternants (e.g. in [ˈʃtymi] ‘stomachs’), but it is neither a sufficient nor a necessary condition to metaphony: first, ­word-­final /i/ may be present in ­non-­metaphonic alternants as well (e.g. in [ˈʃtumi] ‘stomach’); second, the unstressed vowel following the metaphonized one may be /a/ (e.g. in [ˈlɛras] ‘larches’ ), or /u/ (e.g. in [ˈtɛvul] ‘tables’), or be absent altogether (e.g. in [aniˈmɛl] ‘animals’). (2) /a/ → /ɛ/ /ɛ/ → /i/ /e/ → /i/ /u/ → /y/ /ɔ/ → /ø/

Singular [ˈtavul] [aniˈmal] [ˈlaras] [rɛd] [vɛrd] [ˈdebul] [fraˈdel] [aˈlegar] [kuˈlur] [ˈʃtumi] [mɔl] [kɔrp] [ˈkrɔvat]

Plural [ˈtɛvul] [aniˈmɛl] [ˈlɛras] [rid] [vird] [ˈdibul] [fraˈdil] [aˈligar] [kuˈlyr] [ˈʃtymi] [møl] [kørp] [ˈkrøvat]

‘table’ ‘animal’ ‘larch’ ‘net’ ‘green’ ‘weak’ ‘brother’ ‘merry, happy’ ‘color’ ‘stomach’ ‘soft’ ‘body’ ‘fir’

The only stressed vowels not undergoing metaphony are /i/, /y/, and /ø/. However, stressed /ø/ alternates with /y/ in other varieties, for example in ­Menzonio (3). (3) Singular [trøɲ] [ʃpørc]

Plural [tryɲ] [ʃpyrc]

‘thunder’ ‘dirty’



Metaphony in the Ticino Canton and phonological features 

 131

The alternation /a/ → /i/ is also attested (in Cevio as well as in other dialects), but only if the stressed vowel is in a closed syllable and is followed by a nasal consonant (4). (4) /a/ → /i/

Singular [grand] [ˈtantu]

Plural [grind] [ˈtinti]

‘big’ ‘much’

In verbal morphology metaphony is very similar, with only a minor difference. /ɛ/ alternates with /i/ in some verbs—just as in nominal inflection—but it alternates with /e/ in others. Metaphonic alternants are very common in the 2nd person plural form of verbs (5), which had—and frequently still has—a ­word-­final /i/. (5) /a/ → /ɛ/ /ɛ/ → /e/ /ɛ/ → /i/ /e/ → /i/ /u/ → /y/ /ɔ/ → /ø/

1st sg.prs.ind 2nd sg.prs.ind [ˈvali] [ˈvɛli] [imˈpari] [imˈpɛri] [ˈparla] [ˈpɛrla] [ˈbɛvi] [ˈbevi] [pruˈmɛti] [pruˈmeti] [ˈɛvi] [ˈivi] [parˈlɛvi] [parˈlivi] [seg] [sig] [vend] [vind] [risˈpund] [risˈpynd] [adˈɲus] [adˈɲys] [ˈtrɔvi] [ˈtrøvi] [ˈgɔdi] [ˈgødi]

‘to be worth’ ‘to learn’ ‘to speak’ ‘to drink’ ‘to promise’ ‘to ­have-­­pst-­inpf’ ‘to ­speak-­­pst-­inpf’ ‘to saw’ ‘to sell’ ‘to answer’ ‘to know’ ‘to find’ ‘to benefit/to delight’

2.3 Phonology, morphology or both? As mentioned above, metaphony is not fully regular. Some nominal plural forms as well as some verbs do not have a metaphonic alternant (e.g. [ˈnɛti] (no gloss) is both the 1st and the 2nd person of a verb). Furthermore, metaphonic alternants are always the phonological exponents of morphological categories (as number, or person), and they are not (or, more precisely, no longer are) the result of an assimilatory process since the stressed vowel does not assimilate its height or backness to any vowel in the same word—indeed, metaphony often occurs in monosyllables, as several examples in (­2–­5) show. These facts raise the question of the status of Ticinese metaphony; is it a phonological,

132 

 Stefano Canalis

morphological, or morphophonological process? What is its relationship with morphology? Clearly, the issue of the phonology/morphology interface cannot be solved in a single paper, and the main focus of my contribution is on phonological features rather than this topic. I will limit myself to point out that, although not fully systematic, metaphony is a very common alternation in Ticinese, and even today it occurs even in fairly recent loanwords from Italian (as [koˈriʤ] ‘you correct’ vs. [koˈrɛʤ] ‘I correct’, from Italian correggere ‘to correct’), at least for some speakers (Cecconello 2009: 47). Moreover, although the available data show exceptions, the process had a clear phonetic motivation (the presence of a ­word-­final /i/) and probably it was more regular in the past, as comparison with closely related dialects suggests. Therefore, even assuming that it may already have been a phonologically moribund process in the data discussed here, it is likely that it was a regular purely phonological process in ­non-­directly observable previous synchronic states. Furthermore, even adopting this moderately conservative position—i.e. that metaphony as attested in the available data is not a purely phonological ­phenomenon—the phonological properties of the alternations remain a relevant issue. Even if Ticinese metaphony is primarily seen as a mechanism to express some nominal or verbal categories, with the lexicon specifying which roots undergo metaphony and which roots use other morphological exponents (or no exponent at all as in many cases singular and plural forms are identical), the exponent itself is made of phonological material, and therefore needs to be specified in phonological terms. While affixation is the addition of a certain phoneme or phoneme cluster to a lexical root, processes like Ticinese metaphony change (assuming binary features) a certain phonological feature or set of features in the lexical root—or more specifically in its stressed vowel.1 Insofar as these alternations are not phonologically arbitrary pairs of vowels but are phonologically motivated (as I will argue in section 3), a description in terms of phonological features is not only possible but even necessary.

2.4 Absence of the ­word-­final unstressed vowel An issue related to the ­morpho-­phonological status of Ticinese metaphony discussed in the previous section is the absence of a ­word-­final unstressed vowel after the alternating vowel; unlike most other metaphonic processes in other

1 Or add it/them to the root, if monovalent features are adopted—see section 3.3.

Metaphony in the Ticino Canton and phonological features 



 133

Italian dialects,2 a ­word-­final vowel which acts as trigger of the process is usually absent. This may be interpreted as a further argument to argue that the process is morphologized, since it is not the output of an assimilatory process. However, a phonological feature (or any phonological unit, more generally) need not be associated with a segmental unit to exist and spread to segmental positions. Indeed, several analyses of umlaut in Germanic languages—e.g. Lieber (1987), Lodge (1989), Wiese (1996), Anderson (2005)—have assumed a floating/extrasegmental feature as the phonological exponent of umlaut. Ticinese metaphony is broadly similar to umlaut: stressed vowels alternate in certain morphological environments, both raising and fronting occur, and the alternation is no (longer) due to assimilation to an unstressed vowel. In the following sections I will adopt a similar assumption; I will represent Ticinese metaphony as the docking of a floating phonological feature—which is therefore assumed to be a phonological exponent of certain morphological categories, as ‘plural number’ in nouns and adjectives—on the stressed vowel, as in (6). As the main goal of this paper is to understand which feature(s) is (are) involved, for the time being the metaphonic feature is simply indicated as [F], to be understood as a mere ­non-­committal diacritic before its nature is discussed in the following sections. (6)  Exponent

/g a t/ + [F]

→

C V C Meaning

‘cat’ + pl

/g a t/ + [F] → [gɛt] C V C

→

‘cats’

2.5   Transparency of ­word-­final unstressed vowels The assumption discussed in section 2.4 is not sufficient to account for all cases of metaphony. In some words a vowel is present after the stressed vowel, but it does not undergo metaphony (the relevant examples in (2) are repeated in (7)). (7) Singular [ˈtavul] [ˈdebul] [ˈʃtumi]

Plural [ˈtɛvul] [ˈdibul] [ˈʃtymi]

‘table’ ‘weak’ ‘stomach’

2 However, in southern Italy unstressed /i/ and /u/ have turned to schwa in several dialects, so that metaphonic alternations like [ˈrossa] ‘red-f.sg’ vs. [ˈrussə] ‘m.sg’ < *[ˈrussu] < *[ˈrossu] no longer have an overt [+high] trigger.

134 

 Stefano Canalis

Given a standard autosegmental representation like that in (6), if metaphony is triggered by a floating feature following a lexical root, it should be subject to phonological locality conditions; the feature(s) would be predicted to dock on the first available vowel, or on both the final and the stressed vowel, but they should not skip one vowel. Therefore the plural of e.g. the word [ˈtavul] should be *[ˈtavyl] (8) or *[ˈtɛvyl] (9) rather than the actual [ˈtɛvul]. In [ˈtɛvul], apparently, one of the segmental slots on which the floating feature could dock is skipped (10). (8) /t a v u l/ + [F] → /t a v u l/ + [F] → *[ˈtavyl] C V C V C

C V C V C

(9) /t a v u l/ + [F] → /t a v u l/ + [F] → *[ˈtɛvyl] C V C V C

C V C V C

(10) /t a v u l/ + [F] → /t a v u l/ + [F] → [ˈtɛvul] C V C V C

C V C V C

However, vowel transparency is not unattested in vowel harmony systems. More specifically, transparency in Ticinese metaphony is probably related to the broader problem of transparency in ­stress-­dependent harmonic systems; in some Romance metaphonic systems vowel raising skips the penultimate vowel of proparoxytones,3 which behaves as a transparent segment (see e.g. Hualde 1989, Walker 2010).4 Discussing the motivation for this type of transparency would require a paper in itself; we will simply assume that the representation in (6) has to be partially modified, and the domain of Ticinese metaphony is prosodically defined (not an unreasonable assumption as its target is the stressed vowel): the floating feature docks on the closest metrically most prominent syllable node (i.e. the stressed vowel) (11), not on the segmentally closest vocalic slot.

3 Indeed, diachronically Ticinese words as [ˈtavul], [ˈdebul] or [ˈʃtumi] were proparoxytones before their ­word-­final vowel was lost. 4 In other Romance dialects metaphony is blocked in proparoxytones, in still others also the penultimate vowel is raised.

Metaphony in the Ticino Canton and phonological features 



(11)    /t a v u l/ + [F]

→ /t a v u l/ + [F]

C V C V C

C V C V C

* *

 135

→ [ˈtɛvul]

* *

*

*

2.6 Aspects of the alternations If it is to be described informally, Ticinese metaphony is a remarkably clear and simple phonological process: front vowels undergo scalar raising and back vowels are fronted. A way to sum up this pattern is to say that stressed vowels subject to metaphony become one step more similar to /i/ than they are in their ­non-­metaphonic alternant ((12); as mentioned above, the alternation /ø/ → /y/ is not attested in the Cevio dialect, but it is in other metaphonic Ticinese dialects). (12) i y

u e

ø ɛ

ɔ

a   As the examples in (2) and (4) show, sometimes raising is not perfectly scalar; it ‘skips’ one step in nominal inflection (in which /ɛ/ becomes /i/), and two steps if the environment is /aNC/ (metaphonized /a/ becomes /i/ if followed by a nasal consonant and another consonant). Both ‘exceptions’ are probably more apparent than real. As for the first, the /ɛ/ ~ /e/ phonological contrast is ‘weaker’ than the others in the Cevio dialect: it is neutralized in unstressed position (in which only [e] occurs), is asymmetric (the contrast between mid-low and mid-high vowels being absent in back vowels), and few minimal pairs exist. Therefore the possibility of changing both /ɛ/ and /e/ to /i/ seems to follow from the similarity between the two vowels. As for the change from /a/ to /i/ before a nasal coda, in this environment midfront vowels are never attested (Salvioni 1886: 1­ 99–­2 00), so the exception is only apparent.

136 

 Stefano Canalis

3 Which features to use? 3.1 Binary features Once these ‘exceptions’ are explained, the general pattern of raising and fronting is quite straightforward. Yet this simple and intuitive descriptive generalization is lost if the standard features of SPE (= Chomsky & Halle 1968)—or more recent developments of this theory, as Halle, Vaux & Wolfe (2000)—are adopted to represent the Ticinese vowel system and Ticinese metaphony. A fundamental requirement for a theory of phonological features is to be able to capture phonological classes and express phonological processes, but traditional binary features fail to represent Ticinese metaphony as a coherent process. The Cevio vowel system is represented in (13) using binary features (feature specifications not relevant here omitted). (13)

/i/ [+high, −low, +ATR, −back, −round] /y/ [+high, −low, +ATR, −back, +round] /u/ [+high, −low, +ATR, +back, +round] /e/ [−high, −low, +ATR, −back, −round] /ø/ [−high, −low, −back, +round] /ɛ/ [−high, −low, −ATR, −back, −round] /ɔ/ [−high, −low, +back, +round] /a/ [−high, +low, −ATR, −back, −round]

This feature system only allows for three different vowel heights ([−high, +low], [−high, −low] and [+high, −low), therefore a further feature is needed to capture the contrast between /ɛ/ and /e/. This feature could be [ATR] or [tense]; in either case, the use of such features to represent the contrast between mid-low and midhigh in Ticinese (and in Romance languages more generally) can be disputed, as it is not obvious that the difference between these vowels is primarily a matter of tenseness, or advancement of the tongue root, rather than a step in the gradual increase from low to high—but see Calabrese (2007) for arguments in favor of the use of [ATR] to describe Romance vowels. Accepting [ATR] or [tense] as valid ­features for Cevio vowel system, the features required to account for the metaphonic alternations are reported in (14). (14) /a/ → /ɛ/ = [+low] → [−low] /ɛ/ → /e/ = [−ATR] → [+ATR] (or [−tense] → [+tense]) /ɛ/ → /i/ = [−high] → [+high] and [−ATR] → [+ATR] (or [−tense] → [+tense]) /e/ → /i/ = [−high] → [+high] /u/ → /y/ = [+back] → [−back]



Metaphony in the Ticino Canton and phonological features 

 137

/ɔ/ → /ø/ = [+back] → [−back] /ø/ → /y/ = [−high] → [+high] (in the dialects having this alternation) Four different floating features are needed to represent metaphonic alternations: [−low], [+ATR] (or +[tense]), [−back], [+high]. But the problem does not lie in the sheer number of features needed; more importantly, it is not possible to represent all alternations as the result of floating [−low, +ATR, +high, −back] docking on a stressed vowel. The metaphonic outputs /i/ and /y/ are compatible with this set of features; however, the change from /a/ to /ɛ/ cannot involve [+ATR] and [+high]; the change from /ɛ/ to /e/ cannot involve [+high]; neither can the change from /ɔ/ to /ø/. So Ticinese metaphony should be split into three different processes (15). (15) Alternations Floating features needed 1) /ɛ/ → /i/, /e/ → /i/, /u/ → /y/, /ø/ → /y/ [−low], [−back], [+ATR] (or +[tense]), [+high] 2) /ɛ/ → /e/, /ɔ/ → /ø/ [−low], [−back], [+ATR] (or +[tense]) 3) /a/ → /ɛ/ [−low], [−back] SPE phonological features are thus unable to represent Ticinese metaphony as a coherent phonological process; three unrelated changes have to be assumed. Various solutions may be proposed to solve this problem, but none of them is entirely convincing. First, the simplest answer is that the SPE feature system is right: it could be argued that it is correct in positing three different changes, simply because what seems to be a single phonological process could actually be the accidental partial overlapping of three different processes. However, it is difficult to find any independent reason to consider the alternations of Ticinese metaphony as unrelated to each other. They occur in the same phonological environment (i.e. stressed vowels); they occur in the same morphological environments (nominal and verbal inflection); they can all be described as floating features; they have a common diachronic origin, and synchronically their phonological nature (fronting and scalar raising) is clear and shared by all alternants. Finally, in the many varieties in which metaphony is retreating, all alternations are receding; if there were three independent processes, it would be likely for them to have different rates of change and loss in different varieties. Second, one could suppose that in the Ticinese lexicon each root undergoing metaphony is lexically specified for the set of floating features it needs to realize phonologically its inflectional categories; thus /gat/ would select the floating feature [−low] to realize its plural, /mɔl/ would select [−back], and so on. Indeed, a broadly similar proposal has been advanced for a different case of metaphonic scalar raising; in an Asturian dialect stressed /a/ is raised to /e/ if

138 

 Stefano Canalis

the ­word-­final vowel is high ([ˈgetu] ‘male cat’ vs. [ˈgata] ‘female cat’), while /e/ and /o/ are raised to /i/ and /u/ respectively ([kalˈdiru] ‘pot’ vs. [kalˈderos] ‘pots’, Hualde (1989: 791)). Wolf (2006: 3­ 1–32) posits two different floating autosegments as allomorphs of the morphological feature [plural], depending on the quality of the stressed vowel: the floating autosegment would be [−low] if the stressed vowel is /a/, and [+high] if it is /e/ or /o/. This solution, although descriptively adequate, is ultimately close to a diacritic device, and circular: specification in the lexicon ‘explains’ why a certain root selects a certain set of floating features only because we know which set of floating features each root must be associated with. Third, it could be concluded that the failure of the SPE feature system to represent Ticinese metaphony is not a real failure, but a consequence of the ­non-­phonological nature of this process. While feature systems must be able to describe phonological classes and processes, ­ morphologically ­ induced phoneme changes are not subject to the same restrictions; therefore the failure of a feature theory to represent a process in phonological terms may simply reflect the fact that the process in question is not phonological. Against this thesis, it is worth repeating that there is a clear pattern of raising and fronting; therefore, before concluding that the alternations are not phonological, it is useful to test whether other theories of phonological features are able to better capture this ­generalization.

3.2 Scalar height The main problem with an analysis of Ticinese metaphony in terms of standard binary features lies in the representation of vowel height; as the representation of height is assigned to three different features ([high], [low] and [ATR]/[tense]), scalar raising (or lowering) is difficult to represent. This problem is well known, and several alternative models have been put forward to solve it. In general, they abandon the assumption of the binarity of phonological features, and assume ­ ulti-­valued scalar dimension. Various reasons to do so have been that height is a m suggested: among the most important, languages exist with four (and possibly five) vowel heights which do not seem to differ in terms of the features [tense] or [ATR]; the phonetic correlate of vowel height is uniform (height is strongly correlated with the inverse of the frequency of the first formant); perhaps most importantly, fairly numerous scalar lowering or raising processes exist. Among the various proponents of a scalar height feature, Ladefoged (1975, 2006), Lindau (1978), Clements (1991), Clements & Hume (1995), Ladefoged & Maddieson (1996), Parkinson (1996), and Gnanadesikan (1997) may be mentioned.



Metaphony in the Ticino Canton and phonological features 

 139

Obviously, their theories differ in several aspects and details. For instance, Ladefoged (1975, 2006) interprets height as a multivalued feature with ordered values: [High1], [High2], [High3]... [Highn] (or [low], [­mid-­low], [mid], [­mid-­high], [high]). Clements and Hume (1995) propose several tiers of [±open] features; thus /a/ is represented as [+open1, +open2], /e, o/ as [−open1, +open2], /i, u/ as [−open1, −open2], and so on. Gnakadiesan (1997) uses ternary features, while Parkinson (1996) uses multiple occurrences of a monovalent feature [closed]. However their differences are not relevant to the present discussion; they can all represent stepwise vowel raising straightforwardly. Back stressed vowels also become front in Ticinese metaphony, so another feature is needed; according to these models the floating features are two (for example [High+1, −Back] in Ladefoged’s model or [−open, Coronal] in Clement’s model). Still, a general problem with these feature theories is that ­cross-­linguistically the attested upper limit of vowel height contrasts seems to be five, yet nothing in these theories rules out the possibility of more vowel heights, e.g. [High9] or [−open7]. Furthermore, they often mix ­n-­ary, binary and unary features in their feature systems.

3.3   Components and dependency relations A scalar vowel height feature avoids the problems encountered by standard binary features in the representation of Ticinese metaphony. A further alternative to the latter is provided by the theories of Dependency (among others Anderson & Jones 1974; Anderson & Ewen 1987; Anderson 2002), Government (among others Kaye et al. 1985), and Particle (among others Schane 1984) Phonology. Features like those discussed in the sections above—be they binary or ­n-­ary—compose phonemes, but they are not realizable in isolation; they are only realizable when combined with other features. Instead, these three theories assume that (as far as vowel place is concerned) only three atomic components5 exist, which are both independently interpretable as phonemes, and parts of more complex phonemes. These components are i, u, and a6. Their phonetic interpretations are palatality, labiality/roundness, and openness, respectively (or, in acoustic

5 More specifically, they are usually called components in Dependency Phonology, elements in Government Phonology, and particles in Particle Phonology. 6 Adopting a slightly simplified version of the notational conventions of Dependency Phonology, henceforth components will be represented by bold characters, while segments made of such components will be represented by curly brackets.

140 

 Stefano Canalis

terms, r­ espectively, predominance of energy in the higher part of the spectrum— acuteness—, predominance of energy in the lower part of the spectrum—gravity and flatness—, and concentration of energy in the central part of the spectrum— compactness, sonority). Unlike binary features, which have a positive and a negative value, these components have just one value: they can be either present or absent, implying that the negative value of a component is not a phonological entity. When alone, the three components are realized as the extremes of the vocalic triangle, [i], [u], and [a]. Their combinations give rise to all the other vowels: thus in a vowel system /a e i o u/, /o/ is {a,u}, /e/ is {a,i}. In a vowel system including also /y/ and /ø/, /y/ is {i,u}, /ø/ is {a,i,u} (where the order of components in the notation is not relevant: {u,a} and {u,i} would still represent /o/ and /y/, respectively, since the two components in each segment are equally important). The underlying idea is that all vowels other than /a i u/ are composed of the fusion of the properties of the three most basic vowels. For example, /o/ is seen as the result of both roundness and aperture mixed together; it is neither purely rounded nor purely open, but a fusion of the two dimensions. Likewise, /y/ is the ‘sum’ of rounding and palatality. This inventory of components is empirically motivated by several widely attested ­cross-­linguistic phenomena: for example they are consistent with the typological observation that /a i u/ are the most common vowels and form the most basic vowel system7; even in more complex vocalic systems /a i u/ frequently are the only vowels admitted in unstressed position (for example Sicilian allows /a e i o u/ in stressed vowels, but the unstressed vowels neutralize contrasts between high and mid vowels, reducing the inventory to /a i u/). This representation also predicts that vowels composed of two or more components imply the presence of the vowels corresponding to the components they are literally made of; indeed, almost any language having /y/ or /e/ has /u/ and /i/, or /a/ and /i/, as well, but the reverse is very rarely true. Besides vowel inventories, also the possible processes predicted by this theory are empirically motivated. For example vowel coalescence of /aj/ into /e/ and /aw/ into /o/ is common c­ ross-­linguistically, both as a synchronic alternation and as a diachronic change. Many ­non-­occurring or very rare processes, which could nonetheless be expressed by binary features, are ruled out by this notation. For instance fronting of back rounded vowels triggered by a front vowel (u > y / _C0i) is a common process, whereas derounding of a back rounded vowel triggered by a front unrounded vowel (u > ɯ / _C0i) is rarely if ever attested. Both processes are equally possible adopting binary 7 Apart from the few languages only having two vowels, usually /a/ and a higher vowel.



Metaphony in the Ticino Canton and phonological features 

 141

­features, since /i/ is both [−back] and [−round]; but using unary components /i/ is simply {i}; therefore in Dependency Phonology only palatality can spread to /u/, while non-roundness does not exist phonologically. Likewise, in many languages front vowels palatalize consonants, but back vowels rarely if ever velarize them. A further departure from traditional binary features (as well as from most other feature theories) is the idea that also below the segment there exist asymmetric ­head-­dependent relations among components.8 Just as in a foot the stressed syllable is more prominent than the unstressed one, and in a ­syllable the nucleus is more prominent than onset and coda, similarly within a segment one component can be in a more prominent role (the head) than the other(s) (the dependent(s)). It is true that in the two former cases we have a relation among elements which manifest themselves as a word unfolds in time—­syllables within a foot, segments within a syllable—while subsegmental dependencies take place among simultaneous components; but on a more abstract level, all segments are assumed to have a head element more prominent than the other components. In fact, different degrees of prominence are formally necessary if we want to be able to represent every phonological contrast; mere ­co-­occurrence of components, without asymmetric relations among them, can represent only some of them. For example it can represent three levels of height at most: /a/ {a}, /e/ {a,i}, /i/ {i}. But if a language displays a contrast between ­low-­mid and ­high-­mid vowels, as /ɛ/ vs. /e/, such a contrast cannot be represented by means of the presence of a component in one of the segments and its absence in the other vowel, as both vowels include i and a. However, if dependency relations among components are available in this formal apparatus, given that /ɛ/ is more open than /e/, we can say that a dominates i in the former vowel, and the reverse relation takes place in the latter (notationally, if in a segment a component x dominates y it will be represented as {x;y}, so that /ɛ/ is {a;i} and /e/ is {i;a}; more generally, vowel height is inversely correlated with the preponderance of a). Since the head component is defined as the component relatively more prominent/salient than the other(s), by this very definition a segment can only have one head, and the head component must differ from the dependent component(s) (to put the latter statement differently, {x,x} and {x;x} are considered to be merely redundant notational variants of {x}).

8 In Dependency and Government Phonology, but not in Particle Phonology; in that theory degrees of vowel height are represented with multiple occurrences of the a component, much like in Parkinson (1996).

142 

 Stefano Canalis

These general principles define the set of possible phonological contrasts. However, no language fully exploits all these possibilities; rather, every language possesses only a subset of the theoretically possible vowel contrasts. This means that ­language-­specific licensing constraints on the combination of components (Charette & Göksel 1996; Kaye 2001) are necessary to account for the Cevio vowel system. In this dialect u cannot enter into asymmetric relations with other components (cf. (1)), so the constraints *{u;} and *{;u} are active. Since only a and i can entertain asymmetric relations in the Cevio dialect, their relationship with each other is assumed to be tighter than with u, and hence the mid-front rounded vowel /ø/ is interpreted as {(a,i),u} rather than simply {a,i,u}. From the assumption of dependency relations it also follows that, while spreading of a binary feature can only change a feature from its positive to its negative value (or the other way round), adding a component to a vowel may not only insert a previously absent element, but also change its relative prominence within the vowel. The possible minimal changes caused by a new component are three. If a component x is absent, it becomes a dependent element: {y} → {y;x} (or simply {x,y}, if an asymmetric relation is not contrastive for that combination of components)9. If a component is already present as the least salient one, it switches to head: {y;x} → {x;y}. If it is already present as the most salient element, the only way to further promote it is to make it the only element in the segment: {x;y} → {x}10.

3.4 Ticinese metaphony in Dependency Phonology The representation of the Cevio vowel inventory according to Dependency Phonology is shown in (16), while the metaphonic alternations are shown in (17). (16) /i/ {i} /y/ {i, u}

/u/ {u}

/e/ {i ; a} /ø/ {(a, i), u} /ɛ/ {a ; i}

/a/ {a}

/ɔ/ {a, u}

9 In general, at least if a phonological process in a certain language is ­structure-­preserving, ‘minimal change’ in the combination of components is to be understood as ‘minimal change consistent with the underlying component combinations allowed in that language’s phonological inventory’. This means that a change violating universal or l­anguage-­specific licensing constraints would be ‘repaired’ (broadly in the spirit of Calabrese (2005)) to comply with them. 10 This alternation can be seen as x filling the only structural role not yet occupied in {x;y}, therefore creating a segment {x;x}, which is automatically simplified to {x}—see above.



(17) /a/ → /ɛ/ /ɛ/ → /e/ /ɛ/ → /i/ /e/ → /i/ /u/ → /y/ /ɔ/ → /ø/ /ø/ → /y/

Metaphony in the Ticino Canton and phonological features 

 143

{a} → {a ; i} {a ; i} → {i ; a} {a ; i} → {i} {i ; a} → {i} {u} → {i, u} {a, u} → {a, u, i} {(a, i), u} → {i, u}

The whole process can be uniformly expressed as a floating autosegment i that minimally increases the preponderance of the i component in the stressed vowel: i changes from absent to present as dependent (in the alternation /a/ {a} → /ɛ/ {a;i}), or merely to present (in the alternations /u/ {u} → /y/ {u,i} and /ɔ/ {a,u} → /ø/ {a,u,i}, since the relative prominence of i with respect to u is not contrastive in this vowel system); from dependent to head (in the alternation /ɛ/ {a;i} → /e/ {i;a}); from being the head to being the only component (in the alternation /e/ {i;a} → /i/ {i}), or from being present together with a to being the only height component (in the alternation /ø/ {(a, i), u} → /y/ {i, u}—the minimal change would be {(i; a), u}, but since such a combination would violate a ­language-­specific ­constraint—the contrast /œ/ ~ /ø/ does not exist in Ticinese—it is ‘repaired’ to the most similar licit output increasing i, i.e. {i, u}). In some environments both /ɛ/ and /e/ have /i/ as their metaphonic alternant. As discussed in section 2.6, this is possibly due to the peculiar, ‘weak’ status of the /ɛ/ ~ /e/ contrast; if this hypothesis is correct, both mid-front vowels can be represented as {a, i}, and hence the only possible way to increase the preponderance of i is to have /i/ as their metaphonic alternant.

4 Conclusions and final remarks Compared to standard binary features, both multivalued scalar features and theories assuming atomic components and dependency relations offer a more satisfactory account of Ticinese metaphony, as they are able to represent all the alternations as the result of one and the same process. According to the former family of theories, the floating features which realize the exponent of some nominal and verbal morphemes are [high+1] and [−back] (or broadly equivalent features in other theories adopting scalar height features); according to the latter family of theories, the floating element is the component i, which increases its preponderance in the stressed vowel. Unlike in multivalued scalar features, in Dependency Phonology the number of possible vowel heights is limited by the theory: no more than four levels

144 

 Stefano Canalis

(or five, if the dependency relationship of ‘mutual preponderance’ is a possible option—cf. Anderson & Ewen 1987: ­31–­32, ­126–­127) are predicted to exist. A more specific advantage of Dependency Phonology resides in the analysis of varieties (like the one described in section 2.1) in which metaphonic alternations are expressed partly by vowel alternations and partly by diphthongization with a front glide. The process can be straightforwardly described by Dependency Phonology as either fusion with or concatenation to the stressed nucleus of the same i component, whereas the features [high+1] and [−back] could be compatible with any ­non-­low front vowel, not necessarily only non-syllabic /i/. With respect to other problems, the exact relationship between phonology and morphology in Ticinese metaphony, the transparency of the unstressed final vowel in paroxytones, and the mechanism of the process have only been touched upon, and need to be investigated in further research.

References Anderson, John M. 2002. ‘Dependency phonology. Ms., available at http://www.fb10.­ uni-­bremen.de/linguistik/dpng/pdf/Anderson2002IntroDependencyPhonology.pdf. Anderson, John M. 2005. Old English ­i-­umlaut (for the umpteenth time). English Language and Linguistics 9 (2): ­195–­224. Anderson, John and Colin Ewen. 1987. Principles of dependency phonology. Cambridge: ­Cambridge University Press. Anderson, John and Charles Jones. 1974. Three theses concerning phonological representations. Journal of Linguistics 10 (1): ­1–­26. Calabrese, Andrea. 2005. Markedness and economy in a derivational model of phonology. Berlin: Mouton de Gruyter. Calabrese, Andrea. 2007. On the feature [ATR] (and [Tense]) in the Romance languages. In Roberta Maschi, Nicoletta Penello and Piera Rizzolatti (eds.), Miscellanea di studi linguistici offerti a Laura Vanelli da amici e allievi padovani, ­103–­115. Udine: Forum Editrice Universitaria Udinese. Cecconello, Giulia. 2009. La metafonia nel Canton Ticino. MA thesis. Padua: University of Padua. Charette, Monik and Asli Göksel. 1996. Licensing constraints and vowel harmony in Turkic languages. SOAS Working Papers in Linguistics and Phonetics 6: ­1–­25. Chomsky, Noam and Morris Halle. 1968. The sound pattern of English. New York: Harper & Row. Clements, George N. 1991. Vowel height assimilation in Bantu languages. Working Papers of the Cornell Phonetics Laboratory 5: ­37–­76. Clements, George N. and Elizabeth Hume. 1995. The internal organization of speech sounds. In John Goldsmith (ed.), Handbook of phonological theory, 245–­306. Oxford: Basil Blackwell. Gnanadesikan, Amalia Elisabeth. 1997. Phonology with ternary scales. Amherst, MA: University of Massachusetts Amherst dissertation. Halle, Morris, Bert Vaux and Andrew Wolfe. 2000. On feature spreading and the representation of place of articulation. Linguistic Inquiry 31 (3): ­387–­444.



Metaphony in the Ticino Canton and phonological features 

 145

Hualde, José Ignacio. 1989. Autosegmental and metrical spreading in the ­vowel-­harmony systems of northwestern Spain. Linguistics 27 (5): ­773–­805. Kaye, Jonathan. 2001. Working with licensing constraints. In Katarzina ­Dziubalska-­Kołaczyk (ed.), Constraints and preferences, ­251–­268. Berlin: Mouton de Gruyter. Kaye, Jonathan, Jean Lowenstamm and ­Jean-­Roger Vergnaud. 1985. The internal structure of phonological representations: a theory of Charm and Government. Phonology Yearbook 2: ­305–­328. Ladefoged, Peter. 1975. A course in phonetics. New York: Harcourt Brace Jovanovich. Ladefoged, Peter. 2006. Representing linguistic phonetic structure. Unfinished manuscript, http://www.linguistics.ucla.edu/people/ladefoge/PhoneticStructure.pdf (retrieved 14 June 2014). Ladefoged, Peter and Ian Maddieson. 1996. The sounds of the world’s languages. Oxford: Blackwell. Lieber, Rochelle. 1987. An integrated theory of autosegmental processes. Albany: State ­University of New York Press. Lindau, Mona. 1978. Vowel features. Language 54 (3): ­541–­563. Lodge, Ken. 1989. A ­non-­segmental account of German Umlaut: diachronic and synchronic perspectives. Linguistische Berichte 124: ­470–­491. Maiden, Martin. 1991. Interactive morphonology. Metaphony in Italy. London/New York: ­Routledge. Parkinson, Frederick B. 1996. The representation of vowel height in phonology. Columbus, OH: Ohio State University dissertation. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti: Fonetica. Torino: Einaudi. Updated Italian translation of Historische Grammatik der italienischen Sprache und ihrer Mundarten. Lautlehre. Bern: A. Francke. Rossebastiano, Alda. 1995. Prolessi di i e metafonesi nel Basso Canavese. In Emanuele Banfi, Giovanni Bonfadini, Patrizia Cordin, Maria Iliescu (eds.), Italia Settentrionale: crocevia di idiomi romanzi. Atti del convegno internazionale di studi. Trento, ­21–­23 ottobre 1993, ­43–­46. Tübingen: Max Niemeyer Verlag. Salvioni, Carlo. 1886. Saggi intorno ai dialetti di alcune vallate all’estremità settentrionale del Lago Maggiore. I. Annotazioni fonetiche e morfologiche. II. Effetti dell’-i sulla tonica. Archivio Glottologico Italiano 9: ­188–­260. Schane, Stanford. 1984. The fundamentals of particle phonology. Phonology Yearbook 1: ­129–­155. Walker, Rachel. 2010. Nonmyopic harmony and the nature of derivations. Linguistic Inquiry 41 (1): 169–­179. Wiese, Richard. 1996. The phonology of German. Oxford: Oxford University Press. ­ OA-­754 from Wolf, Matthew. 2006. For an autosegmental theory of mutation. Available as R http://roa.rutgers.edu.

Diana Passino

Progressive metaphony in the Abruzzese dialect of Teramo

Université Nice Sophia Antipolis

Abstract: This contribution deals with metaphony, an assimilatory process that has targeted stressed vowels in the evolution from Latin to Romance, particularly widespread in Italo-Romance. First, progressive metaphony is tackled, an ­understudied type of metaphony that concerns only restricted areas of the Italian peninsula located in the Abruzzi and Molise. A description of the process and a reconstruction of the vowel-system of the dialect that best instantiates this phenomenon are provided, followed by a formal account. Both feature-based and element-based analyses are presented, discussed and compared. Then, the inferences from the analysis of progressive metaphony are tested on data concerning regressive metaphony. Finally, the hints gathered from these analyses are exploited to advance an alternative, tentative proposal on the general workings of the metaphonic process, where the element A and the notion of head have a key role. This proposition helps answering some questions that are left open when metaphony is looked at through the usual perspective, and suggests some new paths of research.

1 Introduction The phenomenon detailed in this contribution, here called progressive metaphony, is one of many assimilatory processes concerning vowels in the Italo-Romance dialects (see Sánchez-Miret 1999 for an exhaustive look at these processes). Of these assimilatory processes, regressive metaphony is undoubtedly the most pervasive and the one most discussed in the literature (Rohlfs 1966; Tekavčić 1972; Maiden 1991; Calabrese 1985, 1998, 2011, among others). Progressive metaphony has so far received less attention, perhaps because it concerns a much more restricted area—the Abruzzi and Molise—while regressive metaphony regards almost the entire Italian peninsula, with the noteworthy exception of the Tuscan area, amongst few others. In progressive metaphony, it is word-internal, unstressed high vowels that cause raising of a following stressed /æ/ before undergoing reduction to schwa (Rohlfs 1966 § 25; Giammarco 1979: 29; Sánchez Miret 1999, among others), as illustrated in (1), based on Rohlfs’ data. As is customary in the Romance dialectological tradition, Latin etyma are indicated with capital letters. In addition, the stage preceding the

148 

 Diana Passino

application of progressive metaphony is also indicated between square brackets, since the application of this process is optional, al­though frequent. The forms where progressive metaphony has not applied are also grammatical: (1) Bellante (Rohlfs 1949 [1966: 49–50]) FILARE > fiˈla(re) > [fəˈlæ] > [fəˈli]     ‘to spin’ SUDARE > suˈda(re) > [səˈdæ] > [səˈdi] ‘to sweat’ vs > /aˈma(re)/   [aˈmæ] * [aˈmi]     ‘to love’ AMARE CREARE > /krea(re)/    [krə ˈɦæ]   * [krə ˈɦi]1 ‘to create’ COM(PU)TARE > /konta(re)/ [kunˈdæ]     * [kənˈdi]2 ‘to count’ However, in regressive metaphony word-final unstressed high vowels, which eventually disappeared in some dialects, influenced the stressed word-internal vowels causing raising or diphthongization (see the literature quoted above for a detailed description of the different kinds of regressive metaphony found in the Italian peninsula). An example of regressing metaphony yielding raising is shown in (2) with data from the Abruzzese dialect of Colonnella (Maiden 1991: 206): (2) Colonnella (Maiden 1991: 206) NOVA(M) >   ˈnɔːvə   ‘new-f.sg’ NOVU(M) >  ˈnuːvə  ‘new-m.sg’ Harmonic, assimilatory processes have received both element- and feature-based analyses notwithstanding the concern raised on both element and feature theory with respect to their ability to deal with such phenomena: according to Maiden (1991: 140), generalized vertical adjustments in vowel height, like the ones taking place in regressive metaphony, are the graveyard of binary features analyses. According to Clements (1991), vowel raising is a phenomenon that seems to call for an analysis in terms of features, since element theory is only able to express vowel lowering. Harmonic processes involving raising have however been expressed in terms of I addition (Colman & Anderson 1983 among others) or A demotion in element theory (Maiden 1991; Harris & Lindsey 1995) and as spreading of [+high] in feature theory (Calabrese 1985 et seq.). Accordingly, progressive metaphony, which has never been analyzed in formal terms to the best of my knowledge, can provide a testing ground for competing theories regarding the representation of the internal structure of segments and analyses of ­harmonic ­processes.

1 Vowels in hiatus trigger consonant epenthesis. 2 Postnasal obstruents are voiced.



Progressive metaphony in the Abruzzese dialect of Teramo 

 149

Therefore, as well as providing a more detailed description of progressive metaphony, this contribution compares an element-based analysis of progressive metaphony in terms of A demotion to a feature-based analysis in terms of height assimilation. The analysis is based on the dialect of Teramo, which is detailed in section 2 and for which a vocalic inventory is proposed to refine previous descriptions that I argue to be incomplete. Most relevant to the discussion on progressive metaphony, this inventory includes two series of high vowels, tense and lax, and also opposes tense and lax low central vowels. This contribution is organized as follows. Section 2 further illustrates progressive metaphony while section 3 is dedicated to the vocalic inventory of the dialect of Teramo, which is the focus of this work. In section 4, progressive metaphony is analyzed in terms of both element and feature theory and the two analyses are compared. Section 5 analyzes regressive metaphony in the dialect of Teramo through the prism provided by the findings on progressive metaphony, before exploiting those findings in section 6 to sketch a tentative account of the general process of metaphony.

2 P  rogressive metaphony in the Abruzzese dialects As illustrated above, progressive metaphony (Rohlfs 1966 § 25; Giammarco 1979: 29; Sanchez Miret 1999) is a phenomenon of long-distance assimilation between vowels, whereby a stressed /æ/ assimilates to the preceding unstressed high vowel. The quality of the mentioned vowel is visible under stress, as in 1sg.prs, since vowels undergo reduction to schwa in pretonic position. The process is illustrated again in (3) for the sake of clarity, with data from the dialect of Teramo, on which this study focuses. (3) Teramo FATI(C)ARE > [fatəˈji] ‘work-inf’ cf. [faˈtɪːjə] ‘work-1sg.prs’ COQUINARE > [kʊʧəˈni] ‘cook-inf’ cf. [kʊˈʧɪːnə] ‘cook-1sg.prs’ vs. CLAMARE > [kjaˈmæ] ‘call-inf’ cf. [ˈ kjæːmə] ‘call-1sg.prs’ In (4) I refine the previous descriptions of the phenomenon and show that progressive metaphony does not apply when /æ/ sits in a closed syllable: (4) [kəˈtærrə]  * [kəˈtirrə]   ‘guitar’ (tool to make spaghetti) cf. It. chitarra vs. /pɪɲˈɲæːtə/ >  [pəɲˈɲitə] ‘casserole’ cf. It. pignatta

150 

 Diana Passino

In addition, as mentioned above, progressive metaphony is an optional process. Although it is very frequent and productive, words where the process has not been applied are also grammatical. It may be that this happens under the influence of Italian, although this cannot be ascertained at present. The extent of the area affected by progressive metaphony has not been described in detail. De Giovanni (1973: 76) refers to this kind of assimilation as typical of the Teramo dialect, and most of the available data (cf. Rohlfs 1966 § 25) come from this area (zone 1b of the map in Fig. 1, showing the Upper-Southern dialects of Italy):

Fig. 1: The Upper-Southern dialects of Italy (source of the image https://it.wikipedia.org/wiki/ Dialetti_italiani_meridionali#/media/File:Neapolitan_language.jpg)

However, the influence of i, u < Ī, Ū on the following tonic /æ/, which becomes /e/ or /je/, is also documented in zone Ic, namely in the dialects of Vasto (Rolin 1908) in southern Abruzzo and Agnone in Northern Molise (Ziccardi 1910)3, as well as in zone II in the dialects of Castelmauro, Ripamolisani, Guglionesi (De Giovanni 2003: 90), as exemplified in (5): (5) Vasto   [fikˈke]   cf. It. ficcare  ‘to stuff’ Castelmauro [bbruˈʃe] cf. It. bruciare ‘to burn’

3 The case of Agnone represents a case not entirely coincident to progressive metaphony, though.



Progressive metaphony in the Abruzzese dialect of Teramo 

 151

The linguistic atlases of Italy, ALI and AIS, also document the influence of i, u < Ī, Ū on the following tonic /æ/. The phenomenon is scattered throughout the area around Teramo (Silvi, Bisenti, Campli, Montesilvano, Castelli), in Southern Abruzzo (Sant’Eufemia a Maiella, Vasto, Palmoli), and Molise (Roccasicura, Morrone del Sannio). This study concentrates on the dialect of Teramo, where the phenomenon is most pervasive. There, progressive metaphony has often been described as a phenomenon of class shift, or attraction from the first conjugation (Latin -ARE > Teramo dialect -æ) to the third (Latin -IRE > Teramo dialect -i) since it concerns all infinitives of the first conjugation where the outcomes of Ī, Ū are present in the stem (Savini 1881; D’Ovidio 1886: 86; De Lollis 1890–92). A formal analysis has never been proposed, however, and it will be attempted in this contribution. To perform a formal analysis of progressive metaphony in the dialect of Teramo, precise data concerning the vowel inventory of this dialect and the representation of the internal structure of these vowels are needed. However, as is discussed in next section, there is reason to believe that the available descriptions of this system are not exhaustive. In order to conduct an analysis that is as rigorous as possible, the next section attempts to redefine the vocalic inventory of the Teramo dialect, refining the previous proposals when necessary.

3   The vowel system of Teramo This section illustrates and discusses the limits of existing descriptions of the vocalic inventory of the Teramo dialect4 and puts forward a tentative new proposal.

3.1 The vowel system of Teramo: previous proposals The vocalic inventory of the dialect of Teramo has been described by several scholars, starting with Savini (1881); De Lollis (1890–92); Giammarco (1979); Lepschy and Lepschy (1988), and Scorretti (2012). Savini (1881) offers the reader a comparison between the pronunciation of sounds in the dialect of Teramo and the Italian language. The phonological inventory of the dialect is thus described comparatively by stressing the presence or absence of sounds from the national language in the dialect of Teramo. Savini

4 The vocalic inventories illustrated in this section refer to non-metaphonic contexts unless otherwise stated.

152 

 Diana Passino

also describes the diachronic evolution of the sound inventory from Latin, which is shown in the table in (6) adapted from De Giovanni (1973)5: (6) A sketch of the evolution of the Teramo dialect vocalic inventory based on Savini (1881) and adapted from De Giovanni (1973) Proto-Romance

i

e

ɛ

a

ɔ

o

u

Open syllable

i

æ

ɛ

æ

ɑ /ɔ

ɑ /ɔ

u

Closed syllable

i

æ

ɛ

a

ɔ

o

u

De Giovanni (1973) also summarizes the observations of De Lollis (1890–92), reported in (7): (7)  A sketch of the evolution of the Teramo dialect vocalic inventory based on De Lollis (1890–92) and adapted from De Giovanni (1973) Proto-Romance

i

e

ɛ

a

ɔ

o

u

Teramo dialect

i

a

a

a

ɔ

o

u

Based on De Lollis (1890–92) and data provided by the ALI, Scorretti (2012: 72) proposes a synchronic inventory and represents it with distinctive features, as illustrated in (8): (8) Scorretti’s (2012) proposal for the vowel system of Teramo

æ ɛ e i ɔ o u

±Low

+ + − − + − − 6 ±Constricted + − − + − − + ±Back

− − − − + + +

It is my contention that, for a number of reasons these inventories, which are also not mutually consistent, need refining. None of them can be used as the basis for an analysis of progressive metaphony. Some of the inconsistencies among the systems can be attributed to the use of orthography and some problems are instead connected to the influence of the Italian vowel system, which may have biased the perception.

5 The sound transcription in Savini (1881), De Lollis (1890–92), and De Giovanni (1973) uses phonetic characters of the Romanist tradition but lacks a table stating the equivalences. I have adapted the transcription to IPA interpreting their phonetic characters based on the description provided and my own data. 6 See Scorretti (2012: 51–52) for an illustration and discussion of the feature ±Constricted.

Progressive metaphony in the Abruzzese dialect of Teramo 



 153

One common belief concerning the vocalic inventory of the Teramo dialect holds that the majority of Latin vowels in stressed position evolved into /a/. In fact, Giammarco (1979), for instance, describes a tonic three-vowel system æ, i, u, for the dialect, tracing out an evolutionary path where mid vowels evolved from Ĭ, Ē, Ĕ, and Ŭ, Ō, Ŏ, conflated to /a/, which then underwent palatalization, as shown in (9a), where the slightly different evolution proposed by Lepschy and Lepschy (9b) is also illustrated: (9) a. Giammarco’s (1979: 24) proposal Ī Ĭ Ē Ĕ

i



  i

Ŏ Ō Ŭ Ū

a

o

a > æ

u

Eastern Abruzzese

u

Teramo Abruzzese

b. Lepschy and Lepschy (1988: 48)

Ī



e

ĀĂ

  i

ĬĒ Ĕ

a

ĀĂ

Ŏ Ō Ŭ Ū

o

u

As for the unstressed position, according to Giammarco (1979), the vowel inventory is reduced to a, ə, u in pretonic position, ə, i in postonic position and ə in final position, as illustrated in (10): (10) Vowel inventory proposed by Giammarco (1979) for the Teramo dialect a. Stressed: æ, i, u b. Unstressed:  protonic: a, ə, u  postonic: ə, i  final: ə Giammarco (1979) claims that in the Abruzzese dialect of Teramo the outcomes of Ĭ, Ē, Ĕ, and the outcomes of Ŭ, Ō, Ŏ, conflated to [a], the outcome of Ā, Ă, which then underwent palatalisation. However, only [a] < Ā, Ă underwent palatalisation (PANEM > pænə) as opposed to the outcomes of Ĭ, Ē, Ĕ (MĒNSEM > *mæsə) and those of Ŏ, Ō, Ŭ (FLŌREM> *flærə), which Giammarco also indicates as [a]. It is my contention that from a phonetic point of view, different sounds correspond to what Giammarco indicates as /a/ > [æ], the sounds allegedly conflating many different outcomes in the dialect of Teramo: in addition to [æ], there are also [ɑ] and [ɐ].

154 

 Diana Passino

In addition, from the phonological point of view, by looking at the patterns of reduction in pretonic position, what is described as /a/ by Giammarco, does not pattern uniformly, as shown in (11), where the segment that Giammarco indicates as a displays different reduction trajectories7: (11) Pretonic reduction of /a/ /a/ < Ŭ,Ō,Ŏ reduces to [u]ː /a/ < Ĭ, Ē, Ĕ reduces to [ə]ː /a/ < Ā, Ă no reduction [a]ː

ˈkraʧə / kruʧəˈfissə “cross / crucifix” ˈpandzə / pənˈzatə “I / you think” ˈparlə / parˈlætə  “I / you talk”

This suggests that what is generally indicated as /a/ by Giammarco (1979) is not the same phonological entity, and the description needs to be refined. As mentioned above, I propose that what Giammarco (1979) conflates to /a/ are 3 different sounds, namely [æ], [ɑ] and [ɐ], which have different etymological origins, independent phonemic status, and thus different phonological representation. Accordingly, they undergo a different reduction path. Another refinement to be made concerns the high vowels originating from Ī and uo > u Ĕ > je > i Ŏ > uo > u

So actually, as a result of monophthongization, there are both tense and lax high vowels in the dialect of Teramo, enjoying the same distributional possibilities as shown in (13): (13) vɪːnə dʊːrə

‘wine’ ‘hard’

piːtə puːkə

‘feet’ (metaphonic) ‘few’ (metaphonic)

Incidentally, also Savini (1881: 37) noticed that the sound of the dialectal /i/ corresponding to the Tuscan diphthong [jɛ] was different than other /i/s, and in fact, there are a series of exceptional outcomes in the dialect of Teramo where tense /i/ corresponds to Tuscan [jɛ] < such as [pitrə] and [kkisə] < PĔTRU(M), ECCLĔSIA(M). cf. It. Piɛtro, kkjɛsa. The historical origin of tense vowels explains why there are no minimal pairs containing tense and lax high vowels. Since stressed high lax vowels could not be influenced by unstressed high vowels and undergo metaphony, and the source of tense vowels are mainly metaphonic diphthongs, it is never the case that tense vowels are present in roots where also high lax vowels can be present. In this case the lack of minimal pairs, then, depends on historical reasons. Few other diphthongs, present in exceptional outcomes, also evolved as high tense vowels. This state of affairs could result in the presence of minimal pairs. In this case it is only by chance that the presence of minimal pairs did not stem from these outcomes. From a distributional point of view, however, tense and lax high vowels are not in a complementary distribution and none can be derived by the other by means of a phonological process. As regards their phonetic difference, nowadays, in the dialect of Teramo the difference is remarkable from a perceptual point of view only in the case of the front vowels. Under the influence of Italian, lax [u] is often pronounced as tense. The evidence provided by nearby isolated villages and the evidence discussed above, provided by Romani’s observations, however, hints at a system such as the one I am suggesting in this contribution, where tense and lax vowels co-existed in

156 

 Diana Passino

a recent stage of the language (from about the second half of the 19th century, after the monopthongization of diphthongs had taken place, to the second half of the 20th century, when the linguistic unification of Italy can be said to be accomplished, and when bilingualism and the thorough diffusion of Italian led to the italianization of the dialects and the conflation of tense and lax high vowels in the case of Teramo). The stage prior to italianisation is object of the present analysis. In the next section, in accordance with the observations made above and moving away from previous analyses, I sketch the vocalic inventory that I propose for the Teramo dialect. The sketch is tentative and does not investigate the reasons for exceptional outcomes such as Ŏ, Ŭ > u FŎCUM, FŎRTEM, DIŬRNUM [fukə, furtə, jurnə] and Ĕ > i SĔMPER, VĔNIO [simbrə, vingə], nor does it describe all the intermediate stages between Latin and the recent stage of the language object of investigation since they are not relevant for the analysis. Its main aim is to describe the latter inventory in order to reconstruct the phonological representation of each vowel so that an analysis of the progressive metaphonic process can be put forth, the rest is left to future research.

3.2 The vowel system of the dialect of Teramo: a tentative proposal As was customary in the dialectological tradition of the first half of the 20th century, the evolution of Latin vowels to the system object of investigation is sketched as shown in Tab. 1. (1a) lists the outcomes of Latin vowels in nonmetaphonic contexts whereas (1b) illustrates regressive metaphonic contexts. As is clear from the table, no differentiation of outcomes is recorded at present between open and closed syllables: Tab. 1: Evolution of the vowel system of the dialect of Teramo a. Non metaphonic position Given final A, E, O, U

Open syllable

Closed syllable

ĀĂ>æ Ĕ>ɛ Ē>ɐ Ĭ>ɐ Ī>ɪ Ŏ>ɔ Ō>ɑ Ŭ>ɑ Ū>ʊ

PĀNEM > pæːnə, MĂREM > mæːrə CĂRNEM > kærnə NĀVEM > nævə MĔLEM > mɛːlə PĔLLEM > pɛllə CRĒDO > krɐːtə PĒNNA > pɐnnə NĬVEM > nɐːvə PĬSCEM > pɐʃʃə DĪCO > dɪːʧə DĪXIT > dɪssə BŎNUM > bɔːnə FLŎCCUS > flɔkkə NEPŌTEM > nəpɑːtə PŌNTEM > pɑndə CRŬCEM > krɑːʧə BŬLLAM > bɑllə CRŪDUM > krʊːtə LŪCTUM > lʊttə

Progressive metaphony in the Abruzzese dialect of Teramo 



 157

b. Metaphonic position Given final I, (S)

Open syllable

Closed syllable

ĀĂ>i Ĕ>i Ē>i Ĭ>i Ī>ɪ Ŏ>u Ō>u Ŭ>u Ū>ʊ

PECCĀTI > pəkkiːtə PĔDES > pitə CRĒDIS > kriːtə PĬLI > piːlə FĪLII > fɪːjə BŎNI > buːnə NEPŌTES > nəpuːtə IŬGI > juːgə CRŪDI > krʊtə

PĂNNI > pinnə AUGĔLLI > ʧ illə SĒCCI > sikkə PĬSCES > piʃʃə DĪCIS > dɪʧə FLŎCCI > flukkə PŌNTES > pundə CŬRRIS > kurrə LŪCTI > lʊttə

The vocalic inventory proposed for the tonic position in the Teramo dialect is recapitulated in (14): (14) Evolution of the Latin system in the dialect of Teramo

Ī Ĭ Ē Ĕ

ĀĂ

Ŏ

Ō Ŭ Ū

ɪ

ɐ

ɛ

æ

ɔ

ɑ

ʊ

non-metaphonic contexts

ɪ

i

i

i

u

u

ʊ

metaphonic contexts

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

Having established the vocalic inventory of the Teramo dialect, it is now possible to submit a proposal concerning the representation in terms of the subsegmental structure of these vowels in order to carry out an analysis of progressive metaphony, as illustrated in the next section. Crucial data, some of which are previously unrecorded, have arisen from the above reconstruction, such as the tense character of the metaphonic vowels, and vowels deriving from old diphthongs in general, as opposed to the lax character of other high vowels of the system.

4 Progressive metaphony in the dialect of Teramo Based on the refined inventory of vowels in the dialect of Teramo, two possible analyses of progressive metaphony are discussed next. Progressive metaphony in the dialect of Teramo is a process triggered by lax high vowels targeting /æ/ in tonic open syllables and yielding the tense high front vowel. As mentioned in the introduction, it has been argued that element theory is unable to express raising patterns, since height cannot be expressed as such in element theory. Harmonic processes involving raising, however, have since then been analyzed in terms of

158 

 Diana Passino

A demotion in element theory (Maiden 1991; Harris & Lindsey 1995) or I ­augmentation (Anderson & Durand 1986 among others). Accordingly, progressive metaphony may provide a testing ground for competing theories of the representation of the internal structure of segments. In this contribution the two approaches are compared in order to understand which can provide more insight into the workings of the process. Before illustrating the analyses, the process and explananda are summarized in (15) so the analyses can be compared according to their explanatory power: (15) Progressive metaphony in the Teramo dialect /ʃtrɪl’læ/ > [ʃtrəl’li] ‘to yell’ /abbrʊ’ʃæ/ > [abbrə’ʃi] ‘to burn’ Explananda – Why only [æ] is targeted – Why only in open syllables – Why the outcome is tense

4.1 A feature-based analysis The autosegmental treatment of progressive metaphony using features as subsegmental primes is quite straightforward if we represent /æ/ as [+tense], as shown in (16), where a representation in distinctive features is proposed for the vocalic inventory of the Teramo dialect: (16) Representation in distinctive features of the Teramo dialect inventory æ

ɑ

ɐ

ɛ

ɪ

i

ɔ

ʊ

u

−high

−high

−high

−high

+high

+high

−high

+high

+high

−low

+low

+low

−low

−low

−low

−low

−low

−low

+tense −tense −tense −tense −tense +tense −tense −tense +tense −back +back −back −back −back −back +back

+back

+back

The tense representation has been chosen for this segment in the Teramo dialect inventory, despite the fact that in contemporary English phonology [æ] is usually described as phonologically lax, since it patterns with lax vowels in distribution and undergoes tensing (cf. Labov 1981; Kenstowicz 1994: 17, among others). In the history of the English language, however, [æ] is considered to be the output of a tensing process (Dobson 1968; Kiparsky 1995). Furthermore, in Somali, [æ] is the +ATR counterpart of [ɑ], as can be observed in ATR harmony

Progressive metaphony in the Abruzzese dialect of Teramo 



 159

processes (­Farnetani 1981: 56, among others). In addition, in vowel systems where ­tenseness plays a role, Kiparsky (1974, 1995) classifies [æ] as tense. There is no critical evidence to represent æ as −low. It could also be represented as +low, in which case progressive metaphony should be represented as a process spreading the height features +high and −low. Nothing crucial hinges on this, however, and the simplest representational option has been chosen to illustrate the argument. Height assimilations such as the one studied in this contribution are treated in terms of spreading of the feature [+high]. The triggers of the metaphonic process share the feature [+high] while the target is [−high], as shown in (17): (17) Subsegmental representation of triggers and target Triggers Target ʊ ɪ æ +high +high −high −low −low −low +back −back −back −tense −tense +tense As illustrated in (18), due to the spreading of the feature [+high], the tense low front vowel [æ] becomes [i], while all other features are unaffected and remain equal. Using privative features and underspecification, the essence of the analysis does not change: (18) Progressive metaphony as rightwards spreading of [+high] +high

→

+high

+high

→

+high

+back

−back

−back

−back

−back

−back

−tense

+tense

+tense

−tense

+tense

+tense

−low

−low

−low

−low

−low

−low

V /ʊ/

V /æ/

V /i/

V /ɪ/

V /æ/

V /i/

In some dialects of the Teramo area, both unstressed ʊ and ɪ reduce to schwa, while in others only ɪ reduces to schwa. Processes must be then ordered so that the delinking of features in unstressed position takes place after spreading, because what spreads—[+high]—is a feature that is eventually delinked.

160 

 Diana Passino

That metaphonic spreading operates before reduction is also evident from the fact that only underlying /ʊ/ and not [ʊ] as the outcome of reduced /ɔ, ɑ/ triggers the process, as shown, respectively, in (19): (19) frʊsˈti < frʊsˈtæ  ‘to whip’ cf. jə ˈfrʊstə ‘I whip’ *pʊrˈti < pʊrˈtæ ‘to bring’ cf. jə ˈpɔrtə ‘I bring’ The feature-based analysis accounts for the data in a straightforward way. A low tense front vowel /æ/ becomes high, /i/ by switching the feature [−high] to [+high] due to spreading from the preceding high vowels, /ɪ, ʊ/, which trigger assimilation. What remains to be explained in this perspective is the following: if progressive metaphony is simply a raising process, why are mid-vowels not targeted by the process, as shown in (20)? (20) pinɛtə > * [pənitə] ‘pine forest’ juttɔnə > * [juttunə] ‘glutton’ A reviewer suggests that one reason that would justify this state of affairs could be a requirement both on the target and on the output of the process stating that they be tense. However, considering that the harmonic triggers are lax vowels, this would not be completely consistent with what is commonly held to be a harmonic or metaphonic process, that is, a quality agreement process (Savoia & Maiden 1997: 15). Although it is possible that in harmonic processes the agreement between trigger and target is only partial, as in this case, where the process only demands to agree in height and not in tenseness, a harmonic process demanding that in the outcome of the process a feature, albeit different from the harmonic trigger, must obligatorily disagree, is unheard of to the best of my knowledge. The analysis proposed further in this contribution, moreover, proposes that the disagreement concerning the tenseness value between the trigger and output of the process is only apparent. It concerns the phonetic interpretation, and it is not underlying. Other aspects concerning the context of the assimilation target, namely stress and open syllable, also remain unexplained in this analysis, although harmonies initiated by vowels in weak positions towards vowels in privileged positions, as in this case of progressive metaphony, are recorded (Maiden 1991; Walker 2005), as well as the opposite pattern.

4.2 Element-based analysis In element theory, harmonic processes involving raising are expressed in terms of A demotion/absence/loss (Maiden 1991; Harris & Lindsay 1995; Scheer 1995; Russo 2007) or I addition (Anderson & Durand 1986 among



Progressive metaphony in the Abruzzese dialect of Teramo 

 161

others). The basic idea beyond this proposal is that in harmonies concerning the element A, this latter is sustainable only if it is sanctioned by another A in the domain: a trigger lacking A may provoke delinking or recession of this segment in the target so that this latter becomes more harmonic to the trigger. In the case at hand the analysis in terms of I addition is not possible since one of the triggers, /ʊ/, does not contain I. Accordingly, the former analysis in terms of A demotion is presented next, after shortly introducing element theory. Elements are fully interpretable phonological primes. They are intended as cognitive units, but their definition is acoustic in nature. According to Harris & Lindsey (1995: 53), A, I, and U, the elements representing the corner vowels [a], [i], and [u], are mental object that are mapped to acoustic patterns defined qualitatively as overall quasi-spectral shapes. The pattern of the element A, whose independent interpretation is the vowel [a], is a spectral energy mass in the middle of the sonorant frequency zone, the frequency band containing the most significant information relating to vocalic contrasts, given that energy minima are present at top and bottom of this zone. It is accordingly referred to as mAss by Harris and Lindsey (1995: 53). The pattern of the element I, on the other hand, is indicated as dIp, since it is characterized by energy lower in the middle of the sonorant frequency zone than at either side. The pattern of U, on the other hand, is characterized by a spectral peak at the lower end of the zone, indicated as the rUmp pattern. The element schwa, on the other hand, representing the neutral vowel, is like a baseline to which elemental patterns are superimposed (Harris & Lindsey (1995: 60). It is present in the representation of all vowels but by convention it is only noted when it is the head of the phonological expression. Vowel inventories richer than three vowel systems are obtained through a combination of elements. In a phonological expression, elements are organized in an asymmetric relationship so that one element is the head and the others are operators. By convention, the head of a phonological expression is underlined. In a representation like that found in Harris & Lindsey (1995), for instance, the difference between tense and lax vowels is represented in terms of headedness: lax vowels are headed by schwa, whereas tense vowels are headed by a full vocalic element, as shown in (21), where schwa, although present in the representation of all vowels, is not noted unless it is the head of the phonological expression. In element theory, the elemental composition of a segment can be calculated by looking at the phonetic content, namely the acoustic pattern of the segment, but also by looking at the reduction patterns in systems where

162 

 Diana Passino

segments undergo reduction. Compound elements in particular, when undergoing reduction, may reveal one of the primitive elements that they contain. In the case at hand, we assume that ɔ, ɑ, and ʊ contain U, since they reduce to [ʊ]. In (21) the Teramo dialect system proposed in (14) for tonic vowels is recapitulated, and a representation of this inventory in terms of elements is put forward in Harris & Lindsey’s (1995) element systems. The representation proposed in (21) is discussed next: (21)  Vowel inventory of the Teramo dialect in element theory (Harris & Lindsey 1995)

Ī Ĭ

Ē Ĕ

Ŏ Ō

Ŭ Ū

i ɑ ε ʊ non-metaphonic contexts æ Iə A ə AI AI AU AU ə U ə -------------------------------------------------------------------------------------------- i i i i u u ʊ Iə I I I U U Uə (regressive) metaphonic contexts c

a

  

ĀĂ

From an acoustic point of view the front vowels /æ/ and /ɛ/ contain both I and A. In a system like the one of the Teramo dialect, contrast between these two s­ egments could be expressed by means of tenseness, opposing respectively a phonological expression headed by a full element, namely A, as in (AI) /æ/, to a schwa-headed phonological expression (AIə) /ɛ/. However, it could also be expressed by opposing a A-headed to a I-headed phonological expression, as discussed next. A is a palatalized [a], so it must be A-headed. Conversely I can be assigned as a head to /ɛ/, since there is no contrast of tenseness in mid vowels. The lax pronunciation of (AI) then could be just a matter of phonetic interpretation, not depending on schwa headedness. At the present state of the research there is not enough evidence to choose between these two representations. This is not however crucial for our discussion, for which it is important to distinguish A headed segments from non A-headed segments. We represent it as I headed for symmetry with the other mid vowel, which we represent as U headed, as discussed next. /ɐ/ is represented as Aə and in principle could be headed by ə or A. Since this vowel reduces to schwa, the former representation is chosen here. All these representations are consistent with the patterns of reduction documented. /ɐ/, /ɛ/ reduce to schwa, which is contained in all vowels. /æ/ only reduces to schwa in postonic position, escaping reduction in pretonic position. The possibility for /æ/ to reduce to [a] in pretonic position is also recorded.

Progressive metaphony in the Abruzzese dialect of Teramo 



 163

From a phonetic point of view, the back vowels /ɑ/ and /ɔ/ must contain both A and U and they become U in pretonic position, losing A. Headedness is assigned based on the phonetic content: in order to obtain contrast, in the case of /ɔ/, we represent it as headed by U, since it is a round vowel. We represent /ɑ/ as headed by ə, since we have seen that A-headed segments do not undergo reduction or in any case A is maintained while this segment reduces to U and loses A. We have now factored out /æ/ as the only A headed segment. This move then, in an analysis where progressive metaphony is analyzed in terms of A demotion, can provide an opportunity to better define the process and to understand why only /æ/ is targeted by the process, leaving other segments containing A unaffected. Only this segment is A-headed, so the process could be defined and analyzed as a process of A-demotion restricted to cases where A is the head. A slightly different system is proposed by Backley (2011), who exploits the possibility that phonological expressions need not be necessarily headed. This move allows to reduce overgeneration and obtain contrast with fewer elements. In Backley’s system headed expressions are naturally strong and represent acoustically prominent segments, whereas non-headed expressions are much weaker. Acoustic prominence suggests a headed element, since the physical properties of a head are stronger than those of a non-head. For instance headedness is used to distinguish palatals from coronals, both represented as (I). Also headed U is interpreted as labial resonance and non-headed U as velar resonance. In this way, labials and velars are formally related, and at the same time phonologically distinct. In (22), a proposal concerning the representation of the Teramo dialect inventory in Backley’s system is illustrated, which is discussed next: (22) Vowel inventory of Teramo dialect in element theory (Backley 2011)

Ī Ĭ

Ē Ĕ

Ŏ Ō

Ŭ Ū

i ɑ ε ʊ non-metaphonic contexts æ I A AI AI AU AU U -----------------------------------------------------------------------------------i i i i u u ʊ I I I I U U U metaphonic context c

a

  

ĀĂ

The distinction between tense and lax high vowels is therefore obtained in Backley’s system by opposing headed vowels (tense) to non-headed vowels (lax), as shown in (22). Exploiting Backley’s perspective, it is also possible

164 

 Diana Passino

to propose that among the phonological expressions containing A, only /æ/ is A-headed, on the grounds that this is the only vowel that never undergoes reduction in pretonic position. Headedness thus distinguishes /æ/ from /ɛ/, which also contains AI, as in Harris and Lindsey’s representation, but is not headed in this representation. /ɐ/, however, can be represented as non-headed and only containing A. Both /ɛ/ and /ɐ/ reduce to schwa, which in Backley’s model is a nonheaded A. /ɔ/ and /ɑ/ both contain A U and reduce to U. Contrast in this case can be obtained in several ways. As shown in (22), in /ɔ/, headship is assigned to U, allowing the possibility for /ɑ/ to be non-headed since this move enables /æ/ to stand out as the only A-headed segment in the inventory. Again, in an analysis where progressive metaphony is analyzed in terms of /æ/ demotion, this can provide an opportunity to understand why only /æ/ is targeted by the process, leaving other segments containing A unaffected. Height harmony analysis in element theory holds that A is demoted in a phonological expression if it is not supported by A in other phonological expressions in the domain. By simply demoting A from the elemental content of /æ/ on the grounds that is not licensed by other As in the domain in word forms where progressive metaphony takes place, we obtain a phonological expression containing only I, as shown in (23): (23) Progressive metaphony as A demotion a. Harris & Lindsey’s (1995) model b. Backley’s (2011) model

st ʊ dj æ



>

st ʊ dj i

U

I

U

ə

A

ə

st ʊ dj æ > st ʊ dj i ‘to study’

I



U

I

U

I

A

In Harris & Lindsey’s model, all phonological expressions need to be headed. Therefore, A deletion automatically implies that I becomes a head, which explains why a tense vowel results from the metaphony process. In Backley’s model, it is not clear how I acquires head status unless one wants to stipulate that a headed phonological expression is specified to remain headed after undergoing a process that removes the head element. This stipulation, however, ought to be substantiated empirically. As mentioned above, the A-demotion analysis holds that in harmonic systems where processes of vowel raising are found, A needs to be supported by the presence of another A in the same domain. If A is absent from the rest of the domain, A in a given segment undergoes recession for lack of support,



Progressive metaphony in the Abruzzese dialect of Teramo 

 165

­ ielding vowel-raising. In the case at hand, assuming the representation of y the ­inventory proposed in (21) and (22), progressive metaphony in the dialect of Teramo can be analyzed as a harmonic process whereby A needs to be supported by other As in the string only if it is a head, and /æ/ is the only phonological expression where A is a head, which is why it is the only segment targeted by the process. However, as mentioned above, the analysis in (22) encounters problems in deriving the correct tenseness value of the output of the metaphonic process. Although Harris & Lindsey’s (1995) model, as mentioned above, succeeds in accounting for the synchronic process in a straightforward way, an analysis in Backley’s model, discussed next, can help make sense of the recorded tense output, as well as shed light on some diachronic and synchronic aspects of the process. The analysis is framed in Strict CV (Lowenstamm 1996; Scheer 2004), a theory where phonological constituent structure is represented as a sequence of strictly alternating consonantal and vocalic positions and all syllabic effects are derived from a network of lateral relations, government and licensing, entertained by segments. Government is a regressive lateral relation that inhibits segmental expression. Word-internally, government can only be exerted by full nuclei. As for final empty nuclei, their governing ability is the object of parametric variation. As opposed to government, licensing enhances segmental expression and is also dispensed by full nuclei word-internally or word-finally by a parameter. Government and licensing, lateral regressive antagonistic forces, are solely responsible for the effects commonly attributed to syllable structure, and their existence accounts for empirical data concerning cross-linguistic phonological phenomena and patterns. Researchers working in the Strict CV perspective (Ségeral & Scheer 2001 among others) have also highlighted the possibility that segmental length could be virtual, namely that phonological length need not necessarily be spelled out as phonetic length as long as the phonological behavior of the long segment allows reconstruction of its phonological length. The tense outcome can be then motivated by assuming that underlying length differences are spelt out phonetically as quality differences. Concerning the tenseness opposition among vowels, it has been proposed before for French (Rizzolo 2002) that this surface distinction may in fact spell out a length distinction. Length and tenseness interact closely in several Indo-European languages where only one of these features is distinctive (German, Swedish, English, Latin). In Danish, where short vowels are not laxer than long vowels, both length and tenseness are contrastive.

166 

 Diana Passino

It is proposed here that metaphonic vowels are underlying long vowels, and it will be argued that they derived by former diphthongs. Their length, however, is phonetically expressed through tenseness. This way the element I never acquires headship since the output segment in the metaphonic process has the same underlying representation of the output, as far as the quality is concerned, that of a lax unheaded vowel, as illustrated in (24): (24) CVCVCVC V kʊʧ ɪ n i

< kʊ ʧ ɪ n æ < COQUINARE ‘to cook’

I

Not only does this proposal account for the tense outcome using Backley’s element system but it allows in addition to draw a connection between the synchronic length of the vowels that undergo the metaphonic process, spelled out as tenseness, and A demotion, in a perspective where A has been argued to be structurally, more than elemental (Pöchtrager 2006; Carvalho & Russo 2007; Pöechtrager & Kaye 2010). My argument runs as follows: if A has a structural identity, it is plausible that its demotion has effects on structure. As a consequence, metaphony analyzed as A demotion could motivate the increase of structural space and connect the description of the process to its outcome. A demotion could provide the structural space needed for a lax vowel to branch and be spelled out as tense. I build on the idea that A is relied on in structure proposing, different from Pöechtrager and Kaye (2010), according to whom A is only structural, that A is both structural and elemental and that this depends on its status: being head endows it with structural identity while being operator makes it elemental. Loss of headship for A, as well as delinking of an A head, produces the appearance of a residual V position resulting from A abandoning its structural vest or disappearing by delinking. An analysis proposing that demotion of A from head to operator has effect on structure yielding an empty V position, can also be applied to the diachrony of the process. As illustrated in (25) with written data from a dialectal comedy (Delfico 1891–92), the process originated as breaking: (25) strillie c  ucinie a  zzuffuchie  pegniete

‘to scream’ ‘to cook’ ‘to choke’ ‘casserole’

Progressive metaphony in the Abruzzese dialect of Teramo 



 167

As shown above, the process targets only segments headed by A, namely [æ]. Because it lacks support from other As in the word form, A in /æ/ (AI) loses its headship8 and breaks, as shown in (26): (26) CVCVCVCV kʊʧ ɪ n i

e < kʊ ʧ ɪ n æ < COQUINARE ‘to cook’

I A

Again, the proposal that loss of A headship results in the presence of an extra V position may provide a connection between A demotion and breaking so that a cause and effect relationship that allows further insight on the process may be hypothesized. It has been mentioned above that in the dialect of Teramo all diphthongs have eventually become monophthongs, yielding tense vowels that are opposed to lax vowels. Diphthongs often occupy two vocalic slots. Therefore, an analysis proposing that vowels originating from monophthongization still retain two slots in their phonological representation is also diachronically consistent. Moreover it may explain why the process is still active notwithstanding the diachronic change that the diphthongs have undergone. From a synchronic point of view, given the monophtongization of diphthongs, the progressive metaphony process must have been reanalyzed from a process of A demotion where A loses its head status yielding vowel diphthongization to a process of raising where A head is delinked yielding vowel lengthening (tensing in surface). This state of affairs could also explain why the process only takes place in open syllable. Closed syllables do not generally tolerate long vowels. So far this contribution, as well as providing a more detailed description of progressive metaphony, has compared an element-based analysis of progressive metaphony in terms of A demotion to a feature-based analysis in terms of height assimilation. Despite the linearity of a feature-based analysis, an element-based account in terms of A demotion offers an analysis that helps to make sense of some diachronic and synchronic aspects of the phenomenon: it can help clarify why only one segment of the inventory is targeted by the process. In addition, it opens a path of research that connects A demotion to breaking, suggesting a

8 Since the source of the evidence is written, it is difficult to say whether indicates the raising diphthongs [je/jɛ] or the centering diphthong [iə]. The lack of graphically indicated stress on when is word final makes the latter option more likely. Nothing crucial for the argument hinges on this so I leave this topic for further research.

168 

 Diana Passino

representation for tense vowels in the dialect of Teramo that is diachronically and synchronically consistent with respect to the phonological inventory and to the phonological processes of progressive as well as regressive metaphony. Having provided an analysis of progressive metaphony in terms of A demotion, one can legitimately suppose that such an analysis supports Maiden’s (1991) proposal for regressive metaphony in Italo-Romance. Next, data are provided from the dialect of Casalincontrada, which arguably coincides with an earlier stage of the dialect of Teramo, prior to monophthongization (De Lollis 1890–92). On the one hand these data support the previous findings about the relationship between A demotion from the role of head and the creation of structural space leading to diphthongization; on the other, they show that metaphony in the dialects of Italy does not always consist uniquely of A demotion. The data of Casalincontrada are best analyzed as I addition (cf. Anderson & Durand 1986; Colman & Anderson 1983 for the analysis of I-umlaut in Old English).

5 R  egressive metaphony in the dialects of Teramo and Casalincontrada We analyze here regressive metaphony in the dialect of Casalincontrada, since De Lollis, who provides valuable data on the dialect in his contribution of 1890–92, argues that it represents a stage that was common to the dialect of Teramo prior to monophthongization. In the dialect of Casalincontrada, as it happens in the whole Adriatic Abruzzo, metaphony was only triggered by Proto-Romance final unstressed /i/. Final /i/ was mainly an inflectional ending conveying grammatical information concerning plural in some inflectional classes. As a consequence, the reduction of word-final vowels to schwa, in the same area, led to the morphologization of metaphony so that metaphonic alternations became eventually a morphological exponent of number. As it is apparent from the synoptic table in (27), the analysis that best accounts for these data is one in terms of I augmentation, similar in its essence to the one proposed for Germanic I umlaut (Anderson & Durand 1986). The table also shows the following: 1. m  etaphony manifests itself as diphthongization. Vowels that did not undergo diphthongization in their evolution from Latin do so as an effect of metaphony; 2. metaphony entails the addition of the element I (marked in gray in tableau)9; 3. if A heads a phonological expression, then it is demoted as an effect of ­metaphony. 9 The outcome of Ŏ is the only exception to I addition analysis. Ŏ resistance to metaphony, ­however, is documented also in other dialects, as reported by Savoia & Maiden (1997).

Progressive metaphony in the Abruzzese dialect of Teramo 



 169

(27) Metaphony in the dialect of Casalincontrada (CH) Latin vowel

Proto-Romance outcome

Casalincontrada outcome in context of final A, E, O, U (non-metaphonic)

Casalincontrada outcome in context of final I (metaphonic)

Ă,Ā TRAVEM ‘beam’

Open syllable: a

travə

treivə

A

A I

Ă,Ā PASSUM ‘step’

Closed syllable: a

passə

pjessə

A

A I

Ĕ PĔCTUM ‘breast’ Ē TĒCTUM ‘roof’ Ĭ PILUM ‘hair’

Ĭ PĬSCEM ‘fish’

Closed syllable: ɛ

Closed syllable: e

Open syllable: e

Closed syllable: e

pættə

pjett

A

A

I tættə

I

A

A

I

I

pæilə

A

I

I

pæʃʃə A

Open syllable: ɔ

peilə

A

I Ŏ BOVEM ‘ox’

teittə

peiʃʃə A I

vɑuvə

vəyvə

AU

AU I

170 

 Diana Passino

Latin vowel

Proto-Romance outcome

Casalincontrada outcome in context of final A, E, O, U (non-metaphonic)

Casalincontrada outcome in context of final I (metaphonic)

Ŏ GROSSUM ‘large’

Closed syllable: ɔ

rossə

ruossə

A

A

U

U

Ō FLŌREM ‘flower’

Open syllable: o

fjɑ u rə

fjəyrə

AU

AU I

Ŭ CRŬCEM ‘cross’

Open syllable: o

krɑuʧə AU

krəyʧə AU I

Closed syllable: o Ŭ TŬRDUM ‘mockingbird’ Ū MŪTUM ‘dumb’

Open syllable: u

tordə

təyrdə

A

AU

U

I

məutə

məytə

AU

AU I

Ū JŪSTUM ‘fair’

Closed syllable: u jəustə AU

jəystə AU I

As in the case of progressive metaphony analyzed in previous sections, A is demoted uniquely when this element heads the phonological expression. Otherwise, the phenomenon, exception made for the Ŏ outcome, deploys as a diphthongization with addition of I. The element representation proposed

Progressive metaphony in the Abruzzese dialect of Teramo 



 171

here for the Casalincontrada dialect respects the one proposed for Teramo in (22). Again, it is possible to connect the demotion of A from its head status to diphthongization. Usually in metaphony of the Neapolitan type, namely the diphthongizing one, only open-mid vowels yield diphthongs whereas closed-mid vowels raise, as an effect of metaphony. In a perspective whereby Italo-Romance metaphony consists of spreading of [+high], this drove Calabrese (1985 et seq.) to argue that in Italo-Romance mid vowels there is an ATR opposition and that diphthongs arise as a repair by fission, which avoids the forbidden combination of features [+high −ATR]. The data from the dialect of Casalincontrada, on the contrary, show that diphthongs may arise also from +ATR vowels and that metaphony cannot be always accounted for by an analysis in terms of raising. Interestingly enough, the Ū outcome, namely the diphthong /əu/, also undergoes metaphony, which in this case manifests itself as fronting of a high back vowel, whereas the Ī outcome does not. This is consistent with a process of I addition and cannot be explained by analysing metaphony as a process spreading [+high], considering that [u] in /əu/ is already [+high] and that metaphony manifests itself as fronting. In the present stage of the language, the dialect of Teramo, where diphthongs have monophthongized and there is no difference to be noticed in open and closed syllable, metaphony can be analyzed as A demotion, with the exception of ɐ bŏvem ‘ox’

ɔ

Ō flōrem ‘flower’

A pɐlə ‘hair-SG’ A

o

vɔvə ‘ox-SG’

I pilə ‘hair-PL’ I vuvə ‘ox-PL’

U

U

A

A

flɑrə ‘flower-SG’

flurə ‘flower-PL’

U

U

A

A

According to Savoia & Maiden’s (1997) data, in the dialect of Bellante, 13 kilometers from Teramo, A demotion analysis is possible and exceptionless: plurals of words with stressed ɐ < Ē, Ĭ, that in Teramo get an [i], inconsistently with an analysis in terms of A demotion, get a schwa under metaphony, as shown in (29): (29) Metaphony in Bellante (Savoia & Maiden 1997) sg pl tɐttə təttə ‘roof ’ lɛttə littə ‘bed’ pɔrkə purkə ‘pig’ æːkə iːkə ‘needle’ Schwa, the epenthetic vowel in these dialects, in fact, is what would remain after A delinking in ɐ, that, assuming the same representation as in Teramo, would consist in headless A. Starting from these data, one can synchronically obtain the plural of some inflectional classes by delinking A (cf. D’Alessandro & van Oostendorp 2014). At



Progressive metaphony in the Abruzzese dialect of Teramo 

 173

the present stage of this research I am not engaging in arguing that regressive metaphony is synchronically active in Teramo or Bellante beyond doubt, but it surely has been active long after the Late Latin/Proto-Romance period in which the process originated, although the evidence cannot be discussed here for space reasons. Be that as it may, the synchronic analysis of progressive and regressive metaphony in different stages of the dialect of Teramo let some elements surface that may shed some light on the general phenomenon of Italo-Romance metaphony, as discussed in the next section.

6 R  egressive metaphony: diachrony, open questions and tentative suggestions After showing the possibility of carrying synchronic analyses of metaphonic processes in language stages far with respect to the original process originated in late Latin or Proto-Romance, we can go back to the latter to see if it may benefit what has emerged from the previous discussion. According to Loporcaro (2011), who discusses various types of evidence, metaphony in statu nascendi was a raising process triggered by Ī Ū and their Proto-Romance outcomes, so the first stage of metaphony was the Sabino type of metaphony whereby e, o > i, u and ɛ, ɔ > e, o. The diphthongizing metaphony being a further evolution. The Italo-Romance metaphonic pattern, among other things, differs with respect to several parameters: metaphony can be triggered by /i/ and /u/ or just /i/. It may target all kinds of syllables or just the open ones, it may concern all vowels or just a subgroup. Regarding this, Maiden (1991: 115) draws an implicational hierarchy according to which metaphony of low vowels generally entails that of higher vowels. Furthermore, metaphony of /a/ entails invariably metaphony of mid vowels, metaphony of open-mid vowels presupposes that of midclose vowels (Old Romanesco & Surselvan Romansch being an exception). In addition Maiden (1991: 113) shows that metaphony triggered by /u/ presupposes metaphony triggered by /i/, but the contrary is not true, and metaphony in closed syllables presupposes metaphony in open syllables. The analysis in terms of A demotion (Maiden 1991), as well as the one in terms of spreading of [+high], accounts for metaphony in statu nascendi, represented by Sabino metaphony. However, the history and differentiation of metaphony in Italy, as well as Maiden’s (1991) implicational hierarchy, allow a series of questions to

174 

 Diana Passino

surface that do not find a satisfying explanation in these same analyses. I resume these explananda below: The transformation of the Sabino type raising metaphony in N ­ eapolitan diphthongizing metaphony, and in particular the fact that only open-mid vowels diphthongize cannot be explained by Maiden’s (1991) analysis. Calabrese’s (1985 et seq.) account concerning fission of a [+high, −ATR] disallowed segment resulting from metaphony was criticized by Maiden (1991) among others and questioned by the data coming from the dialect of Casalincontrada analyzed above. Furthermore, these analyses do not explain the cases of metaphony only triggered by /i/. The above proposals predict /i/ and /u/ to both trigger the process since [+high] is present in both segments and they both lack A. In addition, Maiden’s analysis does not explain how [a] metaphony yields [je], [ɛ] or [i], in particular where the element I comes from. Finally, /a/ resistance to metaphony is a problem for Maiden’s (1991) analysis, since [a] should be the privileged target for an A demotion process, whereby A undergoes recession in segments when absent from other segments in the domain. The analysis carried out above on Teramo metaphony allowed some elements to surface that could help to shed light on the essence and on the mechanisms of the general process of metaphony, as it originated. They are recapitulated below: – a relationship between demotion of A heading a phonological expression and the creation of structural space yielding diphthongization; – the presence of a component of A demotion also in cases where metaphony may be analyzed as I addition. Can these elements acquired by analysing progressive and regressive metaphony in the dialect of Teramo be helpful in shedding light on some of the questions raised by the diachrony and the implicational hierarchy? Next I try to exploit these elements to speculate on what analysis can be offered of Italo-Romance metaphony that answers the questions raised above. I will discuss the Sabino raising metaphony first. In this kind of metaphony, A demotion in the mid-close vowels [e] (A, I) and [o] (A, U) results in A delinking, since A is not the head of the phonological expression. The only possible demotion is delinking, which, respectively, yields [i] (I), and [u] (U), as borne out by the data (es. /neru, neri/ > [niru, niri], /roʃʃu, roʃʃi/ > [ruʃʃu, ruʃʃi]). In the open-mid vowels, however, A is the head. A demotion in this case means that A loses headship but it is not delinked, yieding (A I) < [ɛ] (A, I), and (A U) < [ɔ] (A U). These representations do not coincide with the one of the midclose vowels [e] and [o], respectively (A I) and (A U), where I and U are head of the phonological expression.



Progressive metaphony in the Abruzzese dialect of Teramo 

 175

By observing the workings of metaphony in the dialect of Teramo, it has been observed that demotion of A from head status creates structural space. Following this working hypothesis one could imagine that the space provided by A demotion yields vowel tensing from the (A I), (A U) representations by providing length. The close relationship between length and tension in vowels is well known and Latin is precisely one of the languages where the effects of this relationship can be best observed, in particular concerning the Proto-Romance outcomes. A length opposition was reinterpreted as a quality opposition. It can be proposed then that because of this acquired length, the phonological expressions (A I), (A U), resulting from the metaphonic process, are spelled out as tense and therefore as the mid-close [e], [o]. One can hypothesize that this same underlying length connected to A demotion, allows these vowels and not mid-close vowels, to manifest themselves as the diphthongs [je], [wo] in the evolution of the Sabino type metaphony in the Neapolitan type. As for /a/, a vowel which resists metaphony and undergoes this process uniquely in restricted areas, the present perspective could explain the reasons of this state of affairs that, as pointed out above, is not consistent with an A demotion ­analysis. According to the present perspective, A is head in /a/. Accordingly its demotion yields a headless phonological expression also containing A, phonetically ­corresponding to schwa (Backley: 2011) or ɐ. At the same time A demotion provides structural space that, according to mechanisms similar to those explained above for the mid-open vowels, yields a full [a] (cf. Lowenstamm 1991; ­Pöchtraeger & Kaye 2010 for a proposal concerning the spelling out of a long schwa as [a]). Under this perspective, metaphony applies vacuously to /a/, thus /a/ resistance to metaphony is only apparent. Why though, if metaphony applies vacuously to /a/, there are some areas, although restricted, where /a/ undergoes metaphony also on surface? An answer to this question orginates from the observation that [a] metaphony is documented almost uniquely in areas (eastern Abruzzo, Molise, central Romagna, Canton Ticino) where only /i/ metaphony is attested. I stressed above that the presence of metaphony only triggered by one of the high vowels, namely /i/, constitutes a problem for an analysis in terms of [+high] spreading or A demotion. What I propose here is that, in the evolution and spreading of metaphony in the peninsula, there are areas where this process has been re-analyzed, based on raising of the front vowels, as an A demotion process with I addition, as seen for the dialect of Casalincontrada. This would explain why there are areas where /u/ does not trigger the phenomenon and why only in these same areas /a/ undergoes metaphony yielding [je]10.

10 There is reason to believe that [ɛ] and [i] derive from monophthongization of [je].

176 

 Diana Passino

According to the perspective underlying this contribution, this diphthong is what we expect from a process demoting A from its head status, with I addition. A diphthong containing A and I where A is not head. In this contribution, by analysing progressive and regressive metaphony in the dialect of Teramo I underlined the role of A in this process, by adding some observations to the ones made by Maiden (1991) and refining his proposals concerning the role of A in Italo-Romance metaphony. I observed that the notion of head is important in order to predict diphthongization in metaphony, that the behavior of A suggests a structural more than an elemental identity, and that A demotion can also be part of a metaphonic process also including I addition. Metaphony may consist of element diffusion but it is tentatively claimed that the essence of the process consists in A demotion. This also explains why, as observed by Carvalho & Russo (2007), generally there are no lowering metaphonies. No linguistic variation with respect to the element spread in the metaphonic process that allows for A spreading is observed crosslinguistically. Besides, A is not exactly an element. The extension, the several further developments and branching, and the morphologization of Italo-Romance metaphony make it impossible to propose an analysis that is not tentative since any analysis needs to be tested on an enormous amount of synchronic and diachronic data. However, even if this tentative proposal were not backed up by all available data on metaphony, I hope to have shown with this contribution that a formal synchronic analysis on metaphony may be useful both on the descriptive, empirical side and on the theoretical. In this case it allows refinement of the incomplete picture that the literature had drawn of the dialect of Teramo, as well as suggesting a path of research that may perhaps eventually lead to enlarge our understanding of metaphonic processes.

References AIS = Karl. Jaberg & Jakob. Jud, Sprach-und Sachatlas Italiens un der Südschweiz, Zofingen, 1928–40 (8 voll.) ALI = Mario. Bartoli et alii, Atlante Linguistico Italiano, in stampa presso il Poligrafico dello Stato. Anderson, John & Jacques Durand. 1986. Dependency phonology. In Jacques Durand (ed.), Dependency and non-linear phonology, 1–54. London: Croom Elm. Backley, Philip. 2011. An introduction to element theory. Edinburgh: Edinburgh University Press. Calabrese, Andrea. 1985. Metaphony in Salentino. Rivista di grammatica generativa 9–10: 3–140. Calabrese, Andrea. 1998. Metaphony revisited, Rivista di linguistica 10(1): 7–68.



Progressive metaphony in the Abruzzese dialect of Teramo 

 177

Calabrese, Andrea. 2011. Metaphony in Romance. In M. van Oostendorp, C. J. Ewen, E. Hume and Keren Rice (eds.), The Blackwell companion to phonology, 2631–2661. Oxford: Blackwell. Carvalho, Joaquim Brandão de & Michela Russo, 2007. Y a-t-il des métaphonies ouvrantes en roman? In D. Trotter (ed.), Actes du XXIVème Congrès International de Linguistique et Philologie Romane, 43–56. Tübingen: Niemeyer. Clements, George, N. 1991. Vowel height assimilation in Bantu languages. Working Papers of the Cornell Phonetics laboratory 5: 37–76. Colman Fran & John Anderson. 1983. Front Umlaut: a celebration of 2nd fronting, i-umlaut life, food and sex. In Mike Davenport, Erik Hansen and Hans Frede Nielsen (eds.), Current topics in English historical linguistics, 165–190. Odense: Odense University Press. D’Alessandro, Roberta & Marc van Oostendorp. 2014. La metafonia fra fonologia e lessico: il caso dell’ariellese. In D. Pescarini & D. Passino (eds.), Studi sui dialetti dell’Abruzzo. Quaderni di lavoro ASIT 17. De Giovanni, Marcello. 1973. Il vocalismo tonico abruzzese secondo alcuni studi precedenti. Abruzzo 11: 46–84. De Giovanni, Marcello. 2003. Molise. Pisa: Pacini. Delfico, Orazio. 1912–1913. Il medico sensale dei matrimonj, commedia in dialetto ­teramano, Teramo, Rivista Abruzzese, Teramo, Arti grafiche A. De Carolis, (posthumous ­publication). De Lollis, Cesare. 1890–92. Dell’influsso dell’i e del j postonico sulla vocale accentata in qualche dialetto abruzzese, Archivio Glottologico Italiano 12. 1–23, 187–196. D’Ovidio, Francesco. 1886. Ricerche sui pronomi personali possessivi. Archivio Glottologico Italiano IX: 25–101. Dobson Eric. J. 1968. English pronunciation 1500–1700. Oxford: Clarendon Press. Farnetani, Edda. 1981. Dai tratti ai parametri: introduzione all’analisi strumentale della lingua somala. In G.R Cardona e F. Agostini (a cura di) Studi Somali I. Fonologia e lessico. Roma Ministero Affari Esteri. Giammarco, Ernesto. 1979. Abruzzo. Pisa: Pacini. Harris, John & Geoff Lindsey 1995. The elements of phonological representation. In J. Durand & F. Katamba (eds.), Frontiers of Phonology, Harlow, 34–79, Essex: Longman. Kenstowicz, Michael. 1994. Phonology in generative grammar. Oxford: Blackwell. Kiparsky, Paul. 1974. A note on the vowel features. In E. Kaisse and J. Hankamer (eds.), Proceedings of NELS 5. Cambridge MA: Harvard University Linguistics department. Kiparsky, Paul. 1995. Phonological basis of sound change. In J. Goldsmith (ed.), The handbook of phonological theory. Oxford: Blackwell. Labov, William. 1981. Resolving the neogrammarian controversy. Language 57: 267–308. Lowenstamm, Jean. 1996. CV as the only syllable type. Current trends in phonology. Models and methods. In Jacques Durand & Bernard Laks (eds.), 419–441. Salford, Manchester: ESRI. Loporcaro, Michele. 2011. Phonological processes. In Martin Maiden, John Charles Smith and Adam Ledgeway (eds.), The Cambridge history of the Romance languages. Volume 1, structures, 109–154. Cambridge: CUP. Maiden, Martin. 1991. Interactive morphonology. Metaphony in Italy. London and New York: Routledge. Pöchtrager, Markus. 2006. The structure of length. Doctoral dissertation, University of Vienna. Pöchtrager, Markus & Jonathan. D. Kaye. 2010. Dial A for adjunction. Hand out. Old World ­Conference in Phonology 7 Nice 28–30 January.

178 

 Diana Passino

Rizzolo, Olivier. 2002. Du leurre phonétique des voyelles moyennes en français et du divorce entre licenciement et licenciement pour gouverner. Thèse de Doctorat, Université Nice Sophia Antipolis. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti. Fonetica. Einaudi: Torino. Rolin, Gustav. 1908. Die Mundart von Vasto in den Abruzzen. Prager deutsche Studien 8: 477–504. Romani, Fedele. 1907. Abruzzesismi. Firenze: Bemporad & Figlio. Russo, Michela. 2007. La metafonia napoletana: evoluzione e funzionamento sincronico. Bern: Peter Lang. Sánchez Miret, Fernando. 1999. Assimilazione a distanza fra vocali nei dialetti d’Italia: fonetica e spiegazione del cambiamento, Fonologia e Morfologia dell’italiano e dei dialetti d’Italia. Atti del XXXI Congresso della Società di Linguistica Italiana, Paola Benincà, Alberto Mioni & Laura Vanelli (eds.), 269–290. Roma: Bulzoni. Savini, Giuseppe. 1881. La grammatica ed il lessico del dialetto teramano. Loescher: Torino. Savoia, Leonardo & Martin Maiden. 1997. Metaphony. In Martin Maiden and Mair Parry (eds.), The dialects of Italy, 15–25. London: Routledge. Scheer, Tobias. 1995. Halbechte Rektion in germanischem Wandel und althochdeutscher Brechung. Linguistische Berichte 160: 470–511. Scheer, Tobias. 2004. A lateral theory of phonology. Vol 1: What is CVCV, and why should it be? Berlin: Mouton de Gruyter. Scorretti, Mauro. 2012. Il dialetto di Antrodoco. Pescara: Jadran. Ségéral, Philippe & Tobias Scheer. 2001. Abstractness in phonology: the case of virtual geminates. In Katarzyna Dziubalska-Kolaczyk (ed.), 311–337. Constraints and Preferences. Berlin & New York: Mouton de Gruyter. Tekavčić, Pavao. 1972. Grammatica storica dell’italiano, 1ª ed., Bologna, Il Mulino. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language and Linguistic Theory 23: 917–989. Ziccardi, Giovanni. 1910. Il dialetto di Agnone. Zeitschrift für Romanische Phylologie 34: 405–436.

Mirella De Sisto

Metaphony and Raddoppiamento Fonosintattico in plural nouns in the dialect of Airola

Meertens Institute

Abstract: This paper aims to give an overview and an analysis of metaphony and Raddoppiamento Fonosintattico in the southern Italian dialect of Airola. The two processes affect the nominal class of the dialect. In particular, they are both attested in plural forms. Nevertheless, they never co-occur in the same plural noun and appear to have acquired a morphological distinctive value. In fact, metaphony can only take place in masculine nouns, while Raddoppiamento Fonosintattico is attested in feminine plural ones. They are two distinct phenomena, one being phonological and the other being phono-syntactic, and they developed separately. Yet, synchronically speaking, they happen to represent gender distinction.

1 Introduction In the southern Italian dialect of Airola (­north-­eastern part of Campania), the nominal class is affected by two phenomena, namely metaphony and Raddoppiamento Fonosintattico (RF henceforth). The two phenomena appear to be in complementary distribution and to create morphological distinctions in plural nouns: metaphony marking masculine plural and RF marking feminine plural. They are distinct, the former being phonological and the latter being ­phono-­syntactic, and developed separately. Yet, synchronically speaking, they acquired a value of gender distinction. The aim of this article is to explore and give an analysis of metaphony and RF in Airolano in order to give an account of their complementary distribution. The work is based on data from Airolano that was collected in December 2013 and April 2014 by the author. Ten informants were selected in order to form four age groups. The corpus contains common nouns, modern nouns, and nonce words. Firstly, the participants were asked to translate from Standard Italian to Airolano the common and modern nouns preceded by the definite article and to give their plural forms. Then, they were asked to give the plural of the nonce words. All the recordings were transcribed in IPA and they appear in this form in the text. The full set of data is stored in the Italian Dialect archive of Leiden University and in De Sisto (2014).

180 

 Mirella De Sisto

The article is structured as follows. In the next section (section 2), metaphony and RF in Airolano are presented. In section 3, an analysis of the two phenomena takes place. Firstly, some information on the theoretical frameworks the analysis is based on, namely Element Theory and CVCV Theory, is given. Secondly, metaphony is analyzed within Element Theory. Thirdly, a CVCV analysis of RF is presented. In section 4, the complementary distribution of the two phenomena within the nominal class of the dialect is described. Then, their relation in distinguishing gender in the plural of nouns is discussed. Finally, a conclusive section follows.

2 Metaphony and Raddoppiamento Fonosintattico Metaphony and RF are two distinct processes that are both attested in the nominal class of Airolano. They appear to be in complementary distribution by distinguishing masculine plural from feminine plural. In fact, metaphony only takes place in masculine nouns, while RF only affects feminine ones. In this section the two phenomena are described. First, some information about metaphony in Airolano is given (section 2.1). Afterwards, in section 2.2, RF affecting the dialect is presented.

2.1 Metaphony in Airolano Metaphony consists of the raising or diphthongization of a stressed vowel under the influence of a ­non-­adjacent following high vowel (Rohlfs 1966; Maiden 1991; Fanciullo 1994; Ledgeway 2009). In the dialect of Airola, metaphony affects mid vowels, namely /ɔ, o, e, ɛ/, which are raised or diphthongized as follows: (1) metaphony in Airolano: o>u

e>i

ɔ > wo

ɛ > je

As can be observed in the table in (1), the high mid vowels [o] and [e] are raised to [u] and [i], respectively. Diphthongization takes place, instead, in the case of low mid vowels, namely [ɔ] and [ɛ], the former becoming [wo], while the latter becomes [je]. This pattern was already described for other southern Italian dialects (Maiden 1991: 141). Metaphony is attested in different word categories of the dialect, such as nouns, possessive pronouns, verbs, and adjectives. In particular, within the verb

Metaphony and RF in plural nouns in the dialect of Airola 

 181

class, present indicative, past simple, past participle, and imperative present metaphony (De Sisto 2014). The distribution of the phenomenon confirms what was observed for Neapolitan (Ledgeway 2009: 5­ 7–­65) and for other dialects of the area of Benevento (Maturi 2002). Due to a matter of space, only the occurrence of metaphony within the nominal class will be taken into account in the present article. In the nominal word category, metaphony only takes place in masculine nouns. Furthermore, in a group of nouns, it only marks plural forms. This group of nouns originally derived from the Latin third and fifth declension (Ledgeway 2009: 6 ­ 0–­1; Maturi 2002: 140); but synchronically includes also other nouns, such as modern words. In (2), examples of nouns presenting the four types of metaphony are given. (2)  Metaphony in nouns in Airolano a. ˈmɔnəkə b. waˈjonə c. ˈmesə monk-sg boy-sg month-sg ˈmwonəʃə waˈjunə ˈmisə monk-pl boy-pl month-pl 

d. ˈrɛntə tooth-sg ˈrjentə tooth-pl (De Sisto 2014: 13)

In (2), cases of both raising and diphthongization are presented. No w ­ ord-­final high vowel causing metaphony is overtly visible. This is due to vowel reduction, according to which all w ­ ord-­final vowels are reduced to schwa. I assume, following Maturi (2002), that the reduction occurs synchronically. This process will be outlined in the analysis section (2.2). According to the data that were collected (De Sisto 2014), metaphony is proven to be still productive, despite not applying systematically throughout the corpus. The productivity is further supported by the fact that the younger participants extended its occurrence also to modern words. Moreover, taking into account the nonce words of the corpus, the attestation of metaphony appears to be increasing in younger generations (De Sisto 2014: 26).

2.2 Raddoppiamento Fonosintattico in Airolano RF is a sandhi phenomenon which consists of the gemination of a w ­ ord-­initial consonant and is triggered by a preceding vowel. There are two main types of RF, a phonologically triggered one, which is attested in Tuscan dialects and Standard Italian (Nespor & Vogel 1986; Fanciullo 1986; Loporcaro 1997), and a lexically encoded one, typical of Southern Italian varieties (Chierchia 1986; L ­ oporcaro 1997; Borrelli 2002). In the former, an oxytone word causes the phenomenon, while in the latter a specific word or particle is lexically encoded as a trigger. From

182 

 Mirella De Sisto

a diachronic point of view, lexically encoded RF is due to the loss of a w ­ ord-­final consonant which left a trace in the underlying representation of the lexical item (Loporcaro 1997); the empty space is then filled by the following ­word-­initial consonant, which undergoes gemination (Borrelli 2002). L ­ exically-­induced RF only occurs when trigger and target constitute a minimal phrase, which can be defined as “a kind of hierarchically superior word” (Fanciullo 1986: 88). In Airolano, RF is lexically based, hence the process takes place under the influence of specific trigger words. In some instances, however, gemination fails to occur. As observed by Fanciullo (1986: 70), gemination does not affect consonant clusters, already geminated consonants, or those having an already reinforced articulation. Furthermore, some consonants do not only lengthen but also change place and (eventually) mode of articulation, when undergoing RF (Fanciullo 1986: ­70–­71). In Airolano, this is the case of [v], [r], [j], [w] and [ʃ], which turn into [v], [r], [j:], [gw:] and [tʃ], respectively, when RF occurs (see De Sisto 2014: 17). The lexical triggers of RF in Airolano and other Southern Italian dialects are constituted by some prepositions, the numerals ‘two’ and ‘three’, clitic pronouns and mass and feminine plural determiners, demonstrative, indefinite and qualifying adjectives, the third person singular of both auxiliaries, etc. (for a full list see Maturi 2002: 1­ 08–­121). Much research was carried out regarding RF marking mass nouns, and, in particular, when being triggered by the definite article (Loporcaro 1997; Borrelli 2002; Ledgeway 2009). However, not much documentation is available regarding the feminine plural definite article as being a lexical trigger of RF (Iannucci 1948; Agostiniani 1975). Nevertheless this property of the feminine plural definite article is attested in various dialects of Sannio Beneventano (see Maturi 2002) and in Airolano (De Sisto 2014). Diachronically, the feminine plural definite article derived from the Late Latin *ILLAEC (Loporcaro 1997: 49; Borrelli 2002: ­29–­35). The determiners of southern Italian dialects are formed from the final part of the Latin demonstrative: *(ILL)AEC > e. (3) Trace of the lost consonant in the representation of e x x # x x x e

f

art.def.f.sg

e

x x x x m

ǝ n ǝ

woman-pl

As can be observed in (3), the w ­ ord-­final consonant is lost but some traces of it are preserved in the underlying representation. It is this empty slot, then, which causes RF. According to the data that were collected (De Sisto 2014), RF shows to be productive and to apply systematically in all instances in the corpus of a feminine plural definite article followed by a feminine plural noun. No difference in its distribution among the various age groups was attested.

Metaphony and RF in plural nouns in the dialect of Airola 

 183

3 Analysis In this section, the theoretical frameworks, namely CVCV theory and Element Theory, are, firstly, introduced (section 3.1 and 3.2). Second, an ­element-­based analysis of metaphony in Airolano is proposed (section 3.3). Third, in section 3.4, RF is analyzed within CVCV theory.

3.1 CVCV theory CVCV theory assumes that CV constitutes the only syllable type (Lowenstamm 1996). To explain, syllables are considered to consist in ­non-­branching Onsets and ­non-­branching Nuclei and the minimal syllabic unit to be formed by an Onset and a Nucleus together, without Codas or Rhymes (Scheer 2004: 1) (see 4). (4)  Closed syllable

Geminate

Long vowel O N

O N

C

V

O

N

O

N

O N

O N

C

V

C

Ø

C

V

(Scheer 2004: 1)

Given that no branching constituents are possible in the phonological representation, the presence of empty positions for closed syllables has to be assumed. These empty nuclei are regulated by the Empty Category Principle, which states that an empty nucleus might remain empty under precise conditions or if properly governed (Scheer 2004). In particular, ­word-­finally, governing depends on parametric variation (Passino 2013). In Italian and in Italian varieties, including Airolano, ­word-­final empty nuclei are not allowed.

3.2 Element Theory In Element Theory, vowels and consonants are the result of the combination of elements. There are six elements, which are divided in two groups, one as being mainly related to the constitution of vowels, while the other consists of the elements forming consonants (Backley 2011). What is relevant here is the group of elements forming vowels. Vowels are normally composed of one or two of the three elements |I|, |U| and |A|, which indicate palatality, labiality, and height, respectively (Backley 2011: 23). When two elements are combined in the composition of a vowel, one of them occurs to

184 

 Mirella De Sisto

be more prominent and it constitutes, hence, the head (Maiden 1991: 138). The table in (5) shows the composition of vowels, the prominent element being the leftmost one: (5)  Vowels in element theory [a] = |A| [e] = |I, A| [ɛ] = |A, I| [o] = |U, A| [ɔ] = |A, U] [u] = |U| [i] = |I| The three elements, |A|, |I| and |U|, are considered to be monovalent and ­symmetrical. However, some kind of asymmetry between |A| and the other two ­elements has been noticed by various linguists, but it has not been formalized yet. According to Pöchtrager (2006), the element |A|, differently from the other elements, has structural properties (Pöchtrager 2006: 61) and, hence, can allow an adjunction structure (Pöchtrager 2006: 165). A representation of the claim is given in (6): (6) Representation of elements following Pöchtrager (2006) |A| xN1 xN1

→

|I|/|U| xN1 x2

In (6), the structures of |A|, |I| and |U| are given. |A| appears to have a more complex structure. As can be observed, in the representation of |A,| a time slot (a skeletal slot, according to Pöchtrager 2006) is broken up in two levels and the position at the lower level is attached to another slot. A similar postulation is, however, problematic, and it is not completely clear what this structural difference represents. However, the data from Airolano seem to support the hypothesis of structural properties of |A| as consisting in some kind of extra length. The following section will clarify this statement.

3.3 Metaphony as demotion of |A| According to Maiden (1991), metaphony involves the demotion of the element |A| in the vowel in stressed position. The demotion is caused by the absence of the

Metaphony and RF in plural nouns in the dialect of Airola 

 185

element |A| in the following ­post-­tonic vowel (Maiden 1991: 140). In some cases, the element |A| loses its prominent position, while in other cases it is completely lost, as it is illustrated in the table in (7): (7)  Metaphony as demotion of |A| |I, A| (= /e/) → |U, A| (= /o/) → |A, I| (= /ɛ/) → |A, U| (= /ɔ/) →

|I| |U| |I| |I, A| |U| |U, A|

(= /i/) (= /u/) (= /je/) (= /wo/)

(Maiden 1991: 140)

As can be observed, in the first two cases, namely /e/ and /o/, the element |A| is completely lost in the process. In metaphony resulting in diphthongization, the element loses its prominent position; what can seem the addition of an extra element is only the consequence of the demotion of |A| from being the head element. The relative examples will clarify these statements. But first, another aspect of the process needs to be addressed, namely the synchronicity of vowel reduction. As observed in section (2.1), due to reduction, every ­word-­final vowel is reduced to schwa, hence no vowel causing metaphony is overtly visible. Yet, the trigger has to be there. The possibilities can be two: either a floating element is still present or vowel reduction is synchronic and metaphony precedes its occurrence. Following Maturi (2002), this work assumes reduction to be synchronic. Consequently, it is considered to be involved together with metaphony in a process defined by Kiparsky (1973) as “counterbleeding opacity”, which consists of a rule A bleeding a rule B but only after the rule B has taken place (see table in 8). (8)  Counterbleeding opacity ˈʃorə ˈʃurə flower-sg flower-pl

1. Underlying 2. Metaphony 3. Reduction 4. Surface

ˈʃori ˈʃuri ˈʃurə ˈʃurə

In (8), in the underlying form, the ­word-­final vowel is still present; in addition, because metaphony did not occur yet, the ­non-­metaphonic vowel is still attested. In the second phase of the process, metaphony takes place and the vowel is raised, but the final vowel is still preserved. Afterwards, in the reduction phase, the ­word-­final vowel is lost. Finally, in the surface form both processes already occurred, hence the trigger of metaphony is opaque but the metaphonic vowel displays.

186 

 Mirella De Sisto

After this preliminary observation, it is possible now to propose the analysis of metaphony as demotion of |A|. The first case considered is the one of metaphony causing raising. In (9) an example is given: (9) Metaphony consisting of raising ˈʃorə ˈʃurə flower-sg flower-pl |I|

|U| C

V

C

V

∫

o |A|

r

ǝ |A|

|U| →

C

V

C

V

∫

u |A|

r

ǝ |I|

In (9) the CV structure and the elements forming the vowels of the noun ˈʃorə in its singular and plural form are represented. Regarding the w ­ ord-­final vowels, their elements before the occurrence of reduction are considered. In the singular form, two elements are attached to the rightmost V position, namely |I| and |A|; however, in the plural form, only one element is attached to the last V, namely, |I|. As the element |A| is absent from that position, a similar situation has to display in the elements of the V of the stressed syllable. Consequently, the element |A| of the stressed vowel is detached from the V position and is lost. Only afterwards does vowel reduction occur. The case of metaphony resulting in diphthongization consists of a similar process (see 10). (10) Metaphony consisting of diphthongization ˈpɛrə ˈpjerə foot-sg foot-pl |I|

|I| C V

C

p

ε |A|

V

C

V

r

ǝ |A|

|I| →

C

V

p

j

C |A|

V

C

V

e

r

ǝ |I|

(10) shows the CV structure and the elements constituting the vowels of the singular and plural form of ˈpɛrə. As can be seen, also in this case the w ­ ord-­final vowel of the singular noun has two elements attached, |I| and |A|, while the plural one only has |I|. The difference with (9) is that the |A| of the stressed vowel, which is the head element, allows a bigger structure; as a consequence,

Metaphony and RF in plural nouns in the dialect of Airola 

 187

the vowel covers two CV positions. In the plural form, |A| has to be detached from one V position because of the absence of the same element from the ­word-­final V position. As a consequence, it only loses the link to one of the V positions and remains attached to the second one. The |I| element of the stressed position remains attached to both positions but cannot preserve the integrity of the vowel. Therefore, the two V positions split and the vowel breaks into a ­diphthong, resulting into /je/. What is problematic about this approach is the actual trigger of the phenomenon. In fact, in most cases of vowel harmony, the stressed vowel usually causes the change while the unstressed one is the target of it (see Baković 2003; Nevins 2010). Metaphony is triggered by a vowel in ­post-­tonic position, which is the weakest position, according to Walker (2011) (see 11), and the target is the vowel in the strongest position, namely the stressed one. (11)   V/ strong (stressed) > V/Weak (­Pre-­tonic stem) >  > V/­Extra-­Weak (­Post-­tonic, Unstressed clitic)

(Walker 2011: 269)

A possible explanation for this peculiarity of metaphony is proposed by Walker (2005), who claims that high vowels are normally less perceivable and that they become even more difficult to be perceived when occurring in weak position. Consequently, the feature, or in this case the lack of an element in the vowel in weakest position, needs to be spread to the vowel in strongest position. But Walker’s (2005) hypothesis explains feature spreading, while, in metaphony it is the lack of the element that spreads. It is not easy to formalize the “spreading of an absence”. Van der Hulst (forthcoming) proposes the addition of the element A | | in the vowel representation. However, this account presents various problems, which are discussed by the author himself and would require the modification of element theory’s framework, which considers the elements to be autonomously interpretable. Ouddeken (2013) also tries to formulate various strategies but a modification of the theory seems to be the only solution. Nevertheless, despite the problem of accounting for the “spreading of an absence”, the assumption of a particle demotion, to my knowledge, is what gives a better account of the evidence coming from the Airolano data.

3.4 Raddoppiamento Fonosintattico within CVCV framework Passino (2013) proposes an analysis of the different types of RF within the CVCV framework. Focusing on lexically encoded RF and recalling that it is due to

188 

 Mirella De Sisto

some consonant length which is still preserved in the underlying representation of the lexical item, she claims that this trace consists in a ­word-­final empty CV. In other words, the empty CV was once linked to a ­word-­final consonant; when the consonant was lost, the CV was kept and it is still present lexically. But it lacks an element to be attached to. This is also the case of the feminine plural definite article as trigger of RF. Consequently, the empty CV attaches to the following ­word-­initial consonant causing its gemination, as the example in (12) shows: (12) a art.def.f.sg C V C

V#

e

p

pǝtt∫ǝˈrellǝ girl-sg C

V e

C

V

ppett∫ǝˈrellǝ girl-sg

e art.def.f.pl C

t∫

V

C

V

ǝ

r

e

C

V l

C

V ǝ

In (12), the empty CV of the determiner is realized by geminating the initial consonant of the following word, namely /p/. The gemination is created through a link between the empty CV and the following ­word-­initial consonant. This link entails that the consonant is lengthened. The presence of an extra CV in the feminine plural determiner is more evident when the article happens to be isomorphic with the masculine plural one (it was so for two speakers of the data collected in De Sisto 2014), as in (13): (13) o art.def.m.sg a art.def.f.sg

təlefoˈninə cellphone-m.sg ˈpennə pen-f.sg

e art.def.m.pl e art.def.f.pl

təlefoˈninə cellphone-m.pl ˈppennə pen-f.pl

In (13), despite the isomorphism of the two articles, RF is only triggered by the feminine plural determiner preceding the word for pen. This confirms the presence of an extra CV in the case of the feminine plural article and its absence in the representation of the masculine plural one. The empty CV of the feminine plural determiner appears to be never filled by the lengthening of the vowel of the determiner itself. According to Chierchia (1986: 12), in both s­ tress-­induced and lexically encoded RF, this cannot occur because stressed vowels in ­word-­final position cannot be long. In addition, he assumes that, across word boundaries, consonantal lengthening is the only attested repair strategy. In the case of the feminine determiner, I wish to add another reason for the impossibility of the vowel to take over the extra position: the determiner is a clitic in p ­ re-­tonic position, hence, it cannot be long. In fact, even in a

Metaphony and RF in plural nouns in the dialect of Airola 

 189

case where it was lengthened, vowel reduction would take place and the vowel would be shortened again.

4 C  omplementary distribution of metaphony and RF In this section, first, a summary of the distribution of metaphony and RF in the corpus is presented and their interaction discussed. Second, an element-based analysis of the vowel suffixes of masculine and feminine nouns and of their definite articles is proposed. This aims to shed light on the meaning of the morphological distinction, which is synchronically represented by the two processes of metaphony and RF. Third, the element based analysis is combined with the CV representation of the noun phrases and its outcome is described.

4.1 Complementary distribution of metaphony and RF The data that were collected for this research show a pattern of complementary distribution of metaphony and RF. In fact, while metaphony only takes place with masculine plural nouns, RF only occurs with feminine plural ones. Furthermore, the two phenomena cannot ­co-­occur, even when the right conditions for the occurrence of both are met, as shown in (14) and (15): (14) a art.def.f.sg (15) o art.def.m.sg

təleviˈsjonə television-f.sg piˈstonə piston-m.sg

e art.def.f.pl i art.def.m.pl

ttəleviˈsjonə television-f.pl piˈstunə piston-m.pl

In (14) and (15) two nouns preceded by a definite article are given. The noun in (14) is feminine and the one in (15) masculine. Not surprisingly, only the feminine noun in (14) presents RF when preceded by the feminine plural definite article. On the other hand, in both nouns the right conditions for metaphony to occur are met. Both nouns are borrowed from Italian and originally have a suffix -e in the singular and -i in the plural (televisione/televisioni, pistone/pistoni). Therefore, both plural forms in (14) and (15) would be expected to undergo metaphony. This, however, does not occur and metaphony is attested only in the masculine plural noun. As a consequence, the two phenomena appear to be in complementary distribution to mark the plural of nouns: metaphony marking masculine plural

190 

 Mirella De Sisto

and RF marking feminine plural. RF and metaphony have converged into the ­formation of a morphological opposition, despite being completely distinct.

4.2   Analysis of the complementary distribution of metaphony and RF Taking into consideration Passino’s (2009) element-based analysis of Standard Italian inflectional noun suffixes, I propose a similar analysis of the inflectional noun suffixes and of the definite articles of the dialect of Airola. According to Passino (2009: 65), inflectional noun suffixes are composed by one element referring to the word class of the noun and another element indicating number. Regarding the elements referring to number, she assumes that the element |A| indicates singular, while |I| indicates plural (Passino 2009: 66). When applying these assumptions to the Airolano data, the following classification can be observed: (16)   Inflectional suffixes in Airolano



Singular

Plural

Masculine 11

/ɔ/ = |A, U|

/i/ = |I|

Masculine 2

/ɛ/ = |A, I|

/i/ = |I|

Feminine

/a/ = |A|2

/ɛ/ = |A, I| (De Sisto 2014: 41)

The three noun groups in (16) have the element |A| in their singular and |I| in their plural. By comparing the plural forms of the masculine groups with the plural of the feminine one, it can be noticed that, while the formers present only one element, the latter has two elements. In fact, the plural forms of the masculine groups only have the element indicating plural, namely |I|, and lack the one for class. Consequently, the masculine nouns appear to be weaker than the feminine ones in the plural form. To clarify, some examples are given in (17), where the vowels in brackets represent the vowel before undergoing reduction.

1 Some of the nouns are synchronically considered to be part of this group derived from the Latin fourth declension, hence, they end in /u/. But they were reanalyzed as being part of the other group because of the influence of Standard Italian on the dialect. Moreover, they present a ­lexicalized metaphony both in the singular and plural which is not relevant to the current analysis. 2 I assume that two |A| elements, one for class and another for number, mark the feminine singular but, due to the impossibility of two |A|’s to display, only one appears.

Metaphony and RF in plural nouns in the dialect of Airola 

 191

(17) a. ˈpɔrtə (/ɔ/) = |A, U| – ˈpwortə (/i/) = |I| harbor-sg           harbor-pl b. ˈprɛvətə (/ɛ/) = |A, I| – ˈprjevətə (/i/) = |I| priest-sg           priest-pl c. ˈprɛtə (/a/) = |A| – ˈprɛtə (/ɛ/) = |A,I| stone-sg           stone-pl The nouns in (17a) and (17b) are masculine. Their singular forms present two elements, namely |A, U| and |A, I|, respectively, while their plurals only have the element |I|. In the singular of the feminine noun in (17c), the |A| displays and in the plural two elements, |A| and |I| are present. Considering definite articles, the same situation is found (see table in 18). In fact, while the feminine plural determiner has both class and number elements, the masculine plural only presents the element |I|, which indicates number. The mirror situation displays in the singular, where the feminine determiner only has |A|, while the masculine has |U| and |A|. (18)   Elements constituting the definite articles Singular o = |U, A| a = |A|

Masculine Feminine

Plural i = |I| e = |I, A|

The difference between the plural forms of the two genders becomes more evident when combining the element-based analysis with the CV representation of the noun phrase. On a noun level, the difference is shown by the presence of metaphony in the masculine plural and its absence in the feminine plural. In fact, it is the lack of a class element in the masculine plural noun that leads to the attestation of metaphony. On a definite article level, the masculine plural article appears to be two times weaker, in that the feminine plural one is composed of two elements and has an extra CV due to historical reasons. (19) and (20) exemplify the differences: ˈpiʃʃə (19) o ˈpeʃʃə - i fish-m.sg art.def.m.pl fish-m.pl art.def.m.sg |U| |I| C V# C o |A|

p

V C

V

e

∫

C

V ǝ |A|

→

C V#

C

V C

V

C V

i |I|

p

i

∫

ǝ

|I|

192 

 Mirella De Sisto

(20) a art.def.f.sg

ˈkɔʃʃə leg-f.sg

-

e art.def.f.pl

ˈkkɔʃʃə leg-f.pl

|I| C V# C V C V C V C V a k |A|

ɔ

∫

ǝ |A|

→

C

|I|

V C V# C V C V C V C V e |A|

k

ɔ |A|

∫

ǝ |A|

The plural form of the masculine noun in (19) presents one element marking number, while the plural form of the feminine noun has two elements, namely |A| indicating class and |I| for number. The same difference of elements is displayed by the definite articles, but a further difference is added to it, namely the extra CV present in the feminine plural article. It can be concluded that the division of labor of two different phenomena, such as RF and metaphony, acquired morphological value and casually highlighted the asymmetry between masculine plural and feminine plural, showing the weakness of the former. This division of labor became part of the grammar of the dialect and was extended to other cases. For instance, in (14) metaphony fails to take place because it acquired the value of gender marking, namely masculine, and, consequently, cannot mark feminine. To sum up, synchronically speaking, the two processes create morphological distinction by differentiating feminine plural from masculine plural. In addition, their interaction also reveals a structural weakness of masculine plural in comparison to feminine plural.

5 Discussion and conclusion In this article, metaphony and RF attested in Airolano were described. They both affect the noun class of the dialect and mark gender distinctions within the plural. In fact, metaphony occurs in masculine plural, while RF in feminine plural. The two phenomena were analyzed and their division of labor presented. Their occurrence and their complementary distribution was supported by the data collected by the author. The analysis has shown the differences on different levels between masculine and feminine plural. However, further research is needed to pin down the exact cause of this distribution. Both phenomena seem to be connected to some extra material: in the case of RF, this needs to be expressed, while, in the case of metaphony, it needs to be eliminated, at least partially. The intuition, then, would be that RF entails preservation, while metaphony involves weakening or loss.

Metaphony and RF in plural nouns in the dialect of Airola 

 193

Another aspect which would need to be further investigated is the value of the element |A| and its relation to both phenomena. Recalling the hypothesis of it being able to cover two CVs and, in that way, representing length in some way; this seems to be supported by metaphony consisting of diphthongization. However, regarding RF, the |A| element of the determiner would be expected to take over the empty CV. But this could be prevented by other factors, such as, for example, the fact that the determiner is a ­pre-­tonic clitic or that word final vowels cannot be long (Chierchia 1986). Finally, one last problem is constituted by the actual trigger of metaphony as demotion of |A|. As was mentioned above, it is not easy to postulate that the lack of an element needs to spread. The various attempts to solve this problem (Van der Hulst, Ouddeken 2013) only lead to the hypothesis of modifying the element theory framework. However, despite the theoretical flaws, metaphony as demotion of |A|, to my concern, is the approach that, at the moment, can better account for the data collected from Airolano.

References Agostiniani, Luciano. 1975. Rafforzamento sintattico e articolo in alcuni dialetti meridionali. Archivio glottologico italiano 60: ­192–­300. Backley, Philip. 2011. An introduction to Element Theory. Edinburgh: University Press. Bakovic, Eric. 2003. Vowel harmony and stem identity. San Diego Linguistics Papers 1: ­1–­42. http://roa.rutgers.edu/files/­540-­0902/­540-­­0902-­­BAKOVIC-­­0-­6.PDF. Borrelli, Doris. 2002. Raddoppiamento Sintattico in Italian: a synchronic and diachronic ­cross-­dialectal study. London/New York: Routledge. Chierchia, Gennaro. 1986. Length, syllabification, and the phonological cycle in Italian. Journal of Linguistics 8: 5–33. De Sisto, Mirella. 2014. Complementary distribution of Metaphony and Raddoppiamento Fonosintattico in plural nouns in Airolano. Research Master Thesis. Leiden: Leiden ­University. Fanciullo, Franco. 1986. Syntactic Reduplication and the Italian dialects of the ­Centre-­South. Journal of Italian Linguistics 8 (1): ­67–103. Fanciullo, Franco. 1994. ­Morfo-­metafonia. In Cipriano, Di Giovine e Mancini (eds.), Miscellanea di studi linguistici in onore di Walter Belardi, ­571–592. Roma: Il Calamo. II. Iannucci, James. 1948. Gemination of initial consonants and its semantic function in Neapolitan. Romance Philology 2: ­237–39. Hulst, H.G. van der. 2015. Lowering Harmony in Bantu. In Sabrina Bendjaballah, Noam Faust, Mohamed Lahrouchi, Nicola Lampitelli (eds.), The form of structure, the structure of forms: Essays on the realization of linguistic structures, 13–36. Amsterdam: Benjamins ­Publishing Company. Kiparsky, Paul. 1973. Abstractness, opacity and global rules. In Osamu Fujimura (ed.), P ­ honological Representations, ­57–86. Tokyo: Tokyo Institute for Advanced Studies of Language.

194 

 Mirella De Sisto

Ledgeway, Adam. 2009. Grammatica diacronica del Napoletano. Tübingen: Max Niemeyer Verlag. Loporcaro, Michele. 1997. L’origine del Raddoppiamento Fonosintattico. Saggio di fonologia diacronica romanza. Tübingen: Francke Verlag Basel. Lowenstamm, Jean. 1996. CV as the only syllable type. Current trends in phonology. In Jacques Durand and Bernard Laks (eds.), Models and methods, ­419–­441. Salford, Manchester: ESRI. Maiden, Martin. 1991. Interactive Morphonology. Metaphony in Italy. London/New York: ­Routledge. Maturi, Pietro. 2002. Dialetti e sub standardizzazioni nel Sannio Beneventano. Frankfurt: Peter Lang. Nespor, Marina and Vogel, Irene. 1986. Prosodic phonology. Berlin: Mouton de Gruyter. Nevins, Andrew. 2010. Locality in vowel harmony. MIT Press. Ouddeken, Nina. 2013. Metaphony in Italian. An analysis in Element Theory. Research Master Thesis. Leiden University. Passino, Diana. 2009. An ­element-­based analysis of Italian nominal inflection. In ­Montermini, Boyé and Tseng (eds.), Selected Proceedings of the 6th Décembrettes: ­Morphology in ­Bordeaux, ­63–­75. Somerville, Ma: Cascadilla Press. Passino, Diana. 2013. A unified account of consonant gemination in external sandhi in Italian: Raddoppiamento sintattico and related phenomena. The Linguistic review 30 (20): ­313–­346. Pöchtrager, Markus. 2006. The structure of length. Phd Dissertation. University of Vienna. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti. Fonetica. Torino: Einaudi. Scheer, Tobias. 2004. A lateral theory of phonology. Vol 1: What is CVCV, and why should it be? Berlin: Mouton de Gruyter. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language & Linguistic Theory 23: ­917–­989. Walker, Rachel. 2011. Vowel patterns in language. Cambridge: Cambridge University Press.

Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

On integrating different methodologies in phonological research: acoustic, articulatory, behavioral and neurophysiological evidence in the study of a metaphony system

University of Salento and University of Connecticut

Abstract: This work investigates metaphony in the Southern Salentino variety of Tricase. In this variety, the mid stressed vowel [ɛ] is raised to [e] when followed by high unstressed vowels -i and -u, while the mid stressed vowels [ɔ] is raised to [o] when followed by high unstressed -u. We integrate an acoustic-articulatory analysis in production with a behavioral-neurophysiological investigation in perception. Our findings show that the metaphonic process of the Tricase variety involves spreading of two distinct features from the unstressed high vowels to the stressed mid vowels: [+ATR] when the triggers is -i and [+High] when the trigger is -u. In parallel, at the perceptual level, we demonstrate that allophones and phonemes are equally computed in early speech processing and encoded in memory representations, and that therefore the allophonic variation generated by metaphony is part of the phonological grammar acquired by the Tricase speakers.

1 Introduction 1.1 Metaphony and Southern Salento The term metaphony traditionally refers to a vowel assimilation process affecting stressed mid-vowels before high vowels (Savoia & Maiden 1997; Calabrese 1998; Calabrese 2011). Southern Salento varieties (in the province of Lecce) were traditionally considered as being devoid of metaphony (cf. Parlangeli 1953; Rohlfs 1966; Stehl 1988; Mancarella 1998). However, Grimaldi (2003, 2009) demonstrated that a vowel assimilation process raising the stressed mid vowels [ɛ] and [ɔ] to the counterpart ­mid-high [e], [o] when followed by the unstressed high vowels -i or -u is indeed present in this area1. Grimaldi’s study was an acoustic analysis and statistic

1 See also Costagliola (2013) and Romano (2013) for two cases of metaphonic assimilation in Monteròni and Galàtone varieties (Central and Southern Salento, respectively).

196 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

t­ reatment of data collected through fieldwork in 36 localities of Southern Salento by means of a questionnaire. The phenomenon was found in 19 localities of the extreme tip of this area2. While all the Salento varieties spoken in this area share a five vowel phonemic system, i.e. /i, ɛ, a, ɔ, u/, a group of 19 localities modulate the stressed mid vowels [ɛ] and [ɔ], when followed by the unstressed high vowels -i and -u, producing the allophonic variants [e], [o]. In Fig. 1, we can observe the stressed vowels produced by the 19 male speakers characterized by metaphonic ­alternations3. A more detailed statistical analysis of the metaphonic data for each variety showed microvariation in the way in which -i and -u affect the mid vowels.

Fig. 1: F1–F2 scatterplot on logarithmic scale of the five vowels produced by the 19 Southern Salento male speakers (one speaker for each locality) displaying the metaphonic adjustments of mid vowels [ɛ] and [ɔ]. The main assimilation process involves e/_i. Assimilations of e/_u, o/_u and o/_i are also noticeable. In all cases different raising of mid vowels is showed. Ellipses on data, confidence level 68.8%.

2 For each variety, 1 male speaker aged between 50–80 was recorded. The data were elicited by a semi-spontaneous approach on the base of a questionnaire of about 600 stimuli, containing representative samples of the stressed vowels either in open or in closed syllable in all classes of words (nouns, verbs, adjectives, etc.). 3 Fig. 1 was realized using R software (cf. McCloy 2015).



On integrating different methodologies in phonological research 

 197

All 19 ­varieties share the assimilatory process for which the mid front vowel is raised by -i—i.e. [ɛ] → [e] before -i—while varying applications of the process are found in all other conditions, i.e. in the case of [ɛ] before -u, and [ɔ] before -u and -i (cf. Grimaldi 2003: 60–72 for a detailed description of the phenomenon).4 In this work we study one of the 19 localities interested by metaphonic processes, the Tricase variety, integrating two levels of investigation—production and perception—and two methodologies—acoustic-articulatory and behavioral-­ neurophysiological techniques—very rarely exploited together in disentangling the nature of the phonetic-phonological interface (cf. Grimaldi et al. 2010; Miglietta, Grimaldi & Calabrese 2013). The first level is explored by using acoustic analysis of data together with an ultrasound system (US) that reveals the fine-grained action of the a ­ natomical parts involved in the assimilatory process and compares the tongue contours obtained by US imaging. The second level is employed on the one hand by using perceptive tests and on the other hand by electroencephalographic recordings of the auditory evoked potentials (AEPs). We aim to clarify the acoustic-articulatory and perceptual nature of this assimilatory process interpreting it within a cognitive perspective. At the acoustic-articulatory level, our main concern is to investigate the articulatory dynamics of the assimilatory process affecting stressed vowels in the Tricase dialect, and specifically to understand what features are involved in it. At the same time, our main concern at the perceptual level is to investigate the cognitive status of this process. In particular, we test if the allophones generated by this process are encoded in memory representations producing categorical perception. We assume that if they are, the process under analysis cannot be a simple coarticulatory adjustment, but it must be under cognitive control, and, therefore, it has to be considered a component of the phonological grammar of this variety. Before discussing the data, some general explanations of the techniques used are needed.

1.2 The US technique An US machine (i.e. a classical echograph) emits ultra-high frequency sound through a transducer or “probe” containing piezoelectric crystals. When this probe is held against the skin of the neck, the US travels through the tongue and is reflected back to the transducer, resulting in echo patterns from which twodimensional images of the tongue surface are reproduced. These images can be

4 We will address this issue in a future work (see Calabrese & Grimaldi, in preparation).

198 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

viewed continuously on the machine itself for visual feedback, or recorded to video for later analysis. Because ultrasound is not able to image through bone or air, it can only allow visualization of the surface of the tongue and not, for example, the palate, jaw or rear pharyngeal wall. As US is both non-invasive and non-obtrusive, and, therefore, does not affect speech production, the sagittal placement of the US probe under the chin of a speaker provides images of the mid-line of the tongue (sagittally, or along any two-dimensional axis) from the tongue blade to the tongue root at high temporal resolution (typically ranging between 25 and 70 frames/sec) as a bright white line tracing the boundary between the tongue surface and the air above it (cf. Fig. 2). This technique allows exporting US pictures as a continuous video stream and synchronously recording the acoustic signal. In order to prevent uncontrolled head motion that will move the probe to different regions of the tongue during speech— thus impeding the comparison of tongue shapes across tokens—the head of the subject is generally immobilized (cf. Stone 2005; Gick, Wilson & Derrick 2013). The US pictures obtained are then processed with dedicated software in order to get a point-wise trace of the tongue contour (as in Fig. 2) for statistical analysis.

Tongue body

Tongue blade

(b)

Tongue root

(a)

(c)

(a) = Tongue Root (b) = Tongue Body (c) = Tongue Blade

  Fig. 2: On the left: Tongue surface contour (white dots) of [e] before -i in the word [meti] ‘you reaps’. On the right: average tongue contour of ten repetitions of [e] before -i. Tongue root, tongue body, and tongue blade are highlighted.

1.3 The AEP techniques and the mismatch negativity component The AEP technique provides not only a millisecond-precise measurement of information processing in the auditory cortex but also, depending upon the task, can allow to disentangle automatic detection from attentional processes. AEP studies have generally used the so-called ‘oddball paradigm’. It consists of alternating



On integrating different methodologies in phonological research 

 199

repetitive (standard) and infrequent (deviant) sounds (80%–20% of occurrence, respectively) while subjects are distracted from listening by a primary task (e.g. watching a silent movie), to measure the so-called mismatch negativity (MMN) response to different classes of sounds. The MMN is an AEP component, elicited by stimulus change at ≈100–250 ms, mainly generated in the auditory cortex, reflecting the neural detection of a change in a constant property of the auditory environment (Picton 2000; Näätänen et al. 2007; Näätänen, Kujala & Winkler 2011; Winkler & Czigler 2012). In other words, the MMN ­represents the neural detection of a ‘mismatch’ between the deviant and the memory trace formed by the standard in an oddball paradigm. Furthermore, the amplitude and latency peaks of the MMN are directly correlated with the magnitude of the perceived change and, hence, it indicates, at a pre-attentive level, whether the auditory system has distinguished between two stimuli. For example, native speech contrasts elicit larger MMNs than non-native sounds (see Amenedo & Escera 2000; Näätänen 2001). The idea is that elicitation of MMN by infrequently presented deviant sounds indicates that auditory features differentiating the deviant sounds from the standard sound are being detected.

Fig. 3: Mismatch negativity (MMN) in healthy awake adults. One can note that MMN amplitude for standard and deviant sounds is significant with respect to the baseline. Adapted from Sams et al. (1985).

Thus, the basis for MMN elicitation is the contrast between the cortical representation extracted from the auditory regularities occurring in the standard stimulus and the cortical representation extracted from the auditory properties of the deviant stimulus (cf. Fig. 3). The key factors influencing deviance detection in

200 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

the auditory scene are: (1) cortical extraction of the standard regularities from the ongoing acoustic-phonetic input, and (2) cortical representation of these regularities in memory. According to Lahiri & Reetz (2001, 2010) and Eulitz & Lahiri (2004), it is assumed that MMN elicitation consists of the following steps. First, the standard stimulus, a vowel in our study, creates a central sound representation, roughly corresponding to the vowel neural trace stored in the auditory cortex and conveying information about the vowel’s phonological representation. Second, the deviant vowel creates a percept corresponding to the vowel’s phonological representation. Third, MMN is automatically elicited when the phonological representation of the deviant vowel is compared to the phonological representation of the standard vowel and different specifications in their phonological representations are observed at the cortical level. Thus, MMN can be used to assess the extraction of auditory regularities (e.g. the spectral and temporal features characterizing both the deviant and the standard) from the acousticphonetic input sequence and considered a measure of the successful extraction and representation of the auditory regularities. In synthesis, an MMN response— i.e. when MMN amplitude is larger and latency shorter—signals the successful extraction and representation of the phonological properties of the standard and deviant sounds at the cortical level (cf. Sussmann et al. 2013 for discussion).

2 Methods 2.1 Acoustic-articulatory experiment In a sound-proof room, a 54-year-old male speaker of the Tricase variety produced 90 word stimuli containing the stressed target vowel embedded in the frame sentence Ieu ticu moi (I say now), randomly presented on a PC monitor. 10 stimuli for each of the five target vowels were produced: They were designed to elicit the full vowel inventory in both open and closed syllables. For each vowel, we selected surrounding consonantal contexts that did not affect the vowel production: i.e. where possible the stressed vowel was surrounded by labial and/or coronal obstruents. The stressed [ɛ] and [ɔ] vowels were differentiated according to the unstressed vowel context. Then, the speaker produced 10 words containing [ɛ] before by -i, 10 with [ɛ] before by -u, and 10 with [ɛ] before by -e or -a, and so on for [ɔ]. An Aplio XV machine, by Toshiba Medical System Corp. was used to acquire images of tongue contours at 25 Hz. The video stream was synchronously acquired together with the audio signal, by means of an external a/v



On integrating different methodologies in phonological research 

 201

­analog-to-digital acquisition card, and then recorded in real-time on a dedicated PC (cf. Grimaldi et al. 2008; Sigona et al. 2015). The probe was rigidly locked into a fixed position on a microphone stand appropriately adapted. To avoid the transducer pressure on the soft tissue of the jaw, an acoustic standoff eliminating upward pressure of the tissue was used. The two sides of the head of the subject were immobilized with an adjustable wood system built at CRIL that allows the speaker to rest his/her forehead steadily so that the whole tongue contour (i.e. from the tip to the shadow of the hyoid bone) could be captured and analyzed. The elicited speech was recorded by using CSL 4500 (at a sampling rate of 22.05 kHZ), and a Shure SM58-LCE microphone placed at a distance of 20 cm from the speaker. The entire speaker productions were segmented and normalized in peak amplitude by means of Praat 5.2 (Boersma & Weenink 2011). For each vowel, total duration as well as F0, F1, F2 and F3 have been measured. The formant values have been measured in the vowel steady tract (0.025 s) centered at the midpoint (here we focused on the F1 and F2 values only). The US a/v stream has been segmented offline by an automatic software procedure developed at CRIL (cf. Grimaldi et al. 2008), on the basis of audio pulses superimposed to the speech signal before and after each sentence during recording. For each segmented sentence, looking at the acoustic waveform the operator manually placed labels around the time intervals where the relevant vowels occurred, so that the corresponding US pictures could be identified. These pictures were processed with EdgeTrak in order to get a point-wise trace of the tongue contour for the statistical analysis (Li, Kambhamettu & Stone 2005): see Fig. 2.

2.1.1 Statistical analysis For acoustic analysis, an independent t-test was carried out to examine the assimilatory effect of the unstressed vowels -i, -u, and -e,a5 on the mid vowels [ɛ] and [ɔ] (alpha level p < 0.05). Due to the protocol design, samples of the first category are considered independent of any other sample of the second category. So, we compared: − [ɛ] before by -i with [ɛ] before by -e/a; − [ɛ] before by -u with [ɛ] before by -e/a

5 Unstressed -e and -a have been considered together as they do not trigger assimilation.

202  − − − −

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

[ɛ] before by -u with [ɛ] before by -i. [ɔ] before by -i with [ɔ] before by -e/a; [ɔ] before by -u with [ɔ] before by -e/a [ɔ] before by -u with [ɔ] before by -i.

To address the issues of the articulatory features involved in the Tricase metaphony process, we used the Smoothing Spline ANOVA (SS ANOVA) to compare tongue curves (Chong 2002; Davidson 2006; Lilienthal 2009). The SS ANOVA is based on an inferential statistic method used to study similarities and differen­ ces in the shape of two groups of curves. It has already been applied to studies in medical science, environmental science, and epidemiology, and may be applied to compare two groups of tongue curves produced by the same subject: e.g. 10 tongue curves of [ɛ] before by -e/a vs. 10 tongue curves of [ɛ] before by -i, etc. This statistical method has the advantage of determining whether or not there are significant differences between the tongue curves belonging to two groups as well as which sections, i.e. if root, body, or blade of the tongue curves are different (Davidson 2006). Given two groups of tongue curves, the corresponding Smoothing Splines (SS) are computed: they are particular curves that provide the best fit to all the data points belonging to each group. The SS for each data set is termed main group effect, and around each SS the 95% Bayesian confidence interval is constructed. The comparison of the two groups of tongue curves is performed with the interaction diagram, which represents a plot of the difference of the SS for each data set from the SS that is the best fit to all of the data. So, the 95% Bayesian confidence interval is constructed once more, and the difference between the splines is significant when the confidence interval does not encompass the zero on the y axis (cf. Fig. 6(a–d)) where the SS are on the left and the tongue curve (TC) on the right). SS ANOVA analysis has been done by means of a Matlab tool developed at CRIL and the R package “gss” (Chong 2009).

2.2 Behavioral and AEP experiments Twelve right-handed students from the University of Salento (7 females; mean age 21.2, range: 19.1–26.0, s.d. ±2) participated in both behavioral and AEP sessions. No participants reported a history of neurological illness. All of the participants were native speakers of the Tricase dialect, provided written informed consent, and participated in both experiments. The discrimination task and AEP recordings were run within one session with the discrimination task preceding the AEP recordings to prevent the stimuli from being attentively processed before



On integrating different methodologies in phonological research 

 203

the AEP measurements. The experimental procedure received the approval of the local ethics committee. We concentrated our attention only on one assimilation process: [ɛ] → [e] before -i (e/_i in the captions for Figs. 1 and 4): i.e. the process shared by all the speakers of all the 19 localities where metaphonic adjustments were found (Grimaldi 2003, 2009). We used the vowels [ε, e, i] present in the stressed vowel system of the Tricase dialect as experimental stimuli: [ε] and [i] have phonemic status, whereas [e] has an allophonic status. We used three natural speech tokens for each stimulus type to introduce acoustic variability and ensure that the acoustically different tokens were grouped together in a more abstract representation of the speech sound category. A male speaker of Tricase produced a total of 30 pseudowords (10 for each vowel type). The vowels were inserted in the context b[V]b[V] and embedded in the carrier sentence Ieu ticu moi (I say now). The speech signal was recorded in a soundproof room with CSL 4500 and a Shure SM58-LCE microphone with a sampling rate of 44.1 kHz and an amplitude resolution of 16 bits. The acoustic analysis was performed using Praat 5.2 (Boersma & Weenink 2011). The fundamental frequency (F0), first formant (F1) and second formant (F2) values were measured in the vowel steady tract (0.25 s) centered at the vowel midpoint. For every vowel category, we selected three acoustic varying tokens with comparable pitch ([ε] = 174 (±3); [e] = 174 (±7); [i] = 182 (±7)). The F1/F2 average formant values in Hz of the three exemplars were the following: [ε]: F1= 519 (±11), F2 = 1906 (±23); [e]: F1 = 389 (±7), F2 = 1967 (±20); [i]: F1 = 327 (±6); F2 = 2108 (±51). The mean Euclidean distances in mel in the F1-F2 plane (at the vowel mid-point) between all combinations of the three vowel types were [ε−e] 130 mel, [e−i] 88 mel and [ε−i] 212 mel. Lastly, portions containing only the steady-state vowel signal were eliminated from the selected words. All nine stimulus audio files were ramped with 10 ms Gaussian on- and offsets and normalized for duration (200 ms) and peak amplitude (70 dB/SPL).

2.2.1 Behavioral test In an AX (same-different) discrimination task, we assessed the attentive discrimination of the allophonic variation [ε−e]. Each of the three variants of the two vowel categories composing the allophonic variation were combined with one another and the three tokens of the other vowel category that composed the pair. Thus, three pair types were tested in all: [ε−ε], [ε−e], and [e−e]. The inter-stimulus interval was 800 ms, and the trial’s initial silence was 500 ms. Each of the 54 stimulus pairs occurred

204 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

twice. The complete set of 108 stimuli pairs was presented in random order. The ­listeners indicated whether the sounds of a pair were identical or different. The experimental arrangement provided by Praat 5.2 was applied in these tests. The subjects were tested in a sound proof room with a laptop and headphones.

2.2.2 AEP recordings In an oddball paradigm, the MMN responses to the phonemic contrast [e-i] and allophonic pair [ε−e] were recorded. We compared responses to acoustic distinctions generated by the assimilation process with responses to acoustic distinctions associated with phonemic contrasts. By selecting the relevant distinctions, we carefully considered the acoustic distances between the possible stimuli pairs. We used the variant [e] of the mid vowel for the allophonic and phonemic conditions to reduce the difference in acoustic distance between the two conditions. More specifically, we decided to pair the phoneme [i] with the variant [e] to generate a mean acoustic distance (88 mel) more similar to the allophonic [ε−e] (130 mel) than alternative pair [i−ε] (212 mel). This set-up permitted us to significantly match the acoustical deviance between stimuli because this parameter is well known to influence MMN amplitude and latency (Näätänen et al. 1997; Näätänen et al. 2007). In other word, maintaining constant as far as possible the acoustic distance between the contrasts tested, we increase the chances to measure abstract process and not purely physical differences related to stimuli. Together, stimulus sequences of 1000 trials were randomly presented in each block. In each sequence, one vowel type served as the standard (85% of the trials) and the remaining vowel of the sound pair was the deviant. For both vowel pairs, the roles of standard and deviant were reversed in separate blocks. Therefore, a total of four blocks were recorded. Stimulus sequences were presented with a variable inter-stimulus interval of 500 ms to 550 ms. During the recording, the subject sat in an acoustically shielded room watching a silent movie. The subject was instructed to disregard the sounds presented via loudspeakers. The stimuli were presented using ePrime 2.0, and the order of the blocks was counterbalanced between participants. AEPs were recorded (0.1–100 Hz, −2 dB points, sampling rate 250 Hz) with a 64-channel ActiCap system (Brain Products). Vertical eye movements were moni­ tored using electrodes attached above and below the right eye and horizontal movements with electrodes attached to the outer canthi of each eye. The online reference electrode was FCz. Impedance was maintained under 5 Ω. Off-line signal processing was performed with the Brain Vision Analyzer (Brain Products) software package. The EEG was filtered with a bandpass of 1–25 Hz (12 dB/oct), and the raw data were



On integrating different methodologies in phonological research 

 205

re-referenced against the average of the left and right mastoids. Epochs began 100 ms before until 600 ms after stimulus onset. Standard and deviant epochs were averaged and included a pre-stimulus baseline of 100 ms. The ERP responses to the initial three standard stimuli of each block and standard stimuli that immediately followed the deviants were not included in the analyses. The averaged data were baseline corrected over a pre-stimulus interval of 100 ms. Epochs with an amplitude change exceeding 75 μV at any of the electrodes were rejected. All remaining standard and deviant epochs were included in the identity MMN analysis. For the amplitude and peak latencies analysis of the MMN component, we selected a time window based on a visual inspection of the grand average data across all of the subjects. For assessing the MMN component, we adopted the identity MMN (iMMN) approach. The MMN is reflected in a difference waveform calculated by subtracting the AEP response to standard stimuli from the deviant present in the identical block. In contrast, the iMMN is calculated using the recordings of two corresponding blocks. For instance, the standard [e] (of the block [e] standard and [i] deviant) was subtracted from the [e] deviant of the reverse block ([i] standard and [e] deviant). The iMMN approach eliminates variation in ERP morphology that may result from purely acoustic differences and therefore permits the observation of memory representation contributions (e.g. Pulvermüller & Shtyrov 2006). The MMN latency corresponded to the time at which the highest negative amplitude peak in the MMN time window occurred (120–200 ms). This time window was selected based on the grand average across all subjects and was motivated by the expectation to observe the MMN response 100–200 ms after the onset of the deviant sound. The MMN amplitude was obtained by measuring the mean amplitude (μV) contained within a 50-ms time window centered at the MMN latency peak. The analysis was based on the electrodes Fz, Cz, and FCz.

2.2.3 Statistical analysis Discrimination performance was measured in percentage of same and different responses and in terms of d’ (d-prime). This type of analysis was chosen because a percentage analysis of the correct responses on the different pairs alone is not a very meaningful measure of discrimination. The d-prime analysis is based on the Detection Theory proposed by Macmillan and Creelman (2005) which interprets the listeners responses in terms of the listener’s tendency to respond ‘same’ or ‘different’. According to this analysis, four types of response rates are calculated: (1) hit rate; (2) miss rate; (3) false alarm rate; and (4) correct rejection rate. Finally, based on the hit and false-alarm probabilities a d’ value is calculated. Briefly, the best subject maximizes hits and minimizes false alarms, and thus the larger the

206 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

difference between hits and false alarms, the better the subject’s performance. The statistic d’ is a measure of this difference. To analyze the AEPs data, separate one-way ANOVAs were performed with mean amplitude (μV) as the dependent measure and probability (standard vs. deviant) as the independent measure to assess that reliable MMN components were elicited. Separate two-way ANOVAs were performed with latency (ms) and mean amplitude (μV) as the dependent measures to analyze the MMN component. The independent variables were vowel pair status (allophonic vs. phonemic) and electrode (Fz, Cz and FCz). Additionally, the interaction vowel pair status x ­electrode was included in the analyses. Lastly, two separate one-way ANOVAs with an electrode grid of six electrodes (C3, Cz, C4, F3, Fz and F4) were performed to analyze whether hemispheric asymmetries could be observed. The dependent measure was mean amplitude and independent variable laterality (3-line (C3, F3), z-line (Fz, Cz), and 4-line (C4, F4)).

3 Results 3.1 Acoustic-articulatory results The mean formant values in Hz of the Tricase stressed vowels and the allophonic variants are given in Tab. 1. In Fig. 4 all the vowels tokens are plotted in a twodimensional F1-F2 space by recurring to logarithmic scale and ellipses on data (confidence level 68,8%). Fig. 4 was realized using R software (McCloy 2015). In both cases, data concerning the unstressed final high vowels -i and -u triggering metaphonic adjustments are described. Tab. 1: Mean formant values F1-F2 in Hz of the Tricase vowels. DS = Standard Deviation; Min = Minimum; Max = Maximum. F1-F2 Mean DS Min Max

-i

i

e/_i

e/_u

e/_e,a

a

o/_e,a

o/_u

o/_i

u

-u

327 2030 18 57 273 1926

255 2151 18 57 232 2065

406 1930 21 76 393 1821

472 1717 22 42 440 1640

506 1708 18 51 478 1632

665 1349 37 73 616 1266

513 890 29 53 479 790

465 862 25 27 433 8240

500 907 33 62 454 833

322 753 40 79 249 669

347 705 34 96 314 600

370 2147

286 2253

470 2052

515 1774

538 1786

713 1508

577 964

491 918

569 1021

378 960

436 866



On integrating different methodologies in phonological research 

 207

Fig. 4: F1-F2 scatterplot on logarithmic scale of the Tricase stressed vowels. Significant metaphonic adjustments of mid vowels [ε], [ɔ] are shown: i.e. e/_i, e/_u and o/_u. Ellipses on data, confidence level 68.8%.

The results of the independent t-test are given in Tab. 2, where the significant differences of the action of the unstressed high vowels on the mid-stressed vowels are indicated by asterisks. For what concerns F1 values, e/_i and e/_u are different from each other and from e/_e,a. In contrast, while o/_i and o/_u are also different from each other, only o/_u is different from o/_e,a. This reveals the presence of metaphonic adjustments in which [ɛ] is raised before -i and -u but [ɔ] is raised only before -u, confirming previous data obtained (Grimaldi 2003, 2009). Also F2 values showed a significant effect for [ɛ] only: i.e. e/_i is different both from e/_e,a and e/_u. These significant effects are better caught in the graphical representation of data (Fig. 4) where we can observe that [ɛ] is shifted upwards in frequency producing two allophonic variants depending on the unstressed high vowel triggering metaphony

208 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

(i.e. -i or -u). Conversely, [ɔ] is shifted upwards in frequency producing only one allophonic variants due to the action of -u. Finally, the significant effect for F2 is observable in the advancement of e/_i with respect to e/_e,a and e/_u. Tab. 2: Independent t-test (α = 0.05) for F1 and F2 values and for each vowel category considered. Significant effects are highlighted by asterisks. Vowel category e/_i ~ e/_e,a e/_u ~ e/_e,a e/_u ~ e/_i o/_i ~ o/_e,a o/_u ~ o/_e,a o/_u ~ o/_i

F1 t[18] = −10,825 t[17] = −3,563 t[19] = 6,671 t[18] = −0,948 t[18] = −3,877 t[18] = −2,598

F2 p < 0,05* p < 0,05* p < 0,05* p = 0,356 p < 0,05* p < 0,05*

t[18] = 7,184 t[17] = 0,401 t[19] = −7,405 t[18] = 0,669 t[18] = −1,446 t[18] = −2,093

p < 0,05* p = 0,694 p < 0,05* p = 0,512 p = 0,165 p = 0,058

  Fig. 5: F1 Hz statistical distribution of e/_i, e/_u, e/_e,a in the acoustic space with respect to unstressed -i, -u on the left and of o/_i, o/_u, o/_e,a with respect to unstressed -u on the right. On the left it is shown that e/_i and e/_u do not overlap with -i, -u, while on the right that o/_u does not overlap with -u.

Note that the metaphonic raising of [e] before -i, -u and [o] before -u does not generate complete assimilation of mid vowels into the unstressed high vowels (cf. Fig. 4). This process is demonstrated in Fig. 5, where statistical distribution of raised and not raised mid vowels in the acoustic space is shown compared with unstressed vowels triggering metaphonic raising. The F1 parameters distinguishing [e] before -i, -u and [o] before -u do not overlap with the F1 parameters of the unstressed vowel -i and -u that trigger the



On integrating different methodologies in phonological research 

 209

metaphonic adjustments. That is, both for [e] and [o] there is not complete ­assimilation with the unstressed vowels -i, -u. So, the mid vowels are raised but do not become high vowels. Also, Fig. 5 shows clearly that when the unstressed vowels do not trigger metaphonic adjustments, as in the case of [ɔ] before -i and [ɔ] before -e,a, there is an overlapping of F1 parameters. Interestingly, statistical distribution of F1 parameters in Fig. 5 indicates that the distribution of e/_i and e/_u F1 values tends to mirror the distribution of unstressed -i and -u F1 values, respectively. The same happens for o/_u that tends to mirror the distribution of unstressed -u F1 values. This pattern is crucial for the interpretation of the ­assimilatory process and may be clarified by our articulatory data. Regarding the articulatory data, the SS analysis and TCs of the stressed mid vowels together with the unstressed high vowels -i, -u are shown in Fig. 6(a–d): Tongue blade is up on the right, tongue body is up at the center, and tongue root is below on the left of TCs. Observe that the different curves in these figures involve statistical tendencies extracted from actual tongue contours. Thus, what may appear to be millimetric or even submillimetric changes in curve shapes are in fact the result of significantly different anatomical tongue articulations that generate clearly different acoustic effects. In particular, Fig. 6(a) shows that the tongue root and the tongue body of [e] before -i are advanced and raised, respectively, with respect to those of [ɛ] before -e,a. Fig. 6(b) indicates that the tongue root and tongue body of [e] before -u are only slightly advanced and raised, respectively, with respect to those of [ɛ] before -e,a. However, this slight difference generated s­ ignificant variation, as shown on the one hand by SS analysis on the left, and on the other hand by the significant effect noted at the acoustic level for e/_u (cf. Fig. 4 and Tab. 2). The difference in tongue root involvement in the case of [e] before -i with respect to that of [e] before -u is more noticeable in Fig. 6(c). Here one can observe that in the case of [e] before -i (on the left), there is both tongue root advancement and tongue body raising, whereas in the case of [e] before -u (on the right), there is mainly tongue body raising, but little tongue root action. Noticeably, the tongue body of [e] before -i is not as raised as that of unstressed -i: This suggests that tongue body raising results, in this case, as an inertial consequence of tongue root advancement rather than being the critical gesture involved in the metaphonic assimilation. The involvement of tongue root for [e] before -i is better discernible if we look at mid back vowel in Fig. 6(d) where on the right it is shown that the metaphonic adjustment of [o] before -u involves only tongue body raising with no tongue root advancement, when compared with what happens in the case of [o] before -e,a.

210 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

 

(a)

 

(b)

 

(c)

On integrating different methodologies in phonological research 



 211

 

(d) Fig. 6(a–d): (a) SS on the left and TC on the right of [ɛ] before -e,a vs. [e] before -i; (b) SS on the left and TC on the right of [ɛ] before -e,a vs. [e] before -u; (c) TC of [ɛ] before -e/a and [e] before -i compared with unstressed -i (i/e in the caption) on the left; TC of [e] before -i and [e] before -u compared with unstressed -u (u/e in the caption) on the right; (d) SS of [ɔ] before -e/a and [o] before -u on the left and TC of [ɔ] before -e/a and [o] before -u compared with unstressed -u on the right.

3.2 Behavioral and AEP results The percentage analysis of the participants’ ‘same’ and ‘different’ responses indicated that the vowels that composed the allophonic pair [ɛ–e] were judged as ‘‘different’’ at a high rate (94%), whereas both the vowel pairs composed of the identical vowel type showed a high percentage of ‘same’ responses: i.e. [ɛ–ɛ] = 90%; [e–e] = 80% (see Fig. 7). Due to the experimental design, especially given the high number of trials (n = 180) for each subject, a set of z-tests has been used to assess the statistical significance of the ‘same’ and ‘different’ responses. These tests confirmed that the percentages of the ‘same-different’ responses were

[e-e]

Different

[ɛ-ɛ]

Same

[ɛ-e] 0%

20%

40%

60%

80%

100%

Fig. 7: Percentages of same/different responses of the AX discrimination task.

212 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

significantly different within each vowel pair, as well as between the three tested vowel pairs ( p = 0.000). For what concerns the d-prime analysis, used to investigate the listener’s tendency to respond ‘same’ or ‘different’ (cf. Macmillan & Creelman 2005), the mean d’ score was 2.55, which indicated an accurate discrimination between the allophones. Concerning AEP data, significant MMNs were elicited in the allophonic (F (1,66) = 14.592, p < 0.001) and phonemic conditions (F (1,66) = 6.047, p < 0.05). The amplitude ANOVA revealed no significant main effect (vowel pair status: F (1,66) = 1.052, p = 0.31; electrode: F (2,66) = 0.191, p = 0.83) or interaction (vowel pair status x electrode: F (2,66) = 0.204, p = 0.82). In contrast, the ANOVA on the latency of the MMN revealed a significant main effect for vowel pair status (F (1,66) = 6.017, p < 0.05), which indicated that the latency of the phonemic condition was significantly earlier. No main effect was observed for the factor electrode (F (2,66) = 0.283, p = 0.76) or interaction vowel pair status x electrode (F (2,66) = 0.193, p = 0.83). The laterality ANOVAs revealed no significant amplitude difference between the electrodes positioned at the right and left hemisphere for either the allophonic (F (2,69) = 0.236, p = 0.79) or phonemic condition (F (2,69) = 0.518, p = 0.60). In Tab. 3 the grand-average MMN mean peak amplitudes and latencies at Fz, Cz, FCz electrodes are shown (values are averaged across the midline electrodes), whereas in Fig. 8 the averaged ERP responses are represented at the Fz electrode. Tab. 3: The grand-average MMN mean (Fz, Cz, FCz) peak amplitudes and latencies.

Allophonic condition Phonemic condition

Amplitude

Latency

−1.898 (±0.679) −2.004 (±0.895)

182 (±37) 154 (±41)

4 Discussion The aim of this paper was twofold. First, to investigate acoustic-articulatory dynamics of metaphonic process of the Tricase variety, generating allophonic variants. Second, to ascertain how allophonic variants are perceptually computed as regard phonemic contrasts. The acoustic findings confirmed the presence of metaphonic adjustments in the Tricase variety showing that: (i) F1 values of [e] before -i and [e] before -u are lowered and significantly different from those of [ɛ] before -e,a; (ii) F1 values of [o] before -u are lowered and significantly different from those of [ɔ] before -i and [ɔ] before -e,a; (iii) F2 values of [e] before -i are enhanced and significantly

On integrating different methodologies in phonological research 

 μV

μV

–3

–3

–2

–2

–1

–1

–100

0

100

200

300

400

500

ms

–100

1 2

 213

/e/ as deviant /e/ as standard

0

100

200

300

400

500

ms

1

Allophonic condition [ɛ-e]

2

Phonemic condition [e–i]

(a)

(b)

μV

Allophonic condition Phonemic condition

–3 –2 –1

–100

0

100

200

300

400

500

ms

1 2

Fig. 8: Top panel: the group-averaged ERP responses to standards (grey lines) and deviants (black lines) at Fz for the allophonic (left) and phonemic conditions (right) in the 120–200 ms window. Middle panel: Topographic maps showing the iMMN peak latency activation viewed from above: (a) allophonic conditions—iMMN [ɛ] (left) and iMMN [e] (right); (b) phonemic conditions—iMMN [i] (left) and iMMN [e] (right). Bottom panel: the deviant-minus-standard difference wave for the allophonic (grey line) and phonemic conditions (black line) at Fz in the 120–200 ms window.

different from those of [ɛ] before -e,a and [e] before -u (cf. Fig. 4; Tabs. 1 and 2). These facts are better clarified by articulatory data. Comparing SS and TC in Fig. 6(a–d) we observed that in the metaphonic adjustments of [e] before -i, [e] before -u and [o] before -u, there is always tongue body raising, but that only in

214 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

the assimilation process of [e] before -i, advancement of tongue root appears to be a distinct tongue gesture. Hence, as expected, F1 values are always lowered by tongue body raising. When tongue root advancement also occurs, as in the case of [e] before -i, F2 values are also increased generating an advancement in the acoustic space of the vowel (cf. Lindau 1978; Tiede 1996; Stevens 1998; Archangeli & Pulleyblanck 1994; Gick et al. 2006). Overall, our findings suggest that in the Tricase variety stressed vowels F1/F2 cluster into three allophonic variants: i. [e] before -i; ii. [e] before -u iii. [o] before -u Articulatory data lead us to assume that these allophonic variants are differentiated by the distinct feature specifications that are spread by the unstressed high vowels: [+ATR] for [e] before -i and [+high] for [e] before -u and [o] before -u (see Grimaldi et al. 2010 and especially Calabrese and Grimaldi 2013 for a phonological interpretation of these facts). This generates [−high, +ATR] [e̘] when the unstressed high vowel is -i, and [+high, −ATR] [i̙] and [+high, −ATR] [u̘] when the unstressed high vowel is -u, where the [+high, −ATR] vowels are acoustically lower than their [+ATR, −high] counterparts, as in some African varieties (cf. Ladefoged & Maddieson 1996: 305). The resulting superficial stressed vowel system of the Tricase variety is represented in Fig. 9:

i̘ u̘ [+high, +ATR] e̘ [−high, +ATR] i̙ u̙ [+high, −ATR] e o [−high, −ATR] a

Fig. 9: The superficial stressed vowel system of the Tricase variety on the base of metaphonic adjustments active. According with current IPA usage, the diacritics ̘and a̙ re used to indicate Advanced and Non Advanced Tongue Root, respectively.

The AX discrimination test showed that at the attentive level the allophonic pair [ɛ−e] (e and e̘ in Fig. 9) were judged as ‘different’ at a high rate (94%) when compared with the vowel pairs [e−e] and [ɛ−ɛ] composed of identical vowel type that were judged as ‘same’. The electrophysiological study, involving the pre-attentive level and a millisecond-precise measurement of information processing in the brain, provided a more detailed picture of the computations occurring in the perception of allophonic and phonemic pairs. The MMN amplitude analysis revealed no significant difference between the allophonic (i.e. [ɛ−e]) and phonemic (i.e. [e−i]) ­conditions, suggesting



On integrating different methodologies in phonological research 

 215

that the contrastive and non-contrastive vowel pairs are equally computed in early speech processing and encoded in memory representations. If the contrary was true, the amplitude of the allophonic contrast should have been significantly reduced with respect to the phonemic contrast. At the same time, we observed a shorter latency for the phonemic contrast. Generally, the MMN peak latencies are attributed to the acoustic distances between the stimuli: i.e. the MMN latency steadily decreased with increasing acoustic deviation (Näätänen et al. 1997). However, as above noted (in 2.2 and 2.2.2), the Euclidean distance of our phonemic contrast [i–e] was 88 mel, whereas the Euclidean distance of the allophonic contrast [ɛ−e] was 130 mel. Thus, if the difference in the peak latencies elicited by our stimuli was purely due to acoustic reasons, we would have expected that the allophonic contrast should have elicited a shorter latency than the phonemic one. This fact indicates that phonological knowledge was actually accessed and listeners’ MMN responses to the phonemic/allophonic distinctions were not simply an effect of the acoustic distance between the vowel pairs. We suggest that the difference in latencies of the phonemic and allophonic distinctions is due to a difference in perceptual parsing: whereas only contrastive features need to be identified in the phonemic distinctions, both contrastive and non-contrastive features need to be identified in allophonic distinctions. Our hypothesis is that the restriction of the search only to contrastive sound properties may result in faster, less effortful cognitive processing for the phonemic distinction. However, when both contrastive and non-contrastive sound properties must be accessed and parsed, the processing requires additional computational operations and related supplementary neural activations, and therefore is slower. This analysis can be maintained only if mental representations contain not only the contrastive features of vocalic phonemes but also the non-contrastive features of the allophones that are the output of the metaphonic process (cf. Miglietta, Grimaldi & Calabrese 2013). In synthesis, our MMN results demonstrated that the allophonic variants resulting from the assimilatory process—which predictably shifts the target realization of the phoneme /ɛ/ when followed by a high front vowel—are encoded in mental phonological representation. It follows that this process must have a cognitive reality. One can plausibly assume that the same occurs with the other allophonic variants of metaphony discussed above (cf. 1.1). We may conclude that metaphony in the Tricase variety is under cognitive control, and therefore it is part of the phonological grammar of the Tricase variety6. From a general perspective, our results suggest that learners recognize and detect sound alternations conditioned by the linguistic system environment. In

6 See Calabrese (2012) and Miglietta, Grimaldi and Calabrese (2013) for discussion of further theoretical consequences of our perceptual study of Tricase metaphony.

216 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

the case of the Tricase variety, learners converge on five phoneme categories and on rules that predictably shift the target realization of the mid vowels when followed by high vowels. This phonological knowledge provides the language users with the necessary information to produce the appropriate vowel tokens in the assimilatory context.

5 Conclusion In this work we tried to integrate different levels of analysis and different methods in order to provide advancements in the comprehension of the phonetic/phonology interface within a cognitive perspective. Thanks to acoustic-articulatory analysis, using US images of the tongue, we showed that the metaphonic process of the Tricase variety is driven by the spreading of two distinct features from the unstressed high vowels to the stressed mid vowels: (i) [+ATR] when the triggers is -i; (ii) [+High] when the trigger is -u. So, further evidence is given in favor of the use of US in analyzing the articulatory grounding of phonological phenomena. The perceptual level of analysis has been explored using behavioral and electrophysiological methods. Focusing on the allophonic pair [ɛ−e] produced by the metaphonic process and the phonemic contrast [e−i], we showed that MMN amplitudes and latencies are compatible with the idea that allophones and phonemes are equally computed in early speech processing and encoded in memory representations. This finding suggests that predictable vowel allophonic alternations pattern with phonemic contrasts for auditory perception. Overall, our data support a model of phonology in which the acquisition of phonemic categories occurs with the learning of phonetic distribution patterns and their relationships within a grammar.

Acknowledgments The authors would like to thank two anonymous reviewers for their useful suggestions.

References Amenedo Elena & Escera Carles. 2000. The accuracy of sound duration representation in the human brain determines the accuracy of behavioral perception. European Journal of Neuroscience 12: 2570–2574. Archangeli Diana & Pulleybalnck Douglas. 1994. Grounded phonology. Cambridge MA: MIT Press.



On integrating different methodologies in phonological research 

 217

Boersma Paul & Weenink Danile. 2011. Praat: doing phonetics by computer (Computer program), Version 5.2. http://www.praat.org/. Calabrese, Andrea. 1998. Metaphony revisited. Rivista di Linguistica 10: 7–68. Calabrese, Andrea. 2011. Metaphony in Romance. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume & Karen Rice (eds.), The Blackwell companion to phonology, 2631–2661. Wiley-Blackwell. Calabrese, Andrea. 2012. Auditory representations and phonological illusions: a linguist’s perspective on the neuropsychological bases of speech perception. Journal of Neurolinguistics 25: 355–381. Calabrese Andrea & Grimaldi Mirko. 2013. L’interfaccia fonetica-fonologia nella metafonia del Salento meridionale. In Antonio Romano & Mario Spedicato (eds.), Sub voce Sallentinitatis. Studi in onore di p. Giovan Battista Mancarella, 277–288. Lecce: Del Grifo Editore. Calabrese & Grimaldi. In preparation. Microvariation in a metaphonic system: the case of Tricase dialect. Acoustic and articulatory evidence. Chong, Gu. 2002. Smoothing Spline ANOVA Models. Springer: New York. Costagliola, Angelica. 2013. Dialectologie et phonétique expérimentale: Analyse acoustique et articulatoire de certaines variétés du Salentine centrales (Pouilles, Italie du Sud). Unpublished PhD Thesis. Laboratoire de Phonétique et Phonologie, CNRS/Sorbonne Nouvelle – Paris III de Paris. Davidson Lisa. 2006. Comparing tongue shapes from ultrasound imaging using smoothing spline analysis of variance. Journal of Acoustical Society of America 120 (1): 407–415. Eulitz, Carsten & Lahiri Aditi. 2004. Neurobiological evidence for abstract phonological representations in the mental lexicon during speech recognition, Journal of Cognitive NeuroScience 16 (4): 577–583. Gick, Bryan, Douglas Pulleyblank, Fiona Campbell & Ngessimo Mutaka. 2006. Low vowels and transparency in Kinande vowel harmony. Phonology 23: 1–20. Gick, Bryan, Donald Derrick and Ian Wilson. 2013. Articulatory phonetics. Hoboken, NJ: Wiley-Blackwell. Grimaldi, Mirko, Calabrese Andrea, Sigona Francesco, Garrapa Luigina & Sisinni Bianca. 2010. Articulatory Grounding of Southern Salentino Harmony Processes. Paper presented at the 11th Annual Conference of the International Speech Communication Association (ISCA), Interspeech, Spoken Language Processing for All, Japan, Makurai, 26–30 September 2010. 1561–1564. Grimaldi, Mirko. 2003. Nuove ricerche sul vocalismo tonico del Salento meridionale. Analisi acustica e trattamento fonologico dei dati. Alessandria: Edizioni dell’Orso. Grimaldi, Mirko. 2009. Acoustic correlates of phonological microvariations. The case of unsuspected highly diversified metaphonetic processes in a small area of Southern Salento (Apulia). In Danièl Tock & W. Leo Wetzels (eds.), Romance languages and linguistic theory 2006. Selected papers from ‘Going Romance’, Amsterdam 7–9 December 2006, 89–109. Amsterdam/Philadelphia: John Benjamins. Grimaldi, Mirko, Gili Fivela Barbara, Sigona Francesco, Tavella Michele, Fitzpatrick Paul, et al. 2008. New technologies for simultaneous acquisition of speech articulatory data: 3D articulograph, ultrasound and electroglottograph. Paper presented at Proceedings of LangTech, 2008, Rome, Italy. 1–5. Ladefoged, Peter & Maddieson Ian. 1996. The sounds of the world’s languages. Oxford: Blackwell. Lahiri, Aditi & Reetz Henning. 2001. Underspecified recognition. In Carlo Gussenhoven & Natasha Werner (eds.), Labphon VII, 637–675. Berlin: Mouton.

218 

 Mirko Grimaldi, Sandra Miglietta, Francesco Sigona and Andrea Calabrese

Lahiri, Aditi & Reetz Henning. 2010. Distinctive features: phonological underspecification in processing. Journal of Phonetics 38: 44–59. Li, Min, Kambhamettu Chandra & Stone Maureen. 2005. Automatic contour tracking in ultrasound images. Clinical Linguistics and Phonetics 19: 545–554. Lilienthal, Janine. 2009. The articulatory and acoustic impact of Scottish English /r/ on the preceding vowel-onset. In 10th Annual Conference of the International Speech Communication Association (ISCA), Interspeech, Brighton, September 6–10 2009. 2819–2822. Lindau, Mona. 1978. Vowel features. Language 54: 541–563. Mancarella, Giovan Battista. 1998. Salento. Monografia regionale della “Carta dei Dialetti Italiani”. Lecce: Edizioni del Grifo. Macmillan, Neial A. & Creelman C. Douglass. 2005. Detection theory: a user’s guide. Mahaw, NJ: Erlbaum. McCloy, Daniel. 2015. phonR: tools for phoneticians and phonologists. R package version 1.0–1. Miglietta, Sandra, Grimaldi Mirko & Calabrese Andrea. 2013. Conditioned allophony in speech perception: an ERP study. Brain & Language 126 (3): 285–90. Näätänen, Risto, Lehtokoski Anne, Lennes Mietta, Cheour Marie, Huotilainen Minna, Livonen et al. 1997. Language-specific phoneme representations revealed by electric and magnetic brain responses. Nature 385: 432–434. Näätänen, Risto. 2001. The perception of speech sounds by the human brain as reflected by the Mismatch Negativity (MMN) and its magnetic equivalent (MMNm). Psychophysiology 38: 1–21. Näätänen, Risto, Paavilainen Petri, Rinne Teemu & Alho Kimmo. 2007. The mismatch negativity (MMN) in basic research of central auditory processing: a review. Clinical Neurophysiology 118: 2544–2590. Näätänen, Risto, Kujala Teja & Winkler Istvan N. 2011. Auditory processing that leads to conscious perception: A unique window to central auditory processing opened by the mismatch negativity and related responses. Psychophysiology 48: 4–22. Parlangeli, Oronzo. 1953. Sui dialetti Romanzi e Romaici del Salento. Memorie dell’Istituto Lombardo di Scienze e Lettere, classe di Lettere, Scienze Morali e Storiche, 15–16(III), 93–198. Picton, Terrence W, Alain Claude, Otten Leun, Ritter Walter & Achim André. 2000. Mismatch negativity: different water in the same river. Audiology & Neuro-Otology 5: 111–139. Pulvermüller, Friedemann & Shtyrov Yury. 2006. Language outside the focus of attention: The mismatch negativity as a tool for studying higher cognitive processes. Progress in Nueurobiology 79: 49–71. Rohlfs, Gerhald. 1966–1969. Grammatica storica della lingua italiana e dei suoi dialetti, 3 voll., Torino: Einaudi. Or. ed. 1949–1954, Historische Grammatik der Italienischen Sprache und ihrer Mundarten, Bern: A. Francke AG. Romano, Antonio. 2013. Il vocalismo del dialetto salentino di Galàtone: differenze di apertura metafonetiche, tracce isolate di romanzo comune o interferenze diasistematiche? In Antonio Romano & Mario Spedicato (eds.), Sub voce Sallentinitas. Studi in onore di p. Govan Battista Mancarella, 247–276. Lecce: Edizioni del Grifo. Sams, M., Paavilainen, P., Alho, K., and Näätänen, R. 1985. Auditory frequency discrimination and event-related potentials. Clinical Neurophysiology 62: 437–448. Savoia, Leonardo & Martin Maiden. 1997. Metaphony. In Martin Maiden & Mair Parry (eds.), The dialects of Italy, 15–25. London: Routledge.



On integrating different methodologies in phonological research 

 219

Sigona, Francesco, Stella Antonio, Grimaldi Mirko, Gili Fivela Barbara. 2015. MAYDAY: A software for multimodal articulatory data analysis. In A. Romano, M. Rivoira, I. Meandri (eds.), Aspetti prosodici e testuali del raccontare: dalla letteratura orale al parlato dei media, X Convegno Nazionale dell'Associazione Italiana di Scienze della Voce (AISV), Torino, 22–24 gennaio 2014 – Università degli Studi di Torino, 173–184. Roma: Bulzoni. Stehl, Thomas. 1988. “Apulien und Salento”. In Gerhald Holtus, Michael Metzeltin & Carl Schmitt (eds.), Lexikon der Romanistischen Linguistik, Vol. IV, 695–716. Tübingen: Niemeyer. Stevens, Kenneth N. 1998. Acoustic phonetics. Cambridge MA: The MIT Press. Stone, Maureen. (2005). A guide to analyzing tongue motion from ultrasound images. Clinical Linguistics and Phonetics 19: 455–502. Sussman, Elise S., Chen Suan, Sussman-Fort, Jonathan & Dinces Elisabeth. 2013. The Five Myths of MMN: Redefining How to Use MMN in Basic and Clinical Research. Brain Topography 27 (4): 553–564. Tiede, Mark E. 1996. An MRI-based study of pharingeal volume contrasts in Akan and English. Journal of Phonetics 24: 399–421. Winkler, Istavan F. & Czigler Istvan. 2012. Evidence from auditory and visual event-related potential (ERP) studies of deviance detection (MMN and vMMN) linking predictive coding theories and perceptual object representations. International Journal of Psychophysiology 83 (2): 132–143.

M. Rita Manzini and Leonardo M. Savoia

Which syntactic information gets interpreted by phonosyntactic processes? Evidence from /u/ propagation in Southern Italian dialects

Università degli Studi Firenze

Abstract: We address the classical problem of the relation between syntax and phonology starting from the view that phonology, as an interpretive component, works together with morphology and the lexicon as an externalization device (Berwick and Chomsky 2011). We consider a set of phonosyntactic phenomena (i.e. syntactically conditioned phonological phenomena), involving the spreading of /u/ and the preservation of final -a in some Southern Italian varieties. In section 1 we review the data concerning the phonology of /u/ propagation and the syntactic conditions placed on it—as well as treatments suggested for phonosyntactic phenomena within the minimalist framework. In section 2, after providing a phonological analysis within the framework of Savoia (2015, this volume), we argue that the fine variation observed in phonosyntactic contexts of /u/ propagation requires the externalization component to read syntactico-semantic representations. In section 3 we provide a comparison with syntactic conditions on a different phonological process in the same dialects, final -a preservation, and draw some conclusions as to the nature of parametrization.

1 Variation  in /u/ propagation and theoretical background 1.1 Empirical evidence Harmonic processes between the ­pre-­tonic vowel and the stressed vowel involving the [I]/[U] elements are observed in many varieties of the Abruzzi, Lucania, Northern Calabria, and Central Sicily (Tuttle 1985; Savoia 1987). These processes apply word internally, as well as in phonosyntactic contexts (Rizzi and Savoia 1993). Descriptively, propagation is the result of the spreading of [U] properties from an unstressed nucleus to the stressed nucleus (or [a] vowel) immediately to the right (cf. Savoia 2015).

222 

 M. Rita Manzini and Leonardo M. Savoia

In Saracena, unstressed [u] spreads to underlying stressed /a/ to its right in open syllable, when the two nuclei are next to one other, yielding [ɔ:], for instance word internally in (1a). In (1b) underlying /a/ in open syllable has a long palatalized outcome, namely [ɛ:], in the absence of harmony. Stressed vowels other than /a/ are not affected, e.g. [i:] in (1c). The same phonological conditions hold in the phonosyntactic domains D ­ eterminer-­Noun in (2) and C ­ litic-­Verb in (3). Phonologically, (3c) shows that [a] is preserved in closed syllable. A stressed antepenultimate [a] is not affected either, as in (3d)—nor is an unstressed [a], as in (3e). (1) a. [stuˈtɔ:mə] [stuˈtɔ:və] b. [caˈmɛ:mə] [caˈmɛ:və] c. [muˈri:mə]

‘(we) put out’ ‘(I) put out’ ‘(we) call’ ‘(I) called’ ‘(we) die’

(2) a. [u ˈpɔ:nə] [u ˈnɔ:sə]

‘the bread’ b. [ˈpɛ:nə] ‘the nose’ [ˈnɛ:sə]

‘bread’ ‘nose’

(3) a. [u ˈfɔ:jə] [u ˈcɔ:mə] c. [u/a ˈfattsə] d. [u/a ˈcamənə] e. [u faˈtʃi:mə] [u caˈmɛ:mə] 

‘(you) it do’ b. [ˈfɛ:jə] ‘(I) it call’ [a ˈcɛ:mə] ‘(I) it(m./f.) do’ ‘(they) it(m./f.) call’ ‘(we) it do’ ‘(we) it call’

‘(you) do’ ‘(I) it(f.) wash’

Saracena

In Cerchiara in (4)–(6), phonological conditions are less restrictive than in Saracena. Harmonic spreading of ­pre-­tonic /u/ to stressed syllable yields diphthongs (or sequences) such as [ua], taking place both in open and in closed syllable, and independently of penultimate or antepenultimate stress, cf. the (a) examples. Stressed nuclei different from /a/ may also be involved, cf. [uɛ] in (6a) or [ui] in (6b), excluding however nuclei resulting from metaphony, e.g. [i] in (4c), (5c). Harmonic spreading does not take place on unstressed nuclei, cf. (4b), (6c). ­Word-­internal contexts are illustrated in (4), phonosyntactic contexts D ­ -­N in (5), and phonosyntactic contexts ­Cl-­V in (6). (4) a. [fukuˈʁuærə] ‘hearth’ [purˈtuæβə] ‘(I) brought’ [purˈtuæβəsə] ‘(you) brought’ [purˈtuæmmə] ‘(we) brought’ b. [ˈpɔrtənə] ‘(they) bring’ c. [kuˈtʃi:mə] ‘(we) cook’



Which syntactic information gets interpreted by phonosyntactic processes? 

(5) a. [u ˈnuæsə] [u ˈpuɛðə] [kulˈl uætʃənə] c. [u ˈpriβətə] [su bˈbiccə]

‘the nose’ b. [ˈnæɐsə] ‘the foot’ ‘that grape’ [ˈl ætʃənə] ‘the priest’ ‘they are old’

(6) a. [m u ˈðuæjə]

‘(you) me it give’

 223

‘nose’ ‘the grape’

b. [kə mə ˈðæɐjə] ‘what (you) me give?’ [u ˈfuættsə] ‘(I) it do’ [a ˈfættsə] ‘(I) it(f.) do’ [t u ˈðuitʃənə] ‘(they) you it tell’ [t a ˈðitʃənə] ‘(they) you it(f.) tell’ [u ˈkuættʃənə] ‘(they) it chase’ [a ˈkættʃənə] ‘(they) it(f.) chase’ c. [u katˈtʃæβəsə] ‘(you) it chased’ Cerchiara In Stigliano in (7)–(9) (Savoia 1987), harmonic spreading affects all stressed vowels and also ­pre-­tonic /a/. The presence of [ə] between the trigger of the harmony and the stressed nucleus does not block spreading, as in (9d). The outcomes of the harmony for stressed /a/ are [ɔ:] and [wɔ:] following a velar consonant in open syllable, and [wa] in closed syllable, cf. (7)–(9)(a). For p ­ re-­tonic /a/ the outcomes are [wɔ] after velar consonant and [ɔ] in other contexts, cf. (7)–(9)(c). For other stressed vowels we find the simple insertion of a [w] segment; outcomes of metaphony may also be affected. ­Word-­internal contexts are illustrated in (7). In (8) and (9) we display the phonosyntactic contexts ­D-­N and ­Cl-­V, respectively. The relevant harmonic process is triggered by the masculine singular D in (8) and by the masculine singular accusative Cl in (9). While in dialects like Saracena and Cerchiara the vowel that triggers the harmony is overtly realized, specifically as u in the masculine singular determiner and in the accusative clitic, in Stigliano the original *u in unstressed position is neutralized to [ə]. (7) a. [tsəmˈbɔ] ‘to jump’ a’. [ˈtsɔmbə] ‘(I) jump’ [təkˈkwɔ:və] ‘(I) touched’ [ˈtɔkkə] ‘(I) touch’ [addəmˈmwannə] ‘(I) ask’ [nəˈtʃwɛddə] ‘­nut-­dimin’ [ˈneutʃə] ‘nut’ (8) a. [lə ˈnɔ:sə] [lə ˈkwɔ:nə] [lə ˈdwiʃtə] [nə ˈdwɛntə] c. [lə trɔˈpe:ðə]

‘the nose’ b. [ˈna:sə] ‘the dog’ [ˈka:nə] ‘the finger’ [ˈdiʃtə] ‘a tooth’ [ˈdɛntə] ‘the tripod’ c’. [traˈpe:ðə]

‘nose’ ‘dog’ ‘finger’ ‘tooth’ ‘tripod’

224 

 M. Rita Manzini and Leonardo M. Savoia

(9) a. [lə ˈfɔ:jə]

‘(you) it do’

[lə ˈlwaskwə] [lə ˈstrwɛɲdʒə] [lə ˈlui:və] c. [lə kwɔˈnɔskə] [lə fɔrˈre:jə] [lə sɔˈpeimə] d. [lə dəˈvɔ:kə] 

b. [tʃə lə ˈfa:jə] ‘(you) what (for) him do?’ ‘(I) it leave’ [la ˈlaskwə] ‘(I) her leave’ ‘(I) it tie’ [ˈstrɛɲdʒə] ‘(I) tie’ ‘(you) it take away’ ‘(I) it know’ c’. [la kaˈnɔskə] ‘(I) her know’ ‘(I) it would do’ [farˈre:jə] ‘(I) would do’ ‘(we) it know’ [saˈpeimə] ‘(we) know’ ‘(I) it.m empty’ [la dəˈva:kə] ‘(I) it.f empty’ Stigliano

In Stigliano we can differentiate [ə] corresponding to *u from [ə] corresponding to *i,*e,*ɛ precisely in that the former triggers harmony, both ­word-­internally and in phonosyntax, while the latter doesn’t. In phonosyntax, one may say that masculine singular features introduce the phonological specifications [+back, +round]/[U], which though not realized in situ, are manifested through the harmonic process. W ­ ord-­internal contexts are slightly different; however, alternating forms where the [+back, +round]/[U] content overtly surfaces are available, as in (7a’). In this latter instance we will provide the relevant lexical bases with a phonological representation comprehensive of all properties not independently predictable, including [U]. As for the phonological content of segments we adopt an element analysis following in particular Harris and Lindsey (1995, 2000) and the revision in Backley (2011). As detailed by Rizzi and Savoia (1993), the various syntactic contexts that trigger propagation include in a generalized manner singular masculine determiners and the masculine singular object clitic, as illustrated in (1)–(9) in section 1.1. Beyond these, several additional contexts are documented where harmony may occur, though not in all varieties and optionally in some of them. Possible ­DP-­internal triggers include not only determiners but also quantifiers, adjectives, and nouns ­ P-­internal triggers, as as in (10)–(12)(a); AP internal contexts are also relevant. V in (10)–(12)(b) include not only clitics but also auxiliaries, modals, and causative verbs as well as negation. As we already saw in (1)–(9), full realization of /u/ is not required to trigger harmony, which can equally be observed when the trigger (e.g. masculine singular morphology) surfaces as [ə]. In (10)–(12) the + sign indicates obligatory application of harmony, the—sign stands for impossibility and the ± sign corresponds to optionality. (10) a. ­D-­N ­D-­A ­A-­N ­Q-­N

+ ± + ±

[u ˈpɔ:nə] [nu bˈbrɔ:və/bˈbræ:və ˈfeɈɈə] [nu bˈbɛllu ˈkɔ:nə] [ˈtʃində ˈpɔ:nə/ˈpæ:nə]

‘the bread’ ‘a good son’ ‘a nice dog’ ‘a hundred loaves’



Which syntactic information gets interpreted by phonosyntactic processes? 

 225

­Q-­A ± [ɲtʃ ɛ tˈtantu ˈsɔ:nə/ˈsæ:nə] ‘(he) not is so healthy’ ­N-­A ± [nu ˈfeɈɈə ˈsɔ:nə/ˈsæ:nə] ‘a son healthy’ b. ­Cl-­V + [u ˈƔrɔ:pə] ‘(I) it open’ ­Aux-­V + [ˈsʊnnu stæ:tə/stɔ:tə] ‘(I) am (i.e. have) been’ ­Copula-­A ± [su sˈsæ:nə/sˈsɔ:nə] ‘(they) are healthy’ ­Caus-­V ± [m u ˈfɔ:nə ˈfɛ/ˈfɔ] ‘(they) me it make do’ i.e. ‘they make me do it’ ­Mod-­V ± [u ßuˈli:mə ˈfɛ/ˈfɔ] ‘lo vogliamo fare’ ­V-­DP − [tə ˈduɲɲə ˈpæ:n e kˈkæ:sə] ‘(I) to you give bread and cheese’ c. ­DP-­V − [u pittʃuˈliddə ˈfɛr u maˈliɲɲə] ‘the child is naughty’  Saracena (11) a. ­D-­N + [u ˈnuæsə] ­D-­A + [nu bˈbuɛllə ˈkuænə] ­A-­N + [u səˈkunnə ˈfuɪʎʎə] ­Q-­N + [ˈkwandə ˈpuanə] ­Q-­A + [cu gˈgruænnə] ­N-­A + [nu ˈfwɪʎʎə ˈƔuæßəðə] b. ­Cl-­V + [u ˈfuættsə] ­Aux-­V + [ˈanə ˈnuætə ˈji:rə] ­Copula-­A + [su gˈgruæssə] ­Mod-­V + [jɪll u ßɔ ˈfuæ] ­Neg-­V + [ɔnˈn uæddƷə ˈßuistə ˈcu] ­P-­DP + [ku lˈl uækə] ­V-­DP ± [ˈjɪllə ßɔ pˈpuæn e pprəˈsuttə] [ˈjɪllə ɔ pˈpæn e pprəˈsuttə] c. ­DP-­V − [u ˈƔuællə ˈkæntəðə] 

‘the nose’ ‘a nice dog’ ‘the second son’ ‘how much bread’ ‘more big’ ‘a son tall’ ‘lo faccio’ ‘(they) are born yesterday’ ‘(they) are fat’ ‘(he) it wants (to) do’ ‘(I) not have seen anymore’ ‘with the needle’ ‘he wants bread and cheese’

(12) a. ­D-­N + [lə ˈnɔ:sə] ­D-­A + [nə bˈbwɛllə ˈfɪɈɈə] ­A-­N − [nə bˈbrɔːvə ˈfɪɈɈə] ­Q-­A − [ccu gˈgrannə] ­Q-­N − [ˈɔttə ˈka:nə] ­N-­A − [nə ˈsweccə ˈƔrannə] b. ­Cl-­V + [lə ˈfwattʃə] ­Aux-­V − [ˈl addƷə maɲˈdƷa:tə] ­ Copula-­A − [so vvaˈkantə] ­Caus-­V − [tə lə fɔˈtʃejmə ˈfa]

‘the nose’ ‘a handsome son’ ‘a good son’ ‘more big’ ‘eight dogs’ ‘a bucket big’ ‘(I) it do’ ‘(I) it have eaten’ ‘(they) are empty’ ‘(we) you it make do’ i.e. ‘we make you do it’

‘the cock crows’

Cerchiara

226 

 M. Rita Manzini and Leonardo M. Savoia

­Mod-­V ­V-­DP c. ­DP-­V 

− [lə ƔwəˈlØːmə maɲˈdƷa] − [təˈnejmə ˈse:tə] − [lə pəttʃəˈnwennə ˈcaɲdƷə]

‘(we) it want (to) eat’’ ‘(we) are thirsty’ ‘the child cries’ Stigliano

In (13) we provide a schematic representation of the syntactic distribution of harmony. Stigliano, which has laxer phonological constraints, shows the most severely restricted syntactic distribution, since only the contexts D ­ -­N/A and ­Cl-­V are active. In Cerchiara and Saracena harmony applies within the DP, including the ­A-­N context and a large subset of ­VP-­internal contexts, such as Aux/ Modal/­Copula-­V. Saracena nevertheless is more restrictive than Cerchiara, since harmony only variably applies in the ­Q-­N/A contexts and negation is excluded from possible triggers. The ­V-­DP context is variably admitted in Cerchiara, though it is once again excluded by Saracena. (13) a. ­D-­N/A ­A-­N ­N-­A ­Q-­N/A b. ­Cl-­V ­Aux-­V ­Copula-­A Mod/Caus-V ­Neg-­V ­V-­DP c. D ­ P-­V

Saracena + + ± ± + + ± ± − − −

Cerchiara + + + + + + + + + ± −

Stigliano + − − − + − − − − − −

1.2 From /u/ propagation to minimalist syntax The microvariation in (13) was originally analyzed by Rizzi and Savoia (1993). As aptly summarized by Elordieta (2008: 210), Rizzi and Savoia (1993) are among theorists holding that “phonological operations are directly sensitive to syntactic information, in terms of relations of c­ -­command or ­m-­command (i.e. government) holding between the elements participating in phonological processes” (cf. Manzini 1983). This kind of view is opposed by theories holding that “there is a distinct level of representation called Prosodic Structure which mediates between the syntactic and phonological components … Phonological operations themselves do not refer to syntactic constituents, but to the already created prosodic constituents” (cf. Nespor and Vogel 1986 for an elaboration of this view).



Which syntactic information gets interpreted by phonosyntactic processes? 

 227

Rizzi and Savoia characterize the variation in (13) in terms of structural relations of syntactic government between the harmony trigger (containing the licenced unstressed nucleus) and the target (containing the full licencing nucleus). In particular the trigger of propagation may govern the target: (i) as a functional head (­F-­government); (ii) in an agreement configuration (­Agr-­government); and (iii) in a configuration of mutual government (­M-­government). Under this model, the most severe restrictions associated with the variety of Stigliano, are due to the fact that harmony requires the combination of the two conditions of ­F-­and ­M-­government. However, Cerchiara and Saracena are less restrictive, in the sense that ­Agr-­government or ­M-­government are sufficient to trigger harmony. The model also seeks to capture optional environments for harmony. Thus optional application is triggered by simple government in Cerchiara and by ­Agr-­ or ­F-­ or ­M-­government in Saracena, as in (14). (14) Saracena Cerchiara Stigliano Obligatory ­Agr-­gov ­Agr-­gov ­F-­gov or ­M-­gov or ­M-­gov and ­M-­gov Optional ­Agr-­/­M-­/­F-­gov gov Elordieta (1997, 2008) points out empirical problems with the extension of the government model of Rizzi and Savoia to Vowel Assimilation in Basque. In particular “the … branching configurations holding between the participial verb and an inflected auxiliary, a modal particle or a causative verb are identical”, though only the ­ participle-­ auxiliary configuration triggers the rule. Therefore “the government relationships are the same in all cases between the different heads. Since both an inflected auxiliary and a modal particle … are functional heads, ­F-­government does not seem to be a solution, and neither does ­Agr-­government in the sense of Rizzi and Savoia, as this relationship only holds between expressions displaying morphosyntactic agreement in gender and/or number, i.e. in nominal contexts” (Elordieta 2008: 2­ 60–­261). He concludes that the way to capture differences in behavior between functional categories is to make reference to the different feature checking relations they enter into with the lexical heads that they are linearly adjacent to. “Certain phonological phenomena may be specified to apply in the domains or constituents formed by functional and lexical heads related by feature checking”, to be understood as in Chomsky’s (1995) minimalist program (Elordieta 2008: 210). In a similar vein, Roberts (2005) studies Initial Consonant Mutation in Welsh in structures of the type ­Determiner-­Noun, ­Particle-­Verb and ­Verb-­Object. He concludes that the relevant context is defined by head government, i.e. [Y’ Y XP]. In his analysis, an autosegment is associated with a syntactic head, e.g. the autosegment L(enition) to the head v. When the verb V raises to v, the initial consonant

228 

 M. Rita Manzini and Leonardo M. Savoia

of the DP object in [Spec, V] becomes associated with L, which overtly realizes the Accusative feature. In surface structure the verb ultimately raises higher, to a finite agreement position (PersonP) and is not adjacent to its object DP, which nevertheless overtly realizes L. In other words, minimalist feature checking (or Agree) defines phonosyntactic processes. Recent proposals explain the syntactic distribution of phonological phenomena by having recourse to the notion of phase and to the Phase Impenetrability Condition (PIC) of Chomsky (2001, 2007) (Kratzer and Selkirk 2007; Cheng and Downing 2009; D’Alessandro and Scheer 2015; Newell and Piggott 2014 among others). These works share the assumption that syntactic objects constructed by narrow syntax are immediately visible and legible to phonological operations (­Sensory-­Motor interpretation). In other words, there is no theoretical or descriptive necessity to introduce an intermediate prosodic representation which ­re-­organizes the syntactic string in terms of phonological categories, before its interpretation at the interface level. The insight of these works is that there are sequences of syntactic objects—words or morphemes—that form a unique overall domain for phonological processes, though the same processes can be blocked if apparently similar sequences include a phase boundary. D’Alessandro and Scheer (2015: 2) observe that “the very essence of phase theory is to define items that are bigger than the word, and to send them to PF (and LF). Against this backdrop, our first goal is to show that there can only be one ­chunk-­defining device: theory cannot afford to have the same work done twice. We argue that this unification must be in favor of the derivational mechanism: phase theory has independent syntactic motivation, while this is not the case for prosodic constituency on the phonological side”. Similarly, according to Newell and Piggott (2014), a prosodic domain like Phonological Word is constructed on the basis of syntactic phases, though it has a representational nature which is independently preserved after phases are interpreted and thus no more legible to the ­sensory-­motor (SM) interface. As already stressed by Chomsky and Halle (1968), phonology systematically applies inside the word domain, whereas the application spanning two separate constituents typically involves constraints. In general, phonological models assuming cyclical application of phonological rules, for instance prosodic rules, allow for word internal domains, formed by means of morphological derivation, and possibly extended with clitic elements. In Lexical Phonology (Kiparsky 1982a, 1982b) and other cyclic models like Halle and Vergnaud (1987), Halle and Idsardi (1995) phonological rules are sensitive to the properties of affixes in the process of word formation. Newell and Piggott (2014) highlight differences between this type of solution and the ­phase-­based analysis they propose. They argue that only a ­phase-­based analysis can distinguish between different behaviors of the same



Which syntactic information gets interpreted by phonosyntactic processes? 

 229

morphological exponent depending on its point of insertion in a phase domain; a simple cyclical application depending on word internal stratification of affixes is not able to capture this difference. Specifically, they consider hiatus avoidance in Ojibwe, an Eastern Algonquian language. This language disallows vowels in hiatus inside the word and it has recourse to vowel fusion, deletion, or consonant epenthesis to eliminate hiatus contexts created by morphological derivation. However a subset of verbal or nominal affixes tolerates hiatus. Thus we find sequences in which hiatus is resolved by means vowel fusion/deletion, as in (15a), as well as sequences which preserve hiatus, as in (15b). (15) a. niwe:ʒi:na:na:nig ­ni-­we:ʒi:-­in-­a:-ina:­ni-­­Ø-­ag ­1-­­paint-­­final-­ts(3 theme)-­1plural-­­ind-­3plural ‘we paint them’ b. gí:anò kì: gi:-anoki: ‘­past-­work’ ‘he worked, hunted’  (Newell and Piggott 2014: 336, 338) The question is defining the domains which admit or exclude hiatus. Newell and Piggott conclude that for verbs the relevant domain of exclusion is the Event, including a set of possible phases (the verbal root √P, v, Voice, Asp, and Agr) which are suspended in virtue of a condition of ‘phase extension’ requiring that “If the combination of a ­phase-­head and its complement contains an uninterpretable feature, transfer to the interfaces will be delayed until the merger of a later head with the capacity to check this feature. If no such delay is motivated by either the featural properties of H or Z, ­Spell-­out can occur as early as the merger of the ­category-­defining head” (Newell and Piggott 2014: 345, cf. ­Svenonius 2004). The authors further propose that morphological elements, like tense formatives, which admit hiatus are inserted out of the Event domain, in a different phase; the PIC between the two different phases creates an interpretive barrier which justifies the ­non-­application of hiatus resolution, in spite of the phonological similarity of the sequences involved. Analogously, D’Alessandro and Scheer (2015) consider the different behavior of the same auxiliary in the active perfect (transitive/intransitive) and in the passive of the Abruzzo dialect of Arielli. In this dialect, the perfect auxiliary forms are independent of verbal class and voice and only dependent on person. D’Alessandro and Scheer note that in passive configurations they determine Raddoppiamento Fonosintattico of the initial consonant of the following participle, while in the active this process does not apply, as in the examples in (16). In the active, as in (16a), there is a vP phase, blocking Raddoppiamento in the phonology.

230 

 M. Rita Manzini and Leonardo M. Savoia

On the contrary in the passive (16b), v is defective, i.e. it is not a phase head; as a consequence the auxiliary and the participle are comprised in the same phase and Raddoppiamento can take place. (16) a. so  vistə I.am seen ‘I have seen’ b. so    vvistə I.am  seen ‘I am seen’ The most problematic point of a ­phase-­based theory of phonotactic phenomena is how it can account for subtle (and idiosyncratic) differences in phase organization of similar languages, as in the case of different restrictions on propagation analyzed above. As D’Alessandro and Scheer (2015: 2) note, “phase theory is unable to describe all phonologically relevant chunks, as these are too small and too diverse to be delineated in the current working. In order to qualify as the only ­chunk-­defining device in grammar, phase theory needs to be made more flexible: in other words, it needs to be adapted to the demands of phonology”. This means that two major dimensions of variation are introduced by the authors. First, different languages may have different phasal heads, hence phasal domains. Second, the PIC need not be activated both in the syntactic component and in the phonological component, but may be activated only in one of the two. Thus consider the failure for Raddoppiamento to apply in (17), between the auxiliary and an unaccusative verb. Syntactically, we expect the same configuration as in (16b), characterized by no vP phase/a defective vP phase and allowing for promotion of the internal argument to the subject position—yet in the phonology, Raddoppiamento is blocked as in (16a). This ­phonology-­syntax mismatch is resolved by taking vP to be a phase; “a PIC is associated with this access point at PF. In syntax, however … no PIC is associated with v”. According to D’Alessandro and Scheer the parameters they propose are compatible with the ‘­Borer-­Chomsky conjecture’ as to the lexical nature of parametrization. They assume that phasehood is a lexical property associable to certain heads; furthermore for Arielli in (16)–(17), they suggest that “PIC at PF is linked to an active value for the voice feature on v. This feature value seems to be the syntactic correlate to the PIC effect at PF”. (17) so rəmastə ­am-­1st.sg ­stayed-­sg ‘I have stayed’ When it comes to evaluating ­ phase-­ based approaches we note, first, that a ­well-­known divide separates word internal domains from phonosyntactic ones.



Which syntactic information gets interpreted by phonosyntactic processes? 

 231

Thus, Newell and Piggott (2014) for Ojibwe reconstruct the contrast between the application of a phonological rule inside the word, where it is obligatory, and in contexts external to it, where it does not apply or it applies with restrictions, in terms of phases. Our data indeed show that propagation systematically affects the word internal contexts in (1a, 4a, 7a). Modelling the notion of word in terms of the notion of locality/cyclic domain—or phase—is appealing for more than one reason; in particular it would resolve the tension between the listing of morphemes as atomic lexical items and the necessity to predict that words are nevertheless insulated from most processes active within syntax (the lexical integrity hypothesis). In the next section we will be more directly concerned with syntactic domains of application. Capturing the varying conditions on phonosyntactic rules in terms of the sole notion of phase, as D’Alessandro and Scheer (2015) do, involves a considerable enrichment of this notion. Thus the need to account for fine variation means that any projection potentially becomes a phase—and the child who learns a language additionally faces decisions as to whether a PIC attaches to this phase in the phonology or in the syntax. We note also that ±phase or ±PIC are not lexical parameters, since they involve not bona fide properties of lexical items, but rather encode derivational instructions. In general, while the terminology of Chomsky (2001, 2007) is maintained, it is partially voided of its actual content. First, if phases are heavily parametrized, it is not obvious that they can keep regulating movement within the sentence (­A-­movement) and out of it (A’-movement) as well as agreement (from a phase head to its complement and the edge of the lower phase). The parametrization of the PIC should go some way towards resolving this issue in D’Alessandro and Scheer (2013) model. Yet unless it is connected to the syntactic PIC, and the syntactic PIC to ­Spell-­Out, the notion of phase may become empirically vacuous. In other words, in the worst instance, reference to phases may in fact introduce an intermediate level of representation, gathering words/morphemes in chunks legible to the SM interpretation, which configures a sort of surreptitious prosodic organization. Similarly, in the original minimalist conception (Chomsky 2001), probes, i.e. uninterpretable feature sets triggering checking by an interpretable set, coincide with phase heads—i.e. essentially v and C (with I ‘inheriting’ from C). In order to capture the fine parametric variation in phonosyntactic triggers, we may want to elaborate a finely grained repertory of heads against which other heads are checked, as suggested by Elordieta (1997, 2008). This has the advantage of achieving descriptive adequacy—but in the process, the independently needed notion of probe gets complicated so as to become again a way of encoding the facts. In short, notions connected with locality and cyclicity like those of phase, the PIC, and agreement probes are insightfully brought to bear on the definition of phonosyntactic domains by recent minimalist literature. However

232 

 M. Rita Manzini and Leonardo M. Savoia

the restrictive syntactic formulations of such notions are changed in the process— in other words it is not the syntactic notion of phase, PIC, or probe that is being referred to but some different, richer notion. We shall return to specific problems these theories may encounter in accounting for the data in (13) at the end of the next section.

2 The present account 2.1 Phonological analysis In this section we discuss the phonology of propagation including its variation. As we saw, in some varieties (here Saracena, Calabria), only /a/ of the stressed syllable immediately following /u/ undergoes harmony. In other varieties (here ­Stigliano, Lucania and Cerchiara, Northern Calabria), all n ­ on-­ round vowels in stressed position are possible targets of harmony. In the varieties of Lucania ­pre-­tonic [a] is also targeted by the process. The government phonology framework (Kaye  et  al. 1990) allows for an interesting explanation of the observed ­micro-­variation, interpretable as the result of the interaction between different nuclei, the requirements on the segmental content of the positions involved and metrical properties connected to quantity sensitivity. Specifically, propagation applies in the prosodic domain of a stressed nucleus—a fact that we will return to. Consider first the more restrictive Saracena data. As shown in (1b), underlying /a/ in open stressed syllable has the palatalized outcome [ɛ:], as generally attested in South Italian systems, characterized by sensitivity to syllabic structure. In these systems, in the context of open syllable, i.e. a nucleus with two positions, the phonological element [I] is added to /a/, changing it to [A, I] ([ɛ:]). A  more complex and recognizable configuration of the vocal tract is thereby obtained, which strengthens its contrastivity. When /u/ propagation is at work, the [ɔ] result is obtained, which satisfies requirements of a more recognizable phonological content, again in an open syllable context. In (18) we therefore ­tentatively unify velar harmony with the palatalization connected to syllabic structure under the same mechanism. In the contexts of closed syllable or antepenultimate stress, the requirement in (18) is not met, excluding harmony as well as palatalization. (18) Palatalization/Velarization of /a/ (Saracena) [A] associated with a branching nucleus is licenced within the expression including [I/U]



Which syntactic information gets interpreted by phonosyntactic processes? 

 233

(19) accounts for harmonic assimilation. The restriction that defines harmonic requirements between the stressed nucleus and the p ­ re-­tonic vowel works as a ­well-­formedness restriction, filtering in all and only the w ­ ell-­formed representations. In a nutshell, the harmonic process corresponds to licencing, within the word or the clitic group domain, by the head nucleus of the domain. The latter subsumes the [U] property, licencing the corresponding property present on the ­pre-­tonic vowel, as schematized in (20). (19) Propagation [U] in stressed position licences [U] in p ­ re-­tonic position in its immediate domain. (20) Saracena


N

x

u c U

N

N

x x x

x

m

c

x

A

e

N A

>

U

A

Cerchiara in turn displays sensitivity to syllable structure. It presents long stressed vowels, generally diphthongized, in open syllable; in closed syllable and in antepenultimate position it presents short stressed vowels on average, even if reduced diphthongs may occur. Underlying a is generally realized as [æ], i.e. [A, I], which in open syllable is long or diphthongized in [æɐ]. We assume, as in (19), that the spreading of [u] results from the licencing of ­pre-­tonic [u] by the stressed nucleus. When subject to [u] harmony, the outcome of underlying a is [uæ(:)]; it is also possible to find realizations of the type of [uɑ] in which [U] replaces [I]. Harmony affects [−back]/[I] vowels, simply inserting [u] in front of them; the fact that [u] does not become associated with the content of the stressed vowel may be analyzed as a consequence of the fact that the grammar of Cerchiara does not admit *[I, U] vowels, including [y, Ø, œ]. Despite the sensitivity of Cerchiara to ­metrico-­syllabic structure, insertion of [u] into the stressed nucleus does not take the structure of the rhyme into account. Therefore this element becomes associated with the first position of the nucleus, yielding the [uæ] diphthong, both in open and in closed syllable, as shown in (21)–(22).

234 

 M. Rita Manzini and Leonardo M. Savoia

(21) Cerchiara (22) Cerchiara N

N N

A

N

A

N

N A

N R N

x

x

x x x

x

x

x

x

u U

n

u æ s U

e

u U

f

u æ U

>



>

x x

A

N

x

x

ts

e



N

In the Cerchiara variety, harmonic spreading is blocked by a metaphonetic stressed nucleus. In both propagation and metaphony, the stressed vowel licences properties which are present in weak positions. Therefore the stressed vowel licences also the properties associated with metaphony. Moreover, in these varieties, metaphony is at least partially morphologized, in that original ­post-­tonic [+high, +ATR] vowels are no longer or only variably realized. P ­ ost-­tonic vowels are generally neutralized in [ə], as in the examples in (1)–(9). We surmise that metaphony corresponds to a morphologized process whereby the stressed nucleus subsumes and licences the [+high, +ATR] content introduced by the morphological category masculine singular or plural in nouns. If metaphony and propagation can be unified under the same schema in (23), then the restriction against the application of propagation to metaphony contexts depends on the fact that each nucleus can licence only once. More precisely, metaphony applies in the most restricted domain formed by the foot headed by the stressed nucleus, while the domain of propagation includes this inner domain and the phonological material in ­pre-­tonic position. We conclude that if a single constraint is involved, in this variety an ‘Elsewhere’ condition favors its application in metaphony. (23) Metaphony/Propagation An [I/U] head is licenced in the position that governs the prosodic domain. Lastly let us consider Stigliano, where the application of harmonic spreading involves all stressed nuclei, including metaphonetic ones. In Stigliano as well, there is a restriction against *[I,U], whereby the combination of [U] and [I] within the same vowel is excluded. This explains the fact that with input vowels [i e ɛ] the harmonizing element is inserted as the first component of a sequence [wi], [we], [wɛ]. What is more, [w] is inserted independently of the structure of the rhyme, and spreading does not take into account the properties of the onset. In other words, spreading only sees the articulatory properties of the host vowel. The latter must be a phonetically full unrounded vowel. Hence, considering



Which syntactic information gets interpreted by phonosyntactic processes? 

 235

that in this variety phonetically full vowels include [a] or vowels under stress, propagation affects a stressed vowel or a ­pre-­tonic [a]. Unstressed [a] behaves in fact differently from other vowels, in that it can take on the [U] element, yielding [ɔ], as in (24). In closed position, [w] is inserted, creating a [wa] sequence, suggesting that the possibility for [a] to velarize depends on additional factors. Thus the [ɔ] outcome in open syllable satisfies a requirement of the type discussed above for Saracena in (18). In closed syllable/ antepenultimate stressed position, on the contrary, no such requirement holds, and an [U] element in initial position is inserted. What is also noteworthy in (24) is that the [U] trigger is associated with masculine singular morphological content, rather than with a phonetically realized [u] segment. In other words, we are dealing with a phenomenon that, as in the case of metaphony, cannot be reduced to a simple phonological mechanism, but is triggered by morphological restrictions. Thus we assume that a given interpretive content, here masculine (singular) nominal inflection, is associated with a set of phonological properties which are realized on the ­non-­neutralized, stressed vowel. (24) Stigliano N N A

N

A

N

A

N

x

x

x

x x

x

x

e

n

c

s

e

l

msg

U

U

A

A second context where harmonic [ɔ] is realized in Stigliano is the ­pre-­tonic one. Therefore harmonic spreading treats stressed nuclei and unstressed [a] nuclei in the same way, suggesting a finer grained account of the phenomenon. We surmise that [a] shares a set of basic properties with stressed vowels, namely a full and clear acoustic configuration, with a degree of perceptibility not inferior to that which characterizes stressed nuclei. In other words the role of prosodic head is a product of phonetic properties of length, sonority, and perceptibility, which are normally associated with the stressed nucleus. Nevertheless [a] displays these properties to a degree sufficient to be implicated in harmonic processes. In unstressed position, the contrast between open syllable and closed syllable/antepenultimate stressed position is not relevant. The simplest outcome of the harmonic process is then manifested, as velarization in [A,U]. The representation in (25) characterizes sequences where a velar consonant selects the insertion

236 

 M. Rita Manzini and Leonardo M. Savoia

of [w] even in unstressed position, favoring a more autonomous realization of the [U] element. Following Backley (2011), we adopt the hypothesis that the [U] element characterizes a velar consonant. Consonants with properties different from [U] do not interfere with spreading. On the contrary a velar onset consonant concurs to licence the realization of [w] even in the presence of velar outcomes [ɔ], manifesting the [U] property of the velar consonant, as in (25). (25) Stigliano N

N

x

x

x

x x

x x

x

n

s k

c

k w ʔ U U U

e

msg

c



A

e

l

N

N R A N

A

A

The transparent nature of centralized weak vowels [ə], i.e. of the [A] ­non-­dominant type, becomes evident when one or more weak vowels intervene between the ­pre-­tonic trigger and the stressed nucleus. In such contexts, [ə] allows spreading across it, as in (26). We conclude that in the Stigliano propagation, the process is restricted only to full vowels, i.e. those endowed with a dominant element [A]/[I]/[U], excluding thus [ə], associated with a non head [A]. Harmony amounts to the licencing of [U] by the full/stressed nucleus; intermediate weak vowels are simply not legible to the process. (26) Stigliano

N

x

x

x

e

ð

e

msg

U

A

A

N

x

x x x

x

v

k

c



A

e

l

N

N N N R A N

U A



Which syntactic information gets interpreted by phonosyntactic processes? 

 237

Besides the word and the clitic group, exemplified so far, other prosodic strings may be available for the application of phonological processes. For instance in [anu ˈnuætə] ‘(they) have (been) born’ in (27), the rightmost stressed nucleus targeted by the harmonic process, is the nucleus that licences the entire prosodic constituent; in this sense it is an adequate phonological licenser for propagation. (27) Cerchiara N N N

A

N A N

x

x

x x x x x

x

a

n

u n u æ t

e



N A N

In section 1.2 we have discussed recent approaches to the analysis of phonotactic processes, including phase-based ones, agreeing with them on the crucial assumption that phonological interpretation depends on syntactic information, without recourse to intermediate representations. In the structures in (20)–(27), however, it is equally crucial to make reference to the prosodic domain of the stressed nucleus that hosts [U], for which we may conveniently employ conventional labels such as word, clitic group, and so on. This is not to say that syntactic domains are encoded by prosodic objects. Rather, at the SM level a string of syllables gathered in feet forms a phonological word; a sequence of words forms a larger prosodic domain headed by the rightmost stressed vowel. In general, these prosodic domains are defined by the SM component in terms of its own primitives like timing slots, prominence, and prosodic groupings. In this way, the phonological merger of more elementary domains into larger ones creates the structures that make syntactic input legible at the SM level—without any precompiling of the syntactic information itself.

2.2 S  yntactic contexts of obligatory and impossible application The harmonies we are discussing are triggered by syntactic elements which systematically include the masculine singular determiner as well as the masculine singular (object) clitic. What do these contexts share that sets them apart from other contexts subject to variation and optionality?

238 

 M. Rita Manzini and Leonardo M. Savoia

The minimalist model of agreement based on feature checking between an uninterpretable probe and an interpretable goal, is not of easy application in contexts like the Romance DP characterized by agreement of essentially all of the heads present in the structure. Consider standard Italian ­l-­a ­ros-­a ‘­the-­f ­rose-­f ’. If the D (la) is the probe (being presumably the phase head), then its features are uninterpretable; but this runs counter the fact that la bears the referential properties of definiteness and is indeed interpretable in isolation, namely as a (clitic) pronoun. Therefore Manzini and Savoia (2005, 2007, 2011) assume that all features are interpretable. Following Higginbotham (1985), in the expression ­ros-­a ‘­rose-­f ’, ­ros-­ ‘rose’ is a predicate with an open argumental slot (the ­so-­called ­R-­role), denoting the set of individuals who have the property of being ‘rose’. Manzini and Savoia propose that the nominal inflection (N) -a contributes an elementary descriptive content (feminine) towards the saturation of the ­R-­role.1 The determiner concurs with the N inflection to the saturation of the ­R-­role, introducing in this instance reference to a definite individual. In order for the determiner and the N inflection to be interpreted as discontinuously fixing the referent for the argument slot of the noun, agreement must hold between them, i.e. identity (or ­non-­distinctness) of their referential features. If desired, a single copy of the agreement features may be preserved at the LF interface, namely the highest one (leaving the other copies as bound variables), matching minimalist Agree. Going then on to /u/ propagation, the structure in (28) represents the D ­ -­N configuration in one of the relevant varieties, cf. example (12a). Pursuing the line of syntactic analysis sketched above, the lexical base ­nas-­ ‘nose’ has a purely predicative content, while the nominal inflection (N) -ə and the determiner D introduce referentially relevant properties for the saturation of the predicate (masculine). Together D and the N inflection individuate a single referent for the ­R-­role of the predicate, characterized as definite and masculine singular. (28) Saracena

N

cn



uy U

U

:sλy

N

e

D

y

1 For discussion of the notion of nominal inflection N as understood here, see also Savoia (this volume, fn. 1), Franco et al. (2014).



Which syntactic information gets interpreted by phonosyntactic processes? 

 239

The C ­ l-­V context in (29) is similar. Here too we have a predicate ‘wash’ which takes two arguments. One of the two arguments, the external argument, is saturated by the verb finite inflection; recall that the varieties we are considering have p ­ ro-­drop. The other argument, namely the internal one, is saturated by the clitic—whose inflection contributes nominal content N to it. Therefore structures like (28) and (29) can be assimilated on the basis of the fact that in either instance the trigger of /u/ propagation, i.e. the determiner in (28), and the clitic in (29), saturate an argument of the predicative base— and precisely the internal argument in both instances, contributing nominal properties to it. (29) Saracena N



uy U

V lɔ:βλx,λy U

D ǝx

It is worth noting that our data exemplify contexts in which enclitic pronouns in imperatives are affected by propagation, including the most restrictive of three dialects, namely Stigliano. Since for reasons entirely independent of the phenomenon at hand, it is the enclitic that attracts the main stress in the clitic group, the trigger of the spreading is either a vowel in the verb base, as in (30a) or in its inflection, as in (30b); the target is the enclitic. In this instance therefore the ­argument-­of relation is reversed with respect to (28)–(29) (the trigger is the predicate and the target is the argument). (30) a. [[kus] ˈweilə] Stigliano ‘sew it’ b. [[purtamu] nˈ­n-­wellə] Cerchiara ‘bring to.him-it’ In short, in terms of the present proposal, obligatory application of /u/ propagation in all of the varieties considered corresponds to contexts where the trigger and the target are in an argument-of relation, as in (31). This can also be expressed in ­probe-­goal terminology, assuming that D is both a phase head and a probe in (28), while v is a phase head and a probe in (29)–(30). (31) The Trigger is an internal argument of the Target or vice versa (The Trigger probes for the Target or vice versa)

240 

 M. Rita Manzini and Leonardo M. Savoia

A note is in order on the application of harmonic processes word internally. In practice, all of the examples of w ­ ord-­internal /u/ propagation in (1)–(9) involve a morphologically complex word, where the trigger of the spreading is a verbal or nominal base and the target is either a person inflection for verbs or a word formation suffix for nouns. A ­V-­D example is analyzed in (32), where the inflection -ɔmə ‘1pl’ lexicalizes the EPP argument (i.e. the external argument) of the verb ­stut-­ ‘put out’ to which it attaches. Therefore the ­argument-­of relation is once again involved in propagation, as under (31). Since morphological merger creates left branching structures, propagation has the effect of strengthening the PF externalization of the predicative content of the lexical base (the trigger) on the person and number inflection (the target)—essentially as in enclisis examples like (30). (32) Saracena



V stutλx,λy U

D ɔmǝx U

Something similar can be said of an ­ N-­ Suffix example like [nəˈ­ tʃ-­ wɛddə] ‘­nut-­diminutive’ of Stigliano, where the target of propagation, i.e. the diminutive suffix, is effectively a classifier (for ‘small size’), hence part of the system for ultimately fixing the argument satisfying the predicative base of the noun (the trigger). Though here we focus essentially on conditions for phonosyntactic application, it may be worth noting that no unification seems to be possible between (31) and the ­word-­internal facts, if (31) is stated in terms of ­probe-­goal relations. Let us then move to the contexts that are obligatorily activated in Cerchiara and Saracena, though the restrictive variety of Stigliano excludes them. ­DP-­internally these coincide with A ­ -­N, as in (33). The internal structure of the noun is as already detailed in (28), the internal structure of the adjective is also the same (the predicative base satisfied by the inflectional content). Now, the noun and the adjective introduce two separate predicates, in (33) ‘nice’ and ‘dog’. The desired interpretation of the ­A-­N structure is that there is a single individual that satisfies both properties, i.e. lies at their intersection (namely it is both ‘dog’ and ‘nice’ for a dog). The semantic operation that yields this interpretation is called ­theta-­identification by Higginbotham (1985). The syntactic operation corresponding to it is again agreement, understood here as the matching of locally identical ­phi-­feature exponents that concur to denoting a single referent (here controlling two argument slots). Therefore in (33) the trigger x (the N inflection of



Which syntactic information gets interpreted by phonosyntactic processes? 

 241

the adjective) is an argument of the target (the noun)—but via t­ heta-­identification of x with the argument (the N inflection) of the noun. (33) Cerchiara A A bwεllλy

N N N ǝy(msg)=x kuænλx U U

N ǝx

The other context that obligatorily triggers propagation in Saracena and Cerchiara is ­Aux-­V, illustrated in (35). The N inflection of the participle picks up the internal argument of the verb (Burzio 1986). The finite verb inflection, here 1st plural (‘we’), lexicalizes the EPP argument of the sentence. Since in (34) there is a single argument slot, a single argument is individuated by the finite inflection and by the participial one. Given a transitive active lexical verb, however, the participial inflection would pick up its external argument and the finite (EPP) inflection its external one, cf. [æddƷə ˈßuistə] ‘I.have seen’ of Cerchiara. What is true of all A ­ ux-­V configurations is that though two verbs are involved, there is a single event. Therefore again the trigger of propagation (here the ‘we’ inflection) is an argument of the target (the lexical verb)—though only once event identification is taken into account. (34) Saracena V V sʊ



VP D nuy U

...

V V stɔ:tλy U

N ǝy

Given the discussion that precedes, we surmise that obligatory application in Cerchiara and Saracena, but not in Stigliano, is determined by the presence of an ­argument-­of relation between the trigger and the target, except that the relation holds only as a consequence of a unification of the argument structure of the predicate bearing the trigger inflection and the target predicate, along the lines of (35). In terms of ­probe-­goal relations, in (33) the adjective and the noun (trigger and target) can be described as both being goals of a D probe; in (34) we can say

242 

 M. Rita Manzini and Leonardo M. Savoia

that the pro subject is a goal both for the v probe (the lexical verb) and for the I probe (the ­auxiliary). Yet, when a transitive verb is involved, the auxiliary and the verb probe different arguments (cf. [æddƷə ˈßuistə] ‘I.have seen’ of Cerchiara, discussed above). In probe-goal terms it is therefore difficult to think of a natural class encompassing these different contexts and differentiating them from those in (31). (35) The Trigger is an argument of the Target via theta/­event-­identification. Now, certain configurations block /u/ propagation in all varieties, including the less restrictive variety of Cerchiara. Not surprisingly they are independently known to block various other phonosyntactic processes considered by the literature—the most obvious example being the ­subject-­predicate configuration, i.e. ­DP-­IP. In terms of government (Rizzi and Savoia 1993), phonosyntactic rules are prevented from applying in this configuration, since none of the heads embedded under DP governs any of the heads embedded under I (though the N head and the V head may be adjacent). However, the notion of government introduced by Vergnaud (2008[1977]) is eliminated by the minimalist model in favor of more elementary relations. One is feature checking, that connect a superordinate head (the probe) to a lower constituent (the goal). This takes care of the role played by heads in government. The other crucial component of government is locality. Under the classical definition of government a superordinate head X had in its government domain its sister node or a daughter node of its sister. In other words a head X embedding a complement YP has in its government domain the Spec of Y but not the complement of Y. This locality part of the definition of government is taken charge of in minimalist theory by the notion of phase. Indeed, as discussed in some detail in section 1.2, several theorists have studied the relevance of the notion of phase to phonosyntactic phenomena. An approach to the fine variation in /u/ propagation along the lines of (31) and (35) does not exclude that more general constraints will apply, such as the Phase Impenetrability Condition. Indeed phases and the PIC provide an immediate explanation for the impossibility of propagation between DP and IP. In particular if DP is phase, the NP complement of D will have already been sent to ­Spell-­Out when it merges with VP/TP, making it inaccessible, among others, to phonosyntactic rules.2 The interesting question is whether phasal organization

2 Prosodic conditions prevent us from checking the context ­V-­DP, since following the ­V-­final ­inflection we need a stressed syllable (not provided by D). Therefore we have resorted to contexts, where the verb is followed by a D ­ -­less mass noun. It is possible that the optionality in propagation in Cerchiara reflects the possibility of computing or not a ­DP-­phase, based on the presence (or absence) of an empty D. We leave the entire matter open.



Which syntactic information gets interpreted by phonosyntactic processes? 

 243

may be sufficient to define contexts of impossible application in general, as well as those of optional application in Saracena. We may interpret the systematic optionality that characterizes speakers of this variety as the coexistence of two different grammars. Saracena speakers thus entertain a more restrictive grammar where only the contexts of application discussed in section 2.2 are allowed—and a grammar more similar to Cerchiara, where /u/ propagation is potentially excluded only in the presence of intervening phases. This latter generalization may indeed be sufficient to account for Cerchiara. However we note that there is at least one context of application in Cerchiara which does not allow /u/ spreading in Saracena at all, namely ­Neg-­V. No phase presumably intervenes between the Neg clitic head and the V head in T, predicting that propagation will be possible, indeed as in Cerchiara. Therefore we conclude that an account for the contexts of optional and impossible application in Saracena requires looking into the finer details of ­syntactico-­semantic structure.

2.3 Syntactic contexts of optional application We are now faced with the question whether the parametrization schema emerging from (31) and (35) is sufficient to capture the contexts where /u/ propagation applies optionally in Saracena. Within the DP, Saracena displays optionality in the ­Q-­N context. From the point of view of syntactic constituency the Q ­ -­N environment appears to be the same as the D/­A-­N one, as shown in (36). Nevertheless the agreement structure may differ; Q elements may bear a nominal inflection agreeing with the noun, but they may equally not agree, as with the numeral in (36). Furthermore the logical relations vary. Q elements introduce a relation between the N predicate (their restrictor) and the main sentential predicate (their scope). The context for propagation can therefore be described as one in which the trigger (the Q element) is restricted by the target (the embedded noun or adjective). This logical relation allows propagation in Saracena, but does not require it. (36) Saracena

Q t∫indǝ



U

N N pɔ:n/pæ:nλx U

N ǝx

244 

 M. Rita Manzini and Leonardo M. Savoia

In turn, a standard way to construe the sentential negation is as a negative quantification over the event or situation. Thus ‘I didn’t see’ means roughly ‘There is no situation/event of me seeing at (past) time t’. In other words, the IP/VP provides the scope of the event variable introduced by the quantifier. Given the similarities with the ­Q-­N contexts in (36), we may wonder why propagation in the context ­Neg-­I/V doesn’t have some degree of optionality in Saracena, but it is instead impossible. One difference that comes to mind is that Q elements agree with their restrictor N, while Neg does not agree with the verb. It is true that in (36) a ­non-­agreeing Q can trigger propagation on N; nevertheless the ­phi-­features content of N is crucial in licencing the quantification, which requires the presence of [number]. We therefore propose that the context of application relevant for (36) is as in (37). (37) The Trigger quantifies over the Target, where phi-content is involved in the (quantificational) relation Note that in (31) and (35), either the trigger or the target (or both) are also ­characterized by ­phi-­features content. If the trigger has ­phi-­features content, then this is being spread on the target—however the trigger may also spread segmental content of a lexical base onto an item lexicalizing ­phi-­features as for instance with enclitics. This leads us to suspect that in the variety of Saracena the presence of ­phi-­content is always necessary for a syntactic relation to be externalized by /u/ spreading. Within the sentential domain, optionality is associated with the context ­Mod-­V or ­Caus-­V. Both environments involve behaviors characterizing them as complex predicates of sorts, rather than as a sequence of two predicates. Let us consider for instance the M ­ od-­V environment with ‘want’ in (39), corresponding to a core instance of ‘restructuring’ in the sense of Rizzi (1982). The syntactic structuring of the expression is not very different from that of an A ­ ux-­V context. In fact, for Manzini and Roussou (2000) control is an operation directly binding a ­theta-­role, i.e. an argument slot; in other words, no PRO is present. Nevertheless the two structures are far from identical. In A ­ ux-­V environments, the argument structure of the two verbs is shared and there is a single event interpretation. The same is not true of restructuring environments, where an expression like (38), ‘we want to do’, consists of two individuated states/events (a mental state of ‘wanting’ and an event ‘doing’) and the matrix EPP argument simply controls the embedded empty EPP argument. A necessary precondition for restructuring appears to be the lack of independent temporal specifications on the two verbs, in other words the sharing of temporal reference (Wurmbrand 2001). The optionality of /u/ propagation in (38) may then be sensitive to the fact that



Which syntactic information gets interpreted by phonosyntactic processes? 

 245

the EPP argument of Mod (its trigger) is not a direct argument of V, but controls one of the argument slots of V (its target). (38) Saracena

V V ẞulλz



VP D imǝz=x U

>

V fɔ/fɛλx, λy U

Causative contexts, e.g. ‘make(3pl) do’ in (39), are similar to restructuring ones, in that they there are well individuated causation and caused events—though the embedded event lacks independent temporal structure. Therefore, a ­Caus-­V structure does not involve unification of argument structures of the type found in ­Aux-­V environments. However, in causative contexts there is no control from the matrix EPP argument to the embedded one either. Rather all arguments are independently lexicalized; in the absence of a temporal structure, the embedded subject (the causee) is assigned either accusative (if the embedded verb is intransitive)—or else an oblique case (dative). We conclude that the absence of intermediate temporal structures (i.e. of a CP phase) is sufficient to determine that the matrix and embedded V are computed as a complex predicate. Hence the matrix inflection, as an exponent of the EPP argument, can trigger /u/ propagation optionally on the lower V member of the complex predicate (its target). (39) Saracena



[fɔ:λz –nǝz] U >

[fɔ λx, λy U

In general, in Mod/­Caus-­V contexts, the trigger of the spreading, i.e. the finite inflection of Mod/Caus is not an argument of its target, i.e. the lexical V—nor do Mod/ Caus and V denote a single event, in the way of ­Aux-­V contexts. In other words the conditions in (31) and (35) do not hold and this means that propagation is not obligatory. At the same time restructuring/causativization imply a shared higher tense domain between the two predicates—equivalently the absence of a ­C-­T phase associated with the embedded predicate. This licences a weak relation between the matrix

246 

 M. Rita Manzini and Leonardo M. Savoia

finite inflection and the embedded verb, belonging to the same complex event predicate, sufficient for optional application of propagation, along the lines of (40).3 (40) The Trigger is an argument of the Target via Tns identification. Let us go back to one of the contexts that we examined first, namely C ­ l-­V. One of the empirical correlates of restructuring and of causativization is clitic climbing, whereby a clitic associated with the modal/causative verb is interpreted as an argument of the embedded lexical verb. By general consent, the structural representation resulting from clitic climbing is like that of an ordinary ­Cl-­V pair, where Cl is the argument of lexical V. However spreading is not obligatory in ­Cl-­Modal/ Causative contexts, unlike contexts C ­ l-­V, where V is lexical. With restructuring verbs in (41) optionality prevails, though propagation may also not apply at all, in particular with motion verbs in (­41a-­a’). With causative verbs in (42), ­propagation applies again optionally. (41) a. [u/a   ˈvɛ:və a ccaˈmɛ] him/her   I.go    to call ‘I’ll go and call him/her’ a’. [u/a  vɛj  a fˈfɛ] it.m.f  I.go to do ‘I’ll go and to do it’ b. [u ˈsɛ:pə/ˈsɔ:pə fˈfɛ] it   s/he.knows   to.do ‘S/he knows how to do it’ c. [u/a    stɛ    faˈtʃinnə] it.m/it.f s/he.is doing ‘S/he is doing it’ c’. [u ˈstɛ:jə/ˈstɔ:jə faˈtʃinnə] it  you.are    doing ‘You are doing it’ (42) a. [u   ˈfɔ:nə/ˈfɛ:nə ˈdɔrmə] him  they.make   to.sleep ‘They make him sleep’

3 Limitations of space prevent us from dealing in detail with every single context of application. We note that ­N-­A and C ­ opula-­A pattern with Caus/­Mod-­V contexts in being optional in Saracena. This suggests that ­Copula-­A is a raising structure (‘they are healthy’ from ‘are [they healthy]’, along the lines of Moro 1997). Something similar may be true of predicative adjectives, cf. ‘a son (who is) healthy’.



Which syntactic information gets interpreted by phonosyntactic processes? 

b. [n u  ˈfɔ:jə/ˈfɛ:jə ˈfɛ] him it you.make     do ‘You make him do it’ 

 247

Saracena

In ­Cl-­Mod/Caus contexts of the type in (41)–(42) the clitic is not a direct argument of the restructuring/causative verb, unlike in (31), and we correctly exclude obligatory application of propagation. At the same time, in the discussion surrounding (40) we suggested that the temporal structures of the two predicates are identified in Mod/­Caus-­V contexts. One of the correlates of this (or in other words of the lack of an independent temporal structure associated with the embedded predicate) is clitic climbing—i.e. the realization of all clitic material in the sole inflectional domain available, that of the Mod/Caus verb. Under the theory of cliticization of Sportiche (1996), Manzini and Savoia (2005, 2007), clitics are in fact directly merged in a dedicated functional domain. In this sense C ­ l-­Mod/Caus appears to instantiate one of those weaker relations between the trigger of propagation (Cl) and its target Mod/Caus that licences only optional application—as indeed predicted by (40). One may also wonder why within the restructuring set, some verbs do not appear to allow for propagation at all. It is tempting to correlate this with the presence of a prepositional introducer a for the embedded sentence, indicating the presence of P/C phase above the embedded predicate. Kayne (1989) argues that restructuring is a label for a set of processes that do not necessarily pattern together. Just one example from standard Italian is provided in (43). Both ‘want’ and ‘try’ licence clitic climbing. However ‘want’ in (43a) is compatible both with the auxiliary ‘have’ that it selects and with the auxiliary ‘be’ selected by the embedded unaccusative verb. On the contrary ‘try’ in (43b) is not compatible with the ‘be’ auxiliary of the embedded verb. (43) a. Abbiamo voluto/siamo voluti sparire we.have wanted/we.are wanted to.disappear ‘We wanted to disappear’ b. Abbiamo provato/*siamo provati a sparire we.have tried/we.are tried to disappear ‘We tried to disappear’ Following Kayne, we suggest that the different acceptability of propagation in ­ l-­Mod/Asp contexts like (41) is part of a larger set of phenomena, whereby C predicates allowing clitic climbing may or may not allow restructuring proper. Indeed motion verbs maintain their ‘be’ auxiliary independently of the nature of the embedded verb. If we factor away clitic climbing and assume the presence

248 

 M. Rita Manzini and Leonardo M. Savoia

of a C-T phase boundary between the two predicates in (­41a-­a’), the conditions ­required in (40) for spreading from the clitic to the verb do not hold, predicting that spreading is impossible. The data in (41)–(42) are crucial in rejecting some conceivable simplifications of the schema of parametrization that is emerging from (31), (35), (37), and (40). In particular one may consider that optional application in Sara‑ cena is observed whenever trigger or target are endowed with ­phi-­features, as in (37), and the general condition holds that they are not separated by any phase. For instance in (41)–(42), the clitic is not an argument of the modal/ causative verb, excluding obligatory application. However, ­phi-­features are the trigger and there are no intervening phases, which would predict optionality under the alternative account being considered. Yet this is not sufficient to distinguish between (­41a-­a’) and the other examples in (41)–(42). In order to do so, we need reference to the fact that the target of spreading is or is not part of a complex predicate in the sense defined by (40) (lack of independent tense structure associated with the embedded event). Therefore (40) is not dispensable.

2.4 Intermediate conclusions Summarizing so far the account of variation in /u/ propagation that emerges from the partial schemas in (31), (35), (37), (40) is as outlined in (44). (44) ­argument-­of

+Stigl, +Sar, +Cerch

(31)



+Sar, +Cerch,

(35)

­argument-­of via Tns unification

±Sar, +Cerch

(40)



±Sar, +Cerch

(37)

­argument-­of via theta/event identification

quantifier over

The schema of parametrization in (44) implies that propagation is sensitive to the nature of the relation between the trigger and the target, as defined not only by broad locality notions such as phase or broad operations such as feature checking. Satisfaction of an argument predicate structure is a core environment for propagation. Other relations such as ­ operator-­ restrictor ones introduce a degree of optionality. Variation is determined by whether the argument-of relation is direct, or resulting from the deficiency of event or



Which syntactic information gets interpreted by phonosyntactic processes? 

 249

temporal structures, yielding complex predicates with different degrees of internal cohesion. It is also sensitive to whether ­phi-­feature content is being propagated (or target by propagation)—or no such content is present as in ­Neg-­V environments. Therefore segmental harmony processes appear to read directly not only the syntactic tree but also the content of its nodes and the interpretive relations they enter into. The parameters for the application of spreading proposed by Rizzi and Savoia (1993) are compatible with the idea that PF and in particular phonosyntactic processes, only read the abstract node organization of syntax (i.e. sisterhood, dominance and derived notions such as c­ -­command and government), with no access to the LF content of these nodes or to the relations that the node entertain at LF. Hence variation is obtained essentially by refining structural relations, so that government yields ­F-­government, ­M-­government, ­Agr-­Government. Leaving aside empirical matters, this means that the configurations in (44) are not unified under government, but simply encoded into the different notions which government comes to label. Going on to the minimalist framework, we have already noted that ­argument-­of contexts like ­D-­N, cf. (31), can be mimicked by ­phi-­feature checking. Contexts like ­A-­N, cf. (35), can be modelled by something like Multiple Agree— i.e. though A and N do not check one another, both check D. However some of the relations that we saw to be relevant for optional propagation, for instance Q/­Neg-­restrictor/scope do not involve ­phi-­feature checking at all. In ­Caus-­V contexts, the ­phi-­features of Caus are checked by an argument (the causer) not originating in the structure of V at all and hence incapable of checking V, albeit abstractly; a similar argument was made in section 2.2 for the ­Aux-­V context. We are then left with the possibility, notably suggested by D’Alessandro and Scheer (2015), that the notion of phase be refined to capture fine variation in phonosyntactic environments. When such a possibility is applied to the interlinguistic variation depicted in (13), differences in phase organization would mean that for example A ­ -­N and A ­ ux-­V/v are not separated by phases in the Saracena and Cerchiara varieties but they are in the Stigliano variety, that no phase intervenes in ­Neg-­V in Cerchiara though it does in Saracena and Stigliano, and so on, in spite of the overall closeness of these dialects. As for the parameter regarding the PIC it seems to us that a restrictiveness problem arises. Upon noting that a certain configuration blocks phonosyntactic rules, one can postulate a phase head; if the syntax does not seem to detect it, we can say that the PIC is not active in the syntax for that phase head. Vice versa, if we need a head to be a phase in the syntax, without there being any detectable effect at PF, we can say that the PIC is not active in the phonology.

250 

 M. Rita Manzini and Leonardo M. Savoia

By contrast, here we have stuck with the restrictive notion of phase (always associated with a PIC) proposed by Chomsky—and with the restrictive r­ epertory of phase heads justified by locality conditions on movement and Agree (C, v, D). In our view the phasal domains so defined set the limit conditions for the application of phonosyntactic processes (as of syntactic processes). Within this domain parameters may apply that are not themselves defined in terms of phases, as in (45).

3 P  reservation of word final -a and the nature of variation The fine grained variation associated with /u/ propagation summarized in the table in (13) ought to be replicable—either with different languages or with different phenomena in the same languages. Recall that one of the core environments for /u/ propagation (obligatory in all varieties considered) is masculine (singular) agreement between a clitic/determiner and a noun/verb, resulting in realization of the [U] phonological primitive associated with masculine (singular) on the predicative base. In what follows we shall briefly examine a somewhat complementary phenomenon, namely the preservation of word final -a in syntactically conditioned environments, whereas in the absence of such conditions, final unstressed nuclei, including /a/ itself, are weakened to [ə] in the relevant varieties (cf. Rohlfs [1949] 1966: 177). In reality the preservation of final unstressed vowels in phonosyntactic environments may concern /u/ as well; thus in Saracena and Cerchiara, final [u] is found only in the contexts in which propagation is also found. We know from Stigliano that the preservation of /u/ is independent of propagation, since the latter can equally well be triggered in the presence of a final [ə]. Furthermore even in Saracena and Cerchiara the preservation of /u/ only characterizes a subset of triggers for propagation. Specifically the data in (10)–(12) show that final /u/ is never preserved on lexical bases (adjectives, nouns, lexical verbs). In the three dialects under examination, however, the preservation of final /a/ is quite systematic, as shown by the data in (45)–(47). In nominal contexts, [a] is normally preserved as the final inflection for feminine (singular), as illustrated in the (a) examples. ­Non-­etymological final /a/ is also attested, specifically with quantifiers, cf. for instance the ­Q-­N context in Stigliano in (37a). In verbal contexts, Cerchiara and Saracena preserve final /a/ only on clitics; Stigliano preserves it on a set of verbal constituents which include auxiliaries,



Which syntactic information gets interpreted by phonosyntactic processes? 

 251

modals and causatives as well as lexical verbs, as illustrated by the (b) examples. As expected, ­DP-­V(P) environments block the application of the process, cf. the (c) examples. (45) a. ­D-­N ­ D-­A ­A-­N ­Q-­A D ­N-­A ­Q-­N

+ [ˈkɛdda: ˈfɛmənə] ‘that woman’ + [na bˈbɛlla: ˈfIɈɈə] ‘a pretty girl’ + [na bˈbona: ˈfɛmənə] ‘a good woman’ + [ˈtutta: ˈno:və] ‘all new’ + [na ˈfIɈɈa: bˈbra:və] ‘a girl good’ + [ˈpIkka: ˈmɔskə/ˈlɪbbrə/ˈlattə] ‘few flies(f)/books(m)/ milk(m)’ [ˈkwanta: ˈfɛmənə/ˈlɪbbrə/ˈlattə] ‘how many women/ books/milk’ b. ­Cl-­V + [m a: ˈmaɲdƷə] ‘myself it I.eat’ ­Aux-­V ± [l aˈveja:/aˈvejə: ˈmeisə] ‘it I.had put’ ­Mod-­V ± [ˈjIddə vəˈlØja:/vəˈlØjə: ˈdɔrmə] ‘lui voleva dormire’ ­Cop-­A ± [ˈjera:/ə ˈkaudə] ‘it.was hot’ ‘I.had hunger’ ­V-­DP ± [təˈnejə/təˈneja: ˈfa:mə] i.e. ‘I was hungry’ ­V-­AdvP ± [ˈmandʒa:/ˈmaɲdʒə ˈpikkə] ‘(s)he.eats little’ c. ­DP-­V − [ˈfIɈɈəmə ˈcandƷa: ˈsɛmbə] ‘my.daughter cries always’  Stigliano (46) a. ­D-­N + ­ D-­A + ­A-­N + ­N-­A + ­ Q-­N + b. ­Cl-­V + ­Aux-­V − ­Cop-­A − ­Caus-­V − ­V-­­AdvP − c. ­DP-­V −  (47) a. ­D-­N ­D-­A ­A-­N

[ˈkwɪɖɖa ˈfɪmmənə] [na bˈbɛlla ˈfɪmmənə] [kwɪɖɖa bˈbɛlla ˈfɪmmənə] [na ˈfɪmməna ˈƔavətə] [ˈpIkka ˈkarnə] [a ˈƔrɛ:pə] [ˈjerə nˈnɛ:tə] [ˈjerə gˈgautə] [mə faˈtʃi: dˈdɔrmə] [ˈmaɲdƷə pˈpikkə] [ˈ­fiɈɈə-­mə ˈdɔrmə]

+ [ˈkwilla ˈfimmənə] + [ˈkwilla bˈbɛlla ˈfimmənə] + [ˈkwilla bˈbɛlla ˈfimmənə]

‘that woman’ ‘a nice woman’ ‘that nice woman’ ‘a woman tall’ ‘little meat’ ‘it I.open’ ‘I.was born’ ‘(s)he.was tall’ ‘me (s)he.made sleep’ ‘(s)he.eats little’ ‘­daughter-­mine sleeps’ Saracena ‘that woman’ ‘that nice woman’ ‘that nice woman’

252 

 M. Rita Manzini and Leonardo M. Savoia

­N-­A ­Q-­N b. ­Cl-­V ­Aux-­V ­Caus-­V ­Cop-­A ­V-­­AdvP c. ­DP-­V 

+ + + − − − − −

[na ˈfimməna ˈƔavətə] [ˈpIkka: ˈkarnə] [a ˈavə] [aˈvi: nˈnæ:tə] [mə faˈtʃi: dˈdurmə] [ˈjɛrə gˈgavətə] [ˈdurmə pˈpikkə] [kwilla ˈfimmənə ˈdurməðə lˈlæ]

‘a woman tall’ ‘little meat’ ‘it I.wash’ ‘(s)he.had born’ ‘me (s)he.made sleep’ ‘she.was tall’ ‘she.sleeps little’ ‘that woman sleeps there’ Cerchiara

The data in (45)–(47) are tabulated in (48). If we compare (48) with the table in (13) concerning the syntactic contexts for /u/ propagation, it is evident that (13) and (48) only partially overlap in the varieties under study. As it turns out, in Stigliano final /a/ realization has a much wider domain of application than /u/ propagation; the reverse appears to be true in Saracena and Cerchiara. Therefore Saracena/­Cerchiara have fewer contexts of application for /a/ preservation than Stigliano. At the same time, contexts where the realization of final /a/ is optional Stigliano include all those where /u/ propagation is optional in Saracena. In other words, the distribution of /a/ propagation follows in this respect the parameters set out in (44). Furthermore in Saracena and Cerchiara, final /a/ realization applies only within the DP/AP and on clitics, i.e. when /a/ externalizes nominal class (gender). The possibility of such a parameter is implied by reference to ­phi-­features in the table in (44). (48) a. ­D-­N/A ­A-­N ­N-­A ­Q-­N/A b. C ­ l-­V Aux -V ­Copula-­A Mod/­Caus-­V ­V-­DP c. ­DP-­V

Cerchiara/Saracena + + + + + − − − − −

Stigliano + + + ± + ± ± ± ± −

From a phonological point of view, we may treat /a/ as an underlying form which in phonologically and syntactically appropriate contexts is not weakened to [ə] by the rule that otherwise neutralizes unstressed final vowels. What is directly relevant for present purposes are the syntactic conditions on preservation. Relevant data are discussed for the Naples variety by Bafile (1997) and by Ledgeway (2009). Bafile (1997) proposes that the relevant syntactic context is



Which syntactic information gets interpreted by phonosyntactic processes? 

 253

determiner—lexical head, while for Ledgeway (2009) the preservation of -a/-u corresponds to the lexical properties of a ‘restricted number of items’. When these items form a constituent with the following word, the weakening of the vowel to [ə] may not take place, yielding the realization of the full vowel as a result.4 In terms of the parameters developed in (44), Cerchiara and Saracena preserve /a/ only in phonosyntactic environments where /a/ externalizes nominal class/gender. This parameter of variation is easily added to the schema in (44). Furthermore, as already noted, Stigliano has optional preservation of /a/ in all contexts where propagation of /u/ is optional in Saracena, and in the additional ­Aux-­V context. In other words, the fundamental split between an N inflection trigger and others is observed in Stigliano as well, where obligatoriness is reserved for it, thus excluding ­Aux-­V. These conditions are summarized in the table in (49), which both confirms the parameters of table (44) and further refines them. (49) argument of

N class/gender +Cerch/Sar, +Stigl

other ­phi-­features



argument of via theta/event identification +Cerch/Sar, +Stigl

±Stigl



argument of (via other) quantifier of

±Stigl

+Cerch/Sar

Putting the summary of /u/ propagation in (44) together with that of /a/ preservation in (49), we conclude that the realization of final /a/ and the spreading of /u/ jointly insure the externalization of nominal morphology in the key a ­ rgument-­of contexts (­D-­N and ­Cl-­V) in all varieties. Thus as pointed out by Bafile (1997), the realization of final vowels has obvious relevance from a morphological point of view, namely that of “saving from weakening and preserving a minimum of segmental content ([A], [I] o [U]) in vowels corresponding to the endings of certain morpholexical items”. At the same time, in Stigliano the preservation of final /a/ covers contexts where /u/ propagation is not observed. Vice versa, in 4  Bafile (1997) connects the realization of final vowels in phonosyntactic contexts in Neapolitan to a general treatment of unstressed nuclei. Bafile observes that in ­pre-­tonic nuclei, back vowels have a raised outcome [u]; front vowels are realized as [ə] or variably as [i]; in p ­ ost-­tonic position the neutralized outcome [ə] is normally attested. [a] is found both ­pre-­and ­post-­tonically. She argues that the context in which a final unstressed nucleus is not weakened to [ə] corresponds to the phonological phrase; in this context the final unstressed vowel is in a ­pre-­tonic position with respect to the stressed nucleus which heads the prosodic constituent. In other words, a prosodic structure is created analogous to the ­word-­internal structure where ­pre-­tonic nuclei do not neutralize to [ə] but are at least partially preserved.

254 

 M. Rita Manzini and Leonardo M. Savoia

Saracena and Cerchiara obligatory /u/ propagation is found where /a/ preservation is absent. In short, different phonosyntactic processes (here /u/ propagation and final /a/ realization) are sensitive to different features and morphosyntactic relations, within the same basic set. The comparison between the tables in (44) and (49) raises some fundamental questions as to the nature of parametrization. If we associate the parametric values in (44) and (49) with the morphemes /a/ and /u/ in the lexicon, this implies using the lexicon improperly as a repository for computational ­instructions/ constraints—essentially the same problem that we raised for D’Alessandro and Scheer’s (2015) proposal in section 1.2. Incidentally the ­likelihood that parametrization of phases and the PIC is involved in the phonosyntactic phenomena of /a/ preservation and /u/ spreading seems to us further diminished by the need to associate it with the single morphemes /a/ and /u/. We believe that the core insight expressed by the lexical parametrization hypothesis is best reformulated in relation to the crucial role that the lexicon plays in the operation(s) of externalization. We may construe externalization essentially as an instruction to pair some ­sensory-­motor content to some ­syntactico-­semantic content, along the lines of (50). Most of externalization is carried out by the lexicon, which we may think of as a procedure to cut up the conceptual universe into linguistic units. Parameters naturally arise in the course of lexicalization, corresponding to whether certain conceptual properties are or are not lexicalized, and in which way. However if we take the suggestion of Chomky (2001) that linear ordering is also a property of externalization, precedence would also be determined by externalization (dominance relations remaining unchanged). This leads to a less restrictive conjecture about parameters, tying them to externalization, as in (51), of which lexicalization forms crucial part. (50) Externalize Pair phonological information x with ­syntactico-­semantic information y (51) Parametrization Parameters arise in externalization We suggest that the phenomena schematized in (44) and (49) should be seen in the perspective suggested by (50)–(51), as externalizations of morphosyntax by phonology, and as such subject to linguistic variation. We attribute to propagation the status of other harmony processes, whereby a property perceptually vulnerable and in any event relevant for the purposes of recognizability gets extended over a phonological domain, with the result therefore of being amplified and maximized (Kaun 1995, 2004; Walker 2005; Nevins 2010). Kaun (1995) observes that harmonies increase the exposition of the hearer to morphophonological elements



Which syntactic information gets interpreted by phonosyntactic processes? 

 255

characterized by vulnerable phonological properties (Kaun 1995; Walker 2005): “harmony serves to extend the duration of phonetic information which is phonologically important (i.e. distinctive), but which is transmitted by means of relatively subtle acoustic cues” (Kaun 1995, vii). In other words, harmony processes have the effect of extending over a temporal sequence a vulnerable quality, not easily identifiable, increasing the exposure of the hearer to this phonological quality, preserving and increasing its perceptibility. Some of the conclusions of Kaun agree with the restrictions observed in section 2.1. Kaun (2004) observes that rounding is a subtle property from the articulatory and acoustic point of view, at least in the sense that it is sensitive to other properties, so that it is maximally recognizable when associated to back/high vowels. For instance round harmonies tend to avoid front vowels and low vowels, explaining in particular the fact that in the varieties studied here rounding does not normally apply to front vowels. The phenomena that we are concerned with are to be seen as phonological constructs that externalize syntactic informations, namely those concerning inflectional/agreement properties. Parametrization concerns the logical relations being externalized.

4 Conclusion From a phonological point of view, we have studied a harmonic process triggered by the element [U], which according to Kaun (1995, 2004) has a special status in harmonic processes because of its perceptual weakness. Indeed harmony processes have the effect of extending some phonological properties on several positions within a prosodic domain. In present terms a [U] element must be present on the stressed nucleus head of the prosodic domain in order to licence [U] on a ­pre-­tonic nucleus. This leads us to the major stream of our discussion, involving the syntactic conditions on /u/ propagation. As a consequence of these conditions, the net effect of /u/ propagation is that of multiplying acoustic cues leading to the recognition of certain morphosyntactic categories, specifically nominal class/gender and more generally the p ­ hi-­feature complex. This is consistent with the conception of the SM interpretation component as externalization of the syntax within the general framework of the minimalist program (Chomsky 1995) and more specifically of the biolinguistic program (Berwick and Chomsky 2011). Beyond this common core, fine variation is observed across different varieties (and within the same variety) according to which features must or can be involved and to a scale of ­syntactico-­semantic relations of growing generality.

256 

 M. Rita Manzini and Leonardo M. Savoia

Defining the relevant relations has required us to delve in some depth into syntactic analysis—with a view however to addressing the issue that gives the title to this work, i.e. what is externalized exactly at the SM interface. Here we have argued that SM does not simply see syntactic nodes and structural relations among them (as in Rizzi and Savoia 1993). The structural notion of feature checking (e.g. Elordieta 1999, 2008) comes much closer to capturing the relevant environments. In particular given appropriately abstract syntactic structures, and an appropriately abstract notion of feature checking, all of the environments of obligatory and optional application could be conceivably accounted for in terms of feature checking. As the last point of the present work, however, we aimed to push the logically strongest thesis, namely that externalization sees not only morphosyntactic information but has full access to the interpretive properties read by the ­Conceptual-­Intentional (CI) interface. Specifically, following Manzini and Savoia (2005, 2007, 2011), p ­ hi-­feature inflections are construed as elementary ­arguments. Agreement does not delete uninterpretable features (feature checking) but rather matches sequences of locally identical features creating a single discontinuous referent. In this perspective, parameters on /u/ propagations are defined by whether the trigger directly saturates the target or rather the trigger and the target agree in the sense just defined, or some other relation holds between them (e.g. operator, lexical restrictor). The same conditions are active on the phonosyntactic process of /a/ preservation in final position. There is no evidence for specialized buffer components mediating the relation between syntactic structure and SM interpretation. If our stronger thesis is on the right track, such buffer structures are excluded, since intervening between the SM interface and the ­syntactico-­semantic (CI) one, their effect would be to prevent full access of ­Spell-­out (externalization) to CI.

References Backley, Philip. 2011. An introduction to element theory. Edinburgh: Edinburgh University Press. Bafile, Laura. 1997. L’innalzamento vocalico in napoletano: un caso di interazione fra f­ onologia e morfologia. In: Agostiniani, Luciano (ed.), Atti del Terzo Convegno della Società Internazionale di Linguistica e Filologia Italiana, Perugia 1994, 127–148. Napoli: Edizioni ­Scientifiche Italiane. Berwick, Robert and Chomsky, Noam. 2011. The biolinguistic program: the current state of its evolution and development. In Anna Maria Di Sciullo and Cedric Boeckx (eds.), The biolinguistic enterprise, 19–41. Oxford: Oxford University Press. Burzio, Luigi. 1986. Italian syntax. Dordrecht: Kluwer.



Which syntactic information gets interpreted by phonosyntactic processes? 

 257

Cheng, Lisa L.-S. and Laura J. Downing. 2009. Prosodic domains do not match spell-out domains. McGill Working Papers in Linguistics 22 (1): 1–14. Chomsky, Noam. 1995. The minimalist program. Cambridge: The MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: a life in language, 1–­52. Cambridge: The MIT Press. Chomsky, Noam. 2007. Approaching UG from below. In Ulli Sauerland and H. M. Gartner (eds.), Interfaces + Recursion = Language? Chomsky’s minimalism and the view from ­syntax-­semantics, ­1–­29. New York: Mouton De Gruyter. D’Alessandro, Roberta and Tobias Scheer. 2015. Modular PIC. Leiden University – University of Nice, msc. Published as 2015. Modular PIC. Linguistic Inquiry 46: 593–624. Elordieta, Gorka. 1997. Morphosyntactic feature chains and phonological domains, PhD thesis. Los Angeles: University of Southern California. Elordieta, Gorka. 2008. An overview of theories of the ­syntax-­phonology interface. Anuario del Seminario de Filologia Vasca/ International Journal of Basque Linguistics XLII: ­209–­286. Franco, Ludovico, Manzini, M. Rita and Leonardo Savoia. 2014. N class and its interpretation: the neuter in Central Italian varieties and its implications. Paper delivered at CIDSM 8 Padua and Venice, ­20–­22 June, 2014. Halle, Morris and Idsardi, William. 1995. General properties of stress and metrical structure. In Goldsmith John (ed.), The handbook of phonological theory, 403–443. Oxford: Blackwell. Halle, Morris and Vergnaud, ­Jean-­Roger. 1987. Stress and cycle. Linguistic inquiry 18: ­45–­84. Harris, John and Lindsey, Geoff. 1995. The elements of phonological representation. In Jacques Durand and Francis Katamba (eds.), Frontiers of phonology, ­34–­79. London: Longman. Harris, John and Lindsey, Geoff. 2000. Vowel patterns in mind and sound. In Noel BurtonRoberts, Philip Carr and Gerard Docherty (eds.), Phonological knowledge. Conceptual and empirical issues, ­185–­205. Oxford: Oxford University Press. Higginbotham, James. 1985. On semantics, Linguistic Inquiry 16: ­547–­593. Kaun, Abigail. 1995. The typology of rounding harmony. An Optimality Theoretic approach, PhD dissertation, UCLA. Kaun, Abigail. 2004. The typology of rounding harmony. In Bruce Hayes, Robert Kirchner and Donca Steriade (eds.), Phonetically based phonology, 87–­116. Cambridge: Cambridge University Press. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud. 1990. Constituent structure and government in phonology. Phonology 7:­ 293–­231. Kayne, Richard. 1989. Null subjects and clitic climbing. In Osvaldo Jaeggli and Ken Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer. Kiparsky, Paul. 1982a. Lexical phonology and morphology. In I. S. Yang, (ed.), Linguistics in the Morning Calm, ­3–­91. Seoul: Hanshin. Kiparsky, Paul. 1982b. From cyclic phonology to lexical phonology. In H. van der Hulst and N. Smith (eds.), The structure of phonological representations (Part I), 131–­175. Dordrecht: Foris. Kratzer, Angelika and Elizabeth Selkirk. 2007. Phase theory and prosodic spellout: the case of verbs. The Linguistic Review 24: ­93–­135. Ledgeway, Adam. 2009. Grammatica diacronica del napoletano, Tübingen: Niemeyer. Manzini, M. Rita. 1983. Syntactic conditions on phonological rules, ­1–­9. MIT Working Papers in Linguistics 5, Cambridge, Mass. Manzini, M. Rita and Anna Roussou. 2000. A minimalist theory of ­A-­movement and control. Lingua 110: ­409–­447.

258 

 M. Rita Manzini and Leonardo M. Savoia

Manzini, M. Rita and Leonardo M. Savoia 2005. I dialetti italiani e romanci. Morfosintassi generativa, 3 voll. Alessandria: Edizioni dell’Orso. Manzini, M. Rita and Leonardo M. Savoia 2007. A unification of morphology and syntax. London: Routledge. Manzini, M. Rita and Leonardo M. Savoia 2011. Grammatical categories. Cambridge: Cambridge University Press. Moro, Andrea. 1997. The raising of predicates. Cambridge: Cambridge University Press. Nespor, Marina and Irene Vogel 1986. Prosodic Phonology. Foris: Dodrecht. Nevins, Andrew. 2010. Locality in vowel harmony. Cambridge: The MIT Press. Newell, Heather and Glyne Piggott. 2014. Interactions at the ­syntax–phonology interface: Evidence from Ojibwe. Lingua 150: ­332–362. Rizzi, Luigi and Leonardo M. Savoia 1993. Conditions on /u/ propagation in Southern Italian dialects. In Adriana Belletti (ed.), Syntactic theory and the dialects of Italy, 252–­318. Torino: Rosenberg & Sellier.­ Roberts, Ian. 2005. Principles and parameters in VSO languages. Oxford: Oxford University Press. Rohlfs, Gerhard. 1966[1949]. Grammatica storica della lingua italiana e dei suoi dialetti. Fonetica. Torino: Einaudi. Savoia, Leonardo M. 1987. Teoria generativa, modelli fonologici e dialettologia. La propagazione di u in una varietà lucana. Rivista Italiana di Dialettologia XI: ­185–263. Savoia, Leonardo M. 2015. I dialetti italiani. Sistemi e processi fonologici nelle varietà italiane e romance, Pisa: Pacini. Savoia, Leonardo M. (this volume). Harmonic processes in some Italian varieties. ­ eripheries: Svenonius, Peter. 2004. On the edge. In D. Adger, C. de Cat and G. Tsoulas (eds.), P syntactic edges and their effects, ­261–287. Kluwer: Dordrecht. Sportiche, Dominique. 1996. Clitic constructions. In Johan Rooryck and Laurie Zaring (eds.), Phrase structure and the lexicon, 213–­276. Dordrecht: Kluwer­. Tuttle, Edward. 1985. Assimilazione ‘permansiva’ negli esiti ­centro-­meridionali di A tonica. L’Italia Dialettale 48: ­1–­34. Vergnaud, ­Jean-­Roger. 2008. Letter to Noam Chomsky and Howard Lasnik on “Filters and Control”, April 17, 1977. In Robert Freidin, Carlos Otero and Maria Luisa Zubizarreta (eds.), Foundational issues in linguistic theory, 3–­15. Cambridge: The MIT Press. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural language & linguistic theory 23: ­917–­989. Wurmbrand, Susi. 2001. Infinitives: restructuring and clause structure. Berlin/New York: Mouton de Gruyter.

Joan Mascaró

On the typology of metaphony/stress dependent harmony1

Universitat Autònoma de Barcelona

Abstract: The analysis of the stress-dependent harmony systems of Calvello and Servigliano shows that a set of basic typological effects derive from the typological distinction between phonological harmony and morphemic harmony, a distinction determined by the structure of harmonic triggers. Triggers can be morphemes consisting of one or several of features, they can consist of a pure phonological element, or they can be mixed, consisting of both of segmental and floating material. Preservation of underlying contrasts is shown to play a crucial role both in phonological and in morphemic harmony. Several problems and questions that are important for future research are also discussed.

1 Phonological and morphemic harmony The term metaphony is a convenient phonological label that covers a set of quite diverse harmonic/assimilatory phenomena typical of a group of Romance languages. It was coined as a translation of the German Umlaut and it is applied to the raising or diphthongization of stressed vowels triggered by a final high vowel. It is different from canonical cases of vowel harmony because it is stress dependent and, for a subset of cases, because it does not involve strict assimilation but diphthongization. I will consider metaphony as a special case of stress dependent harmony, a term borrowed from Majors (2000), that covers assimilatory phenomena that are triggered or targeted by the stressed position. Several typological classifications of metaphony have been proposed, mostly based on the nature of triggering of vowels, the nature of targeted stressed vowels, the metaphonic change, and morphological conditions (see Maiden 1987, 1991). In this article I propose that a set of basic typological effects derive from the distinction between phonological harmony and morphemic harmony (Finley 2009; Cole 1991 [1987]) and that contrast preservation plays a crucial role in stress dependent harmony. In this section I introduce these two types of harmony. Section 2 examines ­metaphony and morphological gemination in Calvello, a dialect with morphemic

1 This work has benefited from grant FFI2013-46987-C3-2-P from the Spanish Ministry of ­Economy and Competitivity.

260 

 Joan Mascaró

harmony. In section 3 I introduce contrast preservation in the analysis of Calvello, and in section 4 I illustrate phonological harmony with another Italo-Romance dialect, Servigliano. Section 5 summarizes the results and discusses unsolved problems and prospects. In phonological harmony the trigger is a segmental phonological element, as in Turkish vowel harmony (Kabak 2011). Morphemic harmony is an extension of featural affixation (Akinlabi 1996, 2011). A featural affix consists of a floating feature (or a set of floating features). Consider the completive/incompletive alternations in Kanembu (Nilo-Saharan, Chad; Roberts 1994; Akinlabi 1994; Finley 2009). In the completive the vowels can be {a, ə, ɛ, ɔ, ɪ, ʊ}, in the incompletive they become {ɜ, e, o, i, u}:2 (1) ɡɔ́nə̀kɪ̀ dállə̀kɪ̀ bàrɛ́nə́kɪ̀

‘I took’ ‘I got up’ ‘I cultivated’

ɡónɜ̀kì dɜ́llɜ̀kì bɜ̀rénɜ́kì

‘I am taking’ ‘I am getting up’ ‘I am cultivating’

The incompletive morph is not an affix consisting of segmental material, but a floating [+ATR] feature that is linked by a phonological operation to all the vowels of the verbal form. In Akinlabi’s (1994) and in Robert’s (1994) analysis the incompletive has a floating [+ATR] and the completive is unmarked and gets the default [−ATR] value. Within stress dependent harmony, phonological and morphemic harmony can be illustrated with the Italo-Romance dialects of Grado (Maiden 1991; Walker 2005) and Lugo (Savoia & Maiden 1997; Pelliciardi 1977), respectively. (2) Grado méto témpo rómpo fjór

míti tímpi rúmpi fjúri

‘put-1sg.prs.ind/2prs.ind’ ‘time-sg/pl’ ‘break-1sg.prs.ind/2sg.prs.ind’ ‘flower-sg/pl’

(3) spós nér móart kapǽal patáka

spús nír múrt kapɛ́l patǽaka

‘spouse-m.sg/m.pl’ ‘black-m.sg/m.pl’ ‘dead-m.sg/m.pl’ ‘hat-m.sg/m.pl’ ‘simpleton-m.sg/m.pl’

In Grado the height features of the final vowel spread on the preceding stressed vowel, as shown in (2). In Lugo (3) singular and plural are distinguished in

2 Both /a/ and /ə/ become [ɜ]. I have substituted this vowel symbol for Akinlabi’s and Robert’s [ʌ] because Jouannet, at least in Jouannet (1982), uses [ɜ] and says clearly that it is central, in addition to tense.



On the typology of metaphony/stress dependent harmony 

 261

­ asculine nouns and adjectives by the height of the stressed vowel: the plural m morph is a floating autosegment that attaches to the stressed vowel and raises it. The existence of featural affixes follows from the existence of autosegments. Lexical representations can be impoverished in several ways: they can lack phonological content altogether, as in pro or expletives, or their phonological content can be impoverished: they can lack skeletal (C, V) information, as in the case of featural affixes, or they can lack featural specifications (see Svenonius & Bye 2012 for important general observations). In some cases morphs have a mixed ­character: they can consist of both segmental and floating material, or segmental and bare skeletal material (Mc Laughlin 2000; Finley 2009 and references therein). The Romance dialect of Calvello (Basilicata, Italy; Gioscio 1985) that I examine in the next section, illustrates both cases.

2 M  etaphony and morphological gemination in Calvello An instance of morphemic harmony, in which a morph contains melodic material that is not linked to any skeletal position, is metaphony in the Italo-Romance dialect of Calvello, where there is raising of mid close vowels to high, and diphthongization of mid open vowels (é→í, Ó→ú; ɛ́→jé, ɔ́→wé). Before discussing this process I will analyze gemination in this dialect because it provides an interesting case of the opposite situation: a morph that has a skeletal C position which is deprived of any melodic material. In several dialects of Italy the mass-count distinction is marked by the stem final vowel, as in Servigliano, e.g. lu péʃʃu ‘the fish-count’, lo péʃʃo ‘the fish-mass’, or in Antrodoco, here only in the definite article and in demonstratives, lu férru ‘the (piece of) iron’, lo férru ‘iron-mass’ (Maiden 1997: 73–74; for Servigliano, Camilli 1929: 226; Mascaró 2011: 31–33; for Antrodoco, Scorretti 2012: 125–127). In Calvello (Goscio 1985: 58, 38–39) the exponence of mass is not a desinential vowel; mass is marked by gemination of a singular noun’s initial consonant when preceded by the definite article (4a); when the noun begins with a vowel or /w/ the asyllabic form of the article /l/ appears (4d), and there is no explicit mass/count distinction. Gemination also takes place after the feminine plural definite article (4b), and after a closed set of determiners, pronouns, prepositions, conjunctions, adverbs, and verbal forms (4c).3

3 Other examples: ɛ kkwánnə ‘and when’, pə ffá ‘in order to make’. In the case of noun-intial /v/, /ɣ/, /j/ gemination yields [bb], [gg], [ɟɟ], respectively, e.g. játtʃə, lu ɟɟáttʃə ‘ice’. Calvello doesn’t have the phonological gemination triggered by a preceding stressed vowel, as in Italian sarà [bb]uono ‘it will be good’.

262  (4)

 Joan Mascaró

Count a. lu kásə lu pánə lu fjérrə b. f.sg la kásə la mánə c. li múlə d. l attsárə l wóʎʎə

Mass lu kkásə lu ppánə lu ffjérrə f.pl lə kkásə lə mmánə tre mmúlə

‘the cheese-count/mass’ ‘the bread-m.sg.count/mass’ ‘the iron-m.sg.count/mass’ ‘the house-f.sg/pl’ ‘the hand-f.sg/pl’ ‘the mules/three mules’ ‘steel-mass’ ‘oil-mass’

This particular case of gemination is an instance of morphological gemination (­raddoppiamento morfologico);4 following Chierchia’s (1986) analysis for other cases of morphological gemination, this case can be dealt with by positing a melodically empty skeletal position at the end of the word that triggers the gemination of the initial consonant of the following word (see also Passino 2013 and references). Although several other representational systems would give similar results, here I will assume that the root of the tree of segmental structure is C, V, which represent a timing slot and the features that distinguish vocalic from consonantal elements; the root dominates the rest of segmental features, here represented by IPA symbols. The underlying structure of definite articles is the one in (5). The articles l-u, l-a, and l-i have a standard segmental representation; l-ə, the feminine plural, which causes purely phonological gemination (see fn. 2) has a final, melodically null C that triggers gemination through linking of the melodic structure of the initial C of the following word; l-u, the masculine mass article also displays this final C, and differs in this way from the masculine singular count article: (5) Singular

Plural

Mass

Masculine

Feminine

l-u

l-a

C V

CV

l-i

l-ǝ

CV

CV C

l-u CV C

4 Other languages also use gemination to mark morphological features. In Qafar (Cushitic, Ethiopia) the indefinite genitive in feminines is a final bare C that geminates the initial consonant of the following word: saɡa ‘cow’ saɡa ɗɗaylo ‘cow’s offspring’ (Svenonius & Bye 2012: 443).

On the typology of metaphony/stress dependent harmony 



 263

The association of the bare C to the melodic structure of the following C is forced by the interaction of the constraints Dep-Link, *No-Link, *CC, *uw, ij: (6) a. *no-link: Floating features/featureless skeletal elements are disallowed.5 b. Dep-Link: Let S be a segment in the input and F a feature in the input, and S’, F’ their corespondents in the output. Assign a violation mark if S’ is linked to F’ and S is not linked to F.6 c. *CC: Assign a violation mark to any pair of C slots linked to the same segmental features (No geminates). d. *uw, uj: Assign a violaton mark to any VC sequence linked to the same segmental features (No homorganic falling diphthongs). The tableau in (7) illustrates gemination in mass DPs. The geminating c­ andidate (7a) wins because the fully faithfull candidate (7b) satisfies the constraint Dep-Link but violates *No-Link, and because the preference of geminates to falling high diphthongs favors the candidate (7a) over (7c): (7) *No-Link> >Dep Link, *CC; *uw, ij> >*CC l u

ka s ǝ

C VC CVCV a. l u

*No-Link

kas ǝ

C V C CV C V b. l u

ka sǝ

C V C C V CV c. l u

ka sǝ

C V C C V CV

1W

Dep Link

*uw, ij

*CC

1

1

L

L

1

1W

L

I will now return to the case of morphemic harmony in Calvello. This dialect, like all the others discussed here, has the seven vowel system in stressed position shown in roman boldface in (8). Metaphony triggers raising/diphthongization as indicated.

5 Cf. No-Float in McCarthy et al. (2012). 6 Torres-Tamarit (2012: 114, passim). It is similar to NoSpread (McCarthy 2000) and to NoLink (McCarthy 2008: 278).

264 

 Joan Mascaró

(8) i

u

e je ε

o wo ɔ

a

This process shows different kinds of morphologization. First, since vowel reduction merged all the inflective final vowels to [ə], the historical, metaphonytriggering vowels [i] and [u] are not present anymore in the trigger position, and the metaphonic process must be ascribed to a floating feature that is not realized as a normal suffixal element by concatenation with the stem, but through attachment to a prominent position, the stressed syllable. But in this case we do not have a pure featural affix, like [+ATR] in Kanembu (1) or [+high], [+ATR] in Lugo (3); the morph consists of both segmental material, i.e. [ə], and floating material, that I will assume to be [+high], [+ATR]. Metaphony affects elements in the noun, adjective, and verb paradigm. In verbs it distinguishes the second from the first and third person singular in some paradigms: (9)

a. 1sg.prs.ind b. 2sg.prs.ind c. 3sg.prs.ind

‘to reap’ sɛ́rrə sjérrə sɛ́rrə

‘to peel’ mónnə múnnə mónnə

Verbal suffix /ə/ /ə, [+high], [+ATR]/ /ə/

In (9a, c) the morphological structure is /sɛ́rr-ə/, /mónn-ə/7, with a segmental desinential morph. But in (9b) the desinential morph has the structure in (10a), where a node with the morphological features of 2sg.pr.ind dominates an unordered set consisting of [V ∆ ] (where ∆ represents the segmental tree of /ə/), a floating [+high] and a floating [+ATR]; henceforth I will represent this structure in the form in (10b): (10) a. 2sg.pr.ind   V [+high]    ∆ [+ATR]

/



/

b. /ə, [+high], [+ATR]/

Metaphonic raising also distinguishes, in verbs, the second from the first and the third person in the singular of the present subjunctive (11a–c), and in the plural of the past subjunctive (11d–f):

7 Although stress in all these dialects is partly predictable, I mark the stress in the underlying form for convenience.

On the typology of metaphony/stress dependent harmony 



(11)

a. b. c. d. e. f.

‘to put’ 1sg.prs.sbjv 2sg.prs.sbjv 3sg.prs.sbjv 1pl.pst.sbjv 2pl.pst.sbjv 3pl.pst.sbjv

mettéssə mettíssə mettéssə mettéssəmə mettíssəvə mettéssərə

 265

Verbal suffix(es) /é - ss - ə/ /é - ss - ə, [+high], [+ATR]/ /é - ss - ə/ /é -ss - əmə/ /é - ss - əvə, [+high], [+ATR]/ é-ss-ərə

In (11a, c, d, f) the last morph in the verbal desinence is /ə/, but in (11b, e) it is /ə, [+high], [+ATR]/, the floating autosegments causing raising (as in (9b)), here in the thematic vowel of the desinence. In nominals, raising/diphthongization has been morphologized differently in nouns and in adjectives. In nouns it distinguishes masculine singulars from masculine plurals. (12) a. rɛ́ndə vɔ́skə mésə sórətʃə b. pɔ́rtə pérə

rjéndə vwóskə mísə súrətʃə pɔ́rtə pérə

‘tooth-m.sg/pl’ ‘forest-m.sg/pl’ ‘month-m.sg/pl’ ‘mouse-m.sg/pl’ ‘door-f.sg/pl’ ‘pear-f.sg/pl’

The gender-number marker is typically fusional in Italo-Romance varieties: there is a single vowel representing both gender and number values. In Calvello, the exponent in feminines is always /ə/; no singular-plural distinction is possible. In masculines, the masculine singular marker is also /ə/, but the masculine plural marker, which is /ə, [+high], [+ATR]/, causes raising or diphthongization of the stressed vowel. Finally, in adjectives morphologization has taken a different path. Here the morph /ə, [+high], [+ATR]/ is not the exponent of masculine plural, but the general exponent of the masculine, both for singular and plural:8 (13) f (sg and pl): /ə/ apɛ́rtə sɔ́ttsə cénə sólə

m (sg and pl): /ə, [+high], [+ATR]/ apjértə swóttsə cínə súlə

‘open’ ‘flat’ ‘full’ ‘alone’

8 We could also assume that in masculine nouns /ə/ is the gender marker and the floating features are the exponent of plural, and that in adjectives the floating features are the exponent of masculine; similar analyses could be proposed for verbal forms. I owe this suggestion to Eulàlia Bonet.

266 

 Joan Mascaró

What has happened historically is that, since the original raising/diphthongization affected both the masculine singular and the masculine plural (presumably súlu - súli, but sóla - sóle ‘alone’), at the point of morphologization, in adjectives it has been naturally transferred to the masculine feature. For nouns, in the masculine type mústu (← /móstu/) - músti (← /mósti/) ‘must’, with absolute neutralization, the underlying root /most-/ has been reanalyzed as /múst-/, except in a few cases with alternations (píru - píri ‘pear tree-masc/pl’ vs. pérə ‘pear-f.sg/pl’). The cases with regular singular/plural alternations in (12) derive normally from nouns with an original, non-high final singular stem vowel -e, e.g. més-e - mís-i > més-ə - mís-ə. Let us now proceed to the analysis. I will first deal with the simpler mappings e→i, o→u , and then try to account for the diphthongizing mappings ɛ→je, ɔ→wo. According to our previous discussion, the representation of the gendernumber morpheme that triggers metaphony is a morphologically labeled node that dominates the unordered set (ə, [+high], [+ATR]) (14a). This morpheme can be associated to the root /sórətʃ/ ‘mouse’ in (14b), yielding the ordered pair (14c),9 and the constraint hierarchy will force attachment of the floating features to the stressed vowel. (14) a. [Aff, m.pl /ə, [+high], [+ATR]/ ] b. [Root, m sórətʃ] c. [Root, m sórətʃ] [Aff, m.pl /ə, [+high], [+ATR]/ ] In (14c), the floating features remain unordered with respect to the stem vowel /ə/. The constraint *No-Link (6a) disallows unlinked floating material. MaxFloat (15a, Wolf 2007) penalizes deletion of floating material, and FloatStr (15b) requires that floating material be linked to a the prominent stressed position. This is shown in the tableau in (16). (15) a. MaxFloat: Assign a violation mark to any floating feature F in the input that does not have a correspondent feature F’ in the output. (Adapted from Wolf 2007) b. Float-Str: Assign a violation mark to any floating feature F in the input that does not have an output correspondent feature F’ linked to the stressed vowel.

9 Alternatively, morphs could appear unordered lexically, and the prefixal or suffixal charater of affixes could be determined by the constraint hierarchy.

On the typology of metaphony/stress dependent harmony 



(16) /sórətʃ-ə, [+hi], [+ATR]/

Max-Float

[+hi] [+ATR] sórətʃ-ə sórətʃ-ə

*No-Link

 267

Float-Str

*!* *!*

** **

[+hi] [+ATR] *!* sórɨt∫ -ǝ  [+hi] [+ATR] súrǝt∫ -ǝ

3 Stress dependent harmony and contrast The diphthongizing cases are more difficult to account for. One of the challenges of metaphony derives from the fact that a purely harmonic spreading process would give as a result the mappings ɛ→ɪ, ɔ→ʊ, e→i, o→u with [+high] as the harmonic feature, and ɛ→i, ɔ→u, e→i, o→u with [+high] and [+ATR] as harmonic features. Calabrese (1985, 1988: 41–58, 313–316, 1995: 396–402, 1999) derives diphthongization from the fact that associating [+high] to a mid open vowel results in a language-particular illicit [+high, –ATR] vowel, i.e. ɪ, ʊ. This triggers different repair strategies. Let us see how they work in the case of the front vowel. A readjustment process of Fission splits ɪ in two segments, S1 and S2. S1 retains the [+high ] feature of ɪ, and S2 retains its [−ATR] feature. Since S1 is now unspecified for [ATR], and the unmarked value of [ATR] for high vowels is [+ATR], we get an on-glide diphthongal [j]. S2 is [−ATR] and is unspecified for [high]; for [−ATR] vowels the unmarked value of [high] is ‘–’, and we get [ɛ] as the vocalic nucleus. We can adapt Calabrese’s insight to an OT framework without having to resort to an intermediate step, the illicit structure that has to be repaired, and by adding two more ingredients. First, I will assume that there is association of both [+high] and [+ATR] to the stressed syllable. Second, I propose that contrast preservation also plays a crucial role in these cases. The basic idea is that the mappings ɛ→i, e→i, and ɔ→u, o→u, predicted by attachment of [+high] and [+ATR] to the stressed vowel, neutralize an underlying contrast, namely ɛ/e and ɔ/o. A constraint that penalizes neutralization of specific underlying contrasts can derive, as we will see, the desired results. Contrast preservation constraints have been proposed by Łubowicz (2003, 2012), Padgett (2003), Tessier (2004), and Ní Chiosáin and Padgett (2009), (for

268 

 Joan Mascaró

other views on contrast see Flemming 2004, and for an overview, see Hall 2011). Here I will assume that their approach is basically correct, and that constrast preservation constraints play a role in evaluation, but I will not discuss the specifics of each approach. Let me mention, however, a basic problem that contrast preservation constraints face, that should be addressed by future research on this topic. Consider a constraint that militates against loss of segmental contrast of a specific sort. This means that given inputs I1, I2 differing just in the feature value of one segment, we want to prevent their merger, i.e. the mappings I1 → O1 and I2 → O2, O1=O2. Assume we are evaluating the candidates generated by the lexical element I1=/AsB/, where s is the contrasting segment. Assume further that the segments s and t differ just in the values of a set of features, f. The constraint PreserveContrast(f) will penalize the merger of segment s with segment t. This means that for input I1=/AsB/ a candidate Cs’D will violate this constraint just in case it is identical to the winning candidate of a possible input I2=AtB. But this raises a problem of circularity: before we know whether the candidate Cs’D violates PreserveContrast(φ) we have to know the winning candidate of I2, whence whether other candidates violate PreserveContrast(φ). I will illustrate the different effects of constraint interaction with the front vowels. If the association of [+high], [+ATR] to the stressed vowel is not disallowed by higher-ranked constraint requirements, we predict the mappings ɛ→i, e→i. But notice that in these cases the underlying /ɛ/-/e/ contrast is lost in a prominent position, main stress. The constraint PreserveContrast(Stress,mid), to be slightly revised below, militates against the loss of potential ɛ/e and ɔ/o underlying contrasts. (17) a. PreserveContrast(f): Let s, s′, t, t′ be segments, and s and t differ in the values of the set of features f. Assign a violation mark to a candidate Cs’D from input /AsB/, sℜs′, if and only if: a) there is a possible input /AtB/ whose winning candidate is Ct’D; b) t ℜt′ c) s′=t′ b. PreserveContrast(f)(Stress,mid) (PrCont(Str,mid)): Let s, s′, t, t′ be segments, and s and t differ in the values of [high] and [ATR]. Assign a violation mark to a candidate Cs’D from input /AsB/, sℜs′, if and only if; a) there is a possible input /AtB/ whose winning candidate is Ct’D b) t ℜt′ c) s′=t′

On the typology of metaphony/stress dependent harmony 



 269

In the simplified tableau in (18), I give the basic typological options for the outcome of the front mid open vowel ɛ (the other mapping, e→i, is ensured as long as FloatStress dominates Id(hi)). Assume first that Dep-C is high-ranked, disallowing diphthongization (18a–c). Then if float-str, the constraint that requires that floating material be attached to the stressed syllable, dominates PrCont(Str,mid),10 we predict the mapping ɛ→i with both [+high] and [+ATR] associated to the target vowel (18a). This is the case of the dialect of Foggia (Calabrese 1995:400, and references), and the cases of hypermetaphony described by Maiden (1991: 179–187, 232–244). If we get the reverse ordering, PrCont(Str,mid)>>FloatStress, we get at least two different ways in which contrast preservation is enforced by PrCont(Str,mid), (18b, c). The ordering FloatStress>>Id(ATR) derives (18b), ɛ→e, the chain-shift mapping that is found in Sabine metaphony (Servigliano, Ascrea, Antrodoco). Under Id(ATR)>>FloatStress we get ɛ→ɛ (18c) with just mid close to high raising (e→i, o→u), as in Veneto and Grado (Walker 2005: 921–931). Finally, if Dep-C is low-ranked, [+ATR] can be realized on the target vowel and [+high] can be realized as an on-glide at the prize of inserting new material. In this case, if Id(ATR) is ranked below FloatStress, [+ATR] will associate to the target vowel and we will get diphthongs with mid closed vowels (jé, wó, (18d)), as in Calvello, Arpino, and many other varieties. If we have the reverse ordering, Id(ATR)>>FloatStress, the original mid open vowel will be preserved (jɛ́, wɔ́, (18e)), as in Francavilla Fontana (Calabrese 1985, 1988: 41–49). (18) Mappings of /ɛ/ in the context . . . ə, [+high], [+ATR] a.

/ɛ/

Dep-C FloatStress

ɛ→ɛ

*!*

ɛ→e

*!

b.

/ɛ/

*

Id(ATR)

*

*

*! Dep-C PrCont(str,mid) FloatStress

ɛ→ɛ

Id(ATR) Id(hi)

**! *

 ɛ→e ɛ→i ɛ→jɛ

Id(hi)

*

 ɛ→i ɛ→jɛ

PrCont(str,mid)

*!

* *

*

*!

10 Reference to stress is neccessary because of the process of vowel reduction with the effect ɛ→e, ɔ→o in unstressed position.

270 

 Joan Mascaró

c.

/ɛ/

Dep-C PrCont(str,mid) Id(ATR) FloatStress **

 ɛ→ɛ ɛ→e

*!

ɛ→i ɛ→jɛ d.

/ɛ/

*!

*

PrCont(str,mid)

FloatStress

* *

*!

ɛ→ɛ

*!*

ɛ→e

*!

ɛ→i

Dep-C

/ɛ/

Id(hi)

Id(ATR) *

*!

*

*

*

 ɛ→jɛ e.

Id(hi)

PrCont(str,mid)

Id(ATR) FloatStress Dep-C

ɛ→ɛ

*!*

ɛ→e ɛ→i

Id(hi)

*! *!

*

*

* *

 ɛ→je

As an illustration, consider the input /rɛ́ndə/ (12a), and the evaluation of the most relevant candidates. (19)

/rɛ́nd-ə,[+hi],[+ATR]/

PrCont (str,mid)

a. rɛ́nd-ə

Id(ATR)

Dep-C

Id(hi)

*!*

b. rénd-ə c. rínd-ə

Float Stress *!

*!

* * *

d.  rjénd-ə

* *

(19c) violates PrCont(str,mid) because, as shown in (20), there is a possible input /rínd-ə,[+hi],[+ATR]/ mapping into [rínd-ə]. In /rínd-ə,[+hi],[+ATR]/ ɛ́=t in (17b), in [rínd-ə] í=t′; in (19), for /rɛ́nd-ə,[+hi],[+ATR]/ ɛ́=s, for [rínd-ə] í=s′. Since s′=t′, a violation mark is assigned. (20)

/rínd-ə,[+hi],[+ATR]/

PrCont (str,mid)

Float Stress

a. rɛ́nd-ə

*!*

b. rénd-ə

*!

Id(ATR)

Dep-C

*

Id(hi) * *

c.  rínd-ə d. rjénd-ə

*

*

*



On the typology of metaphony/stress dependent harmony 

 271

As formulated in (17b), the contrast preservation constraint should equally apply to /rɛ́nd/=/AsB/ (19) and to /rínd-ə/=/AsB/. In other words, we do not know whether the contrast should be maintained by changing the mapping /rɛ́nd/→[rínd] or by changing the mapping /rínd/→[rínd]. I assume that the more faithful mapping /rínd/→[rínd] is preserved. Therefore, I propose to modify the constraint family by the addition of clause d): (21) a. PreserveContrast(φ): Let s, s′, t, t′ be segments, and s and t differ in the values of the set of features φ. Assign a violation mark to a candidate Cs’D from input /AsB/, sℜs′, if and only if: a) there is a possible input /AtB/ whose winning candidate is Ct’D; b) t ℜt′ c) s′=t′ d) the candidate Cs’D from /AsB/ is more harmonic than the candidate Ct’D from /AtB/. b. PreserveContrast(φ)(Stress,mid) (PrCont(Str,mid)): Let s, s′, t, t′ be segments, and s and t differ in the values of [high] and [ATR]. Assign a violation mark to a candidate Cs’D from input /AsB/, sℜs′, if and only if; a) there is a possible input /AtB/ whose winning candidate is Ct’D b) t ℜt′ c) s′=t′ d) the candidate Cs’D from /AsB/ is more harmonic than the candidate Ct’D from /AtB/. Thus in a situation of potential merger the merger is avoided by letting the most harmonic candidate (with respect to its own evaluation) surface, while the least harmonic one is avoided becuse it violates the contrast preservation constraint.

4 Phonological stress dependent harmony Let us now consider an instance of phonological harmony. The stress dependent harmonic system we find in Servigliano is very different from the morphemic harmony we find in Calvello. Here the harmonic mappings (ɛ→e, ɔ→o, e→i, o→u) are transparent, always triggered by a surface high vowel [i] or [u], and can be analyzed as an instance of phonological harmony. Servigliano has a complex set of harmonic processes. Here we will be interested only in harmony caused by a final [+high], [+ATR] triggers, a regular process with very few lexical exceptions (Camilli 1929; Mascaró 2011: 31–33). A high vowel triggers gradual raising of the preceding stressed mid vowel, i.e. ɛ→e, ɔ→o, e→i,

272 

 Joan Mascaró

o→u. This is illustrated below with the suffixes /-i/ (masculine plural, 2nd singular present indicative, 2nd singular imperative). (22) a. pɛ́d-e rrepprɛ́ts:-o bɔ́n-a mɔ́r-e pés-a métt-e fjór-e bótt-e

péd-i rreppréts:-i bón-i mór-i pís-u mítt-i fjúr-i bútt-i

‘foot-m.sg/pl’ ‘take care-1sg/2sg.prs.ind’ ‘good-f.sg/m.pl’ ‘die-3sg/2sg.prs.ind’ ‘heavy-f.sg /m.sg.count’ ‘put-3sg.prs.ind/2sg.imp’ ‘flower-m.sg/pl’ ‘cask-m.sg/pl’

I follow Mascaró (2011) in analyzing Servigliano as a case of phonological harmony, governed by the constraints Agree(+high, +ATR): (23) a. Agree(+high, +ATR): For every pair of adjacent vowels one of which is [+high, +ATR], assign one violation mark if they are not linked to the same token of [+high] and [+ATR]. b. Ident-Suff(VF): Assign one violation mark for any feature in a suffixal vowel that does not have the same value as its correspondent vowel in the input.  (Mascaró 2011: 28) Agree(+high, +ATR) requires adjacent vowels to agree in [high] and [ATR] if one of them is [+high, +ATR], and Id-Suff(VF) forces regressive harmony by ensuring suffixal faithfulness. However, I make use of PreserveContrast(Stress,mid) instead of Mascaró’s (2011) local conjunction Ident(high&ATR), which disallows the mappings ɛ→i, ɔ→u, also used by other authors (Kirchner 1996; Walker 2005: 959–960, 2011: 260–261); Id-Suff(VF)>>Agree(+high,+ATR), since spreading takes place from right to left, and PrCont(Str,mid)>>Agree(+high,+ATR), in order to prevent fellswop raising to high. Consider the evalution of /mɔr-i/ → [mór-i] ‘die-2sg.prs.ind’: (24)

/mɔ́r-i/

Id-Suff(VF)

PrCont(Str,mid)

*

F mór-i mɔ́r-i

**!

múr-i mɔ́r-e

Agree(+high,+ATR)

*! *!

*

Even if raising harmony in Servigliano is a good instance of regular phonological harmony, it shows some early signs of morphologization. Serviglano has an adjectival paradigm of five elements: singulars (masculine and feminine), plurals

On the typology of metaphony/stress dependent harmony 



 273

­(masculine and feminine), and mass (25a, b). Mass nouns end in the mass marker /-o/, and show metaphonic raising, even though the final vowel is not high (25a–c). But there are also nouns ending in /-o/ that are not mass nouns and show no metaphonic raising (25d). Thus in péʃʃ-o the suffixal -o causes raising, while in tesɔ́r-o the suffixal -o does not. (25) a. b. c. d.

bón-u bɔ́n-a nír-u nér-a lu péʃʃu li péʃʃi tesɔ́ro stɛ́roko

bón-i bɔ́n-e bón-o ‘good-m.sg/f.sg/m.pl/f.pl/mass’ nír-i nér-e nír-o ‘black-m.sg/f.sg/m.pl/f.pl/mass’ lo péʃʃo ‘the fish-count/pl/mass’ ‘treasure-m.sg’ ‘treasure-m.sg’ ‘belch-m.sg’

We can assume that the underlying forms for the examples in (25d) have the simple segmental affix, /-o/, whereas the mass forms in (25a–c) are /bɔ́n/, /nér/, and /pɛ́ʃʃ/ with a complex mass marker with a segmental /-o/ and floating [+high], [+ATR], similar to what we found in the morphologized cases we have discussed for Calvello. Following Mascaró (2011: 32–33) the structure of the mass suffix is as in (26), with floating autosegments, which are similar to the structures in (11) for Calvello. [+hi]

(26) ner-

o

[+ATR]

→

[+hi] [+ATR] ner-

[o]

5 Conclusion and prospects We have examined some cases of stress dependent harmony that indicate that there is an important typological distinction determined by the structure of triggers. They can be morphemes consisting of a floating feature, or a set of features (featural affixation), they can be morphemes formed by both segmental and floating material (segmental and featural affixation), or they can be a pure phono­ logical element, no morphology being involved (pure phonological harmony). We have also seen that preservation of underlying contrasts plays a crucial role both in phonological and in morphemic harmony. There remain many open questions, though. First, as we have seen, the precise formulation of contrast preservation constraints needs further investigation. Also, there are many other cases that will not fit well in this simplified analytic frame. Even if the opaque cases like Calvello need an analysis in terms of featural floating material, it is not totally clear whether more transparent cases, like Servigliano, might also require this approach. The basic question is whether in cases like Servigliano the trigger is

274 

 Joan Mascaró

just a high vowel, or a specific morpheme. Counter-evidence to the morphemic analysis that should be checked in future research would be cases in which harmony is triggered within a morpheme. This means basically that we would have to have a large sample of cases of invariable words (mainly adverbs) with apparent harmonic effects and no case, or very few cases, where such effects were missing. This is difficult in Servigliano because the process is neutralizing, and there is always the possibility, in the case of few intramorphemic examples, of analyzing the morpheme with the underlying phonetic stressed vowel. Camilli (1921) offers just the examples jéri ‘yesterday’ and ódʒːi ‘today’. These words have [ɛ] and [o], respectively, in Vulgar Latin, but in a synchronic analysis of Servigliano they can be represented with their phonetic value in underlying form, i.e. /jéri/, /ódʒːi/. Another possible source of evidence is harmony in proparoxytones. A word like Italian doménik-a ‘Sunday’ has a morpheme-internal post-tonic high vowel that should trigger raising on the tonic. But Servigliano has a regressive process of total vowel copy affecting post-tonic vowels that makes a sequence formed by a possible stressed target, a high vowel, and a non-high vowel like é-i-a impossible; the word for ‘Sunday’ is doménnak-a (pl. doménnek-e; cf. ʃtómuk-u, ʃtómik-i ‘stomach-sg/pl’; prɛ́dok-o, prédik-i, prɛ́dak-a ‘preach-1sg.prs.ind/3sg. prs.ind’). In prédik-i there is a post-tonic high vowel but the raising of the tonic is triggered by the final vowel. Other dialects avoid metaphony altogether in proparoxytones, and in other cases, like in Grado, proparoxytones allow harmony, even if the tonic is not affected, suggesting a more pure phonological spreading process (géndene, gíndini ‘louse’s egg-m.sg/m.pl’; zóvene, zúvini ‘youngsterm.sg/m.pl’; álboro, álburi ‘tree-m.sg/m.pl’; ánzolo, ánzuli ‘angel-m.sg/m.pl’; see Mascaró (submitted)). At the other end we have cases like Lena Asturian (Neira 1955) in which the floating features analysis seems much more justified because the process can proceed regressively from the final vowel to the stressed vowel skipping a potential post-tonic target (péʃaru, páʃaros ‘bird-sg/pl’; birwíβanu, birwéβanos ‘wild strawberry-sg/pl’).

References Akinlabi, Akinbiyi. 1994. Alignment constraints in ATR harmony. Studies in Linguistic Sciences 24: 1–18. Akinlabi, Akinbiyi. 1996. Featural affixation. Journal of Linguistics 32: 219–289. Akinlabi, Akinbiyi. 2011. Featural affixes. In Marc van Oostendorp, Colin J. Ewen, Elizabeth V. Hume & Keren Rice (eds.), The Blackwell companion to phonology,  1945–1971. Chichester: Wiley-Blackwell. Calabrese, Andrea. 1985. Metaphony in Salentino. Rivista di grammatica generativa 9–10: 1–140.



On the typology of metaphony/stress dependent harmony 

 275

Calabrese, Andrea. 1988. Towards a Theory of Phonological Alphabets. PhD dissertation. ­Cambridge: MIT Press. Calabrese, Andrea. 1995. A constraint-based theory of phonological markedness and simplification procedures. Linguistic Inquiry 26 (3): 373–463. Calabrese, Andrea. 1999. Metaphony revisited. Rivista di Linguistica 10: 7–6. Camilli, Amerindo. 1929. Il dialetto di Servigliano. Archivum Romanicum 13: 220–251. Chierchia, Gennaro. 1986. Length, syllabification and the phonological cycle in Italian. Journal of Italian Linguistics 8: 5–33. Cole, Jennifer. 1991 [1987]. Planar Phonology and Morphology. PhD dissertation. Cambridge: MIT Press. Published New York: Garland Publishing. Finley, Sara. 2009. Morphemic harmony as featural correspondence. Lingua 119: 478–501. Flemming, Edward. 2004. Contrast and perceptual distinctiveness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.), Phonetically-based phonology, 232–276. Cambridge: Cambridge University Press. Gioscio, Joseph. 1985. Il dialetto lucano di Calvello. Stuttgart: Franz Steiner. Hall, Daniel Currie. 2011. Contrast. In Marc van Oostendorp, Colin Ewen, Elizabeth Hume & Keren Rice (eds.), The Blackwell companion to phonology, 27–53. Oxford: Wiley-Blackwell. Jouannet, Francis. 1982. Le kanembou des Ngaldoukou, langue saharienne parlée sur les rives sptentrionales du Lac Tchad. Paris: Société d’études linguistiques et anthropologiques de France. Kabak, Baris. 2011. Turkish vowel harmony. In Marc van Oostendorp, Colin Ewen, Beth Hume & Keren Rice (eds.), The Blackwell companion to phonology, 2831‒2854. Oxford: Wiley-Blackwell. Kirchner, Robert. 1996. Synchronic chain shifts in Optimality Theory. Linguistic Inquiry 27 (2): 341–350. Łubowicz, Anna. 2003. Contrast Preservation in Phonological Mappings. Amherst, MA: PhD. University of Massachusetts dissertation. Łubowicz, Anna. 2012. The phonology of contrast. London: Equinox. Maiden, Martin. 1987. New perspectives on the genesis of Italian metaphony. Transactions of the Philological Society 85: 38–73. Maiden, Martin. 1991. Interactive morphonology: metaphony in Italy. New York: Routledge. Maiden, Martin. 1997. Inflectional morphology of the noun and adjective. In Martin Maiden & Mair Parry (eds.), The dialects of Italy, 68–74. London: Routledge. Majors, Tivoli. 2000. A perceptually grounded OT analysis of stress-dependent vowel harmony. In Proceedings from Southwest Workshop on Optimality Theory 4, 29–42. Tucson, AZ: University of Arizona. Mascaró, Joan. 2011. An analysis of sress-dependent harmony in servigliano. Probus 23 (1): 21–55. Mascaró, Joan. Submitted. Myopic harmony and nonlocal trigger-target asymmetyries. McCarthy, John J. 2008. The gradual path to cluster simplification. Phonology 25: 271–319. McCarthy, John J. 2000. Faithfulness and prosodic circumscription. In Joost Dekkers, Frank van der Leeuw & Jeroen van de Weijer (eds.), Optimality theory: syntax, phonology, and acquisition, 151–189. Oxford: Oxford University Press. McCarthy, John J., Kevin Mullin & Brian W. Smith. 2012. Implications of Harmonic Serialism for lexical tone association. In Bert Botma & Roland Noske (eds.), Phonological explorations. empirical, theoretical and diachronic issues, 265–297. Berlin/Boston: De Gruyter.

276 

 Joan Mascaró

Mc Laughlin, Fiona. 2000. Consonant mutation and reduplication in Seereer–Siin. Phonology 17 (3): 333–363. Neira, Jesús. 1955. El habla de Lena. Oviedo: Instituto de Estudios Asturianos. Ní Chiosáin, Máire & Jaye Padgett. 2009. Contrast, comparison sets, and the perceptual space. In Parker, S. (ed.), Phonological argumentation: essays on evidence and motivation, 103–121. London: Equinox Publishing. Padgett, Jaye. 2003. Contrast and post-velar fronting in Russian. Natural Language and Linguistic Theory 21 (1): 39–87. Passino, Diana. 2013. A unified account of consonant gemination in external sandhi in Italian: Raddoppiamento Sintattico and related phenomena. The Linguistic Review 30 (2): 313–346. Pelliciardi, Ferdinando. 1977. Grammatica del dialetto romagnolo. La lèngva dla mi tëra. Ravenna: Longo Editore. Roberts, James S. 1994. Nontonal floating features as grammatical morphemes. Work Papers of the Summer Institute of Linguistics 38: 87–99. Savoia, Leonardo & Martin Maiden. 1997. Metaphony. In Martin Maiden & Mair Parry (eds.), The dialects of Italy, 15–25. London: Routledge. Scorretti, Mauro. 2012. Il dialetto di Antrodoco. Sambuceto: Fondazione Ernesto Giammarco. Svenonius, Peter & Patrik Bye. 2012. Non-concatenative morphology as epiphenomenon. In Jochen Trommer (ed.), The morphology and phonology of exponence, 427–495. Oxford: Oxford University Press. Tessier, Anne-Michelle. 2004. Input “clusters” and contrast preservation in OT. Proceedings of the West Coast Conference on Formal Linguistics 23: 759–772. Torres-Tamarit, Francesc. 2012. Syllabification and opacity in Harmonic Serialism. Bellaterra, Catalonia: Universitat Autònoma de Barcelona dissertation. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language & Linguistic Theory 23 (4): 917–989. Walker, Rachel. 2011. Vowel patterns in language. Cambridge: Cambridge University Press. Wolf, Matthew. 2007. For an autosegmental theory of mutation. In Leah Bateman, Michael O’Keefe, Ehren Reilly & Adam Werle (eds.), University of Massachusetts Occasional Papers in Linguistics 32: Papers in Optimality Theory III, 315–404. Amherst MA: GLSA.

Martin Krämer

Metaphonic chain shifts, vowel height and markedness UiT The Arctic University of Norway

1 Introduction In this paper, I address the chain shift nature of metaphony in the dialects of Italian. In the most pervasive type, a post-tonic high vowel causes raising of the stressed vowel. Dialects differ in the behavior of intervening vowels, in which stressed vowels allow raising, and whether they only raise or diphthongize as well. Dialects differ as well with respect to the fate of the triggering vowel. It may be realized faithfully, reduced to schwa, or deleted. A chain shift can be observed in the targeted vowels. In the most extensive system, that of the Ischia dialect (Maiden 1991), stressed /a/ raises to [ɛ], /ɛ/ to [e], and /e/ raises to [i], preceding an unstressed high vowel. Regarding the targets of raising, Maiden (1991) identified an implicational hierarchy: if a lower vowel raises the higher vowels raise as well, but not vice versa. From the vantage point of Optimality Theory the question arises why do not all vowels simply raise to high. These apparently opaque patterns, or cases of Derived Environment Blocking, will be accounted for here in a parallelist fashion, ­combining Syntagmatic Correspondence (Krämer 1998, 2001, 2003a) with Local Conjunctions (henceforth LC; Prince & Smolensky 1993/2004; Smolensky 1993; ­Kirchner 1996; Crowhurst & Hewitt 1997; Moreton & Smolensky 2002; Lubowicz 2002; Krämer 2003a). While earlier approaches use LC in an opportunistic way (usually some markedness constraint M is conjoined with some faithfulness constraint F into M&F to achieve some desired effect), in this paper we will explore a more principled approach to LC, based on Kirchner’s (1996) account of one-step vowel raising as LCs of faithfulness constraints (F&F LCs). We will consider all LCs of all faithfulness constraints on height and their effect. Furthermore it will be shown that the implicational nature of the typology, that is, that lax mid vowels only raise where tense mid vowels do, and low vowels only raise where lax mid vowels do, is conditioned by constraint rankings that are necessary for the neutralization of the tenseness contrast in mid vowels in unstressed position (mid vowel tensing). The paper is structured as follows. Section 2 introduces the basic pattern and the aspects of it that are relevant here for the discussion of LC and the nature

278 

 Martin Krämer

of the process. Section 3 discusses the problem with the chain shift as well as the markedness paradox that emerges from certain analyses of metaphony and mid vowel tensing in unstressed syllables. The latter has to be accounted for in a fashion compatible with an analysis of metaphony, since it involves the same feature and occurs, by and large, in the same varieties. From this, section 3 will develop an argument to treat metaphony as identity licensing, i.e. emergent via syntagmatic correspondence. Section 4 analyses the chain shift as an effect of local conjunctions of IO-Identity constraints on height features, extending Kirchner’s (1996) LC approach to a similar chain shift involving vowel height. Section 5 gives a brief sketch of the identity licensing using Krämer’s (1998, 2001, 2003) Syntagmatic Correspondence. Section 6 provides a brief speculation on potential other phenomena that could be analyzed with the conjoined faithfulness constraints on vowel height. Section 7 concludes.

2 The patterns In this section I discuss only the aspect of the pattern that is the central concern of this paper. For broader, more detailed, introductions see Krämer (2009), Calabrese (2011) or Mascaró (this volume). In a nutshell, Italian-style metaphony is the raising of non-high stressed vowels by one step on a 4-step height scale (a -ɛ/ɔ -e/o -i/u) in the presence of a following (usually word-final) high vowel (/i/ or /u/) (at least at an abstract level of representation). Maiden (1991: 115f) identifies an implicational hierarchy for the targets of raising. Raising of low vowels implies raising of mid vowels and raising of lax mid vowels implies raising of tense mid vowels. Thus, once one observes a low vowel raising to [ɛ] one can expect that the higher vowels also raise in the same context in this dialect, but not vice versa. (1) Metaphonic chain shift typology (ignoring diphthongization)





 /ˈe/ → [ˈi]/

/ˈɛ/ → [ˈe]/ /ˈa/ → [ˈɛ]/

C(CVC)i,u# C(CVC)i,u# C(CVC)i,u#



Veneto



Southern Umbro



Ischia

These hierarchies, however, are not without exception. In varieties of Lombardian (Poschiavo and Isolaccia) and a few other areas, only lax mid vowels raise (Maiden 1991: 128f; Savoia & Maiden 1997: 17). This reminds of another, potentially related, process in Standard Italian and all these dialects. Lax mid vowels raise to tense mid vowels in unstressed syllables, as illustrated in (2).



Metaphonic chain shifts, vowel height and markedness 

(2) Unstressed vowel neutralization in Standard Italian a. [ortopeˈdiːa] ‘orthopaedics’ b. [ˈlɔdʒika] [ortoˈpɛdiko] ‘orthopaedist / [lodʒikaˈmente] orthopaedic’ ‘hang-1sg’ [ˈkɔːpri] [apˈpɛndo] [appenˈdjaːmo] ‘hang-1pl’ [koˈpriːte] ‘well’ [ˈpɔrto] [ˈbɛːne] ‘benefit’ [porˈtjaːmo] [beneˈfiːtʃo]

 279

‘logics’ ‘logically’ ‘cover-2sg’ ‘cover-2pl’ ‘carry-1sg’ ‘carry-1pl’

Both processes, metaphony and unstressed mid vowel raising, can be connected to the same cause. Vowels in unstressed syllables, as well as word-final vowels are in a phonologically weak position. Weak positions favor neutralization (resulting in smaller inventories), while strong positions, such as stressed syllables, syllable onsets, word-initial syllables, etc., are more favorable in contrast and hence support larger segment inventories, an observation which gave rise to, among other theories, Positional Faithfulness (McCarthy & Prince 1995; Beckman 1997a,b) as well as Positional Markedness or Licensing (Zoll 1996/8; Walker 2005, 2011). An underlying segment that ends up in a weak position on the surface has three options—ignore its weakness and defy markedness, change to the unmarked, or encroach upon a strong position. The lax mid vowels seem to be the marked, say [+RTR]1, in the weak unstressed position and change to a less marked value, i.e. [−RTR]. Something about /u/ and /i/, the high vowels, also seems to be too marked for word-final (or in some dialects, post-tonic) position. This dilemma is solved by assimilating the vowel in the strong stressed syllable, by making it raise towards the height of the vowel in the weak position, rather than changing the vowel in the weak position. In this sense, the two processes are related.

3 Issues with chains and markedness In this paper I focus on three problems with metaphony: First, we are dealing with a chain shift, every vowel in the chain moves only one step rather than all the way, which is a challenge, especially for analyses cast in OT (see Łubowicz 2011). Second, the typology of this chain shift reveals an implicational relation between the involved targets of raising: the higher a vowel is the more likely it is also to raise. Third, there is a markedness paradox if we

1 [RTR] = Retracted Tongue Root, [ATR] = Advanced Tongue Root.

280 

 Martin Krämer

take the foundational tenets of segmental phonology and standard diagnostics of phonological markedness (for an excellent overview see Rice 2007) seriously. The two types of raising observed, i.e. metaphonic raising and raising of lax mid vowels in unstressed syllables, are difficult to reconcile, it seems. I will first discuss the chain shift nature and then elaborate on the markedness paradox.

3.1 The chain shift nature of Italian metaphony The raising observed in metaphony is a typical chain shift. Low-mid and low vowels do not become high, they just move one step up on a scale. (3) Shifting on the height scale a. Chain shift: /a/ →

b. Height scale:

[ɛ] /ɛ/

→

[e]  /e/   →  [i] Low — lax-mid — tense-mid — high

If metaphony occurs because of a demand for licensing of the final high vowel in a prominent position (e.g. Walker 2003, 2011), this one-step raising does not really help the final V that much. Furthermore, because of the nonlocal nature of the process, Walker (2011) classifies metaphony as indirect licensing in her threeway distinction of licensing types. In addition, her typology includes direct licensing and identity licensing. Direct licensing is simply the avoidance of the marked feature in weak positions and its expression exclusively in strong positions (e.g. mid lax vowels occurring only in stressed syllables here). Indirect licensing is achieved via autosegmental spreading of a feature, such that it is linked to the weak as well as the strong licenser position. Identity licensing is achieved by feature duplication in the same representation. This implies some kind of correspondence between the involved weak and strong position, as proposed with Syntagmatic Correspondence (Krämer 1998, 2001, 2003a) or surface correspondence (Walker 2000, 2011 and references therein). We will come back to the issue of formalizing the metaphonic raising process shortly when we consider the second problem. The first issue has to do with the nature of the chain shift itself. Why do these vowels raise only one step rather than going all the way and licensing all the post-tonic vowel’s height features in the strong position? And how does one formalize this in Optimality Theory? Accounts of metaphony usually involve quite some degree of opacity, for example Calabrese’s (1998) analysis involves feature spreading, which is then undone by various repair mechanisms to improve the

Metaphonic chain shifts, vowel height and markedness 



 281

unrealizable structure that resulted from the spreading operation (see Cole 1998 for an insightful discussion). In Optimality Theory we either have to analyze this kind of pattern with special theoretical assumptions (see e.g. Torres-Tamarit & Linke, this volume) or rethink it in a different way. If metaphony is caused by positional markedness and is a licensing strategy it is some kind of spreading of the involved features (i.e. indirect licensing) or identity by correspondence (i.e. identity licensing). However, changing a low or low-mid vowel from [−high, +low, −ATR] to /i/, i.e. [+high, −low, +ATR] changes every height feature there is. There is some threshold beyond which segments cannot be changed. In our case this threshold seems to be maximally one feature change. Anything beyond this is not tolerated. I will elaborate on this below and formalize the insight by way of LC, following a proposal by Kirchner (1996). In a nutshell, faithfulness constraints “gang up” against some orthogonal force. Violating faithfulness constraint A or B is permissible, but violating constraint A and B is not. Before going into the details of this idea and those of the solution to the third problem I proceed to the next problem to be approached here, the implicational nature of the dialect typology. There are dialects in which only the tense mid vowels raise, those in which all the mid vowels raise, and finally those in which all vowels raise. (4) Shifting on the height scale and dialectal variation a. Chain shift: a → ε → e b. Height scale:

→ i

Low — lax-mid — tense-mid — high Veneto Southern Umbro    

Ischia

There is no dialect in which only the low vowel or only the lax vowels raise. (5) Shifting on the height scale—unattested dialectal typology a. Chain shift: a → ε → e → i b. Height scale: Low — lax-mid — tense-mid — high

If we take into consideration that the triggering vowel is a high tense vowel, and hypothesize that the goal is to achieve complete height identity, i.e. a high vowel, raising a low vowel one step only moves it minimally towards the target and the

282 

 Martin Krämer

resultant vowel has missed the target by two features, since it is still not [high] and still not [ATR]. Raising the lax mid vowels by one step misses the target only by one feature, [high]. Raising the tense mid vowel is the most satisfactory operation, since it creates a [high] vowel, complete height identity with the trigger is achieved. Thus, the incentive of raising decreases the lower the target vowel is.

3.2 A markedness paradox As indicated above, dialects with metaphony also have neutralization of the ATR contrast in mid vowels in unstressed syllables. If this is feature delinking and metaphony is feature spreading, these two processes require opposing markedness relations. For unstressed syllable neutralization, going higher is becoming less marked, for metaphony, the stressed, i.e. prominent vowel, has to be raised in the presence of an unstressed high vowel, because being high is marked in unstressed position (according to Walker 2005, 2011). The claim that high vowels are marked in unstressed syllables is actually controversial. Unstressed vowel reduction tends to reduce vowel systems to schwa or the corner vowels [i, u, a], or high vowels and schwa (see e.g. Crosswhite 2001; de Lacy 2006). In general, there is an inversely correlated sonority bias for stressed and unstressed positions. High vowels and schwa are cross-linguistically most welcome in unstressed syllables, while low vowels attract stress in qualitysensitive systems (Kenstowicz 1997). If we analyze unstressed vowel reduction as direct licensing (F can only be expressed in the prominent position and is delinked elsewhere) and metaphony as spreading, i.e. Walker’s indirect licensing, we have the configurations in (6a, b). (6) /ɛ/ → [e] a. Unstressed syllable reduction

b. Metaphony

x [F]

x



 

y [F]

These are two mutually contradicting ways of deriving an [e] from an /ɛ/. The same change is brought about once by delinking a marked feature (cf. a) and once by adding a marked feature (cf. b). One way of distinguishing /e/ and /ɛ/ is by means of the feature ATR (Advanced Tongue Root). /e/ is marked [+ATR] or just [ATR], and /ɛ/ is [−ATR] or unmarked. This analysis entails that unstressed vowel reduction is addition of [ATR] to unmarked /ɛ/. Reduction and other neutralization processes are usually assumed to produce a structure that is less marked than the input, while assimilation usually results in an increase in markedness



Metaphonic chain shifts, vowel height and markedness 

 283

in the targeted segment (see, e.g. Rice 2007). This suggests that the feature active in Italian is rather [RTR] (Retracted Tongue Root) and /ɛ/ is marked [RTR], while /e/ is unmarked. In this analysis, unstressed syllable reduction then can be captured as delinking of [RTR], which results in a less marked vowel (likewise with binary [±RTR], assuming delinking of the marked and subsequent insertion of the default negative value for mid vowels). However, this analysis of reduction causes a problem for the analysis of metaphony. When dealing with metaphony it is desirable to derive an [e] from an /ɛ/ by spreading, while we just derived an [e] from an /ɛ/ by delinking. High vowels cannot be marked for [ATR] if the active tongue root feature in Italian is [RTR]. There is no need for both features in an analysis of a vowel inventory that has a tongue root contrast in mid vowels only. Thus, if the [RTR] analysis of reduction is correct there is nothing in the high vowels that can spread to an /ɛ/ to turn it into an [e]. One option is to assume that the feature is binary and the spreading feature is [−RTR]. We can analyze both processes then: reduction is the delinking of the marked valued feature [+RTR] and consecutive insertion of the default unmarked [−RTR], and metaphonic raising of a lax mid to a tense mid vowel is spreading of [−RTR]. Spreading of a feature with a negative value, though, is not as elegant an analysis as one would strive for, since, as mentioned above, spreading is generally expected/assumed to create a shift to the marked. Alternatively, we could analyze metaphony as identity licensing, establishing a correspondence relation between the post-tonic and the stressed vowel (e.g. Syntagmatic Identity limited to the domain of the foot). With this analysis of metaphony the orientation of the markedness asymmetry does not matter since identity is required regardless of the value of the feature, and reduction can be analyzed as a change towards the unmarked independently of metaphony. Another problem for an analysis of metaphony as feature licensing by spreading is that in the most economic analysis the high vowels do not have any ATR or RTR feature to spread to the vowels for the following reason. Tongue root position is not contrastive in high vowels in Italian, there is no /ɪ/ or /ʊ/. Accordingly the high vowels are contrastively specified for [+high] or privative [high], but do not need a specification for [RTR] or [ATR]. The same problem arises with the raising of the low vowel /a/ to [ɛ]. The feature that presumably changes is [±low]. The stressed vowel is still [−high] and [+RTR] (or [−ATR]), conflicting with the [+high] and [−RTR] specifications of the triggering vowel. Thinking derivationally one could conclude that metaphony applies late, after non-contrastive features are filled in. However, often the trigger is realized as schwa or deleted. Thus, metaphony applies before the reduction or deletion

284 

 Martin Krämer

of the trigger. However, this aspect of the phenomenon is dealt with by TorresTamarit & Linke (this volume).2

3.3 Summary and discussion Metaphonic raising, especially if understood as height licensing, is problematic for two reasons. On the one hand, the raising by one step on the height scale does not improve the licensing of the post-tonic vowel significantly, especially if the step involves changing a feature that is not even contrastive in the post-tonic high vowel, such as [(±)RTR] or [(±)low]. On the other hand, this chain shift is problematic because the vowel height dimension is usually not analyzed as a multi-valued feature (7a) (though see Krämer 2009), but rather as a set of binary or privative height features (7b). (7) Height multi-valued or privative a. Multi-valued height i

[high]

b. Privative height x height

e ε

[RTR]

a

[low]



[RTR]

[high/low]

Considering metaphonic raising and raising in unstressed syllables together creates a markedness paradox under standard assumptions. Metaphonic raising is often analyzed as autosegmental spreading, while neutralizing raising in unstressed syllables is most straightforwardly analyzed as autosegmental feature delinking. One process thus creates an [e] or [o] from /ɛ/ or /ɔ/, respectively, by

2 A reviewer points out that this would be a very abstract derivational analysis and that the usual account would be to posit a floating feature (Finley 2009). In the case of metaphony it is not clear which feature should be posited. It cannot be just [high] since most of the observed raising does not lead to the realization of that feature. One would thus have to explain why the floating feature is not realized anyway. Furthermore, floating features complicate the representational theory, since they are entities at the level of segments, even though they are dependent on segments for their realization, while otherwise they can be considered mere attributes to segments, as implied in the Identity-IO(F) approach to faithfulness.



Metaphonic chain shifts, vowel height and markedness 

 285

spreading (addition of information) and the other creates an [e] or [o] from /ɛ/ or /ɔ/, respectively, by delinking (subtraction of information). The tense mid vowels thus have to be at the same time richer and poorer in feature content than the lax mid vowels. The tense mid vowels should be unmarked because of the unstressed mid vowel raising pattern that changes RTR to ATR. If the two processes are of separate types the paradox is resolved. Thus, I conclude that unstressed mid vowel raising is delinking while metaphonic raising is identity licensing. In the next section I will analyze the chain shift as resulting from locally conjoined faithfulness constraints on vowel height, which militate against changes that involve more than one height feature. The process is triggered by Syntagmatic Identity constraints demanding identity in height between the final/post-tonic vowel and the stressed vowel.

4 Local conjunctions for height and the implicational character of the typology In this section we will look at primitive IO-Identity(F) constraints for vowel height features and their potential combinations. These and Syntagmatic Identity constraints are the two sets of constraints that are in direct conflict in metaphony. While S-Ident(F) constraints, which will be discussed in the next section, demand identity between correspondent segments within a string (i.e. the final or post-tonic and the stressed vowel), satisfaction of this demand is opposed by IO-Ident(F) constraints. They are thus central for any analysis. Combination of constraints of the same type to complex constraints in LC yields “weaker” i.e. more restricted versions of the basic constraints, but it also increases the number of constraints opposing changes between input and output. Since they are ranked higher than their component constraints (Paninian ranking relation of special above general; Prince & Smolensky 1993/2004), they prevent excessive violation of the lower ranked constraints, in this case they ban excessive unfaithfulness to height. In the case of IO-Ident(F) constraints we will see that these complex constraints have similar effects as Gnanadesikan’s (1997) IdentAdjacent(vowel height) constraints and block excessive feature changes. It is not necessary to postulate a new family of constraints, such as Ident-Adjacent (V-height), and a height scale as a phonological entity, or a new set of features of a feature class if the effects can be derived from an independently motivated type of constraint on independently motivated features.

286 

 Martin Krämer

Kirchner (1996) analyzes the three-step raising pattern in Nzɛbi (structurally the same as the Ischia metaphony pattern) with LCs of Parse(F) constraints in the containment model of faithfulness. The analysis and discussion here are based on his approach, though the analysis is recast in the correspondence model of faithfulness (McCarthy & Prince 1995, 1999). McCarthy & Prince (1995, 1999) introduced IO-Ident(F) constraints as one family of correspondence constraints. While they defined them with reference to binary features, the nature of features as binary or privative is a matter of debate but the choice is not crucial here. The only aspect of this issue that is important here is that we do not need scalar or multi-valued features. I will thus define the constraints in a way that treats features as privative.3 Though nothing hinges on this choice. The standardly assumed set of height features consists of [high], [low], and [ATR] or [RTR] (see Krämer 2003a for discussion and references). (8) Input-Output Identity to Vowel Height a. IO-Identity[F]: Correspondent segments in input and output have identical specifications of the feature F. If xy and x is [F], then y is [F] and if x is [ ] then y is [ ], and vice versa. b. IO-Ident[high]: Correspondent segments have identical specifications of the feature [high]. c. IO-Ident[low]: Correspondent segments have identical specifications of the feature [low] d. IO-Ident[RTR]: Correspondent segments have identical specifications of the feature [RTR]. These constraints can be conjoined in local constraint conjunctions. The mechanism is defined in (9). (9) Definition of Local Conjunction (LC) (Smolensky 1993) The LC of C1 and C2 in domain D, [C1 & C2]D, is violated when there is some domain of type D in which both C1 and C2 are violated. For our conjunctions the domain D is defined as the segment, a by and large uncontroversial domain. The operation yields four new constraints, which are given below. (a) through (c) are conjunctions of each single constraint with every other single constraint and (d) is the conjunction of all three constraints.

3 See Blaho (2008) and Krämer (2009, 2012) for some discussion of privativity versus binarity and the function and definition of faithfulness constraints in relation to feature type.

Metaphonic chain shifts, vowel height and markedness 



 287

(10) Height faithfulness conjunctions a. IO-Ident[high]&IO-Ident[RTR]: Assign one violation mark for every segment that violates both constraints. b. IO-Ident[low]&IO-Ident[RTR]: Assign one violation mark for every segment that violates both constraints. c. IO-Ident[high]&IO-Ident[low]: Assign one violation mark for every segment that violates both constraints. d. IO-Ident[high]&IO-Ident[RTR]&IO-Ident[low]: Assign one violation mark for every segment that violates all three constraints. All three basic constraints can be violated in metaphony. The low vowel violates Ident[low] by raising to mid lax. The mid lax vowel violates Ident[RTR] by raising to mid tense. And the mid tense vowel violates Ident[high] by raising to high. However, no single vowel ever violates all three of them or even only two of them. Thus, we can assume that in the most extensive pattern, in which every non-high vowel raises one step, the constraint(s) responsible for the change rank between the simple constraints and their conjunctive derivatives. The latter are always top-ranked. The constraints triggering raising are indicated by the dark column in (11) and in the following tableaux in this section, and will be replaced by proper constraints in the next section. We suspect that whatever is in that black box of constraints prefers complete raising to high rather than just one step. Therefore, a tense-mid vowel followed by a high one incurs one mark, lax-mid followed by high incurs two marks, and low followed by high incurs three violations in all tableaux in this section.

a. /e-i/ → [e-i]

*

 b. /e-i/ → [i-i]

*

 c. /ɛ-i/ → [ɛ-i]

**

 d. /ɛ-i/ → [e-i]

*

e. /ɛ-i/ → [i-i]

*

*

*

 f. /a-i/ → [a-i]

***

 g. /a-i/ → [ɛ-i]

**

*

*

*

*

*

*

h. /a-i/ → [e-i] i. /a-i/ → [i-i]

IO-Id[high]

IO-Id[RTR]

IO-Id[low]

Get high!

IO-Id[low]& IO-Id[RTR]

IO-Id[high]& IO-Id[RTR]

IO-Id[high]& IO-Id[low]

IO-Id[high]& IO-Id[low]& IO-Id[RTR]

(11) Reinforcing IO-faithfulness

* *

*

*

*

*

*

288 

 Martin Krämer

As we can see in this tableau, there is no conjunction that is violated by the one-step change that raises the tense mid vowels to high. The lax mid vowel violates one LC when raising to high by changing two features. The low vowel has to pay by far the highest cost for raising all the way up the height scale. Since such long distance raising is not attested in Italian dialects we can assume that the set of LCs is top-ranked in all dialects. The same incremental effect could not be achieved by ranking one or two of the basic constraints above the black box. First, if we deny the existence of the LCs and randomly allow any of the three constraints to outrank the black box we predict patterns in which only the lax vowels raise. The LCs, at least as long as they are ranked above the raising-inducing force, block complete raising to high from the positions that are more than one change away, i.e. /ɛ/ and /a/. (12) illustrates what happens in the absence of the LCs and high ranking of a randomly chosen IO-Ident constraint, IO-Ident(high). In this hypothetical dialect /a/ and /ɛ/ raise to [e] while /e/ does not change. Another option is that the metaphony constraints outrank all the Identity constraints, resulting in unattested long distance raising (e.g. /a/ → [i/u]). We will see shortly that the simple constraints cannot be ranked freely with respect to each other. (12) No LCs: Non-attested metaphony IO-Id[high] F a. /e-i/ → [e-i]

b. /e-i/ → [i-i]

Get high!

IO-Id[low]

IO-Id[RTR]

* *!

c. /ɛ-i/ → [ɛ-i]

**!

F d. /ɛ-i/ → [e-i]

*

*



e. /ɛ-i/ → [i-i]



 f. /a-i/ → [a-i]

**!*



 g. /a-i/ → [ɛ-i]

**!

*

*

*

*

*

*

*!

F h. /a-i/ → [e-i]

i. /a-i/ → [i-i]

*!

*

However, with the LCs on top, ranking Ident[low] and Ident[RTR] high results in more attested patterns. To account for the pattern in which only /e/ raises, we have to rank IO-Ident[low] and IO-Ident[RTR] above the metaphony constraints and IO-Ident[high].

Metaphonic chain shifts, vowel height and markedness 



 289



a. /e-i/ → [e-i]

*!

F b. /e-i/ → [i-i]

*

F c. /ɛ-i/ → [ɛ-i]

d. /ɛ-i/ → [e-i]



e. /ɛ-i/ → [i-i]

** *! *!

g. /a-i/ → [ɛ-i]



h. /a-i/ → [e-i]



i. /a-i/ → [i-i]

*

*

       F f.  /a-i/ → [a-i]

IO-Id[high]

Get high!

IO-Id[RTR]

IO-Id[low]

IO-Id[low]& IO-Id[RTR]

IO-Id[high]& IO-Id[RTR]

IO-Id[high]& IO-Id[low]

IO-Id[high]& IO-Id[low]& IO-Id[RTR]

(13) The most restricted pattern: only /e/ raising

* ***

*! *!

*

*

**

*!

*

*

*

*

*

* *

However, ranking relations between the simple height constraints are not entirely free. To understand this it is instructive to reconsider unstressed mid vowel tensing. The unstressed syllable neutralization pattern changes RTR in mid vowels, but it does not change the features [high] or [low]. If lax mid vowels are unwelcome in unstressed syllables, for example because of a constraint like *s̆/{ɛ, ɔ} (Crosswhite 2001; Walker 2005, 2011) they could be changed to tense mid vowels or to high, which would imply a change in [RTR] and a change in [RTR] as well as [high], respectively, or to low, which would necessitate changing the feature [low]. (14) Repairing lax mid vowels a.  /ɛ/ → [e]  = [RTR] → [ ] = 1 change b.  /ɛ/ → [i]  = [RTR] → [high] = 2 changes c.  /ɛ/ → [a]  = [RTR] → [low, RTR] = 1 change We can thus assume that raising of /ɛ/ and /ɔ/ to [e] and [o], respectively, rather than lowering them to [a], in unstressed syllables, requires IO-Ident[low] to be ranked above IO-Ident[RTR]. Raising the two lax mid vowels to [high] would be a violation of two Identity constraints on height and thereby also of one LC and is bounded by the other two repair candidates given the current constraint set. We thus cannot say anything about the ranking of Ident[high], independently of what we know from the metaphonic alternations, unless we look a bit further afield.

290 

 Martin Krämer

De Lacy (2006) compiled an interesting typology of reduced unstressed vowel inventories that seems to strive towards sonority reduction and achieves it only to varying degrees. His table summarizing the typology (de Lacy 2006: 310) is reproduced in (15). (15) Harmonically contiguous unstressed syllable inventories4 a,æ

√

ɛ, ɔ

√ √

e, o

i, u

ə

Language

√ √ √

√ √ √ √

√ √ √ √ √

New Zealand English Central Catalan Sri Lankan Portuguese Creole Siuslawan Māori

Among non-contiguous systems we find languages like Russian in which the mid vowels change but the low one does not, resulting in the basic vowel triangle i, u, a, while the same language also displays reduction of the inventory to i, u, ə (see, e.g. Crosswhite 2001 for more details). Thus, even though the low vowels are the most marked in sonority reducing neutralization ­patterns they also are most resistant to change. With these pieces of information we can assemble a universal hierarchy (or formalize it as a stringent set of constraints on vowel height in the same way as de Lacy’s 2006 stringent constraints on consonantal place faithfulness). This set then complements whichever markedness constraint set one assumes to drive vowel reduction (for example de Lacy’s 2006 *−Dw ≥ {ɛ, ɔ} (sloppily said “unstressed vowels should not be higher in sonority than or equal to ɛ and ɔ”)). Faithfulness to the most marked, i.e. [low], is placed highest in the system, while the ranking between the other two faithfulness constraints has to remain undetermined. The resulting partial hierarchy is given in (16). (16) Ranking of height related primitive Identity constraints IO-Ident[low] IO-Ident[RTR]

IO-Ident[high]

4 For references on the individual languages consult de Lacy (2006).

Metaphonic chain shifts, vowel height and markedness 



 291

(17) Only mid unstressed vowels raise IO-Id[low]

*−Dw ≥ {ɛ, ɔ}

IO-Id[RTR]

a. /ɛ/ → [i]

*

F b. /ɛ/ → [e]

*



c. /ɛ/ → [ɛ]



d. /ɛ/ → [a]

*!



e. /a/ → [i]

*!

*



f. /a/ → [e]

*!

*



g. /a/ → [ɛ]

*!

IO-Id[high] *!

*! *

* *

*

F h. /a/ → [a]

*

With this background information on ranking we can return to our typology of metaphonic raising patterns. If only the highest ranked simple IO-Ident constraint, IO-Ident[low], outranks the metaphony-inducing constraints we describe the pattern in which only the low vowel refuses to raise, as shown in the next tableau. (In the following tableaux the LCs and the candidates excluded by them are removed to make tableaux faster accessible.) (18) The medium restricted pattern: /e/ and /ɛ/ raising IO-Id[low]

a. /e-i/ → [e-i]

Get high!

IO-Id[RTR]

*!

F b. /e-i/ → [i-i]



*

c. /ɛ-i/ → [ɛ-i]

**!

F d. /ɛ-i/ → [e-i]

*

F e. /a-i/ → [a-i]

***

f. /a-i/ → [ɛ-i]

IO-Id[high]

*!

*

**

If Ident[low] is ranked below the black box, and still kept above Ident[RTR], as required by the unstressed mid vowel neutralization pattern, we get the most expansive metaphony pattern.

292 

 Martin Krämer

(19) The most expansive pattern: /e/, /ɛ/, and /a/ raising Get high!

a. /e-i/ → [e-i]

IO-Id[low]

IO-Id[RTR]

*!

F b. /e-i/ → [i-i]

c. /ɛ-i/ → [ɛ-i]

F d. /ɛ-i/ → [e-i]

f. /a-i/ → [a-i]

F g. /a-i/ → [ɛ-i]

IO-Id[high]

* **! *

*

***! **

*

A full shift of all vowels to [high] would require a further move of the constraints requiring raising beyond the LCs. It is not reasonable to exclude this entirely just because it is not found in Italian dialects. The grammar here just predicts such patterns to be extremely unlikely. A more interesting issue is the analysis of the Lombardian dialects, in which only the lax mid vowels raise, but neither the tense ones nor the low ones move. Interestingly this is a change in the same feature as observed in unstressed mid vowel neutralization. In the current system this would require ranking of Ident[high] and Ident[low] above Ident[RTR] and the metaphony-­ inducing constraints. (20) Lombardian: only /ɛ/ raising IO-Id[low]

IO-Id[high]

F a. /e-i/ → [e-i]

b. /e-i/ → [i-i]



c. /ɛ-i/ → [ɛ-i]

Get high! *

*! **!

F d. /ɛ-i/ → [e-i]

*

F f. /a-i/ → [a-i]

***



g. /a-i/ → [ɛ-i]

IO-Id[RTR]

*!

*

**

While the discussion on mid vowel raising in unstressed syllables resulted in the conclusion that the ranking should be fixed for Italian, with IO-Ident[low] dominating IO-Ident[RTR] and IO-Ident[high] and remained inconclusive about the other two IO-Identity constraints, the existence of the Lombardian pattern reconfirms that there is no fixed ranking relation between IO-Ident[high] and IO-Ident[RTR].

Metaphonic chain shifts, vowel height and markedness 



 293

However, the “pan-Italian” or maybe universally fixed relation between IO-Ident[low] and the other two height Identity constraints explains why we do not get certain other patterns, that would result from illegal higher positioning of these two constraints, as illustrated by the following two tableaux. (21) Unattested pattern I: Only /a/ raises IO-Id[RTR]

IO-Id[high]

Get high!

F a. /e-i/ → [e-i]

*

b. /e-i/ → [i-i]

*!

F c. /ɛ-i/ → [ɛ-i]

d. /ɛ-i/ → [e-i]



f. /a-i/ → [a-i]

IO-Id[low]

** *!

* ***!

F g. /a-i/ → [ɛ-i]

**

*

(22) Unattested pattern II: Only /e/ does not raise IO-Id[high] F a. /e-i/ → [e-i]

b. /e-i/ → [i-i]



c. /ɛ-i/ → [ɛ-i]

F d. /ɛ-i/ → [e-i]

f. /a-i/ → [a-i]

F g. /a-i/ → [ɛ-i]

Get high!

IO-Id[low]

IO-Id[RTR]

* *! **! *

*

**!* **

*

In the tableaux in this section, I marked violations in the black box as follows. Tense-mid followed by high incurs one mark, lax-mid followed by high incurs two marks, and low followed by high incurs three violations. In some cases these violations were crucial to pick the correct output. We enter the black box now to see if these violation marks are warranted.

5 Into the black box: Syntagmatic Identity licensing One peculiarity of Italian metaphony is that only the high vowels—in some dialects only one high vowel—cause Umlaut. If it were only for the [high] feature in

294 

 Martin Krämer

the post-tonic vowel to cause raising, then changing the mid lax vowels to mid tense or the low vowel to mid lax, as observed, would not give a candidate an advantage over faithful candidates. That the lax vowels change nevertheless shows that in some dialects only [high] is involved, in some [RTR] is active too, and in the last group all three height features. Since the conjoined IO-F constraints block complete height assimilation, these changes are a failing struggle for perfection that is reduced to avoidance of the worst. The definition of S-Identity provided by Krämer (2003a) is given in (23), together with compatible informal definitions of the individual constraints for the three height features. Each of these alone causes harmony patterns involving the respective feature (tongue root harmony, as observed in many Nilo-Saharan languages and height harmony as observed in many Bantu languages) (Krämer 2003a and references there, or any more recent decent handbook overview article on vowel harmony). (23) Vowel Height Syntagmatic Identity a. S-Identity[F] (Krämer 2003a: 73): Let x be an entity of type T in representation R and y be any adjacent entity of type T in representation R, if x is [αF] then y is [αF]. Where T is a segment, mora, syllable, or foot. (A segment/mora/syllable/foot has to have the same value for feature F as the adjacent segment/mora/syllable/foot in the string.) b. S-Ident[high]: Correspondent segments in a surface representation have identical specifications of the feature [high]. c. S-Ident[low]: Correspondent segments in a surface representation have identical specifications of the feature [low] d. S-Ident[RTR]: Correspondent segments in a surface representation have identical specifications of the feature [RTR]. The tableau below shows how the violations of these individual constraints match the marks registered in the black column in previous tableaux. (24) Shedding light on the black box S-Ident[low] a. a-i b. ɛ-i c. e-i

*

S-Ident[RTR]

S-Ident[high]

Get high!

*

*

***

*

*

**

*

*

d. i-i To accommodate the specific type of harmony we find in metaphony one would have to further specify the S-Identity constraints as limited to the



Metaphonic chain shifts, vowel height and markedness 

 295

domain of the foot or referring to the involved vowels. After all, it is only the (historically) high vowels that trigger metaphony and there is no lowering of stressed high vowels to license non-high final vowels. The imbalance that the post-tonic or final (weak) vowel is dominant and the stressed and intermediate vowels assimilate can be due to positional faithfulness, as proposed by Walker (2011, this volume). See also Krämer (2003a,b) on right edge faithfulness in the context of regressive vowel harmony. (25) gives the basic metaphony ranking in (a), and in (b) the internal ranking of the atomic IO-Identity constraints that are placed around the S-Identity block in various ways in different dialects. (25) General metaphony ranking a. LCs, Faith-s]PWd ›› S-Ident(height)/foot ›› Faith-ˈs, Faith- PWd[s b. IO-Ident[low] ›› IO-Ident[RTR], IO-Ident[high] However, the details of the analysis accounting for directionality and dominance are not of major concern in this chapter, so I will leave these issues here and move on to the discussion of more general repercussions.

6 Beyond metaphony In this section I briefly sketch how LCs of height faithfulness potentially account for other scalar phenomena as well. Gnanadesikan (1997) discusses coalescence in Sanskrit (Gonda 1966) as a case which requires Ident-Adjacent(height) constraints. In Sanskrit, junctures of a low and a following high vowel are turned into a long mid vowel. Thus, /a/ + /i/ is realized as [eː] and /a/ + /u/ is realized as [oː]. This pattern emerges in avoidance of diphthongs, which are not found in Sanskrit. Furthermore Sanskrit does not have any short mid vowels. One can thus assume that all mid vowels in Sanskrit are derived from sequences of low and high vowels (McCarthy 2005). Changing an input sequence of [high] and [low] into a mid vowel, i.e. changing both vowels, seems to be an exaggerated repair if changing one vowel would suffice to avoid the marked diphthong. The low vowel could raise or the high vowel could lower to create a long high or long low vowel. However, this would involve two feature changes in one and the same segment, while lowering each of the two vowels by only one step causes two feature changes as well, but each in a different segment. Thus, changing one vowel a lot violates one of our LCs on height, while changing each vowel only once does not. The individual vowels are indexed with a subscript number in tableau (26), which are repeated in the cells under the constraints to associate violation marks with vowel segments.

296 

 Martin Krämer

(26) Sanskrit coalescence *Diphthong a. /a1-i2/ − a1i2

IO-Id[high]& IO-Id[high] IO-Id[low] IO-Id[low]

*!

b. /a1-i2/ − i1i2

*1!

*1

*1

c. /a1-i2/ − a1a2

*2!

*2

*2

*2

*1

F d. /a1-i2/ − e1e2

As a reviewer pointed out, the analysis would be different if this were actual coalescence. As the observant reader surely has detected, the vowels do not merge into one, they only each change one feature, resulting in a fake long vowel. If the mid long vowel in candidate (d) had been a correspondent of vowel 1 and vowel 2 simultaneously it would also violate the LC. However, the coalescence candidate would violate further faithfulness constraints, such as Linearity or Integrity (see McCarthy & Prince for definitions), that militate against changes in the linear order of segments between input and output, rendering a candidate with coalescence inferior to the forms given in the tableau. Going beyond the contrast dimension of vowel height one can generalize the approach by replacing our three height features by abstract features that represent the steps on a scale, e.g. manner or sonority of consonants. Thus, for any dimension of contrast we can assume there to be LCs of the Identity constraints catering for the individual features on the scale. However, future research has to reveal in how far the LC approach can be extended to other phenomena showing scalar effects, such as consonant lenition, and thereby to other class nodes and their dependent features.

7 Conclusion The chain shift nature of Italian metaphony has been explained in this paper as an effect of Local Constraint Conjunctions.5 IO-Ident[F] constraints on vowel height have been systematically conjoined and shown to derive the observed pattern that

5 See Łubowicz (2011) for an alternative approach to chain shifts in terms of Contrast Preservation though.



Metaphonic chain shifts, vowel height and markedness 

 297

a vowel never changes more than one single height feature in metaphonic raising. Metaphonic raising has been considered in the context of unstressed mid vowel tensing and this has given us a convincing argument to analyze Italian metaphony as identity licensing rather than indirect licensing in Walker’s (2011) licensing typology. If metaphony emerges as licensing by correspondence and unstressed mid vowel tensing as markedness reduction by feature delinking the markedness paradox is avoided that arises in an analysis that considers metaphonic raising to be spreading and reduction as delinking. Both processes apparently derive the same output from the same input by opposite means (feature addition versus feature deletion, respectively). Identity licensing has been straightforwardly implemented here as due to a set of Syntagmatic Identity constraints (Krämer 1998, 2001, 2003). This indicates that there is a close relation between Romance-style metaphony and vowel harmony, as found in unrelated languages. Considering unstressed vowel reduction also led to a deeper insight in the relevant rankings of simple faithfulness con­ straints on vowel height that shape the micro-typology of metaphony targets observed in dialects of Italian, i.e. the implicational hierarchy of vowel raising. LC, Syntagmatic Identity, as well as the features used in the analysis here are all independently motivated by theoretical considerations and evidence completely outside the scope of this paper. The approach proposed here is thus more economic than any analysis that makes theoretical assumptions specific to ­metaphony. Future research is needed to investigate how far systematic LCs of faithfulness constraints as well as predetermined rankings or stringent ranking relations between simple faithfulness constraints within a single dimension of contrast play a role in shaping other phonological patterns.

References Beckman, Jill N. 1997a. Positional faithfulness. PhD dissertation. Amherst: University of ­Massachusetts. Beckman, Jill N. 1997b. Positional faithfulness, positional neutralisation and Shona vowel harmony. Phonology 14: 1–46. Blaho, Sylvia. 2008. The syntax of phonology. A radically substance-free approach. PhD dissertation. Tromsø: University of Tromsø. Calabrese, Andrea. 1998. Metaphony revisited. Rivista di Linguistica 10: 7–68. Calabrese, Andrea. 2011. Metaphony in Romance. In van Oostendorp, Marc, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.), The Blackwell companion to phonology, 2631–2762. Oxford: Blackwell. Cole, Jennifer 1998. Deconstructing metaphony. Rivista di Linguistica 10 (1): 69–98. Crosswhite, Catherine. 2001. Vowel reduction in optimality theory. New York and London: Routledge.

298 

 Martin Krämer

Crowhurst, Megan, and Marc Hewitt. 1997. Boolean operations and constraint interactions in Optimality Theory. MS thesis, University of North Carolina at Chapel Hill and Brandeis University. de Lacy, Paul. 2006. Markedness: reduction and preservation in phonology. Cambridge: ­Cambridge University Press. Gnanadesikan, Amalia. 1997. Phonology with ternary scales. PhD dissertation. Amherst: ­University of Massachusetts. Finley, Sara. 2009. Morphemic harmony as featural correspondence. Lingua 119: 478−501. Frigeni, Chiara. 2003. Metaphony in Campidanian Sardinian: a domain-based analysis. Toronto Working Papers in Linguistics 20: 63–91. Kenstowicz, Michael. 1997. Quality-sensitive stress. Rivista di Linguistica 9 (1): 157–188. Kirchner, Robert. 1996. Synchronic chainshifts in optimality theory. Linguistic Inquiry 27: 341–350. Krämer, Martin. 1998. A correspondence approach to vowel harmony and disharmony. SFB 282 Working Paper Nr. 107. Heinrich-Heine-Universität Düsseldorf. Krämer, Martin. 2001. Yucatec Maya vowel alternations − harmony as syntagmatic identity. Zeitschrift für Sprachwissenschaft 20 (2): 175-217. Krämer, Martin. 2003a. Vowel harmony and correspondence theory. Studies in generative grammar 66. Berlin, New York: Mouton de Gruyter. Krämer, Martin. 2003b. What’s wrong with the right side? Edge (A)symmetries in Phonology and Morphology. Unpublished manuscript, University of Ulster. Krämer, Martin. 2009. The phonology of Italian. Oxford: Oxford University Press. Łubowicz, Anna. 2002. Derived environment effects in optimality theory. Lingua 112: 243−280. Łubowicz, Anna. 2011. Chain shifts. In Marc van Oostendorp, Colin Ewen, Beth Hume and Keren Rice (eds.), Companion to phonology. Wiley-Blackwell. Maiden, Martin. 1991. Interactive morphonology. Metaphony in Italy. London: Routledge. Mascaró, Joan. (this volume). The typology of metaphony/stress dependent harmony. McCarthy, John J. & Alan S. Prince. 1995. Faithfulness and reduplicative identity, UMOP 18: 249–384. McCarthy, John J. & Alan S. Prince. 1999. Faithfulness and identity in prosodic morphology. In R. Kager, H. van der Hulst, and W. Zonneveld (eds.), The prosody-morphology interface, 218−309. Cambridge: Cambridge University Press. McCarthy, John J. 2005. Taking a free ride in morphophonemic learning. Catalan Journal of Linguistics 4: 19−55. Moreton, Elliot & Paul Smolensky. 2002. Typological consequences of local constraint conjunction. WCCFL 21. Prince, Alan S. & Paul Smolensky. 1993/2004. Optimality theory: constraint interaction in generative grammar. Technical Report of the Rutgers Center of Cognitive Science, RUCCS-TR 2, Rutgers University, New Brunswick and Technical Report of the University of Colorado Computer Science Department, CU-CS-696-95, University of Colorado, Boulder. Publication in 2004: Malden (Mass.), Oxford: Blackwell Publishers. Rice, Keren. 2007. Markedness in phonology. In Paul de Lacy (ed.), The Cambridge handbook of phonology, 79−97. Cambridge: Cambridge University Press. Savoia, Leonardo & Martin Maiden. 1997. Metaphony. In Martin Maiden & Mair Parry (eds.), The dialects of Italy, 15–25. London: Routledge.



Metaphonic chain shifts, vowel height and markedness 

 299

Smolensky, Paul. 1993. Harmony, markedness, and phonological activity. Paper presented at Rutgers Optimality Workshop-1, Rutgers University, New Brunswick, NJ. [Rutgers Optimality Archive, http://ruccs.rutgers.edu/roa.html.] Torres-Tamarit, Francesc & Kathrin Linke. (this volume). Opacity in vowel merger – metaphony interactions. Walker, Rachel. 2000. Long-distance consonantal identity effects. In Proceedings of WCCFL 19, pages 532−545. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language & Linguistic Theory 23: 917–989. Walker, Rachel. 2011. Vowel patterns in language. Cambridge: Cambridge University Press. Zoll, Cheryl. 1996/1998. Parsing below the segment in a constraint-based framework. PhD dissertation, University of California, Berkeley. [Published by CSLI Publications, Stanford, 1998.]

Rachel Walker

Positional prominence and consonantal interactions in metaphony and post-tonic harmony* University of Southern California

Abstract: This study examines consonantal interactions with two types of vowel harmony in dialects of Italy. In metaphony a stressed vowel undergoes height harmony with a post-tonic vowel. In post-tonic harmony, post-tonic vowels harmonize with the final vowel. These harmonies exhibit distinct typologies of consonantal blocking. Metaphonic harmony may selectively operate across ­palatals and /ll/, or it may be blocked by all coda consonants. However, posttonic harmony may be restricted to transliquid contexts. The insight pursued here is that target contexts that better support cues for the harmonizing features are more prone to allow metaphony. This is analyzed using positional licensing constraints. Post-tonic harmony is proposed to be driven by a constraint that promotes unbounded harmony from vowels in a weak context. Assuming that consonants participate in harmony, nonliquid consonants are proposed to block post-tonic harmony because of a constraint that penalizes features associated across dissimilar segments. The analysis thus posits two sources for blocking. Blocking by nonliquid consonants is attributed to their lack of similarity to vowels, consistent with several patterns of consonantal blocking in the world’s languages. In ­contrast, the patterns of consonantal blocking in metaphony emerge as an ­epiphenomenon of contexts in positional licensing constraints.

1 Introduction In this paper, I consider two types of vowel harmony controlled by the final vowel in Romance dialects of Italy. In metaphony, the vowel in the stressed syllable assimilates in height features to a post-tonic high vowel, classically an inflectional vowel in the final syllable. In post-tonic vowel harmony, vowels following the syllable with main stress assimilate for some or all features with the vowel

* For comments on this work I am grateful to Aaron Kaplan and an anonymous reviewer, and I owe thanks to Mario Saltarelli for assistance with the data. Any errors are my own.

302 

 Rachel Walker

in the final syllable. In some dialects, certain consonants block harmony. These patterns of consonantal interactions are studied to cast light on the nature of the harmony systems. Selective consonant blocking in post-tonic harmony is found to fall in line with the characteristics of blocked feature sharing in vowelconsonant-vowel contexts in diverse languages of the world. Specifically, consonants that are less similar to vowels are more prone to block feature sharing. However, patterns of consonant blocking in metaphony show distinct properties. The observation pursued here is that consonants are prone to block metaphony in contexts where they tend to reduce the robustness of perceptual cues in the target stressed position. Consonants that serve to support cues for the harmonizing features in the stressed position are more likely to allow the transmission of metaphony. These different properties of consonant blocking are in keeping with an understanding of metaphony as driven by positional licensing, in contrast to post-tonic harmony. The organization of this paper is as follows. In section 2, I outline the wellknown systems of metaphony and post-tonic harmony in the dialect of Servigliano, and review a licensing-based account of their analysis, with metaphony analyzed as a positional licensing phenomenon and post-tonic harmony as a phenomenon of maximal licensing. In section 3, I introduce related vowel harmonies, which show the additional complication of selective consonant blocking. The analysis of these phenomena is taken up in section 4, dealing in turn with blocking consonants in post-tonic harmony and metaphony. In section 5, an ­alternative approach to post-tonic harmony driven by positional licensing is considered and rejected. In section 6, I present the conclusion.

2 Background: harmony in Servigliano The Marche dialect of Servigliano presents patterns of metaphony and post-tonic vowel harmony that have been widely studied. The phenomena are reported by Camilli (1929) and their interpretation is informed by numerous generative studies (Kaze 1989; Maiden 1995; Nibert 1998; Canalis 2009; Mascaró 2011; Walker 2011). In this section, the data and principal elements of Walker’s (2011) analysis are reviewed to lay groundwork for related vowel patterns introduced in §3, which show consonantal interactions.

2.1  Metaphony The dialects under study in this paper, including Servigliano, present vowels /i, e, ɛ, a, ɔ, o, u/, whose height feature values are shown in (1). Following

Positional prominence and consonantal interactions in metaphony 



 303

Calabrese (1988), the contrast among /e, o/ versus /ɛ, ɔ/ is represented by [±ATR]. In unstressed syllables, this contrast is neutralized in favor of [e, o]. (1) Height feature specifications for vowels of Servigliano and related dialects [high] [low] [ATR]

[i] + – +

[e] – – +

[ɛ] – – –

[a] – + –

[ɔ] – – –

[o] – – +

[u] + – +

In the metaphony of Servigliano, post-tonic [i, u] trigger raising of a stressed mid vowel. The raising is gradual so that /e, o/ raise to [i, u], as in (2a), while /ɛ, ɔ/ raise to [e, o], as in (2b). (2) a. ˈmett-o ˈpes-a ˈfjoɾ-e ˈloŋg-a b. ˈpɛtten-e tʃiˈlɛstɾ-a biɾiˈkɔkan-a ˈmɔʃ-a

ˈmitt-i ˈpis-u ˈfjuɾ-i ˈluŋg-u ˈpettin-i tʃiˈlestɾ-u biɾiˈkokun-u ˈmoʃ-u

‘I / you put’ ‘heavy-f.sg / m.sg’ ‘flower-m.sg / m.pl’ ‘long-f.sg / m.sg’ ‘comb-m.sg / m.pl’ ‘heavenly / pale blue-f.sg / m.sg’ ‘apricot (tree)-f.sg / m.sg’ ‘dejected-f.sg / m.sg’

The data in (3) confirm that only mid stressed vowels undergo harmony. Stressed high and low vowels do not show height assimilation. (3) aˈmik-a aˈmik-u ˈditʃ-e ˈditʃ-i ˈmut-a ˈmut-i ˈmuɾ-e ˈmuɾ-u ˈpatɾ-e ˈpatɾ-i ˈpann-u ˈpaɲɲ-i

‘friend-f.sg / m.sg’ ‘he says / you say’ ‘mute-f.sg / m.pl’ No gloss-m.pl / m.sg ‘father-m.sg / m.pl’ No gloss-m.sg / m.pl

The analysis of Servigliano metaphony developed by Walker (2011) uses a positional licensing constraint, given in (4).1 As used here, positional licensing constraint refers to a positional markedness constraint that assigns a penalty to a marked structure that lacks association with a specified prominent position. Such constraints prevent distinctive features that are realized solely in nonprominent positions.

1 Servigliano also displays harmony that raises pretonic vowels. The analysis presented here considers pretonic raising to be a distinct harmony process from metaphony.

304 

 Rachel Walker

(4) License([Height] / spost-tonic[+high], ˈs) Assign a violation to any height feature associated with a post-tonic high vowel that is not associated with the stressed syllable.2 Positional licensing constraints may single out features for licensing when they occur in contexts and/or feature combinations that are perceptually difficult. In (4), the structure subject to licensing is the set of height features {[high], [ATR], [low]} when they occur in a post-tonic high vowel. Post-tonic syllables are perceptually weak due to their reduced metrical prominence, and high vowels tend to be lower in amplitude and shorter in duration than lower vowels, which could make them less perceptible.3 The licensing position is specified as the primary stressed syllable, which is maximally prominent in the metrical dimension. In order for metaphony to be driven by one constraint,4 the licensing restriction operates over the class of height features: it is necessary to drive each of [+high] and [+ATR] to spread to the stressed syllable, and [Height] is the minimal class that contains both features (Padgett 2002). As discussed by Walker (2005, 2011), positional licensing constraints can drive harmony for the specified feature(s) from a nonlicensing position to the prominent licensing position and they can drive harmony from the licensing position to a nonlicensing position. They also can drive patterns that eliminate the marked features in nonlicensing positions, without assimilation, or they can drive the specified features to flop to the licensing position. In cases where harmony operates from a nonlicensing position, it is necessary to consider the means of trigger control, that is, what determines that material in the nonlicensing position is preserved at the cost of that originating in the licensing position. The most common mechanism for achieving trigger control is a positional faithfulness constraint (Beckman 1997, 1998). In Servigliano, trigger control for harmony can be attributed to a faithfulness constraint for the final syllable (Hyman 1998; Curtin 2002; Krämer 2003; Walker 2005, 2011; Sasa 2009). For the data considered thus far, it is sufficient to define this constraint in general terms as operating over the vowel in

2 The definition in (4) provides a paraphrase for the formal definition of positional licensing constraints. To address licensing at a distance, the formal definition provided in Walker (2011) operates over feature chains, consisting of chains of corresponding features. Licensing at a distance will not be at issue for the patterns under study here, so for present purposes it is sufficient to refer to features alone. 3 Despite these characteristics that could hinder perceptibility, high vowels have other properties that potentially render them more perceptible than mid vowels. As ‘corner’ vowels, high vowels improve dispersion in the inventory, and their focalization of acoustic energy could make them more salient. High and mid vowels (but not low /a/) thus each present properties that could make them less perceptible. Phonological patterns support a conclusion that the markedness of unstressed high and mid vowels is variable across languages (Walker 2011). 4  See Calabrese (1985) for a unified view of the metaphony-driver in a rule-based framework.



Positional prominence and consonantal interactions in metaphony 

 305

the word-final syllable, as in (5); however, in the next section, this will be adjusted to the final syllable in the maximal prosodic word. (5) Ident-IO-sFinal(high) Let a be a segment in the word-final syllable in the output and b be a correspondent of a in the input. If a is [ghigh], then b is [ghigh]. Although unstressed final syllables are weak, as reflected by the positional licensing constraint in (4), which governs material in post-tonic syllables, they are not the weakest metrical position. Studies of vowel patterns in dialects of Italy support the prominence scale in (6), applicable to Servigliano and various other dialects (Maiden 1995, 1997; Canalis 2007, 2009; Walker 2011). In this scale, |x| signifies “the intrinsic prominence of x”. (6) Vocalic metrical prominence scale (e.g. Servigliano) |primary stressed vowel| > |secondary stressed vowel| > |pretonic stem vowel| > |word-final unstressed vowel| > |post-tonic nonfinal vowel, unstressed clitic vowel| In agreement with the positional licensing constraint, post-tonic vowels fall in the bottom range of the prominence scale. However, unstressed vowels in a final syllable have greater prominence than nonfinal post-tonic vowels. Studies on the prominence of final syllables have shown that they can exhibit mixed effects, with the potential for evidence of both strength and weakness (Barnes 2006; Walker 2011). It is their strength-based properties that enable final syllables to serve as a place where positional faithfulness is enforced. Following Kaplan (2015), I assume that any position that is not minimal in a prominence dimension is eligible to serve as a designated position in a positional faithfulness constraint (see section 5). Optimality Theory (Prince & Smolensky 2004) predicts that positional faithfulness constraints can be differently ranked with respect to each other across languages, so a faithfulness constraint for the final syllable could dominate a positional faithfulness constraint for a stressed syllable. The constraint hierarchy for metaphony in Servigliano ranks the positional licensing constraint, License([Height] / spost-tonic[+high], ˈs), over positional faithfulness constraints for [high] and [ATR] in the primary stressed syllable and their counterparts that are not position sensitive (i.e. nonpositional). In addition, Ident-IO-sFinal(high) dominates Ident-IO-ˈs(high) and Ident-IO-ˈs(ATR) to guarantee control of harmony by the final syllable. The workings of this hierarchy in a word where a mid vowel undergoes metaphonic raising to high are shown in (7). Candidate (7a) is the winner. It spreads height features from the final posttonic vowel to the stressed syllable, violating Ident-IO-ˈs(high). Candidate (7b), which does not show metaphony, violates the positional licensing constraint. In (7c), harmony operates from the stressed syllable, causing the post-tonic vowel to

306 

 Rachel Walker

lower. This obeys positional licensing and stressed syllable faithfulness, but it is ruled out by positional faithfulness for the final syllable. (7) Servigliano metaphony raising a mid [+ATR] vowel /pes-u/

Ident-IOsFinal(high)

Lic([Height] / spost-tonic[+high], ˈs)

Ident-IOˈs(ATR)

*

 a. ˈpisu b. ˈpesu c. ˈpeso

Ident-IOˈs(high)

*! *!

Although ranking Ident-IO-ˈs(ATR) below the positional licensing constraint is not required for the form considered in (7), because the stressed vowel is [+ATR] in the input, this ranking will be necessary for metaphony in forms where [−ATR] /ɛ, ɔ/ raise to [+ATR] [e, o]. Several other details are needed for the constraint hierarchy to capture Servigliano metaphony in full, discussed by Walker (2011). In overview, these include the following. The gradual raising of [−ATR] mid vowels to [+ATR] mid rather than to [+high] is obtained by a conjunction of faithfulness constraints for [high] and [ATR] in the stressed syllable: Ident-IO-ˈs(high) & Ident-IOˈs(ATR) (after Kirchner 1996). This conjunction will be violated by a vowel that alters both its [high] and [ATR] values, and it dominates the positional licensing constraint, limiting metaphony of a [−ATR] mid vowel so that it becomes [+ATR] but not high (i.e. /mɔʃ-u/ → [ˈmoʃu], not *[ˈmuʃu]). The failure of low vowels to undergo metaphonic raising is achieved by ranking Ident-IO-ˈs(low) over the positional licensing constraint (i.e. /pann-u/ → [ˈpannu], not *[ˈpennu] or *[ˈpinnu]). Lastly, Ident-IO-sFinal(high) dominates the positional licensing constraint, so that in word with a final high vowel and a stressed vowel that cannot raise to high, the final vowel will remain high rather than lower to [−high] as a means of satisfying the positional licensing constraint (i.e. /pann-u/ → [ˈpannu], not *[ˈpanno]). The Hasse diagram in (8) summarizes the ranking of the position-sensitive constraints. (8) Ident-IO- σ Final(high) Ident-IO-'σ (low) Ident-IO-'σ(high) & Ident-IO-'σ(ATR) License([Height]/σ post-tonic[+high],'σ) Ident-IO-'σ (high)

Ident-IO-'σ (ATR)



Positional prominence and consonantal interactions in metaphony 

 307

To review, this account references prominent positions using positional licensing constraints and positional faithfulness constraints. Positional licensing constraints are markedness constraints that restrict material in nonprominent positions that lacks an association with a prominent licensing position. Metaphony is driven by a positional licensing constraint requiring height features in a post-tonic high vowel to be associated with the stressed syllable. Positional faithfulness constraints enforce identity in prominent positions. Positional faithfulness is responsible for preservation of height features in the final vowel, so that it controls metaphony.

2.2 Post-tonic vowel harmony Post-tonic vowels in Servigliano exhibit regressive copy harmony, where harmony is enforced for all vowel features. Examples in (9a–c) show copy harmony triggered by a final stem vowel, and those in (9d) show a trigger in an enclitic. Copy harmony does not persist to stressed syllables or pretonic vowels. Some examples in (9) show metaphony in the stressed vowel. The vowel where metaphonic raising features originate may belong to a suffix (9a) or an enclitic (9d). (9) a. ˈpɾɛdok-o ˈpɾɛdak-a ˈpɾedik-i b. ˈstommuk-u ˈstommik-i c. doˈmennak-a doˈmennek-e d. ˈmett-a=tʃa=la ˈmett-e=tʃe=le ˈmett-o=tʃo=lo ˈmitt-u=tʃu=lu ˈmitt-i=tʃi=li

‘I preach’ ‘s/he preaches’ ‘you preach’ ‘stomach-m.sg’5 ‘stomach-m.pl’ ‘Sunday-f.sg’ ‘Sunday-f.pl’ ‘put it-f.sg there’ ‘put it-f.pl there’ ‘put it-mass there’ ‘put it-m.sg there’ ‘put it-m.pl there’

/ɛ/ → [ˈe] by metaphony

cf. ˈmette ‘s/he puts’, tʃe ‘there’

/e/ → [ˈi] by metaphony /e/ → [ˈi] by metaphony

Like metaphony, post-tonic harmony is triggered by the vowel in the final syllable, and trigger control is achieved by faithfulness to the final syllable. Walker (2011: 267–269) argues that Servigliano clitics are affixal and attach by adjunction at the level of prosodic word (PWd) to yield a nested PWd structure as in

5 Camilli (1929) gives both stómmucu ([ˈstommuku]) (p. 225) and stòmmucu ([ˈstɔmmuku]) (p. 269). The latter might have a typographical error, because it is unexpected given metaphony, or it could be a variable pronunciation.

308 

 Rachel Walker

[[[ˈmitt-u]PWd = tʃu]PWd = lu]PWd (see also Loporcaro 2000). Therefore, the vowel that controls post-tonic harmony, and also metaphony, is final in the maximal PWd, that is, it is final in a PWd that is not dominated by a PWd (Itô and Mester 2009). This necessitates restricting the positional faithfulness constraint for final syllables (in (5)) to syllables that are final in PWdmax. In addition, the range of features to which final syllable faithfulness applies must be expanded to all of the vowel features relevant for Servigliano, except [ATR]. Faithfulness for [ATR] is not selectively enforced in the final syllable, because the [ATR] contrast among mid vowels is neutralized in all unstressed syllables. The set of final syllable faithfulness constraints needed to control post-tonic copy harmony will be referred to as Ident-IO-sFinalPWdMax(high)/ (low)/(Color), where Color refers to the set {[back], [round]}. Unlike metaphony, post-tonic harmony does not spread to reach a prominent position. Instead, features spread in an unbounded fashion within the sequence of post-tonic syllables. Post-tonic harmony is proposed to be driven by a maximal licensing constraint (Walker 2011). Following Jiménez and Lloret (2007), unbounded harmony from a weak trigger is conceptualized as involving licensing to which every vowel contributes, not just those in a specific prominent position. When a specified feature or set of features is present in a word, sometimes specifically in a weak context, a maximal licensing constraint assigns a penalty to every vowel to which those features are not associated. The constraint for copy harmony operates over all vowel features in post-tonic vowels, as framed in (10). In this constraint, the weak context in which vowel features are subject to maximal licensing is post-tonic vowels, which corresponds to the bottom two steps on the metrical prominence scale in (6).6 (10) License([V-Feature] / spost-tonic, ∀V) For each vowel feature associated with a post-tonic vowel, assign a violation to any vowel in the same prosodic word to which that feature is not associated. To produce harmony for all features, the maximal licensing constraint must dominate nonpositional Ident constraints for all vowel features, collapsed here as Ident-IO(V-Feature). (Henceforth V-Feature is abbreviated to [V-F].) Although the stressed syllable undergoes raising in metaphony, it does not undergo copy harmony. Therefore, Ident-IO-ˈs(V-F) dominates the maximal licensing constraint.7 Ident-IO-sFinalPWdMax(high)/(low)/(Color) will also dominate the

6 Unstressed proclitics are also included on this low end of the prominence scale, and consistent with their projected weakness, they also display copy harmony in Servigliano. Because the focus here is on harmony controlled by the final syllable, harmony among proclitics is not discussed. 7 In words like [doˈmenneke], where final and stressed vowels are underlying identical, copy harmony still does not persist to pretonic syllables. See Walker (2011: 275) for a proposal to restrict harmony using a crisp edge constraint for a maximal metrical foot.



Positional prominence and consonantal interactions in metaphony 

 309

maximal licensing constraint to prevent the stressed syllable from serving as the trigger for harmony. The rankings are illustrated in (11). To avoid clutter in this and later tableaux, a single violation is shown for each unfaithful vowel and for each set of features in a post-tonic vowel that lacks association to a licensing position. This simplification does not alter the selection for the candidates considered. (11) Post-tonic copy harmony /mett-e=tʃe=la/

Ident-sFinalPWdMax (high)/(low)/(Color)

Ident- License([V-F]/ ˈs(V-F) spost-tonic, ∀V) *

 a. ˈmett-a=tʃa=la b. ˈmett-e=tʃe=le

*!

Ident (V-F) ** *

c. ˈmatt-a=tʃa=la

*!

***

d. ˈmett-e=tʃe=la

*, *!**

The winning candidate, in (11a), displays copy harmony among the post-tonic vowels. This incurs violations of the nonpositional Ident constraint, for lowering of two post-tonic /e/s to [a], and it violates the maximal licensing constraint, because [a]’s features [+back], [+low], and [−ATR] fail to spread to the stressed vowel. Candidate (11b) shows copy harmony among all vowels in the word, with the stressed vowel serving as the trigger. This satisfies the maximal licensing constraint, but it is ruled out by its violation of final syllable faithfulness. ­Candidate (11c) also shows copy harmony among all vowels, but with trigger control by the final vowel. This candidate is ruled out by its violation of stressed syllable ­faithfulness. Finally, candidate (11d) does not enforce Ident-violating harmony. This incurs violations for the three vowels to which features of post-tonic [a] are not associated and for the single vowel to which features of post-tonic [e] are not associated. Violations for post-tonic [e] are minimized assuming that the sequence of syllables containing /e/ share features in this output. An overview of the rankings involved in post-tonic copy harmony is given in (12). (12) Ident-IO-σ FinalPWdMax(high)/(low)/(Color)

Ident-IO-'σ (V-F)

License ([V-F]/σ post-tonic, V) A

Ident-IO(V-F)

The rankings for post-tonic copy harmony and metaphony can be integrated. Final syllable faithfulness is enforced in both patterns, placing Ident-IOsFinalPWdMax(high)/(low)/(Color) in the top tier. The positional licensing ­constraint

310 

 Rachel Walker

that drives metaphony dominates faithfulness constraints for [high] and [ATR] in the stressed syllable. These are among the stressed syllable faithfulness constraints that dominate the maximal licensing constraint which drives vowel copy. The operation of the combined hierarchies over a form with post-tonic harmony and metaphony is illustrated in (13). The function of each constraint or constraint cluster in the pattern is identified at the top of each column. (13) Metaphony and post-tonic copy harmony Final s faith Metaphony /mett-e=tʃe=lu/

*

b. ˈmett-u=tʃu=lu d. ˈmett-e=tʃe=lu

Vowel copy

License Ident-ˈs License IdentsFinalPWdMax ([Height]/ (high)/ ([V-F]/ (high)/(low)/ sp-t[+hi], ˈs) (ATR) sp-t, ∀V) (Color)

 a. ˈmitt-u=tʃu=lu c. ˈmett-e=tʃe=le

ˈs faith

*!

Nonpositional faith Ident (V-F)

*

***

*

**

*!

* *!

*, ***

The input in (13) contains a final high vowel and a mid [+ATR] vowel in the syllable that receives stress. The winning candidate is (13a), with copy harmony among post-tonic vowels and metaphonic raising of the stressed vowel. Candidate (13b) shows post-tonic copy harmony but not metaphony, violating the positional licensing constraint. In (13c), the stressed vowel controls copy harmony among all vowels, violating final syllable faithfulness. Candidate (13d) obeys Ident constraints, although we may assume that features are shared across the sequence of syllables containing [e]. This candidate violates both of the harmonydriving constraints. In summary, post-tonic vowel harmony involves a maximal licensing constraint that reflects the weakness of post-tonic vowels. As with metaphony, harmony is controlled by the vowel in the final syllable, through the activity of a final syllable faithfulness constraint. However, while metaphony must reach the stressed syllable, faithfulness to the stressed position acts a limiter on satisfaction of the maximal licensing constraint, so that copy harmony does not persist beyond the post-tonic vowels. Final vowels show a mix of strength and weakness in both harmonies. They serve as the trigger for each harmony, obtained by ranking Ident constraints for this position in the top tier. Yet final syllables pattern with other post-tonic syllables in being subject to licensing constraints that single out weak elements.



Positional prominence and consonantal interactions in metaphony 

 311

3 C  onsonantal interactions with post-tonic harmony and metaphony In Servigliano, metaphony and post-tonic harmony do not show sensitivity to intervening consonants. However, in some dialects, certain consonants block harmony, opening new vantage points on harmonies controlled by the final vowel and the constraints that shape them. In the Lazio dialect of Sant’Oreste, intervening liquids, [l] and [r], permit copy harmony from the vowel in the final syllable to nonfinal post-tonic vowels, as shown in (14a) (Elwert 1958; Maiden 1995, 1997; Cimarra 1998; Canalis 2009). When another consonant intervenes, a nonfinal post-tonic vowel is neutralized to [i], as in (14b). Historical reconstructions provided by Maiden (1995) and cognates show evidence of the regressive direction of assimilation and neutralization of vowel quality. Pretonic vowels do not show such extreme neutralizations, although [o] is often raised to [u] (14c). (14) a. biʃˈʃɔkala < *beˈstjɔkkola biʃˈʃɔkele < *beˈstjɔkkole ˈrandala Ital. taˈrantola ˈarbulu < *ˈarberu ˈdukulu Ital. ˈɛdukalo ˈarbili < *ˈarberi ˈɔmmara Ital. ˈombra ˈennara Ital. ˈedera setˈtɛmmere Ital. setˈtɛmbre noˈɛmmere Ital. noˈvɛmbre ˈsiguru Ital. ˈsigaro ˈsigiri Ital. ˈsigari b. ˈtrapinu < *ˈtrapanu ˈʃtɛfine < *ˈstɛfanu ˈdʒakimu < *ˈdʒakomo ˈsabbit̞u Ital. ˈsabato ˈskɔmmida Ital. ˈskɔmoda ˈʃimiʃu Ital. ˈtʃimitʃe ˈstɔmmiku Ital. ˈstɔmako ˈʃtrɔliku c. affilaˈrati < *affilaˈrati pekuˈraru < *pekoˈraru lavanˈnara < *lavanˈdara kapuˈmilla Ital. kamoˈmilla

‘lizard’ ‘lizards’ ‘tarantula’ ‘tree’ ‘educate him’ ‘trees’ ‘shade’ ‘ivy’ ‘September’ ‘November’ ‘cigar’ ‘cigars’ ‘drill’ ‘Stephen’ ‘James’ ‘Saturday’ ‘uncomfortable-f’ ‘bug’ ‘stomach’ ‘astrologer, fortune teller’ ‘in a row’ ‘shepherd’ ‘washerwoman’ ‘chamomile’

312 

 Rachel Walker

Examples in Elwert (1958) show post-tonic harmony controlled by final [i, e, a, u]. He does not provide examples with harmony from final [o]; however, there is not reason to expect that [o] would not trigger copy harmony. Final unstressed [o] does occur in Sant’Oreste, e.g. bollino ‘postage stamp’, ceppo ‘base of a plow’, ciamorrito ‘to make cold (pst ptcp)’, but it may be infrequent because of the tendency for unstressed /o/ to raise to [u]. A similar pattern is found in the dialect of Umbertide, with copy harmony from the final vowel across liquids to a nonfinal post-tonic vowel, e.g. [ˈfragwala] / [ˈfragwele] ‘strawberry-sg/pl’, [ˈsigoro] ‘cigar’ (Canalis 2009). As in Sant’Oreste, when a nonliquid consonant blocks post-tonic harmony, a nonfinal post-tonic vowel is neutralized to [i], e.g. [ˈstɔmbiko] ‘stomach’, [ˈmonika] ‘nun’, [ˈpettine] ‘comb’. In some dialects, post-tonic harmony is limited to specific features. In a dialect spoken in some localities of Garfagnana, harmony among post-tonic vowels causes raising only, as shown in (15). This harmony, which is again con­ trolled by the final vowel, apparently operates only across liquids (Venturelli 1974; Maiden 1991a, 1997; Canalis 2009). Venturelli (1974: 104) explicitly identifies [l] and [r] as part of the context for post-tonic harmony, though he does not provide examples that show blocking by nonliquid consonants in post-tonic syllables.8 (15)

ˈkavul-i ˈruttsuli ˈandʒil-i kuˈnijjur-i balˈluttʃiur-i ˈalbir-i

ˈkavol-o ˈruttsolo ˈandʒel-o kuˈnijjor-o balˈluttʃior-o ˈalber-o

‘cabbage-pl/sg’ ‘fragment of firewood to burn-pl/sg’ ‘angel-pl/sg’ ‘rabbit-pl/sg’ ‘boiled chestnut-pl/sg’ ‘tree-pl/sg’

Additional cases of post-tonic raising harmony, without full copy, are noted by Maiden (1988a). An interesting aspect of the Garfagnana dialect described by Venturelli is that it shows raising harmony in various parts of the word, with different properties depending on the vowel’s metrical prominence. In pretonic syllables, mid vowels raise to high when preceding a high vowel (Venturelli 1974; Maiden 1988a, 1991a). Examples in (16) show that in pretonic syllables, raising harmony is not restricted to transliquid contexts. (16) viˈstito durˈmi sinˈtuto dunˈnina

‘suit’ ‘to sleep’ ‘felt (pst.ptcp)’ ‘little woman’

veˈstaʎʎa dorˈmjan senˈtjan donˈnona

‘dressing gown’ ‘we sleep’ ‘we feel’ ‘big woman’

8 Venturelli (1974) does not provide data to indicate the behavior of post-tonic nonfinal /a/ in raising harmony.



Positional prominence and consonantal interactions in metaphony 

 313

ruviˈna ‘ruin’ Ital. rovina tilivisiˈon ‘television’ Ital. televisione addumistiˈkato ‘domesticated’ Ital. addomesticato rapiruntsuˈlin ‘little radish’ Ital. raperonzolino sittiˈmana ‘week’ Ital. settimana umbrilˈlin ‘parasol’ Ital. ombrellino vilˈluto ‘velvet’ Ital. velluto munuˈmento ‘monument’ Ital. monumento Furthermore, this Garfagnana dialect exhibits a metaphony that raises /ɛ/ to [e] in a stressed syllable before final [i] (Venturelli 1974; Maiden 1987; Giannelli 1997).9 This harmony is not restricted by the intervening consonant, as shown in (17). (17)

ˈtemp-i ˈdent-i spaˈvent-i ˈpetts-i omˈbrell-i raˈstell-i

ˈtɛmp-o ˈdɛnt-e spaˈvɛnt-o ˈpɛtts-o omˈbrɛll-o raˈstɛll-o

‘time-pl/sg’ ‘tooth-pl/sg’ ‘fright-pl/sg’ ‘piece-pl/sg’ ‘umbrella-pl/sg’ ‘rake-pl/sg’

Metaphony can show sensitivity to intervening consonants. This tends to be evidenced in contexts where metaphony is less favored in the typology, that is, in contexts where it is more restricted, such as when the stressed syllable is closed, when the triggering vowel is back (/u/) instead of front (/i/), and when the underlying vowel in the stressed syllable is lower in height (Maiden 1987, 1988b, 1991b). In some dialects, metaphony operates in closed syllables only if the intervening consonant is palatal or /ll/, and this restriction may be narrowed to contexts where there are multiple conditions present where metaphony is less favored.10 To illustrate, Maiden (1987) points to the remnants of metaphony of /ɛ/ in the Valsesia dialect described by Spoerri (1918). Traces of metaphonic raising of /ɛ/ by /u/ occur only in the reflex of the suffix [*–ˈɛllu] as [ˈel], as in (18a) (cf. [–ˈɛlla] < [*–ˈɛlla]), or before a palatal consonant, as in (18b). ‘bird’ b. letʃ (18) a. iˈtʃ-el murˈf-el ‘child’s mucus’ petʃ cf. takuˈr-ɛlla ‘padlock’ vetʃ aˈɲ-ɛlla ‘ewe lamb’ cf. ˈvɛdʒ-a

|medial syllable| Kaplan finds that patterns driven by positional licensing constraints appear to privilege only positions that are maximal in prominence in some dimension. For instance, in the dimension of metrical prominence, only positions that are assigned primary stress function as licensors in positional licensing patterns. Likewise, in the dimension of sequential prominence, initial syllables are the only positions singled out by positional licensing. However, the positions accessed by positional faithfulness constraints appear to include any position that is nonminimal in its prominence dimension. For instance, positional faithfulness phenomena that privilege final syllables are attested (see section 2.1), even though final syllables are not maximally prominent on the sequential prominence scale. To incorporate this insight, Kaplan proposes to modify the schema for positional licensing constraints so that the set of possible licensors is defined as consisting of positions that are maximal in a prominence hierarchy. At the same time, he modifies the schema for positional faithfulness constraints so

15 Servigliano provides a further reason to reject a positional licensing account for post-tonic copy harmony. In Servigliano, copy harmony operates not only among post-tonic vowels, but also among proclitics, where it is controlled by the proclitic nearest to the stem. It is reasonable to suppose that copy harmony in both domains is driven by the same constraint (Mascaró 2011; Walker 2011), but there is no evidence to indicate that the stem-adjacent proclitic resides in a prominent position. This favors a maximal licensing approach to copy harmony in Servigliano, and by extension to similar post-tonic harmonies in other dialects of Italy, rather than an account based in positional licensing.



Positional prominence and consonantal interactions in metaphony 

 327

that they may reference positions that are nonminimal in a prominence hierarchy. Kaplan’s proposed revision to the set of possible licensors excludes the alternative positional licensing analysis for post-tonic vowel harmony, because it relies on the final syllable functioning as the licensor. Since the final syllable is not maximal on a prominence hierarchy, it is not available in this capacity. It is noteworthy also that Kaplan’s proposal rules out incongruous licensing patterns, where positions that serve as licensors for material in weaker positions in turn require their own content to be licensed by a more prominent position in their prominence scale.16 In closing, while some properties of post-tonic vowel harmony might at first appear amenable to an analysis based on positional licensing, such an approach is undesirable. A positional licensing account is faced with the problem of an inconsistent understanding of licensing strength in languages where the final position holds trigger control for both post-tonic harmony and metaphony in the stressed syllable. Moreover, an account with positional licensing by the final syllable would open up unwanted typological predictions. Kaplan’s study has shown that a constrained typology of positional privilege derives from a formalism where the licensing positions singled out by positional licensing constrants are solely those that are maximally prominent. Further, the concept of maximally prominent licensors for positional licensing constraints is reinforced by patterns of consonant blocking seen in some patterns of metaphony, where only the stressed syllables that offer maximally robust perceptual cues for the harmonizing features are singled out as licensors. Yet this understanding of such blocking effects also suggests that the maximally prominent positions that can serve as licensors can be restricted to contexts where they offer especially robust cues for the features subject to licensing.

6 Conclusion The patterns of consonant blocking examined in this paper are in keeping with an analysis where metaphony and post-tonic harmony are driven by distinct types of constraints, positional licensing and maximal licensing, respectively. Under-

16 However, this theory does not exclude apparently unattested patterns where a position that is maximally prominent in one scale (e.g. a syllable with primary stress) serves as a licensor for features in weaker positions on its scale, but whose content is in turn subject to licensing by a position that is maximally prominent on another scale (e.g. an initial syllable). Patterns of this kind will need further attention.

328 

 Rachel Walker

standing metaphony as motivated by positional licensing is consistent with the insight that the consonants that are prone to block metaphony are those that are likely to reduce perceptual cues for the harmonizing features in the target stressed syllable. In contrast, consonant blocking in post-tonic harmony shows a similarity effect, a typological characteristic more general to feature sharing in vowelconsonant-vowel contexts—without reference to positional prominence—which is amenable to analysis in terms of gestural unformity constraints. The treatment of these harmonies using distinct constraints permits them to be separately ranked, allowing the possibility that blocking by nonliquids is enforced in posttonic harmony but not in metaphony. This study also bears on the nature of prominence in different positions. The final syllable in these dialects shows strength in wielding trigger control for both metaphony and post-tonic harmony, which is captured using positional faithfulness constraints for this position. However, the final syllable patterns with weak post-tonic syllables in being a context where features are subject to a licensing restriction. This sets it apart from the stressed syllable, which has the capacity to serve as a licensor in positional licensing constraints in addition to showing positional faithfulness effects, consistent with the typological findings of Kaplan (2015). The connections drawn between the respective roles of different prominent positions and scalar positional prominence hierarchies thus provide valuable insight into properties of these vowel harmony systems.

References Barnes, Jonathan. 2006. Strength and weakness at the interface: positional neutralization in phonetics and phonology. Berlin: Mouton de Gruyter. Beckman, Jill. 1997. Positional faithfulness, positional neutralization and Shona vowel harmony. Phonology 14: 1–46. Beckman, Jill. 1998. Positional faithfulness. Amherst, MA: University of Massachusetts, Amherst dissertation. (Published New York: Garland, 1999.) Browman, Catherine & Louis Goldstein. 1988. Some notes on syllable structure in Articulatory Phonology. Phonetica 45: 140–155. Calabrese, Andrea. 1985. Metaphony in Salentino. Rivista di Grammatica Generativa 9–10: 1–40. Calabrese, Andrea. 1988. Towards a theory of phonological alphabets. Cambridge, MA: MIT dissertation. Camilli, Amerindo. 1929. Il dialetto di Servigliano. Archivum Romanicum 13: 220–271. Canalis, Stefano. 2007. Total vowel harmony in two Romance dialects. Handout of paper presented at Phonetics and Phonology in Iberia, Universidade do Minho, Braga, June 25–26, 2007. Canalis, Stefano. 2009. Post-tonic vowel harmony in some dialects of central Italy: the role of prosodic structure, contrast and consonants. In Marina Vigário, Sónia Frota & M. João



Positional prominence and consonantal interactions in metaphony 

 329

Freitas (eds.), Phonetics and phonology: interactions and interrelations, 247–266. Amsterdam: John Benjamins. Cimarra, Luigi (ed.). 1998. Il dialetto di Sant’Oreste: richerche e materiali; con un saggio di W. Theodor Elwert. Sant’Oreste (Roma): Apeiron. Cole, Jennifer & Charles Kisseberth. 1995. Restricting multi-level constraint evaluation: opaque rule interaction in Yawelmani vowel harmony. In Keiichiro Suzuki & Dirk Elzinga (eds.), Proceedings of the South Western Optimality Theory Workshop. 18–38. University of Arizona Coyote Papers. Côté, Marie-Hélène. 2000. Consonant cluster phonotactics: a perceptual approach. Cambridge, MA: MIT dissertation. Crosswhite, Katherine. 2004. Vowel reduction. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.), Phonetically driven phonology, 191–231. Cambridge: Cambridge University Press. Curtin, Suzanne. 2002. Representational richness in phonological development. Los Angeles, CA: University of Southern California dissertation. de Lacy, Paul. 2007. The interaction of tone, sonority, and prosodic structure. In Paul de Lacy (ed.), The Cambridge handbook of phonology, 281–307. Cambridge: Cambridge University Press. D’Imperio, Mariapaola & Sam Rosenthall. 1999. Phonetics and phonology of main stress in Italian. Phonology 16: 1–28. Elwert, W. Theodor. 1958. Die Mundart von S. Oreste. Romanica. Festschrift für Gerhard Rohlfs, 121–158. Halle: Niemeyer. Gafos, Adamantios. 1996. The articulatory basis of locality in phonology. Baltimore, MD: Johns Hopkins University dissertation. (Published New York: Garland, 1999.) Gafos, Adamantios & Linda Lombardi. 1999. Consonant transparency and vowel echo. Proceedings of the North East Linguistic Society 29: 81–96. Giannelli, Luciano. 1997. Tuscany. In Martin Maiden & Mair Perry (eds.), The dialects of Italy, 287–302. New York: Routledge. Howe, Darin & Douglas Pulleyblank. 2004. Harmonic scales as faithfulness. Canadian Journal of Linguistics 49: 1–49. Hyman, Larry. 1998. Positional prominence and the ‘prosodic trough’ in Yaka. Phonology 15: 41–75. Itô, Junko & Armin Mester. 1999. Realignment. In René Kager, Harry van der Hulst & Wim Zonneveld (eds), The prosody-morphology interface, 188–217. Cambridge: Cambridge University Press. Itô, Junko & Armin Mester. 2009. The onset of the prosodic word. In Steve Parker (ed.), Phonological argumentation: essays on evidence and motivation, 227–260. London: Equinox. Itô, Junko, Armin Mester & Jaye Padgett. 1995. Licensing and underspecification in Optimality Theory. Linguistic Inquiry 26: 571–613. Jiménez, Jesús & Maria-Rosa Lloret. 2007. Andalusian vowel harmony: weak triggers and perceptibility. Handout of paper presented at the fourth Old World Conference in Phonology, Workshop on Harmony in the Languages of the Mediterranean. Rhodes, January 18–21, 2007. Kaplan, Aaron. 2015. Maximal prominence and a theory of possible licensors. Natural Language and Linguistic Theory 33: 1235–1270. Kawahara, Shigeto. 2008. On the proper treatment of non-crisp edges. Japanese/Korean Linguistics 13: 55–67. Stanford: CSLI Publications. Kaun, Abigail. 1995. The typology of rounding harmony: an optimality theoretic approach. Los Angeles, CA: UCLA dissertation.

330 

 Rachel Walker

Kaun, Abigail. 2004. The typology of rounding harmony. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.), Phonetically driven phonology, 87–116. Cambridge: Cambridge University Press. Kaze, Jeffery. 1989. Metaphony in Spanish and Italian dialects revisited. Urbana-Champaign, IL: University of Illinois, Urbana-Champaign dissertation. Kirchner, Robert. 1996. Synchronic chain shifts in Optimality Theory. Linguistic Inquiry 27. 341–349. Krämer, Martin. 2001. Yucatec Maya vowel alternations: Harmony as syntagmatic identity. Zeitschrift für Sprachwissenschaft 20: 175–217. Krämer, Martin. 2003. Vowel harmony and correspondence theory. Studies in Generative Grammar 66. Berlin: Mouton de Gruyter. Loporcaro, Michele. 2000. Stress stability under cliticization and the prosodic status of Romance clitics. In Lori Repetti (ed.), Phonological theory and the dialects of Italy, 137–168. Amsterdam: John Benjamins. Mahanta, Shakuntala. 2007. Directionality and locality in vowel harmony with special reference to Assamese. Utrecht: Utrecht University dissertation. Maiden, Martin. 1987. New perspectives on the genesis of Italian metaphony. Transactions of the Philological Society 85: 38–73. Maiden, Martin. 1988a. Armonia regressiva di vocali atone nell’Italia meridonale. L’Italia dialettale 51: 111–139. Maiden, Martin. 1988b. On the dynamics of low mid vowel diphthongization in Tuscan and Gallo-Italian. Canadian Journal of Italian Studies 11: 1–37. Maiden, Martin. 1991a. Armonia regressiva di vocali atone nell’Italia centromeridionale: la sua importanza per la teoria della fonologia prosodica. In Luciano Giannelli, Nicoletta Maraschio, Teresa Poggi Salani & Massimo Vedovelli (eds.), Tra rinascimento e strutture attuali. Saggi di linguistica italiana, 233–239. Torino: Rosenberg and Sellier. Maiden, Martin. 1991b. Interactive morphonology: metaphony in Italy. New York: Routledge. Maiden, Martin. 1995. Evidence from the Italian dialects for the internal structure of prosodic domains. In John Charles Smith & Martin Maiden (eds.), Linguistic theory and the Romance languages, 115–131. Amsterdam: John Benjamins. Maiden, Martin. 1997. Vowel systems. In Martin Maiden & Mair Perry (eds.), The dialects of Italy, 7–14. New York: Routledge. Majors, Tivoli. 1998. Stress-dependent harmony. Austin, TX: University of Texas, Austin dissertation. Mascaró, Joan. 2011. An analysis of stress-dependent harmony in Servigliano. Probus 23: 21–55. McCarthy, John J. 1994. On coronal ‘transparency’. Paper presented at the Trilateral Phonology Weekend, University of California, Santa Cruz, January 22, 1994. McCarthy, John J. 1998. Morpheme structure constraints and paradigm occultation. In M. Catherine Gruber, Derrick Higgins, Kenneth Olson & Tamra Wysocki (eds.), Proceedings of the Chicago Linguistic Society 32, Part 2: The Panels. 123–150. Chicago, IL: Chicago Linguistic Society. Melillo, Giacomo. 1926. I dialetti del Gargano: saggio fonetico. Pisa: Simoncini. Nevins, Andrew. 2010. Locality in vowel harmony. Cambridge, MA: MIT Press. Ní Chiosáin, Máire & Jaye Padgett. 2001. Markedness, segment realization and locality in spreading. In Linda Lombardi (ed.), Segmental phonology in Optimality Theory, 118–156. Cambridge: Cambridge University Press.



Positional prominence and consonantal interactions in metaphony 

 331

Nibert, Holly. 1998. Processes of vowel harmony in the Servigliano dialect of Italian: A comparison of two non-linear proposals for the representation of vowel height. Probus 10: 67–101. Padgett, Jaye. 2002. Feature classes in phonology. Language 78: 81–110. Prince, Alan & Paul Smolensky. 2004 [1993]. Optimality Theory: Constraint interaction in generative grammar. Oxford: Blackwell. (Revision of 1993 technical report, Rutgers University Center for Cognitive Science.) Rohlfs. Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti, Vol. I, II. Turin: Einaudi. Sasa, Tomomasa. 2009. Treatments of vowel harmony in Optimality Theory. Iowa City, IA: University of Iowa dissertation. Sganzini, Silvio. 1928. Osservazioni sul vocalismo dei dialetti della Valle de Blenio. L’Italia dialettale 4: 150–167. Spoerri, Teofilo. 1918. Il dialetto della Valsesia. Rendiconti del reale Istituto lombardo di scienze e lettere II 51: 391–409, 683–698; 732–752. Steriade, Donca. 1995. Positional neutralization. Ms., UCLA. Steriade, Donca. 1999. Phonetics in phonology: The case of laryngeal neutralization. UCLA Working Papers in Linguistics 2: 25–146. Venturelli, Gastone. 1974. Varietà di armonizzazioni vocaliche nella Garfagnana centro-meridionale. Atti del XIV Congresso internazionale de linguistica romanza 3: 101–104. Naples: Macchiaroli and Amsterdam: John Benjamins. Walker, Rachel. 2001. Round licensing, harmony, and bisyllabic triggers in Altaic. Natural Language and Linguistic Theory 19: 827–878. Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language and Linguistic Theory 23: 917–989. Walker, Rachel. 2011. Vowel patterns in language. Cambridge: Cambridge University Press. Wright, Richard. 2004. A review of perceptual cues and cue robustness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.), Phonetically driven phonology, 34–57. Cambridge: Cambridge University Press.

Francesc Torres-Tamarit and Kathrin Linke

Opaque interactions between vowel merger and metaphony

UMR 7023 SFL/CNRS, Paris 8 and University of Vienna

Abstract: Metaphony is a common process in most Italo-Romance varieties, whereby an unstressed high vowel inflectional suffix causes raising of a stressed root vowel. In some varieties, metaphony interacts with a process in which phonetic contrast among a set of suffixes neutralizes (i.e. vowel merger). This paper develops a parallel analysis of two opaque interactions between vowel merger and metaphony in Turbidity Theory, a model assuming containment, combined with privative features and maximal economy in the representation of segments. The basic idea is that metaphony in opaque environments is computed synchronically as a non-local licensing condition of a feature [high] only if it is u ­ nderlyingly present. On the one hand, overapplication of metaphony is due to prohibiting the realization of a feature [high] lexically linked to a suffix, which needs to be licensed non-locally by associating it with a stressed vowel. On the other hand, underapplication of metaphony is due to insertion of a feature [high] in non-low suffixes, which does not need to be licensed because it is not lexical.

1 Introduction Metaphony refers to a phonological process characterizing most Italo-Romance varieties whereby an unstressed high vowel inflectional suffix causes raising of a preceding stressed (mid) root vowel. For instance, Foggiano presents the transparent metaphony case, i.e. that both stressed high-mid and low-mid root vowels alternate with high vowels whenever a high vowel suffix follows (1). (1) Transparent metaphony in Foggiano (Calabrese 2011) ˈkjena ‘full-f.sg’ ˈkjinu ‘full-m.sg’ ˈmo∫∫a ‘soft-f.sg’ ˈmu∫∫u ‘soft-m.sg’ ˈpεte ‘foot’ ˈpiti ‘feet’ ˈgrɔssa ‘big-f.sg’ ˈgrussu ‘big-m.sg’ The process of metaphony interacts with a (diachronic) process of vowel merger in some Italo-Romance varieties. Vowel merger describes a situation in which the phonetic contrast among a set of suffixes spelling out different morphosyntactic features is neutralized. The examples in (2) illustrate vowel merger of two nominal and two verbal inflectional suffixes in the present indicative singular in Salentino.

334 

 Francesc Torres-Tamarit and Kathrin Linke

(2) Vowel merger in Salentino (Calabrese 1985; Ribezzo 1912) Nominal {[masculine] & [plural]} {[feminine] & [plural]} suffixes [i] [i] Verbal {[2nd person] & [singular]} {[3rd person] & [singular]} suffixes [i] [i] For Italo-Romance systems with vowel merger, vowel suffixes might be merged in favor of high vowels or in favor of non-high vowels, usually a schwa. From a serial ordered-rule perspective, in all these varieties metaphony precedes vowel merger. However, metaphony overapplies or underapplies depending on whether a metaphonic trigger becomes opaque or a metaphony-neutral segment becomes an alleged trigger in surface representations (3). (3) Vowel merger—metaphony interactions System A (cf. Arpinate 2.1) Metaphony Vowel merger

System B (cf. Salentino 2.2)

/Underlying trigger/

/Underlying blocker/ — (in favor of a non-high vowel) (in favor of a high vowel) [Surface opaque triggering suffix] [Surface opaque blocking suffix]

Overapplication

Underapplication

In this paper, an analysis of the two systems schematized in (3) will be developed, couched within the formalism of Turbidity Theory (TT, Goldrick 2001; Van Oostendorp 2008), a constraint-based model assuming containment and using parallel computation. In TT, the lexical association between a phonological feature and a root node is always contained in output representations. Associations between lexical ­features and root nodes are represented by ↑ (projection relation); associations between surface features and root nodes are represented by ↓ (pronunciation relation). In (4), an input-output mapping is illustrated, in which the ­phonological feature F is realized locally on the surface, that is, by the root node r that projects the feature F. (4) Local feature-root node association r r → F F We will propose, however, that phonological features can also be realized non-locally, that is, licensed at a distance by root nodes other than those that project them. A non-local input-output mapping is illustrated in (5), where a

Opaque interactions between vowel merger and metaphony 



 335

phonological feature F is pronounced by a root node s although it is projected by a root node r. (5) Non-local feature-root node association r s r s → F F We assume a form of substance-freedom in the following terms: first, phonological features are privative (cf. Harris & Lindsey 1995 and references therein) and the representation of segments is maximally economic, i.e. segments are specified for the number of features necessary to capture their phonological activity and patterning, and not for more features (Dresher 2009). Second, two instances of the same phonetic output may have two different phonological output representations due to whether a segment is derived or non-derived. The basic claim of this paper is that metaphony in opaque environments is computed synchronically as a non-local licensing condition of a feature [high] only if it is underlyingly present, as shown in (5). On the one hand, overapplication of metaphony is due to prohibiting the realization of a feature [high] that is lexically linked to a suffix, which consequently needs to be licensed non-locally by associating it with a stressed vowel. On the other hand, underapplication of metaphony is due to insertion of a feature [high] in non-low suffixes, which does not need to be licensed because it is not lexical. This paper is organized as follows. In section 2, data from two Italo-Romance varieties are presented, namely Arpinate and Salentino, which show different opaque environments. The representational assumptions and the formalization of constraints used in the analysis are explained in section 3. Section 4 is devoted to showing how the opaque data can be accounted for using TT with privative ­features. Section 5 briefly concludes.

2 Data The underlying vowel system of most Italo-Romance varieties consists of seven vowels, with a two-height distinction in the mid region (6). (6) Italo-Romance varieties’ 7-vowel system /i/

/u/ /e/

/o/ /ε/



/ɔ/ /a/

336 

 Francesc Torres-Tamarit and Kathrin Linke

2.1 Arpinate In Arpinate (Lazio), all nominal suffixes except for class I feminine singular correspond to surface schwas. Being non-high suffixes, these schwa morphs are expected not to trigger metaphony, but some of them do, namely both class II suffixes and the class III plural suffix (7). The data are taken from Calabrese (1998), who cites Parodi (1892). (7) Nominal suffixes and metaphonic alternations (Calabrese 1998; Parodi 1892) classes sg pl I (mostly f) -a -ǝ no metaphony no metaphony II (mostly m) -ǝ -ǝ III (both f & m)

metaphony

metaphony

-ǝ

-ǝ

no metaphony

metaphony

Some examples are given in (8) and (9), in which opaque forms in both nominal and verbal paradigms show overapplication of metaphony (within boxes). Highmid vowels raise to high, and low-mid vowels undergo a one-step raising to highmid and also diphthongize. (8) Adjectival endings and metaphonic alternations (class I/II) sg pl sg pl f ˈsola ˈsolǝ ‘alone’ ˈnera ˈnerǝ ‘black’ m

ˈsulǝ

ˈsulǝ

f

ˈbɔna

ˈbɔnǝ

m

ˈbwonǝ

ˈbwonǝ

‘good’

ˈnirǝ

ˈnirǝ

ˈvεkkja

ˈvεkkjǝ

ˈvjekkjǝ

‘old’

ˈvjekkjǝ

(9) Verbal endings and metaphonic alternations (present indicative) 2nd sg 3rd sg 2nd sg 3rd sg ˈvidǝ

ˈvedǝ ‘you see / (s)he sees’ ˈsjentǝ

ˈsεntǝ ‘you feel / (s)he feels’

ˈkurrǝ ˈkorrǝ ‘you run / (s)he runs’ ˈkwoʎʎǝ ˈkɔʎʎǝ ‘you pick / (s)he picks’ In Arpinate, overapplication of metaphony is due to ‘opaque triggers’ (i.e. seeming­ly metaphony-neutral segments due to vowel merger to schwa).1 1 Throughout this paper, the term vowel merger is used instead of vowel reduction because no direct morpho-phonological alternations exist in the domain of inflectional suffixes, as they are always unstressed.

Opaque interactions between vowel merger and metaphony 



 337

2.2 Salentino In Salentino (Puglia), two front high vowel adjectival suffixes exist ([-i]), one for feminine plural and one for masculine plural. However, only the masculine plural suffix causes metaphony (10). The data are taken from Calabrese (1985), who cites Ribezzo (1912). (10) Adjectival endings and metaphonic alternations I sg pl f -a -i no metaphony m

no metaphony

-u metaphony

-i metaphony

Some examples are given in (11), where words within boxes contain a front high vowel suffix that does not cause raising of a preceding mid vowel. In cases where metaphony does apply, high-mid vowels raise to high, and low-mid vowels show diphthongization, as in Arpinate, but do not raise one-step to high-mid. Furthermore, low-mid back vowels ([ˈɔ]) alternate with [ˈwε]. (11) Adjectival endings and metaphonic alternations II sg pl sg pl f

ˈfredda

m

ˈfriddu

ˈfriddi

f

ˈlεnta

ˈlεnti

m

ˈljεntu

ˈfreddi

‘cold’

piˈlosa piˈlusu

‘slow’

ˈljεnti

ˈbɔna ˈbwεnu

piˈlosi

‘hairy’

piˈlusi ˈbɔni

‘good’

ˈbwεni

The same situation is found for nouns, as shown in (12), where singular forms, although ending in a high vowel suffix, do not undergo metaphony. (12) Noun endings and metaphonic alternations sg pl sg

pl

ˈmesi

ˈmisi

‘month(s)’

ˈtori

ˈturi

‘tower(s)’

ˈpεti

ˈpjεti

‘foot / feet’

ˈkɔri

ˈkwεri

‘heart(s)’

In Salentino, metaphony underapplies due to ‘fake metaphonic triggers’ (i.e. deceptive triggers due to vowel merger to high).

338 

 Francesc Torres-Tamarit and Kathrin Linke

3 Theory 3.1 Representations We propose to specify vowels in Arpinate and Salentino as in (13), based on Dyck (1995) for Italo-Romance. (13) Surface vowel feature specifications [front], [high] [front], ([high], [low]) [front], [low]

[i] [e] [ε]

([ǝ])

[u]

[back], [high]

[o]

[back], ([high], [low])

[ɔ]

[back], [low]

[a] [low]

To characterize the surface eight-vowel system in (13), a total of four privative phonological features have been used: two height features, [high] and [low], and two place features, [front] and [back]. Only three features are actually needed to specify a maximally economical eight-vowel inventory (23). However, since the two sets of front and back vowels (i.e. except for ǝ and a) with three height levels each act in complete symmetry, a fourth feature is needed to capture the observed patterning of the vowel system at hand. High-mid vowels are minimally specified by a place feature, either [front] or [back]. However, if derived from low-mid vowels, we assume that besides the place feature, they also contain the feature [low], inherited from low-mid vowels, and the feature [high], from high vowels. Schwas are unspecified. Although we use standard feature labels like [high] and [low], we should emphasize that their phonetic content does not play a direct role in phonological computation. We prefer to think of phonological features as being similar to colors in the sense that they can be mixed to create more complex segments, as elements in Element Theory (Harris & Lindsey 1995; Backley 2011). This is why [high] and [low], for instance, can freely surface together to create a derived intersecting segment, in this case a high-mid vowel.

3.2 Constraints In TT, a family of constraints called Reciprocity (14) ensures that associations between lexical features and root nodes ( projection relations ↑) are realized locally in surface representations ( pronunciation relations ↓) (14).



Opaque interactions between vowel merger and metaphony 

 339

(14) Reciprocity (Van Oostendorp 2008) a. Reciprocit​y​ ​​fr​ ​ (​  ​fr​) ​ : assign a violation mark for every feature F projected by a root node r that is not pronounced by r. b. Reciprocit​yf​​​ r ​​ (​ f​ r ​ ​): assign a violation mark for every feature F pronounced by a root node r that is not projected by r. To allow associations between features and root nodes to be realized non-locally, we propose a new family of constraints, Licensing (15), which is a less stringent version of Reciprocity. (15) Licensing a. Licensing ↑ F ( ↑F): assign a violation mark for every projection ­relation ↑ of a feature F that does not correspond to some pronunciation relation ↓ of F. b. Licensing ↓ F ( ↓F): assign a violation mark for every pronunciation relation ↓ of a feature F that does not correspond to some projection relation ↑ of F. Regarding the operations performed by Gen in TT, recall that projection relations are always contained in output forms; Gen only disposes of the set of operations in (16). (16) Gen operations a. Insertion of a pronunciation relation ↓ b. Insertion of a feature F together with its pronunciation relation ↓ With respect to unfaithful mappings, these can be the result of two different scenarios: non-pronunciation of a projected feature, and pronunciation of a nonprojected feature. On the one hand, non-pronunciation is triggered by the satisfaction of feature co-occurrence constraints of the type *{F,G} with the special case *F (Levelt & Van Oostendorp 2007; Van Oostendorp 2014) (17). (17) Markedness constraints a. *F: assign a violation mark for every pronunciation of the feature F. b. *{F,G}: assign a violation mark for every pronunciation of the features F and G by the same root node. On the other hand, an unfaithful mapping can also arise from the pronunciation of a non-projected feature G, which is enforced by the presence of another feature F. This situation is induced by implicational constraints (Van Oostendorp 2014) (18). (18) Implicational constraint F ⊃ G: assign a violation mark for every root node that pronounces a feature F and does not pronounce a feature G.

340 

 Francesc Torres-Tamarit and Kathrin Linke

3.3 The metaphony-triggering constraint Metaphony has traditionally been treated as a specific case of vowel harmony. In OT, vowel harmony has been attributed to several different constraints: spreading, alignment, sharing, agreement by correspondence (see Gafos & Dye 2011; Rose & Walker 2011 for a review), and feature-validation (Nevins 2010). It is not the purpose of this paper to evaluate these different constraints. An alternative view of metaphony has been developed by Walker (2005). According to her, metaphony has a functional basis in the sense that it improves the perceptibility of the feature [+high] in a posttonic position by associating it with a perceptually stronger position, a stressed syllable (19). (19) Walker’s (2005) metaphony-triggering constraint License([+high])posttonic, σ: [+high] in a posttonic syllable must be associated with a stressed syllable. A functional motivation for metaphony is plausible. However, after (diachronic) vowel merger, the motivation for metaphony as proposed in Walker (2005) cannot be interpreted by language learners as driven by functional forces anymore. We take Walker’s analysis as the starting point of our proposal. However, we claim that, in the synchronic grammar, metaphony is re-interpreted as a nonlocal licensing condition in terms of projection relations, as in (20). (20) Metaphony-triggering constraint as a licensing constraint Licensing↑[high]NonFtHd / FtHd ( ↑[high]NonFtHd): assign a violation mark for every projection relation ↑ of a feature [high] in a word-final, non-foot head position, that does not correspond to some pronunciation relation ↓ of [high] in a foot head position. It seems reasonable to see the chronology of the facts as follows. First, functional motivations like perception pressures drive metaphony. Speakers start producing gradual shifts of articulation in mid vowels when preceding high vowel suffixes as a way to enhance the cues of high vowels in unstressed position. Later on, learners re-interpret the input as a categorical distinction between two sets of root vowels. At this point, the synchronic grammar builds a metaphony-triggering constraint in terms of pronunciation relations. After this, a different process starts operating in the language, namely vowel merger, probably driven by functional motivations, too. Therefore, speakers start merging different sets of suffixes, while still displaying metaphony, as it was acquired as a phonological rule. Following,



Opaque interactions between vowel merger and metaphony 

 341

in the absence of a clear distinction among sets of different suffixes, learners re-formulate the metaphony-triggering constraint as a licensing constraint in terms of projection relations, instead of pronunciation relations. Assuming that such a scenario was true, this particular case of opacity resulting from sound change can be handled synchronically by parallel grammars with the help of TT. Grammars would allow for stating linguistic generalizations at two distinct levels: at an abstract, lexical one, and also at a concrete, surface one. A model like this is reminiscent of Evolutionary Phonology (Blevins 2004), a formal model of sound change in which phonetic variation and misperception play a key role in accounting for attested sound patterns in natural language, in which notions like markedness or naturalness have no place in synchronic grammars. We assume the idea, at least, that phonology has the ability to translate notions such as ‘phonetic perceptibility’ (as put forward in Walker 2005) as a condition on projection relations when sound change is obscuring a structural environment. A projection relation is a pure grammatical, function-free notion. Phonology, as an independent module of grammar, should only make use of its own vocabulary. This position precisely allows for explaining phonological opacity as a mismatch between phonetics and phonology (Hamann 2014).

4 Analysis 4.1 Arpinate Before going into the details of the analysis, the underlying representations of metaphonic triggers and metaphony-neutral segments must be discussed. Recall that in Arpinate, some schwa suffixes trigger metaphony contrary to what one would expect. For transparent, non-metaphonic schwa suffixes, either a representation with a place feature or without any feature at all is consistent with the facts. As long as these kinds of suffixes lack the feature [high], metaphony is not triggered. From now on, where place distinctions are not relevant, we will use the cover feature [Place]. In (21), the underlying representation of a non-metaphonic nominal suffix in Arpinate is represented, with abstract morphosyntactic features such as class and number associated with a root node, which in turn projects its phonological features. Parentheses express the fact that a specific feature can be absent from the representation in accordance with the Richness of the Base hypothesis (Prince & Smolensky 1993).

342 

 Francesc Torres-Tamarit and Kathrin Linke

(21) Underlying representation of a non-metaphonic nominal suffix {Class I, [Plural]} • ([Place]) For a metaphonic triggering suffix in Arpinate, the feature [high] is projected (22). (22) Underlying representation of a metaphonic nominal suffix {Class III, [Plural]} •



([Place])

[high]

In Arpinate, suffixes are restricted to low vowels and schwa. This means that no place features can surface in the suffixal domain. A markedness constraint like *[Place]NonFootHead, prohibiting the pronunciation of a place feature in a wordfinal, non-foot head position, must therefore be top-ranked. In case a place feature is present in the input, as in the tableau in (23), the constraint *[Place]NonFootHead prevents this feature from being pronounced, as shown by candidate (c).2 Candidate (a) and candidate (b) satisfy *[Place]NonFootHead, but candidate (b) fatally violates *[high]NonFootHead, which disfavors the pronunciation of a feature [high] in ­ nderlyingly, word-final, non-foot head position. The feature [high], not present u is therefore inserted in candidate (b). If a feature [high] is pronounced and the place feature is left unpronounced, a central high vowel is generated, which is absent from the sound inventory of Arpinate. In the absence of the feature [high] in word-final, non-foot head position, no metaphony is expected. The most harmonic candidate is candidate (a), in which the place feature is left unpronounced, which incurs a violation of low-ranked Reciprocityv[Place] / NonFootHead, which states that every place feature projected by a vocalic root node should be 2 Tableaux are presented in comparative format, and also include traditional violation marks. A capital W is entered into the cell of a particular loser row if the winner is favored over that loser by the constraint in that column. A capital L is inserted in the opposite situation, that is, if that loser is favored over the winner. Nothing is inserted if neither the winner nor a particular loser are favored by a specific constraint. Every L must be dominated by at least one W in c­ omparative tableaux, mean­ ing that every loser-favoring constraint is dominated by at least one winner-favoring constraint.

Opaque interactions between vowel merger and metaphony 



 343

pronounced by the same vocalic root node. By comparing these candidates, two ranking arguments are discovered: both *[Place]NonFootHead and *[high]NonFootHead dominate Reciprocit​yv ​​​[Place]   ​  ​ / NonFootHead.3 (23) Non-metaphonic (transparent) -ə V

-

     [back] 'o

[Pl] -

a.  [back] 'o b.

V *[Pl]NonFtHd

*[high]NonFtHd RV[Pl] / NonFtHd

ǝ [Pl]

*

ɨ

[back] [high] [Pl]

*W

*

'o - e/o c.    [back]

[Pl]

*W

L

For opaque, metaphonic schwa suffixes, however, an underlying feature [high] is present in the underlying representation. In the tableau in (24), a potential feature [Place] projected by the suffix is not included for clarity (this would be consistent with the Richness of the Base hypothesis). As seen in the previous tableau, the feature [high] can never surface due to top-ranked *[high]NonFootHead, which forces all non-low suffixes to surface as schwa. Thus, *[high]NonFootHead rules out candidate (c), and dominates ​ReciprocityV[high] / NonFootHead, which allows suffixes projecting a feature [high] to leave this feature unpronounced. The difference between candidate (a) and candidate (b) is that the latter violates Licensing↑[high]NonFootHead / FootHead. This is the metaphony-triggering constraint defined in (20), which forces a projected feature [high] in word-final, nonfoot head position, to be non-locally licensed via pronunciation by the stressed root vowel. The winning candidate is therefore candidate (a), with metaphony, which violates ReciprocityV[high], as the stressed root vowel pronounces a 3 A candidate in which both [high] and [Place] are pronounced would also violate the feature co-occurrence constraint *{[high],[Place]}, which conflates the two independently motivated constraints presented in the tableau in (23).

344 

 Francesc Torres-Tamarit and Kathrin Linke

feature [high] that is not projected by its root node. To sum up, the feature [high] lexically associated with the suffix vowel cannot be pronounced in this position due to top-ranked *[high]NonFootHead. However, it must be pronounced by the stressed root vowel in order to satisfy Licensing↑[high]NonFootHead / FootHead. (24) Metaphonic (opaque) -ə V

-

V

     [back] [high] 'u

-

a. [back] 'o

*[high]NonFtHd

ǝ [high]

-

-

*

*

L

*

*

L

ǝ

b.    [back] [high] 'u

↑[high]NonFtHd ​ ​ [high] ​ ​ V​[high]​​ / NonFtHd v ​

*W

ɨ

c.    [back] [high]

*W

4.2 Salentino In Salentino, the nominal metaphonic suffixes contain the feature [high] and a place feature, [front] for masculine singular, and [back] for masculine plural (25). (25) Underlying representation of two metaphonic nominal suffixes {[Masculine], [Singular]}

{[Masculine], [Plural]}

•

•

[front]

[high]

[back]

[high]

As opposed to Arpinate, the feature [high] must be pronounced in a word-final, non-foot head position, because R ​eciprocityV[high] / NonFootHead, which requires the local pronounciation of [high], dominates *[high]NonFootHead. This is shown in the tableau in (26). Once the feature [high] is pronounced, it must also be licensed non-locally to satisfy the metaphony-triggering con­ straint, Licensing↑[high]NonFootHead / FootHead. The most harmonic candidate, ­candidate (a), only violates ReciprocityV[high] and *[high] / NonFootHead.

Opaque interactions between vowel merger and metaphony 



 345

(26) Metaphonic (transparent) high vowel suffix V

-

V

[front] [high] [back] ​ V​[high]​​ / NonFtHd 'i

-

↑[high]NonFtHd ​   ​v[high]  ​ ​ *[high]NonFtHd

u

a.  [front] [high] [back] 'e b.

-

-

*

L

*

L

L

u

  [front] [high] [back] 'e

*

*W

o

c.  [front] [high] [back]

*W

For the opaque, non-metaphonic high vowel nominal suffix, the feature [high] is absent from the underlying representation, which only contains the feature [front]. This suffix corresponds to feminine plural (27). (27) Underlying representation of a metaphonic nominal suffix {[Feminine], [Plural]} • [front] However, this suffix, which only contains the feature [front], surfaces as a high vowel. This is due to the insertion of the feature [high] triggered by the satisfaction of the constraint [Place]⊃[high] / NonFootHead. So far we have shown two different types of vowel merger. The one operating in Arpinate is due to deletion of the feature [high], and the one operating in Salentino is due to insertion of the feature [high]. These two types of vowel merger are the result of two different markedness constraints, *F, and an implicational constraint F⊃G, respectively.4

4 The latter type of vowel merger consisting of addition of a feature contradicts the way vowel reduction processes are understood in Element Theory, which is seen as reduction of the number of elements in prosodically weak positions. However, it is clear that in Salentino the feature [high] is not underlying, as it blocks metaphony.

346 

 Francesc Torres-Tamarit and Kathrin Linke

As illustrated in the tableau in (28), vowel merger to high in Salentino is the result of satisfying the implicational constraint demanding the pronunciation of [high] in the presence of [Place]. This means that in wordfinal, non-foot head positions, all non-low suffixes must be high vowels. This inserted feature [high] has no projection relation, and this is why it vacuously satisfies the metaphony-triggering constraint (not included in the tableau below). However, we still have to account for the fact that metaphony does not overgenerate in this case. The answer is straightforward. Metaphony would imply a double pronunciation of the feature [high], as in candidate (b), which gratuitously violates License↓[high] twice, r­ equiring any pronunciation of the feature [high] to correspond to some projection relation, and ​Reciprocity​​​ V [high]  ​ ​ twice. Thus, candidate (b) is harmonically bounded by candidate (a). In other words, insertion of [high] is minimally required to comply with vowel merger, but metaphony is unmotivated if the feature [high] is inserted (i.e. derived).5

5 The analysis has focused on high-mid vowel targets, which raise to high when affected by metaphony. When the target is a low-mid vowel, however, we get diphthongization in Salentino (/ε, ɔ/ → ['jε, 'jɔ]). This outcome can be analyzed as the effect of a feature co-­occurrence constraint *{[low],[high]}, which prohibits these two features from being pronounced by the same root node. To satisfy *{[low],[high]} and also the metaphony-­ triggering constraint, an extra ­melodic slot is inserted to host the pronunciation of the feature [high]. However, ­diphthongization co-occurs with a one-step raising of the stressed low-mid vowel in Arpinate, that is, from low-mid to high-mid, according to Calabrese (1998)’s phonetic transcriptions. This partial raising could be better understood as a phonetic coarticulation effect induced by the presence of the high front glide. Moreover, given that no (lexical) contrast exists between ['je] and ['jε] in metaphonic environments, it is difficult to see whether this presumed one-step raising is phonological, or a language-­ particular phonetic implementation of the metaphonic process consisting of adding an extra melodic slot. In any case, discussing the topic of metaphonic chain shifts goes beyond the purpose of this paper.

Opaque interactions between vowel merger and metaphony 



 347

(28) Non-metaphonic (opaque) high vowel suffix V

-

V

     [front] [front] 'e

-

v   ​  ​ / NonFtHd ↓[high] ​ ​ [high] [Pl]⊃[high] / NonFtHd ​  ​[Pl] ​ v​ 

i

a. [front] [high][front] 'i b.

-

-

-

**W

**W

L

L

L

L

ǝ

c.    [front] [front] 'e

*

i

[front] [high][front] 'e

*

*W

e

d.     [front] [front]

*W

5 Conclusions This paper has presented a unified account of two types of opaque interactions between vowel merger and metaphony in Arpinate and Salentino, which supports feature privativity and T T. A new family of constraints has been proposed, Licensing, together with markedness constraints of the co-occurrence type and the implicational type. The basic idea put forward in this paper is that ­metaphony in opaque environments is computed synchronically as a non-local licensing condition of a feature [high] only if it is underlying. On the one hand, overapplication of metaphony, as in Arpinate, is due to prohibiting the realization of a feature [high] lexically linked to a suffix, which therefore needs to be licensed nonlocally by being pronounced by a stressed vowel. On the other hand, underapplication of metaphony, as in Salentino, is due to insertion of a feature [high] in non-low suffixes, which does not, therefore, need to be licensed because it is not lexical. This T T analysis is in line with other studies acknowledging the existence of dissimilarities between phonology and phonetics (Hamann 2014; Blaho 2008;

348 

 Francesc Torres-Tamarit and Kathrin Linke

Odden 2006), and it aims at contributing on a theoretical level to the discussion on whether phonological computation abstracts away from phonetic substance.

References Backley, P. 2011. An introduction to Element Theory. Edinburgh: Edinburgh University Press. Blaho, S. 2008. The syntax of phonology. A radically substance-free approach. PhD ­dissertation. Tromsø: University of Tromsø. Blevins, J. 2004. Evolutionary phonology: the emergence of sound patterns. Cambridge: Cambridge University Press. Calabrese, A. 1985. Metaphony in Salentino. Rivista di Grammatica Generativa 9–10: 1–140. Calabrese, A. 1998. Metaphony revisited. Rivista di Linguistica 10: 7–68. Calabrese, A. 2011. Metaphony in Romance. In Van Oostendorp, M., Ewen, C., Hume, E., Rice, K. (eds.), The Blackwell Companion to Phonology, 2631–2661. Malden: Wiley-Blackwell. Dresher, B.E. 2009. The contrastive hierarchy in phonology. Vol. 121. Cambridge: Cambridge University Press. Dyck, C. 1995. Constraining the phonology-phonetics interface, with exemplification from Spanish and Italian dialects. PhD thesis. University of Toronto. Gafos, A., Dye, A. 2011. Vowel harmony: opaque and transparent vowels. In Van Oostendorp, M., Ewen, C., Hume, E., Rice, K. (eds.), The Blackwell Companion to Phonology, 2164–2189. Malden: Wiley-Blackwell. Goldrick, M. 2001. Turbid output representations and the unity of opacity. In Hirotani, M., Coetzee, A., Hall, N., Kim, J.Y. (eds.), Proceedings of the North East Linguistics Society, volume 30, 231–245. Amherst: GLSA. Hamann, S. 2014. Mismatches between phonology and phonetics. Invited talk at the 11th Old World Conference in Phonology. Harris, J., Lindsey, G. 1995. The elements of phonological representation. In Durand, J., Katamba, F. (eds.), Frontiers of phonology: Atoms, structures, derivations, 34–79. Harlow, Essex: Longman. Levelt, C., Van Oostendorp, M. 2007. Feature co-occurrence constraints in L1 acquisition. Linguistics in the Netherlands 24: 162–172. Nevins, A. 2010. Locality in vowel harmony. Cambridge: MIT Press. Odden, D. 2006. Phonology ex nihilo. Talk at the Tromsø Phonology Project Group Meeting. van Oostendorp, M. 2008. Incomplete devoicing in formal phonology. Lingua 118: 1362–1374. van Oostendorp, M. 2014. Representing variation: The view from phonological theory. Talk at Theoretical Issues in Contemporary Phonology: Reading Tobias Scheer. Parodi, E., 1892. Il dialetto di Arpino. Archivo Glottologico Italiano 13: 299–308. Prince, A., Smolensky, P. 1993. Optimality Theory: Constraint interaction in generative grammar. Unpublished ms., Rutgers University & University of Colorado, Boulder. Published 2004, Malden, MA; Oxford: Blackwell. Ribezzo, F. 1912. Il dialetto apulo-salentino di Francavilla Fontana. Martina Franca. Rose, S., Walker, R. 2011. Harmony systems. In Goldsmith, J., Riggle, J., Yu, A. (eds.), The Handbook of Phonological Theory, 240–290. Wiley-Blackwell. Walker, R. 2005. Weak triggers in vowel harmony. Natural Language & Linguistic Theory 23: 917–989.

Roberta D’Alessandro & Marc van Oostendorp

Abruzzese metaphony and the |A| eater Leiden University Centre for Linguistics & Meertens Instituut

Abstract: This paper addresses metaphony in Ariellese masculine nouns as a synchronic phonological process and argues against a morphological or lexical analysis. Metaphony in Abruzzese is analyzed as the addition of a mora to express the plural morpheme. This mora needs to be linked to a segment. The chosen segment is the most sonorant element in the syllable nucleus, i.e. the |A| element. Therefore, |A| is detached from its original mora and this results in the ­metaphonic effect observed in the stressed vowel. This analysis finds support in phrase-medial configurations, in which the |A| element is interpreted.

1 Introduction1 As the reader of this volume will know, many Romance dialects2, in Italy and elsewhere, show a phonological process called metaphony, whereby stressed stem vowels change their quality in certain morphological contexts (Calabrese (1985, 1998), Fanciullo (1994), Savoia & Maiden (1997) and many others). In the Abruzzese examples in (1), the masculine singular form of a noun, featuring a low stressed stem vowel, alternates with a plural form in which the stressed stem vowel is high: (1) singular waglionə mbrɛl:ə mat:ə

plural wagliunə ‘young person’ mbril:ə ‘umbrella’ mit:ə ‘madman’

[Abruzzese]

According to a definition by Savoia & Maiden (1997:15), “metaphony is a type of quality agreement of stressed mid or low vowels with unstressed high vowels ([i]

1 Thanks are due to Kathrin Linke, Diana Passino, Markus Pöchtrager, Jochen Trommer, and Eva Zimmermann for comments on earlier versions of this paper. 2 In this paper, we use the term “Italian dialect”, following the Italian tradition, to refer to linguistic varieties spoken in Italy directly deriving from Latin, which are independent from Italian (Rohlfs 1966–1969, Grassi, Sobrero & Telmon 1997, Loporcaro 2009, et al.).

350 

 Roberta D’Alessandro & Marc van Oostendorp

or [u]) in a following syllable”. Its effects are present in most dialects, at least as a remnant of an earlier change, where the inflectional endings [-i] and [-u] created extensive patterns of paradigmatic alternation in stressed vowels. In dialects in which final unstressed vowels have subsequently merged or fallen, like in the case of Abruzzese in (1), metaphonic alternation may constitute the sole means of differentiating morpho-syntactic properties within paradigms: the differences are no longer seen from the suffixes. For the same reason, metaphony can be argued to have become phonologically opaque in such dialects: the sources of vowel height spreading (the inflectional endings) are no longer high themselves, as they have become schwa. Furthermore, it does not always seem easy to describe metaphonic p ­ atterns in a natural way when we use restrictive models of segmental organization such as Element Theory (Backley 2011). In the examples in (1), the alternation is one of heightening of the stressed vowel in metaphonic context. This is problematic in Element Theory, which does not have an equivalent of the feature [high]. Witness other papers in this volume, the amount of phonological and syntactic microvariation in (Southern) Italian dialects is quite striking. In this study, we will be mainly concerned with Eastern Abruzzese metaphony, and in particular with an intriguing pattern of metaphony targeting masculine nouns in the variety spoken in Arielli, in the province of Chieti3. This dialect displays metaphony in a particularly clear and surprising way; and by doing so, it shows how we can indeed describe this phenomenon without making too many concessions to theoretical restrictiveness. In what follows, we will give an overview of the data in section 1; then we will comment briefly on some previous influential approaches to metaphony in section 2. This section also contains a short introduction to the relevant aspects of Element Theory. In section 3, we will outline our analysis of Abruzzese metaphony. Section 4 analyzes the interaction with a epenthesis, and section 5 gives some conclusions.

3 The data were collected through recording spontaneous speech as well as through elicitation. Metaphony being so pervasive and productive, no variation has been found within this variety, and we therefore do not report it. Unless otherwise stated, all the data are from Ariellese.

Abruzzese metaphony and the |A| eater 



 351

2 The data 2.1 Ariellese metaphony In Ariellese, metaphony targets mid and low stressed root vowels in nouns, adjectives, and verbs. The following examples illustrate the most commonly found metaphonic alternations for this particular dialect: (2) masculine nouns a > i lu canə e > i lu velə ɛ > i lu martɛllə ɔ > u lu / la nəpɔtə o > u lu waʎʎonə

li chinə li vilə li martillə li nəputə li waʎʎunə

(3) masculine adjectives a > i mattə e > i nerə ɛ > i bɛllə ɔ > u turdəllɔnə o > u novə

mittə ‘crazy-sg.pl’ nirə ‘black-sg.pl’ billə ‘beautiful-sg.pl’ turdəllunə ‘dumb-sg.pl’ nuvə ‘new-sg.pl’

‘the dog/s’4 ‘the veil/s’ ‘the hammer/s’ ‘the nephew-niece/s’ ‘the young boy / girl/s’

(4) verbs (1st vs 2nd person sing. alternation)5 a > i magnə mignə ‘I eat / you eat’ e > i vedə vidə ‘I see / you see’ ɛ > i lɛggə liggə ‘I read / you read’ ɔ > u dɔrmə durmə ‘I sleep / you sleep’ sonə sunə ‘I play / you play’ o > u (5) feminine nouns never show any metaphony: la mammə li mammə ‘the mother/s’ a la cestə li cestə ‘the basket/s’ e la cɛndə li cɛndə ‘the belt/s’ ɛ la ɣɔnnə li ɣɔnnə ‘the skirt/s’ ɔ la mojə li mojə ‘the wife / wives’ o

4 We use the IPA only when there is no corresponding sound in Italian and when the nature of the sound is crucial for understanding the argument. Otherwise, we present words in an orthography which is close to that of Italian, as customary in the transcription of Abruzzese. 5 See D’Alessandro & Roberts (2010).

352 

 Roberta D’Alessandro & Marc van Oostendorp

A result of metaphony is that some nouns that can have either masculine or feminine gender show the alternation only in the masculine form; this means that those nouns show an overt gender distinction only in the plural. The word muturettə can, for example, denote a small tractor, when masculine, or a small motorbike, when feminine; the article will show which gender is used in the singular, but in the plural the presence or absence of metaphony will signal the distinction: (6) a. nu / lu muturettə ‘a/the small motor’ b. na / la muturette ‘a/the small motorbike’

ddu / li muturittə ‘two / the small motors’ ddu / li muturettə ‘two / the small motorbikes’

However, feminine adjectives always display the relevant alternation, and therefore coincide with masculine adjectives in the plural: (7) la moja6... a > i ...mattə e > i ...nerə ɛ > i ...bɛllə ɔ > u ...turdəllɔnə o > u ...novə

li moja... ...mittə ...nirə ...billə ...turdəllunə ...nuvə

‘the wife / wives...’ ‘crazy-f.sg / pl’ ‘black-f.sg / pl’ ‘beautiful-f.sg / pl’ ‘dumb-f.sg / pl’ ‘new-f.sg / pl’

At first glance, metaphony looks undeniably morphological in examples such as this: no phonological prediction seems possible, apart from the fact that forms which have a high vowel in the singular obviously do not alternate. Otherwise the relevant factors always seem to be morphological. However, we wish to maintain that this view is incorrect. The manipulation of vowels in this case is completely systematic, and we believe that phonology is the only grammatical module that can manipulate vowels in this way. For this reason, phonology actually does play an important role. Specifically, we hope to show that metaphony is a phonological process that takes place every time the relevant conditions are met. It is of course morphology that sets up those conditions, in the sense that certain affixes trigger metaphony and others do not; but even those conditions are purely phonological. The relation between morphology and phonology is therefore not different in this respect than that between, for instance, the past

6 Observe that the form in isolation is mojə, as in (5). The -a in moja indicates that this noun is part of a phrase together with the adjective. We will return to this -a in section 1.3. and following. Observe furthermore that adjectives in Abruzzese, like in the rest of Romance, are postnominal.

Abruzzese metaphony and the |A| eater 



 353

tense suffix and voicing in English. It is true that the suffix has different shapes for different environments (it is voiced in wanted and voiceless in leaped), but this does not necessarily mean that morphology determines those different shapes: the classical analysis is that morphology just provides one exponence for this suffix, which is then adapted by phonology to the relevant context.

2.2 F ull productivity and interaction with other phonological processes Metaphony is fully productive, also in regions like Abruzzo where the endings have been reduced to schwa. In present-day Abruzzese, it has been generalized to plural formation of words that historically did not undergo metaphony. For instance, it has been extended in adjectives to the feminine plural adjectival paradigm, even though the feminine plural ending never had a high vowel to begin with (­Giammarco 1960, 1973, 1979). (8) masculine (sg / pl) *neru-*neri > *neru-*niri > nerə-nirə > nerə - nirə ‘black’

feminine (sg / pl) *nera-*nere *nera-*nere nerə-nerə nerə- nirə

Protoromance Old Abruzzese Ariellese

Between Protoromance and Old Abruzzese, the features of which are still preserved in some varieties (Giammarco 1973; Avolio 2014), metaphony applied to the masculine plural forms in adjectives, since they were the only ones featuring a high final vowel. After this, presumably the final vowels got reduced, as we have seen. The result of this is that there was no longer a difference between singular and plural in the feminine part of the paradigm. In the current form of the dialect, this difference has been reestablished, by extending the alternation of the masculine paradigm: (8)’ nerə-nirə

nerə-nirə

[Giammarco 1973:40–45]

In the adjectival paradigm, metaphony thus signals plurality. This is different for the nominal paradigm, where metaphony still applies only to masculine plurals and there is typically no difference between singular and plural forms of feminine nouns: (9) surellə-surellə (‘sister/s’ f.sg−f.pl) but not *surillə [see fratellə-fratillə (‘brother’ m.sg−m.pl’)]

354 

 Roberta D’Alessandro & Marc van Oostendorp

Words like fax or telefənə (‘telephone’), which have obviously entered the lexicon quite recently, and in any case long after the neutralization of final vowels to schwa, undergo metaphony, as (10) shows. (10) a. lu faxə - li fixə b. lu tɛlɛfənə - li tilifənə Example (10) also shows that vowel “raising” sometimes spreads to other, nonstressed, vowels in the word: the [ɛ] in both initial syllables of telefənə raise to [i] in the plural. This is visible also in the alternations exemplified in (11). We have underlined the tonic vowel for expository convenience: (11) tələvisjonə / tilivisjunə7 ‘television set / television sets’ f.sg / pl tɛlɛfənə / tilifənə ‘phone / phones’ m.sg / pl’ olandɛsə / ulandisə ‘Dutch’ m.f.sg / pl svədɛsə / svidisə ‘Swedish’ m.f.sg / pl In these cases, it is not just the stressed [o] or [ɛ] which alternates. Also the (mid) vowels preceding that vowel do. While metaphony is fully productive, the vowel harmony illustrated in (11) only concerns a limited number of words, which seem to share the characteristics of being polysyllabic and quite recent. It furthermore poses the interesting puzzle as to why the low vowel [a] does not participate in the alternation. We will leave the particulars of this extension of the original metaphony pattern to future research; in this paper we only concentrate on the alternations of the tonic vowel itself.

2.3 A-insertion Before going into our analysis of metaphony in phonological terms, we wish to introduce another phonological process which takes place more or less in the same geographic area and in very similar morphological contexts. This phenomenon, which we will dub “a-insertion” has not been connected to metaphony before, but in our view it offers a key insight into metaphonic processes. In Abruzzese, all nouns typically end in schwa, as we have seen. In some phrasal contexts, however, for instance preceding a post-nominal adjective, an (epenthetic) [a] appears word-finally, as in example (12).

7 Observe that tələvisjonə is a feminine noun which exceptionally undergoes metaphony.



(12) a. la the-f.sg ‘the house’ b. la  casa the-f.sg house-f.sg ‘the red house’

Abruzzese metaphony and the |A| eater 

 355

casə house-f.sg roscə red-f.sg

This [a] is not always etymological, as it is in this case: it can also appear on nouns that never featured an [a], as example (13) shows. (13) a. lu the-m.sg ‘the tractor’ b. li the-m.pl ‘the new tractors’

mutorə tractor-m.sg mutura tractors-m.pl

nuvə new-m.pl

We have seen that [a] can appear on feminine nouns, where it is most likely an etymological residue (see Passino this volume for a detailed description of the contexts in which [a] can appear). [a] can furthermore also appear between verb and wh- words in questions or exclamatives, possibly as a marker of a phase head (see D’Alessandro & Scheer 2013). Finally, [a] can appear on masculine plural nouns, when in phrasal contexts. The nouns on which such a non-etymological [a] such as the one in (13) can appear given the right phrasal context, are the same that display metaphony. Crucially, [a] never surfaces on masculine singular nouns, not even when followed by a postnominal adjective, as (14) illustrates. (14) a. lu canə the-m.sg dog-m.sg ‘the dog’ b. lu canə cioppə the-m.sg dog-m.sg lame-sg ‘the lame dog’ c. *lu cana cioppə the-m.sg dog-m.sg lame-sg d. li china ciuppə the-m.pl dogs-m.pl lame-pl ‘the lame dogs’ We argue that the fact that a surfaces on masculine plural, but not on masculine singular, nouns is not accidental, but it is the result of metaphony, which consists, we maintain, in moving an A element from the tonic vowel to word-final position.

356 

 Roberta D’Alessandro & Marc van Oostendorp

3 M  etaphony as regressive assimilation: an analysis and its problems The standard formal analysis of metaphony in Italian dialects assumes some form of (regressive) assimilation of a phonological feature [high] (and in some cases [front]; see Savoia and Maiden 1997:18). The autosegmental analysis of assimilation, in turn, is one of spreading; in our case spreading a feature from the suffix to the stem vowel carrying the main stress. This kind of analysis probably fits the diachrony of the process quite well. Historically, metaphony applied to masculine nominal paradigms in which the singular ending was a mid vowel -e or -o, and the plural ending was -i, and in which presumably the height feature of the plural suffix spread to the stressed vowel of the stem [Atlante Italo-Svizzero AIS, Rohlfs 1966; Giammarco 1979; Maiden 1987, 1991, Savoia and Maiden 1997): (15) Singular pede(m) *pɛde *pɛde pedə

Plural pedes *pɛdi (ɛ >i) *pidi (assimilation) pidə (neutralization of endings)

(Vulgar) Latin Eastern Romance Old Abruzzese Ariellese

Many modern synchronic accounts mimic at least the last two steps of this diachronic process, as both the assimilation and the neutralization of endings may be still productive. However, the opacity of the process causes complications for surface-oriented theories such as Optimality Theory or Government Phonology, which do not allow for phonological processes to be extrinsically ordered. Torres-Tamarit and Linke (this volume) take one route by giving a synchronic phonological analysis within OT which is not opaque. Kaye and Pöchtrager (2011) working within the Government Phonology tradition, follow another track, and argue that Germanic Umlaut and Romance metaphony are not phonological, but should rather be generated by another module of grammar, which they term a “paradigm generator”. This generator produces lexical items which are related in form (and possibly meaning). Kaye & Pöchtrager point out that, apart from the opacity, a (Government) Phonological analysis, wedded as it is to an analysis in terms of Element Theory, suffers from another problem: the process involved cannot be described easily in terms of assimilation, as there is no element corresponding to [high] within that theory (see also Ouddeken 2013). An Element-Theoretic analysis of vowels (Backley 2011) assumes that there are three monovalent primitives, appropriately called elements, and represented



Abruzzese metaphony and the |A| eater 

 357

as A, I, and U. These elements can be pronounced on their own, when they will sound as /a/, /i/, and /u/, but they can also be combined. A combination of A and I will sound as /e/, and a combination of A and U as /o/. We thus have: (16) [a] = |A| [i] = |I| [u] = |U| [e],[ɛ] = |A.I| [o],[ɔ] = |A.U| [ə] = ø [no elements] Further distinctions (such as that between [e] and [ɛ] or [o] and [ɔ]) can be derived by refinements of the representational apparatus, but for our purposes this simple version of Element Theory will suffice, as [e]/[ɛ] and [o]/[ɔ] behave the same with respect to metaphony. The schwa vowel can be seen as devoid of content. It should be clear that e.g. the metaphonic change from /e/ to /i/ cannot be the result of spreading; it has to be the result of deletion of something, viz. an A element. (17) a. [a] → [i]  = |A|    → |I| b. [e],[ɛ] → [i]  = |A.I|  → |I| c. [o],[ɔ] → [u] = |A.U| → |U| Such an analysis has indeed been proposed (most famously, by Maiden 1991), but Kaye & Pöchtrager (2011) reject this solution, as randomly deleting underlying elements according to them does not belong to the universal repertoire of phonological processes. The upshot of this is that already the change from Protoromance to Old Abruzzese in (15) above cannot be described as a regular phonological process within Element Theory, and therefore should be considered suspect, at least from the point of view of theories of synchronic phonology (we have to say that diachronic change does not necessarily obey the same constraints as synchronic processes). We follow Kaye & Pöchtrager (2011) in assuming that these two parts of the analysis—the fact that the process is opaque, and the fact that it seems to involve a strange kind of element loss—should be connected to each other. We do not follow them in concluding that the process can therefore not be phonological. The main reason is that we find the alternative they suggest—assuming that there is an independent non-phonological module which can still manipulate phonological elements and delete A’s in certain contexts—even less attractive, given that such an alternative model has an unknown theoretical power: what would be the restrictions on such a module? And why would the human language faculty provide two modules each with independent power to work on phonological representations?

358 

 Roberta D’Alessandro & Marc van Oostendorp

We thus stick to the assumption that if a process involves phonological primitives such as elements (or features), it must be phonological. Furthermore, we assume that since metaphony in Abruzzese is still productive for some syntactic categories, it has to be understood in grammatical terms. Morpho-syntax puts together certain morphemes, and metaphony has to be a regular response to a certain arrangement of such morphemes. The fact that this relation is rather problematic for the theory as it stands is of course unfortunate, but we think the problem can be repaired without destroying the glory of a restrictive view on what phonological computation can and cannot do. Our goal is to be maximally conservative and at the same time create some space for metaphony to fit in.

4 Metaphony as truncation Which tools do we have at our disposal for understanding the process at hand? We have identified two problems with analyzing metaphony in terms of elements: the process is about taking away an A element, and the process seems opaque in derivational terms. We will propose that these are two sides of the same coin; but in order to understand why, it is useful to see how taking away an A element, as a phonological response to a certain morphosyntactic structure, is a problem to begin with, since it is so only under a specific, restrictive view of the interface between morphosyntax and phonology. This is a view which subscribes to an Items-and-Arrangement model of morphology (Hockett 1954) and its interface to phonology, in particular a morpheme-based view of morphology. Under an Items-and-Arrangement view, each feature of the morphosyntactic representation corresponds to some phonological ‘material’ (features, feature bundles, segments, syllables, or larger units). Lexical items have an exponent which is a phonological representation. Deleting material is not an option in such a view, since it would correspond to giving a ‘negative’ exponent to morphosyntactic material. This view contrasts with Items-and-Process models, in which phonological words are built by applying processes to stems; under such a model ‘delete an A’ could be a possible way of deriving a word. The debate between Items-and-Arrangement and Items-and-Process is a long one, and runs across many models of formal linguistics (both could be implemented within OT, for instance). In many cases, the two views might be notational variants, but we prefer Items-and-Arrangement because it seems more restrictive, and furthermore seems supported by many data. A consequence of this is that the morphology-phonology interface should always be additive: if we merge two morphemes with each other, the resulting

Abruzzese metaphony and the |A| eater 



 359

phonological representation should be at least as phonologically complex as the two morphemes on their own. That is to say: if we have reason to believe that a plural form of an adjective has a feature [plural] and the singular has no such feature (there is no feature [singular] in the language), the plural of some adjective should always have at least as much phonological complexity as its singular counterpart. A more special result of this is that truncation should not be allowed in a morpheme-based Items-and-Arrangement model of morphology. This is at first blush problematic, as we do seem to find cases of truncation in natural languages. The perfective in Tohono O’odham is a rather well-known case in point. This perfective seems to be based on the imperfective; yet it is phonologically less complex. (18) imperfective bisck ñeok ma:k dagsp

perfective

[Tohono O’odham]

bisc ‘sneeze’ ñeo ‘speak’ ma: ‘give’ dags ‘press with hand’

(Fitzgerald 1997; Horwood 2001; cited in Trommer & Zimmermann 2014:467) The final k or p in the imperfective in each case really belongs to the stem (it is not part of some ‘imperfective’ morpheme) and it apparently gets deleted in order to form the perfective. The problem then is that we add a morpheme in order to derive the perfective, but in phonology this results in deleting a vowel.8 Trommer and Zimmermann (2014) discuss these data and show how they can be made compatible with a conservative Items-and-Arrangement view of the interface. In their proposal, the perfective morpheme does correspond to phonological material, viz. a mora. This mora however for some reason cannot get incorporated into the prosodic structure of the stem, and therefore stays unpronounced by the phonetics, which only considers material that is fully integrated into the phonological tree, following the basic principles of (Colored) Containment Theory (Prince & Smolensky 2004; Van Oostendorp 2006, 2007-to appear). Trommer & Zimmermann (2014) furthermore adhere to the principles of Autosegmental Phonology, such as the Well-Formedness Condition of Goldsmith (1976, WFC), which claims that every prosodic position should correspond to precisely one segment, not more, not less. They relativize these constraints, as their

8 A further complication arises in words in which the final consonant is preceded by a high vowel which in turn is preceded by a coronal consonant. In those cases, also the high vowel is deleted. We follow Fitzgerald (1997); Harwood (2001), and Trommer & Zimmermann (2014) in treating this as a separate phenomenon.

360 

 Roberta D’Alessandro & Marc van Oostendorp

analysis is captured in terms of OT, but we choose to remain neutral on that particular aspect of the analysis in this paper. Taken together, the principles of Colored Containment and the WFC give us the following principles, which will be essential for our analysis of Abruzzese metaphony (the analysis we present here is slightly simplified over that presented by Trommer & Zimmermann 2014, but does not alter its essence): (19) a. Every mora should be linked to something (either higher-order prosody or segmental material) b. A segment cannot be linked to more than one mora. c. A syllable cannot host more than three mora’s. These principles together give the result that the purely additive morphology of Tohono O’odham sounds like truncation in the phonetic signal. The following picture illustrates why this is the case. Gray elements denote the stem ‘to give’; the dark gray mora in the end is the phonological exponent of the perfective: σ

(20)

μ m

a

μ

μ

μ

k

The perfective is only a mora. According to (19a) it needs to be linked to an other structure in some way; but according to (19b) it cannot get incorporated into the syllable of the stem, because that is already trimoraic. So the only thing it can do in order to satisfy all requirements is attract the k segment at the end of the stem. This segment cannot be linked to two moras, however, because of (19c), and as a result it now loses its link to the third mora of the syllable—that is why we have drawn a dashed line there. The end result is that the k is only linked to the ‘loose’ mora. But since this mora is not linked to any higher-order prosodic structure it will not be interpreted, and neither will the segment hanging under it, which by transitivity is also not linked to the phonological word (or whatever higher nodes we may assume here). So although the phonological output structure in (20) is rather complicated, and indeed more complicated than the imperfective from which it is derived, it contains uninterpretable material at the interface to phonetics, which therefore sounds as truncation. We believe this to be an elegant and interesting analysis of truncation in general, and we wish to observe that the Abruzzese data can also be seen as a form of truncation, albeit not of a whole segment, but just of one element, the

Abruzzese metaphony and the |A| eater 



 361

|A| element.9 In this case, truncation of that element corresponds to addition of a feature [plural] or of the two features [masculine, plural], as we have seen. We propose that this feature (bundle) corresponds to an empty mora in Abruzzese. The difference, however, is that this bundle does not need (or want) to be filled by a full consonant, but only by the element |A|. The reason for this may be that |A| counts as the most sonorous of elements (Hermans and Van Oostendorp 2005; Pöchtrager 2006; Backley 2011). We thus get representations such as the following: (21) mesə / misə (month-m.sg / pl) a. σ σ μ

b.

m

m

μ

|A. I| e

s

σ

σ

μ

μ

μ

ə

a

|A. I| i

s

ə

Again, the mora needs to be linked to some segmental material (schwa-syllables cannot be heavy in Abruzzese, so it cannot be linked to a syllable which already has a mora; we will return to this below). It therefore looks for an |A| element, and finds it in the stressed syllable. Parallel to Tohono O’odham, the |A| element is not allowed to be linked to two moras, and therefore loses its connection to its original host. The end result is that it is now linked to the uninterpretable mora only, and therefore will not be pronounced. The Abruzzese word in metaphony context thus ‘loses’ its lowness in the same way that Tohono O’odham perfectives lose their final consonant: phonologically, it stays where it is, but it gets in a position which will be ignored by the phonetics. The mora thus functions as an ‘|A| eater’, grabbing an underlying |A| in its environment and gnawing it away.

9  In the terminology of Arndt-Lappe and Alber (2012), metaphony as we analyze it here would probably be termed ‘subtraction’, but as we believe that truncation and subtraction are essentially the same thing, we have decided not to make that terminological distinction in the main text.

362 

 Roberta D’Alessandro & Marc van Oostendorp

We do obviously have to take into account that there is a difference between Tohono O’odham and Abruzzese: in the former language, the mora ‘eats’ a whole consonant, but in the latter only one specific element is eaten, |A|. We suppose that this is the result of the interplay of two parameters or rankable constraints. First, some kind of (language-specific) markedness will decide what kind of material can fill the mora: in Tohono, this is consonantal, in Abruzzese it is vocalic. And second, in Tohono, apparently some kind of integrity constraint on segments is operative, so that segments move as a whole, whereas Abruzzese moras are satisfied with eating only the most sonorous element. Notice that both of these kinds of parameters would be independently necessary in the analysis of any spreading process: in some languages consonants spread (geminate) whereas in other languages vowels lengthen and in some languages, segments spread as a whole, while in other languages this only happens to individual elements. Note that in this analysis also the alleged ‘opacity’ disappears. The synchronic analysis does not directly mirror the diachronic order of affairs. In particular, there is no reason to assume a synchronic process of vowel reduction in Abruzzese, let alone an extrinsic ordering of this process with respect to metaphony. There is some kind of ordering of the attraction of |A| to the empty mora and the subsequent ‘deletion’ of that mora, but that ordering is intrinsically given by the way in which the grammar has been set up for independent reasons, since the former process is phonological and the latter a result of phonetic implementation which necessarily follows. The reason why metaphony targets only the stressed vowel is that all vowels following that vowel are reduced in Abruzzese, and therefore will not contain an |A| element. Even if two vowels are following that vowel, the source of |A| will still be the stressed vowel: (22) a. ʹmɔ.nə.kə b. ʹmɔ.b:ə.lə c. ʹjɛ.nə.mə

mu.nə.tʃə (monk-m.sg / pl) mu.b:ə.lə (furniture-m.sg / pl) ji.nə.mə (son-in-law.my-m.sg / pl)

(23) jenəmə, jinəmə a.

j

(son-in-law.my- m.sg / pl)

σ

σ

σ

μ

μ

μ

|A. I| ɛ

n

ə

m

ə

Abruzzese metaphony and the |A| eater 



b.

j

σ

σ

σ

μ

μ

μ

μ

ə

ə

|A. I| i

n

ə

m

 363

In the figures we have given so far, the stressed vowel was always mid front, but a parallel analysis applies to underlyingly back stressed vowels. Again these vowels just lose their |A| to the |A| eater, and the result will be a high back rounded [u], regardless of how many reduced vowels follow, as the following examples illustrate: (24) tonə / tunə a.

t b.

t

(thunder-m.sg / pl)

σ

σ

μ

μ

|A. U| ɔ

n

σ

σ

μ

μ

μ

ə

ə

|A. U| u

n

(25) monəke / munətʃə a. σ

(monk-m.sg / pl)

μ m b.

m

ə

|A. U| ɔ

n

σ

σ

μ

μ

ə

n

ə

σ

σ

σ

μ

μ

μ

μ

ə

ə

|A. U| u

n

ə

tʃ

364 

 Roberta D’Alessandro & Marc van Oostendorp

A more problematic case are forms such as cane ‘dog’ which have an [a] vowel in the singular, i.e. which have only an A element. The result of the empty mora attracting this A element should be that the stressed vowel becomes [ə]: the plural ‘dogs’ should be cənə, but it is chinə instead: apparently an |I| has been inserted: (26) canə, chinə a.

(dog-m.sg / pl)

σ

σ

μ k b.

k

μ |A| a

n

ə

σ

σ

μ

μ

μ

ə

ə

|A|

n

i An important question to ask now obviously is where this |I| comes from. We can think of two possible answers. The first is that schwa is not a possible stressed vowel in Ariellese (or in many other languages of the world). The |I| element would then be a possible repair for this situation, and function as a kind of epenthetic element. Passino (this volume) objects to such an analysis—albeit for a different dialect—observing that it seems more plausible that [a] is the epenthetic vowel in Abruzzese given the kinds of facts we discuss below. Our analysis of those data does not involve epenthesis, so that we are not sure that the objection applies to it, but more generally one can wonder what other evidence there would be for epenthesis of |I| elements in Abruzzese; we are not aware of any. An alternative, then, would be that the plural affix actually contains both a mora and a floating |I| element. The |I| element cannot be linked to the empty mora as they are part of the same morpheme; but it can be linked to the stressed vowel. Yet, if this vowel is underlyingly [e], it already features an |I|, so linking another one has no effect. If the vowel is [o], linking |I| to it would lead to a combination of |I| and |U|, which is ruled out in Abruzzese in general, as the language does not have front rounded vowels. So the only place where the |I| element can be seen to surface is when the stressed vowel is an [a]. This analysis

Abruzzese metaphony and the |A| eater 



 365

might be seen as rather arbitrary when taken for Abruzzese in its own right, but it has as an advantage that it links this particular type of metaphony to that in other languages, in which this is sometimes expressed as fronting, i.e. as I linking more generally.

5 Interaction with [a] ‘epenthesis’ We now return to the interaction of metaphony with another vocalic alternation: that of the epenthesis of [a], illustrated in (14), repeated here for convenience. canə (27) a. lu the-m.sg dog-m.sg ‘the dog’ b. lu canə cioppə the-m.sg dog-m.sg lame-sg ‘the lame dog’ c. *lu cana cioppe the-m.sg dog-m.sg lame-sg d. li china ciuppə the-m.pl dogs-m.pl lame-pl ‘the lame dogs’ The [a] in (27) is not etymological in any way, and there is no reason to suppose it is synchronically underlying in the masculine plural suffix. Since (27d) represents a masculine form, this plural form should etymologically have been a high front [i]; this is the historical reason why metaphony took place here. We thus have a somewhat paradoxical pattern here: the ending which is supposed to have synchronically triggered height metaphony, itself surfaces as low. But notice that the pattern starts making at least some sense under the view presented above: (28) lu mutorə novə, li mutura nuvə ‘the new tractor’ (sg /pl) σ σ σ σ a. μ m

|U| u

t

μ

μ

|A.U| ɔ r

ə

σ

μ n

|A.U| o

μ v

ə

366 

 Roberta D’Alessandro & Marc van Oostendorp

b.

m

σ

σ

σ

μ

μ

μ

|U| u

t

|A.U| u r

μ a

n

σ

σ

μ

μ

μ

ə

ə

|A.U| u

v

Notice that in cases where the [a] surfaces, there is no preceding [ə]. In Element Theory [ə] is the realization of an empty vowel without elements. We submit that this interpretation can be context-sensitive, and will only happen in the absence of an adjacent full vowel. We are thus proposing that there is no such thing as [a] insertion, but that an [a] that is generated by independent mechanisms is phonetically realized only when specific phonotactic conditions are met. Specifically, [a] is realized in this extrametric position as this is an intra-phrasal position, i.e. a strong position. [a] in Abruzzese is never reduced in unstressed position inside a word, but it is at the end of a word, as (29) and (30) illustrate. The intraphrasal position is equivalent to an intraword position, hence [a] is fully realized. (29) cavallə / capabballə / waglionə (‘horse / downhill / boy’) (30) mammə, casə (‘mother / house’) Recall that etymological [a] is also not reduced in the intraphrasal position: (31) la casə / la casa novə

(‘the house / the house new’)

6 Conclusions Metaphony in Ariellese masculine nouns is neither only morphological nor only lexical: it is instead the result of a synchronic phonological process, consisting in adding a mora to the word to express the plural morpheme. This mora needs to be linked to a segment. The stressed vowel, being the most prominent unit, is the obvious candidate. More specifically, the chosen segment is the most sonorant element in the nucleus, i.e. the |A| element. In order to keep a correct syllabic structure, |A| will need to be detached from its original mora. The result will be both the realization of the remaining element in the stressed syllable, bearing a metaphonic “effect”, and, when in phrasal configuration, the realization of the |A| element in the extrametrical position.



Abruzzese metaphony and the |A| eater 

 367

References Avolio, Francesco. 2014. Tratti fonetici, morfosintattici e lessicali tra Italia mediana e Italia meridionale: una prima panoramica. Paper presented at the Workshop Fenomeni grammaticali sul confine mediano/alto-meridionale, University of Viterbo. Arndt-Lappe, Sabine and Birgit Alber. 2012. Templatic and subtractive truncation. In J.Trommer (ed.), The morphology and phonology of exponence: The state of the art. Oxford: Oxford University Press. 289–325. Backley, Phillip. 2011. An introduction to element theory. Edinburgh: Edinburgh University Press. Calabrese, Andrea. 1985. Metaphony in Salentino. Rivista di Grammatica Generativa 9–10: 1–140. Calabrese, Andrea. 1998. Metaphony revisited. Rivista di Linguistica 10: 7–68. D’Alessandro, Roberta and Ian Roberts. 2010. Past participle agreement in Abruzzese: split auxiliary selection and the null-subject parameter. Natural Language and Linguistic Theory 28: 41–72. D’Alessandro, Roberta and Tobias Scheer. 2013. Phase head marking. In Tokizaki, H. & Y. Dobashi (eds.), Linguistic analysis. Special issue on Universal Syntax and ­Parametric Phonology 38 (4): 305–330. Fanciullo, Franco. 1994. Morfo-metafonia. In Cipriano, Di Giovine e Mancini, (eds.), Miscellanea di studi linguistici in onore di Walter Belardi, 571–592. Rome: Il Calamo II. Fitzgerald, Colleen. 1997. O’odham Rhythms. PhD dissertation, University of Arizona. Giammarco, Ernesto. 1960. Grammatica delle parlate d’Abruzzo e Molise. Pescara. Giammarco, Ernesto. 1973. Abruzzo dialettale. Pescara: Istituto di Studi Abruzzesi. Giammarco, Ernesto. 1979. Abruzzo. Serie Profilo dei dialetti italiani. Pisa: Pacini Editore. Goldsmith, John. 1976. Autosegmental phonology. PhD Dissertation, MIT. Grassi, Corrado, Alberto Sobrero and Tullio Telmon. 1997. Fondamenti di dialettologia. Rome-Bari: Laterza. Hermans, Ben J. H. and Marc van Oostendorp. 2005. Against the sonority scale: evidence from Frankish tones. In H. Broekhuis, N. Corver, & R. Huybregts, (eds.), Organizing grammar, 427–437. Berlin: Mouton de Gruyter. Hockett, Charles F. 1954. Two models of grammatical description. Word 10: 210–234. [Readings in Linguistics, vol. I, 386–399]. Horwood, Graham 2001. Antifaithfulness and Subtractive Morphology. Ms.,Rutgers University, available as ROA 466-0901. Kaye, Jonathan and Markus A. Pöchtrager. 2011. What is this thing called phonology? Talk given at the Leiden University Center for Linguistics, 25. November. Loporcaro, Michele. 2009. Profilo linguistico dei dialetti italiani. Rome – Bari: Laterza. Maiden, Martin. 1987. New perspectives on the genesis of Italian metaphony. Transactions of the Philological Society 85: 38–73. Maiden, Martin. 1991. Interactive morphonology. Metaphony in Italy. New York: Routledge. Oudekken, Nina. 2013. Metaphony in Italian. An analysis in Element Theory. ResMA Thesis, Leiden University.

368 

 Roberta D’Alessandro & Marc van Oostendorp

Passino, Diana. This volume. Progressive metaphony in Teraman Abruzzese. Pöchtrager, Markus. 2006. The structure of length. PhD Dissertation. University of Vienna. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti: Fonetica. Turin: Einaudi. Prince, Alan, and Paul Smolensky. 2004. Optimality Theory: Constraint Interaction in Generative Grammar. Blackwell Publishing. Savoia, Leonardo & Martin Maiden. 1997. Metaphony. In Maiden, M. & M. Parry, (eds.), The dialects of Italy, 15–25. London: Routledge. Torres-Tamarit, Francesc and Kathrin Linke. This volume. Opacity in vowel merger – metaphony interactions. Trommer, Jochen & Eva Zimmermann. 2014. Generalized mora affixation and quantitymanipulating morphology. Phonology 31: 463–510. Van Oostendorp, Marc. 2006. A Theory of Morphosyntactic Colours. Ms., Meertens Institute, Amsterdam. http://egg.auf.net/06/docs/Hdt%20Oostendorp%20coulours.pdf. Van Oostendorp, Marc. To appear [2007] Input-Output Relations in Phonology: Coloured Containment Theory. Interface Explorations series. Mouton de Gruyter.

Index A demotion 3, 4, 148, 149, 158, 160, 161, 163, 164, 166, 167, 168, 171, 172, 173, 174, 175, 176, 184, 185, 186, 193 Abruzzese 2, 3, 89, 147, 148, 149, 153, 349, 350, 351, 352, 353, 354, 356, 357, 358, 360, 361, 362, 364, 365, 366 Abruzzian 60, 74 acoustic 6, 14, 34, 37, 97, 139, 161, 162, 163, 195, 197, 198, 201, 203, 204, 205, 208, 209, 212, 214, 215, 235, 255, 304 Agnonese 56, 72, 73, 74, 79 agreement 16, 98, 102, 105, 106, 160, 227, 228, 231, 238, 240, 243, 250, 255, 256, 305, 340, 349 Airola 4, 179, 180, 190 Airolano 179, 180, 181, 182, 183, 184, 187, 190, 192, 193 Àllai 24, 25, 26, 27, 28, 29, 30, 47 allomorphic morphological rule 55, 65 allomorph selection 116 allophone 130 Altamurano 89, 90, 92, 93, 96, 97, 98, 99, 100, 102, 103, 105, 106, 107, 108, 110, 111, 112, 113, 114, 115, 116, 118, 119, 122, 123 amplitude 199, 200, 201, 203, 204, 205, 206, 212, 214, 304 Andria 9, 22, 23 Antrodoco 261, 269 Apricena 314, 323 Apulian 74, 89, 99 Arpinate 5, 107, 334, 335, 336, 337, 338, 341, 342, 344, 345, 346, 347 Arpino 269 articulatory 6, 31, 64, 195, 197, 202, 209, 213, 214, 216, 217, 234, 255 Ascrea 269 Assamese 325 assimilation 1, 3, 4, 14, 15, 31, 32, 91, 133, 149, 150, 160, 167, 195, 196, 201, 203, 204, 208, 209, 214, 227, 233, 259, 282, 294, 303, 304, 311, 325, 356 assimilatory raising 55, 82 ATR metaphony 18, 20

auditory evoked potentials 197 autonomy of morphology 108, 109, 113, 114, 118 auxiliary 227, 229, 230, 242, 247 Bedouin Arabic 316 behavioral 6, 195, 202, 203, 211, 216 Bellante 148, 172, 173 Benevento 181 breaking 166, 167 Calabrian 11, 13, 74, 75 Calvello 259, 260, 261, 263, 265, 269, 271, 273 Canavese  128 Canosa di Puglia 324 Casalincontrada 168, 169, 170, 171, 174, 175 Catalan  59, 290 causative verb 227, 246, 247, 248 Central Catalan 290 chain shift 277, 278, 279, 280, 281, 284, 285, 296 Classical Latin 55, 57 clitic 182, 187, 188, 193, 223, 224, 228, 233, 237, 238, 239, 243, 246, 247, 248, 250, 305, 63 coarticulation 30, 31, 66 consonant blocking 4, 302, 314, 315, 316, 317, 323, 324, 325, 327, 328 constraint conjunction 286, 296 constraint-based  5, 334 containment 5, 286, 333, 334, 359, 360 contextual allomorphy 101, 102 contrast 3, 6, 17, 36, 45, 47, 57, 67, 70, 80, 105, 109, 129, 135, 136, 141, 143, 162, 163, 164, 199, 204, 205, 207, 212, 215, 216, 231, 235, 250, 259, 260, 267, 268, 269, 271, 273, 277, 279, 282, 283, 296, 297, 301, 302, 303, 308, 316, 325, 328, 333, 346 contrast preservation 259, 260, 267, 268, 269, 271, 273, 296 contrastivity 9, 15, 30, 43, 49, 232 copy harmony 307, 308, 309, 310, 311, 312, 314, 317, 319, 320, 321, 326

370 

 Index

counterbleeding opacity 185 crisp edge constraint 308, 322 CVCV Theory 180, 183 Daco-Romanian 80 Dependency Phonology 139, 141, 142, 143, 144 derived environment blocking 277 determiner 188, 191, 193, 223, 237, 238, 239, 250 dialect typology 281 diphthongization 5, 12, 13, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 70, 71, 73, 74, 75, 76, 78, 79, 80, 81, 82, 83, 91, 92, 93, 94, 96, 97, 98, 100, 103, 105, 111, 118, 119, 120, 128, 144, 148, 167, 168, 170, 171, 174, 176, 180, 181, 185, 186, 193, 259, 261, 265, 267, 269, 278, 337, 346 direct licensing 280, 282 Distributed morphology 2, 89, 90, 125 electroencephalographic 197 Element Theory 3, 4, 45, 49, 148, 157, 158, 160, 161, 162, 163, 164, 180, 183, 184, 193, 338, 345, 350, 356, 357, 366 Empty Category Principle 183 evolutionary phonology 341 experimental phonetic 55, 56, 57, 64, 71 externalization 3, 10, 12, 31, 50, 221, 240, 253, 254, 255, 256 featural affixation 2, 260, 273 feature co-occurrence constraint 343, 346 feature privativity 347 floating feature 2, 90, 100, 109, 110, 113, 118, 123, 134, 137, 260, 264, 266, 273, 284 Florentine 56, 62 Foggia 269 Foggiano 333 French 55, 56, 58, 59, 62, 97, 165 fronting 91, 94, 95, 97, 98, 99, 105, 111, 118, 119, 128, 133, 136, 137, 138, 140, 171, 365 Galician-Portuguese 56 Garfagnana 312, 313, 314, 318 gemination 91, 181, 182, 188, 259, 261, 262, 263

gender 11, 21, 69, 179, 180, 192, 227, 252, 253, 255, 265, 352 Generative Phonology 65 German 55, 71, 97, 165, 259 Government 10, 139, 165, 226, 227, 232, 242, 249 Government Phonology 10, 127, 139, 141, 232, 356 Grado 260, 269, 274 harmony 2, 3, 4, 14, 16, 23, 24, 26, 27, 29, 30, 32, 37, 38, 40, 42, 43, 46, 47, 49, 50, 134, 158, 164, 187, 222, 223, 224, 226, 227, 232, 233, 236, 249, 254, 255, 259, 260, 261, 263, 267, 271, 272, 273, 274, 294, 295, 297, 301, 302, 303, 304, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 340, 354 headedness 46, 161, 162, 163, 164 hypermetaphony 106, 107, 113, 269 I addition 148, 160, 161, 168, 171, 174, 175, 176 Iacurso 9, 19, 20, 44 identity licensing 278, 280, 281, 283, 285, 293, 297 implicational constraint 339, 345, 346 Inclusiveness Condition 9 indirect licensing 280, 281, 282, 297 insertion 101, 102, 105, 114, 116, 123, 223, 229, 233, 235, 283, 333, 335, 339, 345, 346, 347, 354, 366 Ischia 277, 278, 281, 286 Isolaccia 278 Italian 3, 9, 12, 15, 22, 41, 43, 44, 55, 56, 57, 58, 60, 62, 63, 65, 70, 71, 72, 79, 82, 89, 90, 97, 99, 107, 118, 132, 133, 147, 148, 150, 151, 152, 154, 155, 156, 179, 180, 181, 182, 183, 189, 190, 221, 232, 238, 247, 261, 274, 277, 278, 279, 280, 283, 288, 292, 293, 296, 297, 323, 349, 350, 351, 356 Italo-Romance 1, 5, 11, 49, 56, 57, 60, 65, 68, 74, 82, 83, 93, 102, 127, 128, 147, 168, 171, 173, 174, 176, 260, 261, 265, 333, 335, 334, 338

Index 

Kanembu 260, 264 Làconi 9, 24, 25, 26, 27, 28, 29, 30, 47, 48 laboratory phonology 64 latency 199, 200, 204, 205, 206, 212, 213, 215 Lena Asturian 274 lengthening process 62, 63, 113 licensing 4, 5, 10, 16, 18, 37, 41, 48, 49, 142, 143, 165, 278, 279, 280, 281, 282, 283, 284, 293, 297, 301, 302, 303, 304, 305, 306, 307, 308, 309, 310, 315, 317, 318, 320, 321, 323, 324, 325, 326, 327, 328, 333, 335, 339, 340, 341, 343, 344, 347 licensing constraints 142, 301, 304, 307, 310, 323, 324, 325, 326, 327, 328 life cycle of phonological rules 64 liquid 315, 321, 322 local conjunction 5, 272, 286 local constraint conjunction 286, 296 locality 102, 114, 134, 196, 231, 242, 248, 250 Logudorese Sardinian 56, 61, 78 Lombard 15, 128, 129 Lombardian 278, 292 Lucanian 74 Lugo 260, 264 Macedo-Romanian 80 Makassarese 316 Māori 290 markedness paradox 278, 279, 280, 282, 284, 297 Mascioni 9, 16, 17, 18, 19, 20, 44, 45 maximal economy 333 maximal licensing 302, 308, 309, 310, 315, 317, 318, 320, 321, 325, 326, 327 maximally economic 335 memorized stem alternants 90, 118 metaphonic trigger 2, 334, 337, 341, 342 Milena 9, 32, 33, 34, 35, 36, 37, 39, 40, 41, 42, 44, 46, 47 mismatch negativity 198, 199 Molisano 68, 89 monophthongization 12, 63, 93, 98, 99, 155, 167, 168, 175 morphemic harmony 259, 260, 261, 271, 273

 371

morphological exponents 11, 132 morphologization 12, 65, 168, 176, 264, 265, 266, 272, morphologized metaphony 22, 41 morphology 1, 2, 10, 11, 31, 89, 90, 100, 101, 108, 109, 113, 114, 116, 118, 131, 132, 144, 205, 221, 224, 253, 273, 352, 353, 358, 359, 360 morphophonological rule 55, 65, 89 morphophonology 2, 89 Neapolitan 56, 57, 59, 60, 62, 63, 65, 67, 68, 69, 70, 71, 79, 80, 83, 96, 171, 174, 175, 181, 253 negation 15, 224, 226, 244 neurophysiological 6, 195, 197 neutralization 11, 22, 27, 31, 48, 98, 111, 266, 267, 277, 279, 282, 289, 290, 291, 292, 311, 319, 320, 324, 325, 354, 356 New Zealand English 290 number 11, 21, 69, 129, 131, 133, 137, 143, 152, 168, 190, 191, 192, 211, 227, 240, 244, 253, 265, 266, 285, 295, 324, 325, 335, 341, 345, 354 Old Aretino 56 Old French 55 Old High German 55 Old Ligurian 128 Old Romanesco 60, 173 Olivone 314 opacity 1, 4, 5, 6, 185, 280, 341, 356, 362, Optimality Theory 5, 7, 277, 280, 281, 305, 356 overapplication 5, 333, 334, 335, 336, 347 Particle Phonology 139, 141 Paulilàtino 9, 16, 17, 18, 19, 21, 27, 29, 45 perception 64, 152, 195, 197, 214, 216, 323, 340 phase 185, 228, 229, 230, 231, 232, 237, 238, 239, 242, 243, 245, 247, 248, 249, 250, 355 phonetic motivation 64, 65, 71, 132 phonological components 226 phonological features 3, 9, 11, 18, 19, 123, 127, 128, 132, 136, 137, 138, 334, 335, 338, 341

372 

 Index

phonological knowledge 215, 216 phonological rule 55, 65, 79, 81, 105, 109, 231, 340 phonosyntax 224 plural 2, 4, 29, 31, 69, 89, 102, 103, 104, 105, 106, 107, 109, 113, 120, 121, 122, 130, 131, 132, 133, 134, 137, 138, 168, 172, 179, 180, 182, 186, 187, 188, 189, 190, 191, 192, 234, 241, 260, 261, 262, 264, 265, 266, 272, 334, 336, 337, 344, 345, 349, 352, 353, 354, 355, 356, 359, 361, 364, 365, 366 Portuguese 55, 58, 60, 61, 62, 66, 290 Poschiavo 278 Positional Faithfulness 279, 295, 304, 305, 306, 307, 308, 315, 319, 320, 322, 326, 328 positional licensing 4, 301, 302, 303, 304, 305, 306, 307, 309, 310, 315, 318, 323, 324, 325, 326, 328 positional markedness 279, 281, 303 post-tonic harmony 4, 301, 302, 307, 308, 311, 312, 314, 315, 316, 317, 318, 319, 320, 321, 323, 324, 325, 327, 328 preservation of -a 253 pretonic reduction 154 privative 3, 5, 10, 43, 159, 283, 284, 286, 333, 335, 336 production 64, 195, 197, 198, 200 projection relation 334, 339, 340, 341, 346 prominence 4, 141, 142, 143, 163, 237, 301, 304, 305, 308, 312, 319, 322, 326, 327, 328 pronunciation relation 334, 339, 340 propagation of /u/ 253 Proto-Balcan-Romance 81 Proto-Romance 55, 60, 62, 72, 152, 168, 169, 170, 171, 172, 173, 175 Provençal 59 Qafar 262 quantifier 244, 248, 253 Raddoppiamento 4, 179, 180, 181, 187, 229, 230, 262 readjustment rules 101, 118, 119 reciprocity 338, 339 restructuring verb 246

Richness of the Base 341, 343 Romansh 3, 13, 60 root node 97, 103, 104, 107, 334, 335, 339, 341, 342, 343, 344, 346 Russian 290 S. Marco d’Alunzio 32 Salentino 5, 56, 74, 75, 76, 77, 78, 79, 195, 333, 334, 335, 337, 338, 344, 345, 346, 347 Salento 6, 69, 75, 77, 78, 79, 195, 196, 202 Sannio Beneventano 71, 182 Sanskrit 295, 296 Sant’Oreste 311, 312, 315, 317, 319 Sardinian 4, 9, 11, 13, 15, 16, 18, 19, 20, 21, 23, 24, 26, 27, 29, 30, 31, 45, 47, 49, 50, 56, 57, 58, 59, 61, 62, 63, 65, 66, 69, 78 Serviglianese 69 Servigliano 60, 259, 260, 261, 269, 271, 272, 273, 274, 302, 303, 304, 305, 306, 307, 308, 311, 315, 317, 319, 326 Settimo S. Pietro 9, 20, 21, 22, 30, 41 Siuslawan 290 sonority reduction 290, 319 sound change 5, 55, 57, 63, 64, 65, 76, 82, 99, 341 Southern Umbro 281 Spanish 1, 55, 57, 58, 63, 259 Sri Lankan Portuguese Creole 290 Standard Italian 56, 58, 72, 179, 181, 190, 238, 247, 278, 279 Stigliano 223, 224, 226, 227, 232, 234, 235, 236, 239, 240, 241, 249, 250, 251, 252, 253 stress 1, 2, 64, 90, 91, 92, 98, 105, 106, 113, 114, 116, 149, 160, 167, 188, 190, 222, 232, 235, 239, 259, 260, 264, 267, 268, 269, 270, 271, 272, 273, 282, 301, 310, 322, 326, 327, 356 stress-dependent harmony 259 stressed nucleus 9, 11, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 30, 32, 37, 38, 39, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 144, 221, 223, 232, 234, 235, 236, 237, 253, 255 Strict CV 165 stringent constraints 290

Index 

substance-freedom 335 suppletion 2, 90, 101, 108, 114, 115, 117, 118 suppletive 89, 90, 114, 115, 116, 117, 118, 119, 123 surface 10, 27, 48, 49, 69, 70, 91, 92, 96, 98, 105, 111, 113, 117, 118, 119, 123, 165, 167, 173, 174, 175, 185, 197, 198, 228, 271, 279, 280, 294, 334, 336, 338, 341, 342, 343, 364 Sursilvan 59 Sutera 9, 32, 37, 46 syllable structure 56, 58, 59, 61, 62, 72, 79, 165, 233 syntagmatic correspondence 5, 277, 278, 280 syntagmatic identity 283, 285, 293, 294, 297 Teramo 3, 147, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 162, 163, 165, 167, 168, 171, 172, 173, 174, 175, 176 Ticinese dialect 314 tongue 6, 17, 62, 63, 97, 136, 197, 198, 200, 201, 202, 209, 213, 214, 216, 279, 282, 283, 294, 314, 323 transparency 64, 65, 133, 134, 144 Tricase 195, 197, 200, 202, 203, 206, 212, 214, 215, 216 Turbidity Theory 5, 333, 334 Tuscan  58, 62, 147, 155, 181 ultrasound 6, 197, 198 Umbertide 312 umlaut 133, 168, 177, 259, 293, 356 unary features 139 underapplication 5, 333, 335, 347

 373

underapplies 334, 337 underlying 5, 10, 11, 14, 15, 21, 34, 42, 43, 70, 91, 92, 95, 98, 102, 110, 111, 113, 116, 119, 120, 124, 140, 142, 160, 165, 166, 175, 176, 182, 185, 188, 222, 232, 233, 252, 259, 262, 264, 266, 267, 268, 273, 274, 279, 308, 313, 334, 335, 341, 342, 343, 344, 345, 347, 361, 363, 364, 365 underlying length 165, 175 unstressed position 38, 46, 129, 135, 140, 153, 159, 223, 235, 236, 269, 277, 279, 282, 340, 366 unstressed vowel neutralization 279 Valsesia 313 Veneto 269, 278, 281 Villapriolo 9, 32, 34, 40, 41, 42, 43, 44 vowel deletion 91, 92, 94, 105, 114 vowel fronting 95, 105, 118, 119 vowel height 3, 5, 127, 138, 139, 141, 148, 277, 278, 284, 285, 286, 290, 294, 296, 297, 350 vowel merger 5, 315, 319, 333, 334, 336, 337, 340, 345, 346, 347 vowel raising 105, 134, 139, 148, 164, 277, 278, 279, 281, 285, 292, 297 vowel reduction 117, 181, 185, 186, 189, 264, 269, 282, 290, 297, 314, 320, 336, 345, 362 vowel system 57, 58, 61, 67, 69, 77, 78, 89, 90, 92, 97, 129, 136, 140, 142, 143, 151, 152, 153, 156, 203, 214, 263, 335, 338 Yucatec Maya 325