Syllable and Word Languages 9783110346992, 9783110343458

Citation preview

Javier Caro Reina, Renata Szczepaniak (Eds.) Syllable and Word Languages

linguae & litterae

Publications of the School of Language & Literature Freiburg Institute for Advanced Studies Edited by Peter Auer, Gesa von Essen, Werner Frick Editorial Board Michel Espagne (Paris), Marino Freschi (Rom), Ekkehard König (Berlin), Michael Lackner (Erlangen-Nürnberg), Per Linell (Linköping), Angelika Linke (Zürich), Christine Maillard (Strasbourg), Lorenza Mondada (Basel), Pieter Muysken (Nijmegen), Wolfgang Raible (Freiburg), Monika Schmitz-Emans (Bochum)

Volume 40

Syllable and Word Languages Edited by Javier Caro Reina and Renata Szczepaniak

ISBN 978-3-11-034345-8 e-ISBN [PDF] 978-3-11-034699-2 e-ISBN [EPUB] 978-3-11-038395-9 ISSN 1869-7054 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliografische Information der Deutschen Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de. © 2014 Walter de Gruyter GmbH, Berlin/Boston Typesetting: Johanna Boy, Brennberg Printing: Hubert & Co. GmbH & Co. KG, Gottingen ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Table of contents Acknowledgements 

 IX

List of abbreviations and symbols 

 XI

Peter Auer Preface   1 Javier Caro Reina and Renata Szczepaniak Introduction: Syllable and word languages 

 8

Part 1: Theoretical issues Stig Eliasson The typology of syllable and word languages and Swedish phonological structure   43 Matthias Heinz Syllable complexity in the diachrony of Romance languages: A center vs. periphery view and the syllable vs. word rhythm paradigm   87 Barış Kabak Pervasive syllables and phonological unity in words 

 112

Guido Seiler and Kathrin Würth Monosyllabic Lengthening in German and its relation to the syllable vs. word language typology   140 Renata Szczepaniak Vowel and consonant epentheses in the history of German from the typological perspective of syllable and word languages   160

Part 2: Diachronic approaches Kurt Braunmüller Scandinavian word phonology: Evidence for a typological cycle 

 183

VI 

 Table of contents

Martin Joachim Kümmel Syllable- and word-related developments in earlier Indo-Iranian 

 204

Damaris Nübling From Christel to Christina, from Klaus to Nico: A diachronic study of German first names (1945–2010) and their shift towards the syllable language type   222

Part 3: Synchronic approaches (Germanic languages) Pia Bergmann Reduction and deletion of glottal stops and geminates at phonological word boundaries in German compounds: Effects of word frequency and accentuation   251 Peter Gilles Phonological domains in Luxembourgish and their relevance for the phonological system   279 Steffen Höder Low German: A profile of a word language 

 305

Beat Siebenhaar Phonological and phonetic considerations for a classification of Swiss German dialects as a word language or a syllable language   327

Part 4: Synchronic approaches (Romance languages) Javier Caro Reina Central Catalan in the framework of the typology of syllable and word languages   349 Uli Reich and Martin Rohrmeier Batidas latinas: On rhythm and meter in Spanish and Portuguese and other forms of music   391



Table of contents 

Stephan Schmid Syllable typology and the rhythm class hypothesis: Evidence from ItaloRomance dialects   421

 VII

Acknowledgements The present volume goes back to the Phonological Typology of Syllable and Word Languages in Theory and Practice workshop held at the Freiburg Institute for Advanced Studies (FRIAS) on March 29–31, 2010. We would like to express our gratitude to all speakers and participants of the workshop and to the authors for their contributions to this volume. We would also like to thank the Freiburg Institute for Advanced Studies (FRIAS) for providing financial and administrative support for the workshop. We are indebted to the editors-in-chief – Peter Auer, Gesa von Essen, and Werner Frick – for the admission of this volume into the linguae & litterae series and to the anonymous reviewers for their very helpful comments. Javier Caro Reina and Renata Szczepaniak

List of abbreviations and symbols The following list contains abbreviations and symbols used in this volume. The examples were transcribed using the International Phonetic Alphabet (IPA). Glossing follows the Leipzig Glossing Rules, which are available at http://www.eva.mpg.de/lingua/resources/glossing-rules. php. Deviations from the conventions are given in the list. word boundary # syllable boundary $ syllable boundary & compound boundary  empty element ∅ * reconstructed form foot boundary ]F ]ω word boundary + morpheme boundary < derives from > becomes † non-existent AFS adjective-forming suffix AN Ancient Nordic ON Old Norse UTR uter (common gender) OHG Old High German Gmc Germanic Av. Avestan BP Brazilian Portuguese C consonant C= unaspirated consonant Cat. Catalan CMP compound marker CNV converb marker Dan. Danish Engl. English ENHG Early New High German EP European Portuguese F foot / fricative Far. Faroese Fin. Finnish Fr. French Friul. Friulian Fs strong foot Fw weak foot G glide Germ. German Icel. Icelandic

It. Italian L liquid Lat. Latin Mex. Sp. Mexican Spanish MHG Middle High German Mod. Sp. Modern Spanish MP Middle Persian MSL Monosyllabic Lengthening N nasal NFS noun-forming suffix NHG New High German Norw. Norwegian OAv. Old Avestan OHG Old High German OIA Old Indo-Aryan OIr Old Iranian OP Old Persian OSax. Old Saxon OSL Open Syllable Lengthening Pen. Sp. Peninsular Spanish PIE Proto-Indo-European PIIr Proto-Indo-Iranian Port. Portuguese PSL prototypical syllable language PW phonological word PWL prototypical word language pword  phonological word Rom. Romanian S sibilant Scand. Scandinavian SL syllable language Sp. Spanish SUF suffix Swed. Swedish T stop V vowel V̅ long vowel Vː long vowel V̆ unstressed vowel

XII 

 List of abbreviations and symbols

Ved. Vedic VFS verb-forming suffix WGmc West Germanic WL word language x main stress plus tonal accent (accent 2) YAv. Younger Avestan

ZHG μ σ σs σw ω

Zurich German mora syllable strong syllable weak syllable phonological word

Peter Auer (University of Freiburg)

Preface

Structuralism started from the assumption that a “language” is a well-defined entity set off from other such entities by its internal structure. This holds in particular for phonology where, according to structuralist analysis, a small number of elements (the phonemes) combine to make up the phonological system. The glue that keeps the phonological system together are the oppositional relationships between the single phonemes. Of course, it was well known that some phonemes take part in more oppositional relationships based on distinctive features than others (for instance, alveolar stops are usually part of more phonological oppositions than laterals or rhotics), and that these phonemes are therefore in a certain sense more embedded in the system, while others are more peripheral. By and large, however, the phonological structure of a language was taken to be dense and compact, in particular when compared to morphology and syntax. Typology as I understand it is concerned with the question of how the elements of a language cohere in a non-trivial way. It is always more than simple classification, which could be based on any single suitable phonological feature. Finding typologically relevant features implies a quest for those traits of a language that make others predictable, or at least likely to co-occur. As long as structuralism prevailed, the answer to the question of how the phonology of a language coheres was considered trivial and phonological typology did not receive much attention. The search for segment-based, absolute or implicational universals (as already in Trubetzkoy 1939) of the type “if a language has n and only n phonemes of type A then these phonemes will be α and β”, or, even more simple, “a language will have no more/less than n phonemes of type A” by and large replaced typology in the proper sense of the word. These universals may be able to reveal certain markedness relationships, but they tell us little about what the phonology of any one language is, beyond whether it includes more or less marked phonemes. In contrast to structuralism, modern phonological theories assume a much more complex, multi-tiered and hierarchical sound structure; in addition to segments, their oppositional relationships and rules for simple linear concatenations, they also specify structures above the segment, based on hierarchically ordered prosodic domains. These domains are considered to be more than secondary (derived) constructs built on top of segments. They are at least as basic for the make-up of a language as the segments. Since rich prosodic structure has become part and parcel of phonological description, completely new ways of doing phonological typology have become relevant.

2 

 Peter Auer

The papers brought together in this volume by Javier Caro Reina and Renata Szczepaniak make a specific statement about phonological typology based on prosodic domains. They claim that the syllable and the phonological word are the “attractors” for a large group of phonological and phonetic structural regularities and processes in most languages of the world and, more importantly, that it is either the syllable or the phonological word that is preferentially profiled in a language: There are languages that raise the syllable to the status of the central prosodic domain around which the sound features of that language are organized, and there are others that prefer the phonological word. Phonological typologies of this type, i.e. based on the prosodic domains of the syllable and the phonological word, look back on 30 years of research, beginning with Donegan and Stampe’s (1983) seminal paper on rhythmic typology. Whereas earlier studies in the 1980s and 1990s (including e.g. Dauer 1983; Auer 1993) applied the typology to samples of genetically maximally distinct languages, the typology was later (starting perhaps around the millenium) found to be equally applicable in the analysis of phonological developments in one single language that changed its type. Studies by Szczepaniak (2007) on German, Heinz (this volume) and Reich and Rohrmeier (this volume) on the development of Romance languages out of Latin, or Kümmel (this volume) on Old Iranian vs. Indo-Aryan and its diverging developments are just a few examples of this thriving field of research. The typology has also been used in highly fruitful ways for the comparison of closely related varieties (cf. Nübing and Schrambke 2004 on Alemannic; Caro Reina, this volume on Catalan dialects; Höder, this volume on Low German vs. Standard German; Schmid, this volume on Italo-Romance dialects; and Reich and Rohrmeier, this volume on Spanish and Portuguese), and even for the comparison of subsystems of the lexicon of a single language (as in Nübling’s paper in this volume, in which the onomastic lexicon of German is shown to diverge from the rest of the lexicon). It seems that the typology is generally well suited to capture constellations of single changes in a holistic way and that it may indeed make it possible to spell out one of the phonological meanings of drift ‒ the way in which “language moves down time in a current of its own making” (Sapir 1921 [1963: 150]). Thus, what in the beginning was conceived as an attempt to find order in the diversity of linguistic sound systems of the world turned out to be at least as useful ‒ perhaps even more useful ‒ for describing the subtleties of dialect differences, language change, and register differentiation. Another development in research on the typology of syllable vs. word languages over the past 30 years concerns the link between phonology and phonetics. The first rhythm-based typologies were an attempt to rescue the distinction between “stress-timing and syllable-timing” from purely phonetic research that was based exclusively on measurements of isochrony (the duration of syllables or

Preface 

 3

feet). The numerous phonetic studies on isochrony (done mainly until the 1980s) had run into methodological difficulties and, at least prima facie, did not support the distinction as proposed by Kenneth Pike or David Abercrombie (see the discussion in the introduction to this volume). Redefining the distinction in terms of “word languages” and “syllables languages” made it possible to phonologize this tradition of research. The well-known interrelations between isochronous rhythm and phenomena such as the reduction of unstressed syllables or the strengthening of stressed syllables, were used to turn the opposition of stress-timing and syllable-timing upside down: Instead of isochrony being seen as the cause of all these phenomena, they (as well as many others) were now analyzed as ways by which a language can profile the phonological word or the syllable as its basic prosodic domain. Yet all these processes and regularities are of course also reflected in the phonetics of a language. This has led to a certain renaissance of phonetic research on prosodic domains, starting perhaps with Ramus, Nespor, and Mehler’s (1999) influential paper (cf. Siebenhaar, this volume). The phonetic differences between word and syllable languages are not found in isochrony according to this research, but rather in the proportion of vocalic and consonantal material. These phonetic measurements obviously reflect phonological structure (shell complexity in particular) more or less directly. Several papers in this volume bridge the gap between phonology and phonetics, particularly Schmid’s paper, in which the phonetic and phonological classifications of the Italo-Romance varieties are compared. In the following, I would like to discuss some of the problematic (or seemingly problematic) issues regarding the typology. (a) First, there is the question of whether all languages can be classified as either word languages or syllable languages unambiguously. The answer is no. Rather undisputedly, the typology is not binary, but defines a continuum. A language is hardly ever “a word language” or “a syllable language” in its extreme, pure form, but more or less of a word or syllable language. As a consequence, the question “Is language X a syllable or word language?” seldom makes sense. Rather, a language will often show traits of both, but the claim is that one of them will dominate. This dominance can be established language-internally (see e.g. the discussion of Swiss German by Siebenhaar, this volume), or, perhaps most promisingly, by identifying the direction in which the language is moving (“drifting” in Sapir’s sense). (b) The typology crucially depends on a definition of the syllable and the phonological word. As Eliasson and others in this volume argue, the two prosodic categories are not always discussed on the same level: The syllable is a phonetic-phonological category, the prosodic word a phonological-morphological one.

4 

 Peter Auer

Definitions of the syllable hence tend to oscillate between more phonetic and surface-related vs. more phonological, “abstract” notions. Despite these differences, a working consensus on a universal definition of the syllable is usually relatively easy to reach, at least among phonologists. This is not necessarily the case when it comes to a definition of the phonological word. The phonological word needs to be distinguished from the morphological word, but of course the two are related. Consequently, the difficulties of defining morphological words in a cross-linguistically valid way will be reflected in the definition of the phonological word. Even more importantly, the sound features that profile a prosodic word differ across languages, and they seem to do so in a systematic way. When working with Indo-European languages we tend to take stress to be the decisive feature for a phonological word, and indeed the way in which stress is assigned is of central importance for the sound structure of these languages. However, in many languages outside this language family word stress is absent or a superficial, phonetic feature only. Not all languages of this kind are syllable languages. There are other ways of profiling the word than stress, such as harmonic processes that do not cross word boundaries (see Kabak, this volume for Turkish and Korean). As a consequence, it may be useful to think of the word vs. syllable typology less in terms of a single continuum of the type syllable phonological word but rather in terms of a multiplex continuum with more than one “word pole”: word1 (stress-related)

  word2 (harmony-related)

               syllable word3 (......)   wordn (..........) (c) Of course, the typology of word and syllable languages is the more useful the more processes and regularities can be shown to be related to the word or syllable pole. It is therefore of prime importance to isolate the phonological features and processes that are diagnostic of the syllable and word prototype, respectively. Several papers in this volume address this question. For instance, Seiler and Würth (this volume) investigate monosyllabic lengthening as a word-related phonological process that gives more weight to monosyllabic words by adding another moraic unit to it. It seems, however, that there are some phonological

Preface 

 5

processes that cannot be assigned to one of the poles at all, as Szczepaniak (this volume) shows. One of her examples is consonant insertion (epenthesis, epithesis) that occurs in word languages and syllable languages alike, although in different environments. Consonant insertion profiles the syllable when it resolves a hiatus (…V$V… > ...V$CV...), but it profiles the word when it adds complexity to the periphery (e.g. ... VC# > ...VCC#). Obviously, we are dealing with two different processes here which just happen to have the same name. (d) Possible counterevidence to the typology could be seen in the interesting finding of several contributions to this volume (Gilles on Luxemburgish, Caro Reina on Catalan, Siebenhaar on Swiss German dialects, Kabak on Turkish and Korean) that some word-related processes at the same time refer to the prosodic domain of the syllable, i.e. in a certain sense, they are both syllable- and word-related. To take just one example, the Luxembourgish rule of final n-deletion in certain segmental environments is restricted to the syllable coda, as described by Gilles (this volume); however, it only applies when the syllable is word-final. Should we consider it to be a word-related or a syllable-related rule? Here, it is useful to remember that the central question for the typology is whether the process or regularity in question profiles the word or the syllable. In the present case, it is obvious that n-deletion profiles the phonological word, since the rule only applies in word-final position (hence, it helps to delimit the word). Since the rule does not apply word-internally, it provides no cue for identifying syllables or syllable boundaries. We also need to take into account that the distinction between word-related and syllable-related processes within a language may not always be categorical. Languages are more probabilistic and category boundaries more gradient than structuralists may have thought. For instance, Kabak (this volume) shows that in Korean and Turkish certain processes (such as resyllabification) are categorical within the domain of the phonological word, hence profiling it, but also apply across words in an optional way. It is often forgotten that this state of affairs is by no means rare and also holds for languages such as English and (standard) German, otherwise considered word languages. This is shown in Bergmann’s paper on word-related phonetic processes such as degemination and glottal stop insertion before a vowel in German. Degemination across word boundaries is not expected in a language such as German, whereas glottal stop insertion should be obligatory. But in reality, degemination may occur across word boundaries as well, and glottal stop insertion may be lacking, albeit much more rarely than within a word (depending, in that case, on accent position and frequencies of co-occurrence). (e) One of the big unresolved questions, reaching out into the field of sociolinguistics, is why some languages move from the pole of a word language toward

6 

 Peter Auer

that of a syllable language, or the other way round, and why languages sometimes stop this drift and return to a previous stage again. The drift from syllable to word prosodies has often been explained by the phonetics of stress in Indo-European languages. But what about the opposite movement toward the syllable pole? Braunmüller (this volume) suggests three possible explanations: new morphological structures (e.g. affixations that change the phonotactics or bring new stressed syllables into the word), language contact with a syllable language, and language cultivation (spelling pronunciation, for instance). The impact of language cultivation certainly cannot be denied, but it is of course restricted to standard varieties and presupposes a strong social and/or political power that can impose the cultivated norm on the language of the masses of the population. Language contact can account for many cases of substrate influence on an L2 turned into an L1, provided the substrate language was a syllable language. One might speculate that in addition to these clear cases, (more) syllable-based variants of a formerly word-based language will emerge even when none of the L1s in play is a syllable language. The reason is that syllable languages are often phonologically and phonetically more “simple”, i.e. they have few allophonic variants due to segmental or prosodic context or register. They are therefore more transparent and easier to process for learners. A language that has many learners and hence L2 speakers may therefore tend to assume features of a syllable language; a lingua franca would be predicted to be of that type. By extension, the argument also applies to standard languages, which are usually learned as L2 by many speakers in the area “roofed” by it (unless, of course, language cultivation erects a strong boundary against such developments). Pidgins, creoles, and koinai by the same token tend to be more syllable-profiling than traditional dialects. (f) Finally, the question remains whether and how the phonological make-up of a language is related to its syntax and morphology. After the daring statements on this topic made by Donegan and Stampe, few linguists have ventured into this field. In sum, this volume provides ample evidence that a phonological typology based on the relevance of prosodic domains such as the syllable or the phonological word continues to have enormous potential for investigating the make-up of a language, diversity among the world’s languages, but also phonological change and the emergence of new varieties under specific sociolinguistic conditions. Research in this area remains vibrant and challenging at the same time.

Preface 

 7

References Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Dauer, Rebecca (1983): Stress-timing and syllable-timing reanalyzed. Journal of Phonetics 11: 51–62. Donegan, Patricia J. and David Stampe (1983): Rhythm and the holistic organization of language structure. In: John F. Richardson, Mitchell Marks and Amy Chukerman (eds.), Papers from the Parasession on the Interplay of Phonology, Morphology and Syntax, Chicago, 22‒23 April 1983, 337‒353. Chicago: Chicago Linguistic Society. Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 129.) Stuttgart: Steiner. Ramus, Franck, Marina Nespor and Jacques Mehler (1999): Correlates of linguistic rhythm in the speech signal. Cognition 73: 265‒292. Sapir, Edward (1921 [1963]): Language. An Introduction to the Study of Speech. London: Rupert Hart-Davis. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Trubetzkoy, Nikolay S. (1939): Grundzüge der Phonologie. (Travaux du Cercle Linguistique de Prague 7.) Prague.

Javier Caro Reina (University of Freiburg) and Renata Szczepaniak (University od Hamburg)

Introduction: Syllable and word languages The phonological typology of syllable and word languages is a prosodically-based model. It focuses on the question of which prosodic category is central for the phonological make-up of a particular language and how the central prosodic category influences the syllable structure, the distribution of the phonetic and phonemic inventory, and phonetic and phonological processes. The differentiation between two prototypes – the syllable language and the word language – is based on an empirical study carried out by Auer (1993). On the basis of a sample containing 34 geographically and genetically different languages, he observes that most of the languages examined tend to have one central prosodic category, either the syllable or the phonological word. This category is used as the domain for phonetic and phonological processes. The existence of a central prosodic domain manifests itself in a series of cross-linguistic correlations of phonetic and phonological characteristics. For example, syllable complexity correlates positively with word-related phonotactic restrictions and processes, but negatively with tone and syllable-related processes while tone and vowel harmony mostly occur in languages with no or only vague word accent (see Auer 1993: 88). Therefore, phonological properties are not arbitrary, but cluster reflecting the grade of relevance of both (as well as other) prosodic categories in a particular language. The phonological typology of syllable and word languages must be kept apart from the phonetic distinction between syllable-timed and stress-timed languages, also known as the “isochrony hypothesis” – one of the earliest attempts to develop a speech-rhythm based classification. It not only gave rise to a series of experimental studies, but was also the starting point for the rhythm class hypothesis as well as the phonological typology of syllable and word languages. The introduction is structured as follows. Section 1 addresses the research literature on rhythm-based classifications. Section 2 presents the phonological typology of syllable and word languages. Section 3 is dedicated to the typological parameters. Section 4 deals with the research areas where the typology has proven to be fruitful and sketches future lines of research. Finally, section 5 provides a summary of the contributions of the volume.



Introduction: Syllable and word languages 

 9

1 Rhythm-based classifications 1.1 Syllable-timed vs. stress-timed languages (“Isochrony Hypothesis”) The distinction between syllable-timed and stress-timed languages is put forward by Pike (1945: 35‒36), who assumes that the rhythmical organization of languages is based on either syllables or stresses occurring in equal time intervals. He classifies English as a stress-timed language and mentions Spanish as an example of a syllable-timed language. The so-called “isochrony hypothesis”, which predicts that both rhythmic types are mutually exclusive, was first formulated by Abercrombie (1967: 97): “[a]s far as is known, every language in the world is spoken with one kind of rhythm or with the other”. He lists Arabic, English, and Russian as stress-timed languages, and French, Telugu (a Dravidian language), and Yoruba (a Kwa language) as syllable-timed languages. Later, a third type of speech rhythm based on mora-timing (e.g. Japanese) was added (see e.g. Ladefoged 1975; Warner and Arai 2001). In the 1970s and 1980s, experimental studies were carried out in order to classify particular languages, most of them focusing on English and Romance languages such as French, Italian, Portuguese, and Spanish (for an overview of the phonetic research see Auer and Uhmann 1988, Bertinetto 1989, and Kohler 2009, among others). These studies show that neither of the rhythmical units in human languages ‒ syllables and feet ‒ are isochronous. Neither English nor Spanish, the languages explicitly mentioned by Pike (1945), could be confirmed as being stress- and syllable-timed, respectively. Against the expectation that in stress-timed languages the duration of phonetic feet ‒ that is, the duration of the stressed syllable and all subsequent unstressed syllables ‒ should be equal, a considerable variation in foot duration was found to occur in English depending on syllable number and structure as well as the position within the intonational phrase (see Faure, Hirst, and Chafcouloff 1980 and Lehiste 1973, among others). Conversely, syllable duration in Spanish, classified as syllable-timed by Pike, varies depending on accent position as well as the syllable’s internal structure and the position within the foot (see Pointon 1980, also for a review of experimental work carried out on Spanish). Although the dichotomy of syllable- and stress-timed languages has been clearly refuted (see Pamies Bertrán 1999), some of the comparative durational measurements suggest that languages can be located on a continuum between a syllable- and a stress-timed prototype. According to Delattre (1966), English and Spanish differ with respect to the durational variability of syllables. In English, syllable duration varies much more than in Spanish and it depends much more on syllable structure (open vs. closed), accentuation, and position within the

10 

 Javier Caro Reina and Renata Szczepaniak

phonetic foot. Note, however, that the results of the comparative study conducted by Roach (1982) do not even suggest a weak version of the rhythmic classification. Roach (1982) calculated the standard deviation of syllable duration and the mean deviation of foot duration from the predicted value in languages classified by Abercrombie (1967) as stress-timed (Arabic, English, and Russian) and syllable-timed (French, Telugu, and Yoruba). With respect to foot duration, variability was higher in English than in languages classified as syllable-timed. Conversely, standard deviation of syllable duration did not differ significantly among the languages. Recently, however, Gibbon and Gut (2001) replicated the measurements of the standard deviation of syllable duration in English and Ibibio and found a much higher standard deviation in English (93 ms) than in Roach’s study (only 86 ms). The striking inconsistency of the results of instrumental studies can be explained by a discrepancy between the production and perception of speech rhythm, as proposed by Allen (1975) and Lehiste (1977, 1979). Lehiste’s experiments indicate perceptual limits since durational differences smaller than 30‒50 ms cannot be perceived (cf. similar results for German in Hoequist, Kohler, and Schäfer-Vincent 1986). In some cases, for example, when feet are in phrase-initial position, the durational difference must be bigger than 100 ms to be perceived. More importantly, perception experiments conducted by Couper-Kuhlen (1993) show that hearers tend to identify isochronous chains in spoken text. The durational difference between such perceptually isochronous patterns can reach 20% in the case of isometric patterns and even 30% in the case of nonisometric patterns. Auer, Couper-Kuhlen, and Müller (1999: 51‒53) suppose that the reconstruction of meaning demands so much cognitive attention that acoustic variability remains unnoticed. Cauldwell (2002) argues that the opposition between irrhythmical speech production and perceptual rhythmicity is necessary for effective communication. The perception of rhythmical differences is crucial also in language acquisition. In several studies it has been demonstrated that children are sensitive to rhythmical differences between syllable-, stress-, and mora-timing languages (see Cutler et al. 1986; Mehler et al. 1988). Nazzi, Berto­ni­ci­ni, and Meh­ler (1998) provide evidence that French newborns can discriminate between unfamiliar languages only when they belong to different rhythmic classes (e.g. English and Japanese), but not between English and Dutch (both stress-timed).



Introduction: Syllable and word languages 

 11

1.2 Rhythm class hypothesis Since it was not possible to capture rhythmic patterns responsible for the perceived rhythmic differences, the focus of many acoustic studies has been shifted to correlates of speech rhythm. Ramus, Nespor, and Mehler (1999) propose a phonetic classification based on two correlates: The average proportion of vocalic intervals (%V) and the average standard deviation of consonantal intervals (∆C). The first parameter, %V, is directly related to the consonant/vowel ratio and is smaller in languages with full and reduced vowels (e.g. English) than in languages without vowel reduction (e.g. Spanish). The second parameter ∆C, reflecting syllable complexity, is higher in English than in Spanish. Hence, both parameters indirectly support a scalar model of rhythmic distinction between stress-, syllable-, and mora-timed languages (for further discussion see Kohler 2009). Language discrimination tests on adults and newborns indicate that %V could be the rhythmic property retrieved by the perceptual system for recognition of speech rhythm type. Following Ramus, Nespor, and Mehler (1999), Frota and Vigário (2001) have provided evidence for rhythmic differences between European and Brazilian Portuguese, which are regarded in the literature as wordtimed and syllable-timed, respectively. Low, Grabe, and Nolan (2000) and Grabe and Low (2002) introduce the Pairwise Variability Index (PVI) as a better indicator of rhythmicity. The vocalic nPVI (normalized PVI) clearly separates languages into stress-timed (British English, Dutch, German) and syllable-timed (French, Japanese, Spanish). The intervocalic rPVI (raw PVI) does not correlate with the vocalic nPVI, but shows instead peripheral values for languages difficult to classify (Polish and Estonian). Bertinetto and Bertini (2008) propose a modification of PVI: The new algorithm, CCI (the Control/Compensation Index), can control the speech tempo dispersion better than both of the earlier models. Patel and Daniele (2003a, 2003b) used nPVI for a comparison of rhythm in language and classical music (of sixteen composers who lived between 1800 and 1900). Applying nPVI to tone sequences, they found that the rhythmical differences of English and French musical themes reflect the speech rhythm of the composers’ native languages. Acoustic analyses combining the methods based on Ramus, Nespor, and Vogel (1999), Varco C, and the Pairwise Variability Index (PVI) have been carried out for Catalan, English and Spanish (Prieto et al. 2012) and Italian dialects (Schmid, this volume).

12 

 Javier Caro Reina and Renata Szczepaniak

2 The phonological typology of syllable and word languages The typology of syllable and word languages was motivated by the idea that the difference between syllable- and stress-timed languages is related to their phonological make-up (Auer 1993, 1994, 2001). Phonological aspects were already considered in the typological proposal by Donegan and Stampe (1983) and, more consistently, by Dauer (1983, 1987). Both models will be briefly introduced in this section because they depart from the distinction between syllable- and stress-timing and represent important studies that paved the way for the development of the phonological typology of syllable and word languages. We refer to Auer (1993: 10‒24) for a detailed account of Dauer’s and Donegan and Stampe’s models as well as other typological proposals such as Gill’s (1986) distinction between iambic and trochaic languages and Pulgram’s (1970) between word languages, nexus languages, and cursus languages. Dauer (1983, 1987) proposes a typological model depending on stress. As she explicitly points out, “the rhythmic differences we feel to exist between languages such as English and Spanish are more a result of phonological, phonetic, lexical, and syntactic facts about language than any attempt on the part of the speaker to equalize interstress or intersyllable intervals” (Dauer 1983: 55). According to this model, all languages are stress-based, but to varying degrees depending on the interaction of various phonetic-phonological features, which include syllable complexity, syllable duration, tone, distribution of vowels and consonants, and free or fixed word stress (see Dimitrova 1997 for Bulgarian). These features can increase the difference between stressed and unstressed syllables. In Dauer’s model, languages are compared with respect to how the prominence relations between syllables is achieved. A syllable-timing (or syllable-based) counter-pole is not included. On the basis of a synchronic and diachronic analysis of two Austroasiatic language families (the Munda and the Mon-Khmer family), Donegan and Stampe (1983) propose a holistic classification including phonetic and phonological aspects which correlate with morphological, syntactic, and semantic aspects. One of the classificatory parameters is the distinction between syllable rhythm and word rhythm. Here, syllable and word are identified as rhythmic units: Languages can be classified rhythmically according to whether they use, as their unit of isochronous speech, the word (isoaccentual rhythm, in measures of at least two beats, or moras) or the syllable (isosyllabic or isomoric rhythm, depending on whether one-beat vs two-beat syllables are distinguished). In mapping these units onto time, only the material in the “rhyme” of the word or syllable ‒ from the accented syllable to the end of the word, from the nucleus to the end of the syllable ‒ is rhythmically relevant. (Donegan and Stampe 1983: 344)



Introduction: Syllable and word languages 

 13

Donegan and Stampe correlate syllable rhythm with the absence of vowel reduction, liaison, simple syllable structure, and long phonological words. Word rhythm is associated with vowel reduction in unstressed syllables, hiatus at word boundaries, complex consonant clusters, and short phonological words. Thus, the authors bring some important correlations into discussion. The seminal work by Auer (1993), which provides the basis for the phonological typology of syllable and word languages, examines whether the phonological traits – mainly those proposed by Dauer (1987) – correlate in the predicted way. The investigation of a geographically and genetically diversified sample of 34 languages representing the major language families of the world yields evidence for a series of correlations (Auer 1993: 88). The correlations are given in (1) and (2) according to their predictive character: (1) Strong correlations: a) a complex syllable structure positively correlates with word-related phonotactics and processes (e.g. stress-dependent vowel reduction or allophonic alternation restricted to a certain position within the word); b) a complex syllable structure negatively correlates with tone and (tentatively) with vowel harmony as well as with syllable-related phonotactics and processes (e.g. vowel epenthesis); c) stress-dependent vowel reduction positively correlates with word-related phonotactics and processes; d) stress-dependent vowel reduction negatively correlates with tonal system or vowel harmony; e) central or back unrounded vowels are rarely found in languages with a simple syllable structure and in languages with no or vaguely defined word stress; f) languages without word stress do not exhibit syllable shells exceeding CC... CC; g) vowel harmony and tone mainly occur in languages with no or vague word stress; h) word-related phonotactics or processes do not occur in languages with non-restricted tone; i) phonemic geminates are rare in languages with stress-dependent vowel reduction. (2) Weak correlations: j) syllable structure optimizing processes occur in languages with simple syllable structures; k) syllable structure deteriorating processes occur in languages with complex syllable structures;

14 

 Javier Caro Reina and Renata Szczepaniak

l) stress-dependent vowel reduction does not occur in languages with low syllable complexity and weak stress rules.

Some of these correlations are confirmed in the analysis of Central Catalan (Caro Reina, this volume). For example, Central Catalan has a complex syllable structure and word-related processes such as vowel centralization and word-final obstruent devoicing, which is consistent with correlations (a) and (c). Vowel harmony is not attested in Central Catalan, which is in accordance with correlation (b). Interestingly, vowel harmony is attested in Valencian, where the scope of syllable structure complexity is even wider than in Central Catalan. There is one prediction, however, that is not borne out. Notwithstanding the presence of stress-dependent vowel reduction, Central Catalan has phonemic geminates (e.g. ball-ar [bəˈʎa] ‘dance-inf’ vs. vetll-ar [bəʎˈʎa] ‘be awake-inf’), thus contradicting correlation (i) (see Wheeler 2005: 36–37 for the phonological status of geminates in Catalan). In the same vein, the survey conducted by Schmid (this volume) on ten Italo-Romance dialects provides evidence supporting correlation (a). For example, there is a positive correlation between a complex syllable structure and vowel centralization, as attested in Romagnolo and Turinese. The opposite, however, does not hold since vowel centralization is also attested in dialects with simple syllable structures such as Feltrino and Milanese. Additionally, Seiler and Würth (this volume) critically discuss the correlations between the word-related processes of monosyllabic lengthening, open syllable lengthening, and degemination in German dialects. While they find no correlation between monosyllabic lengthening and open syllable lengthening or degemination, they observe a correlation between open syllable lengthening and degemination. This is the first attempt to establish correlations between word-related processes. Certainly, more research is needed to systematically examine whether and to what extent there is a correlation not only between the typological parameters (e.g. syllable structure, phonological processes, etc.), but also between the patterns of each single parameter (e.g. word-related processes such as vowel centralization and vowel deletion). In contrast to Dauer’s (1983, 1987) assumption, Auer (1993: 89, 2001: 1398) emphasizes that stress does not play a central role in typological classification. Some languages without word stress and, hence, without distinction between stressed and unstressed syllables, show moderately complex syllables. On the other hand, strong (phonetically well-exposed) word stress is not restricted to languages with complex syllables. Additionally, the languages in the sample display phonotactic restrictions and processes which may refer to the syllable, to the phonological word, or to both domains alike. Based on this observation, Auer (1993: 90) proposes a prosodically-based typological model in which languages are classified according



Introduction: Syllable and word languages 

 15

to the prosodic category dominantly referred to by the phonotactic restrictions and processes. Obviously, the syllable does not serve as the universal central prosodic category (see Auer 1994 for discussion). The languages investigated in Auer (1993) can be distributed on a scale between the syllable language and the word language prototype. Figure 1 illustrates the typological affiliation of some of the languages included in the sample. Some of them (e.g. Vietnamese) can be unambiguously classified as syllable languages while others (e.g. English) can be unambiguously classified as word languages. However, some languages such as Turkish represent a mixed type, where neither the syllable nor the phonological word can be viewed as dominating (see Kabak, this volume). This intermediate stage can be explained as resulting from the phonological change from one language type into another (see Szczepaniak 2007; Gilles, this volume). The typological parameters which can be used for the definition of both prototypes will be discussed in more detail in the following section. syllable language Navaho Vietnamese Yoruba

word language Hausa Japanese Quechua

French Khalka Turkish

!Xóõ Nama Uzbek

Diegueño English Klamath

Figure 1: Typological affiliation of selected languages (Auer 1993: 94)

For the typology, only two prosodic categories are relevant: the syllable and the phonological word. The syllable is one of the lower-level prosodic constituents (Nespor and Vogel 1986: 61‒62). The optimal syllable structure can be defined in terms of phonological principles, formulated as preference laws by Vennemann (1986, 1988). Therefore, the syllable language prototype can be easily described as consisting of only optimal syllables (CVCV...). The prototype of a word language is much more difficult to grasp since the category of the phonological word can only be defined for a specific language. Nespor and Vogel point out that the phonological word is modelled on morphological information: “The phonological word is the lowest constituent of the prosodic hierarchy which is constructed on the basis of mapping rules that make substantial use of nonphonological notions. In particular, the phonological word (ω) represents the interaction between the phonological and the morphological components of the grammar” (Nespor and Vogel 1986: 109). Hence, the reason for postulating the word as a phonological domain is the very frequent lack of isomorphism between the possible size of the phonological word and that of the morpho-syntactic word. A phonological word can be bigger than, the same size as, or even smaller than one morpho-syntactic word. The size of the phonological word varies language-specifically (for examples see Booij 1983; Nespor and Vogel

16 

 Javier Caro Reina and Renata Szczepaniak

1996; Hall and Kleinhenz 1999; Dixon and Aikhenvald 2002; Schiering, Bickel, and Hildebrandt 2010, among others). Additionally, diagnostic criteria for phonological wordhood vary (Hyman 2008; Revithiadou 2011). In German, there is suprasegmental evidence for word-internal phonological words ‒ that is, phonological words are smaller than grammatical (morpho-syntactic) words. Raffelsiefen (2000) points out that German compounds behave differently from simplexes, even though the latter are polysyllabic or polypedal phonological words. First, within a polysyllabic and polypedal phonological word there is only one tense vowel that can be long as in Schokolade [[ˌʃoko]Fw[ˈlaːdə]Fs]ω ‘chocolate’ vs. Steinobst [[ʃtaɪ̯n]F]ω[[ʔoːpst]F]ω ‘stone fruit’. Second, within a polysyllabic and polypedal phonological word all intervocalic consonants are syllabified in the syllable-onset position as in Zauber+er [ˈt͡saʊ̯. bə.ʀɐ]ω ‘wizard’. The resyllabification does not take place at phonological word boundaries, hence [[ʃtaɪ̯n]F]ω[[ʔoːpst]F]ω ‘stone fruit’ and not *Stei.nobst. And finally, whereas in a polysyllabic and polypedal phonological word the last branching foot is the most prominent (e.g. Schòkoláde [[ˌʃoko]Fw[ˈlaːdə]Fs]ω ‘chocolate’), in compounds we normally find the opposite situation (e.g. Stéinòbst [[ʃtaɪ̯n]F]ω[[ʔoːpst]F]ω ‘stone fruit’). Irrespective of the number of syllables, the first foot dominated by a separate phonological word carries the main stress. Interestingly, empirical studies show that phonological word boundaries can be blurred in frequent compounds (see Bergmann, this volume for German). Since the size and internal structure of phonological words are defined language-specifically (Auer 1994: 61), the word language prototype must comprise different subtypes (see Eliasson, this volume). However, the tendency to expose the structure and stabilize the size of the (language-specifically defined) phonological word can be stated as being constitutive for the word language prototype.

3 Typological parameters In order to evaluate the typological affinity of a given language, a series of parameters have been proposed (Auer and Uhmann 1988: 253; Nübling and Schrambke 2004: 284–285; Szczepaniak 2007: 330). The number of parameters found in the literature may differ considerably, but they are all based on Auer (1993, 2001: 1395–1398). Hence, they represent only a selection of possible phonetic and phonological phenomena. In Table 1, we present three general parameters: the syllable structure, the distribution of the vowel and consonant inventory, and phonetic and phonological processes. Singular phenomena such as phonemic geminates or assimilation processes (sandhi) can be then interpreted as belonging to the inventory parameter or to the phonological processes, respectively.



Introduction: Syllable and word languages 

 17

Thus, features such as phonemic geminates should be examined with regard to their symmetrical (syllable-related) or asymmetrical (word-related) distribution. Only a thoroughly conducted analysis allows for a robust typological conclusion. Table 1: Typological parameters Parameters

Syllable language

Word language

Syllable structure

simple

high variable

Vowel and consonant inventory

symmetrical

asymmetrical

Phonetic and phonological processes

syllable-related and syllable-optimizing

word-related and word-optimizing

With regard to the syllable structure, syllable languages are characterized as having a rather simple syllable structure while word languages are characterized as having a high variable syllable structure. Syllable languages display an unmarked syllable type comprising one minimally sonorant consonant (C) and one maximally sonorant vowel (V). The optimal syllable structure CV is shaped by Vennemann’s (1988) preference laws such as Head Law, Coda Law, etc. Deviations from the optimal syllable structure CV imply a violation of the preference laws and, hence, a deterioration of the optimal syllable. Syllable complexity gradually increases with each consonant segment added to the syllable shell (see Maddieson 2011 for a classification of syllable complexity). The maximal grade of syllable deterioration is reached when the Sonority Sequence Principle is violated. This is the case with extrasyllabic consonants. For example, in banks the [s] constitutes an extrasyllabic consonant since it is higher in sonority than the preceding consonant segment. Complex syllables are typical of word languages, where they occur at word and morpheme boundaries, thereby highlighting the morphological constituents. Examples of word-initial and word-final complex syllables are German Sprache [ˈʃpʀaːxə] ‘language’ and Herbst-s [hɛɐ̯pst͡s] ‘autumngen.sg’, respectively. In word languages complex syllable structures are expected to occur in stressed syllables and at word and morpheme boundaries. Additionally, speech tempo may cause the increase of syllable complexity (for examples in Romance languages see Heinz, this volume). While complex syllable structures are indicative of word languages, the opposite does not necessarily hold. That is, a simple syllable structure may also be found in a word language. This is the case with languages such as !Xóõ and Nama, which have a simple syllable structure. However, they demarcate word-initial boundaries by means of phonotactic restrictions. As a consequence, a consonant set may appear exclusively in the

18 

 Javier Caro Reina and Renata Szczepaniak

word-initial onset (Auer 1993: 84). In this respect, the absence of a complex syllable structure obeys strong phonotactic restrictions. In view of the different patterns described for the syllable and word language types, the notion of “marked” and “unmarked” syllable structures comes into a new light. A complex syllable structure is marked in a syllable language. In contrast, a complex syllable structure is unmarked in a word language. With regard to the distribution of the vowel and consonant inventory, we can differentiate between a symmetrical system which is consistent with a syllable language and an asymmetrical system which is consistent with a word language. In syllable languages, the occurrence of vowels and consonants is not dependent on word-related factors such as stress or within-word position. For example, in Spanish the same consonants may appear both in word-initial and word-medial onsets. Similarly, the same vowels may appear in both stressed and unstressed position (Szczepaniak 2009). In syllable languages, phonotactic restrictions are conditioned by the syllable structure, as in Chinese where obstruents only occur in the syllable onset. In contrast, word languages exhibit asymmetrical distributions which help to highlight the phonological word. In this respect, Trubetzkoy (1971: 242) speaks of Grenzsignale (‘boundary signals’). Asymmetrical distributions usually arise from phonological processes. Processes leading to vowel asymmetries include vowel lengthening and diphthongization in stressed syllables and vowel reduction in unstressed syllables. Vowel reduction processes in unstressed syllables imply the neutralization of contrasts of vowel height, quantity, and nasalization (Barnes 2006: 20) such that phonological oppositions (e.g. quantity oppositions) are licensed only in stressed syllables. For example, in German the occurrence of schwa is limited to unstressed syllables as in Blume [ˈbluːmə] ‘flower’ or flieg-en [ˈfliːgən] ‘fly-inf’. On the other hand, processes leading to consonant asymmetries involve aspiration and glottalization of voiceless stops in word-initial position as well as degemination, lenition, and consonant deletion in word-medial position. Interestingly, strong asymmetrical distributions regarding the consonant system do not seem to be common among Indo-European word languages. Additional research would be needed to answer the question whether asymmetrical distributions regarding the vowel system are more widespread in word languages belonging to the Indo-European family (e.g. German) than in word languages belonging to other language families (e.g. !Xóõ). As for the phonetic and phonological processes, a high incidence of processes referring to a specific prosodic domain is diagnostic of its central importance. In syllable languages, processes operate at the level of the syllable regardless of stress (if there is any) and position. For example, vowel nasalization can be classified as a syllable-related process when it applies in both stressed and unstressed syllables as in Portuguese. Additionally, syllable-related processes may optimize



Introduction: Syllable and word languages 

 19

the syllable structure. Instances of syllable repairing strategies include vowel epenthesis in complex consonant clusters and vowel deletion in hiatuses. Syllable-related processes may also deteriorate the syllable structure. In other words, not all syllable-related processes are necessarily syllable-optimizing. This is the case with lenitions that increase the sonority in every syllable onset thereby violating the Onset Law and even the Contact Law (see Szczepaniak 2007: 125‒133 for the High German consonant shift and Caro Reina, this volume for deaffrication in Central Catalan). In word languages, processes operate at the level of the word and are dependent on stress and within-word position. For example, vowel denasalization can be classified as a word-related process when it applies in unstressed position only as in Algarve Portuguese (Sampson 1999: 202), thereby contributing to an asymmetry of the vowel system. It should be noted that there are processes which cannot be exclusively classified as syllable-optimizing or word-optimizing. For example, consonant epenthesis is a syllable-optimizing process when it repairs ill-formed syllable structures but word-optimizing when it highlights the word and morpheme edges (see Szczepaniak, this volume for details). The evidence obtained from applying these parameters to the phonology of a certain language helps to determine the relevance of the prosodic categories of the syllable and the word. We would like to emphasize that the typological affiliation of a language cannot be made exclusively on the basis of one single parameter. For example, a simple syllable structure alone is not compelling evidence for a syllable language. Once strong phonotactic restrictions as the ones found in !Xóõ can be excluded, we can classify a language as a syllable language. Depending on the strategies favored, we can distinguish different types of word languages. In that respect, German and !Xóõ represent two types of word languages. While German exhibits a complex syllable structure, vowel asymmetry and a slight consonant asymmetry, !Xóõ displays a simple syllable structure, vowel asymmetry and a strong consonant asymmetry, as illustrated in Table 2. Table 2: Typological patterns in German and !Xóõ Language

Complex syllable structure

Vowel asymmetry

Consonant asymmetry

German

+

+

+

!Xóõ

–

+

++

In previous proposals of typological parameters (Dauer 1983, 1987), we find the parameter “stress”, which is assumed to be phonetically weak in syllable languages and phonetically strong in word languages. However, as mentioned

20 

 Javier Caro Reina and Renata Szczepaniak

above, stress does not play an essential role in the typology. With regard to stress, it is worth noting that fixed stress, especially word-initial stress as in Finnish or word-final as in Turkish, may function as a word boundary signal. Interestingly, fixed stress is accompanied by vowel harmony in these languages. Whether fixed stress is found to occur more frequently with vowel harmony than with vowel centralization, both of which are word-related processes that strengthen the integrity of the word, requires additional cross-linguistic study.

4 Research areas The typology of syllable and word languages has been hitherto applied in cross-linguistic studies and historical linguistics. With regard to cross-linguistic studies, Nübling and Schrambke (2004) provide a typological overview of Germanic languages (see Figure 2 for the typological affiliation of the Germanic languages analyzed); Schrambke (2007) shows parallel developments in Northern Alemannic and Danish; Szczepaniak (2009, 2010) delivers a contrastive study on German and Spanish and German and Luxembourgish, respectively. Gilles (this volume) and Höder (this volume) contribute to the typological overview given by Nübling and Schrambke (2004). syllable language Afrikaans Norwegian Swedish

word language Luxembourgish

Danish English German

Figure 2: Typological affiliation of selected Germanic languages (Nübling and Schrambke 2004: 286)

With regard to historical linguistics, Szczepaniak (2007) has demonstrated the typological drift in the history of German. While Old High German ‒ the oldest stage of German (700‒1050) ‒ can be unambiguously classified as a syllable language, the subsequent diachronic development shows a gradual optimization of the phonological word. Thus, the phonological evolution of German can be best described in terms of a change of the central prosodic category. Diachronic research hence provides substantial evidence for a typological shift (for the typological drift from Old into Middle Iranian see Kümmel, this volume). In addition, diachronic analysis may also be useful in cases where a language has no clear typological affinity. This is the case with Luxembourgish. The knowledge of previous historical stages can help us to better understand the position of Luxembourgish on the scale of syllable and word languages (see Gilles, this volume).



Introduction: Syllable and word languages 

 21

In what follows, we would like to sketch some lines of research which future investigation may take. Auer (2010) points out that in language change the transition from a syllable language to a word language seems to be unmarked. Two external factors may impede the transition: language standardization and language contact. With regard to language standardization, the typological development of a language may be influenced ‒ and even reversed ‒ by language planning and language standardization. For example, the phonological make-up of standard Italian deviates considerably from Italian dialects (see Schmid, this volume). Similarly, Braunmüller (this volume) shows the influence that language cultivation had on the spelling and pronunciation of Swedish. With regard to language contact, Auer (2010) suggests that varieties emerging in language contact situations favor syllable-related structures (e.g. simple syllable structures, no stress-dependent vowel reduction, etc.). This may be the case with pidgins, creoles, and koinai (see Plag and Schramm 2006 and Klein 2011 for the syllable structure in creole languages). For example, Nübling and Schrambke (2004: 286) explain the syllable-related traits of Afrikaans, which sharply contrast with the word-related traits of Dutch, as a result of language contact. Similarly, the language contact situations briefly described for Barcelona Catalan in Caro Reina (this volume) seem to support the hypothesis that language contact favors syllable-related processes. Forthcoming studies on how language contact may influence the typological drift should also consider the typological make-up of pre-contact varieties. In this respect, a contact situation between a word-oriented language such as Central Catalan and a syllable language such as European Spanish has resulted in the loss of centralized vowels in unstressed position, which can be viewed as an optimization of the syllable (see Szczepaniak 2009 for the typological classification of European Spanish). A preference for syllable-related features has also been reported for L2 varieties of English (Wells 1982: 643‒644, 646‒647, 651), and the multicultural varieties of London (Torgersen and Szakay 2012), Copenhagen (Quist 2008), and Hamburg (Auer and Dirim 2004: 207; cf. Kern 2011: 212‒214 for prosodic differences between Turkish German and German). In this respect, L2 varieties of English and the multicultural variety of London differ typologically from standard British English. However, long-lasting language contact can lead, again, to a transition towards a word language. For example, Auer (1993: 29, 31‒32) points out that language contact may be responsible for the typological differences between Turkish and Uzbek on the one hand and Classical Mongolian and the modern Mongolian dialect Khalkha on the other. In both cases the extensive language contact of Uzbek with Arabic, Farsi, and Russian, and of Khalka with Chinese and Russian

22 

 Javier Caro Reina and Renata Szczepaniak

has resulted in a development towards the word language pole (see Figure 1 for the typological affiliation of Kahlka, Turkish, and Uzbek). Another promising area of study is language variation in standard and non-standard varieties (Caro Reina 2013). While cross-linguistic studies have mainly concentrated on standard languages, the typology could be a useful tool to throw light on the nature of language variation within a dialect group (cf. Schmid 1997 for an account of the syllable structure of Italian dialects, Schrambke 2003 and Nübling and Schrambke 2004 for Alemannic, Noske 2007 for Northern and Southern Dutch, and Caro Reina, this volume for Catalan). Another line of future research might evaluate the interaction between phonology and morpho-syntax. While the syllable is purely phonological in nature, the phonological word incorporates morpho-syntactic information (Nespor and Vogel 1986: 109). The question that arises is how morpho-syntactic information is highlighted in a word language. The development of the highly productive linking element -s- in German compounds is an example of how morpho-syntactic information is mapped phonologically. In German, compounds consist of at least two phonological words (Wiese 1996: 72‒73, 296–302; Raffelsiefen 2000). In compounds such as Geburtstag ‘birthday’, which contains the phonological words Geburt ‘birth’ and Tag ‘day’ ([Geburts]ω[tag]ω), the right edge of the first constituent is demarcated by an ill-formed syllable structure resulting from the linking element ‑s‑. This ill-formed syllable structure enables the listener to recognize morphological boundaries and, thus, to decode morphological information. Linking elements arose in Early New High German when lexicalized nominal phrases such as Teufels Sohn ‘devil’s son’ were reanalyzed as compounds ([[desART.GEN TeufelsN]NP Sohn]NP > [derART.NOM [Teufelssohn]N]NP). Thus, the former inflectional endings were functionalized as a prosodic means to demarcate the phonological word (Nübling and Szczepaniak 2008, 2009). This process is in line with the typological drift experienced by German (Szczepaniak 2012). Furthermore, the implications of the typological affinity for morphology should be investigated (see Plank 1998). Auer (2001: 1395) suggests that mapping rules in syllable and word languages correspond to the patterns of agglutinative languages and inflected languages, respectively. While affixes constitute syllables in agglutinative languages, they may undergo reduction and contribute to syllable complexity in inflected languages, cf. the plural morph in Turkish (e.g. ev-ler ‘house-pl’) and Catalan (e.g. parc-s ‘park-pl’). Moreover, in word languages morphological information may be coded in the stressed syllable. This is the case with umlaut in German (e.g. Ofen [ˈoːfən] ‘oven[sg]’ vs. Öfen [ˈøːfən] ‘oven.pl’).



Introduction: Syllable and word languages 

 23

5 Summaries of the contributions in this volume This volume is divided into four parts, which focus on theoretical issues (part 1), diachronic approaches (part 2), and synchronic approaches in Germanic and Romance languages (part 3 and 4, respectively).

Part 1: Theoretical issues In “The typology of syllable and word languages and Swedish phonological structure” Stig Eliasson takes a critical stance towards the notion of the phonological word. He points to the problem of its cross-linguistic comparability as well as to the lack of agreement among different types of diagnostic criteria for phonological wordhood. He also recalls that the typology of syllable and word languages should take into account the fact that the exact definition of the phonological word can only be given in a specific language. As a consequence, an asymmetrical typological model with multiple continua between one prototypical syllable language and many different types of word languages should be considered. In his analysis of selected phonological features of Central Standard Swedish, Eliasson unveils the great importance of the morpheme as a domain of phonological rules and processes. This observation challenges the primary motivation for establishing a prosodic hierarchy, which is the non-isomorphism between prosodic and morphosyntactic categories. In contrast to the general assumption of prosodic phonology that isomorphism can occur only sporadically, Eliasson observes a regular reference of phonological means to morphological categories in Swedish. First of all, tonal accent (grave accent) often signals polymorphemicity since it occurs predominantly in morphologically complex words. In this respect, the majority of compounds receive tonal accent, which indicates the connectedness of their constituents (e.g. fisk+död [xfis·kˌdøːd] ‘fish death’). Furthermore, both consonant phonotactics and syllable structure frequently point to morpheme boundaries. In Swedish, not the word boundaries per se, but rather the (word-internal and word-external) right edges of root morphemes may be marked by most complex postvocalic clusters such as ‑str‑ or ‑kst‑. Additionally, in many cases vowel quantity provides a clue to the word-internal morpheme boundaries. Here, the vowel length depends on the syllable structure within the root morpheme due to syllable harmony. Thus, long vowels occur when the second mora is not covered by a second coda consonant belonging to the same morpheme (e.g. skjut+s [ɧʉːts] ‘shoot-pass’). As to the typological classification of Central Standard Swedish, Eliasson observes that numerous important properties place this language closer to the word language than the syllable language pole. He emphasizes the orga-

24 

 Javier Caro Reina and Renata Szczepaniak

nizing role of stress and the varying syllable-shell complexity as well as the different vowel inventories in stressed and unstressed position. However, a leaning towards “natural” resyllabification in fast casual speech as well as the uniformity of word-internal and word-external sandhi, especially supradentalization, work against this tendency. Matthias Heinz’s paper “Syllable complexity in the diachrony of Romance languages: A center vs. periphery view and the syllable vs. word rhythm paradigm” focuses on syllable complexity in Romance languages. Syllable complexity is a central parameter of the typology of syllable and word languages. The author approaches syllable complexity from two different perspectives. First, syllable complexity may be related to speech rate. Thus, there are Romance languages such as European Portuguese and French where simple syllables are fully articulated in the lento style while complex syllables are favored in the corresponding allegro style. Complex syllables conditioned by speech tempo result from the deletion of unstressed vowels. For example, French qu’est-ce que tu fais? ‘what are you doing?’ is realized as [kɛskətyˈfɛ] in the lento style and as [kstyˈfɛ] in the allegro style. Second, syllable complexity may be related to the centrality or periphery of the lexicon. Complex syllable structures have been simplified in Spanish and Brazilian Portuguese. However, learned words have conserved the original Latin complex structure as in Spanish abstracción ‘abstraction’. Additionally, the complex syllable structures of learned words may be simplified in allegro style as in Spanish exacto ‘exact’, which is realized as [ekˈsakto] in careful speech and as [eˈsato] in faster speech. Complex syllables are peripheral in these languages since their occurrence is restricted to a well-defined lexical set. In contrast, as a consequence of the vowel deletion processes in the allegro forms, complex syllables have adopted a central character in languages such as European Portuguese and French. On the basis of the categories “style” (allegro vs. lento) and “lexical relevance” (central vs. peripheral), the author examines the patterns of syllable complexity in Brazilian and European Portuguese, French, and Spanish. The aim of Barış Kabak’s paper “Pervasive syllables and phonological unity in words” is two-fold. First, it discusses the implications of the typology of syllable and word languages in light of synchronic, diachronic and psycholinguistic data. Second, it addresses the question of whether and to what extent syllable-based generalizations can co-exist with regularities that refer to the domain of the word. For that purpose, Kabak focuses on Korean and Turkish, which have been described in the literature as syllable-based. The primacy of the syllable as a phonological domain in these languages can be inferred from generalizations that condition morpho-phonological representations and language change. For example, in Korean and Turkish we can observe hiatus avoidance. A hiatus involves a marked syllable contact which is repaired by means of vowel coales-



Introduction: Syllable and word languages 

 25

cence as in Korean (e.g. /ai/ → [ɛː] ‘child’), or by means of suffix allomorphy as in Turkish (e.g. duvar-ı ‘wall-acc’ vs. kapı-yı ‘door-acc’), although in Turkish heterosyllabic vowel sequences are allowed in loan words and in words that have undergone word-medial consonant deletion. Further evidence for the syllable as a main phonological category is provided by experiments involving the perception of speech strings. However, there are also phonological regularities that clearly point to the existence of a higher phonological domain: the phonological word. For example, in Turkish the relevance of the prosodic word is evident from vowel harmony operating at the level of the word, word-final stress, and syllable weight restrictions in monosyllabic words. The author argues that the affinity of a language with a syllable or a word language should be linked to the way mental representations are shaped by syllable-based or word-based representations, respectively. He shows that the phonological word, which is the central prosodic category in word languages, may also be a relevant prosodic domain in Korean and Turkish. The existence of this type of mixed languages is not necessarily indicative of a typological drift. It rather throws light on the role that different prosodic domains may play in the organization of speech units. Thus, vowel harmony and final stress guarantee not only cohesion within the phonological word but also facilitate speech segmentation owing to their demarcative function, as psycholinguistic experiments conducted on French and Turkish have revealed. Guido Seiler and Kathrin Würth’s paper, “Monosyllabic Lengthening in German and its relation to the syllable vs. word language typology” introduces a valuable parameter for the typology: Monosyllabic Lengthening as a quantity-related process that regulates the weight of the word at the moraic level. While the role of the phonological foot as a domain regulating word size has been described by Szczepaniak (2007) for Standard German, the role of the mora has not been systematically explored in the typology so far. The authors describe Monosyllabic Lengthening in German dialects and discuss the word-related nature of this process. On the basis of Zurich German, a High Alemannic dialect, they demonstrate that vowel lengthening is – among other things – a strategy to fulfill minimality requirements. Content words, monosyllabic or disyllabic, must be at least bimoraic; that is, short vowels are not allowed in monosyllabic words with the syllable types CV and CVC. This is achieved by vowel lengthening in Zurich German, which may bring about a paradigmatic quantity alternation. For example, the monosyllabic word Has [hɑːz̥] ‘hare’ has undergone lengthening, thereby becoming bimoraic. On the other hand, the rest of the paradigm contains a short vowel as in the bisyllabic words Has-e [ˈhɑz̥ə] ‘hare-pl’ and Häs-li [ˈhæz̥li] ‘hare-dim’, which are bimoraic too. The patterns described for Zurich German challenge the traditional view that explains Monosyllabic Lengthening (in Standard German)

26 

 Javier Caro Reina and Renata Szczepaniak

as a result of paradigm levelling after open syllable lengthening. Since there is no open syllable lengthening in this dialect, Monosyllabic Lengthening proves to be an independent process. This process, however, is not exclusive to Zurich German. It is also attested in several German dialects such as Bavarian, Silesian, Thuringian, and Upper Saxon. For their survey on Monosyllabic Lengthening in High German dialects, Low German and Standard German, Seiler and Würth distinguish the following four types: 1) vowel lengthening before single nonmoraic consonants (Leichtschlussdehnung I); 2) final consonant gemination (Leichtschlussdehnung II); 3) vowel lengthening before single consonants and (former) geminates (Schwerschlussdehung I); and 4) vowel lengthening before single consonants, (former) geminates, and cluster stems (Schwerschlussdehung II). In a further step, the authors examine possible correlations between monosyllabic lengthening and other word-related, quantity-affecting processes such as open syllable lengthening and degemination. It turns out that Monosyllabic Lengthening combines freely with the other parameters. Having thus shown that Monosyllabic Lengthening is a very weak predictor for other word-related parameters the authors conclude their paper with open questions concerning the typology of word vs. syllable languages in general: Whereas there is no doubt that the typology is a useful heuristic tool to locate a language within a continuum from syllable orientation to word orientation, future research will help to reveal implicational and correlative relationships between the individual parameters that will allow us to assess the predictive power of the typology. Renata Szczepaniak analyzes “Vowel and consonant epentheses in the history of German from the typological perspective of syllable and word languages”. Generally, epentheses are very frequent in the history of German, especially in Old High German (OHG; 700‒1050) and in Early New High German (ENHG; 1350‒1650). During the language’s history, the character of epentheses has changed: In Old High German, the inserted vowels were cluster breakers. They optimized the syllable structure by resolving complex consonant clusters as in OHG wurm > wurum ‘worm’ (the epenthetic element is underlined). At the same time, they destabilized the phonological word by alternating the number of syllables (wurm/wurum) as well as the word edges (wurm/wurum). In contrast, the Early New High German epenthetic schwa was a trochee maker. It appeared directly after the stressed vowel (long vowel or diphthong), creating trochaic phonological words with a high internal sonority degree as in MHG mûre > (E)NHG Mauer ‘wall’. At the same time, they blurred the word-internal syllable structure, rendering the offset of the diphthong ambisyllabic (e.g. [ʊ̯] in [ˈmaʊ̯ɐ]). Similarly, the “weak” consonants (i.e. glides [j, w], r, and h) served as hiatus breakers in Old High German (e.g. OHG bū-an > būw-an ‘live-inf’). Although the consonant epenthesis applied only within the phonological word, it did not mark its inter-



Introduction: Syllable and word languages 

 27

nal structure since all weak consonants also occurred in word-initial onset. In Early New High German, epenthetic consonants served as word edge strengtheners. The epenthesis operated post-consonantally, extending word-final consonant clusters (e.g. MHG mâne > (E)NHG Mond ‘moon’). They emphasized the right edge of the phonological word making it more “visible”. While the Old High German vowel and consonant epentheses optimized the syllable structure, the Early New High German epentheses highlighted the phonological word. The shift of the referred domain which comprises the syllable in Old High German and the phonological word in Early New High German is part of the phonological change of German from a syllable language to a word language type, which has brought about a series of word-optimizing processes: While the Early New High German consonant epenthesis emphasized the right edge of the phonological word by adding a strong consonant and building up word-final consonant clusters, glottal stop and [h] mark the left edge of New High German words.

Part 2: Diachronic approaches In “Scandinavian word phonology: Evidence for a typological cycle”, Kurt Braunmüller observes that some of the Scandinavian languages deviate from the general tendency in the diachrony of the Germanic language family, which goes from predominantly syllable languages in Pre-/Proto-Germanic and ancient Germanic dialects towards word languages at later stages. He assumes that the counter-drift towards a syllable language, which can be traced in the history of Swedish and Norwegian, is basically a consequence of three language-external factors: early language contact, language cultivation, and the language-internal development of the enclitic processes in the nominal and verbal grammar. The general tendency towards the word language type most extremely manifests itself in the history of Danish, where the reduction of non-derived words to monosyllabics is most advanced. Conversely, Swedish is the best counter-example: The drift towards a word language was stopped and even reversed, probably due to early language contact with the genetically and typologically differing Sámi and Fennic languages. Vowel balance and vowel harmony, which are characteristic of those contact languages, can be observed in northern Norwegian and Swedish dialects. Vowel balance applies to inflectional endings so that their vowel quality depends on the accented vowel. Vowel harmony takes place in stressed light syllables and the following unstressed ones. By the end of the eighteenth century, language cultivation had strongly influenced the written forms as well as the pronunciation of modern Swedish. One of the main principles was euphony: The ongoing tendencies in language change were evaluated, and the forms that con-

28 

 Javier Caro Reina and Renata Szczepaniak

tained full vowels were favored in written texts. Finally, the new clitic markers for definiteness and for medio-passive have changed the shape of words by adding unstressed syllables with full vowels. Martin Joachim Kümmel’s paper, “Syllable- and word-related developments in earlier Indo-Iranian” offers a comparative account of the phonology of the two branches of Indo-Iranian: Old Indo-Aryan (Vedic Sanskrit) and Old Iranian (Avestan, Old Persian). While Old Indo-Aryan and Old Iranian share a great amount of morphological and syntactic similarities, they notably differ with regard to their dominant prosodic category. These differences, as Kümmel reveals, can be adequately explained in terms of syllable- and word-related features. Thus, Old Indo-Aryan is rather syllable-oriented. Moreover, syllable-enhancing processes increase from Old to Middle Indo-Aryan so that most modern Indo-Aryan languages are syllable-oriented. By contrast, Old Iranian is rather word-oriented and shows a rise in the incidence of word-related features in Middle Iranian. The following parameters are examined: 1) phonotactic restrictions; 2) syllable structure; 3) syllable-related processes; 4) word-related processes; and 5) the phonological domain of consonant assimilation processes (sandhi). The results of the contrastive study can be summarized as follows. With regard to the distribution of vowels and consonants, Old Indo-Aryan has no stress-related or position-related restrictions while Old Iranian shows vowel quantity neutralization and monophthongization of short diphthongs in word-final position. In transmitted Avestan, there seems to be shortening of unstressed long vowels and lengthening of stressed short vowels. Additionally, Old IndoAryan has geminates that have resulted from assimilation processes that, conversely, have not occurred in Old Iranian. Syllable complexity is attested in Old Indo-Aryan and Old Iranian. However, complex structures are disfavored in IndoAryan compared to Old Iranian. In Middle Indo-Aryan, word-final consonants are not allowed and word-medial clusters are restricted to geminates. Violations of the Sonority Sequencing Principle are removed in Middle Indo-Arian by means of assimilation while they are retained in Middle Iranian. The syllable-related processes described are vowel epenthesis, consonant deletion in consonant clusters, and resyllabification. For example, Proto-Indo-European word-medial consonant clusters containing a laryngeal were resolved in Old Indo-Aryan through vowel epenthesis (cf. Vedic duhit vs. Avestan duɣða ‘daughter’). Simplification of consonant clusters by means of consonant deletion is more common in Old Indo-Aryan than in Old Iranian. The word-related process described is consonant epenthesis at the word margins. For example, Old Iranian has consonant epenthesis in word-initial position as in *šwaš > Avestan xšuuaš ‘six’. Finally, consonant assimilations (sandhi) within and across word and morpheme boundaries are discussed. While internal and external sandhi are found in Old Indo-Aryan,



Introduction: Syllable and word languages 

 29

external sandhi seem to be more restricted in Old Iranian. The evidence supported in the survey suggests that the prosodic category of the syllable is more relevant in Old and Middle Indo-Aryan while the prosodic category of the word is more relevant in Old and Middle Iranian. Damaris Nübling uses the typological model of syllable and word languages to account for the deep, ongoing onomastic change concerning German first names. In her contribution “From Christel to Christina, from Klaus to Nico: A diachronic study of German first names (1945‒2010) and their shift towards the syllable language type”, she concentrates on first names, which in contrast to last names are not inherited. The inventory of German first names is, hence, more likely to fluctuate, all the more from the beginning of the twentieth century, at the latest since 1945, when the transition from nachbenennung, a practice of (bound) naming after the parents or grandparents, the godfathers, saints, monarchs, etc., to free naming was completed. Since then, the inventory has grown considerably and naming has become more individual: While in the 1970s each of the most frequent male names covered 5‒6% of the newborn boys and each of the most frequent female names 3‒4% of the girls, today the most frequent name of both sexes covers only about 1%. Interestingly, the inventory is changing qualitatively as well: In the course of time, the most popular first names formally dissociate from the usual structure of common nouns (and last names). In order to grasp this formal dissociation, which can be seen as an expression of enduring social individualization, the author examines the structure of the most popular male and female first names from 1945 to 2010. The analysis reveals that a complete onomastic replacement has taken place. The top 20 lists from 1945, 1975, and 2005 only have one first name in common (Michael). The most frequent baby names of 1945 ‒ Gertrud, Christel (for girls) and Wolfgang, Klaus (for boys) ‒ did not differ from common nouns while the most frequent baby names of 2010 ‒ Mia, Lea (for girls) and Leon, Noah (for boys) ‒ show completely different structures compared to common nouns. The general tendency for girls’ as well as boys’ names is the shift to more syllable language structures. Today, first names exhibit CV syllables and full vowels both in stressed and unstressed position. Thus, they sharply contrast with the complex phonological structure of common nouns, which show complex consonant clusters and reduced vowels in unstressed position. The growing importance of hiatuses containing two full vowels increase the difference between first names and common nouns, the latter allowing only the combination of a full vowel and a schwa.

30 

 Javier Caro Reina and Renata Szczepaniak

Part 3: Synchronic approaches (Germanic languages) The phonetic investigation conducted by Pia Bergmann is dedicated to the “Reduction and deletion of glottal stops and geminates at phonological word boundaries in German compounds – Effects of word frequency and accentuation”. In contemporary German, which is a word language, the glottal stop and the geminates are important boundary markers. They facilitate the decoding of the speech stream by highlighting the morphological boundaries within compounds: Glottal stops, which are restricted to the vowel-initial onset of stressed (prototypically word-initial) syllable (e.g. Zahn#[Ɂ]arzt ‘dentist’), and geminates, which can arise only at the morphological boundary (e.g. [tː] in Haut#typ ‘skin type’), signal morphological complexity. This study shows that the reduction and deletion of both boundary markers in spoken language proceeds gradually with the increasing token frequency of the compound whereas accentuation ‒ that is, sentence accent ‒ works against it and strengthens the boundary. The investigation is conducted on 1456 test sentences read out loud by 14 speakers of a non-dialectal northwestern German variety. As for the glottal stop, the study reveals a gradient phonetic reduction including an intermediary stage of vowel glottalization. Complete deletion occurs significantly more often in high-frequent words and in unaccented position. Interestingly, in accented position the glottal stop is preserved regardless of the preceding segment. The articulatory make-up of the segments, however, is crucial in unaccented position. Here, the glottal stop reduction occurs most frequently after a vowel (V#V) while after /t/ (t#V) the glottal stop is realized even more often than in accented position. Furthermore, the frequency of the glottal stop after /n/ is higher than in V#V contexts, but lower than after /s/. This is due to the fact that /n/ (in contrast to /s/) shares the articulatory feature of glottal vibration with the following vowel, which favors the deletion of the glottal stop. For the analysis of degemination, besides the absolute duration of the geminate, the relative duration of the segment sequence with respect to the duration of the first two syllables also was measured. While token frequency appears to be decisive for the shortening of the whole word, the relative duration of the geminates depends on accentuation as well as the quality of the preceding vowel. Additionally, the test persons were asked for their auditive judgments about the number of segments at the morphological boundary in compounds such as Haut#typ ‘skin type’. Interestingly, they decided more frequently for one segment than for two in the case of high-frequent words. This tendency is significant in unaccented position. In accented position, the compound boundaries were significantly more often perceived as being produced with two segments. In “Phonological domains in Luxembourgish and their relevance for the phonological system”, Peter Gilles analyzes contemporary Luxembourgish, a West



Introduction: Syllable and word languages 

 31

Germanic language, which has been described as a typological mixed type and, hence, more syllable-based than closely-related Standard German. Luxembourgish has not only been subject to long-term intense language contact with French and German. It also exhibits a lesser degree of standardization. Both aspects make it especially interesting for a typological analysis. Gilles concentrates on selected features in order to determine the importance of the phonological domains of the syllable and the word. In contrast to German, schwa is not restricted to unstressed syllables in Luxembourgish. In other words, it does not highlight a certain word position. However, the vowel inventory is still reduced in unstressed position. Furthermore, the process of schwa epenthesis improving the syllabic structure is no longer productive. Instead, the syllable structure shows a high degree of complexity, including extrasyllabic consonants marking the word edges. The trochaic word structure is supported by a strong tendency to the penultima stress and to the reduction of pretonic syllables as well as posttonic syllables in trisyllabic words (e.g. Fenster-en [ˈfəns.tə.ʀən] > [ˈfəns.tʀən] ‘window-pl’). Furthermore, the clitization of the definite article results in complex consonant clusters profiling the left word edge as in /daːt/+/kɑnt/ > [tkɑnt] d’Kand ‘the child’. Luxembourgish exhibits two prominent cases of external sandhi which render the word boundaries unstable: the so-called n-rule and voicing assimilation across words. However, the visibility of the word boundaries is still guaranteed. With regard to the n-rule, three context-dependent processes must be distinguished: resyllabification of the word-final n when the following word begins with a vowel, n-preservation in front of homorganic consonants [t, d, ʦ] or h, and n-deletion in front of heterorganic consonants. The author observes that the n-rule cannot be interpreted as clearly syllable-related since it affects only the word-final syllables; that is, there is no n-deletion within the phonological word. Additionally, the n-rule is partially sensitive to morphological factors: n-deletion is blocked in words ending with the word-formation suffixes ‑in and ‑ioun as well as in the negative prefix on-. With regard to voice assimilation across words, resyllabification, which blurs the word boundaries by shifting the word-final consonant to the vowel-initial syllable of the next word, brings about voicing assimilation (e.g. Welt /vælt/ ‘world’ vs. Weltall [væl.‿d#ɑl] ‘universe’). Thus, the word boundaries still remain visible. This kind of voicing assimilation is blocked word-internally. In “Low German: A profile of a word language”, Steffen Höder elaborates on the typological differences between Modern Low German, which comprises a group of de-standardized North Low German dialects, and Standard (New High) German. Using the dialects spoken in the vicinity of Hamburg as a showcase, he argues that Modern Low German is even closer to the word language pole than Standard German. First of all, Modern Low German exhibits a higher stress sensitivity. The syllable structure is highly dependent on stress. The uneven distribu-

32 

 Javier Caro Reina and Renata Szczepaniak

tion of the total number of 26 potential syllable nuclei contributes to word coherence: 16 monophthongs and 5 diphthongs can occur in stressed syllables whereas only 5 monophthongs (including the vocalized /r/) are possible in pre-tonic position. Post-tonic syllables can contain only 5 vowels and 4 syllabic nasals and liquids (/m n ŋ l/). The latter are restricted to this position. Only three nuclei (/i ɪ a/) can occur in any type of syllable. Furthermore, both phonemic and allophonic vowel length as well as diphthongs are restricted to stressed syllables, which itself reinforces the culminative function of stress. Low German exhibits a word-related distributional rule for consonants: Two most prominent cases involve the flap [ɾ] and /sː/, both of which cannot occur at word boundaries. The phonological word is also decisive for a series of phonological processes. The Word-Medial Obstruent Voicing rule, which is frequent but optional in most dialects, brings about a word-related alternation between voiced medial and unvoiced word-external consonants as in the Altenwerder dialect /klɔpm/ ‘knock-inf’ → [ˈkʰlɔpm̩], [ˈkʰlɔbm̩]. This process, which marks the word by a word-internal sonority increase and a sonority decrease at the word boundaries, also affects the behavior of obstruents before postverbal ik ‘I’, which is incorporated into the phonological word of the preceding verbal form (e.g. /ˈʃrʊp#ɪk/ ‘scrub I’ → /ˈʃrʊpɪk/ → /ˈʃrʊbɪk/). The notion of the expanding phonological word also explains the phonological shape of the regular contraction of prepositions and articles in Low German. Here, the same word-based processes, among others Word-Medial Obstruent Voicing, apply (e.g. /ɔp/ ‘on’ + /dɛɪ̭/ ‘the’ > /ɔpɪ/ > [ɔbɪ̞] ‘on the’). In contrast to Standard German, Low German displays a word-level suprasegmental Knick phoneme, also called “dragging tone”. Knick occurs exclusively in word-final syllables and is normally restricted to the stressed syllable. For example, there is a prosodemic distinction between /vît/ ‘willow’ (with a slightly different pitch contour and a longer segmental duration) and /vit/ ‘far’. Finally, the author points to the striking typological similarities between Low German and Scandinavian languages, which can be interpreted as a result of contact-induced phonological convergence, worthy of being investigated in further studies. The phonetic-phonological approach chosen by Beat Siebenhaar in “Phonological and phonetic considerations for a classification of Swiss German dialect as a word language or syllable language” shows how durational phonetics can successfully be used for typological research. In contrast to Ramus, Nespor, and Mehler (1999) and Low, Grabe, and Nolan (2000), who calculate durational variation of vocalic and consonantal clusters, this contribution aims at phonetic correlates of typological differences. Siebenhaar’s starting point is the observation that the typological parameters directly influence the temporal organization of speech. By way of example, the unequal duration of syllables marking word structure results from the word-based distribution of different syllable types



Introduction: Syllable and word languages 

 33

whereas the preference for simple syllable structure leads to words comprising syllables of nearly equal length. In order to reexamine the typological status of Swiss German dialects, which have been hitherto classified as (more) syllable-optimizing (than Standard German), Siebenhaar first discusses the main typological features: Swiss German dialects display a very complex syllable structure and allow even fewer optimal syllable onsets than Standard German. However, the unstressed vowel system in Swiss German dialects is reduced to a lesser extent. In contrast to Standard German, where the deletion of schwa leads to the syllabic consonant formation of the word-final n (e.g. mach-en ‘make-inf’ [ˈmaxən] > [ˈmaxn̩]), schwa is retained in Swiss German dialects and word-final n is deleted (e.g. [ˈmaxən] > [ˈmaxə]), thereby optimizing the word-final syllable. Additionally, the word boundaries are not marked by initial glottal stop and word-final obstruent devoicing. Moreover, readjustments on the syllabic level (resyllabification, external sandhi, hiatus resolving epenthesis, and elision) weaken the word boundaries. In the subsequent durational analysis of spontaneous speech data, the author examines the role of the syllable and the word for duration of voiceless fricatives and schwa. When a word-final fricative is resyllabified (e.g. [d̥ɑs#lɑnd̥] > [d̥ɑ.slɑnd̥] ‘the land’), it is also lengthened and reaches the main duration of a word-initial fricative. Both processes, resyllabification and durational readjustment, weaken the word boundaries. In contrast, the duration of schwa is dependent on both the word and the syllable: Schwa is longer in word-final syllables than in non-word-final syllables. At the same time, the duration is, to a lesser degree, dependent on the syllable type, given that schwa is shorter in closed syllables than in open syllables. Finally, the tendency toward a later position of the pitch accent peak, which in Swiss German dialects shifts from the stressed to the following unstressed syllable, levels out the differences between the stressed and unstressed position of the word. In sum, the fine-grained phonetic analysis supports an affinity of Swiss German with the syllable language pole.

Part 4: Synchronic approaches (Romance languages) Javier Caro Reina’s paper “Central Catalan in the framework of the typology of syllable and word languages” offers an analysis of the syllable structure, distribution of the vowel and consonant inventory, and phonological processes. The author applies these parameters by distinguishing between stressed and unstressed position on the one hand and word-initial, word-medial, and word-final position on the other. Central Catalan displays a complex syllable structure that allows two consonants in the onset and a maximum of three consonants in the coda. Complex syllable structures appear more frequently in stressed position

34 

 Javier Caro Reina and Renata Szczepaniak

and they concentrate in word-final position (e.g. parc-s ‘park-pl’). With regard to the distribution of the vowel and consonant inventory, Central Catalan exhibits stress-dependent vowel reduction which comprises the centralization of /a ɛ e/ and the merger of /ɔ o u/ as [u] in unstressed position. In contrast, there are no essential asymmetries regarding the consonant inventory. With regard to the phonological processes, the paper focuses on word-related processes since syllable-related processes are the same as in other Catalan dialects. These processes include resyllabification, lenition of intervocalic voiced stops, coda voice agreement, and hiatus resolution. Interestingly, it is the word-related processes that differentiate Central Catalan from other Catalan dialects. The account of word-related processes in Southern Valencian, Northwestern Catalan, and Central Catalan clearly shows an increase in the incidence of word-related processes in Central Catalan. While some word-related processes are shared by all three dialects (word-final obstruent devoicing, affrication of word-final /ʒ/, obstruent voicing across words, deletion of unstressed vowels, and deletion of word-final n), others only occur in Central Catalan (vowel centralization, merger of back non-high vowels, and reinforcement of word-final r). The gradual increase of word-related processes found in the Catalan dialects reflects the typological development experienced by Central Catalan. These findings support the traditional dialect classification, which comes into a new light since the differences between the different dialects can be accounted for in terms of typological features. The results demonstrate the validity of the typology to explain language variation within a dialect group. Uli Reich and Martin Rohrmeier explore in “Batidas latinas: On rhythm and meter in Spanish and Portuguese and other forms of music” the common principles that organize rhythm and meter in music and language. The authors employ concepts derived from music theory and cognition in order to account for differing prosodic patterns in Portuguese and Spanish. Thus, music theory can contribute to a better understanding of supra-segmental phonology since prosodic differences between languages such as Portuguese and Spanish cannot be explained in terms of diverse configurations of prosodic grammatical functions. Different realizations of metrical structure on the one hand and metrically guided attention or reduction on the other are properties shared by music and language. In that respect, Portuguese and Spanish do not differ in the content-related phonological subsystem but rather in the level of rhythmic performance. The findings can be summarized as follows. First, word stress is part of the lexical subsystem of Portuguese and Spanish. Second, syllables emerge from the interaction of the organization of segments within metrical and gestural restrictions rather than belonging to Lexical Phonology. Third, metrical algorithms ascribe alternating strength to groups of sounds at different levels of the hierarchal organization, which sharply contrasts with music, where metrical strength is cyclical and



Introduction: Syllable and word languages 

 35

unbounded. Fourth, similar to music, moraic structure is not inherent to syllables or metrical systems. And fifth, the model is not viewed as procedural. Thus, the prosodic differences between Portuguese and Spanish result from the selection of the metrical level in the realization of the phonological representation of words. That is, while beats are realized at the base line of the metrical grid in Spanish, they are realized at higher levels of the metrical hierarchy in Portuguese, which brings about a reduction and deletion of unstressed vowels. The paper “Syllable typology and the rhythm class hypothesis: Evidence from Italo-Romance dialects” by Stephan Schmid brings together the typology of syllable and word languages and the rhythm class hypothesis on the basis of an empirical study of selected Italo-Romance dialects. The goal of the contribution is to demonstrate the fruitfulness of combining phonological and phonetic approaches for a typological classification of Italo-Romance dialects. The dialects under consideration are Bitontino, Feltrino, Friulian, Milanese, Neapolitan, Piedmontese, Pisan, Romagnolo, Sicilian, and Venetian. In the phonological approach, we find an account of the syllable in Italo-Romance regarding the number of syllable types, the sonority relations in word-initial and word-final clusters, and vowel reduction processes in unstressed position. One of the hypotheses tested is whether dialects that allow a higher number of syllable types will show marked syllable structures according to the Sonority Sequence Principle. In order to estimate the frequency of occurrence of different syllable types in the Italo-Romance dialects, the author conducted an analysis of the lexicon of dialect dictionaries. With regard to the maximum number of syllable types permitted, Pisan sharply contrasts with Friulian since they have a total number of 18 and 40 different syllable types, respectively. The validity of the Sonority Sequence Principle is assessed in the word-initial and word-final clusters of the dialects with the highest number of syllable types. Interestingly, Feltrino, Friulian, and Milanese have unmarked structures while Romagnolo and Turinese allow marked structures. Numerical complexity in word-final position is determined by diachronic apocope. A further typological feature relevant for the typology involves the differences between the phonemic and phonetic contrasts licensed in stressed and unstressed syllables. With regard to the vowel system, the paper focuses on the oppositions licensed in stressed and unstressed syllables (including vowel quality and quantity), centralization in unstressed syllables, and deletion of schwa in prepausal unstressed syllables. In the phonetic approach, the author performs an acoustic analysis of ten utterances segmented into vocalic and consonantal intervals in nine of the ten selected dialects by applying three different methods: 1) standard deviation measurements following Ramus, Nespor, and Vogel (1999); 2) Varco C, which introduces the factor “speech rate”; and 3) the Pairwise Variability Index (PVI), which measures mean differences between successive pairs of vocalic and consonantal

36 

 Javier Caro Reina and Renata Szczepaniak

intervals. The similar results obtained from both approaches demonstrates the viability of the combination of phonetics and phonology for the typology. Moreover, the dialect continuum resulting from the classification of the Italo-Romance dialects lends weight to the theoretical assumption that syllable and word languages are idealizations and that languages and language varieties may be closer to the syllable or the word pole according to the dominant phonological category.

References Abercrombie, David (1967): Elements of General Phonetics. Edinburgh: Edinburgh University Press. Allen, George D. (1975): Speech rhythm: Its relation to performance universals and articulatory timing. Journal of Phonetics 3: 75‒86. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (1994): Einige Argumente gegen die Silbe als universale prosodische Hauptkategorie. In: Karl Heinz Ramers, Heinz Vater and Henning Wode (eds.), Universale phonologische Strukturen und Prozesse, 55–78. Tübingen: Niemeyer. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Auer, Peter (2010): Word and syllable languages. Thoughts on the typological and diachronic relevance of a prosodic distinction. Paper presented at Phonological Typology of Syllable and Word Languages in Theory and Practice, workshop at the Freiburg Institute for Advanced Studies, University of Freiburg, March 29–31, 2010. Auer, Peter and Susanne Uhmann (1988): Silben- und akzentzählende Sprachen. Literatur­ über­blick und Diskussion. Zeitschrift für Sprachwissenschaft 7/2: 214–259. Auer, Peter, Elisabeth Couper-Kuhlen and Frank Müller (1999): Language in Time. The Rhythm and Tempo of Spoken Interaction. Oxford/New York: Oxford University Press. Auer, Peter and İnci Dirim (2004). Türkisch sprechen nicht nur die Türken. Über die Unschärfebeziehung zwischen Sprache und Ethnie in Deutschland. (Linguistik, Impulse & Tendenzen 4.) Berlin/New York: De Gruyter. Barnes, Jonathan (2006): Strength and Weakness at the Interface. Positional Neutralization in Phonetics and Phonology. (Phonology and Phonetics 10.) Berlin: Mouton de Gruyter. Bertinetto, Pier Marco (1989): Reflections on the dichotomy ‘stress’ vs. ‘syllable-timing’. Revue de Phonétique Appliquée 91/93: 99–130. Bertinetto, Pier Marco and Chiara Bertini (2008): On modeling the rhythm of natural languages. Quaderni del Laboratorio di Linguistica della SNS 7. Booij, Geert E. (1983): Principles and parameters in prosodic phonology. Linguistics 21/1: 249‒280. Caro Reina, Javier (2013): Phonological variation in Catalan and Alemannic from a typological perspective. In: Peter Auer, Javier Caro Reina and Göz Kaufmann (eds.), Language Variation ‒ European Perspectives IV. Selected papers from the Sixth International



Introduction: Syllable and word languages 

 37

Conference on Language Variation in Europe (ICLaVE 6), Freiburg, June 2011, 27–44. (Studies in Language Variation 14.) Amsterdam: John Benjamins. Cauldwell, Richard (2002): The functional irrhythmicality of spontaneous speech: A discourse view of speech rhythms. Apples 2/1: 1‒24. Couper-Kuhlen, Elisabeth (1993): An Introduction to English Prosody. London: Edward Arnold. Cutler, Anne, Jacques Mehler, Dennis Norris and Juan Segui (1986): The syllable’s differing role in the segmentation of French and English. Journal of Memory and Language 25: 385‒400. Dauer, Rebecca (1983): Stress-timing and syllable-timing reanalyzed. Journal of Phonetics 11: 51–62. Dauer, Rebecca (1987): Phonetic and phonological components of language rhythm. In: Proceedings of the 11th International Congress of Phonetic Sciences, vol. 5, 447–450. Tallinn: Academy of Sciences of the Estonian S.S.R. Delattre, Pierre (1966): A comparison of syllable length conditioning among languages. International Review of Applied Linguistics in Language Teaching 4/1‒4: 183‒198. Dimitrova, Snezhina (1997): Bulgarian speech rhythm: Stress-timed or syllable-timed? Journal of the International Phonetic Association 27: 27‒33. Dixon, Robert M. W. and Alexandra Y. Aikhenvald (2002): Word: A typological framework. In: Robert M. W. Dixon and Alexandra Y. Aikhenvald (eds.), Word: A Cross-Linguistic Typology, 1–41. Cambridge/New York: Cambridge University Press. Donegan, Patricia J. and David Stampe (1983): Rhythm and the holistic organization of language structure. In: John F. Richardson, Mitchell Marks and Amy Chukerman (eds.), Papers from the Parasession on the Interplay of Phonology, Morphology and Syntax, Chicago, 22‒23 April 1983, 337‒353. Chicago: Chicago Linguistic Society. Faure, George, Daniel Hirst and Michel Chafcouloff (1980): Rhythm in English: Isochronism, pitch, and perceived stress. In: Linda Waugh, Linda and Cornelis van Schooneveld (eds.), The Melody of Language, 71‒79. Baltimore: University Park Press. Frota, Sónia and Marina Vigário (2001): On the correlates of rhythmic distinctions: The European/Brazilian Portuguese case. Probus 13/2: 247–275. Gibbon, Dafydd and Ulrike Gut (2001): Measuring speech rhythm. EUROSPEECH-2001: 91‒94. Gill, David (1986): A prosodic typology of language. Folia Linguistica 20: 165‒231. Grabe, Esther and Ee Ling Low (2002): Durational variability in speech and the Rhythm Class Hypothesis. In: Carlos Gussenhoven and Natasha Warner (eds.), Laboratory Phonology 7, 515–545. Berlin/New York: Mouton de Gruyter. Hall, T. Alan and Ursula Kleinhenz (eds.) (1999): Studies on the Phonological Word. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Hoequist, Charles E., Klaus J. Kohler and Kurt Schäfer-Vincent (1986): Further experiments on speech rate perception with logatomes. In: Charles E. Hoequist, Klaus J. Kohler and Kurt Schäfer-Vincent (eds.), Speech Rate ‒ Final Report on a Research Project, 7–28. (AIPUK 22.) Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Universität Kiel. Hyman, Larry M. (2008): Directional asymmetries in the morphology and phonology of words, with special reference to Bantu. Linguistics 46: 309–350. Kern, Friederike (2011): Rhythmus im Türkendeutschen. In: Thomas Stehl (ed.), Sprachen in mobilisierten Kulturen: Aspekte der Migrationslinguistik, 209‒298. (Mobilisierte Kulturen 2.) Potsdam: Universitätsverlag. Klein, Thomas B. (2011): Typology of creole phonology: Phoneme inventories and syllable templates. Journal of Pidgin and Creole Languages 26/1: 155–193.

38 

 Javier Caro Reina and Renata Szczepaniak

Kohler, Klaus J. (2009): Rhythm in speech and language. A new research paradigm. Phonetica 66: 29–45. Ladefoged, Peter (1975): A Course in Phonetics. New York: Harcourt Brace & Jovanovich. Lehiste, Ilse (1973): Rhythmic units and syntactic units in production and perception. Journal of the Acoustical Society of America 54: 1228‒1234. Lehiste, Ilse (1977): Isochrony reconsidered. Journal of Phonetics 5: 253‒263. Lehiste, Ilse (1979): The perception of duration within sequences of four intervals. Journal of Phonetics 7: 313‒316. Low, Ee Ling, Esther Grabe and Francis Nolan (2000): Quantitative characterizations of speech rhythm: Syllable-timing in Singapore English. Language and Speech 43/4: 377–401. Maddieson, Ian (2011): Syllable structure. In: Martin Haspelmath, Matthew S. Dryer, David Gil and Bernard Comrie (eds.), The World Atlas of Language Structures (WALS) Online. Munich: Max Planck Digital Library. Available at http://wals.info/chapter/12. Mehler, Jacques, Peter Jusczyk, Ghislaine Lambertz, Nilofar Halsted, Josiane Bertoncini and Claudine Amiel-Tison (1988): A precursor of language acquisition in young infants. Cognition 29/2: 143‒78. Nazzi, Thierry, Josiane Berto­ni­ci­ni and Jacques Meh­ler (1998): Language discrimination by newborns. Toward an understanding of the role of rhythm. Journal of Experimental Human Perception and Perfomance 24/3: 756‒766. Nespor, Marina and Irene Vogel (1986): Prosodic Phonology. (Studies in Generative Grammar 28.) Dordrecht: Foris. Noske, Roland (2007): Schwa on the border between Dutch and French. Two refutations of assumptions about the histories of Dutch and French. Actes des 5èmes Journées d’Études Linguistiques (JEL’2007), Schwa(s): 61‒68. Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 129.) Stuttgart: Steiner. Nübling, Damaris and Renata Szczepaniak (2008): On the way from morphology to phonology. German linking elements and the role of the phonological word. Morphology 18: 1–25. Nübling, Damaris and Renata Szczepaniak (2009): Religion+s+freiheit, Stabilität+s+pakt und Subjekt(+s+)pronomen: Fugenelemente als Marker phonologischer Wortgrenzen. In: Peter O. Müller (ed.), Studien zur Fremdwortbildung, 195–222. (Germanistische Linguistik 197/198.) Hildesheim: Olms. Pamies Bertrán, Antonio (1999): Prosodic typology: On the dichotomy between stress-timed and syllable-timed languages. Language Design 2: 103‒130. Patel, Aniruddh D. and Joseph R. Daniele (2003a): An empirical comparison of rhythm in language and music. Cognition 87: B35‒B45. Patel, Aniruddh D. and Joseph R. Daniele (2003b): Stress-timed vs. syllable-timed music? A comment on Huron and Ollen (2003). Music Perception 21/2: 273‒276. Pike, Kenneth L. (1945): The Intonation of American English. Ann Arbor: University of Michigan Press. Plag, Ingo and Mareile Schramm (2006): Early creole syllable structure: A cross-linguistic survey of the earliest attested varieties of Saramaccan, Sranan, St. Kitts and Jamaican. In:



Introduction: Syllable and word languages 

 39

Parth Bhatt and Ingo Plag (eds.), The Structure of Creole Words: Segmental, Syllabic and Morphological Aspects, 131–150. Tübingen: Niemeyer. Plank, Frans (1998): The co-variation of phonology with morphology and syntax: A hopeful history. Linguistic Typology 2/2: 195–230. Pointon, Graham E. (1980): Is Spanish really syllable-timed? Journal of Phonetics 8: 293‒304. Prieto, Pilar, Maria del Mar Vanrell, Lluïsa Astruc, Elinor Payne and Brechtje Post (2012): Phonotactic and phrasal properties of speech rhythm. Evidence from Catalan, English, and Spanish. Speech Communication 54/6: 681–702. Pulgram, Ernst (1970): Syllable, Word, Nexus, Cursus. The Hague: Mouton. Quist, Pia (2008): Sociolinguistic approaches to multiethnolect: language variety and stylistic practice. International Journal of Bilingualism 12: 43–61. Raffelsiefen, Renate (2000): Evidence for word-internal phonological words in German. In: Rolf Thieroff, Matthias Tamrat, Nanna Fuhrhop and Oliver Teuber (eds.), Deutsche Grammatik in Theorie und Praxis, 43‒56. Tübingen: Niemeyer. Ramus, Franck, Marina Nespor and Jacques Mehler (1999): Correlates of linguistic rhythm in the speech signal. Cognition 73: 265‒292. Revithiadou, Anthi (2011): The phonological word. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 2: Suprasegmental and Prosodic Phonology, 1204‒1227. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Roach, Peter (1982): On the distinction betweeen ‘stress-timed’ and ‘syllable-timed’ languages. In: David Crystal (ed.), Linguistic Controversies. Essays in Linguistic Theory and Practice in Honour of F. R. Palmer, 73‒79. London: Arnold. Sampson, Rodney (1999): Nasal Vowel Evolution in Romance. Oxford: Oxford University Press. Schiering, René, Balthasar Bickel and Kristine A. Hildebrandt (2010): The prosodic word is not universal, but emergent. Journal of Linguistics 46/3: 657–709. Schmid, Stephan (1997): A typological view of syllable structure in some Italian dialects. In: Pier Marco Bertinetto, Livio Gaeta, Georgi Jetchov and David Michaels (eds.), Certamen Phonologicum III, 247–265. Torino: Rosenberg & Sellier. Schrambke, Renate (2003): Der alemannische Sprachraum. Ältere Gliederungen und ein neuer Versuch. Alemannisches Jahrbuch 2001–2002: 161–189. Schrambke, Renate (2007): Changing from syllable-rhythm to word-rhythm: Parallels between Danish and Northern Alemannic. Dialectologia et Geolinguistica 15: 102–115. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Szczepaniak, Renata (2009): Wortsprachliches Deutsch und silbensprachliches Spanisch. Ein phonologisch-typologischer Vergleich. Estudios filológicos alemanes 17: 251–267. Szczepaniak, Renata (2010): Phonologisch-typologischer Wandel des Deutschen und des Luxemburgischen im Kontrast. In: Antje Dammel, Sebastian Kürschner and Damaris Nübling (eds.), Kontrastive Germanistische Linguistik, vol. 1, 85–110. (Germanistische Linguistik 206/209.) Hildesheim: Olms. Szczepaniak, Renata (2012): Lautwandel verstehen. Vom Nutzen der Typologie von Silbenund Wortsprachen für die historische und die synchrone germanistische Linguistik. In: Péter Maitz (ed.), Historische Sprachwissenschaft. Erkenntnisinteressen, Grundlagenprobleme, Desiderate, 85–104. (Studia Linguistica Germanica 110.) Boston: de Gruyter.

40 

 Javier Caro Reina and Renata Szczepaniak

Torgersen, Eivind and Anita Szakay (2012): An investigation of speech rhythm in London English. Lingua 122/7: 822‒840. Trubetzkoy, Nikolaj S. (1971): Grundzüge der Phonologie. 5th ed. Göttingen: Vandenhoeck & Ruprecht. Vennemann, Theo (1986): Neuere Entwicklungen in der Phonologie. Berlin: Mouton de Gruyter. Vennemann, Theo (1988): Preference Laws for Syllable Structure and the Explanation of Sound Change. With Special Reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. Warner, Natasha and Takayuki Arai (2001): Japanese mora-timing. A review. Phonetica 58: 1‒25. Wells, J. C. (1982): Accents of English. Vol. 3: Beyond the British Isles. Cambridge: Cambridge University Press. Wheeler, Max (2005): The Phonology of Catalan. (The Phonology of the World’s Languages.) Oxford: Oxford University Press. Wiese, Richard (1996): The Phonology of German. (The Phonology of the World’s Languages.) Oxford: Clarendon Press.

Part 1: Theoretical issues

Stig Eliasson (University of Mainz)

The typology of syllable and word languages and Swedish phonological structure* Abstract: Selected phonological phenomena of Central Standard Swedish are examined within the typological framework of syllable and word languages. The main observations are: Stress assignment is central, conditioning numerous synchronic phonological rules relevant to the typology. Tonal accent is frequently connective, signaling morphologically complex words. Vowel quantity tends to indicate a following morpheme boundary. Phonetically ambisyllabic consonants, seen as a word-language feature in the model, present a special case of classification, as they mostly reflect underlying geminates. Inner and outer sandhi caused by r-triggered “retroflection” of dentals is structurally uniform, a situation that the model views as characteristic of syllable rather than word languages. Natural syllabification, while in lento and normal speech often blocked on syntactic and morphological grounds, is prominent in fast informal speech. All in all, Swedish is more similar to a prototypical word language than to a prototypical syllable language. Noteworthy from a theoretical angle is that several phonological rules refer to morphosyntactic categories directly, rather than indirectly via the Prosodic Hierarchy. The fact that phonological properties and processes regularly and directly profile distinct grammatical units underscores the important role of the configurative or structure-signaling function in language.

1 Introduction The seminal typology of syllable and word languages (here for short: SL/WL typology) introduced by Auer (1993, 2001) arranges languages along a continuum or scale, whose end-points are prototypical syllable and word languages, respectively. According to Auer (2010), “[a] word language is one in which the prosodic domain of the phonological word is of central importance, [and] a syllable language is one in which the prosodic domain of the syllable is of central importance”, where the importance is established by the fact that “a maximum of phonological regularities make reference to the domain (‘profile it’)” (similarly, Auer 1993: 90–91). In contrast to typologies of phoneme inventories (e.g. Trubetzkoy

* I am greatly indebted to Javier Caro Reina and Renata Szczepaniak for comments on an earlier version of this paper. Any remaining deficiencies and shortcomings are entirely my own.

44 

 Stig Eliasson

1962; Hockett 1955; Maddieson 1984) and cross-linguistic investigations of phonological processes or rules (Bhat 1978; Ferguson 1978; Ultan 1978; etc.), the SL/ WL typology links phoneme systems, phonotactics, and phonological processes as well as the diachronic development of these to a hierarchy of differently-sized “prosodic” constituents or units (cf. Auer 2001: 1393, referring to Nespor and Vogel’s 1986 theory of Prosodic Phonology). Specifically, in his typology founded on prosodic constituents, Auer (1993: 91, 2001: 1393) takes the syllable and the phonological word to be the key entities. This paper attempts to situate certain phenomena in the phonology of Central Standard Swedish within the framework of the SL/WL typology and comments on some descriptive and theoretical issues connected with this application. I limit the discussion to the synchronic facts, referring to Braunmüller (this volume) for a diachronic, pan-Scandinavian, perspective. The available space allows only a brief characterization of the respective Swedish phenomena (for qualifications and details on Swedish phonology, cf. the references provided). As expected from a Germanic language, Swedish by and large adheres to an overall phonological organization similar to that of modern German (on the latter, see Szczepaniak 2007: 269–317), with the phonological word being of high relevance. It must be noted, though, that Nübling and Schrambke (2004: 287, 304) treat Swedish ‒ measured against what they call the accent languages: German, English and Danish ‒ essentially as a syllable language, while Page (2009: 137) declares all Germanic languages without exception to be word languages. The main observations of this paper are as follows: First, stress assignment emerges as a central property in Swedish phonology that conditions numerous other phonological rules that are relevant to SL/WL typology, including tone assignment, the complementary V/C lengthening, low-level phonetic diphthongization, centralization of short vowels (with schwa formation), unvoiced stop aspiration and vowel/zero alternations. To a large extent, this cohort of rules is indicative of a word language. Second, albeit to varying degrees, word-internal and word-external sandhi caused by the extremely productive rule of supradentaliza­tion (r-triggered “retroflection” of dentals) is strikingly uniform, a situation that is considered to be characteristic of a syllable language. Third, while frequently blocked on syntactic and morphological grounds (cf. for German, Szczepaniak 2007: 34, 314–315, etc.), natural phonetic syllabification ‒ presumably a syllable-language feature ‒ does surface, especially in fast informal speech. Fourth, phonological means may often be taken to signal grammatical words and morphemes directly, as when tonal accent (accent 2) marks compounds, e.g. in grammatical words consisting of smaller phonological words, or when a long stressed vowel before a consonant cluster points to a morpheme boundary after the first cluster member. Finally, the fact that many phonological properties and processes appear to directly profile several distinct



The typology of syllable and word languages and Swedish phonological structure 

 45

grammatical units underlines the fundamental importance of structure-signaling means in language and in linguistic communication (cf. Trubetzkoy 1962: 29–30, 241–261; Jakobson and Halle 1956: 9; Auer 2001: 1395). This observation makes it interesting to explore the degree of elaboration of structure-signaling devices cross-linguistically.

2 Theoretical preliminaries 2.1 Prosodic Phonology The theory of Prosodic Phonology employs the term “prosodic” in a different sense from that of traditional phonetics and phonology. As Vogel (2009: 60) puts it: “The term ‘prosodic phonology’ is ambiguous in that it may refer either to (a) phonological phenomena termed ‘prosodic’, as opposed to segmental (e.g., tone, stress, intonation), or (b) the theory of phonology consisting of a hierarchically arranged set of constituents that are not necessarily isomorphic to constituents of other components of grammar”. According to Nespor and Vogel (1986: 11, 16), the Prosodic Hierarchy itself consists of seven units: phonological utterance, intonational phrase, phonological phrase, clitic group, phonological word, foot and syllable. Auer (1993: 90, 2001: 1393) and Vogel (2009: 61) also include the mora in the hierarchy. Prosodic Phonology presumes a double tree structure for sentences, one “prosodic” tree and one syntactic tree, as diagrammed for an Italian example in Nespor (1999: 123).

2.2 The phonological word The syllable (σ) and the phonological word (w or PW), which form the pillars of the SL/WL typology, are both difficult to delimit precisely.1 But whereas the notion of syllable may momentarily be taken for granted, the phonological word (also named prosodic word, pword, or mot) deserves a brief comment. Nespor and Vogel (1986: 109) write: “The phonological word is the lowest constituent of the prosodic hierar­chy which is constructed on the basis of mapping rules that make substantial use of non­phonolog­ical notions. In particular, the phonological word (w) represents the interaction between the phonological and the morphological components of the grammar”. More exactly, they offer the following com-

1 On miscellaneous conceptions of the syllable, cf. e.g. papers in Cairns and Raimy (2011).

46 

 Stig Eliasson

posite definition (Nespor and Vogel 1986: 141; cited here in the slightly modified formulation of Revithiadou 2011: 1209): 2 (1) A. The domain of PW is Q (= any terminal element of the syntactic tree). or B. I. The domain of PW consists of a. a stem; b. any element identified by specific phonological and/or morphological criteria; c. any element marked with the diacritic [+W]. II. Any unattached elements within Q form part of the adjacent PW closest to the stem; if no such PW exists, they form a PW on their own.

Among the miscellaneous elements of this definition, we may for present purposes especially note the stem of a word, which Nespor and Vogel (1986: 18) define as “the underived, uninflected form of a word”, corresponding apparently to the notion of root morpheme.

2.3 Language-particular and cross-linguistic comparability of phonological words The diagnostic criteria for phonological words are of various kinds, but ideally they should concur ‒ within a language as well as across languages.3 Nonetheless, not infrequently there exists in a specific language a “lack of agreement among different types of diagnostic criteria” (Revithiadou 2011: 1221). Hyman (2008: 335) even suggests that “[i]t may in fact be necessary to distinguish different kinds of PW, which may or may not coincide in a given language”. He designates “[s]ome of the possibilities” as follows: (2)

Some possible kinds of phonological words according to Hyman (2008: 335–336): (a) the demarcative word a property marks the beginning or end of the word (b) the culminative word a feature occurs only once per word (c) the harmonic word a feature is realized throughout the word (d) the metrical word a word consists of hierarchically arrayed moras or syllables (e) the minimal word a word must consist of a minimum [number] of moras or syllables

2 Hall (1999) and Dixon and Aikhenvald (2002) discuss varying conceptions of the phonological word as opposed to the grammatical or morphological word. 3 Hall (1999: 3–8), Raffelsiefen (1999) and Revithiadou (2011: 1216–1221), among others, discuss various diagnostics of phonological wordhood.



The typology of syllable and word languages and Swedish phonological structure 

 47



(f) the maximal word a word can consist of a maximum [number] of moras or syllables (g) the phonotactic word a word permits only certain output segments/ sequences (h) the morphophonotactic a word permits only certain input segments/sequences word

More generally, the problem also concerns the “cross-linguistic comparability” of the phonological word – that is, whether the “PW is well-defined across languages” (Hyman 2008: 335). Schiering, Bickel, and Hildebrandt (2010), in a similar vein as Hyman, argue that while some languages such as Vietnamese exhibit no phonological-word domain at all, certain other languages such as the SinoTibetan Limbu exhibit multiple word domains.4 In view of this state of affairs, they conclude: “Facing the fact that even such finer-graded word notions differ across languages, we would go one step further in positing only language-particular prosodic words” (Schiering, Bickel, and Hildebrandt 2010: 701; emphasis added). This then concurs with Auer’s (1993: 91) assessment that “the exact definition of [the phonological word] can only be given for a specific language”. But if the definition of phonological wordhood is language-specific, this means that the phonological words of designated word languages may differ and that a typology based on the phonological word must cope with the question of how to reconcile the differences.

2.4 One or several SL/WL continua The diverging cross-linguistic characterizations of phonological words directly affect the basic set-up of the SL/WL model. Ideally, in this kind of typology differences between more syllable-centered and more PW-centered languages would be organized along a single continuum or parameter as in Figure 1. PSL PWL Figure 1: One single continuum between the prototypical syllable language (PSL) and the prototypical word language (PWL)

The two end-points of the scale in Figure 1 are not equally easy to define, however.

4 Vogel (2009: 63–64) attempts a brief response to these strictures.

48 

 Stig Eliasson

One of the most fundamental features of syllable languages as opposed to word languages is the recurrence and enhancement of the same syllable type throughout the word. Auer writes (italics added): With increasing shell complexity, a language will […] be increasingly likely to have different cluster phonotactics in word-internal and word-marginal syllable shells (left onset of initial syllable and right coda of final syllable). For instance, in a language with a maximal shell complexity of CV, word-internal syllables and word-marginal syllable shells are quite likely to be identical, i.e. we may find words of the type CV&CV; a language with a maximal shell complexity of CVC may have words of the type CVC&CVC, but it may also have restrictions on word-internal clusters, restricting them to single consonants, i.e. the pattern CV&CV(C). In a language of a maximal shell complexity of CCCVCCC, however, it is extremely unlikely that word patterns such as CCCVCCC&CCCVCCC will occur; in all likelihood, word-internal clusters will be less complex (for instance, up to CCCVCCCVCCC), as will be non-marginal syllable shells. This imbalance between word-internal and word-marginal syllable shells, in turn, is a characteristic feature of word languages, while the identity of syllable phonotactics in all environments in the word is typical for syllable languages. (Auer 1993: 92)

The symmetry of syllable phonotactics throughout the word is quite a tangible criterion for defining the prototypical SL pole, but, given the possibility of several kinds of deviations from recurring syllable phonotactics, slightly less efficient in defining a single prototypical WL pole. Furthermore, if criteria for determining phonological wordhood do not necessarily converge within a language or cross-linguistically (see section 2.3), this would result in multiple prototypical word languages, as Figure 2 illustrates. PW L 1 PW L 2 PSL

PW L 3

PW L n Figure 2: Multiple continua between the prototypical syllable language and word languages

Assuming that Figure 2 can be said to better represent the SL/WL model than Figure 1, attention would also have to be paid to the relations between the different kinds of phonological words.



The typology of syllable and word languages and Swedish phonological structure 

 49

2.5 Indirect reference to morphosyntax Insofar as SL/WL typology takes Prosodic Phonology as one of its foundations, it may also lean on the notion of indirect reference to morphosyntactic structure. In Prosodic Phonology, “[i]ndirect reference is a grammatical meta-principle which requires that phonological rules or constraints are not allowed direct access to morphological or syntactic information” (Trommer 2011: 2485). In other words, phonological rules are assumed to draw on such information indirectly via the units of the Prosodic Hierarchy. Presupposing a generative unidirectional (derivational) model of language, Nespor and Vogel put the idea as follows: … once the mapping rules have used the morphological information needed to create the appropriate phonological constituents, this information is no longer available to phonological rules. In other words, any phonological rule that applies in a domain created on the basis of morphological structure may refer only to the phonological domain, not to the morphological elements in the corresponding morpho-syntactic tree. (Nespor and Vogel 1986: 18)

Given that the primary motivation for establishing a prosodic hierarchy alongside the morphosyntactic one is non-isomorphism between prosodic and morphosyntactic categories, they further state that “[t]he superiority of a theory of prosodic phonology that does not directly refer to morpho-syntactic boundaries […] is not undermined by sporadic cases of isomorphism” (Nespor and Vogel 1986: 110; italics added). Interestingly, though, Nespor later assesses the extent of isomorphism a bit differently (italics added): Given the interpretative function of phonology, it is to be expected that the constituents of the prosodic hierarchy are not totally independent from the rest of the grammar. In most cases, in fact, the prosodic constituents (constructed through mapping rules on the basis of specific notions taken from morphology, syntax and semantics) are coextensive with constituents of another component. This is, however, not always the case. The main motivation for having a phonological hierarchy separated from the well-established morphosyntactic hierarchy lies in the fact […] that phonological rules apply within domains that are not necessarily coextensive with the constituents into which a string is analysed morphologically and syntactically. […] What is important is that, in cases of discrepancy, phonology is organized according to the prosodic hierarchy, not to any other constituent analysis. (Nespor 1999: 118–119)

Yet, if prosodic constituents are “[i]n most cases, in fact”, isomorphic with morphosyntactic constituents, this will involve a large-scale duplication of structure that runs counter to the goal of constructing a parsimonious overall model of language. To this extent, the question remains whether phonological rules could actually at times be free to make direct reference to morphological categories.

50 

 Stig Eliasson

2.6 Domain of phonotactic constraints Much debated in linguistics and not easily resolved is the issue of what is the proper unit for the description of phonotactic constraints in languages: syllables, words or morphemes. In Prosodic Phonology, “[t]he PW has been considered an impor­ tant domain for phonotactic constraints, together with the syllable and the foot” (Revithiadou 2011: 1219). A prominent exponent of this view of the prosodic word is Booij (1999), who in an optimality-theoretic vein holds that “there are no statements in the grammar as to which sequences of sounds are possible under­lying forms for morphemes” (Booij 1999: 51), i.e. there are “[no] constraints on underlying forms” (Booij 1999: 47). He further claims that “the relevant constraints pertain to the surface level, i.e. they are output con­straints” (Booij 1999: 47), and that “the phonotactics of a language is to be stated in terms of properties of syllables, feet and prosodic words, and in terms of conditions of alignment between the edges of grammatical and prosodic categories” (Booij 1999: 70) (cf. also Hall 1999: 6–7). Nevertheless, holding to a process model of phonology and given that phonotactics must be stored in the mental lexicon along with distinctive segmental and distinctive prosodic units, it is difficult to relegate the phonotactic description entirely to the phonetic output. But if basic phonotactics is to be accounted for at the underlying rather than the phonetic level,5 this decision eliminates the syllable as a candidate for expressing distributional constraints in many languages. For syllable structure as such is not likely to be specified in the linguistic lexicon on grounds outlined by Blevins (cf. also Blevins 2004: 232–233): Three observations suggest that in the general case, syllable structure is not present in under­lying representations: (1) minimal pairs distinguished by syllabification alone are rare, and are nonexistent in many languages; (2) segments in many languages exhibit syllabicity alternations which can be viewed as the simple result of derived syllabi­fica­tion; (3) individual morphemes often fail to conform to the possible syllable types of a given language, making lexical syllabification infelicitous. (Blevins 1995: 221)

Additionally, as Booij (2011: 2056) observes, “not all phonotactic constraints can be reduced to syllable structure constraints”. As far as words are concerned, their phonotactic structure can be accounted for by morpheme phonotactics together with morphological rules of derivation, inflection and composition, thereby disfavoring the word as the basic phonotactic domain. Finally, since morphemes are the minimal lexical building blocks, they would seem also to constitute the

5 However, cf. Lexical Phonology, which “take[s] the ‘morpheme structure rules’ of a language to be so many rules of its lexical phonology” (Kiparsky 1982: 167).



The typology of syllable and word languages and Swedish phonological structure 

 51

natural domain for stating phonotactic regularities. The strong syllabic effects on morpheme phonotactics, on the other hand, may then be viewed as a result of the process operating in language acquisition of building underlying forms on the basis of phonetic surface forms.

2.7 Direction of processing in the phonological component and configurative function As seen in section 2.5, following generative tradition, Nespor (1999: 118) considers the phonological component of a language to be “interpretative” – that is, phonology is a module that is appended to the “central” component of syntax and whose task is to derive phonetic forms. This premise may, however, be overly simplified. An undercurrent in several phonological or psycholinguistic studies  ‒ though rarely, if ever, expressed by special mechanisms in explicit formal models ‒ is that phonetic form is also used to recover underlying phonological and grammatical structure (e.g. Shin 2007). A similar idea occurs in work on SL/WL typology. Szcze­ paniak (2009: 49) ties in the typological distinction between syllable and word languages with the speaker and hearer perspectives on linguistic structure. A syllable language, she says, renders the speaker’s task easier, since it favors easily pronounceable, similarly formed syllables throughout the word, but the hearer’s task harder, because the phonetic forms will contain fewer clues for inferring morphological boundaries. A word language, in contrast, favors the listener by providing numerous structure-signaling clues, yet producing these puts an added burden on the speaker. Following up on Szczepaniak’s observation, both phonetic properties reflecting lexical phonotactics and phonetic features assigned by certain phonological rules may assist in the identification of the grammatical structure of an utterance. In the cases where structural clues have been assigned by rules, these same rules may also be employed in the recovery of underlying phonological forms, either by themselves or in combination with cognitive strategies (Eliasson 1997). Consequently, operating both ways, phonology will not merely derive, but also recover representations, if need be, in conjunction with cognition. Grossly categorized, rule-derived as well as lexical structure-signaling phonetic properties ‒ called “configurative” by Jakobson and Halle (1956: 9) ‒ are either delimitative (demarcating) or culminative. The delimitative features “indicate the boundary between two units (= close word combinations, words, morphemes)” (Trubetzkoy 1962: 29; translation SE). The culminative features, which Auer (2001: 1395) holds to be possibly the most important ones for the SL/WL typology, indicate “how many ‘units’ (= words, or word combinations) are contained in a given sentence” (Trubetzkoy 1962: 29; translation SE), or, in Auer’s

52 

 Stig Eliasson

formulation, set off “one and only one syllable (or mora) over all others within a unit of the size of the phonological word” (Auer 2001: 1395; translation SE).

2.8 Syllabification A phonological process of special interest to SL/WL typology is syllabification. From the perspective of a process model of phonology, the relation between underlying phonological structure and syllabified speech may be represented as in Figure 3. Underlying structure

Chains of morphemes Syllabification

Pronunciation

↓↑ Syllabified speech

Grammatical boundaries marked Not syllabified (in a phonetic sense)

Desyllabification Grammatical boundaries effaced, but often signaled by phonological means

Figure 3: Relation between sequences of morphemes and syllabified phonetic strings (adapted from Eliasson 2010a: 7)

As Figure 3 is meant to suggest (cf. also section 2.7), in language production speakers convert lexically and grammatically demarcated phonological representations into pronounceable syllabified sequences, while in language comprehension hearers reconvert the latter into grammatically and lexically meaningful chunks. Syllabification patterns may differ between languages (cf. e.g. French and German in this regard), but there are also essential syllabification tendencies that apply in a fairly uniform manner cross-linguistically (section 5).

2.9 Diagnostics for syllable and word languages A central part of the SL/WL framework is the set of criteria for diagnosing syllable and word languages. Auer (2001: 1395–1398) provides a convenient succinct listing and discussion of the most important ones.6 The particular SL/WL criteria employed in this paper are those in Table 1.

6 Auer (1993: 11), Auer and Uhmann (1988: 253), Nübling and Schrambke (2004: 284–285),



The typology of syllable and word languages and Swedish phonological structure 

 53

Table 1: Selected properties of prototypical syllable and word languages (basically following Auer 2001: 1395–1398) Diagnostic feature 1

2

3

4

5

6

Word accent and tone

Usually initial or final, phonetically weak, accent in the phono­logical phrase. Tone possible in every syllable. Syllable structure (esp. Syllable structure is the syllable shell com­plexity) same for all positions in the in different positions within word. the word (Preference for CV sequences in all word positions.) Geminate vs. ambisyllabic Geminates in stressed and consonants unstressed position possible, but no ambi­syllabic consonants. Sandhi

Outer (word-external) sandhi = inner (wordinternal) sandhi. Vowel inventory in accented No accent-dependent phoand unaccented position nemic or phonetic reduction in unaccented syllables (e.g. unaccented long vowel phonemes possible and little centralization). (a) V deletions Deletions optimize CV structure. (b) V or C insertion

7

Syllable language

Word- or syllable-related phonological processes

Insertions get rid of complex syllable edges or prevent hiatus. Syllable-related phono­ logi­cal processes such as resyllabification across grammatical (including syntactic) boundaries, syllable-final weakening, etc.

Word language Dynamic or musical accent empha­sizing a syllable or a mora within the phonological word. Syllable structure differs ac­cord­ing to the location of the syllable within the word. Accented and word-margin sylla­bles exhibit more complex edges. Ambisyllabic consonants allowed; geminates disallowed, or allowed only in certain positions within the phonological word. Outer sandhi ≠ inner sandhi. Accent-dependent phonemic or phonetic reduction in unaccented syllables (e.g. no long unaccented vow­els, centralization of unstressed vowels). Deletions produce complex accented or word-edge syllables. Insertions of consonants add to the complexity of syllable edges. Accent-dependent or word-related processes such as word-initial aspi­ra­ tion, word-initial glottalstop insertion, etc.

Nübling et al. (2006: 21) and Szczepa­niak (2007: 52–53, 330) also provide lists of SL/WL criteria. Szczepaniak (2007: 27–51) discusses all criteria at length.

54 

 Stig Eliasson

The selection of diagnostics in Table 1 leaves out rhythmical adaptation as well as the criterion of vowel harmony, which Auer (2001: 1397–1398) considers inconclusive. The seven diagnostic criteria of Table 1 will be applied to Swedish in turn. Because of the special importance of the phenomenon, a separate section (section 5) will be devoted to natural syllabification, which is one of the processes under point 7 in Table 1.

3 Canonical forms of native Swedish morphemes First, however, a brief background remark on some points of Swedish phonotactics is in order. As we mentioned in section 2.6, if the minimal lexical units are morphemes, morphemes would seem to constitute a convenient domain for stating underlying phonotactic regularities. Beside special types of derivational or inflectional morphemes, Swedish possesses the two basic phonotactic types of free native or nativized root morphemes reproduced in (3) (Eliasson 2009: 59, 84).7 (3) (a) C03VC02(3) (b) C03VC12(3)v̆L

The dominant canonical form of native or nativized root morphemes (favored morpheme templates) is that of (3a) – that is, monosyllables (e.g. tag ‘grip’). Additionally, the language possesses a large array of native morphemes ending in unstressed -v̆l, -v̆r, -v̆n (3b), most of which have monosyllabic allomorphs in word-internal prevocalic contexts due to syncope (i.e. /C03VC12(3)L-/ as in tagl-et horsehair-def.sg < tagel ‘horsehair’ and similar non-compound word forms). Their underlying forms have to be bi- rather than monosyllabic, however, for if we were to posit monosyllabic bases, we could not predict what vowel (/e/, /a/, /o/, /u/ or / ʉ/) to insert in certain bisyllabic allomorphs (gammal ‘old’, morgon ‘morning’, purpur ‘purple’ (n.), etc.) (Eliasson 1972: 183–186). Still another class of examples consists of normally bound /C03VC12(3)L-/ root morphs such as /famml-/ in faml-a-Ø fumble-vfs-inf ‘fumble’. If these are set up as monosyllables, they will require a new phonotactic template in addition to (3a) and (3b). If, on the other hand, they are assumed to contain a pre-L vowel underlyingly (i.e. have

7 In the formulas, C03 means “from zero to three consonants” (and correspondingly for other numerical sub- and superscripts), v̆ is an unstressed, usually fugitive vowel, and L stands for /l/, /r/, or /n/. In (2b), the last or only medial consonant cannot be l, r, n if followed by v̆l, v̆r, v̆n, respectively.



The typology of syllable and word languages and Swedish phonological structure 

 55

the form /C03VC12(3)eL/), they will align with group (3b), although their fugitive vowel will not surface phonetically, since the root morphemes in question do not happen to occur in the environments where the unstable vowel would stay.8 In regard to consonant phonotactics, the phonemes /h/, /ɕ/, and, in most varieties, /ɧ/ are restricted to morpheme- or syllable-initial position. Post-vocalically, two-member consonant clusters in both (3a) and (3b) include sequences of two identical consonants (tagg ‘prickle’, traggel ‘plodding’). Most, though not all, post-vocalic three-consonant clusters in morphemes are of foreign origin (cf. Eliasson 2009: 74–78 for free morphemes in ‑CCC), hence the parentheses in the formula. We should also mention that, while high long vowels are often slightly diphthongized phonetically ([i:j, y:ɥ, ʉ:ß, u:w]), no native morpheme has a nucleus with two contiguous vowel phonemes or a diphthong. Hence, paus ‘pause’, whose vocalic nucleus may be pronounced either as a succession of two vowels or as a diphthong (Hedelin 1997: 862; Garlén 2003: 568), is a straightforward foreign morpheme (cf. Tamm 1887: 4–5). Finally, borrowed morphemes and borrowed, originally polymorphemic words reinterpreted by Swedish speakers as single morphemes frequently violate the native canonical patterns in regard to both consonant and vowel combinatorics.

4 The SL/WL criteria applied to Swedish 4.1 Word stress, tonal accent Let us now turn to the application of the SL/WL criteria mentioned in Table 1 to Swedish, beginning with stress and tone. Auer (2001: 1395) observes that “the culminative emphasis of one and only one syllable (or mora) over all others within a unit of the size of a phonological word […] is typical for word languages” and that this function “can be fulfilled both by a dynamic and by a musical accent” (translation SE).

8 Should need arise, the vowel will show up, though, as in the expression svammel och fammel ‘rambling and fumbling’, retrieved from the internet and which according to its author was created by him (http://gunnarstrandberg.wordpress.com/2012/03/04/646-svammel-och-fammel/; accessed Sep. 10, 2012) (cf. also that SAOB, vol. 8 under the entry famla records the occasional creation of the form fammel). Additionally, we may note that it is unnecessary to postulate monosyllabic underlying forms of certain kinds of ‑eL morphemes in order to prevent tone assignment (see section 4.1 and cf. Wetterlin 2010: 53–54, 71, etc.), since the absence of tone there can largely be predicted by other criteria (Eliasson 2012a).

56 

 Stig Eliasson

Disregarding differences of detail, dynamic stress in Swedish is distributed similarly to stress in German.9 In stark contrast to German, however, Swedish word stress may be coupled with tonal accent (“grave accent”, “accent 2”).10 Like main stress (with marginal exceptions), tonal accent is assigned only once per word. Quite commonly, it signals that two meaningful elements, both of which must contain at least one vowel, belong to the same grammatical word, or, put differently, it points to the presence of a morpheme boundary within the word. Thus, many native vocalic inflectional and derivational suffixes like the verb-forming suffix (vfs) ‑a and the noun-forming suffix (nfs) ‑e induce tonal accent, as in fiska-Ø [xfis·ka] fish-vfs-inf ‘fish’ and fisk-e [xfis·kə] fish-nfs ‘fishing’, both from the root noun fisk [ˈfis·k] ‘fish’.11 The items fiska and fiske are simplexes (non-compound words). But the great majority of all compounds also receive tonal accent as illustrated in (4) (boxes indicate primary- or secondary-stressed syllables, superscript lines specify the domain of tonal accent, as described by Elert 1970: 44; ÷ marks a compound boundary). (4) (a) σσ σσσσσ (b) σσσσσ σσσσσ σσσσσσσ σσσσσσσσ

fisk÷död fisk÷restaurang förståelse÷prov fiskeri÷näring kondoleans÷skrivelse kommunikations÷medier

‘fish death’ ‘fish restaurant’ ‘comprehension test’ ‘fishing industry’ ‘[official] letter of condolence’ ‘communication media’

Compounds with anacrusis as in (4b) show, moreover, that in the total vocabulary of Swedish, native as well as borrowed, neither stress nor tonal accent inherently indicates word edges, but rather joins two different syllables somewhere within a grammatical word. In the contemporary language, then, both phenomena are primarily culminative, not delimitative. The occurrence of tonal accent is not an infallible indicator of polymorphemicity, though. In fact, numerous tonal words exist that are monomorphemic, witness alun ‘alum’, biskop ‘bishop’, koppar ‘copper’, kummin ‘caraway’, sammet

9 Lindqvist (2007: 116­–143) and, more summarily, Garlén (1988: 132–138) review the distribu­tion of Swedish stress. 10 Words that lack tonal accent (traditionally called “acute accent” or “accent 1” words) carry the unmarked sentence intonation. Wetterlin (2010), who retains the traditional assumption of two word accents, argues that accent 1 is the marked or lexically specified accent. For a brief assessment, see Eliasson (2012b). 11 The phonetic transcrip­tion in this paper retains the older IPA usage of a raised cross (x) to symbol­ize main stress plus tonal accent (accent 2).



The typology of syllable and word languages and Swedish phonological structure 

 57

‘velvet’, pilgrim ‘pilgrim’, härold ‘herald’, and so forth. These items have the stress pattern of simplexes, but it is also possible to find monomorphemic polysyllabic word stems or words that take the stress characteristic of compounds. Thus, ingefär-a ginger-nfs ‘ginger’ (< Middle Low German ingever < Old French gingebre) is pronounced [xiŋ:əˌfæ:ra], as if it consisted of two, in reality fictitious, elements *ing-e ?-suf plus *fär-a ?-nfs. The monomorphemic ananas ‘pineapple’ (< Portuguese < Tupi-Guaraní) is most often rendered as [xan:anas(:)] like a compound (also [ˈan:anas]), yet Swedish provides no suitable independent correlates *ann-a and *nass. The similarly monomorphemic albatross [ˌalbaˈtrɔs:] ‘albatross’ (< English, ultimately < Arabic) has the alternative pronunciation [xal·baˌtrɔs:], but neither the existing alb-a ‘(minister’s) alb’, nor tross ‘hawser’ could help unmask it as a compound. In Prosodic Phonology, words of this quasi-compound type would be taken to demonstrate the need for the notion “phonological word”, separate from the grammatical or morphological word; however, in comparison to the veritable mass of real compounds the group of pseudo-compounds is not overly large and its members might to a considerable extent be identified by phonotactics in combination with some lexical marking.12

4.2 Syllable structure in different positions in the word According to Auer (2001: 1396), stressed syllables in word languages tend to be phonotactically more complex than unstressed ones, whereas syllable languages have syllables that are uniform throughout the word (see also section 2.4). In Swedish, stressed-syllable complexity results from three facts: (a) stress in the native vocabulary is routinely assigned to root morphemes as well as many derivational prefixes; (b) root morphemes and stressable prefixes tend to be phonotactically more elaborate than inflectional suffixes; and (c) rule-assigned stress in turn conditions other phonological processes such as quantity, low-level phonetic diphthongization, syncope, and so forth.13 From the point of view of process

12 A much more intricate prosodic regularity that for reasons of space cannot be treated here involves set lexicalized phrases such as Vita huset [ˌvi:taˈhʉ:sət] ‘the White House’, where the first constituent is partly destressed and also loses its tonal accent. Such phrases have the appearance of a word at the phonological level, even though they possess an obvious syntactic structure. Cf. also phrasal verbs like titta fram [ˌtit:a ˈfram:] ‘peep out’, where other words may intervene between the two constituents. 13 Note in particular that quantity ‒ although it is from a perceptual point of view a marker of stress ‒ does not condition, govern or trigger stress. In other words, stress in Swedish is not quantity-sensitive, as is sometimes suggested (e.g. Hulst 2010: 443, 444, drawing on other work).

58 

 Stig Eliasson

phonology, therefore, the ultimate reason for the surface, stressed-syllable complexity in Swedish is actually rooted in diversified morpheme phonotactics and stress assignment. This also means, by the way, that the extra mora, sometimes thought to characterize stressed as opposed to unstressed syllables in Swedish, is not an autonomous unit in the language, since it largely reflects stress-derived surface length.14 Specifically, the additional mora of a long vowel and the mora of a half-long consonant after a short stressed vowel merely represent the varying insertion points of lengthening triggered by stress (Eliasson 2001: 45, 2010a: 12). Consequently, the mora does not form an independent part of a putative prosodic hierarchy for Swedish. Like stressed syllables, word-edge syllables of word languages are said to exhibit more complex margins than other syllables. For left-edge syllables in Swedish, this is certainly often true (section 3). Further­more, we recall that some free morphemes have fairly elaborate right edges, as illustrated in (5a). (5) (a) blixt /blikst/ [ˈblik·st] ‘flash of lightning’ (b) hingst /hiŋst/ [ˈhiŋ·st] ‘stallion’ lärft /lɛrft/ [ˈlær·ft] ‘linen’ räfst /rɛfst/ [ˈrɛf·st] ‘inquisition’

blixt-en blixt-ar hingst-en hingst-ar lärft-en lärft-er räfst-en räfst-er

On the other hand, very many right word edges (in the traditional sense) in Swedish are quite simple as shown, for instance, by the suffixation of the definite article ‑(e)n and the plural endings ‑ar, ‑er in the examples in (5b), where the complex clusters are accommodated word medially. Thus, rather than word edges, complex post-vocalic clusters in non-compound Swedish words fairly often constitute the edges of root morphemes, while in numerous other cases they involve regular suffixation as in tänk-t(-a) think-pst.part(-pl) ‘imagined (thing(s), etc.)’, läng-st(-e) long-superl(-masc.def) ‘(the) longest (one)’, etc. Much less commonly, they are embedded intervocalically within bisyllabic morphemes such as bolster ‘feather bed’ and vänster ‘left’, within a partly opacated or semi-morphemic item such as farstu ‘vestibule’ (contracted from an older clearly composite form), and in unquestionably borrowed morphemes like kvalster ‘mite’ (< Low German qualster) or hamster ‘hamster’ (< High German Hamster) (cf. Eliasson 2009: 80). In formations such as those in (5) and in words like tänkt(a), the morpheme-edge or internal-boundary marking effect of cluster complex­ity seems as strong as its word-edge marking effect. A further issue arises with respect to the popular premise that the syllable is a basic phonotactic unit (section 2.6). Consider, for instance, the intervocalic

14 Cf. also Eliasson (2001: 40–45) on the problematic status of the mora in Faroese.



The typology of syllable and word languages and Swedish phonological structure 

 59

clusters in (6) and the phonetic syllabifications most favored by the subjects in a perceptual pilot test carried out by Gårding (1967) (examples from Gårding 1967: 26, all with the verb-forming suffix ‑a; phonological and phonetic transcriptions by SE; + morpheme boundary, . syllable boundary). (6) -ksl- axla /aksel+a/ [xak·.sla] -spl- haspla /haspel+a/ [xhas·.pla] -rpl- sörpla /sørpel+a/ [xsœ̞r·.pla] -ndl- handla /handel+a/ [xhan·.dla] -kstr- blixtra /blikst+r+a/ [xblik·.stra]

‘shoulder’ (v.) ‘reel’ (v.) ‘slurp’ (v.) ‘shop’ (v.) ‘flash’ (v.)

As these syllabifications demonstrate, complex phonetic clusters like -ksl-, -spl-, -kstr- do not constitute syllable margins, because they are in fact phonetically heterosyllabic, i.e. they straddle, rather than are contained within, syllable boundaries. Instead of complex syllable structure, then, this is a matter of complex allomorph margins (section 3; cf. also 4.6).15 Furthermore, the fact that subjects often vary their syllabifications is one of several arguments suggesting that syllable structure is not specified underlyingly in the lexicon (cf. 2.6). Given these considerations, the morpheme would seem to be the natural domain of underlying phonotactic generalizations in Swedish.

4.3 Geminate vs. ambisyllabic consonants As conceived in SL/WL typology, syllable languages may possess geminates, and word languages ambisyllabic consonants, but, with the exception of certain geminates in word languages, not vice versa. Auer (2001: 1397) states that “[a]s a consequence of the clearly articulated syllable structure, in syllable languages no ambisyllabic elements are to be expected; these are there often matched by geminates” (translation SE). Further, Nübling and Schrambke (2004: 284) remark that whereas geminates may be present in syllable languages, word languages suffer “geminate reduction ‒ except where morphologically relevant, e.g. at a compound boundary” (translation SE). Finally, because ambisyllabic consonants involve diffuse syllable boundaries, they are seen as signaling the cohesion of phonological words, hence they constitute a diagnostic of word languages. The application of the geminateness vs. ambisyllabicity criterion to Swedish carries with it an interesting twist. Phonetically, Swedish has long ambisyllabic

15 Nor, incidentally, do they signal right margins of grammatical words, as they cannot occur there; *aksl, *haspl, *handl as independent words are ungrammatical.

60 

 Stig Eliasson

consonants in words such as illa [xil:a] ‘badly’, pappa [xpap:a] ‘dad’, etc. In this respect, Swedish behaves like a word language. Yet, if we want to account for deeper phonological structure, the picture is more complex. Compare the items in (7a) and (7b). (7)

(a)

(b)

ω σs V

V

[i: l a] /il+a/ ‘hasten’

V

ω

σw

σs

σw C

(c)

ω

C

σs V

[i l: a] /ill+a/ ‘badly’

V

σw C C

V

[i l: a] /ill+a/ ‘badly’

Under the analysis in (7a, b), the short intervocalic consonant of ila (7a) goes with the second syllable, whereas the long intervocalic consonant of illa (7b) is shared by the two syllables. This is precisely as expected in a word language. Nevertheless, beside the special segment-internal syllabification in (7b), distribu­ tional patterns as well as morphophonemic parallels indicate that most fully long consonants in Swedish are actually better treated as underlying geminates (Eliasson 2010a: 10–11).16 We may, therefore, replace the tree of (7b) by that of (7c), where [l:] actually equals [l·l] in that the half-length of [l·] after a short stressed vowel parallels half-length in intervocalic clusters generally, as in spilta [xspil·ta] ‘stall (of horse)’, bjälke [xbjɛl·kə] ‘beam’, etc. Hence, underlying geminate structure and phonetic ambisyllabicity (in our case segment-internal syllabification of phonetically long consonants) do not constitute diametrical opposites, but can coexist in a language. Moreover, this fact should be taken into account in the typological classification of the language.

4.4 Word-external vs. word-internal sandhi According to the SL/WL typology, outer and inner sandhi should match in syllable languages, but tend to diverge in word languages. Szczepaniak (2007) notes that “the similarity of word- and sentence-sandhi rules is […] to be expected only in syllable languages” and that the equation “sentence-sandhi = word-sandhi” is

16 The long consonants in items such as kort ['kɔʈ:] ‘short’, korta [xkɔʈ:a] ‘shorten’ and kors ['kɔʂ:] ‘cross’ (n.), korsa [xkɔʂ:a] ‘cross’ (v.) derive from phonological r plus dental combinations, however (see section 4.4). The velar nasal /ŋ/ after a stressed vowel is also always long without necessarily having to be interpreted as an underlying geminate.



The typology of syllable and word languages and Swedish phonological structure 

 61

a characteristic of syllable languages (Szczepaniak 2007: 36 and 52, respectively; translation SE). Szczepaniak illustrates the point more fully (translation SE): Assimilations that facilitate pronunciation appear in syllable languages both wordinternally (so-called word-sandhi) and at word boundaries (so-called sentence-sandhi): in Spanish, n before g is velarized throughout, for instance, ven.ga [beŋ.ga] (word-sandhi) and con ganas [koŋ.ga.nas] (sentence-sandhi). In German, a similar assimilation is found wordinternally, as in Anker [aŋ.kɐ]. At word boundaries, it is blocked, whereby these remain easily recognizable, for instance, in Kopf an Kopf [kɔp͜f. ʔan.kɔp͜f ]. (Szczepaniak 2009: 51)

In Swedish, the most prominent sandhi process is the typologically unusual, recursive rule of supradentalization, also called postalveolarization or “retroflection” (Eliasson 1986: 277–294, 2000: 39­­–41; Braunmüller 2007: 39–41; Lindqvist 2007: 58–62; Riad 2010). Essentially, this rule turns a dental after a single /r/ or a supradental ([ʈ ɖ ɳ ʂ ɭ]) into a supradental, while the conditioning /r/ drops. The process hardly ever affects dentals after a double r (/rr/). The effects of the rule are illustrated in (8), where the maximal number of possible assimilations have been indicated ( word-internal,  word-external sandhi). (8) (a) Kör så snar-t som Lars säg-er till! ['ɕœ̞:ʂo·'snɑ:ʈʂɔm'lɑ:ʂ:ɛjə'ʈil:]                drive as soon-adv as Lars say-prs particle ‘Go as soon as Lars tells us!’ (b) Bör du inte för-st let-a under stol-ar-n-a? ['bœ̞:ɖʉ̞xin·tə'fœ̞ʂ·ʈxɭe:taѳndəxʂʈu:laɳa]                                         should you not first look-vfs under chair-pl-def-pl  ‘Shouldn’t you first look under the chairs?’

While highly productive, supradentalization varies a great deal as to its degree of obligatoriness. Several different parameters are involved in determining its variability, such as the kind of r-plus-dental cluster (e.g. the assimilation of /rl/ to [ɭ] is always optional, and the assimilation never proceeds beyond the [ɭ] to a following dental), presence and nature of grammatical boundary, stress, tonal accent, speech rate, pausation, sociolinguistic variables, etc. (Eliasson 1986: 281–282).17 Morpheme-internally, supradentalization of /rt, rd, rs, rn/ is nearly always obligatory (Lars ['lɑ:ʂ] (name), not *['lɑ:rs], in Central Swedish; exceptions: urdu ['ɵr·dɵ] ‘Urdu’, etc.), across word-internal morpheme boundaries it is less compulsory (normally stolarna [xstu:laɳa] ‘the chairs’, but [xstu:larna] seems passable), and across word boundaries it is least mandatory (snart som ['snɑ:ʈ ʂɔm] or ['snɑ:ʈ sɔm] ‘soon as’). These differing degrees of optionality

17 On the variability of /rs/-sandhi in northern Swedish, see Lindblad (1999).

62 

 Stig Eliasson

aside, from a purely structural point of view word-external supradentalization in Swedish actually replicates word-internal assimilation. For instance, in (8a) the assimilation of /r#t/ (# = word boundary) in the phrasal verb säger till parallels that of /r+t/ in snart, and in (8b) the assimilation of /r#st/ in the prepositional phrase under stolarna recalls that of /r(+)st/ in först. Furthermore, when the trigger /r/ is word-final, supradentalization often results in resyllabification. For instance, when fully assimilated, the phonological r#dental combinations in kör så, säger till, bör du, and under stolarna in (8) are resyllabified as [.ʂ .ʈ .ɖ .ʂʈ], respectively (see also section 5). Hence, since the process is not confined to phonological words, but may operate across phonological word boundaries and may also result in resyllabification, it constitutes a syllable-language trait in Swedish (Nübling and Schrambke 2004: 288). By the same token, it obscures rather than highlights the division of utterances into words, making Central Swedish in this respect harder to perceive and process than Standard Finland Swedish, which lacks supradentalization. Another productive assimilation rule that sometimes applies across word boundaries is nasal assimilation, frequently turning /n/ into [ŋ] before a velar stop not only in compounds such as ban-gård (lit. ‘track-yard’) ‘railroad station, depot’ and pann-kak-a pan-cake-nfs ‘pancake’, but also in word combinations like en korg ‘a basket’, hon kom ‘she came’, and occasionally even in en man går ‘a man goes’, and similar expressions.

4.5 Vowel inventories in accented vs. unaccented position In section 4.2, we commented on the phonotactics of syllables in various positions in the word. A related question concerns which selection of vowel phonemes appears in different word positions. Whereas syllable languages favor identical vowel inventories in all environments regardless of accent, word languages tend to employ more diversified vowel systems in accented than in unaccented position. Accented syllables in word languages may, for example, harbor phonemic oppositions between long and short vowels, while unaccented syllables may involve allophonic centralization, schwa vowels, and so forth (Auer 2001: 1396). As far as Swedish is concerned, it is convenient to distinguish the matters of (a) underlying vowels, (b) phonetic length in vowels, and (c) low-level phonetic centralization of short stressed and unstressed vowels. The maximal system of nine vowel phonemes appears in Swedish in stressed position, whereas the vowel phoneme inventory in other environments is reduced, as indicated in Table 2.



The typology of syllable and word languages and Swedish phonological structure 

 63

Table 2: Vowel phonemes in different positions in native and borrowed, mono- and polymorphemic, non-compound Swedish words Pre-stress position i e a u

Stress position

Post-stress position 1

Post-stress position 2

i

i

i

e

e

e

a

a

a

e e a o

o

u

u y

y

y

ø

ø

u

u

u

Post-stress position 3

u

u

Given these distributional patterns, a /y/ in a non-compound word helps signal pre-stress, stress or post-stress 1 position as opposed to post-stress positions 2 and 3; an /ø/ signals the pre-stress or stress position as opposed to all post-stress positions; and /i/ and /u/ the first four positions to the exclusion of post-stress position 3. Nevertheless, the degree to which the positional dependencies of vocalic subsystems aids the listener’s comprehension of the speech signal may in practice be somewhat limited, at least as compared to the role of vowel quantity. To see the contribution of vowel quantity to structure recovery, we need to separate three situations: (i) stressed environments that induce vowel lengthening (vis /vis/ [ˈvi:s] ‘wise’), (ii) stressed environments that induce no vowel length (viss /viss/ [ˈvis:] ‘certain’), and (iii) unstressed environments (as the first syllable in visir /visir/ [viˈsi:r] ‘visier; vizor’). The fact that long vowels show up only in case (i), but not elsewhere, does not mean that a vowel length opposition has been neutralized in situations (ii) and (iii), however. Rather, the reverse is true. No minimal phonetic contrasts between long and short vowel phones are ever possible in the language – that is, minimal pairs of the type [ˈvi:s] : *[ˈvis] do not exist, and there are consequently no underlying long vowels that can fill the V position in the formulas (3a) and (3b) above. The long vowel phones arise, instead, in well-defined phonotactic environments through a stress-triggered lengthening rule (Eliasson and La Pelle 1973: 139, 147; Eliasson 1985: 114–115, 120, 2010a: 12–13). Moreover, in contrast e.g. to Faroese, which ties in vocalic quantity with syllable boundaries (Eliasson 2001: 45), stress-induced vowel lengthening in Swedish is often contingent on morpheme boundaries; witness skjuts [ˈɧʉ:ts] shoot-pass ‘is shot’ < bimorphemic /ɧʉt+s/ as opposed to skjuts [ˈɧɵt·s] ‘lift’ (n.)

64 

 Stig Eliasson

< mono­morphemic /ɧʉts/, where length is inserted in the post-vocalic consonant (Eliasson 2010a: 15–17). Thus, the alternation in speech of long stressed vowels (case (i)) and short unstressed vowels (case (iii)) does help configure syllable sequences into phonological words. However, within the stressed syllables themselves, vowel length frequently signals that a morpheme boundary occurs in a following post-vocalic consonant cluster (Eliasson 2010a: 17–19, 47) – that is, it may be said to make direct reference to a morphological category without any obvious need to go via the prosodic word (cf. 2.2, 2.5). Finally, centralization of unstressed as opposed to stressed vowels is seen as a characteristic of word languages (Auer 2001: 1396). In Swedish (like in German), short vowel allophones are centralized regardless of whether they are stressed or unstressed. The centralization of the short allophone of the /ʉ/ phoneme into a mid central vowel [ɵ] is extreme, and an unstressed phonetic schwa-like vowel, usually transcribed [ə], exists in the language.18 In sum, the characteristics of Swedish mentioned in this section are fairly typical WL features. The account of Swedish vowel quantity, moreover, can be given directly in terms of morphemes rather than indirectly via the prosodic word as in Prosodic Phonology.

4.6 Vowel elision and consonant insertions In section 3 we saw that most vowel/zero alterna­tions in modern Swedish may be understood as resulting from syncope. Typical examples are given in (9a) and (9b), listing unsyncopated singular and syncopated plural forms, respectively. (9) (a) axel /aksel/ [ˈak·səl] (b) axlar /aksel+ar/ [xak·slar] ‘shoulder(s)’ aster /aster/ [ˈas·tər] astrar /aster+ar/ [xas·trar] ‘aster(s)’ braxen /braksen/ [ˈbrak·sən] braxnar /braksen+ar/ [xbrak·snar] ‘bream(s)’

Like the analogous cases in (6), the three-member intervocalic consonant clusters of the plural forms in (9b) fail to mark word edges, but do, with a fairly high probability, suggest morpheme edges (main exception: the word type bind-l-ar bind-nfs-pl ‘bandages; armlets’ < bind-el, cf. bind-a-Ø bind-vfs-inf ‘bind’).

18 The vowel phoneme /ʉ/ derives from a complex chain shift (Eliasson 2010b; but see also Braunmüller 2003). On Swedish schwa, see Lyttkens and Wulff (1916: 105) and Hellberg (1974: 177–123) and for its nota­tion Lind­qvist (2007: 83–84). I do not enter here upon the question of whether or not schwa formation should be viewed as separate from centralization.



The typology of syllable and word languages and Swedish phonological structure 

 65

Turning to consonant insertions, these often serve in languages generally to prevent hiatus. Hiatus fillers occur regularly, e.g. in Turkish where the buffer consonants ‑n- and ‑y- separate vocalic case endings from stems as in köprün-ün bridge-n-gen ‘of (the) bridge’ and köprü-y-e bridge-y-dat ‘to (the) bridge’, or in Basque where an epenthetic ‑r- surfaces before certain vocalic endings as in zubi-a-r-en bridge-def-r-gen ‘of the bridge’ and zubi-a-r-i bridge-def-r-dat ‘to the bridge’. In Swedish, consonant insertions to ensure CV syllable structure are not prevalent. Two cases deserve mention here. On a very limited scale, /j/ may serve as an onset filler in instances such as [xni:jə] for nio [xni:ə, xni:u] ‘nine’, [xti:jə] for tio [xti:ə, xti:u] ‘ten’, and [xli:jə] for lie [xli:ə] ‘scythe’ (note also [xnij:ə], [xtij:ə], [xlij:ə] with ambisyllabic [j:]) and the variant pronunciations [vijuˈli:n] for violin [viuˈli:n] ‘violin’ and [erɔˈpɛj:(i)ska ɵn(i)ˈju:nən] for Europeiska unionen [ɛurɔˈpe:iska ɵniˈu:nən] ‘European Union’. This epenthetic /j/ does not appear in aloe [xɑ:luə], [xɑ:lɔə] ‘aloe’, oboe [xo:bɔə] ‘oboe’, or gröe [xgrø:ə] (‘poa’, a kind of grass) with different first vowels. Also very restricted is the process whereby a /g/ is inserted after /ŋ/ in borrowed word formations of the type (10a) as compared to their base forms in (10b) (afs adjective-forming suffix). (10) (a) diftong-er-a [diftɔŋˈge:ra] ‘diphthongize’ (b) diftong [difˈtɔŋ:] ‘diphthong’ dipthong-vfs-vfs triangul-är [triaŋgʉ̞ˈlæ:r] ‘triangular’ triangel [triˈaŋ:əl] ‘triangle’ triangle-afs

Again, the inserted segment serves as a syllable-onset filler. In brief, if interpreted as deriving from syncope, vowel/zero alternations increase the complexity of the accented syllable and hence are a word-language property (cf. Auer 2001: 1396), while j- and g-insertion, which amend onsetless syllables, constitute a syllable-language feature (cf. Auer 2001: 1397).

4.7 Word-related phonological processes We now come to the last of the seven groups of criteria mentioned in Table 1: word-related as opposed to syllable-related phonological processes. A typical syllable-related process is resyllabification (i.e. the movement of a segment or segments from one syllable to another), which will be addressed in section 5 in connection with some comments on syllabification generally. Word-related processes, on the other hand, following Auer’s (2001: 1397) definition, are “phonological processes […] that only appear in a specific position in the phonological word (initially, medially or finally) and that are not accent-dependent” (transla-

66 

 Stig Eliasson

tion SE).19 The kinds of rule that he mentions are final devoicing, medial or final phonetic weakening, initial mutation, postaspiration of initial stops and initial glottal-stop insertion. Of these processes, the first three are missing in modern Swedish, but post­ aspiration of voiceless stops and glottal-stop insertion do exist. According to Garlén (1988: 68), the aspirated stops [pʰ tʰ ʈʰ kʰ] are distributed as in (11a), and the unaspirated [p= t= ʈ= k=] as in (11b).20 (11) (a) (b)

After pause, etc. Initially in word-elements In stressed syllable onset Before pause After homo-morphemic s Before voiceless C Before unstressed V21

topp [tʰ-] ‘top; peak’ träd-topp [tʰ-, -tʰ-] ‘tree-top’ topas [-ˈpʰ-] ‘topaz’ topp [-p:ʰ] ‘top; peak’ stopp [-t=-] ‘stop’ tops [-p·=-] ‘cotton bud’ topp-en [-p:=-] ‘the top’

The first three items in (11a) suggest that the aspiration of voiceless stops in Swedish highlights the beginning of a word or the onset of a syllable with primary or secondary stress, whereas the (articulatorily differing) aspiration in pre-pausal position of the [-p:ʰ] in topp is, obviously, of a different nature.22 The glottal stop in Swedish, which appears to be weaker than the North Standard German one, may like in many other languages occur “before vowels at the beginning of words or in forcible pronunciation” (Malmberg 1968: 75; translation SE). In highly, perhaps overly, emphatic pronunciation of items like those in (12), glottal stops might creep in, as indicated (tentative transcriptions: ( ) optional, (( )) optional and probably less likely).

19 Szczepaniak (personal communication) adds that word-related processes can also be accentdependent. 20 Examples are mine. The (non-IPA) superscript equal sign (=) indicates unaspirated consonants. Garlén’s categories “post-pausally”, “initially in word elements” (ordledsinitialt), and “initially in stressed syllable” overlap. Although Garlén does not illustrate this, initial voiceless stops in words such as trä(d) ‘tree’, klä(d) ‘dress’, etc. are also aspirated (Lyttkens and Wulff 1916: 267; Noreen 1903–1907: 403). Lindqvist (2007: 26) briefly contrasts Swedish and German aspiration, and Helgason and Ringen (2008) report on an instrumental investigation of aspiration in six Central Swedish speakers. 21 Unless the stop is morpheme initial or follows a pause. 22 Further examples of aspiration are (in my pronunciation) tatar [tʰaˈtʰɑ:r] ‘Tatar’, tatuera [tʰat=u̞ˈ((ˀ))e:ra] ‘tattoo (v.)’, tentativ [tʰɛnt=aˈtʰi:v], [ˈtʰɛn·t=at=iv] ‘tentative’, ratatouille [rat=aˈtʰuj:] ‘ratatouille’. Elert (1997: 28; translation SE) offers a different interpretation: “The aspirated stops occur above all before a stressed syllable, which often means at the beginning of a word. In other positions, the stops are unaspirated”.



 67

The typology of syllable and word languages and Swedish phonological structure 

(12) aort-a [(ˀ)ax((ˀ))ɔʈ:a] ‘aorta’ aerogram [(ˀ)a((ˀ))erɔˈgram:], [(ˀ)ærɔ …] ‘air letter, aerogram’ eon [(ˀ)eˈ ((ˀ))o:n] ‘eon’ kaot-isk [khaˈ ((ˀ))u:tisk], [khaˈ ((ˀ))o:tisk] ‘chaotic’ (cf. kaos [ˈkhɑ:us] ‘chaos’) oas-er [(ˀ)uˈ((ˀ))ɑ:sər] ‘oases’ residu-um [rɛˈsi:d(ʉ̞ ((ˀ)))ɵm] ‘residue’ (def. residu-et) vakuum [ˈvɑ:k(ʉ̞ ((ˀ)))ɵm] ‘vacuum’ (def. vakuum-et)

The examples seem to suggest that the glottal stop in Swedish does not just serve as a word boundary marker, but is perhaps also related to otherwise empty syllable onsets. In conclusion, aspiration marks word onsets and signals stress, while glottal-stop insertion may primarily, though not exclusively, be a word boundary marker.23

5 Syllabification Let us proceed to the process of syllabification. As described in section 2.8, syllabification is a feature of speech production that adjusts lexical forms to the exigencies of articulation and frequently leads to incongruities between morphosyntactic and syllabic boundaries. (13) illustrates typical outcomes for Swedish (shading marks discrepancies, clear boxes indicate where morpheme and syllable boundaries actually coincide). (13) (a)

f

i

s

k

. ‘fishing industry’ (b)

m o

n

o

ö

v e .

‘agreement’

r .

+

r

+

.

+ .

. ‘monopolize’

(c)

e

+

e

p

o

i

l

s

+ .

s

t

n

ä

r

s .

ä

m

m .

+

+

i

+

.

i

+

. n

+ .

e

ng

r

+

. e

l

s .

+

a

e

23 Underlying segments that typically signal syllable onsets are /ɧ/ (in many varieties of Swedish), /ɕ/ and /h/ (section 3).

68 

 Stig Eliasson

Disregarding a largely opacated morphological boundary in en(-)s in (13c), each example contains four morpheme boundaries, but only one single morpheme boundary in each item happens to coincide with a syllable boundary. Particularly striking is the syllabification of supradentals. While the nonsupra­dentalizing South Swedish and Standard Finland Swedish syllabify words of the type tå-r-n-a toe-pl-def-pl ‘the toes’ on the pattern VC.CV as [ˈtʰɛo:ʁ.na] and [ˈt=o:r.na], respectively, Central Swedish furnishes the V.CV syllabification [ˈtʰo:. ɳa] in line with syllable divisions considered typical of syllable languages. Moreover, this kind of syllable division extends to supradentals across word boundaries and even across the boundaries between different rhythmic groups, as in the case of the last two supradentals [ʂ] and [ɭ] in (14) (the borders of rhythmic groups are indicated by |; cf. also the analogous examples in (8) in 4.4). (14) Detta händ-er under december-dag-ar, när sol-en stå-r låg-t. This happen-prs during December-day-pl, when sun-def stand-prs low-adv Detta|händer under de|cemberdagar, nä|r solen(|) stå|r lågt. [xdɛt·.ta.ˈhɛn·.də.rɵn.də.ɖɛ.xsɛm·.bɛ.ˌɖɑ:.ga.ɳæ.ˈʂu:.lən.(ˈ)sto(·).ˈɭo:kt] ‘This happens on days in December, when the sun is low.’

Furthermore, like in many other languages, the potential contrast between the syllabifica­tions of phonemically dissimilar word pairs in Swedish is often blurred or even neutralized. A typical instance is (15) (facultative glottal stops not marked in the transcriptions). (15) (a) Plan-en av-vis-a-s. [xɑ:v.ˌvi:.sas] plan-def off-show-vfs-pass (b) Plan-Ø-en av-is-a-s. [xɑ:v.ˌi:.sas] plane-pl-def off-ice-vfs-pass (c) Neutralized form: [xɑ:.ˌvi:.sas]

‘The plan is rejected.’ ‘The aircraft are being de-iced.’ ‘is rejected’ / ‘is being de-iced’

As Gårding observes: Syllabic juncture tends to be ambiguous. […] [T]he recognition of juncture plays a small role in the language as a system of communication. There are very few pairs in which the sequences are also grammatically identical. The grammatical context, then, in addition to the semantic context will make acoustical juncture cues fairly redundant. This makes it possible for a speaker to disregard word- and morpheme-boundaries when he speaks fast and re-divide his speech into syllables conforming to the natural syllabification rules, i.e., to his habits of coordinating articulatory movements. (Gårding 1967: 66; italics added)

Consequently, even in a language such as Swedish with many WL characteristics, syllabification in fast, casual speech often proceeds in ways characteristic of syllable languages. Still, differences between the two language types do exist, as



The typology of syllable and word languages and Swedish phonological structure 

 69

frequently recognized e.g. in the teaching of French ‒ which is more SL-like ‒ to Swedes (Rankka, Barjac, and Axelsson 1978: 18–21). The deviations in Swedish from natural syllabification tendencies reflect a more “explicit” code (Jakobson and Halle 1956: 6), where grammatically determined syllabification is partly upheld (cf. Vennemann 1982: 299–302).

6 Recovery of grammatical boundaries by phonological means While syllabification is a chief agent that destroys border signals, numerous phonotactic and several rule-derived phonological phenomena concur in providing clues to the morphological composition of Swedish words.

6.1 Clues from morpheme structure As we saw in section 3, Swedish has two native canonical types of root morphemes: monovocalic C03VC02(3) and bivocalic C03VC12(3)v̆L (v̆ usually alternating with zero). In native root morphemes, any of the nine Swedish vowel phonemes can occupy the main V position, whereas the v̆ position, like simple affixes, pre­ sents a reduced subset of vowels (section 4.5). The size of the vowel sets, then, is partly correlated with phonotactic position in native bisyllabic morphemes and the division of words into roots and simple affixes. In addition, native morphemes never fill the V position with a vowel cluster or a diphthong underlyingly (section 3). Hence, when vowel combinations occur in the native vocabulary, this situation always entails an intervening morpheme boundary (/V+V/).24 Consonant combinability, moreover, is reasonably evolved in Swedish morphemes, not only before, but also to a certain extent after the vocalic nucleus. The combinatory possibilities in monovocalic morphemes with up to three consonants after the vowel are exemplified in (16).25

24 Extending Dressler, Dziubalska-Kołaczyk and Pestal’s (2010: 52) terminology, the combination /V+V/ would constitute a morphonotactic vowel cluster. These authors define a “morphonotactic consonant cluster” (italics added) as “a cluster which comes into being only through morphotactic operations”. Such clusters “facilitate morphological processing in perception”. 25 The structures CCCVCCC and VCCC, examples of which are missing in the chart, occur in rare surnames of partly uncertain origin such as Strands and Axt (cf. Strands attested e.g. in England and the German surname Axt). As onomastics constitutes no reliable guide to the normal phonotactic patterns of a language, such items are not included here.

70 

 Stig Eliasson (16) Canonical forms of monovocalic Swedish root morphemes with illustrative examples (rendered in phonological transcription [second column] and normal orthography) C C C V C C C C C V C C C b l i k s t blixt ‘flash of lightning’ C V C C C h i ŋ s t hingst ‘stallion’ V C C C C C C V C C s t r a k s strax ‘instantly’ C C V C C k r a k s krax ‘croaking’ (n.) C V C C k a s k kask ‘helmet’ V C C a s k ask ‘ash; box’ C C C V C s p r a k sprak ‘crackle’ (n.) C C V C p l ø s plös ‘tongue (of shoe)’ C V C l ø s lös ‘loose’ V C ø s ös ‘scoop!’ (imper.) C C C V s t r ø strö ‘litter, bedding’ C C V s l ø slö ‘blunt; indolent’ C V ɧ ø sjö ‘lake’ V ø ö ‘island’

The set of examples in (16) may be compared to typical root morphemes in a marked (C)V-syllable language such as the Polynesian (Austronesian) language Samoan in (17).26 (17) Canonical forms of typical native root morphemes in Samoan (in the second column in phonological transcription, in the third column in the variety of Samoan orthography that employs diacritics; the subscripts i j indicate dissimilar, the subscripts i i identical vowels) C V C V n o f o C V i V j n o a C Vi V i l a a V C V a n o V i V j a o V i V i a a

nofo noa lā ano ao ā

‘sit; dwell’ ‘tie, bind’ ‘sun’ ‘pay attention, listen’ ‘cloud’ ‘very; just’

Like the items in (17), the overwhelming majority of all native root morphemes in Samoan are phonologically bivocalic (C)V(C)V-structures.27 As we see, comparing (16) and (17), Swedish, in contrast to Samoan, tends to compress the phonological information of root morphemes into monovocalic units, not infrequently creating consonantal clusters both pre- and post-vocalically. The other major option is to

26 Further examples may be found e.g. in Milner (1966). 27 Native trivocalic morphemes such as aitu ‘ghost, spirit’, fanua ‘land’, and tagata ‘man, human being’ are uncommon.



The typology of syllable and word languages and Swedish phonological structure 

 71

utilize phonotactic building blocks corresponding to coda-carrying monovocalic units with the addition of -v̆L (section 3). The relevant parameters and the differences between the two languages are schematized in Table 3. Table 3: Phoneme inventory size, degree of consonant combinability, and syllabic/vocalic structure of underlying root morphemes in the native lexicons of Samoan and Swedish Structural property

Samoan

Swedish

(a) Phoneme inventory (i) Vowels (ii) Consonants

Small 5 V phonemes 10 (13) C phonemes

Medium-sized 9 V phonemes 18 C phonemes

(b) Consonant combinability

None at all

Medium

(c) Syllabic/vocalic struc­ture of underlying root morphemes

Bivocalic (“bimora­ic”) (most often correspond­ing to two phonetic syllables)

(i) Monosyllabic (ii) Bisyllabic (ending in ‑v̆L)

Table 3 seems to suggest some potential trade-offs between the variables (a) phoneme inventory size, (b) degree of consonant combinability, and (c) root morpheme length. Samoan has a small phoneme inventory (5 basic vowels and 10–13 consonants, three of which are mainly restricted to loanwords), zero consonant combinability, and a phonologically bivocalic minimum for root morphemes. Many Samoan root morphemes therefore surface phonetically as bisyllables, whose root morpheme status is hard to recognize in running speech on the basis of segmental information alone. Swedish, on the other hand, exhibits a much larger phoneme inventory, a reasonably high, though not extreme, propensity for clustering consonants,28 and a moderately limited morpheme length. Hence, in Swedish, the possibility of spotting and correctly identifying root morphemes in running speech on the basis of distinct segmental shape alone is relatively high (cf. lengthwise compressed, but segmentally diversified morpheme shapes like strö, sprak, strax, blixt, hingst). The question to what extent structural trade-offs occur in languages is, of course, still unresolved. Recently, Maddieson examined the belief that complexity in one subpart of a linguistic or phonological system may be compensated for

28 Braunmüller (1995: 11; translation SE) holds that “typologically with respect to its phonotactics, Swedish is still, if anything, similar to languages with predominantly simple patterns”.

72 

 Stig Eliasson

by greater simplicity in another subpart. In stark contrast to that view, he claims that “to a great degree complexity in various sub-systems co-occurs, rather than languages ‘compensating’ for complexity in one area with simplicity in another” (Maddieson 2011: 28). For example, in a cross-linguistic investigation of possible correlations between the simplicity or complexity of miscellaneous phonological parameters in a sample of 515 languages worldwide (Maddieson 2006, 2007), the parameter of consonant inventory size tended to correlate positively rather than negatively with syllable shell complexity. That is, the more consonants a language has, the more likely it is to have a complex syllable structure. The relation syllable shell complexity vs. root morpheme length, on the other hand, was not part of Maddieson’s (2006, 2007) study, but in a later contribution he notes: [T]here is some connection between the richness of a language’s phonological resources and the relative frequency of words of different lengths, with longer words being primarily associated with a smaller inventory of vowels, but also with simpler syllable structures. There is a hint of a compensatory relationship in this result, but it is not an across-the-board phenomenon: A smaller inventory of consonants does not predict longer words. (Maddieson 2011: 32; italics added)

Maddieson compares phonological resources to the number of words of varying lengths, not to canonical types of morphemes and their frequency, which might yield a different result. Nevertheless, the connection that he sees, even if subtle, appears to provide some partial typological support for the possible trade-off between segment inventory, syllable structure, and number of syllables in morphemes that we alluded to above with regard to Swedish and Samoan. This problem aside, however, it is clear that in SL/WL-typological terms (cf. Szczepaniak 2009: 49–50) Swedish is phonotactically much closer to a prototypical word language than to a syllable language and that the relatively elaborate shapes of root morphemes and certain derivational affixes are a powerful aid in localizing parts of the speech signal that are essential from a lexical-semantic point of view.

6.2 Clues from the outputs of rule applications Similarly, many features added to the speech signal through phonological – particularly prosodic (i.e. non-segmental) – rules are effective markers of grammatical structure in Swedish, as brought out by examples (18) and (19).29

29 Eliasson (2008) discusses more fully the general task of recovering underlying phonological structure in Swedish by means of prosody. On the role of vowel quantity, cf. also Raffelsiefen



The typology of syllable and word languages and Swedish phonological structure 

 73

(18) ital-ien-sk-kling-ande ital-suf-afs-sound-prs.part /itali+en+sk+kliŋ+ande/ [(ˀ)i.ta.li.x((ˀ))e:nsk.ˌkhliŋ:.an.də] ‘Italian-sounding’ (19) Pingst-lov in-föll den veck-a-n. Whitsuntide-holiday in-fell that week-nfs-def /piŋst+lov#inn+føll#dɛnn#vekk+a+n/ [xphiŋ·st.ˌlo:v.x(ˀ)in:.ˌfœl:.'dɛn:.xvek:an] ‘The holidays at Whitsuntide took place that week.’

The prosody-aided, phonotactically-supported recovery of boundaries in example (18) is displayed graphically in Figure 4. Stress/tonal accent i t a l i e : n + s k + k l i ng + a n d e V length

C cluster Syllable boundary



Figure 4: Recovery of morpheme boundaries in a sample word (careful pronunciation, rendered in conventional spelling with added signs) on the basis of (a) stress pattern, (b) tonal accent, (c) quantity, and (d) phonotactics. Arrowheads point to recovered boundaries; continuous, dashed, and dotted arrows suggest successively lower degrees of reliability of clues. Only prosodically or phono­tactically recoverable boundaries shown; glottal stops not considered. Syllabification reinforces the division in one case.

In this example, the stress and tonal accent contours encompassing the fourth and the fifth vowel indicate that a morpheme boundary (+) might be hidden somewhere in the intervening consonant sequence. The extension of the sequence (four or five consonants in a row) confirms that suspicion. The length of the vowel [e:] before the complex cluster [nsk] furthermore shows that a morpheme boundary must intervene between [n] and [sk]. Since the [sk], in turn, cannot be accommodated with the following [khl] within a single morpheme (differently: skleros [skleˈro:s] the morpheme hinted at by the prosodic pattern Stress/tonal‘sclerosis’), accent Stress/tonal accent boundary Stress Stress/tonal accent p i ng s t + l o : v # (ˀ)i n + f ö l l # d e n # v e ck + a n (2007: 42–54). The items klingande in (18) and inföll in (19) are often pronounced with ambisyllabic [ŋ:] and [ɱ:], respectively. C cluster V length (+) Syllable boundary

Syllable boundary

74 

 Stig Eliasson

must fall precisely between these two cluster parts. The separation of /-sk/ and /kl-/ is further enhanced by a syllable boundary. Finally, in the latter part of the word, /ŋ/ before a full vowel has statistically a good chance of being followed by a grammatical boundary. The boundary recovery in sentence (19) is diagrammed in Figure 5. Stress/tonal accent

Stress/tonal accent Stress Stress/tonal accent

p i ng s t + l o : v # (ˀ)i n + f ö l l # d e n # v e ck + a n C cluster

V length (+)

Syllable boundary

Syllable boundary

Figure 5: Recovery of word and morpheme boundaries in a sample sentence (careful pronunciation) on the basis of (a) stress pattern, (b) tonal accent, (c) quantity, and (d) phonotactics. Only prosodically or phono­tactically recoverable boundaries shown. Syllable boundaries reinforce division in some cases.

Disentangling the details of the recovery diagrammed in Figure 5, we note that two types of grammatical boundaries are identified: traditional word boundaries (marked by #) and morpheme boundaries (+). First, the succession of stress and tonal accent units shows that there must be three word boundaries internal to the sentence, yielding four words. Since the sequence [xpiŋ·st.ˌlo:v.x(ˀ)in:.ˌfœl:] is pronounced with two occurrences of the stress pattern [ˈ ˌ] (main stress plus secondary stress) as well as with two tonal accents, the first such boundary must be placed at the juncture of the two prosodically marked domains, somewhere in the region of [v] and [i]. In addition, the fact that the /o/ in lov is long ([o:]) and the next vowel is not schwa (as in the morpheme sovel [ˈso:vəl] ‘meat’) points to a possible morpheme boundary at most one step away to the right, thus supporting the tentative word division just made. If, furthermore, the syllabification is [v.i] and/or the [i] of the prosodically demarcated sequence inföll is pronounced with a glottal stop (that is, [.ˀi]), then the word boundary definitely falls between [v] and [i]. Similarly, the second and the third word boundaries will, in view of the succession of stress and tonal patterns, fall uniquely in the places noted in Figure 5: inföll will be separated from den, and den from veckan. As for morpheme boundaries, the only place where the consonant cluster [‑ŋstl‑] in pingstlov can be split up on two phonotactically well-formed morphemes is between [t] and [l]. Likewise, the cluster [‑n:f‑] in inföll must on phonotactic grounds have an internal boundary. Moreover, the concluding sequence veckan is, because of its tonal



The typology of syllable and word languages and Swedish phonological structure 

 75

accent, likely to have a morpheme boundary between the [k:] and the [a]. Finally, syllabification enhances the identification of at least two grammatical boundaries. Hence, Figures 4 and 5 make explicit the remarkable, not to say astonishing, degree to which the type and location of grammatical boundaries in Swedish can be predicted by prosodic, phonotactic, and segmental properties, particularly when these various decoding aids operate jointly in context.

7 Swedish and SL/WL typology: Summary and final remarks This paper has provisionally examined selected features of the phonology of modern Swedish in the light of the typology of syllable and word languages (the SL/WL typology). We have only considered major tendencies in the phonological data, leaving details and exceptions aside. The main descriptive points summarized in section 7.1 concern (a) the organizing role of stress in the language, (b) varying syllable-shell complexity in stressed and unstressed position, (c) compatibility of underlying geminates and phonetic ambi­syllabicity, (d) structural uniformity of word-internal and word-external sandhi produced by the highly productive rule of supradentalization, (e) varying utilization of vowel phonemes in stressed and unstressed position, (f) WL-type vowel deletions and SL-type consonant insertions, (g) special kinds of word-related phonological processes, and (h) natural syllabification. We conclude in section 7.2 with a few brief remarks pertaining to indirect and direct reference to morphology, the domain of phonotactic generalizations, the chief grammatical units “profiled” by phonological means, and configurative function as a principal focus of the typology.

7.1 Descriptive/typological considerations Several phonological processes in Swedish are tied in with the presence or absence of stress, as illustrated in Figure 6.

76 

 Stig Eliasson

Stress

No stress

Tone assignment V lengthening

C lengthening. Hence, lack of vowel length

Diphthongization of long high Vs

Syncope

V centralization (including schwa formation in unstressed syllables)

Postaspiration of unvoiced stops Figure 6: Some synchronic phonological rules in Swedish partially conditioned by the presence or absence of stress (  related to stressed syllables, in unstressed position). Lengthening is for practical reasons diagrammed with two arrows in the figure, but is actually a single process that adds duration to vowels and consonants in mutually complementary contexts.

The assignment of tonal accent and length depends crucially on stress. Neither can occur in environments with lower phonetic degrees of stress than primary (ˈ) or secondary stress (ˌ) (4.1, 4.5). If vowels are long and high, they are also often slightly diphthongized (section 3). Aspiration is above all contingent on stress (4.7); apart from this it is mostly a word-initial boundary signal. Unstressed vowels side with short stressed ones by becoming centralized (4.5). Schwa formation, if in fact separate from centralization (4.5), likewise depends on the absence of stress, as does syncope (4.6). Consequently, while justified doubts may exist about the paramount relevance of word stress to phonological typology (cf. Auer 2001: 1398), the rule-assigned distinction “stress vs. non-stress” occupies a central place in the synchronic phonological structure of Swedish, co-conditioning directly or indirectly a number of other phonological phenomena. Shell complexity is more elaborate in stressed than in unstressed position. Complex post-vocalic clusters often arise through morphological processes, but may also constitute the right edges of morphemes, either word finally or word internally (4.2). Similarly, complex word-internal clusters after a stressed vowel whose last cluster-member is -l, -r, or -n commonly have a morpheme boundary to their right (e.g. bolstr-ar ‘feather beds’). The criterion of phonetic ambisyllabicity seems at first to qualify as an indubitable word-language feature, but when we look at the matter from a phonological



The typology of syllable and word languages and Swedish phonological structure 

 77

angle, weighty arguments make us conclude that most phonetically ambisyllabic consonants are actually underlyingly geminates or else arise from likewise bisegmental combinations of /r/ with voiceless dentals or from /ŋ/ (4.3). While phonetic geminates are a purported characteristic of syllable languages, the question arises how to evaluate phonological geminates within the SL/WL model. Contrary to expectations for a word language, word-external sandhi caused by the immensely productive supradentalization rule is structurally equal to the corresponding word-internal assimila­tions (4.4). The full set of nine vowel phonemes appears in the stressable position of a word, while the set of vowels in non-stressable positions tends to be reduced – sometimes less, sometimes more, depending on the distance from the stressed syllable (4.5). Long vowel allophones in stressed and short vowel allophones in stressed and unstressed position differ with respect to both quality and quantity. From a morphological point of view, vowel quantity is especially important, since it fairly regularly foreshadows a possible morpheme division in a following consonant cluster (the type skjuts ‘is shot’). A prominent WL-type vowel deletion in Swedish is the syncope of unstressed vowels before homomorphemic ‑l, ‑r, ‑n, while SL-type insertions of /j/ and in a few cases /g/ are less common (4.6). On the other hand, aspiration of voiceless stops is a pervasive low-level phonetic process that marks stressed syllable onsets, and by the same token often word onsets, since the stressed syllable is commonly the first syllable in the word (4.7). Glottal-stop insertion, too, often occurs before vowels at the beginning of words, but in creating CV syllables it also produces an SL effect. Syllable division in Swedish, which we treated separately in section 5, may respect and hence signal grammatical boundaries, but with increased speech tempo it deviates in the direction of natural phonetic syllabification. That is, the grammatically regulated positioning of syllable boundaries is fairly important in Swedish, but may be overridden by assimilations and natural syllabifying tendencies, as when VCi.+CiV structures (where i marks identical segments) turn into V.CV structures (the type [xɑ:v.ˌvi:.sa] > [xɑ:.ˌvi:.sa]), or when, under appropriate prosodic conditions, supradentalized r-plus-dental clusters turn into syllable onsets ([.ʈ], [.ɖ], [.ɳ], etc.). In terms of the criteria of Table 1 in section 2.9 above, then, the approximate SL/WL profile for Swedish is the one in Table 4.

78 

 Stig Eliasson

Table 4: SL and WL properties of Swedish in terms of the major criteria of Auer (2001) (gross, tentative estimate only) Diagnostic feature

Swedish SL WL

Main effect of phenomenon

1

Stress and tone

+

Stress profiles words and connects morphemes in compounds. Tone usually connects morphemes (but bi- or polysyllabic tonal morphemes do occur, posing counterexamples to the connective function of tone).

2

Position-dependent syllable structure

+

Cumulation of consonants in stressable (core) syllables of native root morphemes.

3

Geminate vs. ambi­syllabic conso­nants

+

+

Geminates underlyingly, but in the typical cases ambisyllabicity phonetically.

4

Sandhi: supradentalization and also nasal assimilation

+

5

Accented vs. unac­cented vowel inventory

+

Full set of vowel phonemes in stressable syllables in native root morphemes, partly reduced set in non-stressable position.

6

(a) V-deletions

+

Syncope increases surface complexity of codas in underlyingly bisyllabic root-morphemes.

(b) V- or C-insertion 7

Word/syllable-related phonological processes

Supradentalization not restricted to the phonological word. Occasionally also nasal assimilation proceeds beyond a word boundary.

Restricted j- and g-insertion secures CV structure in syllable onsets.

+ (+)

+

Postaspiration of stops often marks stressed syllable onsets, and hence often also word onsets. (Glottal-stop insertion may in addition guarantee CV-type onsets.) Syllabification gravitates towards the WL type, but in fast speech certain tendencies towards natural syllabification become noticeable.



The typology of syllable and word languages and Swedish phonological structure 

 79

The diagnostic criteria of the SL/WL model are not mutually weighted and there exists as yet no graded numerical scale for classifying languages in terms of the model. Moreover, the criteria relied on here might have been divided up differently or converted into more fine-grained features (cf. e.g. Nübling and Schrambke 2004: 284–285), and they might have been supplemented by still other measures. Nevertheless, it seems that, like several other languages that have been subjected to an analysis in SL/WL terms, Swedish emerges as having a mixture of wordand syllable-language features, and that numerous important properties place Swedish closer to the WL than to the SL end of that typological scale.

7.2 A concluding theoretical note The application of the SL/WL typology to Swedish also raises several more fundamental issues, most of which go back to assumptions in Prosodic Phonology (cf. 2.1–2.3, 2.5–2.6). One of these issues involves the idea of indirect as opposed to direct reference to morphosyntax. That is, may phonological rules be permitted to refer directly to morphosyntactic entities (morpheme, grammatical word, etc.), or should they only be allowed to address these entities indirectly via elements of the Prosodic Hierarchy (2.1, 2.5)? Strict adherence to the latter view is complicated by frequent overlaps of morphosyntactic and “prosodic” units. By Nespor’s (1999: 119) own admission, “[i]n most cases, in fact”, the constituents of Prosodic Phonology are isomorphic with traditional morphosyntactic constituents. Swedish, too, offers patterns at the morphosyntax-phonology interface where phonological and higher-level divisions diverge, e.g. pseudo-compounds of the type ingefära ‘ginger’ and certain other kinds of expressions (4.1). These patterns are definitely in the minority, however, and the great majority of cases admits direct reference without any difficulty. Another problem concerns the domain for expressing phonotactic regularities. If we loosen the stipulation that phonological rules can never refer to morphosyntactic structure directly and accept the premise that lexically specifiable phonotactic properties are part of lexical entries rather than of phonetic forms (2.6), we might ask whether, instead of the phonological word, foot, or syllable, the lexical morpheme could be the appropriate phonotactic domain in Swedish. This option would neatly bring out the existence of two basic phonotactic types of native Swedish root morphemes, accounts for part of the dissimilarities between native and borrowed root morphemes often sensed by native speakers, and sorts out certain inflectional and derivational morphemes as phonotactically special. As to the latter point, the SL/WL approach underscores that content morphemes

80 

 Stig Eliasson

in word languages select their syllable nuclei from larger sound inventories and tend to have more complex syllable shells than functional elements. In accordance with this, native Swedish root morphemes are as a group characterized by (a) a core syllable drawing on a fully differentiated vowel inventory (4.5), (b) core syllable shells evincing a heightened propensity for consonant clustering (4.2), and (c) onsets that include special distributionally restricted consonant phonemes (/h, ɕ, (ɧ)/) (section 3). From this gross division of morphemes, from stress assignment and other stress-triggered processes, and from the rules of morphotactics follows the sequencing of different syllable types in Swedish phonological words automatically. Moreover, in a complete description of the language, the underlying phonotactic properties of morphemes must be accounted for anyway. Consequently, it is not obvious that, beyond the morpheme, the description of Swedish phonotactics will also require the notion of the phonological word (2.2). Still another point relates to what higher-level entities are “profiled” (rather than merely referred to) by phonological means. While many phonological processes destroy, others underscore the phonotactic physiognomy of grammatical units. The phonologically profiled grammatical units in Swedish vary. In Prosodic Phonological terms, most of these processes might be taken to relate largely to the phonological word: stress, tone, vowel elision, aspiration, and, commonly, glottal-stop insertion (4.1, 4.6, 4.7). Yet, it seems that the stress pattern and tonal accent in regular compounds rather mark the grammatical word in that they indicate the connectedness of its constituent parts (e.g. [xfis·kˌdø:d] ‘fish death’) (4.1). Furthermore, it is noteworthy that the morpheme plays a distinct role in Swedish, particularly when the possibility of a word-internal morpheme boundary in simplexes is suggested by tonal accent (e.g. fisk-e [xfis·kə] ‘fishing’) (4.1), by vowel lengthening (type sv[ɑ:]n-s swan-gen ‘of (a) swan’) (4.5), by morphotactically induced vowel combinations within words (by-ig [xby:ig] ‘gusty’) (3; 6.1), or when potential morpheme edges are hinted at by complex post-vocalic consonant clusters arising from syncope (such as [j·sl] in mejsl-ar [xmɛj·slar] ‘chisels’ < /mɛjsel/) (4.2). Thus, in the phonological rule system as much as in phonotactics, it would seem advantageous to be able to draw directly on morphosyntactic concepts. If, upon closer inspection, the phonological word should turn out to be a somewhat insecure foundation for defining a prototypical WL pole (cf. the issue of cross-linguistic comparability of phonological words, 2.3), the question arises how to deal with the patent variation in phonological complexity between languages. Part of an answer might derive from the importance that SL/WL typology attributes to the role of configurative means in language. Conceivably, a major typological scale could describe configurative (culminative and delimitative) function in general without necessarily being dependent on specific definitions



The typology of syllable and word languages and Swedish phonological structure 

 81

of phonological words. Distinguishing between configuratively simpler vs. configuratively more elaborate languages would stress the essential role of configurative features in speech communication, it would be in line with Trubetzkoy’s (1962: 29) original proposal for the study of structure-signaling phonological properties (also adopted by Jakobson and Halle 1956: 9), and it accords well with Auer’s insightful remark that the “culminative and delimitative functions of phonology […] are naturally at the center of interest in a typology oriented towards prosodic categories” (Auer 2001: 1395; translation SE).

References Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Auer, Peter (2010): Word and syllable languages. Thoughts on the typological and diachronic relevance of a prosodic distinction. Paper presented at Phonological Typology of Syllable and Word Languages in Theory and Practice, workshop at the Freiburg Institute for Advanced Studies, University of Freiburg, March 29–31, 2010. Auer, Peter and Susanne Uhmann (1988): Silben- und akzentzählende Sprachen. Literatur überblick und Diskussion. Zeitschrift für Sprachwissenschaft 7/2: 214–259. Bhat, D.N.S. (1978): A general study of palatalization. In: Joseph H. Greenberg (ed.), Universals of Human Language. Vol. 2: Phonology, 47–92. Stanford, CA: Stanford University Press. Blevins, Juliette (1995): The syllable in phonological theory. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 206–244. (Blackwell Handbooks in Linguistics 1.) Oxford: Blackwell. Blevins, Juliette (2004): Evolutionary Phonology. The Emergence of Sound Patterns. Cambridge, England: Cambridge University Press. Booij, Geert (1999): The role of the prosodic word in phonotactic generalizations. In: T. Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 47–72. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Booij, Geert (2011): Morpheme structure constraints. In: Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 4: Phonological Interfaces, 2049–2069. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Braunmüller, Kurt (1995): Zur Phonotaktik des Schwedischen: Prinzipien und Tendenzen. In: Kurt Braunmüller (ed.), Beiträge zur skandinavistischen Linguistik, 9–52. (Studia Nordica 1.) Oslo: Novus Forlag. [Reprinted from: Skandinavistik 10, 1980: 29–60.] Braunmüller, Kurt (2003): Die große skandinavische Vokalverschiebung. Resultat einer gesamtnordischen Entwicklung? NOWELE 43: 21–51.

82 

 Stig Eliasson

Braunmüller, Kurt (2007): Die skandinavischen Sprachen im Überblick. 3rd ed. Tübingen/Basel: Francke. Cairns, Charles E. and Eric Raimy (eds.) (2011): Handbook of the Syllable. (Brill’s Handbooks in Linguistics 1.) Leiden: Brill. Dixon, Robert M. W. and Alexandra Y. Aikhenvald (2002): Word: A typological framework. In: Robert M. W. Dixon and Alexandra Y. Aikhenvald (eds.), Word: A Cross-Linguistic Typology, 1–41. Cambridge/New York: Cambridge University Press. Dressler, Wolfgang U., Katarzyna Dziubalska-Kołaczyk and Lina Pestal (2010): Change and variation in morphonotactics. Folia Linguistica Historica 31: 51–67. Elert, Claes-Christian (1970): Ljud och ord i svenskan [Sounds and Words in Swedish]. Stockholm: Almqvist & Wiksell. Elert, Claes-Christian (1997): Inledning [Introduction]. In: Per Hedelin, Norstedts svenska uttalslexikon [Norstedt’s Swedish Pronunciation Dictionary], 10–43. Stockholm: Norstedts. Eliasson, Stig (1972): Unstable vowels in Swedish: Syncope, epenthesis or both? In: Evelyn S. Firchow, Kaaren Grimstad, Nils Hasselmo and Wayne O’Neil (eds.), Studies for Einar Haugen, 174–188. The Hague/Paris: Mouton. Eliasson, Stig (1985): Stress alternations and vowel length: New evidence for an underlying nine-vowel system in Swedish. Nordic Journal of Linguistics 8: 101–129. Eliasson, Stig (1986): Sandhi in Peninsular Scandinavian. In: Henning Andersen (ed.), Sandhi Phenomena in the Languages of Europe, 271–300. (Trends in Linguistics. Studies and Monographs 33.) Berlin/New York/Amsterdam: Mouton de Gruyter. Eliasson, Stig (1997): The cognitive calculus and its function in language. In: Jadranka Gvozdanović (ed.), Language Change and Functional Explanations, 53–70. (Trends in Linguistics. Studies and Monographs 98.) Berlin/New York: Mouton de Gruyter. Eliasson, Stig (2000): Typologiska och areallingvistiska aspekter på de nordeuropeiska språkens fonologi [Typological and areal linguistic aspects of the phonologies of the northern European languages]. In: Ernst Håkon Jahr (ed.), Språkkontakt – Innverknaden frå nedertysk på andre nordeuropeiske språk [Language Contact – The Influence of Low German on Other Northern European Languages], 21–70. (Nord 2000:19.) København: Nordisk Ministerråd. Eliasson, Stig (2001): Färöisk ljudstruktur och fonologisk teori [Faroese sound structure and phonological theory]. In: Kurt Braunmüller and Jógvan í Lon Jacobsen (eds.), Moderne lingvistiske teorier og færøsk [Modern Linguistic Theories and Faroese], 37–65. (Nordisk Sprogråds skrifter.) Oslo: Novus Forlag. Eliasson, Stig (2008): Prosodic clues to word-internal grammatical boundaries in Swedish. Presentation at the Workshop on Prosodic Alignment at the Word Level, Institut für Deutsche Sprache (IDS), Mannheim, November 20–21, 2008. Eliasson, Stig (2009): Über den qualitativen und quantitativen Beitrag des Deutschen zur schwedischen Phonotaktik. In: Lars Wollin, Dagmar Neuendorff and Michael Szurawitzki (eds.), Deutsch im Norden. Akten der nordisch-germanistischen Tagung zu Åbo/Turku, Finnland, 18.–19. Mai 2007, 55–88. (Nordeuropäische Beiträge aus den Human- und Gesellschaftswissenschaften / Scandinavian University Studies in the Humanities and Social Sciences 28.) Frankfurt am Main: Peter Lang. Eliasson, Stig (2010a): German and Swedish quantity: A comparison. In: Antje Dammel, Sebastian Kürschner and Damaris Nübling (eds.), Kontrastive Germanistische Linguistik, vol. 1, 5–55. (Germanistische Linguistik 206/209.) Hildesheim: Olms.



The typology of syllable and word languages and Swedish phonological structure 

 83

Eliasson, Stig (2010b): Kedjeförskjutningen av långa bakre vokaler och svenskans ‘tionde’ vokal. [The chain shift of long back vowels and the ‘tenth’ vowel in Swedish.] In: Maj Reinhammar (ed.), Studier i svensk språkhistoria 11. Förhandlingar vid Elfte sammankomsten för svenska språkets historia i Uppsala 23–24 april 2010 [Studies in Swedish Language History 11. Transactions of the Eleventh Meeting on the History of the Swedish Language in Uppsala, April 23–24, 2010], 127–136. (Acta Academiae Regiae Gustavi Adolphi 113.) Uppsala: Kungl. Gustav Adolfs Akademien för svensk folkkultur. Eliasson, Stig (2012a): Predicting Swedish tone: The recalcitrant class of bisyllabic root-morphemes in ‑el. Paper read at the 45th Annual Meeting of the Societas Linguistica Europaea, Stockholm, 29 August‒1 September 2012. Eliasson, Stig (2012b): Review of ‘Tonal accents in Norwegian. Phonology, morphology and lexical specification’ by Allison Wetterlin. Zeitschrift für Dialektologie und Linguistik 79: 122–124. Eliasson, Stig and Nancy La Pelle (1973): Generativa regler för svenskans kvantitet [Generative rules for Swedish quantity]. Arkiv för nordisk filologi 88: 133–148. Ferguson, Charles A. (1978): Phonological processes. In: Joseph H. Greenberg (ed.), Universals of Human Language. Vol. 2: Phonology, 403–442. Stanford, CA: Stanford University Press. Gårding, Eva (1967): Internal Juncture in Swedish. (Travaux de l’Institut de Phonétique de Lund 6.) Lund: C.W.K. Gleerup. Garlén, Claes (1988): Svenskans fonologi [The Phonology of Swedish]. Lund: Studentlitteratur. Garlén, Claes (2003): Svenska språknämndens uttalsordbok [The Pronunciation Dictionary of the Swedish Language Committee]. Stockholm: Svenska språknämnden/Norstedts Ordbok. Hall, T. Alan (1999): The phonological word: A review. In: T. Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 1–22. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Hedelin, Per (1997): Norstedts svenska uttalslexikon [Norstedt’s Swedish Pronunciation Dictionary]. Stockholm: Norstedts. Helgason, Pétur and Catherine Ringen (2008): Voicing and aspiration in Swedish stops. Journal of Phonetics 36: 607–628. Hellberg, Staffan (1974): Graphonomic Rules in Phonology. Studies in the Expression Component of Swedish. (Nordistica Gothoburgensia 7.) Göteborg: Acta Universitatis Gothoburgensis. Hockett, Charles F. (1955): A Manual of Phonology. (Indiana University Publications in Anthropology and Linguistics. Memoir 11.) Baltimore: Waverly Press. Hulst, Harry van der (2010): Word accent systems in the languages of Europe. In: Harry van der Hulst, Rob Goedemans and Ellen van Zanten (eds.), A Survey of Word Accentual Patterns in the Languages of the World, 429–507. Berlin/New York: Mouton de Gruyter. Hyman, Larry M. (2008): Directional asymmetries in the morphology and phonology of words, with special reference to Bantu. Linguistics 46: 309–350. Jakobson, Roman and Morris Halle (1956): Fundamentals of Language. (Janua Linguarum 1.) The Hague: Mouton & Co. Kiparsky, Paul (1982): From cyclic phonology to lexical phonology. In: Harry van der Hulst and Norval Smith (eds.), The Structure of Phonological Representations (Part I), 131–175. (Linguistic Models 2.) Dordrecht: Foris Publications. Lindblad, Vern M. (1999): Sandhi of /rs/ sequences in North Norrland Swedish. Ph.D. dissertation, University of Washington, Department of Linguistics.

84 

 Stig Eliasson

Lindqvist, Christer (2007): Schwedische Phonetik für Deutschsprachige. Hamburg: Helmut Buske Verlag. Lyttkens, Ivar Adolf and Fredrik Amadeus Wulff (1885): Svenska språkets ljudlära och beteck­ ningslära jämte en afhandling om aksent [Phonetics and Orthography of the Swedish Language Together with a Treatise on Accent]. Lund: C. W. K. Gleerups Förlag. Lyttkens, Ivar Adolf and Fredrik Amadeus Wulff (1916): Ordskatt och ljudförråd i svenska språket [Vocabulary and Sound Inventory of the Swedish Language]. Lund: C. W. K. Gleerups Förlag. Maddieson, Ian (1984): Patterns of Sounds. Cambridge: Cambridge University Press. Maddieson, Ian (2006): Correlating phonological complexity: Data and validation. Linguistic Typology 10: 106–123. Maddieson, Ian (2007): Issues of phonological complexity: Statistical analysis of the relationship between syllable structures, segment inventories, and tone contrasts. In: Maria-Josep Solé, Patrice Speeter Beddor and Manjari Ohala (eds.), Experimental Approaches to Phonology, 93–103. Oxford: Oxford University Press. [Also in UC Berkeley Phonology Lab Annual Report 2005: 259–268. Available at http://linguistics.berkeley.edu/ phonlab/annual_report/annual_report_2005.html]. Maddieson, Ian (2011): Phonological complexity in linguistic patterning. Proceedings of the 17th International Congress of Phonetic Sciences (ICPhS XVII), Hong Kong, 17-21, 2011, 28–34. Available at http://www.icphs2011.hk/resources/OnlineProceedings/PlenaryLecture/ Maddieson/Maddieson.pdf. Malmberg, Bertil (1968): Svensk fonetik [Swedish Phonetics]. Lund: Gleerups. Milner, George B. (1966): Samoan Dictionary: Samoan-English, English-Samoan. London: Oxford University Press. [Reprinted 1993. Auckland: Polynesian Press.] Nespor, Marina (1999): Stress domains. In: Harry van der Hulst (ed.), Word Prosodic Systems in the Languages of Europe, 117–159. (Empirical Approaches to Language Typology. EUROTYP 20.4.) Berlin/New York: Mouton de Gruyter. Nespor, Marina and Irene Vogel (1986): Prosodic Phonology. (Studies in Generative Grammar 28.) Dordrecht: Foris. Noreen, Adolf (1903–1907): Vårt språk. Nysvensk grammatik i utförlig framställning 1 [Our Language. A Comprehensive Modern Swedish Grammar 1]. Lund: C. W. K. Gleerups Förlag. [Pp. 339–542, 557–558, 576–577 reprinted as: Adolf Noreen, Ljudlära [Phonetics]. Lund: Gleerups, 1966.] Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 129.) Stuttgart: Steiner. Nübling, Damaris, Antje Dammel, Janet Duke and Renata Szczepaniak (2006): Historische Sprachwissenschaft des Deutschen. Eine Einführung in die Prinzipien des Sprachwandels. (Narr Studienbücher.) Tübingen: Gunter Narr Verlag. Page, B. Richard (2009): Review of ‘Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache’ by Renata Szczepaniak. Diachronica 26: 134–139. Raffelsiefen, Renate (1999): Diagnostics for prosodic words revisited: The case of historically prefixed words in English. In: T. Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 133–201. (Current Issues in Linguistic Theory 174.) Amsterdam/ Philadelphia: John Benjamins.



The typology of syllable and word languages and Swedish phonological structure 

 85

Raffelsiefen, Renate (2007): Morphological word structure in English and Swedish: The evidence from prosody. In: Geert Booij, Bernard Fradin, Angela Ralli and Sergio Scalise (eds.), On-line Proceedings of the Fifth Mediterranean Morphology Meeting (MMM5), Fréjus 15–18 September 2005, University of Bologna. Available at http://mmm.lingue. unibo.it/mmm-proc/MMM5/209-268-Raffelsiefen.pdf. Rankka, Erik, Jean Barjac and Christiane Axelsson (1978): Kurs i franskt uttal. För användning i lärostudio (LS) [Course in French Pronunciation. For Use in Language Laboratory (LL)]. 2nd ed. Stockholm: AWE/Gebers. Revithiadou, Anthi (2011): The phonological word. In: Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 2: Suprasegmental and Prosodic Phonology, 1204‒1227. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Riad, Tomas (2010): Retroflektering [Retroflection]. In: Kristinn Jóhannesson, Ida Larsson, Erik Magnusson Petzell, Sven-Göran Malmgren, Lena Rogström and Emma Sköldberg (eds.), Bo65. Festskrift till Bo Ralph [Bo65. Festschrift for Bo Ralph], 214–227. (Meijerbergs arkiv för svensk ordforskning 39.) Göteborg: Meijerbergs institut för svensk etymologisk forskning, Göteborgs universitet. SAOB = Svenska Akademiens ordbok [The Dictionary of the Swedish Academy] (1893–): Ordbok över svenska språket utgiven av Svenska Akademien [Dictionary of the Swedish Language Published by the Swedish Academy]. Lund: Svenska Akademien. Schiering, René, Balthasar Bickel and Kristine A. Hildebrandt (2010): The prosodic word is not universal, but emergent. Journal of Linguistics 46: 657–709. Shin, Eurie (2007): How do listeners compensate for phonology. In: Maria-Josep Solé, Patrice Speeter Beddor and Manjari Ohala (eds.), Experimental Approaches to Phonology, 386–404. Oxford: Oxford University Press. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Szczepaniak, Renata (2009): Silbensprachen versus Wortsprachen. Natur & Geist. Das Forschungsmagazin der Johannes Gutenberg-Universität Mainz 25/2: 49–52. Tamm, Fredrik (1887): Fonetiska kännetecken på lånord i nysvenska riksspråket [Phonetic Criteria of Loanwords in Modern Standard Swedish]. (Upsala Universitets Årsskrift 1887. Filosofi, Språkvetenskap och Historiska vetenskaper, I.) Upsala: Akademiska Boktryc­ keriet, Edv. Berling. [Also in facsimile edition. (Acta Universitatis Upsaliensis. Studia Philologiae Scandinavicae Upsaliensia, 5). Uppsala: Almqvist & Wiksells Boktryckeri, 1966.] Trommer, Jochen (2011): Phonological sensitivity to morphological structure. In: Marc van Oostendorp, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 4: Phonological Interfaces, 2464–2489. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Trubetzkoy, Nikolaj S. (1962): Grundzüge der Phonologie. 3rd ed. Göttingen: Vandenhoeck & Ruprecht. Ultan, Russell (1978): A typological view of metathesis. In: Joseph H. Greenberg (ed.), Universals of Human Language. Vol. 2: Phonology, 367–402. Stanford, CA: Stanford University Press. Vennemann, Theo (1982): Zur Silbenstruktur der deutschen Standardsprache. In: Theo Venne­mann (ed.), Silben, Segmente, Akzente. Referate zur Wort-, Satz- und

86 

 Stig Eliasson

Versphonologie anläßlich der vierten Jahrestagung der Deutschen Gesellschaft für Sprachwissenschaft, Köln, 2.–4. März 1982, 261–305. (Linguistische Arbeiten 126.) Tübingen: Max Niemeyer Verlag. Vogel, Irene (2009): Universals of prosodic structure. In: Sergio Scalise, Elisabetta Magni and Antonietta Bisetto (eds.), Universals of Language Today, 59–82. (Studies in Natural Language and Linguistic Theory 76.) Dordrecht: Springer. Wetterlin, Allison (2010): Tonal Accents in Norwegian. Phonology, Morphology and Lexical Specification. (Linguistische Arbeiten 535.) Berlin/New York: De Gruyter.

Matthias Heinz (University of Salzburg)

Syllable complexity in the diachrony of Romance languages: A center vs. periphery view and the syllable vs. word rhythm paradigm [D]ie Aufgabe [der historischen Phonetik] besteht [darin], mit Hilfe phonetischer Kenntnisse den Weg nachzuzeichnen, von dessen Zurücklegung im Bewußtsein der Sprechenden keine Spur bleibt, der aber ganz bestimmt zurückgelegt worden ist.1 (Richter 1977: 369)

Abstract: Although Romance languages display an overall diachronic drift towards the simplification of syllable types, the persistence of complex structural patterns is also evident (especially in languages with a long-standing written tradition such as Catalan, French, Italian, Portuguese and Spanish). Complex syllable shells are found in specialized lexical strata (cf. VCC and CVCC in learned borrowings such as Sp. abstracción ‘abstraction’ and perspectiva ‘perspective’). Syllabic complexity may also depend on speech rate and result in two opposite patterns: On the one hand, varieties with features of the syllable-language type reduce complexity in allegro realizations, while complex syllable types arising from learned forms occur in lento speech only (e.g. Spanish). On the other hand, languages such as European Portuguese and Standard French favor syllabic complexity in allegro forms due to the weakening of unstressed vowels, while lento forms maintain (unstressed) vocalic nuclei. This typological distribution is discussed within a “center vs. periphery” account of the functioning of language systems. According to this view, complex syllable types belong to the system’s periphery in some languages while being part of its center (or core) in others.

1 Introductory remarks This paper focuses on one important parameter of phonological typology, namely syllabic complexity, as found to a varying extent in Romance languages. The

1 ‘The task of historical phonetics consists of reconstructing – with the help of phonetic knowledge – the path that has left no trace in the speakers’ conscience despite having no doubt been taken’ (translation M.H.).

88 

 Matthias Heinz

degree of complexity the syllable structure of a language can attain depends on a variety of phonological and rhythmic-typological factors. With respect to the typological framework proposing a continuum between “syllable-based” and “word-based” languages,2 the different forms of syllabic simplicity and complexity in the Romance languages laid out in the present article are assumed to provide evidence especially for profiling (and in some cases contrasting) the “syllable” end of the continuum. Despite the rich and diverse information regarding the syllable phonology of non-standard, especially diatopic Romance varieties,3 here mainly “macroscopic” (e.g. larger regiolects with common characteristics such as the so-called tierras altas and tierras bajas varieties of Spanish, see below) and standard Romance varieties will be touched upon. The western Romance languages of Spanish, Portuguese (especially the European variety), Catalan and French (which shares some typological features with Catalan) or, within the eastern Romance group, Italian and Romanian, display somewhat different systematic4 tendencies with regard to syllabic complexity. Only a broad overview and some typological generalizations regarding mainly Portuguese, Spanish and (Standard) French5 can be given here. Syllabic complexity is understood in what follows as complexity of the syllabic shell, i.e. consonantal complexity in the onset and/or the coda. Complexity in the left or right margin of a syllable can be said to increase if the onset contains two or more consonantal segments and/or the coda contains one or more segments, resulting in potential syllables of the type (C…)CCV(C)(C…).6 While so-called naked, open syllables made up of a (vocalic) nucleus only (V) are also

2 Cf. Auer (1993: 10–11, 24–34, 41–55 and passim, 2001: 1394–1398). For a detailed overview of parameters within the “syllable vs. word language” framework cf. Szczepaniak (2007: 52–53). 3 Cf. for instance Schmid (2000), who offers an overview of syllable types in a sample of Italo-Romance dialects; cf. also Ledgeway (2009: 36–39) on Neapolitan and, more generally, Loporcaro (2011) on the respective diachronic tendencies in the Romance languages. 4 The adjectives systematic and, further on, systemic are used here distinctly, the former meaning ‘regular, according to a system’‚ the latter ‘of or pertaining to the system in question’. 5 Catalan (see Caro Reina, this volume) requires a more detailed look due to the interesting divergence within its eastern and western dialects. Eastern Romance – Italian and Romanian – present a somewhat mixed picture and cannot be treated in detail here (cf. Heinz 2010b). 6 A general typology of syllable structure in the form of “preference laws” is sketched by Vennemann (1988, drawing on a host of Romance examples, cf. especially 14, 16–19, 69 note 3; cf. also Restle and Vennemann 2001). Many descriptive terms in syllabic typology (e.g. head, shell and others) are indebted to Vennemann’s seminal work (1988: 5–6). Formal accounts of syllabic typology have gained particular momentum within the constraint-based framework known as Optimality Theory (OT), cf. Prince and Smolensky (2004).



Syllable complexity in the diachrony of Romance languages 

 89

found in many languages, the CV syllable type, which combines a consonantal onset and a vocalic nucleus, can be considered as prototype of simple syllable structure.7 The CV pattern is agreed to be a universally-attested structural type of syllable. Jakobson implicitly states this generalization as follows: “There are languages lacking syllables with initial vowels and/or syllables with final consonants, but there are no languages devoid of syllables with initial consonants or of syllables with final vowels” (Jakobson 1962: 526).8 Statistics on syllable types present in Old Spanish (Guerra 1983; Heinz 2008) and Catalan texts (Heinz 2010a and, in more detail, 2010b) show that as expected and in line with earlier research the simple CV type is much more prevalent than all the other types. Nonetheless, the status of some of the more complex syllable types, such as CCVC or CCVCC as in Spanish tras or trans, proves to be interesting. In spite of being much less frequent than CV, those types are still not uncommon in a language like Spanish (and much less so e.g. in Catalan or Portuguese), but seem to occur mainly in specific lexical items; in Catalan and European Portuguese further factors lead to syllabic complexity. Taking a closer look it becomes clear that some of the more complex syllable types have their primary raison d’être in learned borrowings, i.e. certain complex syllable patterns are found mostly – or even exclusively – in specialized lexical strata or in texts belonging to single discourse traditions, e.g. the CVCC and VCC syllable types in Spanish that are characteristic of the learned vocabulary (cf. Latin borrowings such as Sp. abstracción ‘abstraction’ and perspectiva ‘perspective’). The clusters and coda consonants in these Spanish forms are traditionally called grupos cultos (‘learned (consonant) clusters’); they are liable to be reduced in less careful, faster speech but are pronounced clearly in careful articulation. Thus, with regard to variation in phonetic realization two types of behavior can be observed, each of which can be assigned to a peripheral or a central tendency of the language system. The central (or core) area of a language system is made up of a restricted set of structural principles and regularities whereas

7 Cf. also Maddieson’s (2008) grouping of a large sample of languages (485) according to syllabic complexity, i.e. (C)V, (C)(C)V(C), (C)(C)(C)V(C)(C) and beyond. The representation of syllable structure as a sequence of C and V elements features prominently e.g. in the generative approach known as “CV Phonology” (where CV is the temporally defined sceletal tier, cf. Clements and Keyser 1983; it also figures in a different, highly abstract approach derived from Government phonology called “Strict CV phonology”, positing empty elements as part of an underlying CV structure, cf. Lowenstamm 1996; Scheer 2004), though as a means of primary structural description it is found much earlier, cf. Hockett (1947) and Abercrombie (1967: 38–41). 8 Cf. furthermore – among many others – Clements and Keyser (1983: 28), Trask (1996, s.v. syllable), Maddieson (2005), and MacNeilage and Davis (2005).

90 

 Matthias Heinz

its periphery consists of (supposedly irregular) countertendencies, variants, and historical residues. It is the interplay between systemic areas that gives rise to the characteristic feature combination of a specific language. This approach will be explained in further detail in section 3 below. The two types of behavior are exemplified by different Romance languages: In (Standard Peninsular) Spanish complex syllable types in the more learned vocabulary are realized in a careful lento pronunciation style (e.g. Sp. do[k]tor, conce[p]to). Based on the circumscribed lexical occurrences of such syllabic realizations, we could call them peripheral with regard to their position in the system. On the other hand, in languages such as French and European Portuguese (henceforth EP) complex (often homosyllabic) clusters tend to occur in allegro-style pronunciation with typical deletion of unstressed vowels.9 In other words, here it is a systemically central tendency which results in syllable shell complexity, as diachronic tendencies and synchronic frequency indicate that the reduction of unstressed vowels in those languages is clearly part of their core phonological characteristics. The resulting complexity is of a differently conditioned kind, though, for (depending on temporal characteristics of articulatory realization) the deletion of unstressed vowels can lead to the remaining consonant clusters becoming part of stressed syllables, with increased complexity in the left or right margin. The two basically antagonistic tendencies underlying this behavior in Latin and Romance languages, namely syllabic reduction and persistence of shell complexity, will be illustrated in section 2. I will then give a center vs. periphery account aiming to explain the different status of superficially similar syllabic patterns occurring in certain Romance varieties (section 3). A typological grouping of those varieties can be seen to emerge from the two kinds of language-specific preferences regarding complexity (section 4).

9 Compare Fr. devoir, lento: [dəˈvwaʁ], allegro: [dvwaʁ], EP para lento: [ˈpɐɾɐ], allegro: [pɾɐ] and the examples (7) and (8) in section 3.1. Lento vs. allegro articulations may in many instances be gradient in nature. For the sake of descriptive simplification, the two are taken to be dichotomous here, which seems adequate with regard to their forming the two extreme edges of a continuum of potential realizations.



Syllable complexity in the diachrony of Romance languages 

 91

2 Tendencies in Latin and Romance languages: Syllabic reduction vs. persistence of shell complexity 2.1 Syllabic reduction The syllabic typology of Romance languages displays an overall diachronic drift towards the simplification of syllable structure already visible in Latin (cf. Lehmann 2005), resulting in many cases in an optimal CVCV patterning of strings of segments, with CV being indubitably the minimal point of syllabic reduction. As for Romance syllable structure, things seem rather clear. The most common syllable types found in Romance languages are those Clements and Keyser (1983: 28) have described as “core syllable types”, i.e. CV, V, CVC, VC. CV makes up more or less 50% of all syllabic occurrences in Spanish (Guerra 1983; Heinz 2008), Catalan (De Yzaguirre 1995; Heinz 2010a), Portuguese (Frota, Vigário, and Martins 2006), French (Wioland 1985), Italian (Schmid 1999: 103, 2000) and Romanian (cf. Hess 1975: 238–275, who also gives statistics for the other five languages). Likewise, Latin displays an overall tendency towards the reduction of syllabic complexity and a predominance of the CV pattern.10 CVCV sequence optimization is brought about especially by epenthesis (Kiss 1971: 100–107) as a process of consonant cluster resolution by way of vowel insertion.11 Probably the most frequent case of epenthesis in Romance languages is vowel prosthesis, which also has consequences for syllable structure. 12 This phenomenon of inserting a vocalic element (usually /e/ or /i/) before an s impurum (i.e. /s/ followed by a consonant), thus simplifying clusters and creating a syllabic nucleus of its own, is characteristic of many, albeit not all, Romance languages:

10 This tendency is documented early on in the history of Latin, cf. Kiss’s (1971) study on syllable types based on a corpus of Latin texts. On Latin syllable structure cf. also Pulgram (1970), Allen (1973: 27–45), Lahiri, Riad, and Jacobs (1999: 396–399), Lehmann (2005) and Sampson (2010: 41–52). 11 Although relatively frequent in Late Latin, this phenomenon does not seem to have affected the phonological evolution of early Romance (“Malgré sa fréquence relativement forte en latin impérial et même plus tard (par exemple dans les chartes de la Gaule des VIIe–VIIIe siècles), elle ne semble avoir laissé aucune trace en roman” [In spite of its relatively frequent occurrence in Imperial Latin and even later on (e.g. in charters from seventh- and eighth-century Gaul), it does not seem to have left any trace in Romance], Kiss 1971: 100). However, the author (p. 100) mentions Brazilian Portuguese (henceforth BP) as one of the Romance varieties where closed syllables are often avoided by inserting epenthetic vowels; cf. the common rendering of Port. absoluto ‘absolute’ (EP [ˌɐb.su.ˈlu.tu]) in BP as [ˌɐ.bi.so.ˈlu.tu]. 12 Cf. Sampson’s (2010) recent monograph offering a detailed account of this phenomenon within the panorama of the Romance languages at large.

92 

 Matthias Heinz

(1) Lat. schola ‘school’ > a. Port.13, Cat. escola, Sp. escuela, Fr. école b. It. scuola,14 Rom. şcoală

Still, an opposite trend is also visible in frequent vowel syncopation in Late Latin, giving rise either to new complex onset clusters (e.g. o.cu.lus > o.clus ‘eye’) or to closed syllables, often with the result of deteriorated syllable contact (e.g. do.mi.na > dom.na ‘mistress’, cf. Kiss 1971: 14–21; Väänänen 1981: 40–41).15 From a typological point of view Lehmann (2005: 143–144) summarizes the diachrony of Latin syllable structure as follows: “As regards diachrony, Latin represents a stage in a movement that starts in Proto-­Italic and ends in Proto-Romance (or, in some respects and some language, even later and that leads to a simplification of syllable structure in all of its constituents” (emphasis added). Hence there seems to be a long-term trend working towards the simplification of syllable patterns both in Latin and in early as well as, partly, in later stages of the Romance languages. In section 4.4, we will see some cases (e.g. in southern varieties of Peninsular Spanish) where the mo­ve­ment towards simplification is synchronically observable within phonetic variation (especially in so-called allegro realizations). The CV type, closely related to a general rise of open syllables,16 comes to make up more than half of all occurrences in the corpus analyzed by Kiss (53%, cf. 1971: 103), while moderately complex types such as CCV, CVC, CCVC (forms like spes or with muta cum liquida as in tres and -flic- in afflictio) and even more complex types are much less frequent. So, Latin displays a tendency for syllabic simplification (re­garding onset, nucleus, and coda, i.e. all

13 For EP Sampson (2010: 112) rightly observes: “In standard Portuguese, the disappearance of prosthetic vowels at a phonetic level is a recent phenomenon caused by the general weakening of unstressed vowels”. 14 In Old Italian prosthetic /i/ is partly found, surviving in some traditional formulaic expressions of the written language such as in iscuola ‘in school’, per iscritto ‘written down’ (where the word preceding the prosthetic vowel ends in a consonant). 15 Geisler (1992: 44–89) gives a detailed account of the effects of syncope on syllabic and prosodic systems on their way from Latin to the Romance languages. 16 Loporcaro (2011: 105), however, concludes from data on syllable contact in the Romance languages that “late Latin/proto-Romance had no ‘tendency towards open syllables’. Quite the contrary, for stressed syllables, a ‘tendency towards closed syllables’ seems to have been in force”. Furthermore, the author states that “proto-Romance was probably characterized by a ‘tendency towards a closed syllable’” (153), invoking the “proliferación de las sílabas cerradas” [proliferation of closed syllables] (caused especially by widespread apocope), which Catalán (1971: 78) dates for Old Spanish between the 11th and 13th centuries.



Syllable complexity in the diachrony of Romance languages 

 93

syllabic constituents) that is to become even more per­vasive in the Romance languages. At first glance the latter can then be described as having a) rather simple syllable types, b) generally well-defined, constant syllable boundaries, c) sonority sequencing mostly in line with a hierarchy of segments, with sonority peaks in the (vocalic) nucleus,17 d) a tendency towards resyllabification across word boundaries. Spanish is largely representative of these general Romance trends; the following example – many more such examples are found throughout the Romance languages – illustrates that the reduction of syllable structure (here of a coda segment) takes place as Romance vernaculars evolve from Latin:18 (2) CiCj > Cj Lat. Mod. Sp. ipse (dem.pron.m ‘himself’) > ese (dem.pron.m ‘that’) (cf. Bybee 2001: 208)

Some Romance varieties, such as Southern Peninsular Spanish and the so-called Lowland (tierras bajas) varieties of American Spanish, show an even stronger trend towards reduction, especially of coda segments. In (3) syllable-final /s/ undergoes a process of aspiration and sometimes one of subsequent deletion of the remaining coda material: (3) Sp. entonces ‘then’ [enˈtonses] > [enˈtonseh] > [enˈtonse] (Bybee 2001: 209)

In some varieties this tendency can lead to the deletion of almost any consonantal segment admitted in syllable-final position (the deletion of word-final /d/, /n/, /r/, /s/ is commonly observed) and can have consequences for the expres-

17 A “Sonority Sequencing Generalization” (SSG) in this sense is described by Blevins (1995: 210) as follows: “Between any member of a syllable and the syllable peak, a sonority rise or plateau must occur” (cf. also Vennemann 1988: 36; Restle and Vennemann 2001: 1312). The fundamental idea of sonority distribution within the syllable is older and can be found in Sievers (1901: 198; already, though less explicitly, in the first edition, 1876: 111–113) and Jespersen (1904). 18 As is suggested by the presentation of this example in Bybee (2001: 208), gemination is thought to be part of this evolution in some intermediate stage, although one has to be careful to distinguish forms such as Old Sp. with a diacritic digraph rendering the unvoiced sibilant in the Alfonsine castellano drecho orthography (13th century; cf. instances of this form e.g. in the Primera Crónica General, ca. 1270–89, I, chapter 555, ed. by Menéndez Pidal 1977: 308–309) from actual (phonetic) gemination.

94 

 Matthias Heinz

sion of morphological markers.19 All of those cases give rise to open syllables and optimize strings of segments towards the favored CVCV sequences. This behavior is fully in line with the syllable vs. word typology in that those syllable-related processes of word-final consonant deletion operate, as it were, at the expense of the phonological word, affecting word boundaries and possibly making them disappear.20

2.2 Persistence of complex syllable types Nonetheless, it is evident that some complex structural patterns21 persist, especially in Romance languages with long-standing written tradition such as French, Italian, Spanish, Catalan, Portuguese and Romanian.22 As mentioned above, certain complex syllable types are present mainly – or solely, as in the example of Spanish – in specialized lexical strata, e.g. the CVCC and VCC syllable types in Spanish, which are characteristic of the learned vocabulary (cf. borrowings of Latin origin such as the above-mentioned Sp. abstracción, obstruir, perspectiva, etc.). (4) Lat. abstractione(m) ‘abstraction’: Rom. abstracţie (also abstracţiune), It. astra­zione, Fr. abstraction, Sp. abstracción, Cat. abstracció, Port. abstracção (BP: abstração)

The consonant clusters present in Lat. abstractione(m) /bstr/ and /kt/ (plus the glide /j/, leading to the formation of an affricate sound by way of palatalizing the preceding dental) are sometimes simplified in spoken language. However, both a graphic and a phonic reflection of those sequences are found in relatively complex syllable structures, e.g. in Italian, French or Spanish:

19 Cf. Alvar (1996: 242–247), Brown (2009), and Colina (2009: 29) on the deletion of the morphologically frequent syllable-final /s/. Deleted inflectional markers may be reflected in minute though phonologically relevant differentiations as in la vaca [laˈβaka] ‘def.art[sg] cow[sg]’ vs. la-s vaca-s [laˈbaka(ː)] ‘def.art-pl cow-pl’ or tiene [ˈtjene] ‘hold[3sg.pres]’ vs. tiene-s [ˈtjɛnɛ] ‘hold-2sg.pres’. In the latter example the distinction of vowel quality, merely allophonic in Standard (Northern Peninsular) Spanish, is phonologized. 20 Cf. the blurring of word edges through assimilation and deletion phenomena observed by Szczepaniak (2007: 137–138, 146, 333–334) in Old High German with its clear syllable language characteristics. 21 Discerning between phonetic outcomes or mere graphic representations is sometimes difficult for earlier stages of the languages, cf. Heinz (2010b), also Wright (1994). 22 The tradition of learned elements in the lexicon, which in some respects proves to be linked to the occurrence of complex syllable types, is obviously less important in other Romance varieties with only a recent history of standardization (e.g. Sardinian, Romansh, etc.).



Syllable complexity in the diachrony of Romance languages 

(5) a. It. astrazione b. Fr. abstraction c. Sp. abstracción

/as.trat.ˈtsjo.ne/ /aps.trak.ˈsjɔ̃/ /abs.trak.ˈθjon/

 95

VC.CCVC.CaffrGV.CV VCC.CCVC.CGVnas VCC.CCVC.CGVC

Except for the later form of Rom. abstracţie, abstracţiune (first attested in 1821), the Italian, French, Spanish, Catalan and Portuguese forms date between the 13th and 16th centuries (cf. the dictionary of Latin loanwords in Romance by Reinheimer Rîpeanu 2004: s.v. abstractio), i.e. mostly before the fixing of orthographic norms. Although, of course, a Latinized (or in other instances, Grecized) graphic Gestalt was sometimes added or restored in later stages – something rather difficult to deduce from historical dictionaries – we must simply take notice of the presence of such forms in texts. In many cases they may not give an accurate graphic picture of the actual sounds; rather, the pronunciation of restored etymological or even pseudoetymological letters is known to sometimes have been induced by the spelling itself. Still, an explanation is needed for how those clusters have survived in such great numbers and over long periods of time. Those elements might be thought to have never actually been pronounced and to be simply the cultural baggage of re-Latinizing tendencies still carried by part of the vocabulary.23 This, however, proves to be a fairly reductive view once we consider careful pronunciations of Spanish where those so-called grupos cultos are invariably pronounced,24 let alone French, where in a form like absolution [apsɔlyˈsjɔ̃] no deletion of the coda obstruent occurs.25 Words with such clusters have been part of the lexicon of those languages long enough for some consistent usure phonique (Lüdtke 1986), a diachronic “wearing off”, i.e. attrition of sound structure, to be expected. So while writing traditions surely have their part in the pre­ser­vation of such structures (as suggested by Wright 1994: 176, for earlier phases of scriptural development), they cannot wholly explain why clusters, which make syllable structure more complex and counter a clear tendency towards syllabic simplification, are still so vigorous.

23 Variants like Spanish setiembre ‘September’ beside septiembre could be regarded as indicators of a merely graphic reality of complex consonant groups, but it is noteworthy that the two go beyond spelling variants, for the [pt] in septiembre may be realized in careful pronunciation (i.e. in a “dicción culta y enfática”, a marked lento pronunciation style, Navarro Tomás 1967: 140). 24 Indeed, an “etymologizing approach to spelling and pronunciation” (Sampson 2010: 106) began to affect particularly learned words in Spanish from the seventeenth century on, leading to “a conservative trend that developed strongly from the eighteenth century resulting in the restoration of ‘silent’ etymologically justified consonants in the standard pronunciation of more learned words” (Sampson 2010: 106). 25 Rather, the fairly productive assimilation process of coda voice agreement applies ([ap.s‑]; for similar cases in French cf. Pustka 2011: 148).

96 

 Matthias Heinz

3 A center vs. periphery view on syllable type complexity in Romance languages The presence of complex syllable types that seem contrary to the general evolutionary trend towards simplification which would seem so pervasive in the phonological typology of the Romance languages therefore needs explanation. After discussing a few particular cases in French and EP, I will aim to use the “center (or core) vs. periphery” view of the organization of linguistic systems introduced briefly in section 1 above, in order to shed light on the problem.26

3.1 Syllable shell complexity as a typological parameter in Romance languages Learned borrowings are especially prone to the above-mentioned complex patterns, as shown by cases like the outcomes of Latin abstractio in Romanian, Italian, French, Catalan, Spanish and Portuguese. It is loans of learned origin that seem to best preserve certain complex syllable types which are very rare or absent in most other words. Those lexical units can be “carriers” of syllabic complexity even in languages with a clear preference for simple syllable types like Spanish. Alternations between inherited word and newly-formed related lexical items of learned origin are shown in example (6) from Portuguese; in the adjective pluvial (‘rainy’) a muta cum liquida nexus is restored, contradicting the diachronic sound change which had made pl- clusters disappear:27 (6) Lat. pluvia(m) ‘rain’ > Port. chuva (CV.CV) vs. Lat. pluviale(m) ‘rainy’ > Port. pluvial (CCV.CGVC)

26 In distinguishing central from peripheral types it goes without saying that a circular definition – “peripheral because less frequent, less frequent therefore peripheral” – must be avoided. Systemically peripheral complex structural patterns clearly display a degree of phonotactical markedness without necessarily being infrequent, while on the other hand certain marked structures may occur frequently. A general example of this is the inflectional morphology of many European languages where irregular forms occur in high-frequency verb paradigms (cf. e.g. the English verb take with the past tense took and the past participle taken) and seem actually to be preserved as an effect of their relative frequency of use. For the correlation of frequency and cognitive entrenchment cf. Langacker (1991, 1994) and Bybee (2006), on different notions of frequency in linguistics cf. Loiseau (2010). 27 Restle and Vennemann (2001: 1315–1316) explain how a system that has undergone improvements (in this case regarding the parameter of head quality) can be “gestört” (‘disturbed’), as in the above example (6), by the subsequent importation of loan elements.



Syllable complexity in the diachrony of Romance languages 

 97

Apart from complexity imported through loanwords, EP – as a language displaying characteristics of a word language – admits highly complex consonant clusters in the onset and the coda due to frequent schwa deletion (cf. example (8) below). French, showing an essential prosodic organization into phrase units (known as groupes rythmiques; cf. e.g. Dufter 2004) which makes it a very special case among the Romance languages, admits syllabic complexity at either end of a diaphasic spectrum, from carefully pronounced lento style to realizations conditioned by fast and less careful allegro articulation. While consonantally complex patterns like Fr. strict, abstraction are found in the learned vocabulary and tend to be fully articulated in lento (and less probably so in allegro) speech style, allegro forms sometimes diverge considerably from the phonotactics of lento articulations:28 (7) Fr. a. tu l’as vu ‘you have seen him/her’ allegro [tla.ˈvy] vs. lento [ty.la.ˈvy] b. je (ne) te le fais pas dire ‘you can say that again’ [ʃtlə.fɛ.pa.ˈdiʁ] vs. [ʒə.nə.tə.lə.fɛ.pa.ˈdiʁ] c. le voilà qui revient ‘back he comes’ [lvla.kiʁ.ˈvjɛ̃] vs. [lə.vwa.la.ki.ʁə.ˈvjɛ̃]

It is noteworthy that allegro-style-induced vowel deletion in French goes beyond schwa, indeed almost any vowel can be reduced in clitic elements as examples 7a and 7c show. This affects the right margin of phrase-initial and phrase-internal words, whereas complex codas are the result of word-final schwa deletion as in e.g. Fr. table ‘table’ [tabl], arbre ‘tree’ [arbʁ], livre ‘book’ [livʁ], etc.29 If interpreted homosyllabically, many of these clusters (e.g. [ʃtl], [lvl]) present a challenge to the predictions of sonority sequencing (see above, note 15). In EP the multiple deletion of unstressed (pretonic) vowels can result in clusters of remarkable complexity,30 as in the first syllable of (8a–c): (8) a. d(e)p(e)nicar ‘nibble’ b. d(e)spr(e)zar ‘despise’ c. d(e)spr(e)stigio ‘slating’

[dpni-] (CCCV) σ [dʃpɾ̩-] (CCCC) σ [dʃpɾʃ-] (CCCCC)σ (!)

28 These examples are by no means marginal in the reality of spoken French (for examples 7b and 7c, cf. Krötsch 2004: 218). 29 On properties of schwa in French cf. Noske (1993); for a list of vowels liable to be elided in French cf. Pustka (2011: 151). 30 Cf. Mateus and D’Andrade (2000) and the clusters analyzed e.g. by Cunha (2011).

98 

 Matthias Heinz

In such “hyper-complex” syllables as (8b), the liquid [ɾ] can assume the properties of a nucleus: (8b’) EP desprezar [d p . za ] (lento [d .p . za ])

C C C d

p

C

C

V C

z

a

Again, as in the French examples mentioned above, many of these cases defy hierarchical segment ordering according to the predictions of sonority sequencing: (8b’’) Sonority sequencing of Port. [d p ]zar

…

[vibr] [fric] [obstr] [+voice] [obstr] [–voice]

[ d

p

]

Moreover, in much theoretical work on syllable phonology the possible complexity of syllable shells (and, as the quote from Blevins 2004 will show, nuclei) is heavily constrained (cf. for instance Duanmu’s 2009 CVX proposal; cf. also Blevins 2010). Blevins (1995, 2004: 213–214, 2006) rightly mentions “uncommon syllable types” which are not accommodated or, more often than not, excluded in primarily theoretically informed inventories of permissible syllable types: [T]here is nothing intrinsically difficult about the production or perception of VVVV [e.g. the augmentative suffix -kaaei] in Gilbertese or CCCCCC [e.g. the finite verb gvprckvnis] in Georgian, and there is no evidence that these tauto­syl­la­bic sequences are in any way unstable. Rather, the rarity of such systems appears to be the result of the uncommon convergence of significant rules of consonant or vowel loss resulting in long V and C clusters respectively; prosodic systems in which stress-ti­ming, and not syllable-timing, prevails; and unambiguous rules of syllabification. (Blevins 2004: 214)

The overall complexity the syllabic shell can attain in languages like French and EP as well as violations of the sonority hierarchy match the parameters observed in the word-language type.31 The boundaries of the phonological word are less

31 Cf. the overview of the respective typological parameters in Szczepaniak (2007: 52).



Syllable complexity in the diachrony of Romance languages 

 99

sharply profiled, though, as seen in the tendency towards resyllabification across word boundaries in other Romance languages like Spanish and Italian.32 Moreover, French, with its phrase-related prosodic make-up, is somewhat difficult to classify with more clear-cut cases (cf. Dufter 2004).

3.2 Complexity: central vs. peripheral The few examples presented so far (to which many more can be added) show that there is a need to accommodate the tension between the array of generally simple syllable types that seem to be at the center of the phonotactic patterning of many Romance languages and the persistence of more complex types, whose status is less clear. Albeit varying in degree within the different languages, two main factors, one external, one internal, seem to condition the build-up of syllabic complexity, namely the influence of loan phonology and the effects of (different styles of) articulatory realization. Hence, a more general question is raised: How can the interaction of the central, internally governed and the seemingly disparate, externally influenced, peripheral workings of the different levels of linguistic description be modeled? The basic center–periphery conceptualization proposed here is in itself not new. Within structuralism (especially the Prague School), a center–periphery distinction in language systems is acknowledged (cf. Vachek 1966; Daneš 1982). Within the realm of phonological analyses, Fries and Pike (1949) introduce the idea of “coexistent phonemic systems”, applied at first to the speech of monolingual speakers (“two or more phonemic systems may coexist in the speech of a monolingual”, Fries and Pike 1949: 29), while Pilch (1965), in a very brief contribution, sketches a center–periphery distinction for the phonetic level of a language system. In more recent formal approaches to language theory such as Chomsky (1981, 1986), a core grammar and a “periphery of marked elements and constructions” are clearly set apart:

32 In EP the alveolar ([z]) or the postalveolar ([ʃ/ʒ]) realization of the plural marker ‑s in articles, pronouns and adjectives depends on the vocalic or non-vocalic nature of the word-initial segment following it; if followed by a vowel it is syllabified as the head of the following word-initial syllable. French has enchaînement and liaison both resulting in the syllabification of a preceding vowel as the head of the following syllable (in the case of liaison, a silent vowel, underlyingly represented; cf. Encrevé 1988, who also studies cases of liaison without enchaînement; for EP and French cf. Auer 1993: 25–29, 50).

100 

 Matthias Heinz

[E]ach actual “language” will incorporate a periphery of borrowings, historical residues, inventions, and so on, which we can hardly expect to – and indeed would not want to – incorporate within a principled theory of UG [Universal Grammar]. For such reasons as these, it is reasonable to suppose that UG determines a set of core grammars and that what is actually represented in the mind of an individual even under the idealization to a homogeneous speech community would be a core grammar with a periphery of marked elements and constructions. Viewed against the reality of what a particular person may have inside his head, core grammar is an idealization. (Chomsky 1981: 7–8)33

The description of the nature of peripheral elements within a language, “borrowings, historical residues, inventions”, is very much to the point, but while the focus of theoretical linguistic approaches is almost exclusively on what Chomsky calls core grammar, the relevance of the periphery to the functioning of the system may be underestimated.34 Indeed, in my view, for a thorough understanding of how language systems work, both their peripheral and their core components need to be taken into account. The center–periphery perspective bears upon the recognition of the intricate interdependencies between synchrony and diachrony and internally-organized grammatical algorithms – syllabification rules would formally be seen as essentially an algorithm with very few exceptions in the output – vs. textual instantiations of certain patterns belonging to peripheral areas of linguistic subsystems. An increase of variation is seen as one characteristic of the periphery of grammar: “[T]he proportion of variable phenomena increases the closer one approaches the ‘periphery’ of the grammar, hence: syntax < morphology < phonology < phonetics […], although the more peripheral language components are, of course, never entirely variable” (Hinskens 1998: 160). The manifold implica-

33 The wording of a generative definition of core grammar vs. periphery from a linguistic encyclopedia is the following: “Core-grammar is that part of the relatively stable (steady) state of the language faculty (i.e. of the adult I-language) that results from the setting of parameters in UG (the initial state of the language faculty, S0), as opposed to the periphery, which consists of additional, marked, language-specific rules and exceptions” (taken from Glottopedia; cf. also Chomsky 1986: 147–149, 221). 34 The view expressed by Chomsky clearly reflects the tenets of generative theorizing on Universal Grammar as the basis of the human language faculty, the discussion of which would by far exceed the focus of this paper; here no debate of the further implications of the generative approach is intended. I would simply like to draw attention to the basic distinction of central and peripheric areas in language systems, an aspect that can be traced back to both Prague structuralist and to generative modeling of language system organization although it has not, in my view, received the attention it deserves. Cf. also Bertinetto’s (2003) reflections on centro and periferia; the author, however, refers rather to the heuristic function of the broader epistemological figure.



Syllable complexity in the diachrony of Romance languages 

 101

tions of Hinskens’ ordering of descriptive levels cannot be discussed here, but it comes as no surprise that the greatest variability of realization is said to occur on the phonetic level, becoming more restricted on the phonological level, which still displays more variation than the other levels. Summing up the above, in this view a language system can be described within a center–periphery model where 1) the central area of a language system incorporates a core of universal or near-universal (regarding frequency of occurrence) principles and regularities as well as further conditions (or constraints) defining grammatical features of that specific language (which it may or may not share with some other languages); 2) the peripheral area is characterized by additional structural features which are part of the system as a whole but often constitute contrary tendencies, irregularities or rule exceptions.

From this center vs. periphery view in syllable typology an analogy can be drawn to the division of the lexicon proposed by Itô and Mester (1995) and others who subdivide the lexicon of a language into a core of native words and a periphery of assimilated, fully integrated loanwords and unadapted loanwords further on the outside. This view, which takes a center–periphery organization of language-specific inventories of basic syllabic patterns as its point of departure, fits with the idea of a typological continuum of languages giving preference, to varying degrees, to the syllable or the phonological word as the principal rhythmic domains. Obviously, what is peripheral in one system may be central in another. Thus the central area of the syllable-type inventory is to be conceived of as considerably more extensive in a word language with its characteristic syllable structure complexities than in a syllable-language variety, where simple syllable structures prevail across the board (therefore making the peripheral positioning of particular instances of complex syllable types within their inventories all the more plausible). For a typical syllable language (again, Spanish can be seen as largely representative of this type) this kind of scheme could be illustrated as follows: (9) CCCV → (CCV (CVC (CV) V, VC) VCC) ← VCCC

At the very center of this subcomponent of grammar a universal core is situated, which according to most ­views on syllabic typology is constituted by the universally attested CV type (cf. Trask 1996: s.v. universal syllable type; MacNeilage and Davis 2005): (10) … (CV) …

and a somewhat larger (depending on individual constraints on syllabic well-formedness and simplicity) language-specific central area:

102 

 Matthias Heinz

(11) … (CVC (CV) V, VC) …

In a language like Spanish, the peripheral area consists of a transitional zone containing loan elements and thus including certain syllabic patterns that seem to be “foreign” to the central areas of the respective subsystem (made up of cross-linguistic or even universal as well as language-specific “constraints”, “parameters”, “features” or whichever concept of hierarchical organization is chosen). Loan phonology can give information about the systemic status of peripheral elements: Realizations which in a first phase may be no more than citation forms in the use of individual speakers (forms produced by nonce borrowing) can eventually gain phonemic status in a language. An example of this is the phoneme /ŋ/ in French. It occurs in borrowings like parking, shopping and a few minimal pairs like Fr. chopine ‘half-pint’ vs. shopping ‘(window-)shopping’ can be found (cf. Hess 1975: 74–80; Müller 1975). Hinskens (1998: 183–184) describes how the spreading of a sound change in dialects can be induced by borrowed items: “[B]orrowed sound changes will usually be embedded in specific loaned items or morphemes. Initially a sound change thus borrowed will probably entrench itself for a while in the borrowing dialect in the loan words, before starting to spread in a lexically diffuse fashion”. Rather than simply summing up the inventory of syllable types, the phonotactic subsystem in a given language could then be visualized as a center–periphery configuration (the arrows symbolize the effect of external influences such as prolonged language contact resulting in structural transfer): (12) … →

(CCV (CVC

(CV)

V, VC) VCC)

←…

This ordering of elements is illustrated here for syllabic typology as one important parameter beside others in the syllable vs. word rhythm classification. Syllabic typology has been singled out because of its strong influence on the rhythmic shape of many Romance languages.35 The assumption of a center-periphery continuum should not, of course, be applied deterministically in the description of a language system and its components; still, it may be useful for other typological parameters. For the question can be asked throughout a given prosodic system which parameters belong to the central area, and which are part of the periphery. Given their individual combinations in actual languages, a third possibility exists, i.e. some parameters may not be applicable in a specific language system so they cannot be assigned to any of the areas. The phonological word as the

35 In accordance with most accounts of rhythmic typology, accentual constraints seem to play a greater role in Germanic languages such as English and (Modern) German.



Syllable complexity in the diachrony of Romance languages 

 103

relevant domain for the word-language type could then be posited as “not applicable” or systemically peripheral in languages conforming more or less to the Silbensprache as opposed to the Wortsprache type (sensu Auer 2001: 1395–1398 and Szczepaniak 2007: 27–56 and passim). Indeed, in a syllable language, constraints governing the forming of simple syllable structures (preferably of the CV type) may “override” any condition for preserving or strengthening word boundaries; witness the resyllabification across morphemic and lexical boundaries common in most Romance languages (cf. section 3.1 and note 33 above). Consequently, in a word language, the phonological word and the structural constraints it may entail are systemically central; here processes profiling and optimizing the word domain are favored over, or can block, those operating on smaller domains (namely syllables and segments). Again, some parameters may not be applicable in some varieties or show up in the periphery only in others. Especially archaisms, dialectalisms and loanwords are likely to be part of the peripheral elements countering central typological tendencies of a system.36 Such a center–periphery view explains, to take up Spanish again for the sake of illustration, a phenomenon in the phonetics and (allophonic) phonology of the so-called tierras altas37 (Highland) varieties of American Spanish, e.g. in Peru or Mexico. In these varieties, certain syllable structures are common which contain consonant clusters otherwise, in Peninsular Standard Spanish, found only in carefully and/or emphatically pronounced cultismos, i.e. learned borrowings of Greek or Latin origin, whose pronunciation is clearly influenced by spelling. Compare a careful lento-style realization of Sp. exacto with the allegro form, delivered at a higher speech rate and/or with less controlled pronunciation:38 (13) Pen. Sp. exacto ‘exact’ lento [ek.ˈsak.to], [eɣ.ˈsak.to] vs. allegro [e.ˈsa.to]

On the other hand, the tendency to reduce and elide vowels while preserving the consonants can spawn complex clusters in the Highland (tierras altas) varieties of Spanish in Peru and Mexico, e.g. in a Mexican realization like

36 Cf. Szczepaniak (2007) on phonological-typological features of Swiss German (317–325) and of certain loan elements (335) diverging from the general word-language pattern evident in (Modern) Standard German. 37 On the distinction of tierras altas vs. tierras bajas (Henríquez Ureña 1921) or tierras interiores/ de la corte vs. tierras marítimas/de la flota (Menéndez Pidal 1962) as dialectal areas of American Spanish, cf. Sosa’s (2000: 487–491) overview. 38 On different realizations of in Spanish speech cf. Navarro Tomás (1967: 140), on general properties (also phonetic) of cultismos and semicultismos cf. García Gallarín (2007: 51).

104  (14)

 Matthias Heinz

a. bloques para apuntes ‘notepad’ Mex. Sp. (tierras altas) allegro: [ˈbloks.ˈpa.ra.ˈpunts]39 b. bloques para apuntes Pen. Sp. normal/allegro: [ˈblokes.ˈpa.ra(ː).ˈpuntes]

Thirdly, on the far end of a continuum of phonetic realizations we find the tendencies of Southern Peninsular and Lowland (tierras bajas) varieties of Spanish towards deleting final consonants, bringing about a higher number of open syllables and ultimately optimizing syllables in the direction of the universal CV template: (15) bloques para apuntes Southern-Pen./Lowland Sp. [ˈbloke(h).ˈpa.ra(ː).ˈpunte(h)]

The reason for the difference lies in the diverging status of formally similar realizations: In American Highland Spanish the resulting complex syllable types are systemically more central due to the allophonic weakening of vowels, in Peninsular Standard Spanish (and even more so in Southern Peninsular and American Lowland varieties) they are part of the peripheral areas of the system – and can be merely potential or rare realizations belonging to particular diaphasic registers; here the central position is occupied by simple (“core”) syllable types.40

4 Different ways of coping with syllable shell complexity So the behavior of (at least Romance) languages with regard to syllabic complexity emerges as a typological parameter providing evidence (alongside other criteria) for classifying those varieties within a prosodic typology. In languages with a preference for optimizing syllable structure the resulting simple types are examples of the outcome of differing strategies of cluster simplification:

39 This relajación vocálica, a weakening of unstressed vowels with devoicing and eventual vowel deletion, has been described as a common feature of Mexican (Highland) Spanish (though it is subject to variation and diatopical differentiation in distribution), cf. Lope Blanch (1963/64), Kabatek (1994) and Moreno de Alba (2002: 31–41). In many cases, as in (14a), the resulting clusters are word-final; still, word-initial vowels may also undergo the weakening process, cf. forms like the pronoun usted [(u)ˈsteð] ‘you (2nd person sg., polite form of address)’. 40 Referring to the pronunciation of as [ks] Navarro Tomás (1967: 140) speaks of “casos muy marcados de dicción culta o enfática” [strongly marked cases of educated or emphatic pro­ nunciation] as opposed to the “conversación corriente” [colloquial conversation].



Syllable complexity in the diachrony of Romance languages 

 105

(16) Sp. (many varieties) absoluto ‘absolute’ [a.so.ˈlu.to] (17) BP absoluto [ˌɐ.bi.so.ˈlu.tu]

In the case of many varieties of Spanish (notably the Southern Peninsular and American Lowland varieties) the preferred phonological process is the reduction of C clusters (here the consonant deletion is the result of regressive total assimilation of the type illustrated in example 2, i.e. CiCj > Cj), whereas in BP cluster resolution by epenthesis (insertion of [i]) is a common process. In those varieties the tendency to reduce shell complexity is pervasive and can affect even the more peripheral areas with learned lexical items carrying syllabic complexity. In both French and EP, on the other hand, the weakening of unstressed vowels is clearly a central process applying systematically and leading to increased syllabic complexity in fast, fluent speech (as shown also above, examples 7 and 8): (18) a. lento: Fr. qu’est-ce que tu fais? ‘what are you doing?’ [kɛs.kə.ty.ˈfɛ] EP desprezar ‘to despise’ [dɨ.ʃpɾɨ.ˈzaɾ] b. allegro: Fr. [ksty.ˈfɛ] EP [dʃpɾ̩.ˈzɐɾ]

As mentioned above (3.1), in EP (and to a lesser extent in BP), Spanish and French, learned borrowings are another source of complex syllable types. Drawing from each of those sources of syllabic complexity (allegro forms and learned lexical items), the inventories of French and EP present a particularly wide spectrum from simple to complex syllable types, whereas in Peninsular Spanish this spectrum is much more limited. In conclusion, two types of syllable shell complexity can be observed in the Romance languages analyzed here. On the one hand, there is complexity in a synchronically stable periphery of the syllable type inventories (induced by loanword phonology, lexically fixed in learned borrowings) in Spanish, French and Portuguese (EP, BP). On the other hand, in some languages, e.g. in EP and French, complexity is the result of central processes (vowel deletion) in the respective system. According to prototypical syllable structural parameters of the syllable– word continuum, syllable rhythm is defined, among other things, by (1) simple syllable types (CV), (2) well-defined, speech-rate independent syllable boundaries and (3) adherence to the sonority hierarchy. In light of these parameters and focusing on the aspect of shell complexity, the least syllable-oriented cases are French and EP. The diverging dynamics within the Gallo- and Ibero-Romance varieties discussed here is sketched in Figure 1:

106 

 Matthias Heinz

Syllable types (à preferential outcome of speech-style-related processes)

Allegro

Lento

simple

Southern Pen. and tierras bajas Sp., BP

EP, Fr.

complex

EP, Fr., tierras altas Sp. [+central]

Southern Pen. and tierras bajas Sp., BP [–central]

Figure 1: Syllable types resulting from lento vs. allegro style realizations

The single varieties behave differently with regard to the distribution of simple and complex syllable types in lento and allegro (faster speech rate, less careful articulation) realizations. This has consequences for the central or peripheral status of the resulting syllable structures. While unmarked simple syllable types are to be expected and indeed found in all of the varieties and in all speech styles, they are the preferential outcome of allegro style in Southern Peninsular/tierras bajas Spanish and BP (roughly: the faster the speech, the more syllable structure is optimized towards the CV template, mostly by way of weakening and eventually deleting coda segments in both stressed and unstressed syllables, cf. examples 13 and 15). In EP and French, on the other hand, the lento style favors the realization of a higher number of vowels (cf. the lento form in 18). The latter form nuclei of simpler syllable structures compared to those produced when unstressed vowels are omitted and the remaining consonantal material is compressed into consonant clusters of sometimes remarkable complexity. It is with respect to the complex types, then, that a divergence between central vs. peripheral phenomena can be perceived. The [+central] and the [–central] traits are meant to characterize the difference between complex patterns found in Standard41 French and EP on the one hand – complex types resulting from systemically central tendencies as the normal outcome in informal allegro speech  –  and Spanish (Southern Peninsular, tierras bajas varieties) and BP on the other – complex types having a peripheral position in the system because they mainly occur in careful or emphatic realizations of loanwords. The former

41 In southern regional varieties of French, the pronunciation of schwa is preserved in many more contexts than in (northern) Standard French (for a concise description cf. Pustka 2011: 196–197).



Syllable complexity in the diachrony of Romance languages 

 107

two languages display a neat contrast between (highly elaborate) stressed and (highly reduced and centralized) unstressed vowel systems resulting in frequent reduction of unstressed vowels (including, but not limited to, many instances of schwa), whereas on the opposite end of the spectrum Spanish – BP being a somewhat less clear-cut case – has a pentavocalic system of distinct uniformity and simplicity.

References Abercrombie, David (1967): Elements of General Phonetics. Edinburgh: Edinburgh University Press. Allen, W. Sidney (1973): Accent and Rhythm. Prosodic Features of Latin and Greek: A Study in Theory and Reconstruction. Cambridge: Cambridge University Press. Alvar, Manuel (ed.) (1996): Manual de dialectología hispánica. El español de España. Barcelona: Ariel. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Bertinetto, Pier Marco (2003): ‘Centro’ e ‘periferia’ del linguaggio: una mappa per orientarsi. In: Daniele Maggi and Diego Poli (eds.), Modelli recenti in linguistica. Atti del Convegno della Società Italiana di Glottologia, Macerata, 26–28 ottobre 2000, 157–211. Roma: Il Calamo. Blevins, Juliette (1995): The syllable in phonological theory. In: John Goldsmith (ed.), The Handbook of Phonological Theory, 206–244. (Blackwell Handbooks in Linguistics 1.) Oxford: Blackwell. Blevins, Juliette (2004): Evolutionary Phonology: The Emergence of Sound Patterns. Cambridge: Cambridge University Press. Blevins, Juliette (2006): Syllable: Typology. In: Edward Keith Brown (ed.), Encyclopedia of Language and Linguistics, vol. 12, 2nd ed. 333–337. Oxford: Elsevier. Blevins, Juliette (2010): Review of ‘Syllable structure: The limits of variation’ by San Duanmu. Linguistic Typology 14/2–3: 287–291. Brown, Earl K. (2009): A Usage-Based Account of Syllable- and Word-Final /s/ Reduction in Four Dialects of Spanish. München: Lincom. Bybee, Joan L. (2001): Phonology and Language Use. (Cambridge Studies in Linguistics 94.) Cambridge: Cambridge University Press. Bybee, Joan L. (2006): From usage to grammar. The mind’s response to repetition. Language 82/4: 711–733. Catalán, Diego (1971): Entorno a la estructura silábica del español de ayer y del español de mañana. In: Eugenio Coseriu and Wolf-Dieter Stempel (eds.), Sprache und Geschichte. Festschrift für Harri Meier zum 65. Geburtstag, 77–110. München: Fink.

108 

 Matthias Heinz

Chomsky, Noam (1981): Lectures on Government and Binding. (Studies in Generative Grammar 9.) Dordrecht: Foris. Chomsky, Noam (1986): Knowledge of Language: Its Nature, Origin and Use. New York: Praeger. Clements, George N. and Samuel Jay Keyser (1983): CV Phonology. A Generative Theory of the Syllable. (Linguistic Inquiry Monographs 9.) Cambridge, Mass.: MIT Press. Colina, Sonia (2009): Spanish Phonology: A Syllabic Perspective. Washington, D.C.: Georgetown University Press. Cunha, Conceição (2011): Die Perzeption von lexikalischen Plosivclustern und CVC-Sequenzen im Portugiesischen. In: Conceição Cunha, Sylvia Jaki, Tabea Reiner and Ursula Stangel (eds.), Sprachvariation im europäischen Kontext – Language Variation in Europe and Beyond, 71–88. (Journal LIPP 1.) Available at http://ojs.ub.uni-muenchen.de/index.php/ lipp/index. Daneš, František (1982): Zur Theorie des sprachlichen Zeichensystems. In: Jürgen Scharnhorst and Erika Ising (eds.), Grundlagen der Sprachkultur. Beiträge der Prager Linguistik zur Sprachtheorie und Sprachpflege, vol. 2, 132–173. (Sprache und Gesellschaft 8/2.) Berlin: Akademie-Verlag. De Yzaguirre, Lluís (1995): L’estructura sil·làbica del català central. Ph.D. dissertation, Departament de Filologia Catalana, Universitat de Barcelona. Available at http://latel.upf. edu/morgana/altres/cibres/tesi_lym/estrucsil.htm. Duanmu, San (2009): Syllable Structure. The Limits of Variation. Oxford: Oxford University Press. Dufter, Andreas (2004): Ist das Französische eine silbenzählende Sprache? In: Trudel Meisenburg and Maria Selig (eds.), Nouveaux départs en phonologie. Les conceptions subet suprasegmentales, 139–159. Tübingen: Narr. Encrevé, Pierre (1988): La liaison avec et sans enchaînement. Phonologie tridimensionnelle et usages du français. Paris: Seuil. Fries, Charles C. and Kenneth L. Pike (1949): Coexistent phonemic systems. Language 25: 29–50. Frota, Sonia, Marina Vigário and Fernando Martins (2006): FreP: An electronic tool for extracting frequency information of phonological units from Portuguese written text. In: Proceedings of the 5th International Conference on Language Resources and Evaluation, Genoa, 22–28 May 2006, 2224–2230. Available at http://www.lrec-conf.org/proceedings/lrec2006. García Gallarín, Consuelo (2007): El cultismo en la historia de la lengua española. Madrid: Parthenon. Geisler, Hans (1992): Akzent und Lautwandel in der Romania. (Romanica Monacensia 38.) Tübingen: Narr. Glottopedia [http://www.glottopedia.org]. Guerra, Rafael (1983): Estudio estadístico de la sílaba en español. In: Manuel Esgueva and Margarita Cantarero (eds.), Estudios de fonética, vol. 1, 9–112. (Collectanea Phonetica 7.) Madrid: Consejo Superior de Investigaciones Científicas. Heinz, Matthias (2008): La estructura silábica en la historia del español: Aspectos cualitativos y cuantitativos. In: Concepción Company Company and José G. Moreno de Alba (eds.), Actas del VII Congreso Internacional de Historia de la lengua española, Sección temática: Fonética histórica, fonología histórica y grafemática, Mérida (Yucatán), 4–8 septiembre 2006, vol. 1, 275–291. Madrid: Arco Libros. Heinz, Matthias (2010a): La diachronie des structures syllabiques en espagnol et en catalan. Analyses quantitatives et textuelles. In: Maria Iliescu, Paul Danler and Heidi Siller-Rung-



Syllable complexity in the diachrony of Romance languages 

 109

galdier (eds.), Actes du XXVe Congrès International de Linguistique et de Philologie Romanes, Section 4: Description historique et contrastive des langues romanes: Phonétique et phonologie, Innsbruck, 2–8 septembre 2007, vol. 2, 97–106. Berlin/New York: Walter de Gruyter. Heinz, Matthias (2010b): Zentrum und Peripherie in romanischen Silbentypsystemen. Habilitationsschrift (Habilitation thesis), Universität Tübingen. Henríquez Ureña (1921): Observaciones sobre el español de América. Revista de Filología Española 7: 357–390. Hess, Zita (1975): Typologischer Vergleich der romanischen Sprachen auf phonologischer Basis. (Heidelberger Beiträge zur Romanistik 8.) Bern/Frankfurt a.M.: Lang. Hinskens, Frans (1998): Variation studies in dialectology and three types of sound change. Sociolinguistica 12: 155–193. Hockett, Charles (1947): Problems of morphemic analysis. Language 23/4: 321–343. Itô, Junko and Armin Mester (1995): The core-periphery structure of the lexicon and constraints on reranking. In: Jill Beckmann, Laura Walsh Dickey and Suzanne Urbanczyk (eds.), Papers in Optimality Theory, 181–209. Amherst, MA: Graduate Linguistic Student Association. Jakobson, Roman (1962): Phonological Studies. 2nd ed. (Roman Jakobson / Selected Writings 1.) The Hague: Mouton. Jespersen, Otto (1904): Lehrbuch der Phonetik. Leipzig/Berlin: Teubner. Kabatek, Johannes (1994): México frente a Madrid: Aspectos fonéticos del habla de los taxistas en dos capitales hispanas. Iberoamericana 54/2: 5–15. Kiss, Sándor (1971): Les transformations de la structure syllabique en latin tardif. Debrecen: Kossuth Lajos Tudományegyetem. Krötsch, Monique (2004): La formation de groupes de consonnes au début des unités phoniques en français parlé. In: Trudel Meisenburg and Maria Selig (eds.), Nouveaux départs en phonologie. Les conceptions sub- et suprasegmentales, 217–233. Tübingen: Narr. Lahiri, Aditi, Thomas Riad and Haike Jacobs (1999): Diachronic prosody. In: Harry van der Hulst (ed.), Word Prosodic Systems in the Languages of Europe, 335–422. (Empirical Approaches to Language Typology 20/4.) Berlin/New York: Mouton de Gruyter. Langacker, Ronald W. (1991): Foundations of Cognitive Grammar. Vol. 2: Descriptive Application. Stanford, CA: Stanford University Press. Langacker, Ronald W. (1994): Culture, cognition, and grammar. In: Martin Pütz (ed.), Language Contact and Language Conflict, 25–53. Amsterdam/Philadelphia: John Benjamins. Ledgeway, Adam (2009): Grammatica diacronica del Napoletano. (Beihefte zur Zeitschrift für romanische Philologie 350.) Tübingen: Niemeyer. Lehmann, Christian (2005): Latin syllable structure in typological perspective. In: Gualtiero Calboli (ed.), Latina lingua! Proceedings of the Twelfth International Colloquium on Latin Linguistics, Bologna, 9–14 June 2003, vol. 1, 127–147. Roma: Herder. Loiseau, Sylvain (2010): Les paradoxes de la fréquence. Energeia 2: 20–55. Lope Blanch, Juan M. (1963/64): En torno a las vocales caedizas del español. Nueva Revista de Filología Hispánica 27: 1–19. Loporcaro, Michele (2011): Syllable, segment and prosody. In: Martin Maiden, John Charles Smith and Adam Ledgeway (eds.), The Cambridge History of the Romance Languages. Vol. 1: Structures, 50–108. Cambridge: Cambridge University Press. Lowenstamm, Jean (1996): CV as the only syllable type. In: Jacques Durand and Bernard Laks (eds.), Current Trends in Phonology. Models and Methods, vol. 2, 419–441. Salford: European Studies Research Institute.

110 

 Matthias Heinz

Lüdtke, Helmut (1986): Esquisse d’une théorie du changement langagier. La linguistique 22/1: 3–46. MacNeilage, Peter and Barbara L. Davis (2005): A cognitive-motor syllable frame for speech production: Evidence from neuropathology. In: William J. Hardcastle and Janet Mackenzie Beck (eds.), A Figure of Speech. A Festschrift for John Laver, 129–144. Mahwah, NJ: Erlbaum. Maddieson, Ian (2005): Syllable structure. In: Martin Haspelmath, Matthew S. Dryer, David Gil and Bernard Comrie (eds.), The World Atlas of Language Structures (WALS), 54–57. Oxford: Oxford University Press. Mateus, Maria Helena and Ernesto D’Andrade (2000): The Phonology of Portuguese. (The Phonology of the World’s Languages.) Oxford: Oxford University Press. Menéndez Pidal, Ramón (1962): Sevilla frente a Madrid: Algunas precisiones sobre el español de América. In: Diego Catalán (ed.), Estructuralismo e historia, 99–165. (Miscelánea homenaje a André Martinet 3.) La Laguna: Universidad de La Laguna. Menéndez Pidal, Ramón (ed.) (1977): Primera crónica general de España. (Fuentes cronísticas de la historia de España 1.) Madrid: Gredos. Moreno de Alba, José G. (2002): La pronunciación del español en México. México: El Colegio de México. Müller, Bodo (1975): Phonologie und Purismus. In: Wolfgang U. Dressler and František Václav Mareš (eds.), Phonologica 1972. Akten der zweiten Internationalen Phonologie-Tagung, Wien, 5.–8. September 1972, 353–358. München: Fink. Navarro Tomás, Tomás (1967): Manual de pronunciación española. 13th ed. Madrid: Consejo Superior de Investigaciones Científicas. Noske, Roland (1993): A Theory of Syllabification and Segmental Alternation. With Studies on the Phonology of French, German, Tonkawa, and Yawelmani. (Linguistische Arbeiten 296.) Tübingen: Niemeyer. Pilch, Herbert (1965): Zentrale und periphere Lautsysteme. In: Eberhard Zwirner and Wolfgang Bethge (eds.), Proceedings of the Fifth International Congress of Phonetic Sciences, Münster, 16–22 August 1964, 467–473. Basel/New York: S. Karger. Prince, Alan and Paul R. Smolensky (2004): Optimality Theory: Constraint Interaction in Generative Grammar. Oxford: Blackwell. Pulgram, Ernst (1970): Syllable, Word, Nexus, Cursus. (Janua Linguarum / Series Minor 81.) The Hague: Mouton. Pustka, Elissa (2011): Einführung in die Phonetik und Phonologie des Französischen. Berlin: Erich Schmidt. Reinheimer Rîpeanu, Sanda (ed.) (2004): Dictionnaire des emprunts latins dans les langues romanes. Editura Academiei Române: Bucureşti. Restle, David and Theo Vennemann (2001): Silbenstruktur. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1310–1336. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Richter, Elise (1977): Chronologische Phonetik des Französischen bis zum Ende des 8. Jahrhunderts. In: Yakov Malkiel (ed.), Kleinere Schriften zur allgemeinen und romanischen Sprachwissenschaft, 353–373. (Innsbrucker Beiträge zur Sprachwissenschaft 21.) Innsbruck: Institut für Sprachwissenschaft. Sampson, Rodney (2010): Vowel Prosthesis in Romance. A Diachronic Study. Oxford: Oxford University Press.



Syllable complexity in the diachrony of Romance languages 

 111

Scheer, Tobias (2004): A Lateral Theory of Phonology. What Is CVCV, and Why Should It Be? (Studies in Generative Grammar 68/1.) Berlin: Mouton de Gruyter. Schmid, Stephan (1999): Fonetica e fonologia dell’italiano. Torino: Paravia. Schmid, Stephan (2000): Tipi sillabici in alcuni dialetti dell’Italia centro-meridionale. In: Annick Englebert, Michel Pierrard, Laurence Rosier and Dan Van Raemdonck (eds.), Actes du XXIIe Congrès international de linquistique et philologie romanes, Bruxelles, 23–29 juillet 1998. Vol. 3: Vivacité et diversité de la variation linguistique, 343–350. Tübingen: Niemeyer. Sievers, Eduard (1876): Grundzüge der Lautphysiologie: Zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. (Bibliothek indogermanischer Grammatiken 1.) Leipzig: Breitkopf & Härtel. Sievers, Eduard (1901): Grundzüge der Phonetik: Zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. 5th ed. (Bibliothek indogermanischer Grammatiken 1.) Leipzig: Breitkopf & Härtel. Sosa, Juan Manuel (2000): Sobre el consonantismo, el vocalismo y la entonación en el español de América. Zeitschrift für romanische Philologie 116/3: 487–509. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Trask, R. Larry (1996): A Dictionary of Phonetics and Phonology. London/New York: Routledge. Vachek, Josef (1966): Les problèmes du centre et de la périphérie du système de la langue. (Travaux linguistiques de Prague 2.) Praha: Academia. Väänänen, Veikko (1981): Introduction au latin vulgaire. 3rd ed. Paris: Klincksieck. Vennemann, Theo (1988): Preference Laws for Syllable Structure and the Explanation of Sound Change. With Special Reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. Wioland, François (1985): Les structures syllabiques du français. Fréquence et distribution des phonèmes consonantiques. Contraintes idiomatiques dans les séquences consonantiques. (Travaux de linguistique quantitative 31.) Genève: Slatkine. Wright, Roger (1994): Logographic script and assumptions of literacy. In: Roger Wright (ed.), Early Ibero-Romance. Twenty-one Studies on Language and Texts from the Iberian Peninsula between the Roman Empire and the Thirteenth Century, 165–180. Newark, Delaware: Juan de la Cuesta.

Barış Kabak (University of Würzburg)

Pervasive syllables and phonological unity in words Abstract: In light of diachronic, synchronic, and psycholinguistic evidence, this paper shows that syllables are pertinacious units of speech in Korean and Turkish, which are characteristically syllable based. It is demonstrated that syllable-based generalizations in these languages not only shape morpho-phonological representations but also influence the way speech strings are perceived. Furthermore, they determine the fate of morphemes in diachronic change. Despite the primacy of syllables in these languages, it is also the case that they are replete with phonological regularities that are crucially linked to the word (and sometimes its syntactic category information), and that they exhibit static sound regularities that are asymmetrically distributed to word edges. The fact that several phonological phenomena in both Korean and Turkish blur morphological boundaries also comes with a caveat: These regularities simultaneously strive to achieve unity and coherence within prosodic units that are often larger than a lexical word in these languages. Accordingly, it is argued that such processes should be construed as a type of strengthening process since cohesion within a particular prosodic domain, presumably the Phonological Word, creates harmony and unity within that domain, thereby constituting vital signs of wordhood. On the basis of evidence from language processing, phonological unity that arises from regularities such as vowel harmony and stress is shown to facilitate word segmentation for the language user due to their demarcative function. It is suggested that the challenge in the delineation of word- and syllable-based generalizations, as well as evidence that seemingly contradicts the syllable vs. word rhythm typology, are due to crosslinguistic differences concerning the nature of words, as well as the pivotal role prosodic domains and the prosodic hierarchy play in organizing speech units.

1 Introduction In languages that employ syllable-based prosodies, rather than word-based prosodies, the syllable is expected to be the primary prosodic category that underlies the majority of phonological regularities. Beyond its use as a typological utility in characterizing prosodic systems across the languages of the world, the broader implications of the typology proposed by Auer (1993) for the overall phonological system of individual languages and for the language learner and user are



Pervasive syllables and phonological unity in words 

 113

understudied in the literature. Furthermore, the extent to which syllable-based generalizations can coexist with patterns and regularities that refer to other prosodic domains within predominantly syllable-based languages and what that concurrence implies for mental representations and diachronic change remain to be explored. This paper aims to unpack these two understudied research questions in the context of languages that are pervasively syllable based.1 I will first focus on the synchronic, diachronic, and psycholinguistic repercussions of employing a syllable-based prosody for the phonological system as a whole. In particular, I will explore the consequences of being characterized as a syllable language on the nature and dynamics of phonological patterns and regularities, and the way sounds and sound sequences are represented and processed. I hold the view that the extent to which a language can be classified as a syllable- or a word-based language (or a mixture of the two) should emerge from the way syllable-based (or word-based) generalizations shape mental representations in both synchronic and diachronic terms in that language. Furthermore, I will argue that the primacy of syllables that emerges in multiple layers of linguistic organization does not and should not exclude the relevance of other prosodic units to the phonological system of that language. More specifically, I will show that the Phonological Word (PW), which is presumably the vital prosodic unit in “word-based” languages, also constitutes an important prosodic domain in the “syllable-based” languages under exploration here, despite clear signs of syllable dominance. I suggest that when a language exhibits a “mixed” behavior on the surface, this is not necessarily evidence of a diachronic cline from a syllable language to a word language (or vice versa). Rather, it should be construed as evidence of the pivotal role prosodic domains and the prosodic hierarchy play in organizing speech units. Delineation between word-based and syllable-based phonologies often poses a challenge to an analyst because the notion of the PW, as opposed to the notion of the syllable, is vaguer, and it tends to be language specific.2 The observation that prosodic constituents such as the PW are not necessarily isomorphic to morphosyntactic units such as the word has been the impetus behind many prosodic phonology models (e.g. Nespor and Vogel 1986). Accordingly, PWs in some lan-

1 On a par with Auer’s (1993: 102‒111) revised syllable vs. word language typology, I assume that it is the prosodic domains rather than rhythm that are at the core of prosodic typology. Hence, I do not deal with the isochrony hypothesis in this paper. 2 This is not to say that there is no controversy surrounding the notion of the syllable and its universality. For a recent proposal on constraints on syllable size, see Duanmu (2009), and also Blevins (2010) for a review of this new proposal.

114 

 Barış Kabak

guages are often larger than morphological words (see Revithiadou 2011 for an in-depth review of notions and ambiguities surrounding the PW). Hence, what looks like a syllable-related phenomenon can easily be interpreted as a PW-related phenomenon since syllables are subsumed under Phonological Words in the prosodic hierarchy. Furthermore, the syllable languages under investigation protect the integrity of morphemes by way of syllable protection; at the same time they often create cohesion within a cluster of morphosyntactic units by way of predictable allomorphic variation that encapsulates the whole construction, thereby forming a single prosodic constituent. Hence, the rules that mimic or extend syllable-based generalizations can coexist with or be isomorphic to those that simultaneously strive for unity and coherence within these larger prosodic domains. In what follows, I will first examine some of the relevant phonological properties of Korean and Turkish (section 2). Korean and Turkish constitute the archetypes of agglutinative morphology (though see Kabak and Plank 2006 for flexive tendencies in both of these languages) and are argued to be between a word and a syllable language, “where both the syllable and the phonological word are important for their phonological make-up” (Auer 1993: 110). To bolster this observation, I will present some phonological processes that arguably target syllables in these languages, and also highlight where phonological regularities are bound by words or word edges. Section 3 will discuss what I refer to as “emergent” phonological phenomena that are evident in synchronic and diachronic variation as well as in processing related phenomena; these arguably constitute the most vital pieces of evidence to characterize the prosodic typology of a language as they determine the nature of mental representations within and across generations. The first phenomenon under investigation will be emergent segments (vowels and consonants) as a result of syllabically-conditioned phonological parsing, both in perception and production. Second, I will discuss how vowel harmony and fixed, non-lexical stress, which are typically found in syllable languages,3 can be used as cues for word segmentation. Finally, I will show how the evolution of bound morphemes and the lexicalization of phonological processes (e.g. lexical accents) can be shaped by syllable-based generalizations.

3 It should be noted that, due to its deliminative function, vowel harmony is difficult to unambiguously correlate with syllable-based languages (Auer 2001: 1397). Although vowel harmony may equalize the syllable within a word, in another language it may lead to asymmetrical vowel inventory (e.g. German), making the syllable within the word unequal (Szczepaniak, personal communication). Although this certainly constitutes an important typological difference, it is beyond the scope of the current paper.



Pervasive syllables and phonological unity in words 

 115

2 A syllable or a word language? 2.1 Korean The syllable not only lies at the heart of the Korean phonological system but also underlies its centuries-old indigenous writing system, Hangul, which is based on morphophonemic syllables. The characterization of Korean presented below is based on Sohn’s (1999) description of the Korean sound system. The great majority of phonotactic restrictions in Korean is a result of the constraints on the syllable structure, which is relatively simple. Although the on-glide diphthongs complicate the internal structure of the nucleus to some extent, the language can be said to have CV and CVC syllables on the surface. It should be noted that morphophonemic representations of many words and affixes in Korean, and this also applies to Turkish as we will see below, may involve more complex phonemic sequences than what is eventually allowed to surface. Korean employs several obligatory phonological processes that refer to the syllable structure. The coda neutralization process is one such phenomenon, where all tense (indicated by ) and aspirated stops become plain, and /s, sˈ, c, ch, and h/ are neutralized to [t] in the syllable-final position irrespective of its place within the morphological word.4 (1)

Coda neutralization in Korean (Sohn 1999):5 a. /path # kal-i/ [kat.kˈari] ‘field plowing’ b. /os # an/ [o.dan] ‘upper garment’ c. /us # os/ [u.dot] ‘upper garment’ d. /ic-ta/ [it.tˈa] ‘forget’ e. /puəkh # aph/ [pu.ə.gap] ‘kitchen front’

It should be noted that neutralization yields segments or segmental sequences that may be subject to further phonological processes if the relevant conditions are met. For instance, the inter-sonorant voicing process obligatorily applies in (1b, c, and e). We also observe the obligatory post-obstruent tensification process in (1a and d). Furthermore, resyllabification applies across word boundaries, which turns neutralized codas into onsets (1b, c, e). In addition to neutralization, several phonological processes satisfy the syllable structure constraints of the

4 Forms given in slanted brackets correspond to underlying morphophonemic representations and those with square brackets to surface forms. The symbols , , refer to a syllable, morpheme, and word boundary, respectively. Following Sohn (1999), /c/ and its allophone [ɟ] will be used to refer to the voiceless palatal stop and the voiced palatal stop, respectively. 5 Unless otherwise noted, all Korean examples are adapted from Sohn (1999).

116 

 Barış Kabak

language. The examples in (2) illustrate the obligatory consonant cluster simplification process, where the consonantal sequences in morphophonemic (underlying) representations are simplified by segmental deletions (2a, b), or by way of coalescence with the following consonant (2c).6 The motivation for all these processes is to eliminate illicit syllable structures on the surface. (2) Consonant cluster simplification: a. /nəks/ [nək] ‘soul’ b. /paːlp/ [paːp], [paːl] ‘step on’ c. /ilh-ko/ [il.kho] ‘lose and’

In section 3, we will see that the compliance with the native language syllable structure rules is so robust that it also applies in speech perception, causing Korean speakers to hear illusory vowels in non-native speech. Korean also has a number of static phonotactic restrictions and synchronic phonological processes that are linked to the word level or higher-prosodic domains such as the Phonological Phrase (PPh). For instance, the absence of word-initial liquids, which is notorious across Altaic languages, is prevalent in Korean although the liquids can appear elsewhere, albeit subject to allophony ([r] in the onset, [l] in the coda position or as part of a geminate). Also, the [n, l] epenthesis crucially occurs only at the edges of words. More specifically, the process takes place at the beginning of an /i/- or /j/-initial word after a consonant-final word or stem.7 Some examples are provided in (3). (3) [n, l] epenthesis (sensitive to word edges): coda neutralization n/l-epenthesis a. /cis # iki-ta/ → (cit-n-iki-ta) → [ciɲ.ɲi.gi.da] b. /kˈoch # ilim/ → (kˈot-n-ilim) → [kˈoɲ.ɲi.ɾim] c. /səul # jək/ → (səul-n-jək) → [sə.ul.ljək]

‘knead to a mash’ ‘flower name’ ‘Seoul Station’

Korean is often described as having a long-vowel shortening rule that is sensitive to the Phonological Phrase. Central Korean dialects (which include Seoul Korean) have preserved quantity (length) contrasts among vowels in the initial syllable of a word (which then surfaces in phrase-initial contexts) while other Korean dia-

6 Example (2c) demonstrates the coalesence of an underlying /h/ with the following segment (also known as h-aspiration, see Sohn 1999: 166 for details) rather than its complete deletion, since deletion would incorrectly predict the surfacing of the second morpheme /ko/ as [ko] (i.e. with a plain onset obstruent) instead of [kho] (i.e. with an aspirated onset consonant). 7 [l] (or its allophonic variant [ʎ]) surfaces when the consonant-final word ends in /l/, creating a geminate structure (see Sohn 1999: 167 for details).



Pervasive syllables and phonological unity in words 

 117

lects have lost them (see Sohn 1999: 60‒61 for details). Although the contrast is currently disappearing among younger generations, who do not make a systematic distinction between long and short vowels receptively or productively (see Kabak 2004 for details and the references therein), those who preserve the contrast systematically shorten long vowels in non-phrase initial syllables (4). As such, this rule is sensitive to the position of words within larger sequences and is crucially linked to prosodic units higher than the syllable. (4) Long vowel shortening (sensitive to phrases): a. /saːlam/ [saː.ɾam] ‘person’ b. /khin # saːlam/ [khin saɾam] ‘tall person’

With respect to assimilatory processes, Korean employs a considerable number of strictly obligatory and highly predictable consonantal assimilation processes. It should be noted that some of these rules also apply across words with the same magnitude, which then blurs not only suffix but also word boundaries. Conversely, there are processes that preserve such boundaries. This dichotomy is illustrated in Table 1. Table 1: Boundary-enhancing vs. boundary-weakening processes in Korean Boundary-enhancing

8

Boundary-weakening

(i) Tensification: /cap # ci/ à [cap.cˈi] ‘magazine’

(iv) Lateralization: /ciːn # li/ à [ciːl.li] ‘truth’

(ii) Non-application of palatalization: /kath-i/ à [ka.chi] ‘together’ vs. /path # ilaŋ/ à *[pa.chi.ɾaŋ] but [paɲ.ɲiɾaŋ]8 ‘plowed row’

(v) Sibilation: /cəc # so/ à [cə(s).sˈo] ‘milking cow’

(iii) Sai-sios (‘Bindungs-s’, see below for examples)

(vi) Nasalization: /cip # mun/ à [cim.mun] ‘house gate’ (vii) Decoronization (optional): /kas # kˈin/ à [kat kˈin] ~ [kak kˈin] ‘hat string’

Tensification and the absence of palatalization (Table 1: (i) and (ii)) arguably protect the edges of morphemes. The so-called sai-sios phenomenon (also known as ‘Bindungs-s’) inserts an /s/ between morphemes forming words or compounds (Table 1: (iii)). It should be remembered that the coda /s/ is neutralized to [t] on

8 Palatalization fails to apply due to [n-l] epenthesis, which is triggered by the word edges, as explained above.

118 

 Barış Kabak

the surface due to the coda neutralization process discussed above. The insertion of /s/ then prepares the context for several aforementioned assimilation rules such as the intensification of the following lax obstruent (through the post-obstruent tensification process, as illustrated in (5a) below), consonant cluster simplification (5b), and nasalization (5c). (5)

Sai-sios a. /pi # paŋul/ b. /pom # pi/ c. /pi # mul/

{pi-s paŋul} {pom-s pi} {pi-s # mul}

[pit.pˈaŋul] ~ [pip.pˈaŋul] [pom.pˈi] [pin.mul] ~ [pim.mul]

‘rain drops’ ‘spring rain’ ‘rain water’

The effect of sai-sios is apparent in minimal pairs such as those given in (6), where (6a) is a compound construction, and (6b) is morphologically simplex. Saisios applies to the former but not the latter. (6) a. /kaːm-s # ki/ [kaːm.kˈi] ‘winding’ b. [kaːm.ki] ‘flu’

Accordingly, although it is most often the case that morpheme and word boundaries are highly blurred in Korean due to the uniformity of within- and acrossword phonological processes, there are processes that instead seem to strengthen boundaries. Alternatively, one can assume that those processes that blur morphological boundaries achieve quite the opposite: Korean strives to treat sequences larger than a morphosyntactic word as a single prosodic unit (e.g. the Phonological Word). Assimilatory processes can then also be viewed as a type of strengthening process since cohesion within such a prosodic domain creates harmony within that domain, thereby strengthening its presence as a single unit. Crucially, such an alternative view of strengthening is not the same as edge-oriented strengthening. Rather, it is strengthening by way of achieving unity within a particular domain. This can serve to facilitate, for instance, word segmentation for the language user since where cohesion begins and where it ends can provide the language user with important cues for segmentation. We will revisit the effect of cohesion for word segmentation in the case of Turkish below.

2.2 Turkish Unlike in Korean, in Turkish there are no obligatory assimilatory processes across syllables; although voicing alternations take place across syllables, they should be viewed as a result of allomorphic rules, e.g. koş-tu ‘run-pst’ vs. sol-du ‘fade-



Pervasive syllables and phonological unity in words 

 119

pst’.9 Heterosyllabic assimilatory processes are optional, e.g. anla [anla ~ anna] ‘understand’. As expected from a syllable language, geminates are tolerated across syllables (e.g. bak.kal ‘grocer’) as well as across morpheme boundaries (e.g. bit-ti ‘finish-pst’). The language strives to keep its syllables maximally bimoraic (Kabak and Vogel 2001; Kabak and Revithiadou 2009a), hence long vowels in closed syllables, underlying consonant clusters of a particular type, as well as underlying geminates cannot surface at the end of a word. Such illicit syllable types systematically give rise to repair processes such as epenthesis, geminate simplification, and long vowel shortening. Syllable repair strategies, however, are blocked when the option of resyllabification is available. This is illustrated with long vowel shortening in (7a), which applies when no resyllabification option is available. Vowel-initial morphemes create the context for resyllabification due to the onset maximization principle (roughly, syllables must have onsets). However, this systematically happens when the following element is part of the same Phonological Word (7b) or the recursive Phonological Word (PWrec) that stems from verb-auxiliary constructions, where resyllabification is obligatory (7c), and not when it is contained in a Phonological Phrase (7d). (7) Long vowel shortening a. /haraːm/ [haram]PW ‘forbidden’ *[haraːm] b. /haraːm-(y)I/ [haraːm-ı]PW forbidden-acc *[haram-ı] c. /haraːm et-mEk/ [[haraːm]PW [etmek]PW]PW-rec ‘to take away the pleasure’ forbidden aux-inf *[haram etmek] d. /haraːm et/ [[haram]PW [et]PW]PPh ‘forbidden meat’ forbidden meat *[haraːm et]

Likewise, albeit optionally, resyllabification may also take place across words. In those cases too we see the blocking of certain phonological processes. One such process is word-final devoicing (8).

9 In Turkish, there are sequences where voiceless segments follow voiced ones (or vice versa). In those cases, no voicing assimilation is observed (e.g. çanta ‘bag’, yaprak ‘leave’), suggesting that obligatory voicing alternations are allomorphic in nature. Here I mostly adhere to the conventions of Turkish orthography. Accordingly, i represents the high front unrounded vowel. The high back unrounded vowel (IPA: [ɯ]) is indicated with a barred (ı) in lieu of the orthographic dotless (ı). The symbol ü stands for the high front rounded vowel, and ö the non-high front rounded vowel (IPA: [y] and [ø], respectively). The symbol ş represents the voiceless palato-alveolar fricative (IPA: [ʃ]) and y corresponds to the palatal glide (IPA: [j]). The symbols ç and c indicate voiceless and voiced palatal affricates (IPA: [tʃ] and [dʒ]), respectively.

120 

 Barış Kabak

(8) Word-final devoicing UR

Devoicing

Blocking of devoicing

a. /kitab/ [ki.tap] (*[ki.tab]) [ki.ta.b-ı] (*[ki.tap-ı]) ‘book’ book-acc b. /kitab # ev-i/ [ki.tap.e.vi] ~ [ki.ta.be.vi] ‘bookstore’ book#house-cmp

In sum, repair processes preserve the syllable structure. While they are enforced within Phonological Words, they can also apply, albeit with diminished strength, across words within constructions that arguably form Recursive Phonological Words (see Kabak and Revithiadou 2009a). As such, rules that clearly refer to the syllable are crucially restricted to the domain of the Phonological Word, and they can also operate in extended (recursive) Phonological Words.

2.2.1 Vowel harmony and the phonological word in Turkish Turkish has regular vowel harmony (palatal and labial harmony), which operates from one prespecified vowel to another (see Kabak 2011 for details). Although one might not need to refer to a domain like the PW to delimit the extent of vowel harmony, the fact that lexically specified vowels (i.e. vowels that are prespecified for the relevant features that spread to the following vowels) almost always coincide with the word-initial syllable make it obvious that we are dealing with a word-related process. Because word-initial vowels must be prespecified, this leads to asymmetries in the vocalic properties of word-initial vs. word-non-initial positions. More specifically, while all 8 vowels are possible in the word-initial position, low rounded vowels are not expected in non-initial positions since rounding cannot spread to low vowels. That is, rounding harmony is parasitic on the height of the following vowel such that it must be high otherwise rounding does not spread (9). Hence, native Turkish words are not expected to have low round vowels in non-initial syllables (see Kabak 2011 and Kabak and Weber 2013 for exceptions). (9) a. soru *soro ‘question’ b. kömür *kömör ‘coal’ c. soru-lar *soru-lor ‘question- pl’ d. kömür-ler *kömür-lör ‘coal-pl’

What is relevant to the discussion at hand is what vowel harmony actually achieves. While this question deserves an extensive discussion in its own right, I suggest that vowel harmony can be regarded as a process where the vocalic



Pervasive syllables and phonological unity in words 

 121

oppositions within a particular domain, here arguably the PW, are neutralized. This in turn leads to the question: Does it strengthen or weaken word boundaries? Below, I will argue that vowel harmony (and also fixed word-final stress) can indeed serve as a cue for word boundaries in online processing, and hence may be construed as a process that strengthens words. Yet again, I use the term “strengthening” to describe a way of achieving unity within a particular domain.

2.2.2 Minimal word size and category-specific phonology Yet another indication that words are relevant for the phonology of Turkish lies in size restrictions. Monosyllabic Turkish words must be bimoraic (Kabak and Vogel 2001). Exceptions to this principle are one common noun, su ‘water’, and a set of pronominal forms (e.g. o ‘he/she/it’, bu ‘this’, şu ‘that’). Every other subminimal form must either be lengthened (such as the names for musical notes: doː, reː, miː, etc.) or extended via a glide (e.g. for some speakers, ko-y ‘put-imp’, cf. ko-du ‘put-pst’) to meet this restriction. This suggests that although CV syllables are possible in Turkish, wordhood in Turkish has a particular character with specific requirements regarding the type of syllables words can have. Finally, some indirect evidence for word-related phonological phenomena in Turkish comes from the observation that the application of certain phonological processes may crucially depend on the syntactic category of the targeted element, referred to as “category-specific phonology” (e.g. Kabak and Plank 2007; Smith 2011). For example, in Turkish, the stem-final k/zero alternation process in derived environments only applies to nouns (e.g. [durak] ‘bus stop’ vs. [dura-ın] ‘bus stop-gen’) while verb roots do not show this alternation (e.g. [bırak] ‘leave’ vs. [bırak-ır], *[bıra-ır] ‘leave-aor’). In summary, the above described facts of Korean and Turkish show that in what we refer to as syllable languages, wordhood (properties which indicate that a string of sound units is to be interpreted and processed as a word) may be distinct and independent from the properties that sublexical prosodic units, such as the syllable, can have. Furthermore, size and category requirements show signs of phonology interacting with words. However, there are several reasons to maintain that Korean and Turkish are truly syllable languages. In what follows, I will review a set of empirical evidence that suggests that Turkish and Korean show clear signs of being syllable languages where the phonological system is modulated by syllable-based generalizations. I will also highlight specific cases where processes such as vowel harmony and fixed edgemost stress, properties that are typically found in syllable rhythm languages, are indeed used for word segmentation in online speech processing.

122 

 Barış Kabak

3 Pertinacity of syllables: Emergent segments via syllable repair In this section, I will discuss synchronic and psycholinguistic phenomena that arise as a result of the language user’s compliance with the syllable structure properties of the language. In general, these phenomena can be referred to as repair strategies that strive to avoid illicit syllable types, and can be observed both in production and perception. These demonstrate the pivotal role syllables play not only in the phonological system of Korean and Turkish, but also the way the speakers of these languages process speech.

3.1 Vowel hiatus Both Korean and Turkish have processes that conspire against heterosyllabic vowel contact (i.e. V1.V2). This presumably stems from the fact that syllables in principle must have onsets although onsetless syllables are tolerated word initially. Both languages employ different strategies to repair such illicit contacts. In Korean, for instance, a process that is often referred to as vowel coalescence optionally combines the height attribute of the first vowel (V1) and the frontness of the second (V2) either morpheme internally or across morpheme boundaries to create a new vowel that occupies two vacant vowel slots, thereby surfacing as long. (10) Vowel coalescence in Korean (from Kang 1999) a. /ai/ [ɛː] ‘child’ b. /oi/ [öː] ‘cucumber’ c. /po-i-ta/ [pöːda] ‘to be seen’ d. /onui/ [onüː] ‘brother and sister’ e. /thi-i/ [thiː] ‘to be opened’

In Turkish, the effects of vowel hiatus avoidance are conspicuous in suffix allomorphy (e.g. Lewis 1967; Clements and Keyser 1983: 67‒73). The examples in (11) below illustrate the obligatory allomorphic variation of suffixes conditioned by consonant-final (C-final) words vs. vowel final (V-final) words. (11) Hiatus resolution through suffix allomorphy: C-final V-final a. duvar-ı ‘wall’ b. kapı-yı ‘door’ wall-acc door-acc c. yap-ar ‘he/she does’ d. oyna-r ‘he/she plays’ do-aor play-aor



Pervasive syllables and phonological unity in words 

 123

In essence, Turkish suffixes adapt their forms in different ways to avoid vowel hiatus. Despite such systematic resistance against non-optimal syllable types, Turkish has V1.V2 sequences within a word due to two main reasons: (I) borrowings and (II) segmental deletion processes, which are discussed below with examples. I. Borrowings: Turkish borrowed a significant amount of words from several languages, which sometimes resulted in the incorporation of sound strings that do not comply with the native language phonotactic rules. Words that are often pronounced in careful speech with heterosyllabic vowels constitute one such case, as in (12). (12)

a. b. c. d.

kuaför neon mesa:i meal

(French: Coiffeur) (French: Néon) (Arabic: mesa[ʔ]i) (Arabic: me[ʕ]al)

‘hair dresser’ ‘neon’ ‘work’ ‘meaning’

II. Consonantal deletion processes: (i) Words containing (soft-g): Formerly a phone, the so-called soft-g () is now nothing more than an empty consonantal slot in the phonological representation of most speakers of Standard Turkish (e.g. Clements and Keyser 1983: 67‒73; Sezer 1986: 238‒239; Kabak 2007a). When the so-called soft-g is not pronounced, it may lead to heterosyllabic vowel sequences such as those seen in (13). (13)

a. b. c. d. e.

ağır yoğurt öğür sığar ciğer

[aır] [yourt] [öür] [sıar] [cier]

‘heavy’ ‘yogurt’ ‘to retch’ ‘fits’ ‘liver’

(ii) /v/, /y/ and /h/ deletion: In some registers of Turkish, lexical /v/, /y/ and /h/ are subject to deletion (e.g. Sezer 1986). This process not only leads to the surfacing of V1.V2 sequences when the consonant in question is intervocalic, but also to what is known as the compensatory lengthening of the preceding vowel depending on the position of the segment within a syllable. In particular, compensatory lengthening is observed if the deleted consonant occurred in the coda position (14a‒c). In the onset position, however, the consonant does not lead to lengthening (14d‒f).

124  (14)

 Barış Kabak

Compensatory lengthening a. /ahmet/ [aːmet] ‘Ahmet’ b. /seyret/ [seːret] ‘watch’ c. /övmek/ [öːmek] ‘to praise’

No compensatory lengthening d. /tohum/ [toum] ‘seed’ e. /deyim/ [deim] ‘idiom’ f. /över/ [öer] ‘(s)he praises’

(iii) Intervocalic stem-final velar stop deletion in (certain) derived environments: As mentioned in section 2.2.2 above, stem- or morpheme-final velars are deleted in certain derived environments. Although the details are not of concern here, this process also creates heterosyllabic vowel sequences. (15) UR a. /ayak-I/ b. /etek-In/

k→0 [ayaı] ‘foot-acc’ [etein] ‘skirt-acc’

3.1.1 Vowel assimilation: A syllabically conditioned (but word-sensitive) process All the above processes create a vowel hiatus, which is optionally resolved by a process called vowel assimilation. The process operates from left to right such that the second vowel is fully assimilated to the first (Sezer 1986: 242‒243; Kabak 2007a, b). (16) a. ağır [aır] b. yoğurt [yourt] c. öğür [öür]

~ [aar] *[ıır] ‘heavy’ ~ [yoort] *[yuurt] ‘yogurt’ ~ [öör] *[üür] ‘to retch’

Although the process is completely predictable, not all heterosyllabic vowel sequences are subject to vowel assimilation (see Kabak 2007a for details). When the phonological environment does not prepare the context for vowel assimilation, repair segments emerge to resolve the unwanted heterosyllabic vowel contact. These segments are invariably glides, whose place specification is determined by the quality of the neighboring vowels (e.g. 17‒19, taken from Kabak 2007a). (17) Coronal glide epenthesis takes place when the first vowel in the sequence is coronal, or when the second vowel is /i/: C→0 glide insertion a. /şehir/ [şeir] [şeyir] ‘city’ b. /eğit/ [eit] [eyit] ‘educate’ c. /kafein/ -- [kafeyin] ‘caffeine’ (18) Labial glide epenthesis conditioned by a round vowel: C→0 glide insertion a. soğan [soan] [sowan] ‘onion’ b. tavuk [tauk] [tawuk] ‘chicken’



Pervasive syllables and phonological unity in words 



c. doğa d. dua

[doa] -----

 125

[dowa] ‘nature’ [duwaː] ‘prayer’

(19) Velar glide epenthesis between back vowels: C→0 glide insertion a. sığar [sıar] [sıɰar] ‘(it) fits’ b. sıhhat [sıat] [sıɰat] ‘health’ c. saat ------ [saɰat] ‘clock, hour’ d. sığır ------ [sıɰır] ‘cattle’

In summary, glides [y, w, ɰ] are inserted as a repair strategy to avoid heterosyllabic vowel contact.10 Interestingly, glide insertion is reminiscent of vowel epenthesis in Turkish. In particular, epenthetic vowels emerge to resolve illicit consonantal sequences violating the syllable structure, and they are uniformly high vowels (e.g. Yavaş 1980; Kaun 2000). The examples in (20) illustrate this process in loan words that do not fit the syllable structure patterns of Turkish. (20) Vowel epenthesis in Turkish: a. [tiren] (Fr. train) ‘train’ b. [tırafik] (Fr. traffic) ‘traffic’ c. [buronşit] (Fr. bronchite) ‘bronchitis’ d. [fülüt] (Fr. flute) ‘flute’

Closer inspection reveals that the glides [y, w, ɰ] and the epenthetic vowels [i, u, i] are different sides of the same coin, especially judging by their phonological similarity. Both sets are the most sonorous segments in the language, and they are all high. A parsimonious account should be able to capture not only the similarity but also the apparent complementary distribution. Accordingly, epenthetic segments (glides and vowels) in Turkish are variants of an underlying sonorant, which is used as a default buffer segment. The sonorant appears in hiatus resolution contexts and, depending on the demands of the syllable structure, is assigned either to a syllable nucleus (thereby appearing as a vowel), or a non-nucleus slot (thereby appearing as a glide). The high sonorant in question takes its further place specification from the neighboring sounds.

10 See Kabak (2007a) for a discussion of the coexistence of deletion and the subsequent insertion of consonants as a repair.

126 

 Barış Kabak

[sonorant] ([high]) (nucleus positions) [u] [ı] [i] [ü]

(non-nucleus positions) [w] [ ɰ] [y]

Figure 1: Syllabically-conditioned epenthesis

Heterosyllabic vowel sequences may also arise across words. Compared to within-word vowel assimilation, however, the direction of assimilation is reversed. That is, the first vowel assimilates to the second one (Kabak 2007b). Some examples are provided in (21), where left-to-right assimilation yields ill-formed repairs. (21)

Vowel hiatus and assimilation across words (data from Kabak 2007b): a. ne ayıp! [naayıp] *[neeyıp] ‘how shameful!’ b. ne okuyor? [nookuyo(r)] *[neekuyo(r)] ‘what does s/he study?’ c. ne özel! [nöözel] *[neezel] ‘how special!’ d. anne anne [annaane] *[anneene] ‘maternal grandmother’ e. iyi akşamlar [iyaakşamlar] *[iyiikşamlar] ‘good evening’ f. yirmiüç [yirmüüç] *[yirmiiç] ‘twenty-three’ g. ortaokul [ortookul] *[ortaakul] ‘middle school’ h. soru işareti [soriişareti] *[soruuşareti] ‘question mark’

The repair strategy is again syllabically conditioned. That is, the combined sequences form a prosodic unit (presumably a recursive Phonological Word) in which resyllabification applies. Onsetless syllables are “rescued” by vowel assimilation. However, the repair does not apply to any syllable, but syllables that coincide with a word boundary. Hence, compared to within-word vowel assimilation, the directionality of assimilation here is reversed. One can construe that the change in the direction of assimilation is due to the primacy of word onsets for lexical access and word recognition purposes (e.g. Gow and Gordon 1995; McClelland and Elman 1986). Furthermore, typologically, word-initial positions are known to be immune to phonological processes, showing less variation in surface realization (e.g. Bergman 1968; Casali 1998). Hence, a word-final vowel is more likely to undergo change than a word-initial vowel. What is of relevance here is the fact that Turkish crucially respects the primacy of word onsets in the context of a process that is truly syllabically conditioned.



Pervasive syllables and phonological unity in words 

 127

3.2 (In)Tolerance for consonant clusters and word edge asymmetries As mentioned above, both Turkish and Korean do not permit onset consonant clusters. Speakers of these languages often insert an epenthetic vowel into these clusters in loan word adaptations or in second-language productions (22). (22) Vowel epenthesis in Turkish and Korean Turkish: [tiren] ‘train’, [kıredi] ‘credit’, [fülüt] ‘flute’ Korean: [siwitshi] ‘switch’, [ph illæʃi] ‘flash’ (from Kang 2010: 230)

Strikingly, however, both languages allow consonant clusters as well as other segmental sequences that contain offending syllable structures (long vowels as well as geminates in the case of Turkish, see, for example (7) above) in native morphophonemic representations. That is, underlying representations of morphemes must be assumed to contain these sequences since they surface when the appropriate syllable structure conditions are met (e.g. a vowel-initial suffix is added). Crucially, these stem alternations are observed at the right edge, and not at the left edge, which again is indicative of asymmetry at the level of the word (23). (23) Surface-illicit sound sequences at the right edge (Gloss: Korean, -i: nominative, -ta: ‘and/also’; Turkish, -ta/-da: locative, -i: accusative): UR Surface alternations C-initial suffix V-initial suffix a. kaps kap-ta kap.s-i ‘price’ (Korean) ‘soil’ (Korean, Seoul) b. hilk hik-ta hilk-i c. hakk hak-ta hak.k-ı ‘right’ (Turkish) d. zamaːn zaman-da zamaːn-ı ‘time’ (Turkish) e. aln alın-da aln-ı ‘forehead’ (Turkish)

Having shown the asymmetries between the right and left edge of words, we will now turn to the way syllable structure requirements influence listeners’ perception.

3.2.1 Syllabically conditioned perceptual epenthesis In a series of classic speech perception studies, listeners whose native language does not tolerate consonant clusters were shown to hear illusory vowels in consonant clusters (e.g. Dupoux et al. 1999). In a phoneme detection task, Japanese listeners systematically heard the vowel [u] between the two adjacent consonants in a word like ebzo (which suggests that they heard the word as ebuzo). Furthermore, they could not distinguish words with consonant clusters such as ebzo from

128 

 Barış Kabak

their vowel-containing counterparts such as ebuzo. These studies convincingly show the way native language phonotactics shapes the way speech is processed. The case of Japanese, however, does not explain whether epenthesis is due to purely sequential restrictions on sound combinations (e.g. [b.z] is not attested in Japanese) or syllable structure restrictions (e.g. no syllable can end in a [b] in Japanese). In an attempt to tease apart linearly stated co-occurrence restrictions from those that stem from syllable structure restrictions, Kabak (2003) conducted a discrimination study with native speakers of Korean, which tolerates certain types of heterosyllabic contact (see also Kabak and Idsardi 2007). The hetero­ syllabic consonantal contact types used in the study were: [k.m, l.n, c.m, c.t, ɟ.m, ɟ.t].11 Although these sequences are possible in English, they are considered to be illicit in Korean for various reasons. While [k.m, l.n] contain perfectly legitimate consonants in their respective positions (i.e. [k] and [l] are both fine as codas, and [m, n] are fine as onsets), the combination of these sounds are not permitted due to assimilation processes that obligatorily apply to them (i.e. /k.m/ must surface as [ŋm] and /l.n/ must surface as [n.n] or [l.l]). The clusters [c.m, c.t, ɟ.m, ɟ.t], on the other hand, are illicit because they contain a consonant that is not allowed to occur in the coda position. Specifically, [c] and its variant [ɟ] can never arise in the coda position due to the obligatory neutralization process discussed in section 2.1 above while the following consonants ([m] and [t]) are perfectly legitimate in the onset position. In sum, there are different ways in which a sequence [VC1C2V] can be illicit in Korean: (i) C1C2 causes a contact violation (*[k.m], *[l.n]) or (ii) C1 causes a coda violation (*[c.] *[.ɾ] *[h.]).12 Example (24) below lists the types of consonantal contact used in the stimuli of the perception study. (24) Heterosyllabic consonantal contacts used in Kabak (2003): C1+C2 Possible contact: [k.th], [l.th] Illicit contact: *[k.m], *[l.n] Illicit coda: *[c.th], *[ɟ.th], *[c.m], *[ɟ.m]

In a “same-different” discrimination task, nonsense words containing these heterosyllabic clusters were contrasted with their vowel-containing counterparts (e.g. pak.tha vs. pakutha, pak.ma vs. pakuma, pacta vs. pacita) with the assumption that if listeners hear epenthetic vowels due to first-language phonotactic restrictions, they should not be able to discriminate between word pairs that contain sequences that are not attested in Korean (i.e. all illicit sequences, either

11 The Korean [c] and [ɟ] here stand for voiceless and voiced palatal affricates, respectively. 12 A third possibility is that C2 cannot be an onset (*[.l] *[.ŋ]).



Pervasive syllables and phonological unity in words 

 129

due to the nature of the contact or the coda consonant). If, however, the syllable structure restrictions influence speech perception, only those contacts that involve illicit coda consonants should yield perceptual epenthesis. The control group was English speakers, who were not expected to have any difficulty differentiating between the pairs in question. The results of the study are shown in Figure 2 with d-prime scores (discriminability scores based on Signal Detection Theory). The higher the d-prime, the less difficulty participants had with the cluster in question.

Figure 2: Discriminability scores of heterosyllabic consonant clusters in Kabak (2003)

As can be seen in Figure 2, words that contained illicit codas were poorly discriminated by Korean listeners, while those that involved illicit contacts were distinguished as successfully as the English controls. The results are summarized in (25) below. (25) Illicit codas vs. illicit contacts Illicit codas: *[phac.ma] *[phaɟ.tha] *[phaɟ.ma] *[phac.tha]

= [pha.chi.ma] = [pha.ɟi.tha] = [pha.ɟi.ma] = [pha.chi.tha]

d’=0.31 d’=0.47 d’=0.49 d’=0.89

Illicit contacts: *[pak.ma] *[pal.na]

≠ [pakuma] ≠ [paluna]

d’=2.28 d’=3.42

130 

 Barış Kabak

Overall, the results suggest that contact violation is irrelevant in speech perception. Instead, illicit codas are repaired as onsets, which lead to perceptual epenthesis. In other words, perceptual epenthesis emerges to create well-formed syllables and not due to linear phonotactic restrictions in Korean. The results have serious consequences for certain theories, which undermine the role of syllables in phonology but instead resort to linear or string-based generalizations to capture phonotactic patterns (e.g. Steriade 1999; Blevins 2003). In general, these views assume that syllable-based markedness and faithfulness constraints should be stated in string-based terms. To that end, Blevins (2003: 383), for instance, suggests that “what determines positions of contrast or neutralization is the presence/absence and quality (sonorant, obstruent, etc.) of the immediately following or preceding segment”. In this study, however, we see that syllable structure violations – not linear co-occurrence restrictions – have an impact on speech perception. The effect is so robust that listeners hear illusory vowels as a repair strategy. The syllabically conditioned nature of modifications is further evinced in loan word adaptations. In particular, [l] does not surface in the onset position in Korean. If a foreign word contains an onset that begins with [l], listeners are observed to be faithful to it. Due to the ban on the surfacing of [l] in the onset position, the segment can only be faithfully realized in the coda position. However, this leads to a complication since Korean strives to maximize onsets. Hence, speakers instead double the [l] to meet both restrictions (26, example from Kabak 2003). (26) Syllabically conditioned repairs in loan word adaptations in Korean: L.I. (Linguistic Inquiry): [ɛ.lay] is not optimal in Korean because it contains an [l] in the onset. The consonant cannot simply move to the coda position because syllables must have onsets (i.e. *[ɛl.ay]). Hence, the consonant is doubled to yield the most optimal structure: [ɛl.lay]. Igloo: Same restrictions as above: *[ig.lu], *[igl.u]. When doubling applies, it creates a coda cluster in the first syllable, which is not allowed (*[igl.lu]). Hence, epenthesis additionally applies, yielding [igil.lu].

These observations clearly suggest that the syllable structure is highly pertinacious in Korean leading to perceptual illusions as well as surface representations that are not isomorphic to the source, at the cost of proliferating the number of segments.



Pervasive syllables and phonological unity in words 

 131

4 Syllables and beyond: Phonology for word-segmentation The culminative nature of accent (Liberman and Prince 1977) signals to the language user the presence of a lexical word for every (primary) stress-bearing syllable. When coupled with the edge demarcation function of stress in a given language, stress should then signal to the language user the existence of word onsets or offsets relative to the position of the stressed syllable in the speech string. Figure 3, where the capital X marks a stress-bearing syllable, graphically represents how a parser can postulate word boundaries when scanning two words in a speech string based on stress information. While in a leftmost word stress language (e.g. Finnish) boundary demarcation is done regressively upon the perception of a stressed syllable, the mirror image of this process is expected in a rightmost stress language (e.g. Turkish), where the stress cue progressively signals that a word boundary is expected. Leftmost word stress (Finnish) Word1 Word2 Xxx Xxx

Rightmost word stress (Turkish, French) Word 1 Word2 xxX xxX

Figure 3: Demarcative function of edgemost stress (from Kabak, Maniwa, and Kazanina 2010).

Furthermore, vowel harmony regulations can also help demarcate individual words since they dictate which vowel features are allowed to co-occur within a word. Exploiting this abstract knowledge, the language user should then be able to postulate a word boundary whenever a sequence of two adjacent vowels is disharmonic. Kabak, Maniwa, and Kazanina (2010) tested these predictions in an online word spotting experiment, where Turkish and French native speakers were asked to decide whether the word they saw on the screen was contained in the speech string they heard by pressing buttons that corresponded to “yes” or “no”. An example set of conditions used in the experiment is provided in Table 1. It should be noted that both Turkish and French have rightmost fixed stress, which more closely matches the accentual characteristics described in Auer (1991: 296‒297) for syllable rhythm languages as opposed to word rhythm languages, while Turkish but not French employs vowel harmony. Accordingly, the hypothesis is that stress should signal a word boundary immediately after the stressed syllable (i.e. in stress-3 condition), thereby facilitating the detection of a target non-word that follows such a boundary in both languages. In addition, target word detection should yield faster response latencies and/or lower error

132 

 Barış Kabak

rates in the harmony-mismatch condition only in Turkish (and not in French since the language does not have systematic vowel harmony). Table 2: A full sample set of conditions used in Kabak, Maniwa, and Kazanina (2010). The target word is paVO. The stressed syllables are in capitals. Syllables with front vowels are in bold. Harmony-match

Harmony-mismatch

Stress-2

goLUshopaVO

gøLYshøpaVO

Stress-3

goluSHOpaVO

gølySHØpaVO

The results of Kabak, Maniwa, and Kazanina (2010) strongly support the hypothesis. More specifically, whenever the target non-word was immediately preceded by a stressed syllable, its detection was faster than when there was no preceding stressed syllable in both languages. Vowel disharmony also robustly facilitated target detection, albeit this was the case only in Turkish, as expected: Targets were detected significantly more accurately in the harmony-mismatch conditions than in the harmony-match conditions. This study showed that predictable edgemost stress can serve as a word boundary inducer in that it was found to operate as a regressive, as Suomi, McQueen, and Cutler (1997) and Vroomen, Tuomainen, and de Gelder (1998) found for Finnish, and as a progressive cue, as Kabak, Maniwa, and Kazanina (2010) showed for Turkish and French. The relevance of these findings for our purposes in this paper is straightforward: Vowel harmony and edgemost fixed (demarcative non-lexical) stress, both of which are typically found in syllable rhythm languages (e.g. Auer 1991: 296–297), are indeed word-relevant processes as they are used by native speakers for word segmentation. Regarding the contradiction that vowel harmony creates, Auer (1991: 297) makes the observation that “its existence in syllable rhythm languages only seemingly contradicts the basic assumption that it is the syllable, and not the phonological word, which is the central prosodic category in such a language”. On the basis of psycholinguistic evidence reviewed above, including the experimental findings from Finnish, yet another edgemost stress language (e.g. Suomi, McQueen, and Cutler 1997; Vroomen, Tuomainen, and de Gelder 1998), I suggest that edgemost non-lexical stress should also be added to the list of contradictions, since a process that is characteristically found in syllable rhythm languages indeed serves to mark word boundaries.



Pervasive syllables and phonological unity in words 

 133

5 Pervasive syllables in morphological and prosodic change Can the primacy of syllable-based generalizations determine the course of action in diachronic morpho-phonological change? In this section, I will approach this question from the point of view of lexical accent genesis. In a predictable word-final stress language like Turkish (Kabak and Vogel 2001, 2011), where the regular stress assignment rule predicts stress on the final syllable of a word (27), accent-bearing morphemes (lexical accents) are not expected. However, there are several morphemes that affect stress assignment and disturb the appearance of stress on the word-final syllable (28). (27) Rightmost regular stress: a. kitáp b. kitap-lár c. kitap-lar-ı́m d. kitap-lar-ım-dán

‘book’ ‘books’ ‘my books’ ‘from my books’

(28) Lexically accented morphemes (underlined): Pre-accenting morphemes: a. yáp-ma-dı ‘(s)he did not do’ b. anné-ymiş ‘she is a mother’ (hearsay) Locally accented morphemes (underlined): c. gel-íyor ‘(s)he is coming’ d. yap-ámaz ‘(s)he cannot do’

Stress is right-edge oriented irrespective of the length of the word and the weight of the syllable, hence no accent needs to be assumed as part of the lexical representation of morphemes in (27). In (28), however, we have two types of morphemes for which accent needs to be lexically pre-specified: (i) those that are “pre-accenting” (28 a, b) and (ii) those that are “locally accented” (28 c, d). Closer inspection reveals that lexical accent-bearing morphemes in (28) are a result of a grammaticalization process in which independent morphemes become weak and turn into suffix-like elements in compound-like structures over time. Crosslinguistically, the grammaticalization of such tightly coherent constructions and collocations is a common process that proliferates the number of lexical accents within a language. Here, the conflicting directionality of compound stress assignment in comparison to word stress assignment is one of several factors that determines whether an otherwise edgemost stress language will develop lexical accents (see Kabak and Revithiadou 2009b for details). Let us illustrate this prediction in the case of Turkish. In Turkish, where the regular word stress is rightmost, the compound stress assignment has a conflicting directionality to

134 

 Barış Kabak

word stress. The primary stress of the leftmost Phonological Word is promoted (i.e. leftmost accent wins). This is illustrated in (29), where the syllables marked in bold represent the most prominent (i.e. compound stress-bearing) syllable in the entire compound construction. (29) Compound stress in Turkish (leftmost accent is promoted) a. [kitáp] [ev-í] ‘bookstore’ book house-cmp b. [kurú] [fasulyé] ‘dried beans’ dry bean

As mentioned above, lexical accent-bearing morphemes in Turkish came from compositional constructions. In particular, the second word was weakened and became cliticized to the first one, causing the stress to appear immediately before it since the first word continues to bear word-final stress. This leads to the emergence of pre-accenting morphemes, as illustrated by postpositional constructions in (30).13 Interestingly, the bound morpheme undergoes vowel harmony and syllabifies with the host that it attaches to, proving yet again that an extended phonological word can emerge out of two independent words. (30) Pre-stressing morphemes that stem from postpositions ise: conditional, ile: comitative a. [adám] # [isé] [melék] # [isé] b. [adám] # [ilé] [melék] # [ilé] c. [pará] # [isé] [dedé] # [isé] d. [pará] # [ilé] [dedé] # [ilé]

> > > > > > > >

adám=sa ‘as for the man’ melék=se ‘as for the angel’ adám=la ‘with the man’ melék=le ‘with the angel’ pará=ysa ‘as for money’ dedé=yse ‘as for the grandfather’ pará=yla ‘with money’ dedé=yle ‘with the grandfather’

What is relevant to the discussion here is that the precise shape of the bound morpheme emanates from the primacy of syllables in Turkish. It appears that an entire syllable of the disyllabic postpositive word is deleted in (30 a, b) when it is morphologized. However, this is not the case in (30 c, d), where faithfullness to all of the segmental units in the disyllabic word becomes conspicious. The [y] appearing with the bound forms in (30 c, d) has nothing to do with the so-called buffer consonants that emerge in the hiatus resolution processes discussed above.

13 It should also be noted that Modern Turkish is now at a transitory stage such that both free and bound variants of the postpositions occur.



Pervasive syllables and phonological unity in words 

 135

Closer inspection reveals that it emerges for reasons unrelated to syllable repair. That is, [y] needs to be inserted when there is a vowel hiatus context, but there is no such context here. That is, just -sa/-se and -la/-le would have been phonologically sufficient. What seems to be the case here is that the bound form is maximally faithful to the content features of the full form unless a syllable structure violation is incurred. Accordingly, =yle and =yse are maximally similar (but not identical) to ile and ise, respectively. The deletion of the glide in post-consonantal contexts therefore has a plausible explanation: [y] would create a complex onset following consonants, leading to a syllable structure violation (e.g. *melek.yle). Therefore, it must be deleted, yielding -sa/-se and -la/-le as optimal allomorphs (see Kabak 2006 for similar processes in Turkish). In certain cases, the final suffix of the first word and the following element were reanalysed as one morphological unit. Unlike in the previous case where we observe a pre-stressing effect, these typically carry a lexical accent on their initial syllable. This is illustrated in (31), where the converb marker attached to the verb has been analysed as a single unit together with the following postverb (auxiliary).14 This yielded suffix complexes such as ‑Adur and -Iyor, where the lexical stress is fixed on the initial syllable of the morpheme. (31) Locally accented morphemes a. *[yaz-á] # [tur] write-cnv aux b. *[gel-é] # [yorı̈r] come-cnv aux c. *[gel-í] # [ver] come-cnv aux

>

yaz-ádur

>

gel-íyor

>

gel-íver

Here, too, we observe the influence of the syllable structure on the mental representation of the morpheme. Interestingly, all such locally accented morphemes are disyllabic in Turkish (see Kabak and Revithiadou 2009b for a similar tendency in Basque) because syllables were strictly protected when reanalysis took place, leading to no elision or fusion. In an otherwise pervasively rightmost stress language, stress on a monosyllabic morpheme is easily interpreted as an instance of regular rightmost accent when that morpheme is word final. Hence, no locally accented monosyllabic morpheme has emerged in the history of the language. That is, just as in the previous case, faithfulness to the original syllabic compo-

14 The function of such postverbs is, among other things, to mark temporal and aspectual distinctions. See Johanson (1998: 113‒114) for details.

136 

 Barış Kabak

sition of morphemes played a significant role in lexical accent genesis and their maintenance in mental representations.

6 Discussion and conclusion In this paper, I have provided several pieces of evidence that suggest that syllables not only shape morpho-phonological representations but also influence the way speech strings are perceived in Korean and Turkish. The resolute tenacity of syllables in the phonological systems also determines the fate of morphemes diachronically. I have shown that these languages also make use of phonological processes that crucially refer to the word (and sometimes its syntactic category information) and exhibit static sound regularities that are asymmetrically distributed to word edges. I have argued that several harmonizing processes (e.g. assimilation and neutralization rules) in both Korean and Turkish may appear to be blurring morphological boundaries. These processes instead should be seen as regularities that strive to achieve unity and coherence within prosodic units that are often larger than a lexical word in these languages. Although they may not necessarily be construed as edge-oriented strengthening phenomena, such harmonizing processes facilitate speech segmentation since cohesion within a speech string most often accrues to the demarcation of the beginning and end of a domain. As such, vowel harmony and fixed rightmost stress assignment are truly word-related phenomena, constituting vital signs of wordhood in Turkish for processing reasons. Coupled with the fact that syllabification and other assimilatory processes may concurrently take place across words in certain morphosyntactic constructions, the implication of these observations for prosodic typology is that words can also serve as important anchors and targets for phonological processes in languages that show true signs of syllable-based prosodies. This should come as no surprise since one of the fundamental assumptions in Prosodic Phonology states that prosodic domains are ordered in a hierarchy and organized into layers (e.g. Selkirk 1984; Nespor and Vogel 1986). More specifically, exhaustive parsing of utterances into syllables, which are then contained in Phonological Words due to layering, allows Phonological Words to (optionally) inherit low-level properties, and also develop unique properties of their own. As such, there is no theoretical motivation to expect a single prosodic category to carry primary importance in determining the overall character of a phonological system. As the delineation between the syllable and the word in some languages is not always clear, primarily owing to the fact that the notion of word is not fully construed due to various uncertainties and ambiguities surrounding it, the ques-



Pervasive syllables and phonological unity in words 

 137

tion remains as to how the nature and dynamics of wordhood in pervasively syllable-based languages differ from those in pervasively word-based languages.

References Auer, Peter (1991): “Stress-timing” vs. “syllable-timing” from a typological point of view. In: Bohumil Palek and Přemysl Janota (eds.), Proceedings of the Conference Linguistics and Phonetics: Prospects and Applications (LP’90), vol. I, 292‒305. Prague: Charles University Press. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Bergman, Richard (1968): Vowel sandhi in Igede and other African languages. Unpublished M.A. thesis, Hartford Seminary Foundation. Blevins, Juliette (2003): The independent nature of phonotactic constraints: An alternative to syllable-based approaches. In: Féry Caroline and Ruben van de Weijer (eds.), The Syllable in Optimality Theory, 375‒403. Cambridge: Cambridge University Press. Blevins, Juliette (2010): Review of ‘Syllable structure: The limits of variation’ by San Duanmu. Linguistic Typology 14/2-3: 287‒291. Casali, Roderic F. (1998): Resolving Hiatus. New York/London: Garland. Clements, George N. and Samuel Jay Keyser (1983): CV Phonology: A Generative Theory of the Syllable. Cambridge, MA: MIT Press. Duanmu, San (2009): Syllable Structure: The Limits of Variation. Oxford: Oxford University Press. Dupoux, Emmanuel, Kazohiko Kakehi, Yuki Hirose, Christophe Pallier and Jacques Mehler (1999): Epenthetic vowels in Japanese: A perceptual illusion? Journal of Experimental Psychology: Human Perception and Performance 25/6: 1568‒1578. Gow, David and Peter C. Gordon (1995): Lexical and prelexical influences on word segmentation: Evidence from priming. Journal of Experimental Psychology: Human Perception and Performance 21/2: 344‒359. Johanson, Lars (1998): The history of Turkic. In: Lars Johanson and Éva Á. Csató (eds.), The Turkic Languages, 81‒125. (Routledge Language Family Descriptions.) London/New York: Routledge. Kabak, Barış (2003): The perceptual processing of second language consonant clusters. Ph.D. dissertation, University of Delaware. Kabak, Barış (2004): Acquiring phonology is not acquiring inventories but contrasts: The loss of Turkic and Korean primary long vowels. Linguistic Typology 8/3: 351‒368. Kabak, Barış (2006): An obstacle to the morphologization of postpositions. Studies in Language 30: 33‒68. Kabak, Barış (2007a): Hiatus resolution in Turkish: An underspecification account. Lingua 117/8: 1378‒1411.

138 

 Barış Kabak

Kabak, Barış (2007b): Vowel assimilation across words in Turkish. Turkic Languages 11: 181‒195. Kabak, Barış (2011): Turkish vowel harmony. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 5: Phonology across Languages, 2831‒2854. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Kabak, Barış and William Idsardi (2003): Syllabically conditioned perceptual epenthesis. Proceedings of the Annual Meeting of the Berkeley Linguistics Society 29: 233‒244. Kabak, Barış and William Idsardi (2007): Perceptual distortions in the adaptation of English consonant clusters: Syllable structure or consonantal contact constraints? Language and Speech 50/1: 23‒52. Kabak, Barış, Kazumi Maniwa and Nina Kazanina (2010): Listeners use vowel harmony and word-final stress to spot nonsense words: A study of Turkish and French. Laboratory Phonology 1: 207‒224. Kabak, Barış and Frans Plank (2006): Where flexion encroaches on agglutination in Turkish and Korean. In: Yong-Kun Ko et al. (eds.), Whither Morphology in the New Millennium?, 123‒152. (Morphology Monograph Series 1.) Seoul: Pagijong Press. Kabak, Barış and Frans Plank (2007): Phonology and word class distinctions. Paper presented at the 7th Biennial Meeting of the Association for Linguistic Typology, Paris, France. Kabak, Barış and Anthi Revithiadou (2009a): An interface approach to prosodic word recursivity. In: Janet Grijzenhout and Barış Kabak (eds.), Phonological Domains: Universals and Deviations, 105‒133. (Interface Explorations 16.) Berlin/New York: Mouton de Gruyter. Kabak, Barış and Anthi Revithiadou (2009b): From edgemost to lexical stress: Diachronic paths, typology and representation. The Linguistic Review 26: 1‒36. Kabak, Barış and Irene Vogel (2001): The phonological word and stress assignment in Turkish. Phonology 18: 315‒360. Kabak, Barış and Irene Vogel (2011): Exceptions to stress and harmony: cophonologies or prespecification? In: Horst Simon and Heike Wiese (eds.), Expecting the Unexpected: Exceptions in Grammar, 59‒94. Berlin/New York: Mouton de Gruyter. Kabak, Barış and Silke Weber (2013): Markedness, context, and directionality in vowel harmony: A corpus study on vowel co-occurence patterns. Dilbilim Araştırmaları Dergisi [Journal of Linguistic Research] 2013: 53‒85. Kang, Ongmi (1999): A correspondence analysis on hiatus resolution in Korean. Studies in Phonetics, Phonology and Morphology 5: 1‒24. Kang, Yoonjung (2010): The emergence of phonological adaptation from phonetic adaptation: English loanwords in Korean. Phonology 27/2: 225‒253. Kaun, Abigail (2000): Epenthesis-driven harmony in Turkish. Proceedings of the Annual Meeting of the Berkeley Linguistics Society 25/2: 95‒106. Lewis, Geoffrey (1967): Turkish Grammar. Oxford: Oxford University Press. Liberman, Mark Y. and Alan Prince (1977): On stress and linguistic rhythm. Linguistic Inquiry 8/2: 249‒336. McClelland, James L. and Jeffrey L. Elman (1986): The TRACE model of speech perception. Cognitive Psychology 18: 1‒86. Nespor, Marina and Irene Vogel (1986): Prosodic Phonology. (Studies in Generative Grammar 28.) Dordrecht: Foris. Revithiadou, Anthi (2011): The phonological word. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 2:



Pervasive syllables and phonological unity in words 

 139

Suprasegmental and Prosodic Phonology, 1204‒1227. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Selkirk, Elizabeth O. (1984): Phonology and Syntax: The Relation Between Sound and Structure. Cambridge, MA: MIT Press. Sezer, Engin (1986): An autosegmental analysis of compensatory lengthening in Turkish. In: Leo Wetzels and Engin Sezer (eds.), Studies in Compensatory Lengthening, 227‒250. (Publications in Language Sciences 23.) Dordrecht: Foris. Smith, Jennifer L. (2011): Category-specific effects. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 4: Phonological Interfaces, 2439‒2463. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Sohn, Ho-Min (1999): The Korean Language. Cambridge: Cambridge University Press. Steriade, Donca (1999): Alternatives to syllabic analysis of consonantal phonotactics. In: Osamu Fujimura, Brian Joseph and Bohumil Palek (eds.), Proceedings of the 2nd Linguistics and Phonetics Conference, 205‒245. Prague: Charles University Press. Suomi, Kari, James M. McQueen and Anne Cutler (1997): Vowel harmony and speech segmentation in Finnish. Journal of Memory and Language 36: 422‒444. Vroomen, Jean, Jyrki Tuomainen and Beatrice de Gelder (1998): The roles of word stress and vowel harmony in speech segmentation. Journal of Memory and Language 38: 133‒149. Yavaş, Mehmet S. (1980): Borrowing and its implications for Turkish phonology. Unpublished Ph.D. dissertation, University of Kansas.

Guido Seiler (University of Freiburg) and Kathrin Würth (University of Teacher Education Luzern)

Monosyllabic Lengthening in German and its relation to the syllable vs. word language typology Abstract: Many German dialects show clear evidence for a vowel lengthening process in monosyllabic words. Monosyllabic Lengthening has attracted little attention in the literature on both German dialectology and phonological theory. While a number of other phonological properties have already been discussed in the literature as being indicative of either the syllable or the word type of language, Monosyllabic Lengthening has not yet been analyzed with regard to the syllable/word language typology. The present paper pursues four goals: First, we will discuss the prosodic-phonological traits of Zurich German (High Alemannic) in detail, showing that Monosyllabic Lengthening is an essential component of Zurich German phonology. Second, we will look at Monosyllabic Lengthening processes in German more generally, arguing that Monosyllabic Lengthening processes are crucial, but often overlooked in most High German dialects. Third, we will argue that Monosyllabic Lengthening operates at the level of the phonological word. Therefore, it is clearly indicative of the word-centered type of language. And fourth, we will examine the predictive power of Monosyllabic Lengthening for other syllable vs. word language parameters, focusing on parameters within the domain of quantity systems. We will discuss the theoretical and typological implications of the fact that Monosyllabic Lengthening freely combines with other parameters.

1 Introduction At the core of linguistic typology lies the assumption that we find implicational relationships between certain linguistic properties. It is thus the aim of linguistic typology to classify languages according to their structural properties and investigate whether and to what extent the occurrence of such a property is likely to correlate with further properties. For a typology that distinguishes between syllable and word languages, the basic observation is that languages tend to organize their rhythmic structure in a particular, i.e. isochronous, way and that this has further consequences on a language’s phonological shape. In syllable-based languages, the syllable is the basic prosodic category for phonological processes, whereas in word-based languages the basic category is



Monosyllabic Lengthening in German  

 141

the phonological word. A number of parameters have been proposed that serve as diagnostics to classify languages as syllable languages or word languages (cf. Auer 1993, 2001). Nübling and Schrambke (2004: 282) point out that the syllable and word language division is an idealization. They consider it to be two poles on a continuum where languages arrange themselves close to either of the poles. From a diachronic perspective, it has been proposed that German developed from a syllable language in Old High German towards a word language in New High German (cf. Szczepaniak 2007). Looking at the dialect areas in more detail, Nübling and Schrambke (2004) conclude that southern dialects adhere more to the syllable-language type than northern varieties. This paper consists of two main parts. First, we will introduce an additional parameter that has not been taken into account so far, namely Monosyllabic Lengthening (MSL). MSL is a process by which words undergo some kind of augmentation in order to fulfill language-specific word size requirements. We argue that this process affects the prosodic word and is thus expected in word languages. Interestingly, a number of dialects that are assumed to belong to the syllable-type side of the continuum have introduced MSL as an innovation. Second, we will discuss the consequences of MSL for the syllable/word language typology in more general terms. We will take MSL as a starting point in order to investigate the inner structure, the “geometry” of the presumed continuum from syllable to word orientation. It is not clear at this point whether the proposed indicators of syllable or word orientation (cf. Auer 1993, 2001) are completely independent of each other or whether they interact in specific ways such as hierarchies, correlations, and implications (which, of course, would significantly enhance the predictive power of the syllable vs. word language typology). Therefore, we will look at several other quantity-related processes found in the German-speaking area and examine whether the presence of one feature is dependent on another. The paper is organized as follows: To illustrate the word-orientated character of MSL, we take Zurich German (ZHG) as an example. In section 2, we will present the relevant characteristics of ZHG phonology and relate them to MSL, arguing that MSL can best be accounted for under the assumptions of moraic theory (Hayes 1989). Section 3 provides further data in order to give a more general picture of the distribution of MSL in German dialects. The predictive power of the MSL parameter will be examined in section 4. Focusing on parameters that are presumably related to quantity, we will investigate whether they are implicational. It turns out that MSL freely combines with other parameters. Section 5 concludes this paper by discussing the theoretical and typological implications of this result.

142 

 Guido Seiler and Kathrin Würth

2 Monosyllabic Lengthening in Zurich German To illustrate the workings of MSL in more detail, we will look at ZHG, a High Alemannic dialect spoken in Switzerland, roughly in the area of the canton of Zurich. In ZHG we find paradigmatic quantity alternations where the monosyllabic form contains a long vowel while the vowel is short in the disyllabic (e.g. plural) form:1 (1) [hɑːz̥] ‘rabbit’ vs. [hɑz̥ə] ‘rabbits’

A closer inspection of these alternations reveals that it is in fact a lengthening of the monosyllabic form. This phenomenon has been termed Leichtschlussdehnung (lit. ‘lengthening of light endings’, Bohnenberger 1953: 155) in Alemannic dialectology and, more recently, MSL. We will come back to it in section 2.2.

2.1 Moraic theory (Hayes 1989) MSL in ZHG is best analyzed in terms of moraic theory as proposed by Hayes (1989). Moraic theory uses the traditional notion of the mora as a means to distinguish light and heavy syllables. In Hayes’s proposal, heavy syllables bear two moras and light syllables bear one mora. Short vowels are associated underlyingly with one mora; long vowels have two moras. Short consonants, on the other hand, have no mora underlyingly, whereas geminates are inherently moraic, cf. (2): (2) Vowels

Short vowel Long vowel

1 μ Consonants 2 μ

Single consonant Geminate

– 1μ

Moraic theory accounts for language-specific differences in syllable weight by means of how these languages assign moraic structure. In particular, languages vary in whether they consider CVC syllables to be heavy (such as e.g. Classic

1 The transcription adopted here deviates from the phonetic conventions as it suggests a contrast between ZHG geminate obstruents and single devoiced obstruents. This notation of the single obstruent as devoiced goes back to the assumptions of traditional dialectology that ZHG obstruents contrast in intensity rather than length (i.e. consonants were either lenis or fortis, cf. Winteler 1876; Baur 1939; Weber 1948). Recent studies have shown that all ZHG obstruents are underlyingly voiceless, and they only contrast in their length (cf. Willi 1996; Würth 2002; Fleischer and Schmid 2006; Seiler 2009). We kept the lenis notation only because it helps the (German) reader to recognize the etyma.



Monosyllabic Lengthening in German  

 143

Latin) or light (e.g. Lardil).2 CVC syllables that behave like heavy syllables – e.g. in stress assignment or metrics – thus must be bimoraic. This is achieved by language-specific rules of Weight-by-Position: In languages like Latin (but not Lardil), consonants are assigned a mora by the grammar if they are in a particular syllable position, usually the coda. In languages with a Weight-by-Position rule, we thus find two types of moraic consonants, cf. (3):3 (3) a. underlyingly moraic: “geminate”

b. moraic by rule: “Weight-by-Position” σ

µ | C

µ µ | | C V C

In moraic theory, lengthening is achieved either by adding a mora that was not present in the underlying representation or by reassociating a mora formerly assigned to another segment. Compensatory Lengthening is a prime example of the latter where an element is deleted on the segmental level while the moraic tier remains intact. Hayes (1989: 260) cites the diachronic change from Indo-European *casnos to Latin cānus as an example: (4) (4)

σ

k

σ

µ

µ

a

s

σ

n

σ

µ

µ

u

s

µ

µ

k

σ

µ

µ

a

s

n

µ

µ

u

s

σ

µ

µ

k a n u s

2 Cf. Hayes (1989: 255) and references therein. 3 The fact that the surface structures of (3a) and (3b) are similar suggests that the contrast is neutralized, too. In a pilot study on ZHG, this prediction has been borne out, cf. Seiler and Würth (2008).

144 

 Guido Seiler and Kathrin Würth

In Compensatory Lengthening, a mora that is segmentally unassociated after the deletion of the segment is relinked to the preceding vowel. Crucially, there are no changes on higher prosodic levels. In fact, Compensatory Lengthening ensures that the prosodic structure remains unaltered. By contrast, the addition of a (formerly absent) mora changes the prosodic structure fundamentally. Mora addition must thus be well motivated. It normally comes into play as a repair strategy in order to fulfill demands of higher prosodic categories, i.e. minimality constraints.

2.2 MSL in Zurich German: Quantity contrasts and word minimality ZHG has both contrastive vowel length and consonant length (cf. Weber 1948; Keller 1961; Würth 2002). Sonorant consonants have no underlying length distinction. As for the obstruents, length is the only phonetic cue for the contrast (Willi 1996). All obstruents in the ZHG consonant system are voiceless and unaspirated. The minimal pairs given in (5) show that both vowel and consonant quantity combine freely. (5) [z̥ib̥ə] ‘seven’ [Vɑtːə] ‘cotton wool’ [Z*ɑd8ə] ‘damage’ [huːb̥ə] ‘cap’

[z̥iːb̥ə] ‘to strain’ [Vɑːtːə] ‘to wade’ [Z*ɑtːə] ‘shadow’ [huːpːə] ‘to honk’

As can be gleaned from the examples above, ZHG allows short vowels in open stressed syllables. Contrary to many other German varieties – including Standard German – ZHG has not undergone Open Syllable Lengthening (OSL) (see section 3.1). However, while ZHG allows CV syllables in disyllabic words, we do not find them in monosyllabic (content) words. Codaless monosyllabic words always contain a long vowel, as illustrated in (6). (6) [roː] ‘raw’ [z8eː] ‘lake’ [mɑː] ‘man’, *[mɑ] (but: [mɑnə] ‘men’)

The picture becomes somewhat more complex if we look at the full range of possible monosyllabic words as presented in (7):



Monosyllabic Lengthening in German  

 145

(7) a. *CVC *[hAz8] (but [hAːz̥] ‘rabbit’) b. CVCC [hund8] ‘dog’ c. CVːC [huːz8] ‘house’ d. CVCː [hAsː] ‘hatred’ e. CVːCː [huːtː] ‘skin’ f. CVCCː [g*umpː] ‘jump’

Before we discuss the data in (7) in more detail, let us first point out four important observations: First, light stressed syllables are permitted in ZHG in principle, e.g. (5). Second, ZHG prohibits words with the CV syllable structure, e.g. (6). Third, words with the CVC syllable structure (7a) are prohibited, too, thus they behave like CV words. And fourth, ZHG allows word-final geminates (7d–f). Let us now have a closer look at the moraic structure of the words in (7). We argue that ZHG imposes a minimality requirement on words. The mora count for short and long vowels as well as for geminates follows straightforwardly from the theoretical assumptions given in (2). The words in (7c–f) do not violate minimality.4 From the fact that ZHG prohibits both CV (6) and CVC (7a) words, we conclude that they have the same moraic structure, namely that they are monomoraic. At first glance, the fact that CVC syllables are light may suggest that ZHG has no Weight-by-Position rule. Such an analysis, however, cannot account for CVCC words like the one in (7b): The existence of CVCC words is clear evidence that they do not pattern with CV(C) and are thus heavy, i.e. bimoraic. To account for these seemingly paradoxical structures, we adopt a very well-established parameter, namely final consonant extrametricality (cf. Hayes 1982, 1995: 57): Word-final single consonants are ignored in that position and thus invisible for the Weight-by-Position rule. Therefore, CVC words count as monomoraic. In CVCC words like (7b), on the other hand, Weight-by-Position does apply: Because CVCC words do not pattern with CVC (CVCC is allowed, CVC is not), we conclude that non-final coda consonants are assigned one mora by rule: CVCC counts as bimoraic and thus fulfills the constraints on minimal word size. ZHG thus has (i) a minimality requirement on words with two moras, (ii) Weight-by-Position, and (iii) final consonant extrametricality. The prohibition of subminimal, i.e. monomoraic, words leads to length alternations in the paradigm, as depicted in (8) where the relevant forms are highlighted.

4 Some of the examples – (7e) and (7f) – consist of three moras. Trimoraic syllables are considered rare in the world’s languages and languages often impose a bimoraic upper limit per syllable. This, however, is not relevant for the purposes of the present paper.

146 

 Guido Seiler and Kathrin Würth

(8) singular a. /hAz̥/ b. /Z*lAg*/ c. /huːz̥/

plural hAːz̥ Z*lAːg* huːz̥

diminutive hAz̥ə hQz̥li Z*lEːg* Z*leg*li hyːz̥ər hyːz̥li

‘rabbit’ ‘stroke’ ‘house’

The examples in (8) show that lengthening takes place whenever a word would otherwise be monomoraic. This alternation is purely phonological as e.g. monosyllabic plural forms are lengthened, too (8b). Note that it is clearly a lengthening process. Words with an underlyingly long vowel do not undergo shortening but remain long throughout the paradigm (8c).5 No lengthening takes place when the word is closed by a geminate, e.g. (7d). This is predicted by moraic theory since geminates are inherently moraic. In monosyllabic words ending in a cluster (7b), no lengthening occurs either: The word is bimoraic because the non-final coda consonant is moraic (via Weight-by-Position). However, CVC words undergo lengthening because Weight-by-Position does not apply, due to final consonant extrasyllabicity. In sum, words that are underlyingly CV(C) lengthen their vowel by adding a mora. This lengthening is motivated by word minimality restrictions. Minimal word constraints and minimality restrictions in general are common in the languages of the world (cf. Hayes 1995: 47–48, 88; Hall 1999: 105–106). As in a number of languages, ZHG (content) words must be minimally bimoraic. Monosyllabic Lengthening applies if an underlying form does not meet this requirement. Moraic theory forces us to analyze MSL as a process aiming to optimize the word size. The domain where the minimality restrictions hold is the phonological word (cf. Dixon 1977; Hall 1999; Dixon and Aikhenvald 2002). This makes MSL a valuable parameter for the syllable vs. word language typology at issue. Moreover, if we look back at the history of German, we find that MSL is absent both in Old High German (OHG) and in Middle High German (MHG). This makes MSL a rather recent innovation. Processes whose domain is the phonological word are interpreted as diagnostics for word languages (cf. Auer 2001; Szczepaniak 2007). While our case of MSL fits the general diachronic tendency presented by Szczepaniak (2007) – i.e. drift from syllable language to word language – it is highly unexpected for ZHG since High Alemannic is described as syllable-oriented in a very consistent way (cf. Nübling and Schrambke 2004; Szczepaniak 2007). The following section gives an overview of how widespread MSL is in the German-speak-

5 The ZHG pattern is not singular in Alemannic: Similar patterns have been described by Spaelti (1994) for the canton Glarus dialect, Kraehenmann (2003) for Thurgovian, and Auer (1989) for the dialect of Constance.



Monosyllabic Lengthening in German  

 147

ing area. The question of its relation to other word-oriented innovations will be addressed in section 4.

3 More on Monosyllabic Lengthening in High German varieties Having introduced the basic workings of MSL and its most adequate explanation as a means of avoiding subminimal words, we will set out to investigate MSL and related quantity processes in the German-speaking area. The purpose of this section is twofold: First, we will show that MSL is an individual phonological process that can be found in a much greater area than usually attested. This is due to the fact that the effects of MSL are often considered to be a mere result of paradigm levelling. We will challenge this view in section 3.1. Second, we will give an overview of several quantity-related phonological processes that can all be explained as instances of MSL, namely the enhancement of a subminimal word form by adding a second mora. These types of MSL will be presented in section 3.2 and brought together with other well-known prosodically related processes, namely Open Syllable Lengthening and Degemination.

3.1 Monosyllabic Lengthening or paradigm levelling? In the New High German (NHG) standard language, monosyllabic forms which contained a short vowel in MHG are lengthened, too. Thus, the following examples are highly reminiscent of the ZHG pattern discussed above: (9) MHG > [tak] [slak]

NHG [thaːk] ‘day’ [∫laːk] ‘stroke’

Furthermore, and most strikingly, NHG displays lengthened vowels exactly in the same environments as ZHG does. Lengthening is excluded in words ending in an etymological geminate or cluster in MHG: (10) MHG > [flɛkk] [hunt]

NHG [flɛk], *[fleːk] [hunt], *[huːnt]

‘stain’ ‘dog’

The standard explanation for these lengthened vowels in NHG is very different, though. It goes back to Paul (1884, in particular p. 119) and has hardly been questioned in German historical linguistics ever since (cf. however Hotzenköcherle

148 

 Guido Seiler and Kathrin Würth

1986; Auer 1989; Kusmenko 1995; Seiler 2009). According to Paul, the only genuinely phonological lengthening process is OSL. Once OSL applies, it creates vowel quantity alternations in inflectional paradigms: (11) Singular [tak] Plural [taːgə], long vowel due to OSL

The introduction of a long vowel in the monosyllabic, uninflected form is explained as an instance of paradigm levelling, i.e. analogy: The alternation has been levelled in favor of the long vowel. In other words, there is no MSL at all if understood as a genuinely phonological process. Instead, long vowels have been “borrowed” from other positions in the same inflectional paradigm. Or, as Wilmanns (1911: 319) puts it: “Der regulierende Faktor war offenbar die Silbenscheide: in offener Silbe trat Dehnung ein, in geschlossene drang sie nur durch Systemzwang und durch Formübertragung”.6 The standard explanation based on paradigm levelling is not applicable to ZHG: ZHG did not undergo OSL; therefore, there are no long vowels in disyllabic forms which the analogy could be based on, see (1). Given the levelling hypothesis, the ZHG pattern is not only unexpected, it even contradicts the hypothesis because vowel lengthening in monosyllabic forms introduces a paradigmatic quantity alternation (instead of eliminating one). One might therefore ask whether the NHG pattern is really based on levelling or perhaps just another instance of MSL. In fact, the NHG pattern in isolation admits both hypotheses as viable explanations since (as compared to MHG) all instances of lengthened vowels of monosyllabic forms can in principle still be regarded as the product of analogy. However, since ZHG forces us to introduce a process such as MSL anyway, one might suspect that MSL is perhaps much more widespread in High German. If this turns out to be correct, the assumption of paradigm levelling becomes vacuous and therefore simply unnecessary. As a matter of fact, clear, positive evidence in favor of the MSL hypothesis can be found in many other High German varieties. By positive evidence for MSL we mean those monosyllabic forms where (i) the vowel is lengthened as compared to MHG and (ii) which cannot possibly be analyzed as the result of analogical levelling. This is the case whenever the paradigm does not contain any other, disyllabic form with a long vowel (i.e. whenever the base for the analogy is missing). As we have demonstrated above, the ZHG lengthened monosyllabic forms pre-

6 Obviously, the regulating factor was the syllable boundary: lengthening took place in open syllables, whereas it has been extended to closed syllables only by means of system pressure and form transfer (translation GS and KW).



Monosyllabic Lengthening in German  

 149

cisely meet the above-mentioned criteria: The paradigm does not contain disyllabic forms with long vowels; therefore the long vowel of the monosyllabic form cannot be borrowed from there. Whereas ZHG lengthens only before a single singleton but not before clusters or geminates, other dialects lengthen in the latter environments, too. (12) Other instances of MSL which cannot be attributed to OSL: [loː*] – [lexxɑ] ‘hole, holes’ (Middle Bavarian: Pfalz 1913: 27) [nɔː*d9] – [nɛxd̥ə] ‘night, nights’ (Thuringian: Thüringisches WB IV: 797)

3.2 A brief synopsis of High German quantity developments The behavior of isolated forms is not revealing enough, however. The full extent of MSL within High German can be estimated only if different types of MSL are distinguished from each other and more quantity-related processes are taken into account. In order to establish MSL as a central, widespread process in High German, it must be shown that it is much more prevalent than expected and covers an area that goes beyond ZHG. Additionally, it must be made clear that the forms in question cannot be explained away entirely on the basis of analogical levelling. We therefore attempt to give a brief synopsis of the relevant quantity developments of High German dialects (for details see Seiler 2009). The examples in (12) suggest that MSL shows up in different ways and to different extents, depending on the particular (groups of) dialects. Let us therefore first focus on different versions of MSL before we turn to the more general picture which includes other quantity-related processes as well. The ZHG type with lengthening only before a single consonant has been called Leichtschlussdehnung (‘light-end lengthening’) by Bohnenberger (1953), who distinguishes it from Schwerschlussdehnung (‘heavy-end lengthening’). We will use his terminology for our categorization, too. The first of four MSL types in total is represented by dialects where the vowels of monosyllabic forms have not been lengthened at all. Failure of MSL is a common property of Low German dialects and also of the Northern pronunciation of Standard German (Paul 1884: 112; Keller 1961: 316). In High German varieties, retention of vowel shortness in monosyllabic words is far less common, yet it does occur in Alemannic dialects of the Bodensee region, as well as in the Alemannic–South Franconian transition zone around Rastatt (SDS II/49, SSA III/2.010). Moreover, the examples demonstrate that in these dialects there is no OSL either:

150 

 Guido Seiler and Kathrin Würth

(13) No MSL and no OSL in the Bodensee region and the Alemannic–Franconian transition zone: [grɑb̥], [greb̥ər] ‘grave, graves’ (SDS II/49) [rɑd̥], [red̥ər] ‘wheel, wheels’ (SSA III/2.010)

The second MSL type corresponds to the ZHG pattern described above: Lengthening occurs only before single nonmoraic consonants but not before clusters or geminates. We will call it Leichtschlussdehnung I; see example (1) for illustration. This type is very widespread. Apart from ZHG, it is found in most other Alemannic dialects, in West Central German and Upper Saxon (see Seiler 2009: 239–251). Our next type, Leichtschlussdehnung II, operates in the same environments as Leichtschlussdehnung I, but with different results. Here, it is not the vowel but the final consonant which is lengthened. Although this is, of course, not a vowel lengthening process but rather consonant gemination, we include it in our collection of MSL types because its effects on the moraic level are identical. Final gemination is found in the Highest Alemannic dialect of Valais: (14) Leichtschlussdehnung II = final gemination (Valais, Highest Alemannic: SDS II/45): MHG [glas] > [glɑss]

The third type of MSL is Schwerschlussdehnung I. Here, not only words ending in simple consonants but also those ending in (former, i.e. MHG) geminates are lengthened (whereas lengthening does not occur in words ending in a cluster). This type is represented by the Silesian dialect of Seifhennersdorf: (15)

Schwerschlussdehnung I (Silesian: Michel 1891: 5–7): simple consonant: MHG [grap] > [g*rɑːb̥] geminate: MHG [sakk] > [z̥ɑːg*] cluster: MHG [naχt] > [nɑχd̥]

Finally, other dialects lengthen monosyllabic forms across the board, i.e. simple consonant, (former) geminate and cluster stems. We call this type of MSL Schwerschlussdehnung II. It is represented by Bavarian and Thuringian, e.g. Middle Bavarian of Marchfeld: (16)

Schwerschlussdehnung II (Middle Bavarian: Pfalz 1913: 9, 12, 39): simple consonant: MHG [rat] > [rɔːd̥] geminate: MHG [bokk] > [b̥oːg*] cluster: MHG [naχt] > [nɔːχd̥]

The different MSL types can easily be traced back to different interactions between prosodic parameters we have established already, i.e. minimal weight of words and/or stressed syllables, Weight-by-Position, and tolerance for geminates (in



Monosyllabic Lengthening in German  

 151

final position and/or elsewhere) (Seiler 2009). For example, consider Thuringian again (example 12). In Thuringian a constraint on the minimal weight of a word is active; therefore MSL applies in principle. This is all we have to know for lengthening of Rad-type words. As for other monosyllabic words, the other word-prosodic parameters must be taken into account. The fact that cluster stems are lengthened, too, can be expressed by means of the Weight-by-Position parameter. If we assume that Thuringian does not have Weight-by-Position, no consonant can count as moraic in words like Nacht. Therefore, the minimality constraint can be satisfied only by means of vowel lengthening (thus Nācht). Finally, Thuringian does not tolerate geminates at all. A degeminated form like *[v̥iʒ̊] would violate word minimality as well (the form goes back etymologically to a geminate stem, e.g. ZHG [v̥i∫∫] where the geminate is retained). Let us now broaden the picture and include other quantity-related processes such as OSL and the development of geminates in our short survey. Following Auer (2001: 1397), Nübling and Schrambke (2004: 284) and Szczepaniak (2007: 52–53), both OSL and Degemination are expected in the development towards word languages (whereas the lack thereof serves as a diagnostic for the syllable-oriented type). Table 1 displays a greater number of High German dialects (thus following the synopsis in Seiler 2009: 249). Furthermore, we also included Low German and Standard German. In the first column we subsumed all dialects which lengthen vowels in monosyllabic words before a simple consonant (Leichtschlussdehnung I). The second column contains all dialects where vowels of monosyllabic words are lengthened before clusters (Schwerschlussdehnung II). As for OSL, we simply distinguished between dialects which have undergone OSL from those which have not. With regard to the further development of the MHG geminates, we distinguished between three types of dialects: dialects with full conservation of geminates (most of Alemannic), dialects with Degemination in final position only (Bavarian), and dialects which degeminate across the board.7 For the sake of clarity, we refrained from distinguishing between the Leichtschlussdehnung types I and II. Also, we did not include the Schwerschlussdehnung I type, since this type (represented e.g. by Silesian) automatically follows from the interaction of the other parameters: If a given dialect lengthens before simple consonant and disallows geminates in final position, it follows that this dialect also lengthens in words like Fisch, Bock, etc. (to our knowledge, no dialect retains the geminate and lengthens the vowel in front of it). Since Table 1 is a summary of the results

7 The development of vocalic quantities also suggests that dialects with total degemination first shortened geminates in final position only. Total degemination must have taken place later, namely after the introduction of MSL. See Seiler (2009: 264–266) for details.

152 

 Guido Seiler and Kathrin Würth

of Seiler’s (2009) dialect-comparative investigation, we refer to that paper for details: Table 1: Synopsis of quantity-affecting processes (following Seiler 2009: 249) MSL before MSL before simple cluster consonant Alemannic of Bodensee region

no

no

High Alemannic

yes

Uri (Highest Alemannic)

no

Low Alemannic

OSL

Word-Final Total Degemination Degemination

no

no

no

no

no

no

no

no

yes

no

no

yes

no

yes

no

no

Swabian

yes

no

yes

no

yes

Bavarian

yes

yes

yes

yes

no

West-Central German

yes

no

(yes)

no

yes

Thuringian and East Franconian

yes

yes

yes

yes

yes

Silesian

yes

no

yes

yes

yes

Upper Saxon

yes

no

yes

no

yes

Low German

no

no

yes

n/a

yes

Standard German

yes

no

yes

no

yes

As a result, it is worth noticing that there is an implicational relationship between the presence of MSL before simple consonants and clusters: The latter occurs only if the former occurs as well, but not necessarily vice versa. In other words, whereas MSL before simple consonants can occur on its own, MSL before clusters cannot, but it must always be accompanied by MSL before simple consonants. This is a nice result as it is predicted by our moraic representation in a straightforward way: The Weight-by-Position parameter can distinguish only between languages where one mora is assigned positionally to coda consonants and languages where positional mora assignment does not occur. Therefore, sequences of consonants may either be heavier than single consonants (as in a language with Weight-by-Position) or they may be of equal weight (as in a language without Weight-by-Position). Crucially, this representation does not allow single consonants to be heavier than sequences of consonants (with the exception of geminates, whose weight is, of course, not positionally assigned but underlying, i.e. lexically contrastive).



Monosyllabic Lengthening in German  

 153

Let us now briefly come back to the standard explanation of MSL based on paradigm levelling, now in light of the results given in Table 1. The analogical levelling hypothesis does not account for MSL dialects without OSL (High Alemannic), nor is it a viable explanation for dialects with some type of Schwerschlussdehnung (Bavarian, East Franconian, Silesian, Thuringian). It still is, however, at least theoretically, an option for those varieties where MSL of the Leichtschlussdehnung type co-occurs with OSL (Low Alemannic, Swabian, Upper Saxon). Here, and only here, all instances of MSL can be considered the result of levelling. However, we question whether such an explanation is desirable since it does not capture the relatedness of the quantity processes. In sum, MSL and OSL give an impressively consistent picture: As a rule of thumb, MSL appears to be a southern, OSL a northern innovation. Therefore, in the southernmost and northernmost areas either MSL (High Alemannic) or OSL (Low German) often occurs on its own. In between, i.e. in rather central areas, the geographical extents of MSL and OSL overlap, which is entirely expected. As a consequence, the innovations which were already established by necessity on independent grounds (OSL and MSL in its different versions) are fully sufficient to account for the Standard German (and Low Alemannic, Swabian and Upper Saxon) pattern. This does not necessarily falsify the analogical levelling hypothesis, but makes it simply dispensable. Applying Occam’s razor as a methodological principle, we now feel free to abandon the levelling hypothesis entirely. In other words, for the majority of dialects it is obvious that MSL occurs independently of OSL. This fact not only establishes MSL as a genuinely phonological process, it also has consequences for the proposed syllable vs. word language typology. The issue will be discussed in the following section.

4 Doing typology: In search of correlations and implications In the previous sections, we established MSL as a genuine phonological process whose domain of application is clearly the phonological word. We have used ZHG as an example of MSL and extended the analysis to other varieties of German. Within the context of the syllable vs. word language typology, MSL must be interpreted as an optimization of the phonological word and therefore as indicative of the word-oriented type. This is remarkable since Nübling and Schrambke (2004) have characterized Alemannic as a syllable language (where we would not expect word-size regulating processes). We then extended our analysis to other quantity-affecting processes, whose application is, following Auer (2001), Nübling and Schrambke (2004), and Szczepaniak (2007), also typical of word languages, namely OSL and Degemination. We will now take a closer look at possible

154 

 Guido Seiler and Kathrin Würth

interactions between the parameters MSL, OSL, and Degemination. Under the assumption that a typology is supposed to be more than a mere enumeration of loosely related individual features which can be freely turned on or off in a language, we may expect correlations, or at least implications, between at least some of the parameters. Let us first examine possible interactions between MSL and OSL. The two parameters and their combined positive or negative values generate a variation space of four logically possible language types, namely {+MSL +OSL}, {-MSL -OSL}, {+MSL -OSL}, and {-MSL +OSL}. The first two language types are in full accordance with the syllable vs. word language typology: The first type combines two word language parameter values, the second type combines two syllable language parameter values. In our sample, both of these “well-behaved” types occur. {+MSL +OSL} is represented by Low Alemannic, Bavarian, and East Central German, {-MSL -OSL} by South Franconian around Rastatt and Alemannic dialects of the Bodensee region. However, odd feature combinations occur, too. The High Alemannic dialect of Zurich displays MSL but no OSL, thus {+MSL -OSL}. If this were the only odd feature combination, one could conclude that the presence of MSL is a necessary but not sufficient condition for the presence of OSL. To put it differently, one might argue that OSL implies MSL, but not vice versa. Unfortunately, this is not the only odd combination we find. Low German as well as the Highest Alemannic dialect of Uri display exactly the inverse feature combination. Whereas they clearly show the results of OSL, monosyllabic words are not lengthened, thus {-MSL +OSL}. The results can be summarized as follows: Table 2: Interactions of MSL and OSL –OSL

+OSL

–MSL

South Franconian, Bodensee Alemannic

Low German, Uri Highest Alemannic

+MSL

Zurich High Alemannic

Low Alemannic, Bavarian, East Central German

Let us now turn to possible interactions of MSL with the other putative word language property, Degemination. Again, four feature combinations are logically possible: {+MSL +degem}, {-MSL -degem}, {+MSL -degem}, and {-MSL +degem}. The first type is fully word-oriented with regard to the two properties in question, and is represented by Swabian and Central German in our sample. The second type displays a fully syllable-oriented feature combination, which we identified in the Alemannic dialects of the Bodensee area and Uri. However, if odd feature combinations also exist, the relationship between MSL and Degemination is cer-



Monosyllabic Lengthening in German  

 155

tainly not a strict correlation. Moreover, if both odd feature combinations occur, the relationship between MSL and Degemination is not even implicational. Again, both odd combinations are attested: {+MSL -degem} is the pattern we found in High and Low Alemannic and Bavarian, {-MSL +degem} is represented by South Franconian around Rastatt and Low German. Table 3 shows that all logically possible combinations are attested within the ensemble of German dialects: Table 3: Interactions of MSL and Degemination –Degemination

+Degemination

–MSL

Bodensee Alemannic, Uri Highest Alemannic

Low German, South Franconian

+MSL

High and Low Alemannic, Bavarian

Swabian, Central German

The results of our comparative examination of three quantity-affecting, word-oriented processes can be summarized as follows. MSL does not correlate with OSL or Degemination, nor is it possible to identify any implicational relationships between MSL and OSL or Degemination. Instead, MSL freely combines with OSL and/or Degemination.8

5 Conclusion and outlook On the basis of Zurich German vowel lengthening patterns, we have established that Monosyllabic Lengthening is a genuine phonological process whose domain of application is the phonological word. Given the assumptions of Moraic Theory, MSL is motivated by the satisfaction of a constraint on the minimal weight of content words, which we interpreted as an optimization of the phonological word. We then showed that MSL is in fact much more widespread within the ensemble of German varieties than traditionally assumed, and that traditional work on German language history has insufficiently acknowledged this fact. In particular, we demonstrated that the traditional explanation of vowel lengthening in monosyllabic words (i.e. levelling in favor of lengthened vowels due to OSL) does not hold wherever MSL introduces (instead of eliminates) quantity alternations within paradigms. We then sketched a brief typology of the three quanti-

8 Interestingly, however, there is an implicational relationship between OSL and Degemination. The application of both Word-Final or Total Degemination seems to be dependent on the application of OSL since the combination {-OSL +degem} does not occur.

156 

 Guido Seiler and Kathrin Würth

ty-affecting processes MSL, OSL, and Degemination, all of which are indicative of a more word-oriented language type. It turns out that the application of MSL does not correlate with either of the two other processes, nor does it enter an implicational relationship with either of them. Let us finally draw some conclusions from the findings with regard to the more general perspective of the syllable vs. word language typology, and propose possible pathways for future research. First, we conclude that the already rich collection of parameters which are claimed to be indicative of syllable vs. word orientation must be supplemented by another parameter, namely MSL, or, to put it more generally, by a constraint on the minimal size of content words. We believe that processes like MSL cannot be understood as anything but an optimization of the phonological word. Cross-linguistically, many languages show robust evidence in favor of minimal weight requirements of words. In light of the syllable vs. word language typology, at least one property of word orientation can be attributed to these languages. One word of caution is in order, however. Like any other word-oriented property, MSL depends on a clear understanding of the notion of the phonological word. Strictly speaking, a typology based on the syllable vs. word orientation of phonological processes is preliminary as long as a universally applicable definition of the phonological word is missing. As Schiering, Bickel, and Hildebrandt (2010) point out, such a definition is difficult, if not impossible to establish. Rather, the phonological word appears to be an emergent category within individual languages with different formal manifestations. If this is right, it is difficult to use word orientation as an applicable diagnostic for the position of a language in typological space cross-linguistically. Future research must carefully examine the possible consequences of these findings for the syllable vs. word language typology. Second, it has been stated in previous work (Auer 2001; Szczepaniak 2007) that syllable and word orientation constitute two poles of a continuum rather than a strict binary division of language types. This may be expected as diachronic typological shifts from syllable orientation towards word orientation (and vice versa?) must be possible in principle, and language change is gradual. Consequently, it is expected that very closely related languages/dialects occupy subtly different positions on that continuum. Our own investigation of German quantity is in accordance with this general picture. However, future research will have to go a step further and ask what exactly the structure of the continuum is: Do languages randomly combine syllable- and word-oriented parameters, or do certain parameters tend to co-occur with others with greater frequency? What is the predictive power of the presence of one parameter for the presence of others? As long as questions like these remain unanswered, the syllable vs. word language typology is merely able to locate a single language in a typological space



Monosyllabic Lengthening in German  

 157

within which “anything goes” in principle. Yet the typologically interesting phenomena are precisely those where not anything goes, but certain structural properties cluster with a certain class of others. As Auer (2001: 1398) states, the syllable vs. word language typology (or: hypothesis, as Auer calls it) first requires an empirical examination of correlations between the involved phonological properties. Only then can the move from a mere localization of languages in typological space towards a predictive theory be made. Our third concluding remark is methodological. In order to capture the precise nature of the continuum between syllable and word orientation, it is necessary to rigorously examine the predictive power of each of the proposed syllable/word language parameters for the other parameters. Of course, this is a very ambitious goal; however, it defines a concrete and at the same time theoretically appealing agenda for future research. In the present paper we have studied possible interactions between only three structurally affine processes, within a small set of genetically closely related dialects, but we hope that future research will undertake similar examinations of other parameters. As for our investigation of MSL in German varieties, it has been shown that MSL is not a reliable predictor for the presence of OSL or Degemination even within the limitations of our sample, which is at first glance not very encouraging. Nevertheless, we believe that micro-comparative studies are an ideal tool for further refinements and assessments of the syllable vs. word language typology: Closely related, very similar languages formidably show which structural parameters vary independently of each other and which ones pattern together; (only) in the latter case is it necessary also to determine whether the co-variance is robust in a broader cross-linguistic context. In sum, the syllable vs. word language typology is undoubtedly a useful heuristic tool, a “checklist” to uncover, name, and classify similarities and differences between languages in a particular, relatively broad systematic field. However, we suggest also looking at the syllable vs. word language typology as a (possibly falsifiable) hypothesis about the nature of languages (in accordance with Auer 2001: 1398). Therefore, it is an enlightening enterprise to test the predictive power of each of the relevant parameters for others in an unbiased, even mechanical way. Following this path, we will find answers to questions such as: Can the proposed syllable- vs. word-oriented traits be weighted? Which co-occurrences of traits are necessary, which ones are random? Can the number of indicators for syllable vs. word orientation be reduced to a small set of principle indicators? In the context of our investigation it seems that MSL is one of the rather marginal indicators of word orientation, but this assessment remains ad hoc as long as the predictive power of other traits is not examined in a similarly rigorous way. Future research will have to undertake exactly this kind of examination.

158 

 Guido Seiler and Kathrin Würth

References Auer, Peter (1989): Zur Dehnung im Alemannischen (am Beispiel des Konstanzer Stadtdialekts). Zeitschrift für Dialektologie und Linguistik 56/1: 32–57. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Baur, Arthur (1939): Schwyzertüütsch: Praktische Sprachlehre des Schweizerdeutschen. Zürich: Rigi-Verlag. Bohnenberger, Karl (1953): Die alemannische Mundart. Umgrenzung, Innengliederung und Kennzeichnung. Tübingen: Mohr. Dixon, Robert M. W. (1977): A Grammar of Yidiɲ. Cambridge: Cambridge University Press. Dixon, Robert M. W. and Alexandra Y. Aikhenvald (2002): Word: A typological framework. In: Robert M. W.Dixon and Alexandra Y. Aikhenvald (eds.), Word: A Cross-Linguistic Typology, 1–41. Cambridge/New York: Cambridge University Press. Fleischer, Jürg and Stephan Schmid (2006): Zurich German. Journal of the International Phonetic Association 36/2: 243–253. Hall, T. Alan (1999): Phonotactics and the prosodic structure of German function words. In: T. Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 99–131. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Hayes, Bruce (1982): Extrametricality and English stress. Linguistic Inquiry 13/2: 227–276. Hayes, Bruce (1989): Compensatory lengthening in moraic phonology. Linguistic Inquiry 20/2: 253–306. Hayes, Bruce (1995): Metrical Stress Theory. Principles and Case Studies. Chicago/London: University of Chicago Press. Hotzenköcherle, Rudolf (1986): Aspekte und Probleme der Vokalquantität im Schweizerdeutschen. In: Robert Schläpfer and Rudolf Trüb (eds.), Dialektstrukturen im Wandel. Gesammelte Aufsätze zur Dialektologie der deutschen Schweiz und der Walsergebiete Oberitaliens, 319–333. Aarau: Sauerländer. Keller, Rudolf E. (1961): German Dialects. Manchester: Manchester University Press. Kraehenmann, Astrid (2003): Quantity and Prosodic Asymmetries in Alemannic. (Phonology and Phonetics 5.) Berlin/New York: Mouton de Gruyter. Kusmenko, Jurij K. (1995): Die mittelgermanische Quantitätsverschiebung im Lichte der modernen germanischen Dialekte. In: Hans Fix (ed.), Quantitätsproblematik und Metrik: Greifswalder Symposion zur germanischen Grammatik, 73–102. Amsterdam: Rodopi. Michel, Reinhart (1891): Die Mundart von Seifhennersdorf. Beiträge zur Geschichte der deutschen Sprache und Literatur 15: 1–69. Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik – Beihefte 129.) Stuttgart: Steiner.



Monosyllabic Lengthening in German  

 159

Paul, Hermann (1884): Vokaldehnung und vokalverkürzung im neuhochdeutschen. Beiträge zur Geschichte der deutschen Sprache und Literatur 9/1: 101–134. Pfalz, Anton (1913): Die Mundart des Marchfeldes. In: Joseph Seemüller (ed.), Deutsche Mundarten 4. (Sitzungsberichte der Kaiserlichen Akademie der Wissenschaften in Wien. Philosophisch-Historische Klasse 170/6.) Wien: Hölder. Schiering, René, Balthasar Bickel and Kristine A. Hildebrandt (2010): The prosodic word is not universal, but emergent. Journal of Linguistics 46/3: 657–709. SDS = Hotzenköcherle, Rudolf (ed.) (1962–1997): Sprachatlas der deutschen Schweiz. Bern/ Basel: Francke. Seiler, Guido (2009): Sound change or analogy? Monosyllabic Lengthening in German and some of its consequences. Journal of Comparative Germanic Linguistics 12/3: 229–272. Seiler, Guido and Kathrin Würth (2008): On two types of moraic consonants. Winteler’s Law in the light of Moraic Theory. Talk at the 16th Manchester Phonology Meeting (22–24 May 2008), University of Manchester. Spaelti, Philip (1994): Weak edges and final geminates in Swiss German. Manuscript, University of California, Santa Cruz. ROA-18. Rutgers Optimality Archive, http://roa.rutgers.edu. SSA = Steger, Hugo, Eugen Gabriel and Volker Schupp (eds.) (1989–2011): Südwestdeutscher Sprachatlas. Marburg: Elwert. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Thüringisches WB = Institut für Deutsche Sprache und Literatur an der Deutschen Akademie der Wissenschaften zu Berlin (1966–1970) and the Sprachwissenschaftliche Kommission der Sächsischen Akademie der Wissenschaften zu Leipzig (since 1971) (eds.): Thüringisches Wörterbuch. Auf der Grundlage der Sammlung von Victor Michels und Herman Hucke bearbeitet von Band IV bis Band VI unter Leitung von Karl Spangenberg, fortgesetzt unter Leitung von Wolfgang Lösch, weitergeführt von Susanne Wiegand. Berlin: Akademie Verlag. Weber, Albert (1948): Zürichdeutsche Grammatik. Zürich: Schweizer Spiegel. Willi, Urs (1996): Die segmentale Dauer als phonetischer Parameter von ‚fortis‘ und ‚lenis‘ bei Plosiven im Zürichdeutschen. Eine akustische und perzeptorische Untersuchung. (Zeitschrift für Dialektologie und Linguistik – Beihefte 92.) Stuttgart: Steiner. Wilmanns, Wilhelm (1911): Deutsche Grammatik. Vol. 1: Lautlehre. 3rd ed. Strassburg: Trübner. Winteler, Jost (1876): Die Kerenzer Mundart des Kantons Glarus in ihren Grundzügen dargestellt. Leipzig/Heidelberg: C.F. Winter’sche Verlagshandlung. Würth, Kathrin (2002): Zur phonologischen Klassifikation der Fortis/Lenis-Opposition im Zürichdeutschen. M.A. thesis, University of Zurich, Switzerland.

Renata Szczepaniak (University of Hamburg)

Vowel and consonant epentheses in the history of German from the typological perspective of syllable and word languages Abstract: This paper analyses vowel and consonant epentheses in the history of German from the typological perspective of syllable and word languages (Auer 1993, 2001; Szczepaniak 2007a). The prosodic analysis reveals a shift of the epenthetic domain from the phonological syllable in Old High German (OHG) towards the phonological word in Early New High German (ENHG): The OHG epentheses improved syllables by simplifying consonant clusters, e.g. OHG durh > duruh ‘through’ (CVCC > CV.CVC) or resolving hiatuses as in OHG sāen > sāwen ‘to sow’ (CV.VC > CV.CVC). In contrast, epentheses operating since ENHG have highlighted the phonological word by producing hiatuses and, thus, raising the word-internal sonority, e.g. MHG bûr(e) > (E)NHG Bauer ‘peasant’ (CVC > CVṾVC) or by producing complex word-external syllable codas, e.g. MHG mâne > (E)NHG Mond ‘moon’ (CVCV > (CVC) > CVCC). This domain shift is part of the phonological change of German from a syllable to a word language type, which has brought about a series of word-optimizing processes.

1 Syllable and word as typologically relevant phonological domains In the history of German, vowel and consonant epentheses are very frequent. This paper argues that epenthesis can ease as well as impair articulation. This depends on the prosodic domain in which the epenthesis applies. Furthermore, it will be shown that the character of vowel and consonant epentheses has indeed changed during the history of German from syllable-optimizing and, thus, facilitating articulation, which apply regardless of the morphological boundaries, to word-emphasizing epentheses, which frequently also impair articulation. This is on par with the general phonological change of German from a syllable to a word language type (see Szczepaniak 2007a). The following analysis is based on the phonological hierarchy, as established by Nespor and Vogel (1986). It contains at least two low-level categories, the phonological syllable (σ) and the phonological foot (F), which are constructed on phonological information exclusively: The internal structure of the phonological syllable is regulated by universal principles (preference laws) described by Ven-



Vowel and consonant epentheses in the history of German  

 161

increasing sonority

nemann (1988). They show that the optimal syllable contains one segment with the highest degree of Consonantal Strength (a strong consonant = plosive) in the onset and a vocalic (i.e. sonorous) nucleus. The deviation from the CV ideal leads to different grades of complexity (see also Maddieson 2005). As segments can be more easily added to the onset than to the coda, elaborated coda clusters impair the syllable structure more than onset clusters (see Auer 1993: 43). Besides the quantity also the quality of the involved segments are of high relevance for the syllable complexity. The violation of sonority sequencing by means of extrasyllabic consonants, e.g. s in Werk-s ‘work, factory-gen.sg’, raises the complexity of the syllable, rendering its articulation more difficult, see Figure 1.

v

ɛ

ɐ̯

ks

Figure 1: The qualitative increase of syllable complexity by extrasyllabic consonants (Werk-s ‘work, factory-gen.sg’)

The syllables are arranged into phonological feet according to their relative prominence (stress structure), whereby the prominent (i.e. stressed) syllable forms the foot head, e.g. [zeː] in segelte ‘(she) sailed’ [ˈzeː.gəl.tə]F. The feet in turn are integrated into the phonological words. The edges of the phonological word, which is the next higher category, are aligned with morpheme boundaries, e.g. Segel+tuch ‘sailcloth’ [[ˈzeː.gəl]F]ω[[tuːx]F]ω (Nespor and Vogel 1986: 109; Hall 1999). In contrast to syllables and feet, phonological words are based on morphological information. This makes it more difficult to define the optimal structure of the phonological word. It is easier to speak about phonetic and phonological processes and rules highlighting its structure. Thus, there is no optimal phonological word, but there are different degrees of its “visibility”, which in turn makes the morphological structure easier to decode (Szczepaniak 2009). The “visibility” of the phonological word can be achieved at two different levels (see Figure 2, for details on the selected processes see Nübling and Szczepaniak 2008: 13–16): 1) The size of the phonological word can be subject to regulations. Here, the tendency to regulate the number of feet and syllables within the word is possible. Thus, the stabilization of the size can be seen as a strategy to optimize the phonological word. The “optimal” size itself is language-specific. This kind of optimiza-

162 

 Renata Szczepaniak

tion is known from the historical phonology of German, where the trochaic ideal, the monopedal disyllabic word, has developed (see Szczepaniak 2007a). 2) The optimization of the shape of the phonological word is more complex. Here, the internal structure can be differentiated by word position or accentbased distribution of vowels and consonants as well as of different syllable types. Because of the special character of the phonological word, which is directly linked to the morphological structure, the internal phonological word structure is expected to reflect its internal morphological structure, e.g. by segmental or structural emphasis of the stem syllable.

increasing consonantal strength

processes increasing the sonority within the phonological word: – word-medial consonant weakening – contraction processes increasing the difference between the stressed and unstressed syllable: – qualitative and quantitative reduction of unstressed vowels – lengthening of stressed vowels

left edge

right edge

size of the phonological word processes strengthening the left edge: – consonant aspiration – extrasyllabic consonants

processes strengthening the right edge: – Auslautverhärtung – consonant epenthesis – extrasyllabic consonants

Figure 2: The optimization of the phonological word (selected processes from the history of German)



Vowel and consonant epentheses in the history of German  

 163

2 Syllable and word language as opposite typological prototypes The typology of syllable and word languages, developed by Auer (1993, 2001), is based on the observation that the phonological make-up of languages varies as a consequence of their central phonological domain. So, the relevance of both phonological constituents as domains for phonological processes and rules vary cross-linguistically. Hence, a given language can be more syllable- or more word-optimizing (see Gilles, this volume). In Auer (1993, 2001), some phonological phenomena with high diagnostic value are discussed. For a thorough typological analysis, it will, however, not be sufficient to “detect” those phenomena in a given language. For example, according to Auer (1993) geminates usually correlate with a simple syllable structure. This does not mean that geminates represent reliable diagnostics for a syllable language. Here, it is instead necessary to prove whether the geminates are restricted to a certain position within the word (e.g. word-medial geminates) or can appear in every position. Generally, word-based restrictions make the word “visible”. In word languages, where the phonological word is optimized to a high degree, the morphological structure is well pronounced. Hence, word languages provide clues for the hearer to decode the morphological information more easily. In syllable languages, where the tendency to optimize the syllable is strong, there is little chance to mark the morphological structure, since phonological words (and even phonological phrases) contain sequences of (almost) equal syllables, which in contrast are very easy to articulate. In other words, word languages benefit the hearer by facilitating decoding, whereas syllable languages bring about an ease of articulation to the speaker’s favor (Szczepaniak 2010: 86–88). Epenthetic vowels and consonants facilitate the articulation only if their insertion simplifies consonant clusters (CCV > CVCV) or resolves hiatuses (CVV > CVCV).1 In such cases the epenthesis operates on the syllable level and improves the structure leading in optimal case to CV syllables. However, the same processes can also impair the syllable structure when operating on the word level. In such case, new consonant clusters are built or enhanced by epenthetic consonants (CVC > CVCC), while epenthetic vowels create new hiatuses (CVC > CV.VC). The following analysis will also investigate the quality of the epenthetic segments and their impact on the syllable and word structure.

1 In this article epenthetic vowels and consonants will be underlined. Syllable boundaries will be marked with a dot which in case of ambisyllabic segments (vowels as well as consonants) is set directly under the segment in question.

164 

 Renata Szczepaniak

3 Vowel epentheses in Old and Early New High German In the history of German, two different vowel epentheses are documented. First, there is a strong tendency to use vowel epenthesis during the OHG period (see section 3.1), which interestingly has not left any marks on the Standard German vocabulary. An epenthetic schwa, originating from them, is however still present in Middle German dialects (Auer 1997) as well as in Luxembourgish, where it is not productive anymore (Szczepaniak 2010; Gilles, this volume).2 In ENHG, a second layer of epenthetic vowels appeared (see section 3.2). They differ from the first layer concerning the vowel quality and the context, which will be explained with the typological shift from a syllable towards a word language and, thus, the growing relevance of the phonological word.

3.1 The Old High German cluster breakers In his comprehensive study of OHG (as well as of Old Low German) documents until 1250, Reutercrona (1920) distinguishes three areal groups of epenthetic vowels (see Table 1). Generally, the epenthetic vowels occur only within consonant clusters. All OHG dialects share the epenthesis in word-medial w-clusters as well as in word-medial and word-final lh- and rh-clusters (highlighted in grey). In Old Upper German dialects, the epenthesis is possible in many more contexts, including word-initial w- and L-/N-clusters as well as clusters of two obstruents, mostly in foreign words. Old Low German, which displays more epenthetic contexts than Old Middle German, but not word-initially and between obstruents, will be not considered in this paper.

2 An epenthetic schwa is still productive in contemporary Dutch (Booij 1995). It is, however, word optimizing. Since the inserted schwa vowel differs in quality from the stressed vowel, it optimizes the shape of the word. According to Kuijpers and van Donselaar (1997), speakers are more likely to insert the schwa if this will create a trochaic word and, thus, optimize the word size, e.g. [kalm] > [kaləm] ‘quiet’ (see section 1).



Vowel and consonant epentheses in the history of German  

 165

Table 1: The OHG vowel epenthesis word-initial

word-medial

word-final

C+L/L+C

chraft > charaft ‘power’

malha > malaha ‘bag’

durh > duruh ‘through’

C+N/N+C

snepfa > senepfa ‘snipe’

senwa > senawa ‘sinew’

(not documented)

C+w

swarz > sowarz ‘black’

farwa > farawa ‘color’

(not documented)

C+C

pisitech ‘parakeet’ (from Lat. psittacus)

smaragede ‘emerald’ (from Lat. smaragdus)

kisiht > kisihit ‘face, sight’

According to Reutercrona (1920: 172), the OHG epenthetic vowels are more likely to appear in complex consonant clusters, e.g. OHG durft > duruft ‘necessity’. In his study of the Old Alemannic Rule of Saint Benedict from the ninth century, Seiler (1874: 430–433) observes that the vowel epenthesis applies almost regularly in tautosyllabic clusters, e.g. werah ‘work’, while the heterosyllabic clusters exhibit alternation, e.g. between wer.ch-um and werach-um ‘work-dat.pl’. Furthermore, homorganic clusters such as ld, rd, nd, or mb do not trigger epenthetic vowels. It can be stated that epenthesis appears more frequently when the consonant clusters are articulatorily more challenging and helps, thus, to resolve them (see Sievers 1901: 294–295; Howell 1991: 104). OHG vowel epenthesis can be, consequently, analyzed as a syllable-optimizing process originating from the speaker’s need for easier articulation. As to the role of the phonological word, Table 1 shows that the epenthetic vowels can appear in all word positions only in Old Upper German. In Old Middle German (highlighted in grey in Table 1), the phonological word plays a minor role, because the epenthesis is restricted to word-medial and word-final clusters. The word-initial clusters remain unaffected.3 The question remains whether the phonological word serves as the construction domain of the epenthesis. This question will be treated below. Here, it should be pointed out that the epenthesis usually applies within a phonological word. Reutercrona (1920: 164) observes,

3 A similar difference in degrees of sensitivity to word position between Upper and Middle German can be observed for the Second Consonant Shift. In contrast to Middle German, where Germanic plosives in the word-initial position can only be affricated, Old Alemannic allows affricates as well as fricatives in every word position, also word-initially (see Sonderegger 2003: 262; Szczepaniak 2007a: 126–131).

166 

 Renata Szczepaniak

however, some instances of epentheses in consonant clusters occurring at the boundary between prefix and stem, e.g. for OHG ur+loub ‘permission’, for OHG ur+chundi ‘testimony’. In such cases, the clusters rl and rch belong to two adjacent phonological words [ur]ω[loub]ω and [ur]ω[chundi]ω, which behave like the word-internal clusters, e.g. [durh]ω ~ [duruh]ω ‘through’. So, the epenthesis of o and i is not sensitive to the word boundaries. At this point, a distinction must be made between the construction domain and the domain of optimization. The phonological word definitely was not optimized by the OHG vowel epenthesis, which applies irrespective of word size and, thus, does not make it better recognizable. Instead, the epenthesis destabilizes the word size by adding a syllable to mono-, di- or trisyllabic words, see (1). This word size alternation in OHG is contrary to the later tendency towards trochaic words (see section 3.2 as well as footnote 2 for Dutch). (1) OHG vowel epenthesis and word size alternation CVCC > CV.CVC wurm ~ wurum ‘worm’ CVC.CV > CV.CV.CV farwa ~ farawa ‘color’ CV.CVC.CV > CV.CV.CV.CV bifilh-u ~ bifilih-u ‘order-1sg.pres’

The epenthetic vowels can appear after a short stem vowel or (only in Old Upper German) in the word-initial cluster directly before the stressed vowel (Reutercrona 1920: 172; Szczepaniak 2007a: 111–112). So, the posttonic epenthetic vowel can be added only after a light syllable (L) and carries then the second mora within a resolved moraic trochee, i.e. disyllabic foot head with monomoraic stress syllable, e.g. wu in wurum ‘worm’ or fa in farawa ‘color’ (cf. Dresher and Lahiri 1991). Hence, the construction domain is not the phonological word, but the phonological foot (see Figure 3). Still, it is the minimal foot domain and, hence, the most prominent word position, which is affected by the posttonic vowel epenthesis. The pretonic (Upper German) epenthetic vowels are also adjacent to the prominent word position. (a)

(x ) μ μμ L H du ruh

.) (b) (x μ μ μ L L L se ne wa

Figure 3: OHG vowel epenthesis within the minimal foot domain

Despite the difficulties connected with the fragmentariness of the historical data, it can be stated that the OHG posttonic epenthetic vowels tend towards vowel harmony. According to Reutercrona (1920: 174), the quality of the inserted vowel



Vowel and consonant epentheses in the history of German  

 167

is [a]. Hence, OHG belongs to the group of languages with the unmarked low vowel (see Lombardi 2003). However, the OHG records reveal that the inserted vowels tend to assimilate to the preceding stressed stem vowel in backness, cf. durh vs. durah, duruh ‘through’ as well as in height, cf. senwa vs. senawa, senewa ‘sinew’. The features of the stressed vowels spread only to the epenthetic vowels, while the following vowels remain unaffected. So, the domain of the OHG vowel harmony is not the phonological word, but, again, the resolved moraic trochee (see Figure 3). However, the epenthetic vowels can also assimilate in place to the following consonant. For this reason, it is the vowel u which occurs most frequently before [w], e.g. farwa vs. farawa, faruwa ‘color’ or witwa vs. wituwa ‘widow’. The pretonic epenthetic vowels behave similarly: They adapt the features from the following stressed vowel (e.g. snepfa vs. senepfa ‘snipe’) or from the consonant (e.g. swarz vs. sowarz ‘black’). Such a complex interaction of competing processes controlling the quality of epenthetic vowels is also described by Uffmann (2006), who investigates the word-final epenthesis in English loanwords in the Bantu language Shona (Zimbabwe), the creole language Sranan (Surinam) and the genetically unrelated Oceanic language Samoan (Samoan Islands). It should be mentioned here that the vowel assimilation processes in OHG are not restricted to the epenthetic vowels. In addition to the well-known OHG i-umlaut (not vowel harmony) affecting the stem vowel by palatalization caused by the following (syllabic or non-syllabic) i, vowel harmony applies also to the word-internal vowels which copy the features of the word-final vowel, e.g. OHG wuntar ‘miracle’ vs. wuntor-o ‘miracle-gen.pl’ (see Szczepaniak 2007b). Hence, there is a range of vowel assimilations going in different directions and affecting the “weak” vowels in OHG. I call them “weak” because of their prosodic position – they are unaccented word-medial vowels (so-called Mittelsilbenvokale) – and not because of the information carried by the syllable, which can be e.g. the comparative suffix -ir as in OHG jung-ir ‘young-comp’ vs. jung-or-o ‘young-compnom.masc.sg’. While the phonological word is the prototypical domain of vowel harmony (Hulst and Weijer 1995: 501; Krämer 2003: 24–26), the qualitative levelling in OHG takes place in a much smaller domain. Hence, it does not enhance the coherence of the phonological word, as it does e.g. in Finnish, where the vowel harmony helps to detect the word boundaries (see Vroomen, Tuomainen, and De Gelder 1998). Because of the different directions of feature copying by epenthetic and medial vowels, OHG speakers could not conclude the word position from the harmonized vowel string. It must have been especially difficult in Old Upper German, where the epenthetic vowels also occur in the pretonic position. Additionally, the harmonized vowel strings do not reliably mark morphological boundaries: An epenthetic vowel can copy the features of the stressed stem vowel, but it also can assimilate to the following consonant, e.g. farwa vs. farawa,

168 

 Renata Szczepaniak

faruwa ‘color’. So, the harmonized string does not necessarily correlate with the lexical information. In the case of the so-called Mittelsilbenvokale, the morphological boundary can be blurred by the vowel assimilation, e.g. between the comparative suffix ‑ir and the following morpheme as in OHG jung-ir ‘young-comp’ vs. jung-or-o ‘young-comp-nom.masc.sg’. Hence, the OHG epenthesis brings about an ease of articulation by resolving complex consonant clusters and levelling the vowel quality. It does not enhance the coherence of the phonological word and, additionally, it destabilizes the word size.

3.2 The Early New High German trochee makers The ENHG epenthesis applied only after a range of important word-optimizing processes starting in Middle High German (MHG). Here, the deletion of unstressed (presumably schwa-like) vowels, which started in MHG, has to be pointed out. The apocopes and syncopes have contributed to the predominance of trochaic words and to the development of the trochaic ideal (see Eisenberg 1991; Löhken 1997; Szczepaniak 2007a: 166–181):4 Table 2: The reduction of the size of the phonological word in MHG [σσσ]ω MHG MHG MHG

fa.te.r-es kel.be.r-en en.ge.l-e

[σσ]ω > > >

fa.ter-s ‘father-gen.sg’ kel.b-er-n ‘calf-pl-dat.pl’ en.gel ‘angel[gen.pl]’

According to Moser and Stopp (1970: 267–280), the ENHG epenthetic schwa mainly appears in the structure VV9r(V), i.e. between a stressed long vowel î, iu [yː], û, which diphthongizes as /aɪ/, /oɪ/, /aʊ9/ respectively, and the sonorant r, which can be ᵔ ᵔ followed by a schwa as shown in (2).5 Less frequently, schwa is inserted between a diphthong and l, e.g. MHG vûl > ENHG fawel/faul ‘lazy’. The insertion is also documented before an obstruent, e.g. MHG leip > ENHG leieb/leyb ‘body’.

4 In MHG, vowel deletion was restricted by the Sonority Sequencing Principle. This restriction was then gradually lifted giving rise to syllables with extrasyllabic consonants, e.g. st in gib-st ‘give.pres-2sg’ (< MHG gibest) in ENHG (see section 4.2). 5 According to Moser and Stopp (1970: 267), schwa epenthesis applies after the ENHG diphthongization. They observe, however, the opposite order (i.e. epenthesis followed by diphthongization) in Middle German documents (see also Steffens 1988: 98–100).



Vowel and consonant epentheses in the history of German  

 169

(2) The main prosodic contexts of the ENHG epenthesis MHG Vːr MHG gîr MHG Vːrə MHG mûre

> ENHG VV9r > VVər > ENHG VV9r(e) > VVər >

> > > >

NHG VV9ər NHG Geier ‘vulture’ NHG VV9ər NHG Mauer ‘wall’

Contrary to Paul (1998: 144), I argue that there is a substantial difference between the West Germanic vowel insertion (e.g. Germ *fugl` > OHG fogal ‘bird’) and the ENHG epenthesis. While the former resolves the Germanic consonant clusters and installs a vocalic syllable nucleus, which is in favor of the syllable, the latter does not optimize the syllable structure. Instead, the ENHG epenthesis creates new trochees with stressed diphthong (or long vowel) followed by a naked schwa syllable. The emerging syllable structure, containing a hiatus, is less optimal than the input. The syllable deterioration is particularly apparent by the transition from MHG Vː.rə > (E)NHG VV9ər, where the second syllable changes from CV into VC. Since the second (schwa) syllable is naked, the second part of the preceding diphthong (e.g. ɪ in Geier [ˈgaɪɐ]) becomes ambisyllabic in order to serve as coda ᵔ ᵔ of the first and as head of the second syllable. So, the ambisyllabicity additionally blurs the syllable boundary. The ENHG epenthesis not only creates trochees (cf. Hall 2011: 1577), it also increases the word-internal sonority (ENHG mauer, fawel, leieb) making the whole phonological word better recognizable.6

4 Consonant epentheses in Old and Early New High German In this section, the distribution of the epenthetic consonants in OHG and ENHG will be analyzed. The different motivation of the consonant epenthesis lies not only in the quality, but also in the context and prosodic effect: While the OHG intervocalic epenthetic consonants resolve the hiatuses (see 4.1), the ENHG consonants appear post-consonantically highlighting the right edge of the phonological word (see 4.2). Finally, the insertion of non-contrastive glottal stop, which presumably started in MHG, and the distribution of h, which changed in MHG, contribute to the drift towards a word language (see 4.3).

6 A similar kind of word optimizing schwa epenthesis can be observed in contemporary Luxembourgish. Here, the word pattern with the epenthetic vowels originating from OHG, e.g. Lux. Wurem ‘worm’ (< OHG wurum), is gradually replaced by the new word pattern with the epenthetic schwa in the hiatus position, e.g. Wuerm (see Szczepaniak 2010; Gilles, this volume). This process is part of the word-oriented phonological development in Luxembourgish.

170 

 Renata Szczepaniak

4.1 The Old High German hiatus breakers In OHG, the hiatus position is frequently filled by “weak” consonants: glides [w, j], r and h.7 All these consonants (except r)8 appear in verba pura, i.e. verbs with vowel-final stems as bū+an ‘to live’, and their nominal derivations (Armborst 1979: 27). Hence, the epenthesis resolves the hiatus between a long stem vowel (or a diphthong) and the vowel-initial (inflectional or derivational) suffix. The quality of the epenthetic consonant varies in the OHG dialects (see Table 3). Table 3: The OHG hiatus breaker OHG (except East Franconian)

East Franconian

sāen > sāhen ~ sājen ‘to sow’ bluoan > bluohan ~ bluowan ‘to blossom’ hīen > hījen ~ hīwen ‘to marry’ būan > būwan ‘to live’

sāen > sāwen ‘to sow’ bluoen > bluowen ‘to blossom’ hīen > hīien ‘to marry’ būan > būwan ‘to live’

Before examining the quality of the OHG inserted consonant, I should emphasize that we are dealing here with historical dialectal data with a dominating phonetic-phonological principle of writing.9 In all OHG dialects (except East Franconian), h is most frequently found in verba pura with non-high vowels ā and uo (Braune 2004: 147), while w is documented after high vowels (ū and ī) and j after non-velar vowels (ī and ā) (Braune 2004: 110–111). In East Franconian, the distribution of the hiatus breaker correlates with backness: w follows non-palatal vowels (ā, uo and ū), while palatal stem vowel (ī) demands j. Interestingly,

7 Weak consonants (w, j, r) are placed close to the vowels on the universal sonority scale proposed by Vennemann (1986: 36), where h is not considered because of the fact that it does not combine with other consonants within a syllable head or coda in many languages. Taking into account the synchronic and diachronic phonotactics in word-initial and word-final consonant clusters in Slavic languages (here Polish and Croatian) which do not include any morphological boundary, I hypothesize a (weak) position of h between liquids and fricatives. This assumption is supported by the articulatory properties of h, especially the low degree of impediment of the pulmonic air flow. 8 Since the OHG r epenthesis does not occur very frequently, I will not discuss it here. Braune (2004: 277, 291) mentions some instances of r insertion in hiatus position in the preterite form of strong verbs, e.g. scrir-um ‘shout.pst-3pl’ from scrīan ‘to shout’. 9 The phonetic value of written in the hiatus context is disputed (see Rauch 1973: 777; Armborst 1979: 27). However, Braune (2004: 147) argues that the OHG epenthetic h has the phonetic quality of an aspirate h, since it triggers the same vowel changes as the “old” h leading to the stressed vowel shortening in the dialect of Notker (Braune 2004: 151).



Vowel and consonant epentheses in the history of German  

 171

the quality of the hiatus breaker in East Franconian varies within the paradigm depending on umlaut: bluow-en ‘blossom-inf’ (without umlaut) vs. [ˈblyεjιt] ‘blossom-3sg.pres.ind’ (with umlaut; see Armborst 1979: 29). It is not surprising that glides are used in OHG for hiatus resolution, since they are typical hiatus breakers. Originally, the quality of hiatus breakers was phonetically motivated: The formant transition between adjacent vowels is interpreted as an intermediary glide: i.V > i.jV vs. u.V > u.wV. Although such epenthetic consonants copy the place features from the adjacent vowel and are, thus, minimally contrastive to their environment, they optimize the syllable structure by filling the empty syllable onset. Regarding the quality, the OHG consonant insertion differs in one point from the epenthesis classification proposed by Blevins (2008). She suggests that consonant epenthesis adjacent to a vowel can only have two “natural histories”, where the phonetic properties and the prosodic domain are interrelated. According to her proposal, only the evolution of intervocalic glides is a natural history. “Natural” epenthetic laryngeals (glottal stop and h) are not expected intervocally, but only at prosodic boundaries (of the phonological word or the phonological phrase). The rise of East Franconian consonant epenthesis could then be interpreted as a “natural history”, with an analogical use of w after ā. The quite regular occurrence of an epenthetic h after non-high vowels in other OHG dialects, however, needs further explanation. Obviously, their system of epenthetic consonants is already a result of analogical extension. It can still be assumed that it was h which originally occurred after ā, while the appearance of j as in sājen is due to the later context extension.10 Interestingly, OHG is not the only language exhibiting an intervocalic epenthesis of glottal sounds. It is also reported in certain dialects of Shona, a Bantu language, where glottal stop is inserted in front of a (Mudzingwa 2010). Brunner and Żygis (2011) observe that low vowels are likely to occur with glottal sounds. In their study, they show that glottalized vowels are perceived as lower. They also mention that the cross-linguistic occurrence of low vowels with a glottal sound can be explained by the general property of human articulation, more precisely as a consequence of laryngeal settings, especially a retracted tongue root and higher larynx. Hence, we can assume that the intervocalic epenthesis of glottal sounds is also phonetically motivated.

10 Sonderegger (1959: 152) proposes the following chronology of the OHG consonant epenthesis: 1) w in the eighth/ninth century; 2) h since the ninth century; and 3) j as the youngest epenthetic consonant.

172 

 Renata Szczepaniak

The OHG consonant epenthesis optimizes the syllable by resolving the hiatus position.11 The role of the phonological word is restricted to the domain in which the epenthesis operates. However, the syllable onsets filled by epenthetic consonants do not mark the internal position of the phonological word, since similar onsets, e.g. onset with glides or h are also allowed in the word-initial position, e.g. wurm ‘worm’, jung ‘young’, himil ‘sky’. So, the OHG consonant epenthesis does not make the phonological word more “visible”.

4.2 The Early New High German word edge strengtheners The ENHG consonant epenthesis operates post-consonantally and comprises the insertion of dental, labial and velar plosives (Moser 1951). I will concentrate on the most frequent dental epenthesis, which has also strongly influenced the Standard German vocabulary.12 Since the thirteenth century, dental plosives occur at the direct end of morphosyntactic words (see 3a) as well as at the word boundary in word formations (see 3b), mostly after n, r, s and ch, less frequently also after f, seldom after b/p, and sporadically also after m, k or a vowel (Moser 1951: 49). In the second half of the sixteenth century, the new variants mostly displaced the original forms. (3) lists the prosodic contexts of ENHG dental epentheses which have become part of the Standard German lexicon. (3) The prosodic context of the ENHG dental epenthesis a) At the direct end of a morphosyntactic word MHG ackes > ENHG axt ‘axe’ MHG habech > ENHG habecht ‘hawk’ MHG mâne > ENHG mand/mond ‘moon’ MHG nieman > ENHG niemand ‘nobody’ (lit. ‘no man’) MHG saf > ENHG saft ‘juice, sap’ MHG selbes > ENHG selbst ‘self’ b) Within a morphosyntactic word (before the suffix ‑lich) MHG eigenlîch > ENHG eigentlich ‘actually’ MHG ordenlîch > ENHG ordentlich ‘orderly’

11 In OHG, hiatuses are also avoided in compounds, i.e. at the phonological word boundaries. Gröger (1911: 64) observes that the linking vowels are frequently deleted when following a light syllable in compounds with vowel-initial second part, e.g. klasa.ougi > ‘glass eye, leukoma’. 12 For more details on all ENHG epentheses see Moser (1951).



Vowel and consonant epentheses in the history of German  

 173

The ENHG dental epenthesis occurs only at the right edge of a phonological word, no matter whether it is within or at the end of a morphosyntactic word. Generally, the plosive insertion deteriorates the syllable structure by increasing the complexity of the coda. Looking at the preceding consonant, which is not necessary homorganic with the epenthetic consonant, it can be stated that the insertion is not “natural” in Blevins’ words in all cases. Inserted after f or ch as in ENHG saft ‘juice, sap’ and habecht ‘hawk’ respectively, the dental epenthesis leads to heterorganic complex syllable codas. Generally, the epenthesis highlights the right word edge and makes the phonological word boundaries more easily perceivable: NHG [zaft]ω ‘juice’. The present analysis supports the observation made by Murray (1989) that consonant epenthesis often leads to less preferred syllables. In these cases, it is worthwhile to look at the higher prosodic domains. In the examples given in (3b), t is inserted at the morphological boundary being a phonological boundary at the same time: MHG ordenlîch > ENHG [ordent]ω.[lich]ω ‘orderly’. Interestingly, t is not resyllabified into the left end of the second phonological word, which would lead to an optimal syllable contact n.tl. As a consequence, the morphological structure is emphasized: [ˈɔɐ9dənt]ω[lIç]ω ‘orderly’. The fact that the phonotactic context nl strongly supports the dental epenthesis (nl > ndl) can be explained by the organization of articulatory gesture (Brownman and Goldstein 1991; Page 1997). The perception of insertion is based on the fact that the velic opening precedes the release of the alveolar closure gesture when followed by a liquid. In the ENHG dialects, this insertion is not restricted to the prosodic context given in (3b). The dental appears also in the nl-cluster within a phonological word, e.g. MHG [e.nel]ω > [e.nl`]ω ENHG [en.del]ω ‘grandfather’, marking the boundary between the stem and the suffix ‑el (< MHG ‑el(e) < OHG ‑ilo, ‑ila, ‑ala; Moser 1951: 29). Interestingly, these epenthetic consonants have not been integrated into the Standard German lexicon. The preservation and reanalysis of the epenthetic t as part of the lexical representation in the prosodic context given in (3b), e.g. before the suffix -lich, can be explained with the growing relevance of the phonological word. It is the phonological tendency to strengthen the right word edge by transition from a friction gesture to a total closure which highlights the morphological boundary.

4.3 The glottal stop and h as word edge markers in contemporary German It is difficult to determine when exactly the epenthesis of the glottal stop started. Indirectly, the wide range of external sandhi phenomena indicates at least the non-existence of a word-initial glottal stop in OHG. The word-external elision (gibu ih > gibuh ‘I give’) is viewed by Wilmanns (1911: 352) as indirect evidence

174 

 Renata Szczepaniak

that the glottal stop did not exist in OHG. Clausing draws the same conclusion from the existence of the Notker’s Anlautgesetz in Late OHG: We can therefore infer […] that Notker’s German must have lacked precise syllable and word boundaries, that the language was more flowing and less choppy than modern German is today. This, in turn, leads me to conclude that the glottal stop was not present in Notker’s time, since the use of a glottal stop would institute a pause which would make sound assimilation arising out of close juncture impossible. (Clausing 1979: 363–364)

Generally, the glottal stop insertion is limited to the foot-initial onsetless position in contemporary German (Wiese 2000: 58–60). It is, however, only a vowel-initial strong foot which can be strengthened with a glottal stop. According to Auer (1994), it appears regularly at the left edge of phonological words (4a), even if the first syllable is unstressed (4b). The insertion applies also at the left edge of the strong foot in morphologically transparent complex words (4c) as well as (in the northern variety of Standard German) in hiatus position in loans (4d). The glottal stop is not used in hiatus position within a phonological foot (4e) and in morphologically opaque loans (4f). Hence, the appearance of the glottal stop highlights the most prominent position in the phonological word and is word optimizing (see also Alber 2001). (4) The insertion of glottal stop in contemporary German (Auer 1994: 75) (a) word-initially: [[/aː.təm]F]ω Atem ‘breath’, [[/ap]F]ω[[/ɛḅən]F]ω abebben ‘to ebb away’ (b) word-initially: [[/ɛnt]Fw[/aI9.gnən]Fs]ω enteignen ‘to expropriate’13 (c) foot-initially: [[/aːt.mən]F]ω beatmen ‘to ventilate’ (d) foot-initially: [[kaː]Fw[/oː.tIS]Fs]ω chaotisch ‘chaotic’ (in the northern variety of Standard German) (e) syllable-initially: *[[ˈboː./a]F]ω Boa ‘boa’, *[[ˈkaː./ɔs]F]ω Chaos ‘chaos’, *[[seː./ən]F]ω sehen ‘to see’ (f) *[[poly]Fw[/ɛs.tɐ]Fs]ω Polyester ‘polyester’, *[[ad./ap]Fw[tiːRən]Fs]ω adaptieren ‘to adapt’

The prosodic context of glottal stop insertion comprises not only the word-initial, but also the foot-initial position. Hence, the generalization made by Blevins (2008) should be corrected. She assumes that that the laryngeal epenthesis (of glottal stop or h) is limited to word or phrase boundaries, which must be related to the fact that the pitch contours are initiated by a laryngeal mechanism. In NHG, the foot-initial position frequently aligns with the word-initial position, since

13 For the prosodic status of unstressed prefixes in German, see Raffelsiefen (2000).



Vowel and consonant epentheses in the history of German  

 175

most phonological words are monopedal. Additionally, the phonetic study conducted by Żygis and Pompino-Marschall (2011) shows that glottal stop appears in 85% of words starting with a stressed naked syllable, but only in 60% of words with an unstressed initial syllable. Hence, the insertion of glottal stop is less frequent when the word does not start with the strong foot. The typological contrast between OHG and ENHG consonant epenthesis supports Uffmann’s (2007) division between two different epenthesis strategies depending on the degree of contrast. According to Uffmann (2007: 458), glottal stops are inserted to maximize the contrast with the following vowel, whereas glides guarantee a minimal contrast and, thus, an easy articulation. From the typological perspective, the distributional change of the second glottal sound h, which took place in the course of MHG, is also relevant because this sound also develops the word edge marking function in contemporary German. The OHG aspirate h results from the Germanic guttural spirant χ in the syllable-initial position (Braune 2004: 145). The old spirant χ is preserved in the syllable coda as well as in the geminate. The epenthetic h, discussed in 4.1, conforms with the general syllable-based distribution of both sounds. (5) The syllable-based distribution of the aspirate h and the spirant χ in OHG a) h in the syllable-onset [h]ano ‘rooster’, [h]lūt ‘loud’, se[h]an ‘to see’ b) χ in the syllable-coda nā[χ] ‘close’, blu[χχ]en ‘to burn’, fa[χ]s ‘hair’, wa[χ]san ‘to grow’14

In MHG the general loss of intervocalic h (including the epenthetic h) leads to new foot-internal hiatuses or even to a total contraction, e.g. MHG slahen > slan. The loss of h, a kind of word-internal lenition, increases the foot-internal sonority contributing to the structural emphasis of monopedal phonological words. Today, the aspirate h is restricted to the foot-initial position and, hence, to the left boundary of monopedal phonological words. Its current distribution is similar to the glottis stop (for details see Wiese 2000: 60).15 Interestingly, both glottal sounds, the glottis stop and the aspirate h, are subject to deletion when the phonological word boundaries initiated with one of these sounds are weakened. As observed by Wiese (2000: 60), h is realized in this position only by careful

14 When followed by s, the spirant χ is subject to total assimilation in OHG (twelfth century), e.g. OHG folwa[χ]san > folwassan ‘grown-up’, OHG wahsmo > wasmo ‘growth’, OHG fla[χ]s > flas ‘flax’ (Braune 2004: 150). 15 For word-based development of the OHG semivowels u8 and i8, see Szczepaniak (2007a: 257– 261).

176 

 Renata Szczepaniak

pronunciation. The deletion of h is accompanied by resyllabification of the syllable-final consonant irrespective of existing morphological boundaries. This indicates the weakening of the existing phonological word boundary (see 6a). The phonetic investigation conducted by Bergmann (this volume) reveals that the reduction and deletion of glottal stop indicating the boundary weakening significantly correlates with high lexical frequency (see 6b). Frequent items are more prone to complete deletion of the phonological word boundaries. This “behavior” of both glottal sounds strengthened their word-edge marking function. (6) The loss of h and Ɂ at the phonological word boundaries a) [baːn]ω[hoːf]ω > [baː.nof]ω ‘train station’ b) [baʊ9]ω[ʔamt]ω > [baʊ̣amt]ω ‘building control department’

5 Summary The character of the epentheses occurring in the history of German changes dramatically. In OHG, the inserted vowels and consonants optimize the syllable structure by breaking the hiatuses (e.g. OHG būan > būwan ‘to live’) and resolving the consonant clusters (e.g. OHG wurm > wurum ‘worm’). This goes in line with other OHG phonological processes including assimilations and consonant deletion within complex consonant clusters. Such articulatory simplifications occur not only within, but also across word boundaries, e.g. OHG [guot]ω[līh]ω > [guol]ω [līh]ω ‘glorious’, OHG [friunt]ω[holt]ω > [friun]ω[hold]ω ‘friendly’ (see Gröger 1911; Szczepaniak 2007a: 136–139). What all these processes have in common is not just syllable optimization, which is of course an important argument for classifying OHG as a syllable language. I would like to emphasize that they also destabilize the phonological word by changing the number of syllables (quantitative dimension) as well as the word edges (qualitative dimension). The phonological word, however, can be assumed to exist in OHG – it serves as the operative domain for many processes –, but it is not subject to optimization. The ENHG epenthetic vowels and consonants highlight the phonological word by producing trochees with a high degree of word-internal sonority (MHG mûre > (E)NHG Mauer ‘wall’) and by strengthening the word edges (MHG mâne > (E)NHG Mond ‘moon’). They play an important role in the typological development of German towards a word language. Starting in MHG, word size and word shape became two of the most important parameters for phonological processes: The regulation of the word size was brought about by a series of syncopes and apocopes leading to the rise of the trochaic ideal. Hence, the ENHG vowel epenthesis can be analyzed as an output-oriented (more precisely: word-ori-



Vowel and consonant epentheses in the history of German  

 177

ented) process. As early as MHG the word shape was optimized by reduction of unstressed vowels. Thus, the ENHG vowel epenthesis also contributes to the emphasis of the phonological word. The structural emphasis in these new trochees is additionally reached by the high degree of word-internal sonority. This parameter gains importance in the course of MHG and ENHG: The MHG weakening (i.e. spirantization) of the word-medial plosives (see Simmler 1983: 1122; Auer 1999; Szczepaniak 2007a: 204–207) and the consonant strengthening at the word edges (e.g. the foot-/word-final voice neutralization, so-called Auslautverhärtung) increase the make-up of the phonological words. The ENHG consonant epenthesis emphasizes the right edge of the phonological word by adding a strong consonant and building up word-final consonant clusters, while the glottal stop and h mark its left edge.

References Alber, Birgit (2001): Regional variation and edges: Glottal stop epenthesis and dissimilation in standard and southern varieties of German. Zeitschrift für Sprachwissenschaft 20/1: 3–41. Armborst, David (1979): Old High German (East Franconian) w as a hiatus-filling glide. An example of the implications of rule unordering in the synchronic grammar of Old High German. Zeitschrift für Dialektologie und Linguistik 46/1: 26–32. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (1994): Einige Argumente gegen die Silbe als universale prosodische Hauptkategorie. In: Karl Heinz Ramers, Heinz Vater and Henning Wode (eds.), Universale phonologische Strukturen und Prozesse, 55–78. Tübingen: Niemeyer. Auer, Peter (1997): Areale Variation und phonologische Theorie: Überlegungen am Beispiel der mitteldeutschen ‘Epenthese’. In: Gerhard Stickel (ed.), Varietäten des Deutschen. Regional- und Umgangssprachen, 46–87. (Jahrbuch des Instituts für deutsche Sprache 1996.) Berlin/New York: Walter de Gruyter. Auer, Peter (1999): Variabilität der intervokalischen Position in deutschen Trochäen. Germanistische Linguistik 141–142: 304–333. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Blevins, Juliette (2008): Consonant epenthesis: natural and unnatural histories. In: Jeff Good (ed.), Linguistic Universals and Language Change, 79–107. Oxford: Oxford University Press. Booij, Geert (1995): The Phonology of Dutch. Oxford: Clarendon Press. Braune, Wilhelm (2004): Althochdeutsche Grammatik I. Laut- und Formenlehre. Edited by Ingo Reiffenstein. 15th ed. Tübingen: Niemeyer. Brownman, Catherine and Louis Goldstein (1991): Gestural structures: Distinctiveness, phonological processes, and historical change. In: Ignatius G. Mattingly and Michael

178 

 Renata Szczepaniak

Studdert-Kennedy (eds.), Modularity and the Motor Theory of Speech Perception, 313–338. Hillsdale, NJ: Erlbaum. Brunner, Jana and Marzena Żygis (2011): Why do glottal stops and low vowels like each other? In: Wai-Sum Lee and Eric Zee (eds.), Proceedings of the XVIIth International Congress of Phonetic Sciences, 376–379. Hong Kong: University of Hong Kong. Clausing, Stephen (1979): Notker’s Anlautgesetz. Journal of English and Germanic Philology 78/3: 358–373. Dresher, B. Elan and Aditi Lahiri (1991): The Germanic foot: Metrical coherence in Old English. Linguistic Inquiry 22/2: 251–286. Eisenberg, Peter (1991): Syllabische Struktur und Wortakzent. Prinzipien der Prosodik deutscher Wörter. Zeitschrift für Sprachwissenschaft 10/1: 37–84. Gröger, Otto (1911): Die althochdeutsche und altsächsische Kompositionsfuge mit Verzeichnis der althochdeutschen und altsächsischen Composita. (Gesellschaft für Deutsche Sprache Zürich. Abhandlungen 11.) Zürich: Zürcher & Furrer. Hall, Nancy (2011): Vowel epenthesis. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 3: Phonological Processes, 1576–1596. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Hall, Tracy Alan (1999): The phonological word: A review. In: Tracy Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 1–22. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Howell, Robert B. (1991): Modern evidence for ancient sound changes: Old English breaking and Old High German Vowel epenthesis revisited. In: Elmer H. Antonsen and Hans Henrich Hock (eds.), Stæfcræft. Studies in Germanic Linguistics: Selected Papers from the First and the Second Symposium on Germanic Linguistics, University of Chicago, 24 April 1985 and University of Illinois at Urbana-Champaign, 34 October 1986, 103–113. (Current Issues in Linguistic Theory 79.) Amsterdam/Philadelphia: John Benjamins. Hulst, Harry van der and Jeroen van de Weijer (1995): Vowel harmony. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 495–534. Oxford: Blackwell. Krämer, Martin (2003): Vowel Harmony and Correspondence Theory. (Studies in Generative Grammar 66.) Berlin/New York: Mouton de Gruyter. Kuijpers, Cecile and Wilma van Donselaar (1997): The influence of rhythmic context on schwa epenthesis and schwa deletion in Dutch. Language and Speech 41: 87–108. Löhken, Sylvia C. (1997): Deutsche Wortprosodie. Abschwächungs- und Tilgungsvorgänge. (Studien zur deutschen Grammatik 56.) Tübingen: Stauffenburg. Lombardi, Linda (2003): Markedness and the typology of epenthetic vowels. Unpublished manuscript, University of Maryland, College Park, MD. [Available on Rutgers Optimality Archive, ROA-578]. Maddieson, Ian (2005): Syllable structure. In: Martin Haspelmath, Matthew S. Dryer, David Gil and Bernard Comrie (eds.), The World Atlas of Language Structures, 54–57. Oxford: Oxford University Press. Moser, Hugo and Hugo Stopp (1970): Vokalismus der Nebensilben I (Vokalschwund und Vokalzusatz). (Grammatik des Frühneuhochdeutschen 1.) Heidelberg: Winter. Moser, Virgil (1951): Frühneuhochdeutsche Grammatik. Band I, 3: Lautlehre (Konsonanten, 2. Hälfte (Schluss)). Heidelberg: Winter. Mudzingwa, Calisto (2010): Shona morphophonemics: Repair strategies in Karanga and Zezuru. Ph.D. dissertation, University of British Columbia.



Vowel and consonant epentheses in the history of German  

 179

Murray, Robert W. (1989): On epenthesis. Folia Linguistica 23/3–4: 293–316. Nespor, Marina and Irene Vogel (1986): Prosodic Phonology. (Studies in Generative Grammar 28.) Dordrecht: Foris. Nübling, Damaris and Renata Szczepaniak (2008): On the way from morphology to phonology. German linking elements and the role of the phonological word. Morphology 18: 1–25. Page, B. Richard (1997): Articulatory phonology as a tool for explanation in historical phonology: The case of stop epenthesis in Germanic. In: Irmengard Rauch and Gerald F. Carr (eds.), Insights in Germanic Linguistics II. Classic and Contemporary, 175–188. (Trends in Linguistics. Studies and Monographs 94.) Berlin/New York: Mouton de Gruyter. Paul, Hermann (1998): Mittelhochdeutsche Grammatik. Edited by Peter Wiehl and Siegfried Grosse. 24th ed. Tübingen: Niemeyer. Raffelsiefen, Renate (2000): Evidence for word-internal phonological words in German. In: Rolf Thieroff, Matthias Tamrat, Nanna Fuhrhop and Oliver Teuber (eds.), Deutsche Grammatik in Theorie und Praxis, 43–56. Tübingen: Niemeyer. Rauch, Irmengard (1973): Old High German vocalic clusters. In: Braj B. Kachru, Robert B. Lees, Yakov Malkiel, Angelina Pietrangeli and Sol Saporta (eds.), Issues in Linguistics. Papers in Honor of Henry and Renée Kahane, 774–779. Urbana: University of Illinois Press. Reutercrona, Hans (1920): Svarabhakti und Erleichterungsvokal im Altdeutschen bis ca. 1250. Heidelberg: Winter. Seiler, Friedrich (1874): Die althochdeutsche Übersetzung der Benediktinerregel. Beiträge zur Geschichte der deutschen Sprache und Literatur 1: 402–485. Sievers, Eduard (1901): Grundzüge der Phonetik. Zur Einführung in das Studium der Lautlehre der indogermanischen Sprachen. 5th ed. (Bibliothek indogermanischer Grammatiken 1.) Leipzig: Breitkopf & Härtel. Simmler, Franz (1983): Konsonantenschwächung in den deutschen Dialekten. In: Werner Besch, Ulrich Knoop, Wolfgang Putschke and Herbert Ernst Wiegand (eds.), Dialektologie. Ein Handbuch zur deutschen und allgemeinen Dialektforschung, 1121–1129. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 1/2.) Berlin/New York: Walter de Gruyter. Sonderegger, Stefan (1959): Review of ‘Althochdeutsche Grammatik’ by Wilhelm Braune. Anzeiger für deutsches Altertum und deutsche Literatur 71/4: 145–157. Sonderegger, Stefan (2003): Althochdeutsche Sprache und Literatur. Eine Einführung in das älteste Deutsch. Darstellung und Grammatik. 3rd ed. Berlin/New York: Walter de Gruyter. Steffens, Rudolf (1988): Zur Graphemik domanialer Rechtsquellen aus Mainz (1315–1564). Ein Beitrag zur Geschichte des Frühneuhochdeutschen anhand von Urbaren. (Mainzer Studien zur Sprach- und Volksforschung 13.) Stuttgart: Steiner. Szczepaniak, Renata (2007a): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Szczepaniak, Renata (2007b): Vokalharmonie im Althochdeutschen und im Walserdeutschen. Ein Fall von phonologisch-typologischer Kontinuität. Zeitschrift für Dialektologie und Linguistik 74/1: 38–60. Szczepaniak, Renata (2009): Silbensprachen versus Wortsprachen. Natur & Geist. Das Forschungsmagazin der Johannes Gutenberg-Universität Mainz 25/2: 49–52.

180 

 Renata Szczepaniak

Szczepaniak, Renata (2010): Phonologisch-typologischer Wandel des Deutschen und des Luxemburgischen im Kontrast. In: Antje Dammel, Sebastian Kürschner and Damaris Nübling (eds.), Kontrastive Germanistische Linguistik, vol. 1, 85–110. (Germanistische Linguistik 206/209.) Hildesheim: Olms. Uffmann, Christian (2006): Epenthetic vowel quality in loanwords: Empirical and formal issues. Lingua 116/7: 1079–1111. Uffmann, Christian (2007): Intrusive [r] and optimal epenthetic consonants. Language Sciences 29/2–3: 451–476. Vennemann, Theo (1986): Neuere Entwicklungen in der Phonologie. Berlin/New York: Mouton de Gruyter. Vennemann, Theo (1988): Preference Laws for Syllable Structure and the Explanation of Sound Change. With Special Reference to German, Germanic, Italian, and Latin. Berlin/ New York: Mouton de Gruyter. Vroomen, Jean, Jyrki Tuomainen and Beatrice de Gelder (1998): The roles of word stress and vowel harmony in speech segmentation. Journal of Memory and Language 38: 133–149. Wiese, Richard (2000): The Phonology of German. 2nd ed. (The Phonology of the World’s Languages.) Oxford: Oxford University Press. Wilmanns, Wilhelm (1911): Lautlehre. 3rd ed. (Deutsche Grammatik. Gotisch, Alt-, Mittel- und Neuhochdeutsch 1.) Strassburg. Żygis, Marzena and Bernd Pompino-Marschall (2012): Glottale Markierungen vokalinitialer Wörter: Deutsch und Polnisch im Kontrast. In: Lutz Gunkel and Gisela Zifonun (eds.), Deutsch im Sprachvergleich. Grammatische Kontraste und Konvergenzen, 75–98. (Institut für Deutsche Sprache. Jahrbuch 2011.) Berlin: De Gruyter.

Part 2: Diachronic approaches

Kurt Braunmüller (University of Hamburg)

Scandinavian word phonology: Evidence for a typological cycle* Abstract: The purpose of this paper is to show that there is evidence for the existence of a typological cycle in the history of the Scandinavian languages which counteracts, at least in part, the general drift in other Germanic languages from a syllable to a word language. The reason for this opposite development is that new syllable language structures came into being due to clitization in these languages, both in the noun and in the verb phrase. The result is a mixed type of language (viz. a predominantly word language type in the lexeme and a syllable language with regard to the new clitics). These enclitic pro­cesses emerged during the early Middle Ages and reflect structures in definite noun phrases and in (synthetic) passive verb phrases that are – again – more characteristic of syllable than word languages. The conclusion to be drawn from these observations is that morphological fusion processes from analytic to synthetic structures may stop and even reverse the typological drift from syllable to word lan­guages, at least partially. Moreover, outcomes of language contact and especially restructurings due language cultivation may also have a considerable influence on the evolution of languages.

1 The Germanic languages: Some general structural patterns and drift phenomena This contribution deals with the issue of whether there is evidence for a typological cycle in the Scandinavian languages that deviates from the general tendency or drift from predominantly syllable languages in Pre-/Proto-Germanic and ancient Germanic dialects towards word languages in the modern Germanic languages. The Scandinavian languages have also undergone an opposite development or coun­ter-drift, giving rise to new structures that must be acknowledged as characteristic (again) of syllable languages. These new structures are mainly due to

* I would like to thank Stig Eliasson, Steffen Höder and the editors for their helpful comments and suggestions, which have improved my paper. All remaining deficiencies and errors are, of course, mine.

184 

 Kurt Braunmüller

two (independent?) clitization processes and affect both the nominal and verbal systems of these languages. With the rise of enclitic definite articles in noun phrases and the emergence of a multifunctional medio-pas­sive suffix, a new typological cycle seems to have come into being, comparable to the observation made by von der Gabelentz ([1891] 1972: 255‒258) that he called “De[n] Spirallauf der Sprachgeschichte” [The helix in language history]. He further pointed out that there has always been a competition between the ease of articulation and the need to keep relevant distinctions alive.1 Common to all Germanic languages are, among others, the following wordbased/word-optimizing phonological developments: (a) Systematic consonantal shift from Indo-European /b, d, g/ to /p, t, k/, from /bh, dh, gh/2 to /b, d, g/ and from /p, t, k/ to either voiceless /f, θ, χ/ or voiced fricatives /β, ð, ɣ/, depending on divergent stress patterns in Pre-Germanic (Gmc)/ Indo-European, gen­erally known as Grimm’s Law and Verner’s Law, respectively.3 (b) The subsequent and default placement of stress on the first syllable of a word (often accompanied by a plosive aspiration) which is still the default pattern in Modern Icelandic and, to a certain extent, in Faroese. In one of the Mainland Scandinavian languages, Danish, this typical Germanic stress pattern still produces further phonological consequences and has gi­ ven rise, among other things, to new affricates (due to the so-called “Akzent­ballung” [accumulation of

1 “Nun bewegt sich die Geschichte der Sprachen in der Diagonale zweier Kräfte: des Bequemlichkeitstriebes, der zur Abnutzung der Laute führt, und des Deutlichkeitstriebes, der jene Abnutzung nicht zur Zerstörung der Sprache ausarten lässt. Die Affixe verschleifen sich, verschwinden am Ende spurlos; ihre Functionen aber oder ähnliche bleiben und drängen wieder nach Ausdruck. Diesen Ausdruck erhalten sie, nach der Methode der isolirenden Sprachen, durch Wortstellung oder verdeutlichende Wörter. Letztere unterliegen wiederum mit der Zeit dem Agglutinations­ prozesse, dem Verschliffe und Schwunde, und derweile bereitet sich für das Verderbende neuer Ersatz vor: periphrastische Ausdrücke werden bevorzugt […]” (von der Gabelentz 1972: 256 [all emphases are mine]). [The history of languages vacillates along the vector of two forces: that one of establishing ease, which leads to the wear of sounds on the one hand, and that one of establishing distinctiveness, which keeps wearing forces away from destroying the language. The affixes get slurred and finally disappear without a trace; their functions or similar forces remain, however, and try to gain expression again. They obtain such an expression along the method of isolating languages, by word order or by explanatory words. The latter, however, are subject to the agglutinating process as time goes by, i.e. slurring and loss, whereas new replacing elements arise in order to substitute the ruined forms: periphrastic forms are preferred […]; translation and emphases, K.B.] 2 These sounds/phonemes are only attested in eastern Indo-European dialects. 3 /p, t, k/ > /f, θ, χ/ or /β, ð, ɣ/, which is in my opinion a bifurcating development based on Pre- or (latest) Early Germa­nic stress placement: /b, d, g/ > /p, t, k/ and /bʰ, dʰ, gʰ/ > /β, ð, ɣ/ (> /b, d, g/ in West Germanic).



Scandinavian word phonology: Evidence for a typological cycle  

 185

accent]; cf. Bandle ([1973] 2011: 67 and map 15); e.g. Gmc [th-] before a stressed /i/ becomes [d̥s-], as in Dan. tidlig ‘early’). But also geographically more restricted developments such as the (early medieval) Old High German (OHG) sound shift with its development of aspirated consonants, voiceless fricatives and especially its affricates, might be seen as a consequence, or at least a by-product, of this ancient Germanic stress placement rule (cf. Swiss Germ. Kind [ˈkXɪnd̥h] ‘child’ < Gmc *kinþa- or Germ. Pfaffe ‘[pejorative for] priest’ < Lat. papa). (c1) The subsequent reduction of vowels in unstressed syllables to a sub­set of the vowels of stress­ed syllables (such as in Gothic to /i, a, u/ or, even more reduced, in spoken Modern Faroese to /e [< ɪ, ʊ], a/4 where the reduced final /e/ has taken over other weak-stressed allophones). (c2) In many cases, this common Germanic de­velopment has led to a general re­duction of unstressed vowels to schwa (/ə/) or even the complete loss of these vowels (viz. apocope: -∅). Other stress patterns occur as well but are – in almost all cases – restricted to loan words and due to foreign influence (e.g. from Latin, Classical Greek, Anglo-Norman, French or Italian) or later became distinctive in some word formation patterns, as illustrated by the following examples from German where the vowels in bold represent a (contrasting) word accent: übersetzen ‘to put over’ vs. übersetzen ‘to translate’. Most verbal prefixes, however, are unstressed (cf. Germ. bezahlen, Dan. betale ‘to pay’; Germ. erzählen, Dan. fortælle ‘to tell’, Germ. verstehen, Swed. förstå ‘to understand’ etc.). Nominal prefixes are normally stressed, however (cf. e.g. Engl. uncertain, indispensable, non-standard, semiprofes­sional). If this phonological drift had continued unimpeded, accompanied by syncope in unstressed syllables, it would, in principle, have led to the dominance of monosyllabic words with complex consonant clusters in the coda. Moreover, the loss of all inflectional markers that terminate in vowels may be one consequence of such a development and may, in some cases, give rise to the origin of (simple) tone languages because there would no longer be any way to avoid the occurrence of (new) homonyms. There is evidence for such a devel­op­ment in some south-east­ern Jutish dialects (see Braunmüller [1987] 1995b),5 in Low

4 When reading aloud, a morphologically conditioned distribution may occur as well: {-ur} in singular and {-ir} in plu­ral forms (for more details see Selås 1997). 5 E.g. hus ‘house.sg’ vs. hus ‘house.pl’ with tonal accent, drek! ‘drink.imp’ vs. drek ‘drink.inf’ with tonal accent, to mention only two of a total of nine instances. Even parts of speech can be kept apart by using tone distinctions in south-eastern Jutish dialects: gul [ˈguʎ] ‘gold.n.sg’ vs. gul [guʎ] ‘golden.adj.sg.utr’ (utr = common gender which represents both m and f).

186 

 Kurt Braunmüller

German and in some Norwegian dialects, as well, but a ra­di­cal drift towards a complete typological change in the Germanic (or Scandinavian) languages is still a far way off. Apo­co­pe or, at least, the mer­ger of unstressed vowels into schwa /ə/, however, is quite typical for many modern Germanic languages, especially in their spo­ken varieties (cf. Low and High German dialects, Dutch, Danish and Dano-Norwegian). The only exceptions are Ice­landic and, to a lesser degree, Faroese, New Norwegian and Swedish (due to language planning; cf. section 6.4).

2 Intervening factors As is well known, languages do not develop according to only one single morpho-typological principle. Most languages reflect a mixture of more than one word structure type. English for example may be classified as overwhelmingly isolating in type but still shows some features of an inflectional or agglutinative type (cf. the plural forms -(e)s, the preterit and past participles -ed, etc.). The following factors may, however, interfere with a straightforward cyclical typological development, as suggested by von der Gabelentz (cf. footnote 1). In the case of the Scandinavian languages, the following developments also play an important role. All of them have consequences for the phonology of these languages: (I) The emergence of new clitics, both in the nominal and the verbal part of grammar. In Ancient Norse (AN), the demonstrative pronoun hinn/hin/ hit ‘this.nom.sg.m/f/n’ became cliticized as -inn/-in/-it ‘the.nom.sg.m/ f/n’, which have become a definite article and two personal pronouns in the accusative as well. AN mik ‘me.1.acc.sg’ and sik ‘him-/herself.3.acc. sg’ became cliticized as ver­bal affixes in Old Norse (ON): -mk and -sk, which later developed into -sk/-st. This clitic marker then ac­quired the function of a medio-pas­sive marker. In the Scandinavian languages these forms which developed into -s/(-st) later took on multiple functions and are used today to indicate passive, reflexive, reciprocal and in some cases verb deponent functions as well. (II) Language contact at various stages of their history which led, among other things, to vowel harmony (probably due to contact with Sámi dialects in central Scandinavia) and new word-formation elements (mostly borrowed from Middle Low German during the era of the Hanseatic League in the late Middle Ages). (III) Language cultivation and language plan­ ning which inhibited or even reversed some phonological developments, especially those in



Scandinavian word phonology: Evidence for a typological cycle  

 187

unstressed syllables. This factor plays an important role in the history of Icelandic, Faroese, New Norwegian and even Swedish. All these factors contribute to impeding an in­herent pho­nological drift in Germanic to a more or less monosyllabic language with a rich inventory of nuclei (many vowels, several diphthongs and even tones), con­so­nant clusters both word initially/in the onset (1–3 consonants [Cs]) and, to an even greater extent, in the coda (with a maximum of 5 Cs in word-final position) but normally not within morphemes.

3 Splitting up Proto-Germanic: Mainly a case of vowel change? One of the salient phenomena when Proto-Germanic developed into distinct dialects was the relative con­tinuity of the consonantal frames while the vowels, both the stressed and unstressed ones, changed quite radically. The only exception is (Old) High Ger­man where unvoiced plosives developed into fricatives and affricates (often in comple­mentary distribution), the so-called Second or Old High German Consonant Shift (cf. OSax./WGmc appul > OHG aphul [ˈap͡fʊl] ‘apple’ or OHG zimb(o)rōn [ˈt͡sɪmb(ɔ)roːn] ‘to build’; cf. Eng. tim­ber). However, in eastern Germanic and especially in the northern Germanic varieties, the main changes oc­cur­red in the vowel systems (such as umlaut) where the loss of vowel length was accompanied by a qualitative reduction of long vowels to phonetically divergent short vowels. As a consequence of this development, reductions such as apocope and syncope occurred in unstressed syllables. The following examples give a short paradigmatic overview of this development in Early Germanic. The starting point (1) is the inscription of Gallehus (about 425 AD, found in Denmark near the German border, written in the so-called older futhark). It is clearly readable and its interpretation is quite uncontroversial. (1) ek hlewagastiR holtijaR horna tawido6 1sg Hlewagast.nom.sg.m Holt.gen.sg.m horn.acc.sg.n make.1.sg.pst.ind

‘I Legest, son of/from Holt, made [the] horn’.

The Proto-Germanic reconstruction of this sentence is rendered in (2) (following Niel­sen 2000: 78‒79, but with some modifications and corrections; the glosses and the translations remain the same in (2)–(7)):

6 The final R still represents [z] and not yet [r], as it did in later times.

188 

 Kurt Braunmüller

(2) *ek[a] hlewaχastiz hultijaz hurnan tawiðōn (Proto-Germanic)

The hypothetical Gothic (3), Medieval Old Norse (4) and Old English (5) versions are as follows: (3) *ik hliugasts hulteis haúrn tawida (Gothic) /horn/ (4) *ek hlégestr hyltir horn g{e|ø}rða7 [**táða]8 (Old Norse/Old West Scandinavian) (5) *ic hléogiest hylte horn tāwode (Old English) /ik/

Only Old High German (6) differs considerably from the other old Germanic dialects due to the Old High German Con­sonant Shift, which marks this variety as clearly divergent from all other Germanic dialects. The deviance of the modern Danish translation in (7) is mostly due to other non-phonological factors, however (the only new phonolo­gi­cal change is the so-called Pan-Scandinavian breaking: Gmc e > ja /__ a: *eka > jak > /iɛ{k|g}/ or > /jai/ jeg, with a double consonantal weakening of /k/ > /g/ > /i/). Moreover, it shows the strongest reductions in phonology: (6) *ih [hleo]gast hulzi horn zawita (Old High German) /iç/ /hultsi/ /t͡savita/ (7) Jeg Legest lavede [dette] horn (Modern Danish) ‘I Legest made this horn’.

These examples clearly show that apocope (-a, -(a)n > -∅), umlaut (u > o; u > y; a > e), syncope – especially the loss of inter-vocalic consonants (-ewa- > -eo/-ē; -awi > a; -ija[z]/r > ‑ir/-eis) – and the reduction of long vowels to miscellaneous short ones in unstressed positions (-ō > a, -a(n) > e [ɪ |ə]?) are mainly responsible for the divergences observed. All these developments in North Germanic fit into a continuous and very strong shift from a syllable to a word language. The typological turnover must have taken place by the end of the Viking Age (i.e. in the late eleventh century) but did not become visible before the end of the twelfth century when writing on

7 There are sometimes two (ablauting) stems in North Germanic varieties (cf. also the relative particle sem/som). 8 This verb is only attested in the oldest runic inscriptions and not represented in all Germanic dialects.



Scandinavian word phonology: Evidence for a typological cycle  

 189

parchment with Roman letters became the default means for writing documents of any kind. The re-assignment of pho­nemic length from (predominantly) the vowel to the syllable level during the later Middle Ages has led to several far-reach­ing restructurings in word structure as well. The ruling doctrine9 says that a so-called Great Vo­wel Shift, equi­va­lent to a similar development in English language history, took place sometime be­tween the thirteenth and the mid-fourteenth century and started out by changing the vowel quality of the low vowel /a/ with subsequent and direct consequences for the back vowels /o/ and /u/. To give an example: The originally long a /a:/ (normally written á) in Old Scandinavian became either /o/ or /o:/ (written aa or å) in Mainland Scandinavian, depending on the syllable structure (long vowels after a single (C) or no consonants but short vowels after two consonants (C1C2) or a long consonant (Cː)), but /au/ or /auː/ in Icelandic and /ɔa/ or /ɔ/ in Faroese (in both cases written á), respectively. Old long /oː/ then developed – according to the ruling doctrine – into /u(ː)/, and /u(ː)/ finally into /ʉ(ː)/, where this shift termina­ted; but /ʉ(ː)/ did not merge with the fronted u due to earlier umlauting processes: /y(ː)/. Bear in mind now that the Great Vowel Shift has taken place, the length of the nucleus vowel depends on the phonological structure of the stressed syllable: In open syllables and in syllables with only one word-final consonant in the coda, the vowel is normally long; in all other cases the vowel is short. Exceptions, e.g. long vowels before consonant clusters, are in almost all cases due to additional morphological affixation or inflectional marking: Swed. svans ‘swan.g.sg’ [ˈsva:ns] < svan ‘swan.n.sg’ [ˈsva:n] + -s ‘g.sg’ and not *[ˈsvans]. svans [ˈsvans] ‘tail.n.sg’ also exists but as a completely different word where the s is part of the lexeme. The main purpose for keeping vowel length unchanged is based on paradigmatic reasons: The vowel length should not change within a paradigm. Or, to put it in

9 We do not agree with the widespread view that a so-called push chain development has taken place which made the vowel quality in the Mainland Scandinavian languages change, beginning with the shift of the (originally long and allophonic dark) /a/ to /o/, then from /o/ to /u/ and finally from /u/ to a fronted /ʉ/, with many exceptions, however. A more detailed survey of all relevant arguments and the ruling doctrine can be retrieved from Eliasson (1992). This typological account is, in our view, not able to give an explanation of this shift and describes by and large only what is obvious and what can be observed when comparing the findings at the beginning and at the end of this restructuring process. This account, however, is not able to explain why, in which direction (can a drag chain development definitely be excluded?) or how this development has taken place on the phonetic level. For a phonetic- and cognition-based approach that is able to explain the parallel developments in the Insular Scandinavian languages as well, see Braunmüller (2003).

190 

 Kurt Braunmüller

a different way, word phonology principles override syllable structure principles (for a far more detailed account of this very complex issue, see Eliasson 2010, where quantity in Swedish and German is contrasted). Divergent developments in realigning phonemic length after the so-called Great Vowel Shift can also be observed, apparently caused by the competing application of phonological rules when restructuring the syllable in late Medieval Scandinavian, as seen in examples (8a) and (8b): (8) Common Medieval Scandinavian skip /skip/ ‘ship’ became either a. skib /ski:Ɂb/ in Danish or skip /ʃi:p/ in Norwegian and Faroese but b. skepp /ʃep:/ in Swedish.

This means that either the vocalic or the consonantal parts of the stressed syllable were lengthened in order to obtain only long syllables: Vː(C), [VV]D(C); VCː, VCC. Genuine extra-long or “heavy” syllables are still possible, however (e.g. in West Norwegian dialects, e.g. haust C[VV]DCC ‘autumn, harvest’), but in most cases these extra-long syllables are due to gram­matical af­fixation, e.g. in Swed. fint [ˈfi:nt] ‘fine.adj.n’ + t ‘n.sg/adv’ or the svans example mentioned above (cf. also Braunmüller 2007: 43‒45).

4 On the phonotactics and word structure of Proto-Germanic and Ancient Nordic Proto-Germanic and Ancient Nordic, the ancestral language of all modern Scandi­ na­vian varieties, predominantly show features that are common to syllable languages but also have acquired some features which are typical of word languages: (1) All vo­wels can occur in any syllable type, i.e. there are no different sets of stres­sed and un­stressed vowels, which is typical of syllable languages (cf. e.g. Finnish). (2) Consonant clusters do occur but they are still less com­plex compared to Old Norse (and Medieval Scandinavian in gen­eral) and less fre­quent. In the word-initial onset for instance we find the following clusters: (2) a. #CC–: fl-, fr-, gl-, hl-, wr-, þr-[θr-]. Clusters with #CCC–: str- are not attested in Ancient Nordic sources. In the interlude, in word-medial position, more than two con­sonants are not possible:



Scandinavian word phonology: Evidence for a typological cycle  

 191

(2) b. –CC–: e.g. -lg-, -kl-, -ng-, -rb-, -rk-, -st- (some of them can be split up into two syllables, with the consequence that these clusters disappear), whereas in the coda cluster or in word-final position two and sometimes even three consonants may occur, a development which became more frequent in Old Norse: (2) c. –(C)CC#: e.g. -fR, -kl, -lt, -mR, -ng, -rl, -rn, -st, -tR and -htr, -lfs, where the last conso­ nants (-r and -s, respectively) are due to inflectional mor­pho­logy (often -r ‘nom.sg.m’ < -ur and -s ‘gen.sg.m/n’ < -as/-is). (3) a. The already-mentioned dynamic stress on the first syllable,10 which later be­came ty­pi­cal of all Germanic languages, was introduced very early but without immediate conse­ quences for the vocalism of the unstressed syllables. (3) b. In Scandinavian, vowel reduction in unstressed syllables did not occur (visibly) before the com­plete restruc­tur­ing of the whole runic al­pha­bet had come to an end in the early ninth century. In other words, apocope and syncope may have occurred much earlier but not have been marked in writing yet. (3) c. In the High Middle Ages, we find ap­pro­­xi­mate­ly the same conso­nan­tal struc­tures in Old Norse/Medieval Scandina­vian and in its contact language, Middle Low German (for details see Braunmüller 1989: 18‒24), but there are now many more clusters and long conso­ nants at­tes­ted.

5 West Scandinavian: A typological change from a predominantly syllable language to a word language West Scandinavian language history has been greatly influenced by normative and very con­servative writing practices which obscure “natural” phonological developments that would have taken place during the last six to seven centuries, e.g. the reduction or at least the cen­tralization of unstressed vowels and the collapse of minor phonological op­po­sitions with low functional load.11 Modern Icelandic reflects this bifurcating development in an ex­treme way: The writing has by and large not changed since the High Middle Ages (dis­re­gar­d­ing some mar-

10 It should be emphasized, however, that word accents themselves cannot automatically be regarded as a typical feature of word languages. Finnish, for example, also has a word accent on the first syllable but must be considered a clear case of a syllable language due to its syllable structure which, among other things, prevents consonant clusters. 11 Cf. Icelandic in Canada where the fusion of mid-high and mid-low vowels, called flámæli ‘[lit.] flat way of pronouncing’, occurs as a default. Teachers in Iceland, however, are trying to fight against this more or less “natural” phonological change (for more details see Braunmüller 2007: 239‒240).

192 

 Kurt Braunmüller

ginal orthographic changes such as ok > og ‘and’ or ‑it > -ið in the de­finite clitic neuter article) due to strict language culti­vation which prohibited any form of pho­nological in­novation. The current pro­nunciation, how­ever, di­verges tremendously from this medieval form of writing (and reading aloud) Icelandic texts. Moreover, weak-stressed vowels repre­sent clear cases of spelling pronunciation. The reconstruction of some Swedish word-final syllables from the dialects which took place in much later times will be dis­cussed in section 6.4. Almost the same can be said about the Faroese phonological history as well: Faroese received a historically-based orthography in the mid-ninteenth century which was based primarily on the normative writing of Old Norse texts (such as Icelandic) with some exceptions (þ/Þ- has been replaced by t/T- and -ð(-) has been retained though it is no longer pro­nounced in any Faroese dialect). Due to the lack of any script for this West Scandinavian variety for more than half a millennium, the phonological development of Faroese went even further (cf. Braunmüller 2007: 288‒297). Moreover, the Icelandic language and language policy has been considered the guiding principle by most teachers and researchers on the Faroe Islands. One of the results is that unstressed vowels are pronounced when read aloud as they are written in the script, notwithstanding all reductions in colloquial speech. Medieval West Scandinavian and its modern successors, Icelandic and Faroese, show signs of developing from typical syllable to word languages, among others the following developments: (A) Consonant clustering at the word edges: Palatalization in the onset before mid and high front vowels may reduce consonant clustering in word-initial position, as in ON skip [ˈskiːph] > Far. [ˈʃiːph] ‘ship’, but it can also give rise to new phonemes (affricates) which increases the inventory of consonantal phonemes; cf. e.g. West Scand. gef/va ‘to give’ [+voice] and köt­tur/ket­ta ‘cat’ [-voice], are pronounced /gj- ~ ɟ-; kj- ~ c-/ in Icelandic but /ʤ-; ʧ-/, /ʃ-/ in Fa­ro­ese. Instances of consonant clustering in the coda of mono­syl­la­bic words in Faroese are kúgv ‘cow’ or plógv ‘plow’ /-igv; -ɛgv/, where the ori­­gin­ally long high and mid vowels are fronted and short­ened: /uː > i; oː > ɛ/ which has been called “sharpening” (Schärfung, viz. the stabilizing of the consonantal coda). Sharpening thus changes both the quantity and quality of the nucleus and shows that strengthening the coda may have direct and far-reaching consequences for the nucleus of these (monosyllabic) words. Even con­sonant scrambling in co­da clus­ters is found, e.g. in Far. danskt ‘Dan­ish.nom-acc.n.sg/adv’ or feskt ‘fresh. nom-acc.n.sg/adv’: -skt > /-kst/. Danish has abolished the -t in these cases, and Swedish drops the inter­me­diate -k- in -skt- patterns, as for example in /hemskt/ >



Scandinavian word phonology: Evidence for a typological cycle  

 193

/hemst/ ‘hor­rible/y.nom-acc.n.sg/adv’ or in /svenskt/ > /svenst/ ‘Swedish.nomacc.n.sg/adv’. (B) The occurrence of complex syllables (with consonant clusters in the onset and coda or long consonants with or without pre-aspiration in the interlude before e.g. un­voiced long con­so­nants), e.g. in Icel. hattur ‘hat’ /-h.t-/ vs. the earlier unaspirated finna /-nː-/ ‘to find’. In Faroese, however, there is no such phonemic split of consonant clusters. The in­di­cator for the fact that the syllable boundary (.) is situated at the end of the consonant cluster, viz. at the end of the lexeme itself, can be observed from pro­nun­ciation: Simple con­sonants will always lead to a long, or in the case of hattur, to a di­phthongal rea­lization of the preceding nucleus, which does not take place in cases like Far. hattur ‘hat.nom.m.sg’, which is pronounced /haht.ur/ (and not */hɛah.tur/). (C) More complexity in stressed than in unstressed syllables (cf. ON/West Scand. fiskr ‘fish.nom.m.sg.’ vs. fisku­num ‘fish.pl-the.dat.m.pl’ or ON vetr ‘winter.sg.m’ > Icel./Far. vetur). (D) Long vowels and long consonants can only occur in stressed syllables in all Medieval Scandinavian languages. In other words, length is reinterpreted as a feature derived from stress. In Pre- and An­cient Germanic this correlation between stress and vocalic length did not yet exist. (E) Tendency towards a reduction of unstressed inflectional syllables, but only in spoken Fa­ro­ese (cf. section 1.2, c1). Reductions in names of islands containing oy (or the older form ey) may also occur, e.g. Nólsoy /-i/ and their inhabitants such as in Føroyingur /ˈfø:riŋgur/. (F) Long syllables are found quite frequently, many of them due to the lengthening of syl­lab­les/words in the Late Middle Ages (cf. vowel lengthening in words like koma /o > oː/ ‘to come’). When long syllabic consonants which originated from progressive as­si­milation (-lR > -ll or -nR > -nn) occur, they are devoiced and split­up into two very distinct consonants, which might be called “consonantal di­phthon­gization”. (See Icel. kalla ‘to call’ /-dl-/ in me­dial and jökull ‘glacier’ /-d̥l̥/ and steinn ‘stone’ /-d̥l̥/ in word-final position. The only difference is final devoicing which is one of the most natural choices in this position.) (G) Hiatus occurs in some Faroese dialects but most dialects have inserted glides (cf. blá­ur ‘blue’: [ˈblɔa(v)ʊɹ]). In contrast to these West Scandinavian varieties, the second vowel is often omitted in Swedish, a language without genuine pho-

194 

 Kurt Braunmüller

nematic diphthongs, as in e.g. Europa ‘Europe’: [eˈruːpha] or augusti ‘August’: [a(ɵ)ˈgɵsˑti]. Other pos­sibilities are to insert a glide (/j;w:[v]/): [evˈruːpha] or a junc­ture ([...+...]): neutral ‘neu­­tral’ [ne+ɵˈthraːl] which splits the nucleus into two vocalic parts. (H) There is no vowel harmony in any western Scandinavian language or dialect.

6 East Scandinavian: Evidence of a typological cycle and counter-drift 6.1 The consequences of retroflex consonants in unstressed syllables As shown in section 5, the evolution of the West Scandinavian languages runs in the same direction as that of most other Germanic languages, e.g. German (see Szczepaniak 2007a). Although all Scandinavian languages developed an enclitic definite article and a synthetic medio-passive voice during the Viking Age, the resulting phonotactics in the western and eastern language groups are rather different. An important precondition for this divergent development is a special kind of sandhi, i.e. the formation of retroflex consonants (henceforth R), viz. clusters containing an apical r before /t, d, n, s, l/ which is realized as [ʈ, ɖ, ɳ, ʂ, ɭ]. There are also ten­dencies towards retroflex formation in modern Faroese but they are weak (cf. Hagström 1970). Normally clusters like -rn- are pronounced distinctly, ei­ther as /‑rn/ as in torn ‘tower’ and inflections such as -irnir, -rnar ‘def. pl’, or as /-d̥n̥/ (cf. horn ‘horn’, bjørn ‘bear’). These pronunciations are probably due to influence from the second (formerly national) language spoken on the Faroe Islands, Danish. In other cases, the preceding r is omitted, e.g. in words like torskur ‘cod’ /toskur/, even in the definite ar­ticles in their plural forms (cf. Far. hundarnir ‘the dogs’ > /hunda­­nir/). This is, partially, equivalent to a simplification or reduction of a re­tro­flex cluster, as this can be observed in both Norwegian varieties as well (cf. Da­no-Norw. hundene and New Norw. hundane ‘dog.pl.def’ < -e.pl-ne.def and -ar.pl-ne.def). This form of contact assimilation must be seen as a clear case of sandhi because it crosses syllable, morpheme and even word boundaries (cf. the Swedish compound jord|skalv ‘earth|quake’ /-ɖ | ʂ-/12 and the Swedish sentence Han kommer snart hem ‘He comes home soon’, which con­tains both intra-:

12 The Danish equivalent jord | skælv is pronounced completely differently because of the velar



Scandinavian word phonology: Evidence for a typological cycle  

 195

/ʈ/  and trans-word retroflex consonants: /ʂ/ . More details on this very complex phenomenon can be retrieved from Eliasson (1986).

6.2 The rise of new enclitic inflections: A starting point for a new typological cycle The rise of a second inflectional paradigm in the nominal inflection marks the beginning of a new ty­po­logical cycle with unreduced, unstressed vowels and predo­mi­nant­ly ba­lanced sequences of consonants and vowels, which are characteristic of syllable lan­guages, cf. Old Norse: (11a) Singular: dagr-inn (< dagr + [h]inn) /-in/13: -VC# day.nom.sg.m –the.art.def.nom.sg.m day.nom.sg.m + this.dem.nom.sg.m dags-ins /-ins/: -VCC# day.gen.sg.m –the.art.def.gen.sg.m degi-([h]i)num /-inum/: -VCVC# day.dat.sg.m –the.art.def.dat.sg.m dag-inn /-in/: -VC# day.acc.sg.m –the.art.def.acc.sg.m ‘the day’ (11b) Plural: dagar-nir /-nir/: -CVC# day.nom.pl.m–the.art.def.nom.pl.m daga-nna /-n(ː)a/: -CV# day.gen.pl.m–the.art.def.gen.pl.m dǫgu-num < dǫgum + ([h]in)um /-num/: -CVC# day.dem.pl.m–the.art.def.dem.pl.m but in Old Swed. døgum-in /-in/: -VC# day.dat.pl.m.–the.art.def.dat.pl.m daga-na < daga + (hi)na /-na/: -CV# day.acc.pl.m–the.art.def.acc.pl.m

Moreover, the newly de­vel­oped inflection exhibits an iconic mar­k­ing in the singular pa­radigm: All forms contain an i (marked in bold) as an overt definite singular marker, which still can be seen in the modern West Scandi­na­vian languages (cf. Braunmüller 2001: 79).

r [ʀ] and the deletion of d in rd codas: /ˈjoʀˌskɛl(v)Ɂ/. In the phonetic transcription the r becomes vocalized: [ˈjɔɐˌsɡ̊ɛl(v)Ɂ]. 13 Long consonants normally do not occur in unstressed syllables (thus -nn becomes /-n/ in this position).

196 

 Kurt Braunmüller

The result of this process in the East Scandinavian languages (cf. the Swedish examples in (12) and (13)) is that these inflected forms now show an absolutely balanced sequence of vowels and consonants in their definite inflection: (12) Swed. båt-en (-VC#) ‘boat.sg.utr-the.sg.utr.def’14 and båt-ar-na (-VCRV#)15 ‘boat.pl.utr-the.pl.utr.def’ (13) Swed. fågel-n (...VCR-CR#), ‘bird-the.sg.utr.def’ and fågl-ar-na(-s) ‘bird.pl.utr-the.pl.utr.def(.g)]’ (-VCRV + C#)16

In Dan­ish, Dano-Norwegian (Bokmål) and even in New Norwegian (Nynorsk)17 retroflex consonants by and large do not occur in inflections. Moreover, Danish, which only has a velar r [ʀ] in most of its varieties, has no retroflex consonants at all. (14) Dan. båd-en (-VC#) ‘boat.sg.utr-the.sg.utr.def’ and båd-ene ‘boat.pl.utr-the.pl.utr.def]’ (-VCV#).

In other cases, the r has disappeared, such as in the plural forms of ‘boat’ or ‘bird’: (15) Dano-Nor­w. båt-en ‘boat.sg.utr-the.sg.utr.def’ and båt-ene ‘boat.pl.utr-the. pl.utr.def’ (< båt-er ‘boat.pl.utr’ as well as in New Norw. båt-en sg.utr.def and båt-ane ‘boat.pl.utr-the.pl.utr.def’: ‘the boats’ (< båt-ar ‘boat.pl.utr’: /-VC#; -VCV#/ (cf. also Norw. fugl-[e/a]ne ‘bird.pl.utr.‑the.pl.utr.def’ < fugl-[e/a]r ‘bird. pl.utr’).

The main point to be observed here is that new morphological developments, such as the emergence of a new inflectional system in the unstressed parts of a word, may initiate a new typological cycle which follows the principles of a syllable language: The drift from a syllable to a word language draws to a (preliminary) halt and a new morpho-pho­nemic development starts up – with far-reach­ ing typological consequences.

14 “utr” stands for uter, a common gender for both masculine and feminine which is typical of the Mainland Scandinavian language (except New Norwegian). 15 “CR” stands for “resonant” (C = consonant). 16 The subscript R stands for “resonant /r, l – m, n/”. For more details on the phonotactics of Swedish and the potential influence of Finnish see Braun­mül­ler (1995a: 36‒40; 41‒45). 17 Nynorsk, before 1929 called Landsmaal, represents the other written standard language of Norway. It was created by Ivar Aasen and is based predominantly on West­ern Norwegian dia­­ lects. In these regions [ʀ] prevails.



Scandinavian word phonology: Evidence for a typological cycle  

 197

6.3 Language contact: A typological counter-drift towards a more syllablerelated language Language contact caused other intervening factors, viz. vowel balance/har­mony, as seen in Finno-Ugrian languages (Sámi, Finnish). Both in northern Norwegian and Swedish dialects we observe vowel balance (Swed. vokal­balans, Nor­w. jamvekt), which means that the inflectional endings are taken from two distinct subsets of vowels (high /i – u/ vs. middle /e – o/ and low /a/), depending on the vo­wel of the accented syllable, normally the lexical morpheme. In these dialects, we may also find vowel harmony (Norw. jamvekt med tiljamning), which means that the vowel quality of the stressed syllables is also projected onto the un­stressed ones.18 This development, however, is restricted to so-called light syllables that con­tain­a short vowel and a short consonant (not long consonants or consonant clus­ters), such as ON baka CVCV ‘to bake’, which developed e.g. into Norw. båkå and does not get a final schwa or apocope, as opposed to those lexemes that consist of so-called heavy syllables, such as ON kasta CVCCV ‘to throw, to cast’, which became either kaste or kast-∅ in Norwegian dialects. Vowel balance and vowel harmony in non-heavy syllables prevent the reduction of the coda and pre­serve the syllable-oriented language features of the modern Scandinavian languages. As Kusmenko (2008: chapters 11 and 12) convincingly argues, these phonological developments in the vernaculars may be due to language contact with Sámi dialects. The main precondition for his argumentation, however, is that the Germanic-speaking Scandinavians and the Sámi met on equal terms, which seems probable. Language contact strengthened the syllable type, which might be regarded as a coun­ter-drift back towards a more syllable-oriented language type that keeps unstressed and thus non-reduced vowels in unstressed syllables. Kusmenko’s hypothesis explains not only why there are still reduced vowels (viz. schwa: /ə/, at least in colloquial Swedish) in unstressed syllables/inflectional endings but also why there are still many instances of vowel harmony in Norwegian and Swedish rural dia­lects. The non-reduced vowels /a, u, ʉ/, written a, o and u, in the inflectional endings of Standard Swedish which are audible in distinct pronunciations, however, have their origins in linguistic conservatism and language cultivation in the early nineteenth century (see section 6.4). The same in principle also holds for New Norwegian but not for Dano-Norwegian, which must be considered a Danish creoloid according to Trudgill’s (1992:

18 The same process has also taken place in dialects of Old High German (cf. Szczepaniak 2007b).

198 

 Kurt Braunmüller

21–22) definition. This vernacular only succeeded in introducing four new grammatical morphemes, descending from Norwegian dialects, all of them ending on -a, which mark (1) the enclitic definite feminine singular article (cf. jent-a ‘girl.f.sg-the.def.f.sg’, bok-a ‘book.f.sg-the.def.f.sg’, regering-a ‘government.f.sg-the.def.f.sg’) (2) the enclitic definite neuter plural article (cf. år-a ‘year.pl19-the.def.pl’, språk-a ‘language.pl-the.def.pl’) (3) the preterit and, with identical forms, (4) the perfect par­ticiple of regular verbs (cf. vask-a ‘wash-ed.ipfv/pst.ptcp’, which is no longer morpho-phonolo­gical­ly related to Dan. vask-ede ‘wash-ed.ipfv’ and vask-et ‘wash-ed.pst.ptcp’). This distinction does not exist in the (conservative) creoloid Dano-Norwegian: vask-et ‘wash-ed.ipfv/pst.ptcp’.

Moreover, accent 2, a distinctive prosodic feature which became strengthened obviously due to language contact with a (phonologically balanced) syllable language like Sámi, represents a tonal movement that only occurs in plurisyllabic words.20 Accent 2 seems to play an important role for the retention of non-reduced vowels in unstres­sed syl­lab­les. In so-called accent 2 words, the pitch rises (again) in the second syllable with its peak somewhere on the nucleus, while the dynamics of the primary accent of the first syllable fades out. The exact pitch patterns may vary con­siderably however from one dialect to another (for details see Gårding 1978 and Riad 2000) but this has no in­fluence on the quality of these nuclei. It is only important that the pitch in the un­stressed syllable rises (again) and prevents its nucleus from being reduced to schwa or even zero. Only in some south-eastern Norwegian varieties, especially in the area of Oslo, do we find a second stress intensity peak in accent 2 words on the se­cond, nor­mally un­stres­sed syllable.21 The consequence of this prosodic development is

19 There is no formal difference between the singular and the plural forms in neuter: år ‘year’ – år ‘years’. 20 There are exceptions, however: Swedish bisyllabic words such as vinter ‘winter’ or kommer ‘come.prs’ do not have an accent 2 pattern because their second syllables did not contain nuclei, an epenthetic schwa, when the period of the accent 2 distribution was taking place. Their Old Scandinavian coun­ter­parts were monosyllablic (cf. vetr ‘winter’ and kemr ‘come.3.sg.prs’, respectively). Many of these bisyl­labic words are due to the clitization of definite article and may now form new phonetically minimal pairs, e.g. and|en ‘duck-the’ vs. ande|n ‘spirit-the’. According to Elert (1981: 61‒68) there are (at least) 356 bisyl­labic pairs whose only difference is the presence or lack of this prosodic feature. 21 This phenomenon also occurs in modern colloquial Faroese – for reasons yet unknown/unexplored.



Scandinavian word phonology: Evidence for a typological cycle  

 199

that most Norwegian and Swe­d­ish dialects (if they have not been influenced by dialects with vowel reduc­tion, e.g. those in Bergen [Norway] or southern Sweden) have kept their full vo­wels in many syllable types: in Norway in jamvekts­ord and in Sweden in all dia­lects where apocope did not occur (i.e. in non-heavy syllables). The typological parallels to the phonotactics of Fin­nish (and Sámi) are striking, as already pointed out in Braunmüller ([1980] 1995a: 41-46): These languages also re­tained full un­stressed vowels but they never had a distinct and heavy dynamic stress on the first syllable although the default accent had been on the first syllable, too. Whether language contact in northern, especially central northern Scandinavia was definitively re­spon­sible for these parallel developments is still widely unexplored and has been ignored by most scholars in Scandinavia with a North Germanic mother tongue. The presupposed, time-honoured su­periority of the Germanic-speaking people has prevented any unprejudiced investigations in this direction.22

6.4 Language cultivation: Introducing spelling pronunciation in modern Swedish Language cultivation is another force which may have helped prevent the PanGer­manic drift towards a word language from being carried to its end. This development – in com­­bination with a strong tendency to favor spelling pronunciations when reading texts aloud – seems to play an import role in the history of modern Swedish. As early as 1612, the Royal Chancellery in Stockholm was given the order to write a final -a in infi­ni­tives and no longer -e. This was an attempt to distinguish Swedish from Danish, which was written with an -e (cf. Svensson 1981 and Te­leman 2002: 65-81). By the end of the eighteenth century, language cultivation had led to a new codification of the written norm. As Teleman (2002: 189-197) reports, authors of bilingual dictionaries and grammatical essays such as Sahl­ stedt (1747–53, 1769), but also Botin (1777) and Leopold (1801), became more influential. One of their principles was euphony (in Swedish “väl­ljuds­regel”), which means that they evaluated the variants and tendencies in current language change and decided which forms sounded best. Since zero plurals decreased in

22 Two (characteristic) loans from Finnish, however, are acknowledged in Swedish language history: piga ‘maid, servant girl’ < Finn. piika (cf. Dan. pige) ‘girl’ and pojke ‘boy’ < Finn. poika ‘son, boy’.

200 

 Kurt Braunmüller

the first half of the eighteenth century, they favored the -(n)a plural forms in the case of bisyl­labic neuters, such as in äpplena ‘apple.pl.n-the.def.pl.n’ (vs. äpple-n ‘apple.pl.n’) or hjärtana ‘hearts-the.def.pl.n’ (vs. hjärta-n ‘heart.pl.n’). In the case of monosyl­la­bic neuters, -en ‘pl.n’ became the de­fault form (cf. husen ‘house.pl.n-the.def.pl.n’). The definite marker -en, now a common gender (often called “uter”) singular form, also occurs in loan words from English, such as in jeans-en ‘jeans.sg.utr.-the.def.sg.utr’. To­gether with non-reduced uter plurals ending on -or [-ʊr], -ar and -er [-ɛr], de­finite plu­ral suffixes with retroflex consonants occurred: -orna/arna/erna ‘pl. utr.|+def.pl.utr’ which show a balanced syllable structure: -CRV#. According to the eu­phonic principle, the original­ly feminine adjective declension ending on -a was preferred: Today the -a inflectional marker is the default form. In older, more conservative and West Swedish texts, however, one can still find an -e, but only toge­ther with nouns explicitly referring to males (cf. Käre/Bäste Ulf! ‘Dear. adj.m.sg. Ulf!’). As far as the verbal paradigm is concerned, the dominant plural suffix was -a by the end of the eighteenth century; -e, -om and -o were clearly in the minority. Thus, -a [-a] became the uniform plural inflection marker which existed in some conservative publi­cations until the middle of the twentieth century (cf. hava ‘have.prs.pl’ or äro ‘be.prs.pl’ vs. har ‘have.prs.sg’ and är ‘be.prs.sg’). At the same time, an overt mor­phological marking between the past participle in­flection on -en (cf. skriven ‘[is/are] written.ptcp.prf’) and the supine inflection on -it (cf. skrivit ‘[has (been)] written.sup.(prf)’) of irregular verbs became ob­li­gatory, with the consequence that the vowel quality (/e/ vs. /i/) became an indicator of a grammatical distinction, something which is unique in the verbal systems of the main­land Scandinavian languages. The corresponding forms of the regular verbs, when phonotactically possible, always have an -a in their suffixes (cf. spelat ‘played.sup./ptcp.prf’) as well as the suffixes of their infinitival forms (here: spela ‘to play.inf’).

7 Conclusion Scandinavian word phonology has confirmed the general tendency in Germanic that there is a development from syllable to word languages. The cause for this uniform development is the dynamic accent on the first (stressed) syllable, most extremely manifested in modern Danish,23 and the subsequent reduction of the

23 Cf. Bandle (2011: 67): “dänische Akzentballung” and map 15, which shows the direct con-



Scandinavian word phonology: Evidence for a typological cycle  

 201

following syllables, both in their vocalism and consonantism. But there are other factors which may stop or even reverse this drift towards a word language. As has been shown, (a) language contact, together with (b) the retention or introduction of tonal accents and (c) language cultivation may impede this inherent tendency. Clitization can also be considered a typical outcome of the dynamic accent of the first (stressed) syllable in Germanic which leads to univerbation and may inhibit the drift towards a word language. The best counter-example to the present overall tendency is Swedish where the drift towards a word language was stopped during the eighteenth century, at least in distinct pronunciations and written forms. But Swedish and Norwegian dia­lects also resisted this drift, probably due to early language contact with genetically and typologically diverging varieties, i.e. the Sámi and Fennic languages. Danish, on the other hand, can be considered the most typical Germanic language because it executes this Germanic drift to its “bitter” end, due to radical apocope, syncope, consonant vocalization and the reduction of non-derived words to monosyllabic items with an extreme primary stress accompanied by an immediate decline of all phonological items which occur after the nucleus, both in mono- and polysyllabic words. Some south-eastern Jutish varieties (e.g. certain Low German and Norwegian dialects) even show tenden­cies towards becoming genuine (simple) tone languages, the definite end of any inflexional or agglutinative type of language which Germanic represented at its beginning.

References Bandle, Oskar (2011): Die Gliederung des Nordgermanischen. (Beiträge zur Nordischen Philologie 47.) Tübingen/Basel: Francke. First published 1973 (Beiträge zur Nordischen Philologie 1.) Basel: Helbing. Botin, Anders af (1777): Svenska språket i tal och skrift. Stockholm. Braunmüller, Kurt (1989): Voraussetzungen für die Übernahme mittelniederdeutscher Sprach­strukturen in die skandinavischen Sprachen. In: Karl Hyldgaard Jensen, Vibeke Winge and Birgit Christensen (eds.), Niederdeutsch in Skandinavien II. Akten des 2. Nordischen Symposions ‘Niederdeutsch in Skandinavien’ in Kopenhagen, 18.–20. Mai 1987, 9‒29. (Beihefte zur Zeitschrift für Deutsche Philologie 5.) Berlin: Erich Schmidt.

sequences of the heavy dynamic accent on the first syllable from 1100 A.D. onwards. One of the consequences of this development was the origin of the widely-known Danish stød, a (near) glottal stop occurring after voiced nuclei or voiced consonants in monosyllabic words (cf. Braunmüller 2007: 111‒115).

202 

 Kurt Braunmüller

Braunmüller, Kurt (1995a): Zur Phonotaktik des Schwedischen: Prinzipien und Tendenzen. In: Kurt Braunmüller (ed.), Beiträge zur skandinavistischen Linguistik, 9–52. (Studia Nordica 1.) Oslo: Novus. First published 1980 Skandinavistik 10: 29–60. Braunmüller, Kurt (1995b): Dialekt, Sprachverwandtschaft und ‚Drift‘: Zur Stellung des Søn­derjysk in der germanischen Sprachfamilie. In: Kurt Braunmüller (ed.), Beiträge zur skandinavistischen Linguistik, 288‒302. (Studia Nordica 1.) Oslo: Novus. First published 1987, conference paper. Braunmüller, Kurt (2001): Morfologisk typologi og færøsk [Morphological typology and Faroese]. In: Kurt Braunmüller and Jógvan í Lón Jacobsen (eds.), Moderne lingvistiske teorier og færøsk [Modern Linguistic Theories and Faroese], 67‒88. Oslo: Novus. Braunmüller, Kurt (2003): Die große skandinavische Vokalverschiebung. Resultat einer gesamtnordischen Entwicklung? NOWELE 43: 21–51. Braunmüller, Kurt (2007): Die skandinavischen Sprachen im Überblick. 3rd ed. Tübingen/Basel: Francke. Elert, Claes-Christian (1981): Ljud och ord i svenskan 2 [Sounds and Words in Swedish 2]. Umeå: Universitetet i Umeå. Eliasson, Stig (1986): Sandhi in Peninsular Scandinavian. In: Henning Andersen (ed.), Sandhi Phenomena in the Languages of Europe, 271–300. (Trends in Linguistics. Studies and Monographs 33.) Berlin/New York/Amsterdam: Mouton de Gruyter. Eliasson, Stig (1992): Cognition in phonological change. In: Günter Kellermann and Michael D. Morrissey (eds.), Diachrony within Synchrony: Language History and Cognition. Papers from the International Symposium at the University of Duisburg, 26‒28 March 1990, 283‒308. Frankfurt am Main: Lang. Eliasson. Stig (2010): German and Swedish quantity: A comparison. In: Antje Dammel, Sebastian Kürschner and Damaris Nübling (eds.), Kontrastive Germanistische Linguistik, vol. 1, 5–55. (Germanistische Linguistik 206/209.) Hildesheim: Olms. Gabelentz, Georg von der (1972): Die Sprachwissenschaft, ihre Aufgaben, Methoden und bisherigen Ergebnisse. 2nd ed. (Tübinger Beiträge zur Linguistik 1.) Tübingen: TBL. First published 1891, Leipzig: Weigel. Gårding, Eva (1978): The Scandinavian Word Accents. (Travaux de l’Institut de Phonétique de Lund 11.) Lund: C.W.K. Gleerup. Hagström, Björn (1970): Supradentaler i färöiskan. Ett bidrag till beskrivningen av färöiskt uttal [Supradentals in Faroese. A contribution to the description of Faroese pronunciation]. Fróðskaparrit 18: 347‒360. Kusmenko, Jurij (2008): Der samische Einfluss auf die skandinavischen Sprachen. Ein Beitrag zur skandinavischen Sprachgeschichte. (Berliner Beiträge zur Skandinavistik 10.) Berlin: Nordeuropa Institut der Humboldt Universität. Leopold, Carl Gustaf af (1801): Afhandling om Svenska Stafsättet. Svenska Akademiens handlingar ifrån år 1796. Första delen. Stockholm. Nielsen, Hans Frede (2000): The Early Runic Language of Scandinavia. Studies in Germanic Dialect Geography. Heidelberg: Winter. Riad, Tomas (2000): The rise and fall of Scandinavian accent. In: Gudrún Thórhallsdóttir (ed.), Proceedings of the Tenth International Conference of Nordic and General Linguistics, University of Iceland, June 6–8, 1998, 15‒33. (The Nordic Languages and Modern Linguistics 10.) Reykjavík: University of Iceland. Sahlstedt, Abraham (1747): Försök Till en Swensk Grammatika, Efter det nu för tiden brukeliga sättet at tala ok skrifwa. Stockholm. Revised edition 1969.



Scandinavian word phonology: Evidence for a typological cycle  

 203

Sahlstedt, Abraham (1753): Anmärkningar Öfwer Swenska språket. Jämte Et Bihang af Swenska orthographien. Stockholm. Selås, Magnhild (1997): Endingsvokalar i talemålet i Tórshavn [Vowels in inflectional end­ings in the spoken variety of Tórshavn]. Nordica Bergensia 13: 69‒97. Svensson, Lars (1981): Ett fall av språkvård under 1600-talet [A Case of Language Cultiva­tion in the Course of the 17th Century]. Lund: Ekstrand. Szczepaniak, Renata (2007a): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Szczepaniak, Renata (2007b): Vokalharmonie im Althochdeutschen und im Walserdeutschen. Ein Fall von phonologisch-typologischer Kontinuität. Zeitschrift für Dialektologie und Linguistik 74/1: 38–60. Teleman, Ulf (2002): Ära, rikedom och reda. Svensk språkvård och språkpolitik under äldre nyare tid [Honour, Wealth and Order. Swedish Language Cultivation and Language Policy in Older Modern Times]. Stockholm: Norstedts Ordbok. Trudgill, Peter (1992): Introducing Language and Society. London: Penguin.

Martin Joachim Kümmel (University of Jena)

Syllable- and word-related developments in earlier Indo-Iranian Abstract: Old Indo-Aryan (Vedic Sanskrit) and the Old Iranian languages (Avestan, Old Persian) are closely related and still quite similar, especially in morphology and syntax. Phonologically, their differences are much more pronounced. IndoAryan seems to show a drift towards a more syllable-oriented language type. For example, Old Indo-Aryan has heavy restrictions on final consonants and is known for its well-developed sandhi across word boundaries, and in mainstream Middle Indo-Aryan, nearly all consonant clusters are assimilated. Most modern Indo-Aryan languages are considered prototypical syllable languages. In contrast, Old and Middle Iranian allow many more consonant clusters and generally show more word-related features. In this paper, the most ancient members of the two languages families are compared regarding syllable- and word-related features and processes.

1 Introduction The aim of this paper is to compare syllable- and word-related features and processes in two closely cognate language families, Indo-Aryan and Iranian. Many of the divergences between these two families seem to be related to these typological features. Already the earliest attested stages show these differences, and the focus of this paper will mainly lie on the oldest historical languages belonging to these families, i.e. Old Indo-Aryan (OIA) and the Old Iranian (OIr) languages.

1.1 The typological parameters To assess the differences within the framework of syllable versus word language typology, the parameters shown in Table 1 are used (for the general typology cf. Auer 1993; Szczepaniak 2007; see Table 1 in the introduction to this volume). The choice of these parameters is partly due to the availability of data in historical corpora. Features that cannot be deduced from the transmitted texts were not used. For example, arguments presupposing knowledge of synchronic stress are difficult to apply to Old Iranian since accentuation is not transmitted. We can distinguish between features related to single segments and their distribution (1 and 2 for vowels and consonants, respectively) and features related to larger units, i.e.



Syllable- and word-related developments in earlier Indo-Iranian  

 205

syllable structure within the word (3) and syllable structure within the phrase or sentence (4). Table 1: Typological parameters Parameter

W

S

1a

Vowel system

stress-sensitive

uniform

1b

Vocalic processes

deletion when unstressed

epenthesis

2a

Geminates

stress-sensitive

uniform

2b

Consonantal processes

epenthesis at borders

deletion in clusters

3a

Syllable structure: (onset)

complex

simple

3b

Codas

complex

simple

3c

Sonority hierarchy

not necessarily obeyed

obeyed

4a

Resyllabification

no

yes

4b

Sandhi

no/internal

yes/external

1.2 The older Indo-Iranian languages and their attestation Old Indo-Aryan (Sanskrit, cf. Cardona 2003; Jamison 2004a) has been preserved in India as a high-status language until today. Its earliest stage is called Vedic (Ved.), named after the so-called Vedas, the earliest and holiest texts of Hinduism. They were composed from ca. 1600 until 500 BC, i.e. centuries before the oldest Indian inscriptions. Because of their sacred character and the magical importance of correct recitation they were orally transmitted with the help of a very sophisticated transmission system and therefore preserved with astonishing accuracy (cf. Jamison 2004a: 673). The earliest text corpus is the Rigveda, a collection of 1028 hymns probably dating from the last centuries of second millennium BC. It was followed by other collections until the end of the Vedic period around 500 BC. The late Vedic language was described by native grammarians and phoneticians, and the version described in the work of the famous grammarian Pāṇini became the canonical literary language, Classical Sanskrit. Old Iranian (cf. Schmitt 1989a; Skjærvø 2009) is attested by two different languages: Avestan (Av.) and Old Persian (OP). Avestan (see Hale 2004) was the holy language of the Zoroastrians, and texts in that language were also orally transmitted, but with less accuracy than the Vedas. In Sasanian times they were written down in a special alphabet designed for that purpose (see Hoffmann and Narten 1989). The resulting text is phonetically very exact but seems to contain some

206 

 Martin Joachim Kümmel

secondary changes in liturgical pronunciation that were probably not part of the original texts at the time of their composition (for more detailed information see de Vaan 2003: 5–15).1 The surviving text corpus is composed in two clearly distinct language varieties: Old Avestan (OAv.) and Younger Avestan (YAv.),2 probably dated around the Vedic period. Old Persian (see Schmitt 1989b, 2004) was the official language of the Achaemenid kings, attested in inscriptions (ca. 500–300 BC) written in a special simple cuneiform script invented to write this language. The corpus is rather small and the script leaves room for interpretation but thanks to the cognate languages just mentioned, and later Persian, it is nevertheless not too difficult to read them. Other Old Iranian languages (cf. Schmitt 1989c) are attested merely by isolated words and names and can only be reconstructed from their later successors. Old Indo-Aryan and Old Iranian are very close in terms of grammar and lexicon, so close that one can directly convert texts from one language to another. Their most marked differences are found in phonology. Due to their similarities, their common ancestor, Proto-Indo-Iranian (PIIr), can be reconstructed to a large extent. This language represents one subfamily of Indo-European, and features of Proto-Indo-European (PIE), the reconstructed parent language of the whole family, will be mentioned in the following. Spoken Indo-Aryan apparently underwent important changes already in the Vedic period. In the third century BC, the inscriptions of King Ashoka found in all parts of his empire show a language quite different from Vedic, and the same is true for Pāli, the literary language of early Buddhism. Both of these represent the earliest stage of Middle Indo-Aryan (cf. Oberlies 2001, 2003; Jamison 2004b). Further changes can be observed in the later so-called Prakrits (considered natural languages in opposition to Sanskrit, the holy language preserving the shape of late Vedic; cf. Bubenik 2003). Nearly all Indo-Aryan languages were written by means of scripts originally designed for Middle Indo-Aryan (the languages of the earliest inscriptions) and only later adapted to Old Indo-Aryan; these scripts as a rule provide rather accurate phonological information. Since the last centuries BC, we also have attestations of different Middle Iranian languages (for an overview see Schmitt 1989c) that are conventionally

1 Most importantly, many epenthetic vowels are transmitted that cannot have been part of the original text, as the meter shows (cf. Hale 2004: 752–753; Skjærvø 2009: 64–65). Such secondary vowels are transcribed here by superscript letters; e.g. Old Avestan pərəsā ‘I am asking’ = original disyllabic pǝrsā, later pronounced pǝrǝsā. 2 Some short passages are probably composed in a more archaic form of Younger Avestan, cf. Tremblay (2006).



Syllable- and word-related developments in earlier Indo-Iranian  

 207

grouped into two main branches: Western Iranian is represented by Middle Persian (MP) in the South and Parthian in the North. The main Eastern Iranian languages are Khwarezmian, Bactrian, Sogdian, Khotanese, and the scarcely attested Sarmatic-Alanic predecessor of Ossetic (cf. Bielmeier 1989). Middle Iranian languages are written by means of quite different scripts, sometimes providing only very limited information on phonological details, e.g. the majority of them do not regularly indicate short vowels, and many letters can be ambiguous. But the comparison between different writings and other indirect information often helps to interpret insufficient direct evidence.

1.3 Overview of the phonological systems For convenience, I first give an overview of the phonological systems that can be reconstructed for these languages (see table 2 and table 3 for Old Indo-Aryan and table 4 and table 5 for Avestan; cf. Hale 2004: 745–753; Jamison 2004a: 675–681; Schmitt 2004: 723–727). In the tables the usual philological transliteration is given in italics; the phonological correspondence in IPA letters follows. Sounds in brackets are only allophones; sounds in curly brackets are innovations of later stages. Table 2: Segmental system of Old Indo-Aryan (vowels) Monophthongs

Diphthongs

short high

long

front

back

front

back

i i-ɪ

u u‑ʊ

ī i:

ū u:

e e:

o o:

low

a ɐ

low-high

front

back

ai ɐi ̯

au ɐu̯

ā a:

Table 3: Segmental system of Old Indo-Aryan (consonants) labial

dental

retroflex

palatalveolar/palatal

velar

plain stops

pp

tt

ṭṭ

c tʆ

kk

voiced stops

bb

dd

ḍḍ

j dʓ

gɡ

aspirated stops

ph pʰ

th tʰ

ṭh ṭʰ

ch tʆʰ

kh kʰ

breathy stops

bh b̤ʱ

dh d̤ʱ

ḍh ḍ̈ʱ

jh {d̤ʓ̈ʱ}

gh ɡ̈ʱ

ss

ṣṣ

śʆ

fricatives glides/liquids

v ʋ-w

l l 

rɾ

yj

nasals

mm

nn

ṇṇ

(ñ ɲ)

glottal

(ḥ h) hɦ ṅŋ

208 

 Martin Joachim Kümmel

All consonants except r and h could be phonetically and phonologically geminated. r and rarely l had syllabic allophones normally transcribed as r̥ and l̥ respectively; in this case I will deviate from the philological tradition and follow the IPA and write r̩, l̩. Table 4: Segmental system of Avestan (idealized) (vowels) Monophthongs

Diphthongs

short

long

front high

i i-ɪ {e ɛ}

low

ǝə

back

front

u u‑ʊ

ī i:

{o ɔ}

{ē e:}

a ɐ

back

front

back

ū u:

short

aē/ōi ɐi̯

ao/ə̄u ɐu̯

{ə̄ ə:}

{ō o:}

long

āi a:i̯

āu a:u̯

ā a:‑ɑ:

{ā̊ ɔ:}

ą ɐ̃:

nasalized

Table 5: Segmental system of Avestan (idealized) (consonants) labial

dental

alveolar

postalveolar

palatal velar

voiceless stops

pp

tt

c tɕ

kk

voiced stops

b b

d

j dʑ

g ɡ

{š́ ɕ}

xx

voiceless fricatives f f

θ

ss

š ʃ-ṣ

voiced fricatives

(β β)

(δ ð)

zz

ž ʒ-ẓ

glides/liquids

v,uu ʋ‑w

nasals

mm

rr nn

glottal

hh

(ɣ ɣ) y, ii j (ń ɲ)

(ŋ ŋ)

Consonants could not be geminated.

1.4 Accent and stress The accentuation in Vedic was well transmitted by the oral tradition of the texts and described by the native phoneticians and grammarians. Accent was free, and every word form normally had one accented syllable, but enclitic particles, non-initial vocatives and verbs in main clauses were not accented. The accent seems to have been mainly characterized by (rising) high pitch on the accented syllable and falling pitch on the following syllable (Jamison 2004a: 680). At the time of late Rigveda, short prevocalic high vowels generally became non-syllabic, and if they had been accented, the falling pitch of the following syllables was phonologized. Thus later Vedic had acquired a pitch accent system with two contrasting intonations (cf. Cardona 2003: 108, 121), viz. in havyá- ‘sacrificial drink’



Syllable- and word-related developments in earlier Indo-Iranian  

 209

vs. vasavyà- ‘wealth’. In later Indo-Aryan, this pitch accent system was lost, and a weak non-phonological stress developed according to syllable weight (cf. Masica 1991: 183). In general, the development of mainstream Indo-Aryan reduced the importance of word stress. Stress in Old and Middle Iranian was not directly transmitted, but indirect traces point to a similar development: The older system with free stress was first preserved (presupposed by some specifically Younger Avestan developments, cf. Hoffmann 1986: 165–167 = 1992: 839–841; de Vaan 2003: 577–602; Schmitt 2004: 725) but then was lost in favor of different systems mostly according to syllable weight. However, these can only be inferred from Middle Iranian data so it is difficult to reconstruct them for Old Iranian.

2 Phonological inventories and their relation to words and syllables In both branches of Indo-Iranian we normally do not find significant differences between the phonemic inventories of different syllable types. Neither accentuation nor the position within the word seem to impose restrictions. This also largely applies to segmental quantity which was synchronically independent of the accent or the position within the word. This can be seen in short vowels in accented open syllables: Ved. mánas ‘mind’ = Av. manō, Ved. hímā ‘winter’; long vowels in unaccented syllables: Ved. varāhá- ‘boar’ = Av. varāza-, Ved. bhárāmi ‘we bear’ = YAv. barāmi; and long consonants in Ved. ánna- ‘food’, panná- ‘fallen’, devátta- ‘given by the gods’, sádassu ‘at the seats’ (cf. below). In Old Iranian, however, there was one exception: The quantity distinction was neutralized for word-final vowels. In Old Avestan as well as Old Persian, they are transmitted as long.3 In Younger Avestan, this only applies to monosyllables and the vowels ō, ɔ̄ and ə̄, while all other original final vowels are always short (cf. Hale 2004: 747; Skjærvø 2009: 55). (1) Cf. OAv. = OP. idā = YAv. iδa = Ved. ihá ‘here’ with original *-ă vs. OAv. = OP. yaθā = YAv. yaθa = Ved. yáthā ‘how’ with original *-ā

3 Cf. de Vaan (2003: 161–162); Hale (2004: 747); Schmitt (2004: 725); Skjærvø (2009: 55, 58). In Old Persian, a synchronically short a can appear at the end of words, if originally followed by a consonant (t, h or n); it is not totally clear whether the consonants were still preserved but simply not written or had been lost with the subsequent phono­logization of the quantity contrast between old *-a > -ā and old *-aC > -a.

210 

 Martin Joachim Kümmel

There is an additional vocalic restriction for word-final position in Younger Avestan: While short diphthongs *ai, *au are preserved word internally and in accented monosyllables, they were monophthongized at the end of polysyllabic words and clitics, so that the inventory of vowels is reduced in non-initial syllables before a word boundary: In this position, only a, e, i, u; ə̄, ō, ɔ̄ can occur, while ā, ē, ī, ū and short diphthongs are impossible. This feature must be interpreted as word related. In the transmitted Avestan texts, shortening of (probably unaccented) long vowels as well as lengthening of (probably accented) short vowels seems to occur (cf. the evidence discussed by de Vaan 2003: 31–160, 205–312). However, these effects are most certainly secondary due to their later transmission, and thus they rather reflect the influence of Middle Iranian speech habits, and it is not advisable to take them as linguistically real Old Iranian processes. In Younger Avestan, at least one case of allophonic alternation between vowels depends on syllable structure: The diphthong *ai split into something like *əi in closed syllables (before a coda) vs. *ai in open syllables, eventually yielding ōi vs. aē in the extant texts (see de Vaan 2003: 340–348). While all Old (and Middle) Indo-Iranian languages distinguish between long and short vowels, consonantal quantity was not relevant in Old Iranian (and most of Middle Iranian): Geminates are absent from the system. This state of affairs is probably inherited from Proto-Indo-Iranian and Proto-Indo-European for which no geminates can be reconstructed (Hoenigswald and Woodard 2004: 538; Kobayashi 2004: 32). In Old Indo-Aryan, however, contrastive geminates are clearly present from the beginning, but apart from /t:/, /dʱ:/ and /n:/,4 other geminates are quite rare and mainly occur in onomatopoetic words and foreign elements, such as e.g. kukkuṭá- ‘rooster’; notable exceptions are vr̩kká- ‘kidney’ and uccá- ‘high’. The Indo-Aryan geminates are clearly secondary; they arose from original clusters by assimilation or simplification: (2) ánna- ‘food’ < *ád+na cítti- ‘thinking’ < PIIr *citsti-5 < *cit+ti- (cf. OIr cisti-) addhā́ ‘manifest’ < PIIr *adz-dʱā́ < *adʱ+tā́, cf. OIr azdā (synchronically opaque)

4 These geminates mainly occur at morphological boundaries between the root and derivational as well as inflectional affixes. 5 In heterosyllabic clusters of two dental stops, an epenthetic sibilant was inserted already in PIE, and this rule persisted in PIIr, cf. Kobayashi 2004: 37–38.



Syllable- and word-related developments in earlier Indo-Iranian  

 211

In some relic forms, the older rule of geminate simplification at morpheme boundaries is preserved. The productive rule in Old Indo-Aryan, however, was that such geminates are retained (Kobayashi 2004: 45–46): (3) -as+su (locative plural) in sádas-su ‘in the seats’ (in contrast to Av. *-asu > -ahu) vs. the inherited archaism ás+si > ási ‘you (sg.) are’ = Av. ahi

Non-phonological gemination was also frequent since first consonants in most clusters could be lengthened, so that etymological kṣat-trá- ‘knife’ and kṣa-trá‘rule, reign’ were both syllabified as /kṣat.trá-/ (see Wackernagel 1896: 112–114; Kobayashi 2004: 31–33, also on possible exceptions in Early Vedic and north-western dialects). Phonological geminates and non-phonological cluster gemination in clusters are probably interrelated, and accordingly, it is not surprising that no such phonetic gemination rules have been observed in Iranian. It is not clear, however, what these important differences between Indo-Aryan and Iranian mean in terms of the syllable vs. word language typology. In general, the inventories of all Old Indo-Iranian languages are uniform across all syllables. In Old Iranian, however, word-final syllables show some restrictions which can be considered a word-related feature.

3 Syllable structure 3.1 Clusters and shell complexity Indo-European languages typically allow various consonant clusters (up to four consonants), and Indo-Iranian is no exception to this (for an overview of possible clusters in Old Indo-Aryan see Masica 1991: 161–162). We may differentiate between the development in the onset and in the coda. According to Auer (1993: 43), coda complexity seems to be more clearly related to word-based languages than onset complexity; this, however, probably just means that more languages highlight the right edge rather than the left. For languages highlighting the left edge, onset complexity is most relevant.

3.1.1 Onset clusters In the onset of words, the older stages are rather conservative, preserving the general IE type ST(R)- and even some TS clusters, viz. Ved. strī́ ‘woman’, kṣatrám ‘rule, reign’ = Av. strī, xšaθrəm. Only Iranian also shows FST clusters (see below and cf. Kobayashi 2004: 33). This may be due either to the simplification of an older cluster in Indo-Aryan or to consonantal epenthesis in Iranian (see below).

212 

 Martin Joachim Kümmel

Table 6: Complex word-initial onset clusters in Ancient Indo-Iranian Cluster

Old Indo-Aryan

Old Iranian

Proto-IndoIranian

Vedic

Avestan

*#d͡ʓn

#d͡ʓɲ

*#ʃn > #xʃn

*dʓnā‘to know’

jñā-

xšnā-

*#pst

#st

#fʃt

*pstā̆n‘breast’

stána-

fštāna-

*#kt

#t

#t

*kturíja‘fourth’

turī́ya-

tūiriia-

Word internally, ST clusters do not occur in Old Indo-Aryan, but TR- and also TS- are possible, cf. mán.tra- ‘solemn utterance’, chan.tsat ‘will appear’. In Old Iranian, ST clusters are found word internally; they could even be sometimes syllabified as onsets after vowels in internal position in contrast to Indo-Aryan, cf. the following case: (4) Ved. pa.vás.ta- vs. OP pa.va.stā- ‘skin’ (presupposed by MP pōst < *ˈpa.wa.stā instead of †pawast < †*pa.ˈwas.tā, Klingenschmitt 2000: 215)

Such syllabification is also presupposed by vocalic alternations sensitive to syllable structure in Younger Avestan (as mentioned above): (5) YAv. aē.šma- ‘wrath’, maē.sman- ‘urine’ vs. šōiθ.ra- ‘dwelling’

Thus it seems that Old Iranian allows and partially prefers more complex onsets in word-initial as well as internal position compared to Indo-Aryan.

3.1.2 Coda clusters If we compare identical words or identical morphological structures with syllable- and word-final clusters, the difference between Old Indo-Aryan and Old Iranian is very clear:



Syllable- and word-related developments in earlier Indo-Iranian  

 213

Table 7: Proto-Indo-Iranian coda clusters in Old Indo-Iranian (hyphens indicate morpheme boundaries; in the examples, a dot means a syllable boundary) Proto-Indo-Iranian

Cluster Old IndoAryan

Old Iranian

Old IndoAryan (Vedic)

Avestan

*sarwá-tāt-s ‘wholeness’

*ts#

t#

s#

sarvátāt

hauruuatās

*uts-tāná-dʓʱasta‘with hands’

*ts‑t

t$t

$st

ut.tānáhasta-

u.stānazasta-

*jáu̯g(-d) ‘yoked’ / *máu̯g(-d) ‘saved’

*g#

k#

g(d?)#

á-mok

yaogət̰

*cár-d ‘made’ *wárt(-t) ‘turned’

*r-t# *rt-t#

r# rt#

rd# ?

kár a-vart

cōrət̰ ?

*wāk-ʃ ‘voice’

*kʃ#

k#

xʃ#

vā́k

vāxš

*bʱak-ʃ-ta ‘received’

*kʃ-t

kt

x$ʃt

á-bhak.ta

bax.štā

*wakʃ-t ‘grew’ / *dʱā́gʱ-ʃ-t ‘burnt’

*kʃt#

k#

xʃt#

dhā́k

vaxšt

*-r̩ʃ Gen. Sg. r-stems

*r̩ʃ#

ur#

ərʃ#

-ur

-ərəš/-arš

*darʃ-tā́ ‘who sees’

*Vrʃ-t

rVṣ$ṭ

Vr$ʃt

draṣ.ṭā́

-dar.šta

*twárʃ-tā ‘shaper, creator’

*wVrʃ‑t

wVṣ$ṭ

wVr$ʃt

tváṣ.ṭā

θβōrə.štā

*dʱār-ʃ-t ‘held’ / *bʱā́r-ʃ-t ‘carried’

*rʃt#

r#

rʃt#

bhā́r

dārəšt

Obviously, Old Indo-Aryan imposes heavy restrictions resulting in cluster simplification and loss of morphological information; word-final codas with more than one consonant were only allowed for consonants belonging to the root (cf. Masica 1991: 162–163; Jamison 2004a: 680). This may be taken as a word-based feature even if its occurrence was quite rare (only two verbal roots are actually concerned). In internal position, clusters of two consonants were always syllabified as C$C. Clusters of three consonants were normally syllabified as C$CC, but CC$C could appear if a resonant (nasal or r) was followed by two stops, normally velar + dental; similarly, clusters of four were allowed only in this case: (6) Ved. paṅk.tí- ‘group of five’, yuṅk.te ‘yokes (for himself)’; vark.tam ‘you (two) turn’; yuṅg.dhvám ‘you yoke (for yourselves)’, pāṅk.trá- ‘an animal (mouse?)’

However, the ancient phonetic handbooks and the manuscripts show an optional loss of velar stops after the velar nasal in codas: yuṅ.tē. Nasals preceding sibilants probably had no oral closure (so-called Anusvāra, transliterated ṃ) so that

214 

 Martin Joachim Kümmel

they cannot be considered consonants, and thus were part of the nucleus rather than the coda; therefore, words like saṃs.thá- ‘standing together’ did not have a complex coda NS but a complex nucleus VṼ (cf. Cardona 2003: 110). In Middle Indo-Aryan, all word-final consonants were disallowed (Masica 1991: 170–171; Oberlies 2003: 170; Jamison 2004b: 704), and internal codas were restricted to geminates and homorganic nasals. In earliest Old Iranian, i.e. Avestan, word-final codas were obviously much more complex than in Indo-Aryan, freely allowing all kinds of CC and even CST. Old Persian seems to have had only simple consonants word finally (Schmitt 2004: 725), but the limited material does not allow certainty. Internal codas were probably much more restricted. As stated above, more obstruent clusters could be syllabified as onsets, so the preceding syllable had a less complex coda (if any), cf. the examples bax.štā, -dar.šta in the table and OAv. xraf.stra-. Therefore, the partial preference for $CCV structures instead of C$CV does not clearly increase shell complexity in word-internal position. On the other hand, clusters of more than two consonants could be preserved in more cases than in Indo-Aryan.

3.1.3 Conclusion Consonantal clusters and complex onsets were increasingly disfavoured in Indo-Aryan, while in Old Iranian they were mostly tolerated and even preferred in many cases. Both branches have a slight preference for complexity at word boundaries as opposed to internal position: While in Indo-Aryan this mainly concerns onsets (#ST vs. $C, $TR, $TS), it is valid for both codas and onsets in Iranian (#CST vs. $ST, CST# vs. C$).

3.2 Sonority Hierarchy The only exception to the sonority sequencing principle6 in Old Indo-Aryan is represented by initial ST clusters where extrasyllabicity of the sibilant might be assumed (cf. Kobayashi 2004: 41–43). Original Indo-Iranian TST clusters and final TS had lost the sibilant, and different stops were not allowed in tautosyllabic positions (no †pt-, †kt-, †db-, etc.). In Middle Indo-Aryan, ST clusters were assimilated so that no exceptions were left.

6 This principle in IE syllabification means that within any syllable, sonority must increase towards the nucleus. For more on this principle and its possible violations see Byrd 2010: 87, 100, 2012.



Syllable- and word-related developments in earlier Indo-Iranian  

 215

In Old Iranian, tautosyllabic as well as heterosyllabic ST clusters are generally preserved; as a result of the Iranian change of stops into fricatives before consonants, original TST clusters were preserved as FST clusters, viz. Av. fštāna‘breast’, xštuua- ‘sixth’, vaxšt ‘grew’. Voiceless TT clusters were simplified in word-initial position and elsewhere changed to FT clusters, viz. *ptərwija- ‘father’s brother’ > Av. tūiriia-; *kturíja- ‘fourth’ > Av. tūiriia- vs. ā­-xtūiriia-. Voiced TT clusters, however, seem to have been preserved at least in OAv. dǝbąz- ‘to support’ and dəbu- ‘to deceive’; their number was increased in Avestan by the change *dwi- > dbi-, cf. OAv. daibitiia- = YAv. t̰bitiia- ‘second’. Thus Indo-Aryan tends to remove all violations of the sonority sequencing principle, while in Iranian such a trend is not apparent.

3.3 Vowel epenthesis and loss The two branches also show some differences with regard to secondary vowels. Iranian languages normally developed an epenthetic schwa vowel before original syllabic r which then was syllabified as a coda, producing a new consonant cluster. In Indo-Aryan, however, syllabic r was first preserved and then replaced by a simple short vowel (Masica 1991: 167; Jamison 2004b: 703), thus preserving the open syllable. In some internal consonant clusters,7 an epenthetic vowel is present in Indo-Aryan (cf. Jamison 2004a: 680–681) but lacking in Iranian (Hale 2004: 747–748). Normally, it is assumed that an over-short vowel ɪ̆ arose in Proto-Indo-Iranian and was later lost in Iranian, but Werba (2005: 711–716) proposes that epenthesis was an Indo-Aryan innovation in these cases (cf. also Kobayashi 2004: 136–138). In any case, this process again produced simpler syllable structures in Indo-Aryan as opposed to Iranian. Except for the possible loss of the epenthetic vowel, deletion of unaccented vowels (syncope/apocope) does not seem to occur anywhere in Old Indo-Iranian.

7 This applies to clusters with *H, the Proto-Indo-Iranian reflex of Proto-Indo-European socalled laryngeals, probably similar to a back fricative or [h]. These sounds were most often lost in the attested languages and have to be reconstructed for the protolanguage (cf. Hoenigswald and Woodard 2004: 537). For initial *CHC clusters, the development is disputed because Iranian has variation, viz. in a word like *pHtā́ > Av. patā ~ pita ‘father’, dative *pHtráj > *ptrai ~ *pitrai > OAv. fǝδrōi ~ piθrē.

216 

 Martin Joachim Kümmel

Table 8: Epenthetic vowels in Indo-Iranian ProtoIndoEuropean

ProtoIndoIranian

Old IndoAryan

Old Iranian

ProtoIndo-European

Vedic

Avestan

*Cr̩C

*Cr̩C

Cr̩C

CərC

*mr̩.tá- ‘dead’

mr̩.tá-

mər.ta-

*CHC

*CH(ь)C

CiC

CC

*dʱugʱ.H(ɪ̆)tā́ ‘daughter’

du.hi.tā́

dugə.dā/duɣ.ða

*támH(ɪ̆).sra‘dark’

tá.mis.ra-

*tan.sra‑ > tą. θra‑

Thus, some vowel-related processes in Indo-Aryan enhance the optimal (open) syllable, while Old Iranian tends to preserve more complex syllable structures.8

3.4 Consonantal processes Due to the Old Indo-Aryan preference for simpler clusters, consonants were lost in different contexts, as shown by the examples in Table 6 and Table 7: Stops were lost in initial position before obstruents other than simple sibilants (i.e. whenever the sonority hierarchy would not have been obeyed), sibilants were lost between stops, and nearly all consonants were lost word finally after other consonants. Only one of these processes is also no diacritic below g found in Old Iranian, namely the loss of original voiceless stops preceding stops in word-initial position (see Avestan tūiriia- ‘fourth’). Conversely, Old Iranian sometimes exhibits consonantal epenthesis at word boundaries, e.g. initial *šC became xšC: (7) *dʓnā- > *šnā- > Av. OP xšnā- ‘to know’; *šwaš > Av. xšuuaš ‘six’

In word-final position, Avestan writings like -hāgət̰ ‘following’ for original *hāg may point to an epenthetic stop after g, but there may be a purely graphic explanation (Skjærvø 2009: 56).

8 Secondary epenthesis occurs in the transmission of the Avestan texts (cf. above). Since it did not belong to the original text, it is not relevant here.



Syllable- and word-related developments in earlier Indo-Iranian  

 217

4 Sandhi and resyllabification Old Indo-Aryan is famous for its obligatory sandhi rules that apply across word boundaries (cf. Cardona 2003: 110–122; Jamison 2004a: 677–678), and we still refer to the phenomenon by the term coined by the Indian grammarians. Already in ancient times, the so-called padapāṭha (‘word-reading’) was developed for the transmission of the holy texts. In this version, all external sandhi are dissolved, and the words are given in the form appropriate before a pause. The assimilation rules applied between words are very similar to internal phonological rules, but there are slight differences mainly due to the fact that some rules for word-final position are applied before external sandhi: Most importantly, final clusters were simplified, and final obstruents became neutralized to laryngeal features which are taken over from the initial sound of the following word, viz. (8) *ā́raik-t ‘(has) left + 3s’ → ā́raik + u → ā́raig u9 *yás nas ‘who (rel.) … us’ → *yáz nas → yó naḥ (loss of voiced *z)

In word-internal sandhi, laryngeal features are only assimilated if the following sound is an obstruent, but not elsewhere, viz. (9) ririk-vā́n ‘leave-perfect+participle’ → ririkvā́n ‘having left’ (v denotes a resonant [ʋ]) ás-mi ‘to be+1s present’ → ásmi ‘I am’ (not †*azmi > †āmi/†emi)

In Early Vedic, we find more relics of internal-like rules, and there is even initial mutation (e.g. retroflexation) in some shorter phrases (cf. especially Hale 1999: 147–150): (10) diví + sthás ‘in the sky + you (dual) are’ → diví ṣṭhás ‘you are in the sky’ svàr + ná ‘sun + like’ → svàr ṇá ‘like the sun’

In later Vedic and Classical Sanskrit these special rules were lost. As far as we can tell from the metrical evidence, resyllabification has always been obligatory within an utterance. The Old Iranian sources are normally only transmitted like a padapāṭha, i.e. words are given in their prepausal form. External sandhi, however, are sometimes found within shorter phrases including monosyllabic word forms:

9 OIA u is a connective sentence particle obligatorily attached to the first word.

218 

 Martin Joachim Kümmel

(11) Av. yō ‘who (rel.)’ → yasə tē … ‘who (rel.) for you …’ (with preserved s) Av. yat̰ ‘that, which (relative)’ → *yat na > yaθənā/yaθna … ‘which indeed’ (with θ < t before a consonant)10

One might therefore assume that sandhi were mainly internal. However, some cases of secondary reanalysis show that external sandhi were more widespread in spoken Old Iranian than the transmitted texts tell us: (12)  *āt + im … > *ād im … ‘then him …’ = ā$dim … → reanalysis ā(d) dim … → variant di- of the clitic pronoun i- (Younger Avestan and Old Persian)

This implies that the final dental of such words was voiced before a following vowel as in Vedic, and it also presupposes resyllabification which is otherwise difficult to prove since Avestan meter is not based on syllable quantity. Since such cases of reanalysis seem to be confined to shorter phonological phrases (cf. Vedic above), we can conclude that Old Iranian had external sandhi within these shorter phrases. Whether external sandhi were also applicable on the larger sentence level, as it was in Vedic, is not clear. It is thus possible that external sandhi were more restricted in Old Iranian than in Old Indo-Aryan.

5 General conclusions The features discussed above can be summarized in Table 9 and Table 10 for the individual languages, following the scheme given in Table 1 (S = syllable-related, W = word-related). Table 9: Old Indo-Aryan (Vedic Sanskrit) 1a

Vowel system uniform

S

1b

Vowel epenthesis

S

2a

Geminates possible, uniform

S

2b

Consonant deletion in clusters

S

3a

Onsets: $CC / #STR

(W)

3b

Codas: C$, rarely RT$

(S)

3c

SH mostly obeyed

S

4a

Resyllabification: yes

S

4b

Sandhi: yes

S

10 Cf. Hintze 2007: 66–68.



Syllable- and word-related developments in earlier Indo-Iranian  

 219

Table 10: Old Iranian (Avestan and Old Persian) 1a

Vowel system uniform except word finally

S W

1b

Less vowel epenthesis or deletion

(W)

2a

No geminates

?

2b

Consonant epenthesis

W

3a

Onsets: $STR / #FSC

W

3b

Codas: CC$ / FST#

W

3c

SH often violated

W

4a

Resyllabification: yes?

S?

4b

Sandhi: less than OIA

S (W)

The investigation shows that the two branches of Indo-Iranian clearly developed in different directions. Old Indo-Aryan exhibits a rather high rate of S features (which was further increased in Middle Indo-Aryan) at all levels (though less clearly in syllable structure). Old Iranian, on the other hand, seems to have a tendency towards W-related features from the beginning (though only weakly in the case of sandhi and vowels). It seems that these tendencies continue in the later history of both families, but this requires a separate investigation.

References Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Bielmeier, Roland (1989): Sarmatisch, Alanisch, Jassisch. In: Rüdiger Schmitt (ed.), Compendium Linguarum Iranicarum, 236–245. Wiesbaden: Reichert. Bubenik, Vit (2003): Prākrits and Apabhraṁśa. In: George Cardona and Dhanesh Jain (eds.), The Indo-Aryan Languages, 204–249. (Routledge Language Family Series 2.) London/ New York: Routledge. Byrd, Andrew Miles (2010): Reconstructing Indo-European Syllabification. Ph.D. dissertation, Univer sity of California, Los Angeles. Available at http://uknowledge.uky.edu/lin_facpub/55/. Byrd, Andrew Miles (2012): Predicting Indo-European syllabification through phonotactic analysis. In: Benedicte Nielsen Whitehead, Thomas Olander, Birgit Anette Olsen and Jens Elmegård Rasmussen (eds.), The Sound of Indo-European: Phonetics, Phonemics and Morphophonemics, 33–52. (Copenhagen Studies in Indo-European 4.) Copenhagen: Museum Tusculanum. Cardona, George (2003): Sanskrit. In: George Cardona and Dhanesh Jain (eds.), The Indo-Aryan Languages, 104–160. (Routledge Language Family Series 2.) London/New York: Routledge.

220 

 Martin Joachim Kümmel

de Vaan, Michiel (2003): The Avestan Vowels. (Leiden Studies in Indo-European 12.) Amsterdam/New York: Rodopi. Hale, Mark (1999): ha: so-called ‘metrical lengthening’ in the Rigveda. In: Heiner Eichner and Hans Christian Luschützky (eds.), Compositiones Indogermanicae in Memoriam Jochem Schindler, 143–151. Praha: Enigma. Hale, Mark (2004): Avestan. In: Roger D. Woodard (ed.), The Cambridge Encyclopedia of the World’s Ancient Languages, 742–763. Cambridge: Cambridge University Press. Hintze, Almut (2007): A Zoroastrian Liturgy: The Worship in Seven Chapters (Yasna 35–41). Wiesbaden: Harrassowitz. Hoenigswald, Henry M. and Roger D. Woodard (2004): Indo-European. In: Roger D. Woodard (ed.), The Cambridge Encyclopedia of the World’s Ancient Languages, 534–550. Cambridge: Cambridge University Press. Hoffmann, Karl (1986): Avestisch ṣ̌. In: Rüdiger Schmitt (ed.), Studia Grammatica Iranica. Festschrift für Helmut Humbach, 163–183. (Münchener Studien zur Sprachwissenschaft: Beihefte, Neue Folge, 13.) München: Kitzinger. Hoffmann, Karl and Johanna Narten (1989): Der Sasanidische Archetypus. Untersuchungen zur Schreibung und Lautgestalt des Avestischen. Wiesbaden: Reichert. Jamison, Stephanie W. (2004a): Sanskrit. In: Roger D. Woodard (ed.), The Cambridge Encyclopedia of the World’s Ancient Languages, 673‒699. Cambridge: Cambridge University Press. Jamison, Stephanie W. (2004b): Middle Indic. In: Roger D. Woodard (ed.), The Cambridge Encyclopedia of the World’s Ancient Languages, 700‒716. Cambridge: Cambridge University Press. Klingenschmitt, Gerd (2000): Mittelpersisch. In: Bernhard Forssman and Robert Plath (eds.), Indoarisch, Iranisch und die Indogermanistik. Arbeitstagung der Indogermanischen Gesellschaft vom 2. bis 5. Oktober 1997 in Erlangen, 191–230. Wiesbaden: Reichert. Kobayashi, Masato (2004): Historical Phonology of Old Indo-Aryan Consonants. (Study of Languages and Cultures of Asia and Africa. Monograph Series 42.) Tokyo: Tokyo University of Foreign Studies. Masica, Colin P. (1991): The Indo-Aryan Languages. (Cambridge Language Surveys.) Cambridge: Cambridge University Press. Oberlies, Thomas (2001): Pāli: A Grammar of the Language of the Theravāda Tipiṭaka. With a Concordance to Pischel’s Grammatik der Prakrit-Sprachen. (Indian Philology and South Asian Studies 3.) Berlin/New York: Walter de Gruyter. Oberlies, Thomas (2003): Aśokan Prakrit and Pāli. In: George Cardona and Dhanesh Jain (eds.), The Indo-Aryan Languages, 161–203. (Routledge Language Family Series 2.) London/New York: Routledge. Schmitt, Rüdiger (1989a): Die Altiranischen Sprachen im Überblick. In: Rüdiger Schmitt (ed.), Compendium Linguarum Iranicarum, 25–31. Wiesbaden: Reichert. Schmitt, Rüdiger (1989b): Altpersisch. In: Rüdiger Schmitt (ed.), Compendium Linguarum Iranicarum, 56–85. Wiesbaden: Reichert. Schmitt, Rüdiger (1989c): Andere altiranische Dialekte. In: Rüdiger Schmitt (ed.), Compendium Linguarum Iranicarum, 86–94. Wiesbaden: Reichert. Schmitt, Rüdiger (2004): Old Persian. In: Roger D. Woodard (ed.), The Cambridge Encyclopedia of the World’s Ancient Languages, 717–741. Cambridge: Cambridge University Press. Skjærvø, Prods Oktor (2009): Old Iranian. In: Gernot Windfuhr (ed.), The Iranian Languages, 43–195. (Routledge Language Family Series.) London/New York: Routledge.



Syllable- and word-related developments in earlier Indo-Iranian  

 221

Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin: De Gruyter. Tremblay, Xavier (2006): Le pseudo-gâthique. Notes de lecture avestiques II. In: Antonio Panaino and Andrea Piras (eds.), Proceedings of the 5th Conference of the Societas Iranologica Europæa held in Ravenna, 6–11 October 2003. Vol. 1: Ancient & Middle Iranian Studies, 233–282. Milano: Mimesis. Wackernagel, Jacob (1896): Altindische Grammatik. Band I: Lautlehre. Göttingen: Vandenhoeck & Ruprecht. Werba, Chlodwig H. (2005): Sanskrit duhitár- und ihre (indo-)iranischen Verwandten: Zur ‘Vokalisierung’ der Laryngale im Ur(indo)arischen. In: Günter Schweiger (ed.), Indogermanica. Festschrift Gert Klingenschmitt. Indische, iranische und indogermanische Studien dem verehrten Jubilar dargebracht zu seinem fünfundsechzigsten Geburtstag, 699–732. Taimering: Schweiger VWT-Verlag.

Damaris Nübling (University of Mainz)

From Christel to Christina, from Klaus to Nico: A diachronic study of German first names (1945–2010) and their shift towards the syllable language type Abstract: German underwent a typological change from a syllable language in Old High German towards a word language today (Szczepaniak 2007). Proper names followed this development until the last century (cf. Christel, Gertrud, Klaus, Wolfgang). Some of the most popular German first names from 2010, however, such as Mia, Lea, Leon, Noah, show completely different structures compared to common nouns. In sharp contrast to common nouns, first names dispose of CV-structures, full vowels in unstressed syllables and different accent positions. Thus, there must have been a deep-rooted onomastic change. The most frequent baby names of 1945 were still in harmony with the usual word structures. This article shows that the decrease of transgenerational transmission of first names led to a departure fom native phonological structures. The following factors are analyzed: the number of syllables; accent position; and the number of consonant clusters, hiatuses, schwa and unstressed full vowels. It will be demonstrated that the phonological distance between first names (particularly female names) and common nouns has increased over time and that there is an increasing tendency for names to contain syllable language structures. Thus, a typological difference developed between these two nominal classes. The reason behind this change can be found in the individualizing function of proper names and social indi­ vidualization over time.

1 German proper names and their changing structures In most phonological studies, proper names are excluded or simply forgotten, possibly due to their exceptional status: Proper names fulfill special functions, the most important of which is to identify solely one person or object without semantic content. Proper names also diverge with respect to their formal proper­ ties: If we consider first names, company and product names and others, they clearly differ from the usual structures of common nouns. They are often longer, they have different accent patterns and frequently full vowels in unstressed syllables. Thus, they do not use lexical material, nor do they even correspond to native patterns; rather they seek to incorporate deviant structures. This can be

From Christel to Christina, from Klaus to Nico  

 223

best observed with names that are invented (product names, company names) or borrowed from other languages (first names). Therefore, it can be suggested that if we are free to create new words, we do not make use of traditional material. However, we could, for instance, use word-formation. This was the case in earlier times when companies were named Drogerie- und Farbwarenhandlung Gehe & Comp. (1835), Haarmanns Vanillin­fabrik and Norddeutsche Affinerie. Today, the names of these companies are Celesio, Symrise and Aurubis. Older dishwashing detergents were called Spüli (< spülen ‘to wash’); today they are named Sunil, Elina Clean and Palmolive. In the past, children were named Sieglinde, Gertrud, Herbert and Wolfgang, but there have not been any Germanic first names among the top twenty name lists since the nineteen-seventies. Today, children are called Mia, Lina, Lilly (girls) and Noah, Luca, Leon (boys). In the following article, it is hypothesized that if people are indeed free to create new words, they look for the greatest difference to the common vocabulary. If the common vocabulary belongs to the word language type, it is expected that syllable language structures will be produced. Evidence for this hypothesis is provided by Ronneberger-Sibold (1993, 1995), who examined product names and word shortenings (acronyms, clippings, truncations, abbreviations) such as Kita ‘daycare center’ and Ersti ‘first-term student’ (instead of Kindertagesstätte and Erstsemester). She found that German short words diverge from common nouns with respect to the properties listed in Table 1. The features of short words clearly tend to be more syllable-language-like: There are predominantly CV syllables, i.e. open syllables, there is no discrepancy between stressed and unstressed vowels and the most frequent vowels use the whole oral cavity (o, i, a – instead of e, i, a with common nouns). Common nouns, in contrast, belong to word language structures: They prefer closed syllables; they have developed a binary vowel system with separate inventories, which is why schwa is the most typical and frequent unstressed vowel; and the most frequent full (stressed) vowels do not contain rounded vowels. Table 1: Phonological differences between shortened and unshortened common nouns (according to Ronneberger-Sibold 1993: 424) Short words (clippings)

Common nouns

1.

preference for open syllables

preference for closed syllables

2.

closed syllables occur word-initially

closed syllables occur word-finally

3.

(almost) no [ə]

[ə] most frequent vowel

4.

most frequent vowels: o–i–a

most frequent stressed vowels: e–i–a

224 

 Damaris Nübling

Fahlbusch (2010, 2011) studied the development of more than 450 names (including 291 renamings) of the 160 most important stock-listed German companies over a period of 160 years. While in earlier times these names usually contained the family names of the company owners or information about the product, they gradually began avoiding these features by getting more opaque and more imaginative, e.g. Manufactur-, Confections- und Tuchgeschäft C. Karstadt > ARCANDOR AG (see Figure 1).

Figure 1: Percentage of anthroponomastic (company owners’ names) and invented components (fantasy) in German company names 1850–2009 (Fahlbusch 2011: 57)

Furthermore, they lost lexical transparency (Figure 2): Drogerie- und Farbwarenhandlung Gehe & Comp. later became Celesio AG, and Morphosys Gesellschaft für Proteinoptimierung GmbH simply became MorphoSys AG.

From Christel to Christina, from Klaus to Nico  

 225

Figure 2: Percentage of fully transparent company names 1850–2009 (Fahlbusch 2011: 71)

As a third finding, the proportion of loaned elements increased: Haarmanns Vanillinfabrik became Symrise AG. Mechanische Werkstätten Harkort & Co. became Demag Cranes AG.

Figure 3: Percentage of foreign company names 1850–2009 (Fahlbusch 2011: 64)

226 

 Damaris Nübling

Even in writing, company names contained an increasing number of graphical anomalies, e.g. with unorthographical punctuation between letters or with capital writing within the word. Many of these changes were unlawful and later legalized. It can be concluded that the distance between the common vocabulary and company names increased over time with regard to lexical, phonological and graphical features.

2 German first names and their historical development The question arises whether personal names also follow these diverging tendencies. We now turn to first (or given) names in Germany. In the following section, we concentrate on the 40 most frequent ones. As a basis of comparison, I took the 40 most frequently occurring common nouns of spoken language (most of the frequency dictionaries are based on written language which were therefore excluded). Ruoff (1990) (Häufigkeitswörterbuch gesprochener Sprache) is the most well-known frequency dictionary of spoken language; it is based on interviews with Southwest German speakers from the 1970s. Most of the informants told about their lives as farmers and their experience in the Second World War which explains why Krieg ‘war’ constitutes the tenth most frequent noun. However, clear dialect words occur such as Bub ‘boy’ and Mädlein ‘girl’ which makes the corpus less representative. Furthermore, plural forms were not counted separately, but listed as singular (base) forms. This corpus does not reflect the real occurrences of common nouns. Another corpus, the so-called Pfeffer corpus (Pfeffer and Lohnes 1984), shows that the plural Kinder ‘children’ occurs more often than the singular Kind ‘child’; the same applies to Dinge(n), Sachen ‘things’, Menschen ‘people’ and other nouns. Although proper names usually occur only in the singular, they should be compared with common nouns as they typically appear in spoken language. Therefore, the Pfeffer corpus was taken as a basis (Table 2, right side).

From Christel to Christina, from Klaus to Nico  

 227

Table 2: Frequency list of the 40 most frequent common nouns according to Ruoff (1990) and Pfeffer and Lohnes (1984) Ruoff (1990) (not considered) Position 1–14 15–28 29–40 Jahr ‘year’ Holz ‘wood’ Stück ‘piece’ Tag ‘day’

Ding ‘thing’

Bruder ‘brother’

Leute ‘people’

Kuh ‘cow’

Zeit ‘time’ Haus ‘house’

Winter ‘winter’ Maschine ‘machine’ Hand ‘hand’ Woche ‘week’

Mann ‘man’

Frau ‘woman’

Wald ‘forest’ Kind ‘child’

Vieh ‘cattle’ Mark ‘d-mark’ Stunde ‘hour’

Vater ‘father’

Landwirtschaft ‘agriculture’ Mädlein ‘girl’ Sommer ‘summer’ Morgen ‘morning’ Feld ‘field’ Teil ‘part’

Pfeffer and Lohnes (1984) (considered) Position 1–14 15–28 29–40 Zeit ‘time’ Hause Tage ‘days’ ‘(go/be) home’ Beispiel Teil ‘part’ Schüler ‘pupil’ ‘example’ Schule ‘school’ Beruf ‘job’ Uhr ‘clock’ Kinder ‘children’

Mann ‘man’

Winter ‘winter’

Jahr ‘year’ Stadt ‘city’

Frau ‘woman’ Menschen ‘men’ Vater ‘father’

Dinge ‘things’ Leben ‘life’

Leute ‘people’

Abend ‘evening’

Jahre ‘years’ Jahren ‘years (dat.)’ Herr ‘mister’

Bauer ‘peasant’ Arbeit ‘work’

Bub ‘boy’

Arbeit ‘work’

Geschäft ‘shop’

Hof ‘farm’

Stall ‘stall’

Haus ‘house’

Sache ‘thing’

Geld ‘money’

Tag ‘day’

Mutter ‘mother’

Schule ‘school’

Dingen ‘things (dat.)’

Krieg ‘war’

Sport ‘sport’ Familie ‘family’ Lehrer ‘teacher’ Sommer ‘summer’ Mutter ‘mother’ Eltern ‘parents’ Sachen ‘things’

Sprache ‘language’ Geld ‘money’ Wetter ‘weather’ Frage ‘question’ Mädchen ‘girl’ Art ‘sort’

The Pfeffer corpus represents colloquial spoken language from 1961. It is based on audio recordings of nearly 80 hours of mono- and dialogues in 57 German, Austrian and Swiss towns (185 women, 218 men). The dark-colored cells in Table 2 show the overlapping vocabulary of both frequency lists: This constitutes more than half of the words, so the differences are not too considerable. This confirms that we have a realistic representation of the top nouns. In this study, the top 40 given names list includes the top 20 girls’ and boys’ names. The time span from 1945 to 2010 was divided into 14 five-year subperiods which resulted in a total of 560 single names. As we are concentrating only on the top 20 girls’ and boys’ names, this is just the tip of the iceberg. Table 3 shows the 40 most frequent first names of 1945, 1975 and 2005. These three time segments do not share any common names except one: Michael (bold) in 1945 and 1975. Thus, a complete onomastic replacement has taken place. Even more, it becomes obvious that the phonological type of names changed dramatically, which will be shown in section 3.

228 

 Damaris Nübling

Table 3: Frequency list of the 40 most frequent first names of 1945, 1975 and 2005 1945

1975

2005

girls

boys

girls

boys

girls

boys

1

Renate

Hans

Sandra

Christian

Leonie

Lukas

2

Monika

Peter

Stefanie

Markus

Hanna

Leon

3

Karin

Klaus

Nicole

Michael

Anna

Luka

4

Ursula

Wolfgang

Kathrin

Stefan

Lea(h)

Finn

5

Brigitte

Jürgen

Tanja

Andreas

Lena

Niklas

6

Bärbel

Uwe

Anja

Thomas

Laura

Jonas

7

Elke

Bernd

Ivonne

Alexander

Emilie

Tim

8

Ingrid

Karl

Julia

Sven

Lara

L(o)uis

9

Helga

Horst

Claudia

Thorsten

Sophie

Jan

10

Christa

Dieter

Melanie

Jan

Marie

Paul

11

Gisela

Günther

Katja

Matthias

Julia

Felix

12

Hannelore

Heinz

Nadine

Frank

Sarah

Jannick

13

Jutta

Rainer

Silke

Martin

Lilli

Julian

14

Barbara

Michael

Andrea

Jens

Emma

Max

15

Heike

Manfred

Sonja

Sebastian

Lina

Philipp

16

Christel

Rolf

Susanne

Marco

Johanna

Maximilian

17

Marion

Gerhard

Bettina

Oliver

Ne(e)le

Ben

18

Erika

Werner

Daniela

Andre/é

Alina

Moritz

19

Angelika

Gerd

Sabine

Mark

Luisa

Nico

20

Anke

Helmut

Alexandra

Daniel

Sophia

Tom

First, the question of how many children have one of the most frequent names should be clarified. This depends on the period of time. It is well established that over time people have made use of an increasingly ample inventory of names, i.e. naming has become more individual (Debus 1977; Seibicke 1991, 2008; Gerhards 2003: 101–125). It is estimated that in the nineteen-seventies, each of the most frequent male names covered 5–6% of all newborn boys and each of the most frequent female names 3–4% of the girls. Today, the most frequent name covers only about 1% with regard to both sexes (girls: 0.95%, boys: 1.04%), the tenth most common female name 0.66% and male name 0.76% (according to www.beliebte-vornamen.de). This directly corresponds to social individualization (according to Gerhards 2003: 101–125). In order to understand the following developments, some remarks on the history of name-giving in Germany are necessary. It was not until the nine-

From Christel to Christina, from Klaus to Nico  

 229

teenth century that people had the freedom to name their children anything they wanted. In former times, a strict transmission of first names was the only practice. This means that a child’s name was already determined before it was born. In most cases children were named after their parents or grandparents, godparents, saints, monarchs, etc. The English term for this transmission of names is the German loan word nachbenennung. The practice of nachbenennung was used for many centuries. Due to this long tradition of onomastic recycling, the name inventory became smaller and smaller – which eventually led to the emergence of family names. Thus, the diversity of first names was very restricted. Sometimes, more than 20% of a population was called Johannes or Margarete (Kunze 2003). In those times, there was no such thing as “fashionable” names because people did not have much choice regarding the names of their children. This custom of nachbenennung is also called bound naming. Since the nineteenth century, bound naming gradually gave way to free naming. At the same time, the motivation for choosing a name changed. Today, the most dominant factor in naming is euphony. Since 1900, at the latest since 1945, naming is considered to be free. This process of decreasing nachbenennung is shown in Figure 4, taken from Gerhards (2003). On the basis of 100 spot tests of two baptism registers from 1890 to 1994, he found that until the nineteen-sixties, the rate of nachbenennung after the parents ranged between 20% and 30%. Since 1970, the rate has gone quickly down to almost zero today.

Figure 4: Transgenerational transmission (nachbenennung) of first names in Germany 1890– 1994 (Gerhards 2003: 91)

Today, nachbenennung occurs very seldom and depends on several social and regional factors. Firstly, more boys than girls are subject to this custom which can be explained by the “son and heir” principle. Secondly, nachbenennung is practiced more often in the South than in the North of Germany. Thirdly, it is an upper-class phenomenon and, fourthly, it occurs more frequently in Catholic than Protestant areas. Fifthly, if a child has one or more middle names, which

230 

 Damaris Nübling

generally happens more often in the South than in the North, nachbenennung moves to the second (or third) name, cf. Leonie Maria, Leon Peter (Simon 1991; Kunze 2003: 55). Figure 4 only refers to the first given names. Figure 5 is taken from Debus’ (1985) investigation of nachbenennung after parents, grandparents, other relatives, and godparents in Kiel from 1957 to 1966.

Figure 5: Nachbenennung in Kiel from 1957 to 1966 (Debus 1985)

His study is based on interviews about parents’ naming motives. Although he investigated only a short period (10 years), the downward trend is very obvious. Furthermore, Debus differentiated according to sex, with the clear result that boys were more frequently named after relatives and godparents than girls. Simon’s (1989, 1991) onomastic studies of Westphalia paint a similar picture.

From Christel to Christina, from Klaus to Nico  

 231

Now the question arises whether the decrease of nachbenennung is reflected in the structure of the personal names. Was this change in tradition followed by a change in name types? As we saw in the beginning, people do not reproduce the usual linguistic structures (which in German correspond to a word language type) if they are allowed to create (or to borrow) new words. The new structures clearly contain more syllable language features. The prototypical phonological word in German comprises one foot and one to two syllables, e.g. Tag ‘day’, Haus ‘house’, Mutter ‘mother’, Leute ‘people’ (Eisenberg 1991, 2006; Hall 1999; Wiese 2000; Szczepaniak 2007). Inflected words clearly tend to be disyllabic: Tage ‘days’, Häuser ‘houses’. They constitute trochees with a final reduced syllable which contains schwa ([ə] or [ɐ]) (Figure 6). Proper names, however, diverge in many respects from these structures. This is demonstrated in Figure 7 which analyzes a typical German girl name, Michaela. This word, in contrast, contains two feet, the second one even being the stronger one (the full stress is on the third syllable of the whole phonological word). Furthermore, there is no reduced syllable (no schwa). Usually, first names have simpler syllable structures (CV) which is also the case here.

[ 't

ω

ω

ω

ω

F

F

F

F

σ

σ

σ

aː [ 't k ] aː

[ 't

k]

σ σ

σ

aː[ 't g aː əg]

ə]

Figure 6: The phonological structure of Tag ‘day’ and Tage ‘days’

Fw σ [ mi '

ω

ω

Fw

Fs

σσ

σσ

ç ia [ m '

ç'eːa

Fs σσ

σ

'eːl a ]

la]

Figure 7: The phonological structure of Michaela [ˌmiçaˈeːla]

232 

 Damaris Nübling

Proper names such as Elke, Christel or Peter as well as monosyllabic names such as Klaus and Heinz fully correspond to the common phonological word. Today, they are completely out of fashion – but not, as one might assume, because they were overused in previous generations and are now considered old-fashioned. It will be argued that it is their phonological word quality which disqualifies them as first names: First names by and by become more linguistically distant to common nouns. Today it can be observed that some old-fashioned names are coming back in style, such as Emma, Emilia, Laura, Anna. This, however, is not only due to a revival of old traditions. Rather, it is their onomastic quality which is making them popular again. Other old-fashioned names such as Wilhelm and Friedrich or Gertrud and Sieglinde are not coming back in style. This will be discussed in the next section.

3 Linguistic comparison of first names and common nouns since 1945 As mentioned above, our analysis is based on the top 20 girls’ and boys’ names from 1945 to 2010 (in five-year subperiods). In a first step, all proper names and common nouns were transcribed using IPA. Then, the average number of syllables was calculated, followed by the most common sounds, and so on. The transcription method is a compromise between a phonological and a phonetic transcription. In German, every [r] in final position (Peter) and before a consonant (Werner) is vocalized and thus counted as schwa [ɐ]. Prevocalic [r] and [r] after a consonant were counted as a voiceless fricative [χ], as in Marion or Christina. Two adjacent vowels belonging to different syllables were counted as hiatuses: Marion, Florian, Fabian, Michael. Some Germans pronounce a glottal stop in Michael [ˈmiçaˌɁeːl]. This was not considered here. In the following section, the changes of several onomastic features compared to common nouns will be presented. With respect to the common nouns, no change over the past 65 years is assumed as the elementary vocabulary should stay more or less the same over time (see Table 2, which obviously contains elementary everyday vocabulary such as ‘child’, ‘day’, ‘time’, ‘year’, ‘house’). Therefore, only a simple (straight) line was inserted in the following figures which is contrasted with the rapidly changing onomastic structures from 1945 to 2010.

From Christel to Christina, from Klaus to Nico  

 233

3.1 Number of syllables The average common noun consists of 1.7 syllables as calculated from the 40 most frequent items. Table 4: Number of syllables of the 40 most frequent common nouns (Pfeffer corpus) Number of syllables

Tokens

1

13

2

26

3

1

Average

1.7

If we compare this average with first names, we get the following result (Figure 8): Proper names are clearly always longer than common nouns. In 1945, they started with 2.1 syllables. In the seventies, they suddenly became much longer, even up to 2.6 syllables. Interestingly, they have been getting shorter again since 1985. Today (2010), they have reached the same value as 1945. All in all, first names are clearly longer than common nouns.

Figure 8: The number of syllables in first names and common nouns

234 

 Damaris Nübling

If we consider the sexes separately, clear differences can be found (Figure 9): Boys’ names started out having the same number of syllables as common nouns, whereas girls’ names were already considerably longer than common nouns in 1945. After the sixties, they converged with the male names by getting shorter, whereas, at the same time, male names became longer so that both adopted a rather similar length. Since 1970, a clear androgynous development in length can be observed. It should be kept in mind that this strongly corresponds to the end of bound naming. As the “son and heir” principle did not apply to girls (they usually married into another family), their names clearly already diverged more from the common noun structures than male names in 1945.

Figure 9: The number of syllables in girls’ and boys’ names

3.2 Accent positions The most radical differences, both between common nouns and names and between female and male names, can be stated with respect to the accent positions (Figures 10–12).

From Christel to Christina, from Klaus to Nico  

 235

Figure 10: Differences in length and accent positions of common nouns (left column) and female first names

Figure 10 illustrates the different features of female names. The far-left striped column shows the values for the common nouns: One third are monosyllabic, and the rest are almost entirely disyllabic with initial stress. In contrast, there are almost no monosyllabic female names at all, except one, Kim, in 1995. Most of the female names are di- and trisyllabic. The disyllabic names mostly have initial stress (dark grey). Prototypical examples are Elke (with schwa) in former times and Lea currently (without schwa but with a hiatus). We will later return to these phonological differences. Since the nineteen-seventies, there have also been final stressed disyllabic names such as Nadine [naˈdiːn] and Yvonne [iˈvɔn]. This type does not appear at all in the most popular names for boys. The typical female name is trisyllabic with initial stress (Monika, light grey) or non-initial stress (white, as in Brigitte). There have been many changes during these sixty-five years; at the moment, the disyllabic names are coming back in style. In every period, the distance to the common nouns is considerable. The male names (Figure 11), however, are closer to the common nouns (striped left column). We find many monosyllabic names (Heinz, Karl, Horst, Klaus, black) as well as trochaic disyllabic names, such as Peter (in former times, with schwa) and Leon (today, without schwa but with a hiatus). As we already know, the boys’ names have become longer since the nineteen-seventies (trisyllabic names are colored in white and light grey); here, we can observe a growing distance

236 

 Damaris Nübling

Figure 11: Differences in length and accent positions of common nouns and male first names

Figure 12: Differences in accent positions of common nouns and first names

From Christel to Christina, from Klaus to Nico  

 237

to the common noun structures. Until the nineteen-seventies, male names were more tightly bound to intrafamilial transmission. The girls were already mostly unbound from these traditions, which can be seen in Figure 5. Furthermore, non-initial stress seems to be a feature of female names. Due to this function, the overall onomastic picture in Figure 12 is not very significant. However, some changes can be observed over the sixty-year time span. To sum up, in the beginning of the studied time period, female names already had rather different accent positions and syllable numbers compared to common nouns, whereas boys’ names commenced with nearly the same structures of common nouns and later dissociated over time.

3.3 Consonant clusters To count the amount of consonant clusters, a very simple method was used: Every combination of two consonants – irrespective of its position in the syllable structure – was counted as a cluster, except the combination of [r] + consonant (Gerhard), as [r] is vocalized in this position. Combinations of three consonants (which are very rare) were counted as two clusters, as in Heinz [haɪnts]: [nt] + [ts]. The consonantal structure of the most frequent common nouns is rather simple; in sum, there are 14 clusters within the 40 top common nouns (Figure 13). The first names, however, started with higher values, but then reduced their complexity considerably. Today they are even less complex than common nouns. This radical change towards CV structures started in the late nineteen-seventies. At the same time, there were dramatic changes on the vocalic level, which is the topic of section 3.6. Regarding consonant clusters, there are large gender-related differences: The boys’ names started with more consonant clusters and still contain more, whereas the girls’ names use more CV structures (Figure 14). Since 2005, the top female names have been completely free of clusters, whereas in 2010 five of the top male names still contained clusters (Maximilian, Max, Felix, Niclas, Moritz). To get an impression of this deep-rooted change, some examples of the top 20 names of 1945 and 2010 are juxtaposed:

238 

 Damaris Nübling

Figure 13: The development of consonant clusters in first names

Figure 14: Sex-related differences in the development of consonant clusters in first names

From Christel to Christina, from Klaus to Nico  

 239

Male names 1945: Hans, Horst, Heinz [nts], 2010: Leon, Ben, Finn, Luis, Tim, Klaus, Rolf, Bernd, Jan, Ma[ks], Elias, Noah, Wolfgang, Manfred … Julian, Jonas, Luca … Female names 1945: Christa, Elke, Ingrid, 2010: Anke; Brigitte, Christel Helga, Angelika …

Mia, Hanna, Lena, Lea, Leonie, Lina, Lilly, Laura, Lara, Emma, Maja …

Few consonant clusters mean that there are many CV structures. Again, it becomes obvious that the first names started with native structures and later incorporated more syllable language features. The relatively low amount of clusters in common nouns can be explained by the absence of compounds. In contrast, names such as Wolfgang and Manfred consist of compounds which go back to word formation in Old High German or even earlier times. The high amount of onomastic consonant clusters in the nineteen-sixties and -seventies is caused by rather long names such as Stefan/Stefanie, Kathrin, Claudia, Alexandra/Alexander, Andrea/ Andreas and Sebastian. Figures 13 and 14 only contain the absolute number of clusters and not their share relative to the number of syllables.

3.4 Hiatuses Hiatuses of two full vowels are rather seldom for common nouns. Native hiatuses always consist of a stressed diphthong au [au], ei [ai] and eu/äu [ɔi] + schwa [ə] or [ɐ]: Bauer ‘farmer’, Reiher ‘egret’, teuer ‘expensive’. Other hiatuses always belong to foreign words (Linguistik, Theater, Theologie). In our sample of the top 40 of common nouns, there is only one example, the loan word Familie [faˈmiː.li.ə], where the hiatus [i.ə] occurs in unstressed position (in fast speech, it is contracted to [jə]). The first names – here, both sexes behave similarly – start with the same values as common nouns, but they increase their number of hiatuses dramatically, again in the seventies (Figure 15).

240 

 Damaris Nübling

Figure 15: The development of hiatuses in first names from 1945 to 2010

A closer look at the accent position reveals interesting changes: The hiatuses move from unstressed to stressed syllables, from the background to the foreground, or, in linguistic terms, they become more salient. This is not visible in Figure 15. Therefore, some examples are provided in Figure 16.

unstressed hiatuses

1945

ˈMarion – ˈMichael –

1980 ˈJulia, ˈClaudia ˈChristian, Seˈbastian, ˈMichael, ˈDaniel, ˈFlorian 2010 – (?Eˈmilie? / ˈEmil[iː]) Maxiˈmilian, ˈJulian

stressed hiatuses

Daniˈela Maˈtthias, Anˈdreas, Toˈbias ˈMia, ˈLea, ˈLeonie, Soˈfia, Luˈisa ˈLeon, ˈLuis, Eˈlias, ˈNoah

Figure 16: Increase of hiatuses and their transition into stressed positions

From Christel to Christina, from Klaus to Nico  

 241

First of all, there are no native hiatuses as there are no schwas involved. The only exception is Emilie which, together with Emily, was ranked ninth in 2010. It is very likely that both are pronounced with initial stress and final [iː] but the older German pronunciation Eˈmil[iə] cannot be excluded. However, this is rather improbable. The fact that both writings share the same position indicates that there is only a graphical and not a phonological difference. Thus, every onomastic hiatus consists of two full vowels. In 1945, they were few (solely two out of 40 names) and occurred exclusively in unstressed position. In 1980, they increased in number and also spread into the stressed position. Today (2010), most of the hiatuses are stressed (mostly on the first vowel and thus without a glottal stop in between) and not accompanied by further syllables: The current first names nearly exclusively contain one hiatus (Mia, Lea, Noah). Concerning the question of syllable versus word language typology, on the one hand hiatuses do not improve the syllabic quality because they cause a less preferred syllable contact. On the other hand, hiatuses provide more similar syllables, i.e. they diminish the discrepancy in sonority between the syllables within a word. Contrary to the German word language type, they violate the principle of highest sonority in the stressed syllable since they produce similar sonority degrees on both syllables. Thus, this onomastic development towards stressed hiatuses with full vowels is not only a departure from native structures – it is a development towards syllable language structures. As German is a highly developed word language, nearly all onomastic changes result in more syllable language structures. Therefore, Germans use the onomastic inventories of syllable languages when they are looking for euphonic names.

3.5 Final schwa A complementary picture to the increase of hiatuses is provided by the development of final schwa: Here, a dramatic downfall to zero can be observed over time (Figure 17). Again, a glance at the behavior of both sexes is revealing (Figure 18). As usual, the boys’ names started with more common noun structures (trochees, second syllable with schwa, see Peter, Jürgen, Uwe, Dieter, Günther, Rainer, Werner), whereas the girls’ names started with lower values, i.e. they had fewer names with common noun structures (Anke, Elke, Renate, Brigitte, Bärbel, Christel). However, it must be pointed out that in Figure 18 the number of schwas is related to the total number of unstressed vowels. Boys’ names quickly reduced their schwas after 1965 and replaced them, as we will see in section 3.7, with full vowels. Here, the male names underwent a “feminization”. Since the seventies

242 

 Damaris Nübling

again, both sexes have converged and further reduced this vowel, which is the most frequent vowel in common nouns.

Figure 17: The ratio of schwa compared to all unstressed vowels

Figure 18: The ratio of schwa compared to all unstressed vowels in common nouns and in male and female names

From Christel to Christina, from Klaus to Nico  

 243

The onomastic loss of schwa clearly supports the trend towards syllable language features, which usually includes a symmetric vowel inventory occurring in every syllable, be it stressed or unstressed. Good examples are Italian and Spanish where nearly every vowel can appear in every syllable (Szczepaniak 2009). This was similar in Old High German. In Middle High German at the latest, two vowel inventories developed: 23 stressed vowels opposed to one unstressed, schwa [ə] (Szczepaniak 2007: 188); later, a second central vowel [ɐ] came along. All in all, this still applies for New High German with 18 stressed vowels and two central vowels. Not surprisingly, [ə] is the most frequent vowel in German nouns. This extreme vocalic discrepancy is overridden in first names (as well as in other names and in short words): In 1945, schwa was still included in many first names. Step by step, it became less common – until there were none in the top names of 2010. Both sexes show a similar decrease.

3.6 Full vowels in unstressed positions Knowing that the average number of syllables in names never falls below 2 (Figure 8), we get a lot of unstressed vowels. Since 1980, they only very rarely consist of schwa. This leads to the question of which vowels are replacing them. Figure 19 shows the ratio of full vowels in unstressed positions.

Figure 19: Unrounded unstressed vowels in common nouns and first names

244 

 Damaris Nübling

Complementary to Figure 18 (development of schwa), there is a high increase in unrounded unstressed vowels in Figure 19, above all [a] and [i]. Today, 82% of all unstressed vowels consist of these two vowels. Generally there are only few rounded vowels, even more in female than in male names. Most important is the fact that final -a is a very strong sex indicator for female names. In recent years, the male names have converged even with respect to this important feature (cf. Noah, Luca, furthermore Jonas, Elias).

4 Conclusion Most of the contemporary discrepancies between common nouns and first names correspond to the differences between word versus syllable languages. Even the high amount of stressed and unstressed hiatuses, which blur the syllable contact instead of producing clear CV structures, support syllable language features because they create more similar syllable nuclei with regard to sonority degrees. Thus, first names make use of completely different hiatuses which – in contrast to the native ones – never contain schwa and which also can occur in unstressed position (ˈMichael). By this means, they clearly distinguish themselves from the common nouns by disobeying the sequence of full and reduced syllables or, in other words, the alternation of high and low sonority degrees. The other onomastic peculiarities fully conform to syllable languages (see Figure 20). Thus, German separates the two functionally diverging noun classes of common nouns versus proper names by typological means. This does not mean that the German language as a whole is moving in this direction: Proper names stand in many respects outside the language system. They are often not bound to the native vocabulary, so it is not astonishing that they use different means. Even in writing, proper names strongly diverge from common nouns. They are not even subject to the same orthographic rules. As Figure 20 shows, female names diverge more than male names from native structures, even in the beginning of the studied time period in 1945. Gender differences seem to be not only of high social relevance, they also have to be coded unambiguously on the name as it is even regulated by law in Germany. Both sexes have converged over time with respect to some features (e.g. the average number of syllables, full unstressed vowels, loss of schwa, high amount of hiatuses), but with respect to others, they have clearly diverged: Monosyllabic names are overwhelmingly male (Ben, Finn), iambic names are female (Naˈdine, Yˈvonne), there are more open final syllables and less consonant clusters in female names. This complex of gender up- and downgrading between 1945 and 2009 was investigated by Nübling (2009a, b) with the result that girls’ and boys’ names have never been as similar

From Christel to Christina, from Klaus to Nico  

Word language type Features of common nouns in German: • lexical (initial) stress • monopedal words • monosyllabics and trochees • reduced syllables (schwa) • strengthened word edges (clusters) • alternation of full + reduced vowels

 245

Syllable language type Features of contemporary first names: • different stress patterns • polypedal words • polysyllabic words (up to 5 syllables) • full unstressed vowels (no schwa) • CV structures (no clusters) • full vowels in every syllable

male names 1945 Klaus Norbert

female names 1945 ˈChristel Barbara

2010 Nico Noah

2010 Chrisˈtina Leonie

Figure 20: Overview of the most important developments

as they are today, i.e. they became rather androgynous over time. In these studies, more linguistic aspects were considered, e.g. the overall amount of sonority. This fact of convergence can be explained by the feminist movement since the sixties and the equal treatment of both sexes starting in the seventies and eighties. Here, we only state that male names use more native structures than the female ones. This can best be explained by social gender differences which ascribe men the role of son and heir in contrast to women who traditionally left their family after marriage. This led to the persistent tradition of male nachbenennung within the family. This tight connection is demonstrated in Figure 21 which reveals a parallel between the social practice of nachbenennung (holding for both sexes) and one linguistic development, the loss of schwa in first names. The seventies can be identified as a turning point. a) (1946–1994) b) Decrease of of schwa schwa (1945–2010) (1945–2010) a)Decrease Decrease of of nachbenennung nachbenennung (1946–1994) b) Decrease

Figure 21: Parallel developments of social and linguistic changes

246 

 Damaris Nübling

Looking back to Figure 20, it must be pointed out that only special syllable language features have entered German first names, for example different accent positions which are not determined lexically and which diverge from the rather strict trochaic stress pattern of German. The same holds for the length of the pword (1–2 syllables) which is often ignored in first names. Conversely, foreign syllable language features such as geminates and weak or tonal accents will probably never enter German first names. Assuming that a name like Anna was loaned from Italian, the geminate would be replaced by an ambisyllabic short [n]. Most of the syllabic properties of German first names consist of purely phonotactical reorderings and simplifications (CV structures, full vowels in unstressed syllables) and in the exceedance of the pword length. Qualitatively new features or even phonological processes (like nasals or other non-native sounds, vowel epenthesis, vowel harmony) will never penetrate the onomastic system. The sound inventory was not enlarged, but rather the combination of the sounds. Thus, foreign names do not primarily fulfill syllable language requirements; they first of all have to contrast with the native system. This task is best performed by typologically divergent structures. Summing up, the typological change of first names directly corresponds to the non-linguistic decline of nachbenennnung. First names clearly reflect social changes. The decline of nachbenennnung led to the dissociation of common nouns and first names. Possibly, the real reason why we choose so many foreign names could be that they provide us with a pool of diverging structures. We are probably less interested in imitating or adopting foreign cultures than in escaping our native word language structures in order to mark a special, perhaps the most important, class of words: proper names.

References Debus, Friedhelm (1977): Soziale Veränderungen und Sprachwandel. Moden im Gebrauch von Personennamen. In: Hugo Moser (ed.), Sprachwandel und Sprachgeschichtsschreibung im Deutschen, 167–204. (Schriften des Instituts für deutsche Sprache 41.) Düsseldorf: Schwann. Debus, Friedhelm (1985): Zur Pragmatik von Namengebung und Namengebrauch in unserer Zeit. Beiträge zur Namenforschung 29: 305–343. Eisenberg, Peter (1991): Syllabische Struktur und Wortakzent: Prinzipien der Prosodik deutscher Wörter. Zeitschrift für Sprachwissenschaft 10/1: 37–64. Eisenberg, Peter (2006): Grundriss der deutschen Grammatik. Vol. 1: Das Wort. Stuttgart: Metzler. Fahlbusch, Fabian (2010): Von KarstadtQuelle zu Arcandor, von MTU Friedrichshafen zu Tognum. Prinzipien des diachronen Wandels von Unternehmensnamen. Master’s thesis, University of Mainz.

From Christel to Christina, from Klaus to Nico  

 247

Fahlbusch, Fabian (2011): Von Haarmanns Vanillinfabrik zu Symrise, von der Norddeutschen Affinerie zu Aurubis: Prinzipien des diachronen Wandels von Unternehmensnamen. Beiträge zur Namenforschung 46: 51–80. Gerhards, Jürgen (2003): Die Moderne und ihre Vornamen. Eine Einladung in die Kultursoziologie. Wiesbaden: Westdeutscher Verlag. Hall, T. Alan (1999): The phonological word: A review. In: T. Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 1–22. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Kunze, Konrad (2003): dtv-Atlas Namenkunde. Vor- und Familiennamen im deutschen Sprachgebiet. 4th ed. München: Deutscher Taschenbuch Verlag. Nübling, Damaris (2009a): Von Monika zu Mia, von Norbert zu Noah: Zur Androgynisierung der Rufnamen seit 1945 aus prosodisch-phonologischer Perspektive. Beiträge zur Namenforschung 44: 67–110. Nübling, Damaris (2009b): Von Horst und Helga zu Leon und Leonie: Werden die Rufnamen immer androgyner? Der Deutschunterricht 5: 77–83. Pfeffer, Alan and Walter Lohnes (eds.) (1984): Grunddeutsch. Texte zur gesprochenen deutschen Gegenwartssprache. Einführungs- und Registerband. (Phonai 28.) Tübingen: Niemeyer. Ronneberger-Sibold, Elke (1993): On different ways of optimizing the sound shape of words. In: Henning Andersen (ed.), Historical Linguistics 1993. Selected papers from the 11th International Conference on Historical Linguistics, Los Angeles, 16–20 August 1993, 421–432. (Current Issues in Linguistic Theory 124.) Amsterdam/Philadelphia: John Benjamins. Ronneberger-Sibold, Elke (1995): Die Optimierung von Lautgestalten durch synchrone Wortkürzung und durch langfristigen Sprachwandel. In: Norbert Boretzky, Wolfgang Dressler, Janez Orešnik, Karmen Teržan and Wolfgang U. Wurzel (eds.), Natürlichkeitstheorie und Sprachwandel, 31–44. Bochum: Brockmeyer. Ruoff, Arno (1990): Häufigkeitswörterbuch gesprochener Sprache. (Idiomatica 8.) Tübingen: Niemeyer. Seibicke, Wilfried (1991): Vornamen. 2nd ed. Frankfurt am Main: Verlag für Standesamtwesen. Seibicke, Wilfried (2008): Die Personennamen im Deutschen. 2nd ed. Berlin/New York: De Gruyter. Simon, Michael (1989): Vornamen wozu? Taufe, Patenwahl und Namengebung in Westfalen vom 17. bis zum 20. Jahrhundert. (Beiträge zur Volkskultur in Nordwestdeutschland 67.) Münster: Coppenrath. Simon, Michael (1991): Der Pate als Namengeber. Rheinisch-westfälische Zeitschrift für Volkskunde 36: 215–227. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Szczepaniak, Renata (2009): Wortsprachliches Deutsch und silbensprachliches Spanisch. Ein phonologisch-typologischer Vergleich. Estudios Filológicos Alemanes 17: 251–267. Wiese, Richard (2000): The Phonology of German. 2nd ed. Oxford: Clarendon Press.

Part 3: Synchronic approaches (Germanic languages)

Pia Bergmann (University of Freiburg)

Reduction and deletion of glottal stops and geminates at phonological word boundaries in German compounds: Effects of word frequency and accentuation* Abstract: In the typology of syllable and word languages, the standard variety of modern German is regarded as a word language, which implies that morphological structure is systematically highlighted by phonological means. Next to phonotactic properties like complex syllable onsets or codas at word edges, ambisyllabic consonants, and others, specific boundary markers serve to strengthen the edges of phonological words which regularly correlate to lexical words. However, word boundary marking may be subject to systematic weakening under certain structural and usage-based conditions. This paper focusses on the influence of lexical frequency and accentuation on boundary strength by means of an acoustic-phonetic production study. The edge marking phenomena under investigation are glottal stop deletion or reduction, and the deletion or reduction of geminates across the boundary of two adjacent phonological words. The results lead to the conclusion that both structural and usage-based features indeed exert a systematic influence on acoustic reduction at phonological word boundaries.

1 Introduction The standard variety of New High German is regarded as a word language as opposed to a syllable language like Swiss German or even earlier stages of German (Old High German) (cf. Szczepaniak 2007; Nübling and Szczepaniak 2008; Siebenhaar, this volume). Word languages and syllable languages differ in their phonological structure inasmuch as the former optimize the prosodic domain of

* The present study was part of the project “Edge-marking in German compounds: Effects of frequency and prosodic constituents” (AU 72/18-1), funded by the German Research Foundation as part of the Priority Programm 1234 “Phonological and phonetic competence: between grammar, signal processing and neural activity”. We thank the editors of this volume for many valuable comments on an earlier version of this paper. We further want to thank Sascha Wolfer for his support with the statistical analysis, as well as the student assistants of the project: Meike Stärke, Sonja Fischer and Katharina Kurz.

252 

 Pia Bergmann

the phonological word (pword) whereas the latter optimize the syllable. One way to emphasize the pword is to strengthen its boundaries by means of additional consonantal material, as has happened in the history of High German in words like nie.man > nie.mand ‘nobody’ or je.man > je.mand ‘somebody’ (cf. Szczepaniak 2007, this volume). The epenthesis of the consonant /t/ leads to a more complex syllable coda which means that the syllable structure is more marked, because it departs more strongly from the unmarked CV structure. The enhancement of the pword thus works at the expense of an optimal syllable structure. Other ways to strengthen the boundary of a pword are glottal stop insertion, stop aspiration, blocking of degemination across the word boundary and blocking of assimilations across the word boundary (cf. Auer 1994, 2001; Nübling and Szczepaniak 2008). Despite the fact that from a phonological point of view the boundary strengthening tendencies are well documented, from the viewpoint of phonetics, we encounter a great deal of variation in the realization of the prosodic boundaries. This occurs, for example, when frequent words like Hauptbahnhof ‘main station’ are realized with deleted /t/ ([ˈhaupbaːnhoːf]), whereas infrequent words like Hauptpastor ‘main priest’ keep the /t/ (cf. Bergmann 2012b). Against this backdrop, the main interest of the present study lies in the question of how the prosodic boundary of a pword is realized and what factors systematically influence its production. We examine two phenomena – (a) glottal stop reduction or deletion and (b) degemination across the pword boundary in binary compounds and particle verbs in German – by means of an acoustic-phonetic production study. The influencing factors under investigation are absolute word frequency and accentuation. In the generative framework of prosodic phonology, the pword (phonological or prosodic word, ω) is the domain that maps morphological entities onto phonological/prosodic entities (Löhken 1997; Nespor and Vogel 2007). Researchers agree that, in German, syllabification does not cross pword boundaries and therefore serves as a good indicator of these boundaries. For instance, the morphologically complex word gier+ig ‘greedy’ is syllabified as gie.rig, despite its internal morphological boundary, and thus is considered to constitute one phonological word: (gie.rig)ω. In the word lieb+lich ‘mellow’, however, resyllabification across the morphological boundary is blocked (*lie.blich), so that the string is analyzed as consisting of two separate phonological words: (lieb)ω(lich)ω (cf. Wiese 1996; Löhken 1997; Hall 1999). In cases like the latter, resyllabification is discouraged by devoicing and maybe aspiration of the final stop, both of which do not occur when the cluster [bl] is in the onset position. They therefore serve to mark the final edge of the first constituent and set apart the constituents of the morphologically complex word. Comparably, and turning to the edge-marking phenomena of interest in the present paper, the glottal stop serves to mark the initial bound-



Reduction and deletion of glottal stops and geminates  

 253

ary of a pword. As a result, its deletion makes it difficult to determine whether or not the morphological constituents constitute one pword each. In the word Zahn#arzt ‘dentist’, for instance, the second constituent starts with a vowel in an onsetless syllable, which leads to glottal stop insertion. As a consequence, syllabification coincides with the (morphological or phonological) word boundary: (Zahn.)ω(Ɂarzt)ω. If the glottal stop is deleted, however, the coda of the first syllable can and should be resyllabified as the onset of the second syllable, taking into account that onsetless syllables carrying word stress are dispreferred in German (cf. Löhken 1997). In this case, the original morphological and phonological word boundaries are blurred: Zah.narzt. Since the pword is assumed to be the domain for syllabification, as stated above, resyllabification is clearly problematic for an analysis of the string into two pwords. However, the assumption of resyllabification is theoretical in nature, derived from the principle of onset maximization and the dispreference of onsetless syllables with word stress in German. The question of whether or not the words are indeed realized and perceived as being produced with a resyllabified coda consonant is not pursued in the present study. Additional investigation of acoustic and/or articulatory data as well as perception data would be necessary to answer this question. So, while the question of resyllabification must remain unanswered, glottal stop deletion can quite uncontroversially be interpreted as a weakening of boundary strength. Through deletion of the glottal stop, the beginning of the second element in the compound is less pronounced compared to second elements that are initiated by a glottal stop. Thus, even if it is not possible to draw conclusions about the prosodic constituency of the words under investigation, glottal stop deletion is a clear indicator of a diminished signalling of the prosodic boundary. The same is true for cases of degemination. By geminates we refer to sequences of the same consonant across the morphological boundary of the compound; real geminates (within one morpheme) do not exist in New High German. Degemination is one of the phonological processes whose applicability is supposed to be constrained by a pword boundary in German. Thus, the inflectional ending of the second person singular (-st) is produced with only one s when attached to an s-final stem (e.g. du hass+st > du hasst ‘you hate’). In complex words which consist of two pwords, however, same consonants across the boundary are retained (e.g. Schrift.tum ‘literature’) (cf. Wiese 1996). Like glottal stop deletion, degemination can render resyllabification possible, namely in cases where the first constituent of the word has a short vowel, e.g. Stadt#tour ‘city tour’. Here, degemination leads to an ambisyllabic t (Sta.dt.our), which obscures the boundary between the elements. In words with a long vowel in the first constituent, no ambisyllabicity is at play (e.g. Brot.teig > Bro.teig ‘dough’). Since the phonological process of degemination is not supposed to take

254 

 Pia Bergmann

place across the pword boundary, its systematic occurrence is an indicator that boundary marking is weakened. Describing the glottal stop as being “inserted” or “deleted” implies that we are dealing with a categorical choice. This assumption is supported explicitly or implicitly by many phonologists in their accounts of the occurrence restrictions of the glottal stop in German, which include prosodic structure (domain boundaries and stress), speech rate and regional varieties (cf. Wiese 1996; Hall 1999; Alber 2001). Phonetic accounts of the phenomenon that were undertaken on the basis of various speech data show that the categorical view does not mirror the acoustic and articulatory reality of spoken language. Instead of clear deletions of the glottal stop, articulatory reduction strategies leading to glottalization are common. Moreover, these studies point out further factors that influence the occurrence of glottal stops and/or glottalization (cf. Pierrehumbert and Talkin 1992; Kohler 1994; Dilley, Shattuck-Hufnagel, and Ostendorf 1996; Kohler and Rehor 1996; Rodgers 1999). Word frequency belongs to these factors and is mentioned by Umeda (1978) for American English and Kohler (1994) for German. Systematic accounts of a larger database do not exist, however. Relating back to the glottal stop in its function as a boundary marker, these studies indicate (1) that gradient weakening of pword boundary marking seems to be a better way to describe the linguistic reality, and (2) that edge-marking by glottal stops/glottalization is subject to many influencing factors, including speech style, speech rate, segmental structure, prosodic structure and frequency. Gradient rather than categorical changes are evidenced not only for the reduction of glottal stops, but for other edge-marking phenomena too. A case in point is assimilation, which from a phonological point of view is a categorical process, whereas articulatory and acoustic studies have proved it to vary gradiently and systematically (cf. Hardcastle 1995; Nolan, Holst, and Kühnert 1996; Ellis and Hardcastle 2002; Stephenson and Harrington 2002; Stephenson 2003; Ernestus et al. 2006; Bergmann 2012a).1 We therefore hypothesize that the same is true for the reduction of geminates, although literature on the phenomenon is scarce (cf. Kohler 2001 for the only short account of phonetic realizations of geminates in German). In the present paper, we focus our attention on lexical frequency as one important influencing factor for the reduction of boundary-marking phenom-

1 It must be pointed out that Ellis and Hardcastle (2002) add another aspect to the discussion of whether assimilation should be considered graded or categorical: On the basis of their articulatory study they show that there are speaker-specific preferences for one or the other. Similar conclusions are suggested by the results of Mücke, Grice, and Kirst (2008) and Bergmann (2012a).



Reduction and deletion of glottal stops and geminates  

 255

ena. Until now, frequency has been widely acknowledged as having influence on speech production.2 Synchronic as well as diachronic studies exemplify its influence for word languages like Dutch and English (cf. Bush 2001; Jurafsky et al. 2001; Stephenson and Harrington 2002; Keune et al. 2005; Pluymaekers, Ernestus, and Baayen 2005; Ernestus et al. 2006; Phillips 2006; Stephenson 2003), as well as German (Jaeger and Hoole 2007, Bergmann 2008, 2012a, Kirst 2008, Mücke, Grice & Kirst 2008). The main theoretical background of these studies stems from usage-based approaches to language which assume that language experience has an impact on linguistic structure and linguistic knowledge. Accordingly, increased experience with a lexical item will lead to different treatment in language production and perception (and probably to a specific mental representation) (cf. Bybee 2001; Bybee and Hopper 2001; Pierrehumbert 2001). From a phonetic point of view, this assumption is already borne out in Lindblom’s (1990) so-called H&H-theory. According to Lindblom, phonetic variation can be explained by a model of the speaker who moves on a scale from hypo- to hyperarticulation (and back). The scale is defined by two opposing types of constraints: production constraints that strive for ease of articulation, and perception constraints that support and facilitate the hearer’s decoding of the message. Since high frequency in the model is supposed to facilitate lexical access, it weakens the perception constraints and allows the speaker to hypoarticulate. The expectation that high frequency leads to hypoarticulation is corroborated by the empirical findings mentioned above. One possible motivation for hypoarticulation may thus be articulatory economy in interplay with perceptual recoverability. Next to lexical frequency, we are interested in the impact of accentuation on the reduction of boundary marking. Accentuation refers to the sentence accent, more precisely the nuclear accent of the intonation phrase. Studies have shown that prosodic structure is phonetically encoded, i.e. edges and heads of prosodic domains are phonetically strengthened in order to serve speech stream segmentation and information retrieval (cf. Fougeron and Keating 1997; Keating et al. 2003; Cho, McQueen, and Cox 2007). We therefore hypothesize that accentuation will prevent the pword boundary from being strongly reduced or deleted. With our empirical production study, we aim to gain insights into the phonetic realization of the boundary-marking phenomena “glottal stop” and “geminates” as determined by lexical frequency and accentuation. We will present evidence for gradient reductions rather than categorical deletions of boundary marking. Section 2 will introduce materials and methods of the study. In section 3

2 This is even more true for research on speech perception (cf. e.g. Dahan, Magnuson, and Tane 2002 and references therein).

256 

 Pia Bergmann

the results will be presented. Section 4 will give a summary and a general discussion.

2 Materials and methods 2.1 Speech material The speech material of our investigation consists of 1456 test sentences for degemination, and 1288 test sentences for glottalization. 14 speakers (7 female, 7 male) of a non-dialectal northwestern German variety read the test sentences out loud in a soundproof booth at the University of Freiburg. Test sentences were presented in random order on a computer screen. The recording was done with Adobe Audition via a USB condenser microphone SAMSON C01U. The stimuli included high-frequency and low-frequency binary noun compounds (Comp) and particle verbs (Part) with different segmental contexts across the internal ω boundary, e.g. Zahn#arzt (‘dentist’), or Tat#ort ‘crime scene’. In addition to frequency and lexical category, we varied vowel quantity in the first constituent of the word, which was always monosyllabic. Care was taken that the test items resembled each other phonetically as much as possible. This was done to prevent any coarticulatory effects that could impede the interpretation of the results. Speech materials for glottalization include four different segmental contexts. The initial consonant of the second constituent is of course always a vowel; the final consonant of the first constituent is a vowel (V#V), an alveolar nasal (n#V), an alveolar stop (t#V) or an alveolar fricative (s#V). The stimuli consist of the following test items: Table 1: Test items for glottalization High-frequency V:

Low-frequency

V

V:

V

Stau#amt ‘traffic jam department’ Stau#ordnung ‘traffic jam regulations’

0

V#V Comp

Bau#amt ‘building department’ Bau#ordnung ‘building regulations’

0



Reduction and deletion of glottal stops and geminates  

High-frequency Part

 257

Low-frequency

V:

V

V:

V

zu#ordnen ‘to allocate’ zu#arbeiten ‘to do the preliminary work for so.’

0

bei#ordnen ‘to appoint’ bei#arbeiten ‘to mend/add to building materials’

0

n#V Comp

Ton#art ‘key (music)’ Zahn#arzt ‘dentist’

Sonn#abend ‘Saturday’

Bahn#arzt ‘med. doctor on a train’ Bahn#art ‘type of track’

Spann#art ‘type of stringing’ Bann#arzt ‘type of med. doctor in Third Reich’

Part

ein#ordnen ‘to arrange’ ein#atmen ‘to inhale’

an#ordnen ‘to arrange’ an#eignen ‘to acquire’

rein#ordnen ‘to arrange sth. into sth.’ rein#atmen ‘to breathe into sth.’

hin#ordnen ‘to arrange’ an#atmen ‘to breathe on so./sth.’

t#V Comp

Tat#ort ‘crime scene’ Haut#arzt ‘dermatologist’

Stadt#amt ‘municipal offices’ Stadt#ansicht ‘townscape’

Saat#ort ‘place of sowing’ Tat#art ‘modus operandi’

Stadt#ort ‘type of cantons in the Old Swiss Confederacy’ Blatt#art ‘leaf type’

Part

0

mit#arbeiten ‘to contribute’

0

mit#atmen ‘to breathe with so.’ mit#ackern ‘to slog away with so.’

s#V Comp

Maß#arbeit ‘made-to-measure, precision work’ Haus#arbeit ‘housework, homework’

Guss#eisen ‘cast iron’ Pass#amt ‘passport office’

Glas#arbeit ‘work of art made of glass’ Glas#amt ‘glass department’

Pass#arbeit ‘process of fitting sth. accurately’ Fass#arbeit ‘a barrel as a work of art’

Part

aus#atmen ‘to exhale’ aus#arbeiten ‘to perfect’

0

los#atmen ‘to start breathing’ los#eilen ‘to hurry away’

0

258 

 Pia Bergmann

We collected six different segmental contexts for the study of degemination: t#t, s#s,3 f#f, n#n, m#m and l#l. The test items are given in the following table: Table 2: Test items for degemination High-frequency V:

Low-frequency

V

V:

V

t#t Comp

Brot#teig ‘dough’ Haut#typ ‘skin type’

Stadt#tour ‘city tour’ Stadt#turm ‘city tower’

Draht#tunnel ‘wire tunnel’ Draht#typ ‘wire type’

Watt#tunnel ‘mudflat tunnel’ Blatt#typ ‘leaf type’

Part

0

mit#tanzen ‘to attend a dance’ mit#teilen ‘to inform’

0

mit#tanken ‘to tank up simultaneously’ mit#taumeln ‘to stagger along with s.o.’

s#s Comp

Groß#segel ‘main sail’

Fluss#säure ‘hydrofluoric acid’

Fuß#sehne ‘tendon in the foot’

Fluss#senke ‘river valley’

Part

aus#sondern ‘to eliminate sth.’

0

aus#sorgen ‘to become set for life’

0

f#f Comp

Brief#fenster ‘envelope window’

Schiff#fenster ‘ship window’

Hof#form ‘form of a (noble) court’

Stoff#form ‘shape of cloth’

Part

auf#fallen ‘to attract attention’

0

auf#fackeln ‘to be ablaze’

0

Bahn#nutzung ‘utilization of a track’

Spann#netz ‘safety net’

n#n Comp

Bahn#netz ‘railroad network’

Brenn#nessel ‘stinging nettle’

3 Due to final devoicing in German and voiced alveolar fricatives in onset position in the chosen variety, the segments differ with respect to voicing.



Reduction and deletion of glottal stops and geminates  

High-frequency Part

 259

Low-frequency

V:

V

V:

V

ein#nehmen ‘to take (medicine)’ ein#nisten ‘to nest’

hin#nehmen ‘to put up with’

ein#nagen ‘to gnaw at (wood)’ ein#nageln ‘to nail sth. in’

hin#neigen ‘to lean in towards so./sth.’

Team#messung ‘team measurement’

Schwimm#messung ‘measurement of swimming performance’

m#m Comp

Strom#messung ‘current measurement’

Schwimm#meister ‘pool attendant’

Part

0

um#melden ‘to register change of address’

um#merken ‘to change one’s mental note of sth.’

l#l Comp

Wahl#leiter ‘election supervisor’ Stuhl#lehne ‘back of a chair’

Schall#loch ‘sound hole’ Fall#linie ‘fall line, slope line’

Stahl#leiter ‘steel ladder’ Stahl#lenker ‘steel handlebar’

Stall#leiter ‘head of stable’ Woll#lappen ‘a woollen cloth’

Part

0

voll#laufen ‘to drink like a fish’

fehl#leiten ‘to misdirect’ fehl#leisten ‘to pay an incorrect amount of money’

voll#lachen ‘to laugh at so.’

All test items were embedded into carrier sentences which varied with respect to prosodic structure. Each test item occurred in accented position as well as in unaccented position. Sentence position of the test or control item was kept constant so as not to interfere with positional effects, especially final lengthening or glottalization at the IP boundary. In order to enhance the comprehensibility of the test sentences, we constructed question-answer pairs with the test item being part of the answer. Question and answer were presented visually on a computer screen. The following question-answer pairs exemplify the test sentences for the high-frequency particle verb ein#nehmen ‘to take (medicine)’:

260 

 Pia Bergmann

Table 3: Test sentences for the test item ein#nehmen accented

(Wie äußert sich deine Allergie gegen das Medikament?) (‘How does your allergy to the medicament manifest itself?’) Wenn ich es EINnehme, bekomme ich MAGENkrämpfe. when I it take get I stomach cramps ‘When I take it, I get stomach aches.’

unaccented

(Warum schluckst du ein Medikament, das du nicht gut verträgst?) (‘Why do you take a medicament you don’t tolerate?’) Wenn ich es NICHT einnehme, geht es mir NOCH schlechter. when I it not take goes it me even worse ‘When I don’t take it, I feel even worse.’

In addition to the test items, we recorded reference items containing the same segments across a syntactic boundary. These were embedded into questions, e.g. Willst du dir einen Wein nehmen? ‘Do you want to take a glass of wine?’ as a reference for the test item ein#nehmen.

2.2 Acoustic analysis and segmentation The acoustic analysis and segmentation (including labeling) of the data was done with Praat (cf. Boersma and Weenink 2009). Spectrograms were analyzed by visual inspection. Relevant labels and analysis categories for glottal stop/glottalization were: –– the occurrence of acoustic silence + a burst which was categorized as a glottal stop; –– the occurrence of irregular glottal vibration in the initial vowel of the second constituent which was categorized as glottalization; –– the duration of glottalization in the vowel. Relevant labels and analysis categories for degemination were: –– the absolute duration of the segment sequence; –– the relative duration of the segment sequence with respect to the duration of the first two syllables; –– in addition to the acoustic analysis it was determined whether there were one or two segments audible at the word boundary.



Reduction and deletion of glottal stops and geminates  

 261

All measurements and segmentations were labelled in Praat and automatically exported via a script.

2.3 Statistic analysis Statistic analysis was carried out with SPSS and R. The statistic analysis is based on 1684 items for glottal stop/glottalization and 2086 items for degemination. Statistical calculations were pooled across speakers.

2.4 Frequency counts Frequency checks were obtained for each lexical item by means of Google (German pages only: “Seiten auf Deutsch”). Words with more than 100,000 hits were categorized as high-frequency words, words with less than 1,000 hits as low-frequency words.

3 Results In the following sections we will present the results of the statistical analysis. We will start with the results for the deletion and reduction of the glottal stop (section 3.1), followed by the results for the reduction of geminates (section 3.2).

3.1 Glottal stop and glottalization The investigation of deletions and reductions of the glottal stop included two different types of analysis. We first combined the occurrence or non-occurrence of glottal stops and glottalization in four categories: “n/n” for neither glottal stop nor glottalization, “n/y” for no glottal stop but glottalization, “y/n” for glottal stop without glottalization into the vowel and “y/y” for both glottal stop and glottalization. The distribution of the four categories on the independent variables was then calculated and statistically evaluated. We then calculated a logistic regression for the dependent variable “glottal stop”.

262 

 Pia Bergmann

3.1.1 Combined realization categories The results for the combined variables first of all show that category “y/n”, i.e. glottal stop without glottalization, is virtually absent (cf. Table 4). Out of 1684 valid cases only 20 items (1.2%) are characterized by a glottal stop that is not followed by glottalization into the following vowel. Next, the category “n/n” yields 4.6%, followed by “n/y” with 30.9% and “y/y” with 63.3%. Table 4: Overall distribution of glottal stops and/or glottalization Glottal stop / Glottalization n/n overall distribution (n=1684)

n/y

y/n

y/y

absolute

78

520

20

1066

%

4.6%

30.9%

1.2%

63.3%

The distribution shows that the overall amount of clearly marked boundaries is very high when compared to the results of Kohler (1994), who found only 27% (= 946 cases) of the “y/y”-category and 37% (= 1285 cases) of the “n/y”-category. Complete glottal stop deletion occurs in 21% (= 718 cases) and glottal stop without glottalization in 15% of all the cases in his data. Since the data of the present study stem from read speech in a controlled laboratory setting, whereas Kohler’s results are based on semi-spontaneous speech, we hypothesize that this difference is at least partly due to stylistic differences (cf. Rodgers 1999 for significant effects of speaking style on glottal stops and glottalization). Although the overall number of non-deleted glottal stops in the present study is about twice as high as the number of deleted glottal stops (1086 items vs. 598 items), we still have clear evidence for a gradient reduction of the glottal stop via glottalization in the category “n/y”. Tokens of this category constitute cases of a weakened boundary marking as compared to those of the “y/y” and “y/n” categories. Deletion of word-internal boundary marking is represented by the “n/n” category.

3.1.1.1 Frequency The tokens of the “n/n” category are of special interest with respect to their distribution among high-frequency and low-frequency words. We expect complete deletion of the boundary to occur more often in high-frequency words than in low-frequency words. Likewise, boundary weakening is also expected to occur more often in high-frequency words than in low-frequency words. The opposite holds for the categories “y/n” and “y/y”. Moreover, we expect the low-frequency



Reduction and deletion of glottal stops and geminates  

 263

test items to be pronounced like the reference items across a syntactic boundary as opposed to the high-frequency items; this would indicate that the lexical constituents of the low-frequency nominal compounds or particle verbs are produced as two independent lexical items unlike those of the high-frequency words, which are produced as a single lexical item. The following table shows the resulting distribution of the test and reference items in the four categories. Additionally, standardized and corrected residuals are given in rows in order to make the distribution more easily interpretable. The Chi-square test proved the distribution to be significant (χ2(6) = 69.839, < .001). “Ref” refers to the reference items, which contain the same segments as the test items across a syntactic boundary. Table 5: Distribution of glottal stops/glottalization due to frequency Glottal stop/Glottalization Frequency HF

LF

Ref

n/n

n/y

y/n

y/y

absolute

55

224

11

325

expected

28.5

189.9

7.3

389.3

corrected residuals

6.4

3.7

1.7

-6.8

absolute

15

194

5

455

expected

31.0

206.6

7.9

423.5

corrected residuals

-3.8

-1.4

-1.4

3.3

absolute

8

102

4

286

expected

18.5

123.5

4.8

253.2

corrected residuals

-2.9

-2.7

-.4

3.9

Looking at category “n/n”, the corrected residuals indicate that high-frequency words are produced without any boundary marking significantly more often than expected (tokens are e.g. Sonn#abend ‘Saturday’, Haus#arbeit ‘housework, homework’, aus#arbeiten ‘to perfect sth.’, zu#arbeiten ‘to do the preliminary work for so.’). In the same category, low-frequency words yield a comparable significant deviation from the expected distribution as reference items. We can thus conclude that (1) high-frequency words are more prone to complete deletion of the boundary than low-frequency words and reference items, while (2) low-frequency words and reference items tend to retain their boundaries.

264 

 Pia Bergmann

With respect to boundary weakening (“n/y”), we find similar effects for high-frequency words and reference items. Although the low-frequency items deviate from the expected number of occurrences, this deviation is not significant. Deviations in the expected direction are observable in the “y/y” category. The “y/n” category constitutes a problem because tendencies are contrary to our expectations. The deviations do not yield a significant level, however, and 13 of the 20 cases belong to the n#V context. From an articulatory point of view, the alveolar nasal is ambiguous with respect to its favoring of glottal stops/glottalization. Being a sonorant, it has glottal vibration, thus favoring the deletion or reduction of the glottal stop. On the other hand, it involves a complete oral closure, which favors a simultaneous glottal closure because oral and glottal gestures are coordinated (cf. Kohler 1994). The high proportion of /n/ in the glottal stop category would hint at a higher emphasis of the latter feature, i.e. gestural coordination of oral and glottal closures. This line of reasoning will be taken up later in the discussion of the results on the influence of the segmental contexts on glottal stop deletion. First, we want to go into more detail on category “n/n”. Although the tendencies support our hypothesis, the occurrence of low-frequency words and reference items in this category calls for an explanation. In addition to the fact that overall percentages are very low (0.9% and 0.5% of all items) and could probably be dismissed as random “speech errors”, it is interesting that more than half of the low-frequency items under discussion have the constituent “-arbeit/en” as their second part (e.g. Glas#arbeit, bei#arbeiten). This is an unexpectedly high proportion, because only four words out of all 24 low-frequency words have “-arbeit/en” as their second constituent. This could mean that not only the frequency of the whole lexical item plays a role for deletion/reduction, but also the frequency of the single constituents. This were not controlled in the present study.

3.1.1.2 Accentuation In addition to frequency, the influence of accentuation on the occurrence of glottal stop deletion and reduction was tested (χ2(3) = 19.82, p < .001). We hypothesized that accentuation leads to boundary strengthening so that we encounter more accented items in category “y/y” than in the other categories. Within this category, we expect to find fewer unaccented items than accented items. Moreover, we expect more unaccented items in the weakened category “n/y” than in category “y/y”, and hypothesize that there are relatively more unaccented than accented items in the “n/y” category. Figure 1 presents the distribution of glottal stops/glottalization due to accentuation of the test items. The distribution including standardized and corrected residuals is given in Table 6.



Reduction and deletion of glottal stops and geminates  

 265

Figure 1: Distribution due to accentuation

Table 6: Distribution due to accentuation Glottal stop/glottalization Accentuation

n/n

n/y

y/n

y/y

unaccented (n = 643)

absolute expected corr. residuals

47 35.1 2.9

232 209.3 2.7

9 8 .5

355 390.6 -4.1

accented (n = 641)

absolute expected corr. residuals

23 34.9 -2.9

186 208.7 -2.7

7 8 -.5

425 389.4 4.1

The distribution shows that our hypotheses are only partly confirmed. It is true that more accented items accrue in category “y/y” than in all other categories. Additionally, we find significantly more accented items in this category than predicted by random distribution. Although there are consequently fewer unaccented items in category “y/y” than predicted, we still find that most of the unaccented items are produced with a glottal stop and glottalization (n = 355). The weakened boundary with a deleted glottal stop occurs in only 232 unaccented realizations. Thus, both accented and unaccented items most frequently realize strong boundary marking by a glottal stop or glottalization. Complete deletion of the boundary (“n/n”) occurs significantly more often in unaccented items than in

266 

 Pia Bergmann

accented items. Finally we can state that in relation to accentuation, unaccentuation leads to an increased weakening of the word-internal boundary. The high overall amount of strong boundary marking may be due to the high formality of the speech situation, as mentioned above.

3.1.2 Deletion of the glottal stop We carried out a logistic regression analysis for the independent variables “Frequency”, “Accentuation” and “Segmental context” due to the significant results Kohler (1994) found for this variable. The analysis yielded the following results: With respect to the occurrence or non-occurrence of the glottal stop we found significant main effects in the expected directions for frequency (χ2(6) = 33.38, p < .001) and accentuation (χ2(6) = 30.68, p < .001). Thus, we find significantly more deletions of the glottal stop in high-frequency words and unaccented items. We also found a significant main effect for segmental context (χ2(9) = 28.46, p < .001), as well as a significant interaction for accentuation and segmental context (χ2(3) = 13.64, p < .01). All other interactions proved to be statistically non-significant. Since results were significant for the segmental context and its interaction with accentuation, we will go into more detail for this independent variable. Due to their articulatory specifics, we expect the fewest occurrences of glottal stops between vowels, i.e. in category V#V. Most glottal stops are expected in the t#V context. These expectations are based on the findings of Kohler (1994), who establishes two relevant articulatory features for the influence on the occurrence/ non-occurrence of glottal stops: glottal vibration and oral closure. As mentioned above, the existence of an oral closure may favor the occurrence of a glottal stop because of gestural coordination. Glottal vibration, on the other hand, favors the deletion or reduction of the glottal stop due to the tendency for coarticulation between the two sonorant segments, which is a case of articulatory economy. Both the n#V context and the s#V context are ambiguous in this respect. The /n/ has an oral closure and glottal vibration, whereas the /s/ has no oral closure and no glottal vibration. In Kohler’s study, sonorants group with vowels, while fricatives group with stops under condition accented, and with sonorants and vowels under condition unaccented. The following figures illustrate the results of the present study. Figure 2 presents the percentage of produced glottal stops per segmental context, while Figure 3 shows the segmental context in its interaction with accentuation; the segmental context is plotted on the x-axis, and the two lines represent unaccented and accented items.



Reduction and deletion of glottal stops and geminates  

Figure 2: Main effect of segmental context on glottal stop realization

Figure 3: Interaction of segmental context and accentuation on glottal stop realization

 267

268 

 Pia Bergmann

The influence of the segmental context as a main effect is as expected for the contexts V#V and t#V (Figure 2). Moreover, both the n#V and the s#V contexts indeed render an equally high amount of glottal stops below the t#V context and above the V#V context. The ambiguous articulatory setting thus seems to lead to a midlevel realization of glottal stops compared to the unambiguous settings. Turning to the interaction between accentuation and segmental context (Figure 3), we can first observe that the tendencies of the main effect do not hold for the accented condition. Instead, accentuation seems to level out the segmental differences by increasing the amount of glottal stops to a nearly equal level in all contexts. Only in the unaccented condition do the differences between segments prevail. We can furthermore observe that accentuation has the strongest effect on n#V and V#V. In s#V words the difference between accented and unaccented condition is less conspicuous. Finally, and surprisingly, in t#V words there is no significant difference between the conditions, but unaccented items have slightly higher values than accented items. It thus seems that if the articulatory makeup of the segment strongly favors the occurrence of the glottal stop (= /t/), the influence of accentuation has no further effect on articulatory strengthening. But how can the strong influence of accentuation on n#V and V#V, compared to the weaker influence on s#V, be explained? The n#V and V#V contexts are characterized by glottal vibration, but differ with respect to oral closure. A possible explanation for the similar behavior of these segmental contexts is that the features “glottal vibration” and “coordination of oral and glottal closure” are affected by accentuation to varying degrees. Assuming that accentuation leads to an increase of articulatory effort, and taking into account that the maintenance of a (regular or irregular) vibration of the glottis means less effort than the interruption of a complete closure, we can conclude that the feature “glottal vibration” is more intimately intertwined with the occurrence of accentuation. On the other hand, it is difficult to see how the “coordination of oral and glottal closure” could be directly linked to articulatory effort. As a consequence, those contexts that are characterized by glottal vibration exhibit stronger effects of accentuation than the other contexts.

3.1.3 Summary Summarizing section 3.1.2, we can state that high-frequency words lead to significantly fewer glottal stops at the word boundary than low-frequency words. Deletion or weakening of boundary marking is thus significantly more common in high-frequency words than in low-frequency words. Accentuation and segmental contexts cannot be judged independently. Expected differences (t#V > s#V > n#V > V#V) are only valid in unaccented condition, but are leveled out in accented



Reduction and deletion of glottal stops and geminates  

 269

condition. Accentuation affects the segmental contexts differently (V#V, n#V > s#V > t#V), which was explained by articulatory specifics. The analysis of the combined realization of the word boundary with or without a glottal stop and glottalization showed, firstly, that boundary weakening by glottalization without a glottal stop is a common option for speakers, although not the most common realizational variant. Speakers most often produced a glottal stop and glottalization. Secondly, high-frequency words were shown to lead to significantly more boundary weakening and deletion than low-frequency words and word strings across a syntactic boundary. Thirdly, accented items exhibit less boundary weakening and deletion than the same items in unaccented position.

3.2 Degemination In addition to glottal stop deletion and glottalization, we analyzed degemination as a phenomenon of pword boundary weakening. We assume that the reduction of cluster duration leads to the segment becoming more similar to a singleton consonant so that its perception as two segments becomes less and less clear. We did acoustic measurements of cluster duration (section 3.2.1) and auditive judgements to determine whether one or two segments were audible (section 3.2.2).

3.2.1 Cluster durations An ANOVA was carried out for the independent variables “Frequency”, “Accentuation” and “Vowel quantity”. The dependent variable was the absolute duration of the cluster across the word boundary. We expected shorter durations in frequent items, in unaccented condition and after long vowels. There were significant main effects for all of the tested variables in the expected direction: frequency (F(1,1442) = 12.7975, p < .001), accentuation (F(1,1442) = 38.4018, p < .001), and vowel quantity (F(1,1442) = 51.9687, p < .001). It can be concluded that high-frequency leads to a significant shortening of the segment sequence across the boundary. The same is true for items in unaccented condition compared to accented condition, and items with phonological long vowel quantity compared to short vowel quantity. In addition to absolute durations we calculated relative durations of the cluster with respect to word duration. For each token, the duration of the first two syllables (which always span the pword boundary) was measured, and the ratio of the cluster to these two syllables was calculated. This was done to prove whether the durational reduction that was found in cases of high frequency, unaccentuation and long vowel quantity was really due to the durational reduction of the segments at the word boundary only, or rather to an overall shortening of the

270 

 Pia Bergmann

whole word. Results indicate that the latter seems to be the case for frequency, but not for accentuation and vowel quantity: Table 7: Relative cluster durations Independent Variable Frequency

Accentuation

Vowel quantity

Mean relative duration (%)

high (n = 700)

28.19

low (n = 756)

28.16

unacc (n = 728)

26.84

acc (n = 728)

29.51

long (n = 728)

25.95

short (n = 728)

30.4

t-test t(1430.605) < 1, p = .948

t(1454) = 5.622, p < .001

t(1397.168) = 9.538, p < .001

The mean relative durations of the clusters in high-frequency words and low-frequency words hardly differ at all. In relation to the significant absolute durational differences for frequency, one possible explanation for the different results is that frequency does not affect the duration of the cluster as an indicator for boundary strength, but rather the duration of the whole lexical item. We additionally carried out a linear regression analysis with “Frequency” as a metrically scaled independent variable. The results are significant (F(1, 2084) = 6.536; p < .05), but in an unexpected way: Relative durations increase slightly with frequency. The positive correlation is extremely weak, however (r = .056, r2 = .003), so that the significant result may be due to the high number of tokens. Still, the discrepancy between the results for absolute and relative cluster durations due to frequency indicates that it is the overall duration of the word that is reduced, and not the boundary spanning cluster. Turning to accentuation, the results are as expected: Absolute as well as relative durations are significantly longer in the accented condition than in the unaccented condition. The same is true for vowel quantity, which has the strongest effects on the relative duration of the cluster, with much higher proportions after a short vowel than after a long vowel. This confirms the expectation that segments are shorter after long vowels and can be explained as a compensatory lengthening/shortening with respect to the domain of the syllable. In sum, absolute segment durations across the pword boundary proved to vary significantly with frequency, accentuation and vowel quantity. For the unaccented items and the items with long vowel quantity we additionally found a significant reduction of relative durations. The hypothesis that the high frequency of words leads to durational reduction was confirmed. It was shown, however,



Reduction and deletion of glottal stops and geminates  

 271

that the durational reduction due to frequency refers not only to the cluster, but to the whole lexical item, too. When discussing the effect of high frequency on the weakening of the pword boundary, therefore, it must be taken into account that the shortening effects have to be subtracted from the overall shortening of the word. Furthermore, we found additional evidence that durational characteristics must be considered in interplay with other factors which relate to syllable structure: Consonantal reduction is conspicuously weakened after syllables with a short vowel.

3.2.2 Categorical evaluation In addition to the evaluation of gradient durational reductions, auditive judgements about the number of segments at the pword boundary were carried out and evaluated for their distribution in high-frequency and low-frequency words, as well as their distribution in accented and unaccented items. We expect a higher number of “one segment” ratings for high-frequency words and for items in unaccented position. Figure 4 demonstrates the distribution due to lexical frequency. The distribution including standardized and corrected residuals is given in Table 8.

Figure 4: “Number of segments” due to lexical frequency

272 

 Pia Bergmann

Table 8: “Number of segments” due to lexical frequency

HF

LF

One segment

Two segments

absolute

660

39

expected

650.8

48.2

stand. residuals

.4

-1.3

corr. residuals

1.9

1.9

absolute

689

61

expected

698.2

51.8

stand. residuals

-.3

1.3

corr. residuals

1.9

1.9

First of all, it can be observed that the overall auditive judgement of the segment sequence leads to many more “one segment” answers than “two segment” answers. This indicates that the boundary of the pword is weak and not clearly perceptible in general. It must be borne in mind, however, that acoustic cues for the separation of the two segments across the boundary vary greatly. While the segments in the t#t context may be clearly separated by a burst and aspiration, the other segments do not possess such perceivable features. Listeners may therefore more often tend towards a “one segment” answer. Nevertheless, the distribution due to frequency shows a trend in the expected direction (χ2(1) = 3.673, p = .055). That is, in high-frequency words the segment sequence is less often perceived as consisting of two separate segments.

Figure 5: “Number of segments” due to accentuation



Reduction and deletion of glottal stops and geminates  

 273

The distribution in accented and unaccented items is highly significant (χ2(1) = 24.535, p < .001). Items in unaccented condition are significantly more often perceived as being produced with one segment than predicted by random distribution (cf. Figure 5 and Table 9) Table 9: “Number of segments” due to accentuation Accentuation unacc

acc

One segment

Two segments

absolute

697

26

expected

673.1

49.9

stand. residuals

.9

-3.4

corr. residuals

5.0

-5.0

absolute

652

74

expected

675.9

50.1

stand. residuals

-.9

3.4

corr. residuals

-5.0

5.0

In summary, we can state that the segment sequence across the pword boundary is considerably more often perceived as one rather than two segments in general. Sequences in high-frequency words show a trend towards being perceived as one segment rather than two; in unaccented condition this difference is significant.

4 Summary and discussion This paper investigated the influence of lexical frequency and accentuation on pword boundary marking in complex words in German. The boundary marking phenomena of interest were deletion and/or reduction of the glottal stop and the reduction of geminates. While accentuation is nowadays widely considered to have an influence on the strength of prosodic edge-marking, systematic studies of frequency effects have only recently begun to increase. With respect to the production of glottal stops and geminates in German, systematic accounts of frequency effects do not exist. The results of our study show that lexical frequency is an important factor for the regulation of boundary strength. We found highly significant effects of frequency on the combined occurrence of glottal stops and/or glottalization (section 3.1.1), as well as on the deletion of glottal stops (section 3.1.2). In high-frequency items, significantly more boundary weakening can be observed than in low-frequency items or across a syntactic boundary. Whereas the relation of phonetic

274 

 Pia Bergmann

reduction to boundary weakening is quite clear for the phenomenon of glottal stops/glottalization, things are more complicated for the phenomenon of degemination. Our results indicate that the durational reduction of the boundary-spanning cluster must be viewed with regard to the overall durational reductions of the word in question. On the basis of our analysis, we can only hypothesize that high frequency leads to a gradient durational reduction of the word, but not of the cluster specifically. This was concluded from the discrepancy between significant reductions of absolute durations compared to stable relative durations of the cluster in the word (section 3.2). In view of this relative durational stability, boundary marking does not seem to be affected by frequency. The auditive categorical analysis, however, yielded a trend for clusters in high-frequency items to be perceived as “one segment” more often than “two segments”. In the case of degemination, it thus seems that categorical characteristics are a better indicator of frequency effects on boundary marking than gradient characteristics, where we could not detect direct effects of frequency on pword boundary strength. It should be noted, however, that the analysis of degemination can still be refined: Auditive analysis of the number of segments should be supplemented by an acoustic analysis, and the future evaluation should take into account possible differences between segmental contexts that were evaluated together in the present study. Our hypothesis with respect to the influence of accentuation on boundary strength were confirmed for both degemination and glottal stops/glottalization. It was shown for glottal stops/glottalization, however, that accentuation interacts with the segmental context in a significant way. Accentuation differences were not present in the stop#vowel context, weaker in the fricative#vowel context and strongest in the sonorant/vowel#vowel context. We explained this finding by examining the articulatory characteristics of the segments. A recent study by Mitterer, Yoneyama, and Ernestus (2008) highlights the fact that – besides other factors – the auditory salience of a phonetic contrast plays an important role in the perception of reduced segments. Kohler (1976) hypothesizes that auditively salient acoustic parameters are protected against articulatory reduction. There is thus evidence that perceptual constraints should not be neglected when dealing with articulatory reductions. Although the present study focusses on the production of the pword boundary, we assume that boundary marking is “done” for the benefit of the listener. The fact that the different segmental contexts are affected by accentuation in such a different manner should therefore at least lead to the consideration that perceptual constraints might play a role. From this point of view, the stop#vowel context would not need further enhancement under accentuation because – due to the burst and the articulatorily favored subsequent glottal stop – the boundary would be easily detectable. In the sonorant/vow-



Reduction and deletion of glottal stops and geminates  

 275

el#vowel contexts, the auditive cues are not as strong, and thus in the strengthening condition, the glottal stop needs to be introduced despite being articulatorily disfavored. However, this conclusion verges on banality if it is not underscored by (1) further research into acoustic details of the production (we did not analyze the actual production of the final segment of the word’s first constituent), and (2) studies of how the segment sequences and whole words are actually perceived. Next to frequency, accentuation and the segmental context, syllable structure was found to have an influence on the production of the pword boundary. This factor was only investigated for the geminates. Durational variation of the segment cluster varies significantly with the phonological vowel quantity of the word’s first constituent. We interpret this effect as a compensatory shortening or lengthening with respect to the domain of the syllable. All in all, we conclude from the presented results that the observed boundary reduction is affected by several factors belonging to linguistic structure (prosodic and segmental structure) as well as language use (frequency, coarticulation). We were able to show that boundary marking is a graded phenomenon. This is at least true for the glottal stop, which, from a phonological point of view, is described as being categorically inserted or deleted. The results of our production study suggest, however, that it is a gradient phenomenon, with a high amount of in-between cases, namely glottalization. Although nothing can be said about the way the investigated items are mentally represented, from a theoretical point of view, our study supports the assumption of usage-based approaches that language use may influence linguistic structure: We find high-frequency lexical items leading to more and stronger boundary reduction than low-frequency items. This systematic link between frequency of occurrence and boundary strength indicates that elements that are traditionally supposed to have the same prosodic structure (i.e. all compounds and all particle verbs) are not treated as such by speakers. Instead, speakers modulate prosodic structure in a gradient fashion. For the classification of New High German as a word language, this does not pose a severe problem. Even if pword boundaries are weakened, and even if they are deleted under the influence of (non-)accentuation and frequency, it must be borne in mind that weakening is only one side of the coin, the other one being strengthening of the boundary. Rather than constituting a problem for German’s classification as a word language, the systematic use of boundary strength to signal word structure exemplifies that the pword is a prosodic constituent of vital importance in the investigated variety of German. As expected for a word language as compared to a syllable language, morphological structure is highlighted by phonological means. What the results of the present study show, is that it is not merely morphological constituency that exerts an influence on phonological and phonetic word structure. Clearly, additional struc-

276 

 Pia Bergmann

tural and usage-based factors have to be taken into account. The study thereby contributes to an in-depth phonetic and phonological investigation of the typology of syllable and word languages and opens up promising perspectives for future research in the field.

References Alber, Birgit (2001): Regional variation and edges: Glottal stop epenthesis and dissimilation in standard and southern varieties of German. Zeitschrift für Sprachwissenschaft 20/1: 3–41. Auer, Peter (1994): Einige Argumente gegen die Silbe als universale prosodische Hauptkategorie. In: Karl Heinz Ramers, Heinz Vater and Henning Wode (eds.), Universale phonologische Strukturen und Prozesse, 55–78. Tübingen: Niemeyer. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Bergmann, Pia (2008): Assimilation within complex words in German. Poster presented at the Consonant Clusters and Structural Complexity Workshop, IPS Munich, July 31– August 2. Bergmann, Pia (2012a): Articulatory reduction and assimilation of n#g sequences in complex words in German. In: Philip Hoole, Lasse Bombien, Marianne Pouplier, Christine Mooshammer and Barbara Kühnert (eds.), Consonant Clusters and Structural Complexity, 311–344. (Interface Explorations 26.) Berlin/Boston: Mouton de Gruyter. Bergmann, Pia (2012b): Die phonetische Realisierung prosodischer Grenzen in der Spontansprache. Oral presentation at the 8th Conference for Phonetics & Phonology (P&P), Jena, October 12– 13. Boersma, Paul and David Weenink (2009): Praat: doing phonetics by computer (Version 5.1.05) [Computer program]. Available at http://www.praat.org. Bush, Nathan (2001): Frequency effects and word-boundary palatalization in English. In: Joan Bybee and Paul Hopper (eds.), Frequency and the Emergence of Linguistic Structure, 255–280. (Typological Studies in Language 45.) Amsterdam: John Benjamins. Bybee, Joan (2001): Phonology and Language Use. (Cambridge Studies in Linguistics 94.) Cambridge: Cambridge University Press. Bybee, Joan and Paul Hopper (eds.) (2001): Frequency and the Emergence of Linguistic Structure. (Typological Studies in Language 45.) Amsterdam: John Benjamins. Cho, Taehong, James M. McQueen and Ethan A. Cox (2007): Prosodically driven phonetic detail in speech processing: The case of domain-inital strengthening in English. Journal of Phonetics 35/2: 210–243. Dahan, Delphine, James S. Magnuson and Michael K. Tane (2002): Time course of frequency effects in spoken-word recognition: Evidence from eye movements. Cognitive Psychology 42/4: 317–367. Dilley, Laura, Stefanie Shattuck-Hufnagel and Mari Ostendorf (1996): Glottalization of word-initial vowels as a function of prosodic structure. Journal of Phonetics 24/4: 423–444.



Reduction and deletion of glottal stops and geminates  

 277

Ellis, Lucy and William J. Hardcastle (2002): Categorical and gradient properties of assimilation in alveolar to velar sequences: evidence from EPG and EMA data. Journal of Phonetics 30/3: 373–396. Ernestus, Mirjam, Mybeth Lahey, Femke Verhees and R. Harald Baayen (2006): Lexical frequency and voice assimilation. Journal of the Acoustic Society of America 120/2: 1040–1051. Fougeron, Cécile and Patricia A. Keating (1997): Articulatory strengthening at edges of prosodic domains. Journal of the Acoustical Society of America 101/6: 3728–3740. Hall, Tracy Alan (1999): The phonological word: A review. In: Tracy Alan Hall and Ursula Kleinhenz (eds.), Studies on the Phonological Word, 1–22. (Current Issues in Linguistic Theory 174.) Amsterdam/Philadelphia: John Benjamins. Hardcastle, W. J. (1995): Assimilations of alveolar stops and nasals in connected speech. In: Jack Windsor Lewis (ed.), Studies in General and English Phonetics: Essays in Honour of Professor J. D. O’Connor, 49–67. London/New York: Routledge. Jaeger, Marion and Phil Hoole (2007): Articulatory features influencing regressive place assimilation in German. In: Jürgen Trouvain and William J. Barry (eds.), Proceedings of the 16th International Congress of Phonetic Sciences (ICPhS XVI), Saarbrücken, Germany, 6–10 August 2007, 581–584. Saarbrücken: Universität des Saarlandes. Jurafsky, Daniel, Alan Bell, Michelle Gregory and William D. Raymond (2001): Probabilistic relations between words: Evidence from reduction in lexical production. In: Joan Bybee and Paul Hopper (eds.), Frequency and the Emergence of Linguistic Structure, 229–254. (Typological Studies in Language 45.) Amsterdam: John Benjamins. Keating, Patricia, Taehong Cho, Cécile Fougeron and Chai-Shune Hsu (2003): Domain-initial strengthening in four languages. In: John Local, Richard Ogden and Rosalind Temple (eds.), Phonetic Interpretations, 145–163. (Papers in Laboratory Phonology 6.) Cambridge: Cambridge University Press. Keune, Karen, Mirjam Ernestus, Roeland van Hout and R. Harald Baayen (2005): Variation in Dutch: From written MOGELIJK to spoken MOK. Corpus Linguistics and Linguistic Inquiry 1–2: 183–223. Kirst, Raphaela (2008): Einfluss von Wortfrequenz und Informationsstruktur auf die Assimilation in alveolar-velaren Sequenzen – eine elektropalatographische Studie. Unpublished Master Thesis, University of Cologne. Kohler, Klaus J. (1976): Die Instabilität wortfinaler Alveolarplosive im Deutschen: eine elektropalatographische Untersuchung. Phonetica 33: 1–30. Kohler, Klaus J. (1994): Glottal stops and glottalization in German. Data and theory of connected speech processes. Phonetica 51: 38–51. Kohler, Klaus J. (2001): Variability of opening and closing gestures in speech communication. In: Klaus J. Kohler (ed.), Sound Patterns in German Read and Spontaneous Speech: Symbolic Structures and Gestural Dynamics, 33–96. (AIPUK 35.) Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Universität Kiel. Kohler, Klaus J. and Claudia Rehor (1996): Glottalization across word and syllable boundaries. In: Klaus J. Kohler, Claudia Rehor and Adrian P. Simpson (eds.), Sound Patterns in Spontaneous Speech, 195–206. (AIPUK 30.) Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Universität Kiel. Lindblom, Björn (1990): Explaining phonetic variation: A sketch of the H&H theory. In: William J. Hardcastle and Alain Marchal (eds.), Speech Production and Speech Modelling, 403–439. Dordrecht: Kluwer.

278 

 Pia Bergmann

Löhken, Sylvia C. (1997): Deutsche Wortprosodie. Abschwächungs- und Tilgungsvorgänge. (Studien zur deutschen Grammatik 56.) Tübingen: Stauffenburg. Mitterer, Holger, Kiyoko Yoneyama and Mirjam Ernestus (2008): How we hear what is hardly there: Mechanisms underlying compensation for /t/-reduction in speech comprehension. Journal of Memory and Language 59/1: 133–152. Mücke, Doris, Martine Grice and Raphaela Kirst (2008): Prosodic and lexical effects on German place assimilation. In: Rudolph Sock, Susanne Fuchs and Yves Laprie (eds.), Proceedings of the 8th International Seminar on Speech Production, 8–12 December 2008, Strasbourg, France, 225–228. Available at http://issp2008.loria.fr/proceedings.html. Nespor, Marina and Irene Vogel (2007): Prosodic Phonology. (Studies in Generative Grammar 28.) Berlin: Mouton de Gruyter. Nolan, Francis, Tara Holst and Barbara Kühnert (1996): Modelling [s] to [ʃ] accomodation in English. Journal of Phonetics 24/1: 113–137. Nübling, Damaris and Renata Szczepaniak (2008): On the way from morphology to phonology. German linking elements and the role of the phonological word. Morphology 18: 1–25. Phillips, Betty S. (2006): Word Frequency and Lexical Diffusion. Basingstoke: Palgrave Macmillan. Pierrehumbert, Janet B. and David Talkin (1992): Lenition of /h/ and glottal stop. In: Gerard J. Docherty and D. Robert Ladd (eds.), Gesture, Segment, Prosody, 90–127. (Papers in Laboratory Phonology 2.) Cambridge: Cambridge University Press. Pierrehumbert, Janet B. (2001): Exemplar dynamics: Word frequency, lenition and contrast. In: Joan Bybee and Paul Hopper (eds.), Frequency and the Emergence of Linguistic Structure, 137–157. (Typological Studies in Language 45.) Amsterdam: John Benjamins. Pluymaekers, Mark, Mirjam Ernestus and R. Harald Baayen (2005): Lexical frequency and acoustic reduction in spoken Dutch. Journal of the Acoustic Society of America 118/4: 2561–2569. Rodgers, Jonathan (1999): Three influences on glottalization in read and spontaneous German speech. In: Klaus J. Kohler (ed.), Phrase-Level Phonetics and Phonology of German, 173–280. (AIPUK 34.) Kiel: Institut für Phonetik und digitale Sprachverarbeitung, Universität Kiel. Stephenson, Lisa and Jonathan Harrington (2002): Assimilation of place of articulation: Evidence from English and Japanese. In: Proceedings of the 9th Australian International Conference on Speech Science & Technology, Melbourne, December 2002, 592–597. Melbourne: Australian Speech Science and Technology Association. Stephenson, Lisa (2003): An EPG study of repetition and lexical frequency effects in alveolar to velar assimilation. In: Proceedings of the 15th International Conference of Phonetic Sciences, Barcelona, Spain, August 3–9, 2003, 1891–1894. Barcelona: Universitat Autònoma de Barcelona. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Umeda, Noriko (1978): Occurrence of glottal stops in fluent speech. Journal of the Acoustic Society of America 64/1: 88–94. Wiese, Richard (1996): The Phonology of German. (The Phonology of the World’s Languages.) Oxford: Clarendon Press.

Peter Gilles (University of Luxembourg)

Phonological domains in Luxembourgish and their relevance for the phonological system* Abstract: Situated in the recent discussion on syllable languages and word languages, the relevance and impact of phonological domains for the phonology of Luxembourgish will be discussed. Analyzing various aspects of the phonological organization of Luxembourgish, this paper addresses the  question of whether there is one main prosodic domain, i.e. either the phonological word or the syllable, that shapes the overall phonological structure of the language. The features presented here concern the distribution of schwa, syllable complexity, the behavior of unstressed syllables, the role of the trochaic foot and the impact of language contact with French. It turns out that Luxembourgish cannot be classified entirely as a syllable language or a word language. Instead, features originating from both types seem to be mixed in an intricate way. Furthermore, frequent sandhi phenomena like the n-rule and voicing assimilation show that both phonological domains are indeed necessary to describe the structures appropriately.

1 Introduction Inspired by recent work on the phonological system of Luxembourgish by Nübling (2005) and Szczepaniak (2010), in this paper I shall address the relevance of two phonological domains, i.e. the syllable and the phonological word, for the phonological system of Luxembourgish. Following the proposal initiated by Donegan and Stampe (1983) and elaborated by Auer (1993, 1994, 2001) and Szczepaniak (2007), languages can be classified typologically as syllable or word languages according to the relevance of the prosodic domains of the syllable or the phonological word, respectively. Specifically, a language “favors” either the syllable or the phonological word in sound change or synchronic phonological rules, processes and alternations, i.e. the phonology of a given language is organized around one main prosodic domain (see Table 1 in Caro and Szczepaniak,

* I thank Cristian Kollmann, Fränz Conrad and the editors for valuable comments on an earlier draft of this article.

280 

 Peter Gilles

this volume for a systematization of the core parameters). These typological parameters may manifest themselves in various ways in phonetics and phonology. Table 1, which is based on Auer (1993), gives examples of how these parameters can be implemented in word languages and syllables languages, respectively. This list is far from complete and has to be regarded rather as an open list, illustrating the typological distinction. Table 1: Central characteristics of word languages and syllable languages Word languages

Syllable languages

complex syllable structure; sonority hierarchy partially not respected

simple syllable structure (preferably CV); sonority hierarchy respected

differences between accented and unaccented syllables

only minor differences between accented and unaccented syllables

central vowel phonemes possible

no central vowels

phonetically strong word accent

phonetically weak word accent

reduction of non-accented syllables

no accent-dependent reduction

vowel deletion leading to complex clusters

vowel epenthesis to achieve CV structure

resyllabification only within a phonological word

resyllabification across word boundaries

external sandhi ≠ internal sandhi

external sandhi = internal sandhi

frequent assimilations

few assimilations

A prototypical syllable language fulfills as many of the criteria on the right side of the table as possible. Italian, for example, is characterized by a simple syllable structure (alongside phonological processes aiming at the dominance of CV syllables), geminates and a phonetically rather weak word accent. German or Danish, on the other hand, represent word languages, because they have, inter alia, developed a complex syllable structure and a phonetically strong word accent. It should be emphasized that this typological distinction is organized as a continuum with ideal(ized) endpoints, allowing several intermediate steps. Conceptualized as a direct opposition, a feature which strengthens the phonological domain of the word may weaken the domain of the syllable (and vice-versa). How a variety can be located on this continuum is shown by Caro Reina (this volume) for various Catalan varieties. Here, the phonology of Luxembourgish is tested against some of these criteria in order to assign the language a place on the continuum of the syllable language/word language distinction. The analysis of the structural distance between Luxembourgish and the closely-related Standard German as a prototypical word language will also be of particular interest.



Phonological domains in Luxembourgish  

 281

2 Luxembourgish Luxembourgish (Lëtzebuergesch) is a small West Germanic language (approximately 300,000 to 400,000 speakers) of Central Franconian origin and today the national language of the Grand Duchy of Luxembourg.1 It is used mainly as a spoken language for a wide range of situations, both formal and informal. In this multilingual society Standard German and Standard French also form part of a complex triglossic situation with a high degree of language contact. For Luxembourgish, contact with German and French has resulted in numerous lexical borrowings with various degrees of phonological integration. Language standardization is at a medium level for Luxembourgish: Language codices such as dictionaries and grammars exist, but Luxembourgish is still not taught in a serious way in the educational system. As most written domains are occupied by French or German, the status of Luxembourgish as a written language is still weak. In contrast with fully-fledged standard languages like French or German, Luxembourgish phonology has therefore been less influenced by written language to date. This has led to a situation where considerable regional and individual variation is preserved. Nevertheless, the variety spoken in the geographical center of the country (Zentralluxemburgisch) has been functioning more and more as an emerging standard variety. The data presented in the following also comes from this variety. In this scenario Luxembourgish is an ideal candidate to test how a phonological system develops when it is not subject to strong standardization. With regard to the dichotomy of word and syllable languages, Nübling (2005) and Szczepaniak (2010) discuss several phonological features of German and Luxembourgish in a contrastive way. In her analysis of several sound changes, Szczepaniak identifies both syllable-language- and word-language-related traits in Luxembourgish. Among the syllable-related features are schwa epenthesis, umlaut, resyllabification, the n-rule, voice assimilation and the presence of schwa in stressed syllables (in addition to its presence in unstressed syllables). On the other hand, the author also presents several features relating to word languages (e.g. final obstruent devoicing, development of ambisyllabic consonants, complete apocope of final schwa, decrease of schwa epenthesis, dissimilation of [st], [sp] > [ʃt], [ʃp] to indicate the beginning of a phonological word). These findings suggest that Luxembourgish exhibits typological characteristics of both language types and it is consequently classified as “phonologically mixed”.

1 For an overview of the phonetics/phonology of Luxembourgish see Keller (1961), Newton (1990), Gilles (1999, 2009) and Gilles and Trouvain (2013).

282 

 Peter Gilles

The present study will partly rediscuss Szczepaniak’s findings, but will also bring further data and phonological argumentation into the debate. In order to answer the far-reaching question of the typological status of Luxembourgish, it is crucial to take the whole phonological system into account. However, this study will only concentrate on aspects of syllable and word structure, word stress and external sandhi. This ensures that the most pertinent aspects of the phonological system are dealt with. As we will see, the Luxembourgish data represent an interesting test case in the discussion of how preferences for a phonological domain such as the syllable or the phonological word shape the phonology of a language. Moreover, the discussion will also reveal that an unambiguous attribution of certain phonological processes to either the syllable or the word “pole” of the continuum is not always straightforward and requires careful consideration of several facets. The following sections will discuss features of Luxembourgish phonology that clearly relate to the syllable (section 3) and to the phonological word (section 4). Section 5 presents two cases of external sandhi that pose certain problems, as they point to the relevance of the two phonological domains at the same time. In the final section 6 the results will be summarized.

3 (Potential) Syllable language traits The following critical discussion will be devoted to the distribution of schwa and schwa epenthesis as potential traits to classify Luxembourgish as a syllable language.

3.1 Distribution of schwa The distribution of schwa in Luxembourgish is not restricted to unaccented syllables, but can also occur in accented syllables (see also Szczepaniak 2010). In other words, it is not the domain of the phonological word that is relevant for the distribution of schwa, but rather the syllable. Following van Oostendorp (1998), this kind of schwa can be labeled “stable schwa” as it forms part of the underlying representation of the word and cannot be regarded as the result of vowel reduction or epenthetic processes that facilitate syllabification. As with all central vowels, its phonetic realization varies according to contextual influences. In most cases, it can be perceived as a slightly fronted or rounded central vowel, but the articulatory difference to a schwa in unaccented syllables remains hardly noticeable. In Luxembourgish orthography, this stressed schwa is written with (see examples in 1).



Phonological domains in Luxembourgish  

 283

(1) Schwa in stressed syllables Zëmmer [ˈʦəmɐ] ‘room’ sëtz-en [ˈzəʦən] ‘sit-inf’ Lëtzebuerg [ˈləʦəbuəɕ] ‘Luxembourg’ ënnen [ˈənən] ‘below’ schëdd-en [ˈʃədən] ‘pour-inf’ këddel-en [ˈkədələn] ‘tickle-inf’

If Luxembourgish belonged to the word language type, the phonology would strive to maximize the contrast between the vowel inventory of accented and unaccented syllables by avoiding schwa (as the most important vowel of unstressed syllables) in stressed syllables. Since this is not the case with regard to schwa, one could argue that this feature characterizes Luxembourgish as belonging more to the syllable language type. Despite the fact that schwa can occur in stressed and unstressed syllables, one must keep in mind that the two vowel inventories still remain rather different: The vowel inventory for unstressed syllables is much more restricted compared to the stressed syllables2 – which is a feature of a word language. Thus, the described distribution of schwa is probably only a weak indicator towards a syllable language. This claim finds further support if the sound history of the “stressed” schwa is taken into account. Historically, this schwa originates from West Germanic /i/ or (subsequently unrounded) /y/, which were affected by a general and regular sound change of vowel lowering ([i] > [e] > [ə]). In this general and regular sound change, not only polysyllabic words like the ones in (1) but also monosyllabic words like Fësch ‘fish’, wëll ‘wild’ or Kënn ‘chin’ were affected (see Keller 1961: 257; Sturm 1988: 69‒77). It can be concluded that the evolution of this schwa and word stress are not related. That this schwa is found in stressed syllables seems thus a mere by-product of an unrelated segmental sound change.3 This somewhat special status of schwa is further supported by the phonological integration of French loans, which are generally very frequent in Luxembourgish. The words in (2) demonstrate that after phonological integration a

2 This only holds true for the Germanic part of the Luxembourgish lexicon. The numerous loan words from French allow a wide range of vowels in unstressed syllables even after phonological integration (e.g. Premier [ˈpʀəmjeː] ‘prime minister’, Bopa [ˈboːpaː] ‘grand-father’ (< fr. bon-papa), Poteau [ˈpɔtoː] ‘pillar’, Bijou [ˈbiʒuː] ‘jewellery’). The integration of vowels from French into Luxembourgish are only partly related to the prosodic dichotomy and will thus be excluded from the discussion. 3 To complicate this aspect even further, in segmental-phonological terms this schwa can be regarded as an allophone of the phoneme /e/: While [e] is realized only before (historically) velar consonants, [ə] is realized in all other contexts (see Gilles and Trouvain 2013).

284 

 Peter Gilles

former full vowel [e] is reduced to schwa and this syllable also carries the word accent. (2) Schwa in stressed syllables French Luxembourgish télé [teˈle] > Tëlee [ˈthəleː] ‘TV’ vélo [veˈlo] > Vëlo [ˈvəloː] ‘bike’ mélasse [meˈlas] > Mëlass [ˈməlas] ‘molasses’ béton [beˈtõ] > Bëtong [ˈbətoŋ] ‘concrete’ pétrole [peˈtʀol] > Pëtrol [ˈphətʀol] ‘petrol’

In the course of the phonological integration of these words, the word accent jumps to the first syllable to comply with the general tendency for trochaic word accent (Gilles 2010) and at the same time the former full vowel of the stressed syllable undergoes vowel reduction to schwa. These words thus again seem to demonstrate that the distribution of schwa is not sensitive to stress, which then can be regarded as a characteristic of a syllable language. The next feature to be discussed concerns the schwa vowel in unaccented syllables (see also Krier 2008: 104; Conrad 2010). Contrary to e.g. Standard German, unstressed schwa in Luxembourgish shows a strong resistance against syncope in word-final unaccented syllables (3). Even in faster speaking styles, schwa is phonetically realized in word-final syllables (in most cases morphological endings like -en). Reducing these syllables to syllabic sonorants (often with subsequent assimilation of place of articulation), quite frequent in Standard German, is not possible at all in Luxembourgish. (3) Schwa resisting syncope in final unaccented syllables

Ënn-en bak-en komm-en gleew-en Fuedem Kärel

[ˈənən] *[ˈən̩ː] ‘onion-pl’ [ˈbaːkən] *[ˈbaːkŋ̩] ‘bake-inf’ [ˈkomən] *[ˈkom̩] ‘come-inf’ [ˈgleːvən] *[ˈgleːvn̩] ‘believe-inf’ [ˈfuədəm] *[ˈfuədm̩] ‘wire’ [ˈkɛːʀəl] *[ˈkɛːʀl̩] ‘guy’

According to Nübling (2005), the resistance of syncope can be regarded as a syllable language feature, as every syllable in a word retains its own vocalic nucleus. It remains to be shown, however, whether or not this resistance is merely a by-product of phonotactic constraints. Moreover, as will be shown in section 5, not all schwas are resistant against syncope.



Phonological domains in Luxembourgish  

 285

3.2 Schwa epenthesis The process of schwa epenthesis can be regarded as a classic feature of syllable languages. Hall (2011: 1576) notes that “in most cases, the function of vowel epenthesis is to repair an input that does not meet a language’s structural requirements”. In Luxembourgish, schwa epenthesis takes place to break up coda clusters consisting of a sonorant and an obstruent or certain sonorant-sonorant combinations. The process originated in certain Old High German (OHG) dialects, but in Luxembourgish epenthesis has been active for a longer period and has developed a greater number of instances compared to German dialects (Nübling 2005: 151). In words like Kallef ‘calf’, originating from a monosyllabic form *kalf (< Germanic root *kalƀaz-), an intrusive schwa-like vowel may have arisen due to a mistiming of articulatory gestures of the sonorant l and the following obstruent f (Auer 1997). By turning a formerly monosyllabic word into a disyllabic one, the syllable structure is optimized according to the preferences of a syllable language: A former CVCC sequence changes into a more preferable CVCVC sequence (probably with an ambisyllabic consonant). The list in (4) contains the most common words (in orthographical form) of present-day Luxembourgish that still show epenthesis (see also Krier 2008; Szczepaniak 2010; Conrad 2010). (4) Schwa epenthesis in present-day Luxembourgish

/lf/ > [ləf] Kallef ‘calf’, hallef ‘half’, Hëllef ‘help’, eelef ‘eleven’, zwielef ‘twelve’, Wollef ‘wolf’, Folleg ‘episode’, Sallef ‘salve’ /lɕ/ > [ləɕ] Mëllech ‘milk’, wellech ‘whichever’, sëllech ‘several’ /lk/ > [lək] Wollek ‘cloud’, Kallek ‘lime’, Vollek ‘people’ /lm/ > [ləm] Hallem ‘culm’ /nf/ > [nəf] fënnef ‘five’ /rl/ > [ʀəl] Kärel ‘guy’, Pärel ‘pearl’, Märel ‘blackbird’, Charel ‘Charles’

Epenthesis takes place in coda clusters consisting of a sonorant and either an obstruent or a second sonorant. As in various Central German dialects exhibiting epenthesis, the process is blocked if the second member of the coda is a sibilant. Hence, the vast majority of final clusters ‑ls, ‑lʃ, ‑ns, ‑mʃ, ‑nʃ, -rʃ have not developed epenthesis. Sibilant obstruents are regarded as extrasyllabic consonants (in this position they violate the sonority hierarchy) and thus cannot trigger epenthesis (Auer 1997: 61). The historical coda cluster -rn also represents a dispreferred coda structure and could have been affected by epenthesis as well. However, with the exception of gären (< OHG gern), which exists alongside gär, the solution applied in the relevant words is instead the deletion of the final nasal, and after this cluster simplification the r-final forms in (5) emerge.

286 

 Peter Gilles

(5) Non-application of schwa epenthesis due to final n-deletion OHG > Luxembourgish dorn Dar *Daren ‘thorn’ horn Har *Haren ‘horn’ kern Kär *Kären ‘core’

(cf. Germ. Dorn) (cf. Germ. Horn) (cf. Germ. Kern)

In general, schwa epenthesis seems to have affected all words that meet the described coda structures. Nevertheless, it has decreased noticeably in present-day Luxembourgish. First, several possible contexts that are documented in older dictionaries no longer exhibit epenthesis (6).4 Interestingly, most instances have an /r/ as the first member of the cluster.5 (6) Loss of former epenthesis /nɕ/ /lp/ /lm/ /rm/ /rf/ /rɕ/

[ˈmunəɕ] > [ˈmə.nəɕ] > [ˈtuləp] > [ˈpɑləm] > [ˈgɑləm] > [ˈaːʀəm] > [ˈvuːʀəm] > [ˈtuːʀəm] > [ˈduːʀəf] > [ˈfaːʀəf] > [ˈkuːʀəf] > [ˈmiːʀəf] > [ˈkiːʀəɕ] > [ˈduːʀəɕ] > [ˈmuːʀəɕ] >

[munɕ] munch ‘some’ [mənɕ] Mënch ‘monk’ [tulp] Tulp ‘tulip’ [pɑlm] Palm ‘palm’ [gɑlm] Galm ‘stench’ [aːɐ̯m] aarm/Aarm ‘poor’/‘arm’ [vuəm] Wuerm ‘worm’ [tuəm] Tuerm ‘tower’ [duəf] Duerf ‘village’ [faːəf] Faarf ‘color’ [kuəf] Kuerf ‘basket’ [miːəf] mierf ‘crumbly’ [kiəɕ] Kierch ‘church’ [duəɕ] duerch ‘through’ [muəɕ] Muerch ‘marrow’

Second, new words entering the language, mostly as loans from Standard German, are no longer subject to epenthesis, which yields strong evidence that the process is no longer active (7).

4 Older dictionaries until the mid-20th century list a far greater number of words with epenthesis (cf. LWB, WLM). The WLM of 1906 contains e.g. the loan Ongeren < Ungarn ‘Hungary’, where schwa epenthesis took place during phonological integration. 5 After the loss of the schwa, the former syllable-initial [ʀ] moved into the coda, which in turn led to r-vocalization (e.g. Wuerm [ˈvuː.ʀəm] > [vuːʀm] > [vuːəm] > [vuəm] ‘worm’). Whether r-vocalization was a consequence of the loss of schwa epenthesis or an independent process in the restructuring of the syllable coda still remains an open research question.



Phonological domains in Luxembourgish  

 287

(7) New loans without epenthesis /lm/ Golf, Film, Alm ‘alp’, Helm ‘helmet’, Palm ‘palm’, calm ‘quiet’ /lɕ/ Elch ‘elk’, Molch ‘newt’, Mulch ‘mulch’, Clinch /rf/ Worf ‘throw’, Harf ‘harp’ /rm/ Daarm ‘intestine’, Schierm ‘umbrella’, ferm ‘strong’, Form, Norm, enorm /rn/ extern, intern, modern, Fernseh ‘TV’

Although the rise of schwa epenthesis in Old High German must be seen as an optimization of the syllable structure, this process did not continue into present-day phonology. Instead, the word language feature of a more complex syllable coda is gaining ground. Today, schwa forms part of the lexical representation of the few words in (4) above. In other words, schwa epenthesis has been lexicalized. From the perspective of the syllable language/word language dichotomy it can be concluded that a syllable language feature has been lost in the course of language history. The discussion so far has shown that Luxembourgish offers a few features that could be attributed to a syllable language. However, the evidence is either weak (for schwa in stressed or root syllables) or disappearing (schwa epenthesis). It seems justified to classify the features discussed only as potential syllable language features. In order to characterize Luxembourgish appropriately, the next section will explore the word language traits.

4 Word language traits Apart from well-known word language features such as final obstruent devoicing (bleiw-en [ˈblɑɪ ̯vən] ‘stay-inf’ > (du) bleif-s [blɑɪ f̯ s] ‘(you) stay.pres-2sg’) or ambisyllabic consonants (Hunneg [ˈhuṇəɕ] ‘honey’, Bidden [ˈbiḍən] ‘bin’), Luxembourgish is subject to several further processes which have not yet fully figured in the discussion. The syllable structure, especially of the onset, shows great accordance with Standard German. Consonant clusters with up to three elements in the word-initial onset (Sproch ‘saying’, Strof ‘punishment’) or the word-final coda (buet-s ‘bath.pres-2sg’, Uebst ‘fruits’, Stëbs ‘dust’, (hien) gaaps-t ‘(he) yawn.pres-3sg’) form an integral part of the system (see Bertram 2011). These complex coda clusters can adequately be analyzed as containing extrasyllabic consonants, which are allowed only at the edges of phonological words (see Wiese 2000 for German). It is their function to render the boundaries of the word more prominent. Thus, the word language criterion of a complex syllable structure is met. Word stress plays a crucial role in Luxembourgish and falls in most cases on the penultimate syllable creating trochaic feet (cf. Gilles 2010). From the fol-

288 

 Peter Gilles

lowing examples it will become clear that the trochaic foot is achieved and/or strengthened by various phonetic, phonological and morphological processes. By rendering the stressed syllable articulatorily and perceptively more prominent, surrounding unstressed syllables are accordingly more vulnerable for various kinds of reduction processes. This holds true for the native lexicon, where reduction of vowels in unstressed syllables is frequently observable, especially in the speech of the younger generation (e.g. kanneresch [ˈkɑnɐʀeʃ] > [ˈkɑnəʀəʃ] ‘childish’, wichteg [ˈviɕteɕ] > [ˈviɕtəɕ] ‘important’, Samschdeg [ˈzɑmʃdeɕ] > [ˈzɑmʃdəɕ] ‘Saturday’). The same holds partially true for the integration of loan words. While the French loans in (2) keep the full vowel in the final open syllable, after the word stress has moved to the penultimate syllable, the examples in (8) show that under the same circumstances a now unstressed closed syllable is reduced to an open central vowel [ɐ]. When the plural suffix -en is attached to these words the r-vocalization is cancelled and [ɐ] is realized as schwa. (8) French Luxembourgish Singular Plural chauffeur coiffeur tracteur docteur but: auteur

> > > > >

Chauffer(en) Coiffer(en) Trakter(en) Dokter(en) Auteur(en)

‘driver(pl)’ ‘hairdresser(pl)’ ‘tractor(pl)’ ‘doctor(pl)’ ‘author(pl)’

[ˈʃoːfɐ] [ˈʃoːfəʀən] [ˈkwɑfɐ] [ˈkwɑfəʀən] [ˈtrɑktɐ] [ˈtrɑktəʀən] [ˈdoktɐ] [ˈdoktəʀən] [ˈoːtœʀ] [ˈoːtœʀən]

The last example in (8), however, shows that the integration process of French loans has not reached its endpoint yet. Some words may retain the original French syllable structure. These examples from loan phonology demonstrate how the imposition of the preferred trochaic foot favors vowel reduction in unstressed syllables. Taking all of these processes together, Luxembourgish strives to render the domain of the phonological word more prominent (at the expense of the syllable). Vowels in pretonic syllables may undergo deletion and the remaining consonantal material may be moved to the onset of the stressed syllable where the complexity of the onset cluster is increased. As can be seen in (9), vowel loss affects not only inherently weak schwas but also full vowels. Given that all resulting clusters consist of obstruent + sonorant, they are all characterized by an increase in sonority and therefore comply with the Sonority Sequencing Generalization (cf. Blevins 1995). With the exception of the affricate + vibrant cluster in [ʦʀek] ‘backwards’, all clusters are allowed in Luxembourgish.



Phonological domains in Luxembourgish  

(9)

villäicht zeréck6 virun dorun doran duerop

[fiˈlæɪ̯ɕt] [ʦəˈʀek] [fiˈʀʊn] [doːˈʀʊn] [doːˈʀɑn] [duəˈʀɔp]

> > > > > >

vläicht zréck vrun drun dran drop

 289

[flæɪ̯ɕt] ‘perhaps’ [ʦʀek] ‘backwards’ [fʀʊn] ‘ahead’ [dʀʊn] ‘at it’ [dʀɑn] ‘in that’ [dʀɔp] ‘on it’

In most cases, the previous structure with a CV syllable was more optimal in terms of a syllable language, which normally would impede such a reduction. Consequently, this process strengthens the domain of the phonological word. This process supposedly has been active in Luxembourgish phonology for a long time, as it is also found in certain derivations in family names, e.g. the family name Mréches derives from the diminutive form of Mariachen ‘Marydim’. Further complex onset clusters arise in the interplay of the proclitic definite article. When immediately adjacent to the noun, the article for feminine singular déi /dəɪ̯/, neuter singular dat /daːt/ and plural /daːt/ is pronounced [t] or [d̥], depending on the voicing feature of the following consonant. Orthographically these clitics are spelled as d’. The examples in (10) show that the clitic is incorporated in the following syllable onset. (10) Fem. Sg. Fem. Sg. Fem. Sg. Neut. Sg. Neut. Sg. Pl. Pl.

/dəɪ̯/+/fʀaː/ [tfʀaː] /dəɪ̯/+/mɑm/ [d̥mɑm] /dəɪ̯/+/ˈbɑnɑn/ [ˈd̥bɑnɑn] /daːt/+/ʀəɪ̯/ [d̥ʀəɪ̯] /daːt/+/kɑnt/ [tkɑnt] /dəɪ̯/+/ˈkʀiɕəʀ/ [ˈtkʀiɕɐ] /dəɪ̯/+/pləɪ̯/ [tpləɪ̯]

d’Fra d’Mamm d’Banann d’Réi d’Kand d’Krich-er d’Plé-i

‘the women’ ‘the mother’ ‘the banana’ ‘the deer’ ‘the child’ ‘the war-pl’ ‘the plough-pl’

Contrary to the examples in (9) this condition of external sandhi gives rise to several otherwise disallowed syllable onsets such as *tfʀ-, *tk-, which violate the sonority hierarchy. Again, the most important result of this process is to profile the left-hand side of the phonological word. Thus, in Luxembourgish rendering the left boundary of the word more prominent seems to be preferable to adhering to the principle of keeping syllable structures as simple as possible. If the domain of the syllable were the guiding phonological category in this sandhi, then one would expect some kind of repair to take place. However, vowel insertion as a repair strategy for this violation is not attested.

6 In Luxembourgish orthography the does not indicate stress but represents a closed short [e].

290 

 Peter Gilles

Comparable developments have taken place for the syllable coda. In line with several Central Franconian dialects, Luxembourgish was subject to an unexceptional apocope of word-final schwa, which turned former Middle High German CVCV words into CVC words (e.g. MHG mitte > Mëtt ‘center’, wuche > Woch ‘week’, suoche > Sich ‘search’, hecke > Heck ‘hedge’, seife > Seef ‘soap’). By this development, a previously open syllable was changed into a closed one, which is dispreferred in syllable languages. Several contexts are traceable where complex word-final codas arose after the schwa was syncopated in the morphological suffix -es (11). (11) [gəˈkɑɪ̯məs] Gekeimes [gəˈtʀæɪ̯pəs] Geträipes [ænəs] (leschten) Endes

> [gəˈkɑɪ̯ms] > [gəˈtʀæɪ̯ps] > [æns]

Gekeims Geträips Enns

‘germination’ ‘bowel’ ‘after all’

The superlative suffix -st is invariably attached to all adjectives even when highly marked codas are the consequence (12). Whereas even a word language like Standard German has developed a phonologically conditioned allomorph containing schwa to facilitate syllabification (‑st ~ ‑est), Luxembourgish accepts coda clusters with three elements. (12) Luxembourgish German [ʃəɪ̯nst] schéin-st [bʀeːtst] breet-st

[ʃøːns.tə] [bʀaɪ̯.təs.tə]

schön-ste ‘beautiful-superl’ breit-este ‘wide-superl’

When taking fast speech forms into account, even more complex codas may occur (cf. variants like zwanzeg [ˈʦwɑnʦəɕ] > [ˈʦwɑnʦɕ] ‘twenty’, fofzeg [ˈfofʦəɕ] > [fofʦɕ] ‘fifty’). Furthermore, Luxembourgish has developed a pseudo-suffix -s (with yet unclear semantics and morphological function), which is quite frequently added to function words. Again, this consonant adds to the complexity of the coda (13). (13)

an engem > an engems ‘into one’ schonn > schonns ‘already’ obschonn > obschonns ‘although’ nodeem > nodeems ‘after’ duerop > drops ‘afterward’

A peculiar case of schwa loss is observable in polysyllabic, inflected words containing two unstressed syllables following the stressed initial syllable. While the full form, which is also the orthographically correct form, retains all syllables and exhibits a fairly simple syllable structure, a new and very frequent variant has emerged in which the penultimate syllable is syncopated (14); the syllable boundary is indicated by a period.



Phonological domains in Luxembourgish  

(14)

Fenster-en Eemer-en dréchen-en wichteg-en trëppel-en

[ˈfəns.tə.ʀən] [ˈeː.mə.ʀən] [ˈdre.ɕə.nən] [ˈviɕ.tə.ʑən] [ˈtʀə.pə.lən]

> [ˈfəns.tʀən] > [ˈeːm.ʀən] > [ˈdreɕ.nən] > [ˈviɕt.ʑən] > [ˈtʀəp.lən]

 291

‘window-pl’ ‘bucket-pl’ ‘dry-pl’ ‘important-pl’ ‘trip-inf’

Conrad (2010) has shown that the disyllabic form already dominates for certain words and this underlines the importance of the trochee for Luxembourgish. Again, we are confronted with the abandonment of a syllable language feature (CV syllable structure) in favor of a word language feature (more complex syllable structure). What is instructive here is that it is always the syllable immediately adjacent to the stressed syllable that gets reduced, while the final syllable always stays intact. Two word-related aspects are responsible for this asymmetry. First, due to the preference for trochees, pretonic or post-tonic syllables are affected first. Syllables that are further away are protected. Second, the final syllable carries important morphological information (e.g. number, case or person), which in Luxembourgish seems to be protected from reduction. It should become clear from these examples that word-related aspects such as morphological structure shape the phonological form. If Luxembourgish were to belong to the syllable language type, one would expect a stronger tendency to protect unstressed syllables as well as to avoid complex onsets or codas. Instead, one finds a partial increase in syllabic complexity. The last feature to be discussed in this section refers to the linking-element -s, which has been analyzed by Nübling and Szczepaniak (2008) as a means of optimizing the phonological word in Standard German. More specifically, the linking -s enables clearer rendering of the juncture in compounds. It turns out that the linking -s in general affects even more compounds in Luxembourgish than in German (15a). Furthermore, while in Standard German a linking -s is systematically not allowed when the first word of the compound ends with an open syllable, this blocking does not exist in Luxembourgish (15b). (15) Luxembourgish German a. Äerdbier-s-kuch Erdbeer-kuchen ‘strawberry cake’ Fussball-s-terrain Fussball-platz ‘football field’ b. Vodka-s-glas Vodka-glas ‘glass for vodka’ Büro-s-artikel Büro-artikel ‘office supplies’ Konto-s-stand Konto-stand ‘account balance’ Auto-s-assurance Auto-versicherung ‘car insurance’ Auto-s-steier Auto-steuer ‘car tax’ Allé-s-match n.a. ‘first leg’

This indicates that Luxembourgish benefits more from the linking -s, which brings it closer to a word language.

292 

 Peter Gilles

5 Phenomena related to the syllable and the phonological word at the same time This section is devoted to two prominent cases of external sandhi which demonstrate the relevance of the domains of the syllable and the phonological word at the same time. It will be discussed how these features could be related to the syllable language/word language dichotomy. Both phenomena are rather frequent and therefore pertinent and important features of Luxembourgish phonology.

5.1 n-rule The first feature concerns the final alveolar nasal, which is deleted according to the structure of the onset of the following word (the so-called n-rule or n-deletion). In a simplified version of this phonological rule, every word-final -n is protected from deletion only if the following word starts either with a vowel or with the consonants t, d, ts, n or h. If this criterion is not met, the nasal gets deleted. Some examples of preservation (bold face) and deletion (underscore) are given in (16). (16) n-rule in Luxembourgish Preservation a. den Auto ‘the car’ den Dëlpes ‘the idiot’ klengen Tuerm ‘small tower’ en Zuch ‘a train’ en Haus ‘a house’ b. Frittendëppen ‘chip pan’ Zoppenteller ‘soup plate’ c. Wäin drénk-en ‘wine drink-inf’ (si) ginn heem ‘(they) go home’ kann drénk-en ‘can drink-inf’ hunn ugefaangen ‘have started’ unzefänken7 ‘in order to start

Deletion de_ Patron ‘the employer’ de_ Bauer ‘the farmer’ klenge_ Kueb ‘small raven’ e_ Land ‘a country’ e_ Kaffi ‘a coffee’ Fritte_bud ‘snack bar’ Zoppe_läffel ‘soup spoon’ Wäi_ sich-en ‘wine get-inf’ (si) gi_ virun ‘(they) go further’ ka_ maach-en ‘can do-inf’ hu_ gesicht ‘have searched’ u_fänk-en ‘start-inf’

This phonological rule applies across the board to all word classes and in most cases affects the morphological suffix -en (16a). The n-rule is also found in compounds (16b) and in syllables containing vowels other than schwa (16c). It is thus always the right-hand context that determines whether the nasal is preserved or

7 Form of the so-called “extended infinitive” used in infinitive clauses.



Phonological domains in Luxembourgish  

 293

deleted. Although this is a postlexical rule, speakers adhere to it almost without fail. Given that -en belongs to the most frequent syllable types, the n-rule operates quite frequently. The phonological explanation given in Gilles (2006) assumes that final -n is always extrasyllabic in Luxembourgish, i.e. the nasal is realized phonetically only if it can be phonologically integrated into a syllable. In order to render this phonological licensing possible, various conditions have to be met: –– The following word begins with a vowel: In cases like en Apel ‘an apple’, the extrasyllablic nasal can be integrated into the following syllable due to postlexical resyllabification ([ən] + [aː.pəl] > [ə.naː.pəl]), a feature similar to enchaînement in French. This is possible because Luxembourgish, in contrast e.g. to Standard German, lacks the glottal stop in vowel initial syllables. After resyllabification the nasal forms part of the onset of the following syllable. It is no longer regarded as extrasyllabic and is therefore protected from deletion. A similar effect occurs when the word at the right-hand side begins with the fricative [h], which is phonetically weak in Luxembourgish and phonologically transparent for resyllabification. –– The following word begins with a homorganic obstruent [t, d, ʦ]: In combinations like den Tuerm ‘the tower’, the nasal is retained because, together with the following alveolar obstruent, it constitutes a partial ambisyllabic geminate [nt]. This is possible because [n] and [t] share all C-place features (cf. Clements and Hume 1995 for a discussion of the relevant feature geometries). Thereby the nasal belongs at least partially to the following syllable, and this is sufficient to syllabify this former extrasyllabic consonant. Deletion is again blocked. –– The following syllable begins with a heterorganic consonant, e.g. [p, k, ʃ]: In instances like de_ Chauffer ‘the driver’, the following onset is occupied by a consonant, which does not allow the formation of a partial geminate because its C-place features are different. Therefore resyllabification is impossible and the nasal retains the status of an extrasyllabic consonant. In the phonological representation the nasal is thus not licensed and is subsequently deleted as stray material. This phonological description focuses on the structural notion of the syllable, i.e. on syllabic licensing and resyllabification. In fact, the n-rule cannot be described properly without crucial reference to the phonological properties of the syllable. It seems logically consistent to assume that the n-rule is thus an example of the particular importance of the syllable in Luxembourgish. Nevertheless, if one takes a closer look at the domain of application, the predominant importance of the syllable becomes less obvious. Of course, this

294 

 Peter Gilles

external sandhi takes place at a syllable boundary, but this boundary is always a word boundary, too. In order to attribute the n-rule to the domain of the syllable only, one must find contexts without an interfering word boundary. Word-internal combinations of -n + heterorganic, heterosyllabic consonant are limited to the coronal consonants, i.e. z, ʃ, l, ɕ, because word-internal clusters of nasal and non-coronal consonants are always homorganic (Bank [bɑŋk] ‘bank’, pompel-en [ˈpompələn] ‘pump-inf’, Gromper [ˈgʀɔmpɐ] ‘potato’). In the coronal context, it turns out that the nasal is always retained word-internally regardless of the consonant the next syllable begins with. A form like Ronschel [ʀon.ʃəl] ‘pucker’ is correct although the n-rule is violated: *[ʀo_.ʃəl] is clearly ungrammatical. The further examples in (17) demonstrate that the n-rule is not applicable at word-internal syllable boundaries. (17) Blocking of n-rule inside of the phonological word Zënsen [[ʦən]σ[zən]σ]ω *[[ʦə_]σ[zən]σ]ω Bensin [[bæn]σ[zin]σ]ω *[[bæ_]σ[zin]σ]ω Ronschel [[ʀon]σ[ʃəl]σ]ω *[[ʀo_]σ[ʃəl]σ]ω klunsch-en [[klun]σ[ʃən]σ]ω *[[klu_]σ[ʃən]σ]ω perséinlech [[pæɐ]σ[zəɪ̯n]σ[ləɕ]σ]ω *[[pæɐ]σ[zəɪ̯_]σ[ləɕ]σ]ω Ween-chen [[veːn]σ[ɕən]σ]ω *[[veː_]σ[ɕən]σ]ω

‘interest rate’ ‘fuel’ ‘pucker’ ‘sway-inf’ ‘personally’ ‘cart-dim’

These word-internal contexts indeed are the clearest indicator of the domain of application of the n-rule: This phonological rule applies only to the right edge of the phonological word, which, of course, always coalesces with the syllable boundary too. Furthermore, for this phonological analysis one must keep in mind the crucial assumption of this phonological analysis, namely that final -n in Luxembourgish is extrasyllabic. According to the assumptions of Lexical Phonology, those extrasyllabic segments can only be attributed to the edges of words and are not permitted word-internally (Wiese 2000: 47‒49). Hence, the word-internal nasals in (17) are not subject to the n-rule anyway. The evidence that the n-rule does not exclusively depend on the presence of any syllable boundary is further supported by its partial sensitivity to morphological factors. Words ending with the feminine suffix -in (18a), the negation prefix on- (18b) or the word formation suffix ‑ioun (18c) block the n-rule. Here, regardless of the following context, the nasal is retained. If the syllable was the decisive domain of application, then this dependence on morphological factors would not be relevant for the n-rule. (18) Blocking of n-rule for word formation affixes -in, on-, -ioun a.

Léier-in geschwat Manger-in verluer

*Léieri_ geschwat *Manageri_ verluer

‘talked to the teacher-fem’ ‘lost the manager-fem’



Phonological domains in Luxembourgish  

b. c.

 295

onméiglech *o_méiglech ‘impossible’ ongerecht *o_gerecht ‘unjust’ Natioun gegrënnt *Natiou_ gegrënnt ‘founding a nation’ Regioun verloossen *Regiou_ verloossen ‘leaving the region’

In concluding this discussion on the n-rule, we are left with an intricate interwovenness of the domains of the syllable and the word. Without any doubt, deletion or preservation of final -n must be explained with regard to the notion of the syllable, because it simplifies the syllable coda. As for the domain of application, however, the n-rule is categorically blocked within phonological words and occurs only across the boundaries of phonological words. Although syllable structure is optimized by the n-rule, it cannot unambiguously be described as a syllable-related feature only. Thus, for the discussion here, it seems that the n-rule can neither be considered a typological feature of a syllable language nor of a word language. In further research it remains to be discussed how the domain of optimization and the domain of application are related.

5.2 Resyllabification and voicing assimilation The next feature relates to resyllabification from a broader point of view. Resyllabification can occur on different levels of the phonological system (lexical and postlexical), whenever the syllable structure of the base form is restructured due to affixation or in external sandhi. This form of “continuous syllabification” (Kenstowicz 1994: 269‒271) is dealt with here under the umbrella term “resyllabification”. Luxembourgish allows resyllabification in many more cases than e.g. Standard German. Similar to Standard German, resyllabification takes place within the phonological word, when a schwa-initial, inflectional or derivational suffix is added. As shown in (19), a syllable-final consonant moves into the onset of the following syllable (or forms an ambisyllabic consonant, if the preceding syllable ends with a short vowel). (19b) also shows that obstruent final devoicing (Auslautverhärtung) is cancelled after resyllabification and the underlying voiced consonant shows up. (19) Word-internal resyllabification a. Kraaft Sprooch Reef Präis op

/kʀaːft/ /ʃpʀoːx/ /ʀeːf/ /pʀæɪ̯s/ /op/

[kʀaːft] [ʃpʀoːx] [ʀeːf] [pʀæɪ̯s] [op]

+ /-en/ + /-en/ + /-er/ + /-er/ + /-en/

Kräften [kʀæf.tən] ‘power(-pl)’ Sproochen [ʃpʀoː.xən] ‘language(-pl)’ Reefer [ʀeː.fɐ] ‘circle(-pl)’ Präisser [pʀæɪ̯.sɐ] ‘price(-pl)’ open [oṗən] ‘open(-pl)’

296 

 Peter Gilles

b. Scheif Blat héich

/ʃɑɪ̯v/ /blaːd/ /heiʑ/

[ʃɑɪ̯f ] [blaːt] [həɪ̯ɕ]

+ /-en/ Scheiwen [ʃɑɪ̯.vən] ‘pad(-pl)’ + /-er/ Blieder [bliə.dɐ] ‘leaf(-pl)’ + /-en/ héijen [həɪ̯.ʑən] ‘high(-inf)’

Due to the absence of a syllable-initial glottal stop – and contrary to e.g. Standard German – Luxembourgish also shows across-word resyllabification, which is almost compulsory when the following syllable begins with a vowel (cf. an older account in Goudaillier 1987). This kind of resyllabification is known in several languages, e.g. French and English (see Gussenhoven and Jacobs 1998: 167, where the process is called “liaison”). (20) presents examples where the final consonants move across the word boundary into the empty syllable onset of the following word. This modification eventually leads to a misalignment of the syllable and the phonological word. Resyllabification is indicated with a “‿”, the boundary of a phonological word with a “#”. (20) Resyllabification across word boundaries a. Sprooch ausbau-en [ʃpʀoː.‿ɣ#æʊ̯s.bɑʊ̯.ən] Reef apak-en [ʀeː.‿v#ɑ.paː.kən] Scheif ausmol-en [ʃɑɪ̯.‿v#æʊ̯s.moː.lən] Kees opmaach-en [keː.‿z#op.maː.xən] Kraaft ubidd-en [kʀaːv.‿d#u.biḍən] b. war och [vaː.‿ʀ#ox] gëschter Owend [gəʃ.tə.‿ʀ#oː.vənt] awer ech [aː.və.‿ʀ#əɕ]

‘language elaborate-inf’ ‘circle wrap-inf’ ‘pad color-inf’ ‘cash register open-inf’ ‘power offer-inf’ ‘was-2sg also’ ‘yesterday evening’ ‘but I’

The instances in (20b) are decisive in showing that it is indeed resyllabification that is occurring here: The vibrant r has (at least) two allophones depending on the position in the syllable. When r is located syllable-finally it is vocalized to [ɐ], whereas r is realized as a vibrant [ʀ] when located syllable-initially. The non-vocalized realizations in (20b) thus clearly indicate that r is no longer part of the preceding syllable but actually forms part of the syllable onset, i.e. it was resyllabified.8 Although clearly a postlexical process, this resyllabification is almost obligatory in Luxembourgish. The filling of otherwise empty onsets demonstrates that syllable structure is optimized at the expense of marking the word boundary, which is blurred instead. This type of resyllabification can thus be regarded as a feature of a syllable language.

8 This process is similar to the so-called “linking r” of some varieties of English (see Wells 1982).



Phonological domains in Luxembourgish  

 297

However, as can already be seen in the examples in (20a), resyllabification is always associated with the concomitant process of voicing assimilation (see Zsiga 2011 for a general discussion). Moreover, as the further examples in (21) show, every resyllabified consonant undergoes voicing in external sandhi at the same time. The left-most column in (21) contains the phonological forms with the underlying coda consonants. The phonetic transcription in the second-to-last column shows the resyllabified and voiced consonant when a vowel-initial phonological word is directly adjacent. (21) a. /ʦæɪ̯t/ Zäit a Raum [ʦæɪ̯.‿d#ɑ ʀæʊm] ‘time and space’ /vælt/ Weltall [væl.‿d#ɑl] ‘universe’ /kux/ Kuch iess-en [ku.‿ɣ#iə.sən] ‘cake eat-inf’ /kaːf/ kaf Äppel [kaː.‿v#æ.pəl] ‘buy.imp[sg] apple.pl’ b. /kleːd/ Kleed undo-en [kleː.‿d#un.doː.ən] ‘dress put on-inf’ /ʃʀɑʊ̯v/ Schrauf andréi-en [ʃʀɑʊ̯.‿v#ɑn.dʀəɪ̯.ən] ‘screw turn-inf’ /taːz/ Taass eidel [taː.‿z#ɑɪ̯.dəl] ‘cup empty’

From (21b) it also becomes obvious that final underlying voiced obstruents, which otherwise would be subject to final devoicing (see 19b), regain (or rather keep) voicing in this context (/kleːd/ + /undoːən/ > [kleːt] + [un.doː.ən] > [kleː.‿d#un. doː.ən]). The process of voicing assimilation consequently overrides the effect of final devoicing.9 Voicing assimilation itself has a natural phonetic basis, because voicing can be easily maintained between two vowels. Voicing assimilation of this kind occurs across the board and can be found for all syllable types, between words of all word classes and also within compounds. Similar patterns of resyllabification combined with voicing assimilation have been discussed for Dutch, Flemish and Limbourgish (Grijzenhout and Krämer 2000; Ernestus 2003; Hinskens 2007; Mascaró and Wetzels 2008; Noske 2008), as well as for Catalan (Lloret and Jiménez 2009; Caro Reina, this volume). What is unique about Luxembourgish is that the joining of two phonological words always must be marked by voicing assimilation (plus resyllabification). Furthermore, if the syllable coda contains more than one obstruent or an affricate, all sounds are affected by voicing assimilation by creating a fully voiced intervocalic sequence stretching over a word boundary (22).

9 Note that in most cases, final devoicing is also part of the spelling rules of Luxembourgish (hence Schrauf ‘screw’ in contrast to Schrauw-en ‘screw-pl’), whereas across-word voicing is not marked in spelling.

298 

 Peter Gilles

(22) Resyllabification and voicing assimilation in complex coda clusters a. aacht [aːxt] ‘eight’ Gesiicht [gə.ziːɕt] ‘face’ bleif-s [blɑɪ̯fs] ‘stay.pres-2sg’ Uebst [uə̯pst] ‘fruits’ ganz [gɑnʦ] ‘whole’ b. aachtanachzeg [aːɣ.‿dɑn.ɑx.ʦəɕ] ‘eighty-eight’ (lit. ‘eight and eighty’) Gesiicht ukuck-en [gə.ziːʑ.‿du.ku.kən] ‘face look at-inf’ bleif-s op [blɑɪ̯v.‿zop] ‘stay up.pres-2sg’ Uebst a Geméis [uə̯bz.‿dɑ.gə.məɪ̯s] ‘fruits and vegetables’ ganz elleng [gɑn.‿ʣə.læŋ] ‘all alone’

As far as resyllabification is concerned, it is interestingly only the last consonant of the coda that moves into the next syllable, while the remaining consonant(s) stay in its (their) original syllabic slot. This behavior provides important evidence that voicing assimilation in fact operates independent of resyllabification. If resyllabification created the input for voicing assimilation, it would be impossible to explain why the whole coda is voiced and not only the resyllabified consonant. By analyzing the behavior of obstruents word-internally, however, it becomes evident that voicing is not possible under this condition. In a word like déck /dek/ ‘thick’, the voiceless consonant is retained in the inflected form décken [dekən], although a vowel-initial syllable follows, but is voiced when followed by a vowel-initial phonological word as in déck an dënn [de.‿gɑn.dən] ‘thick and thin’. This contrast is demonstrated in (23). Phonologically speaking, the feature “voice” is neutralized across words (external sandhi), but it is preserved word-internally (internal sandhi). According to Szczepaniak (2007), this incongruity of internal and external sandhi is indicative of a word language. (23) Resyllabification without and with voicing assimilation a. word-internally déck-en [de.kən] ‘thick-pl’ Equip-en [e.ki.pən] ‘team-pl’ Strooss-en [ʃtʀoː.sən] ‘street-pl’ b. across words déck an dënn [de.‿gɑn.dən] ‘thick and thin’ Equipe ass do [e.ki.‿bɑs.doː] ‘team is there’ Strooss an Haus [ʃtʀoː.‿zɑn.hæʊ̯s] ‘street and house’

It can be concluded that voicing assimilation is determined by the phonological word. Only when two phonological words join in external sandhi can voicing take place. The identical word-internal context does not trigger voicing, while resyllabification can occur at every syllable boundary. Like the case of the n-rule



Phonological domains in Luxembourgish  

 299

discussed above, we are confronted with two different phonological domains of application. Voicing assimilation thus would be attributed as a feature of a word language, while resyllabification can be regarded as a syllable language feature. It has yet to be clarified how the two processes are related to each other. Since voicing is involved, the effect of final devoicing has to be taken into account as well. The derivations in (24) present two sandhi constellations and illustrate the sequential order of the processes involved. (24) Phonological derivations of two sandhi sequences a. b.

1. Final Devoicing 2. Voicing Assimilation 3. Resyllabification 1. Final Devoicing 2. Resyllabification 3. Voicing Assimilation

bleif op bleif-s op ‘stay up.imp[sg]’ ‘stay up.pres-2sg’ /blɑɪ̯v/+/op/ /blɑɪ̯vs/+/op/ [blɑɪ̯f]+[op] [blɑɪ̯fs]+[op] [blɑɪ̯v.op] [blɑɪ̯vz.op] [blɑɪ̯.‿vop] [blɑɪ̯v.‿zop] [blɑɪ̯f]+[op] [blɑɪ̯fs]+[op] [blɑɪ̯.‿fop] [blɑɪ̯f.‿sop] [blɑɪ̯.‿vop] n.a. → *[blɑɪ̯f.‿sop]

First of all, final obstruent devoicing (1) as a process of the lexical stratum clearly has to apply before all postlexical processes. Next, voicing assimilation (2) takes places by voicing all obstruents before a vowel-initial word boundary. After that, resyllabification (3) moves the last obstruent into the onset of the following syllable. That voicing assimilation applies before resyllabification follows from the behavior of the complex coda clusters in cases like bleif-s op ‘stay up.pres-2sg’: As becomes evident from (24b), voicing assimilation cannot apply in a resyllabified sequence like [blɑɪ̯f.‿sop] because the condition of a vowel-initial phonological word is not met. Hence, this possible output form *[blɑɪ̯f.‿sop] would be ungrammatical. Resyllabification in Luxembourgish at first glance resembles enchaînement in French (see e.g. Côté 2011) and one could arguably evaluate the Luxembourgish case as a language contact feature. However, resyllabification in French has no influence on the voicing of the liaison consonant. In a compound like sens unique ‘one-way road’, the final s remains voiceless in French, despite the voicing-favoring context of the following vowel (i.e. [sɑ̃.‿sy.ˈnik]). However, when this loan is integrated into Luxembourgish, voicing assimilation takes place (i.e. [sɑ̃.‿zy.ˈnik]; see also Schanen 2006: 513). Thus, the similarity between French and Luxembourgish is only superficial and the phonological processes in the two languages are different. In fact, Luxembourgish shows striking resemblances to certain varieties of Dutch regarding the various kinds of voicing processes (Booij

300 

 Peter Gilles

1999; Ernestus 2003). A relationship with Germanic varieties north of Luxembourg is much sounder from a language historical point of view and deserves further investigation.

6 Discussion In this paper, central aspects of Luxembourgish phonology have been investigated with regard to the typological distinction between syllable languages and word languages. Summarizing the results, it is possible to complement the findings of Szczepaniak (2007) in several ways (Table 2). Table 2: Syllable language and word language traits in Luxembourgish Traits of a syllable language

Traits of a word language

–– schwa in stressed syllables (native and loan lexicon) –– resistance to delete schwa in unstressed final syllables –– schwa epenthesis (decreasing, no longer productive)

–– complex syllable structure –– strong influence of penultimate word stress –– apocope of final schwa has closed previously open syllables, thus reducing the number of CV syllables –– cliticization leading to additional complex onset clusters –– non-syllabic suffixes creating complex coda clusters –– reduction or deletion of pretonic and posttonic syllables –– schwa deletion in non-final unstressed syllables –– increased use of linking -s

It becomes obvious that in present-day Luxembourgish pertinent features pointing towards a syllable language are few in number or are decreasing (schwa epenthesis). In contrast, this survey reveals far more evidence of word language features. It thus seems that the phonology of Luxembourgish is on the same track as German, whose historical development from a syllable language in Old High German towards a word language today was convincingly shown by Szczepaniak (2007). Considering this overwhelming evidence, one could classify Luxembourgish as a word language as well. As has already been indicated by Szczepaniak (2010), some features exhibit an even stronger orientation towards the word language type than than their Standard German counterparts (e.g. increased use of clitics, linking -s and non-syllabic suffixes).



Phonological domains in Luxembourgish  

 301

However, at least two external sandhi phenomena discussed in section 5.3 pose serious challenges to the application of the typological parameter under discussion. The n-rule and resyllabification/voicing assimilation both exemplify how the phonological domains of the syllable and the phonological word interact with and rely on each other at the same time. While these two phenomena are clearly related to syllable structure and weaken boundaries of two adjacent syllables by resyllabification, the domain of application brings the relevance of the phonological word into play again, as both processes apply at the boundaries of phonological words only. Thus, at least for these two rather frequent processes, both domains are needed to capture the phonological structure adequately. Consequently, unambiguous attribution to either the syllable language or word language type seems unfeasible. It remains an open issue for further discussion how the mismatch encountered in the Luxembourgish data could be reconciled with the typological differentiation of syllable languages and word languages. At least three options seem conceivable: (1) It has been proposed in Szczepaniak (2010) that Luxembourgish represents a mixed type, where both phonological categories are crucially relevant. Although appealing, this solution would imply a revision of the current theory which consists basically of one continuum with the prototypical syllable language and the prototypical word language as the respective poles. Instead, two typological continua would be necessary, one to determine the syllable language features and another one for the word language features. A language would then be classified typologically with regard to these two continua. (2) Individual features are weighted according to how prototypical they appear for either a syllable language or a word language. (3) If a process is attributed to more than one domain, the higher domain in the prosodic hierarchy would overrule the lower domains. In the present case, both processes of external sandhi would then be classified as word languages features. Further phonological as well as cross-linguistic research is needed to determine the interrelatedness of the syllable and the phonological word and the relevance of this interrelatedness for the phonological architecture as a whole.

References Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (1994): Einige Argumente gegen die Silbe als universale prosodische Hauptkategorie. In: Karl Heinz Ramers, Heinz Vater and Henning Wode (eds.), Universale phonologische Strukturen und Prozesse, 55–78. (Linguistische Arbeiten 310.) Tübingen: Niemeyer.

302 

 Peter Gilles

Auer, Peter (1997): Areale Variation und phonologische Theorie: Überlegungen am Beispiel der mitteldeutschen ‘Epenthese’. In: Gerhard Stickel (ed.), Varietäten des Deutschen. Regional- und Umgangssprachen, 46–87. (Jahrbuch des Instituts für deutsche Sprache 1996.) Berlin/New York: Walter de Gruyter. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Bertram, Julia (2011): Silbenstruktur des Luxemburgischen und Deutschen im Vergleich. Zur typologischen Bedeutung der Resilbifizierung. In: Peter Gilles and Melanie Wagner (eds.), Linguistische und soziolinguistische Bausteine der Luxemburgistik, 99–113. (Mikroglottika 4.) Frankfurt am Main: Peter Lang. Blevins, Juliette (1995): The syllable in phonological theory. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 206–244. Oxford: Blackwell. Booij, Geert (1999): The Phonology of Dutch. (The Phonology of the World’s Languages.) Oxford: Oxford University Press. Clements, George N. and Elizabeth V. Hume (1995): The internal organization of speech sounds. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 245–306. Oxford: Blackwell. Conrad, François (2010): Das luxemburgische Schwa. Zentraler Vokal einer jungen germanischen Sprache. Magister thesis, Universität Luxemburg/Universität Bamberg. Côté, Marie-Hélène (2011): French liaison. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 5: Phonology across Languages, 2685–2710. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Donegan, Patricia J. and David Stampe (1983): Rhythm and the holistic organization of language structure. In: John F. Richardson, Mitchell Marks and Amy Chukerman (eds.), Papers from the Parasession on the Interplay of Phonology, Morphology, and Syntax, 337–353. Chicago: Chicago Linguistic Society. Ernestus, Mirjam (2003): The role of phonology and phonetics in Dutch voice assimilation. In: Jeroen van de Weijer, Vincent J. van Heuven and Harry van der Hulst (eds.), The Phonological Spectrum. Vol. 1: Segmental Structure, 119–144. (Current Issues in Linguistic Theory 233.) Amsterdam/ Philadelphia: John Benjamins. Gilles, Peter (1999): Dialektausgleich im Lëtzebuergeschen. Zur phonetisch-phonologischen Fokussierung einer Nationalsprache. (Phonai 44.) Tübingen: Niemeyer. Gilles, Peter (2006): Phonologie der n-Tilgung im Moselfränkischen (‘Eifler Regel’). Ein Beitrag zur dialektologischen Prosodieforschung. In: Claudine Moulin and Damaris Nübling (eds.), Perspektiven einer linguistischen Luxemburgistik. Studien zu Diachronie und Synchronie, 29–68. (Germanistische Bibliothek 25.) Heidelberg: Winter. Gilles, Peter (2010): Wie Französisch, nur Ton 1 – Der Wortakzent des Deutschen und Luxemburgischen im Kontrast. In: Antje Dammel, Sebastian Kürschner and Damaris Nübling (eds.), Kontrastive Germanistische Linguistik, vol. 1, 111–142. (Germanistische Linguistik 206/209.) Hildesheim: Olms. Gilles, Peter and Jürgen Trouvain (2013): Luxembourgish. Journal of the International Phonetic Association. Goudaillier, Jean-Pierre (1987): Einige Spracheigentümlichkeiten der lëtzebuergeschen



Phonological domains in Luxembourgish  

 303

Mundarten im Lichte der experimentellen Phonetik. In: Jean-Pierre Goudaillier (ed.), Aspekte des Lëtzebuergeschen, 197–230. (Beiträge zur Phonetik und Linguistik 55.) Hamburg: Buske. Grijzenhout, Janet and Martin Krämer (2000): Final devoicing and voicing assimilation in Dutch derivation and cliticization. In: Barbara Stiebels and Dieter Wunderlich (eds.), Lexicon in Focus, 55–82. (Studia Grammatica 45.) Berlin: Akademie Verlag. Gussenhoven, Carlos and Haike Jacobs (1998): Understanding Phonology. London: Arnold. Hall, Nancy (2011): Vowel epenthesis. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 3: Phonological Processes, 1576–1596. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell. Hinskens, Frans (2007): Sandhi voicing and opacity in Limburg dialects of Dutch: Towards a formal account. In: Sandra Döring and Jochen Geilfuß-Wolfgang (eds.), Von der Pragmatik zur Grammatik, 75–85. Leipzig: Leipziger Universitätsverlag. Keller, Rudolf E. (1961): Luxemburgish. In: Rudolf E. Keller (ed.), German Dialects. Phonology and Morphology. With Selected Texts, 248–297. Manchester: Manchester University Press. Kenstowicz, Michel (1994): Phonology in Generative Grammar. (Blackwell Textbooks in Linguistics 7.) Cambridge, MA/Oxford: Blackwell. Krier, Fernande (2008): Réductions et amplifications phonétiques en luxembourgeois. In: Thomas Stehl (ed.), Kenntnis und Wandel der Sprachen. Beiträge zur Potsdamer Ehrenpromotion für Helmut Lüdtke, 93–107. Tübingen: Günter Narr. Lloret, Maria-Rosa and Jesús Jiménez (2009): Phonological variation in voicing across word boundaries. Paper presented to Old World Conference in Phonology 6, University of Edinburgh. Available at http://www.uv.es/foncat/cat/OrdArr.wiki. LWB = Luxemburgische Wörterbuchkommission (ed.) (1950‒1977): Luxemburger Wörterbuch. 5 vols. Luxemburg: Linden. Available at http://engelmann.uni.lu:8080/portal/wbb/woerterbuecher/lwb/wbgui. Mascaró, Joan and Leo Wetzels (2008): An OT analysis of the basic voicing typology and voice assimilation in Dutch. In: Esther Herrera Z. and Pedro Martín Butragueño (eds.), Fonología instrumental. Patrones fónicos y variación, 129–150. México: El Colegio de México, A.C. Newton, Gerald (1990): Central Franconian. In: Charles V.J. Russ (ed.), The Dialects of Modern German. A Linguistic Survey, 136–209. London: Routledge. Noske, Roland (2008): L’accent en proto-français : arguments factuels et typologiques contre l’influence du francique. In: Congrès Mondial de Linguistique Française 2008, Paris, France. Available at http://www.linguistiquefrancaise.org/10.1051/cmlf08238. Nübling, Damaris (2005): Das Lëtzebuergesche als Herausforderung für die Linguistik. In: Institut Grand-Ducal, Section de Linguistique, d’Ethnologie et d’Onomastique and Centre National de Littérature (eds.), Lëtzebuergesch: Entwicklungstendenzen und Forschungsperspektiven einer jungen Sprache: Beiträge zum Workshop Lëtzebuergesch, November 2001, Luxemburg und Mersch, 147–168. (Beiträge zur luxemburgischen Sprach- und Volkskunde 33.) Luxemburg: Institut Grand-Ducal, Section de Linguistique, d’Ethnologie et d’Onomastique and Centre National de Littérature. Nübling, Damaris and Renata Szczepaniak (2008): On the way from phonology to morphology. German linking elements and the role of the phonological word. Morphology 18: 1–25. Schanen, François (2006): Prosodie luxembourgeoise. In: Susanne Craemer, Enrica Yvonne Dilk, Heinz Sieburg and Ferdinand Stoll (eds.), Europäische Begegnungen: Beiträge zur

304 

 Peter Gilles

Literaturwissenschaft, Sprache und Philosophie. Festschrift für Joseph Kohnen, 509–522. Luxemburg: Saint-Paul. Sturm, Hansjörg (1988): Die Entwicklung des Vokalsystems im Luxemburgischen. Magister thesis, Universität Freiburg. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Szczepaniak, Renata (2010): Phonologisch-typologischer Wandel des Deutschen und des Luxemburgischen im Kontrast. In: Antje Dammel, Sebastian Kürschner and Damaris Nübling (eds.), Kontrastive Germanistische Linguistik, vol. 1, 85–110. (Germanistische Linguistik 206/209.) Hildesheim: Olms. van Oostendorp, Marc (1998): Schwa in phonological theory. Glot International 3/5: 3–8. Wells, John Christopher (1982): Accents of English: An Introduction. Cambridge: Cambridge University Press. Wiese, Richard (2000): The Phonology of German. (The Phonology of the World’s Languages.) Oxford: Oxford University Press. WLM = Wörterbuch der luxemburgischen Mundart (1906): Luxemburg: Huss. Available at http:// engelmann.uni.lu:8080/portal/WBB2009/WLM/wbgui_py?mainmode=&lemid=&prefix=a&mode=&openwb=1. Zsiga, Elizabeth C. (2011): Local assimilation. In: Marc van Oostendorp, Colin J. Ewen, Elisabeth Hume and Keren Rice (eds.), The Blackwell Companion to Phonology. Vol. 3: Phonological Processes, 1919–1944. (Blackwell Companions to Linguistics Series.) Malden, MA: Wiley-Blackwell.

Steffen Höder (University of Kiel)

Low German: A profile of a word language* Abstract: This contribution claims that Modern Low German (as represented by North Low German dialects) is a rather prototypical word language according to the model provided by Auer (2001) and others. The interaction between syllable structure, stress, and phonemic alternations in different contexts is better explained as a consequence of word-related as opposed to syllable-related rules and restrictions. Apart from the relatively high complexity of possible consonant clusters at word boundaries, this view is supported by (a) the stress sensitivity of vocalic and consonantal syllable nuclei, including a highly differentiated vowel system, (b) word-level phonological processes such as word-medial obstruent voicing, and (c) the existence of a word-level suprasegmental phenomenon similar to a pitch accent. On the whole, Low German is even closer to the word language pole of the continuum between word and syllable languages than Standard German. The findings are also relevant in a wider perspective. First, it is of general importance to include dialectal or non-standard varieties in cross-linguistic typological studies and theoretical models. Second, some of the features found in Low German are also found in other non-standard varieties of (Northern) Germany as well as in neighboring languages, such as Danish (including South Jutlandic) and other Scandinavian and Circum-Baltic languages, which suggests an areal or contact-induced relation.

1 Introduction In this contribution, I investigate whether and to what degree Modern Low German is a word language according to the model by Auer (2001). The guiding hypothesis is that Low German has rather typical word-language features, similar to its closest relative and neighbor, High German. The study is based on a synchronic and intralingual analysis of the language’s phonology, although the focus is on features that are different from Standard German and other varieties of High German. While concentrating on dialects in the vicinity of Hamburg, I assume that the results are also representative for the larger proportion of the North Low German dialects, at least those spoken in Hamburg and Schleswig-Holstein. The * I wish to thank Stig Eliasson and Renata Szczepaniak for their valuable comments on an earlier version, and Jasmin Bliesemann for her help with my English. All remaining errors are, of course, mine.

306 

 Steffen Höder

empirical evidence comes mainly from a detailed survey of the dialect of Altenwerder (Höder 2010; for similar dialects cf. Kloeke 1913; von Essen 1958, 1964). While Middle Low German served as the official language of the Hanseatic League and a transnational lingua franca, today’s Low German is, from a socio­ linguistic point of view, a group of de-standardized and relatively diverse dialects. After a successive language shift towards High German throughout the last centuries, Low German is now mostly restricted to domains of everyday life, and it is normally not used as a written language.1 A recent study shows that 14% of the population (about 2.6 million) in the traditional Low German-speaking areas claim to be active speakers (Möller 2008: 33; for a comprehensive overview of the sociolinguistic situation cf. Föllner 2004), all of whom of course are bilingual in some variety of North High German. One consequence of this situation is that communication in a bilingual mode, code-switching, and the use of High German loanwords are very common. While this contribution does not focus on such language contact phenomena, established loanwords both from High German and from other languages are included. Recent additional contact-induced changes in Low German, which also affect the phonology, are discussed in Höder (2011b).

2 Low German as a word language 2.1 Criteria While the distinction between syllable and word languages is also relevant from the perspective of general phonetics and phonology (a possible question would be: is there a universally valid hierarchy of phonological units such as the phrase, the word, the foot, the syllable, the mora?), this volume aims at a typological classification based on a set of characteristic features of word and syllable languages. The distinction between the two types basically comes down to whether the syllable or the phonological word (abbreviated as ω) is the most prominent and/or relevant unit in the phonology of a language. Syllable and word languages are thought of as constituting a continuum. Prototypical syllable and word languages are expected to exhibit differences with respect to their word- and syllable-internal phonological structure, their phoneme inventories, the existence of suprasegmental features, and the scope of phonological rules and processes. Drawing on Trubetzkoy’s (1989 [1939]: 29) distinction between the delimitative and the culmi-

1 As a consequence, there are competing orthographies of Low German. All examples in this contribution are given in phonological transcription only.



Low German: A profile of a word language  

 307

native function of phonological units, different word-level features can be classified as boundary markers (such as consonant clusters in High German) or as markers of word coherence (such as the distribution of vocalic nuclei in a word), depending on whether they highlight the word by strengthening its edges or by reinforcing its internal structure. I use a selection of the criteria proposed, among others, by Auer (2001) and Szczepaniak (2007), chosen according to their relevance for and applicability to Low German. These criteria are summarized in Table 1. Table 1: Syllable vs. word languages prototypical syllable language

prototypical word language

stress

distinctive on phrase level

distinctive on word level

syllable structure

simple (preferably CV)

complex (example: consonant clusters)

vowel system

uniform

differentiated (e.g. stresssensitive)

phonological processes

syllable-related (example: resyllabification across word boundaries)

word-related (example: wordmedial allophones, invulnerable word boundaries)

tone (if existing)

syllable-related (one toneme per word-related (one toneme per syllable) word)

quantity (if existing)

uniform (distinctive in all syllables)

stress-sensitive or word-related (distinctive in stressed syllables)

long consonants

exist

do not exist

2.2 Phoneme system As a point of reference, the Low German phoneme inventory is given in Tables 2–4 (for the details of allophonic realization cf. Höder 2010). The consonant inventory is nearly identical to Standard German, except for some minor differences including the absence of phonemic affricates (since Low German has not undergone the High German Consonant Shift). While /r/ is classified as a vibrant in Table 2, it is frequently and increasingly realized as a uvular vibrant [ʀ], or a uvular or even velar fricative [ʁ ɣ]. The alveolo-palatal fricative /ʑ/ is the Low German equivalent to Standard German /j/ and has a wide range of allophones, including at least [j ʝ ʑ ʒ]. Therefore, it is not an exact voiced counterpart of /ʃ/, which is postalveolar and sometimes labialized ([ʃʷ]).

308 

 Steffen Höder

Table 2: Low German consonants labial

alveolar

plosives

p b

t d

fricatives

f v

s, sː z

nasals

m

n

lateral

l

vibrant

r

palatal

velar

glottal

k ɡ ʃ ʑ

x

h

ŋ

As for the vowels, the picture is more complex and deviates more strongly from Standard German. This is mainly due to the fact that the original vowel quantity is no longer distinctive. The phonemic organization of the vowel system is, as a consequence, based on qualitative distinctions. Compared to Standard German, Low German has six additional vowel qualities (another two distinctive heights and both front and back open vowels). Vowels that were long historically (/i y u e ø o/) still have long allophones in certain contexts. Generally, allophonic length (denoted by a length mark in brackets in Table 3) is restricted to stressed vowels (a) in positions before voiced consonants, (b) in word-final position, or (c) bearing the Knick phoneme (Knick is a suprasegmental phoneme that has been described as an additional quantity or a tonal feature; for details see section 2.5). In addition, the vocalization of postvocalic /r/ has led to the emergence of secondary vowel length, which is distinctive for /a/ and /ɔ/ (cf. /haːt/ ‘hard’ and its Standard German equivalent, hart /hart/). Finally, there are two more diphthongs than in Standard German (/eo̯/ and /ɛɪ̯/).2 Table 3: Low German monophthongs front [–rounded]

front [+rounded]

back

close

i(ː)

y(ː)

u(ː)

near-close

ɪ

ʏ

ʊ

close-mid

e(ː)

ø(ː)

o(ː)

open-mid

ɛ

œ

ɔ, ɔː

open

a, aː

ɒ(ː)

2 The quality of the diphthong /eo̯/ is a rather marked feature of the Altenwerder dialect. Other Hamburg dialects have /ɛo̯/; dialects in Schleswig-Holstein normally have /əo̯/.



Low German: A profile of a word language  

 309

Table 4: Low German diphthongs front

back

close-mid

eo̯

open-mid

ɛɪ̯

ɔi̯

open

au̯

ɑi̯

2.3 Stress sensitivity in segmental phonotactics 2.3.1 Syllable structure, word structure, and stress If, as hypothesized, Low German is a word language, then syllable structure should be stress-sensitive. It should be impossible to determine the structure of a possible syllable without reference to its position within the phonological word (or its position relative to stress). At first glance, this is obviously the case: There are countless examples of monosyllabic words such as /teo̯/ ‘too, overly’ or /dvaːx/ ‘dwarf’, which by definition represent both possible syllables and possible words, and we can, of course, construct examples that violate both syllable-related and word-related phonotactic rules (*/kmø/, */srofp/). There are also sound sequences that represent possible syllables, but impossible words (cf. the second syllables in /kri.ɡŋ/ ‘get-inf’ and /tsɪ.bl/ ‘onion’).3 Furthermore, some sound sequences clearly represent possible words, but it is hard, and controversial, to determine how many syllables they contain (such as /kom̂/ ‘come-inf’ or /fɪn̂/ ‘find-inf’; the circumflex here denotes the Knick phoneme discussed in section 2.5). At first sight, Low German syllable structure does not differ very much from Standard German. The most salient features are the consonant clusters /st sp sv sm sn sl/, which correspond to clusters beginning with /ʃ/ in Standard German, as well as some unique clusters, viz. /dv tv vr/. All of these clusters occur only word-initially or at the beginning of stressed syllables: (1) /stɛɪ̯n/ /sprok/ /svat/

‘stone’ ‘language’ ‘black’

3 Neither stress nor syllabicity is marked in the phonemic transcription throughout this article, as neither is considered phonologically distinctive; see below for details. Stress is marked, though, in words or utterances consisting of two or more phonological words. Glossing follows the Leipzig Glossing Rules (Comrie, Haspelmath, and Bickel 2008).

310 

 Steffen Höder

/smitn/ /snakŋ/ /slaŋ/ /dvaːs/ /tvɛɪ̯/ /vrak/

‘throw-inf’ ‘speak-inf’ ‘snake’ ‘across’ ‘two’ ‘wreck’

Furthermore, syllabic nasals and liquids (/m n ŋ l/ and /r/, realised as [ɐ]4) are very common. Unlike the traditional and presumably still default interpretation of syllabic consonants in Standard German, these cannot be analysed as */əC/ sequences in Low German. The two main arguments for a monophonemic ana­ lysis are (a) that there is no phonemic /ə/ due to a merger of older /ə/ and /ɪ/ (cf. /dɪkɪ/ ‘thick-f.sg’ and the corresponding Standard German dicke /dɪkə/), and (b) that no vowel is ever pronounced before the consonants in question, not even in very slow and careful speech. (2)

/zupm/ /zɪtn/ /zɛɡŋ/ /noɡl/ /votr/

‘drink-inf [alcohol]’ ‘sit-inf’ ‘say-inf’ ‘nail’ ‘water’

Generally, the syllabicity of these consonants is predictable (and hence non-phonemic) from the segmental context, viz. the fact that they are preceded by a consonant and followed by a word boundary or one or two obstruents (cf. /noɡlt/ ‘nail-3sg.pres’). However, the syllabicity of /l/ is difficult to predict in certain words such as /vrøɡlɪx/ ‘grumpy’, which is not syllabified according to the principle of onset maximization as *[ˈvrøː.ɡlɪ̞ç], but is rather trisyllabic ([ˈvrøːɡl̩ɪ̞ç]). In such examples, either the syllabicity of /l/ or the morpheme boundary before the adjectival derivative suffix /-ɪx/ must be granted phonemic status.5 A more detailed analysis reveals further differences between Standard German and Low German. One is that the overall system of syllable nuclei in Low German is more differentiated and stress-sensitive than the corresponding system

4 In some dialects, this postvocalic and syllabic allophone of /r/ has apparently merged with /a/, which not only gives rise to a different distribution of vocalic and consonantal nuclei (since more syllables contain a phonemic vowel /a/ than a phonemic consonant /r/), but may also lead to an expanded diphthong system (with a possible new class of monophonemic diphthongs and triphthongs emerging from tautosyllabic /Vr/ sequences, e.g. /oa̯ eo̯a̯/ < /or eo̯r/). It remains to be investigated how such a development affects the typological classification of Low German as a word or syllable language. 5 Alternatively, onset maximization could be seen as restricted to word-initial onsets.



Low German: A profile of a word language  

 311

in Standard German. The total number of potential syllable nuclei (abbreviated as X̩) is 26, including 16 monophthongs, 5 diphthongs, and 5 syllabic consonants. Of these phonemes, all but the consonants can occur in stressed syllables (21 nuclei), although stressed /ɒ/ is rare. In pre-tonic syllables, however, no more than five are possible, including four monophthongs and one consonant (/i ɪ ʊ a/ and /r/ [ɐ]): (3)

/bilʏtn/ /bɪkikŋ/ /tʊfel/ /kanɛɪ̯l/ /frɡetn/

[biˈlʏ̞tn̩] [bɪ̞ˈkʰikŋ̩] [tʰʊ̞ˈfeːl̞] [kʰaˈnɛɪ̯l] [fɐˈɡetn̩]

‘gradually’ ‘look.at-inf’ ‘too much, too many’ ‘cinnamon’ ‘forget-inf’

In post-tonic syllables, yet another group of nine potential nuclei occurs, including all of the consonants as well as four vowels (/i ɪ a ɒ/): (4) /truriɡɪ/ /lʏtɪ/ /fɪrmɒ/

[ˈtʰruːriɡɪ̞] [ˈlʏ̞tɪ̞] [ˈfɪ̞ɐ̯mɒ]

‘sad-f.sg’ ‘small-f.sg’ ‘company, firm’

As a result, both phonemic and allophonic vowel length as well as diphthongs (which could be analysed as bimoraic and hence as equivalent to long vowels in some underlying way, provided one adopted such a view) are restricted to stressed syllables. Only three nuclei can occur in any type of syllable, viz. /i ɪ a/. Table 5 shows the distribution of nuclei relative to stress (boldface indicates nuclei that are restricted to one position): Table 5: Stress-sensitive nucleus inventory

monophthongs

pre-tonic

stressed

post-tonic

iɪʊa

i y u ɪ ʏ ʊ e ø o ɛ œ ɔ ɔː a aː ɒ

iɪaɒ

eo̯ ɛɪ̯ ɔi̯ au̯ ɑi̯

diphthongs consonants

r

mnŋlr

Of course, names and loanwords (including loans from High German varieties) may exhibit different patterns, such as in the names of the months /au̯ˈɡʊs/ ‘August’ and /ɔkˈteo̯br/ ‘October’, where vowels that are in principle restricted to stressed syllables nevertheless appear in unstressed positions. However, the differences usually decrease with increasing phonological integration. These differences may be illustrated by the alternative forms of names and loanwords given in (5), where structural deviations are gradually reduced by adapting pre- and

312 

 Steffen Höder

post-tonic vowels as well as, in some cases, shortening the words to monosyllabic or trochaic feet: (5) /maːɡɒˈrɛɪ̯tɒ/ (equivalent to English Margaret) → /maɡaˈrɛɪ̯tɒ/, /ɡrɛɪ̯tɪ/, /mɛɪ̯tɒ/ /ʑeo̯ˈhan/ (equivalent to English John) → /ʑʊhan/ [ʑʊ̞ˈhan], /ʑɪhan/ [ʑɪ̞ˈhan] /preo̯ˈfɛsor/ (‘professor’) → /prʊfɛsr/ [pʰrʊ̞ˈfɛsɐ], /prfɛsr/ [pʰɐˈfɛsɐ] /kɔnfɪrmɒˈtʃeo̯n/ ‘Confirmation (Lutheran Church)’ → /kʊnfrmatʃeo̯n/ [kʰʊ̞nfɐmaˈtʃeo̯n]

The uneven distribution of nuclei between the different syllable types serves as a marker of word coherence. It also reinforces the culminative function of stress: The occurrence and position of the prominent syllable within a polysyllabic word can normally be predicted on the basis of the distribution of syllable nuclei, as only few nuclei can be both stressed and unstressed. However, it is possible to find or construct a few examples where the position of stress is unpredictable. The general rule in such cases is that the stress goes on the first syllable, but there are still some exceptions, most of which include a morpheme boundary (such as / bɪˈstɪk/ ‘cutlery’, with the unstressed prefix /bɪ-/): (6)

/bɪlɪx/ → [ˈbɪ̞lɪ̞ç] ‘cheap’ /bɪˈstɪk/ → [bɪ̞ˈstɪ̞k] ‘cutlery’ /balas/ → [ˈbalas] ‘ballast’ /paˈlas/ → [paˈlas] ‘palace’

Thus, the stress sensitivity of the syllable nuclei somewhat paradoxically results in the fact that stress itself, at least within single phonological words, is only marginally distinctive in today’s Low German.

2.3.2 Word-medial consonants Apart from the distribution of nuclei, there is further evidence to support the idea that the phoneme inventory of Low German is best described with reference to the phonological word rather than the syllable, namely the existence of word-based phonological rules. It is apparent that simple syllable-related rules can also have implications for the occurrence or non-occurrence of certain consonants at word boundaries. For example, as in Standard German, a final obstruent devoicing rule (Auslautverhärtung) prohibits voiced obstruents in the coda of a syllable, including word-final syllables. Still, there is no need to invoke an additional word-based rule to explain such restrictions. Such rules are needed, though, to account for the restriction of /h/ to word-initial onsets or the impossibility of word-initial /ŋ/, as in Standard German.



Low German: A profile of a word language  

 313

In addition, Low German has restrictions on word-medial consonants, i.e. consonants that cannot occur at word boundaries. One such example is the alveolar flap [ɾ], which can be interpreted as a word-medial allophone of /d/, even if it is in complementary distribution with both /d/ and /r/ and corresponds diachronically to either sound. It is restricted to a specific context, namely ˈV_Vω or ˈV_L̩ω, where L̩ stands for a syllabic liquid: (7)

/fadr/ ‘father’ → [ˈfaɾɐ] /mʊdr/ ‘mother’ → [ˈmʊ̞ɾɐ] /lɛdɪx/ ‘empty’ → [ˈlɛɾɪ̞ç] /haːdr/ ‘shepherd’ → [ˈhaːɾɐ] /tydln/ ‘tie-inf, tether-inf’ → [ˈtʰyɾl̩n]

A similar example is the phoneme /sː/, which has evolved from and is in complementary distribution with the sequence /st/. Phonemic long /sː/ is a characteristic feature of some dialects in the Hamburg area. In other areas, allophonic long [sː] may occur as a realisation of /st/ and alternate with [st]. This phoneme can only occur immediately after a stressed vowel and is always followed by an unstressed nucleus within the same word (ˈV_X̩ω): (8) /vɛsːn/ ‘west’ /zʏsːr/ ‘sister’ /bɛsːɪ/ ‘best-f.sg’

The restriction of this phoneme to a particular word-related, stress-sensitive position certainly qualifies as a typical word-language feature according to the criteria listed in Table 1. At the same time, however, long consonants are rather untypical features of prototypical word languages. While long /sː/ can be seen as a negligible quantity from a synchronic perspective, it shows that (and how) even in a word language typical syllable-language features can re-emerge diachronically, even if /sː/ is only one consonant and does not (yet) form part of a larger inventory of geminates as in classic syllable languages. All in all, the segmental phonotactics supports the hypothesis of Low German being a word language. The delimitative function of certain consonant clusters as markers of word boundaries is roughly equal to their Standard German equivalents. The stress sensitivity of the phoneme inventory and related phenomena, however, mark the coherence of the word as a phonologically relevant unit slightly more strongly than in Standard German.

314 

 Steffen Höder

2.4 Phonological processes 2.4.1 Word-medial obstruent voicing In addition to word-related distributional rules, there are also word-related phonological processes in Low German. One example is the word-medial obstruent voicing rule, a process which is similar to lenition rules in other German dialects: Voiceless obstruents are replaced by corresponding (or at least homorganic) voiced ones if preceded by a stressed vowel and followed by a syllable nucleus (ˈV_X̩ω). Word-medial obstruent voicing is frequent but optional in most dialects, including Altenwerder: (9) /klɔpm/ ‘knock-inf’ → [ˈkʰlɔpm̩], [ˈkʰlɔbm̩] /ritn/ ‘tear-inf’ → [ˈritn̩], [ˈridn̩] /dɪkɪ/ ‘thick-f.sg’ → [ˈdɪ̞kɪ̞], [ˈdɪ̞ɡɪ̞]

In other dialects, this process can be said to be obligatory. An example is the traditional dialect of Finkenwerder (cf. Kloeke 1913), where all voiceless obstruents in the relevant contexts have become voiced. This change has tremendous consequences within the phoneme system and leads, among other things, to the development of tertiary vowel length, since allophonic length before voiced consonants becomes phonemically distinctive. The Altenwerder contrast between / ritn/ [ˈritn̩] or [ˈridn̩] (with optional word-medial obstruent voicing) ‘tear-inf’ and /ridn/ [ˈriːdn̩] ‘ride-inf’ thus corresponds to a distinction in vowel length in Finkenwerder, whereby /ridn/ (with obligatory obstruent voicing) ‘tear-inf’ is still distinguished from /riːdn/ ‘ride-inf’. Additional restrictions and processes can apply which may vary across dialects. In Altenwerder, for instance, the voicing of /p/ (→ [b]) is restricted to contexts where it is preceded by a phonemically short monophthong (ˈV̆_X̩ω): (10) /pipm/ ‘cheep-inf’ → [ˈpʰipm̩], [ˈpʰibm̩] /klɔpm/ ‘knock-inf’ → [ˈkʰlɔpm̩], [ˈkʰlɔbm̩] /leo̯pm/ ‘run-inf’ → [ˈleo̯pm̩], *[ˈleo̯bm̩]

A similar rule restricts the voicing of /t/ → [d] to occurrences in prenasal position, whereas /t/-voicing results in [ɾ] in other contexts: (11) /smitn/ ‘throw-inf’ → [ˈsmidn̩] /smitr/ ‘thrower’ → [ˈsmiɾɐ]

From a word-based perspective, the alternation between voiced and unvoiced variants can be explained if we assume that there is a tendency towards voiced medial consonants in phonological words or, at least, word-internal trochaic feet



Low German: A profile of a word language  

 315

with medial consonants (ˈVCX̩ω). If such a word-based phonological template exists, it also highlights the coherence of the word and supports its importance as a salient phonological unit. Furthermore, a tendency towards voiced word-medial segments in turn implies that voiceless segments tend to be restricted to the word boundaries – a possible parallel to the role of sonority in syllable structure. The interesting question, then, is whether trochees with consonants other than stops also agree with the proposed pattern. It is worth mentioning that a narrow majority of all consonants are voiced per se and thus already fit into it. As for the remaining consonants, voiceless /h/ does not occur word-medially, and voiceless /sː/, /ʃ/ and /x/ do not have exact voiced counterparts. Word-medial /f/ is quite rare and mostly restricted to loanwords. Thus, it seems indeed reasonable to interpret word-medial obstruent voicing as a process that improves ˈVCX̩ω structures by affecting all voiceless obstruents that have an exact voiced counterpart, i.e. /p t k f s/, but not /sː ʃ x h/.

2.4.2 Stem-final consonants and postverbal /ɪk/ Additional evidence for such a word-based view comes from the case of unstressed postverbal /ɪk/, i.e. the first-person singular subject pronoun. Example (12) shows the infinitives of some verbs as compared to the first-person singular in the present tense: (12) /fâln/ ‘fall-inf’ /ʃrʊbm/ ‘scrub-inf’ /lezn/ ‘read-inf’ /zɛɡŋ/ ‘say-inf’

/ɪk ˈfâl/ ‘I fall-1sg.pres’ /ɪk ˈʃrʊp/ ‘I scrub-1sg.pres’ /ɪk ˈlês/ ‘I read-1sg.pres’ /ɪk ˈzɛx/ [ɪ̞k ˈzɛç] ‘I say-1sg.pres’

Generally the infinitive is formed by adding a nasal suffix, while the first-person singular has a zero affix. In some verbs, the infinitive or the finite form is also marked suprasegmentally (as in /lês/; see section 2.5). Apart from that, we see that there is a regular morphophonological alternation between stem-final voiced and unvoiced obstruents, similar but not identical to the alternations caused by the final devoicing rule in Standard German. While the /l/ does not change,6 stem-final /b/ in the infinitive alternates with /p/ in the present tense where it is syllable-final, /z/ alternates with /s/, and /ɡ/ alternates with /x/. However, if the subject pronoun /ɪk/ is placed after the finite verb, the stem-final consonants take on yet a different set of forms:

6 Stem-final /l/ does not alternate with another phoneme, but there is allophonic variation. Postvocalic /l/ is realised as a non-lateral approximant ([l̞]), i.e. without the articulators touching each other. This can be interpreted as the preliminary stage of a vocalization process.

316  (13)

 Steffen Höder

/ɪk ˈfâl/ ‘I fall’ /ɪk ˈʃrʊp/ ‘I scrub’ /ɪk ˈlês/ ‘I read’ /ɪk ˈzɛx/ ‘I say’

/ˈfal ɪk/ ‘fall I’ /ˈʃrʊb ɪk/ ‘scrub I’ /ˈlez ɪk/ ‘read I’ /ˈzɛx ɪk/ [ˈzɛç ɪ̞k] ‘say I’

Within a syllable-based model, we would have to explain this phonological difference between constructions with preverbal and postverbal /ɪk/ by some kind of two-step resyllabification process across the word boundary. First, the stem-final consonant would be reanalysed as the onset of the second syllable, according to the principle of onset maximization. In a second step, the final devoicing rule would be reversed (or simply would not apply), since the stem-final consonant is not in the coda (any longer): (14) /ˈfâl#.ɪk/ → /ˈfa.l#ɪk/ /ˈʃrʊb~p#.ɪk/ → /ˈʃrʊ.b~p#ɪk/ → /ˈʃrʊ.b#ɪk/ /ˈlêz~s#.ɪk/ → /ˈlê.z~s#ɪk/ → /ˈle.z#ɪk/

Such an approach works well for most verbs and consonantal alternations. However, it fails to account for the deviant behavior of /ɡ/ and /x/, since it predicts forms like */ˈzɛɡ ɪk/ instead of the correct /ˈzɛx ɪk/: (15) /ˈzɛɡ~x#.ɪk/ → ?/ˈzɛ.ɡ~x#ɪk/ → */ˈzɛ.ɡ#ɪk/

From a word-related perspective, the behavior of obstruents before postverbal /ɪk/ can be explained more neatly. A word-based approach would not assume a resyllabification followed by a reversal of final devoicing, but rather an expansion of the phonological word so as to incorporate the postverbal pronoun. Consequently, the word-medial obstruent voicing rule applies to the stem-final consonant where possible. Such a process not only results in the correct realisations of stem-final /b~p/ and /z~s/, but also leads to the correct unvoiced form of stem-final /x/, as this consonant is not affected by word-medial obstruent voicing:7 (16) /ˈfâl#ɪk/ → /falɪk/ /ˈʃrʊp#ɪk/ → /ʃrʊpɪk/ → /ʃrʊbɪk/ /ˈlês#ɪk/ → /lesɪk/ → /lezɪk/ /ˈzɛx#ɪk/ → /zɛxɪk/

7 The same rules also apply for parallel morphophonological alternations between /d/ and /t/ (/ lodn/ ‘load-inf’ → /lôt/ ‘load-1sg.pres’) as well as between /b/ and /f/ (/ʃribm/ ‘write-inf’ → /ʃrîf/ ‘write-1sg.pres’). Whether /b/ alternates with /p/ or /f/ is predictable from the preceding vowel. Other dialects have /v~f/ instead of /b~f/; in Altenwerder, /v/ is generally restricted to the onset of stressed syllables. In any case, word-medial obstruent voicing applies before postverbal /ɪk/.



Low German: A profile of a word language  

 317

Additional processes apply in these cases. Some of those, such as the absence of a glottal stop before the vocalic onset of /ɪk/ (cf. /falɪk/ [ˈfalɪ̞k] ‘fall I’ vs. /fâl#ˈɪk/ [fâl̞ˈʔɪ̞k] with contrastive stress on the pronoun), can be predicted by both the syllable-based and the word-based approach. Other phenomena, however, can only be explained as a result of an expansion of the phonological word, such as the loss of Knick in first-person forms with postverbal /ɪk/ (cf. /ɪk ˈlês/ ‘I read’ vs. /lezɪk/), which is in accordance with a phonotactic rule that restricts Knick to word-final syllables (see 2.5).

2.4.3 Expanding phonological words: Function words The notion of the expanding phonological word also proves useful for the analysis of other regular processes in which function words are integrated into a larger phonological unit. One example is the contraction of prepositions and articles (for similar mechanisms in other German varieties, cf. Kabak and Schiering 2006). Low German allows contraction of prepositions and articles in a much more systematic and regular way than Standard German. The Standard German inventory includes only a relatively small and fixed set of lexicalized contractions (e.g. im < in dem ‘in the-dat.sg.m’, aufs < auf das ‘on the-acc.sg.n’), all involving some form of the definite article. In contrast, almost any Low German preposition can be contracted with any oblique form of the definite or the indefinite article. As a result, there are regular paradigms, as illustrated in Table 6: Table 6: Preposition + article contractions def.m.sg (free form: /dɛn/)

def.f.sg, def.pl (free form: /dɛɪ̯/)

def.n.sg (free form: /dat/)

indf (free form: /ɛɪ̯n/)

/bi/ ‘at, by’

/bin/

/biɪ/

/bit/

/bin/

/ʏm/ ‘around’

/ʏm̂/

/ʏmɪ/

/ʏmt/

/ʏm̂/

/ɔp/ ‘on’

/ɔpm/

/ɔpɪ/

/ɔpt/

/ɔpm/

/mɪt/ ‘with’

/mɪtn/

/mɪtɪ/

/mɪt/

/mɪtn/

/ɪn/ ‘in’

/ɪn̂/

/ɪnɪ/

/ɪnt/

/ɪn̂/

Here, the article forms are phonetically reduced in a way that enables their incorporation into the preceding phonological word. The different forms illustrate different possibilities of achieving this aim. The neutral form /dat/ is reduced to /t/ which is attached to the coda of the last syllable of the preposition except where this coda already ends in /t/. The feminine and plural form /dɛɪ̯/ is reduced to the vowel /ɪ/, one of the vowels that are possible in post-tonic syllables. Word-medial obstruent voicing can apply optionally where possible (e.g. in /ɔpɪ/ [ˈɔbɪ̞] ‘on the’

318 

 Steffen Höder

or /mɪtɪ/ [ˈmɪ̞ɾɪ̞] ‘with the’). The most complex mechanism affects the masculine and indefinite forms /dɛn/ and /ɛɪ̯n/, which are (a) reduced to /n/ and attached to the preposition as a coda consonant if the preposition ends in a vowel (/bin/), or (b) reduced to a syllabic nasal and attached to a consonantal coda if it ends in an obstruent, including assimilation to homorganic stops (/mɪtn/, /ɔpm/), or (c) realised as a Knick on a preceding nasal (/ɪn̂/, /ʏm̂/). In any case, the results are perfectly normal phonological words, and the reduction of the articles can best be explained by means of the word-based phonotactic patterns that the contractions conform to. Similar processes are at work in the case of other contraction phenomena, such as the cliticization of definite articles after the conjunction /dat/ ‘that’ (/dat/ + /dɛɪ̯/ → /datɪ/ [ˈdaɾɪ̞], /dat/ + /dat/ → /datat/ [ˈdaɾat]) or the cliticization of demonstrative /dat/ ‘dem.n.sg’ after finite verb forms (/ɪs/ ‘be.3sg.pres’ + /dat/ → /ɪsːat/, /ʃas/ ‘shall.2sg.pres’ + /dat/ → /ʃasːat/). To sum up, the word-related phonological processes illustrated in this section show that the phonological word is indeed an important phonological unit, and even more so than in Standard German, as some morphophonological alternations cannot be explained sufficiently on the basis of syllable-related rules. However, it is also clear that the phonological word cannot be equated with the lexical word in all cases, since rules for the cliticization and contraction of certain function words imply an expansion of the phonological word beyond the boundaries of the lexical word.

2.5 Word-level suprasegmental: The Knick phoneme In addition to the segmental phenomena discussed so far, there is also a wordlevel suprasegmental phoneme in (North) Low German, which has been labelled “dragging tone” (German Schleifton), “overlength”, “accent 2”, or “tone 2”. This suprasegmental, a reflex of an apocopated or syncopated syllable in older forms, is found in lexical items such as /vît/ ‘willow’, which is phonetically identical to /vit/ ‘far’ in segmental terms, but suprasegmentally marked by at least a slightly different pitch contour and a longer segmental duration. This is reflected in the impressionistic label “Knick” (‘bend’), which is taken from several speakers’ description of the phenomenon.8 While it is commonly agreed that the proso-

8 For convenience, Knick is indicated by a circumflex (ˆ). Although this diacritic is identical to the IPA symbol denoting a falling contour, which would indeed be one possible interpretation of Knick, this usage is not meant to rule out other analyses.



Low German: A profile of a word language  

 319

demic distinction in Low German dialects is phonologically similar to the tonal system in, among others, Limburgish and Ripuarian dialects (cf. Gussenhoven and Peters 2004; Peters 2006), the exact phonological interpretation of this feature is rather controversial. The proposed approaches to the phonological interpretation differ in (a) whether the longer segmental duration or the characteristic pitch contour is considered distinctive, (b) whether it is assumed to affect only vowels or also nasal consonants, and (c) whether the vowel system in general is thought of as organized in terms of quantitative or qualitative distinctions. The classic view, as introduced by Bremer (1927), regards the pitch contour as the phonemically distinctive feature. As vowel length in general is seen as phonemic in this interpretation, it results in a system with binary vowel quantity (“long” vs. “short”) and a tonal distinction between two tonal accents or tonemes (“push tone” vs. “dragging tone”, or “toneme 1” vs. “toneme 2”). A similar view is held by Jakobson (1962 [1931]: 235), Ternes (2001: 180, 2006), and Prehn (2007).9 In contrast, a majority of studies support an interpretation which is based on a ternary quantitative contrast, including Knick as a third “overlong” quantity instead of a tonal feature (cf. von Essen 1958: 110–112, 1964: 10–11; Ternes 1981; Auer 1991: 24–25; Chapman 1993). In yet another analysis, Kohler (1986, 2001) argues that the vowel system is mainly organized in terms of qualitative distinctions and thus reduces the ternary quantity to a binary system, in which vowels bearing a Knick are analysed as “long” and hence equivalent to the “overlong” or “dragging tone” vowels in previous analyses. I myself have proposed an analysis that combines a qualitatively organized vowel system (as presented in section 2.2) with a prosodemic distinction which includes both tonal and durational aspects (cf. Höder 2010 for details). This more neutral point of view is reflected in the terminology used (“prosodeme 1” vs. “prosodeme 2 [= Knick]”). In this analysis, which is also applied here, Knick even includes the phonetically long nasals in words like /kom̂/ ‘come-inf’, as they also exhibit the characteristic pitch contour found in prosodemically marked vowels. A similar approach to long nasals is proposed by Prehn (2010). Even von Essen’s (1958: 111) analysis can be considered parallel, as it interprets long nasals as equivalent to “overlong” vowels.

9 In a subsequent study, though, Prehn (2011) claims not to find significant tonal differences and hence argues for the durational contrast to be distinctive. However, as Prehn’s work is based on recent recordings of (elderly) speakers from different places within the North Low German area, her findings may also reflect interdialectal differences or a (recent) loss of the tonal distinction. Further research is needed to clarify this issue.

320 

 Steffen Höder

As for the distribution of Knick, the most important rule states that Knick occurs exclusively in word-final syllables and is normally restricted to the stressed syllable: (17) /lŷt/ ‘people’ /bɪvîs/ [bɪ̞ˈvîːs] ‘prove-1sg.pres’ /klɪdôʃ/ [kʰlɪ̞ˈdôːʃ] ‘clothes, dress’

Its occurrence in syllables without (primary) stress is limited to compounds or pseudo-compounds, i.e. words whose (exceptional) phonological structure resembles compounds (regarding the distribution of nuclei, among other things) and which therefore could be analysed as consisting of two phonological words (cf. Raffelsiefen 2000): (18) /ˈʏnrˌɛ̂lf/ ‘Lower Elbe [part of the river]’ (< /ʏnr/ ‘under; lower’ + /ɛ̂lf/ ‘Elbe’) /ˈʃufˌlôt/ ‘drawer’ (< /ʃuf-/ ‘push-’ + /lôt/ ‘drawer, chest’) /ˈaːˌbɑ̂i̯t/ ‘work-1sg.pres’ (vs. /ˈaːˌbɑi̯t/ ‘work [noun]’) /ˈhaːˌbâːx/ ‘hostel, shelter’

Furthermore, Knick is subject to rather intricate phonotactic constraints on the segmental context. It is restricted to long monophthongs (/aː ɔː/), monophthongs with a long allophone (/i y u e ø o ɒ/), and diphthongs, and it can occur in open syllables as well as before fricatives, the stop /t/, or nasals: (19) /bâːx/ ‘recover [a ship]-1sg.pres’ vs. /baːx/ ‘mountain’ /vît/ ‘willow’ vs. /vit/ ‘far’ /lôt/ ‘drawer, chest’ vs. /lot/ ‘late’ /mêo̯t/ ‘fashion’ vs. /meo̯t/ ‘courage’

In addition, Knick can appear on any vowel if followed by (tautosyllabic) /l/: (20) /âl/ ‘all-pl’ vs. /al/ ‘already’

Finally, Knick can occur in syllable-final nasals, resulting in possible minimal triplets as its position within the syllable is distinctive in certain contexts: (21) /fɪn̂/ ‘find-inf’ vs. /fɪn/ ‘find-1sg.pres’ /min̂/ ‘mine-pl [weapons]’ vs. /mîn/ ‘mine [weapon]’ vs. /min/ ‘my-nom’

Thus, while Knick is attached to a specific phonetic segment within a syllable (the tone-bearing unit in a narrow sense) and its position within the syllable can vary, the scope of this suprasegmental feature is still the whole phonological word, i.e. the prosodemic contrast is only distinctive once per word. This is another indicator that the position of the word is the dominant phonological unit in Low



Low German: A profile of a word language  

 321

German. Furthermore, as Knick usually coincides with stress, it can also be said to reinforce the culminative function of stress. Knick plays a minor role in the inflectional morphology (cf. also section 2.4.2). In addition to (and often in combination with) other affixes, it serves as a plural marker, an infinitive marker, and a person-number marker: (22)

/dax/ ‘day’ vs. /dôx/ ‘day-pl’ (combined with vowel alternation) /rat/ ‘wheel’ vs. /rø̂t/ ‘wheel-pl’ (combined with vowel alternation) /kom̂/ ‘come-inf’ vs. /kôm/ ‘come-1sg.pres’ /left/ ‘live-3sg.pres’ vs. /lêft/ ‘live-pl.pres’

As most existing affixes in Low German are non-vocalic, i.e. they consist of coda consonants or consonantal nuclei, the use of a suprasegmental phoneme for inflectional purposes can be seen as part of a general tendency to avoid vocalic suffixes.10 The overall result is that inflected forms usually do not exhibit phonological structures that are different from those of uninflected words, which in turn results in an even more homogeneous structure of the phonological word. In summary, it can be said that the distribution of the Knick phoneme cannot be described or explained adequately without reference to the phonological word, and the occurrence of Knick can be interpreted as a coherence marker.

3 Areal perspective: North German, Scandinavian, and beyond The typological profile of Low German as a word language rather than a syllable language is, first of all, of descriptive value. Moreover, as it deals with a non-standardized, scarcely written variety, it provides a useful addition to the available descriptions of Germanic languages, in particular with respect to cross-linguistic studies. While non-standard varieties have generally been neglected by typologists in favor of more easily accessible standard languages, there is now a growing interest in dialectal data, and the importance of non-standard evidence for typological studies is increasingly acknowledged (cf. Auer 2004). This is evidently of particular relevance within the field of phonology and phonetics, as it is the non-standard varieties rather than the codified standards that are actually used by the speakers in everyday oral communication. On top of that, the word-language features of Low German, as outlined in section 2, are most likely not limited to the original Low German dialects, but probably apply to High German varieties spoken in Northern Germany as well.

10 The only exception is adjectival inflection, where we find a suffix /ɪ/ (with several functions).

322 

 Steffen Höder

Low German has been in constant contact with High German throughout its existence, and speakers have been shifting to High German, which they acquired as an L2, since early modern times. The majority of the population eventually shifted to High German in the middle of the 20th century (with many of their descendants becoming monolingual in High German). As a result, the colloquial North High German variety is heavily influenced by Low German structures in all parts of the grammar and the lexicon. The organization of the vowel system in North High German, for example, clearly follows the Low German model and represents what I have labelled the “diasystematic intersection” of both systems (Höder 2011b). The analysis of the Low German features may thus stimulate further research into North High German. Another field for which the phonological features of Low German and its status as a word language are highly relevant is the study of the areal or contact-linguistic relations between the German-speaking area and its northern neighbors (cf. Höder 2011a). North Low German is located in a transitional zone between High German and Danish. The Danish-German border region has been a region of intensive bi- and multilingualism, language contact, and language shift from at least the early Middle Ages until today, with a high degree of contact-induced convergence between the languages involved (High German, Low German, Standard Danish, the regional Danish variety South Jutlandic, and North Frisian). Indeed, there are striking phonological similarities between Danish and Low German which may, at least in part, be due to language contact. Parallel word-language features include for instance (a) differentiated, highly stress-sensitive vowel systems (cf. the analysis in Grønnum 1998), (b) some form of word-medial obstruent voicing, (c) the existence of a word-level suprasegmental phoneme (Low German Knick, Danish stød). South Jutlandic, while being stød-less, even employs a prosodemic distinction that is very similar to the Low German Knick, although it is normally referred to as a tonal contrast (cf. Bjerrum 1948; Ringgaard 1973: 25; Ejskjær 2005: 1723). In a wider perspective, Low German also forms part of a historical contact zone that encompasses the larger part of Northern Europe. The North Germanic subfamily of the Germanic languages is relatively closely related to Low German, and the contact between Low German and Continental Scandinavian in particular has been very influential in the history of the Scandinavian languages (Braunmüller 2004, this volume). Intriguingly, again, there are typological similarities between the phonology of Low German and the Scandinavian languages that could well be the result of contact-induced convergence, among them the relatively large vowel inventories and the pitch accent systems of Norwegian and Swedish. Furthermore, both Low German and Continental Scandinavian belong to the so-called Circum-Baltic languages, a proposed linguistic area consisting



Low German: A profile of a word language  

 323

of Germanic, Baltic, and Finnic languages, among others (for an overview cf. Koptjevskaja-Tamm and Wälchli 2001). Eliasson’s (2000) survey shows that these languages share some rather uncommon phonological features, including ones that are relevant to the distinction between word and syllable languages, such as large (and highly differentiated) vowel inventories and word-related systems of vowel and consonant quantity. So far, an explanation of these parallels is a desideratum. However, they too might be due to the intense contact within the Circum-Baltic area in historic and prehistoric times, including contact with Middle Low German. A more detailed knowledge of the phonology of its modern successor could provide more insight into the range and origin of this areal group.

4 Conclusion Low German indeed turns out to be quite a characteristic word language. On the whole, it seems to be even closer to the word language pole of the continuum between word and syllable languages than Standard German. It meets all the criteria summarized in Table 1, except for the fact that there is one long consonant, /sː/ (which, however, only occurs word-medially), and the observation that stress is only marginally distinctive. Its word-language features include (a) slightly more complex word-initial consonant clusters than in Standard German, (b) a fairly complex vowel system, which is more differentiated than the Standard German one, (c) a highly stress-sensitive system of syllable nuclei, (d) word-related phonological processes, particularly in relation to word-medial consonants, where in some cases a resegmentation of words is at work rather than resyllabification processes, and (e) a word-based suprasegmental distinction. The main difference between Low German and Standard German lies in the greater relevance of word coherence phenomena in Low German, while the marking of word boundaries is approximately equivalent in both languages. The findings are also relevant in a wider perspective. Some of the features found in Low German are also found in other non-standard varieties of (Northern) Germany as well as in neighboring languages, such as Danish (including South Jutlandic) and other Scandinavian and Circum-Baltic languages, which suggests an areal or contact-induced relation. This result underlines the importance of including dialectal or non-standard varieties in typological and cross-linguistic studies and may stimulate further research in this area as well.

324 

 Steffen Höder

References Auer, Peter (1991): Zur More in der Phonologie. Zeitschrift für Sprachwissenschaft 10: 3–36. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Auer, Peter (2004): Non-standard evidence in syntactic typology. Methodological remarks on the use of dialect data vs spoken language data. In: Bernd Kortmann (ed.), Dialectology Meets Typology. Dialect Grammar from a Cross-Linguistic Perspective, 69–92. (Trends in Linguistics. Studies and Monographs 153.) Berlin/New York: Mouton de Gruyter. Bjerrum, Marie (1948): Felstedmaalets tonale Accenter [The tonal accents of the Felsted dialect]. (Humanistiske Studier 3.) Århus: Universitetsforlaget. Braunmüller, Kurt (2004): Niederdeutsch und Hochdeutsch im Kontakt mit den skandinavischen Sprachen. In: Horst H. Munske (ed.), Deutsch im Kontakt mit germanischen Sprachen, 1–30. (Reihe Germanistische Linguistik 248.) Tübingen: Niemeyer. Bremer, Otto (1927): Der Schleifton im Nordniedersächsischen. Niederdeutsches Jahrbuch 53: 1–32. Chapman, Carol (1993): Überlänge in North Saxon Low German: Evidence for the metrical foot. An approach to vowel length based on the theory of metrical phonology. Zeitschrift für Dialektologie und Linguistik 60/2: 129–157. Comrie, Bernard, Martin Haspelmath and Balthasar Bickel (2008): Leipzig Glossing Rules. Conventions for interlinear morpheme-by-morpheme glosses. Available at http://www.eva. mpg.de/lingua/resources/glossing-rules.php. Ejskjær, Inger (2005): Dialects and regional linguistic varieties in the 20th century III: Denmark. In: Oskar Bandle, Kurt Braunmüller, Ernst Håkon Jahr, Allan Karker, Hans-Peter Naumann and Ulf Teleman (eds.), The Nordic Languages. An International Handbook of the History of the North Germanic Languages, vol. 2, 1721–1741. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 22/2.) Berlin/New York: Walter de Gruyter. Eliasson, Stig (2000): Typologiska och areallingvistiska aspekter på de nordeuropeiska språkens fonologi [Typological and areal linguistic perspectives on the phonology of the northern European languages]. In: Ernst Håkon Jahr (ed.), Språkkontakt. Innverknaden frå nedertysk på andre nordeuropeiska språk. Forskingsprogrammet Norden og Europa [Language contact. The influence from Low German on other languages in northern Europe. The research programme ‘The North and Europe’], 21–70. (Nord 2000/19.) Copenhagen: Nordisk Ministerråd. Essen, Otto von (1958): Die Vokale der niederdeutschen Mundart von Kirchwerder. Zeitschrift für Phonetik und allgemeine Sprachwissenschaft 11: 105–118. Essen, Otto von (1964): Kirchwerder bei Hamburg. (Lautbibliothek der deutschen Mundarten 33/34.) Göttingen: Vandenhoeck & Ruprecht. Föllner, Ursula (2004): Zum Gebrauch des Niederdeutschen in der Gegenwart – soziolinguistische und pragmatische Aspekte. In: Dieter Stellmacher (ed.), Niederdeutsche Sprache und Literatur der Gegenwart, 99–148. (Germanistische Linguistik 175/176.) Hildesheim/ Zürich/New York: Olms.



Low German: A profile of a word language  

 325

Grønnum, Nina (1998): Fonetik og fonologi. Almen og dansk [Phonetics and phonology. General and Danish]. Copenhagen: Akademisk forlag. Gussenhoven, Carlos and Jörg Peters (2004): A tonal analysis of Cologne Schärfung. Phonology 21: 251–285. Höder, Steffen (2010): Das Lautsystem des Altenwerder Platt. Eine phonetisch-phonologische Bestandsaufnahme. Niederdeutsches Wort 50: 1–27. Höder, Steffen (2011a): Dialect convergence across language boundaries: A challenge for areal linguistics. In: Frans Gregersen, Jeffrey K. Parrott and Pia Quist (eds.), Language Variation – European Perspectives III. Selected papers from the 5th International Conference on Language Variation in Europe (ICLaVE 5), Copenhagen, June 2009, 173–184. (Studies in Language Variation 7.) Amsterdam/Philadelphia: Benjamins. Höder, Steffen (2011b): Niederdeutsch und Norddeutsch: ein Fall von Diasystematisierung. Niederdeutsches Jahrbuch 134: 113–136. Jakobson, Roman (1962): Über die phonologischen Sprachbünde. In: Selected Writings. Vol. 1: Phonological Studies, 137–143. ’s-Gravenhage: Mouton. First published in Réunion Phonologique Internationale tenue à Prague, Prague 1931. Kabak, Barış and René Schiering (2006): The phonology and morphology of function word contractions in German. Journal of Comparative Germanic Linguistics 9: 53–99. Kloeke, Gesinus (1913): Der Vokalismus der Mundart von Finkenwärder bei Hamburg. Hamburg: Gräfe & Sillem. Kohler, Klaus J. (1986): Überlänge und Schleifton im Niederdeutschen. Zusammenfassung der Ergebnisse aus vier Dialektuntersuchungen. In: Klaus J. Kohler, Regina Tödter and Michael Weinhold (eds.), Phonetische Forschung in der niederdeutschen Dialektologie, 5–17. (Arbeitsberichte des Instituts für Phonetik der Universität Kiel 23.) Kiel: Universität Kiel, Institut für Phonetik. Kohler, Klaus J. (2001): Überlänge im Niederdeutschen? In: Robert Peters, Horst P. Pütz and Ulrich Weber (eds.), Vulpis Adolatio. Festschrift für Hubertus Menke zum 60. Geburtstag, 385–402. (Germanistische Bibliothek 11.) Heidelberg: Winter. Koptjevskaja-Tamm, Maria and Bernhard Wälchli (2001): The Circum-Baltic languages. An areal-typological approach. In: Östen Dahl and Maria Koptjevskaja-Tamm (eds.), The Circum-Baltic Languages. Typology and Contact. Vol. 2: Grammar and Typology, 615–750. (Studies in Language Companion Series 55.) Amsterdam/Philadelphia: Benjamins. Möller, Frerk (2008): Plattdeutsch im 21. Jahrhundert. Bestandsaufnahme und Perspektiven. (Schriften des Instituts für niederdeutsche Sprache 34.) Leer: Schuster. Peters, Jörg (2006): The dialect of Hasselt. Journal of the International Phonetic Association 36: 117–124. Prehn, Maike (2007): Schwa loss and its results in Low German. Tone or overlength? Linguistics in the Netherlands 2007: 187–198. Prehn, Maike (2010): Die langen finalen Nasale im Nordniedersächsischen. Ihre Phonetik und phonologische Repräsentation. In: Matthias Katerbow and Alexander Werth (eds.), Moderne Regionalsprachen als multidimensionales Forschungsfeld, 187–208. (Germanistische Linguistik 210.) Hildesheim/Zürich/New York: Olms. Prehn, Maike (2011): Vowel quantity and the fortis/lenis distinction in North Low Saxon. Amsterdam: Landelijke Onderzoekschool Taalwetenschap. Raffelsiefen, Renate (2000): Evidence for word-internal phonological words in German. In: Rolf Thieroff, Matthias Tamrat, Nanna Fuhrhop and Oliver Teuber (eds.), Deutsche Grammatik in Theorie und Praxis, 43–56. Tübingen: Niemeyer.

326 

 Steffen Höder

Ringgaard, Kristian (1973): Danske dialekter. En kortfattet oversigt [Danish dialects. A concise survey]. Copenhagen: Akademisk forlag. First published Århus 1971. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Ternes, Elmar (1981): Über Herkunft und Verbreitung der Überlänge in deutschen Dialekten. In: Wolfgang U. Dressler, Oskar E. Pfeiffer and John R. Rennison (eds.), Phonologica 1980. Akten der Vierten Internationalen Phonologie-Tagung. Wien, 29. Juni–2. Juli 1980, 379–386. (Innsbrucker Beiträge zur Sprachwissenschaft 36.) Innsbruck: Universität Innsbruck, Institut für Sprachwissenschaft. Ternes, Elmar (2001): Ansätze zu einer Phonemtypologie deutscher Dialekte. In: Margret Bräunlich, Baldur Neuber and Beate Rues (eds.), Gesprochene Sprache – transdisziplinär. Festschrift zum 65. Geburtstag von Gottfried Meinhold, 171–182. (Hallesche Schriften zur Sprechwissenschaft und Phonetik 5.) Frankfurt am Main: Lang. Ternes, Elmar (2006): Tone reversal in Franconian and elsewhere. North-Western European Language Evolution 48: 91–109. Trubetzkoy, Nikolaus S. (1989): Grundzüge der Phonologie. 7th ed. Göttingen: Vandenhoeck & Ruprecht. First published Prague 1939.

Beat Siebenhaar (University of Leipzig)

Phonological and phonetic considerations for a classification of Swiss German dialects as a word language or a syllable language [b̥lɪt͜st͜st͜st͜sʊ:g̊] Blitzt s z Zug? ‘Is there lightning in (the town of) Zug?’

Abstract: In his seminal publication, Auer (1993) asks whether the model of word languages and syllable languages can be reshaped such that reference to duration is avoided altogether. This paper answers the question in the negative, as segmental aspects of the different sound classes, their interaction, and the interaction with prosodic aspects have a clear influence on the temporal structure of languages and therefore on prosody. As a result, phonetic aspects can give additional evidence for the typological distinction of word languages and syllable languages. After a discussion of phonetic correlates of phonological differences and considering the typological status of the Swiss German dialects, this is exemplified with some analyses of a corpus of spontaneous Swiss German dialect data. Resyllabification without qualitative modification, the duration of schwa, and intonation peak alignment are examined. The analyses show that in the Swiss German dialects phonetic aspects of processes that strengthen the syllable are stronger than the processes that mark the word boundary, albeit the results are not entirely straightforward. Moreover, the definition of the syllable has a considerable impact on the typological assignment of the Swiss German dialects (and other languages) to syllable or word languages.

1 Introduction Swiss German dialects have often been interpreted as syllable languages, as opposed to Standard German which is said to be a word language (Nübling and Schrambke 2004; Szczepaniak 2007: 317–325). The example above demonstrates the problems that arise with the classification of a language as a word language or a syllable language, and it raises the question of whether a language with a sentence of four words that contains only two vowels can be a syllable language. I will try to obtain some answers on the basis of phonological and phonetic differences between Standard German and Swiss German dialects. Furthermore, some

328 

 Beat Siebenhaar

prosodic analyses will show that phonetics should again be considered for the typological discussion. Nübling and Schrambke (2004) present a distinction of Northern Alemannic and Southern Alemannic based not on perceptually or historically important isoglosses (as it is usually done), but instead on aspects representing a typological difference between word and syllable languages. The distinction fits very well into my timing model for a dialectal speech synthesis system, emerging from the concept in the model for Swiss Standard German (Siebenhaar, Zellner Keller, and Keller 2001). The statistical model of Swiss German dialects’ timing in this project was more effective when the syllable boundaries based on the sonority hierarchy alone were taken into account, rather than a phonologically defined syllable boundary. Moreover, considering word boundaries did not alter the model fit (Siebenhaar 2004a, 2004b; Leemann and Siebenhaar 2007). Therefore, this statistical model gives additional evidence for a definition of Swiss German dialects as syllable languages. However, this is a phonetic model while Nübling and Schrambke’s (2004) distinction is based more on phonological differences. I will therefore go back to the roots of the distinction of syllable-timed and stress-timed languages. I do not, however, adhere to a purely phonetic model as do Ramus, Nespor, and Mehler (1999) or Low, Grabe, and Nolan (2000), who calculate duration relations of vocalic and consonantal segments, but instead focus on the phonetic differences of the phonological aspects that are thought to distinguish word languages from syllable languages. Finally, some phonetic evidence is given to support the classification of Swiss German dialects as syllable languages. In his seminal publication, Auer (1993) presents a list of prosodic aspects that can account for typological differences between word and syllable languages. The following list (Table 1) is enhanced by features mentioned by Nübling and Schrambke (2004: 284–285) and Szczepaniak (2007: 52–53). The features mentioned at the top are phonological in nature, while features at the bottom of the list are more phonetic consequences. This ordering takes into account that a clear-cut differentiation of phonetics and phonology is not possible (Siebenhaar and Leemann 2012).



Phonological and phonetic considerations  

 329

Table 1. Prosodic aspects defining a typological difference between word and syllable languages Prosodic aspects defining a typological difference between word and syllable languages syllable type more phonological

variable or fixed syllable boundary respecting the sonority hierarchy differences of the vowel system in stressed and non-stressed syllables use of word-delimiting allophones epenthesis used to optimize the word or syllable boundaries vowel elision used to optimize the word or syllable boundaries presence of geminates resyllabification in syllable languages presence of external sandhi phenomena

more phonetic

regular or irregular reanalysis difference or lack of difference between heavy and light syllables lexicalized or rule-based stress attribution presence of word stress compression of syllables within a foot or within a syllable isochrony on stress or syllable level different reduction phenomena different phonological processes with an increase of the speech rate

2 Phonetic correlates of phonological differences To define the typological difference between syllable languages and word languages, this paper focuses on phonological aspects. Therefore, it goes back to the phonetic roots of the distinction between syllable-timed and stress-timed languages (Pike 1946; Abercrombie 1967); in other words, it supports the phonological theory with phonetic evidence on the prosodic level. It is obvious that most phonological differences are reflected in phonetics, and many of the phonetic differences also affect prosody: The majority of the typological differences are related to syllable structure, but there are still some that are entirely segmental. But both suprasegmental and segmental aspects may have an inherent influence on the temporal dimension of speech. Concerning the typological difference, segmental and suprasegmental aspects should therefore not be separated. A well-known example illustrates the segmental impact on prosody. Since the 1950s it has been known (cf. the summaries of Klatt 1976 and van Santen

330 

 Beat Siebenhaar

1992) that high vowels have a shorter intrinsic duration than low vowels, and high vowels are generally uttered with a higher pitch than low vowels. These effects of vowel height are segmentally conditioned, and therefore are within the realm of segmental phonetics. However, the segmental aspect has a clear influence on the temporal structure of languages and therefore on prosody, which may be taken into account when making a typological distinction. In the following, I discuss which prosodic consequences result from the phonologically defined differences listed in Table 1: syllable type, sonority hierarchy, vocalism differences in stressed and unstressed syllables, phonotactics, epenthesis and elision, and sandhi. Resyllabification and the reduction of unstressed syllables are discussed in section 3, where prosodic consequences of the phonologically defined differences are empirically tested. The syllable type directly affects the temporal structure; in word languages, the syllable structure tends to be more variable and complex in stressed syllables compared to a relatively simple structure in unstressed syllables. Syllable languages, on the other hand, tend to have identical syllable structures in stressed and unstressed syllables. As a consequence, the different relation between vowels and consonants has been taken up by different models to test the isochrony hypothesis on a purely temporal level (Ramus, Nespor, and Mehler 1999; Low, Grabe, and Nolan 2000). Using controlled laboratory data it was shown that languages of different rhythmic types (syllable-timed, stress-timed, and mora-timed languages) have different relations to the succession of vowels and consonants. While syllable languages more or less respect the sonority hierarchy, word languages violate the sonority hierarchy in various ways. This distinction is mainly observable at the word level. However, the phonological word is often unstable in cases of complex words or utterances. Furthermore, it is crucial to look at utterances of spontaneous speech because communication happens in utterances, not isolated words. When word boundaries are disregarded in spontaneous speech in favor of the sonority hierarchy, it results in many resyllabification processes, which can resolve some – but not all – of the perturbing syllable contacts. Additionally, they can be affected by sandhi phenomena (cf. Moulton 1986). However, resyllabification can also be found without sound quality changes on the segmental level. In these cases, a word-final consonant is delinked from the phonological structure of the word without being removed from the melodic stream and relinked with the following syllable. Thus, [ɛr hɛt əs ˈoʊ̯to] ‘he has a car’ becomes [ɛr hɛ.tə.sˈoʊ̯.to].1 A word-final consonant therefore becomes a syllable-initial

1 The Handbook of the International Phonetic Association (1999) proposes placing the accent sign before the accented syllable. However, the definition of the syllable boundary is one of the



Phonological and phonetic considerations  

 331

consonant of the following word. Conversely, this resyllabification process also occurs when a word-initial consonant, mainly a fricative before a stop, is linked to the coda of the preceding word. Thus, [ɛr g̊ˈɑːt ɪ ʃtˈɑl] ‘he goes into the stable’ becomes [ɛr g̊ˈɑː.tɪʃ.tˈɑl]. The position of a consonant in the onset or the coda of a syllable has a considerable influence on its duration because syllable-onset consonants are shorter than syllable-final consonants, as shown in section 3.1. Consequently, a different observance of the sonority hierarchy directly affects the temporal structure. If the vocalism is different in stressed and non-stressed syllables in word languages, this has an influence on the temporal structure. In German, long and short vowels are found in stressed syllables, but in unstressed syllables there are only short vowels. If a long vowel of a stressed syllable shifts into an unstressed position, it is phonologically shorted. Phon [foːn] is shortened when it occurs in unstressed position: Phonologie [fonologˈiː]. The duration of stressed syllables is therefore more variable than unstressed syllables; the temporal structure is directly affected. Word languages show word delimiting markers, which are usually different allophones or extrasyllabic consonants (in German, mainly [s]) in word-internal or word-final position. In Standard German, all obstruents in word-final position are unvoiced. As the voiced obstruents have a shorter intrinsic duration than their unvoiced counterparts (Nocchi and Schmid 2006; Kuzla, Cho, and Ernestus 2007), devoicing directly affects the temporal structure. Epenthesis and vowel elision have different functions in word and syllable languages. Depending on the language type, they optimize the word boundaries or the syllable structure. In Standard German, a word-initial vowel is usually preceded by an initial glottal stop while in word-internal position a glottal stop only occurs under specific circumstances. The glottal stop epenthesis affects the rhythmic structure of voiced and unvoiced parts of an utterance (Dellwo, Fourcin, and Abberton 2007). In word languages, vowel deletion is the extreme reduction of non-accented syllables. The reduction of Standard German [rˈɛtən] to [rˈɛtn̩] ‘to save’ is a classic example. This vowel deletion results in complex consonant clusters at the word boundary that reduce the duration of the voiced part of a word and, consequently, the rhythmic structure.

problems dealt with in this article. Therefore, the accent sign is set before the vowel of the accented syllable as it is partially done in the SAMPA transcription and also in the new wordbook on German orthoepy (Krech et. al. 2009).

332 

 Beat Siebenhaar

If a language has word-internal geminates, there are a greater number of phonetically longer consonants than in languages without geminates. Again, the temporal relation of the vowel-consonant succession is different. If we have external sandhi, which is typical of syllable languages, latent word-final consonants are reinstalled immediately before a vowel, as in French liaison: ton père vs. ton oncle ([tõ pɛʁ] ‘your father’ vs. [tõ.nõŋ.klə] ‘your uncle’). Before a consonant, assimilation processes can replace one segment with another across word boundaries: Swiss German si händ ball gschpilt /z̥i hænd̥ b̥ɑl kʃpilt/ becomes [z̥i hæm pɑl kʃpilt] ‘they were playing (foot)ball’ where first the two lenis consonants fuse to a fortis plosive and then the place of articulation of the nasal is assimilated (Moulton 1986; Fleischer and Schmid 2006: 248–249). Consonant epenthesis as in the French example changes the rhythmic succession of consonants and vowels. Consonant assimilation in the Swiss German dialect example also changes the rhythmic structure, as different segments have different intrinsic durations. Without reiterating all of the aspects which differentiate word languages from syllable languages listed by Auer (1993) and Nübling and Schrambke (2004), it becomes clear that segmental differences and varying phonotactic restrictions have a direct influence on the temporal structure of a language. However, it is not clear whether these segmental aspects result in a different rhythmic concept, or whether the rhythmic concept of a language entails the phonological structure. This seems to be the question of which came first, the chicken or the egg, which cannot be satisfactorily answered. These comments on the temporal effect of the different features are in some way a direct answer to Auer’s (1993: 89) question of whether the model of word languages and syllable languages can be reshaped such that reference to duration is avoided altogether. I do not think that this link to phonetics should be abandoned. The correlation to phonetics, as exemplified in section 3, strengthens the model.

3 Swiss German dialect – A syllable language? Nübling and Schrambke (2004: 304) assign the Swiss German dialects to the pole of syllable languages. Szczepaniak (2006, 2007: 317–325) supports this position and shows that Standard German has undergone a typological drift toward a word language. However, the assignment of the two varieties to the different poles is not straightforward throughout. Some divergent examples of Standard German and Swiss German dialect equivalents are discussed below with regard to the word/syllable language type differentiation.



Phonological and phonetic considerations  

 333

3.1 Word-related features in Swiss German dialects Comparing Standard German and Swiss German dialects with regard to syllable structure, Swiss German dialects often do not behave like syllable languages. Syllable languages tend to have simple CV syllables. However, Swiss German dialects show even more complex consonant clusters than Standard German. The following examples are cases where Swiss German dialects behave more like word languages than Standard German. The nearly systematic syncope of schwa in Swiss German word-initial syllables results in word-initial onsets that are more complex than in corresponding Standard German words ([gəʃpˈɛnst] vs. [kʃpæŋʃt] ‘ghost’, [gəfʁˈaːkt] vs. [kfrˈœːg̊t] ‘asked’). Also, the coda is not necessarily simpler in the Swiss German dialects than it is in the Standard German equivalent: The affricate [k͜x], which is – besides [x] – a usual variant of Standard German [k], is more complex than the simple [k] of Standard German ([gəʃtrˈɪkt] vs. [kʃtrɪk͜xt] ‘knitted’). Moreover, schwa + /r/ is not vocalized in Swiss German dialects while there is a rule-based vocalization in Standard German (Krech et al. 2009: 87). In word-final position of Swiss German /ər/, the schwa is even often elided ([pʰˈeːtɐ] vs. [pʰˈeːtər]/[pʰˈeːtr̩] ‘Peter’). In addition, many Standard German bisyllabic plural forms have open syllables while the corresponding Swiss German plural form is often monosyllabic with a closed syllable ([ʃtˈyːlə] vs. [ʃtyə̯l] ‘chairs’; [hˈʊndə] vs. [hʏnd̥] ‘dogs’). The Swiss German dialects even have grammatical words without a vowel ([das] vs. [d̥s]/[s] ‘the’ (nom/acc.sg.neut); [t͜suː] vs. [t͜s] ‘to, in’). The neuter article ds or s and the preposition z are therefore common proclitics (Nübling 1992: 240), resulting in very complex syllable onsets ([das gəʃpˈɛnst] vs. [d̥skʃpæŋʃt] ‘the ghost’, [t͜suː t͜suːk] vs. [t͜st͜sʊːg̊] ‘in Zug’). Word languages show more violations of the sonority hierarchy than syllable languages. Therefore, if Swiss German dialects are indeed closer to a syllable language, they should follow the sonority hierarchy more than Standard German. The examples mentioned above show that Swiss German violates the sonority hierarchy just as Standard German does, especially in syllable onsets where the consonant cluster is even more complex than in Standard German ([gəʃpˈɛnst] vs. [kʃpæŋʃt] ‘ghost’; [gəʃtʁˈɪkt] vs. [kʃtrɪk͜xt] ‘knitted’). In the syllable coda, Swiss German dialects and Standard German show similar violations ([hɛʁpst] vs. [hɛrb̥ʃt] ‘autumn’). However, if the fricatives [s] and [ʃ] are defined as syllabic consonants and therefore as syllable nuclei, or as part of a semisyllable as found in Bella Coola, Georgian, and Polish (Cho and Halloway King 2003), this would eliminate many of these violations in the Swiss German dialects and, to a lesser extent, Standard German. Therefore, depending on the definition of the syllable, Swiss German respects the sonority hierarchy less than or equal to Standard German.

334 

 Beat Siebenhaar

Word languages have reduced vowel systems in stressed and unstressed syllables. The vowel of unstressed syllables in Standard German is normally [ə]. Exceptions are full vowels in derivational suffixes such as -heit, ‑bar, etc., in the determinatum of compounds which consist of more than one phonological word and some pretonic vowels in three-syllable words such as Forelle [foʁˈɛlə] ‘trout’. In many cases where schwa is not elided, the corresponding Swiss German vowel of Standard German schwa is also a schwa, albeit with a less central quality: [a, ɑ, æ, ɘ, ɜ, ɵ] (SDS III/1). Furthermore, Swiss German dialects have [i] in inflection endings (e.g. as a conjunctive marker in mer säit, er machi Ferie ‘one says he makes (pres.conj.3.sg.act) holidays’ or in e schööni Souerei ‘a nice (nom/acc. sg.fem) mess’). Moreover, the alpine dialects have a very complex distribution of different vowel qualities in reduced syllables. Besides reduced schwa, the linguistic maps on the plural endings of normal verbs (SDS III/34) give evidence for -i, -a, -ä ([æ]) and -u. Thus, Swiss German dialects, like a word language, have a reduced vowel system in unstressed syllables. However, the reduction is not as advanced as in Standard German.

3.2 Syllable-related features in Swiss German dialects On the other hand, the Swiss German dialects correspond more to a syllable language regarding word-final closed schwa syllables. In this position, Standard German has a schwa syncope in most cases, which results in a suboptimal syllable nucleus with a syllabic consonant. Conversely, Swiss German dialects have a systematic apocope of word-final -n resulting in open syllables, which optimizes the syllable structure ([zˈaːgn̩] vs. [z̥ˈæg̊ə] ‘to say’). In spoken Standard German, the non-optimal word-final syllable is very often subject to further reduction and assimilation processes that result in a monosyllabic word ([zˈaːgn̩] > [zˈaːgŋ̍] > [zaːŋ] ‘to say’). Resulting in a CVC structure, these processes strengthen the word boundary. In contrast, the Swiss German form has two prototypical CV syllables ([z̥ˈæg̊ə] ‘to say’). Therefore, in this first central aspect of syllable structure differentiating both German varieties, Swiss German dialects are not clearly positioned as syllable languages. Word languages tend to alternate allophonic and neutralized variants of a phoneme depending on their position within a word. These word-related processes, such as the classic example of final-obstruent devoicing in Standard German (Hall 2011: 51–54; for other languages see Auer 1993: 72–74), accentuate the boundaries of a word. Syllable languages do not show different allophonic or neutralized variants at word boundaries. Standard German has these differ-



Phonological and phonetic considerations  

 335

ent allophones at both edges of a word: Word-initial vowels are usually marked with a glottal stop, which can be seen as a glottalized allophone of a vowel or an epenthesis (see next paragraph), and in word-final position all obstruents are devoiced. Both processes highlight the word boundaries. Swiss German dialects do not have a glottal stop at all and they do not show final obstruent devoicing as they completely lack voiced obstruents. The lack of these word-boundary-marking allophones in Swiss German dialects triggers or allows a massive resyllabification process across word boundaries. In some cases, these resyllabification processes following a syllable-optimizing n-epenthesis even result in the reanalysis of nouns (e.g. en Eber, de Eber > e Neber, de Neber ‘a boar, the boar’, en Igel, de Igel > e Nigel, de Nigel ‘a hedgehog, the hedgehog’). In word languages and syllable languages, elision and epenthesis have different functions. In word languages they optimize the word, while in syllable languages they optimize the syllable. The examples here show that Standard German and Swiss German dialects both show epenthesis and elision. However, Swiss German shows clear tendencies toward syllable optimization while Standard German tends toward an optimization of the word. Standard German has glottal stop epenthesis before word-initial stressed vowels. In the Swiss German dialects, there is a syllable-initial epenthesis of n and r. The dialectal article [ə] before word-initial consonant ([aɪ̯n.hʊnt] vs. [ə.hʊnd̥] ‘a dog’) turns into [ən] before a word-initial vowel, while the vowel in Standard German is preceded by a glottal stop ([aɪ̯n.ʔˈaː.bn̩t] vs. [ə.nˈɑː.b̥ə] ‘an evening’). Sometimes the dialectal hiatus is avoided by inserting [r] ([deːn.ʔˈa.pfl̩] vs. [d̥ə.rˈœ.pfəl] ‘the apple’). This process is not a reinstallation of a historical sound; it is purely phonologically motivated. This is obvious in the example [d̥ə.rˈœ.pfəl] where there is an insertion of r also in the accusative, where historically there was an n. Though the Standard German glottal stop is mainly a strengthening of the word boundary, it is consequently also a strengthening of the word-initial syllable. In Swiss German, the glottal stop is not part of the sound system. Here, the hiatus is avoided either by the elision of one of the adjacent schwa sounds or by inserting an epenthetic n ([ʔeːɐ̯.gˈiːb̥t.ʔiːɐ̯.ʔaɪ̯n.gə.ʃˈɛŋk] vs. [ɛr.g̊ˈɪ.tə.rə͜əs.kʃˈæŋk͜x]/[ær.g̊ˈɪ.tə.rə.nəs.kʃˈæŋk͜x] ‘he gives her a present’). In both cases, the syllable is optimized and the morpheme structure becomes opaque. Concerning the word-final boundary, in Standard German schwa elision strengthens the final syllable and results in complex syllable codas or non-optimal (consonantal) syllable nuclei. The Swiss German dialects have open syllables in these cases ([aɪ̯n.ʔˈaː.bn̩t] vs. [ə.nˈɑː.b̥ə] ‘an evening’). The western Swiss German dialects show vocalization of word-final ‑əl to ‑ʊ, resulting in optimal syllable structures ([dɛn.ʔˈa.pfl̩] vs. [d̥ə.rˈœ.pfʊ] ‘the.acc apple’). The transition

336 

 Beat Siebenhaar

of the historical form [g̊ɑrn] > [g̊ˈɑ.rən] > [g̊ˈɑː.rə]2 ‘yarn’ in the Alpine dialects shows different syllable strengthening processes. First, there is an epenthetic schwa that breaks up the consonant cluster; then the final n is deleted. Both processes strengthen the syllable. Epenthesis and elision are therefore the processes that clearly separate syllable-profiling Swiss German dialects from word-profiling Standard German. Swiss German dialects have geminates within a foot, which is a criterion for syllable languages. Because all obstruents are voiceless and the distinction between fortis and lenis is maintained as an opposition of long and short consonants (Willi 1996). This can be interpreted as a distinction between singletons and geminates (Krähenmann 2003). The minimal pairs [ˈɑfːə] ‘ape’ vs. [ˈɑfə] ‘just’ or [vˈɑd̥̥ə] ‘calf’ vs. [vˈɑtə] ‘cotton’ exemplify this opposition. Most Swiss German dialects also have nasal geminates. In contrast to Standard German, the Swiss German dialects show external sandhi (Moulton 1986; Nübling and Schrambke 2004; Szczepaniak 2006) that blurs the word boundaries very much in favor of syllable optimization ([nɪçt.fiːl] vs. [nɪ.pfɪl] ‘not much’; [diː fraʊ̯] vs. [pfræʊ̯] ‘the woman’). In word languages, stress is lexically assigned, whereas syllable languages have a more rule-based stress attribution. In general, Standard German and Swiss German dialects have the same stress attribution. However, there are some exceptions, and they mainly follow the same pattern. Swiss German is more radical in attributing the traditional Germanic word-initial stress to loan words and abbreviations ([filˈeː] vs. [fˈile] ‘filet’; [t͜sedˈeː] vs. [t͜sˈeːd̥eː] ‘CD’). The most striking examples can be found in the Alpine Valais dialect ([militˈɛːɐ̯] vs. [mˈilitɛːr] ‘military’; [matʰematʰˈiːk] vs. [mˈatematik͜x] ‘mathematics’). The more rule-based stress attribution of the Swiss German dialects is a further indication of their being a syllable language while Standard German, with its more lexically-based stress attribution, is closer to a word language type. Overall, Swiss German fulfills some requirements for a syllable language while for others the situation is not so clear. The Swiss German dialects have typical syllable language features such as geminates, and their word boundaries are not marked with specific allophones. Moreover, their epenthesis and elision processes, as well as syllabification and sandhi phenomena, optimize the syllable. In other cases, the classification is not as clear: In word languages, stress is generally determined by the lexicon. However, compared to Standard German,

2 Vowel lengthening before /r/ is very common in most German dialects. As it is not clear whether this vowel lengthening took place before or after the n-elision, this step of the explanation is omitted here.



Phonological and phonetic considerations  

 337

Swiss German displays a more rule-based stress assignment. The same holds for its different inventory of vowels in stressed and unstressed syllables, which is typical of word languages. However, the inventory in unstressed syllables is broader than the Standard German inventory. The strongest argument against a classification of Swiss German as a syllable language is the syllable structure and the violation of the sonority hierarchy, both of which are core aspects of this language type. Swiss German has even more complex consonant clusters in the syllable onset than Standard German, which, moreover, violate the sonority hierarchy. These violations, exemplified in the sentence at the beginning of this chapter, raise the question of how the syllable should be defined. Besides the traditional explanation of extrasyllabic consonants, one should consider an interpretation of the voiceless fricatives [s] and [ʃ] in interconsonantal position as syllabic consonants or, following Cho and Halloway King (2003), as semisyllabic consonants. With such an interpretation that splits obstruents into two different sonority classes, fricatives and plosives, the classification of Swiss German as a syllable language would be much more consistent. This would also alter the syllable structure of Standard German. However, the change would be much more important for the Swiss German dialects, as interconsonantal fricatives are much more common due to the syncope of schwa in prefixes.

4 Prosodic evidence on the phonetic level As mentioned in section 1, phonological differences have a direct impact on the temporal organization of speech. This influence will be shown on the basis of a manually segmented corpus of more than 130,000 segments of spontaneous speech from four Swiss German dialects.3 The data consist of approximately two hours of spontaneous speech. Forty twenty-year-old subjects from four different dialect regions of German-speaking Switzerland were interviewed. Speakers (5 females and 5 males per dialect) from two Alpine varieties, Valais (VS) and Grisons (GR), and two Midland dialects, Bern (BE) and Zurich (ZH), were recorded in spontaneous interviews. Approximately three minutes per speaker were manually segmented, transcribed and labeled on a segmental and syllabic level and analyzed for temporal aspects. Word boundary was labeled for only half of the data, since the main goal of the analysis was to build models for timing and

3 The data were collected in the framework of a Swiss National Science Foundation (SNSF) project (Quantitative Approaches to Geolinguistics of Swiss German Prosody 2005‒2008) (cf. Leemann and Siebenhaar 2007, 2010).

338 

 Beat Siebenhaar

intonation. Because word boundaries did not alter the model, their labeling was largely omitted. Ramus, Nespor, and Mehler (1999) and Low, Grabe, and Nolan (2000) have developed different algorithms that allow a distinction between syllable-timed languages and stress-timed languages based on a quantification of consonantal and vocalic variability. With controlled data, they show that languages of different rhythmic types have different characteristics regarding the succession of vowels and consonants. This concept has been further reduced to a voiced-unvoiced ratio by Dellwo, Fourcin, and Abberton (2007). In both approaches the phonetic distinction between syllable-timed, stress-timed, and mora-timed languages is transferred to another phonetic level. Unpublished calculations of the Swiss German data used for the present article with the Ramus, Nespor, and Mehler (1999) algorithm located Swiss German at a much more stress-timed position than Standard German. The conclusion was that the spontaneous data do not fit into this rhythmic pattern. This corresponds to Arvaniti’s (2009) results on Greek, Spanish, and English spontaneous speech. She discloses that spontaneous data do not fit into this pattern that is based on read speech because there are too many different and interacting factors which cannot be controlled and affect segmental duration. Spontaneous speech must be calculated in a different way than read speech. Moreover, Arvaniti concludes that rhythm cannot only be based on timing and that other phonetic and psycholinguistic factors which define patterns of prominence must be taken into account. The following section exemplifies some of the mentioned aspects of timing that are more complex than the temporal relation of consonants and vowels. The selection of the phenomena is more or less arbitrary, since until now there has been no research on phonetics relating to the typological distinction of word languages and syllable languages besides the mentioned temporal effect of the succession of vowels and consonants. The focus of the following discussion is on timing – the duration of segments in the Swiss German database. The first aspect is the effect of resyllabification in cases where there is no articulatory assimilation, so that the only indication on resyllabification is prosody, in the case of the voiceless fricative duration. The second aspect is the duration of schwa, which never has lexical stress. I show that its duration depends on the position within the syllable and the word. The discussion should show which factor – word boundary or syllable boundary – is more important for duration changes. The third aspect is only an outlook on intonation. The empirical research on the data is given in Leemann (2012), but here the results are discussed with regard to the differentiation between syllable languages and word languages. Intensity should be examined as well; unfortunately, however, the data are not reliable as microphone distance was not controlled for.



Phonological and phonetic considerations  

 339

4.1 Resyllabification and the duration of fricatives In Swiss German dialects, most consonant classes show a moderate influence of the word boundary on duration: For most consonant classes we find non-significant lengthening in word-initial and word-final position. Yet the influence of the position within a syllable is more consistent; the coda consonant is usually longer than the onset consonant. This can be interpreted as a sign of syllable stability, which supports a classification of Swiss German as a syllable language. However, fricatives show another distribution. Figure 1 shows the duration of fricatives depending on their position in the word. To avoid distortions due to phrase-final and phrase-initial lengthening, the fricatives in first, penultimate, and last syllables are excluded (cf. Siebenhaar 2004a; Kuzla, Cho, and Ernestus 2007). The general distribution shows that word-medial and word-final fricatives have more or less the same duration while word-initial fricatives have a significantly longer duration (F (2,2577) = 3.4473, p < 0.05). At the same time, the duration of the fricatives varies depending on their position within the syllable. Swiss German fricatives are longer in syllable onsets than in syllable codas (F (1,2578) = 9.7836, p < 0.01). However, these two aspects have opposite effects due to resyllabification processes across word boundaries. With a phonetic definition of the syllable based on the sonority hierarchy alone, there is a resyllabification when a word-final fricative precedes a word starting with a nasal, liquid, or vowel. In these cases, the word-final fricative moves into the syllable onset of the following syllable ([d̥ɑs#lɑnd̥] > [d̥ɑ.slɑnd̥] ‘this land’). Conversely, a fricative in the onset before a plosive moves to the coda of the preceding syllable ([d̥ə#ʃtɛrn] > [d̥əʃ.tɛrn] ‘the star’). The additional lines mark the difference of onset and coda consonants for the different positions in the word. While the difference between onset and coda fricatives is not significant for word-initial and word-medial fricatives, the difference for word-final fricatives is significant (F (1,807) = 15.7897, p < 0.0001). This distribution shows that syllable structure and word structure interact. It seems that word-initial fricatives are generally longer, while in word-final position the resyllabification process is dominant and alters the duration of the consonant. When a word-final fricative moves to the syllable onset, it is lengthened and reaches the mean duration of word-initial fricatives. As this is the predominant position for resyllabification, it can be interpreted that in the Swiss German dialects the syllable type processes are stronger than the processes that mark the word boundary. However, this still needs to be compared to Standard German data.

340 

 Beat Siebenhaar

Figure 1: Duration in ms and confidence interval of fricatives in phrase-central syllables depending on the position in the word: initial, medial, final. Additional connecting lines display the position in the syllable (n=2588)

Compared to the other consonant classes, fricatives seem to be quite complex, since word-related aspects have more influence on them than on the other consonants, which are mainly influenced by syllable-related aspects. The analysis of the fricative duration, however, shows that the phonological process of resyllabification has an important impact on consonant duration, thus a phonetic measure. Overall, the variation of consonant duration tends to support a classification of Swiss German dialects as syllable languages.

4.2 Duration of schwa The duration of schwa, whether it is positioned at a syllable border, i.e. syllable-final in an open syllable, or nucleus of a closed syllable, shows an equally problematic distribution. Again, schwas in syllables at phrase boundaries are excluded from the analysis. At first glance the situation seems to be clear. Schwas in an open syllable are significantly longer than schwas in closed syllables (compare the non-overlapping diamonds in Figure 2; F (1,1271) = 22.2925, p < 0.0001). However, the distribution that is due to the position of the schwa-syllable in the word (F (2,1270) = 55.8423, p < 0.0001) is much more important. With respect to both aspects, the influence of the position of the schwa in open or closed syllables is radically reduced. Schwas in open syllables are still longer than those in closed syllables but the main difference is due to the different posi-



Phonological and phonetic considerations  

 341

tion in the word. A General Linear Model, which takes both aspects into account, shows that the syllable boundary only has an F-ratio of 4.5948 (DF: 1) while the corresponding value for the position within a word is 46.3560 (DF: 2). The connecting lines in Figure 2 also show that for a schwa in a word-final syllable, the position within the syllable has no influence, while this influence is significant for word-initial (F (1, 113) = 4.5141, p < 0.05)) and word-medial syllables (F (1, 580) = 5.2034, p < 0.05)).

Figure 2: Duration and confidence interval of schwas in phrase-central syllables depending on the position in the syllable. Additional connecting lines display the position in the word (n= 1272)

In this respect, the temporal structure of the Swiss German data does not clearly favor a classification of Swiss German as a syllable or word language, as lengthening cannot clearly be linked to the word or the syllable. The duration of schwa is dependent on the position within the word, which indicates an assignment to a word language. At the same time, the duration is dependent on the syllable type to a lesser degree; schwa in a closed syllable is shorter than in an open syllable. However, this effect is less important, so that most of the evidence supports a classification of Swiss German as a word language.

4.3 Intonation contour The last aspect to be mentioned is the anchoring of the intonation contour. Different publications (Atterer and Ladd 2004; Kleber and Rathcke 2008) have shown

342 

 Beat Siebenhaar

that the anchoring of the intonation contour in the syllable shows a regionally different distribution in Standard German read speech. In general, the pitch accent is earlier in northern regions and later in southern regions. Gilles’ (2005) work on spontaneous data in eight German towns generally supports this regional distribution, and especially in the southwestern city of Freiburg the pitch peak delay is very strong. Leemann (2012) has analyzed the present Swiss German dialect data regarding intonation. His analyses show that the pitch accents are earlier in the Zurich and Grisons dialects, later in Bern, and variable in Valais, where there are often additional accents in unstressed and even in schwa syllables. As the analysis is done with a Command-Response model, the results are not directly comparable to Atterer and Ladd (2004) or Kleber and Rathcke (2008), but the distribution within the dialects maintains the tendency for later peaks in the southern German dialect area and within the Swiss German dialects in the west. This difference of earlier and later peaks may also be a reflex of the typological difference of syllable and word languages because the late accent gives a higher value to the non-accented syllables, as the peak shifts from the stressed to the non-stressed syllable. The Valais dialect is even more clearly positioned as a syllable language because even schwa-syllables often have an intonation accent, a feature that levels out the difference between accented and non-accented syllables even more.

5 Conclusion Some aspects of Swiss German dialects characterize them as syllable languages. The most striking aspects are the presence of geminates and phonotactics as Swiss German has no word-delimiting allophones. Moreover, resyllabification and sandhi phenomena are very common. Some aspects, on the other hand, do not clearly point to Swiss German being a syllable language. Like a word language, it distinguishes between full vowels in stressed syllables and reduced vowels in unstressed syllables. However, vowels in unstressed syllables show more phonetic variation than Standard German, which only has schwa in this position. The syllable coda is less complex than in Standard German. These phonological differences have direct implications in the temporal domain. Concerning intonation, the Swiss German dialects allow intonational accents on unstressed and even schwa syllables, which is not possible in Standard German. The Swiss German dialects have some properties of a word language. The syllable structure in the onset is very complex; there are even words (articles and prepositions) without vowels that are only used as clitics. The proclisis may be interpreted as a strengthening of the boundaries of the phonological word.



Phonological and phonetic considerations  

 343

Generally, the analysis has shown that the definition of the syllable, as well as the definition of the word, has a considerable impact on the assignment of the Swiss German dialects to a syllable rather than a word language. These basic definitions must be made clear. This is even more important when we analyze spontaneous connected speech, as many processes of temporal reorganization are an expression of communicative attitudes which cannot be analyzed in laboratory speech. Extrasyllabic consonants are problematic for an analysis of connected speech. Therefore, I postulate a purely phonetic definition following the sonority hierarchy that includes the definition of syllabic consonants as syllable nuclei. For Swiss German, this means accepting even [s] and [ʃ] as syllabic consonants. Furthermore, it is worth looking at phonetic reflexes of different aspects of word and syllable languages. The exemplary and exploratory results discussed in section 3, the changing duration of consonants and especially fricatives as a consequence of resyllabification, the dependence of the duration of schwa on position within the syllable and within the word, and the different anchoring of intonational contours show the phonetic consequences of phonologic processes, which generally correlate with word language vs. syllable language criteria. The analysis supports an assignment of the Swiss German dialects to the pole of the syllable languages. In a general linguistic perspective, future work on phonetic evidence may strengthen a phonological typological distinction.

References Abercrombie, David (1967): Elements of General Phonetics. Edinburgh: Edinburgh University Press. Arvaniti, Amalia (2009): Rhythm, timing and the timing of rhythm. Phonetica 66: 46–63. Atterer, Michaela and D. Robert Ladd (2004): On the phonetics and phonology of “segmental anchoring” of F0: evidence from German. Journal of Phonetics 32/2: 177–197. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Cho, Young-mee Yu and Tracy Halloway King (2003): Semisyllables and universal syllabification. In: Caroline Féry and Ruben van de Vijver (eds.), The Syllable in Optimality Theory, 183–212. Cambridge: Cambridge University Press. Dellwo, Volker, Adrian Fourcin and Evelyn Abberton (2007): Rhythmical classification of languages based on voice parameters. In: Jürgen Trouvain and William J. Barry (eds.), Proceedings of the ICPhS XVI, 1129–1132. Saarbrücken: Universität des Saarlandes. Available at http://www.icphs2007.de/conference/Papers/1169/1169.pdf. Fleischer, Jürg and Stephan Schmid (2006): Zurich German. Journal of the International Phonetic Association 36/2: 243–253. Gilles, Peter (2005): Regionale Prosodie im Deutschen. Variabilität der Intonation von Abschluss und Weiterweisung. (Linguistik – Impulse & Tendenzen 6.) Berlin/New York: De Gruyter.

344 

 Beat Siebenhaar

Hall, T. Alan (2011): Phonologie. Eine Einführung. (De Gruyter Studium.) 2nd ed. Berlin/New York: Walter de Gruyter. International Phonetic Association (1999): Handbook of the International Phonetic Association: A Guide to the Use of the International Phonetic Alphabet. Cambridge: Cambridge University Press. Klatt, Dennis H. (1976): Linguistic uses of segmental duration in English: Acoustic and perceptual evidence. Journal of the Acoustical Society of America 59/5: 1208‒1221. Kleber, Felicitas and Tamara Rathcke (2008): More on the “segmental anchoring” of prenuclear rises: Evidence from East Middle German. In: Proceedings of the Fourth Conference on Speech Prosody 2008, 583–586. Available at http://sprosig.isle.illinois.edu/sp2008/ papers/id123.pdf. Kraehenmann, Astrid (2003): Quantity and Prosodic Asymmetries in Alemannic. Synchronic and Diachronic Perspectives. (Phonology & Phonetics 5.) Berlin/New York: Mouton de Gruyter. Krech, Eva-Maria, Eberhard Stock, Ursula Hirschfeld and Lutz Christian Anders (2009): Deutsches Aussprachewörterbuch. Berlin/New York: Walter de Gruyter. Kuzla, Claudia, Taehong Cho and Mirjam Ernestus (2007): Prosodic strengthening of German fricatives in duration and assimilatory devoicing. Journal of Phonetics 35/3: 301–320. Leemann, Adrian (2012): Swiss German Intonation Patterns. (Studies in Language Variation 10.) Amsterdam/Philadelphia: Benjamins. Leemann, Adrian and Beat Siebenhaar (2007): Intonational and temporal features of Swiss German. In: Jürgen Trouvain and William J. Barry (eds.), Proceedings of the ICPhS XVI, 957–960. Saarbrücken: Universität des Saarlandes. Available at http://www.icphs2007. de/conference/Papers/1468/1468.pdf. Leemann, Adrian and Beat Siebenhaar (2010): Statistical modeling of F0 and timing of Swiss German dialects. In: Proceedings of Speech Prosody 2010, 1–4. Available at http://speechprosody2010.illinois.edu/papers/100180.pdf. Low, Ee Ling, Esther Grabe and Francis Nolan (2000): Quantitative characterizations of speech rhythm: Syllable-timing in Singapore English. Language and Speech 43/4: 377–401. Moulton, William G. (1986): Sandhi in Swiss German dialects. In: Henning Andersen (ed.), Sandhi Phenomena in the Languages of Europe, 385–392. Berlin/New York: Mouton de Gruyter. Nocchi, Nadja and Stephan Schmid (2006): Labiodentale Konsonanten im Schweizerdeutschen. In: Hubert Klausmann (ed.), Raumstrukturen im Alemannischen. Beiträge der 15. Arbeitstagung zur alemannischen Dialektologie. Schloss Hofen, Lochau (Vorarlberg), 19.–21.9.2005, 25–35. (Schriften der Vorarlberger Landesbibliothek 15.) Graz-Feldkirch: Neugebauer. Nübling, Damaris (1992): Klitika im Deutschen. Schriftsprache, Umgangssprache, alemannische Dialekte. (Script Oralia 42.) Tübingen: Gunter Narr. Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 129.) Stuttgart: Steiner. Pike, Kenneth Lee (1945): The Intonation of American English. Ann Arbor: University of Michigan Press. Ramus, Franck, Marina Nespor and Jacques Mehler (1999): Correlates of linguistic rhythm in the speech signal. Cognition 72: 1–28.



Phonological and phonetic considerations  

 345

SDS = Hotzenköcherle, Rudolf (ed.) (1962–1997): Sprachatlas der deutschen Schweiz. Bern/ Basel: Francke. Siebenhaar, Beat (2004a): Comparing timing models of two Swiss German dialects. In: Britt-Louise Gunnarsson, Lena Bergström, Gerd Eklund, Staffan Fridell, Lise H. Hansen, Angela Karstadt, Bengt Nordberg, Eva Sundgren and Mats Thelander (eds.), Language Variation in Europe. Papers from ICLaVE 2, 353–365. Uppsala: Uppsala University Press. Siebenhaar, Beat (2004b): Berner und Zürcher Prosodie. Ansätze zu einem Vergleich. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 419–437. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 129.) Stuttgart: Steiner. Siebenhaar, Beat and Adrian Leemann (2012): Methodological reflections on the phonetic-phonological continuum, illustrated on the prosody of Swiss German dialects. In: Andrea Ender, Adrian Leemann and Bernhard Wälchli (eds.), Methods in Contemporary Linguistics, 21‒44. (Trends in Linguistics. Studies and Monographs 247.) Berlin: Mouton de Gruyter. Siebenhaar, Beat, Brigitte Zellner Keller and Eric Keller (2001): Phonetic and timing considerations in a Swiss High German TTS system. In: Eric Keller, Gérard Bailly, Alex Monaghan, Jacques Terken and Mark Huckvale (eds.), Improvements in Speech Synthesis, 165–175. Chichester: Wiley. Szczepaniak, Renata (2006): Konsonantenassimilationen im Alemannischen aus phonologisch-typologischer Perspektive. In: Hubert Klausmann (ed.), Raumstrukturen im Alemannischen. Beiträge der 15. Arbeitstagung zur alemannischen Dialektologie. Schloss Hofen, Lochau (Vorarlberg), 19.–21.9.2005, 61–72. (Schriften der Vorarlberger Landesbibliothek 15.) Graz-Feldkirch: Neugebauer. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. van Santen, Jan (1992): Contextual effects on vowel duration. Speech Communication 11: 513–546. Willi, Urs (1996): Die segmentale Dauer als phonetischer Parameter von “fortis” und “lenis” bei Plosiven im Zürichdeutschen. Eine akustische und perzeptorische Untersuchung. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 92.) Stuttgart: Steiner.

Part 4: Synchronic approaches (Romance languages)

Javier Caro Reina (University of Freiburg)

Central Catalan in the framework of the typology of syllable and word languages* Abstract: This paper presents a synchronic approach to the phonology of Central Catalan applying the typology of syllable and word languages. Selected parameters, including the syllable structure, phonotactic restrictions, and phonological processes, are systematically examined in order to determine the existence and relevance of the prosodic categories of the syllable and the phonological word. Although the evidence supports the existence of both domains, the relevance of the phonological word seems to be greater in Central Catalan than in other Catalan varieties. While the Catalan varieties share the same syllable-related phonological processes, with the introduction of word-related processes such as vowel centralization and merger of back vowels in unstressed syllables Central Catalan has undergone a phonological drift which profiles the phonological word. It is argued that the phonological differences traditionally described in Catalan dialectology can be explained more adequately in terms of typological features.

1 Introduction Catalan is a Romance language spoken in eastern Spain (Catalonia, the Valencian Community, the Balearic Islands, and parts of Aragon and Murcia), Andorra, Roussillon, and the Sardinian city of Alghero (see Map 1).

* My thanks to Peter Auer, Marco García, Maria-Rosa Lloret, and Robert W. Murray for insightful comments on previous versions of this paper.

350 

 Javier Caro Reina

Map 1: Catalan-speaking areas. Reprinted with permission of the Generalitat de Catalunya

Catalan is traditionally divided into two main dialect groups: Western Catalan (català occidental) and Eastern Catalan (català oriental). Western Catalan consists of the North-Western Catalan (català nord-occidental) and Valencian (valencià) dialects, whereas Roussillon Catalan (rossellonès), Central Catalan (català central), Balearic (balear), and Alghero Catalan (alguerès) belong to Eastern Catalan. Map 2 depicts the areas where the different Catalan dialects (and subdialects) are spoken.

 Central Catalan in the framework of the typology of syllable and words languages  

 351

Map 2: Modern Catalan dialects (from Nuet, Bernat, and Torres 1992: 6). Reprinted with permission

The structural differences between Western and Eastern Catalan have been attributed to historical factors such as the pre-Latin substrate (Sanchis 1956), the Roman colonization patterns (Badia 1981), and the new settlements which were built after the Christian reconquest (Ferrando 1989). Veny (2002: 22) does not dispute that these factors may have played a role in the formation of Catalan,

352 

 Javier Caro Reina

but he points out that many of the diverging phonological and morphological processes of Eastern Catalan evolved during a later period. In Central Catalan, for example, centralization of /a ɛ e/ in pretonic position can be traced back to the twelfth century (Rasico 1982: 236). Additionally, the merger of unstressed /ɔ o u/ to [u] was completed in Central Catalan in approximately the sixteenth century (Rasico 1982: 106). Therefore, the first attested forms showing centralization and merger do not support the theories based entirely on historical factors. Veny (2001b: 200) further argues that processes such as vowel centralization cannot be the result of a substrate language if similar processes are also found in other Romance languages in which the substrate influence assumed for Eastern Catalan is not tenable. In this paper, I propose that the phonological differences between Western and Eastern Catalan can be accounted for in a more satisfactory way by applying the typology of syllable and word languages. I argue that Eastern Catalan, contrary to Western Catalan, underwent a language-internal development towards the word pole on the scale of syllable and word languages. The distinction between Western and Eastern Catalan is made on the basis of phonological, morphological, and lexical criteria (see Viaplana 1984 and Veny 2001b: 197–198 for a critical discussion). The phonological differences between both dialect groups are displayed in Table 1. A closer look at the thirteen isoglosses in the table reveals that features 1 (vowel centralization in unstressed position), 2 (merger of back vowels in unstressed position), 8 (deaffrication) and 10 (reinforcement of word-final r) in Eastern Catalan are associated with phonological processes characteristic of word languages such as vowel reductions in unstressed syllables (features 1 and 2), lenitions which decrease the consonantal strength in every syllable onset thereby violating the Onset Law and Contact Law (feature 8), and consonant epentheses which highlight the margin of the phonological word (feature 10). Moreover, if features regarding one single lexical item (features 4, 11, 12, 13) are left out, a considerable portion of the phonological features traditionally used to distinguish Western from Eastern Catalan can be better understood from the perspective of the typology of syllable and word languages.

 Central Catalan in the framework of the typology of syllable and words languages  

 353

Table 1: Phonological features of Western and Eastern Catalan (Veny 1991: 244–245, 2002: 19–20)1 Western Catalan

Eastern Catalan

1

No centralization of /a ɛ e/ > [ə] in unstressed position: palla [ˈpaʎa] ‘straw’ – pall-et-a [paˈʎeta] ‘straw-dim-fem’ pell [peʎ] ‘skin’ – pell-et-a [peˈʎeta] ‘skin-dim-fem’

Centralization of /a ɛ e/> [ə] in unstressed position: palla [ˈpaʎə] ‘straw’ – pall-et-a [pəˈʎɛtə] ‘straw-dim-fem’ pell [peʎ] ‘skin’– pell-et-a [pəˈʎɛtə] ‘skin-dim-fem’

2

No merger of /ɔ o u/ to [u] in unstressed position: coca [ˈkoka] ‘cake’ – coqu-et-a [koˈketa] ‘cake-dim-fem’ cuca [ˈkuka] ‘bug’ – cuqu-et-a [kuˈketa] ‘bug-dim-fem’

Merger of /ɔ o u/ to [u] in unstressed position: coca [ˈkokə] ‘cake’ – coqu-et-a [kuˈkɛtə] ‘cake-dim-fem’ cuca [ˈkukə] ‘bug’ – cuqu-et-a [kuˈkɛtə] ‘bug-dim-fem’

3

Vulgar Latin /e/ (< Latin ē, ĭ, œ) > /e/: cepa > ceba [ˈseβa] ‘onion’

Vulgar Latin /e/ (< Latin ē, ĭ, œ) > /ə/ > /ɛ/: cepa > ceba [ˈsɛβə] ‘onion’

4

Vulgar Latin ŭ in jŭncu > u: junc [d͡ʒuŋ] ‘reed’

Vulgar Latin ŭ in jŭncu > o: jonc [ʒoŋ] ‘reed’

5

Retention of stressed diphthongs guá-, quá-: quattŭor > quatre [ˈkwatɾe] ‘four’ coac(u)la > guatlla [ˈɡwaʎʎa] ‘quail’

Tendency guá-, quá- > [gɔ]-, [kɔ]-: quattŭor > quatre [ˈkwatɾə], [ˈkɔtɾə] ‘four’ coac(u)la > guatlla [ˈɡwaʎʎə], [ˈɡɔʎʎə] ‘quail’

6

Retention of unstressed diphthongs -gua, -qua: aqua > aigua [ˈajɣwa] ‘water’ lĭngua > llengua [ˈʎeŋɡwa] ‘tongue’

Tendency -gua, -qua > ‑[ɣə], -[gə], -[kə]: aqua > aigua [ˈajɣwə], [ˈajɣə] ‘water’ lĭngua > llengua [ˈʎeŋɡwə], [ˈʎeŋɡə] ‘tongue’

7

Vulgar Latin lj, c’l, g’l > [ʎ]: palea > palla [ˈpaʎa] ‘straw’

Vulgar Latin lj, c’l, g’l > [ʎ], [j]: palea > palla [ˈpaʎə], [ˈpajə] ‘straw’

1 For the sake of simplicity, the transcriptions given for Western and Eastern Catalan correspond to North-Western Catalan and Central Catalan, respectively. In the transcriptions for Western Catalan the realization of word-final a as [ɛ] typical of the area around Lleida was not considerd (see Veny 1986: 112 for a map with the distribution of this process).

354 

 Javier Caro Reina

Western Catalan

Eastern Catalan

8

Retention of [t͡ʃ] in the onset: xinxa [ˈt͡ʃiɲt͡ʃa] ‘bed bug’ panxa [ˈpaɲt͡ʃa] ‘belly’

Deaffrication of [t͡ʃ] in the onset: xinxa [ˈʃiɲʃə] ‘bed bug’ panxa [ˈpaɲʃə] ‘belly’

9

cs, sce, i > [jʃ]: cacsa > caixa [ˈkajʃa] ‘box’ pisce > peix [pejʃ] ‘fish’

cs, sce, i > [ʃ]: cacsa > caixa [ˈkaʃə] ‘box’ pisce > peix [peʃ] ‘fish’

10

No reinforcement of word-final r: cor [kɔɾ] ‘heart’

Reinforcement of word-final r through epenthesis: cor [kɔɾt] ‘heart’

11

Epenthetic n in llagost, llagosta, llagostí (< *lagŭsta) ‘locust’: [ʎaŋˈgost], [ʎaŋˈgosta], [ʎaŋgosˈti]

No epenthetic n in llagost, llagosta, llagostí (< *lagŭsta) ‘locust’: [ʎəˈɣost], [ʎəˈɣostə], [ʎəɣusˈti]

12

hedera > hedra [ˈeðɾa] ‘ivy’

hedera > heura [ˈewɾə] ‘ivy’

13

die mart(i)s > dimarts [diˈmat͡s] ‘Tuesday’

die mart(i)s > dimarts [diˈmaɾs] ‘Tuesday’

This paper is the first to approach the phonology of Central Catalan in the framework of the typology of syllable and word languages as introduced by Auer (1993, 1994, 2001). Central Catalan was chosen as the research focus because this dialect shows a higher concentration of word-related processes than other Eastern Catalan dialects. This typology has been hitherto successfully tested by Auer (1993) in a cross-linguistic study based on a geographically and genetically diversified sample of 34 languages. One disadvantage of cross-linguistic studies lies in the fact that they do not consider the wide scope of (non-standard) varieties of a given language. That is, they usually rely on descriptions made in grammars of standard written varieties (see Auer 1993: 40–41 for a discussion). Certainly, it would be more illuminating to focus on different varieties of a certain language, both from a diachronic and synchronic point of view, as done for standard German and Alemannic: Szczepaniak (2007) shows that standard German has undergone a typological change developing from a syllable language in Old High German into a word language in New High German, while Nübling and Schrambke (2004) point to typological differences between North and South Alemannic. Similar studies have not been carried out for Catalan yet and this contribution aims to fill this gap.

 Central Catalan in the framework of the typology of syllable and words languages  

 355

The paper is structured as follows. Section 2 presents the typological parameters that will be applied to Central Catalan. Section 3 is devoted to the syllable structure and deals with the frequency of the underlying syllable types with regard to stress and within-word position, the frequency of the surface word-final syllable types, and the existence of extrasyllabic elements. Section 4 focuses on the distribution of vowels and single consonants. Section 5 gives an account of selected word-related processes drawing on data from other Catalan dialects such as North-Western Catalan, Valencian, and Majorcan Catalan.

2 Determining the phonological domains In order to determine the phonological domains of the syllable and the phonological word in Central Catalan, I will not use the schema containing syllable-related and word-related features as found in Szczepaniak (2007: 330). This schema is especially suited for German but not necessarily for other languages, which may result in a deceptively simple picture of what syllable languages and word languages are (or may be). Instead, I propose an alternative theoretical working basis involving the systematic study of the patterns of selected parameters including the syllable structure, phonotactic restrictions, and phonological processes according to stress (stressed vs. unstressed syllable) and within-word position (word-initial, word-medial, and word-final), as illustrated in Figure 1. stress-related (stressed vs. unstressed)

position-related (word-initial, word-medial, word-final)

Syllable structure Phonotactic restrictions Phonological processes Figure 1: Selected parameters in relation to stress and within-word position

In a syllable language, the syllable structure, phoneme inventory, and phonological processes are not sensitive to stress or position within a larger phonological domain (e.g. the phonological word). Thus, the syllable structure shows the universal template CV, there are no word-related phonotactic restrictions, and the phonological processes have the function of improving the syllable. These features occur in an ideal syllable language which does not necessarily represent any natural language in particular. In contrast, in a word language we may

356 

 Javier Caro Reina

observe substantial differences regarding stress and position in the phonological word. With respect to the stressed and unstressed syllable, we can expect syllable complexity in stressed syllables, restrictions such as neutralization of vowel height, nasalization, and quantity in unstressed syllables, and phonological processes such as vowel lengthening in stressed syllables. With regard to the position in the word, we can expect complex syllable structures at word and morpheme boundaries, restrictions such as the absence of word-final voiced obstruents, and phonological processes such as fortitions at the word edges and/ or word-medial lenitions. It should be noted that phonotactic restrictions may be the result of phonological processes that are not likely to be observed synchronically. In Central Catalan, for example, the non-occurrence of the non-high non-back vowels in unstressed position is due to the neutralization of vowel height. Similarly, complex syllable structures may have arisen as a consequence of unstressed vowel deletion. The deletion of Vulgar Latin unstressed word-final -e and ‑o, which took place in Pre-Old Catalan, led to the appearance of consonant clusters in word-final position in Old Catalan (e.g. Vulgar Latin largo > Old Catalan llarc [ʎaɾk] ‘long[masc]’). Once we have applied the theoretical working basis outlined in Figure 1 to Central Catalan, we will be able to determine the existence and relevance of the phonological domains of the syllable and the word before we can classify, in a second step, Central Catalan as a syllable language or a word language. For the typology it would be valuable to carefully evaluate the word- or syllable-optimizing strategies used in a particular language. This research may yield important new knowledge regarding the question of whether there is a correlation between the typological parameters and, more importantly, how they interact with each other. As for correlations, Auer (1993: 88) observes in his cross-linguistic study a positive correlation between shell complexity and word-related processes. This correlation is borne out for Central Catalan as well, as we will see in ensuing sections. A complex syllable structure helps highlight the word boundaries. For example, it may be found only word-initially as in literary Georgian, only word-finally as in Catalan (excluding some prefixes of Latin origin), or at both margins of the word as in German and English. Interestingly, a simple syllable structure is not, as usually assumed (Auer 1993: 66, 2001: 1396; Szczepaniak 2007: 33–35), necessarily in conflict with a word language. For instance, Auer (1993: 74) reports for !Xóõ, a Khoisan language spoken in Botswana and Namibia, a simple syllable structure and, crucially, phonotactic restrictions permitting the phonemic contrast of 116 consonants in the word-initial onset and six consonants in the word-medial onset (see Auer 1994: 62–64 for a discussion and Traill 1985: 164–166 for details). In that respect, a word-initial complex syllable structure as in German and phonotactic restrictions as in !Xóõ are two completely different strategies

 Central Catalan in the framework of the typology of syllable and words languages  

 357

that equally highlight the left margin of the word. Forthcoming cross-linguistic studies might shed light on which strategy is more frequently used and whether these strategies are even language-family specific. In view of such theoretical assumptions, we can assert that, contrary to the existence of an ideal syllable, there is no ideal phonological word, since its shape is language specific or even morphologically conditioned. In Central Catalan, for instance, masculine nouns and masculine adjectives may have a complex syllable structure in word-final position while infinitives lack word-final syllable complexity. Compare the adjective llarg [ʎaɾk] ‘long[masc]’ with a complex syllable structure in word-final position and the infinitive allarg-ar [əʎəɾˈɣa] ‘lengthen-inf’ with a simple syllable structure in word-final position and vowel centralization in the pretonic syllables. This example shows that the grammatical category involved seems to be relevant.

3 Syllable structure In the World Atlas of Language Structures (WALS), Maddieson (2013) distinguishes between three types of syllable complexity: simple, moderately complex, and complex. Simple syllable structures have the canonical syllable CV and/ or V. Moderately complex syllable structures have the canonical syllable CCV, CVC, and/or CCVC (the second consonant of the onset being a liquid or a glide). Complex syllable structures have no restrictions regarding the number or nature of the consonants (see Kouwenberg 2010: 365–367 for a critical discussion). For the remainder of this paper, I will follow this classification. The different syllable templates proposed by Maddieson are schematized in (1). (1) Syllable templates according to syllable complexity simple: (C)V moderately complex: (C)(C)V(C) complex: (C)(C)(C)V(C)(C)(C)(C)

The syllable structure of Central Catalan can have a maximum of two consonants in the onset and a maximum of two or three consonants in the coda depending on whether they appear word-medially or word-finally. According to Maddieson’s classification, this syllable structure can be considered complex. The template of Central Catalan, which applies for all Catalan dialects as well, is depicted in (2).

358 

 Javier Caro Reina

(2) Central Catalan surface syllable structure

σ Onset

Rhyme Nucleus

(C) (C)

V

Coda (C) (C) (C)

For the frequency of the syllable types allowed in Central Catalan, I will refer to De Yzaguirre (1995). His survey is based on the Diccionari de la Llengua Catalana. This dictionary contains 68,551 words which are transcribed phonologically. In order to analyze the Catalan syllable structure, the syllable boundaries and the consonant and vowel segments were systematically computerized. The total number of syllables elicited is 2,414,824. Additionally, nouns, adjectives, and verbs were inflected.2 Table 2 presents the frequency of the underlying syllable types according to their occurrence in stressed vs. unstressed position and in word-final vs. non-word-final position. The syllable types are arranged according to their absolute frequency.3 De Yzaguirre’s findings allow the following observations: (1) the universal syllable CV, with 1,159,160 tokens (48%), is the most common type in Catalan; (2) open syllables are more common than closed syllables (approximately 60% vs. 40%); (3) simple syllable structures are preferred to complex syllable structures (97% vs. 3%); (4) the complex syllable structures (C)(C)VCC and (C)(C)VCCC appear more frequently in stressed than in unstressed position (9% of all stressed syllables vs. 0% of all unstressed syllables); and (5) the complex syllable structures concentrate in word-final position (10% of all word-final syllables vs. 0% of all non-word-final syllables). These results confirm the cross-linguistic attested preference for open syllables and syllables with simple syllable structures. They further reflect the predicted occurrence of complex syllables in stressed position and at the margins of the phonological word (Auer 1993: 66, 74–75).

2 The approximate number of inflected forms was 595,000 and comprised around 8,400 verbs at a rate of 51 forms per verb, 20,000 masculine nouns, 20,000 feminine nouns, and 20,000 adjectives (De Yzaguirre, personal communication). 3 Percentages are rounded and may not always add up to exactly 100%.

 359

 Central Catalan in the framework of the typology of syllable and words languages  

Table 2: Frequency of the underlying syllable types (De Yzaguirre 1995: 67–69) Syllable type

Frequency

stress-related stressed

CV

position-related

unstressed

non-word-final

word-final

1,159,160 (48%)

327,316 (53%)

831,844 (46%)

1,018,519 (58%)

140,641 (21%)

CVC

654,027 (27%)

174,379 (28%)

479,648 (27%)

309,630 (18%)

344,397 (52%)

VC

223,101 (9%)

11,332 (2%)

211,769 (12%)

169,011 (10%)

54,090 (8%)

V

157,730 (7%)

21,255 (3%)

136,475 (8%)

126,218 (7%)

31,512 (5%)

CCV

116,748 (5%)

19,936 (3%)

96,812 (5%)

103,690 (6%)

13,058 (2%)

CVCC

48,013 (2%)

41,200 (7%)

6,813 (0%)

920 (0%)

47,093 (7%)

CCVC

35,032 (1%)

8,898 (1%)

26,134 (1%)

24,361 (1%)

10,671 (2%)

VCC

10,861 (0%)

6,228 (1%)

4,633 (0%)

4,685 (0%)

6,176 (1%)

CVCCC

4,974 (0%)

4,941 (1%)

33 (0%)

22 (0%)

4,952 (1%)

CCVCC

4,664 (0%)

2,242 (0%)

2,422 (0%)

186 (0%)

4,478 (1%)

VCCC

325 (0%)

325 (0%)

0 (0%)

2 (0%)

323 (0%)

CCVCCC

189 (0%)

189 (0%)

0 (0%)

0 (0%)

189 (0%)

2,414,824 (100%)

618,241 (100%)

1,796,583 (100%)

1,757,244 (100%)

657,580 (100%)

Total:

Note that Table 2 lists the frequency of the underlying syllable types. When focusing on Central Catalan, we must consider phonological rules specific to this dialect. The stress-related and position-related frequency of the surface syllable types could not be calculated using De Yzaguirre’s data. However, the data allow us to study the frequency of the surface syllable types in word-final posi-

360 

 Javier Caro Reina

tion regardless of the onset. These results are given in Table 3. The syllable types are arranged with regard to the number of consonants in the word-final coda. In Central Catalan, the number of complex syllables occurring in word-final position is reduced as a consequence of phonological rules such as simplification of word-final homorganic consonants (see section 5.6). In addition, in order to investigate to what extent the frequency of word-final syllable types in Central Catalan is affected by processes which simplify the word-final coda, I calculated the frequency of word-final syllable types in Valencian, where word-final processes such as the simplification of homorganic consonants do not occur. We find that the word-final complex syllable structure (C)(C)VCC is slightly less frequent in Central Catalan than in Valencian (7% vs. 10%). Table 3: Frequency of word-final surface syllable types in Valencian and Central Catalan4 Syllable type

Underlying frequency

Surface frequency Valencian

Central Catalan

(C)(C)V

185,211 (28%)

185,211 (28%)

197,342 (30%)

(C)(C)VC

409,158 (62%)

409,158 (62%)

416,828 (63%)

(C)(C)VCC

57,761 (9%)

62,773 (10%)

42,984 (7%)

(C)(C)VCCC

5,483 (1%)

471 (0%)

459 (0%)

657,613 (100%)

657,613 (100%)

657,613 (100%)

Total:

The syllable in Catalan is generally organized according to the Sonority Sequencing Principle (see Pons 2011: 113–114 for a discussion on the sonority scale). However, ill-formed syllables containing appendices (i.e. extrasyllabic elements)

4 The total number of underlying and surface syllable types is slightly different (657,580 vs. 657,613). This is due to the fact that De Yzaguirre (1995: 119) treated word-final /mp/ and /mps/ separately. These cases have been taken into account for the table. Common to all Catalan dialects are the following phonological processes occurring in the word-final syllable type (C)(C) VCCC: /nts/ > [ns] (4,349 cases), /ŋks/ > [ŋs] (334 cases), /ɾts/ > [ɾs] (254 cases), /lts/ > [ɫs] (56 cases), and /mps/ > [ms] (19 cases). After applying these processes, we obtain the surface frequency given for Valencian. Additionally, in Central Catalan we find the following processes affecting the underlying word-final syllable types (C)(C)VCCC, (C)(C)VCC, and (C)(C)VC: /lps/ > [ps] (12 cases); /nt/ > [n] (15,034 cases), /ɾs/ > [s] (4,115 cases), /nk/ > [ŋ] (557 cases), /lt/ > [ɫ] (69 cases), /mp/ > [m] (14 cases), and /lp/ > [p] (12 cases); /ɾ/ > [∅] (12,131 cases). The deletion of word-final r and the simplification of word-final homorganic consonants are described in sections 5.5 and 5.6, respectively. Further simplification processes are not treated. For practical reasons, the combination of /t/ with the plural morpheme /s/ was considered as two single segments.

 Central Catalan in the framework of the typology of syllable and words languages  

 361

abound. In Central Catalan, syllable appendices are only found in the coda, as opposed to German and English, in which they can appear in the onset as well. The only consonant that functions as a suppendix in Central Catalan is the coronal segment [s] both word-medially in Latin prefixes (e.g. explicació ‘explanation’ [əksplikəsiˈo]) and word-finally as the realization of either the plural morpheme (e.g. corb-s [kɔɾps] ‘raven-pl’, parc-s [paɾks] ‘park-pl’) or the second-person singular (e.g. sap-s [saps] ‘know-2sg.pres.ind’). Figures 2 and 3 display the sonority of the segments in the syllable of the items corb-s [kɔɾps] and parcs [paɾks]. As can be observed, the word-final sibilant is more sonorous than the preceding stop, thus violating the Sonority Sequencing Principle. Vowel Rhotic Lateral Nasal Fricative Plosive

•

• k

ɔ

•

ɾ

• p

Figure 2: Sonority of corb-s ‘raven-pl’

• s

Vowel Rhotic Lateral Nasal Fricative Plosive

•

• p

a

•

ɾ

• k

• s

Figure 3: Sonority of parc-s ‘parc-pl’

The high frequency of the underlying complex syllable structures in non-final position (see Table 2) results mainly from Latin prefixes such as ex-, trans- (when followed by a consonant belonging to the next onset as in explicació [əks.pli.kə.si. ˈo] ‘explanation’) or in-, ob-, sub-, etc. (when followed by a sibilant belonging to the same coda as in inspector [ins.pək.ˈto] ‘inspector[masc]’). For example, the non-final syllable structure VCC has a frequency of 4,685 tokens, of which 4,493 correspond to the prefix ex-. De Yzaguirre (1995: 23, 121) points out that the dialectal phonetic transcriptions made for the Diccionari Català-Valencià-Balear (DCVB) in the early twentieth century show that word-medial complex syllable structures had been subject to simplification. He further argues that they must have been reintroduced later under the influence of the written language. In this sense, a language-external factor prevented complex codas from occurring mostly in word-final position, which would have contributed to their function as word boundary markers. The simplification of word-medial complex syllable structures lies in the loss of the first segment of the consonant cluster (e.g. [ks] > [s], [ns] > [s], [ps] > [s]). Lloret (2002: 217) reports that this simplification, if not already lexified, correlates with the stylistic context involved. These structures are retained in formal speech. In casual speech, however, they are simplified, especially in frequent words. Examples from Central Catalan are given in (3).

362 

 Javier Caro Reina

(3) Realization of word-medial complex codas in formal and casual speech (Lloret 2002: 217) explicació ‘explanation’ [əksplikəsiˈo] ~ [əsplikəsiˈo] inspector ‘inspector[masc]’ [inspəkˈto] ~ [ispəkˈto] obscuritat ‘darkness’ [upskuɾiˈtat] ~ [uskuɾiˈtat] transmet-re ‘transmit-inf’ [tɾənsˈmɛtɾə] ~ [tɾəsˈmɛtɾə]

4 Phonotactic restrictions 4.1 Stress-related restrictions The vocalism of Central Catalan comprises the vowels [i e ɛ a ɔ o u] in stressed position and the vowels [i ə u] in unstressed position. The number of oppositions permitted in stressed position sharply contrasts with the reduced number of possible oppositions in unstressed position. This seven-to-three inventory reduction results from the centralization of /a ɛ e/ and the merger of /ɔ o u/ to [u] (see Mascaró 2002 for a detailed description and Recasens 1996: 112–115, 144–145 for exceptions). Thus, in Central Catalan height contrasts are neutralized in unstressed syllables. As a consequence, the set of full vowels /e ɛ a ɔ o/ may appear only in the stressed syllable of the phonological word. According to Barnes (2006: 20), the neutralization of vowel height belongs to the most common asymmetries between stressed and unstressed vowels. Map 3 and Map 4 are based on the Atles lingüístic del domini català (ALDC). This atlas is embedded in the traditional dialectology. The data were elicited mainly between 1964 and 1978 in a network of 190 localities and represent the language of non-mobile, older, rural males. Map 3 charts the centralization of /a/ in pretonic position in the lexical item carrer ‘street’. The centralization of /a ɛ e/ in unstressed position is a feature shared by all Eastern Catalan dialects except for Alghero Catalan. Sociolinguistically speaking, this map shows a rather simplified picture. I will refer to some cases in which sociolinguistic factors as well as dialect and language contact situations can add more information. Plaza (1995) carried out a sociolinguistic study in L’Espluga de Fracolí, a locality situated in the transitional area between North-Western Catalan and Central Catalan. The subdialect spoken here is termed Xipella (Veny 2002: 38–39) and is traditionally characterized as having centralization solely in pretonic position. However, Plaza (1995: 131–137) shows that centralization in posttonic position is increasing among the youngest generations due to the influence of Central Catalan. Interestingly, centralization of pretonic /a/ has been observed in parts of North-Western Catalan.

 Central Catalan in the framework of the typology of syllable and words languages  

Map 3: Vowel centralization in unstressed syllable

 363

364 

 Javier Caro Reina

Map 4: Merger of back vowels in unstressed syllable

 Central Catalan in the framework of the typology of syllable and words languages  

 365

For example, the pretonic vowel in the lexical item carrer ‘street’ was phonetically transcribed as [ɐ] in Sort and Àger (localities 95 and 107 on the map). It is not clear whether this process is motivated by a language-internal change or by the influence of Central Catalan (see Recasens 1996: 92 for a discussion). On the other hand, there is a current loss of centralization in the Central Catalan variety spoken in Barcelona, which is conditioned by language contact, especially with Spanish (Lleó, Benet, and Cortés 2008). Similarly, the language contact situation between Alghero Catalan and Sardinian must have also led to the loss of the centralized vowel (PALDC I: 39–40, IV: 463–465). In both cases the centralized vowel [ə], absent in Spanish and Sardinian, has been replaced by the full vowel [a]. Further, Map 4 depicts the distribution of the merger of /ɔ o/ to /u/ in unstressed position in the lexical item port-ar ‘bring-inf’ [puɾˈta]. In Eastern Catalan the distinction between unstressed /o/ and /u/ is made, with some exceptions, in the Balearic variety of Majorcan Catalan (mallorquí, see Map 2). The centralization and merger of back vowels can be synchronically observed in verbal inflection and derivation when verbal endings and suffixes carry the stress. Centralization as a consequence of the stress shift in verbal inflection is exemplified in Table 4. The table displays the infinitive form and the present indicative paradigm of the verbs parl-ar ‘speak-inf’, pens-ar ‘think-inf’, and menj-ar ‘eat-inf’ (see section 5.5 for the deletion of word-final r). The infinitive ending (‑ar) and the ending of the first and second person plural (-em, -eu) are stressed and trigger the centralization of the preceding vowel in the stem (/a/, /ɛ/, and /e/, respectively). The instances of centralization in the stem are shaded in the table. Examples of centralization and merger resulting from derivational processes are found in features 1 and 2 in Table 1, in which the diminutive suffix -et is stressed.5

5 Note that the diminutive suffix is pronounced [et] in Western Catalan and [ɛt] in Eastern Catalan (except for Alghero Catalan).

366 

 Javier Caro Reina

Table 4: Present indicative paradigm of the verbs parl-ar ‘speak-inf’, pens-ar ‘think-inf’, and menj-ar ‘eat-inf’ inf.

parlar

[pəɾˈla]

pensar

[pənˈsa]

menjar

[məɲˈʒa]

pres. ind.

parlo

[ˈpaɾlu]

penso

[ˈpɛnsu]

menjo

[ˈmeɲʒu]

parles

[ˈpaɾləs]

penses

[ˈpɛnsəs]

menges

[ˈmeɲʒəs]

parla

[ˈpaɾlə]

pensa

[ˈpɛnsə]

menja

[ˈmeɲʒə]

parlem

[pəɾˈlɛm]

pensem

[pənˈsɛm]

mengem

[məɲˈʒɛm]

parleu

[pəɾˈlɛw]

penseu

[pənˈsɛw]

mengeu

[məɲˈʒɛw]

parlen

[ˈpaɾlən]

pensen

[ˈpɛnsən]

mengen

[ˈmeɲʒən]

Stress-dependent vowel reduction is not rare among Romance languages. Selected cases are given in Table 5. Not every vowel reduction in unstressed position necessarily implies a vowel centralization or a merger of back vowels. A further reduction of the vowel inventory in unstressed position takes place when mid lax vowels may not occur in unstressed syllables, and are realized as tense. Such reductions apply, for example, in North-Western Catalan and standard Italian. The dotted line in the table separates this kind of vowel reduction from the vowel centralization processes found in Central Catalan, Romansh, European Portuguese, and Neapolitan. In Central Catalan, centralization occurred after mid lax and tense vowels had merged as tense in unstressed position. In this sense, the unstressed vocalism of North-Western Catalan represents an older stage of Central Catalan.­For further examples of centralization in Italian dialects see Schmid (this volume). Table 5: Stress-dependent vowel reduction in selected Romance languages Variety

Stressed syllable

Unstressed syllable

Sources

North-Western Catalan

i, e, ɛ, a, ɔ, o, u

i, e (< e, ɛ), a, o (< ɔ, o), u

Mascaró (2002)

Italian

i, e, ɛ, a, ɔ, o, u

i, e (< e, ɛ), a, o (< ɔ, o), u

Schmid (1999: 133)

Central Catalan

i, e, ɛ, a, ɔ, o, u

i, ə (< e, ɛ, a), u (< ɔ, o, u)

Mascaró (2002)

Romansh

i, e, ɛ, a, ɔ, o, u

i, ə (< e, ɛ, a), u (< ɔ, o, u)

Montreuil (1999: 527)

European Portuguese

i, e, ɛ, a, ɔ, o, u

i, ə (< e, ɛ), ɐ (< a), u (< ɔ, o, u)

Mira (2003)

 Central Catalan in the framework of the typology of syllable and words languages  

 367

Variety

Stressed syllable

Unstressed syllable

Sources

Neapolitan

i, e, ɛ, a, ɔ, o, u

pretonic: i, a, ə (< e, ɛ), u (< ɔ, o, u) posttonic: ə (< i, e, ɛ, a, ɔ, o, u)

Ledgeway (2009: 71–83)

With regard to the occurrence of diphthongs in stressed and unstressed position, Recasens (1996: 123–125) reports the simplification of the unstressed diphthongs [jV], [wV], [Vj], and [Vw] in Central Catalan, especially in allegro forms. Some examples from the ALDC are presented in Table 6. While the unstressed diphthongs [aw ew] are retained in North-Western Catalan in the lexical items abeurador [aβewɾaˈðo] ‘water trough’ and albarda [awˈβaɾða] ‘pack-saddle’ (with l-vocalization), the unstressed diphthong [əw] may be simplified to [u] in Central Catalan giving [əβuɾəˈðo] and [uˈβaɾðə], respectively. It should be noted that this simplification occurs exclusively in unstressed position and hence behaves differently than the contextually conditioned simplification which takes place when the onglide of the diphthong [wV] is absorbed by a preceding velar consonant, regardless of whether the onglide is in a stressed or unstressed syllable. Features 5 and 6 in Table 1 refer to this process (see Recasens 1996: 303–305 for details). Table 6: Simplification of diphthongs in unstressed position North-Western Catalan

Central Catalan

ALDC II/249 abeurador ‘water trough’

[aβewɾaˈðo]

[əβəwɾəˈðo] ~ [əβuɾəˈðo]

ALDC IV/933 albarda ‘pack-saddle’

[aɫˈβaɾða] ~ [awˈβaɾða] (with l-vocalization)

[əɫˈβaɾðə] ~ [əwˈβaɾðə] (with l-vocalization) ~ [uˈβaɾðə]

4.2 Position-related restrictions Trubetzkoy (1968: 43–45) distinguishes four types of word boundary markers: (1) phonematic and aphonematic, (2) individual and group signals, (3) word and morpheme boundary markers, and (4) positive and negative boundary signals. In what follows, I will concentrate on the positive and negative boundary signals. Trubetzkoy speaks of a “positive signal” when a sound (or group of sounds) appears exclusively in a certain position, for example in the word-initial onset, thereby indicating the left margin of the word. A “negative signal”, in contrast, does not highlight any position of the word because the sound involved may

368 

 Javier Caro Reina

appear in different positions (e.g. in the word-initial and word-medial onset). I will use the term “negative signal” here to refer to sounds that do not appear exclusively in one certain position. Certainly, positive signals provide a better mechanism for associating the presence of a sound with a particular position in the word. In order to examine whether Central Catalan highlights word-initial, word-medial and/or word-final positions by means of phonotactic restrictions, the consonant inventory of Central Catalan, as given in Table 7, is arranged in Table 8 according to the occurrence or non-occurrence of single consonants in the word-initial onset, the word-medial onset and coda, and the word-final coda (geminates and consonant clusters are not treated). Table 7: Central Catalan consonants (simplified from Julià 2002: 56) Bilabial Stop

p

Labiodental

b

Fricative

f

Affricate Nasal

Prepalatal

t

d

s

z

ʃ

ʒ

t͡s

d͡z

t͡ʃ

d͡ʒ

Palatal

n

ɲ

ɫ

ɫ

Tap

ɾ

Trill

r w

β

ð

Velar k

Lateral

Approximant

m

v

Alveolar

j

g

ŋ

ɣ

Thus, [β] is a positive signal since it only occurs word-medially (e.g. avi [ˈaβi] ‘grandfather’) but not word-initially or word-finally. Conversely, [b] is a negative signal since it may occur word-initially (e.g. bota [ˈbɔtə] ‘boot’) and word-medially (e.g. canvi [ˈkambi] ‘change’) but not word-finally. In Central Catalan, [β ð ɣ v d͡z d͡ʒ] are positive signals while [b d g z ʒ t͡s ŋ ɾ] are negative signals. Some of these restrictions can be accounted for by phonological processes. For example, the non-occurrence of the voiced obstruents [b d ɡ z ʒ] in word-final position is due to the word-final obstruent devoicing rule (see section 5.1). Crucially, the positive and negative signals described for Central Catalan only function in words in isolation. In the speech stream, however, these signals mostly disappear. As for the positive signals, [β ð ɣ] may also occur word-initially when preceded by a vowel (e.g. la bota [ləˈβɔtə] ‘art.det.fem boot’), and [v d͡z d͡ʒ] may also be found in word-final position when word-final [f t͡s t͡ʃ] become voiced through resyllabification (see section 5.3). As for the negative signals, [b d ɡ z ʒ] may also be found in

 369

 Central Catalan in the framework of the typology of syllable and words languages  

Table 8: Distribution of single consonants in Central Catalan according to the position in the word (adapted and simplified from Julià 2002) word-initial onset #C

word-medial onset and coda C.C, V.C (onset)

C.C (coda)

word-final coda C#

[p]

+

+

+

+

[b]

+

+

+

–

[t]

+

+

+

+

[d]

+

+

+

–

[k]

+

+

+

+

[ɡ]

+

+

+

–

[f]

+

+

+

+

[v]

–

–

+

–

[s]

+

+

+

+

[z]

+

+

+

–

[ʃ]

+

+

–

+

[ʒ]

+

+

–

–

[t͡s]

–

+

+

+

[d͡z]

–

+

–

–

[t͡ʃ]

+

+

–

+

[d͡ʒ]

–

+

–

–

[m]

+

+

+

+

[n]

+

+

+

+

[ɲ]

+

+

–

+

[ŋ]

–

–

+

+

[ɫ]

+

+

+

+

[ʎ]

+

+

–

+

[ɾ]

–

+

+

+

[r]

+

+

+

+

[β]

–

+

–

–

[ð]

–

+

–

–

[ɣ]

–

+

–

–

[j]

+

+

–

+

[w]

+

+

–

+

370 

 Javier Caro Reina

word-final position when word-final [p t k s ʃ] are followed by a voiced consonant thereby becoming voiced according to the coda voice agreement rule (e.g. cap riu ‘no river’ /ˈkap#ˈɾiw/ → [ˈkabˈriw]). Only the negative signals [t͡s ŋ ɾ] remain unaltered in the speech stream. Phonotactic restrictions allowing consonants to appear exclusively in certain positions of the word is a strategy fully exploited in languages such as !Xóõ (see section 2). However, this does not apply in Central Catalan, where stress-related restrictions are more effective than position-related restrictions. The stress-related restrictions involve vowel reduction and the simplification of diphthongs in unstressed position.

5 Word-related processes The word-related processes that will be analyzed are word-final obstruent devoicing, affrication of word-final /ʒ/, obstruent voicing across words, deletion of unstressed (centralized) vowels, deletion of word-final n and r, simplification of word-final homorganic consonants, and reinforcement of word-final r through consonant epenthesis. Vowel reduction processes have already been discussed in section 4.1. The syllable-related processes that exist in Central Catalan are also found in other Catalan varieties and need not be treated in detail. These processes include resyllabification, lenition of intervocalic voiced stops, coda voice agreement, and hiatus resolution.

5.1 Word-final obstruent devoicing As shown in Table 8, voiced obstruents are not permitted word-finally. This phonotactic restriction results from the word-final obstruent devoicing rule which applies in all Catalan dialects. Examples from Central Catalan are given in (4). We can observe that underlying /b d ɡ z ʒ d͡ʒ/ become devoiced when they occur word-finally. Note that the underlying voiced stops /b d ɡ/ are realized as the approximants [β ð ɣ] in intervocalic position (or between vowel and liquid). As we will see in section 5.2, underlying /ʒ/ in roj-a [ˈrɔʒə] ‘red-fem’ undergoes affrication in word-final position.

 Central Catalan in the framework of the typology of syllable and words languages  

 371

(4) Word-final obstruent devoicing sab-en [ˈsaβən] ‘know-3pl.pres.ind’ sap [sap] ‘know[3sg.pres.ind]’ perd-en [ˈpɛɾðən] ‘lose-3pl.pres.ind’ perd [pɛɾt] ‘lose[3sg.pres.ind]’ grog-a [ˈɡɾɔɣə] ‘yellow-fem’ groc [ɡɾɔk] ‘yellow[masc]’ pagès-a [pəˈʒɛzə] ‘farmer-fem’ pagès [pəˈʒɛs] ‘farmer[masc]’ roj-a [ˈrɔʒə] ‘red-fem’ roig [rɔt͡ʃ] ‘red[masc]’ mitj-a [ˈmidd͡ʒə] ‘half-fem’ mig [mit͡ʃ] ‘half[masc]’

Underlying voiced stops in word-final position remain voiceless even when resyllabification occurs (e.g. sap anglès [ˈsa pəŋˈɡlɛs] ‘know[3sg.pres.ind] English’). ᵕ In other words, they do not behave as they do in word-medial intervocalic position (e.g. sab-en [ˈsaβən] ‘know-3pl.pres.ind’). Substantial evidence that clitics belong to the phonological word is provided by the example in (5). Compare rep una carta [ˈrɛ punəˈkaɾtə] ‘s/he receives a letter’ and rep-ho [ˈrɛβu] ‘receive it!’. ᵕ While the former shows obstruent final devoicing of underlying /b/, the latter does not. Crucially, word-final obstruent devoicing is overruled by coda voice agreement, a syllable-related assimilation process whereby obstruents, both underlying voiced and voiceless, are voiced or devoiced depending on whether the following consonant is voiced or voiceless. Thus, underlying /p/, for example, is realized as voiceless in cap tarda [ˈkapˈtaɾðə] ‘no afternoon’ and voiced in cap riu [ˈkabˈriw] ‘no river’. (5) Word-final obstruent devoicing and clitics rep-ho [ˈrɛβu] ‘receive[2sg.imp]-it’ rep una carta [ˈrɛ punəˈkaɾtə] ‘receive[3sg.pres.ind] a letter’ ᵕ

372 

 Javier Caro Reina

5.2 Affrication of word-final /ʒ/ The examples in (6) show that underlying /ʒ/ in roj-a [ˈrɔʒə] and rog-et [ruˈʒɛt] undergoes affrication when occurring word-finally. (6) Affrication of word-final /ʒ/ roj-a [ˈrɔʒə] ‘red-fem’ rog-et [ruˈʒɛt] ‘red-dim[masc]’

roig [rɔt͡ʃ] ‘red[masc]’

Bonet and Lloret (1998: 105–108) draw attention to the fact that affrication takes place before word-final obstruent devoicing. They argue that a deaffrication rule in intervocalic position cannot be assumed since it would have also applied in cases such as lletjor [ʎədˈd͡ʒo] ‘ugliness’, mitj-o-s [ˈmidd͡ʒus] ‘half-masc-pl’, and rebutj-ar [rəβudˈd͡ʒa] ‘reject-inf’, which would have resulted in *[ʎəˈʒo], *[ˈmiʒus], and *[rəβuˈʒa], respectively. These cases, which involve geminates, are to be treated separately. Certainly, the affrication process assumed to occur in word-final position as shown in (6) is highly marked. Deaffrication in word-medial position, in turn, is widely attested in the languages of the world. A comparison of the realization of word-medial [ʒ] in Central Catalan, Southern Valencian (valencià meridional, see Map 2), and Majorcan Catalan (mallorquí, see Map 2) will throw light on the nature of word-final affrication. Table 9 shows that the conservative dialects Southern Valencian and Majorcan Catalan have word-medial affricate [d͡ʒ] whereas Central Catalan has [ʒ]. These observations lead us to conclude that Southern Valencian and Majorcan Catalan have retained the affricate while Central Catalan has experienced a deaffrication process. Therefore, strictly speaking, we should assume for Central Catalan a word-medial /d͡ʒ/ that undergoes deaffrication in word-medial position and devoicing in word-final position. The affrication of word-final /ʒ/ in Central Catalan as described by Bonet and Lloret (1998: 107) and Pradilla (2002: 301–302) – and its interpretation as a word-related process – only holds from a synchronic point of view. Table 9: Counterparts of Central Catalan [ʒ] in selected Catalan dialects (transcriptions taken from the DCVB) Example

Central Catalan

Southern Valencian

Majorcan Catalan

passej-ar ‘walk-inf’ passeig ‘walk’

[pəsəˈʒa] [pəˈsɛt͡ʃ]

[paseˈd͡ʒaɾ] [paˈset͡ʃ]

[pəsəˈd͡ʒa] [pəˈsət͡ʃ]

roj-a ‘red-fem’ roig ‘red[masc]’

[ˈrɔʒə] [rɔt͡ʃ]

[ˈrɔ̞d͡ʒa] [rɔ̞t͡ʃ]

[ˈrɔ̞d͡ʒə] [rɔ̞t͡ʃ]

 Central Catalan in the framework of the typology of syllable and words languages  

 373

5.3 Obstruent voicing across words As we have seen in section 5.1, voiced obstruents become devoiced in word-final position due to the word-final obstruent devoicing rule (see (4) for examples). However, in Central Catalan the word-final voiceless fricatives [f s ʃ] and affricates [t͡ʃ t͡s] undergo voicing when they are resyllabified, as illustrated in (7) (resyllabification is indicated with [ ]). For example, underlying word-final /s/ ᵕ in cos /kɔs/ ‘body’ remains voiceless (cf. the plural form coss-os [ˈkɔsus] ‘bodypl’) while underlying word-final /z/ in gris /ˈɡɾiz/ ‘grey[masc]’ becomes devoiced (cf. the plural form gris-os [ˈɡɾizus] ‘grey-pl’). When cos and gris are followed by a word beginning with a vowel, resyllabification takes place and the voiceless sibilant becomes voiced as in cos estrany ‘body foreign[masc]’ /ˈkɔs#esˈtɾaɲ/ → [ˈkɔ zəsˈtɾaɲ] and gris intens ‘grey intense[masc]’ /ˈgɾiz#inˈtɛns/ → [ˈgɾi zinˈtɛns]. ᵕ ᵕ (7) Obstruent voicing across words (examples taken from Prieto 2004: 213–217) fotògraf [fuˈtɔɣɾəf] fotògraf estranger ‘photographer’ [fuˈtɔɣɾəᵕvəstrəɲˈʒe] ‘photographer foreign[masc]’ cos [kɔs] ‘body’ cos estrany [ˈkɔᵕzəsˈtɾaɲ] ‘body foreign[masc]’ gris [ɡɾis] ‘grey[masc]’ gris intens [ˈɡɾiᵕzinˈtɛns] ‘ grey intense[masc]’ feix [feʃ] ‘burden’ feix enorm-e [ˈfeᵕʒəˈnoɾmə] ‘burden great-masc’ passeig [pəˈsɛt͡ʃ] ‘walk’ passeig agradabl-e [pəˈsɛᵕd͡ʒəɣɾəˈðab.blə] ‘walk pleasant-masc’ mig [mit͡ʃ] ‘half[masc]’ mig humit [ˈmiᵕd͡ʒuˈmit] ‘half[masc] humid[masc]’ despatx [dəsˈpat͡ʃ] ‘office’ despatx ampl-e [dəsˈpaᵕˈd͡ʒamplə] ‘office wide-masc’

Note that consonant clusters containing a sibilant as the realization of a plural morpheme or a second-person singular morpheme (e.g. [ps], [ts], and [ks]) also undergo voicing, as can be observed in (8): (8) Obstruent voicing across words a. sap-s [saps] sap-s anglès [ˈsaᵕbzəŋˈɡlɛs] ‘know-2sg.pres.ind’ ‘know-2sg.pres.ind English’ b. estimat-s [əstiˈmat͡s] estimat-s amic-s ‘dear[masc]-pl’ [əstiˈmaᵕd͡zəˈmiks] ‘dear[masc]-pl friend-pl’ c. amic-s [əˈmiks] ‘friend-pl’ amic-s estimat-s [əˈmiᵕɡzəstiˈmat͡s] ‘friend-pl dear[masc]-pl’

374 

 Javier Caro Reina

The Catalan dialects behave differently regarding the obstruent voicing process. Table 10 shows the patterns of this process in selected Catalan dialects. Interestingly, stops do not undergo voicing in Central Catalan as they do in Southern Valencian. Compare sac ampl-e ‘sack spacious-masc’ pronounced [ˈsaᵕˈkamplə] in Central Catalan and [ˈsaᵕˈɣample] in Southern Valencian (example taken from Veny 2002: 112). Table 10: Obstruent voicing across words in selected Catalan dialects (Lloret and Jiménez 2009: 42) SibFric#V

Affr#V

f#V

Stop#V

Southern Valencian

+

+

+

+

Central Catalan

+

+

+/–

–

Alghero Catalan

+

–

–

–

Central Valencian

–

–

–

–

Obstruent voicing poses a problem in part because resyllabification is a syllable-related process, in part because voicing applies at the word level. Resyllabification is a syllable-related process occurring at the level of the phonological phrase which optimizes the syllable structure; the word-final consonant (or semivowel) of a word fills the word-initial empty onset of the next word, thereby blurring the word and morpheme boundaries. In German, for example, resyllabification is avoided through the addition of a glottal stop. In contrast, Catalan allows resyllabification. Compare New High German um elf [ˈɁʊmˈɁɛlf] ‘at eleven o’clock’ and Central Catalan a les onze [ələᵕˈzonzə] ‘at eleven o’clock’. The inevitable question that arises is whether there are any mechanisms that highlight the word boundaries despite resyllabification. Obstruent voicing might be a possible mechanism; that is, resyllabification, though usually regarded as a syllable-related feature, is not necessarily at odds with a word language when processes such as obstruent voicing are at work. This process clearly helps distinguish between els avis [əlᵕˈzaβis] ‘the grandparents’ and els savis [əlˈsaβis] ‘the wise ones’. Note that in Catalan a sequence of two identical fricative sounds does not result in a geminate. Instead, the sequence is simplified to one single segment. This process is termed fusion (fusió) (Dols 2002: 338). We refer to Contreras, Mateu, and Pons (2004: 528–529) and Pons (2009: 49–50) for a discussion on the issue of whether obstruent voicing has a word or morpheme delimitative function. For examples of obstruent voicing across words in Luxembourgish, see Gilles (this volume).

 Central Catalan in the framework of the typology of syllable and words languages  

 375

5.4 Deletion of unstressed vowels The deletion of unstressed vowels is consistent with word languages since it can lead to complex syllable structures that might serve as word boundary signals. Thereby new consonant clusters, which were phonotactically not permitted, may arise. A striking case is literary Georgian. The word-initial complex syllable structure in the stem /prckvna/ ‘to peel’ resulted from vowel deletion, as a comparison with other related Kartvelian languages clearly shows. For example, the corresponding form in Megrelian is /purckon-u-a/ (Butskhrikidze 2002: 148–150). In Central Catalan, deletion processes affecting unstressed centralized vowels occur especially in pretonic position between stop and r, as shown in (9) (deletion processes occurring in word-initial position are not treated). The deletion of unstressed vowels in Central Catalan does not result in new consonant clusters that highlight the word margins (e.g. català ‘Catalan[masc]’ [kətəˈla] > *[ktla]). In other words, the resulting syllable structure is well formed. (9) Deletion of centralized vowels barana ‘banister’ [bəˈɾanə] > [ˈbɾanə] taronja ‘orange’ [təˈɾɔɲʒə] > [ˈtɾɔɲʒə] veritat ‘truth’ [bəɾiˈtat] > [bɾiˈtat]

Unstressed vowel deletion is also attested in Catalan varieties which lack centralized vowels in unstressed syllable. For this reason, we cannot assume that deletion is always preceded by centralization processes. Nevertheless, deletion between stop and r are more frequently reported for Central Catalan and Roussillon Catalan than for any other Catalan dialect (Recasens 1996: 116–118). In that respect, centralized vowels are more prone to deletion than full vowels. The deleted forms given in (9) may actually co-occur with the corresponding non-deleted forms, depending on the speech tempo (e.g. barana ‘banister’ [bəˈɾanə] ~ [ˈbɾanə]). The ALDC charts the pronunciation of lexical items such as baranes ‘bars [of the cart]’ (ALDC IV/914), caramell ‘sweet’ (ALDC III/682), and paret ‘wall’ (ALDC II/217). The deletion observed on these maps may already be lexified.

5.5 Deletion of word-final n and r The deletion of word-final n and word-final r will be treated together for practical reasons. I will only give a brief overview and refer to Bibiloni (2002) for a detailed description. Examples of deletion of word-final n and r in Central Catalan are shown in (10) and (11), respectively (deletion is indicated with ∅). Note that only deletion of word-final n is reflected in the writing. Deletion of word-final n and

376 

 Javier Caro Reina

r occurs in most Catalan dialects except for the subdialect Ribagorçà (see Map 2) in the case of deletion of word-final n and Valencian in the case of deletion of word-final r (Badia 1994: 241–245; Moll 2006: 115–116; PALDC I: 66–67). If deletion of r were a syllable-related process, we would expect all instances of r to be deleted both in the word-medial and word-final coda. For example, in the lexical item port-ar [puɾˈta∅] ‘bring-inf’ only the r in the word-final coda is deleted (and not in the word-medial coda, which would have resulted in *[pu∅ˈta∅]). The same applies for word-final n. For example, in the lexical item cançó [kənˈso∅] ‘song’ only the n in the word-final coda is deleted while the n in the word-medial coda is retained. According to De Yzaguirre’s data, deletion of word-final r occurs in 12,131 cases. As a consequence, the surface word-final type (C)(C)V is 2% more frequent in Central Catalan than in Valencian (see Table 3). (10) Deletion of word-final n cançó [kənˈso∅] ‘song’ camí [kəˈmi∅] ‘way’

cançon-et-a [kənsuˈnɛtə] ‘song-dim-fem’ camin-ar [kəmiˈna] ‘walk-inf’

(11) Deletion of word-final r primer [pɾiˈme∅] ‘first[masc]’ primer-a [pɾiˈmeɾə] ‘first-fem’ primerenc [pɾiməˈɾɛŋ] ‘early[masc]’ ferrer [fəˈre∅] ‘smith’ ferreria [fərəˈɾiə] ‘smithy’

Further evidence of clitics belonging to the phonological word is presented in (12). Underlying /ɾ/ is deleted when it occurs word-finally as in portar la roba [puɾˈta∅ləˈrɔβə] and fer negocis [ˈfe∅nəˈɣɔsis]. However, when followed by a clitic (or a clitic group) as in portar-la [puɾˈtaɾlə] and fer-ne [ˈfeɾnə], underlying /ɾ/ behaves as in the word-medial coda. (12) Deletion of word-final r and clitics a. port-ar la roba [puɾˈta∅ləˈrɔβə] ‘bring-inf art.det.fem[sg] clothes’ port-ar-la [puɾˈtaɾlə] ‘bring-inf-pron.fem[sg]’ b. fer negoci-s [ˈfe∅nəˈɣɔsis] ‘do.inf business-pl’ fer-ne [ˈfeɾnə] ‘do.inf-part’

 Central Catalan in the framework of the typology of syllable and words languages  

 377

5.6 Simplification of word-final homorganic consonants The simplification of consonant clusters is a process that cannot be clearly ascribed to a syllable language or a word language. On the one hand, simplification enhances the syllable structure, which is a feature typical of a syllable language. On the other hand, this simplification occurs when complex syllable structures are given, a feature typical of a word language. For this reason, Auer (2001: 1397) speaks of a secondary (or indirect) word language feature when dealing with assimilation processes at word boundaries. Assimilation processes involving the simplification of word-final homorganic consonants are found in North-Western Catalan and Central Catalan. In these dialects the homorganic groups /mp nt ŋk lt/ are simplified in word-final position to [m n ŋ ɫ], respectively (Badia 1994: 249–252; see Dols 2002: 331–340 for a detailed description). Map 5 charts the simplification of word-final [nt] in the lexical item font ‘fountain’ on the basis of the ALDC. Examples from Central Catalan are given in (13). Note that simplification of homorganic consonant clusters only applies word-finally as in camp [kam] ‘field’ and (regardless of resyllabification) camp és [ˈkaᵕˈmɛs] ‘field be.3sg.pres.ind’, and not word-medially as in camp-et [kəmˈpɛt] ‘field-dim[masc]’. This and other simplification processes that occur word-finally are the reason why the word-final syllable type (C)(C)VCC is 3% lower in Central Catalan than in Valencian (see Table 3). The possible motivation for this simplification will be treated in the discussion. (13) Simplification of word-final homorganic consonants camp [kam] ‘field’ camp-et [kəmˈpɛt] ‘field-dim[masc]’ camp és [ˈkaᵕˈmɛs] ‘field be.3sg.pres.ind’ font [fɔn] ‘fountain’ font-et-a [funˈtɛtə] ‘fountain-dim-fem’ font és [ˈfɔᵕˈnɛs] ‘fountain be.3sg.pres.ind’ tronc [tɾoŋ] ‘trunk’ tronqu-et [tɾuŋˈkɛt] ‘trunk-dim[masc]’ tronc és [ˈtɾoᵕˈŋɛs] ‘trunk be.3sg.pres.ind’ molt [moɫ] ‘much[masc]’ molt-e-s [ˈmoɫtəs] ‘much-fem-pl’ molt especial [ˈmoᵕɫəspəsiˈaɫ] ‘very special’

378 

 Javier Caro Reina

Map 5: Simplification of word-final homorganic consonants

 Central Catalan in the framework of the typology of syllable and words languages  

 379

5.7 Word-final consonant epenthesis Epenthetic consonants can have a twofold function. First, they can enhance illformed syllable contacts as in the development from Old Catalan genre [ˈd͡ʒɛnɾe] into Central Catalan gendre [ˈʒɛndɾə] ‘son-in-law’. In Central Catalan, the consonant [ɣ] and the semivowels [j w] are usually used for hiatus resolution as in du-e-s ‘two-fem-pl’ [ˈduəs] > [ˈduɣəs] (see Recasens 1996: 246, 300, 302 for details). And second, they can make a simple syllable structure complex, for instance word-finally, thereby demarcating the right margin of the phonological word. Whereas the former is a typical syllable language process, the latter is found in word languages. I will concentrate on consonant epenthesis processes that enhance the phonological word. We can distinguish between consonant epenthesis processes that reinforce word-final open syllables (epítesi postvocàlica) and those that reinforce word-final closed syllables (epítesi postconsonàntica) (see Recasens 1996: 224–226, 247 for details). There is a tendency common to all Catalan dialects to add an epenthetic ‑[t] when words end in a posttonic -i, and an epenthetic -[t] or ‑[k] when words end in a posttonic -e. As a result, word-final open syllables become closed. Examples from Central Catalan are presented in (14) (see Martí 2006: 167–170 for further examples). (14) Reinforcement of word-final open syllables through consonant epenthesis api ‘celery’ [ˈapi] ~ [ˈapit] col·legi ‘school’ [kuɫˈɫɛʒi] ~ [kuɫˈɫɛʒit] tave ‘horsefly’ [ˈtaβə] ~ [ˈtaβət] ~ [ˈtaβək] rave ‘radish’ [ˈraβə] ~ [ˈraβət] ~ [ˈraβək]

The epenthesis of a stop (‑[t] or -[k]) in monosyllabic and bisyllabic words that have not undergone deletion of word-final r takes place exclusively in Central Catalan and Roussillon Catalan (see feature 10 in Table 1). Some examples are given in (15). Interestingly, this epenthesis is not attested in North-Western Catalan, in which deletion of word-final r occurs. (15) Reinforcement of word-final r through consonant epenthesis (Moll 2006: 153) cor ‘heart’ [kɔɾ] > [kɔɾt] mar ‘sea’ [maɾ] > [maɾt]

Map 6 shows the distribution of the epenthetic consonant in the lexical item cor ‘heart’ on the basis of the ALDC. It should be noted that this process has not affected all instances of word-final r in the same way. For example, the epenthesis in the lexical item ahir ‘yesterday’ (ALDC III/641) is attested in a smaller area. Badia (1994: 259) and Moll (2006: 153) explain this process as a reinforcing epenthesis (epèntesi com a reforç, t de suport). Veny (2001a: 130, 2013; PALDC I:

380 

 Javier Caro Reina

Map 6: Word-final consonant epenthesis

 Central Catalan in the framework of the typology of syllable and words languages  

 381

65) claims that the epenthetic consonant -[t] originated with the simplification of word-final -[ɾt] > -[ɾ], which is mainly found in Central Catalan (see Recasens 1996: 222 for details). The alternation in words with etymological -rt (e.g. fort ‘strong[masc]’ [fɔɾt] ~ [fɔɾ], part ‘part’ [paɾt] ~ [paɾ], etc.) would have motivated the epenthesis in words with (preserved) etymological -r as in cor [kɔɾt] and mar [maɾt]. This, however, only holds for epenthetic -[t]. Crucially, epenthetic -[k] in words such as mar [maɾk] ‘sea’ (see the DCVB for its distribution) cannot be explained as a result of the simplification of word-final ‑[ɾk] since this consonant cluster is not subject to simplification in the Catalan dialects (Recasens 1996: 243; Prieto 2004: 233). The fact that both epenthetic consonants are found in the same area leads us to the conclusion that they pursue the same goal. Not only do they reinforce word-final r preserving it from deletion, they also function as word boundary markers. For examples of word-related epenthetic consonants in German see Szczepaniak (this volume).

6 Summary and discussion This paper has approached the traditional classification of Catalan dialects from the perspective of the typology of syllable and word languages. While theories based on extralinguistic factors such as the pre-Latin substrate, the Roman colonization, or the Christian reconquest fail to satisfactorily explain the phonological differences between the Western and Eastern Catalan groups, the typology succeeds in showing that Eastern Catalan has undergone a further phonological development which has not taken place in the Western Catalan dialects. Focusing on Central Catalan, I have presented evidence supporting the existence of the syllable and the phonological word as prosodic domains by analyzing three different parameters: the syllable structure (section 3), the phonotactic restrictions on the phoneme inventory (section 4), and phonological processes (section 5). Further, I have studied these parameters according to two aspects: the stressed and unstressed position and the word-initial, word-medial and word-final position. These aspects have proved particularly fruitful for determining the prosodic domains of the syllable and the phonological word. With regard to the syllable structure, Central Catalan has the surface syllable template (C)(C)V(C)(C)(C), which counts – following Maddieson’s (2013) classification – as complex. Underlying complex syllables were found, as is typical of word languages, most commonly in stressed (9%) and word-final (10%) position. An account of the frequency of the surface syllable types in word-final position has shown that 7% of all word-final syllables are complex. Appendices occur mainly word-finally (e.g. corb-s [kɔɾps] ‘raven-pl’), especially when word-medial

382 

 Javier Caro Reina

appendices of Latin prefixes are simplified (e.g. explicació ‘explanation’ [əks.pli. kə.siˈo] ~ [əs.pli.kə.siˈo]). The influence of the written language must have led to the reintroduction of these complex syllable codas, thereby modifying the typological drift. Phonotactic restrictions were mainly found in the vocalism. Central Catalan is characterized as having a seven-to-three vowel reduction in unstressed syllables. The vowel height contrasts are neutralized by means of centralization and merger. The occurrence or non-occurrence of single consonants in specific positions in the word only takes place when phonological words appear in isolation. In the speech stream, the positive and negative signals disappear (except for the negative signals [t͡s ŋ ɾ]). In this regard, only restrictions found in the vocalism help highlight the phonological word. The word-related stress-dependent processes are centralization of /a ɛ e/ in unstressed position, merger of /ɔ o/ and /u/ to [u] in unstressed position, simplification of diphthongs in unstressed position, and deletion of centralized vowels (especially between stop and r). The word-related processes regarding the position in the word are word-final obstruent devoicing, affrication of word-final /ʒ/ (provided /d͡ʒ/ is not assumed as the underlying form), obstruent voicing across words, deletion of word-final n and r, simplification of the word-final homorganic consonants /mp nt ŋk lt/, and reinforcement of word-final r. Finally, syllable-related processes such as resyllabification, lenition of intervocalic voiced stops, coda voice agreement, and hiatus resolution clearly point to the existence of the syllable as a further prosodic domain. The strategies that highlight the phonological word are summarized in Table 11, following the model depicted in Figure 1. Summing up, the syllable and the phonological word are prosodic domains that coexist in Central Catalan. From the examples in (5) and (12) we can conclude that the size of the phonological word does not coincide with the size of the morphological word since clitics are integrated into the phonological word. Interestingly, the domain of the syllable has lost out to the domain of the phonological word only with regard to the word-final syllable structure and phonological restrictions such as vowel reduction in unstressed position. The syllable-related phonological processes, on the other hand, have been retained. Consequently, phonological processes can operate both on the syllable and the word level without necessarily implying a collision between both domains.

 Central Catalan in the framework of the typology of syllable and words languages  

 383

Table 11: Syllable structure, phonotactic restrictions, and phonological processes according to stress and position in Central Catalan stress-related (stressed vs. unstressed)

position-related (word-initial, word-medial, word-final)

Syllable structure

1) underlying complex syllable structures in stressed position

1) complex syllable structures in wordfinal position (7% of all word-final surface syllable structures) 2) appendices mainly in word-final position (when codas in Latin prefixes containing appendices are simplified)

Phonotactic restrictions

1) reduced vowel inventory in 1) single consonants seldom function unstressed position (seven-toas word boundary signals (not in the three reduction) speech stream) 2) reduced number of diphthongs in unstressed position

Phonological processes

1) centralization of /a ɛ e/ in unstressed position 2) merger of /ɔ o u/ in unstressed position 3) simplification of diphthongs in unstressed position 4) deletion of unstressed (centralized) vowels

1) word-final obstruent devoicing 2) affrication of word-final /ʒ/ 3) obstruent voicing across word boundaries through resyllabification 4) deletion of word-final n 5) deletion of word-final r 6) simplification of word-final homorganic consonants /mp nt ŋk lt/ 7) reinforcement of word-final r through consonant epenthesis

The evidence provided shows that in Central Catalan the right margin of the phonological word seems to take on a special meaning for two reasons. First, complex syllable structures and appendices mainly appear in this position. And second, the processes related to the position in the word operate exclusively word-finally. The correlation between complex syllable structure and word-related processes observed by Auer (1993: 88) has been thus borne out for Central Catalan. If we compare the phonological processes found in Central Catalan with those in other Catalan varieties such as Southern Valencian and North-Western Catalan, the picture we obtain gives us greater insight into the particular phonological stage reached by Central Catalan. Word-final obstruent devoicing, affrication of word-final /ʒ/, obstruent voicing across words, and deletion of word-final n are phonological processes shared by these three varieties. However, the dele-

384 

 Javier Caro Reina

Table 12: Word-related processes in Central Catalan compared to other Catalan dialects Phonological processes

Southern Valencian

North-Western Catalan

Central Catalan

word-final obstruent devoicing

+

+

+

affrication of word-final /ʒ/

+

+

+

obstruent voicing across words

++

+

+

deletion of unstressed /a ɛ e/ (centralized in Central Catalan)

+

+

++

deletion of word-final n

+

+

+

deletion of word-final r

–

+

+

simplification of word-final homorganic consonants

–

+

+

centralization of /a ɛ e/ in unstressed position

–

–

+

merger of /ɔ o u/ in unstressed position

–

–

+

reinforcement of word-final r

–

–

+

tion of word-final r and the simplification of word-final homorganic consonants take place only in North-Western Catalan and Central Catalan. Reinforcement of word-final r, in the cases in which this consonant was not deleted, is found only in Central Catalan. In addition to these word-related processes operating word-finally, Central Catalan shows further word-related processes such as centralization of non-back non-high vowels in unstressed position, deletion of centralized vowels (deletion of /a ɛ e/ in unstressed position is also found, albeit less frequently, in Western dialects), and merger of back non-low vowels in unstressed position. In this respect, Central Catalan has branched off from Western Catalan dialects by using new strategies related to the vocalism in unstressed position. Strategies related to the stressed position such as vowel lengthening are not attested. Interestingly, the increase of the word-related processes in Central

 Central Catalan in the framework of the typology of syllable and words languages  

 385

Catalan has not triggered a decrease of the syllable-related ones. An overview of the phonological processes occurring in Southern Valencian, North-Western Catalan, and Central Catalan is presented in Table 12. Features that have a wider scope, as in the case of obstruent voicing in Southern Valencian (see Table 10) or that occur more frequently as in the case of unstressed vowel deletion in Central Catalan, are represented with ++. Once evidence has been provided for the existence of the syllable and the phonological word as prosodic domains in Central Catalan, the inevitable question remains as to which prosodic domain is of more relevance. In other words, on a scale going from the syllable language type on the left to the word language type on the right, where exactly would we find Central Catalan? This issue cannot be addressed until we establish whether the syllable-related or word-related character of a given variety is determined on the basis of absolute or comparative indicators.6 For example, a complex syllable structure is, in absolute terms, a word-related strategy. However, we must ask ourselves to what extent this strategy can optimize the phonological word. Thus, a complex syllable structure at the margins of a phonological word has the function of marking the word boundaries. In contrast, a complex syllable structure occur­ ring in word-initial, word-medial, and word-final onsets cannot contribute in the same way to the demarcation of the phonological word. Moreover, we cannot carry out an objective analysis without considering the whole picture of strategies operating in connection with each other. For example, if we compare Valencian and Central Catalan with regard to syllable structure and vocalism in unstressed syllables, we can make the following observations. First, the scope of syllable structure complexity is wider in Valencian than in Central Catalan since Central Catalan has simplified the word-final homorganic consonants /mp nt ŋk lt/ (see Table 3 for differences between the frequency of word-final syllable types in Central Catalan and Valencian). Compare Central Catalan camp [kam] ‘field’, font [fɔn] ‘fountain’, alt [aɫ] ‘old[masc]’, and banc [baŋ] ‘bench’ with Valencian [kamp], [fɔ̞nt], [aɫt], and [baŋk] (see Map 5 for the distribution of the simplification of word-final /nt/ in the lexical item font ‘fountain’). Second, Valencian has retained the vowel height oppositions of unstressed vowels while Central Catalan has neutralized them. In this respect, Valencian represents an earlier stage of Central Catalan. The differences described for both dialects raise some challenging questions. Why did Central Catalan simplify word-final homorganic consonants? And is there a correlation between vowel centralization and the simplifi-

6 I have taken these terms from Siegel (2008: 20), used there to discuss morphological simplicity in pidgin and creole languages.

386 

 Javier Caro Reina

cation of word-final homorganic consonants? If this is the case, we could propose the following hypothesis: Considering that word-final complex syllables occur with a low frequency (10% without applying simplification rules) and that vowel centralization is a more effective strategy affecting a higher number of words, we could assume that complex syllables were no longer needed once unstressed vowels were centralized. The first consonant clusters to be simplified would be the homorganic ones. A compelling piece of evidence for this hypothesis would be the first diachronically attested forms showing centralization and simplification of homorganic consonant clusters. If centralization is attested earlier than simplification, the assumed correlation could be confirmed. These and related questions will be answered precisely by studying the diachronic evolution of Catalan in more detail. Comparative indicators would certainly help assess the scope and degree of the syllable-related or word-related strategies. We could compare word-related strategies such as vowel centralization and complex syllable structure in two (or more) varieties that are closely related, distantly related, or not related at all (for example Central Catalan and Valencian, Central Catalan and German, etc.). Alternatively, we could compare these strategies in different stages of the same variety (for example Central Catalan and Old Catalan). With regard to vowel centralization, we can state that German has exploited this strategy more extensively than Central Catalan. In Middle High German, all unstressed vowels underwent centralization (Szczepaniak 2007: 155–166) while in Eastern Catalan only the nonback non-high vowels were subject to centralization. Determining the degree of centralization becomes more difficult when analyzing Romance languages (see Table 5). With regard to syllable structure complexity, Central Catalan and Valencian share the syllable template (C)(C)V(C)(C)(C). However, Central Catalan has a lower number of complex syllables than Valencian because the simplification of homorganic consonants did not occur in this variety. The frequency of complex consonant clusters in word-final position is approximately 7% in Central Catalan and 10% in Valencian, compared to 17% in standard German (Duanmu 2009: 213). Again, this typological strategy is more widely used in German than in Central Catalan. Finally, in light of the phonological processes highlighted in Table 12, we can conclude that Central Catalan is closer to the word language pole than North-Western Catalan while Valencian is closer to the syllable language pole than North-Western Catalan, as illustrated in Figure 4.

 Central Catalan in the framework of the typology of syllable and words languages  

syllable language

 387

word language Central Catalan North-Western Catalan Valencian

Figure 4: Selected Catalan varieties on the scale of syllable and word languages with regard to word-related phonological processes

References ALDC = Veny, Joan and Lídia Pons (2001‒): Atles lingüístic del domini català [Linguistic Atlas of the Catalan Area]. 6 vols. Barcelona: Institut d’Estudis Catalans. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (1994): Einige Argumente gegen die Silbe als universale prosodische Hauptkategorie. In: Karl Heinz Ramers, Heinz Vater and Henning Wode (eds.), Universale phonologische Strukturen und Prozesse, 55–78. (Linguistische Arbeiten 310.) Tübingen: Niemeyer. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Badia, Antoni (1981): La formació de la llengua catalana. Assaig d’interpretació històrica [The Formation of Catalan. An Attempt at Historical Interpretation]. 2nd ed. Barcelona: Abadia de Montserrat. Badia, Antoni (1994): Gramàtica històrica catalana [Historical Grammar of Catalan]. 3rd ed. València: Climent. Barnes, Jonathan (2006): Strength and Weakness at the Interface. Positional Neutralization in Phonetics and Phonology. (Phonology and Phonetics 10.) Berlin: Mouton de Gruyter. Bibiloni, Gabriel (2002): Elisió de -n i -r, distribució de les ròtiques i altres fenòmens consonàntics en el mot [Deletion of -n and -r, distribution of the rhotic consonants and other consonant processes in the word]. In: Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.), Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia. Morfologia [Grammar of Modern Catalan. Vol. 1: Introduction. Phonetics and Phonology. Morphology], 271–285. Barcelona: Empúries. Bonet, M. Eulàlia and Maria-Rosa Lloret (1998): Fonologia catalana [Phonology of Catalan]. Barcelona: Ariel. Butskhrikidze, Marika (2002): The Consonant Phonotactics of Georgian. LOT: Utrecht. Contreras, Joan Miquel, Jaume Mateu and Clàudia Pons (2004): Review of ‘Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia. Morfologia’ by Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.). Llengua & Literatura 15: 523–556.

388 

 Javier Caro Reina

DCVB = Alcover, Antoni M. and Francesc de B. Moll (1930–1968): Diccionari català-valencià-balear [Dictionary of Catalan, Valencian and Balearic]. 10 vols. Palma de Mallorca: Moll. Available at http://dcvb.iecat.net. De Yzaguirre, Lluís (1995): L’estructura sil·làbica del català central [The syllable structure of Central Catalan]. Ph.D. dissertation, Departament de Filologia Catalana, Universitat de Barcelona. Available at http://latel.upf.edu/morgana/altres/cibres/tesi_lym/estrucsil. htm. Dols, Nicolau A. (2002): Fenòmens en grups consonàntics [Processes in consonant clusters]. In: Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.), Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia. Morfologia [Grammar of Modern Catalan. Vol. 1: Introduction. Phonetics and Phonology. Morphology], 319–343. Barcelona: Empúries. Duanmu, San (2009): Syllable Structure. The Limits of Variation. Oxford: Oxford University Press. Ferrando, Antoni (1989): La formació històrica del valencià [The historical formation of Valencian]. In: Antoni Ferrando (ed.), Segon Congrès Internacional de la Llengua Catalana, VIII, Àrea 7. Història de la Llengua [Second International Congress of Catalan, VIII, Area 7. Historical Linguistics], 399–428. València: IFV. Julià, Joan (2002): Els sons del català [The sounds of Catalan]. In: Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.), Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia. Morfologia [Grammar of Modern Catalan. Vol. 1: Introduction. Phonetics and Phonology. Morphology], 37–87. Barcelona: Empúries. Kouwenberg, Silvia (2010): Creole studies and linguistic typology: Part 2. Journal of Pidgin and Creole Languages 25/2: 359–380. Ledgeway, Adam (2009): Grammatica diacronica del Napoletano. (Beihefte zur Zeitschrift für romanische Philologie 350.) Tübingen: Niemeyer. Lleó, Conxita, Ariadna Benet and Susana Cortés (2008): Contact-induced phonological changes in the Catalan spoken in Barcelona. In: Peter Siemund and Noemi Kintana (eds.), Language Contact and Contact Languages, 185–212. (Hamburg Studies on Multilingualism 7.) Amsterdam/Philadelphia: Benjamins. Lloret, Maria-Rosa (2002): Estructura sil·làbica [Syllable structure]. In: Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.), Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia. Morfologia [Grammar of Modern Catalan. Vol. 1: Introduction. Phonetics and Phonology. Morphology], 195–249. Barcelona: Empúries. Lloret, Maria-Rosa and Jesús Jiménez (2009): Phonological variation in voicing across word boundaries. Paper presented to Old World Conference in Phonology 6, University of Edinburgh. Available at http://www.uv.es/foncat/cat/OrdArr.wiki. Maddieson, Ian (2013): Syllable structure. In: Matthew S. Dryer and Martin Haspelmath (eds.), The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. Available at http://wals.info/chapter/12. Martí, Joaquim (2006): Processos fonètics catalans en el DECat de Joan Coromines. II. Consonantisme [Phonetic processes of Catalan in the DECat of Joan Coromines. II. Consonants]. Estudis de Llengua i Literatura Catalanes 53 (Homenatge a Joseph Gulsoy 1.): 147–185. Mascaró, Joan (2002): El sistèma vocàlic. Reducció vocàlica [The vowel system. Vowel reduction]. In: Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.), Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia.

 Central Catalan in the framework of the typology of syllable and words languages  

 389

Morfologia [Grammar of Modern Catalan. Vol. 1: Introduction. Phonetics and Phonology. Morphology], 89–123. Barcelona: Empúries. Mira, Maria Helena (2003): Fonologia [Phonology]. In: Maria Helena Mira Mateus, Ana Maria Brito, Inês Duarte and Isabel Hub Faria (eds.), Gramática da língua portuguesa [Grammar of Portuguese]. 6th ed., 987–992. Lisboa: Caminho. Moll, Francesc de B. (2006): Gramàtica històrica catalana. [Historical Grammar of Catalan] Valencia: Universitat de València. Montreuil, Jean-Pierre (1999): The Romansch syllable. In: Harry van der Hulst and Nancy A. Ritter (eds.), The Syllable. Views and Facts, 527–550. Berlin: Mouton de Gruyter. Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik – Beihefte 129.) Stuttgart: Steiner. Nuet, Josep, Francesc Bernat and Jaume Torres (1992): Mapes per a l’estudi de la llengua catalana [Maps for the Study of Catalan]. 2nd ed. Vic: EUMO. PALDC = Veny, Joan (2007–): Petit atles lingüístic del domini català [Small Linguistic Atlas of the Catalan Area]. 4 vols. Barcelona: Institut d’Estudis Catalans. Plaza, Carme (1995): Lleialtat lingüística, edat i nivell educatiu. La e posttònica a la Conca de Barcerà [Linguistic loyalty, age and educational attainment. Posttonic e in La Conca de Barcerà]. In: M. Teresa Turrell Julià (ed.), La sociolingüística de la variació [Variationist Sociolinguistics], 117–138. Barcelona: PPU. Pons, Clàudia (2009): Català universal, català particular. Català no marcat, català marcat [Universal Catalan, particular Catalan. Unmarked Catalan, marked Catalan. In: Johannes Kabatek and Claus D. Pusch (eds.), Variació, poliglòssia i estàndard. Processos de convergència i divergència lingüístiques en català, occità i basc [Variation, Polyglotism and Standard. Convergent and Divergent Linguistic Processes in Catalan, Occitan and Basque], 43–62. Aachen: Shaker. Pons, Clàudia (2011): It is all downhill from here: The role of syllable contact in Romance languages. Probus 23/1: 105–173. Pradilla, Miquel-Àngel (2002): Ensordiment, espirantització i fenòmens que afecten les sibilants [Devoicing, spirantization and processes affecting the sibilant consonants]. In: Joan Solà, Maria-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (eds.), Gramàtica del català contemporani. Vol. 1: Introducció. Fonètica i fonologia. Morfologia [Grammar of Modern Catalan. Vol. 1: Introduction. Phonetics and Phonology. Morphology], 287–318. Barcelona: Empúries. Prieto, Pilar (2004): Fonètica i fonologia. Els sons del català [Phonetics and Phonology. The Sounds of Catalan]. Barcelona: UOC. Rasico, Philip D. (1982): Estudis sobre la fonologia del català preliterari [Studies on the Phonology of Preliterary Catalan]. Barcelona: Abadia de Montserrat & Curial. Recasens, Daniel (1996): Fonètica descriptiva del català. Assaig de caracterització de la pronúncia del vocalisme i consonantisme del català al segle XX [Descriptive Phonetics of Catalan. An Attempt at a Description of the Vocalism and Consonantism of Catalan of the 20th Century]. 2nd ed. Barcelona: Institut d’Estudis Catalans. Sanchis, Manuel (1956): Factores históricos de los dialectos catalanes. In: Consejo Superior de Investigaciones Científicas (ed.), Estudios dedicados a Menéndez Pidal, vol. 6, 151–186. Madrid: Patronato Marcelino Menéndez y Pelayo.

390 

 Javier Caro Reina

Schmid, Stephan (1999): Fonetica e fonologia dell’italiano. Torino: Paravia. Siegel, Jeff (2008): The Emergence of Pidgin and Creole Languages. Oxford: Oxford University Press. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Traill, Anthony (1985): Phonetic and Phonological Studies of !XÓÕ Bushman. (Quellen zur Khoisan-Forschung 1.) Hamburg: Buske. Trubetzkoy, Nikolaj S. (1968): Introduction to the Principles of Phonological Descriptions. The Hague: Nijhoff. Veny, Joan (1986): Introducció a la dialectologia catalana [Introduction to Catalan Dialectology]. 2nd ed. Barcelona: Enciclopèdia Catalana. Veny, Joan (1991): Katalanisch: Areallinguistik. In: Günter Holtus, Michael Metzeltin and Christian Schmitt (eds.), Lexikon der Romanistischen Linguistik, Vol. 5,2: Französisch, Okzitanisch, Katalanisch, 243–261. Tübingen: Niemeyer. Veny, Joan (2001a): Diatopia i llengua estàndard [Diatopic variation and standard language]. In: Joan Veny (ed.), Llengua històrica i llengua estàndard [Historical Language and Standard Language], 119–171. València: Universitat de València. Veny, Joan (2001b): Les varietats dialectals i els estudis dialectològics [Dialect varieties and dialectological studies]. In: Joan Veny (ed.), Llengua històrica i llengua estàndard [Historical Language and Standard Language], 181–210. València: Universitat de València. Veny, Joan (2002): Els parlars catalans. Síntesi de dialectologia [The Catalan Dialects. Synthesis of Dialectology]. 13th ed. Mallorca: Moll. Veny, Joan (2013): Sobre l’epítesi dental de -r > -rt en català: Del cor al *cort [On the dental epenthesis of -r > -rt in Catalan: From cor to *cort]. Estudis romànics 35: 391–399. Viaplana, Joaquim (1984): En relació a la distinció oriental-occidental. Notes crítiques sobre el mapa dialectal del català [On the eastern-western distiction. Critical notes on the dialect map of Catalan]. Estudis de Llengua i Literatura Catalanes 10 (Miscel·lània Antoni M. Badia Margarit 2.): 27–36.

Uli Reich (Free University of Berlin) and Martin Rohrmeier (University of Cambridge)

Batidas latinas: On rhythm and meter in Spanish and Portuguese and other forms of music* Abstract: Our contribution investigates the formal principles which organize rhythm and meter in music and language. We claim that some aspects of the rhythmical differences between languages are comparable to rhythmical differences between different interpretations of the same piece of music, although these are generally more routinized in languages and varieties than in music. Building on evidence from Spanish and Portuguese, we find that main differences occur with regard to where full sonority is performed on the metrical grid. The same formal decision is observed in the difference between two versions of a piece performed by two jazz pianists. This leads us to a proposal which tries to align Metrical Phonology with metrics in music by foregrounding the temporal nature of moraic structure in the construction of linguistic feet.

1 Introduction: Language is music distorted by semantics Rhythm is the organization of human behavior in time and meter is the abstract representation of beats and their prominence structure. Both notions refer to formal principles in music and language. Intuition and common scientific practice frequently assign music and language to different domains of our cognitive architecture, though there are as many commonalities as differences. Both language and music organize sound events to communicate restrictions for the possibilities of linear ordering of primitives. Both language and music are special, if not unique, in comparison to communicative systems of other animals, and both language and music employ duration, grouping and prominence to establish rhythm. This suggests that meter and rhythm in music and language are not produced by particular, hermetically closed systems at the cognitive level, but

* We thank Jonah Katz (Massachusetts Institute of Technology) for his valuable comments and suggestions.

392 

 Uli Reich and Martin Rohrmeier

can be understood as classes of utterances which combine some common generating principles in different ways. The goal of this paper is to explore an aspect of this potential overlap on theoretical grounds by comparing formal approaches to rhythm and meter in language and music and trying to use concepts from music theory and cognition to account for certain linguistic facts, such as prosodic differences between Romance languages. We can sum up the theoretical claim we want to develop in our contribution in the following three points. (i) The prosody of linguistic utterances serves at least two different goals: the construction and discourse-related packaging of content and the optimization of euphonic quality. (ii) If we fail to take into account either of them, we miss important aspects of linguistic structure building. (iii) The interaction of the formal rules which serve these two goals explains the different prosodic shapes of styles and norms in languages. While the prosody of content building and content management has been studied extensively in linguistics, the optimization of the euphonic quality of linguistic signs has remained at the dark side of the moon in this discipline. However, it is the central object of interest in music theory, since the optimization of euphonic quality without the need for the construction of content in the form of truth-conditional propositions is what defines music – and sets it apart from language.1 We feel that important clarifications of otherwise opaque linguistic facts can be achieved if we admit that music theory may help linguistics to better understand some features of supra-segmental phonology. Our contribution sketches an explanation for those prosodic differences between Romance languages which cannot be attributed to different configurations of the grammatical functions of prosody. We will ground this explanation on music-theoretical grounds. To our understanding, the features discussed in recent approaches to the rhythm-based typology of Romance and other languages can be related to two aspects. First, differences in the grammaticalized usage of prosodic features, such as timing, accentuation, intonation, sonority and phrasing, serve in different configurations for the expression of lexical distinction, morphological rules, information structure and possibly other domains of grammar. Second, rhythmic differences also show up between languages and dialects which share the type and degree of grammatical functions expressed by prosody. This is true for the

1 In some of the music cognition literature, certain kinds of musical meanings are subsumed under the heading of semantics. In order to avoid terminological confusion, we use “meaning” as an umbrella term for all kinds of denotations and connotations and reserve the term “semantics” for propositions and the lexical primitives they are made of. See Jackendoff and Lerdahl (2006), among many others, for discussion.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 393

Ibero-Romance languages discussed in this article. In this case and only in this case, we argue that the same formal differences between metrical patterns and time-span reduction in musical performance may also be a fundamental principle behind different prosodic shapes of linguistic utterances in different traditions of languages. These differences cannot be excluded from the scope of linguistics proper since they play an important role in linguistic variation and change. This view is in line with work on the comparison between formal structures in music and in language which holds that many of the abstract formal relations in music and language are identical in nature; moreover, it is the construction of propositions – by merging lexical primitives into syntactic constituents and by combining them with semantic operators – that makes an utterance linguistic and not musical (Lehrdahl and Jackendoff 1983; Patel 2008; Katz and Pesetsky 2011; Rohrmeier 2011; Rebuschat et al. 2012). It is the need for semantics that sets language apart from music and, as we shall see, it is the grammaticalization of prominence and timing for structural functions serving semantics and pragmatics that makes language less rhythmic than music. In short, language is music distorted by semantics.

2 Facts of prosodic structure in Spanish and Portuguese In early work on linguistic rhythm (Pike 1945; Abercrombie 1967), Romance languages were treated as belonging to the “syllable-timed” type of rhythm, in opposition to the “accent-timed” rhythm, which would characterize, for example, Germanic languages. Many studies later underpinned what intuition had already suggested: This is clearly not the case (see Dufter and Reich 2003 among many others). The striking prosodic differences suggest that Romance languages are distributed over at least three different types. In this article, we focus on Spanish and Portuguese and try to sketch an explanation for the prosodic differences we find in and between these languages.2

2 The names “Spanish” and “Portuguese” stand for the linguistic competence and performance of hundreds of millions of speakers all around the world. Here they should be understood as slightly idealized languages which show sharp contrasts in their prosodic configuration and correspond more or less to the European variety of Portuguese spoken especially in the northeast of Portugal, Brazilian Portuguese as it is spoken in São Paulo and Urban Andean Spanish, as spoken in Bogotá or Lima. Dialectal and stylistic differences can approximate the prosodic configuration to the type given here as characteristic for another language (cf. for example Kabatek 1994 for salient differences between Castilian and Mexican Spanish). Style and social factors also play a major role in prosodic variation, a topic which urgently needs empirical investigation.

394 

 Uli Reich and Martin Rohrmeier

2.1 Vowel reduction Some Portuguese dialects reduce and elide vowels in unaccented syllables, leading sometimes to complex syllable structures, as consonants are couched in fewer syllables than they would in Spanish, a language which in most dialects shows a remarkably stable sonority of vowels in all positions. Urban Brazilian Portuguese shows less reduction than most dialects spoken in Portugal, especially in pretonic positions, but more than most dialects of Spanish: (1)

Three variants of Portuguese desistiu / Spanish desistió ‘(he) gave up’:3 a. European Portuguese: [ˌdəs:.is.ˈtju] ~ [dəsː.ˈtju] b. Brazilian Portuguese: [ˌdi.sis.ˈtʃju] c. Spanish: [ˌde.sis.ˈtjo]

These phonetic variants are related also to the phonetic reality of non-primary accents in these languages: Positions of secondary stress block processes of vowel reduction and elision. While they are overtly realized in Spanish utterances, secondary accents in Portuguese are evident as abstract positions of prominence, but their implementation in spoken utterances is rare.

2.2 Lexical and morphological functions In both Portuguese and Spanish, word accent is free to occur in any of the last three syllables, where it is lexically and morphologically distinctive: (2) Distinctive accents in Portuguese and Spanish: a. Portuguese: sábi-a ‘thinker-fem’ vs. sabiá ‘nightingale’ b. Spanish: cant-é ‘sing-1sg.perf.ind’ vs. cant-e ‘sing-1sg.pres.subj’

To sum up, we could say that in Spanish and Portuguese, primary accent is part of the phonological representation of the word and also a morphonological rule. Secondary accents are attributed by the binary alternation of strong and weak syllables, but they never override the position of the primary accent in both Ibero-Romance languages, where it is distinctive at the word level. These two lan-

Facts in other Romance languages and dialects cannot be considered here, but are on our agenda for future research. 3 The sound file of the European Portuguese example is accessible in two possible pronunciations as an online appendix to the IPA description: http://web.uvic.ca/ling/resources/ipa/handbook/.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 395

guages seem to have selected different prosodic domains as their main category for rules and processes. Portuguese optimizes the Phonological Word, while Spanish, due to its salient stability with respect to the sonority of vowels in all positions, optimizes syllables and feet, since the alternation of strong and weak syllables requires the existence of non-prominent syllables which vanish once their nucleus is elided. Thus, syllable optimization always enhances the salience of metrical feet. Urban vernaculars of Brazilian Portuguese seem to be situated “between” Spanish and European Portuguese since Brazilian Portuguese reduces less than the variety spoken in Portugal, but more than Spanish (see Abaurre and Galves 1998 and Frota and Vigário 2000, 2001 for details about the differences between Portuguese varieties).

2.3 Accents by lexical marks Spanish and Portuguese show prosodic configurations which we could call “frozen Latin”.4 This epithet is due to the general retention of the Latin stress system in those word forms for which it is impossible to derive the stressed syllable with a single set of algorithms. The most common patterns are penultimate stress if the last syllable is monomoraic (3) and ultimate stress if the last syllable has two moras due to consonants in coda position (4a, b), or nasalized (4c) or diphthongized (4 d, e) nuclei: (3) Portuguese: stress on the penultimate syllable ˈgato, ˈlivro, paˈlavra, boˈnito (4)

Portuguese: stress on the ultimate syllable a. naˈriz, feˈliz, paˈís b. aˈmor, tuˈtor, meˈnor c. irˈmã, aˈtum, armaˈzem d. aˈnel, toˈnel, paˈpel e. heˈrói, chaˈpéu, deˈgrau

These basic patterns are easily derived by a moraic trochee with the end rule set to right, cf. (5):

4 For accounts of the Latin stress system, see Mester (1994) and Roca (1999).

396 

 Uli Reich and Martin Rohrmeier

(5) ( x ) ( x ) (x ) σ µ µ µ µ ga to Paenultima (CV.CV.) na riz Ultima (CV.CVC.)

One core problem is that there are common cases of different stresses for phonological words which show exactly the same syllabic and moraic configuration: (6)

a. ˈlápis, ˈpíres, ˈvírus b. ˈflúor, caˈráter, ˈdólar, aˈçúcar c. ˈímã, ˈórfã, ˈhómem, d. ˈtúnel, ˈfácil, ˈágil, ˈútil e. ˈjóquei, ˈvôlei, ˈmédio

Additionally, many words have gone through less prosodic restructuring than the words in (3) and (4) and still show the Latin antepenultimate stress: (7) a. ˈcórrego, aˈbóbora, ˈfósforo, ˈárvore b. ˈdúvida, ˈmáquina, ˈmédico

Brazilian Portuguese also shows a remarkable preference for stress on ultimate monomoraic syllables, not only in borrowed words (8a), but also in words with Latin etymology (8b) and especially in the case of colloquial nicknames (8c): (8) a. caˈfé, uruˈbú, jacaˈré b. aˈvô, aˈvó c. Caˈfú, Peˈlé, Kaˈká

As mentioned above, stress is also distinctive in the lexicon and in the morphology, both in Spanish and in Portuguese (see examples in 2). Verbal morphology in particular shows considerable deviations from the moraic trochee, since bimoraic final syllables do not attract stress as predicted by the algorithm in (5). The location of main stress is completely unpredictable in these cases, exemplified here with some forms of the verb falar ‘to speak’: (9)

a. ˈfalas b. ˈfalam c. faˈlámos d. faˈlassemos e. faˈlaramos f. falaˈriamos g. falaˈremos



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 397

Primary stress is attributed to stems (9a, b), thematic vowels (9c, d, e) and suffixes (9f, g). We cannot see any uniform way to derive these patterns by metrical algorithms. Clearly, Portuguese and Spanish have phonologies in which stress is assigned in different subsystems of the language: lexical, morphological, pragmatic and metrical rules interact to construct the prosodic form of linguistic utterances.

2.4 Metrical alignment Foot construction plays a substantial role in the variation and change of the segmental form of phonological words. Consider these data from vernacular Portuguese: (10) Moraic alignment of stress in vernacular Portuguese traditional vernacular a. abóbora [aˈbɔbora] [aˈbɔbra] accent: × × moras: µ µ µ µ µ µ µ syllables: σ σ σ σ σ σ σ b. homem accent: moras: syllables:

[ˈɔme͂j͂] × µ µ µ σ σ

[ˈɔmi] × µ µ σ σ

c.

cantaram accent: moras: syllables:

[ka͂ˈtara͂w͂] × µ µ µ µ µ σ σ σ

[ka͂ˈtaru] × µ µ µ µ σ σ σ

d.

grandão [gra͂nˈda͂w͂] [gra͂nˈda͂w͂] accent: × × moras: μ μ μ μ μ μ μ μ syllables: σ σ σ σ

Words are restructured to fit into the metrical pattern derived by the moraic trochee at the final periphery. Syncopation (10a), denasalization (10b) and monophthongization (10c) are the processes at work to adjust words which in their traditional, normative forms appear as unpredictable by the metrical algorithm. It is interesting to note that all processes apply to material outside of the syllable which carries primary stress. Word stress is maintained on the same syllable in all examples. The fact that these processes are metrically driven and not

398 

 Uli Reich and Martin Rohrmeier

instances of more general processes is evident in light of the comparison with examples like (10d), in which the last syllable has a nasalized diphthong like in (10c), but it is stressed just as the moraic foot predicts and, therefore, the process does not apply.

2.5 Importance of meter in perception There is additional empirical evidence that supports the importance of linguistic meter for perception. Frota and Vigário (2000) report a perception experiment which sheds light on the role of meter in the processing of speech. They recorded four Portuguese sentences read by Brazilian and Portuguese speakers with varying nuclear accents and asked Brazilian and Portuguese informants to count the stresses they heard, in both their native and foreign varieties. The result is stunning: Brazilians not only count twice as many accents in their native variety as the Portuguese count in the European variety, a result to be expected given the facts outlined above, but Brazilians also count twice as many stresses in the sentences read by Portuguese as the Portuguese informants, and the Portuguese informants count only half of the accents which the Brazilians count in their native variety. Normalized metrical patterns, then, persist independently from the reality of the speech signal. This important insight leads to the assumption that meter is above all an abstract pattern which organizes the perception of events and is related to the music-theoretical concept of meter.

3 Rhythm and meter in music There is a core distinction between musical rhythm and meter, the two parameters that govern the temporal dimension of music. Rhythm, in short, describes durational patterns of notes, percussive and other event onsets (in terms of interonset intervals, IOIs), and accents.5 Meter in turn encompasses the perception and anticipation of a sequence of categorically isochronic beats that involve relative stress patterns. The perception of metrical structure contributes to the “feel”

5 The durations, however, refer to the intervals between note onsets (interonset intervals, IOI) and not their absolute durations: notes may be played staccato or legato, while the implied rhythmical structure is identical. In other words, when attending to rhythm we direct our musical perception to be beginnings of note events, and take these as form-constituting features, and not their ends.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 399

or “groove” within which a rhythm appears; and the same rhythmical pattern may sound different in the context of different meters. Generally, rhythm and meter are cross-cultural musical universals (Stevens 2004; Stevens and Byron 2009) even though they exhibit considerable variety (see also Stobart and Cross 2000) up to complex polyrhythms and polymeters. Though suggested by (Western) musical notation, musical meter is not merely a feature of the musical signal. London stresses: meter is not just a part of the ‘representation of reality’, a means of temporally indexing musical events. It is thus a cognitive structure. Rather, meter is one of the ways in which our senses are guided in order to form representations of musical reality. Meter provides a way of capturing the changing aspects of our musical environment as patterns of temporal invariance. (London 2004: 5)

Through the cognitive process of entrainment (Clayton, Sager, and Will 2005), metrical structure enables listeners and performers to synchronize their attention towards perception of or movement with the music. Meter also constitutes a cognitive ability that governs the temporal synchronization of various domains other than music or language (e.g. co-ordinated motor action and synchronized personal interaction, see Clayton, Sager, and Will 2005). Metrical structure assigns patterns of stress and shapes the way a rhythmical pattern is heard; this has the effect that the same rhythm may be heard differently within different metrical contexts (see below). The exploration of the cognitive foundations behind meter helps us to understand the nature of metrical stress patterns. Dynamical system models characterize the process of metrical entrainment through the interaction of coordinated and reinforcing oscillations which result in regular peaks of expectancy (Large and Jones 1999; Large 2000; Large and Palmer 2002). These become stronger once the rhythmical musical surface establishes a clear sense of beat. Large and Palmer (2002) further show that the different levels of metrical stress can be modeled through phase-locked oscillators at simple multiples of a period. The strong attentional peaks that are generated by dynamic oscillating systems thus form the basis of metrical beats, accents as well as phenomena such as loud rests – i.e. a moment in the music may be heard with stress (or accent) although there is no physical note sounding. Thus the establishment of metrical stress happens in the listener, rather than the music, and metrical stress patterns do not merely assign stress to musical notes. This corresponds with the findings for Portuguese in the aforementioned linguistic experiment led by Frota and Vigário (2000): In the same target sentence, Brazilians perceive twice as many accents as Portuguese do, even when the sentence is pronounced by Portuguese speakers and the relevant positions of prominence are not realized in the speech signal.

400 

 Uli Reich and Martin Rohrmeier

This underpins the idea that meter is a perceptual structure rather than a feature of linguistic utterances. Similar to linguistics, early accounts of musical rhythm have likened its stress patterns to poetic scansion and applied poetic feet such as iamb, trochee and dactyl to musical rhythm (Koch 1787; Cooper and Meyer 1960; Houle 1987). Such a scansion-based assignment of stress has been extended hierarchically in a way that higher-order poetic feet projected their patterns onto entire grouped poetic feet underneath (Cooper and Meyer 1960). This analysis resulted in a formalism in which a whole group of bars or notes could be seen as the weak or strong part of an iamb, for instance. Such formalizations have, however, largely disappeared today, giving way to a recursive hierarchical grid representation that was modeled in analogy to linguistic approaches.

3.1 The grid and metrical well-formedness Since its formalization as part of the Generative Theory of Tonal Music (GTTM, Lerdahl and Jackendoff 1983), one common way to represent musical meter analytically is with a grid structure (12) that is designed in analogy to phonological theories of meter (Liberman and Prince 1977; Halle and Vergnaud 1987; Hayes 1995; van der Hulst 1999). The grid involves one continuous fundamental beat (or tactus) level that is assigned to nominally isochronic time points (and not necessarily musical events). On higher (or lower) levels, the grid displays metrical structure at regular intervals of the underlying structure and represents the strength of metrical stress. One notable difference to phonological meter, however, is that the spacing of stress patterns consequently entails binary and ternary spacings as well as their flexible combination at the same level of representation. The relevance and existence of ternary meters has never been doubted theoretically (unlike historical doubts about the “impossibility” of irregular meters, like 5/4), which may be due to the frequent use of triple meter in Western music. From a cognitive perspective, there seems to be hardly any reason to prefer binary groupings to others. The structure of possible metrical grids has been defined recursively by the GTTM on the basis of metrical well-formedness rules. (11) Metrical well-formedness rules (MWFR; cited from Lerdahl and Jackendoff 1983: 347) MWFR 1: Every attack point must be associated with a beat at the smallest metrical level present at that point in the piece. MWFR 2: Every beat at a given level must also be a beat at all smaller levels present at that point in the piece.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 401

MWFR 3: At each metrical level, strong beats are spaced either two or three beats apart. MWFR 4: The tactus and immediately larger metrical levels must consist of beats equally spaced throughout the piece. At subtactus metrical levels, weak beats must be equally spaced between the surrounding strong beats.

MWFR 1 assures that every musical event is represented in the grid, but allows that there may be beats that are not associated with musical events. MWFR 2 assures that there are no higher-level metrical beats (e.g. on half-note level) that are not also beats on a lower metrical level (e.g. quarter note level). MWFR 3 defines metrical stress and its binary or ternary spacing, so that strong beats cannot be adjacent or too far apart. MWFR 4 enforces the regularity of the grid pattern. While the first two rules are general, the last two are culturally specific to Western tonal music. Based on the cognitive foundations of meter and its limits and on the understanding of meter as cyclical peaks of attention in continuous time, London (2004) develops a novel cyclical and time-continuous representation of meter that incorporates these features and easily extends to non-Western meters. His well-formedness rules incorporate minimal beat durations (100ms), maximal metrical cycle span (5s) as well as a more flexible way of formalizing higher-level metrical accent which generalizes MWFR 3 and 4. London’s formalism or a loosened version of the formalism above may account for complex meters such as 7/8 involving non-isochronous subdivision (2/8+2/8+3/8), or irregular, non-isochronous subdivisions of simple meters, such as 4/4 divided into a 3-3-2 pattern (e.g. found in Khatchaturian’s Sonatina for Piano, 1959), or Brubeck’s Blue Rondo à la Turk, which divides 9/8 into a 2-2-2-3 pattern. In general, musical meter is not restricted to double, triple or quadruple meters. Eastern European, Middle Eastern or Indian musical traditions involve much more complex metrical structures (e.g. Clayton 2000). Meters such as 5/4 or 7/8 are common in Bulgarian music, for instance. However, such complex meters are frequently constructed from the addition of duple and triple parts such as 7/8 realized as 2+2+3 or 5/4 realized as 3+2 or 3/8 + 3/8 + 2/8 + 2/8 (as in Brubeck’s Take Five). The examples in (12) illustrate the difference of binary and ternary meters on the different metrical levels. While in example (12a) the lowest metrical level is ternary, all superordinate levels have binary structure. In particular, the bar level has a quadruple structure, similar to a 4/4 meter with a ternary subdivision. In example (12b) the lowest metrical base level is binary, while the next higher level (bar level) is ternary. The higher metrical levels are – as in virtually all musical cases – organized in a binary way.

402 

 Uli Reich and Martin Rohrmeier

(12) Metrical structure in music a. Händel, Suite in G minor, HWV 439, Gigue, mm. 1–2

b. from Händel, Suite in G minor, HWV 439, Sarabande, mm. 1–4

Since meter is not assigned to musical fragments (in the way that Metrical Phonology derives word stress), but to entire phrases and pieces and repeated cyclically, the difference between right- and left-peripheral accents vanishes. While from the cyclical understanding of meter both structures are identical but shifted in phase, the fact that pieces of music frequently begin on strong beats and anacruses can be construed as optional (in the sense of an omission test) and that strong final events in music are not short but fill entire bars (or more) implies a left-peripheral perspective on musical meter. Compare these characteristics to linguistic phonology: The construction of feet is cyclical, but in most types of linguistic utterances, it must respect distinctive stresses which correspond to lexical or morphological phonology, which are not cyclical, since they apply only once to their domain. Thus, Metrical Phonology and Lexical Phonology are very different in nature, the former being comparable to aspects of musical structure, the latter being completely absent in music. Similarly, a cyclical repetition of the same word or groups of words in language shows that the distinction between left- and right-peripheral metrical accents can only be maintained when the metrical stress is not repeated cyclically. For instance,



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 403

Deutsch’s famous “it sometimes behaves so strangely” example (e.g. Deutsch, Henthorn, and Lapidis 2011), which consists of a recording of this phrase repeated over and over, gradually “morphs” into music and exhibits metrical, rhythmical and melodic qualities. In this case, the constant cyclical repetition may be the reason why the example exhibits perceived metrical and rhythmical regularities.

3.2 The establishment of metrical structure While metrical well-formedness rules merely algorithmically define the logical shape of the grid, they still do not specify how the metrical grid is aligned to a piece or a segment of music (or rather, how the listener infers meter). The establishment of metrical structure is based on various rhythmical, accent, phrase, tonal or other musical features and not every musical surface affords an unambiguous or even any inference of meter. For instance, the fact that rhythmical IOI durations exhibit mostly simple integer-based duration ratios, such as 2:1, 3:1, 4:1, 6:1, etc., makes the inference of a regular beat and metrical structure possible. However, highly irregular IOI patterns, such as at the opening of Babbitt’s Composition for Twelve Instruments (London 2004: 24), conceal their metrical structure. In the GTTM, metrical analysis (assumed as the result of the cognitive process, not necessarily a representation of the cognitive process) is characterized by a constraint-based approach rather than an explicit algorithmic/procedural approach. A set of 10 preference rules define constraints that guide the preference of some solutions of metrical alignment over others. However, the GTTM is imprecise by omitting the specification of an exact weighting or decision process to resolve conflicts in the assignment of metrical structure. The metrical preference rules employ the information of features such as grouping structure, parallelism, stress, duration, as well as specific musical and Western tonal features such as cadence, suspension and bass. Metrical stress may be inferred from/expressed through melodic, rhythmic accents, duration, expressive timing, scale degree, etc. (13) Metrical Preference Rules in the GTTM (cited from Lerdahl and Jackendoff 1983: 347–348) MPR 1 (Parallelism): Where two or more groups or parts of groups can be construed as parallel, they preferably receive parallel metrical structure. MPR 2 (Strong Beat Early): Weakly prefer a metrical structure in which the strongest beat in a group appears relatively early in the group. MPR 3 (Event): Prefer a metrical structure in which beats of level L that coincide with the inception of pitch-events are strong beats of L.

404 

 Uli Reich and Martin Rohrmeier

MPR 4 (Stress): Prefer a metrical structure in which beats of level L that are stressed are strong beats of L. MPR 5 (Length): Prefer a metrical structure in which a relatively strong beat occurs at the inception of either (i) a relatively long pitch-event, (ii) a relatively long duration of a dynamic, (iii) a relatively long slur, (iv) a relatively long pattern of articulation, (v) a relatively long duration of a pitch in the relevant levels of the time-span reduction, (vi) a relatively long duration of a harmony in the relevant levels of the time-span reduction (harmonic rhythm). MPR 10 (Binary regularity): Prefer metrical structures in which at each level every other beat is strong.

While a detailed discussion of the metrical preference rules would exceed the limits of this article, some important features shall be noted. The metrical structure is assigned according to the core salient features of the music: onset (attack), duration, bass and stress. MPR 2 creates the left-peripheral nature of musical meter. While there can be rests at metrically strong positions, in general metrical positions become strong through the events at their points of time (and their musical importance, such as harmonic stability). It further follows that the simple proportions of rhythmical durations are required to establish a sense of meter. It is important to note that while binary and ternary meters are allowed, MPR10 expresses a preference for binary subdivision, which at higher metrical levels is more frequent even when the base meter is triple. While these rules are mainly based on and specific to Western music, Stobart and Cross (2000) give a musical example from the Andes in which the cultural metrical perception of strong– weak stress patterns seems to be reversed. London (2004) argues, however, that due to the Western specificity of the MPRs and their underspecification, a large amount of implicitly acquired knowledge (Rohrmeier and Rebuschat 2012) of statistical, schematic and prototypical patterns must be assumed for the inference of metrical structure. Such a statistical/implicit learning account would further be much more easily adaptable to explaining the cognition of non-Western and other complex meters rather than different sets of MPRs. When we experience rhythm shaped by meter, we do not experience a pattern of onsets/IOIs only, we hear it structured by the metrical cycle as well as its accents, which affords metrical “feel”. A waltz or a 4/4 marching song sound different even though the series of pitches and durations may be identical. Conversely, the same rhythm (in the sense of a mere sequence of durations) sounds differently once different metrical structure is laid underneath it. (14) illustrates an example in which an identical melody sounds different in the context of a 4/4 (14a) or 3/4 (14b) meter. In the context of a performance, the musical performer strengthens the intended metrical structure with expressive means, such as timing, accent, small delays or anticipations – which may serve as intentional



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 405

stress of an event –, and the small shortening or lengthening of beats. Expressive performance makes it possible to convey metrical structure and, accordingly, communicate metrically ambiguous melodies (such as example 14) unambiguously, i.e. in a way that the listener will hear an unambiguous metrical structure. Musical markers for metrical structure may conflict and intentionally or unintentionally cause metrical flux or instability (compare the analysis of Beethoven’s Fifth in London 2004: 89–99). For specific cases in which structural beginnings and endings create metrical overlap, the GTTM also discusses rules for metrical deletion. (14) Example of a melodic phrase that can be heard in 4/4 or 3/4

3.3 Attentional focus within metrical structure Different metrical levels may receive attentional focus during listening, performance, improvisation or composition. During a jazz performance, for instance, the point of attention and the level of abstraction of the performer may be at different metrical timescales. Such different levels of metrical attention are underpinned in reverse by the finding that musicians and non-musicians tap to music on different metrical levels (Drake, Penel, and Bigand 2000). While a performer may choose to elaborate events at all four beats in a 4/4 meter, they may equally have their improvisational focus at other (slower) levels and elaborate only one or two events per bar, based on a time-span reduction of the musical structure, or change this level of abstraction (i.e. the metrical hierarchy). A comparison between Keith Jarrett’s and Bill Evans’ playing of Cole Porter’s What is this thing called love? illustrates this difference (cf. example 15, see also Evans’ perfor-

406 

 Uli Reich and Martin Rohrmeier

mance of Blue in Green on the same album). While Jarrett’s performance is based on a quarter note level (and closely matches the original theme), Evans’ interpretation operates on a time-span reduction at a much higher metrical level of entire bars (or four quarter notes), in which the shorter notes merely have ornamental or appoggiatura character with respect to the stressed long core notes of the reduced theme at the high tempo. Such a difference may constitute an analogue of the linguistic example in (1). Such different realizations of metrical structure and metrically-guided attention/reduction do not constitute structures or cognitive processes that are unique to music. This seems to be the case also if we compare European Portuguese to Brazilian Portuguese and Spanish: Segmentally identical phonological words with the same main stress (cf. example 1) differ with respect to the sonority of syllabic peaks, reducing or eliding vowels from positions that do not carry word accents. If we compare realizations of sound chains which correspond to phonological representations of identical or very similar words in Spanish and of Brazilian and European Portuguese, we find a difference which looks like that of (15). (15) Keith Jarrett’s (from Whisper not, 1999, top) vs. Bill Evans’ (from Portrait in Jazz, 1959, bottom) interpretation of the beginning of the theme of What is this thing called love? The arrows indicate the heard/performed metrical level.6



In music, differences in the interpretation of the metrical grid like those illustrated in this example reflect some differences between individual performances, whereas in language, as we shall see in 4.3, the same type of differences may define the norms of dialects and languages: The same abstract representation of sound strings is performed with different choices of the metrical level realized

6 Note that the Jarrett transcription has one additional (first) bar displaying the anacrusis of the phrase. The anacruses starting the subsequent phrase at the final bar of both transcriptions were omitted.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 407

in the corresponding utterances. In short, we can pronounce a word in Spanish, Portuguese or Brazilian rhythm just as we can play What is this thing called love? at two different metrical levels of reduction. There is absolutely no difference in the content-related phonological subsystems, but at the level of rhythmic performance the differences are salient. This is one core feature that linguistic rhythm shares with musical meter and time-span reduction.

4 Beats, prominence and cognitive networks Based on the discussion of musical and linguistic rhythm, it seems possible to define metrical algorithms in a uniform way as the grouping of beats in time and the construction of prominence relations which delimit and differentiate these groups. In types of utterances we normally associate with language, these prominence relations interact in language-particular (or typological) ways with accentual patterns and timing rules which are generated by form-function pairs which serve the semantic construction and pragmatic management of propositional content. In types of utterances we normally associate with music, these related functional domains are not the intended communicative goals: Rather, it is the aesthetic balance of formal harmony which we experience as beauty, the expression of emotions and social indexicality that is at stake (cf. Scruton 1999; Cross 2005; Miell, MacDonald, and Hargreaves 2005; Cross and Woodruff 2009). The syntactic relations in music do not construct phrases and sentences which carry semantics, but define the well-formedness of harmonic progressions and structural (intra-musical) meanings such as opening, departure, delay, tension or closure (Lerdahl 2001; Rohrmeier 2011). Which of these functional domains is foregrounded at the cost of the others depends on what type of music is being played. In a cello suite by Bach or a piece by Thelonious Monk, e.g. the expressive play with syntactic relations is optimized, while in a song by the Sex Pistols, emotion and social indexicality are pivotal, just as in techno and salsa the rhythmic-metrical parameters are foregrounded to facilitate entrainment, synchronization and dance. The hierarchical ordering of these different functions defines genres and styles and is presumably also responsible for different forms of music in other cultures than ours.7 In our view, it is a pitfall projected by language that leads us to assume that language and music are compact, separate and closed systems at the cognitive

7 For descriptions of communicative functions and their relation to musical forms, see Cross and Woodruff (2009).

408 

 Uli Reich and Martin Rohrmeier

level: Having different names for both categories of communicative utterances leads to the assumption that the formal principles which govern their shapes correspond to different cognitive systems. We prefer to view linguistic and musical utterances as compositions made up of many different formal devices and collections of primitives, some of which are shared and some of which are particular to one of these categories of utterances.8 From a linguistic perspective, this leads to a view of phonology as being made up of different subsystems whose outputs must be mediated by hierarchical ordering in order to pronounce a successful communicative act.

4.1 Differential Phonology Differential Phonology (Reich 2007) is a way to disentangle standard theories of suprasegmental phonology. It owes much to Autosegmental Phonology as proposed by Goldsmith (1990) and Metrical Phonology as advanced by Hayes (1995), but tries to relate systematically formal phonological domains to the functions they convey. In this view, phonology consists of at least six subsystems which are basically autonomous, but which must be related in a way to be specified by language and discourse type. Here is a preliminary list: (16) Six phonological subsystems (i) Lexical Phonology: form–meaning pairs which can also specify the primary accent of a word. (ii) Morphological Phonology: morphological rules can be coded by ablaut, stress, etc. (iii) Pragmatic Phonology: the projection of pragmatic functions (illocutions, information structure, modality). (iv) Syntactic Phonology: syntactic phrases are projected to phonological phrases. (v) Metrical Phonology: the grouping of sound strings in time and the attribution of recurring prominence patterns. (vi) Indexical Phonology: the expression of social relations by phonological clichés.

Aspects of the formal principles governing these subsystems form the body of the literature on phonology, but to our knowledge their interaction in different types of utterances has been somewhat neglected. In our view, it is the ranking of the different formal restrictions projected by these principles which determines the phonetic realities of utterances in different types of communication. In most

8 See Patel (2008) for surveys of work exploring shared and particular formal principles. See also Katz and Pesetsky (2011).



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 409

types of communication, we should get rankings which prefer the first four subsystems to Metrical Phonology, since the main goal of communication is to make utterances understandable. There are a few types of communication, though, where this is not the case. In routinized prayers, for example, the content is already known and repeated many times, and is less important than the metrical alignment of its rhythm, which has a strong effect of social entrainment. Another example is counting-out rhymes, which serve their communicative intention by keeping a very strict sequence of beats which are normally accompanied by synchronized pointing to the participants. In these types of communication, the formal principles of the subsystems (i–iv), which construct content, are ranked below the metrical restrictions which must be met in order to accomplish their communicative intention. In metricized poetry, all restrictions must be met: To be well formed, the metrical pattern of the poem must be in harmony with the content-related accents and time structures.9 In this view, music and language share many formal principles and it is the construction of truth-conditional propositions and its management that sets them apart. Metrical Phonology, then, should work with notions developed in music theory and describe their relation to the structures of prominence and timing (and potentially time-span reduction) at other levels of linguistic expression using the systems (i)–(iv) in (16).

4.2 Moras as points in time We would like to present an account of linguistic meter that treats the projection of prominence on beats in time in a similar way to music theory, since we assume that the metrical contribution to linguistic rhythm is generated by the same structure-building principles and that it is the interaction with semantic and pragmatic prominence and timing that sets language apart from music. The correctness of this approach is corroborated by the fact that this intention of formal correspondence is achieved rather easily by a few reinterpretations of already existing formal models: First, we must strengthen the insight that the primitives of Metrical Phonology are moras which should be understood as points in time, as characterized by Hayes: [...] weight can be thought of as a property of the time dimension: a syllable is heavy because it is long. This is the viewpoint of moraic theory: the moras form an abstract characterization

9 An even more telling case is “bad poetry”: We meet the metrical restrictions but violate lexical, syntactic and pragmatic restrictions.

410 

 Uli Reich and Martin Rohrmeier

of a syllable’s phonological duration. Under the Iambic/Trochaic Law, a canonical iambic foot is canonical because of its uneven durational form, that is, /μ µ ˈμ/; and a canonical moraic trochee is likewise canonical because of its even durations; /μ ˈµ/. (Hayes 1995: 271)

Mora-based syllables thus correspond directly to musical notation which denotes temporally specified tones in full, half, quarter notes and so on. Moraic feet can easily be rewritten as full and half notes, using the symbols ● and ○, in analogy to musical notation. The well-known foot inventory from Hayes (1995) is repeated here with suggestions for the equivalent values of musical notes: (17) Foot inventory and beats in time a. Syllabic trochee b. ( x . ) μ μ ● ● σ σ c. (Moraic) iamb (. x) μμ ● ● ○

Moraic trochee ( x . ) µ µ ● ● two quarter notes or two light syllables ○ one half note or one heavy syllable

two quarter notes or two light syllables one half note or one heavy syllable

An acoustic event which carries a linguistic accent is always a half note: Accented peaks of sonority cannot be reduced (cf. the moraic law in Vennemann 1988, which postulates that an accented syllable is preferred if it is bimoraic). In this view, syllables are not heavy or light, but long or short. A moraic trochee, then, is an algorithm which projects light syllables to one point in time and heavy syllables to two points in time. It differs from moraic iambs in the usual sense: The first or the second beat is prominent. A syllabic trochee is an algorithm which compresses syllables in time: They are projected to one beat independently of the material they insert in the syllable structure. The big difference to music is the relation of feet to the edges of words and phrases. Since there are no words in music, but phrases of related sound events without double articulation, the meaning of left/right periphery from Metrical Phonology is lost, since we cannot tell if an accent is realized at the beginning or end of a unit which is defined at a different level of structure. Unstressed events at the beginning of a musical piece appear as anacruses and we perceive accents as initial for the domain of their application. The interaction of the cyclically attributed algorithms of stress with the phonology of meaningful words is one of the interfaces of phonological subsystems which shape linguistic utterances. It is this interface which defines trochees and iambs (and possibly ternary feet) as



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 411

patterns for word stress, etc. Only the position of stressed events in relation to the (edges of) words indicates left-/right-headedness in Metrical Phonology.

4.3 The rhythmic difference between Spanish and Portuguese Let us take a closer look at the differences between Spanish and European Portuguese. These two Ibero-Romance languages serve as examples because of their very transparent rhythmic difference. As we have seen in sections 2 and 3.2, this difference is related to the sonority of vowels in unaccented positions: In Spanish styles, the sonority of vowels in all positions is retained, while in Portuguese styles they are reduced or elided. This can be captured by the very same difference as the one visible in the notation of (15): We pronounce at different metrical levels and the selection of that level can be part of the normal use of a language by a speech group. The following example illustrates our theory using a preliminary formal sketch of the interaction between the aforementioned phonological subsystems. We chose word forms which are very similar in Spanish and Portuguese and invented a simple sentence in which they could occur. It is important to note that we do not intend to put these different phonological structure-building principles in a sequence, as in the early generative phonologies of rule ordering. Rather, we understand them in a similar sense as the independent layers in Autosegmental Theory (Goldsmith 1990), as independent formal principles which must be associated when we pronounce a meaningful linguistic utterance in discourse: (18) a. Portuguese: b. Spanish:

Percebeste a desaceleração? ¿Percibiste la desaceleración? ‘Did you notice the deceleration?’

The verb and the noun have segmental representations of their stems in the lexicon, specifying also stress in these languages: (19) Segmental representation a. Portuguese: / perseb / b. Spanish: / perseb /

/ desaseleraˈsãw͂ / / desaseleraˈsjoŋ /

To fit into the sentence, these words must have the morphological form required by their syntactic position. This is done by morphonological rules corresponding to syntactic functions, attributing inflectional morphology with agreement features to the verb or merging the noun with a determiner to realize the categories of definiteness, number and gender:

412 

 Uli Reich and Martin Rohrmeier

(20) Morphonological rules (person/number agreement, tense, mood, aspect and definiteness) a. Portuguese: / persebeste / / a desaseleraˈsãw͂ / b. Spanish: / persebiste / / la desaseleraˈsjoŋ /

Morphonological rules need not be concatenative, as in the example discussed here, but may also project prominence, as in the case of the Spanish verb forms mentioned in (2). Syntactic phonology will enforce the alignment of phonological tones and possibly delimitation processes at the edges of syntactic phrases: (21) Alignment of phonological phrases with syntactic phrases T% a. Portuguese: [ / persebeste / / a desaseleraˈsãw͂ /] b. Spanish: [ / persebiste / / la desaseleraˈsjoŋ / ]

Pragmatic phonology projects the type of the border tone to indicate illocutions and assign the nuclear accent tone to the focussed constituent, among many other pragmatic contents which we cannot specify here:10 (22) Nuclear accents and illocutions a. Portuguese: [/ persebeste / b. Spanish: [/ persebiste /

LH* H% / a desaseleraˈsãw͂ /] / la desaseleraˈsjoŋ / ]

These are the basic subsystems of phonology which construct meaning in discourse. As outlined above, eurhythmic principles do not contribute to content building or content management, but organizes sound chains in time and attributes non-distinctive prominence, normally referred to as secondary accents. The algorithmic attribution of strength to beats in time (moras) constructs metrical patterns which interact with the content-building phonologies according to language and discourse type. Another important metrical aspect of phonology, of course, is syllabification: Segmental chains are organized in the cyclical crescendo and decrescendo of sonority which reaches from onsets to nuclei and from there to codas and onsets of following syllables. This has been called the sonority principle and forms the main motivation of preference laws for syllable

10 Note also that our rough preliminary sketch does not consider broader crosslinguistic variation and, at best, only captures some common insights on differences between SAE-languages. We are aware that the picture might change rather sharply in other types of languages. Also, the specification LH* depends on the dialect in question. See Sosa (1999), Moraes (1998), and Frota (2000) for some basic insights into the intonational grammars of both languages.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 413

structure, together with gestural restrictions. The particular restrictions on the organization of segmental material in syllables are surely among the most salient phonological features of a language and also form one of the core parameters for rhythm typology as discussed for example in Auer (1993). Summing up, the big differences of our model compared to standard theories of Metrical Phonology are the following rearrangements: 1. In Spanish and Portuguese, word stress is part of the lexical subsystem of phonology. This is motivated by the observable impossibility of deriving all configurations of stress in these languages algorithmically and the correlated distinctive function of stress. 2. Syllables are not part of Lexical Phonology, but arise out of the interaction of the organization of segments with metrical and gestural restrictions. 3. Metrical algorithms attribute alternating strength to groups of sounds at different levels of hierarchical organization. They are cyclical in nature and apply at phrasal levels, but must respect lexical and morphological stress in normal speech styles. This is a major difference to music: Since there is neither lexical nor morphological phonology in music, metrical strength is fully cyclical and unbounded. 4. Moraic structure is not an inherent property of syllables or metrical systems, but is defined by the relation between syllables and points in time, just as in music. A bimoraic syllable is a syllable which is projected to two moras, understood as points in (perceived) time; a monomoraic syllable, then, is a syllable which is projected to one mora. 5. The model is not conceived as procedural. The phonological subsystems are independent and interact in hierarchies of constraints which are determined by particular languages and communicative functions, just as discussed above.11 The prosodic difference between Spanish and Portuguese, or rather between speech styles which are normalized differently in these two traditions of language, can be understood as a difference in the selection of the metrical level that is realized in the pronunciation of the phonological representation of words. While Spanish realizes all beats at the base line of the metrical grid, which can be understood as the abstract chain of possible sonority peaks as specified in

11 Our model corresponds to some extent with the architecture of grammar outlined in Jacken­ doff (2002: 111–132), who also posits that different subsystems with independent principles generate structure, contrary to standard generative theories which focus on only one structure-generating principle.

414 

 Uli Reich and Martin Rohrmeier

Lexical Phonology, many normal Portuguese speech styles select the second and third level. This leads to the reduction and elision of vowels in weak positions and the compression of segmental material into complex syllables, as suggested by the traditional typological difference between syllable and word languages. We insist that these different processes cannot be attributed to the language-specific phonologies of Portuguese or Spanish, but rather to stylistic differences within these languages which have been normalized differently. This is corroborated by empirical findings: In a perception experiment, Dufter and Reich (2003) filtered segmental information from speech sounds in different communicative styles and asked experienced informants to attribute the resulting stimuli to Romance Languages. In the careful reading style, European Portuguese was systematically mistaken for Spanish. The explanation for this is straightforward: Portuguese speakers pronounce vowels at the base line of the metrical hierarchy when they want to make their speech more transparent. At the same time, there are dialects and styles of Spanish which, with regard to rhythm, sound more like what we call Portuguese here (see Kabatek 1994). The differences sketched for the three Romance languages in the preceding sections correspond to styles rather than to languages. It is only the lack of word accent or possibly even the lack of a phonological word as a domain for any phonological process or rule in French that seems to be a systematic difference compared to the other languages. Why and how these stylistic options have been normalized for the standards we associate with one language or dialect is a question we cannot answer. The following representations illustrate interactions between different phonological subsystems and the difference which arises out of the selection of different metrical levels which are pronounced. Syllable structure and the association to moraic structure are direct consequences of this selection, as we will show in the following examples. In a further step, we will model these facts in the framework of Optimality Theory. We strongly believe, however, that every formalization in such a theory needs to discuss first the structure-building principles of the generator (or better: generators) if we want to do more in our theorizing than rewrite given facts in formal representations. The graphic model for the example in (18) sketches a theoretically grounded representation of what has been called the word rhythm of European Portuguese:



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 415

(23) Portuguese

Example (23) shows the restructuring of segmental representations of words by syncopation and other processes. These processes arise out of the selection of the second and third, but not the first level of metrical strength for the realization of beats. The elision or skipping of vowels of low metrical strength leads to the insertion of consonants into adjacent syllables. These complex syllables will be projected to two moras or, in other languages, to just one mora, which would give us the configuration of syllabic feet in systems we call “not sensitive for quantity” in the standard parameterized version of Metrical Phonology (Hayes 1995: 54). Just like the musicians in (15) may vary in their performance by playing all of the notes (Jarrett) or only a reduced set of notes (time-span reduction) which are at higher levels of the metrical hierarchy (Evans), normal European Portuguese only “plays” vowels at higher levels, while careful urban Spanish varieties play all of the vowels represented in Lexical Phonology: (24) Spanish

416 

 Uli Reich and Martin Rohrmeier

The pronunciation of all vowels represented in the phonological form of the word is due to the base selection of the metrical hierarchy. The relations of lexical vowels, syllables and moras are far more direct than in (23) and make the distinctive potential of the word more transparent. This corresponds to a possible explanation of these facts in terms of Natural Phonology (Donegan and Stampe 1979), which would identify a base level style as “hearer-oriented” and a higher level style as “speaker-oriented” since the pronunciation of all phonemes of a word form enhances the distinctive potential of this form, while processes which suppress the realization of segmental contrasts lead to more economic articulatory gestures. Note that in (23) and (24) we find two phenomena which should be mentioned explicitly, leaving a complete theoretical account for future work. First, clitic elements like the article in (19) and (20) could be regarded as “extrametrical by nature” in languages like Spanish: Clitics are prosodically deficient, meaning that they do not carry any prominence. Thus, they are not parsed by an alternating metrical algorithm.12 More challenging for the elaboration of our theory is the competing projection of edge-based prominence which renders the first syllable of desaceleración prominent and thus leaves an unparsed syllable in the middle of the word, since the lexical stress of the last syllable must be respected. This looks like what has been called weak local parsing (Hayes 1995: 308) but must be carefully studied on empirical grounds.

5 Conclusion Our paper suggested that the fundamental structure-building principles of meter are the same in musical and linguistic utterances and that rhythmic variation such as the difference observed between Spanish and Portuguese corresponds directly to the rhythmical differences in performances of the same piece of music by different musicians. Besides the different use the language-specific grammars make of the main prosodic dimensions of prominence and timing, it is the selection of the metrical level we perform in real communicative acts that makes up the prosodic difference between languages which show the same prosodic configuration in pure grammatical aspects. The first difference sets French apart from the Ibero-Romance languages, while the second difference shapes the phonetic forms of utterances we attribute to Spanish or Portuguese.

12 See Reich (2004) for differences in the treatment of clitics in Spanish and Portuguese.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 417

It is very important to note that rhythm and meter are different in language and music because of the fundamental functional difference between these types of communication: The main difference between utterances we call music and utterances we call language consists in the construction of a proposition out of lexical primitives which are merged by syntactic operations. In order to guarantee communicative success, we must respect those dimensions of timing and prominence which serve to construct the semantic content and its pragmatic management in discourse. This observation does not hold in music, where melody, harmony, meter and rhythm follow structural relations of well-formedness which obey aesthetical rather than semantic restrictions. The picture of our cognitive architecture softens the notion of a rigid barrier between language and music as compact, secluded generators of the particular structures of the respective system. Rather, we propose the structure-building generators to be independent and interactive with other generators in ways determined by our communicative goals: If we need to make propositions transparent, the generator or alternating prominence must respect content-building prominence and timing, while it is free to cyclically realize metrical patterns in systems without lexical, morphological or pragmatic restrictions. From this perspective, our cognitive architecture for the communicative organization of sound consists of at least a structured list of form-function pairs such as words and constructions, a morphology which derives words out of words, a system of metrical algorithms, a set of syntactic principles which merges words to complex phrases and an indexical system which marks outputs as preferred or not preferred for social purposes. It is the alignment of the formal restrictions projected independently by each of these subsystems following a hierarchy determined by the conditions of communicative acts that makes an utterance musical or linguistic or both.

References Abaurre, Maria Bernadete and Charlotte Galves (1998): As diferenças rítmicas entre o Português Europeu e o Português Brasileiro: uma abordagem otimalista e minimalista [The rhythmic differences beween European Portuguese and Brazilian Portuguese]. Documentação de Estudos em Lingüística Teórica e Aplicada [Documentation of Studies in Theoretical and Applied Linguistics] 14/2: 377–403. Abercrombie, David (1967): Elements of General Phonetics. Edinburgh: Edinburgh University Press. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Clayton, Martin (2000): Time in Indian Music. Rhythm, Metre, and Form in North Indian Rāg Performance. (Oxford Monographs on Music.) Oxford: Oxford University Press.

418 

 Uli Reich and Martin Rohrmeier

Clayton, Martin, Rebecca Sager and Udo Will (2005): In time with the music: The concept of entrainment and its significance for ethnomusicology. European Meetings in Ethnomusicology 11: 3–75. Cooper, Grosvenor W. and Leonard B. Meyer (1960): The Rhythmic Structure of Music. Chicago: University of Chicago Press. Cross, Ian (2005): Music and meaning, ambiguity and evolution. In: Dorothy Miell, Raymond MacDonald and David J. Hargreaves (eds.), Musical Communication, 27–43. Oxford: Oxford University Press. Cross, Ian and Ghofur Eliot Woodruff (2009): Music as a communicative medium. In: Rudolf Botha and Chris Knight (eds.), The Prehistory of Language, 77–98. (Studies in the Evolution of Language 11.) Oxford: Oxford University Press. Deutsch, Diana, Trevor Henthorn and Rachael Lapidis (2011): Illusory transformation from speech to song. Journal of the Acoustical Society of America 129/4: 2245–2252. Donegan, Patrica Jane and David Stampe (1979): The study of Natural Phonology. In: Daniel A. Dinnsen (ed.), Current Approaches to Phonological Theory, 126–173. Bloomington: Indiana University Press. Drake, Carolyn, Amandine Penel and Emmanuel Bigand (2000): Why musicians tap slower than nonmusicians. In: Peter Desain and Luke Windsor (eds), Rhythm: Perception and Production, 245–248. Lisse: Swets & Zeitlinger. Dufter, Andreas and Uli Reich (2003): Rhythmic differences within Romance: identifying French, Spanish, European and Brazilian Portuguese. In: Daniel Recasens, Maria-Josep Solé and Joaquín Romero (eds.), Proceedings of the 15th International Congress of Phonetic Sciences, Barcelona, August 3–9, 2003, 2781–2784. Barcelona: Universitat Autònoma de Barcelona. Frota, Sónia (2000): Prosody and Focus in European Portuguese. Phonological Phrasing and Intonation. New York: Garland. Frota, Sónia and Marina Vigário (2000): Aspectos de prosódia comparada: ritmo e entoação no PE e no PB [Aspects of compared prosody: rhythm and intonation in European Portuguese and Brazilian Portuguese]. In: Rui Vieira Castro and Pilar Barbosa (eds.), Actas do XV Encontro Nacional da Associação Portuguesa de Linguística [Proceedings of the 15th Meeting of the Association of Portuguese Linguistics], Faro, 29–30 de Setembro e 1 de Outubro de 1999, vol. 1, 533–555. Braga: Associação Portuguesa de Linguística. Frota, Sónia and Marina Vigário (2001): On the correlates of rhythmic distinctions: The European/Brazilian Portuguese case. Probus 13/2: 247–275. Goldsmith, John A. (1990): Autosegmental and Metrical Phonology. Oxford: Blackwell. Halle, Morris and Jean-Roger Vergnaud (1987): An Essay on Stress. Cambridge, MA: MIT Press. Hayes, Bruce (1995): Metrical Stress Theory: Principles and Case Studies. Chicago: University of Chicago Press. Houle, George (1987): Meter in Music, 1600–1800: Performance, Perception, and Notation. (Music / Scholarship and Performance.) Bloomington: Indiana University Press. Hulst, Harry van der (1999): Word accent. In: Harry van der Hulst (ed.), Word Prosodic Systems in the Languages of Europe, 3–115. (Empirical Approaches to Language Typology 20/4.) Berlin/New York: Mouton de Gruyter. Jackendoff, Ray and Fred Lerdahl (2006): The capacity for music: What is it, and what’s special about it? Cognition 100: 33–72. Kabatek, Johannes (1994): México frente a Madrid: Aspectos fonéticos del habla de los taxistas en dos capitales hispanas. Iberoamericana 54/2: 5–15.



Batidas latinas: On rhythm and meter in Spanish and Portuguese  

 419

Katz, Jonah and David Pesetsky (2011): The Identity Thesis for Language and Music. Available at http://ling.auf.net/lingBuzz/000959. Koch, Heinrich Christoph (1787): Versuch einer Anleitung zur Composition. Leizpig: Böhme. Large, Edward W. and Mari Riess Jones (1999): The dynamics of attending: How people track time-varying events. Psychological Review 106/1: 119–159. Large, Edward W. (2000): On synchronizing movements to music. Human Movement Science 19/4: 527– 566. Large, Edward W. and Caroline Palmer (2002): Perceiving temporal regularity in music. Cognitive Science 26/1: 1–37. Lerdahl, Fred and Ray Jackendoff (1983): A Generative Theory of Tonal Music. Cambridge, MA: MIT Press. Lerdahl, Fred (2001): Tonal Pitch Space. Oxford: Oxford University Press. Liberman, Mark and Alan Prince (1977): On stress and linguistic rhythm. Linguistic Inquiry 8/2: 249–336. London, Justin (2004): Hearing in Time: Psychological Aspects of Musical Meter. Oxford: Oxford University Press. Mester, Armin R. (1994): The quantitative trochee in Latin. Natural Language and Linguistic Theory 12: 1–61. Miell, Dorothy, Raymond MacDonald and David J. Hargreaves (eds.) (2005): Musical Communication. Oxford: Oxford University Press. Moraes, João Antônio de (1998): Intonation in Brazilian Portugese. In: Daniel Hirst and Albert Di Cristo (eds.), Intonation Systems: A Survey of Twenty Languages, 179–194. Cambridge: Cambridge University Press. Patel, Aniruddh D. (2008): Music, Language, and the Brain. Oxford: Oxford University Press. Pike, Kenneth Lee (1945): The Intonation of American English. Ann Arbor: University of Michigan Press. Rebuschat, Patrick, Martin Rohrmeier, John A. Hawkins and Ian Cross (eds.) (2012): Language and Music as Cognitive Systems. Oxford: Oxford University Press. Reich, Uli (2004): Ritmo, saliencia prosódica y clitización en español y portugués. In: Trudel Meisenburg and Maria Selig (eds.), Nouveaux départs en phonologie. Les conceptions subet suprasegmentales, 125–137. Tübingen: Narr. Reich, Uli (2007): Differentielle Phonologie und polyglotte Kompetenz. Prinzipien und der Fall des Französischen in Dakar. Habilitationsschrift (Habilitation thesis), Universität zu Köln. Roca, Iggy M. (1999): Stress in the Romance languages. In: Harry van der Hulst (ed.), Word Prosodic Systems in the Languages of Europe, 659–811. (Empirical Approaches to Language Typology 20/4.) Berlin/New York: Mouton de Gruyter. Rohrmeier, Martin (2011): Towards a generative syntax of tonal harmony. Journal of Mathematics and Music 5/1, 35–53. Rohrmeier, Martin and Patrick Rebuschat (2012): Implicit learning and acquisition of music. Topics in Cognitive Science 4/4: 523–553. Scruton, Roger (1999): The Aesthetics of Music. Oxford: Oxford University Press. Sosa, Juan Manuel (1999): La entonación del español. Su estructura fónica, variabilidad y dialectología. Madrid: Cátedra. Stevens, Catherine (2004): Cross-cultural studies of musical pitch and time. Acoustical Science and Technology 25/6: 433–438. Stevens, Catherine and Tim Byron (2009): Universals in music processing. In: Susan Hallam, Ian Cross and Michael Thaut (eds.), The Oxford Handbook of Music Psychology, 14–23. Oxford: Oxford University Press.

420 

 Uli Reich and Martin Rohrmeier

Stobart, Henry and Ian Cross (2000): The Andean anacrusis? Rhythmic structure and perception in Easter songs of Northern Potosí, Bolivia. British Journal of Ethnomusicology 9/2: 63–92. Vennemann, Theo (1988): Preference Laws for Syllable Structure and the Explanation of Sound Change. With Special Reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter.

Stephan Schmid (University of Zurich)

Syllable typology and the rhythm class hypothesis: Evidence from Italo-Romance dialects* Abstract: This contribution explores the extent to which a phonological typology based on the dichotomy of “syllable languages” vs. “word languages” (Auer 1993, 2001) may be combined with the so-called “rhythm class hypothesis”, an experimental approach that aims at grasping the rhythmic properties of languages by applying different “rhythm metrics” (cf. Ramus, Nespor, and Mehler 1999; Grabe and Low 2002; Dellwo 2006). Accordingly, the study brings together two lines of research that have moved from the same starting point, namely the failure of the traditional “isochrony hypothesis” (Pike 1945; Abercrombie 1967). The empirical basis of the present study consists of data from ten Italo-Romance dialects. Two different types of analysis are carried out. First, a syllabic typology is sketched in terms of phonotactic markedness, comparing the ten dialects with regard to the number of segments that may occur in syllables and with regard to the sonority relations that exist within syllable constituents. Second, different rhythm metrics are calculated for nine of the ten dialects; the acoustic data consist of ten utterances from each dialect which have been segmented into vocalic and consonantal intervals. The results show that the Italo-Romance dialects investigated can be ranged along a continuum of increasing syllable complexity and that dialects with similar phonotactic properties tend to group together in the different rhythm-metric topo­logies.

1 Rhythm-oriented typology in the new millenium The bulk of present-day scientific literature on speech rhythm takes as its starting point the so-called “isochrony hypothesis” which was originally proposed by Pike (1945) and then restated by Abercrombie (1967). As is well known, the * Some aspects of this research have been presented on three previous occasions, namely i) at the workshop on Empirical Approaches to Speech Rhythm (London, UCL, 28 March 2008), ii) at the workshop on Consonant Clusters and Structural Complexity (Munich, LMU, 2 August 2008), iii) at the congress of the Associazione Italiana di Scienze della Voce on The Temporal Dimension of Speech (University of Zurich, 4 February 2009). I am indebted to many persons who have discussed some of the issues raised in this paper; needless to say, I alone am responsible for any shortcomings herein.

422 

 Stephan Schmid

classical dichotomy distinguishes between two major types of languages labeled “syllable-timed” and “stress-timed” (or “isosyllabic” and “isoaccentual”); French and Spanish are examples of the former type, while German and English belong to the latter. Occasionally, a third rhythm type has been taken into account, i.e. the so-called “mora-timed” languages such as Japanese or Telugu (e.g. Hoequist 1983; Murty, Otake, and Cutler 2007). The isochrony hypothesis makes two fundamental claims: i) every language belongs to one particular rhythm type; ii) rhythm types are defined in terms of a timing unit (syllable, foot, mora), which is supposed to occur in sequences of regular intervals with equal durations (cf. Auer and Uhmann 1988: 217).1 To be more precise, the main assumption shared in the actual debate on speech rhythm is not the isochrony hypothesis per se, but rather what is generally considered the failure of the strong version put forward by Abercrombie. In effect, a number of experimental investigations carried out with different languages have proved that i) in the alleged syllable-timed languages the duration of syllables varies according to the number of their segments, and ii) in the alleged stress-timed languages the duration of feet varies according to the number of their syllables (cf. e.g. Auer and Uhmann 1988: 219–237). The impasse created by the experimental confutation of the isochrony hypothesis (at least in its strict sense) led to several attempts either to save the original idea, which continued to be intuitively plausible, or to reformulate it by modifying some of its tenets (Auer and Uhmann 1988: 237–255; Bertinetto 1989: 103–120). Departing from the traditional isochrony framework, two lines of research moved away from speech production towards other domains of linguistic inquiry, i.e. speech perception or phonology; subsequently, a third line returned to acoustic phonetics measuring units of speech other than the syllable and the foot. The “perceptual illusionists” – as Bertinetto (1989: 101–102) called them – immediately assumed that since syllables and feet do not show uniform durations in the acoustic signal, the phenomenon of isochrony might well lie in the way listeners perceive speech (cf. Auer and Uhmann 1989: 241–243). Later, psycholinguists tested the ability of newborn infants and adults to discriminate between languages of different rhythm types, using for instance delexicalized stimuli (e.g.

1 For reasons of space, these introductory remarks are very reductive from both a qualitative and quantitative point of view. Reviews of the scientific literature on the isochrony debate towards the end of the last century can be found in Auer and Uhmann (1988), Bertinetto (1989) and Ramus, Nespor, and Mehler (1999: 265–270); also consider the bibliography on timing and rhythm in speech provided by Roach (2003). The rhythm metrics developed in the new millenium are presented in Barry (2010).



Syllable typology and the rhythm class hypothesis  

 423

Ramus, Nespor, and Mehler 1999: 275–287; Ramus, Dupoux, and Mehler 2003: 338–341). But perhaps the most valuable insights in this respect – often neglected in the rhythm debate – derive from experiments on word recognition: It has indeed been shown that, in order to segment the speech chain, listeners rely on different phonotactic cues depending on the rhythm type of their native language (e.g. Murty, Otake, and Cutler 2007; Kim, Davis, and Cutler 2008). Compared to the perceptual line of investigation, the so-called “phonological illusionists” – to use another label created by Bertinetto (1989: 108–110) – were more prodigious. The phonological turn appeared in the 1980s (e.g. Dauer 1983, 1987) and was developed by several scholars (cf. Auer and Uhman 1988: 244–253; Bertinetto 1989: 108–110). There are two basic tenets of this view: First, rhythm types are not to be found in the speech signal itself, but rather derive from a bundle of properties of the phonological system, the most important being the complexity of syllable structure and the reduction of unstressed vowels (a synoptic view of some phonological factors proposed in the literature can be found in Schmid 1997: 249); second, rhythm classes are not absolute categories, but rather constitute poles of a typological continuum, allowing for mixed or intermediate types. In line with this way of thinking, the most elaborate proposal was formulated by Auer in his pioneer study (Auer 1993; cf. also Auer 2001: 1395–1398). The strength of this model lies in i) the empirical evidence grounded on a typological survey of 34 languages, ii) the great number of phonological phenomena taken into account, iii) the conceptual shift from the stress vs. syllable dichotomy towards two new typological classes (i.e. “syllable languages” and “word languages”), and iv) the impact of this approach on dialectological and diachronic studies (e.g. Nübling and Schrambke 2004; Szczepaniak 2007). I will not discuss this framework in detail, which is better explained in other contributions to this volume. Instead, some words should be spent on the so-called “rhythm class hypothesis”, another line of research which has gained an increasing interest in the last decade and can be regarded as complementary to the word vs. syllable language typology. The two approaches differ mainly in the research techniques employed (according to the classical division of labor between phonetics and phonology) and – to some extent – in the linguistic phenomena under analysis. Nevertheless, the rhythm class hypothesis does have several points in common with the phonological typology of syllable and word languages. Besides the mere rejection of the traditional isochrony hypothesis (the common starting point, so to speak), a more important shared assumption lies in the prototypical conception of rhythm classes, where mixed or intermediate types are possible. Moreover, the elaboration of the new “rhythm metrics” was, at least at its beginning, a direct consequence of the phonological turn. This can be seen as an instance of fruitful

424 

 Stephan Schmid

interaction between the study of phonetics and phonology, where the phonological interpretation of phonetic facts (the lack of syllable and foot isochrony in the languages of the world) leads in a heuristic circle to the formation of new hypotheses for phonetic research. In an influential study, Ramus, Nespor, and Mehler (1999) argue that linguistic rhythm is not an expression of isochrony in terms of syllable or foot duration, but that other units of speech have to be measured, namely vocalic and consonantal intervals. If, according to the phonological turn, languages basically differ in terms of syllable com­plexity and vowel reduction, the following acoustic rhythm measures should be calculated: i) the percentage of the duration of vocalic intervals in a sentence (%V), ii) the standard deviation of the duration of vocalic intervals in a sentence (∆V), and iii) the standard deviation of the duration of consonantal intervals in a sentence (∆C). The main argument is that these rhythm measures constitute acoustic correlates of typologically relevant phonological properties. More precisely, %V (i) and ∆C (iii) refer to syllable complex­ity: For word languages, it is predicted that the numerous syllable types (in particular, the existence of heavy consonant clusters) will reduce the duration of the vocalic part in the overall duration of the speech signal (%V) and increase the durational variability of consonantal intervals (∆C); on the contrary, syllable languages should display a higher overall proportion of vocalic intervals (%V) and a lower durational variability of consonantal intervals (∆C). With regard to ∆V (ii), this rhythm measure reflects another phono­logical property which divides word languages from syllable languages, i.e. vowel reduction processes: It is predicted that word languages will display higher ∆V values, given the greater durational differences between stressed and un­stressed vowels; on the contrary, syllable languages should have lower ∆V values, since the duration of unstressed vowels is generally less reduced than in word languages. The most popular outcome of this approach lies in the visualization of the rhythm measures in three graphs, where languages with similar rhythmic properties appear to be grouped together in the same area of a bidimensional plane: %V–∆C, %V–∆V, ∆V–∆C (see Ramus, Nespor, and Mehler 1993: 273–374). While such a representation has the advantage of also illustrating the continuum nature of rhythmic properties, one might object – from a phonological point of view – that the graphs constitute a “topology” rather than a typology of rhythm, given that a bidimensional plane permits us to grasp at once only two factors within a rather complex interplay of prosodic and phonotactic phenomena. The rhythm class hypothesis has been formulated in different variants (cf. Barry 2010 for an overview). For instance, the finding that in particular ∆C varies noticeably as a function of speech rate (Dellwo and Wagner 2003) led some researchers to apply variation coefficients (Varco) to vocalic and conso-



Syllable typology and the rhythm class hypothesis  

 425

nantal intervals (rather than standard deviations) in order to neutralize speech rate effects on rhythm measures (see Dellwo 2006). While this proposal can be regarded as a statistical refinement of the model elaborated by Ramus, Nespor, and Mehler (1999), the so-called Pairwise Variability Indices (PVI) introduces a substantial change in the view of speech rhythm (see e.g. Grabe and Low 2002). Most of the work done with this metric shares Ramus’ assumption that vocalic and consonantal intervals are the basic units of segmentation in the acoustic signal, but this approach refuses to consider %V as a component of speech rhythm. In fact, the proportion of vowels within the duration of an utterance is a more paradigmatic aspect of the sound shape of a given language, whereas the dynamic nature of rhythm should focus instead on the syntagmatic axis; with the PVI, this is achieved by calculating the averaged differences between the durations of pairs of subsequent vocalic and consonantal intervals (hence the term Pairwise Variability Index; see Grabe and Low 2002: 519–527 for details). Still another view of speech rhythm lies at the basis of the Control/Compensation Index (CCI) introduced by Bertinetto and Bertini (2008). The CCI algorithm constitutes a further development of the PVI, but this model is explicitly more phonological in nature in that it takes another element into account, namely the number of segments which make up the vocalic and consonantal intervals. The theory thus predicts that rhythm classes are formed by either “controlling” or “compensating” languages, according to the degree of durational variability they allow for at the segmental level. The CCI model deserves more attention in the future, but it will not be applied to our data – partly for lack of space, partly because it seems that for the time being the algorithm has not been tested on many languages (with the exception of Mairano and Romano 2010: 95). Finally, the Rhythm Class Hypothesis itself has been questioned by a number of scholars (see e.g. the special issue of Phonetica 66.1-2): In particular, it has been argued that “filling phonological structures with duration measurements may be reflecting segmental duration patterns to a certain extent, but they are unable to capture rhythmic movement patterns evolving with a degree of regularity over time” (Kohler 2009: 33–34). In the empirical part of this study, data from nine Italo-Romance dialects will be analysed by means of the correlates proposed by Ramus, Nespor, and Mehler (1999), i.e. %V, ∆C, and ∆V (cf. section 4.2), also applying the Varco metric suggested by Dellwo (2006) (cf. section 4.3); furthermore, the Grabe and Low (2002) method will be exploited by calculating the respective PVI values (cf. section 4.4). Before I present the acoustic analyses, some phonotactic properties of the dialects under examination will be described. Of course, syllable structure constitutes only one of the several factors considered in the typological framework of Auer (1993: 41–87; cf. also Szczepaniak 2007: 52–53), but it appears to be the most

426 

 Stephan Schmid

promising for an attempt to bring together the phonological typology of syllable and word languages and the rhythm class hypothesis. However, before passing to the empirical section, it is useful to briefly review some previous work that has dealt with isochrony and rhythm in Italian and Italo-Romance dialects.

2 Italian and Italo-Romance dialects 2.1 Italian From a phonological point of view, (Standard) Italian does meet several requisites of a syllable language, even if in a less prototypical form than Spanish. For instance, with respect to the difference between accented and non-accented syllables (cf. Auer 1993: 83) we note that Spanish has the same five vowels /i e a o u/ both in stressed and unstressed syllables, whereas Italian presents seven stressed vowels /i e ɛ a o ɔ u/, but only five unstressed vowels /i e a o u/, with mid lax vowels becoming tense; moreover, the core lexicon does not present non-accented /u/ in word-final position (cf. Schmid 1999a: 129–133). Of course, this paradigmatic reduction in the vowel inventory is phonemic in nature and as such less pervasive than the phonetic reduction of unstressed vowels – both on the spectral and the durational level – in a language like English (cf. Szczepaniak 2007: 52); nevertheless, slight shortening and centralization phenomena may also occur in Italian, in particular in fast speech (cf. Savy and Cutugno 1997). Rather than reducing unstressed vowels, Italian lengthens stressed vowels in word-internal open syllables: /ˈpane/ ‘bread’ → [ˈpaːne] (cf. Schmid 1999a: 165–167); note that such an allophonic rule is lacking in Standard European Spanish. In a sense, this allophonic vowel lengthening of Italian constitutes a feature of a word language, despite its fundamental character as a syllable language. To put it in more formal terms, the lengthening of stressed vowels in open syllables can be interpreted as an instantiation of a bimoric constraint on stressed syllables, which appears in a complementary form in the well-known rule of raddoppiamento fonosintattico: If an Italian word ends with a stressed final vowel and the following word begins with a single consonant, the latter is doubled, e.g. /kanˈtɔ/+/ˈbɛne/ ‘(he) sang well’ → /kan.ˈtɔ b.ˈbɛ.ne/ → [kanˈtɔ ˈbːɛːne]; cf. Schmid 1999a: 170–173). Again, phonosyntactic doubling is lacking in Spanish, as are lexical geminates, which in turn are a prominent element of Italian phonology, clearly shown in numerous minimal pairs like /ˈfa.to/ ‘fate’ ~ /ˈfat.to/ ‘made’ (cf. Schmid 1999a: 167–170). In Italian, however, geminates occur both after stressed and unstressed vowels (e.g. in /af.ˈfat.to/ ‘at all’), which is a feature of syllable languages, whereas typical word languages have geminates only after stressed vowels (cf. Auer 1993: 67).



Syllable typology and the rhythm class hypothesis  

 427

One last trait of a syllable language evident in Italian lies in the existence of allomorphs depending on syllable structure (cf. Auer 1993: 50, 2001: 1396–1397). For instance, the singular definite masculine article has three allomorphs. Nouns beginning with a single consonant or a tautosyllabic cluster select the form il (e.g. il libro ‘the book’, il treno ‘the train’), whereas nouns beginning with a vowel select the consonantal form l’ – e.g. in l’albero ‘the tree’ – in which a process of postlexical resyllabification occurs (/ˈlal.be.ro/). Finally, nouns beginning with an “intrinsic” geminate or a heterosyllabic cluster select the form lo, e.g. in lo scemo ‘the fool’ or lo straniero ‘the stranger’ (where again postlexical resyllabification takes place, since the first half of the geminate and the “extrasyllabic” /s/ occupy the coda position of the preceding syllable). Indeed, the syllable-based rhythm of Italian has long been acknowledged (cf. e.g. Bertinetto 1981: 168–191; Schmid 1999a: 178–179). As a matter of fact, it can be inferred from poetic verse, which is based on syllable counting. Moreover, the syllable-timed character of Italian speech rhythm has long been maintained in the linguistic literature: Not occasionally, the expression “syllabic blood” appears in the title of the first experimental rhythm study on Italian, which reports evidence from duration measurements of feet and syllables in the speech read by two speakers from Turin (Bertinetto 1977). Some following studies – e.g. Vayra, Avesani, and Fowler (1984) – were more sceptical with regard to this classification, but in general many scholars have continued to regard Italian as a syllable language. In particular, in the studies following the rhythm class hypothesis, Italian is often plotted together with the other (presumably syllable-based) Romance languages. This is clearly the case in the three rhythm topologies that Ramus, Nespor, and Mehler (1999: 273–274) provided on the basis of the three correlates %V, ∆V, and ∆C; the acoustic data were taken from sentences read by four speakers of unknown geographical provenance. Schmid (2001) applied Ramus’ metrics to three Germanic and three Romance languages (Standard German, Swiss German, English; French, Spanish, Italian), measuring read speech of one speaker per language; with respect to %V and ∆C, the speaker from central Italy had the most syllable-based values. Mairano and Romano (2007) measured the Northwind and the Sun read by two Italian speakers and obtained similar results for the male speaker, whereas the female speaker attained greater ∆V and ∆C values;2 to some extent, the difference between the two speakers could be nor-

2 The female speaker, brought up in Rome, had been recorded for the phonetic transcription in Rogers and Arcangeli (2004), whereas the male speaker – recorded by Mairano and Romano (2007) – is a trained phonetician of southern origin, reading in Standard Italian.

428 

 Stephan Schmid

malized when Mairano and Romano (2010) applied the Varco and PVI metrics to the same data (note that Italian was not part of the comparative PVI study of Grabe and Low 2002). Several researchers have investigated the question of whether the rhythm of Italian differs from one regional variety to the other. Barry et al. (2003) compared speakers from Bari, Naples, and Pisa with speakers of German and Bulgarian, also analysing the influence of speech rate: The study revealed that for some metrics (e.g. ∆C) the Italian values overlapped with those of the speakers of other languages, whereas the Italians performed differently from the German and Bulgarian speakers with regard to %V; in particular, they seemed to reduce the vocalic intervals less as tempo increased. Looking at prosodic differences in the Italian of speakers from Sicily and Veneto, White, Payne, and Mattys (2009) were surprised to find relatively homogeneous behavior among the two varieties. Finally, Giordano and D’Anna (2010) calculated rhythm measures for 15 regional varieties of Italian, again finding a considerable variability in the applied rhythm metrics (∆V, PVI, etc.); nevertheless, in all regional varieties the %V values are consistent with an overall syllable-based pattern (ranging from 41.0 to 48.4). At this point, two observations are in order. First, the very fact that regional varieties may exhibit different rhythm patterns is at least partially at odds with a basic tenet of the rhythm class hypothesis, namely that correlates such as %V and ∆C directly reflect the syllable structure constraints of a language. Since regional varieties of a given language basically share the same phonotactics, rhythmical differences must be due to other prosodic factors or to rules of speech, such as allophonic vowel lengthening or shortening. Second, in order to understand the purpose of the present study, it is crucial to distinguish between regional varieties of Italian and Italo-Romance dialects. The former are varieties of the national standard language of Italy, sharing the same morphology, syntax, and lexicon, whereas the latter are independent Romance languages with substantial differences on all structural levels; for instance, as far as phonology is concerned, they have different phoneme and syllable type inventories (cf. Schmid 1999b).

2.2 Italo-Romance dialects Not surprisingly, scholars have attempted to classify Italian dialects in terms of isochrony. For instance, Trumper, Romito, and Maddalon (1991: 348) proposed a continuum of Italian dialects, ranging from a rather syllable-timing pole (Veneto and Tuscan) towards a rather stress-timing pole (Apulian), with Calabrian somewhere in the middle. While this proposal is to some extent influenced by the prototypical view of speech rhythm initiated in the phonological turn, its empirical



Syllable typology and the rhythm class hypothesis  

 429

basis is clearly phonetic as it lies in the measurements of the durations of stressed word-final vowels. Still in a prototypical vein, a more phonological argumentation was developed in the diachronic research of Mayerthaler (1996), who analysed the phonological processes affecting the syllable structure of Italian dialects as they moved away from Vulgar Latin (syncope, epenthesis, gemination, etc.). Considering a larger number of dialects, this study establishes a typological and geographical continuum according to the degree of syllable complexity, where phonotactic markedness is interpreted in the sense of Vennemann’s (1988) preference laws for syllable structure (Mayerthaler 1996: 208–211). Basically, the classification contains four dialect areas: i) at the pole with a rather simple syllable structure we find the dialects of the “southern area” (Sicily, Salento, and southern Calabria); ii) less simple syllable structures appear in the “central-southern area” which comprises the regions of Apulia, Abruzzo, Basilicata, Campania, and southern Lazio; iii) the complexity of syllable structures is assumed to increase as we move towards the “central area” (Tuscany, Umbria, Marche, Northern Lazio); finally, the pole with the most complex syllable structures is represented by the northern dialects of the “Gallo-Italian area”, among which Mayerthaler includes Romagnolo and Piedmontese. The differences between the two dialect continua sketched by Trumper, Romito, and Maddalon (1991) on the one hand, and by Mayerthaler (1996) on the other, lie not only in the phonetic or phonological nature of the data (vowel durations vs. syllable structure), but also in the number of dialects considered. The former analyse four dialectal varieties, whereas the latter considers a much larger number of dialects; still, even Mayerthaler’s sample is not complete, since she does not take into account two other Gallo-Italian dialects (Lombard and Ligurian) or the Veneto dialects. Moreover, the two proposals differ with respect to the collocation of certain dialects. Apulian is located at the stress-timed pole in the first classification, but it is intermediate or rather isosyllabic in the second classification; similarly, Tuscan represents the syllable-timed pole in the first case, whereas in the second case it is again classified as intermediate, but less syllable-based. Finally, an important difference between the two proposals resides in the fact that Trumper, Romito, and Maddalon (1991) do not relate their continuum to geolinguistic space, whereas Mayerthaler’s (1996) classification explicitly combines linguistic geography with phonological typology, in that the phonotactic continuum gradually moves from the syllable-based south towards the word- and stress-based north of Italy. Given the rather contrasting pictures that arise from these two continua, let us now examine i) whether a synchronic analysis of syllable structure can provide additional evidence that permits us to elucidate the rhythm typology of Italo-Ro-

430 

 Stephan Schmid

mance dialects (section 3), and ii) whether the application of the new rhythm metrics reveals results similar to the phonotactic analysis (section 4).

3 A sketch of syllabic typology of Italo-Romance Within the typological framework of syllable and word languages, syllable structure plays a prominent role, given that the first three parameters of the model all deal with phonotactic markedness (cf. Auer 1993: 41–44): i) the number of segments in the syllabic shell, ii) coda complexity, and iii) sonority relations in the syllable shell. Parameters i) and ii) may thus be considered quantitative in nature (number of segments), whereas parameter iii) is more qualitative, as it concerns the acoustic timbre of syllable constituents. The theory predicts that syllable languages permit a smaller amount of numerical syllable complexity than word languages, and that syllable languages tend to order their segments within syllable constituents according to the universal preferences for sonority relations (in the sense of Vennemann 1988). In other words, syllable structure is supposed to be less marked in a syllable language than in a word language (cf. Szczepaniak 2007: 30–35, 52–53). In the following description, both numerical complexity and sonority relations will be discussed, but the numerical aspect is dealt with in a somewhat more holistic approach, i.e. in terms of syllable type inventories (cf. 3.2). Obviously, the size of a syllable type inventory depends on the complexity of the syllable constituents (heads, nuclei, codas), which therefore will be taken into account for several points of the analysis (in particular with regard to sonority relations, cf. 3.3). Nevertheless, a systematic account of the structure of syllable constituents lies outside the scope of this contribution (see Schmid 1998 and 2000 for separate descriptions of northern and southern Italian dialects). Also, the rather diverging vowel systems of the ten dialects cannot be described in detail here (see Schmid 1999b for a typological overview), but some aspects, such as the existence of distinctive vowel quantity and/or centralized unstressed vowels, will be briefly mentioned (cf. 3.4), given that the phenomenology of vowel reduction plays a crucial role in the phonological typology of word and syllable languages.

3.1 Syllable types: Data collection and analysis A syllable type is a sequence of categorical symbols (e.g. CV, CVC, CGV, etc.), with three major categories: consonants (C), vowels (V), and glides (G). Treating glides as a separate category has the advantage of highlighting the existence of rising



Syllable typology and the rhythm class hypothesis  

 431

and falling diphthongs; in fact, the structure of the nucleus also contributes to syllable complexity, an aspect that is sometimes treated in terms of segment inventory complexity (cf. Auer 1993: 84). A syllable type inventory lists the different syllable types that occur in the lexicon of a given language, usually ranking them in order of increasing complexity (CV… CCCVCCC). A first comparison of syllable structure complexity is achieved by simply ordering the different languages or dialects according to the number of syllable types they allow.3 For this study, the syllable type inventories of ten Italo-Romance dialects were established manually through the consultation of dictionaries. In Table 1, the dialects are presented in a rough geographical order, which basically proceeds from the northeast to the southwest of the national territory. The table also indicates the sources (i.e. the dictionaries) which were consulted and adds some general geolinguistic information about the dialects, referring to the town and/or region where the dialects are spoken, as well as to the corresponding typological macro-areas postulated by Mayerthaler (1996), i.e. “southern”, “central-southern”, “central”, and “Gallo-Italian”. Table 1: The ten Italo-Romance dialects analysed in the phonotactic study Dialect

Town

Friulian

Region

Area according to Mayerthaler (1996)

Source

Friuli-Venezia Giulia

(not considered)

Faggin (1985)

Venetian

Venice

Veneto

(not considered)

Boerio (1856)

Feltrino

Feltre

Veneto

(not considered)

Migliorini and Pellegrini (1971)

Milanese

Milan

Lombardy

(not considered)

Arrighi (1896)

Turinese

Turin

Piedmont

Gallo-Italian

Gavuzzi (1891)

Emilia-Romagna

Gallo-Italian

Morri (1840)

Tuscany

central

Malagoli (1939)

Campania

central-southern

Altamura (1968)

Apulia

central-southern

Saracino (1957)

Sicily

southern

Traina (1868)

Romagnolo Pisan

Pisa

Neapolitan Naples Bitontino Sicilian

Bitonto

3 An attempt to assign languages to three discrete categories defined in terms of syllable structure complexity has been made by Maddieson (2005b), who distinguishes between “simple syllable structure”, “moderately complex syllable structure”, and “complex syllable structure”; for instance, languages with a simple syllable structure basically allow CV syllables, whereas in a language with a moderately complex syllable structure the most elaborate template is CCVC, with the second element of the onset being a liquid or a glide. Recently, a scale with eight different degrees of syllable complexity has been proposed (Maddieson 2011).

432 

 Stephan Schmid

3.2 Syllable types: The numerical complexity of syllable structure Figure 1 offers a comparison of the syllable type inventories in the dialects examined. On the one hand, a considerable diversity emerges, as the maximum number of 40 syllable types in Friulian is more than twice the minimum number of 18 syllable types in Pisan. On the other hand, the dialects can hardly be classified in two groups (syllable-timed vs. stress-timed), since the number of syllable types slowly increases from left to right; therefore, it is rather difficult to say where a division line between isosyllabic and isoaccentual dialects might be drawn. Tentatively, we could assign the five dialects to the left (from Pisan to Bitontino) to the “syllable dialects” and allocate the five dialects to the right (from Milanese to Friulian) to the “word” or “accent sphere”, but there is no “quantal” difference between the least syllable-based dialect, Bitontino (26 syllable types), and the least word-based dialect, Milanese (28 syllable types). Thus, a paradigmatic description of syllable structure in terms of numbers of segments corro­borates the prototypical understanding of rhythm classes, an essential element of the phonological typology of syllable and word languages.

Figure 1: Number of syllable types in ten Italo-Romance dialects

Comparing our data with previous rhythm typologies of Italo-Romance (cf. 2.2), we note a general accordance, but also some minor discrepancies. Roughly speaking, the five dialects to the left have been classified previously as either isosyllabic or intermediate. For instance, Trumper, Romito, and Maddalon (1991) consider both Venetian (24 syllable types) and Pisan (= Tuscan) syllable-timed,



Syllable typology and the rhythm class hypothesis  

 433

the latter indeed having the smallest inventory (18 syllable types); according to Mayerthaler (1996), however, Pisan belongs to an intermediate type. In her continuum, the syllable-based pole is occupied by Sicilian (23 syllable types), whereas Neapolitan and Bitontino (21 and 26 syllable types, respectively) belong to an intermediate area, despite their rather simple syllable structure. With regard to the classification of the five dialects on the right (which all allow more complex syllable types), there is less disagreement in the literature, since not all of them were considered in the two studies mentioned above. More precisely, only Turinese (= Piedmontese) and Romagnolo are taken into account by Mayerthaler (1996), who locates them at the accent-based pole of the dialectal continuum; our analysis does indeed reveal a high degree of numerical syllable complexity for these two dialects (both have 35 syllable types). Note, however, that the numerical syllable complexity is highest in Friulian (40 syllable types). Towards the center of the continuum we find two dialects with a lower numerical complexity, i.e. Milanese (28 syllable types) and Feltrino (29 syllable types); interestingly, Milanese is Gallo-Italian (as much as Romagnolo and Turinese), whereas Feltrino belongs to the same dialect group as Venetian, i.e. the Veneto dialects (cf. Schmid 1996). The difference between Venetian and Feltrino is essentially due to the fact that the latter has dropped all unstressed final vowels except -a in its evolution from Vulgar Latin, thus experiencing the same diachronic development as the Gallo-Italian dialects (Milanese, Turinese, and Romagnolo). Synchronically, variation in numerical syllable complexity is determined by different coda constraints. With regard to the syllable-based varieties (i.e. the five dialects on the left) and Standard Italian, codas in underlying forms basically contain one consonant (a nasal, liquid, or /s/) and occur word-internally (e.g. tempo ‘time’, porta ‘door’, festa ‘party’); words normally end in a vowel. Quite differently, stress-based dialects also allow obstruents and consonant clusters in word-final position (cf. Table 2 below); ultimately, the phonotactic constraints that operate synchronically are the result of diachronic apocope phenomena. Furthermore, stress-based dialects may also exhibit a higher complexity in the nucleus as a consequence of diachronic diphthongization processes which have enhanced the perceptual gestalt of accented syllables. This partly explains the high number of syllable types in Friulian, where the nucleus often contains a diphthong even in closed syllables (compare Friul. fuart ‘strong’ and biel ‘beautiful’ with It. forte and bello), while Italian diphthongs are derived only from short Latin vowels in open syllables (e.g. in buono ‘good’ < bŏnu(m)). Conversely, the numerical complexity of syllable onsets – or of the syllable “head”, following Vennemann’s (1988) terminology – appears to be less diversified in the Italo-Romance dialects. The general pattern comprises C, CC, and CCC

434 

 Stephan Schmid

heads. In Standard Italian, for instance, a biconsonantal head typically consists of an obstruent followed by /r/ or an approximant (e.g. It. treno ‘train’ and piano ‘slow’); word-initially, an “extrasyllabic” /s/ may be added to the left, creating a CCC head (e.g. It. strada ‘street’). Only two stress-based dialects – Romagnolo and Turinese – seem to allow word-initial heads also with four segments (cf. Table 2 below). Summarizing the findings on the numerical complexity of syllable types, one may conclude that the three constituents are not all of equal importance for the typology of syllable and word languages: In Italo-Romance, coda complexity is the most important parameter, followed by nucleus complexity, whereas heads are of lesser importance. Note that this view slightly differs from the original typology, where the complexity of syllable structure was analysed on the basis of the number of segments in the syllable “shell”, i.e. the union of the syllable’s head and coda (cf. Auer 1993: 41).

3.3 Sonority relations in syllable heads and codas In the preceding section, we gradually departed from a purely numerical analysis of syllable complexity and moved instead towards another aspect, namely the ordering of different segment classes within syllable constituents. As mentioned above, the typology of syllable and word languages does in fact add another important relevant parameter to the analysis of syllable structure, i.e. the sonority relations between the segments that constitute the head and the coda of a syllable (cf. Auer 1993: 43; Szczepaniak 2007: 52). Thus, we need a closer insight into the phonotactic diversity of Italo-Romance dialects. This is partially achieved in Table 2, which illustrates possible consonant clusters in the five stress-based dialects (Obs = obstruent, Son = sonorant, Plo = plosive, Fri = fricative, Aff = affricate, N = nasal, S = sibilant, L = liquid); the analysis is limited to word-initial heads and word-final codas. In the first column of Table 2, the five “stress-based” dialects (the ones that occupy the rightmost side of Figure 1) are divided into three groups, which in turn are ranked according to the increasing markedness of sonority relations in codas and heads. In the first row we have Milanese and Feltrino, which – as shown above – turn out to be less marked also in terms of numerical syllable complexity (28 and 29 syllable types, respectively). Instead, we observe that, in terms of sonority relations, Friulian is less marked than Turinese and Romagnolo (both 35 syllable types), even if it is the dialect with the greatest numerical complexity (40 syllable types).



Syllable typology and the rhythm class hypothesis  

Table 2: Consonant clusters in stress-based Italo-Romance dialects Word-initial clusters (heads)

 435

4

Word-final clusters (codas) CC

Obs+/j/

trop ‘too much’ strak ‘tired’ pian ‘plain’

Obs+/l/ S+Obs+/l/

glesie ‘church’ sclop ‘burst’ (n.)

CC Son+Obs CCS Son+Obs+S

fuart ‘strong’ timps ‘times’

Plo+/l/ Fri+/l/ Plo+N Fri+N Plo+Fri Plo+S Plo+Plo Plo+Aff Fri+Fri Fri+Plos

tlè ‘loom’ slè ‘saddler’ pnel ‘brush’ (n.) fnocc ‘fennel’ dvent ‘become’ bsogn ‘need’ (n.) pkè ‘sin’ pcit* ‘small’ vsen ‘near’ vdend ‘seeing’

CC

Son+Obs S+Plo L+N

camp* ‘field’ rest ‘rest’ (n.) calm* ‘quiet’

Plo+Plo

salvätk ‘savage’

CCC

S+N+N S+Plo+Plo Plo+S+Plo Plo+Aff+N

smnussè ‘cut’ sbdel ‘hospital’ pskè ‘fish’ (verb) pznen ‘child’

N+Plo+S

ends ‘decoy’

CCCC

Fri+S+Plo+L Plo+S+Plo+L

vspre ‘wasp’s nest’ pstren ‘mill’

Milanese (S)CC and Feltrino

Obs+/r/

Friulian

CC SCC

Romagnolo CC or Turinese4

Son+Obs S+Plo L+N

grant ‘big’ nost ‘our’ forn ‘oven’

Starting our examination with the third column of Table 2, i.e. the role of word-final codas, we have already highlighted the role of diachronic apocope processes. The loss of unstressed final vowels (except -a) in the transition from Vulgar Latin to northern Romance initiated a typological shift (cf. Loporcaro 2011: 67–69) with major consequences on the rhythm of these dialects. The importance of this process lies not only in the mere existence of word-final codas containing up to two consonants (and in the corresponding increment of syllable types), but also in the transformation of disyllabic content words into monosyllabic words which display a relatively high frequency in the lexicon (cf. the Italian words tempo ‘time’ and forte ‘strong’ with the corresponding forms temp and fort, quite common in the dialects of northern Italy). Now, even if apocope has dramatically altered the phonotactic and prosodic shapes of words in most northern dialects

4 Most of the examples are from Romagnolo; the three Turinese words are marked with *.

436 

 Stephan Schmid

(with the exception of Venetian and Ligurian), we observe that stress-based dialects exhibit more or less the same coda structures. Regarding the number of possible syllable types, a striking feature of Friulian is the presence of word-final CCC codas. The rightmost position in the cluster (the syllable offset) is basically restricted to a sibilant with a clear morphological function, i.e. the plural marking of masculine nouns and adjectives, as in timp-s ‘time-pl’. Otherwise, Friulian has the same biconsonantal codas as Feltrino and Milanese: These sequences may consist either of a sonorant (or a sibilant) and a voiceless obstruent (e.g. grant ‘big’, nost ‘our’), or a liquid and a nasal (e.g. forn ‘oven’, calm ‘quiet’). So far, the word-final codas of Feltrino, Milanese, and Friulian are ordered in terms of sonority sequencing along the preferences predicted by Vennemann’s (1988: 21) “coda law”. Note that sonority sequencing is also fostered by a phonotactic constraint which bans voiced obstruents from the word-final position; for instance, final devoicing appears in the masculine form of grant ‘big’ (as opposed to the feminine form granda). Basically, the same holds true also for Turinese and Romagnolo, which numerically have a slightly more complex syllable structure than Feltrino and Milanese, but depart in two respects from them. The first difference is the marginal occurrence of CCC codas, which diachronically result from of the posttonic syncope of unstressed vowels (compare ends ‘decoy’ with It. èndice). Diachronic syncope is also responsible for the second difference, a phonotactic pattern of Romagnolo which increases its markedness in terms of sonority relations: Indeed, the adjective salvätk ‘savage’ (< salvatĭcu(m); source: Rohlfs 1949: 227) exhibits a sequence of two stops, i.e. a sonority plateau. The occurrence of marked sonority relations – a typical feature of Romagnolo and Turinese – is even more pervasive in the structure of syllable heads. In fact, if we now turn to the word-initial consonant clusters illustrated in the second column of Table 2, the diversity among the three groups of stress-based dialects (Feltrino and Milanese, Friulian, Romagnolo and Turinese) seems to be much greater. As we can see in the first row, Feltrino and Milanese basically show the general Italo-Romance pattern illustrated in 3.2, shared by Standard Italian and the five syllable-based dialects on the left in Figure 1: In this pattern, an obstruent (or a sibilant+obstruent cluster) is followed by /r/ or /j/. However, the combination of an obstruent+lateral is banned from the core lexicon, since the Latin clusters cl- and pl- have undergone a weakening process which transformed the alveolar lateral in the approximant /j/ (e.g. clave(m) ‘key’ > chiave, planu(m) ‘plain’ > piano). This process reduces the markedness of the syllable according to parameter b) of the “head law” stated by Vennemann (1988: 13–14), in that it enhances the sonority contrast between the two components of the syllable head. Now, the second row of Table 2 reveals that Friulian permits clusters of an obstru-



Syllable typology and the rhythm class hypothesis  

 437

ent+lateral (note that in any Italo-Romance the existence of an Obstruent+/l/ cluster implies the existence of an Obstruent+/r/ cluster, but not vice versa); from a diachronic point of view, Friulian turns out to be more conservative in this respect than Standard Italian. cl- and pl- clusters inherited by Friulian from Latin are slightly more marked compared to the clusters with an approximant, but they still adhere to the sonority sequencing principle. Conversely, in Romagnolo and in Turinese sonority relations within syllable heads are much less constrained. As shown in the second column of Table 2, a coronal lateral may also be preceded by a homorganic obstruent, as in tlè ‘loom’ or slè ‘saddler’. Such clusters are not inherited from Latin, but have arisen from pretonic vowel deletion (cf. the corresponding Italian words telaio and sellaio). These dialects also allow CC sequences composed of an obstruent and a nasal (e.g. pnel ‘brush’) or a fricative (e.g. dvent ‘become.pst.part’); again, such forms are the result of diachronic syncope, as becomes evident from the corresponding Italian forms which preserve the unstressed pretonic vowel (domani, diventato). So far, the clusters are not optimal, but they do conform to the commonly accepted “sonority scale”. However, we also find sequences with two voiced fricatives (vsen ‘neighbor’) or two unvoiced plosives (pkè ‘sin’), the least sonorous segments of all; the Italian cognates vicino and peccato reveal the pretonic syncope that gave rise to these typologically rather marked sonority plateaus.5 Pretonic syncope is also responsible for the still more marked word-initial CCC and CCCC clusters; in sbdel ‘hospital’ (It. ospedale) and pznen ‘child’ (It. piccinino), even two pretonic vowels have been dropped. In several examples, for instance in pskè ‘to fish’, a sonority reversal occurs.6 What becomes clear, then, is that well-formed syllables play a minor role in the phonology of Romagnolo and Turinese. Diachronically, both pretonic and posttonic vowels have been dropped, assigning more prominence to the accented syllable, which in many cases turns out to be the only syllable of the word: Thus, Romagnolo and Turinese may be regarded as word languages.

5 Diachronically, some cases of word-internal posttonic syncope occurred in Vulgar Latin, as is shown by Italian word forms such as verde ‘green’ < viride(m); to a large extent, the phonotactic markedness of Romagnolo and Turinese is due to pretonic syncope. 6 As is well known, syllabification becomes more difficult in increasingly complex consonant clusters. The uncertainty of syllable counting and of placing syllable boundaries has repeatedly been reported as a typical feature of stress and/or word languages. According to theoretical preferences, one could adopt a different phonotactic analysis, e.g. by positing the existence of underlying “empty nuclei” or assigning the status of nuclei to sibilants on the basis of their acoustic prominence. However, it is not the purpose of this paper to discuss the theoretical status of the syllable and the formal nature of syllable boundaries.

438 

 Stephan Schmid

Summarizing the syllabic typology sketched so far, the two basic syllable-related para­meters furnished by the framework of syllable and word languages – numerical complexity and sonority relations within phonotactic constituents – have proven very fruitful for a phonological typology of Italo-Romance dialects. Numerical complexity (or the number of possible syllable types) seems to be mainly determined by coda constraints, which in turn are the result of diachronic apocope; moreover, coda complexity is a major criterion that enables the assignment of a single dialect to either the syllable-based or the stress-based sphere (in our case, the division line falls roughly between Bitontino and Milanese). The adherence to universal sonority preferences allows us to trace a subdivision within the stress-based dialects; three of them (Milanese, Feltrino, and Friulian) basically show unmarked sonority relations (except the word-initial SC-clusters which are generally accepted in Italo-Romance and rather widespread cross-linguistically), whereas Romagnolo and Turinese allow consonant clusters that are heavily marked. Finally, if the markedness of word-final codas is overall a matter of numerical complexity, the markedness of word-initial heads is also determined by sonority plateaus and reversals.

3.4 Some remarks on vowel reduction The primary focus of this contribution is on the relationship between phonotactic complexity and rhythm metrics. Nevertheless, as these two research paradigms – i.e. the phonological typology of syllable and word languages (Auer 1993) and the rhythm class hypothesis (e.g. Ramus, Nespor, and Mehler 1999) – explicitly refer to vowel reduction as a basic parameter for phonological typology (cf. 1), a few descriptive observations on the investigated Italo-Romance dialects are in order. Table 3 compares the ten analysed Italo-Romance dialects with regard to five typologically relevant parameters of vowel systems: i) the number of vowel qualities in stressed syllables, ii) the number of vowel qualities in unstressed syllables, iii) the occurrence of schwa in unstressed syllables, iv) the possibility of schwa dropping in prepausal unstressed syllables, and v) the occurrence of distinctive vowel length; the synopsis is drawn from the typological survey given in Schmid (1999b and references therein).7 In terms of Maddieson’s (2005a) complexity scale

7 It is not easy to find reliable sources on unstressed vowels in all dialects, since many structuralist descriptions only enumerate the stressed vowel phonemes. Also, Mayerthaler’s (1996: 208) treatment of the unstressed vowels oversimplifies dialectal diversity: For instance, she claims [u ə] to be the only unstressed vowels in the “area centro-meridionale”, whereas a closer look at the



 439

Syllable typology and the rhythm class hypothesis  

for vowel systems, none of the ten dialects has a “small vowel inventory” (2-4 vowel qualities) and only Sicilian has an “average vowel inventory” (5-6 vowel qualities); all the other dialects belong to the class with a “large vowel inventory” (7-14 vowel qualities), but note that six out of nine dialects present the smallest number within this category. In fact, the seven accented vowels /i e ɛ a o ɔ u/ are characteristic of Italo-Romance: They appear not only in Standard Italian (cf. 2.1) and in the closely related Tuscan dialect of Pisa, but also in the two Veneto dialects of Venice and Feltre as well as in the neighboring Friulian; moreover, the same stressed vowel system is found in the two southern dialects of Naples and Bitonto. Table 3: Typological traits of ten Italo-Romance vowel systems Stressed vowel qualities

Unstressed vowel qualities

Unstressed schwa

Prepausal apocope

Distinctive vowel quantity

Pisan

7

5

–

–

–

Neapolitan

7

4

+

+

–

Sicilian

5

3

–

–

–

Venetian

7

5

–

–

–

Bitontino

7

2

+

+

–

Milanese

8

4

–

–

+

Feltrino

7

5

–

–

–

Turinese

9

6

+

–

–

Romagnolo

9

5

+

–

+

Friulian

7

5

–

–

+

As mentioned in section 1, in prosodic typology there seems to be a widely shared agreement that vowel reduction processes contribute substantially to the establishment of speech rhythm and rhythm classes. Now, the very term “vowel reduction” may have different meanings depending on whether it refers to i) the paradigmatic difference in the number of phonemes between the stressed and the unstressed vowel system, ii) the articulatory laxing and spectral centralization of unstressed vowels, iii) the temporal shortening of unstressed vowels, or iv) the (diachronic or synchronic) dropping of unstressed vowels. Some of these

two varieties considered in our sample reveals [a ə] for Bitontino (Valente 1975: 15) and [i ə a u] for Neapolitan (Maturi and Schmid 1999: 1393).

440 

 Stephan Schmid

processes – e.g. centralization and shortening – may go hand in hand and, diachronically, these two are likely to precede the complete dropping of unstressed vowels. The main focus of prosodic typology seems to be on the differences between the stressed and unstressed vowel systems within particular languages, but one could also posit the number of vowel qualities to be a typological parameter that differentiates between rhythm classes. As Szczepaniak (2007: 35, 52) points out, if word languages are supposed to enhance the phonemic and phonetic contrasts in the stressed and unstressed vowel systems, this means that, in general, word languages also are supposed to have a more complex stressed vowel system, i.e. one with a higher number of phonemes. At a microscopic level, this prediction is borne out by the ten dialects illustrated in Table 3, where a slightly higher number of stressed vowels is found in the word- or stress-based Gallo-Italian dialects (Milanese, Turinese, Romagnolo). Note, however, that this finding is in line with recent insights from phonological typology: The statistical analysis of large language samples points to a correlation between vowel inventory size and syllable complexity in that languages with few vowels are likely to allow only simple syllable types and languages with a larger vowel inventory tend to have more complex syllable structures (Maddieson 2011: 30). Now turning to the difference between the stressed and unstressed vowel systems, all dialects in Table 3 show a reduction in the number of phonemes (against the prototypical model of a syllable language like Spanish, which has the same vowel set in accented and non-accented syllables; cf. Auer 1993: 67). In the most frequent Italo-Romance vowel pattern, four dialects show only a weak reduction from 7 to 5 vowel qualities (Pisan, Venetian, Feltrino, Friulian), none of which is centralized. Instead, two southern dialects reduce more dramatically from 7 to 4 vowels (as in the case of Neapolitan) or even to 2 vowels (Bitontino); among the unstressed vowels of these two dialects we find schwa, in line with an often observed cross-linguistic pattern – the smaller the set of unstressed vowels, the more probable the appearance of the prototypically “reduced” vowel – and also in line with with a central claim of the typology of word and syllable languages (Auer 1993: 66). Not surprisingly, schwa also belongs to the unstressed vowel systems of the two dialects with underlyingly marked phonotactic structures: Turinese and Romagnolo. It is true that these two dialects have a relatively high number of unstressed vowels, but they also show a noticeable reduction of the vowel qualities (from 9 to 6 and 5, respectively). Among the southern dialects, we find two different paths of vowel reduction, i.e. centralization and peripheralization: Bitontino, the least syllable-based dialect, has only central unstressed vowels [a ə], whereas Neapolitan reduces both towards the center [ə] and towards the periphery [i u]; Sicilian, the most syllable-based southern dialect, only shows



Syllable typology and the rhythm class hypothesis  

 441

the peripheralization pattern, reducing the five stressed vowels /i ɛ a ɔ u/ to the three unstressed vowels /i a u/ (Rohlfs 1949: 214). Note that Neapolitan and Bitontino also share the last step of vowel reduction in the form of a postlexical rule, by which schwa undergoes apocope in prepausal contexts (cf. Schmid 2000: 349; Russo and Barry 2010: 208). For instance, the Neapolitan word gatto ‘cat’ appears both with and without a final schwa depending on the prosodic context (cf. [o ˈɣatːə sta ˈlːa] ‘the cat is there’ vs. ‘the cat’ ([o ˈɣatː]). Postlexical vowel reduction processes of this kind may thus dramatically alter the rhythmic pattern of a particular dialect: Neapolitan and Bitontino are located on the syllable-based side or at the center of the phonotactic continuum in Figure 1 (based on citation forms of words), but when it comes to the phonosyntactic implementation of such words in actually uttered sentences, these dialects are likely to present heavy consonant clusters.8 The last typological parameter highlighted in Table 3 is distinctive vowel quantity, which – at least within the Italo-Romance area – appears to be typical of the rather word-based varieties. As is well known, the distinctive vowel quantity of Latin collapsed in the transition to early Romance (cf. Auer 1993: 25), due to the generalized allophonic lengthening of stressed vowels in open syllables; the rise of a new distinction of vowel quantity in some varieties of northern Italo-Romance is driven by forces of rhythmic compensation and strictly related to “syllable-deteriorating” processes such as degemination and apocope (see Loporcaro 2011). Finally, from the point of view of the typology of syllable and word languages, it is important to note that in Italo-Romance dialects phonemically long vowels only occur in accented syllables, a feature which is predicted for word rhythm, but not for syllable rhythm (Auer 1993: 67).

4 Applying rhythm metrics to Italo-Romance dialects: A pilot study As we have seen in 2.1, a number of scholars have calculated different rhythm metrics for regional varieties of Standard Italian, but to our knowledge only few studies have applied the same method to Italo-Romance dialects. This is some-

8 It is interesting to note that in Neapolitan and Bitontino, centralization and apocope of unstressed vowels represent two ordered rules of synchronic variation, but the same phonological processes must have occurred consecutively – as instances of diachronic sound change – in northern Italy as well, as seen in the existence of centralized vowels in isolated dialects (cf. Loporcaro et al. 2007).

442 

 Stephan Schmid

what surprising, since we expect to find a much greater rhythmical diversity between dialects than between regional varieties: The latter share the same phonotactics, whereas the former vary considerably with regard to syllable structure constraints, as we have seen in the preceding section. Thus, according to the rhythm class hypothesis, as originally stated in Ramus, Nespor, and Mehler (1999), Italo-Romance dialects should display a substantial variation with regard to metrics such as %V and ∆C. Some acoustic evidence already exists, however. In a first study on the dialects of Ischia and Pozzuoli (from the Neapolitan area), Russo and Barry (2004) illustrated by means of spectrograms how diphthongization and schwa elision contribute to create the effect of a rather stress-timed rhythm, despite the alleged syllable-timed nature of Neapolitan that one can postulate on the grounds of its relatively unmarked syllable structure (cf. 2.2 and 3.2, but also 3.4). For ten fluent sections of speech from the Ischia dialect, Russo and Barry (2008: 319–320) calculated “Pairwise Variability Indices” (cf. 1) and found their dialect speaker plotted with stress-timed languages like German and English rather than with varieties of Italian. Recently, the same authors have extended their analysis to six Campanian speakers (two each from Capri, Ischia, and Pozzuoli), for each of whom they calculated PVI values out of ten inter-pause stretches (Russo and Barry 2010: 214–219): Again, five of the six speakers yielded PVI values similar to those of German and English, but for the Capri speakers both the vocalic and the intervocalic PVIs varied noticeably depending on the single utterances (ultimately, on the phonotactics of the lexical items in the ten utterances). Regarding northern Italy, Romano, Mairano, and Pollifrone (2010) applied four types of rhythm metrics (∆C/∆V, Varco, PVI, CCI) to six Piedmontese and Ligurian dialects; their findings confirm the alleged stress rhythm of Piedmontese and the syllable rhythm of Ligurian. Evidence for the stress-based nature of Piedmontese had already been found in the first study that applied the metrics of Ramus, Nespor, and Mehler (1999) to data from six Italo-Romance dialects. Schmid (2004) calculated %V, ∆C, and ∆V values for Turinese, Milanese, Bitontino, Neapolitan, Venetian, and Pisan. The present study enlarges the preceding analysis by adding data from other three varieties (i.e. Friulian, Feltrino, and Sicilian) and by calculating Varco and PVI metrics as well. As the reader will have noticed, these nine dialects have all been subjected to the phonotactic analysis in section 3; for this reason, an acoustic study appears to be particularly promising for the rhythm class hypothesis.



Syllable typology and the rhythm class hypothesis  

 443

4.1 Data and method It is more difficult to find acoustic data for dialects than for a standard language. In our case, the measurements were made with publicly available audio data that were not elicited for the purpose of rhythm analysis. Instead of read speech, the data consist of semi-spontaneous spoken language; in these “ethnotexts”, stories are told about historical facts or aspects of social life. Most of the recordings were gathered in the 1970s and 1980s within the project Profilo dei dialetti italiani, namely for Turinese (Berruto 1974), Venetian and Feltrino (Zamboni 1974), Bitontino (Valente 1975), Pisan (Giannelli 1976), Friulian (Frau 1984), and Milanese (Massariello Merzagora 1988). For Sicilian, the dialect from Enna was chosen among the recordings published by Ruffino (1997), whereas for Neapolitan unpublished oral history texts were used that were made available by the University of Naples (courtesy of Pietro Maturi). Unfortunately, no published acoustic data could be found for one of the most interesting dialects, Romagnolo. Table 4: Number of measurements, proportion of vocalic intervals (%V), standard deviation of vocalic and consonantal intervals over a sentence (∆V, ∆C), averaged by dialect, and the respective standard deviations (SD) Dialect

Vocalic intervals (N)

Consonantal intervals (N)

%V (SD)

∆V (SD)

∆C (SD)

Friulian

185

185

43.4 (3.8)

67.0 (20.6)

85.0 (20.0)

Turinese Milanese

138

141

47.2 (3.4)

55.0 (16.6)

48.0 (11.7)

143

143

48.4 (4.3)

65.3 (16.7)

57.0 (20.0)

Sicilian

207

198

49.7 (4.3)

42.0 (11.1)

41.0 (9.7)

Neapolitan

122

122

53.1 (7.2)

46.0 (10.4)

38.0 (9.4)

Feltrino

185

176

52.0 (3.4)

47.9 (15.0)

46.4 (10.0)

Bitontino

141

154

52.3 (3.5)

72.0 (20.7)

48.3 (9.4)

Venetian

194

191

53.8 (3.7)

42.0 (11.6)

34.0 (5.5)

Pisan

131

125

60.3 (6.1)

62.0 (22.1)

33.0 (5.8)

The analogue recordings were converted into .wav-files and manually segmented in order to obtain duration measurements of vocalic and ­consonantal intervals, according to the guidelines provided by Ramus, Nespor, and Mehler (1999: 271– 272); for instance, prevocalic glides were treated as consonants, whereas postvocalic glides were treated as vowels. For each of the nine dialects, ten utterances

444 

 Stephan Schmid

of different length were analysed in order to obtain the corpus illustrated in Table 4.9

4.2 Italo-Romance dialects according to the metrics of Ramus, Nespor, and Mehler (1999) Figure 2 shows the Italo-Romance dialects in the most classical of all rhythm topologies, i.e. the %V/∆C plane (allegedly syllable-based dialects are designed as circles, whereas allegedly stress- or word-based dialects appear as diamonds; for reasons to be explained below, Bitontino is characterized as stress-based). According to the predictions of Ramus, Nespor, and Mehler (1999), stress-based languages are expected to be situated in the upper left-hand area (given the higher variability of consonant clusters and the higher number of closed syllables), whereas syllable-based languages would tend to be collocated in the center (given the lower variability of consonant clusters and the higher number of open syllables in the lexicon).

Figure 2: Italo-Romance dialects in the %V/∆C plane

9 The presentation of the data follows the model of Ramus, Nespor, and Mehler (1999: 272). Dialects are ordered on the basis of their growing %V values, but the ∆V and ∆C values are not multiplied by 100.



Syllable typology and the rhythm class hypothesis  

 445

A comparison between Figures 1 and 2 does indeed reveal some interesting coincidences. With regard to ∆C, Friulian appears to be an outlier, a fact that could at least partially be due to the high numerical complexity of its syllable shell. Moreover, Milanese, Turinese, and Feltrino are also collocated in the upper area, in line with their stress-based tendency in the syllable type inventory continuum (cf. Figure 1). Quite interestingly, the most complex among the syllable-based dialects, Bitontino, occupies a very similar position to that of stress-based Feltrino in the %V/∆C plane. With regard to %V, these two intermediate dialects tend more towards the rightward sphere, where we find the syllable-based dialects (Neapolitan, Venetian, and Pisan). The y-axis of Figure 3 adds another element to the rhythm in the nine dialects, i.e. the durational variability of vocalic intervals; as pointed out in section 1, ∆V is supposed to be an acoustic correlate of the reduction of unstressed vowels (as well as of the lengthening of stressed vowels).

Figure 3: Italo-Romance dialects in the %V/∆V plane

Again, in line with the predictions of Ramus, Nespor, and Mehler (1999) we find stress-based dialects in the upper left-hand area and syllable-based languages towards the center of Figure 3. In the %V/∆V plane, however, Friulian remains much closer to Milanese and Turinese, and the similar vocalic variability of these three varieties seems to testify in favor of their alleged stress-timed character (besides, as pointed out in 3.4, Friulian and Milanese have distinctive vowel length in stressed syllables). Instead, the two intermediate dialects clearly stay apart: With regard to ∆V, Feltrino behaves like the syllable-based dialects (Sicilian, Neapolitan, Venetian), whereas Bitontino goes even beyond Friulian and

446 

 Stephan Schmid

Milanese. This can partly be explained by the many diphthongs that characterize syllable nuclei in Bitontino (cf. Schmid 2000: 347), but a postlexical process of vowel lengthening may also contribute to this result; remember that Trumper, Romito, and Maddalon (1991) classified Apulian as stress-timed precisely on the grounds of their vowel duration measurements.

4.3 Rhythm and speech rate in Italo-Romance dialects: Varco C One temporal factor of speech production that has been shown to interfere with rhythm metrics is speech rate. Accordingly, using the variation coefficient rather than the standard deviation permits us to control this factor and even yields a clearer picture which enhances the differentiation between rhythm classes (Dellwo 2006). Figure 4 presents the same data as Figure 2, but it substitutes ∆C with Varco C on the y-axis.

Figure 4: Italo-Romance dialects in the %V/Varco-C plane

The first effect achieved through the substitution of ∆C with Varco C concerns Friulian, which loses its outlier position and approaches the sphere of Milanese and Turinese. In other words, Figure 4 is supposed to represent the rhythm of Friulian in a more appropriate way: It still turns out to be the variety with the highest durational variability of consonantal intervals (as was to be expected on the grounds of its coda complexity), but the extreme position in Figure 2 is clearly due to the very low speech rate characteristic of this particular speaker. Using the variation coefficient instead of the standard deviation for consonantal intervals has some other interesting repercussions on the configuration



Syllable typology and the rhythm class hypothesis  

 447

of the Italo-Romance dialect continuum. Comparing Figures 2 and 4, we observe that the Varco metrics makes two other dialects move upwards on the C axis: As in Figure 3, Bitontino again excedes even Milanese and Turinese, thus assuming the character of a stress-based dialect with respect to consonantal variability as well. However, another central-southern dialect also passes from the syllable-based towards the stress-based area: Neapolitan. Indeed, the acoustic analysis of the speech signal reveals that, despite its relatively simple underlying syllable structure (cf. Figure 2), Neapolitan also exhibits some traits of stress-timing, such as the centralization/shorte­ning of word-final unstressed vowels and the apocope of word-final unstressed vowels in prepausal position (cf. 3.4). Further evidence for this hypothesis comes from the last metrics we will calculate for the Italo-Romance dialects: the PVI.

4.4 Italo-Romance dialects in the light of the Pairwise Variability Index (PVI) As already mentioned in section 1, the approach proposed by Grabe and Low (2002) excludes %V as a measure of speech rhythm. Moreover, in opposition to the metrics applied by Ramus, Nespor, and Mehler (1999), the PVI model does not rely on the standard deviations of all vocalic and consonantal intervals by calculating their mean duration in a number of utterances, but rather focuses on the mean differences between successive pairs of vocalic and intervocalic (= consonantal) intervals. The most widespread version of this model uses a normalized PVI (nPVI) for vocalic intervals and a raw PVI (rPVI) for intervocalic intervals (see Grabe and Low 2002 for technical details); vocalic nPVIs are normally represented on the y-axis of the rhythm graph, consonantal rPVIs on the x-axis. For a PVI-like topology the rhythm class hypothesis predicts that syllable-based languages should tend towards the origo of the Cartesian diagram (i.e. the lower left-hand area of the bidimensional plane), whereas stress-based languages would be collocated in the upper right-hand part of the rhythm area. However, the analyses provided so far indicate that rhythm classes are much better discriminated by the vocalic PVI than by the consonantal PVI. For instance, according to the measurements of Grabe and Low (2002: 530), Spanish and German appear to have roughly the same variability of consonantal intervals. Figure 5 shows the results of the PVI metrics applied to our Italo-Romance data (with inverted axes as compared to the standard representation). Regarding the durational variability of the consonantal intervals, the non-normalized scale of the y-axis again leaves Friulian in an outlier-like position, due to the low speech rate of the recorded subject. As a general tendency, the diamonds of the

448 

 Stephan Schmid

Figure 5: Italo-Romance dialects in the PVI plane

allegedly stress-based dialects stay in the upper part of the plane, but Turinese displays a lower PVI-C value than Neapolitan and Sicilian. It seems that, for the Italo-Romance data as well, the normalized vocalic PVI allows for a differentiation which fits better with the expectations derived from a phonotactic analysis (cf. 3.2 and 3.3): The diamonds of the allegedly stress-based dialects tend towards the right-hand part of the plane (the rather stress-based character of Bitontino now being taken for granted), whereas the circles of the three typically syllable-based dialects (Pisan, Venetian, Sicilian) are dispersed towards the left-hand area of the diagram. However, one syllable-based dialect has moved right into the stress-based sphere, due to its high vocalic PVI: Neapolitan, which has already revealed the same tendency on the basis of Varco C (cf. Figure 4). All in all, it is reasonable to interpret the similar position of Bitontino and Neapolitan in Figures 4 and 5 as a consequence of vowel reduction processes rather than as a reflection of their underlying syllable structure (cf. 3.4).

5 Conclusion Summarizing our discussion of the Italo-Romance data, we may conclude that both the phonological and phonetic analyses have provided valuable hints for a prototypical classification of these dialects. With regard to numerical syllable complexity, the Italo-Romance dialects can indeed be ranked along a continuum which comprises a pole with a relatively simple phonotactics towards a pole



Syllable typology and the rhythm class hypothesis  

 449

where complex consonant clusters are allowed. Rather than a neat division into two classes, however, we should conceive of this typology in terms of two opposite tendencies, given the very gradual increase in number of syllable types in the dialect continuum. Moreover, the Italo-Romance dialects have nicely shown that phonotactic markedness is not only a matter of numerical complexity, but that sonority relations within syllable constituents do play a major role. In fact, among the dialects oriented towards the stress-based pole, three subgroups can be individuated: i) Milanese and Feltrino show the same segment ordering as the syllable-based dialects; ii) Friulian allows some more marked consonant clusters, but still obeys the sonority sequencing principle; iii) Romagnolo and Turinese heavily violate such universal preferences by allowing marked patterns, particularly sonority plateaus. In the case of the Romance languages, a synchronic analysis of phonotactic constraints can easily be combined with a diachronic study of phonological processes. Indeed, the reasoning outlined in this contribution is not only complementary to the research of Mayerthaler (1996), but it is also compatible with its fundamental claims. For many dialects, the assignment of single dialects to one or the another type is rather similar, but there is one general issue where our data do not confirm her classification: There appears to be no correlation between the typological continuum and the geolinguistic distribution of the dialects (in the sense that we would gradually proceed from the syllable-based south towards the stress-based north). Within one and the same region, Veneto, we find two dialects with opposite tendencies, i.e. the clearly syllable-based Venetian and the rather stress-based Feltrino. Actually, Venetian is surrounded by dialects with a marked syllable structure, since it borders on both Friulian and Romagnolo. On the other hand, there is a sharp contrast between the stress-based Romagnolo and the syllable-based neighboring Tuscan (in our case: Pisan). Moving further south, we find the intermediate central-southern dialects (Neapolitan and Bitontino), but in the extreme south again a rather syllable-based type appears (Sicilian). Interesting insights have also been gained from the acoustic analysis. The “rhythm class hypothesis” proves to be a promising research paradigm for the Italo-Romance dialects, even if the research presented in this contribution still remains somewhat approximative. Among the methodological problems to be considered are the differences between read and spontaneous speech (the latter, as in our case, often showing much more vocalic variability due to hesitation phenomena), and, more importantly, the possibility that speech rhythm might to some extent also be idiosyncratic (cf. Russo and Barry 2010: 215); therefore, more data – particularly recordings from more than one speaker – are needed for every dialect.

450 

 Stephan Schmid

Still, the results of the acoustic analysis are in line with the basic claims of the rhythm class hypothesis: The dialects tend to group together in certain areas of the various rhythm topologies, and these topological groupings are in accordance with the predictions we were able to formulate on the grounds of their phonotactic patterns. For instance, the %V/∆V plane (Figure 3) plots the four tendentially stressed-based dialects (Friulian, Turinese, Romagnolo, Bitontino) in a similar sphere as the %V/Varco C plane (Figure 4); the case of Friulian proves that Varco C discriminates between rhythm classes better than ∆C. Finally, it appears that the competing metrics grasp different aspects of speech rhythm: So the PVI highlights some stress-based characteristics of Neapolitan which are related to vowel reduction processes, despite its rather simple syllable structure. This finding indicates that the interpretation of rhythm planes cannot be carried out in the light of phonotactic complexity alone, as has been done in the present contribution; rather, a complete phonological analysis should include allophonic rules that contribute to creating the rhythm of speech (e.g. the lengthening of stressed vowels in open syllables in Italian). Moreover, the phonotactic analysis should not only discover the underlying constraints on syllable structures, but also take into account the token frequency of particular syllable types (cf. Dauer 1983: 56; Auer and Uhmann 1988: 247); as shown by Russo and Barry (2010: 217) with data from a southern Italian dialect, the PVI values may vary considerably between different utterances according to the words they consist of. Finally, of particular relevance is, of course, the reduction of unstressed vowels – a crucial element which is explicitly mentioned both in the rhythm class hypothesis (Ramus, Nespor, and Mehler 1999: 267) and in the typology of syllable and word languages (Auer 1993: 6). To conclude, a comparative analysis of syllable structure constraints and rhythm metrics in Italo-Romance dialects shows a rather solid coincidence in the basic findings, which manifests itself in the ranking of the single dialects on a typological continuum of increasing phonotactic complexity as well as in their collocation in the various rhythm topologies. From a more theoretical point of view, the present contribution proves that the rhythm class hypothesis and the phonological typology of syllable and word languages have many points in common, such as the view of typology as a continuum with intermediate languages and the importance attributed to syllable structure as well as to vowel reduction and lengthening processes. Communalities, but also divergences, appear in terminological practice: It is true that labels like “stress-timed” vs. “syllable-timed” (or “isosyllabic” vs. “isoaccentual”) tend to be generally avoided, as the original isochrony hypothesis is no longer defendable. However, the terms “syllable-based” vs. “stress-based” and the very notion of “rhythm class” refer to a view of prosody as a temporal and prominence-related structure; in the dichot-



Syllable typology and the rhythm class hypothesis  

 451

omy of syllable and word languages, quite differently, rhythm itself plays a minor role, as the basic typological cue resides in “the selection of the prosodic category which is central for the make-up of a language’s phonology” (Auer 1993: 95).

References Abercrombie, David (1967): Elements of General Phonetics. Edinburgh: Edinburgh University Press. Altamura, Antonio (1968): Dizionario dialettale napoletano. 2nd ed. Napoli: Fiorentino. Arrighi, Cletto (1896): Dizionario milanese-italiano. Milano: Hoepli. Auer, Peter (1993): Is a rhythm-based typology possible? A study of the role of prosody in phonological typology. (KontRI Working Paper 21.) Universität Konstanz. Available at http://paul.igl.uni-freiburg.de/auer/userfiles/downloads/Phonotypo_Kontri1.pdf. Auer, Peter (2001): Silben- und akzentzählende Sprachen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals, 1391–1399. (Handbücher zur Sprach- und Kommunikationswissenschaft / Handbooks of Linguistics and Communication Science 20/2.) Berlin/New York: Walter de Gruyter. Auer, Peter and Susanne Uhmann (1988): Silben- und akzentzählende Sprachen. Literaturüberblick und Diskussion. Zeitschrift für Sprachwissenschaft 7/2: 214–259. Barry, William J. (2010): Rhythm measures in retrospect. Reflections on the nature of spokenlanguage rhythm. In: Stephan Schmid, Michael Schwarzenbach and Dieter Studer (eds.), La dimensione temporale del parlato, 3–12. Torriana: EDK. Barry, William J., Bistra Andreeva, Michela Russo, Snezhina Dimitrova and Tanja Kostadinova (2003): Do rhythm measures tell us anything about language type? In: Maria-José Solé, Daniel Recasens and Joaquín Romero (eds.), Proceedings of the 15th International Congress of Phonetic-Sciences, 2693–2696. Barcelona: Futurgraphic. Berruto, Gaetano (1974): Piemonte e Valle d’Aosta. (Profilo dei dialetti italiani 1.) Pisa: Pacini. Bertinetto, Pier Marco (1977): “Syllabic blood” ovvero l’italiano come lingua ad isocronismo sillabico. Studi di grammatica italiana 6: 69–96. Bertinetto, Pier Marco (1981): Strutture prosodiche dell’italiano. Firenze: Accademia della Crusca. Bertinetto, Pier Marco (1989): Reflections on the dichotomy ‘stress vs. syllable-timing’. Revue de Phonétique Appliquée 91/93: 99–130. Bertinetto, Pier Marco and Chiara Bertini (2008): On modeling the rhythm of natural languages. In: Plinio Barbosa, Sandra Madureira and César Reis (eds.), Proceedings of the 4th International Conference on Speech Prosody, 427–430. São Paulo: Capes. Boerio, Giuseppe (1856): Dizionario del dialetto veneziano. Venezia: Cecchini. Dauer, Rebecca (1983): Stress-timing and syllable-timing reanalyzed. Journal of Phonetics 11: 51–62. Dauer, Rebecca (1987): Phonetic and phonological components of language rhythm. In: Proceedings of the 11th International Congress of Phonetic Sciences, vol. 5, 447–450. Tallinn: Academy of Sciences of the Estonian S.S.R. Dellwo, Volker (2006): Rhythm and speech rate: a variation coefficient for DeltaC. In: Paweł Karnowski and Imre Szigeti (eds.), Language and Language Processing: Proceedings of the 38th Linguistics Colloquium, 231–241. Frankfurt: Lang.

452 

 Stephan Schmid

Dellwo, Volker and Petra Wagner (2003): Relations between language rhythm and speech rate. In: Maria-José Solé, Daniel Recasens and Joaquín Romero (eds.), Proceedings of the 15th International Congress of Phonetics Sciences. vol. 1, 471–474. Barcelona: Futurgraphic. Faggin, Giorgio (1985): Vocabolario della lingua friulana. Udine: Del Bianco. Frau, Giovanni (1984): Friuli. (Profilo dei dialetti italiani 6.) Pisa: Pacini. Gavuzzi, Giuseppe (1891): Vocabolario piemontese-italiano. Torino: Le Roux. Giannelli, Luciano (1976): Toscana. (Profilo dei dialetti italiani 9.) Pisa: Pacini. Giordano, Rosa and Leandro D’Anna (2010): A comparison of rhythm metrics in different speaking styles and in fifteen regional varieties of Italian. Speech Prosody 2010 100826. Available at http://speechprosody2010.illinois.edu/papers/100826.pdf. Grabe, Esther and Ee Ling Low (2002): Durational variability in speech and the Rhythm Class Hypothesis. In: Carlos Gussenhoven and Natasha Warner (eds.), Laboratory Phonology 7, 515–545. Berlin/New York: Mouton de Gruyter. Haiman, John and Paola Benincà (1992): The Rhaeto-Romance Languages. London/New York: Routledge. Hoequist, Charles J. (1983): Syllable duration in stress-, syllable- and mora-timed languages. Phonetica 40: 203–237. Kim, Jesus, Chris Davis and Ann Cutler (2008): Perceptual tests of rhythmic similarity: II. Syllable rhythm. Language and Speech 51: 343–359. Kohler, Klaus J. (2009): Rhythm in speech and language. A new research paradigm. Phonetica 66: 29–45. Loporcaro, Michele (2011): Syllable, segment and prosody. In: Martin Maiden, John Charles Smith and Adam Ledgeway (eds.), The Cambridge History of the Romance Languages. Vol. 1: Structures, 50–108. Cambridge: Cambridge University Press. Loporcaro, Michele, Rachele Delucchi, Nadia Nocchi, Tania Paciaroni and Stephan Schmid (2007): Schwa finali sull’Appennino emiliano: il vocalismo del dialetto di Piandelagotti. In: Veronica Giordani, Valentina Bruseghini and Piero Cosi (eds.), Scienze vocali e del linguaggio. Metodologie di valutazione e risorse linguistiche, 57–76. Torriana: EDK Editore. Maddieson, Ian (2005a): Vowel quality inventories. In: Martin Haspelmath, Matthew S. Dryer, David Gil and Bernard Comrie (eds.), The World Atlas of Language Structures, 14–17. Oxford: Oxford University Press. Maddieson, Ian (2005b) Syllable structure. In: Martin Haspelmath, Matthew S. Dryer, David Gil and Bernard Comrie (eds.), The World Atlas of Language Structures, 54–57. Oxford: Oxford University Press. Maddieson, Ian (2011): Phonological complexity in linguistic patterning. In: Wai-Sum Lee and Eric Zee (eds.), Proceedings of the 17th International Congress of Phonetics Sciences, vol. 1, 28–34. Hong Kong. Mairano, Paolo and Antonio Romano (2007): Inter-subject agreement in rhythm evaluation for four languages. In: Jürgen Trouvain and William J. Barry (eds.), Proceedings of the 16th International Congress of Phonetics Sciences, vol. 2, 1149–1152. Saarbrücken. Mairano, Paolo and Antonio Romano (2010): Un confronto tra diverse metriche ritmiche usando Correlatore. In: Stephan Schmid, Michael Schwarzenbach and Dieter Studer (eds.), La dimensione temporale del parlato, 79–100. Torriana: EDK. Malagoli, Giuseppe (1939): Vocabolario pisano. Firenze: Accademia della Crusca. Massariello Merzagora, Giovanna (1988): Lombardia. (Profilo dei dialetti italiani 3.) Pisa: Pacini.



Syllable typology and the rhythm class hypothesis  

 453

Maturi, Pietro and Stephan Schmid (1999): Phonetically conditioned allomorphy of functional words in a dialect of southern Italy. In: John J. Ohala, Yoko Hasegawa, Manjari Ohala, Daniel Granville and Ashlee C. Bailey (eds.), Proceedings of the 14th International Congress of Phonetics Sciences, vol. 2, 1393–1396. San Francisco. Mayerthaler, Eva (1996): Stress, syllables, and segments: their interplay in an Italian dialect continuum. In: Bernhard Hurch and Richard A. Rhodes (eds.), Natural Phonology: The State of the Art, 201–221. (Trends in Linguistics. Studies and Monographs 92.) Berlin/New York: Mouton de Gruyter. Migliorini, Bruno and Giovan Battista Pellegrini (1971): Dizionario del Feltrino rustico. Padova: Liviana. Morri, Antonio (1840): Vocabolario romagnolo-italiano. Faenza: Conti. Murty, Lalita, Takashi Otake and Ann Cutler (2007): Perceptual tests of rhythmic similarity: I. Mora Rhythm. Language and Speech 50: 77–99. Nübling, Damaris and Renate Schrambke (2004): Silben- versus akzentsprachliche Züge in germanischen Sprachen und im Alemannischen. In: Elvira Glaser, Peter Ott and Ruedi Schwarzenbach (eds.), Alemannisch im Sprachvergleich. Beiträge zur 14. Arbeitstagung für alemannische Dialektologie in Männedorf (Zürich) vom 16.–18.9.2002, 281–320. (Zeitschrift für Dialektologie und Linguistik ‒ Beihefte 129.) Stuttgart: Steiner. Pike, Kenneth (1945): The Intonation of American English. Ann Arbor: University of Michigan Press. Ramus, Franck, Marina Nespor and Jacques Mehler (1999): Correlates of linguistic rhythm in the speech signal. Cognition 73: 265–292. Ramus, Frank, Emmanuel Dupoux and Jacques Mehler (2003): The psychological reality of rhythm classes. In: Maria-José Solé, Daniel Recasens and Joaquín Romero (eds.), Proceedings of the 15th International Congress of Phonetics Sciences, vol. 1, 337–342. Barcelona: Futurgraphic. Roach, Peter (2003): A bibliography of timing and rhythm in speech. Available at http://www. personal.rdg.ac.uk/~llsroach/timing.pdf. Rogers, Derek and Luciana Arcangeli (2004): Italian. Journal of the International Phonetic Association 34: 117–121. Rohlfs, Gerhard (1949): Historische Grammatik der italienischen Sprache und ihrer Mundarten. I. Lautlehre. Bern: Francke. Romano, Antonio, Paolo Mairano and Barbara Pollifrone (2010): Variabilità ritmica di varietà dialettali del Piemonte. In: Stephan Schmid, Michael Schwarzenbach and Dieter Studer (eds.), La dimensione temporale del parlato, 101–112. Torriana: EDK. Ruffino, Giovanni (1997): I nomi della trottola e la memoria del gioco. Palermo: Centro di studi filologici e linguistici siciliani. Russo, Michela and William J. Barry (2004): Interaction between segmental structure and rhythm. A look at Italian dialects and regional Standard Italian. Folia Linguistica 38: 277–296. Russo, Michela and William J. Barry (2008): Measuring rhythm. A quantified analysis of southern Italian dialects’ stress time parameters. Language design, Special Issue 2: 315–322. Russo, Michela and William J. Barry (2010): Il Pairwise Variability Index (PVI e PVIs): valori ritmici per i dialetti italiani e per l’italiano regionale. Implicazioni tipologiche. In: Michela Russo (ed.), Prosodic Universals. Comparative Studies in Rhythmic Modeling and Rhythm Typology, 185–226. Roma: Aracne. Saracino, Giacomo (1957): Lessico dialettale bitontino. Bari: Scuola tipografica orfanotrofio salentino.

454 

 Stephan Schmid

Savy, Renata and Francesco Cutugno (1997): Hypospeech, vowel reduction, centralization: how do they interact in diaphasic variations? In: Bernard Caron (ed.), Actes du XVe Congrès International des Linguistes, Paper ID 0185. Oxford: Pergamon. Schmid, Stephan (1996): La struttura della sillaba nei dialetti veneti. In: Giovan Battista Pellegrini (ed.), Terza raccolta di saggi dialettologici in area italo-romanza, 125–166. Padova: Centro di Studio Dialettologia Italiana-CNR. Schmid, Stephan (1997): A typological view of syllable structure in some Italian dialects. In: Pier Marco Bertinetto, Livio Gaeta, Georgi Jetchov and David Michaels (eds.), Certamen Phonologicum III, 247–265. Torino: Rosenberg & Sellier. Schmid, Stephan (1998): Tipi sillabici nei dialetti dell’Italia settentrionale. In: Giovanni Ruffino (ed.), Atti del XXI Congresso Internazionale di Linguistica e Filologia Romanza. Vol. 5: Dialettologia, geolinguistica, sociolinguistica, 613–625. Tübingen, Niemeyer. Schmid, Stephan (1999a): Fonetica e fonologia dell’italiano. Turin: Paravia. Schmid, Stephan (1999b): Per un’analisi del vocalismo italo-romanzo in chiave tipologica. In: Paola Benincà, Alberto Mioni and Laura Vanelli (eds.), Fonologia e morfologia dell’italiano e dei dialetti d’Italia, 249–267. Roma: Bulzoni. Schmid, Stephan (2000): Tipi sillabici in alcuni dialetti centro-meridionali. In: Annick Englebert, Michel Pierrard, Laurence Rosier and Dan van Raemdonck (eds.), Actes du XXIIe Congrès international de linguistique et philologie romanes. Tome III: Vivacité et diversité de la variation linguistique, 343–350. Tübingen: Niemeyer. Schmid, Stephan (2001): Un nouveau fondement phonétique pour la typologie rhythmique des langues? Poster presented at the 10th anniversary of the Laboratoire d’analyse informatique de la parole. University of Lausanne. Schmid, Stephan (2004): Une approche phonétique de l’isochronie dans quelques dialectes italo-romans. In: Trudel Meisenburg and Maria Selig (eds.), Nouveaux départs en phonologie. Les conceptions sub- et suprasegmentales, 109–114. Tübingen: Narr. Szczepaniak, Renata (2007): Der phonologisch-typologische Wandel des Deutschen von einer Silben- zu einer Wortsprache. (Studia Linguistica Germanica 85.) Berlin/New York: De Gruyter. Traina, Antonino (1868): Nuovo vocabolario siciliano-italiano. Palermo: Pedone Lauriel. Trumper, John, Luciano Romito and Marta Maddalon (1991): Double consonants, isochrony and raddoppiamento fonosintattico: some reflections. In: Pier Marco Bertinetto, Michael Kenstowicz and Michele Loporcaro (eds.), Certamen Phonologicum II, 329–360. Torino: Rosenberg and Sellier. Valente, Vincenzo (1975): Puglia. (Profilo dei dialetti italiani 15.) Pisa: Pacini. Vayra, Mario, Cinzia Avesani and Carol Fowler (1984): Patterns of temporal compression in spoken Italian. In: Marcel P. R. van den Broecke and Anthony Cohen (eds.), Proceedings of the 10th International Congress of Phonetic Sciences, 541–546. Dordrecht: Foris Publications. Vennemann, Theo (1988): Preference Laws for Syllable Structure and the Explanation of Sound Change. With Special Reference to German, Germanic, Italian, and Latin. Berlin: Mouton de Gruyter. White, Laurence, Eleonor Payne and Sven Mattys (2009): Rhythmic and prosodic contrast in Venetan and Sicilian Italian. In: Marina Vigário, Sónia Frota and Maria João Freitas (eds.), Phonetics and Phonology: Interactions and Interrelations, 137–158. Amsterdam: Benjamins. Zamboni, Alberto (1974): Veneto. (Profilo dei dialetti italiani 5.) Pisa: Pacini.