The Emergence of Creole Syllable Structure: A Cross-linguistic Study 9783110339567, 9783110339314

Citation preview

Mareile Schramm The Emergence of Creole Syllable Structure

Linguistische Arbeiten

Edited by Klaus von Heusinger, Gereon Müller, Ingo Plag, Beatrice Primus, Elisabeth Stark and Richard Wiese

Volume 554

Mareile Schramm

The Emergence of Creole Syllable Structure A Cross-linguistic Study

DE GRUYTER

ISBN 978-3-11-033931-4 e-ISBN (PDF) 978-3-11-033956-7 e-ISBN (EPUB) 978-3-11-039530-3 ISSN 0344-6727 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published be the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2015 Walter de Gruyter GmbH, Berlin/Munich/Boston Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Acknowledgments This book has been a long time in the making and I would like to say a heartfelt thank-you to all those people who have, in one way or another, supported me in my studies. I am especially grateful to Ingo Plag, who introduced me to the study of creole languages and encouraged me to embark on this project. He could always be counted on to provide guidance when I threatened to get lost in the details. His constructive criticism, insightful comments and constant availability were much appreciated. Special thanks also go to Parth Bhatt for his permanent interest in my work. His open-minded approach to alternative theoretical analyses in general and his endless enthusiasm for creole syllables in particular have been a great source of inspiration. I very much enjoyed our talks, creole-related and otherwise. Various people have helped me improve my work by giving me feedback on individual analyses, papers presented at conferences or earlier versions of this book. One person I owe particular thanks to is Anne-Marie Brousseau, not only for reading and commenting on my analyses of the two French-based creoles, but also for sharing her knowledge of 17th and 18th century varieties of French. Thanks are due also to my colleagues and friends at Siegen and elsewhere. The many great colleagues I have had over the years have made work enjoyable even when the conditions kept getting worse. My friends outside of linguistics helped me by taking my mind off work from time to time and by reminding me that the world does not revolve (exclusively) around creole syllables. But most of all, I want to thank my parents for their unquestioning support of what must have seemed a very strange undertaking to them. Their constant belief in me was the greatest help possible.

Contents Acknowledgments | V Abbreviations and notational conventions | XI 1

Introduction | 1

2 2.1 2.2 2.3 2.4 2.5

Creole genesis and syllable structure | 4 Introduction | 4 Sources of creole structure and mechanisms in creolisation | 4 Creolisation and syllable structure | 5 Theories of syllable structure | 8 Summary | 13

3 3.1 3.2 3.3 3.3.1 3.3.2 3.3.3 3.3.4 3.3.5 3.3.6 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.5 3.5.1 3.5.2 3.5.3 3.6 3.6.1 3.6.2 3.6.3 3.6.4

Data and Methodology | 14 Introduction | 14 Selection of creole languages | 14 The creoles and their historical background | 17 Berbice Dutch | 18 Negerhollands | 19 Saramaccan | 19 St Kitts | 21 Guiana FC | 22 Trindad FC | 23 The creole corpora | 24 Source texts for the Dutch-based creoles | 24 Source texts for the English-based creoles | 25 Source texts for the French-based creoles | 26 The resulting corpora | 27 Identification of etyma | 29 The main lexifier varieties | 30 Cognate words and the choice of etymon | 31 Historical and variational aspects in etymon pronunciation | 32 Data coding and analytical procedure | 34 Coding of structural variables | 35 Coding of process variables | 38 Processes not considered in the analyses | 40 The CHAID analyses | 40

VIII | Contents 3.7 4 4.1 4.2 4.3 4.4 4.4.1 4.4.2 4.4.3 4.4.4 4.5 4.5.1 4.5.2 4.5.3 4.5.4 4.6 5 5.1 5.2 5.3 5.4 5.4.1 5.4.2 5.4.3 5.4.4 5.5 5.5.1 5.5.2 5.5.3 5.5.4 5.6 6 6.1 6.2

Summary | 43 Syllable structure and phonotactic restructuring in the Dutch-based creoles | 45 Introduction | 45 Previous studies | 45 Methodological issues | 47 Results I: Berbice Dutch | 52 Word-initial onsets in Berbice Dutch | 52 Word-final codas in Berbice Dutch | 65 Word-internal structures in Berbice Dutch | 78 Summary: Berbice Dutch | 86 Results II: Negerhollands | 86 Word-initial onsets in Negerhollands | 87 Word-final codas in Negerhollands | 96 Word-internal structures in Negerhollands | 107 Summary: Negerhollands | 113 Comparison: Berbice Dutch vs. Negerhollands | 114 Syllable structure and phonotactic restructuring in the English-based creoles | 116 Introduction | 116 Previous studies | 116 Methodological issues | 118 Results I: Early Saramaccan | 121 Word-initial onsets in Early Saramaccan | 122 Word-final codas in Early Saramaccan | 131 Word-internal structures in Early Saramaccan | 143 Summary: Early Saramaccan | 151 Results II: Early St Kitts | 152 Word-initial onsets in Early St Kitts | 152 Word-final codas in Early St Kitts | 159 Word-internal structures in Early St Kitts | 166 Summary: Early St Kitts | 171 Comparison: Early Saramaccan vs. Early St Kitts | 172 Syllable structure and phonotactic restructuring in the French-based creoles | 174 Introduction | 174 Previous studies | 174

Contents

6.3 6.4 6.4.1 6.4.2 6.4.3 6.4.4 6.5 6.5.1 6.5.2 6.5.3 6.5.4 6.6 7

| IX

Methodological issues | 176 Results I: Guiana French Creole | 181 Word-initial onsets in Guiana FC | 181 Word-final consonants in Guiana FC | 188 Word-internal structures in Guiana FC | 195 Summary: Guiana FC | 200 Results II: Trinidad French Creole | 201 Word-initial onsets in Trinidad FC | 202 Word-final consonants in Trinidad FC | 210 Word-internal structures in Trinidad FC | 220 Summary: Trinidad FC | 228 Comparison: Guiana FC vs. Trinidad FC | 229

7.4.1 7.4.2 7.4.3 7.4.4 7.5

Syllable structure in the six creoles: Similarities and differences | 232 Introduction | 232 Cross-creole similarities and differences in syllable types | 232 Cross-creole similarities and differences in restructuring rates | 235 Retention and restructuring of word-initial structures | 236 Retention and restructuring of word-final structures | 238 Retention and restructuring of word-internal structures | 241 Summary: Comparison of restructuring rates | 243 Cross-creole similarities and differences in targets of restructuring | 244 Targets of restructuring among word-initial structures | 245 Targets of restructuring among word-final structures | 247 Targets of restructuring among word-internal structures | 249 Summary: Comparison of targets of restructuring | 251 Summary: Cross-creole comparison | 252

8 8.1 8.2 8.3 8.4 8.4.1 8.4.2 8.4.3 8.4.4

Explaining creole phonotactic restructuring | 254 Introduction | 254 Two approaches to the emergence of creole phonology | 255 Substrate syllable structure | 262 Testing SLA approaches to creolisation | 265 Mechanisms in creolisation: Berbice Dutch | 269 Mechanisms in creolisation: Negerhollands | 272 Mechanisms in creolisation: Early Saramaccan | 275 Mechanisms in creolisation: Early St Kitts | 279

7.1 7.2 7.3 7.3.1 7.3.2 7.3.3 7.3.4 7.4

X | Contents 8.4.5 8.4.6 8.4.7 8.4.8 8.4.9 8.4.10 8.5 8.6 9

Mechanisms in creolisation: Guiana French Creole | 281 Mechanisms in creolisation: Trinidad French Creole | 284 Mechanisms in creolisation: The overall picture | 286 Similarities and differences across creoles: What causes the observed patterns? | 294 Creole phonotactic restructuring: What motivates repair choices? | 296 Summary | 304 Creole syllable structure in phonological theory: previous approaches and future challenges | 305 Summary | 306 Creole syllable structure: A final assessment | 309

Bibliography | 315

Abbreviations and notational conventions A.

Abbreviations of language names BD ESA ESK GFC NH TFC

B.

Berbice Dutch Early Saramaccan Early St Kitts Guiana French Creole Negerhollands Trinidad French Creole

General terms CHAID IL IPA L1 L2

C.

Chi-squared Automatic Interaction Detection (a method used to build classification trees) interlanguage International Phonetic Alphabet a speaker’s native language a language learned subsequently to the L1

Symbols and notational conventions C V # . σ < > / / [ ]

consonant vowel word boundary syllable boundary syllable orthographic representation phonological (i.e. underlying) representation phonetic representation

1 Introduction Pidgins and creoles were long regarded as defective or broken forms of their respective lexifier languages. The realisation that the observed deviations from the lexifier norm are in fact the product of independent linguistic systems led to the establishment of pidgin and creole studies as an academic discipline in the second half of the 20th century. Since then, it has been shown that the study of pidgins and creoles can contribute significantly to discussions in different areas of linguistics, including language contact, language change, second language acquisition and sociolinguistics (cf. Holm 2000: 3f). In the last decades, the body of literature on creoles has grown considerably. On a theoretical level, the question of creole genesis has been – and continues to be – the subject of intensive debate. On a more descriptive level, numerous studies have addressed different grammatical structures of creoles and thus advanced our understanding of the nature of contact languages. However, not all aspects of creole structure were equally well covered in either the more descriptively oriented works or the more theoretical discussions. In the area of creole phonology, most research has focused on segmental inventories, whereas syllable structure, like other suprasegmental aspects, has received relatively little attention. One reason for this state of affairs may well be the long-standing claim that creole syllables tend to follow a simple CV pattern (Holm 1988, 2000, Romaine 1988), a claim that also shaped the discussion concerning the origin of creole syllable structure. In recent years, more detailed studies of creole syllables have shown that most creoles do not in fact restrict their inventory to CV syllables (e.g. Stolz 1986, Sabino 1990, 1993, Aceto 1996, Meade 1995, Singler 1996, Lipski 2000, Plag and Schramm 2006, Bhatt 2007). Instead, creoles typically allow also more marked structures such as V, VC, CVC, CCV and CCVC (Klein 2011). The range of attested syllable structures in creoles is thus much wider than commonly assumed. Moreover, some creoles are considerably more permissive than others (e.g. Plag and Schramm 2006, Klein 2011). Apart from rendering the simplicity claim untenable, these findings stress the importance of an empirically adequate description as a basis for the discussion of creole formation. They also raise the question as to how the observed cross-creole variation can be explained. Which creole structures do we find for different sets of contributing languages, to what extent can these structures be traced back to the respective superstrate or substrate languages and which role do universals of language acquisition and development play in shaping creole syllables? Although recent years have seen an increase in studies addressing either particular aspects of syllable structure in individual languages or particular phe-

2 | 1 Introduction nomena of phonotactic restructuring (especially vowel paragoge, cf. e.g. Plag and Uffmann 2000, Uffmann 2008), there is still a lack of both comprehensive accounts for individual languages and large-scale comparative studies investigating the emergence of creole syllable structure. Such studies are important as they allow us to determine the degree to which different combinations of potentially influential factors lead to different structures in the creole (see, for instance, Plag and Schramm 2006 on the effects that variation in the kind of contact had on syllable structure in English-based Caribbean creoles). Cross-linguistic studies can thus help us evaluate the relative impact of different factors on the outcome of creolisation. The present study is concerned with the emergence of syllable structure in six Caribbean creoles. It extends and refines the comparative approach taken by Plag and Schramm (2006), covering data from two English-based, two Dutch-based and two French-based creoles. This selection makes it possible to investigate both variation across creoles with the same lexifier and variation across creoles with different lexifiers. Concentrating on Caribbean creoles rather than including creoles from different regions has two important advantages. First, reliable data from early stages is available for a number of Caribbean creoles. Such data is of particular interest in the discussion of creole genesis as language-internal developments may obscure the original structures in later stages of the creole (cf. e.g. Aceto (1996) on syllable onsets in Saramaccan). Second, the conditions under which the Caribbean creoles developed were very similar. Concentrating on this region therefore reduces the number of additional variables which might have influenced the creole outcome. The aim of this study is two-fold. The first objective is to determine which syllable structures are attested in the different creoles and whether or not they deviate from those in the respective superstrate and substrate languages and from structures in creoles with similar contributing languages. The second aim is to investigate what factors determine the creole outcome. Current claims about the emergence of creole phonology (Plag 2009, Uffmann 2009), which have not yet been tested on a broader empirical basis, will be evaluated in the light of the findings on creole syllables. This study will thus offer insights into the mechanisms at work in creole formation. It will also have implications for the question of whether or not creoles should be considered as being typologically different from non-creole languages in terms of syllable structure. We will also see that creoles with similar contributing languages may show not only considerably different syllable structures but also different preferences in terms of repair strategies for illicit structures. Empirical support will be provided for the position that the main mechanisms in the formation of creole sylla-

1 Introduction

| 3

bles are transfer of substrate grammatical structures, substrate levelling, regular language transmission and acquisition of superstrate structures. It will be shown that both markedness and perceptual salience influence the probability of retention of different structures in the creole. With respect to phonotactic restructuring I will argue that adaptations of input structures can take place in perception as well as production. The linguistic data for the empirical investigation are taken from six different sources, ranging in date from the 18th through the 20th century, depending on the availability of reliable material for the individual creoles. Sources from similar periods were used for pairs of creoles with the same lexifier. The book is organised as follows. In the next chapter, I will give a brief outline of the different positions researchers have taken in the discussion of creole formation in general and the emergence of creole syllable structure in particular. The chapter also includes an overview of existing models of syllable structure and comments on the categories and terminology chosen for this study. In chapter 3, I will motivate the choice of creole languages and specific source texts used in this study and provide an overview of the historical background of the individual creoles. Methodological issues concerning data coding and analysis will also be covered. Chapters 4 through 6 will be concerned with the results of the empirical investigation of syllable structure and phonotactic restructuring in the individual creole languages, covering the Dutch-based, English-based and French-based creoles in turn. I will establish which kinds of structural categories are targeted by repair mechanisms in the creole and identify cross-category differences in both the rate and the nature of phonotactic restructuring. Chapter 7 will then bring together and compare some of the findings from the preceding chapters, focussing on cross-creole similarities and differences with regard to constraints on syllable structure. In chapter 8, I will discuss the extent to which the observed creole patterns can be explained by current theories of creolisation that ascribe an important role to processes of second language acquisition. The final chapter contains a general conclusion concerning both the emergence and the nature of creole syllable structure.

2 Creole genesis and syllable structure 2.1 Introduction This chapter introduces some of the central issues in the discussion of creole formation, focusing on those aspects which are most relevant to creole syllable structure and creole phonology in more general. I will first give a brief outline of different approaches to creolisation and then relate the most prominent claims advanced in the literature about the nature and emergence of creole syllable structure in particular. I will restrict myself to a sketch of the general positions here. A detailed description of the two proposals which are tested in this study will be given in chapter 8. In section 2.4, I will first provide an overview of different syllable models that have been suggested in the literature and then explain which approach was chosen for this study. A short summary of the chapter is given in section 2.5.

2.2 Sources of creole structure and mechanisms in creolisation The question of how creoles come into being, of the mechanisms involved in their creation, is one of the central issues in the study of creole languages and has been the subject of ongoing debate for decades. In their search for the origin of creole features, researchers have taken different, sometimes strongly opposing views, attributing the most important role in creolisation alternatively to the superstrate, to the substrate languages or to universals of language acquisition and development. In the discussion of creole syllables, the universalist and the substratist positions have been most prominent. The most influential proposals at the universalist end is the language bioprogram hypothesis (Bickerton 1981, 1984). Supporters of this hypothesis view creole structures as the result of first language acquisition by children who receive only inadequate input in the form of a pidgin. According to the theory, such children resort to strategies provided by some innate capacity for language, a core grammar that Bickerton calls the language bioprogram, to create a fully-fledged language, the new creole. In the absence of positive evidence in the input for more complex structures, this new language will be characterised by a strong tendency towards unmarked structures. This view has been challenged by proponents of the substratist hypothesis who regard adults, not children, as the main agents of creolisation. Substratists

2.3 Creolisation and syllable structure

| 5

assume that the majority of creole features have their origin in the native languages of the slaves. In support of this position, numerous researchers have pointed out parallels between creole structures and structures found in the substrate languages (e.g. Boretzky 1983, Migge 1998, Lefebvre 1998, Parkvall 2000, Singler 2000, Siegel 2003, see also the contributions in Lefebvre 2011). Especially where marked features are concerned, it has been argued that the universalist approach fails to account for these parallels. Under the substratist hypothesis, the main processes assumed to be at work in creolisation are transfer or retention of grammatical structures from the substrate languages and relexification, i.e. the replacement of native lexical items by phonetic strings from the superstrate (Lefebvre 1998). In addition, reanalysis and dialect levelling are claimed to play a role. In recent years, the focus in the study of creole genesis has shifted away from identifying the single most important factor in creole formation to determining how different factors might interact (Kouwenberg and Singler 2008a). A growing number of researchers thus take a more integrative view, regarding creolisation as a complex process in which several factors play a role (Siegel 1999, Smith 2001, Migge 2003, Uffmann 2003, 2009, Steele and Brousseau 2006, Braun 2009, Plag 2009). Most creolists today agree that processes of second language acquisition are of central importance in creole formation. Proposals focusing on SLA parallels often treat creolisation as a particular case of L2 acquisition, namely one involving limited access to the target language. It is not always clear, however, how many of the creole structures can in fact be attributed to SLA, which other mechanisms might be involved and how they interact in determining the creole outcome. The next section gives a brief overview of claims that supporters of different creolisation theories have advanced with respect to the nature and development of creole syllable structure.

2.3 Creolisation and syllable structure Suprasegmental aspects of creole phonology have long been neglected in descriptive as well as theoretical accounts. In the past, general statements on syllable structure often did not extend beyond the claim of simplicity. Thus, the most commonly stated generalisation was that CV syllables, that is, open syllables with no more than a single consonant preceding the vowel, dominate in creole languages. A particularly extreme statement comes from Romaine (1988) who maintains that “[c]reoles have no initial or final consonant clusters. They have a simple syllable structure which consists of alternating consonants and vowels, e.g.

6 | 2 Creole genesis and syllable structure CVCV” (Romaine 1988: 63). Cases of phonotactic restructuring in creoles such as the ones in (1) were taken as evidence for the CV hypothesis. (1)

Creation of CV syllables in creoles (data taken from Holm 2000: 141ff) Creole form Etymon Language kini < D. knie Negerhollands méze < P. mês Príncipe CP sisa < E. sister Sranan taki < E. talk Sranan vale < F. avaler Trinidad CF

Supporters of a universalist approach to creole genesis attributed such modifications to a preference for CV as the most unmarked syllable type. Substratists, on the other hand, interpreted the creation of CV syllables as the result of transfer of phonotactic restrictions from substrate languages. For instance, Holm (2000: 239) advances that “[p]erhaps the single most important factor shaping the phonology of many of the Atlantic creoles was the retention of the substrate phonotactic rules tending to give syllables a CV structure.” More detailed studies of syllable structure in individual creole languages have shown, however, that the CV claim is empirically not well-supported. Rather than showing only the most unmarked type of syllable, many creoles also allow more complex structures such as word-initial clusters or word-final singleton consonants (see e.g. the findings by Tinelli 1979, Stolz 1986, Sabino 1990, Singler 1996, Lipski 2000, Plag and Schramm 2006, Bhatt 2007, Klein 2011). In a typological study of syllable templates in creole and non-creole languages, Klein (2011: 183) comes to the conclusion that “creoles cluster in the higher end of the typological middle” and that “no creole language features exclusively CV syllables”. Crucially, the latter statement is true even for early varieties of radical creoles, such as the Surinamese creoles Sranan and Saramaccan (cf. Plag and Schramm 2006). This renders an explanation of non-CV structures as effects of decreolisation or later borrowings, as suggested, for instance, by Alleyne (1980) and Boretzky (1983), unlikely. While it seems that the simplicity hypothesis has to be rejected, it may be true that the range of syllable structures we find in creoles is overall smaller than the range we find in other languages. Comrie (2011) picks up on this point. Looking at creole tone systems and syllable codas from a typological perspective, he finds that complex structures are considerably less frequent in creoles than in

2.3 Creolisation and syllable structure

| 7

non-creole languages.¹ Comrie (2011: 602) tentatively concludes that “phonology may be one area where simplification can be seen as a general driving force in the formation of creoles.” Typological studies thus suggest that, on the one hand, creole syllable structure is not maximally unmarked. On the other hand, it usually does not occupy the upper end of the complexity scale, either. Whether we can conclude from this, as Comrie suggests, that simplification is the driving force in the emergence of creole phonology, remains to be seen. Most recent non-typological studies of individual creoles (e.g. Brousseau and Nikiema 2006, Bhatt 2007, Uffmann 2008) take a more substrate-oriented view in the explanation of phonotactic restructuring. In these studies, creole syllable structure is regarded as a compromise between restrictions imposed in the substrate languages and those imposed in the superstrate. However, since substrate and superstrate languages may exhibit vastly different structures, this leaves a lot of room for variation. In a comparative study of early varieties of four English-based creoles, Plag and Schramm (2006) show that even creoles with the same lexifier and similar substrates may vary considerably in terms of syllable structure, some being closer to the substrates and others being closer to the superstrate. Mostly, this variation concerns structures attested in the superstrate, but not in the substrates, i.e. different creoles may retain a different range of superstrate structures. This is not the only conceivable type of compromise, however. It has also been suggested that under certain conditions structures which are possible in the substrates but unattested in the superstrate may not occur in the creole (Brousseau and Nikiema 2006, Uffmann 2009). It is thus far from clear which creole syllable structures will result from the contact of a given set of substrate and superstrate languages and whether an explanation that is in accordance with the findings for one creole will also hold for the potentially quite different structures of another. To complicate matters further, the results of existing studies are often difficult to compare, because different studies make use of different theoretical frameworks, which can lead to quite diverse interpretations of similar surface structures. For instance, a wordfinal consonant in a CVC sequence, which might be interpreted as a syllable coda in one framework, might be analysed as the onset of an empty-headed syllable in another. Conclusions about creole syllables which were drawn on the basis of different theoretical approaches therefore cannot be compared directly.

1 His remarks on syllabe codas are made with some reservations since the frequency of complex codas in creoles could only be compared to the frequency of complex syllables in general in noncreole languages.

8 | 2 Creole genesis and syllable structure Given this situation, more cross-linguistic studies are needed to determine which types of structures we find in creoles and whether cross-creole differences can be attributed to differences in the contributing languages or to non-linguistic factors such as the kind of contact between substrate and superstrate speakers. Many recent claims made in this respect have not yet been tested on a broader empirical basis. Existing proposals on the emergence of creole syllable structure also require critical assessment and refinement in another respect. The degree of complexity is not the only aspect in which creoles can differ. Also in terms of repair strategies, we find variation. For instance, a structure that is repaired by vowel epenthesis in one creole may be simplified by consonant deletion in another. In general, some creoles show abundant cases of vowel epenthesis, whereas the same process is rare or non-existent in other creoles (cf. Mühlhäusler 1986, Holm 1988). Even within a single creole, the choice of strategy may vary across different structures (Alber and Plag 2001). Given that most studies deal only with particular aspects of syllable structure rather than provide a complete account, this variation is hardly ever systematically investigated. To conclude, more comprehensive accounts of creole syllable structure and phonotactic restructuring are needed to determine both the range and the sources of attested creole structures and repair choices. The present study contributes to filling this gap, providing a systematic investigation of syllable structure and restructuring mechanisms in six creole varieties with three different European lexifiers. The results of the empirical study will be used to test the predictions made by two current creolisation hypotheses, those by Plag (2009) and Uffmann (2009), both of which claim that processes of L2 acquisition feature prominently in creole formation. As pointed out above, the theoretical assumptions underlying an analysis of syllable structure strongly influence its results. In the next section, I will outline some of the existing controversies regarding assumptions about the internal structure of the syllable and motivate the approach chosen for this study.

2.4 Theories of syllable structure Linguists have attempted to define the syllable as both a phonetic and a phonological unit. The search for a defining articulatory or acoustic phonetic characteristic, however, has proved difficult. Thus, it has been suggested that the syllable can be described in reference to lung movements, namely as the unit produced by a single chest pulse (Stetson 1951). Another common phonetic definition is based on auditory characteristics, instead, defining a syllable as (the unit around) a promi-

2.4 Theories of syllable structure

|

9

nence peak (e.g. Jones 1950). However, as several authors remark, neither of these definitions seems to systematically correlate with speaker intuitions about syllables. For instance, some words such as going or being, which speakers intuitively analyse as consisting of two syllables, may be produced by a single chest pulse (cf. e.g. Duanmu 2009: 36). Counterexamples to the prominence theory include words such as English stop. As pointed out by Gimson and Cruttenden (1994: 49f), even though the sequence involves two prominence peaks, the initial [s] and the vowel, speakers analyse the word as consisting of a single syllable. While a convincing definition of the syllable as a phonetic unit may be lacking, most linguists agree that we should assume the existence of the syllable as a phonological unit.² There is surprisingly little agreement, however, with respect to the internal structure of the syllable, the maximal size of the syllable and – in models which assume subsyllabic constituents – the maximal size of individual syllable constituents. Numerous different syllable models have been suggested in the literature. Useful overviews are given, for instance, in Blevins (1995) and Bosch (2011). I will concentrate here on the differences between existing models and the consequences these differences have for the interpretation of surface sequences of segments. Some syllable models do not involve any subsyllabic constituents (e.g. Kahn 1980, Clements and Keyser 1983). In such flat structure models, segments are either dominated immediately by the syllable node or by C and V nodes of an intervening CV tier. Other models postulate the existence of additional constituents on a level between the segments and the syllable. Thus, many phonologists assume that syllables consist of an onset, a nucleus³ and a coda. If no further constituents are involved (e.g. Hockett 1955), a ternary branching syllable model as in (2) results, in which no special relationship exists between any of the syllable constituents. They are all immediately dominated by the syllable node and each of the syllable constituents may be subject to constraints on size and segmental content. (2)

Syllable model with ternary branching σ onset

nucleus

coda

2 For an overview of traditional arguments in favour of the syllable as a constituent, the reader is referred to Blevins (1995). 3 Some models use the term peak instead of nucleus. This difference in terminology is ignored here. I will use the term nucleus in all cases.

10 | 2 Creole genesis and syllable structure Other models assume binary branching and a hierarchical order of constituents. Two alternative structures of this type have been suggested in the literature. In so-called body-coda models (McCarthy 1979), the syllable node branches into body and coda. The body itself is formed by the onset and the nucleus. Onset and nucleus thus stand in a special relationship. In onset-rhyme models, on the other hand, it is the nucleus and the coda which form an intermediate constituent, the rhyme. Of the two types of binary branching models, onset-rhyme models are the more widely used by far (e.g. Pike and Pike 1947, Fudge 1969, Selkirk 1982). The corresponding structure is displayed in (3). (3)

Onset-rhyme model σ onset

rhyme nucleus

coda

In terms of surface structure interpretation, the models in (2) and (3) do not necessarily differ. Phonotactic constraints may be expressed differently, however. Proponents of an onset-rhyme structure often argue that some restrictions operate at rhyme level and thus concern the unit of nucleus and coda taken together rather than each of them individually. We find more variation in surface structure interpretation if we compare the above models to syllable models employed by linguists working in the framework of government phonology (Kaye et al. 1990, among others). Here, we have to distinguish between standard government phonology and the so-called CVCV approach, which emerged from government phonology, but makes different claims about syllable structure. Standard government phonological approaches are similar to the above-mentioned onset-rhyme models in that onsets and rhymes are recognised as syllable constituents. They differ from the above models, however, in that neither syllable nor coda nodes occur in representations of syllabic structure. The term coda may still be used, but then simply refers to consonants that are immediately dominated by the rhyme node. In the CVCV approach, on the other hand, coda consonants do not exist at all. A syllable is taken to consist of only an onset and a nucleus, neither of which are branching (e.g. Scheer 2004). Different assumptions exist not only with respect to syllable-internal structure, but also with respect to the maximum size of the syllable and/or individual syllable constituents. While the analysis of simple CV sequences is usually uncontroversial, many diverging interpretations have been suggested for post-vocalic consonants and consonant sequences in general.

2.4 Theories of syllable structure

|

11

Let us consider the most extreme positions first. On the one hand we have the CVCV approach, which imposes the tightest restrictions on syllable size, requiring a syllable to contain at most a single (prevocalic) consonant and a vowel. Under such an approach, postvocalic consonants and consonantal sequences that occur at the surface level are accounted for by postulating ‘defective’ or ‘degenerate’ syllables involving empty nuclei. Thus, a CVC sequence is interpreted as consisting of a regular CV syllable followed by a degenerate syllable in which only the onset is realised, whereas the nucleus position remains empty. Since, by definition, no syllable contains more than one consonant, all consonant sequences that occur at the phonetic level are analysed as belonging to separate syllables at the phonological level. The opposite extreme involves the assumption that all prevocalic or postvocalic consonants are part of the same syllable as the vowel they precede or follow and that they are incorporated in either the onset or the coda of that syllable. Given such an approach, a CVC sequence is analysed as a single syllable with a simple onset and a simple coda. Surface consonant sequences are taken to represent complex onsets, complex codas or coda-onset sequences without any intervening empty syllable constituents. Many phonologists advocate some intermediate position, suggesting varying limits on the maximum size of individual syllable constituents. For instance, some phonologists take the maximum size of syllable constituents to be what we find in word-internal position. Others base their assumptions about syllable constituents on syllables not containing any affixes. Yet others use yet other subsets of syllables. What all approaches of this kind have in common is that additional consonants that may occur only at word edges are not counted as regular onset or coda segments. Instead, such consonants are often accounted for by either postulating some form of appendix at word edges (e.g. Fudge 1969)⁴ or by assuming that the additional consonants are not in fact separate segments, but form complex segments with the following or preceding consonant (e.g. Duanmu 2009). To complicate matters further, size limits are not the only factors which determine whether consonant sequences are interpreted as belonging to the same syllable or syllable constituent. To give just one example, even though branching onsets consisting of two segments are allowed in many syllable models, not all prevocalic sequences of two consonants are necessarily taken to be complex onsets. Depending on the nature of the segments and their behaviour in phono-

4 Segments in an appendix are usually assumed to be immediate constituents of the syllable node and as such are not part of either onset or coda.

12 | 2 Creole genesis and syllable structure logical processes, the edgemost segment may alternatively be analysed as being in an appendix or as belonging to a separate syllable headed by an empty nucleus. As demonstrated above, the choice of syllable model has considerable influence on the interpretation of surface structures. In studies of individual languages, the researcher chooses the model which best explains the observed facts. In cross-linguistic studies, this choice becomes more problematic. A theoretical model that is well suited for the description of one language may not be the most appropriate one to account for the facts in another language.⁵ For a comparative study, however, we need a set of categories that can be applied to all the languages involved (cf. the discussion of ‘comparative concepts’ in Haspelmath (2010)). As a consequence of the wide applicability, the categories used in language comparisons do not necessarily correspond to those that give us the most elegant account of a particular language. To focus on the present study, I needed categories that would allow me to compare constraints on sound structures above the segmental level in six different creoles, their respective lexifier languages and the major substrate languages involved. I was aiming for a description that is as theory-neutral as possible while at the same time providing results that can be used as the basis for more theoretically-oriented accounts of individual languages. Unfortunately, theoryneutral terminology is not easy to come by. I have therefore chosen to adopt the terminology of a syllable model that includes subsyllabic constituents, but neither appendices nor empty nuclei, i.e. a model in which the divergence between surface structure and underlying phonological structure is kept to a minimum. Thus, the terms ‘onset’ and ‘coda’ will be used for prevocalic and postvocalic (sequences of) consonants, respectively, and phonotactic constraints will be described in terms of restrictions on syllable onsets and codas.⁶ While the decision on what is prevocalic and what is postvocalic is straightforward at word edges, the same is not necessarily true for word-internal positions. In particular, word-internal consonant sequences present a problem. In order to describe such sequences in terms of onset and coda structures, syllable boundaries would have to be posited. However, even in models not assuming empty nuclei, there may be disagreement about syllable boundary placement. Although

5 After all, if a single syllable model existed which is ideal for the description of all languages, there wouldn’t be any reason for the wide variety of different assumptions in the phonological literature. 6 Note that the use of this terminology is not supposed to imply that the corresponding syllable model is necessarily the model which can best account for all of the observable facts on a theoretical phonological level. Whether or not that is the case is a question that lies beyond the scope of the present study and will have to be determined in future research.

2.5 Summary |

13

not all types of sequences are equally problematic,⁷ I decided against a systematic postulation of syllable boundaries in items with intervocalic consonant clusters and opted instead for a description of possible (or impossible) word-medial consonant sequences in general. As a result, many of the descriptions of creole sound structure given in this book concern the level of the phonological word as much as the level of the syllable. Nonetheless, I will mostly talk about syllable structure since it is claims about creole syllables that I intend to relate my results to.

2.5 Summary In this chapter, different claims about mechanisms in creolisation and the structure of creole syllables were introduced. A short overview of existing syllable models was then given and it was illustrated that the choice of syllable model has far-reaching consequences for the theoretical interpretation of surface structures. Based on these considerations, I argued that the categories used for the comparisons in this study should be as theory-neutral as possible. The resulting choices were outlined only briefly. More details on the categories and terminology used will be given in the next chapter, which covers all relevant methodological aspects.

7 For instance, most obstruent-approximant clusters are likely to be analysed as complex onsets and liquid-obstruent clusters will typically be regarded as heterosyllabic coda-onset sequences.

3 Data and Methodology 3.1 Introduction The aim of this chapter is to provide an overview of the various methodological issues and procedures involved in data selection, coding and analysis. The most important questions that are addressed are the following. Which languages were included in the present study and why were these languages chosen over others? Under which circumstances did the chosen creoles develop and which languages were involved in the contact? Where do the data for individual creoles come from? Which properties were investigated and how were the data coded for the subsequent analyses? What kinds of analyses were conducted and what was the rationale behind these procedures? The chapter is organised as follows. In section 3.2 I will introduce the criteria that were applied in the selection of languages to be investigated, both at the level of lexifier languages and at the level of individual creoles. Section 3.3 will provide a brief overview of the historical background of each of the creoles chosen for investigation, covering in particular assumptions regarding the most important substrate languages and aspects which are relevant to determining the intensity of the contact between substrate and superstrate speakers. Information on the corpora for the individual creoles will be given in section 3.4, including a description of the nature of the source texts, comments on the reliability of the data, and a discussion of what I considered different creole word types. Section 3.5 will be concerned with the problems involved in identifying the (likely) form of the input. More specifically, I will address the question of main lexifier varieties and introduce additional factors that were taken into account in the decisions on selecting particular etymon forms. Details of data coding and the general analytical procedure will be provided in section 3.6. Here, I will give an overview of the categories employed in the classification of lexifier and creole structures on the one hand and processes of phonotactic restructuring on the other hand. I will outline the steps involved in data analysis and introduce the rationale I followed in the interpretation of the results. The chapter concludes with a summary in section 3.7.

3.2 Selection of creole languages In the selection of creole languages I aimed at minimising the number of sociohistorical and linguistic variables as far as possible. Two important such variables were the context in which the creoles developed and the general region in which

3.2 Selection of creole languages

| 15

they are (or were) spoken. All of the investigated languages are Caribbean creoles that developed in the context of colonisation, usually in plantation settings. They emerged from the contact between the slaves’ native African languages and the European languages of the colonial masters. A further goal of keeping the region constant was to reduce the range of substrate languages. Although it was not always possible to ensure a strict comparability of substrates across creoles, the substrate languages involved at least show strong similarities in terms of syllable structure and phonotactics (cf. section 8.3 below for information on substrate syllable structure). Concerning the choice of lexifier languages, it was, of course, desirable to cover as many different superstrates as possible. However, an additional condition had to be met. In order to evaluate some of the claims made in the literature concerning the relation between the kind of contact and the relative influence of the lexifier, pairs of creoles with the same lexifier were needed. Thus, a given superstrate was included in the comparison only if two different creoles could be found which inherit the major share of their lexical items from that language. Where possible, two creoles were chosen which shared their main lexifier, but differed with respect to the duration or intensity of contact they had had with the superstrate. The major European lexifiers for Caribbean creoles include English, French and Dutch as well as Spanish and Portuguese. The two Ibero-Romance languages, however, failed to fulfill the above-mentioned criterion. Although two creole languages exist in the area which can be counted as Spanish-related in their present state, it is far from clear whether the same is true for the initial stages of these languages. The candidates in point are Papiamento, spoken on the islands of Aruba, Bonaire, and Curaçao, and Palenquero, the (not strictly Caribbean) creole spoken in Palenque de San Basilia, Colombia. For both languages, it is still debated whether Portuguese should not be considered the original lexifier rather than Spanish. Theories that favour a Portuguese origin usually attribute the modern, Spanish look of the two creoles to their prolonged co-existence with Spanish, assuming either decreolisation or relexification.¹ Unfortunately, data from the very early stages of these creoles, which could have helped resolve the issue, is missing. The earliest available source of any substance for Papiamento is from the second half of the 19th century (Van Name 1869). For Palenquero, no reliable source dates back earlier than the 20th century. Due to these problems, the Ibero-Romance lexifiers were excluded from the present investigation.

1 For a summary of the different positions on the origin of Papiamento and Palenquero, the reader is referred to Lipski (2008).

16 | 3 Data and Methodology The remaining lexifier languages were less problematic. Although not all of them are equally well represented in the Caribbean, suitable data from at least two creoles could be found for each of the lexifiers in question. The quality and quantity of the available data determined the choice among individual creoles with the same lexifier. In the following I will outline the problems involved in dealing with many of the existing data and introduce the criteria that were taken into account in the selection of source texts. As is often the case in creole studies, the kind of data that would be ideal for the purpose of the present study do not exist. In the light of the development of creoles as oral languages it is not surprising that written documents should be scarce to begin with, especially so for the respective earliest, formative period. Those who were native speakers of the creole or at least competent in the language usually did not write it, either because they were illiterate, or because the association of the creole with low social prestige prevented them from doing so. Thus the texts which survived were mostly written by native speakers of European languages who had been educated in Western Europe and were for some reason or other interested in portraying the creole language. We find variation in genre, register and style across different documents. In situations where several creole varieties co-existed (e.g. in the case of Negerhollands), variation between so-called high creole, typically spoken by whites and more closely modelled on the lexifier language, and more basilectal varieties also occurred. Several difficulties arise from this situation. First of all, many of the authors were not familiar with the creole and referred to it as some sort of distorted or corrupted English, Dutch or French. Some of them only noted features that were especially striking or frequent in creole speech, such as individual, creole-particular lexical items, the lack of inflection, or regular substitution patterns (e. g. [θ, ð] > [t, d]). Others were more accurate and recorded details of phonology. However, even those authors who noticed the subtler differences between their own pronunciation and creole usage were faced with the problem of how to represent the noted features. Since the creoles were purely oral languages at the time, no established writing system existed that the authors could have adhered to. In most cases, they made use of the orthographic conventions they were used to from their own native language, usually the main lexifier language of the respective creole, sometimes modifying only single elements to reflect the characteristic features they observed in creole speech. Such writer-specific orthographic modifications are often far from unambiguous as to the pronunciation they might represent. They therefore cannot be interpreted on the basis of single forms, but must be analysed with regard to other, similar instances in the same text. From the above problems, three criteria for the selection of source texts emerged. First and foremost, the notation had to be reliable with respect to

3.3 The creoles and their historical background |

17

phonological aspects. If a given writer had made no observable attempt at systematically portraying deviations from lexifier pronunciation, the text was discarded. Second, early sources were preferred to later ones unless they failed the reliability criterion or portrayed a high creole variety rather than a basilectal one (cf. the discussion of Negerhollands sources in section 3.4.1 below). Finally, any given source text had to be long enough to identify idiosyncratic spelling conventions. For each of the chosen creoles, data from only a single source text was used in order to eliminate further variation.² The application of these criteria led to the following language choices. The group of documented Dutch-based Caribbean creoles is much smaller than that of either English-based or French-based creoles. Apart from the comparatively well-documented Negerhollands, only two other creoles are attested, Berbice Dutch and Skepi Dutch. Both the documentation and the available data proved to be far better for Berbice Dutch. Among the English-based creoles, there were three potential candidates for investigation, Sranan, Saramaccan³, and St Kitts. The data available on these creoles fulfil all the criteria for source texts outlined above. Given that the Surinamese creoles Sranan and Saramaccan are not only closely related, but also highly similar in terms of syllable structure and degree of contact to the superstrate (cf. Plag and Schramm 2006), I chose to include only one of these languages, Saramaccan, in the study and compare it to the island creole St Kitts.⁴ For many of the French-based creoles, it is hard to come by early data that reliably indicate phonological aspects. The choice fell on the two languages for which the earliest reliable sources of acceptable length were available, Guiana French Creole and Trinidad French Creole.

3.3 The creoles and their historical background In this section, I will give a short overview of the circumstances in which each of the chosen creoles developed, including also information on the origin of slaves and the principal substrate languages assumed in the literature. Note that,

2 For a description of the individual source texts see section 3.4 below. 3 Saramaccan is treated as English-based creole here since the majority of etyma are of English origin. It should be noted, however, that the Saramaccan lexicon also contains a considerable share of Portuguese-derived items. 4 The reason for choosing Saramaccan over Sranan was a practical one. Although digitised and partially coded data for both creoles were available from previous studies, the Saramaccan data base could be more easily adapted to the format needed for the present investigation.

18 | 3 Data and Methodology wherever classifications of African languages are given, these are based on Lewis (2009).

3.3.1 Berbice Dutch The last known competent speaker of Berbice Dutch died in 2005 and the language has officially been declared extinct. The creole developed on plantations along the Berbice river in the former Dutch colony of Berbice, an area which today is part of Guyana. According to Robertson (1994), the Dutch colony was slow to grow from the first settlement on the Berbice river in 1627 to no more than a dozen plantations in 1710. Slave supplies during this early period were low and did not satisfy the demand so that Berbice colonists had to rely heavily on slaves born in the colonies (cf. Robertson 1990: 55). As far as languages spoken by the slave population are concerned, it has been argued (e.g. by Smith et al. 1987, Kouwenberg 1994, 2012) that substrate influence on the Berbice Dutch lexicon and grammar can be traced back to a single West African dialect cluster, Eastern Ijo. Regarding the time of development, Robertson (1994: 69) states that Berbice Dutch must have been well-established at least by the time of the 1763 Berbice slave rebellion, since the area in which the creole survived into the 20th century was never resettled by the Dutch afterwards. Even before that rebellion, by the 1740s, settlement within the Berbice colony had started shifting from upriver to downriver and coastal areas (Robertson 1990). The rebellion further sped up that development and, as a consequence, the creole communities in upriver locations became relatively isolated. English colonists became more numerous especially along the coast, starting a gradual transition from Dutch to English control over the colonies. By the end of the 18th century, the British had gained more extensive control of Berbice, Demerara, and Essequibo. The three colonies were officially taken over in 1814, although Dutch continued to be spoken alongside English for some time afterwards. Robertson (1990) assumes that language use among the slaves must also have gradually shifted as a consequence of the above changes. Especially in the areas where English was most dominant, also the slaves must have gradually shifted from Dutch- to English-related creole varieties. Berbice Dutch survived longest in Amerindian communities, which used the creole for communication with the Europeans even after the shift to English supremacy (cf. Robertson 1990).

3.3 The creoles and their historical background |

19

3.3.2 Negerhollands Negerhollands, like Berbice Dutch a creole with a mainly Dutch-derived lexicon, used to be spoken in what today are the US Virgin Islands (formerly the Danish West Indies). Negerhollands emerged around 1700 (cf. van Rossem and van der Voort 1996: 3) and died out only in the late 20th century. Although the Virgin Islands were a Danish colony at the time when Negerhollands emerged, the plantation owners at the formative period were mostly of Dutch origin, which explains how a Dutch-based creole could develop in the first place. Besides varieties of Dutch, also English and Danish were among the European languages spoken in Virgin Islands households, and lexical traces of these can be found in the creole. We know that, from very early on, several varieties of the creole must have existed, including a ‘high creole’, which was modelled more closely on the superstrate and was used by the whites, and a more basilectal variety of the creole spoken by the slaves (e.g. Sabino 1990, van Rossem and van der Voort 1996). It is generally assumed that the last speakers of Negerhollands spoke a continuation of the basilectal rather than a high creole variety (see also section 3.4.1 on the selection of source texts for Negerhollands). According to the demographics reported by Sabino (1990) for the island of St. Thomas, the majority of substrate speakers must have had comparatively good access to the superstrate during the formative period. Many lived in mixed households including both superstrate and substrate speakers, with a ratio of superstrate to substrate speakers no worse than 1:2 around the time Negerhollands is supposed to have crystallised (Sabino 1990: 32ff). Sabino (1990: 45) argues that during this period, speakers of three Kwa languages, Akan, Gã, and Ewe, were dominant among the slaves. Also Stolz and Stein (1986) mention the dominance of these languages among the groups involved in the relevant slave tradings.

3.3.3 Saramaccan Saramaccan is a creole language that today is still spoken by about 26000 people (cf. Lewis 2009), most of them living in Suriname. Saramaccan is usually classified as an English-based creole, but its lexicon also has a substantial Portugueserelated component. While the origin of this Portuguese component is still being debated, most creolists agree that the dominance of English-derived items in the lexicon dates back to the early stages of the creole (e.g. Smith 1987, McWhorter 2004, Smith and Cardoso 2004, Bakker 2009, Good 2009b). Saramaccan is generally believed to be genetically related to another English-based creole language

20 | 3 Data and Methodology of Suriname, Sranan (cf. e.g. Smith 1987, Migge 1998, Good 2009a,b). A precursor of these two creoles developed in the second half of the 17th century on sugar plantations in current Suriname. The first European plantations in the area were established in 1651, when the British claimed the area along the Suriname river as their colony. The colony grew rapidly and by the mid 1660s it is reported that there were between 40 and 50 sugar plantations. Arends (2002: 116f) suggests that the particular characteristics of sugar plantations, such as factory-like operation and a large labour force, would have caused the contact between substrate and superstrate speakers to be more restricted than on other types of plantations. The period of English rule over the area did not last very long. In 1667, the Dutch took over the colony. Another twelve years later, most of the British settlers had left Suriname, effectively removing the main lexifier language from the contact setting. This means that an English-based contact language must already have begun to stabilise by 1680 (see also, e.g., Plag 1993).⁵ As far as the Portuguese contribution to the emerging creole is concerned, a few years prior to the change in power, in 1665, around 200 Portuguese Jews settled in Suriname. The language spoken on their plantations might have been the source of the Portuguese element in Saramaccan (cf. e.g. Arends 2002).⁶ The beginnings of the Saramaccan tribe are usually dated to 1690, when the first mass escape from Suriname plantations took place (cf. Price 1976). The language of the early fugitive slaves would thus have formed the basis of Early Saramaccan with further lexifier influence being reduced by the lack of contact. Subsequent groups of fugitives slaves joined those already living in the rain forest during the following two decades. After 1712, the last date given by Price (1976) for escapes of sizable groups of slaves in the region, substantial outside influence on Saramaccan becomes more unlikely. It has been established in the literature that dialects of Gbe, Kikongo and, to a lesser extent, Twi are likely to have been most prominent among the slaves’ native languages during the formative period of the Surinamese creoles (e.g. Smith 1987, Plag 1993, Arends 1995, 2002, Migge 1998). Among the suggested substrates, Gbe is often assumed to have had the greatest influence on Saramaccan.

5 Arends (2002: 120f) dates the removal of English slightly later, quoting the claim of a Dutch historian that speakers of English formed an important part of the European population in Suriname up to the 1690s. Arends consequently also assumes a larger window for the development of an English-based creole in Suriname. 6 This view is not uncontested, though. See, for instance, Smith (1999) for alternative suggestions concerning the origin of the Portuguese element in Saramaccan.

3.3 The creoles and their historical background |

21

3.3.4 St Kitts St Kitts Creole English, the modern variety of the creole under discussion here, is still spoken today by about 39,000 people living in the islands of St Kitts and Nevis (estimate according to Lewis 2009). St Kitts was the first Caribbean island to be colonised by the Europeans in the 17th century and played an important role in the subsequent colonisation of neighbouring islands. Apart from being the point of departure for English and French colonisers, St Kitts also became a “seasoning ground” for newly arrived slaves, which were later transported to other colonies (Cooper 1999: 379). Baker (1999) assumes that an English-based contact language must have developed soon after the initial settling of St Kitts. According to Parkvall’s (1999a) summary of the early demographics of St Kitts, permanent European settlement began with the arrival of 13 English colonists in 1624 and 400 more in 1626. French settlers arrived shortly after the English and the island was peacefully divided among the two groups. In 1629, both French and English colonists were forced to leave when the Spanish took over St Kitts. However, most of them returned later and joined forces against the Spanish. Until the middle of the 17th century, the population of St Kitts grew rapidly. The second half of the century, however, saw a fast decline again, caused by emigration of settlers to areas offering more profitable opportunities as well as by a series of wars between the French and the English (cf. Parkvall 1999a). Apart from being highly destructive, the wars also caused large-scale fluctuation in the population. Only after 1713, when St Kitts was ceded completely to the English, the population grew again and continued to do so for the rest of the century. Although the proportion of slaves among the population of St Kitts grew steadily, it did not increase as fast as in some other colonies. According to the figures provided by Parkvall (1999a), slaves made up only around 30% of the population after four decades of settlement, crossing the 50% line no earlier than the late 17th century. What is further worth noting is that the percentage of children among the slaves on St Kitts appears to have been relatively high (26.9% and 30.1% in censuses of 1678 and 1708 respectively). As far as the dominant languages among the slave population are concerned, Parkvall (1999a) proposes that five different African language families need to be considered: Atlantic, Mande, Kru, Kwa and Bantu. The figures Parkvall provides on slave shipments, however, indicate that speakers of Atlantic, Mande and Kru did not make up a major part of the slave population of St Kitts during the early stages (up to about 1690). Kwa speakers, on the other hand, dominated between 1640 and 1690. The importance of Kwa speakers is acknowledged also by Cooper

22 | 3 Data and Methodology (1999), who explicitly lists Akan (Fanti/Twi) and Gbe as well as Yoruba.⁷ With respect to the latter, Parkvall (1999a) notes that, up to 1740, Yoruba was spoken only by a minority ( ∅ ∅ > C V > ∅ CC > CVC #C(C) > #VC(C) C# > CV# σ > ∅ VC > CV no restructuring

steer > tiri ants > hansi afraid > fredde sloth > silò rice > alisi soap > sopu remember > membre work > wrokko below > bilò

As the table shows, three different types of vowel insertions are distinguished. The label ‘vowel epenthesis’ is used in cases in which a vowel is inserted between two consonants. Such cases were thus kept apart from insertions at the beginning or end of words (labelled as ‘vowel prothesis’ and ‘vowel paragoge’, respectively). This distinction was necessary, because both edge and non-edge vowel insertions are potential repair strategies for initial and final consonant clusters. Pooling these two processes could have obscured the picture. This problem did not exist for cases of what I termed ‘metathesis’. The label was used not only for the reordering of vowel-consonant sequences as illustrated above, but also for the inversion of purely consonantal sequences. The latter process, however, is rare in the creole data and was therefore not listed above. In this case, the ambiguity of the label is unproblematic since the two types of metathesis apply to different structures. In none of the subsets emerging from the analyses do we find a pooling of the two processes. As noted above, processes were classified as initial, internal or final depending on the position of the structure affected by the change. In many cases, this definition is sufficient. For a subset of changes, however, we need to be more specific. Cases of metathesis, in particular, are somewhat problematic as they affect structures in different positions. For instance, consonant-vowel metathesis in parteira > plattiri (Portuguese > Saramaccan) could be interpreted as either word-initial (for creating a non-etymological word-initial cluster) or word-internal (for eliminating the word-internal coda-onset sequence [rt]). In such situations, the process was associated with the position of the most likely target of restructuring. In the given example, the word-internal coda consonant [r] was identified as more likely target, the alternative being an unmarked simple onset. Metathesis was therefore classified as word-internal process in this case.

40 | 3 Data and Methodology Wherever it proved necessary to recognise more than one process for a single position, the combination of the (two) respective values is given as a variable value. For instance, a pair such as haste > hessi (English > Saramaccan) would be coded for final restructuring as follows: ‘consonant deletion’ for the loss of the etymonfinal [t] and ‘vowel paragoge’ for the addition of a final non-etymological vowel [i], resulting in the variable value ‘consonant deletion + vowel paragoge’.¹⁶

3.6.3 Processes not considered in the analyses In general, I investigated only those processes which could be argued to be triggered by restrictions on syllable structure. I do not comment, therefore, on regular phonological sound changes that apply irrespective of environment, such as the systematic substitution of [t, d] for the interdental fricatives [θ, ð] or similar changes at the segmental level. Segmental substitutions are dealt with in the analyses only if they are context-specific and can be suspected to be induced by particular syllabic structures. Similar reasoning applies to vowel quality and quantity. Although the vowels in creole forms may differ from those in the respective etyma, we usually find systematic correspondences between lexifier and creole vowel qualities and quantities that are induced by differences in the vowel inventories rather than by restrictions on syllable structure. Since the present study is not concerned with systematic segmental correspondences, the nature of individual vowels was usually ignored. In the few cases where it seemed of interest to the discussion, I manually inspected the distribution of vowel qualities in subsets of the data in order to determine whether it might have had an influence on the investigated process. Relevant observations were included in the discussion.

3.6.4 The CHAID analyses CHAID analyses (exhaustive CHAID) were conducted for each of the following subsets of structures: word-initial complex onsets and onsetless structures, wordfinal simple and complex codas and word-internal consonant clusters. Cases of hiatus were not be subjected to a CHAID analysis, since no specific segmental val16 In fact, all cases were double-coded, but all codings that included ‘none’ in second position were subsequently simplified to display only the first value. Thus, an original value such as ‘consonant deletion + none’ would end up as ‘consonant deletion’, and a value of ‘none + none’ as simple ‘none’.

3.6 Data coding and analytical procedure

| 41

ues were coded for vowels. Consequently, it could not be tested whether these properties influence the treatment of hiatus in the creole. The data sets for the individual CHAID analyses comprised all cases involving etyma with the structure under investigation.¹⁷ In the analyses, the process variable for the relevant position was used as dependent variable and the pertinent structural variables as predictor variables. The analyses thus tested in how far the occurrence and the choice of phonotactic restructuring processes (as coded in the process variable) can be predicted from the structural properties of the etyma. More particularly, the CHAID algorithm calculates the best predictor variable on the basis of chi-square tests performed on all independent variables and divides the data into significantly different subsets. The procedure is repeated recursively over subsets, yielding a tree-like structure. Additional splits are created for significant predictors as long as the size of the data subsets does not fall below a predetermined minimum number of cases. I used the same settings for all CHAID analyses, working with a minimum size of three types for daughter nodes. Thus, splits which created subsets comprising at least three word types were allowed, but splits which singled out only two cases were prevented. The limit was set this low in order to allow an investigation not only of frequent, but also of comparatively rare structures. Due to the fact that my data are type-based, rare structures such as, for instance, word-initial plosive-nasal clusters (e.g. in Dutch knie) or word-final liquid-nasal clusters only occur in a small number of cases. Setting a higher limit for final subsets would have prevented the CHAID algorithm from separating such rare clusters from the rest of the data even if they do in fact behave differently. The results of such a CHAID analysis might have been confusing rather than insightful, because data would have been pooled which do not really behave alike. Although a limit of three cases for terminal nodes is certainly not ideal, either, I considered it the best option available for my

17 Etyma from different donor languages were not analysed separately. A reviewer suggested that this might be helpful, especially for creoles with a fairly mixed lexicon. There are two main reasons why this was not done. As described in section 3.5.2, in cases in which the set of potential source languages included related languages or dialects, there were often several potential etyma from different languages for a single creole form. Numerous of these alternative etyma were identical in terms of syllable structure. If the data were to be subdivided according to the origin of the etymon, a single source language would have to be selected in such situations. However, it is unclear which criteria this decision should be based on and how the choices might influence the results of donor-specific analyses. But even in situations in which the identification of the source language is unproblematic, a subdivision according to the lexifier language did not seem desirable. Given that some of the data sets for individual analyses are already quite small, a further division might have reduced them to such an extent as to make a meaningful analysis impossible.

42 | 3 Data and Methodology data and purposes. If three word types in the data show the same structure and all are treated in the same way, then this is at least an indication as to whether this particular structure is acceptable (in case of retention) or restricted (in case of repair) in the creole. One problem that arises from the decision to set a low limit for the size of terminal nodes is that the results of the CHAID analyses might not be reliable in some cases. Chi-square tests are known to be problematic when low cell counts are involved, in which case chi-square values may be inflated. In order to avoid false positive results, additional statistical tests were run over the subsets established in the CHAID analyses. A given split was maintained only if the differences between subsets were significant also according to a Fisher’s exact test, which provides more reliable results than a chi-square test when (some of the) cell counts are low. The results of the statistical analyses were interpreted as follows. If significant predictors of restructuring emerged from a given analysis, it was assumed that the properties coded by these predictor variables underly tighter restrictions in the creole than in the lexifier. The observed phonotactic changes were thus interpreted as repair mechanisms that apply to bring etymon structures in accordance with the phonotactics of the creole.¹⁸ Some additional comments on terminology are in order here which concern differences in the homogeneity of subsets emerging from the analyses. These subsets, although significantly different from each other, are rarely completely homogeneous in the sense that 100% of the cases show the same behaviour. In order to reflect different degrees of homogeneity of behaviour in different subsets of the

18 A reviewer mentioned that also a variable coding the donor language of the respective etymon could have been used as a predictor variable in CHAID analyses, ideally even in the analysis of a pooled data set including all six corpora, so as to allow conclusions about the role of the lexifier in creole phonotactic restructuring. The coding of such a predictor variable would have involved the same methodological problems as separate analyses for etyma derived from different source languages (cf. my earlier comment). A pooled CHAID analysis would have been problematic for a number of reasons, among them differences in the size of the six corpora, which would have led to an unequal overall impact of data from different creoles. Apart from that, it is also unclear how a potential effect of the variable DONOR should have been interpreted. Should it be taken to mean that items from some lexifier language are generally more restricted in the creoles than items from another language? But what could possibly make words from one language less acceptable than words from another language? A plausible explanation might be that etyma from different lexifiers do not have the same structural properties and are therefore treated differently in the creoles. In that case, however, it would be more informative to use the structural properties themselves as predictor variables.

3.7 Summary |

43

data, I distinguished categorical, tendencial, and truly variable behaviour or treatment. Allowing for a certain amount of random noise in the data, I regarded the application of a given process as categorical if it applied to 90% or more of the data under discussion. The remaining cases were taken to be exceptions unless further subregularities could be established for these minority choices. Distributions in which none of the options reached categorical level were described either in terms of tendencies or as ‘truly variable’ (or simply ‘variable’). The latter label was applied if no clear preference could be established among the majority options in a given subset. More specifically, the choice between two alternative options was termed ‘truly variable’ if neither option exceeded 60% or fell below 40%.

3.7 Summary This chapter addressed the various methodological considerations that guided the decisions on creole languages for investigation and the selection of source texts for individual creoles. The chosen languages are all Caribbean creoles which developed in contact situations involving European lexifier languages and African substrate languages. Pairs of creoles with the same lexifier were chosen and the kind of contact varied within a pair where possible. Early creole data were used wherever pertinent sources were available and reliably indicated phonological characteristics. The corpora used in the present study were compiled from the creole word types in the source texts ignoring purely orthographic variation as well as variation in terms of vowel quality and quantity. The chapter also dealt with the question of which language varieties should be considered as having been dominant in the contact settings. The assumptions on influential lexifier varieties were relevant for the coding of etymon forms in the corpora. The main substrate languages identified here and their role in the formation of the creoles will be considered in more detail in chapter 8 below. Finally, I introduced the basic methodology employed in the empirical investigation of creole syllable structure. The data were coded according to two different classes of variables, structural variables representing the phonological structure of attested creole words and assumed etyma, and process variables representing different processes of phonotactic restructuring that relate the given creole word to its respective purported etymon. In the coding of these variables I took care to avoid an overinterpretation of the data. Where alternative possibilities existed regarding the phonological structure of the etymon and/or restructuring processes, as a general rule, I opted for the more conservative choice, assuming

44 | 3 Data and Methodology minimal changes in the creole. The two classes of variables thus coded entered the statistical analyses conducted for subsets of the corpora. More specifically, it was tested whether the creole treatment of some etymon structure (as coded in process variables) can be predicted on the basis of its phonological properties (as coded in structural variables for the etyma). In these analyses, the application of a given process was considered categorical if it applied in at least 90% of the pertinent cases, allowing for a certain degree of random noise in the data. With these methodological issues covered, we can now turn to the results of the data analysis. The three following chapters deal with syllable structure and phonotactic restructuring in the Dutch-based, English-based and French-based creoles, respectively.

4 Syllable structure and phonotactic restructuring in the Dutch-based creoles 4.1 Introduction This chapter offers a detailed discussion of syllable structure and phonotactic restructuring in the two Dutch-based creoles under investigation, Berbice Dutch (BD) and Negerhollands (NH). We will see that Berbice Dutch readily permits certain types of complex onsets as well as consonant sequences in word-internal position, but is highly restrictive with respect to word-final codas. Negerhollands tolerates different types of complex constituents in all positions. Most restrictions on syllable structure prove to be probabilistic rather than categorical for this creole. The chapter is structured as follows. First, I will give an overview of the relevant literature on syllable structure in Berbice Dutch and Negerhollands in section 4.2. Section 4.3 will then cover methodological aspects that go beyond the general remarks in chapter 3 above and apply to the Dutch-based creoles in particular. The results of the investigation of phonotactic patterns and restructuring in Berbice Dutch and Negerhollands follow in sections 4.4 and 4.5, respectively. These sections will give a detailed account of both the retention of lexifier structures and the adjustments to lexifier syllables which we find in the creoles. The chapter ends with a brief comparison of syllable structure in the two Dutch-based creoles (4.6).

4.2 Previous studies The sources for my data on Negerhollands and Berbice Dutch both include descriptions of the syllable structure of the respective creole. There were several reasons why these accounts could not simply be adopted to compare them to those of the other creole languages investigated in this study. The description of BD syllable structure in Kouwenberg (1994) does not include any quantitative results, nor does it systematically address the relation between lexifier and creole structures. Since Kouwenberg’s focus is on the creole facts in their own right, neither the extent of restructuring nor the types of changes that applied to the etyma are dealt with. The results on Negerhollands syllables given in Sabino (1990) present a different case. The first problem that arises is that the author does not indicate which

46 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles etyma were assumed for individual creole words. The appendix includes correlates from various potential donor languages for each NH root, but no indication as to the most likely etymon. The second problem is that Sabino (1990) is interested only in those parts of the data which show variation. Consequently, items with invariant realisations in the creole go undiscussed. For a comparison of lexifier and creole syllables as intended in this study, however, all data need to be taken into account. Two additional studies on Negerhollands syllable structure are worth mentioning here, Sabino (1993) and Stolz (1986). In the following, I will give a brief outline of each of these studies and point out aspects which rendered the respective results unsuitable for use in the present study. Sabino (1993) presents the results of an investigation of Negerhollands word structure in general and initial clusters in particular. Regarding the first aspect, in order to test whether words with simple (C)V syllables are indeed the preferred structure in Negerhollands (and its substrate languages), Sabino compares the frequencies of words with strictly alternating structures in the creole with those she finds for all relevant superstrate and substrate languages. Her counts show a clear predominance of words with (C)V syllables in Negerhollands (80%). In fact, the count is disproportionately higher than in the superstrate languages (6-16%) and higher even than in some of the substrate languages (56-91%) (cf. Sabino 1993: 39). This apparent stark contrast between creole and superstrate is at odds not only with my own analyses, but also with rates of cluster preservation reported by Stolz (1986), which Sabino herself cites. Only for word-final clusters does he find retention rates that fit Sabino’s numbers. For complex onsets or simple codas, on the other hand, the frequency of preservation is much higher (ranging from 62% to 92%). The explanation for this discrepancy lies, it seems, in Sabino’s methodology. From what she says about it, I gather that she compares token counts for Negerhollands with type counts for the other languages. The discussion of the second aspect, word-initial clusters, follows a more qualitative approach. Here, Sabino (1993) provides an overview of all attested word-initial complex onsets and compares the attested cluster types to those of the substrate languages. In addition, three variable rules of onset modification are discussed. Unfortunately, Sabino does not systematically supply application rates for these rules. Apart from a general comment that these were minor rules, more exact quantifications are given for only one of them, so that the overall impact on the creole’s phonotactics remains unclear. Stolz (1986) is a more comprehensive study of Negerhollands phonotactics. Stolz offers an in-depth treatment of the phonological system of Negerhollands as documented in the data collected by de Josselin de Jong (1924, 1925). In his discussion of Negerhollands phonemes and phonological processes, Stolz is primarily

4.3 Methodological issues

| 47

interested in a qualitative account. Thus, although he covers various restructuring mechanisms and gives examples of their application, he does not usually indicate how frequent individual processes are. More statistically oriented is the section on morpheme structure (cf. Stolz 1986: 100ff). Concentrating on the fate of (sequences of) consonants, the author provides detailed lists of corresponding structures in lexifier and creole, complete with type counts. Stolz’s account gives a good overview of the respective overall degrees of preservation and modification of lexifier structures. What is unfortunate, however, is the high degree of generalisation. The only qualitative distinction that is made in this section is between consonants and vowels. Thus, for instance, highly diverse cluster types such as plosive-liquid and [s]-plosive sequences are pooled, obscuring potentially diverse behaviour. Consequently, it remains unclear to what extent the nature of the individual consonants or clusters plays a role for the creole outcome.

4.3 Methodological issues This section addresses language-specific aspects which were considered in the segmental codings. I will first discuss the more general problem of which forms the creole verbs should be derived from, before turning to variation in and historical development of the main lexifier. Which forms are the verbs derived from? In his discussion of Negerhollands etyma, Stolz (1984) argues that the Dutch imperative is the most likely source for the majority of creole verbs. The main piece of evidence for his analysis comes from the creole reflexes of prefixed Dutch verbs, which "keep the prefix in Dutch infinitives, but separate the verb stem and prefix in imperatives [. . . ]" (Stolz 1984: 48f). In Negerhollands, these former prefixes invariably follow the verb stem, i.e. they are in a position which is typical for a Dutch imperative form. Two representative examples from Negerhollands are given in (1) with both the Dutch infinitive and imperative forms for comparison.

48 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (1)

Root stikui paso

Dutch infinitive uitsteken oppassen

Dutch imperative steek uit! pas op!

Although not discussed by Stolz, Berbice Dutch provides comparable facts. Given these observations, I have assumed the imperative form to be the relevant etymon for most Dutch-derived verbs in both Negerhollands and Berbice Dutch. Only very few exceptions exist for which the phonological form of the creole word is more similar to the Dutch infinitive or first person singular form than to the imperative. In these few cases I have followed the general rationale of conservative interpretation and have coded the phonologically closest form as etymon. Word-initial epenthetic [h] Weijnen (1991: 124f) reports [h]-insertion before initial stressed vowels in some dialects of Dutch, including Zeeland. A small number of relevant cases were coded as containing initial [h] already in the etymon and simply maintaining it in the creole. Syllabic consonants In modern Dutch, we find variation between syllabic sonorants and sequences of [ə] plus non-syllabic sonorant in certain unstressed positions. The data on the Dutch-based creoles include numerous etyma which could potentially have contained such syllabic consonants in word-final position. Unfortunately, none of the historical records on Dutch that I consulted discusses the issue, making it impossible to establish for certain whether or not syllabic consonants were common already in the 17th century. Stolz (1986: 68) assumes this when he says that the syllabic consonants in the Dutch-based creoles represent adoptions of Dutch phonetic variants of the respective schwa-C sequences. Hypothesising variation in lexifier pronunciation is consistent with the creole facts. Both Negerhollands and Berbice Dutch show attestations of syllabic [l] ˈ and [n], but also many cases in which derivation from variants with [əl] and [ən] ˈ seems more likely (cf. e.g. regen > rign ~ regən ‘rain’ and middel > midl ~ midəl ‘middle’). Crucially, none of the attested syllabic consonants in the creoles occurs in an environment where it would not have been possible in Dutch.¹ Basically the same type of variation can be assumed for English etyma with underlying word-final schwa plus sonorant. Here, we even have historical evi-

1 Assuming, again, that the regularities were basically the same in the 17th century as they are now.

4.3 Methodological issues

| 49

dence that the phonetic realisation varied already in 16th and 17th century pronunciations (cf. Dobson 1968: 889ff; see also section 5.3 below for a more detailed discussion). In order to avoid an overinterpretation of the data, I coded the pertinent English and Dutch etyma as containing a syllabic consonant if (and only if) the creole reflex also featured such a syllabic segment. Vocalised liquids Sabino (1990) discusses another aspect of variation that affects underlying wordfinal /ər/ and /əl/ in Dutch items. Some of the Negerhollands reflexes of such etyma exhibit final [u] instead of a syllabic liquid or a schwa-C sequence. Sabino (1990: 141f) argues that these vocalisations, as she labels them, can be traced back to variation in lexifier realisations. I followed her suggestion in my coding of the Negerhollands data and treated the pertinent etyma as vowel-final. Absence of the final liquid in these items was thus not attributed to phonotactic restrictions imposed by the creole. Simple velar nasal or nasal plus homorganic stop? Words such as lang ‘long’ or ding ‘thing’, which in modern Standard Dutch end in a velar nasal [ŋ], developed from earlier forms with a final velar stop following the nasal. For Middle Dutch, Goossens (1974) still considers [ŋk] the regular variant.² He makes no clear statement on the period he refers to, so that the only thing we can say with considerable certainty is that in the 15th century, the final stop must have been regularly present. Whether this pronunciation prevailed in 16th century Dutch in general cannot be clearly established. The loss of the plosive in forms with underlying /nɡ/ in the emerging standard Dutch could be either a 16th century or an early 17th century development. For 17th century Dutch, Vekeman and Ecke (1992: 120) assume regular absence of the plosive. Note, however, that reduction to [ŋ] did not apply equally consistently in all dialects. For instance, Zeelandic appears to have maintained [ŋk] longer than Standard Dutch at least in some lexical items. Weijnen (1991) assumes lexical diffusion for the process. For individual lexical items (e.g. ding), he finds even 20th century attestations of [ŋk] for underlying /ŋɡ/ in some areas of the Netherlands, including Zeeland.

2 The voicelessness of the final stop results from final devoicing. It is important to distinguish these forms from those which also underlyingly contain a nasal plus voiceless plosive /nk/. The latter cluster type was not affected by reduction and surfaces as [ŋk] also in modern Dutch.

50 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles Given this situation, I assumed that most of the pertinent etyma featured final [ŋ] rather than a nasal-plosive cluster. Exceptions were made only in individual cases in which there was evidence for a Zeelandic variant with final [ŋk]. An analogous argument can be made for word-internal position, where Dutch used to allow several different sequences of a nasal consonant followed by a homorganic voiced stop. Today, the only surviving sequence of this type is [nd]. The formerly equally acceptable [mb] and [ŋɡ] have undergone reduction since Middle Dutch times and have become [m] and [ŋ], respectively. Again, as in the case of word-final [ŋɡ], individual lexical items may show retention of the velar cluster in Zeelandic dialects and were coded accordingly. In all other etyma I assumed absence of the cluster. Intrusive schwa in final clusters In modern Dutch, variable schwa epenthesis is attested in coda clusters consisting of a liquid and a non-homorganic consonant (cf. Booij 1995: 127f.). The process is reported to be dialect-group specific today (Hinskens, p.c.) and probably has been for a long time. This ties in well with lexical evidence from the appendix in Sabino (1990), which suggests that Zeeland was among those areas where vowel epenthesis occurred. Thus, not in all cases in which we find apparent vowel insertion need this be a creole innovation. Turning to the creole facts, in the Berbice Dutch data, only two of the pertinent cluster types show vowel-epenthesis, namely [lk] and [rm]. What is more, the two cluster types are resolved in all instances. In contrast, not a single case of epenthesis is attested for any of the other potential candidates, such as [lp], [rp], [lf] or [rk]. Since such a restriction is not attested for Dutch and seems unlikely to arise by chance, I have interpreted the insertions as restructuring processes taking place in Berbice Dutch. For Negerhollands, the situation is different. Epenthetic vowels variably occur in several different types of liquid-initial clusters, which renders an etymological explanation more likely than in the case of Berbice Dutch. However, among the affected clusters we also find one occurrence of [lt], which, being a homorganic cluster, should not be affected by vowel epenthesis in Dutch. This, of course, casts doubt also on the etymological interpretation of the remaining epenthetic vowels. In the absence of more reliable information about the frequency of vowel insertions in 17th century dialects I coded the insertion of a vowel as lexifier process only if I had lexical evidence for the presence of such a vowel (usually in a Zeelandic variant). In all other items the vowel was treated as a creole innovation.

4.3 Methodological issues

| 51

Cluster reduction Weijnen (1991) mentions assimilation and deletion processes in several types of clusters in relevant lexifier dialects. One of these processes is loss of postvocalic [r] before coronal plosives and stridents, i.e. mostly before [t,d] and [s,z] (cf. Weijnen 1991: 140ff). The majority of etyma which potentially contained such clusters surface without the rhotic in Berbice Dutch and Negerhollands. In these cases, loss of [r] was assumed to have taken place in the lexifier rather than in the creole. The pertinent etyma were thus coded as ending in a simple consonant, not a cluster. Similarly, variable [t]-deletion in the realisations of the underlying final clusters /ts/ and /st/ was not uncommon in 17th century Dutch, although not as frequent as it is today (cf. Weijnen 1991: 117ff). Den Besten (1987: 73) claims that final [t]-deletion after non-sonorants was wide-spread in colloquial Hollandic Dutch. For underlying /d/, Vekeman and Ecke (1992: 119) report possible loss of the plosive in word-final /ld/, /rd/ and /nd/. For all of the above sequences, absence of the plosive was assumed in the etymon if the creole word showed no evidence of its presence. Thus, root-etymon pairs such as faʃi < vast ‘tie, fasten’ were classified as exhibiting vowel paragoge, but no consonant deletion. Plosive deletions are reported also for word-medial clusters. According to Weijnen (1991: 136,140), medial liquid-[d] as well as [nd]-clusters were affected by assimilation processes which effectively removed the plosive. [t]-deletion often applied in cases in which [t] was preceded by an obstruent and followed by another consonant (i.e. in obs-[t]-C sequences). Creole realisations lacking a plosive in pertinent contexts were interpreted as faithful copies of an already simplified lexifier input. Apart from plosives, also fricatives could be dropped in certain environments. The above-mentioned change [ts] > [s] is in fact part of a more general process of assimilation that, according to Vekeman and Ecke (1992: 120), variably applied also to other postvocalic obstruents preceding [s]. Consequently, in the coding of the data, simplified realisations such as [s] for underlying /χs/ or /fs/ (e.g. in stisli < stijfsel ‘starch’) were ascribed to reduction processes in the input rather than in the creole. Other reduction processes A reduction process that sets modern Dutch apart from Middle Dutch is the loss of intervocalic [d] after long vowels or diphthongs and before schwa. Goossens (1974: 85f) reports that also the following schwa was usually lost if it was in final position, so that, effectively, we are dealing with the loss of the entire final syllable in these cases (e.g. koude > kou, schade > scha). Loss of [də] is also attested in items in which a consonant other than [n] followed the schwa, leading to contractions

52 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles such as beudel > beul and roeder > roer. Before [n], schwa was usually retained, as, for instance, in snijden [snaɪdən] > snijen [snaɪən]. In some dialects of Dutch, items of the latter type show a glide [j] in place of the lost plosive. Another process that is worth mentioning concerns word-final singleton rhotics. According to Weijnen (1991: 143), loss of the rhotic occurs (and occurred) in Zeelandic pronunciations of non-inflective words such as weer ‘again’ or maar ‘but’. In all cases in which the creole words lacked the pertinent sound(s), I interpreted this as a reflection of simplified lexifier structures, not as a change in the creole. The etyma were coded accordingly.

4.4 Results I: Berbice Dutch In this section, I will give a detailed account of the syllabic structures we find in the Berbice Dutch data and describe the processes that relate the creole words³ to their assumed etyma. Overall, it will be shown that Berbice Dutch is highly restrictive with respect to the right word-edge, but more permissive with regard to structures in word-initial and word-internal positions. I will first look at the left and right word-edges in sections 4.4.1 and 4.4.2 respectively, before turning to the investigation of word-internal positions in section 4.4.3. A brief summary of the findings will be provided in section 4.4.4.

4.4.1 Word-initial onsets in Berbice Dutch The bar plot in figure 4.1 gives an overview of onset structures in Berbice Dutch roots and their assumed etyma. The following three types of structures are distinguished: simple onsets, complex onsets consisting of two or three consonants and cases in which the initial syllable starts in a vowel, i.e. has no onset. The frequencies of the respective structures can be read off the vertical axis.⁴ In addition, the percentages are given for each of the subsets. The bar plot reveals a striking similarity in the distribution of onset structures in the two data sets. The minimal differences in numbers are non-significant (Pear3 The term ‘roots’ will alternatively be used for creole words in the following. 4 In this and all following graphs, the frequency count for a particular structure equals the number of (type-based) corpus entries in which that structure occurs. An analoguous interpretation applies to the frequency counts for restructuring processes given in tables and diagrams in later sections. For more details on corpus entries, see chapter 3.4.4.

4.4 Results I: Berbice Dutch |

400 300

frequency

70.4%

100

200

300 200

71.4%

100

frequency

400

500

Berbice Dutch etyma

500

Berbice Dutch roots

53

21.4%

20.3% 9.3%

0

0

7.1% simple

complex type of onset

no onset

simple

complex no onset type of onset

Fig. 4.1: Word-initial structures in Berbice Dutch roots and etyma (N=700)

son’s χ2 =2.23 df = 2, p=0.3287). In both roots and etyma, the most frequent type of word-initial structure is a simple onset. It occurs in over 70% of the cases in creole and lexifier words. Although much less common overall, complex onsets are also attested in considerable numbers in both data sets (21.4% and 20.3% respectively). Vowel-initial words constitute the smallest group in the creole as well as in the lexifier data (under 10% in either set). At this level of abstraction, thus, Berbice Dutch onset structures are comparable to the onset structures occurring in the input. If we were interested only in a rough typological comparison that focusses on the macro-level distinctions used above, then we could stop the analysis at this point. There is no reason to believe that the creole is different from the lexifier in this respect, at least not as far as word-initial structures are concerned. However, such a comparion gives us only part of the story and neglects aspects which are important for the purposes of the present study. First, it focusses on the outcome of creolisation, but tells us nothing about the processes involved. As we will see, comparable distributions at this very general level cannot necessarily always be interpreted as a mere retention of lexifier structures, but may instead result from a combination of different restructuring processes. In other words, the distribution of structures alone does not allow us to draw any conclusions on whether or not lexifier words were restructured in creolisation. Second, a comparison of the frequencies of such broad structural categories as initial clusters and singleton consonants may be overly influenced by common structures. For instance, if

54 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles a certain cluster type occurs only rarely in the input and is lost completely in the creole, then this loss will not have a strong impact on the overall distribution. Yet, knowledge of such such developments is important if we want to find out which structures are permitted in the creoles and which ones are not. To give a hypothetical example, it may be relatively uninteresting to learn that 5 out of 100 obstruent-approximant clusters are restructured. If, on the other hand, 5 out of 5 plosive-nasal clusters are restructured, that surely is remarkable. Yet, the impact (or lack thereof) on the difference in overall distributions is the same in both cases. For the reasons outlined above, nonsignificant differences in the overall distributions of structures in the creole words and their etyma cannot automatically be taken as an indication that no noteworthy changes took place in creolisation. Only a more detailed analysis can uncover whether or not that is the case. Chisquare statistics will still be given for the comparison of distributions in this as well as in later chapters, because they may be of interest from a typological point of view. However, more detailed analyses will be carried out regardless of the outcome of these chi-square tests. Returning to the situation at hand, what the overview in figure 4.1 tells us about complex onsets in Berbice Dutch is that they are not ruled out per se, but instead are about as frequent in the creole as in the etyma. What the overview does not tell us, on the other hand, is whether the complex onsets in the creole are in fact reflexes of the complex onsets in the lexifier, i.e. whether the creole structures are simply faithful copies of input structures. A closer look at the correspondences between onset structures in the etyma and onset structures in the creole suggests a slightly more complex situation. The bar plot in figure 4.2 provides a graphical representation of these correspondences. It reveals that, despite the comparable numbers overall, there is a certain degree of mismatch in structures with some roots showing onset structures that are different from those of their respective etyma. The correspondence plot can be interpreted as follows. Different types of onset structures in the etyma are plotted as separate bars. Within each bar, sections with different shadings indicate the different types of onset structures that occur in the corresponding creole reflexes. The rate at which Berbice Dutch roots surface with the same type of onset structure as their etyma (henceforth called ‘correspondence rate’) is given as percentage for each of the lexifier categories. For instance, the left-hand bar provides information about word-initial structures in creole words whose etyma have a simple onset. The given correspondence rate of 95.7% indicates that 95.7% of all lexifier words with simple onsets have creole reflexes that also exhibit a simple onset. The remaining 4.3% of etyma in this set have creole reflexes with non-corresponding onset structures, i.e. the creole

500 400

simple complex no onset

55

300

creole onsets

200

95.7%

100

frequency

4.4 Results I: Berbice Dutch |

92.3%

0

73.8% simple

complex

no onset

onsets in etyma

Fig. 4.2: Correspondences between word-initial structures in Berbice Dutch roots and etyma (N=700)

reflexes show a different type of onset structure, in this case mostly complex onsets (indicated by middle grey shading). Two main observations can be made on the basis of this bar plot. First, the lowest correspondence rate occurs with vowel-initial lexifier words (73.8%, cf. the rightmost bar). Over a quarter of such items end up with a simple onset (light shading) in Berbice Dutch instead of retaining the vowel-initial input. Second, we can observe a kind of trade-off in numbers between the groups of words with simple and complex onsets, respectively. Neither simple nor complex onsets have a correspondence rate of 100%. On the one hand, the middle bar shows that a small number of etyma with complex onsets surface with a simple rather than a complex onset in Berbice Dutch (light shading, 7.7%, N=11). On the other hand, the leftmost bar reveals also changes in the opposite direction. A subset of lexifier words with simple onsets have creole reflexes with complex onsets (middle grey shading, 4.3%, N=19). While the changes from complex to simple onset readily evoke structural explanations, the opposite change seems more puzzling. Why would a language create new, complex onsets where the input has simpler structures? With regard to onset structures alone, these changes appear unmotivated. Thus, we must look to other structures for a possible trigger. A closer look at the pertinent data reveals that the newly created complex creole onsets can indeed be interpreted as a by-product of repair mechanism targeting other (i.e. non-onset) structures

56 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles in the etyma, in particular unstressed vowels and coda consonants. Representative examples are listed in (2).⁵ Unless indicated otherwise, the etyma are Dutch words. Note that Ijo stands for Eastern Ijo here and in the remainder of the chapter. (2)

Newly created complex onsets in Berbice Dutch Root Etymon Gloss a. bwaru < bewaren ‘preserve’ froto < verrot ‘rotten, rot’ gliki < gelijk ‘on a level’ b. frugɛtɛ < vergeten ‘forget’ frukopu < verkopen ‘sell’ frustan < verstaan ‘understand’ c. bwa < búó ‘leg, foot’ (< Ijo) dwesn < duizend ‘thousand’ mjɛ/mja < mie ‘do, make’ (< Ijo)

In the data in (2a), the complex root onsets result from deletion of the first, unstressed lexifier vowel. Both the preceding and the following consonants are retained and form a complex onset in the creole word. Crucially, this occurs only with etyma that have neither a coda consonant in the first syllable nor a complex onset in the second syllable. The set of examples in (2b) illustrate a slightly different case. All of the pertinent etyma include the Dutch prefix ver- followed by a syllable with either a simple or a complex onset. Consequently, all etyma in this set exhibit a word-internal coda-[r]. In the corresponding roots, the rhotic has been removed from coda position through metathesis and is syllabified as the second element of a new complex onset [fr]. Finally, a third group of etyma show evidence of glide formation (cf. (2c)). All etyma in this subset feature either a diphthong or a heterosyllabic vocalic sequence after the initial onset consonant. In the creole words, the first vocalic element in the sequence is realised as a glide [w] or [j], resulting in new obstruent-glide onsets. It is important to note that all newly created complex onsets are sequences of either obstruent-glide or obstruent-liquid. As I will show below, these types of onset structures are readily retained in Berbice Dutch also where they occur in the etyma.

5 Throughout this book, creole words will be presented in source text notation. Phonetic representations will be given only in cases where the sound structure cannot readily be inferred from this notation. The use of orthographic and phonetic representations of lexifier words follows the same reasoning.

4.4 Results I: Berbice Dutch |

57

Having clarified the puzzling case of complex onset creation, we can now turn to those changes which can in fact be interpreted as being triggered by dispreferred onset structures, namely changes that lead to the reduction of (certain types of) complex onsets and onsetless syllables. I will deal with each in turn.

4.4.1.1 Complex onsets The data set for the investigation of complex onsets consists of 142 etyma and their creole reflexes. As we saw in the overview above, the overwhelming majority of complex lexifier onsets is retained in Berbice Dutch. Only in 9.9% of the pertinent cases (14 out of 142) do we find restructuring. Note that this number differs slightly from that of non-corresponding structures in the correspondence plot in figure 4.2 above (7.7% for complex lexifier onsets). This discrepancy results from the different degrees of detail in the two analyses. In particular, the correspondence plot treats all complex onsets alike, regardless of the number of segments involved. There are, however, lexifier onsets that undergo reduction, but still surface with a complex onset in Berbice Dutch. This can happen with trisegmental lexifier onsets that are reduced to disegmental clusters in the creole. In the correspondence plot, these cases are counted as showing corresponding structures (both have a complex onset). In the investigation of phonotactic restructuring, on the other hand, the same cases fall into one of the groups showing repair strategies.⁶ The table in (3) provides a detailed account of the changes that apply, listing the respective frequencies and percentages of the individual restructuring processes, including the category ‘none’ to represent retention of lexifier structures. Note that the percentages listed for individual processes add up to only 99.9% rather than the 100% given as sum. This is merely an effect of rounding and was therefore ignored.

6 Analoguous cases of discrepancies between correspondence plots and the retention rate in the more detailed analysis of phonotactic restructuring occur also in later sections. Unless they are of particular interest, these cases will not be commented on in detail.

58 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (3)

Processes affecting complex lexifier onsets in BD Process

N

%

none consonant deletion vowel epenthesis metathesis

128 10 3 1

90.1 7.0 2.1 0.7

Σ

142

100.0

Among the repair strategies, deletion of one of the onset consonants is the preferred option, but we also find a small number of cases in which an epenthetic vowel is inserted between the two members of the complex onset. Apart from these two patterns, only a single case of metathesis is attested. The root-etymon pairs in (4) illustrate the observed restructuring mechanisms, listing examples of retention of the complex onset in (4a), for consonant deletion in (4b) and for vowel insertion in (4c). For the sake of completeness, (4d) gives the single case of metathesis, which, however, will be shown below to be exceptional.⁷ (4)

Treatment of complex lexifier onsets in Berbice Dutch Root Etymon Gloss a. fluru < vloer ‘floor’ spigri < spijker ‘nail’ smelti < smelten ‘melt’ twe < twee ‘two’ b. pɛlɛ < spelen ‘play’ ten < steen ‘stone’ c. kini < knie ‘knee’ kunopu < knôôpe ‘button’ d. purbɛrɛ < proberen ‘taste’

In order to find out whether the treatment of complex lexifier onsets is dependent on their structural make-up, I conducted a CHAID analysis (cf. chapter 3.6) with

7 Here, as in all other discussions of restructuring processes throughout this book, only a limited number of examples is given for each category discussed. The purpose of these examples is to illustrate the observed processes and to give the reader an idea of the types of etyma affected by the process under discussion. I aimed for a more or less balanced selection. Therefore, also smaller subsets are not listed in their entirety. The number of examples was enlarged for a given category only if it was felt that the wide range of different structures or treatment options involved could not otherwise be adequately represented.

4.4 Results I: Berbice Dutch |

59

restructuring process as the dependent variable. All variables encoding structural properties such as manner and place of articulation of the segments involved were fed into the model as potential predictor variables.⁸ The tree-like structure in figure 4.3 illustrates the results of the CHAID analysis. It shows that the data set can indeed be split into subsets that differ significantly with regard to the restructuring of complex lexifier onsets. The manner of the second consonant emerges as the best predictor. Restructuring of word-initial complex onsets Node 0 Process % none 90.14 C-deletion 7.04 V-epenthesis 2.11 metathesis 0.70 Total (100.00)

N 128 10 3 1 142

Manner of second cluster member Adj. P-value=0.0001, Chi-square=27.5195, df=6 approximant Node 1 Process % N none 96.70 88 C-deletion 1.10 1 V-epenthesis 1.10 1 metathesis 1.10 1 Total (64.08) 91

obstruent Node 2 Process % N none 80.43 37 C-deletion 19.57 9 V-epenthesis 0.00 0 metathesis 0.00 0 Total (32.39) 46

nasal Node 3 Process % N none 60.00 3 C-deletion 0.00 0 V-epenthesis 40.00 2 metathesis 0.00 0 Total (3.52) 5

Fig. 4.3: Treatment of complex lexifier onsets in Berbice Dutch

The figure can be read as follows. The information given in node 0 corresponds to the overview of restructuring processes in (3) above. The majority choice is highlighted by grey shading in this and all other nodes. Underneath the box for node 0, we find the variable which best predicts the choice of treatment, in this case the manner specification of the second consonant in the cluster. The data set can thus be subdivided into three subsets which are characterised by different manner specifications of C2 and which show significant differences with respect to

8 Recall from the methodology chapter, too, that in all analyses involving consonant clusters, only sequences of two consonants could be taken into account. Therefore, in the small number of trisegmental onsets, only the two edgemost consonants were considered for the coding of structural properties.

60 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles restructuring choices. The first thing that is important to note is that retention of lexifier structures is the majority option in all groups (highlighted by grey shading). Thus, irrespective of their structural make-up, complex onsets have a greater chance of surfacing in Berbice Dutch than of being repaired. However, retention is not equally likely within all subsets, and, additionally, there is variation in the preferred repair option across groups. Complex onsets in which the second consonant is an approximant (cf. node 1) show the highest rate of retention. In fact, restructuring occurs only in a few exceptional cases (including the above-mentioned item with metathesis), so that retention of the complex onset can be considered categorical in this subset. This group includes obstruent-liquid and obstruent-glide sequences. Nasal-glide onsets, which also have an approximant C2 and are permitted in the lexifier, do not occur in the set of etyma. However, based on the fact that the cluster [mj] is among the newly created onset sequences (cf. (2)), we can hypothesise that nasal-glide onsets are also tolerated in the creole. The second major group of onset sequences (node 2) is characterised by an obstruent C2 and overall shows a slightly lower rate of retention than clusters with an approximant in second position. If restructuring applies, the option of choice is consonant deletion. This group of clusters almost exclusively consists of [s]obstruent clusters. In cases of consonant deletion, it is typically the initial [s] that is lost. The analysis yielded no further distinctions in this subset, so that we have to assume that the remaining variation is not caused by differences in the structural make-up of the etyma. For this group of clusters we thus find a preferred, but not a categorical choice of treatment. The final subset to be discussed comprises sequences in which the second consonant is a nasal (cf. node 3), i.e. obstruent-nasal onsets. The analysis suggests variable retention and repair by insertion of an epenthetic vowel between the two consonants. Three different clusters occur in this group, namely [kn], [sn] and [sm]. The two attested cases of vowel insertion both apply to [kn]-clusters, whereas the three instances of [s]-nasal onsets are retained in the Berbice Dutch reflexes (see (5) below for examples). Unfortunately, these five types are the only ones with pertinent clusters among the etyma. Due to the small numbers we cannot say for sure whether onsets of the type obstruent-nasal do in fact form a homogeneous group or whether restructuring should only be postulated for the sequence [kn]. In order to avoid an overinterpretation of the data I will assume that [s]-nasal onsets do not trigger restructuring in Berbice Dutch. The list in (5) gives examples of all of the above groups. The majority of rootetymon pairs is repeated here from (4), regrouped according to the results of the CHAID analysis and complemented by some additional examples to allow for a better comparison of the subsets. The data in (5a) illustrate categorical cluster

4.4 Results I: Berbice Dutch |

61

retention in obstruent-approximant clusters. (5b) exemplifies the two alternatives that occur with initial [s]-obstruent clusters, retention and simplification. Finally, (5c) gives one example each for all attested types of obstruent-nasal onsets and their creole reflexes. (5)

Treatment of complex lexifier onsets in Berbice Dutch – regrouped Root Etymon Gloss a. obstruent-approximant fluru < vloer ‘floor’ trapu < trap ‘stairs’ twe < twee ‘two’ b. [s]-obstruent skrifu < schrijven ‘write’ spigri < spijker ‘nail’ pɛlɛ < spelen ‘play’ ten < steen ‘stone’ c. obstruent-nasal kini < knie ‘knee’ smelti < smelten ‘melt’ snorka < snurken ‘snore’

Summarising the findings for complex onsets, although the overall frequency of restructuring is low and most complex onsets are retained in the creole, two welldefined categories of targeted sequences emerge. Consonant deletion applies to a minority of [s]-obstruent onsets, while initial [kn]-clusters are broken up by vowel epenthesis.

4.4.1.2 Vowel-initial words Having investigated the treatment of complex word-initial onsets, I now turn to the opposite case, lexifier words lacking an onset in the initial syllable, i.e. vowelinitial words. The total number of pertinent cases in the data set is 65. We already saw above in the correspondence plot in figure 4.2 that over 25% of such words undergo restructuring and end up with a simple onset. The table in (6) additionally provides details about the attested repair options and their respective frequencies.

62 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (6)

Processes affecting vowel-initial etyma in BD Process

N

%

none vowel deletion consonant epenthesis

48 13 4

73.8 20.0 6.2

Σ

65

100.0

Again, we find two alternative options besides the majority choice of retention (‘none’), and, again, the deletion of material is preferred to insertion. The initial vowel is lost from 20% of the relevant etyma, whereas epenthetic consonants only occur in around 6% of the cases. Examples of the retention of the initial vowel and the two repair strategies are given in (7). The donor language of each etymon is listed in the rightmost column since a considerable number of the pertinent words are not derived from Dutch, but from Eastern Ijo (abbreviated as ‘Ijo’ below). (7)

Treatment of vowel-initial etyma in Berbice Dutch Root Etymon Gloss Donor a. apara < ápárá ‘skin’ Ijo iʃiri < ijzer ‘iron’ Dutch oli < olie ‘oil’ Dutch b. funggru < ofúnguru ‘mouse, rat’ Ijo kiba < íkíbá ‘piece, short’ Ijo nanaʃi < ananas ‘pineapple’ Dutch c. juku < íkú ‘louse’ Ijo juru < ure ‘hour’ Dutch

As discussed earlier (cf. chapter 3.6), vocalic segments were not coded for different articulatory properties, so that a statistical analysis of the kind conducted for complex onsets is not possible for vowel-initial words. Consequently, it cannot be established whether vowel quality has an influence on the treatment of the respective lexifier word. However, two other properties are worth considering in this connection. The first one is stress. It is a common phenomenon in many languages that vowels in syllables not carrying lexical stress are reduced or even completely omitted in casual or fast speech (e.g. Peterson and Lehiste (1960), Delattre (1969), Rietveld and Koopmans-van Beinum (1987), van Bergem (1993), Beckman (1996)). It might therefore be interesting to test whether lexical stress placement in the etyma plays a role in the choice between retention and repair of initial vowels in the creole.

4.4 Results I: Berbice Dutch |

63

The investigation of stress-influenced processes in Berbice Dutch is complicated by the fact that Eastern Ijo has a tonal rather than a stress system. In words of Ijo origin, we do not find a distinction between lexically stressed and unstressed syllables, but between syllables with high tones and syllables with low tones. What these two oppositions have in common is that they both involve pitch differences. For many languages, pitch is claimed to be one of the acoustic correlates of stress. According to standard reference works, stressed syllables are usually characterised (among other things) by a higher pitch than unstressed syllables (cf. e.g. Laver 1994, Hammond 1999, Odden 2005 for English).⁹ Regarding pitch variation in Kalabari-Ijo, Harry (2004: 22) reports that, apart from the two basic tone levels, we also find a gradual downstep or downdrift among the high tones in a given word. As a consequence, the syllable bearing the first high tone is usually realised with the highest pitch. Given these observations, I will treat the first hightoned syllable in Ijo words as the closest equivalent to the main stressed syllable in European etyma. For the sake of simplicity, I will refer to both alike as ‘stressed’ syllables below. A second factor that could potentially be important in the restructuring of vowel-initial words is word length. Note that vowel deletion in a simple V syllable, as we find it in Berbice Dutch, effectively constitutes the deletion of the entire syllable. Thus, what is triggering the process of vowel deletion in the creole could be a dispreference for longer words (in terms of the number of syllables) rather than a restriction on initial onsetless syllables. Steele and Brousseau (2006) suggest an interpretation along these lines for initial syllable deletion in Haitian Creole and attribute it to a preference for words to maximally consist of a disyllabic foot. Phonologists have proposed that such binary feet or Minimal Words are prosodically unmarked (McCarthy and Prince 1996, 1994). A CHAID analysis with main stress position and the number of syllables as potential predictor variables returned a significant difference in the treatment of mono- or disyllabic words as compared to longer words. The tree-structure in figure 4.4 illustrates the results.¹⁰ In node 1 we can see that short words of only one or two syllables are unlikely to undergo restructuring. Only around 5% of the mono- and disyllabic words sur-

9 Note, however, that such claims may hold only for accented words as most studies of the acoustic correlates of stress do not investigate unaccented words, i.e. words not carrying a pitch accent. For a discussion of the problem, see, e.g., Braun et al. (2011), Plag et al. (2011). 10 For technical reasons the order of categories in the tree diagram is not identical to the one in the frequency table above. The same is true for some other tree diagrams in later sections and chapters.

64 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles Restructuring of vowel-initial words Node 0 Process % N V-deletion 20.00 13 none 73.85 48 C-epenthesis 6.15 4 Total (100.00) 65 Number of syllables in etymon Adj. P-value=0.0000, Chi-square=30.2727, df=2 1-2 Node 1 Process % V-deletion 0.00 none 94.87 C-epenthesis 5.13 Total (60.00)

3-4

N 0 37 2 39

Node 2 Process % V-deletion 50.00 none 42.31 C-epenthesis 7.69 Total (40.00)

N 13 11 2 26

Fig. 4.4: Treatment of vowel-initial words in Berbice Dutch

face with modified onset structures in the creole. If we compare this to the percentages in node 2, we find that lexifier words consisting of three or more syllables show a much more variable behaviour. Retention and restructuring are about equally frequent among these words. Another difference between the two subsets can be seen in the choice of repair strategy. Whereas vowel deletion is the much preferred option in three- or four-syllable words, the only repair that occurs in short words is consonant insertion.¹¹ What might seem puzzling is the occurrence of consonant epenthesis also with longer words. If vowel deletion is motivated mainly by a dispreference for longer words, then why should we find words with three or more syllables in which the onsetless initial syllable is strengthened by insertion of an onset consonant? At this point, it is worth noting an aspect that does not emerge in the tree. Although stress is not a significant predictor for the treatment of vowelinitial etyma, it is not entirely without effect. In the few etyma which gain an onset through consonant epenthesis, the first syllable is always stressed. What we observe here is thus not a repair of onsetless syllables in general, but the

11 For monosyllabic words, of course, vowel deletion would not have been a viable option in the first place.

4.4 Results I: Berbice Dutch |

65

occasional strengthening of initial stressed syllables, a process that may well be grounded in perception. Note also that the two longer words which show consonant epenthesis do not in fact surface with the same number of syllables in Berbice Dutch. In addition to consonant epenthesis, they undergo also other processes of restructuring and end up disyllabic in the creole. The list in (8) provides a rearranged list of examples, ordered according to the two categories that emerged in the CHAID analysis. Retention and occasional consonant epenthesis in short words are illustrated in (8a), whereas variable retention and vowel deletion in longer words are covered by the examples in (8b). (8)

Treatment of vowel-initial etyma in Berbice Dutch – regrouped Root Etymon Gloss Donor a. etymon with one or two syllables iʃiri < ijzer ‘iron’ Dutch oli < olie ‘oil’ Dutch juku < íkú ‘louse’ Ijo juru < ure ‘hour’ Dutch b. etymon with three or more syllables apara < ápárá ‘skin’ Ijo opropo < óprópo ‘pig’ Ijo funggru < ofúnguru ‘mouse, rat’ Ijo kiba < íkíbá ‘piece, short’ Ijo

Summarising the little we can say about the restructuring of vowel-initial words, retention of the onsetless structure is preferred with short etyma of no more than two syllables. In longer words, we find variation between deletion and retention of the initial vowel. Consonant epenthesis appears to be restricted to etyma with initial stress, but is rare even within this subset of the data.

4.4.2 Word-final codas in Berbice Dutch The interpretation of word-final structures is not always unproblematic. In particular, different types of consonantal sequences may not have the same status. Recall from chapter 3 above that I coded as word-final all those segments that follow the final vowel. As a consequence, word-final sequences include possible codas as well as combinations of consonants which involve an onset segment followed by a syllabic consonant in the nucleus and can be schematised as V.CC#. In the first step of my analysis, these different types of sequences are pooled. Inˈ view of this situation, I do not use the label ‘coda’ in the general overview below, but instead employ the more neutral ‘consonant sequences’. Figure 4.5 thus illustrates

66 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles the distribution of word-final consonants and consonant sequences. Three types of structures are distinguished: words that end in an open syllable, i.e. words with no final consonant (‘none’), words with a single word-final consonant (‘C’) and words with two final consonants (‘CC’).¹² Berbice Dutch etyma

500 400 100

100

200

300

frequency

400

1.9%

C

CC

0

9% none

54.6% 35.4% 10%

0

300

89.1%

200

frequency

500

600

600

Berbice Dutch roots

none

C

CC

Fig. 4.5: Word-final consonants in Berbice Dutch roots and etyma (N=700)

Unlike what we saw for word-initial structures, for word-final structures, the difference in distributions across roots and etyma is very highly significant (Pearson’s χ2 =429.95, df=2, p parafin ‘kerosene’). Crucially, this final consonant is a nasal in all three cases. We will see below that nasals are the only type of coda consonant tolerated in word-final position in Berbice Dutch. The middle and rightmost bars in figure 4.6 show much more substantial changes to lexifier structures. They depict the two categories of simple and complex codas, respectively. As we might have expected, the correspondence rate is considerably higher for single word-final consonants than for consonant sequences (14.9% compared to 3.4%). Complex codas, as we saw in the discussion of figure 4.5 above, can be considered illicit in Berbice Dutch. What is worth noting, however, is the fact that practically none of the etyma with complex codas surface with even a simple coda in the creole. Although retention of a simple coda is at least possible in Berbice, reducing a complex coda to a simple one apparently is not a viable repair strategy. Instead, the vast majority of pertinent creole reflexes end in a vowel. In other words, a more complex input appears to yield a simpler output. This observation may seem puzzling at the present level of abstraction. It becomes explicable only if we look at the data in more detail. In the following, I will therefore discuss etyma with simple and complex codas in turn and describe their respective treatment in Berbice Dutch.

4.4.2.1 Simple codas in Berbice Dutch The data contain a total of 382 lexifier words with simple word-final codas. As the comparison in figure 4.5 showed, this type of structure is frequent in the lexifier words, but rare in Berbice Dutch roots. About 85% of all pertinent etyma undergo restructuring. The table in (9) gives an overview of the attested repair strategies which apply to simple codas, starting with coda retention (labelled ‘none’) and listing the remaining processes in order of frequency.

4.4 Results I: Berbice Dutch |

(9)

69

Processes affecting simple codas in BD Process

N

%

none vowel paragoge metathesis vowel paragoge + consonant epenthesis consonant deletion

57 262 45 10 8

14.9 68.6 11.8 2.6 2.1

Σ

382

100.0

The vast majority of cases show vowel paragoge. A non-etymological vowel is added at the end of the word, which causes the former coda consonant to resyllabify in the onset of a newly created syllable. A second type of restructuring process that occurs in a considerable number of etyma is what I have labelled ‘metathesis’. This describes cases in which a coda liquid and the preceding vowel swap places so that the liquid is syllabified as the second member of a complex onset.¹⁴ The two remaining processes in the list are relatively infrequent. Some etyma show an epenthetic consonant in addition to the common paragogic vowel. Finally, in a small number of etyma we observe deletion of the coda consonant. Examples of all attested strategies can be found in the list in (10), with the categories arranged in the same order as in the frequency table above.

14 Note that an alternative interpretation of the data is possible. Instead of assuming liquidvowel metathesis, i.e. a process different from the one attested in the majority of etyma, one could hypothesise that what we observe here is in fact the same process, namely vowel paragoge. In order to explain the word-internal complex onset, we would then have to argue that the vowel in the formerly final syllable is deleted once paragoge has taken place.

70 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (10)

Treatment of simple word-final codas in Berbice Dutch Root Etymon Gloss a. klup < klopp(en) ‘hit, knock’ pin < pijn ‘pain’ b. grasa < gras ‘grass’ hari < haar ‘hair’ pasopo < pas op ‘take care, be careful’ c. diflu < duivel ‘devil’ spigri < spijker ‘nail’ d. pali[ŋɡ]i < pali[ŋ] ‘eel’ trombo < trom ‘drum’ e. giri < gierig ‘stingy’ wabli < watblief ‘beg (your) pardon’

I conducted a CHAID analysis in order to find out whether the structural make-up of the etyma determines the choice of repair strategy. The tree diagram in figure 4.7 illustrates the first resulting split into significantly different subsets. For the sake of clarity, subsequent splits will be presented in separate diagrams. Node 0 repeats the information from (9) above, only with an abbreviated label, ‘V-paragoge +’, for the combination of vowel paragoge and consonant Restructuring of simple codas Node 0 Process % V-paragoge 68.59 metathesis 11.78 V-paragoge + 2.62 C-deletion 2.09 none 14.92 Total (100.00)

N 262 45 10 8 57 382

Manner of coda consonant Adj. P-value=0.0000, Chi-square=294.2844, df=8 obstruent Node 1 Process % N V-paragoge 90.86 179 metathesis 0.00 0 V-paragoge + 0.51 1 C-deletion 4.06 8 none 4.57 9 Total (51.57) 197

liquid Node 2 Process % N V-paragoge 60.68 71 metathesis 37.61 44 V-paragoge + 0.00 0 C-deletion 0.00 0 none 1.71 2 Total (30.63) 117

Fig. 4.7: Treatment of simple codas in Berbice Dutch - first split

nasal Node 3 Process % N V-paragoge 17.65 12 metathesis 1.47 1 V-paragoge + 13.24 9 C-deletion 0.00 0 none 67.65 46 Total (17.80) 68

4.4 Results I: Berbice Dutch |

71

epenthesis. The best predictor variable is given underneath the node. In this case, the manner of articulation of the coda consonant is the most decisive factor for the choice of treatment. The data can be sorted into three main subsets, consisting of etyma with final obstruents, liquids and nasals, respectively. For final obstruents (cf. node 1), the default strategy is vowel paragoge. It applies in over 90% of the pertinent cases. In what follows I will assume that vowel paragoge is categorical with obstruent codas in Berbice Dutch. This analysis is further supported by a remark made by Kouwenberg (1994) with respect to the small number of roots which retain obstruent (or liquid) codas. She argues that these forms do not reflect the earliest form of Berbice Dutch, but result from the loss of a formerly present paragogic vowel and occur only with a specific group of speakers (cf. Kouwenberg 1994: 294). The second node in the tree diagram singles out etyma that end in a liquid (cf. node 2). Liquid codas, like final obstruents, also show a tendency to undergo vowel paragoge, but we observe no categorical behaviour. For this group of coda consonants, metathesis is an alternative option, which is attested in more than a third of all liquid-final etyma. What all obstruent- and liquid-final etyma have in common is that retention of the final consonant is exceptional. We see a rather different picture in node 3, which depicts the treatment of nasal-final etyma. Here, retention of the coda constitutes the majority option (highlighted by grey shading). Although not as frequent as in the other two groups, vowel paragoge is attested also for nasal codas. It affects about one third of the pertinent etyma overall (30.89%), sometimes as the only repair (17.65%) and sometimes in combination with an additional process, the above-mentioned insertion of an epenthetic consonant (13.24%). In the following I will look in turn at final liquids and nasals in more detail and present separate sub-trees that shed more light on the distribution of repair strategies. Liquid-final etyma can be further subdivided into two subsets which differ significantly with respect to their treatment in Berbice Dutch. Figure 4.8 depicts the relevant branch of the tree, reduced for better readability to those restructuring options that occur in this subset. The tree shows that stress acts as a predictor of the choice between different repair strategies. We find a neat division into words with a stressed final syllable (cf. node 4), which almost without exception undergo vowel paragoge, and words

72 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles Restructuring of simple liquid codas Node 2 Process % N V-paragoge 60.68 71 metathesis 37.61 44 none 1.71 2 Total (100.00) 117 Final syllable of etymon stressed? Adj. P-value=0.0000, Chi-square=117.3186, df=2 yes Node 4 Process % N V-paragoge 95.71 67 metathesis 1.43 1 none 2.86 2 Total (59.83) 70

no Node 5 Process % V-paragoge 8.51 metathesis 91.49 none 0.00 Total (40.17)

N 4 43 0 47

Fig. 4.8: Treatment of simple liquid codas in Berbice Dutch

with an unstressed final syllable (cf. node 5), for which metathesis can be considered categorical.¹⁵ For nasal codas, we find an equally clear split. Here, the place of articulation of the nasal is the decisive factor. The tree diagram in figure 4.9 depicts the relevant results of the CHAID analysis. Nasals with dorsal place of articulation are separated from the joined group of labial and coronal nasals. In other words, the velar nasal [ŋ] receives significantly different treatment from the other two members of this class, [m] and [n]. The only available option for [ŋ] (cf. node 6) is a combination of vowel paragoge and insertion of an epenthetic consonant, more precisely, a homorganic stop. In contrast, final [m] and [n] are preferably retained without modification (cf. node 7). A minority of cases show vowel paragoge, but only a single form is attested which additionally surfaces with an epenthetic homorganic stop (trom > trombo).

15 The fact that metathesis occurs almost exclusively in unstressed syllables could be seen as support for the alternative interpretation mentioned earlier, vowel paragoge followed by vowel deletion. All cases of vowel deletion postulated under that hypothesis involve unstressed vowels and we have already seen that deletion of such vowels is, under certain conditions, possible in Berbice Dutch (cf. the creation of new word-initial clusters). In the absence of conclusive evidence, I will maintain ‘metathesis’ as a descriptive label here.

4.4 Results I: Berbice Dutch |

73

Restructuring of simple nasal codas Node 3 Process % V-paragoge 17.65 metathesis 1.47 V-paragoge + 13.24 none 67.65 Total (100.00)

N 12 1 9 46 68

Place of coda consonant Adj. P-value=0,0000, Chi-square=42,9817, df=3 dorsal Node 6 Process % N V-paragoge 0.00 0 metathesis 0.00 0 V-paragoge + 100.00 8 none 0.00 0 Total (11.76) 8

labial, coronal Node 7 Process % V-paragoge 20.00 metathesis 1.67 V-paragoge + 1.67 none 76.67 Total (88.24)

N 12 1 1 46 60

Fig. 4.9: Treatment of simple nasal codas in Berbice Dutch

A rearranged list of examples is given in (11) with each group corresponding to one of the subsets that emerged as significantly different in the analysis. The typical treatment of obstruent codas is illustrated in (11a). (11b) and (11c) show the two (stress-dependent) options for liquid-final etyma, and (11d) and (11e) illustrate the difference in treatment of final [ŋ] compared to the other two nasals.

74 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (11)

Treatment of simple word-final codas in Berbice Dutch – regrouped Root Etymon Gloss a. obstruent grasa < gras ‘grass’ pasopo < pas op ‘take care, be careful’ b. liquid in stressed syllable hari < haar ‘hair’ parisola < parasol ‘umbrella’ c. liquid in unstressed syllable diflu < duivel ‘devil’ spigri < spijker ‘nail’ d. velar nasal [ŋ] pali[ŋ]gi < pali[ŋ] ‘eel’ sa[ŋ]ki < za[ŋ] ‘hymn’ e. other nasal nam < naam ‘name’ pin < pijn ‘pain’ doni < dun ‘thin’

Summarising the results for simple codas, we can conclude that only the nasals [m] and [n] are tolerated word-finally in Berbice Dutch. All other structures are repaired, mostly by vowel paragoge. Only liquids in unstressed syllables undergo metathesis instead. Finally, velar nasals are targeted by a double repair, vowel paragoge and insertion of a homorganic stop.

4.4.2.2 Complex codas in Berbice Dutch Of all lexifier words in the corpus, only 59 display complex codas. In the correspondence plot in figure 4.6 we already saw that such etyma typically have creole reflexes that end in an open syllable. Apart from two exceptional cases in which the cluster is retained, we find no violations of the restrictions on final codas we found above. As we will see, however, the constraints on coda consonants in wordinternal position are less rigid. Regarding repair strategies, vowel paragoge is clearly the preferred option also for complex lexifier codas. As the overview of frequencies in (12) shows, other, or additional restructuring processes are comparatively rare. The list starts with retention of lexifier structures (i.e. no restructuring, indicated by the label ‘none’) and then goes from the most to the least frequent repair mechanism.

4.4 Results I: Berbice Dutch |

(12)

75

Processes affecting complex codas in BD Process

N

%

none vowel paragoge vowel epenthesis V-paragoge + C-deletion V-paragoge + V-epenthesis V-paragoge + metathesis

2 47 3 3 2 2

3.4 79.7 5.1 5.1 3.4 3.4

Σ

59

100.0

Apart from the two exceptional cases in which the cluster surfaces in the creole, all remaining lexifier words with complex codas undergo restructuring. Vowel paragoge applies almost across the board, often as the sole repair (79.7%), but sometimes also in combination with an additional restructuring mechanism (11.9% combined). Only three cases (5.1%) are exempt from the default repair. They undergo vowel epenthesis instead and thus retain a single final consonant in the creole. In all three cases, this final coda consonant is a nasal, resulting from broken-up final [rm]-sequences. Examples illustrating the different processes are listed in (13a-f), given in the same order as in (12) above. IPA symbols are given in some cases where the quality of the segments would not have been obvious from the orthographic representation. (13)

Treatment of complex codas in Berbice Dutch Root Etymon Gloss a. neks < niks ‘nothing’ (colloq.) sent < cent ‘cent’ b. akti < a[xt] ‘eight’ forku < vork ‘fork’ c. darum < darm ‘intestines, bowels’ worum < worm ‘worm’ d. frutu < vru[xt] ‘fruit’ somtiti < somtij[ts] ‘sometimes, perhaps’ e. soloko < zulk ‘such’ wɛlɛkɛ < welk ‘which’ f. balitʃɛ < beslist ‘decided’ hondrutu < honder[t] ‘hundred’

A CHAID analysis was conducted with restructuring process as the dependent variable and the different variables encoding the structural properties of the two

76 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles coda consonants as predictors. The result suggests that three types of complex codas should be distinguished. Figure 4.10 shows the split into subsets. The tree is a simplified version of the actual output. Note that only one category ‘V-paragoge +’ appears in the list of processes. This label covers all those cases in which a second repair applies (cf. the list in (12) above). The analysis was conducted for the uncombined categories, but since all instances of double repairs occur in the same group of etyma (see node 2), no information is lost by collapsing the three pertinent categories. They were combined in the tree in order to give a clearer overview. The variable that emerges as the best predictor represents the combination of manner features in the complex coda, distinguishing between nasal and oral consonants only. The original labels have been replaced here by more informative ones to show which subtypes of clusters occur in the data.¹⁶ The first group of coda clusters (node 1) consists of sequences of a nasal followed by a (homorganic) obstruent, usually a plosive. These types of clusters categorically undergo vowel paragoge, but no further restructuring. The corresponding Berbice Dutch roots thus end up with a syllable contact between word-internal Restructuring of complex codas Node 0 Category % V-paragoge 79,66 V-epenthesis 5,08 none 3,39 V-paragoge + 11,86 Total (100,00)

n 47 3 2 7 59

Manner combination in cluster Adj. P-value=0,0002, Chi-square=33,3281, df=10 nas-obs Node 1 Category % n V-paragoge 96,15 25 V-epenthesis 0,00 0 none 3,85 1 V-paragoge + 0,00 0 Total (44,07) 26

liq-nas ([rm])

liq-obs, obs-obs Node 2 Category % n V-paragoge 73,33 22 V-epenthesis 0,00 0 none 3,33 1 V-paragoge + 23,33 7 Total (50,85) 30

Node 3 Category % V-paragoge 0,00 V-epenthesis 100,00 none 0,00 V-paragoge + 0,00 Total (5,08)

n 0 3 0 0 3

Fig. 4.10: Treatment of word-final complex codas in Berbice Dutch

16 Essentially the same division can also be expressed in terms of two variables encoding manner for the first and second coda consonant separately. The result would be a tree structure with two splits, one for manner in the second member and another for manner in the first member. The above representation has the advantage of being more compact.

4.4 Results I: Berbice Dutch |

77

nasal coda and a following onset consonant. Considering that nasal codas are also allowed word-finally, this result does not come as a surprise. More surprising is what we see in the second group (node 2). This category comprises all sequences of two non-nasal consonants, i.e. approximant-obstruent and obstruent-obstruent clusters.¹⁷ Judging from the findings for final codas, we might have expected these structures to be subjected to more rigid repair mechanisms than nasal-initial clusters. However, the majority of pertinent clusters again show vowel paragoge. Only in about every fourth etymon an additional repair applies. Conversely, this means that in three quarters of the pertinent cases, the consonant cluster itself is retained and surfaces in the creole as a sequence of word-internal non-nasal coda plus simple onset. As illustrated in the examples in (13) above, even obstruent-obstruent syllable contacts as in ak.ti are allowed word-internally in Berbice Dutch. Only in 7 out of 31 cases, both coda consonants are removed (cf. ‘V-paragoge +’). We can thus conclude that Berbice Dutch has much stronger restrictions on word-final than on word-internal codas. The final group of clusters to be discussed is rather small, but well-defined. It is classified as non-nasal plus nasal consonant and comprises the abovementioned three instances of [rm]-clusters. As node 3 illustrates, vowel epenthesis is the only option attested for this type of complex coda. The rhotic is syllabified in the onset of a new syllable, and the nasal remains in word-final coda position. The list in (14) provides a rearranged selection of examples, grouped according to the results of the CHAID analysis. (14a) shows the treatment of nasalobstruent sequences, (14b) illustrates majority and minority options for liquidobstruent and obstruent-obstruent clusters, and (14c) exemplifies the behaviour of the small group of liquid-nasal codas.

17 Clusters of two approximants, such as [rl], do not occur in the set of etyma.

78 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (14)

Treatment of complex codas in Berbice Dutch – regrouped a. nasal-obstruent da[ŋɡ]i < dank ‘thanks’ lampu < lamp ‘lamp, light’ pontu < po[nt] ‘pound (in weight)’ b. C-obstruent akti < a[xt] ‘eight’ forku < vork ‘fork’ frutu < vru[xt] ‘fruit’ hondrutu < honde[rt] ‘such’ c. liquid-nasal worum < worm ‘worm’ darum < darm ‘intestines, bowels’

To summarise the findings for word-final complex codas, the analysis revealed different categorical treatments for nasal-obstruent and liquid-nasal clusters. Nasalobstruent clusters undergo vowel paragoge, whereas liquid-nasal sequences are broken up by the insertion of an epenthetic vowel. For non-nasal complex codas, we find a preference for mere vowel paragoge over more severe restructuring.

4.4.3 Word-internal structures in Berbice Dutch There are at least two types of word-internal structures that could in principle be targeted by repair mechanisms. Firstly, sequences of consonants, and secondly, sequences of vowels in consecutive syllables, i.e. cases of hiatus. Both types of structure occur in the etyma, although the latter is rare. The data set for the investigation of word-internal structures is a subset of the complete data. All monosyllabic etyma were excluded from the analysis, since, by definition, they cannot contain any of the pertinent structures. Also excluded were disyllabic items which end in a syllabic consonant. These were already discussed in the section on wordfinal structures. The two bar plots in figure 4.11 give the frequencies and percentages of word-internal structures in all remaining etyma with two or more syllables and the corresponding creole reflexes. The total number of cases in this subset is 336. The difference between the distributions of medial structures in roots and etyma is statistically significant (Pearson’s χ2 =19.48, df=2, p C[w]V or C[i].V > C[j]V). Due to its very nature, this process is a possible option only with etyma in which the first of the two vowels is either [i] or [u]. Further restrictions might apply concerning the nature of the preceding consonant. It is conceivable that not all combinations of C-glide might be acceptable complex onsets, so that, for some constellations (e.g. with liquid consonants), glide-formation might be ruled out despite the presence of a high vowel. The second attested repair option in (19), vowel deletion, is not necessarily restricted to certain vowels or vowel sequences. With a total number of 4 pertinent cases, the data are too scarce to establish whether any preferences exist.²¹ Apart from glide formation and vowel deletion, no recurrent patterns emerge from the data. The two remaining cases neither fit any of the above categories, nor do they allow for further generalisations. In the absence of a better label, the relevant restructuring mechanisms were subsumed under ‘others’. There is little we can conclude from the heterogeneous treatment of wordinternal vocalic sequences. What seems clear is that restructuring is preferred to hiatus retention. It is not entirely clear, on the other hand, what governs the choice of repair strategies. Both vowel deletion and glide formation appear to be regular options.

21 Note that what I have called vowel deletion could also be interpreted as coalescence of the two adjacent vocalic segments. Since the adjacent vowels in the pertinent etyma are always similar to each other, there is no conclusive evidence in favour of either assumption.

86 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles 4.4.4 Summary: Berbice Dutch The above analysis has shown that Berbice Dutch repairs many lexifier structures, but still allows considerably more complex structures than simple CV. In general, the phonotactic restrictions imposed on structures at the right edge of words are tighter than those holding in word-initial or word-internal positions. With regard to the left word-edge, we found that complex structures are not ruled out per se. Most complex lexifier onsets are retained in the creole. Only two types of word-initial clusters can be shown to be restricted in Berbice Dutch. A minority of [s]-obstruent onsets are simplified by deletion of the initial consonant, and initial [kn]-clusters are generally broken up by vowel insertion. Among vowelinitial words, retention of the onsetless structure is preferred with short etyma. In longer words, we find variation between deletion and retention of the initial vowel. Consonant epenthesis appears to be restricted to etyma with initial stress, but is rare even within this subset of the data. The investigation of the right word-edge revealed a very different picture. With the exception of the nasals [m] and [n], neither simple nor complex codas are tolerated word-finally in Berbice Dutch. Vowel paragoge emerges as the default repair strategy for word-final coda structures. Finally, in word-internal positions we found that, in addition to nasal codas, also non-nasal coda consonants can occur, although they are not completely unrestricted. The treatment of most word-internal clusters turns out to be probabilistic in nature. The rate of retention varies across different cluster types, with manner of articulation and homorganicity being decisive factors. In contrast to consonant sequences, word-internal vocalic sequences generally show a tendency towards restructuring. Cases of hiatus are preferably repaired by either vowel deletion or glide formation.

4.5 Results II: Negerhollands This section comprises a detailed description of Negerhollands syllable structure and the restructuring mechanisms that apply to etymon structures. It will be shown that the syllabic structures of Negerhollands roots in most cases strongly resemble those of the lexifier words. Very few types of lexifier structures are banned completely from the creole. In most cases, we thus find probabilistic rather than categorical behaviour, with the rate of retention of lexifier structures depending on the exact type of structure. In general, the probability of restructuring is higher for word-final than for word-initial or word-internal positions.

4.5 Results II: Negerhollands

| 87

I will start by looking at word-initial onsets in section 4.5.1, then turn to word-final codas in section 4.5.2 and finally deal with word-internal structures (section 4.5.3). Section 4.5.4 summarises the findings for Negerhollands.

4.5.1 Word-initial onsets in Negerhollands The bar plot in figure 4.14 provides a first overview of the types of word-initial structures we find in Negerhollands roots and the corresponding etyma. Simple onsets are distinguished from complex onsets of two or more segments. The label ‘no onset’ is used for words whose initial syllable starts in a vowel. The vertical axis gives the frequency of occurrence of the different structures. Additionally, the resulting percentages are given for each category.

500 400 70.5%

100

100

200

300

frequency

400 300

70.2%

200

frequency

500

600

Negerhollands etyma

600

Negerhollands roots

21.6%

21.2% 8.3%

0

0

8.2% simple

complex type of onset

no onset

simple

complex no onset type of onset

Fig. 4.14: Word-initial structures in Negerhollands roots and etyma (N=863)

The distribution of onset structures in the creole is comparable to that in the lexifier words (Pearson’s χ2 =0.03 df = 2, p=0.9828). In both data sets, we find a vast majority of words with simple onsets and a smaller, but still considerable, number of complex onsets. Words lacking an initial onset are the least frequent of the three types of structures. Based on this overview we would not expect many changes to apply to lexifier onset structures. The correspondence plot in figure 4.15 largely confirms this expectation, but also reveals some structural changes which were masked in the

88 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles

600 500

simple complex no onset

300

400

creole onsets

96.2%

100

200

frequency

distribution plot. In the correspondence plot, etymon structures are given on the horizontal axis as separate categories, whereas the structures of the corresponding creole reflexes are indicated by sections with different shadings within each bar. The rate at which creole words surface with the same word-initial structure as the respective etyma is given as percentage for each category of etymon structures.

93.4%

0

84.7% simple

complex

no onset

onsets in etyma

Fig. 4.15: Correspondences between word-initial structures in Negerhollands roots and etyma (N=863)

The correspondence rates for both simple and complex lexifier onsets are well above 90%, i.e. over 90% of all etyma with simple and complex onsets have creole reflexes that exhibit the same type of structure. For vowel-initial words, the rate is slightly lower, but still remarkably high at 84.7%. Apart from the percentage of corresponding structures, the plot also shows which onset types occur in the non-corresponding parts of the data. The bars for complex onsets and vowel-initial words hold little surprise as to the result of structural changes. The light sections in these bars indicate that a minority of etyma with either complex or missing onsets have creole reflexes with a simple onset. More unexpected is what we find in the leftmost bar. Although the vast majority of cases show structural correspondence, we also find two small sections with different shading in the bar, indicating changes to more marked structures in the creole. In 2.1% (N=13) of all cases in which the etymon exhibits a simple onset, we find complex onsets in the creole reflexes (middle grey shading). Newly created onsetless structures also exist, but are even less frequent (cf. dark shaded

4.5 Results II: Negerhollands

| 89

section). They are attested in 1.6% (N=10) of the relevant cases. The numbers are admittedly small. Nonetheless, the question remains why we should find such cases at all. From the point of view of onset structures, any change away from a simple onset is a change for the worse. And indeed, a closer look at the relevant data suggests that the attested changes are motivated by factors other than onset structure complexity. I will first discuss cases cases of newly created word-initial clusters and then comment on non-etymological initial vowels. The list in (20) gives representative examples of pairs of etyma and creole word in which only the creole word has a word-initial complex onset. The data are grouped into three subsets that differ with respect to the source of the new initial cluster. Note that, throughout this chapter, the donor language is listed explicitly in examples only if the etymon is of neither Dutch nor English origin. Dutch etyma are always listed with a gloss, English etyma are always listed without one. (20)

Newly created complex onsets in Negerhollands Root Etymon Gloss a. klet < gekleed ‘dressed’ sta < laa sta ‘leave’ b. blof < beloven ‘promise’ gwen < gewennen ‘get used to’ zwel < zowel ‘as well’ c. fruko < verkopen ‘sell’ glɔs < guls ‘greedy’

The first subset, illustrated in (20a), comprises cases in which the complex onset itself is not a new creation. What is new is only that the cluster occurs in wordinitial position. All pertinent lexifier words involve initial unstressed syllables and a complex onset in the first stressed (or main stressed) syllable. In the creole, all material preceding this stressed syllable is deleted, so that the complex onset ends up in word-initial position. What we observe here, thus, could be interpreted as a dispreference for word-initial unstressed syllables. The data in (20b) show a similar tendency. Here, too, the etyma start in an unstressed syllable which is reduced in Negerhollands. However, in these items, only the unstressed vowel is lost. Both the preceding and the following consonants are retained in the creole word. Together, they form a new, complex word-initial onset. Stress does not appear to play a role in the third and final subset (20c). Instead, the complex onsets in the creole result from repair mechanisms targeting coda consonants. The initial syllable of the affected etyma contains a liquid in coda position. In Negerhollands, this liquid is moved to the onset, forming a new word-initial cluster with the etymological onset consonant.

90 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles Based on the data above, various types of complex onsets appear to be permissible in Negerhollands. While all newly created onsets are of the type obstruent-liquid or obstruent-glide, among the retained clusters we also find [s]-plosive sequences. The second type of unexpected change in word-initial structures, the insertion of a non-etymological vowel at the beginning of a word, is likely not to be related to phonotactics at all. The affected etyma are either prepositions or interrogative pronouns, whose creole reflexes show variable epenthesis of an initial vowel [ɑ]. Typical examples are given in (21). (21)

Newly created vowel-initial structures in Negerhollands Root Etymon Gloss a. abini < binne(n) ‘in, inside’ abit < buite(n) ‘out, outside’ abobo < bove(n) ‘above, on top of’ b. awa < wat ‘what’ awi < wie ‘who’

Sabino (1990: 104) analyses this initial vowel as a locative particle and concludes that the alternation between vowel-initial and consonant-initial variants (e.g. wa ~ awa ‘what’) is morphologically rather than phonologically conditioned. Having discussed changes which (somewhat unexpectedly) apply to simple onset structures , I will now turn to the discussion of the more marked etymological onset structures. Sections 4.5.1.1 and 4.5.1.2 will be concerned with the treatment of complex lexifier onsets and vowel-initial etyma, respectively.

4.5.1.1 Complex onsets in Negerhollands As we saw above in the correspondence plot, the vast majority of onset clusters are retained in the creole. Of the 183 root-etymon pairs in this data set, only 7.1% (N=13) show a simplification of word-initial consonant sequences. The repair mechanisms targeting complex onsets are listed in (22) in order of frequency. The category ‘none’ covers cases in which no structural changes take place.²²

22 Note that the numbers differ slightly from those in the correspondence plot in figure 4.15 above, where the correspondence rate for complex onsets is at 93.4% (not 92.9%). This discrepancy results from the fact that also Negerhollands roots which undergo restructuring may still end up with a complex onset, albeit a simplified one. The Negerhollands word tret from English straight is a case in point.

4.5 Results II: Negerhollands

(22)

| 91

Processes affecting complex onsets in NH Process

N

%

none consonant deletion vowel epenthesis others

170 9 3 1

92.9 4.9 1.6 0.5

Σ

183

100.0

Among the restructuring processes, deletion of one of the consonants in the complex onset is the preferred option. All remaining repairs are extremely rare. An epenthetic vowel is inserted to break up the onset cluster in three cases, and a single etymon undergoes a change that I have classified as metathesis of the initial consonant to word-internal position. The pertinent root-etymon pair is listed in (23d) following examples of the retention of complex onsets (23a) and the more straightforward repair processes (23b-c). (23)

Treatment of complex lexifier onsets in Negerhollands Root Etymon Gloss a. plim < pluim ‘feather’ skɔp < schoppen ‘kick’ smok < smook ‘smoke’ twalv < twaalf ‘twelve’ b. krew < schreuuwen ‘scream’ ʃip < zweep ‘whip’ tɔp < stoppen ‘stop’ c. filis < vlees, vleis ‘meat’ kɔnɔbə < knôôpe ‘button/knot’ ʃini < snijden ‘cut’ d. pistakəl < spektakel ‘spectacle’

In a CHAID analysis with restructuring process as the dependent variable, the manner of articulation of the second consonant emerged as the best predictor of cluster treatment. As the tree diagram in figure 4.16 shows, three major groups of complex onsets can be distinguished. Clusters with an approximant C2 , an obstruent C2 , and a nasal C2 , respectively, differ in the rate of retention as well as with regard to the choice of repair mechanism. The highest rate of retention is found for sequences in which the second consonant is an approximant (cf. node 1). This group includes all obstruent-liquid

92 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles Restructuring of complex onsets Node 0 Process % N none 92.90 170 C-deletion 4.92 9 V-epenthesis 1.64 3 others 0.55 1 Total (100.00) 183 Manner of second cluster member Adj. P-value=0.0001, Chi-square=28.4772, df=6

approximant Node 1 Process % N none 98.35 119 C-deletion 0.83 1 V-epenthesis 0.83 1 others 0.00 0 Total (66.12) 121

obstruent Node 2 Process % none 84.21 C-deletion 14.04 V-epenthesis 0.00 others 1.75 Total (31.15)

nasal

N 48 8 0 1 57

Node 3 Process % N none 60.00 3 C-deletion 0.00 0 V-epenthesis 40.00 2 others 0.00 0 Total (2.73) 5

Fig. 4.16: Treatment of word-initial complex onsets in Negerhollands

and obstruent-glide clusters.²³ Restructuring occurs only in a very small number of exceptional cases, so that for this type of complex lexifier onset retention can be considered categorical. The second group of clusters are characterised by an obstruent C2 (cf. node 2). More specifically, what we find in this subset of the data are [s]-obstruent sequences. Like obstruent-approximant onsets, also [s]-obstruent clusters show a strong tendency towards retention. A minority of the pertinent cases (14.0%, N=8), however, undergo consonant deletion. In these cases, the initial [s] is lost in the creole reflexes. Finally, node 3 gives an account of the treatment of obstruent-nasal onsets. They are rare even in the etyma, the only attested types in the data being [sm], [sn] and [kn], resulting in a total of five relevant root-etymon pairs. The tree diagram suggests variable behaviour for these clusters with both retention and insertion of an epenthetic vowel between the two consonants as possible options. It should be noted, however, that the two instances of [sm] which occur in the lexifier also surface unaltered in Negerhollands. For initial [sn], on the other hand, there is evidence for the postulated variation. Two different creole reflexes occur for the same etymon, one featuring an unmodified onset cluster, the other a simplified structure with an epenthetic vowel. The only remaining etymon in this subset shows an initial [kn]-cluster,

23 Nasal-glide clusters are not attested in the data.

4.5 Results II: Negerhollands

| 93

which is restructured in the creole. In the absence of evidence to the contrary, I will assume that [kn]-clusters are at least dispreferred, if not illicit onsets in Negerhollands. For [s]-nasal sequences, I will maintain the hypothesis of variable treatment with a tendency towards retention of the cluster. An adjusted list of examples is given in (24a-c) below, illustrating the treatment options for the different subsets of onset clusters as they emerged in the analysis. (24)

Treatment of complex lexifier onsets in Negerhollands – regrouped Root Etymon Gloss a. obstruent-approximant frai < fraai ‘good’ plim < pluim ‘feather’ twalv < twaalf ‘twelve’ b. [s]-obstruent spel < spelen ‘play’ stɔp ~ tɔp < stoppen ‘stop’ krew < schreuuwen ‘scream’ c. obstruent-nasal kɔnɔbə < knôôpe ‘button, knot’ smak < smaken ‘taste’ sni ~ ʃini < snijden ‘cut’

To summarise the above discussion, we find few instances of restructuring overall, but a non-random distribution of those processes which are attested in the data. A minority of [s]-obstruent onsets are simplified by consonant deletion, whereas an epenthetic vowel may be inserted to break up obstruent-nasal onsets.

4.5.1.2 Vowel-initial words in Negerhollands Words lacking an initial onset occur in both lexifier and creole data. The total number of pertinent etyma in the data is 72. The majority of corresponding creole reflexes retain the onsetless structure. In those cases where restructuring applies, the result is a simple onset in the creole word. Repair strategies include deletion of the initial, onsetless syllable and insertion of an epenthetic consonant in onset position. The table in (25) gives the frequencies and percentages for retention and the attested repair mechanisms.

94 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles (25)

Processes affecting vowel-initial words in NH Process

N

%

none syllable deletion consonant epenthesis

61 8 3

84.7 11.1 4.2

Σ

72

100.0

As the table shows, syllable deletion is preferred to consonant epenthesis, which applies in only three cases overall. Among the data showing deletion, we find both cases in which the initial syllable consists of only the vocalic nucleus and cases in which the vowel is followed by a coda consonant.²⁴ Examples illustrating all attested treatment options are given in (26) in the same order as in the frequency table above. In those cases where no gloss is given, the etymon is an English word. (26)

Treatment of vowel-initial words in Negerhollands Root Etymon Gloss a. altəvel < al te veel ‘too much, too many’ eŋges < engaged oli < olie ‘oil’ b. braideri < embroidery genz < against merike < Amerika ‘America’ c. jai < eye jet < eten ‘eat’

Since the data were not coded for vowel quality, only two factors could be tested for influence on the creole outcome, namely the position of main stress and word length in terms of the number of syllables in the etymon. Figure 4.17 illustrates the results of a CHAID analysis taking these two factors as potential predictor variables and restructuring process as the dependent variable. As the tree shows, word length emerges as the best predictor. We find a three-way distinction between monosyllabic etyma, disyllabic etyma, and etyma with more than two syllables. Retention of the onsetless lexifier structure is preferred in all subsets, but we find differences in the probability of retention and in repair strategies. For monosyllables (cf. node 1), syllable deletion is obviously not

24 This distinction is potentially interesting for word-internal structures, since deletion of an initial VC-syllable entails the elimination of a coda-onset sequence.

4.5 Results II: Negerhollands

| 95

Restructuring of vowel-initial words Node 0 Process % syllable deletion 11.11 none 84.72 C-epenthesis 4.17 Total (100.00)

N 8 61 3 72

Number of syllables in etymon Adj. P-value=0.0262, Chi-square=11.0331, df=4

1 Node 1 Process % N syllable deletion 0.00 0 none 80.00 8 C-epenthesis 20.00 2 Total (13.89) 10

2 Node 2 Process % N syllable deletion 7.14 3 none 90.48 38 C-epenthesis 2.38 1 Total (58.33) 42

3 or more Node 3 Process % syllable deletion 25.00 none 75.00 C-epenthesis 0.00 Total (27.78)

N 5 15 0 20

Fig. 4.17: Treatment of vowel-initial etyma in Negerhollands

an option. Instead, a minority of onsetless syllables are repaired by consonant epenthesis. In disyllabic etyma (cf. node 2), retention can be considered categorical. Words with more than two syllables show the highest rate of restructuring. In one quarter of the cases, the initial syllable is lost. The list in (27) provides a rearranged selection of examples, grouped according to word length. Retention of onsetless structures and occasional consonant epenthesis in monosyllables is illustrated in (27a). (27b) gives examples of the preservation of vowel-initialness in disyllabic etyma and (27c) shows initial syllable retention and deletion in longer words. (27)

Treatment of vowel-initial words in Negerhollands – regrouped Root Etymon Gloss a. monosyllabic etymon aks < axe jai < eye b. disyllabic etymon alen < alleen ‘only’ eŋges < engaged oli < olie ‘oil’ c. etymon with three or more syllables enestə < eenigste ‘any’ imatal < immortal braideri < embroidery

96 | 4 Syllable structure and phonotactic restructuring in the Dutch-based creoles Summing up the section on vowel-initial words, we have seen that retention of onsetless structures is generally preferred to restructuring in the creole. Only a minority of etyma are restructured. Monosyllables may gain a non-etymological onset consonant and etyma that consist of three or more syllables may lose the initial, onsetless syllable.

4.5.2 Word-final codas in Negerhollands We saw above that only few word-initial structures are targeted by repair mechanisms in Negerhollands. In word-final position, we find more pronounced differences between lexifier and creole words, although even complex codas may surface intact in Negerhollands. Figure 4.18 gives an overview of the distribution of word-final consonants and consonant sequences in etyma and roots. ‘CC(C)’ stands for consonantal sequences of any type, including both complex codas and sequences of an onset followed by a syllabic consonant.

400 300

frequency

100

39.5%

61.4%

200

300 200

53.7%

100

frequency

400

500

Negerhollands etyma

500

Negerhollands roots

28.4% 10.2%

0

0

6.8% none

C

CC(C)

none

C

CC(C)

Fig. 4.18: Word-final consonants in Negerhollands roots and etyma (N=863)

As the bar plots show, single word-final consonants are the most common structure in both lexifier and creole data, although the preponderance of this type of structure is stronger in the lexifier than in the creole. In the etyma, simple codas occur more than twice as often as open syllables (61.4% final ‘C’ compared to 28.4% ‘none’), whereas the distribution of the two types of structure is somewhat

4.5 Results II: Negerhollands

| 97

more balanced in the creole (53.7% compared to 39.5%). Another difference can be observed by comparing the right-hand bars in the two plots. The etyma of Negerhollands contain more word-final consonant sequences than surface in the roots (10.2% and 6.8%, respectively). Thus, while neither single consonants nor consonant sequences are categorically banned from Negerhollands, an overall tendency towards simpler wordfinal structures is observable. The resulting difference in the distributions of wordfinal structures in roots and etyma is highly significant (Pearson’s χ2 =25.97, df=2, p tenteh ‘to aim’), but retain final rhotics in other words such as Portuguese-derived nouns (e.g. mulher > mujêre ‘woman’) or words with Dutch origin (e.g. plezier > plessiri ‘pleasure’). Based on the above indications and the creole facts, all Portuguese verbal etyma were coded as ending in a vowel rather than a rhotic. Epenthesis and loss of word-initial consonants For the 15th and 16th centuries, Dobson (1968: 993ff) reports that epenthetic glides occurred word-initially before (Middle English) long mid front vowels and after initial [h]. Two lexical items, for which Dobson gives evidence of epenthetic [j], were coded accordingly in the data (eat and hear). None of the words with Middle English high front vowels were affected, which leads Dobson to conclude that they had already begun to diphthongise in the initial stages of the Great Vowel Shift.

5.3 Methodological issues

| 119

The initial glide in the St Kitts reflex of English eye is thus regarded as a creole innovation. No evidence is given for a lexifier origin of word-initial epenthetic [h], which occurs in some creole words. Quite to the contrary, initial [h] could be dropped in English in vulgar or dialectal speech. This process is reported as much more common in unstressed than in stressed syllables. The only words which show [h]-deletion in the English-based creoles, however, involve loss in stressed positions. Moreover, no lexical evidence could be found for the pertinent items. Consequently, [h]-loss (like [h]-insertion) was interpreted as a change in the creole, not a feature of lexifier pronunciation. Rhoticity Modern Standard British English has lost coda [r] as singleton consonant as well as in clusters. In the Early Modern English varieties that served as input to the creoles, however, non-rhoticity was not yet as widespread. Dobson (1968) states that loss of the rhotic did not occur simultaneously in all positions. Assimilatory loss of [r] before [s] and [ʃ] in particular, but also before other coronal consonants, is given as the earliest development with attestations for reduced forms from the 14th century onwards (cf. Dobson 1968: 966f). The second process Dobson mentions is the vocalisation of singleton [r] in unstressed syllables, which would have led to vowel-finality in words such as sister and water. Examples of this type of simplification are rare before 1700, but become more frequent in the 18th century (cf. Dobson 1968: 914f). In stressed syllables, on the other hand, postvocalic [r] was more likely to be maintained even in the 18th century (cf. Dobson 1968: 992f). Based on these historical records, we can conclude that the English model available to creole speakers very likely showed variation between [r]-less and [r]full forms at least in the first two contexts mentioned, [r] before coronal consonants and singleton coda [r] in unstressed syllables. Absence of coda [r] in stressed syllables, on the other hand, would have been more exceptional. The creole facts accord well with this assumption. Both Saramaccan and St Kitts show evidence for the presence of [r] in reflexes of English stressed monosyllables ending in a rhotic (e.g. ESA tîri < steer or ESK cleary < clear), whereas in other contexts [r] may be either present or absent (e.g. ESA membre < remember and bronn < burn versus sissa < sister and kossi < curse). In all of the pertinent cases, the rhotic could have been lost already in the lexifier variety. Other authors have come to similar conclusions, also assigning variability in the presence of the rhotic in creole words to variability in the English model (e.g. Smith 1987: 341, Plag 1999: 181-183).

120 | 5 Syllable structure and phonotactic restructuring in the English-based creoles In view of the above situation, English-derived etyma were coded as being nonrhotic unless the creole reflex suggested that the rhotic was in fact present. This procedure was used for both word-final and word-internal codings. Simple velar nasal or nasal plus voiced homorganic stop? Diachronically, English has seen a development from final [ŋɡ] to [ŋ]. Completion of this process must have varied in time across different dialects. Dobson (1968: 963ff) regards the cluster as the regular Middle English pronunciation and assumes that it persisted, at least in Standard English, into the 16th century. However, the historical records suggest that [ɡ]-less pronunciations were acceptable also in Standard English around 1600, and occurred regularly in the 17th century.¹ Given the historical records, I assumed final [ŋ] to be the regular variant in the pertinent English etyma. The only exception are words ending in unstressed -ing such as e.g. nothing. These words most likely featured final [ɪn] rather than [ɪŋ] (cf. Dobson 1968: 963) and were coded accordingly in my data. The Saramaccan data also include some items of Dutch origin which used to end in a velar nasal-plosive cluster. Like the English etyma, these, too, were usually coded as containing final [ŋ]. For a detailed discussion and possible exceptions see chapter 4.3 on coding in the Dutch-based creoles. Syllabic consonants According to Dobson (1968: 889ff), syllabic consonants occurred in 16th and 17th century English pronunciations, like they do in present-day English, in free variation with sequences of [ə] plus non-syllabic consonant. There is sound evidence for the existence of [l] and [n] alongside [əl] and [ən]. It is thus not unlikely that ˈ ˈ creole speakers were confronted with both variants. The creole words, however, do not reflect such variation. In the St Kitts corpus, most pertinent roots maintain the standard English orthography or deviate from it only in aspects not relevant to the discussion of syllabic consonants. Such spellings could represent either of the two variants (cf. e.g. settle > settle, devil > debbil, parson > passon). Lacking evidence for a syllabic consonant, I coded etyma and creole words alike as ending in .CVC# rather than .CC#. Also for the only root in St Kitts that shows evidence of change, an etymon ˈ a simple word-final coda was assumed. The item in point is uncle > unco. with

1 In some dialects, this change was completed much earlier. Dobson (1968) dates it back to the 14th century for Eastern dialects, including vulgar London English. Retention of final [ŋɡ] after the 17th century is reported only for Northern and West Midland dialects and only in stressed positions.

5.4 Results I: Early Saramaccan

| 121

Absence of the liquid in the creole word was interpreted as consonant deletion in this case.² Comparable assumptions were made for the Saramaccan data. Since there was no compelling evidence for syllabic consonants, the etyma were coded as containing final .CVC# sequences in all pertinent cases.³ [t,d]-deletion in English-derived words Loss of word-final coronal stops is documented for dialectal and vulgar speech from the 14th century onwards, but is reported as applying only variably and only in particular phonetic environments. Dobson (1968: 961f) states that final [t] could be dropped after the voiceless coronal fricative [s] and after the plosives [k] and [p]. Other cases of [t,d]-deletion involved preceding coronal sonorants such as [n] or [l]. After the lateral, occasional loss of the voiced plosive is documented. After the nasal, deletion of both [d] and [t] is mentioned, although the latter is said to apply only rarely. The creole reflexes show that [t,d]-deletion certainly did not apply across the board even in environments in which the process is commonly recognised. For instance, Saramaccan roots typically retain formerly final [nd]-cluster as wordinternal sequences (e.g. bend > bendi, sand > sandu). In view of the variable nature of [t,d]-deletion, it is difficult to judge in which cases it should or should not be assumed to have taken place in the lexifier. I therefore decided to resort to the criterion of lexical evidence. The pertinent etyma were coded as involving simplification only if Dobson (1968) lists evidence for the particular lexical item and if the creole form does not contradict this assumption.

5.4 Results I: Early Saramaccan In this section I will discuss the syllabic structures that we find in the Saramaccan data as well as the processes that relate the creole words to their assumed etyma. It

2 Alternatively, we could have hypothesised that what we observe here is no creole innovation, but simply a preservation of a vocalised lexifier realisation of [l]. However, there is no indication ˈ that such vocalisations occurred in the lexifier. The historical records mention vocalisations only in positions where the lateral has been lost completely in modern English (e.g. in half, walk, cf. Dobson 1968: 988ff). 3 This applies to both English and Dutch etyma. For more details on the presence or absence of syllabic consonants in Dutch lexical items, see also section 4.3 in the chapter on the Dutch-based creoles above.

122 | 5 Syllable structure and phonotactic restructuring in the English-based creoles will be shown that Early Saramaccan is relatively permissive with respect to onset structures, but imposes tight restrictions on syllable codas. We will first look at the left and right word-edges in sections 5.4.1 and 5.4.2, respectively. Structures in word-internal positions will be dealt with in section 5.4.3. I will conclude with a short summary of the findings in section 5.4.4.

5.4.1 Word-initial onsets in Early Saramaccan The bar plots in figure 5.1 show the distribution of onset structures in Early Saramaccan roots and etyma.

400 300

frequency

69.4%

100

200

300 200

74.9%

100

frequency

400

500

Saramaccan etyma

500

Saramaccan roots

18.1% simple

complex type of onset

no onset

17.2%

13.4%

0

0

7%

simple

complex no onset type of onset

Fig. 5.1: Word-initial structures in Saramaccan roots and etyma (N=641)

Words with simple onsets make up the largest share of both roots and lexifier words (74.9% and 69.4%, respectively), followed by a considerable number of words with word-initial clusters (around 17-18%). Words starting in an onsetless syllable are the least frequent type in both sets. However, we also find some differences between the two distributions. While the proportion of simple onsets is slightly higher in the creole than in the set of lexifier words, we find the reverse situation for vowel-initial words, which are almost twice as frequent in the etyma (7% in the creole compared to 13.4% in the lexifier). The overall difference between the two distributions is statistically significant (Pearson’s χ2 =14.32, df = 2, p kiljà ‘bring up, nurse’). This process will be shown below to be exceptional. Examples of the more common treatment options, retention of the consonant cluster, consonant deletion and vowel epenthesis are given in (4a-c). (4)

Treatment of complex lexifier onsets in Saramaccan Root Etymon Gloss Donor a. dringi < drink English pleti < plate English skapu < schaap ‘sheep’ Dutch b. kutjù < crouch English koto < skirt English pinda < mpinda ‘peanut’ Kikongo c. sitteh < stay English supon < spoon English

In order to find out whether retention and restructuring of complex lexifier onsets are determined by the structural make-up of the cluster, I conducted a CHAID analysis with restructuring process as the dependent variable and the various seg-

4 The slight discrepancy between the percentage given here and the correspondence rate in figure 5.2 (75.5%) is due to a small number of etyma with trisegmental onsets which surface with only two consonants in the creole. These cases were counted as showing correspondence in figure 5.2 since both root and etymon have a complex onset. In the investigation of restructuring processes conducted here, however, they fall into the category of consonant deletion.

5.4 Results I: Early Saramaccan

| 127

mental codings (cf. chapter 3.6) as predictor variables. The results are illustrated in the tree diagram in figure 5.3. Restructuring of complex onsets Node 0 Process % C-deletion 22.73 none 72.73 V-epenthesis 3.64 others 0.91 Total (100.00)

N 25 80 4 1 110

Sonority profile of the cluster Adj. P-value=0.0000, Chi-square=61.3217, df=3

falling Node 1 Process % N C-deletion 74.07 20 none 14.81 4 V-epenthesis 11.11 3 others 0.00 0 Total (24.55) 27

level or rising Node 2 Process % N C-deletion 6.02 5 none 91.57 76 V-epenthesis 1.20 1 others 1.20 1 Total (75.45) 83

Fig. 5.3: Treatment of complex lexifier onsets in Saramaccan

The factor that emerges as best predictor of cluster treatment is the sonority profile of the cluster. Complex onsets in which sonority drops towards the syllable peak are treated differently from complex onsets with a level or rising sonority profile. Those clusters which do not obey sonority sequencing (cf. node 1) show a strong tendency towards restructuring (about 85% overall). Consonant deletion is the most common repair option for this group of lexifier words. It comprises English etyma with initial [s]-plosive clusters and words of African origin which feature an initial sequence of nasal plus homorganic stop.⁵ In all cases of deletion, it is the edgemost segment that is lost in the creole, so that the pertinent roots all start in a simple plosive consonant. For a minority of [s]-plosive onsets, vowel epenthesis is attested as an alternative repair option. Here, both lexifier consonants are

5 Recall from the methodology chapter 3 that such cases were coded as two segments, although their status in the donor language may be that of a single, complex segment.

128 | 5 Syllable structure and phonotactic restructuring in the English-based creoles retained, but the cluster is broken up by an intervening vowel, so that the two consonants end up in separate syllable onsets in Saramaccan. In node 2, we see a very different picture. Retention of the cluster can be considered categorical for this second group. Here, we find mostly obstruent-sonorant onsets, but also a small number of [sx]-clusters of Dutch origin. Interestingly, the latter surface as [sk]-onsets in Saramaccan, a structure which, as we just saw, is usually repaired if it occurs in the etymon. The table in (5) gives a rearranged list of examples, illustrating the difference in treatment of complex lexifier onsets with falling sonority on the one hand (5a) and level or rising sonority on the other (5b). (5)

Treatment of complex lexifier onsets in Saramaccan – regrouped Root Etymon Gloss Donor a. falling sonority profile koto < skirt English pinda < mpinda ‘peanut’ Kikongo sitteh < stay English speli < spell English b. level or rising sonority profile dringi < drink English pleti < plate English skapu < [sx]aap ‘sheep’ Dutch smêri < smell English

In conclusion, Saramaccan retains the majority of complex onsets. The only lexifier clusters which are preferably simplified are those in which sonority drops towards the syllable peak.

5.4.1.2 Vowel-initial words in Early Saramaccan Vowel-initial words are the least frequent of word-initial structures. The data set comprises a total number of 86 pertinent root-etymon pairs, on which the following analysis is based. As the table in (6) shows, the treatment of vowelinitial etyma is variable. There is no clear preference among retention and repair, although the latter is slightly more frequent.

5.4 Results I: Early Saramaccan

(6)

| 129

Processes affecting vowel-initial etyma in ESA Process

N

%

none syllable deletion consonant epenthesis

36 43 7

41.9 50.0 8.1

Σ

86

100.0

Over 40% of all pertinent cases preserve the onsetless structure in Saramaccan. Among the repair options, we find a strong preference for deletion of the initial, onsetless syllable.⁶ The only other attested restructuring process is consonant epenthesis, which applies to a minority of vowel-initial etyma. [h] and [j] occur as epenthetic onset consonants in the corresponding roots. Representative examples of the preservation of etymon structures and the two repair options are given in (7), following the same order as in the frequency table above. (7)

Treatment of vowel-initial etyma in Saramaccan Root Etymon Gloss a. allêki < alike awitti < avìtì ‘mouse trap’ ougri < ugly b. nuffe < enough panta < espantar ‘frighten, scare’ télu < inteiro ‘whole’ c. hansi < ants jabri < abrir ‘open’

Donor English Gbe English English Portuguese Portuguese English Portuguese

As outlined before, the factors that can be tested as possible triggers for restructuring in vowel-initial words are limited. A CHAID analysis with the number of syllables and the position of the main stressed syllable in the etymon as predictor variables returned word length as the more influential factor. The tree diagram in figure 5.4 shows the resulting split into words with a maximum of two syllables on the one hand and words of three or more syllables on the other hand. Short words (cf. node 1) tend to retain the initial onsetless syllable. Among those roots which show restructuring in comparison to the lexifier word, we

6 Note that the label ‘syllable deletion’ covers both cases in which the initial syllable in the etymon consists of only the vocalic nucleus and those in which an etymological coda consonant is lost along with the nucleus.

130 | 5 Syllable structure and phonotactic restructuring in the English-based creoles Restructuring of vowel-initial words Node 0 Process % none 41.86 syllable deletion 50.00 C-epenthesis 8.14 Total (100.00)

N 36 43 7 86

Number of syllables in etymon Adj. P-value=0.0000, Chi-square=43.6553, df=2

1 or 2 Node 1 Process % N none 68.75 22 syllable deletion 9.38 3 C-epenthesis 21.88 7 Total (37.21) 32

3 or more Node 2 Process % none 25.93 syllable deletion 74.07 C-epenthesis 0.00 Total (62.79)

N 14 40 0 54

Fig. 5.4: Treatment of vowel-initial lexifier words in Saramaccan

find consonant epenthesis more often than syllable deletion. For mono-syllabic words, syllable deletion is not a possible option at all, for obvious reasons. The few attested disyllabic etyma which undergo syllable deletion, however, are interesting to investigate in so far as they all have non-initial stress. The pertinent etyma are afraid, along, and enough, which lose their initial unstressed vowel and end up as fredde, nanga, and nuffe in Saramaccan. Stress thus appears to have at least a secondary effect on restructuring. Among etyma with more than two syllables (cf. node 2), syllable deletion is the majority choice and applies to almost three quarters of all pertinent cases. The remaining words preserve their onsetless initial structure. Again, stress may be an additional, supporting factor. With the possible exception of a handful of Africanderived items for which main stress (or the syllable with the highest pitch) could not be determined, all tri- and polysyllabic etyma in the data set have non-initial stress. The table in (8) gives a rearranged list of examples illustrating the treatment options for short and long words in Saramaccan.

5.4 Results I: Early Saramaccan

(8)

| 131

Treatment of vowel-initial etyma in Saramaccan – regrouped Root Etymon Gloss Donor a. etymon with one or two syllables abra < over English allêki < alike English hansi < ants English nuffe < enough English b. etymon with three or more syllables awitti < avìtì ‘mouse trap’ Gbe panta < espantar ‘frighten, scare’ Portuguese télu < inteiro ‘whole’ Portuguese

In sum, we found that onsetless word-initial structures are not ruled out per se in Saramaccan. The likeliness of restructuring grows with the number of syllables in the etymon. In shorter words, retention is more frequent than restructuring. In case of restructuring, consonant epenthesis is clearly preferred in this subset. In longer etyma, on the other hand, we find a pronounced tendency towards deletion of the initial, typically unstressed syllable.

5.4.2 Word-final codas in Early Saramaccan The discussion of word-initial structures has revealed that Saramaccan syllables can clearly be more complex than simple CV. In this section we will see that the relative permissiveness that Saramaccan shows with respect to the left word-edge does not extend to word-final structures. Here, we find much more radical restructuring. Figure 5.5, showing the distribution of word-final structures in Saramaccan roots and etyma, gives a first impression of the extent of restructuring at the right word-edge. The difference in distributions is very highly significant (p bassia ‘go down(stairs)’ and molhar > muija ‘to wet, be wet’. The motivation for these changes remains unclear. However, they support the earlier assumption that hiatus is tolerated in Early Saramaccan.

5.4 Results I: Early Saramaccan

| 147

attested restructuring mechanisms, their frequencies and percentages, including the option ‘none’ to indicate retention of the cluster. (17)

Processes affecting medial clusters in ESA Process none consonant deletion metathesis vowel epenthesis consonant deletion + vowel epenthesis segmental change Σ

N

%

66 20 12 12 1 1

58.9 17.9 10.7 10.7 0.9 0.9

112

100.0

Three main strategies for word-internal consonant clusters can be established. Deletion of one of the cluster members is the most common option, accounting for 17.9% of the data. As alternatives, we find insertion of an epenthetic vowel, which breaks up the cluster, and metathesis, which typically moves the first member of the cluster from the coda to the onset of the syllable. A single case shows a combination of consonant deletion and vowel epenthesis, and another individual etymon undergoes a segmental change. Examples of retention and all of the more frequent processes are given in (18a-d). The two more exceptional cases are listed together in (18e).¹³

13 Note that the latter of the two, although classified here as segmental change, could equally well be assumed to involve deletion of the liquid. The development of a non-etymological velar nasal could then be analysed analogously to the cases discussed in the preceding section.

148 | 5 Syllable structure and phonotactic restructuring in the English-based creoles (18)

Treatment of medial clusters in Saramaccan Root Etymon Gloss a. kondre < country sangla < sangrar ‘to bleed’ b. plitti < splinter sissa < sister c. drummi < dormir ‘to sleep’ kruboì < goodbye d. kássika < casca ‘bark, skin’ passamau < pasmado ‘astouned’ e. teréjah < estrela/estrella ‘star’ munga < amolgar ‘to buckle’

Donor English Portuguese English English Portuguese English Portuguese Portuguese Port./Span. Portuguese

A CHAID analysis allows a more differentiated discussion of cluster repairs. The tree diagram in figure 5.13 illustrates the results. The manner of articulation of the first consonant emerges as the best predictor of cluster treatment. Three main groups of clusters can be distinguished, those with a nasal consonant as first member, those with an obstruent C1 and those which start in a liquid. Of these three groups, only nasal-initial clusters are categorically retained (cf. node 1). These are typically sequences of a nasal and a homorganic stop or fricative. At the other end of the retention scale we find liquid-initial clusters (cf. node 3). More precisely, the consonant sequences in this group are of the type liquid-obstruent or liquid-nasal, i.e. the consonants are heterosyllabic in the input. The only cluster in this group for which (variable) retention is attested is [rɡ]. This is reminiscent of the results of the previous section, where [rk], another rhotic-initial cluster with a velar obstruent as C2 , emerged as possible word-internal sequence. Unfortunately, none of the etyma contain medial [rk], so that a direct comparison of the two clusters and their respective treatment is not possible. I will therefore assume that both [rk] and [rɡ] may be variably preserved in medial position. Retention is not a possible option for any of the other liquid-initial sequences. The majority of the remaining clusters show metathesis of the liquid and the preceding vowel. Other restructuring mechanisms are rare. Vowel epenthesis is attested in only two cases, and a segmental change repairs a single LC sequence. Turning to the third group, obstruent-initial clusters (cf. node 2), we find not only an intermediate rate of cluster retention (32.56%), but also greater variability in the choice of repair strategies. The subsequent splits in the tree diagram reveal that this variability results from the fact that the group of obstruent-initial clusters comprises some highly diverse cluster types. Thus, in the second split, two subsets are formed according to the manner of articulation of the second cluster member. Obstruent-sonorant clusters are separated from obstruent-obstruent clusters. The

5.4 Results I: Early Saramaccan

| 149

Restructuring of word-internal clusters Node 0 Process % none 58.93 segmental 0.89 C-deletion 17.86 V-epenthesis 10.71 metathesis 10.71 C-del.+ V-ep. 0.89 Total (100.00)

N 66 1 20 12 12 1 112

Manner of first cluster member Adj. P-value=0.0000, Chi-square=94.6198, df=10

obstruent

nasal Node 1 Process % N none 92.59 50 segmental 0.00 0 C-deletion 7.41 4 V-epenthesis 0.00 0 metathesis 0.00 0 C-del.+ V-ep. 0.00 0 Total (48.21) 54

liquid

Node 2 Process % N none 32.56 14 segmental 0.00 0 C-deletion 37.21 16 V-epenthesis 23.26 10 metathesis 4.65 2 C-del.+ V-ep. 2.33 1 Total (38.39) 43

Node 3 Process % N none 13.33 2 segmental 6.67 1 C-deletion 0.00 0 V-epenthesis 13.33 2 metathesis 66.67 10 C-del.+ V-ep. 0.00 0 Total (13.39) 15

Manner of second cluster member Adj. P-value=0.0000, Chi-square=26.5444, df=4

obstruent

sonorant Node 4 Process % N none 72.22 13 segmental 0.00 0 C-deletion 16.67 3 V-epenthesis 5.56 1 metathesis 5.56 1 C-del.+ V-ep. 0.00 0 Total (16.07) 18

Node 5 Process % N none 4.00 1 segmental 0.00 0 C-deletion 52.00 13 V-epenthesis 36.00 9 metathesis 4.00 1 C-del.+ V-ep. 4.00 1 Total (22.32) 25 Place of cluster members Adj. P-value=0.0176, Chi-square=11.9692, df=4

homorganic Node 6 Process none segmental C-deletion V-epenthesis metathesis C-del.+ V-ep. Total

% N 0.00 0 0.00 0 81.82 9 9.09 1 0.00 0 9.09 1 (9.82) 11

not homorganic Node 7 Process % N none 7.14 1 segmental 0.00 0 C-deletion 28.57 4 V-epenthesis 57.14 8 metathesis 7.14 1 C-del.+ V-ep. 0.00 0 Total (12.50) 14

Fig. 5.13: Treatment of word-internal consonant sequences in Saramaccan

150 | 5 Syllable structure and phonotactic restructuring in the English-based creoles former group, mostly obstruent-approximant sequences, are preferably retained in the creole (cf. node 4). Occasionally, we find deletion of the approximant. Other repairs appear to be exceptional. For obstruent-obstruent sequences, a different picture emerges (cf. node 5). Here, restructuring applies categorically, with consonant deletion and vowel epenthesis as the two main repairs. The final split shows that the preference among these two options depends on the place of articulation of the two cluster members. Homorganic obstruent-clusters (cf. node 6) are preferably simplified by deletion of one of the consonants, whereas non-homorganic clusters (cf. node 7) are more often broken up by the insertion of an epenthetic vowel. It is worth noting that the overall structure of the word also has an influence on the choice of repair strategy among obstruent clusters. In particular, Portuguese etyma with initial esC- merit a separate discussion. These forms typically lose the initial vowel. If no further repair applied, the formerly word-internal clusters would thus end up in word-initial position. It may therefore not be surprising that clusters in this type of etymon are treated differently from clusters in other types of etyma. All cases of deletion in non-homorganic obstruent clusters occur in etyma with initial esC-, where the sibilant may be dropped. No attestations could be found for this process in other types of etyma, in which the preceding vowel is retained. Consonant deletion in non-homorganic obstruent clusters may thus be a case of word-initial rather than medial repair. The group of homorganic obstruent clusters further strengthens this position. Only two of the homorganic clusters occur in etyma with esC-. However, these two cases behave differently from all others. One of them shows vowel epenthesis, the other one deletion of the sibilant, repairs which are attested also for word-initial [s]-obstruent clusters. All remaining homorganic clusters lose the second member of the cluster instead and retain the sibilant in intervocalic position (cf. estanho > sitanja ‘tin’ and estrela > teréjah ‘star’ on the one hand and cases such as sister > sissa and costumar > kossuma ‘to usually do sth.’ on the other hand). Based on these findings, I will assume that true word-internal obstruent clusters undergo deletion of the second cluster member if homorganic and are broken up by vowel epenthesis if non-homorganic. To round off the discussion, a rearranged list of examples is provided in (19), which illustrates the various treatment options for the subgroups that emerged from the analysis.

5.4 Results I: Early Saramaccan

(19)

| 151

Treatment of medial clusters in Saramaccan – regrouped Root Etymon Gloss Donor a. nasal-obstruent kondre < country English limba < limpar ‘to clean’ Portuguese b. liquid-C syllable contact drummi < dormir ‘to sleep’ Portuguese flatta < faltar ‘to miss’ Portuguese djaralì < jardim ‘garden’ Portuguese vergêti < forget English c. obstruent-approximant kubri < cubrir ‘to cover’ Portuguese ougri < ugly English bulja < embrulhar ‘to wrap, pack’ Portuguese d. obstruent-obstruent, homorganic kussuma < costumar ‘to usually do sth’ Portuguese sissa < sister English e. obstruent-obstruent, not homorganic kássika < casca ‘bark, skin’ Portuguese klossibai < close by English

To summarise, nasal-obstruent clusters are the only word-internal consonant sequences which are categorically retained in Saramaccan. Permissible onset sequences of the type obstruent-sonorant also tend to be preserved, but all remaining medial clusters are either strongly dispreferred or categorically banned from the creole. With respect to the majority repair option, we find variability across different types of clusters.

5.4.4 Summary: Early Saramaccan The above analyses have shown that Saramaccan imposes comparatively tight restrictions on syllable structure. However, also in this creole, CV-syllables are not the only permissible type. Word-initially, most complex onsets are retained. Only clusters with a falling sonority profile show a strong tendency towards restructuring. Mostly, the edgemost consonant is lost, but a subset of [s]-plosive sequences are broken up by an epenthetic vowel instead. Vowel-initial words show varying preference patterns depending on the number of syllables. Retention of onsetless structures is the pre-

152 | 5 Syllable structure and phonotactic restructuring in the English-based creoles ferred option with short words, whereas deletion of the initial syllable applies to the majority of longer words. At the right word-edge, we find more substantial changes to lexifier syllable structure. The only permissible word-final codas are [m] and [n]. All remaining simple or complex codas undergo repair. Vowel paragoge applies almost across the board and is partly supplemented by other, additional repairs. The findings for word-internal position largely confirm the restrictions found at the word edges, although the preferred repair strategies may differ. What is important to note, however, is that the range of possible coda consonants is slightly extended in medial position as compared to the end of words. Thus, the velar nasal [ŋ] is permitted alongside the other two nasal consonants wordinternally. We also find evidence of variable retention of the rhotic [r] in coda position, but only before the velar plosives [ɡ] and [k]. Due to the small number of cases, it is impossible to determine any concrete preference pattern for these two cluster types. Also the non-homorganic obstruent cluster [ks] appears to be permissible in word-medial position. The only evidence for this comes from the result of complex coda repairs, however. Unlike consonantal clusters, word-internal vocalic sequences show a highly consistent behaviour. They are categorically preserved in the creole.

5.5 Results II: Early St Kitts In this section we turn to syllable structure and phonotactic restructuring in the second English-based creole investigated here, Early St Kitts. It will be shown that many lexifier structures are not ruled out completely in St Kitts. Although phonotactic restrictions exist, we find that in each position certain types of clusters surface intact in the creole, making St Kitts one of the more liberal creoles in the present set. I will deal with word-initial and word-final structures in sections 5.5.1 and 5.5.2, respectively, before turning to the investigation of word-internal positions in section 5.5.3. The final section will provide a short summary of the results.

5.5.1 Word-initial onsets in Early St Kitts The bar plots in figure 5.14 provide a first overview of the distribution of wordinitial onset structures in Early St Kitts roots and etyma. Simple onsets are distinguished from complex onsets with two or more consonants and from vowel-initial words (‘no onset’).

5.5 Results II: Early St Kitts

300 250 200

frequency

76.8%

50 7.1%

complex

no onset

0

9.2% simple

type of onset

11.8%

11.3%

0

50

100

150

250 200 150

83.7%

100

frequency

153

St Kitts etyma

300

St Kitts roots

|

simple

complex no onset type of onset

Fig. 5.14: Word-initial structures in St Kitts roots and etyma (N=380)

As the figure shows, simple onsets make up the vast majority of word-initial structures in both the creole and the lexifier words. The two remaining types of structure are comparatively infrequent, but neither complex onsets nor onsetless first syllables are categorically banned from St Kitts. The differences in numbers between the two distributions are not very pronounced, but just reach significance level (Pearson’s χ2 =6.02 df = 2, p ungry and hear > irry, an etymological [h]-onset is lost, resulting in creole words with onsetless word-initial structures. In the other two affected etyma, belong and believe, the vowel of the initial, unstressed syllable is dropped, but both the preceding and following onset consonants are retained. The corresponding roots, brom and bree, thus end up with word-initial clusters.

300

154 | 5 Syllable structure and phonotactic restructuring in the English-based creoles

creole onsets

150

98.6%

50

100

frequency

200

250

simple complex no onset

58.1%

complex

no onset

0

71.1% simple

onsets in etyma

Fig. 5.15: Correspondences between word-initial structures in St Kitts roots and etyma (N=380)

With the exception of these four root-etymon pairs, all non-corresponding parts of the data show changes towards less marked structures. Almost 30% of all etyma with onset clusters surface in St Kitts with only a simple onset (cf. the light-shaded section in the middle bar). For vowel-initial words, the rate of correspondence is still lower. 41.9% of the pertinent lexifier words feature a single word-initial consonant instead (cf. the light-shaded section in the rightmost bar). I will discuss the relevant subsets of the data in turn, dealing first with complex lexifier onsets and then with vowel-initial etyma and their treatment in the creole.

5.5.1.1 Complex onsets in Early St Kitts The St Kitts data contain a total number of 45 etyma with word-initial complex onsets. As we saw above, the majority of these clusters are retained in the creole, but we also find a considerable number of cases with restructuring. The only attested repair strategy is consonant deletion. The table in (20) provides the frequencies and percentages of the two treatment options, retention and consonant deletion, as well as an example of each group.

5.5 Results II: Early St Kitts

(20)

|

155

Processes affecting word-initial complex onsets in ESK Process

N

%

none consonant deletion

32 13

71.1 28.9

Σ

45

100.0

Example clebber < clever kin < skin

In a CHAID analysis with restructuring process as the dependent variable and the structural properties of the etyma as predictor variables, the sonority profile of the onset cluster emerged as the best predictor of cluster treatment. The tree diagram in figure 5.16 illustrates the result. Complex onsets with a falling sonority profile are treated differently from complex onsets in which sonority rises towards the syllable peak. Clusters with falling sonority (cf. node 1) are categorically simplified. More precisely, they lose the first, edgemost consonant. This group comprises mostly [s]-plosive onsets, but also two etyma with prenasalised stops. The latter end up as plain stops in St Kitts. Node 2 shows almost the reverse situation. Complex onsets with a rising sonority profile are typically preserved in the creole. The two single cases of conRestructuring of complex onsets Node 0 Process % none 71.11 C-deletion 28.89 Total (100.00)

N 32 13 45

Sonority profile of the cluster Adj. P-value=0.0000, Chi-square=38.8910, df=1

falling Node 1 Process % none 0.00 C-deletion 100.00 Total (24.44)

rising

N 0 11 11

Node 2 Process % N none 94.12 32 C-deletion 5.88 2 Total (75.56) 34

Fig. 5.16: Treatment of complex lexifier onsets in St Kitts

156 | 5 Syllable structure and phonotactic restructuring in the English-based creoles sonant deletion can be regarded as exceptions.¹⁴ An extended list of examples illustrating the treatment of the two types of initial clusters is given in (21). The donor language is only given for words not derived from English. (21)

Treatment of word-initial complex onsets in St Kitts – regrouped Root Etymon Gloss Donor a. falling sonority profile bockra < mbakara ‘white person’ Efik kin < skin perrit < spirit b. rising sonority profile bruk < broke(n) clebber < clever sweet < sweet

We can conclude that St Kitts does not ban complex onset structures per se, but readily allows word-initial clusters as long as they obey sonority sequencing.

5.5.1.2 Vowel-initial words in Early St Kitts The data set for the investigation of vowel-initial structures consists of 43 rootetyma pairs. As we saw above, this is the lexifier structure for which we find the highest rate of restructuring. The table in (22) gives an overview of the repair options which apply, listed in order of frequency. (22)

Processes affecting vowel-initial words in ESK Process

N

%

none syllable deletion consonant epenthesis

25 15 3

58.1 34.9 7.0

Σ

43

100.0

14 Note, however, that these deletions differ from the ones just discussed for [s]-plosive clusters in that, here, the second consonant is lost rather than the first one (e.g. quickly > cookly).

5.5 Results II: Early St Kitts

|

157

Among those etyma which undergo restructuring, the majority (15 out of 18) lose the first, onsetless syllable.¹⁵ An alternative repair is attested in a mere three cases, which feature an epenthetic word-initial onset consonant in the creole word. Examples illustrating each of the three treatment options are provided in (23). (23)

Treatment of vowel-initial etyma in St Kitts Root Etymon a. ebry < every opin < open b. fraid < afraid leben < eleven c. hansaw < answer jie < eye

In order to find out whether the choice between retention of onsetless structures and the two attested repair options depends on the structure of the etymon, I conducted a CHAID analysis with restructuring process as the dependent variable and the number of syllables and the position of main stress in the etymon as predictor variables. The results suggest that the choice among retention of onsetless structures and the two repair options is mainly stress-governed. The tree diagram in figure 5.17 illustrates the split into two subsets with significantly different treatment. Words in which the initial syllable carries main stress behave differently from words with an unstressed first syllable. Stressed onsetless syllables (cf. node 1) show a strong tendency towards retention. The only viable repair for such syllables is the insertion of an epenthetic consonant, with attestations only for [h] and [j]. Deletion of the initial syllable, on the other hand, can be considered exceptional within this subset of the data. Etyma with non-initial stress (cf. node 2) show the opposite distribution of strategies. Here, we find categorical deletion of the initial, unstressed syllable. Consonant epenthesis is not a possible repair for these etyma, and retention of the onsetless structure is exceptional.¹⁶ Examples

15 Usually, this syllable consists of no more than the initial vowel. Only a single etymon additionally features a coda consonant, which is lost along with the vocalic nucleus (cf. almost > mose). 16 Whether this is solely an effect of constraints on syllable structure, however, is not entirely clear. Recall in this connection that we also saw a small number of unstressed vowel deletions in initial syllables which contain an onset (cf. the discussion of newly created initial clusters). Word-level constraints on prosodic structure may constitute an additional factor at work here.

158 | 5 Syllable structure and phonotactic restructuring in the English-based creoles Restructuring of vowel-initial words Node 0 Process % none 58.14 C-epenthesis 6.98 syllable deletion 34.88 Total (100.00)

N 25 3 15 43

Initial syllable in etymon stressed? Adj. P-value=0.0000, Chi-square=39.8731, df=2

yes Node 1 Process % N none 85.71 24 C-epenthesis 10.71 3 syllable deletion 3.57 1 Total (65.12) 28

no Node 2 Process % none 6.67 C-epenthesis 0.00 syllable deletion 93.33 Total (34.88)

N 1 0 14 15

Fig. 5.17: Treatment of vowel-initial lexifier words in St Kitts

of the relevant treatment options of etyma with initial and non-initial stress are given in (24a) and (24b), respectively. (24)

Treatment of vowel-initial etyma in St Kitts – regrouped Root Etymon a. initial syllable stressed ebry < every opin < open hansaw < answer yie < eye b. initial syllable unstressed fraid < afraid leben < eleven nuff < enough

In summary, the above analysis showed that onsetless word-initial structures are tolerated in Early St Kitts provided that the pertinent syllable is stressed. If this condition is met, then the only alternative to preservation is consonant epenthesis, which, however, applies only rarely. In contrast, unstressed vowel-initial syllables are categorically deleted. The result in the creole is either a simple or a complex onset, depending on the structure of the second syllable in the respective etymon.

5.5 Results II: Early St Kitts

|

159

5.5.2 Word-final codas in Early St Kitts In the previous sections, we saw that in Early St Kitts we find a clear division between permissible and categorically banned word-initial structures. In the following sections, I will show that the restrictions imposed on structures at the right edge of words are of a more probabilistic nature. The bar plots in figure 5.18 give an overview of the distributions of word-final structures in St Kitts roots and etyma. Open syllables (labelled ‘none’) are distinguished from simple codas and sequences of two or more consonants. Note that the data contain no final clusters involving syllabic consonants, so that all consonantal sequences can be considered complex codas.

150

57.6%

100

frequency

55.5%

100

frequency

150

200

St Kitts etyma

200

St Kitts roots

41.3% 50

0

3.2% none

C

CC(C)

8.9% 0

50

33.4%

none

C

CC(C)

Fig. 5.18: Word-final consonants in St Kitts roots and etyma (N=380)

The two distributions exhibit similarities in the relative frequency of the different types of structures, but show significant differences with respect to the absolute frequencies (Pearson’s χ2 =13.84 df = 2, p turraw), for the other, we find variation between simplification and retention of hiatus (society > society ~ sissety). What we can say, thus, is that hiatus is at least not unrestricted in St Kitts. However, due to the very small number of cases, it is impossible to draw any more definite conclusions.

17 A single exceptional case is attested, in which the deletion of an unstressed medial vowel leads to the creation of a new consonant cluster. The pertinent root-etymon pair is pickney < piccaninny ‘child’.

168 | 5 Syllable structure and phonotactic restructuring in the English-based creoles

creole structures

60 40

98.9%

74.4%

20

frequency

80

only VCV CC(C) VV

0

33.3% only VCV

CC(C)

VV

word−internal structures in etyma

Fig. 5.23: Correspondences between word-internal structures in St Kitts roots and etyma (N=130)

5.5.3.1 Word-internal consonant sequences in Early St Kitts Overall, 39 etyma contain word-medial consonant sequences. The table in (31) shows that only two regular options exist for such structures in St Kitts. (31)

Processes affecting medial clusters in ESK Process

N

%

none consonant deletion metathesis

26 12 1

66.7 30.8 2.6

Σ

39

100.0

Word-internal clusters are either retained without modification or simplified by deletion of one of the cluster members. Only a single etymon undergoes a different repair, metathesis of the first cluster member with the preceding vowel. The pertinent root-etymon pair is given in (32c). The examples in (32a) and (32b) illustrate retention and consonant deletion, respectively.

5.5 Results II: Early St Kitts

(32)

|

169

Treatment of medial clusters in St Kitts Root Etymon a. after < after ebry < every sunting < something b. binness < business nawsy < nasty c. yentremon < gentleman

The results of a CHAID analysis reveal that the manner of articulation of the first cluster member acts as the best predictor of cluster treatment. For part of the data, place of articulation also plays a role. The tree diagram in figure 5.24 illustrates the resulting splits into subsets. Restructuring of word-internal clusters Node 0 Process % C-deletion 30.77 none 66.67 metathesis 2.56 Total (100.00)

N 12 26 1 39

Manner of first cluster member Adj. P-value=0.0339, Chi-square=15.7382, df=4

nasal; plosive Node 1 Process % C-deletion 16.00 none 84.00 metathesis 0.00 Total (64.10)

fricative

[l]

Node 2 Process % N C-deletion 54.55 6 none 45.45 5 metathesis 0.00 0 Total (28.21) 11

N 4 21 0 25

Node 3 Process % N C-deletion 66.67 2 none 0.00 0 metathesis 33.33 1 Total (7.69) 3

Place of cluster members Adj. P-value=0.0022, Chi-square=9.4166, df=1

homorganic Node 4 Process % N C-deletion 85.71 6 none 14.29 1 metathesis 0.00 0 Total (17.95) 7

Fig. 5.24: Treatment of medial clusters in St Kitts

not homorganic Node 5 Process % C-deletion 0.00 none 100.00 metathesis 0.00 Total (10.26)

N 0 4 0 4

170 | 5 Syllable structure and phonotactic restructuring in the English-based creoles At the top level, word-internal clusters which start in either a nasal or a plosive are distinguished from clusters with a fricative C1 and from lateral-initial sequences. Nodes 1 through 3 show a decrease of cluster retention from close to categorical for nasal- and plosive-initial sequences to unattested for clusters with a lateral C1 . Among the typically retained clusters (node 1), we find mostly plosive-liquid, plosive-glide and nasal-obstruent sequences. Consonant deletion applies only rarely to these clusters. The group at the other end of the acceptability scale (cf. node 3) in fact comprises only a single cluster type, the coda-onset sequence [lm]. Two of the etyma with the pertinent cluster undergo consonant deletion, the remaining one shows metathesis. All coda laterals are thus eliminated in medial positions. However, since evidence is lacking for clusters other than [lm], it remains unclear whether other lateral-initial sequences such as, for instance, [l]-plosive would behave similarly. In the following, I will therefore assume dispreference only for the attested cluster type. The one remaining group, fricative-initial clusters, shows variable behaviour overall (cf. node 2). Here, a second split allows for a more differentiated analysis. As the diagram illustrates, this part of the data can be further divided into two subsets with almost contrary behaviour. The choice among preferred simplification (node 4) and categorical preservation (node 5) is determined by the place features of the cluster members. Homorganic clusters usually undergo consonant deletion. Besides the common [st]-clusters, this group also includes an instance of the fricative-nasal sequence [zn]. In contrast to these homorganic sequences, all attested non-homorganic clusters surface intact in St Kitts. Only three different cluster types occur in this group, [ft], [sm], and [vr].¹⁸ The table in (33) provides a rearranged list of examples, illustrating the treatment options for all different subsets of medial clusters which emerged from the analysis. The order follows the numbering of the terminal nodes in the tree diagram.

18 In the latter case, regular substitution of [b] for [v] applies, so that the cluster surfaces as plosive-liquid sequence [br] in St Kitts.

5.5 Results II: Early St Kitts

(33)

|

171

Treatment of medial clusters in St Kitts – regrouped Root Etymon Gloss a. C1 = lateral: [lm] gor-a-mity < God-all-mighty yentremon < gent[əlm]an b. C1 = nasal or plosive between < between hungry < hungry sanny < sandy c. C1 = fricative, homorganic cluster sister < sister binness < business nawsy < nasty e. C1 = fricative, non-homorganic cluster after < after musmillin < mu[sm]elon ‘muskmelon’

To summarise, St Kitts readily permits certain types of word-internal clusters, including plosive-approximant and nasal-obstruent sequences, but strongly disfavours others. The two most clearly dispreferred sequences are the homorganic obstruent-cluster [st] and the all-sonorant cluster [lm].

5.5.4 Summary: Early St Kitts Overall, St Kitts emerged as a relatively liberal creole language. In both onset and coda position, certain types of complex structures are allowed. Nonetheless, neither syllable position is completely unrestricted. At the left word edge two types of structures are banned, complex onsets which do not obey sonority sequencing and unstressed initial vowels. Word-finally, only a minority of non-nasal codas are repaired in St Kitts. Even complex codas are not ruled out completely, although acceptability varies greatly across different cluster types. Restructuring mechanisms target in particular complex codas that end in a coronal plosive, whereas labial or dorsal (nasal-plosive) clusters or clusters ending in a fricative are more likely to be preserved. The only attested all-sonorant cluster, [rm], never surfaces intact in the creole. In medial positions we find the same types of complex onsets that are also permitted word-initially. Coda-onset sequences consisting of a nasal and homorganic obstruent are also readily tolerated in the creole. However, not all word-internal

172 | 5 Syllable structure and phonotactic restructuring in the English-based creoles clusters are equally acceptable. Especially the homorganic obstruent-cluster [st] and the all-sonorant cluster [lm] are avoided in St Kitts. Not much can be said about the preferred treatment of hiatus in St Kitts. Both simplification and retention are attested among the pertinent items, but the cases are simply too few to allow for generalisations.

5.6 Comparison: Early Saramaccan vs. Early St Kitts In the above analyses, Saramaccan has emerged as the clearly more restrictive of the two English-based creoles. Especially in word-final position, the Surinamese creole imposes much tighter restrictions on syllable structure. Word-initially, on the other hand, we found remarkable parallels between Early St Kitts and Early Saramaccan. With respect to word-initial complex onsets, the two English-based creoles show comparable behaviour both in what kinds of onset structures they allow or disallow and with regard to processes of restructuring that apply to dispreferred lexifier structures. Most types of complex lexifier onsets are routinely preserved. Only onset clusters with a falling sonority profile, i.e. [s]-plosive clusters and prenasalised stops, are targeted by repairs. Restructuring is categorical in St Kitts and strongly preferred in Saramaccan. In case of repair, it is without exception the edgemost segment that is lost, resulting in a simple plosive in all pertinent creole words. Also with respect to vowel-initial words we find similarities between St Kitts and Saramaccan to the effect that both creoles variably retain and repair such onsetless structures. In both creoles, the most common repair mechanism is deletion of the initial, onsetless syllable. St Kitts and Saramaccan differ, however, in what determines the choice between retention and restructuring. In St Kitts, stress emerges as the best predictor of etymon treatment. Onsetless initial syllables are deleted only if they do not carry main stress in the etymon. In contrast, empty onset positions are usually tolerated in stressed first syllables. In Saramaccan, word length appears to play a greater role for the treatment of vowel-initial words. Syllable deletion is common only in etyma of three or more syllables. Turning to the right edge of words, we find much greater differences between the creoles than in initial position. Here, St Kitts and Saramaccan show strongly divergent behaviour. In Saramaccan, the only consonants that are permitted in word-final position are the nasals [m] and [n], and even these only surface intact regularly if the final syllable of the etymon is stressed. Word-final consonants other than [m] and [n] as well as final clusters are categorically banned and typically undergo vowel paragoge. In the case of clusters, a second, additional restruc-

5.6 Comparison: Early Saramaccan vs. Early St Kitts

|

173

turing mechanism applies in the majority of cases. Contrary to Saramaccan, St Kitts allows both nasal and non-nasal singleton consonants in word-final position with only a small minority of repairs applying to non-nasal codas. Also clusters are possible in St Kitts, although not all types of clusters are equally acceptable. The most clearly dispreferred sequences are clusters ending in a coronal plosive, which show a strong tendency towards deletion of that plosive. Word-internally, we find overall more similarities than differences between St Kitts and Saramaccan. Regarding the acceptability of hiatus, the situation is unclear for St Kitts. The few attestations show both retention and repair, but do not allow us to draw any conclusions on what might be the preferred option. In Saramaccan, cases of hiatus are usually tolerated. Clear parallels between the two creoles can be found with respect to wordinternal complex onsets. Consonant clusters which are readily permitted in wordinitial position also tend to surface intact word-medially in both creoles. Etymological coda-onset sequences are subject to a more diverse treatment. Whereas homorganic nasal-obstruent sequences are allowed in both St Kitts and Saramaccan, other types of medial clusters hardly ever surface intact in Saramaccan. St Kitts is slightly more permissive in this respect, tolerating non-homorganic fricative-initial sequences. However, also the more liberal island creole imposes restrictions on medial clusters. Homorganic [st]-sequences are typically reduced to the sibilant and instances of the all-sonorant cluster [lm] are repaired by either liquid-deletion or metathesis.

6 Syllable structure and phonotactic restructuring in the French-based creoles 6.1 Introduction This chapter deals with syllable structure and phonotactic restructuring in the two French-based creoles, Guiana French Creole (Guiana FC, GFC) and Trinidad French Creole (Trinidad FC, TFC). We will see that the two creoles overall show remarkable similarities in their treatment of lexifier structures. Both permit certain types of complex onsets, variably repair onsetless structures, avoid coda rhotics and categorically ban word-final clusters. Differences between Guiana FC and Trinidad FC can be found mainly in the types of onset clusters that are targeted by restructuring mechanisms. The chapter opens with a short overview of the pertinent previous studies. Section 6.3 will then introduce aspects of historical development and variation that were considered in the coding of the French creole data. In sections 6.4 and 6.5, I will provide detailed accounts of the phonotactic patterns and restructuring mechanisms attested in Guiana FC and Trinidad FC respectively, before concluding the chapter with a brief comparison of the two creoles in section 6.6.

6.2 Previous studies Both of the sources for my data on the French-based creoles are 19th century grammar sketches. Both include comments on sound structure, but the extent of such comments and the amount of detail varies. Since none of the two works is empirical in nature, quantifications are given only insofar as certain sounds or structures are said to be absent entirely from the creole. In his study on the grammar of Guiana FC, Saint-Quentin (1872) includes a discussion of the sounds that occur in the creole when he introduces his orthographic conventions. Apart from these segmental issues, not much is said about phonology. The only observations on syllable structure concern syllable-final rhotics, which Saint-Quentin (1872: 171) says are eliminated without exception when they occur in word-final position and in most cases when word-medial. Apart from deletions, Saint-Quentin reports also processes of metathesis of a rhotic with the preceding vowel. Thomas (1869) is much more detailed in his investigation of creole sound structure. He describes the phonetic system of Trinidad FC as spoken in the mid-

6.2 Previous studies

| 175

19th century, commenting also on differences between the sound inventories of the French lexifier and the creole. In addition, he discusses changes such as sound substitutions, metathesis, epenthesis and deletion of sounds. As AubBüscher (1969) states in her introduction to Thomas’s grammar reprint, not all of the described changes can be taken at face value. For instance, Thomas does not distinguish in terminology between sounds and letters, so that not every modification that he entitles epenthesis actually signifies the addition of a nonetymological sound. Some of the listed cases are examples of changes in sound quality, instead (e.g. agacer [aɡase] > angacer [˜ ɑɡase]). Problems occur, too, with a number of etyma he assumes for the creole words. Not in all cases does he arrive at the most plausible conclusion (cf. comments by Aub-Büscher). Despite the above-mentioned problems, the grammars by Saint-Quentin and Thomas are the most comprehensive and most careful accounts of the early varieties of GFC and TFC. In fact, no later study that I am aware of provides independent descriptions of the sound structure of the early stages of either creole. Where later accounts even exist, they cover later stages of the creoles. An example of this is Saint Jacques Fauquenoy (1972), who gives a synchronic account of the phonology of 20th century Guiana FC. An early comparative study, Van Name (1869), uses the information provided in Thomas (1869) and other early texts on creoles to give an overview of “the principle features of Creole grammar” (Van Name 1869: 126). Apart from Trinidad FC, four other creoles with French-derived lexicons are considered. The survey includes a short passage on phonetic differences between the French-based creoles and the lexifier, where both regular substitution patterns for individual sounds and common restructuring processes are considered. Van Name aims at providing a qualitative account of the main features of the French creoles. No single creole is therefore discussed with all its characteristics nor are detailed quantifications given for restructuring processes. A 20th century cross-linguistic study is provided by Stein (1984), who deals with the French-based creoles and their relation to the superstrate. Both Guiana FC and Trinidad FC are included in the set of languages discussed. Stein focuses on the features shared by (the majority of) the creoles, covering aspects of the sound system, morphology and syntax. The author explicitly states that his aim is not to describe any given creole in its entirety, but to point out the main, regular patterns that characterise the French creoles as a group. His information on the individual

176 | 6 Syllable structure and phonotactic restructuring in the French-based creoles creoles is taken from various different sources, which, according to Stein (1984: 3) himself, vary considerably with respect to both quality and quantity.¹ Comparative studies involving French-based creoles have also been conducted by Parkvall (1995, 1999b, 2000). Parkvall (1995) compares selected features of French-lexicon creoles, including some phonological aspects such as the presence or absence of postvocalic rhotics and particular segmental substitution patterns. Apart from segmental aspects, also syllable structure in French creoles is addressed briefly in Parkvall (1999b) and Parkvall (2000). The discussion remains very general, however, and no distinctions are made between individual creoles in this section. Parkvall notes that both onsets and codas are often simplified in French creoles as compared to the lexifier, but that there is no evidence to suppose that any French creole ever allowed only CV and V syllables (cf. Parkvall 2000: 53). Concerning restructuring processes, he remarks that, while consonant deletions are widely attested, vowel paragoge is never employed as a repair strategy for illicit word-final structures in French-lexicon Atlantic creoles.

6.3 Methodological issues This section discusses language-specific aspects which were considered in the coding of the data. Before dealing with individual cases of phonetic variation and historical developments in the pronunciation of lexifier words, I will briefly comment on the more general question of which forms the creole verbs should be derived from. Which forms are the verbs derived from? In the choice of verbal etyma I followed the assumptions made by Thomas (1869: 44ff), who distinguishes four different cases. The first, most common case involves derivation from the infinitive as, for instance, in accomplir > accomplî ‘to fulfil’ or craindre > craine ‘to fear’. Not all creole forms, however, are consistent with this hypothesis. A number of creole verbs are more closely related in form to French past participles, indicatives or imperatives. Thus, môr ‘to die’ is more likely to represent a continuation of the past participle mort than an adaptation of the infinitive mourir. Similarly, pé ‘to be able’ can straightforwardly be derived from 1st

1 As Stein does not go into more detail and the exact sources for individual languages are not indicated, I was unable to obtain more detailed information on the nature of his sources and their limitations.

6.3 Methodological issues

| 177

or 3rd person singular indicative forms peux/peut [pø], whereas a relation to the infinitive pouvoir seems highly doubtful. Regarding the phonetic realisation of individual verb forms, the most notable question is that of the presence or absence of the final rhotic in the French infinitives. According to records on French historical phonetics (cf. e.g. Thurot 1881-83), there used to be variation across different classes of verbs with some of them typically being [ʁ]-less and others usually maintaining the rhotic. The relevant details are given below in the section that deals with word-final rhotics in general. Replacement of [ʎ] by [j] At least up to the 18th century, studies on the pronunciation of French report a palatal lateral approximant [ʎ] in words such as travail or fille, where today we would expect to find [j]. Rickard (1989: 126f) assumes that usage of [j] had become general by the middle of the 19th century. The creole input thus likely still contained the so-called l mouillé [ʎ] and was coded accordingly. The pronunciation of the rhotic /r/ According to Fox and Wood (1971: 49), the pronunciation of the rhotic changed during the course of the 18th century, at least in Parisian speech. The formerly regular alveolar trill [r] started to give way to uvular [ʁ], which then spread to varieties outside Paris. The historical records thus do not provide unequivocal evidence for either pronunciation. Both variants could in principle have been present in the creole input. The creole facts support such an assumption, but point to the uvular variant as the more influential one. In his small grammar of Guiana FC, Saint-Quentin (1872) attests to the existence of different variants of /r/ in the Guyana region. He says: ”[...] il y a des localités où l’on prononce l’ ɾ comme en Espagne et dans le Midi de la France; mais, généralement, à Cayenne, on le prononce comme à Paris” (1872: 138). Given the contrasting of Spanish and Parisian pronunciations, uvular [ʁ] must have been well-established in Paris at the time Saint-Quentin wrote his grammar. His observation that guttural realisations of /r/ were more commonly used than alveolar ones also in the creoles suggests that a similar predominance of [ʁ] can be assumed already for the 18th century varieties of French which served as input to the creoles.² I have therefore coded the rhotic in the French etyma as dorsal rather than coronal.

2 For comparable assumptions about the pronunciation of the rhotic in French etyma see also Steele and Brousseau (2006), Russell Webb (2010), Brousseau (2011).

178 | 6 Syllable structure and phonotactic restructuring in the French-based creoles Concerning the exact nature of the creole rhotics, Stein (1984: 25f) states that the typical realisation in the French-based creoles is not uvular [ʁ], but velar [ɣ] (see also Parkvall (2000), Brousseau (2005, 2007, 2011), Steele and Brousseau (2006)). Uvular variants are attested in certain words in the modern varieties of the creoles, but, according to Stein (1984: 26), the pertinent lexical items were borrowed from French only at a later stage. Following this argument, I generally use [ɣ] as representation of the rhotic in the cited creole forms. Like all other regular segmental substitutions, also the replacement of uvular [ʁ] by velar [ɣ] will not be investigated further. Reinterpretations caused by liaison or enchaînement A number of nouns that, in French, start in a vowel, have been reinterpreted by creole speakers as containing an initial consonant [n] or [z]. These consonants are not creole innovations, but direct continuations of French phonetic realisations involving liaison or enchaînement. Regular French pronunciation of articlenoun sequences such as un homme ‘a man’ or des herbes ‘some herbs’ would have involved syllabification of the article-final consonant in the onset of the following syllable, resulting in the apparent nominal forms [nɔm] and [zɛʁb], respectively. The creole forms nomme ‘man’ and zêbe ‘grass’ as well as all analogously derived forms simply maintain the onset structure of the phonetic input and were therefore coded as not involving any initial restructuring. Word-initial [h] Whereas, in modern French, words such as haler ‘to pull, haul’ and hardes ‘clothes’ start in a vowel, the glottal fricative [h] was still pronounced in 18th century Parisian French.³ It is unclear at which point in time weakening started. The majority of sources cited in Thurot (1881-83: 391ff), however, indicate that aspiration persisted into the 18th century. Similarly, Walter (1993) assumes that the glottal fricative was in fact still fully realised (see also Brousseau (2005, 2007, 2011)). Following these sources, all words starting with ‘h aspiré’ were coded as [h]-initial.

3 Note that this does not apply to all French words spelled with initial , but only to those with so-called ‘h aspiré’. In contrast to initial ‘h muet’ (e.g. in hôtel ‘hotel’ and heure ‘hour’), initial ‘h aspiré’ prevents élision and liaison, i.e. words with initial ‘h aspiré’ pattern with consonantinitial words. Words with ‘h muet’, on the other hand, behave like all other vowel-initial words and were vowel-initial in pronunciation also in 17th and 18th century French.

6.3 Methodological issues

| 179

Word-final /t/ Up to the 18th century, final /t/ was still commonly realised before a pause if it followed a short stressed vowel (cf. Thurot 1881-83: 92f). Pertinent lexifier words whose creole reflexes show final [t] were therefore coded as consonant-final. Word-final /r/ In earlier varieties of French, the rhotic could be elided not only in words which have lost it entirely in modern French, but also in words that today show consistent realisation. Verbal infinitives ending in (e.g. manger ‘to eat’) are an example of the former group, being vowel-final in modern French pronunciation. Verbal infinitives in (e.g. partir ‘to leave’), on the other hand, end in a rhotic in modern French. In 16th and 17th century pronunciations, both groups of verbs commonly lacked the final rhotic (cf. e.g. Fox and Wood 1971, Rickard 1989, Thurot 1881-83). It was realised only when a pause followed.⁴ Pierret (1981) and Rickard (1989) add infinitives in (e.g. savoir ‘to know’) to the list of forms with regular [ʁ]-less pronunciation. For all remaining verbal infinitives, on the other hand, realisation of the rhotic appears to have been predominant and was therefore assumed in the coding of the data. Apart from verbs, nouns are also reported to show loss of the final rhotic in the Parisian speech of the time. Thurot (1881-83) states that /r/ often failed to be realised in nouns ending in and as well as in deverbal nouns in . Furthermore, deverbal masculine forms ending in with a corresponding feminine form in had lost the rhotic, presumably due to confusion with pairs in . Latin-derived nouns ending in , on the other hand, typically maintained final [ʁ] (e.g. in auteur, malheur, cf. Thurot 1881-83: 164f). Although pronunciation of the final rhotic was later on gradually restored in many of the above classes, [ʁ]-less forms are still reported for the 18th century (cf. e.g. Deslandes 1781, Walter 1994). Since the rhotic is also systematically absent in the corresponding creole words, the pertinent etyma were coded as showing an [ʁ]-less pronunciation, thus avoiding an over-interpretation of the data. Word-final C-/r/ and C-/l/ Thurot (1881-83) documents variation in the realisation of underlying word-final consonant-liquid sequences. Final liquids could be dropped after another consonant, a process that is also widespread in modern French, at least in infor-

4 It also continued to occur in liaison if the following word started in a vowel. In the latter case, the rhotic would have been syllabified into the onset of the next syllable, a position in which it would have been likely to be interpreted as part of the following word rather than the verb.

180 | 6 Syllable structure and phonotactic restructuring in the French-based creoles mal contexts (Brousseau, p.c.). It remains unclear, however, how general the phenomenon was in 17th and 18th century French. The phonetic context would certainly have influenced the choice between full and reduced forms with retention of the liquid being more likely if the following word started in a vowel and deletion applying more readily before another consonant. Both Guiana FC and Trinidad FC show consistent absence of the liquid in the pertinent forms. Given that liquid deletion must have been variable in the input, the question remains which etyma should be assumed to have featured a final cluster and which ones shouldn’t. For lack of more detailed information on the frequency of liquid deletion in final C-/r/ and C-/l/ clusters, I chose to base my codings on lexical evidence. Absence of the liquid was interpreted as reflecting lexifier pronunciation only if the historical records mentioned simplification of the cluster for the particular lexical item. Items explicitly listed are, for instance, autre, quatre, table, double (cf. Thurot 1881-83: 266ff, 280ff). Even though this procedure may not result in a perfect replication of the phonetic facts in the lexifier, it gives us at least an approximation to these facts, including both full and reduced forms in the pool of etyma. We cannot, of course, exclude the possibility that also some of the other, remaining items entered the creole in their simplified form. The quantifications in the pertinent part of the analysis should therefore be viewed with some reservations. Other cluster simplifications For a small number of lexical items, Thurot (1881-83) reports the use of two different variants, one including a coda [l], the other lacking it. In the interest of conservative coding, I have assumed that the lateral approximant was absent already in the pertinent etyma if the creole reflexes show no trace of it. Lexifier words in point are malgré ‘in spite of’, quelque ‘some’, and soldat ‘soldier’, (cf. Thurot 188183: 258ff). A process of assimilation is documented for underlying /rl/, which, according to the historical sources, was realised as simple [l], for instance, in parler ‘to speak’ (cf. Thurot 1881-83: 289) and was coded as such in my data. Simplification in the lexifier likewise was assumed for medial /ks/-C sequences (e.g. in extrème or excuser), for which Thurot (1881-83: 339ff) cites common reduction to [s]-C, especially in frequent words. In word-final position, there appears to have been variation in the realisation of underlying /kt/. Opposing claims in the historical records suggest that simplifications occurred alongside unmodified realisations. Both loss of [k] and loss of [t] are documented as alternatives. The few pertinent creole reflexes, however, all

6.4 Results I: Guiana French Creole |

181

show simplification to [k], never to [t]. Since the creole data do not reflect the variation in lexifier pronunciation reported in Thurot (1881-83), I assumed that the French input contained the original cluster. Finally, underlying /sk/ in words such as puisque, presque, and jusqu’(à) could be simplified to [k] in pronunciation. The different quotations in Thurot (1881-83: 19f) show that there was variation between full and reduced forms. Reduction to [k] is attested in both creoles, but is categorical in neither. I therefore assumed that those items which show this reduction were simplified already in the input. The remaining etyma were coded as containing a final cluster.

6.4 Results I: Guiana French Creole In this section, I will give a detailed description of permissible syllable structures in Guiana FC as well as discuss attested restructuring mechanisms that apply to illicit lexifier structures. It will be shown that Guiana FC imposes tighter restrictions on word-final than on word-initial structures. Most rigorously avoided are coda rhotics in general and consonant clusters in word-final position. Marked structures at the left word edge, in contrast, are at least variably retained. Sections 6.4.1 and 6.4.2 will be concerned with word-initial and word-final structures, respectively, and restrictions on different word-internal structures will be dealt with in section 6.4.3. Finally, I will provide a brief summary of the findings on Guiana FC in section 6.4.4.

6.4.1 Word-initial onsets in Guiana FC The two bar plots in figure 6.1 give a first overview of the types of word-initial structures that we find in the two sets of creole words and assumed etyma, respectively. As before, simple onsets are distinguished from sequences of two or more consonants (labelled ‘complex’) and from vowel-initial words (labelled ‘no onset’). The distribution of word-initial structures in the two data sets does not differ significantly (Pearson’s χ2 =1.86 df = 2, p=0.3941). In both roots and etyma, simple onsets are the most common word-initial structure by far, but complex onsets and onsetless structures are also not ruled out completely. Given these numbers, one might expect that most word-initial lexifier structures are retained without modification in Guiana FC, with changes applying only to vowel-initial words. A closer look at the relation between creole and etymon structures, however, paints a slightly different picture. The bar plot in figure 6.2 illustrates to what degree creole words exhibit the same type of word-initial structure as their respective etyma.

182 | 6 Syllable structure and phonotactic restructuring in the French-based creoles

100

frequency

100

76.1% 50

78.9%

0

14.1% simple

complex type of onset

7% no onset

13.1%

10.8%

0

50

frequency

150

Guiana FC etyma

150

Guiana FC roots

simple

complex no onset type of onset

Fig. 6.1: Word-initial structures in Guiana FC roots and etyma (N=213)

Lexifier structures are given as separate bars, whereas the resulting structures in the creole are represented by different shadings within each bar. Correspondence rates between creole and etymon structures, i.e. the percentage of creole words that surface with the same type of structure as their assumed etyma, are given for each of the lexifier categories. For instance, a correspondence rate of 96.3% for simple onsets (cf. the leftmost bar) means that in 96.3% of all cases in which the etymon has a simple onset, the creole form also has a simple onset. In the remaining 3.7% of cases with simple onsets in the etyma, however, the word-initial structures in the creole forms differ from those in their assumed etyma. Comparing the correspondence rates in the plot to our expectations from above concerning targets of restructuring, we find that only vowel-initial structures (cf. the rightmost bar) behave as expected. About one third of the corresponding creole words exhibit a change in onset structures and end up with a single onset consonant (indicated by light shading of the relevant section). The correspondence rates for simple and complex onsets are higher than for onsetless structure, but lower than we would have hypothesised on the basis of the distribution plot in figure 6.1. What we observe here instead of plain preservation is a kind of trade-off in numbers between the two categories. A small percentage of etyma with simple onsets have creole reflexes that show a complex onset instead (cf. the middle grey section in the leftmost bar). On the other hand, some of the clusterinitial etyma end up with a simple onset in Guiana FC (cf. the light section in the middle bar). Of these two changes, only the latter can be explained as caused by

6.4 Results I: Guiana French Creole |

183

creole onsets

100

96.3% 50

frequency

150

simple complex no onset

65.2%

complex

no onset

0

82.1% simple

onsets in etyma

Fig. 6.2: Correspondences between word-initial structures in Guiana FC roots and etyma (N=213)

restrictions on onsets. Why a language should create complex onsets from simple ones, on the other hand, is not quite as obvious. We will therefore take a closer look at the pertinent data, before dealing with complex onsets and vowel-initial words in more detail. The data in (1) illustrate the two different mechanisms that create onset clusters from simple lexifier onsets. (1)

Newly created complex onsets in Guiana FC Root Etymon Gloss a. dromi < dormir ‘to sleep’ srèk < cercle ‘circle’ b. wlé < voulez ‘to want, wish’

Most non-etymological clusters evolve through metathesis of a coda rhotic into the onset of the syllable (cf. (1a)). Out of six pertinent cases overall, five display this type of change. The remaining etymon shows an isolated development. It is not entirely clear which process should be assumed to relate the creole word to its etymon in (1b). The etymological sequence [vu] could have been reinterpreted by creole speakers as the glide [w]. Alternatively, the change could have involved deletion of the initial consonant [v] and a subsequent change of syllabic [u] to non-syllabic [w]. While the motivation for the change in (1b) remains unclear, the development of non-etymological complex onsets as in (1a) can be adduced to restructuring

184 | 6 Syllable structure and phonotactic restructuring in the French-based creoles mechanisms targeting other, non-onset structures. In this case, metathesis most likely applies to remove the word-internal rhotic from coda position. Having discussed these more a-typical cases, we can now turn to the treatment of potentially dispreferred onset structures. In the following two sections, I will deal with complex onsets and vowel-initial words in turn.

6.4.1.1 Complex onsets in Guiana FC The analyses of complex onsets in Guiana FC are based on a set of 28 etyma and their creole reflexes. The table in (2) lists the attested repair mechanisms in order of frequency, including the option ‘none’ to indicate preservation of the etymological structure. (2)

Processes affecting complex onsets in GFC Process

N

%

none consonant deletion

23 5

82.1 17.9

Σ

28

100.0

Over 80% of all lexifier onset clusters are retained without modification in Guiana FC. The remaining complex onsets are simplified by deletion of one of the cluster members. Note that, in trisegmental onsets, the result of simplification may still be a cluster, albeit one consisting of two instead of three consonants. Examples of retention of complex onsets and repair by consonant deletion are provided in (3a) and (3b) respectively.⁵

5 The pronunciation [vwɛ] for the etymon voir represents a variant which is still found in both France and Quebec (Brousseau, p.c.) and was assumed based on the vowel quality in the creole word.

6.4 Results I: Guiana French Creole |

(3)

185

Treatment of complex lexifier onsets in Guiana FC Root Etymon Gloss a. [bj˜ɛ] < [bj˜ɛ] bien ‘good’ [ɡɣ]onde < [ɡʁ]onder ‘to growl’ [tɣwa] < [tʁwa] trois ‘three’ b. [wɛ] < [vwɛ] voir ‘to see’ [pi] < [pɥi] puits ‘well, spring’ [pimal] < [plymal] plus mal ‘worse’

In order to find out whether the structural make-up of the etymological cluster determines its treatment in the creole, I conducted a CHAID analysis with restructuring mechanism as the dependent variable and the various structural variables as predictors. The chi-square-based analysis returned homorganicity of the cluster as well as the place of articulation of the first cluster member as significant predictors, suggesting that consonant deletion is common only in homorganic labial clusters. However, the subsequent application of a Fisher’s exact test showed that the difference between the suggested subsets was not in fact significant. We thus have no reason to believe that the structural properties of the onset clusters determine the likelyhood of simplification in Guiana FC. Consequently, I will assume in the following that complex onsets in general show a strong tendency towards retention in the creole, but may occasionally undergo consonant deletion.

6.4.1.2 Vowel-initial words in Guiana FC No more than 23 etyma in the data start in a vowel. The data base for the investigation of onsetless structure is thus relatively small. Nonetheless, an analysis of the data yields interesting results. The table in (4) provides an overview of all attested repairs that apply to vowel-initial etyma. (4)

Processes affecting vowel-initial words in GFC Process

N

%

none vowel deletion consonant epenthesis

15 6 2

65.2 26.1 8.7

Σ

23

100.0

As the table shows, almost two thirds of the onsetless structures are preserved in Guiana FC. The majority of those words that undergo repair lose the initial vowel,

186 | 6 Syllable structure and phonotactic restructuring in the French-based creoles which in all cases constitutes the entire syllable. Two etyma show an alternative change. They gain a non-etymological onset through consonant epenthesis. Retention of onsetless structures as well as the two repair options are illustrated by the examples in (5). (5)

Treatment of vowel-initial words in Guiana FC Root Etymon Gloss a. amba < en bas ‘below, under’ onz < onze ‘eleven’ ot < autre ‘other’ b. kouté < écouter ‘to listen’ lò < alors ‘then’ c. wot < autre ‘other’ [j]é < eux ( ous ‘you’.

6.5.1.1 Complex onsets in Trinidad FC The data set for the investigation of complex onsets in Trinidad FC comprises 112 etyma with initial clusters and their creole reflexes. The table in (17) gives an overview of all attested restructuring mechanisms, their frequencies and percentages. (17)

Processes affecting complex onsets in TFC Process none segmental change consonant deletion vowel epenthesis Σ

N

%

68 31 11 2

60.7 27.7 9.8 1.8

112

100.0

Unmodified retention of the lexifier structure (‘none’) is the most common option, attested in just over 60% of the cases. Note that this percentage is much lower than the 90.2% correspondence rate for complex onsets given in figure 6.12 above. This apparent discrepancy results from the inclusion of certain segmental changes in

6.5 Results II: Trinidad French Creole |

205

the counting of repairs. As the table in (17) shows, ‘segmental change’ is listed as the most frequent restructuring mechanism. The reason for including this type of change here may not be obvious. Recall from chapter 3 that segmental changes were usually ignored in the investigation of phonotatic restructuring if the respective substitution applies to the segment regardless of context (e.g. systematic replacement of [ð] by [d]). The segmental changes we are dealing with here, however, are of a context-sensitive nature. More specifically, the etymological rhotic [ʁ] surfaces in the creole as glide [w] in certain phonological contexts, but not in others. Apart from these segmental changes, we also find repair mechanisms at work which reduce the complexity of the lexifier onset. Mostly, these are cases of consonant deletion. Only two cases show evidence of an alternative strategy. Here, the onset cluster is broken up by the insertion of an epenthetic vowel. Examples of unmodified retention of complex onsets and all three restructuring processes are given in (18), maintaining the same order as in the frequency table above. For reasons of clarity, IPA-symbols are used for some clusters or complete etyma. (18)

Treatment of complex lexifier onsets in Trinidad FC Root Etymon Gloss a. plime < plume ‘pen’ trahî < trahir ‘betray’ [zj]ex < [zjø] les yeux ‘eyes’ b. [bw]ef < bref ‘brief’ [tw]op < trop ‘too much’ c. cochi < crochu ‘crooked’ touver < trouver ‘find’ d. fonagile < fragile ‘fragile’ gouroupier < croupier ‘servant’

It is not a coincidence that all examples of restructuring given above involve the rhotic as second cluster member. A CHAID analysis revealed that this is in fact the only type of complex onset which regularly triggers repair mechanisms. The top-level split in the tree diagram in figure 6.13 illustrates this finding. The manner of articulation of the second cluster member acts as the best predictor of cluster treatment. For onsets in which C2 is either a glide or a lateral approximant [l], retention can be considered categorical (cf. node 1). In contrast, clusters with a rhotic in second position show more variable behaviour (cf. node 2). The latter group can be further divided into subsets.

206 | 6 Syllable structure and phonotactic restructuring in the French-based creoles Restructuring of complex onsets Node 0 Process % segmental change 27,68 C-deletion 9,82 none 60,71 V-epenthesis 1,79 Total (100,00)

n 31 11 68 2 112

Manner of second cluster member Adj. P-value=0.0000, Chi-square=46.7859, df=3

rhotic

glide; lateral Node 1 Process segmental change C-deletion none V-epenthesis Total

Node 2 Process segmental change C-deletion none V-epenthesis Total

% n 0,00 0 8,70 4 91,30 42 0,00 0 (41,07) 46

% n 46,97 31 10,61 7 39,39 26 3,03 2 (58,93) 66

Place of first cluster member Adj. P-value=0.0000, Chi-square=55.9592, df=3

labial Node 3 Process segmental change C-deletion none V-epenthesis Total

coronal; dorsal

% n 90,00 27 6,67 2 0,00 0 3,33 1 (26,79) 30

Node 4 Process segmental change C-deletion none V-epenthesis Total

% 11,11 13,89 72,22 2,78 (32,14)

n 4 5 26 1 36

Fig. 6.13: Treatment of complex lexifier onsets in Trinidad FC

The second split in the tree diagram separates clusters with a labial C1 from clusters in which the first member has a coronal or dorsal place of articulation. Sequences of labial consonant plus rhotic (cf. node 3) are the primary target for segmental changes. The etymological rhotic surfaces as a bilabial glide [w] in Trinidad FC.Within this subset of the data, alternative repairs are rare. What is even more interesting, however, is that none of the pertinent clusters surfaces intact in the creole. This situation is quite different from the one we find in node 4 for clusters with coronal and dorsal C1 s. Here, the majority (72.22%) of clusters remain intact. Those repairs which occur are divided between segmental changes and consonant deletions with no clear preference for either strategy. A closer look at the data reveals that all of these changes apply in similar environments.

6.5 Results II: Trinidad French Creole |

207

Without exception, they occur in lexifier words in which the rhotic is followed by a labial segment, either a bilabial glide [w] or a rounded vowel.⁹ Again, repair is categorical for the pertinent structures. We can thus generalise that clusters with a rhotic C2 undergo restructuring if either the preceding or the following segment has a labial place of articulation. The table in (19) provides examples of all relevant subsets of onset clusters as discussed above. Retention of non-rhotic clusters is illustrated in (19a). (19b) shows repair options for rhotic clusters with a labial segment either preceding or following the rhotic, and (19c) documents the retention of rhotic clusters in nonlabial contexts. (19)

Treatment of complex lexifier onsets in Trinidad FC – regrouped Root Etymon Gloss a. C2 = glide or lateral plime < plume ‘pen’ [zj]ex < [zjø] les yeux ‘eyes’ b. C2 = rhotic, labial context [bw]ef < bref ‘brief’ [tw]op < trop ‘too much’ fomaie < fromage ‘cheese’ c. C2 = rhotic, non-labial context craine < craindre ‘to fear’ trîbe < triple ‘triple’

Overall, we can conclude that most complex onsets are retained without modification in Trinidad FC. The only sequences which are categorically repaired are those of a rhotic and a preceding or following labial segment. The affected onset clusters either undergo segmental modification and surface as C-[w] onsets or lose their second member altogether.

6.5.1.2 Vowel-initial words in Trinidad FC The data set on vowel-initial words consists of 84 root-etymon pairs. As we saw earlier, onsetless structures are not ruled out completely in Trinidad FC, but they are not unrestricted, either. The table in (20) provides a complete list of all restructuring processes that apply to vowel-initial etyma, ordered from most to least frequent.

9 Since vowel quality was not coded in the data (cf. section 3.6), this generalisation could not emerge in the CHAID analysis.

208 | 6 Syllable structure and phonotactic restructuring in the French-based creoles (20)

Processes affecting vowel-initial words in TFC Process

N

%

none syllable deletion consonant epenthesis

64 19 1

76.2 22.6 1.2

Σ

84

100.0

Among those vowel-initial structures which are not retained in the creole, we find little variation as to the type of attested repair. The default option is to delete the initial, onsetless syllable (simply a vowel or, more rarely, a vocalic nucleus followed by a coda consonant). The result is either a simple or a complex onset, depending on the structure of the second syllable in the etymon. Only a single word shows an alternative repair. Here, an epenthetic consonant is inserted wordinitially, thus creating a non-etymological onset. All attested treatment options are illustrated by examples in (21a-c), including the single case of consonant epenthesis for the sake of completeness. (21)

Treatment of vowel-initial etyma in Trinidad FC Root Etymon Gloss a. aimèn < aimer ‘to love’ estomac < estomac ‘stomach’ obèn < ou bien ‘or’ b. craser < écraser ‘to smash’ sitot < aussitôt ‘as soon as, soon’ vêtî < avertir ‘to warn’ c. [j]eaux < eux ‘they, people, folk’

In a CHAID analysis with the number of syllables and the position of main stress in the etymon as predictor variables and restructuring process as the dependent variable, the number of syllables emerged as the best predictor. The tree diagram in figure 6.14 illustrates the resulting split into subsets. In node 1, we see that short words of one or two syllables categorically preserve the onsetless structure. Longer words of three or four syllables, on the other hand, show variation between retention and deletion of the initial syllable (cf. node 2). This distinction raises the question of whether the initial onsetless syllable is indeed the trigger of restructuring here or whether the overall prosodic structure of the word does not play at least an equally important role. In their discussion of Haitian Creole, Steele and Brousseau (2006) ascribe a similar process of initial syllable deletion in longer words to a preference for words to consist of max-

6.5 Results II: Trinidad French Creole |

209

Restructuring of vowel-initial words Node 0 Process % N none 76.19 64 syllable deletion 22.62 19 C-epenthesis 1.19 1 Total (100.00) 84 Number of syllables in etymon Adj. P-value=0.0001, Chi-square=22.9304, df=2

1-2 Node 1 Process % N none 95.12 39 syllable deletion 2.44 1 C-epenthesis 2.44 1 Total (48.81) 41

3 or more Node 2 Process % none 58.14 syllable deletion 41.86 C-epenthesis 0.00 Total (51.19)

N 25 18 0 43

Fig. 6.14: Treatment of vowel-initial lexifier words in Trinidad FC

imally a two-syllable foot. If we assume such as preference to be at work also in Trinidad FC, then it should apply not only to vowel-initial, but also to consonantinitial words with three or more syllables. An inspection of the pertinent data revealed that such examples do indeed exist (e.g. demander > mander), but they are extremely rare, making up less than 5% of all pertinent etyma. There might be additional factors at work, of course, which prevent initial syllable deletion in most consonant-initial items. A further investigation of this issue is beyond the scope of the present study. Note, however, that the prime candidate for influencing syllable deletion patterns, stress, does not play a role, here. The French input shows hardly any variation in stress-placement. With very few exceptions, stress falls on the final syllable. The only impact of stress placement that can be noted is the fact that none of the few stressed first syllables are ever deleted, regardless of onset structures. Given the above findings, I will assume that the restructuring patterns we find for words with onsetless first syllables result from the interaction of constraints on both syllable structure and the overall prosodic structure of words. The different treatment choices for mono- and disyllabic words on the one hand and words of three or more syllables on the other hand are illustrated by the rearranged list of examples in (22).

210 | 6 Syllable structure and phonotactic restructuring in the French-based creoles (22)

Treatment of vowel-initial etyma in Trinidad FC – regrouped Root Etymon Gloss a. mono- or disyllabic etymon aimèn < aimer ‘to love’ ente < entre ‘between’ obèn < ou bien ‘or’ b. etymon with three or more syllables angacer < agacer ‘to provoke, tease’ estomac < estomac ‘stomach’ craser < écraser ‘to smash’ sitot < aussitôt ‘as soon as, soon’

To sum up, Trinidad FC readily tolerates onsetless initial structures in mono- and disyllabic words. In longer words, the initial (unstressed) syllable may be variably deleted.

6.5.2 Word-final consonants in Trinidad FC This section discusses syllabic structures and repair mechanisms that occur in word-final position. The bar plots in figure 6.15 provide an overview of the attested final structures in Trinidad FC roots and their assumed etyma. Here, as was the case in the corresponding section on Guiana FC, we distinguish between final open syllables, final singleton coda consonants, and sequences of two or more consonants, some of which may not be analysed as complex codas in the lexifier language (for a discussion of this issue see section 6.4.2 above). In both roots and etyma, open syllables are the most common word-final structure, occurring in more than half of the items. The second largest group consists of words ending in a single coda consonant (just under 40% in either set). The main difference between root and etymon structures lies in the categorical ban on consonant clusters in the creole. None of the etymological consonant sequences surfaces in Trinidad FC. What the creole loses in clusters, it gains in open syllables (61% in the creole as compared to 52.6% in the etyma). The resulting difference between the two distributions is statistically highly significant (Fisher’s exact test, p [maʃan]). The remaining six etyma are all derived from Spanish and involve a final unstressed vowel. This vowel is dropped in Trinidad FC, causing the former onset consonant to take a coda position. This change, evidently not triggered by syllable structure constraints, may be motivated by word-level prosodic preferences instead. The stress pattern of the affected Spanish words differs from the stress pattern that the vast

400

212 | 6 Syllable structure and phonotactic restructuring in the French-based creoles

creole structure

200

98.2%

77.3%

100

frequency

300

no coda simple coda cluster

0

0%

no coda

simple coda

cluster

final structure in etyma

Fig. 6.16: Correspondences between word-final structures in Trinidad FC roots and etyma (N=720)

majority of French words show. In French etyma, stress almost invariably falls on the final syllable. The Spanish words undergoing final vowel deletion, on the other hand, have non-final stress. The triggering factor thus might be a dispreference for unstressed final syllables in general.¹⁰ Some examples illustrating this loss of final unstressed vowels are given in (23). (23)

Loss of final unstressed vowels in Trinidad FC Root Etymon Gloss consuè[l] < consuelo ‘comfort, solace’ golè[t] < goleta ‘schooner’ so[ɡ] < soga ‘rope’

Donor Spanish Spanish Spanish

Having addressed the more exceptional changes, we now turn to the discussion of those parts of the data which show more substantial restructuring. The following two sections deal with simple word-final codas and consonant clusters, respectively.

10 Of the few etyma with non-final stress, only two survive in the creole.

6.5 Results II: Trinidad French Creole |

213

6.5.2.1 Simple word-final codas in Trinidad FC Overall, 286 etyma contain simple word-final codas. These etyma and their creole reflexes form the data base for the analyses in this section. The table in (24) gives an overview of the types and frequencies of repair mechanisms that affect simple codas. (24)

Processes affecting simple codas in TFC Process

N

%

none consonant deletion lenition other

205 65 13 3

71.7 22.7 4.5 1.0

Σ

286

100.0

In the vast majority of cases (over 70%), the coda consonant is simply maintained in Trinidad FC. Among the attested restructuring strategies, consonant deletion is the preferred option. Alternative repairs are relatively rare. A small percentage of final consonants undergoes lenition, thus preserving the coda, but changing its segmental content. Finally, three items show yet a different type of restructuring. Here, the label ‘other’ has been applied to cases in which the etymon ends in a voiced plosive, whereas the creole reflex ends in a nasal consonant with the same place of articulation.¹¹ in the creole.The plosive in the etymon is always preceded by a nasalised vowel. Two different, equally plausible interpretations of the observed changes suggest themselves. We could assume that the final plosive is deleted and that the nasal consonant that we find in the etymon is a reflex of the nasalisation of the etymological vowel. Alternatively, the change from etymon to creole word could be interpreted as a change in manner of articulation that applies to the final plosive in the context of the preceding nasalised vowel. For reasons of simplicity, I will refer to changes of this type as nasalisation of the plosive in the following. It should be born in mind, however, that this is not the only possible

11 Although all three pertinent cases among the simple codas involve the alveolar plosive [d], we will see later on in the discussion of word-final clusters that the process is not restricted to this place of articulation. Evidence to that effect is also given by Stein (1984: 30) for a number of other French-based creoles.

214 | 6 Syllable structure and phonotactic restructuring in the French-based creoles interpretation.¹² Examples of this and all other processes, including coda preservation, are given in (25a-d). The order of categories is the same as in the frequency table above. (25)

Treatment of simple codas in Trinidad FC Root Etymon Gloss a. la[k] < la[k] ‘lake’ simai[n] < semai[n] ‘week’ b. [fizi] < [fizil] ‘gun’ tantamâ < tintama[ʁ] ‘great fuss’ c. ca[j] < ca[z] ‘house’ foma[j] < froma[ʒ] ‘cheese’ d. mou[n] < m[˜ɔd] ‘people, they, one’ via[n] < vi[˜ ɑd] ‘flesh’

I conducted a CHAID analysis in order to find out whether the choice of treatment can be predicted on the basis of structural factors. The manner of articulation of the coda consonant emerges as the best predictor, according to which the data can be divided into four subsets. The tree diagram in figure 6.17 illustrates this split. The groups are ordered according to the respective rate at which coda consonants are preserved without modification (option ‘none’), ranging from categorical for nasal and plosive consonants (cf. node 1) to strongly disfavoured for final rhotics (cf. node 4). The latter type of coda is preferably deleted in Trinidad FC (85.1%). Moving to the intermediate groups, we find that lateral approximants have an overall higher probability of surfacing in the creole than fricatives. However, the preferred repairs differ for the two sets. Whereas laterals, if modified at all, are lost completely (cf. node 2), the typical repair for fricatives is lenition rather than deletion (cf. node 3). For the subset of fricatives, further distinctions emerged in the CHAID analysis. They are illustrated in the tree diagram in figure 6.18. As the diagram shows, voicing and place features play a role in the choice of coda treatment, too. Voiceless fricatives (cf. node 5) are in fact treated no differently than nasals or plosives. Preservation can be considered categorical for this group. Among the voiced fricatives, we find considerably more variation. The final split reveals that

12 Note that nasalisation of final voiced obstruents is also widespread in Quebec French (Brousseau, p.c.). We cannot therefore exclude the possibility that it already occurred in the French spoken by 17th and 18th century settlers. However, lacking concrete historical evidence (the process is not mentioned in any of the historical records), I assumed a final plosive in the pertinent etyma.

6.5 Results II: Trinidad French Creole |

215

Restructuring of simple codas Node 0 Process % lenition 4.55 none 71.68 C-deletion 22.73 other 1.05 Total (100.00)

N 13 205 65 3 286

Manner of coda consonant Adj. P-value=0.0000, Chi-square=236.2785, df=9

nasal; plosive Node 1 Process % N lenition 0.00 0 none 97.32 109 C-deletion 0.00 0 other 2.68 3 Total (39.16) 112

lateral Node 2 Process % N lenition 0.00 0 none 85.71 36 C-deletion 14.29 6 other 0.00 0 Total (14.69) 42

fricative Node 3 Process % N lenition 20.00 13 none 76.92 50 C-deletion 3.08 2 other 0.00 0 Total (22.73) 65

rhotic Node 4 Process % N lenition 0.00 0 none 14.93 10 C-deletion 85.07 57 other 0.00 0 Total (23.43) 67

Fig. 6.17: Treatment of simple word-final codas in Trinidad FC - first split

this variation occurs mostly with voiced coronal fricatives (cf. node 7). Almost 40% of such coda consonants undergo lenition. Voiced labial fricatives, to the contrary, are preferably retained in the creole (cf. node 8). The modified list of examples in (26) illustrates the treatment options for the different categories of coda consonants that emerged in the analysis. Each category in (26) corresponds to a terminal node in the tree diagram above. Thus, (26a) covers simple nasal and plosive codas, (26b) illustrates the treatment of lateral approximants, and (26c) shows the behaviour of word-final rhotics. Analogously, the examples in (26d-f) are representative for the subsets singled out in nodes 5, 7, and 8 in the tree diagram, respectively.

216 | 6 Syllable structure and phonotactic restructuring in the French-based creoles Restructuring of simple fricative codas Node 3 Process % n lenition 20,00 13 none 76,92 50 C-deletion 3,08 2 Total (100,00) 65 Voicing of coda consonant Adj. P-value=0.0005, Chi-square=15.4042, df=2

voiced

voiceless Node 5 Process % lenition 0,00 none 96,15 C-deletion 3,85 Total (40,00)

Node 6 Process % lenition 33,33 none 64,10 C-deletion 2,56 Total (60,00)

n 0 25 1 26

n 13 25 1 39

Place of coda consonant Adj. P-value=0.0145, Chi-square=8.4671, df=2

coronal Node 7 Process % lenition 39,39 none 60,61 C-deletion 0,00 Total (50,77)

labial

n 13 20 0 33

Fig. 6.18: Treatment of word-final fricatives in Trinidad FC

Node 8 Process % n lenition 0,00 0 none 83,33 5 C-deletion 16,67 1 Total (9,23) 6

6.5 Results II: Trinidad French Creole |

(26)

217

Treatment of simple codas in Trinidad FC – regrouped Root Etymon Gloss a. nasal or plosive la[k] < la[k] ‘lake’ simai[n] < semai[n] ‘week’ b. lateral approximant vi[l] < vi[l] ‘town’ [fi] < [fiʎ] ‘girl, daughter’ c. rhotic ritoû < retou[ʁ] ‘return’ vè[ʁ] < ve[ʁ] ‘glass’ d. voiceless fricative boue[f] < bre[f] ‘brief’ grai[s] < grai[s] ‘fat’ e. voiced coronal fricative chimi[z] < chemi[z] ‘shirt’ foma[j] < froma[ʒ] ‘cheese’ f. voiced labial fricative boua[v] < bra[v] ‘brave’ sa < sa[v] ‘to be able, to know’

In summary, although many simple codas surface intact in Trinidad FC, two main targets of restructuring can be identified. Firstly, rhotics show a strong tendency towards deletion, and, secondly, voiced coronal fricatives (i.e. etymological [z] and [ʒ]) may undergo lenition and surface as glides in the creole.

6.5.2.2 Word-final consonant clusters in Trinidad FC Overall, 55 etyma in the data feature word-final clusters. One of them had to be excluded from the investigation, because the pronunciation of its creole reflex was unclear.¹³ The remaining 54 etyma and their creole reflexes form the basis for the analysis of final clusters. The etymological clusters are either complex codas or

13 The item in point is the English word warrant. Thomas (1869) uses the same spelling for the corresponding creole word. If the etymon were French, etymological spelling could be interpreted as reflecting regular lexifier pronunciation. Since the etymon is English, however, the situation is more complicated. Thomas might have had in mind a French-based correspondence of spelling and pronunciation rather than an etymological English one. Etymological pronunciation is in fact not very likely as it would make warrant the only creole word ending in a consonant cluster. In any case, in the absence of reliable information on pronunciation, the item was excluded.

218 | 6 Syllable structure and phonotactic restructuring in the French-based creoles so-called offsets (for a discussion see section 6.4.2 above). Whatever their status in the lexifier, none of the sequences surfaces intact in Trinidad FC. As the table in (27) shows, consonant deletion applies across the board to repair the etymological structure. (27)

Processes affecting final clusters in TFC Process

N

%

consonant deletion consonant deletion + other

43 11

79.6 20.4

Σ

54

100.0

Almost 80% of the etyma lose one of the cluster members and retain the other as a simple coda. In the remaining 20.4%, the consonant that remains after simplification also undergoes a qualitative change, turning into a nasal.¹⁴ Both options are illustrated by examples in (28). (28)

Treatment of final clusters in Trinidad FC Root Etymon Gloss a. dire[k] < dire[kt] ‘direct’ lafô[s] < la fo[ʁs] ‘strength’ pouê[t] < prê[tʁ] ‘priest’ b. châ[m] < ch[˜ ɑbʁ] ‘chamber, room’ repô[n] < rep[˜ɔdʁ] ‘to answer’

As the examples in (28a) show, there is some variation as to which cluster member is deleted. In the majority of the cases, it is the second, edgemost consonant, that is lost, but the alternative is not exceptional. For a more detailed analysis, the data were therefore recoded to distinguish between the two deletion options (labelled ‘C1 -deletion’ and ‘C2 -deletion’ in the following). The results of a CHAID analysis, illustrated by the tree diagram in figure 6.19, reveal that the data can be neatly divided into two subsets with contrasting treatment preferences. The manner of articulation of the first cluster member emerges as the best predictor. Node 1 singles out clusters with a rhotic as first member. In all of the pertinent clusters, the rhotic is lost, whereas the second consonant, an obstruent, is main-

14 We observed the same process of nasalisation, albeit without previous consonant deletion, in a small number of etyma with simple codas.

6.5 Results II: Trinidad French Creole |

219

Restructuring of word-final clusters Node 0 Process % C2-deletion 61.11 C2-deletion + other 20.37 C1-deletion 18.52 Total (100.00)

N 33 11 10 54

Manner of first cluster member Adj. P-value=0.0000, Chi-square=51.7499, df=2

rhotic

obstruent

Node 1 Process % N C2-deletion 0.00 0 C2-deletion + other 0.00 0 C1-deletion 100.00 10 Total (18.52) 10

Node 2 Process C2-deletion C2-deletion + other C1-deletion Total

% 75.00 25.00 0.00 (81.48)

N 33 11 0 44

Voicing of first cluster member Adj. P-value=0.0000, Chi-square=18.9870, df=1

voiceless Node 3 Process % N C2-deletion 100.00 22 C2-deletion + other 0.00 0 C1-deletion 0.00 0 Total (40.74) 22

voiced Node 4 Process % C2-deletion 50.00 C2-deletion + other 50.00 C1-deletion 0.00 Total (40.74)

N 11 11 0 22

Fig. 6.19: Treatment of word-final clusters in Trinidad FC

tained in coda position.¹⁵ In contrast, clusters with an obstruent in first position (cf. node 2), typically lose the second consonant in the sequence. This latter group comprises the so-called offsets as well as a number of obstruent-clusters. In 75% of the pertinent cases, the result in the creole is a simple obstruent coda (former C1 ). Only in the cases with double repair does the creole word feature a nasal coda instead. The second split in the tree diagram shows that the likeliness of this additional change can be predicted from the voicing specification of the first cluster member. With voiceless C1 s (cf. node 3), any repair beyond the default C2 -deletion can be considered exceptional. For clusters that start in a voiced obstruent, how-

15 Clusters with a lateral in first position did not occur in the data, so that it remains unclear whether C1 -deletion is indeed rhotic-specific or can be generalised to all liquid-clusters in wordfinal position.

220 | 6 Syllable structure and phonotactic restructuring in the French-based creoles ever, we find more variable behaviour (cf. node 4). In addition to C2 -deletion, over two thirds of the former C1 s in these sequences undergo the change to a nasal in Trinidad FC. More specifically, voiced plosives that are preceded by a nasalised vowel may lose their oral character and surface as a nasal consonant, maintaining the original place of articulation.¹⁶ The attested changes include the changes [d]>[n] and [b]>[m]. To round off the discussion, the list in (29) provides examples illustrating the treatment of all cluster types that emerged as significantly different in the CHAID analysis. The order of categories follows the numbering of terminal nodes in the tree diagram. (29)

Treatment of final clusters in Trinidad FC – regrouped Root Etymon Gloss a. C1 = rhotic lafô[s] < la fo[ʁs] ‘strength’ zê[b] < des he[ʁb] ‘grass’ b. C1 = voiceless obstruent dire[k] < dire[kt] ‘direct’ pouê[t] < prê[tʁ] ‘priest’ c. C1 = voiced obstruent cha[m] < ch[˜ ɑbʁ] ‘chamber, room’ repô[n] < rep[˜ɔdʁ] ‘to answer’ sa[b] < sa[bl] ‘sand’

The findings which we can formulate on the basis of the preceding analysis, then, are the following. Trinidad FC categorically repairs word-final consonant sequences by deleting either the first or the second member of the cluster. If rhotics occur as C1 , the edgemost consonant survives, whereas the reverse is the case in clusters with an obstruent C1 . Voiced plosives as first cluster members show a somewhat special behaviour in that they may surface as nasal consonants in the creole if immediately preceded by a nasal vowel in the etymon.

6.5.3 Word-internal structures in Trinidad FC For the discussion of word-internal structures, the data were reduced to those cases in which the etymon contained at least two vocalic nuclei. This reduction resulted in a data base of 481 root-etymon pairs with medial structures overall.

16 Note that, since the data were not coded for vowel quality, this finding could not emerge in the analysis.

6.5 Results II: Trinidad French Creole |

221

The bar plots in figure 6.20 provide an overview of the frequencies of different word-internal structures in the subsets of creole and lexifier words, respectively. Words with strict alternation of vowels and consonants (labelled ‘only VCV’) are distinguished from words which contain consonantal sequences (‘CC(C)’) or cases of hiatus (‘VV’).

300 200

frequency

73%

100

200

83.8%

100

frequency

300

400

Trinidad FC etyma

400

Trinidad FC roots

26.2% 15.6% only VCV

CC(C)

VV

0.8%

0

0

0.6%

only VCV CC(C)

VV

Fig. 6.20: Word-internal structures in Trinidad FC roots and etyma (N=481)

The distribution of word-internal structures in roots and etyma is similar in the sense that strictly alternating structures make up the clear majority of all cases, whereas hiatus is extremely rare both in the lexifier and in the creole. In terms of absolute frequencies, however, the two distributions differ significantly (Fisher’s exact test, p [ja]). Complex onsets with a liquid or glide as second element are allowed.

5 Note that Twi is treated as representative of Akan languages in general here. Most statements are also true for Fante. I will differentiate between the two only where relevant differences are mentioned in Christaller’s (1875) account. 6 The language only has one liquid phoneme. Christaller (1875) gives [r] realisations for Twi, but mentions that [l] occurs in some Fante dialects.

264 | 8 Explaining creole phonotactic restructuring Gbe allows only open syllables. The final vowel may be either oral or nasal. Vowelinitial words are permitted as are certain types of onset clusters. The most common cluster type involves a liquid in second position, but also sequences in which C2 is a glide [j] are attested. According to Capo (1991), C-[w] clusters are common only in Fongbe, but rare or non-existent in other lects. Also in Fon, they are restricted to word-internal position (cf. also Brousseau and Nikiema 2006), with the glide [w] always deriving from the initial back rounded vowel of an underlying VV sequence. Heterosyllabic vocalic sequences, i.e. cases of hiatus, are possible in Gbe, but may be resolved by vowel deletion or glide formation (cf. Schlegel 1856, Westermann 1930). In Ijo, word-final syllables mostly end in a vowel (either oral or nasal). According to Harry (2004), Kalabari-Ijo also allows the nasal [m] in word-final position. Word-internally, also other nasal consonants can occur in coda position, provided that a homorganic plosive occupies the following onset. Syllables lacking an onset are permitted word-initially as well as word-internally (hiatus). Hiatus may be resolved in certain contexts. With respect to onset clusters, both sources I consulted agree that such structures do not exist at the underlying level in Ijo. Surface clusters are attested, however. Williamson (1965) reports some clusters that result from vowel deletion in sequences of a voiceless plosive followed by a high vowel and a rhotic (e.g. piri ~ pri, Williamson cf. 1965: 20). According to Williamson, the frequency of such clusters varies across different dialects of Ijo. Harry (2004) also mentions surface sequences of an obstruent followed by a glide, where the glide is a high vowel underlyingly. Kikongo is said to permit only open syllables (e.g. Bentley 1887, Laman 1912). In non-final position, we find sequences of a nasal followed by another consonant, but the status of these sequences is unclear. Several of the early grammars treat these sequences as single segments (cf. Laman 1912, Laman and Meinhof 1928). Seidel and Struyf (1910), on the other hand, report that, when such sequences occur in word-initial position, the nasal becomes syllabic, a change that would be highly unexpected if the NC sequences really were complex segments. In line with this, Uffmann (2008) notes that more recently suggestions have been made to analyse the sequences in question as coda-onset clusters rather than prenasalised stops. Under such an approach, Kikongo would have nasal codas word-internally, but only open syllables at the end of words. As far as onset structures are concerned, the only consonant clusters that are permitted in onset position have a glide as their second element. Underlyingly, this glide may be a vowel. Onsetless syllables can occur word-initially, but are often avoided word-internally. Potential hiatus situations are resolved by processes of vowel deletion or coalescence. Sequences of two vowels may also be repaired when

8.4 Testing SLA approaches to creolisation

| 265

occurring across word boundaries, making also word-initial vowels a target for variable restructuring (cf. e.g. Seidel and Struyf 1910: 11f). The description of syllable structure in Mande languages varies slightly across the two sources I consulted. Steinthal (1867) states that syllables are either open or end in a nasal consonant. According to him, only [ŋ] is possible in word-final coda position, whereas word-internally also [n] and [m] occur as coda consonants. Delafosse (1929) mentions also [m,n] in word-final position, but only in words that have lost their final vowel. The two sources converge on their accounts of onsetless syllables. Vowel-initial words are permissible, although both Steinthal (1867) and Delafosse (1929) describe the structure as infrequent. Hiatus is possible, but may be resolved by processes of vowel deletion, coalescence, or glide formation (cf. e.g. Steinthal 1867: 33). Where onset clusters are concerned, both sources recognise clusters of the type obstruent-liquid which result from deletion of a formerly intervening vowel (CVLV > CLV). Delafosse (1929) also mentions a glide [j] as possible C2 in such contractions. Non-contracted clusters involving a glide in second position or C-[w] surface clusters are not mentioned in either of the sources. I will therefore assume in the following that only C-liquid and C-[j] sequences are possible in Mande.

8.4 Testing SLA approaches to creolisation This section sets out to investigate whether the predictions of the two creolisation hypotheses introduced earlier are supported by the data. As discussed in section 8.2 above, Plag (2009) and Uffmann (2009) predict only a subset of all theoretically possible constellations of syllable structure restrictions in superstrate, substrate and creole. In this section it will be tested how well these predictions correspond to the patterns we do in fact find in the data. For this purpose, I will compare syllable structure restrictions in the creoles and their respective contributing languages and establish which creole outcomes are attested for different constellations of restrictions in superstrate and substrate languages. I will use the categories introduced in section 8.2 to distinguish between different combinations of creole behaviour and restrictions in the contributing languages. Some methodological issues need to be addressed at this point. The first one concerns the set of structural categories to be used in the comparison. I will employ the same structural categories here that emerged as undergoing significantly different treatment in the analyses in chapters 4-6. If a creole imposes tighter restrictions on one category than on another or uses different repairs for different structural categories, then this is something that a model of creolisation needs to address. However, as pointed out earlier in section 8.2, the choice of

266 | 8 Explaining creole phonotactic restructuring repair strategy is not immediately relevant to the testing of the two creolisation hypotheses. In a first step, I will therefore concentrate here on a general evaluation of how well interlanguage approaches to creolisation can account for the restrictions we find in the creoles. Whether the concrete restructuring mechanisms attested in the data support or contradict this general evaluation will be discussed only at a later point. In view of this later, process-specific discussion, structural categories which show comparable retention rates but differ in terms of repair choices will be maintained as separate categories here. The second methodological problem concerns the permissibility of different types of syllable structures in the superstrate and substrates. For the creole, we have decided to use a three-way distinction between systematic retention, systematic repair, and more variable treatment. For the superstrate and substrate, however, the same distinctions are not always applicable or even helpful. Especially the available information on substrate syllables as summarised in section 8.3 above does not usually extend beyond a distinction of what is attested or unattested in the respective language. In addition, the authors sometimes mention the variability of some structure, but without any quantifications it is impossible to translate such remarks into the permissibility categories employed for the creole. Similar problems arise for variable superstrate structures. Variability in the input languages poses problems also in a different respect. Neither Plag (2009) nor Uffmann (2009) discuss variable structures in superstrate or substrate and consequently no predictions are available for such cases. The only distinction which is recognised in the hypotheses – and which can therefore also be used in their evaluation – is between structures which are allowed in the superstrate or substrate and structures which are not allowed. I have applied this distinction to superstrate and substrate structures as follows. For the superstrates, I counted as permissible all structures which occur in the assumed etyma from that language. Such an approach seemed acceptable since variation in the superstrates was already taken into account in the coding of the etyma, rendering it unlikely that highly exceptional realisations were included.⁷ As far as non-superstrate structures are concerned, since such struc-

7 Of course, a high degree of variability in the lexifier could also have led to variable treatment of a given structure in the creole that goes beyond a mere copying of input forms. Given sufficient contact to the superstrate, creole speakers might have acquired such a variable superstrate process. Where such superstrate variation exists, it might thus be that variable creole treatment should be interpreted as successful acquisition rather than just partial acquisition, as it would usually be classified (provided the structure in question is not permitted in the substrates). Since no great differences result for the testing of the two creolisation hypotheses, such details will be ignored in the comparisons below.

8.4 Testing SLA approaches to creolisation

| 267

tures could occur only in lexical items derived from a language other than the main lexifier, categories which were classified as impermissible in the superstrate are rare. Therefore, in the comparison of syllable structure following below, no separate column is included in the comparative tables for acceptability in the superstrate. Instead, all structural categories listed should be understood as permissible in the superstrate. Exceptions from this rule are marked as such. For the substrates, a different approach was necessary. Here, I interpreted as permissible all structures which are attested and not described as exceptional in the sources. Structures were counted as impermissible, on the other hand, if they are either completely unattested or described as highly exceptional. For a small number of structures, simplifications of unclear frequency were mentioned in the sources, rendering it impossible to decide whether or not they should be regarded as exceptional. This applied in particular to instances of hiatus in various substrate languages. Since the substrate category of attested structures cannot be directly equated to either of the permissibility categories employed for the creole, separate labels will be used to indicate acceptability in the substrates and in the creoles. The table in (5) gives the labels which will be used for attested and unattested structures in the substrates. The additional label ‘var.’ will be used for those structural categories which were described as undergoing frequent simplifications and whose acceptability could therefore not be determined. (5)

Permissibility categories for substrate structures phonological category in substrate?

label

attested unattested or highly exceptional indeterminable degree of variability

yes – var.

The table in (6) shows the labels employed for the acceptability of structures in the creoles. Apart from the creole outcome, also the corresponding phonological interpretation is given.

268 | 8 Explaining creole phonotactic restructuring (6)

Permissibility categories for creole structures creole outcome

phonological category in creole?

label

no variable yes

– + ++

systematic restructuring variable restructuring no restructuring

In the following sections, I will provide a table for each of the investigated creoles, comparing the acceptability of individual structural categories in the creole itself with that in its contributing languages. The format of these comparative tables is illustrated in (7). (7)

Format of comparative tables

structural category 1 structural category 2 structural category 3 structural category 4a structural category 5 a

creole

substrate 1

substrate 2

pattern

– ++ (+) – +

– yes/– – yes var.

– yes/– – yes var.

1a 2c/1c 1b 3a n.a.

This structure is not permitted in the superstrate.

Unless indicated otherwise, the leftmost column lists all structural categories which emerged from the analysis of the respective creole in chapters 4 through 6. The middle part of the table shows the permissibility labels assigned to the individual structural categories in the creole and each of its main substrate languages. If a given structure is unacceptable in the superstrate, this is indicated in a table note (cf. structural category 4). If no such note is included, the category was treated as permissible in the superstrate. Finally, in the rightmost column of the table, each structural category is assigned a pattern label corresponding to one of the nine theoretically possible combinations of restrictions in superstrate, substrate and creole as discussed in section 8.2 above. In addition to the pattern labels (1a)-(3c), the label ‘n.a.’ (read: ‘not applicable’) is used in cases in which substrate acceptability of the structure could not be determined (cf. structural category 5). Two further modifications to the permissibility labels need to be mentioned here. First, if a creole label occurs in parentheses (cf. structural category 3), this indicates that there are only few attestations for the pertinent structural category. Second, some of the cells for acceptability in the substrates contain the label

8.4 Testing SLA approaches to creolisation

| 269

‘yes/–’, i.e. a combination of the labels for attested and unattested structures (cf. structural category 2). In these cases, the substrate in question shows different permissibility levels for different sub-classes of the structural category.⁸ If comparable sub-distinctions hold across substrates, separate pattern labels were assigned for these sub-classes. In the example above, the creole systematically retains structures belonging to structural category 2. The substrates, on the other hand, allow only a subset of the pertinent structures and disallow the remaining ones. The pattern labels which reflect these restriction constellations are (2c), for creole retention of a (sub)set of structures which is possible in both superstrate and substrates, and (1c), for creole retention of a (sub)set of structures which is allowed in the superstrate but not in the substrates. The comparative tables thus serve to give an overview of the attested patterns of restrictions in superstrate, substrates and creole. In a second step, it will be shown for each creole which of the attested patterns are expected under the hypotheses by Plag (2009) and Uffmann (2009) and which ones are not. This will be discussed in detail only for the first language, Berbice Dutch. In the sections on the remaining five creoles, comments will be restricted to aspects that are particularly noteworthy. The six language-specific sections are then followed by a more general evaluation of the two proposals and a discussion of what the findings can tell us about mechanisms at work in creolisation.

8.4.1 Mechanisms in creolisation: Berbice Dutch The table in (8) compares the permissibility of all investigated structures in the creole, its superstrate and its main substrate language, Easter Ijo.

8 On the level of segmental content, this could, in principle, be argued in many cases. For instance, even if both superstrate and substrate allow clusters consisting of an obstruent and a liquid, the precise combinations of segments attested in each language might differ. However, in most cases, I concentrated on the permissibility of classes of segments or clusters rather than individual segments or segment combinations. Admittedly, this involves a certain degree of simplification. Generalisations were avoided, though, in cases in which they might have been misleading. For instance, if a substrate allowed only rhotics, but no laterals in a given position, then the category of liquid received the split label ‘yes/–’.

270 | 8 Explaining creole phonotactic restructuring (8)

Syllable structure in Berbice Dutch and its contributing languages BD

Ijo

pattern

word-initial position: V, etymon = σ(σ) V, etymon = σσσ(σ) obstruent-approximant sibilant-nasal plosive-nasal [s]-obstruent

++ + ++ (++) (–) ++

yes yes yes/– – – –

2c 2b 2c/1c 1c 1a 1c

word-final position: [m] [n] [ŋ] obstruent liquid, stressed liquid, unstressed nasal-obstruent other C-obstruent liquid-nasal

++ + – – – – – – –

yes – – – – – – – –

2c 1b 1a 1a 1a 1a 1a 1a 1a

word-internal position: V.V obstruent-liquid nasal-obstruent liquid-C, not homorganic liquid-C, homorganic obstruent-obstruent

– ++ ++ + + +

var. yes/– yes – – –

n.a. 2c/1c 2c 1b 1b 1b

All of the structural categories listed are allowed in the superstrate. For the most part, the categories correspond to those that emerged from the analysis in chapter 4.4. One exception was made with respect to word-final nasals. No significant difference was found between the creole treatment of [m] and [n]. Ijo, however, makes a distinction. Also, in terms of the permissibility categories used here for the creole, retention of [m] reaches systematic level, whereas [n] falls into the category of variable restructuring. The originally combined category for the two nasals was therefore split in the table. Note also that two structural categories have double classifications for the substrate. These classifications are based on the fact that we find obstruent-glide and some obstruent-rhotic sequences in Ijo, but no obstruent-lateral clusters. A single structural category, that of hiatus, could not be assigned a pattern label due to an uncertain degree of variability in the substrate.

8.4 Testing SLA approaches to creolisation

| 271

The table in (9) summarises the frequency of attested constellations as identified in (8) above. Frequency is used here in the sense of the number of structural categories for which a given constellation of restrictions in superstrate, substrate and creole is attested. In addition to the pattern labels, also their respective interpretation is included again in (9) for convenience. (9)

Constellations of restrictions in the contributing languages and restructuring in Berbice Dutch superstrate

substrate

creole restructuring

no. of categories

type 1a type 1b type 1c

no restr. no restr. no restr.

restr. restr. restr.

systematic variable none

8 4 4

type 2a type 2b type 2c

no restr. no restr. no restr.

no restr. no restr. no restr.

systematic variable none

– 1 5

type 3a type 3b type 3c

restr. restr. restr.

no restr. no restr. no restr.

systematic variable none

– – –

As is evident from the table, some constellations are more common than others. Not all differences are equally remarkable, however. Thus, the fact that type 3 constellations are entirely absent is not so surprising if we keep in mind that they involve restrictions on the part of the superstrate. The pertinent structures thus could only have occurred in substrate lexical items. More interesting for the testing of the hypotheses are the attested creole outcomes for type 1 and type 2 constellations. Type 1 constellations are characterised by the substrate imposing tighter constraints on some structure than the superstrate. All three theoretically possible creole outcomes, systematic restructuring, variable restructuring and systematic retention of the structure, are attested. Among these patterns, the first, most common one (1a) is compatible with the assumption of full transfer or, in Uffmann’s terms, substrate retention. Patterns (1b) and (1c), on the other hand, would be expected if (partial or complete) acquisition of superstrate structures had taken place. For type 2 constellations, the creole shows less variation in outcome. Structures which are restricted in neither superstrate nor substrate are never systematically restructured. In fact, they are usually systematically retained (pattern 2c), an

272 | 8 Explaining creole phonotactic restructuring outcome which is expected under the transfer hypothesis, but equally compatible with the assumption of successful superstrate acquisition. The only attested pattern that is not expected under any of the predictions by Plag (2009) and Uffmann (2009) is pattern (2b). For a single structural category which is permitted in both superstrate and substrate, we find variable restructuring in the creole. This is the case for vowel-initial words of three or more syllables. As pointed out earlier in chapter 4.4, creole restructuring in this particular case seems to be motivated by word structure rather than syllable structure constraints. Since word structure in superstrate and substrate was not investigated, it remains unclear whether this pattern can be explained in terms of the mechanisms assumed by Plag and Uffmann. The table in (10) summarises these results, dividing the attested patterns into those which are expected and those which are not expected under any hypothesis and providing a pooled count of the pertinent categories for each of the suggested interpretations.⁹ (10)

Expected and unexpected patterns in Berbice Dutch

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c

8 5 8

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition

2b

1

–

8.4.2 Mechanisms in creolisation: Negerhollands The comparison of syllable structure restrictions in Negerhollands and its contributing languages in (11) includes three different substrate languages. They are similar, but not identical in terms of syllable structure. Where the substrates do not agree with respect to the permissibility of some structure, I worked with the majority choice. There is no indication that any of the three substrate languages was highly predominant, so we can assume that whichever structures occurred in two of the main substrates would likely have been common enough to make

9 Note that in this and all following, similar tables the interpretation listed for pattern 2c is slightly abbreviated. ‘L1 transfer or L2 acquisition’ should be understood as meaning ‘L1 transfer or successful L2 acquisition’.

8.4 Testing SLA approaches to creolisation

| 273

it into the levelled grammar. Word-final nasals are a case in point. Gbe does not allow nasal consonants in word-final position, but Twi and Gã do. For the comparison with superstrate and creole, it was therefore assumed that word-final nasals are permitted in the levelled substrate grammar. Nasal-obstruent sequences in word-internal position present a similar situation. Twi allows sequences involving a voiceless obstruent, but voiced plosives do not occur following a nasal. This subdistinction was ignored in the comparison since the two remaining substrates do not impose such restrictions. A different methodological problem arose with the category of hiatus. In all three substrate languages, instances of hiatus are subject to variable restructuring. Since the frequency of this restructuring is impossible to judge on the basis of the available descriptions, no pattern label could be assigned. The category therefore had to be excluded from consideration in the evaluation of the creolisation hypotheses. Also the category of monosyllabic vowel-initial words requires a comment. Although onsetless initial syllables are possible in all three substrates, according to the sources I consulted, these structures always contain vocalic prefixes, which rules out monosyllabic words. In the comparison with superstrate and creole structures, the pertinent category was therefore counted as illicit in the substrates.

274 | 8 Explaining creole phonotactic restructuring (11)

Syllable structure in Negerhollands and its contributing languages NH

Twi

Gbe

Gã

pattern

word-initial position: V, etymon = σ V, etymon = σσ V, etymon = σσσ(σ) obstruent-approximant [s]-nasal [s]-obstruent plosive-nasal

++ ++ + ++ (+) ++ (–)

– yes yes yes – – –

– yes yes yes/– – – –

– yes yes yes – – –

1c 2c 2b 2c 1b 1c 1a

word-final position: nasal sibilant [s] plosive non-sibilant fricative lateral [l] rhotic liquid-C nasal-plosive nasal-fricative plosive-fricative fricative-obstruent

++ ++ ++ + ++ + + + ++ + (+)

yes – – – – – – – – – –

– – – – – – – – – – –

yes – – – – – – – – – –

2c 1c 1c 1b 1c 1b 1b 1b 1c 1b 1b

word-internal position: V.V obstruent-approximant nasal-obstruent obstruent-obstruent liquid-obstruent rhotic-approximant

(++) ++ ++ ++ + (–)

var. yes yes/– – – –

var. yes yes – – –

var. yes yes – – –

n.a. 2c 2c 1c 1b 1a

The table in (12) gives the category count for all constellations of restructuring in Negerhollands and structural restrictions in its contributing languages.

8.4 Testing SLA approaches to creolisation

(12)

| 275

Constellations of restrictions in the contributing languages and restructuring in Negerhollands superstrate

substrate

creole restructuring

no. of categories

type 1a type 1b type 1c

no restr. no restr. no restr.

restr. restr. restr.

systematic variable none

2 8 7

type 2a type 2b type 2c

no restr. no restr. no restr.

no restr. no restr. no restr.

systematic variable none

– 1 5

type 3a type 3b type 3c

restr. restr. restr.

no restr. no restr. no restr.

systematic variable none

– – –

Systematic restructuring is comparatively rare in Negerhollands. Most structures are either variably restructured or systematically retained. The table in (13) shows which patterns can be accounted for in an SLA approach. The single unexpected pattern (2b) occurs with the same structural category that also showed unexpected behaviour in Berbice Dutch, vowel-initial words with more than two syllables. Again, the motivation for (variable) restructuring does not appear to be a constraint on syllable structure as such, since vowel-initial structures are systematically retained in shorter words. (13)

Expected and unexpected patterns in Negerhollands

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c

2 5 15

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition

2b

1

–

8.4.3 Mechanisms in creolisation: Early Saramaccan The table in (14) illustrates the relation between phonotactic restructuring in Early Saramaccan and restrictions on syllable structure in superstrate and substrate languages. In cases where the substrates differed from each other with respect to

276 | 8 Explaining creole phonotactic restructuring the restrictions they impose on syllable structure, I followed the same rationale as before and considered only the majority choice. The list of structural categories given here for word-initial clusters differs slightly from the categories established in the analysis in chapter 5.4. The creole distinguishes only between sequences with a rising sonority profile and sequences with a falling sonority profile. The substrates, on the other hand, make rather different distinctions, and also the superstrate imposes restrictions on a small subset of word-initial consonantal sequences. The more general creole categories were therefore split to make the comparison more readily interpretable than it would have been with multiple classifications for both superstrate and substrate acceptability. Comments are in order also with respect to two classifications. First, the restrictions on subsets of word-initial obstruent-approximant clusters in Gbe and Kikongo were ignored in the classification. Gbe lacks word-initial C-[w] clusters and Kikongo allows no obstruent-liquid sequences. Each of those cluster types, however, is possible in the remaining two substrates. Second, the double classification for vowel-initial words with no more than two syllables is based on the fact that monosyllabic words never start in a vowel in the substrates.

8.4 Testing SLA approaches to creolisation

(14)

| 277

Syllable structure in Early Saramaccan and its contributing languages ESA

Gbe

Kikongo

Twi

pattern

+ + ++ (++) ++ – –

yes/– yes yes/– – – – yes

var./– var. yes/– – – – yes

yes/– yes yes – – – yes

2b/1b 2b 2c 1c 1c 1a 3a

word-final position: nasal [m,n], stressed nasal [m,n], unstressed nasal [ŋ] obstruent liquid, stressed liquid, unstressed nasal-obstruent other CC, homorganic other CC, not homorganic

+ – – – – – – – –

– – – – – – – – –

– – – – – – – – –

yesb yesb yes – – – – – –

1b 1a 1a 1a 1a 1a 1a 1a 1a

word-internal position: V.V obstruent-approximant nasal-obstruent [rɡ,rk] other liquid-C obstruent-obstruent, homorganic obstruent-obstruent, not homorganic

++ + ++ (+) – – –

var. yes yes – – – –

var. yes/– yes – – – –

var. yes yes/– – – – –

n.a. 2b 2c 1b 1a 1a 1a

word-initial position: V, etymon = σ(σ) V, etymon = σσσ(σ) obstruent-approximant [s]-nasal [s]-fricative [s]-plosive nasal-plosivea

a b

This structure is not permitted in the superstrate. Twi allows [m,n] in general. A distinction between stressed and unstressed syllables cannot be made since Twi is a tone language and does not have lexical stress.

The table in (15) gives the category count for the nine theoretically possible combinations of restrictions in Early Saramaccan, its superstrate and substrate languages.

278 | 8 Explaining creole phonotactic restructuring (15)

Constellations of restrictions in the contributing languages and restructuring in Early Saramaccan superstrate

substrate

creole restructuring

no. of categories

type 1a type 1b type 1c

no restr. no restr. no restr.

restr. restr. restr.

systematic variable none

12 3 2

type 2a type 2b type 2c

no restr. no restr. no restr.

no restr. no restr. no restr.

systematic variable none

– 3 2

type 3a type 3b type 3c

restr. restr. restr.

no restr. no restr. no restr.

systematic variable none

1 – –

Three things are particularly noteworthy. First, the majority of those structures which are ruled out in the substrates also underly systematic restrictions in Early Saramaccan. Second, among the structures permitted in both superstrate and substrate, three categories show an unexpected outcome in the creole, namely variable restructuring. The pertinent categories include those for vowel-initial words and obstruent-approximant sequences in word-internal position.¹⁰ Third, the single structural category for which we find a type 3 constellation is systematically restructured in Early Saramaccan. This outcome is expected under Uffmann’s (2009) hypothesis that grammatical restructuring takes place in language transmission from one creole generation to the next if the input does not provide sufficient positive evidence for non-superstrate contrasts. The table in (16) summarises the available interpretations for the patterns found in the Saramaccan data. Note that ‘language transmission’ is used here as an abbreviation for grammatical restructuring in language transmission.

10 Note that the retention rate of medial obstruent-approximant sequences comes close to, but does not quite reach the level of systematic retention.

8.4 Testing SLA approaches to creolisation

(16)

| 279

Expected and unexpected patterns in Early Saramaccan

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c 3a

12 2 5 1

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition language transmission

2b

3

–

8.4.4 Mechanisms in creolisation: Early St Kitts The table in (17) compares the restrictedness of syllable structures in Early St Kitts and its contributing languages. As was the case before in the comparison for Saramaccan, the creole category for initial clusters has once again been split to facilitate the comparison with superstrate and substrate restrictions. All but one of the resulting structural categories are permitted in the superstrate, the exception being word-initial sequences of a nasal and a plosive. The substrates are clearly more restrictive. For the purpose of hypothesis testing, a structure was counted as unrestricted in the substrate only if both substrates allowed it. The reasoning behind this was that if either of the two main substrate languages disallowed the given structure, the more restrictive option would have been likely to occur in the levelled substrate grammar (see Uffmann (2003) for findings on vowel inventories that support such an assumption). As a consequence, the fact that Gbe has no C-[w] clusters in word-initial position led to a double classification of initial obstruent-approximant clusters.

280 | 8 Explaining creole phonotactic restructuring (17)

Syllable structure in St Kitts and its contributing languages ESK

Twi

Gbe

pattern

word-initial position: V, stressed V, unstressed obstruent-approximant [s]-plosive nasal-plosivea

++ – ++ – (–)

yesb yesb yes – yes

yesb yesb yes/– – yes

2c 2a 2c/1c 1a 3a

word-final position: nasal non-nasal C rhotic-nasal C-fricative nasal-plosive[labial/dorsal] C-plosive[coronal]

++ ++ (–) (+) (++) –

yes – – – – –

– – – – – –

1c 1c 1a 1b 1c 1a

word-internal position: V.V obstruent-approximant nasal-obstruent lateral-nasal fricative-C, not homorganic fricative-C, homorganic

(+) ++ ++ (–) ++ –

var. yes yes/– – – –

var. yes yes – – –

n.a. 2c 2c/1c 1a 1c 1a

a a

This structure is not permitted in the superstrate. Twi and Gbe are tone languages and do not distinguish between lexically stressed and unstressed syllables. Both languages allow vowel-initial words, however.

The category count for each pattern of restrictions in St Kitts and its contributing languages is given in (18).

8.4 Testing SLA approaches to creolisation

(18)

| 281

Constellations of restrictions in the contributing languages and restructuring in Early St Kitts superstrate

substrate

creole restructuring

no. of categories

type 1a type 1b type 1c

no restr. no restr. no restr.

restr. restr. restr.

systematic variable none

5 1 6

type 2a type 2b type 2c

no restr. no restr. no restr.

no restr. no restr. no restr.

systematic variable none

1 – 4

type 3a type 3b type 3c

restr. restr. restr.

no restr. no restr. no restr.

systematic variable none

1 – –

Variable restructuring is rare in St Kitts. Most structures are either systematically retained or systematically restructured, retention being slightly more frequent overall. For one of the categories which show systematic restructuring this outcome is unexpected. This is the case for word-initial unstressed vowels, which are not ruled out in any of the contributing languages. The table in (19) summarises the findings concerning expected and unexpected patterns in the St Kitts data. (19)

Expected and unexpected patterns in Early St Kitts

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c 3a

5 4 7 1

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition language transmission

2a

1

–

8.4.5 Mechanisms in creolisation: Guiana French Creole As outlined in chapter 3.3, the main substrate language to be considered for Guiana French Creole is Gbe. Bantu languages (represented in (20) by Kikongo) and the Mande language Bambara are assumed to have played a less important role. Consequently, even if only Gbe, but none of the other substrates imposed

282 | 8 Explaining creole phonotactic restructuring restrictions on a given structure, this structure might still have been unacceptable for the majority of substrate speakers. Such a constellation, however, was not found for any of the structural categories. In all cases in which structural permissibility could be established, Gbe patterns with at least one of the other substrate languages. The minority choice was ignored in such cases, but is, of course, still given alongside all others in the table in (20). A few comments are in order with respect to the structural categories listed here as C-approximant (word-initially as well as word-medially). In word-initial position, this is the only cluster type that occurs in the set of etyma. Wordinternally, the vast majority of clusters with a non-rhotic C1 belong to the category C-approximant. The only two exceptions are nasal-obstruent sequences. The latter are not listed in (20), because evidence from the creole is too sketchy to allow for any assumptions on the acceptability of such clusters (one case of retention and one case of repair). Note that the retention rate for medial C-approximant clusters was recalculated after exclusion of the two NC-sequences. The acceptability label given in (20) for medial C-approximant sequences is thus not based on the percentage given in the tree diagram in chapter 6.4.3, but reflects a slightly higher retention rate of 83.3% (10 out of 12 cases with the pertinent cluster type). C-approximant clusters are noteworthy also because the substrate languages impose rather diverse restrictions on subsets of this category. Although, as outlined in section 8.3 above, C-approximant sequences are possible in all three substrates, not all cluster types occur in all languages. In fact, only C-[j] sequences are allowed in all three substrate languages. C-liquid sequences are acceptable in the majority of substrates, namely in Gbe and Bambara. Word-initial C-[w] clusters, on the other hand, occur only in a single substrate language, Kikongo, but are not attested in either Gbe or Bambara. This led to a double classification for word-initial C-approximant clusters. The situation is slightly different for wordinternal C-approximant clusters. In this position, C-[w] clusters are possible not only in Kikongo, but also in Gbe (at least in Fon lects, (cf. Capo 1991)), making the restrictions in Bambara a minority choice. Medial C-approximant sequences were therefore regarded as acceptable in the majority of the substrates.

8.4 Testing SLA approaches to creolisation

(20)

| 283

Syllable structure in Guiana FC and its contributing languages GFC

Gbe

Kikongo

Bambara

pattern

word-initial position: V, etymon = σ V, etymon = σσ(σ)(σ) C-approximant

+ + ++

– yes yes/–a

– var. yes/–b

– yes yes/–a

1b 2b 2c/1c

word-final position: rhotic other C rhotic-C obstruent-rhotic

– ++ – –

– – – –

– – – –

– yes/– – –

1a 1c 1a 1a

word-internal position: hiatus C-approximant rhotic-C

(–) ++ –

var. yes –

var. yes/–b –

var. yes/–a –

n.a. 2c 1a

a b

C-[w] clusters do not occur. C-liquid clusters do not occur.

The table in (21) summarises the frequencies of the different restriction constellations for all those categories for which a pattern could be established. (21)

Constellations of restrictions in the contributing languages and restructuring in Guiana FC superstrate

substrate

creole restructuring

no. of categories

type 1a type 1b type 1c

no restr. no restr. no restr.

restr. restr. restr.

systematic variable none

4 1 2

type 2a type 2b type 2c

no restr. no restr. no restr.

no restr. no restr. no restr.

systematic variable none

– 1 2

type 3a type 3b type 3c

restr. restr. restr.

no restr. no restr. no restr.

systematic variable none

– – –

In most cases, Guiana FC either places systematic restrictions on a given structure or systematically retains it. Variable restructuring is less common. It applies to only two structural categories and for one of them represents an unexpected outcome. Vowel-initial words that consist of more than one syllable variably lose

284 | 8 Explaining creole phonotactic restructuring their initial vowel despite the fact that this type of structure is possible in both superstrate and substrates. It is not entirely clear whether the motivation for the attested restructuring lies in syllable structure or some other property such as word structure or morphological interpretation. The data were not subjected to an analysis along these lines, so in the context of the present investigation the pattern remains unexplained. However, I will return to this issue in the overall evaluation of the two creolisation hypotheses. The table in (22) summarises findings for the Guiana FC data. (22)

Expected and unexpected patterns in Guiana FC

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c

4 2 3

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition

2b

1

–

8.4.6 Mechanisms in creolisation: Trinidad French Creole The table in (23) compares syllable structure restrictions across Trinidad FC and its main contributing languages. For the sake of this comparison, Twi represents Akan languages and Kikongo serves as a representative of Bantu languages. The category of vowel-initial words of at most two syllables received a double classification based on the fact that vowel-initial monosyllabic words do not occur in the substrates. Restrictions on subsets of word-initial obstruent-approximant clusters in Gbe and Kikongo, on the other hand, were ignored in the classification. Gbe lacks word-initial C-[w] clusters and Kikongo allows no obstruent-liquid sequences. Each of those cluster types, however, is possible in the remaining two substrates. Some structural categories could not be classified with respect to their acceptability in the substrate majority. The categories in point are C-rhotic clusters in word-initial as well as medial positions. The problem with these sequences is that it is not clear which substrate categories they should be compared to. The French rhotic is usually replaced by [ɣ] in Trinidad FC. The most important substrate language, Gbe, has both a rhotic and a velar fricative /ɣ/, but they differ widely in their distributions. The rhotic [r] can occur as C2 in clusters with a coronal C1 . The velar fricative /ɣ/ on the other hand has two allophones, [ɣ] and [w]. According to Brousseau (2011), [w] occurs before back (rounded) vowels, [ɣ] elsewhere. Gbe

| 285

8.4 Testing SLA approaches to creolisation

does not have any consonant sequences in which [ɣ] occurs as C2 , however. As for the remaining two substrates, Twi has C-rhotic sequences, but no voiced velar fricative [ɣ] (cf. Christaller 1875) and Kikongo, according to my sources, has neither a rhotic nor a voiced velar fricative. Kikongo is thus the only substrate for which a clear statement on the (un)acceptability of the pertinent clusters can be made. Given this situation, C-rhotic clusters were excluded from further consideration. (23)

Syllable structure in Trinidad FC and its contributing languages TFC

Gbe

Twi

Kikongo

pattern

word-initial position: V, etymon = σ(σ) V, etymon = σσσ(σ) C-[l], C-glide C-rhotic, labial context C-rhotic, non-labial context

++ + ++ – ++

yes/– yes yes/– ? ?

yes/– yes yes ? ?

var./– var. yes/– – –

2c/1c 2b 2c n.a. n.a.

word-final position: nasal; plosive lateral rhotic fricative[−voice] fricative[+voice, coronal] fricative[+voice, labial] rhotic-C obstruent[−voice] -C obstruent[+voice] -C

++ ++ – ++ + ++ – – –

– – – – – – – – –

yes/– – – – – – – – –

– – – – – – – – –

1c 1c 1a 1c 1b 1c 1a 1a 1a

word-internal position: hiatus C-[l], C-glide C-rhotic, labial context C-rhotic, nonlabial context [s]-plosive; [l]-C plosive.C rhotic-C[labial] rhotic-C[coronal/dorsal]

(+) ++ – ++ ++ + – –

var. yes/– ? ? – – – –

var. yes ? ? – – – –

var. yes/– – – – – – –

n.a. 2c n.a. n.a. 1c 1b 1a 1a

The category count for all theoretically possible constellations of restrictions and restructuring in superstrate, substrates and creole is given in the table in (24).

286 | 8 Explaining creole phonotactic restructuring (24)

Constellations of restrictions in the contributing languages and restructuring in Trinidad FC superstrate

substrate

creole restructuring

no. of categories

type 1a type 1b type 1c

no restr. no restr. no restr.

restr. restr. restr.

systematic variable none

6 2 6

type 2a type 2b type 2c

no restr. no restr. no restr.

no restr. no restr. no restr.

systematic variable none

– 1 3

type 3a type 3b type 3c

restr. restr. restr.

no restr. no restr. no restr.

systematic variable none

– – –

Trinidad FC shows more systematic than variable behaviour. All patterns except one are in accordance with either the transfer hypothesis or the L2 acquisition hypothesis. Only variable restructuring of vowel-initial words with more than two syllables is not accounted for under either of the hypotheses. The fact that onsetless structures are systematically retained in shorter words indicates that the creole restriction here is not motivated primarily in syllable structure. Word length appears to be a factor at play, but the question remains where such a restriction on word length might originate. The patterns found in the Trinidad FC data are listed again in (25), grouped according to the hypothesis under which they are accounted for, if any. (25)

Expected and unexpected patterns in Trinidad FC

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c

6 3 8

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition

2b

1

–

8.4.7 Mechanisms in creolisation: The overall picture The preceding sections dealt with the restriction patterns found in the individual creoles and their contributing languages and addressed the question in how far

8.4 Testing SLA approaches to creolisation

| 287

these patterns can be accounted for by the creolisation hypotheses by Plag (2009) and Uffmann (2009). In this section, I will generalise across these languagespecific findings and attempt a more general evaluation of the two hypotheses. I will also discuss one aspect of the proposals which has not been addressed yet, the role that is ascribed to the conditions under which a given creole developed. Although the language-specific accounts in the previous sections revealed some differences with respect to the relative importance of different factors, there are some general conclusions we can draw that apply to all creoles investigated here. First, a strict transfer-only approach leaves substantial parts of the data unaccounted for. Not one of the six creole languages displays only patterns consistent with transfer. Thus, if creoles do indeed start out with full transfer of substrate grammatical structures, then this initial stage has not been preserved entirely in any of the languages investigated here. This is not surprising, seeing that the data investigated here do not date back to the starting point of creolisation, but rather to some point after stabilisation. What we can say with respect to the two proposals, though, is that the strict version of Plag’s early-interlanguage hypothesis, predicting equivalence of substrate and creole grammars, is not well-supported by the data when not supplemented by other factors. Second, an approach that allows both L1 transfer and L2 acquisition is much more successful than a transferonly approach. These two interpretations cover the creole treatment of the clear majority of structural categories in any of the creoles investigated here. Third, patterns not expected under either of the above hypotheses are overall not very common in the creoles. Unexpected patterns are not the only ones which are rare. The same is true also for expected patterns consistent with grammatical restructuring in language transmission (pattern 3a). These generalisations are also reflected in the pooled category count for all six data sets given in (26). (26)

Overall count of expected and unexpected patterns in the six data sets

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c 3a

37 21 46 2

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition language transmission

2a 2b

1 7

– –

The creole outcome for 58 out of a total of 114 structural categories can be accounted for under the assumption that creoles allow the same kinds of struc-

288 | 8 Explaining creole phonotactic restructuring tures that are also permissible in their substrates (patterns 1a and 2c). In other words, about half of the observed phenomena remain unexplained if no additional factor is assumed. However, if we add those parts of the data which are consistent with an interpretation in terms of L2 acquisition, the number of categories accounted for rises to an impressive 104 (out of 114). Two further categories are covered by Uffmann’s factor of grammatical restructuring in language transmission. Uffmann’s (2009) creolisation scenario leaves the creole outcome of a mere 8 structural categories unexplained. Plag’s approach fares almost as well with 10 categories not accounted for. Even though the number of pertinent categories is small, the question arises which potential explanation there might be for those restriction patterns which are not covered by any of the above mechanisms, but which are nonetheless attested. Why is it the case that some structures which are possible in both superstrate and substrates are repaired in the creole either variably (pattern 2b, 7 categories) or systematically (pattern 2a, 1 category), both outcomes which are not predicted by either Plag or Uffmann. One potential explanatory factor that comes to mind is the alleged tendency of creoles to display unmarked structures. As outlined in chapter 2, supporters of the universalist approach to creolisation assume that phonotactic restructuring in creoles results from a preference of unmarked structures rather than from a transfer of L1 restrictions. In fact, given that markedness effects occur in L1 as well as L2 acquisition and given that both L1 and L2 acquisition are assumed to be part of the creolisation process (at least according to Uffmann 2009), it should not surprise us very much to find markedness effects also in creoles. However, a closer look at the structural categories for which the unexpected patterns are attested casts doubt on the assumption that restructuring applies in order to avoid marked syllable structures. All but one of the pertinent cases concern vowel-initial words. If the lack of a syllable onset were the motivating factor behind restructuring, we would expect restructuring to apply to all vowel-initial words alike. What we find instead is that many of the creoles make further distinctions to the effect that initial vowels are targeted by repairs only if the initial syllable is unstressed (Early St Kitts) or if the etymon consists of three or more syllables (Berbice Dutch, Negerhollands, Trinidad FC). If these are markedness effects, then they are more likely to be connected to word structure than syllable structure. For instance, as mentioned earlier in the relevant sections of chapters 4 and 6, deletion in longer words might be triggered by a preference for words to maximally consist of a binary foot, a structure which is supposed to be prosodically unmarked (McCarthy and Prince 1994, 1996).

8.4 Testing SLA approaches to creolisation

| 289

Apart from constraints on prosodic structure, also other, alternative explanations for the deletion of initial vowels are conceivable. Recall that the range of factors that could be taken into account as potential predictors for the creole treatment of vowel-initial words was very limited. In particular, vowel quality was not systematically investigated. As a consequence, vowel quality was ignored also in the restriction classification for the substrates. In many of the substrates, words can only start in a vowel if that vowel is a prefix. Since not all vowel qualities are covered by prefixes, it is often only a subset of the language’s vowels that is attested in word-initial position. Apart from that, also the prefix status of initial vowels in the substrates could have influenced etymon treatment. If the initial vowel of an etymon was interpreted by substrate speakers as representing a prefix, this could have led to deletion in the creole form (cf. Steele and Brousseau (2006: 336ff) for a similar suggestion regarding initial syllable deletion in Haitian Creole). There are thus several potential explanations for the unpredicted repairs of onsetless word-initial structures. Only a detailed analysis of the relevant phenomena could show what triggers deletion in the pertinent cases. Syllable structure preferences do not appear to be the most likely factor here, although they might, of course, have interacted with other factors to produce the observed outcome. At the rather general level of restriction constellations, the predictions by Plag (2009) and Uffmann (2009) thus turn out to be correct for large parts of the data. We could argue, of course, that this is not very surprising since many of the theoretically possible patterns are in fact predicted to occur. In particular, for type 1 constellations (restrictions only in the substrates), all three possible creole outcomes can be accounted for under either the transfer or the L2 acquisition hypothesis. The relative frequency of transfer and L2 acquisition patterns is not expected to be the same in all creoles, however. The relevant prediction that addresses this issue concerns the role of the conditions under which a given creole developed. Both Plag (2009) and Uffmann (2009) assume that the mechanisms at work in creolisation do not vary greatly across different creoles, but that their relative impact on the emerging creole may do so, depending on the respective social conditions. More specifically, greater availability of or more prolonged contact with the superstrate is assumed to correlate with a larger L2 acquisition component. Conversely, L1 transfer would be expected to feature more prominently in creoles developing in situations where contact with the superstrate was more limited. In order to test this, we need to compare the transfer and L2 acquisition components in creoles with similar superstrate and substrate structures. Given that the substrates show only minor differences in terms of syllable structure, crosscreole differences with respect to the contributing substrates will be disregarded here. They are unlikely to cause large-scale differences regarding the relative impact of L1 transfer. As far as similarity of superstrate structures is concerned,

290 | 8 Explaining creole phonotactic restructuring only creoles with the same main lexifier will be compared. I will discuss the pairs of Dutch-based, English-based and French-based creoles in turn. If the two creoles in a given pair differ with respect to the availability of the superstrate, then we would expect them to show differences in the relative impact of transfer and acquisition on creole structures, too. A note of caution is in order here with respect to the comparison of restriction patterns across creoles. Due to the sometimes vast differences in the nature and size of structural categories, the frequency of a particular pattern does not necessarily correlate with the share of the data that it covers. For instance, simply because L1 transfer and L2 acquisition can account for the patterns of an equal number of structural categories in Berbice Dutch (cf. the table in (27) below), that does not mean that both account for equal proportions of the data. Both mechanisms do, on the other hand, account for an equal number of structural distinctions made in the creole. This latter interpretation is what is intended whenever I comment on the relative importance of different mechanisms across creoles in the following. Given this basis, the comparisons in this section will be impressionistic rather than properly statistical. Among the two Dutch-based creoles, Berbice Dutch has clearly had more limited exposure to the superstrate. The area in which the creole developed was abandoned by the Dutch around the middle of the 18th century in the context of slave rebellions and a general shift of settlements towards the coast (cf. section 3.3). For Negerhollands, on the other hand, continued presence of the superstrate is documented and access to the superstrate appears to have been comparatively good. Consequently, we should find a larger L2 acquisition component in Negerhollands than in Berbice Dutch. The tables summarising expected and unexpected patterns for both languages (repeated in (27) and (28) for convenience) show that this is indeed the case. In Berbice Dutch, L1 transfer and L2 acquisition can explain the creole outcome for a comparable number of structural categories (8 categories each, leaving aside pattern 2c, for which no definite decision can be made). In Negerhollands, the category count is 15 to 2 in favour of L2 acquisition, i.e. most structural categories show patterns consistent with partial or complete L2 acquisition rather than full transfer.

8.4 Testing SLA approaches to creolisation

(27)

Expected and unexpected patterns in Berbice Dutch

expected

unexpected

(28)

| 291

patterns

no. of categories

interpretation

1a 2c 1b, 1c

8 5 8

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition

2b

1

–

Expected and unexpected patterns in Negerhollands

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c

2 5 15

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition

2b

1

–

Also the two English-based creoles differ in terms of superstrate availability. Here, Early Saramaccan is the one with more limited access to the superstrate, English having been removed from the contact setting after no more than a few decades. Again, we can observe a tendency in the predicted direction (cf. the tables in (29) and (30)). Early Saramaccan shows less evidence of L2 acquisition than Early St Kitts. Unlike the Dutch-based creoles, however, none of the English-based creoles show an overall predominance of superstrate acquisition. In Early St Kitts, the category count for structures interpreted as reflecting transfer and L2 acquisition, respectively, is roughly equal. In Early Saramaccan, L1 transfer accounts for the largest group of structural categories by far (12 as opposed to 5 which can be attributed to L2 acquisition).

292 | 8 Explaining creole phonotactic restructuring (29)

Expected and unexpected patterns in Early Saramaccan

expected

unexpected

(30)

patterns

no. of categories

interpretation

1a 2c 1b, 1c 3a

12 2 5 1

transfer from L1 L1 transfer or L2 acquisition (partial or complete) L2 acquisition language transmission

2b

3

–

Expected and unexpected patterns in Early St Kitts

expected

unexpected

patterns

no. of categories

interpretation

1a 2c 1b, 1c 3a

5 4 7 1

L1 transfer L1 transfer or L2 acquisition (partial or complete) L2 acquisition language transmission

2a

1

–

The French-based creoles present a different case from the Dutch- and the Englishbased creoles. Both Guiana FC and Trinidad FC developed under continued superstrate presence. Whether the degree of contact differed nonetheless, perhaps to a less dramatic degree, cannot easily be established. On the basis of the two creoles’ historical backgrounds we certainly would not expect huge differences in the degree of superstrate influence (see also McWhorter (2004) for a similar argument). A look at the assumed transfer and L2 components in Guiana FC and Trinidad FC (cf. the tables in (31) and (32)) confirms this expectation. Neither Guiana FC nor Trinidad FC show a clear predominance of either L1 transfer or L2 acquisition. The two creoles can be regarded as similar with regard to the relative impact the superstrate had on their respective syllable structure.¹¹

11 It ought to be kept in mind, however, that the analysis in terms of restriction patterns was less complete for Trinidad FC than for the other creoles since some structural categories with indeterminable substrate restrictions had to be excluded. Those parts of the data might have tipped the balance in one or the other direction, so the above conclusion is not made without reservations.

8.4 Testing SLA approaches to creolisation

(31)

Expected and unexpected patterns in Guiana FC

expected

not expected

(32)

| 293

patterns

no. of categories

interpretation

1a 2c 1b, 1c

4 2 3

L1 transfer L1 transfer or L2 acquisition (partial) L2 acquisition

2b

1

–

Expected and unexpected patterns in Trinidad FC

expected

not expected

patterns

no. of categories

interpretation

1a 2c 1b, 1c

6 3 8

L1 transfer L1 transfer or L2 acquisition (partial) L2 acquisition

2b

1

–

At the level of restriction patterns it thus seems that a second language acquisition approach to creolisation along the lines of Plag (2009) or Uffmann (2009) is highly successful. The above results suggest that L1 transfer and L2 acquisition are indeed the main factors at play in the shaping of creole syllables and that the degree of superstrate acquisition may vary across creoles depending on the availability of the superstrate. Some evidence could be found also for grammatical restructuring in language transmission. Finally, substrate levelling was assumed to take place in creolisation, and the findings do not contradict that assumption. It should be noted, though, that clear evidence for levelling effects did not emerge. Given the strong similarities of syllable structure restrictions in different substrates and given the fact that all creoles allow a greater range of structures than their substrates, this result is not unexpected. Some questions remain unanswered, however. The two SLA approaches were tested above by comparing the structures of individual creoles with those of their contributing languages. The relation between those structures is, of course, the focus of creolisation theories. Nonetheless it might be interesting to take a more comparative view at this point and re-examine the findings of the cross-creole comparisons in chapter 7 from the perspective of SLA approaches to creolisation. How can we explain the observed similarities and differences if we assume that L1 transfer and L2 acquisition are the main mechanisms at work in creolisation? That is the question to which I turn in the next section.

294 | 8 Explaining creole phonotactic restructuring 8.4.8 Similarities and differences across creoles: What causes the observed patterns? In what follows, I will consider mostly the general tendencies identified in chapter 7 and comment only briefly on possible interpretations of creole-particular restrictions (i.e. distinctions between subsets of some structure which occur only in that creole). A detailed discussion of all these individual patterns is beyond the scope of this book. Two of the common tendencies identified in chapter 7 were the comparatively high retention rates for complex onsets of the type obstruent-approximant and simple nasal codas. These preferences on the part of the creoles may well reflect preferences in the substrate languages. All of them allow at least a subset of obstruent-approximant clusters. If word-final codas occur at all, then only nasals are permitted in that position. Word-internally, all substrates allow some types of nasal-obstruent sequences. Also the dispreference of complex codas in the creoles could be attributed to the ban on such structures in the substrates. However, not all structures which are ruled out in the substrates are equally likely to be repaired in the creoles. For instance, four of the creoles retain a large percentage of simple non-nasal codas, but are much less liberal with respect to complex codas, even though both types of structures are equally unacceptable in the substrates. If this type of within-creole variation is not a transfer effect, can it receive an explanation instead that is in accordance with L2 acquisition? In the case of simple and complex codas, an explanation in terms of markedness suggests itself. Markedness has been shown to influence interlanguage development. With respect to syllable structure, interlanguage studies have found that shorter, less marked, syllable constituents are acquired before longer, more marked, ones (cf. Carlisle 2001: 10, see also Andersen 1987). The constraints in the different creoles may thus reflect different intermediate stages of superstrate acquisition. This interpretation is also in line with the observation that creoles which had comparatively good access to their respective superstrates retain more (and more complex) superstrate structures when compared to creoles with similar contributing languages but more limited superstrate access. While Negerhollands and Early St Kitts preserve most simple codas and some complex codas, Berbice Dutch and Early Saramaccan repair even most simple codas. The fact that rhotics in coda position often form major targets of restructuring cannot be related to markedness as easily. Rather, this dispreference could be interpreted as an effect of perceptual salience (or lack thereof). Salience is another factor which plays a role in interlanguages. We would expect perceptu-

8.4 Testing SLA approaches to creolisation

| 295

ally salient structures to be learned before structures with weak perceptual cues.¹² Coda rhotics clearly fall into the latter category. In other words, the greater likelihood of repair for rhotics as opposed to other simple codas (cf. Negerhollands and the two French-based creoles) is also in line with an interpretation of creoles as interlanguages. A similar argument could be made for the rather general observation that the creoles allow overall more complex structures in onset position than in coda position. Acoustic cues for consonants are stronger in prevocalic than in postvocalic position. From the point of view of salience, onsets are thus preferable to codas. Note, however, that this preference could equally well be the result of substrate transfer. In fact, it might even be the case that all of the above-mentioned factors conspired in creating the observed pattern: L1 transfer would have led to an initial acceptability of (some) obstruent-approximant clusters and a ban on most types of codas. Subsequent L2 acquisition would have caused some of the constraints to be gradually relaxed, starting with constraints on perceptually salient and/or relatively unmarked structures. It thus seems that most of the general tendencies identified in chapter 7 can be accounted for under an SLA approach to creolisation.¹³ As far as cross-creole differences are concerned, only variation in the degree of restrictiveness was mentioned above. But what about creole-particular distinctions between subsets of a given structure? How could they be explained? Some of these restrictions may be substrate-induced and the lack of parallel distinctions in other creoles may then be due to different constellations of substrates. Other patterns are likely to receive an explanation in terms of L2 acquisition, including markedness and/or salience effects. Perceptual salience for one often seems to play a role also in distinctions which are made only in individual creoles (for examples cf. the discussion in the following chapter). What may be impossible to answer at present, however, is the question why a given acquisition-induced distinction occurs in one creole and not in another. Differences in the (detailed) properties of the lexifier input could play a role, of course, but there may also be cases for which no such argument can be made. But even if it turns out that we cannot tell why one creole shows one kind of SLA-compatible restriction and another creole shows another, this should not lead us to discard the entire approach. After all, interlanguages are not completely uniform, either. In fact, a certain amount of variation is expected in L2 develop-

12 The role of perceptual salience in phonotactic restructuring will be discussed in more detail in the following section. 13 As already mentioned in the previous section, the creoles’ treatment of onsetless syllables may or may not present an exception.

296 | 8 Explaining creole phonotactic restructuring ment (cf. e.g. Hancin-Bhatt 2008). In order to decide whether or not the variation observed in creoles is similar to the variation that occurs in interlanguages, we might have to investigate L2 phonological systems more thoroughly first. Only if we know what to expect and what not to expect in such L2 systems can we test SLA approaches to the emergence of creole phonology right down to the finest detail. Despite some questions left to future research, this section has shown that also general tendencies that hold across creoles are in accordance with the tested hypotheses as are cross-creole differences in terms of restrictiveness. Cross-creole differences in the distinctions between subsets of a given structures, on the other hand, were not discussed in detail. Note, though, that some of the relevant restrictions will be dealt with in the following section, which focusses on processes of phonotactic restructuring. Having found support for the tested hypotheses in the patterns of syllable structure restrictions observed in the creoles, we now need to address the question of whether the concrete restructuring mechanisms that occur in the data are also consistent with the suggested interpretations.

8.4.9 Creole phonotactic restructuring: What motivates repair choices? The only predictions we have on specific restructuring processes come from Plag (2009). Based on restructuring choices in interlanguage and loanwords, he expects vowel epenthesis to be the typical repair in cases in which nonnative structures are adapted to an L1 system that imposes tighter restrictions on syllable structure. As briefly discussed earlier, it is not clear from the literature whether vowel epenthesis is always the preferred choice in interlanguage varieties. This has been challenged at least for more advanced stages of second language acquisition. In loanword adaptations, on the other hand, vowel epenthesis is indeed more widely attested than consonant deletion (cf. Kang 2011). We could thus reasonably expect to find a high rate of vowel epenthesis among the repairs at least for those creoles with the greatest predicted influence of L1 transfer, i.e. the two creoles with the most limited access to the superstrate, Berbice Dutch and Early Saramaccan. We already saw above that, according to the analysis in terms of restriction patterns, these two creoles do indeed exhibit a comparatively high degree of L1 transfer.¹⁴ We can now test whether also Plag’s process-specific hypothesis that “we should find more epenthesis in contact situations where the social conditions for L2 acquisition are less favorable” (Plag 2009: 132) is correct.

14 “Comparatively high” should be understood here as high in comparison with the other creole in the respective pair.

8.4 Testing SLA approaches to creolisation

| 297

This certainly seems to be the case at least for the languages investigated here. Although other processes of restructuring are not completely absent from either Berbice Dutch or Early Saramaccan, the vast majority of illicit structures in the two creoles are repaired by insertion of an epenthetic vowel (cf. the results in chapters 4.4 and 5.4). Negerhollands and Early St Kitts offer a more balanced mix of vowel epenthesis and consonant deletion, the two main repair strategies (cf. the results in chapters 4.5 and 5.5). While these findings suggest that vowel epenthesis does indeed prevail in those creoles where we would expect comparatively more L1 transfer, the data also show that vowel epenthesis is not the only process that applies to structures interpreted as showing effects of full transfer (pattern 1a). The French-based creoles offer a good example here. Despite the fact that the above analysis suggested the retention of substrate restrictions for several structural categories in both Guiana FC and Trinidad FC, neither of the two creoles employs vowel epenthesis to repair any of the impermissible etymon structures in my data.¹⁵ We might be tempted to conclude that the two French-based creoles are so heavily influenced by the superstrate that they do not show mechanisms typical of L1 transfer any more. That, of course, would mean that our above analysis in terms of restriction patterns is seriously flawed since it suggested some degree of L1 transfer in both Guiana FC and Trinidad FC. More importantly, however, this conclusion would also go against suggestions that have been made in the literature on French-based creoles exhibiting processes comparable to those attested for Guiana FC and Trinidad FC. For instance, in studies on Haitian Creole, transfer from the substrate has been argued to be responsible for substitution of a glide [w] for an onset rhotic before rounded vowels (Steele and Brousseau 2006, Russell Webb 2010) as well as for deletion of rhotics in coda positions (Russell Webb 2010), i.e. processes which are attested in highly similar or even identical form also in my data. The absence of vowel epenthesis from Guiana FC and Trinidad FC thus cannot simply be equated with an absence of transfer-induced restructuring. Rather, we must conclude that the absence of vowel epenthesis does not tell us anything about the presence or absence of L1 transfer.

15 Other French creoles are similar in this respect. As far as I am aware, there is only a single type of structure for which vowel epenthesis, or, more precisely, vowel prothesis, has been reported as preferred repair in other French-based creoles, namely word-initial [s]-obstruent clusters (cf. e.g. Steele and Brousseau 2006). Unfortunately, such structures did not occur in my data, so I cannot comment on them.

298 | 8 Explaining creole phonotactic restructuring Findings from loanword adaptation studies support this latter assumption.¹⁶ Even though very common in loanwords, vowel epenthesis is not the only attested repair mechanism. Some languages also employ consonant deletion to bring words with nonnative phonotactics into accordance with L1 structures. Attested cases include, for example, deletion of illicit word-final coda consonants and the simplification of word-initial clusters by the deletion of either C1 or C2 (cf. Kang 2011). The target of deletion varies across positions as well as across different cluster types. For instance, initial [s]-C clusters may be simplified by deletion of the sibilant, whereas clusters with a sonorant in second position lose that sonorant rather than the initial consonant. These preferences correspond to what we find in the creoles. It is interesting to note, though, that, according to Kang (2011), no language uses only consonant deletion in loanword adaptations, whereas vowel epenthesis may be used across the board. Apart from vowel epenthesis and consonant deletion, also processes of segmental substitution are common in loanword adaptations. While many of these modifications operate purely on the segmental level, some may have an impact on phonotactics, too. What follows from the above findings on loanwords is that, if we accept that loanword adaptations are similar to cases of transfer in creolisation settings, then we also have to take into account processes other than vowel epenthesis as potential effects of substrate transfer. Seeing that we cannot a priori divide the restructuring mechanisms into strategies possible in transfer situations and strategies not possible in transfer situations, a mere comparison of the frequencies of different repairs cannot be used to test the suggested interpretations of creole phonotactic restructuring. Instead, I suggest that it may be more helpful to focus on the factors that have been claimed to motivate the choice of the observed restructuring mechanisms in loanwords and to see whether we can find comparable factors at work in the creoles. If such parallels can be found, this would lend further support to the assumption of initial full transfer of substrate restrictions. Loanword adaptations are generally agreed to constitute minimal transformations, i.e. the chosen repair involves the smallest possible change that brings the input in accordance with L1 structures. There is an ongoing debate, however, on

16 Since we are focusing on L1 transfer here, loanword adaptations present a better case for comparison than processes attested in second language acquisition. In the case of loanwords, we can be fairly certain that it is indeed the L1 phonological system into which the nonnative words are integrated. In second languages, this may be true for the initial stages, but not necessarily for later ones. I will therefore concentrate on the loanword parallels here. For similar assumptions on parallels between loanword adaptation and creole phonotactic restructuring see also Uffmann (2007), Boersma and Hamann (2009), Russell Webb (2010).

8.4 Testing SLA approaches to creolisation

| 299

the questions of whether the transformations themselves are phonologically minimal (e.g. Paradis and Lacharité 1997, Uffmann 2001, 2007) or phonetically minimal (e.g. Fleischhacker 2001, 2005, Peperkamp and Dupoux 2003, Peperkamp 2005, Peperkamp et al. 2008) and whether they take place in production or in perception. A growing number of researchers argue for models which take both perceptual and production factors into account (e.g. Silverman (1992), Yip (1993), Kang (2003), Kenstowicz and Suchato (2006), Rose and Demuth (2006), Boersma and Hamann (2009)). Recent studies of nonnative speech perception support the assumption that L2 acoustic input may not be perceived faithfully. More particularly, speakers may use their L1 perceptual processing patterns in the interpretation of nonnative auditory-phonetic input forms. Research has shown that L1 perceptual processing patterns, once established, are highly automatic and can be difficult to overcome even after several years of exposure to an L2 with different contrasts (cf. Strange and Shafer 2008). Where such differences exist between L1 and L2, the processing of an acoustic L2 input on the basis of L1 routines can lead to a non-targetlike interpretation of that input. Evidence for such phenomena comes from attested cases of ‘phonological deafness’ to nonnative contrasts. For instance, native speakers of Korean, which only have a single liquid phoneme in their native language, have extreme difficulty discriminating between English [l] and [r]. This is paralleled by segmental substitutions in Korean loanword adaptations, where input [l] may surface as [r] (cf. e.g. Peperkamp and Dupoux 2003). Similarly, perception experiments have shown that, if confronted with some input that does not conform to their native phonotactics, speakers may perceive vowels that are not actually in the acoustic input (e.g. Dupoux et al. 1999, Kabak 2003, Peperkamp and Dupoux 2003, Peperkamp et al. 2008, Hallé 2008). For example, Dupoux et al. (1999) showed that speakers of Japanese were unable to discriminate between the sounds [ebzo] and [ebuzo], which suggests that they perceive both input forms as the phonological form /e.bu.zo/ with an illusory vowel breaking up the nonnative sequence [bz]. Crucially, structures resulting from alternative repairs, such as deletion of one of the consonants, would involve more serious deviations from L1 mappings of auditory input to phonological surface structure. In other words, mapping the phonetic input [ebzo] to a phonological structure including an illusory vowel follows the native perceptual patterns more closely than mapping the input [ebzo] to a phonological structure in which one of the consonants does not have a correspondent. We have thus good reason to believe that nonnative perceptual processing plays an important role in the adaptations we observe in loanwords and that vowel epenthesis may be preferred to other processes because in many cases the resulting sequence constitutes the best native match for the nonnative phonetic input.

300 | 8 Explaining creole phonotactic restructuring That is not to say that all restructuring necessarily takes place in perception. I am inclined to side with researchers such as Boersma and Hamann who recognise adaptations in both perception and production. In fact, some processes we find in both loanword adaptations and creole phonotactic restructuring are very unlikely to have their origin in perception. The simplification of word-initial [s]-C clusters by deletion of the sibilant is a case in point.¹⁷ [s] is a highly salient sound with strong acoustic cues. Mapping an input containing these cues onto a structure lacking a corresponding segment is unlikely to constitute the optimal choice in perception and certainly does not qualify as a phonetically minimal transformation (see also Fleischhacker 2005, Kang 2011). Obviously, there must be other, additional factors at work. A detailed discussion of such factors is beyond the scope of this book. For the purpose of the present discussion, I will therefore simply assume that some, but not all of the adaptations we observe in loanwords are perceptual and that additional transformations may take place in production.¹⁸ Returning to the creoles, if the same mechanisms also operate in creole phonotactic restructuring, we should find a considerable number of processes which can be interpreted as originating in perception, but also others that may call for a more production-oriented approach. In the following, I will first point out some adaptations which have been argued to have their source in perception. I will then turn to cases of phonotactic restructuring in loanwords and creoles which are just as likely (or even more likely) to take place in production. Vowel epenthesis is probably the most commonly discussed process in research on nonnative perception. Illusory vowels have been suggested, for instance, to result from the vocalic reinterpretation of consonantal release (e.g. Blevins and Garrett 1998, Peperkamp et al. 2008, Boersma and Hamann 2009), high frequency noise spectra of sibilants (Fleischhacker 2001) or nonvocalic transitions between consonants (Dupoux et al. 2011). Parallels in the creoles include the wide-spread word-final paragoge in Berbice Dutch and Early Saramaccan, but also epenthetic vowels breaking up [kn]-onsets in the Dutch-based creoles, liquid-nasal codas in Berbice Dutch and Early St Kitts and liquid-initial coda clusters in more general in Negerhollands. Vowel epenthesis into non-homorganic obstruent clusters in Early Saramaccan and Early St Kitts is another case in point.

17 According to Kang (2011), [s]-deletion applies in loanword adaptations in Teluga. In my creole data, the process is attested in Early St Kitts and Early Saramaccan. 18 Boersma and Hamann (2009) suggest that, in addition to perception and production, also lexical recognition and phonetic implementation have to be considered as processes during which adaptations can take place. The above assumptions may thus still constitute a simplification of the actual state of affairs.

8.4 Testing SLA approaches to creolisation

| 301

Also consonant-vowel metathesis has been argued to have its origin in perception. This has been suggested for metathesis as a diachronic process rather than as a repair strategy triggered by L1 transfer. Similar interpretations may be applicable also in the latter context, however. The type of consonant-vowel metathesis for which a perceptual explanation has been advanced involves consonants which are “cued by one or more drawn-out perceptual features” (Blevins and Garrett 1998: 512). If these features spread across the neighbouring vowel, this can cause ambiguity as to the origin of the consonantal feature, which, in turn, can lead to a reinterpretation of the consonant as originating on the other side of the vowel.¹⁹ According to Blevins and Garrett, the types of consonants that may be involved in perceptual metathesis include liquids, laryngeals, pharyngeals and glides (Blevins and Garrett 1998: 513). Among the six creoles investigated in the present study, Berbice Dutch and Early Saramaccan show substantial patterns of liquidvowel metathesis involving coda liquids, particularly if these liquids occur in unstressed syllables. Individual cases of liquid-vowel metathesis are also attested in the other creoles. None of the substrates allow liquids in coda position, so if the linear origin of the acoustic features for coda liquids was ambiguous in the input, this input would likely have been perceived as (C)LV rather than (C)VL.²⁰ A proposal for a perceptual origin of certain segmental substitutions comes from Russell Webb (2010), who discusses restructuring in Haitian Creole. In Haitian, etymological onset rhotics surface as [w] in front of rounded vowels, but not elsewhere. Russell Webb suggests that substitution takes place because the French rhotic is produced with anticipatory lip rounding in this position, i.e. as [ʁw ], a sound which, according to Russell Webb, does not occur before rounded vowels in the main substrate, Gbe. The glide [w] is chosen because its acoustic properties are highly similar to those of a labialised dorsal continuant [ʁw ]. In contrast to the latter, however, [w] is allowed in front of rounded vowels in the substrate. Trinidad FC shows a substitution pattern that is similar to the

19 Russell Webb and Bradley (2009) take a slightly different approach to rhotic-vowel metathesis in French and Spanish, focusing on gestural alignment rather than the elongation of phonetic cues and arguing that complete overlap of rhotic and vowel gestures made a reinterpretation of the linear origin of the rhotic possible. 20 Recall, however, that the most common cases of ‘metathesis’ in the creoles, those involving liquids in unstressed final syllables, can alternatively be interpreted as a combination of vowel paragoge followed by vowel deletion. If this interpretation applies, then adaptation may be only partially perceptual in nature. Thus, a perceptual explanation might apply to the paragogic vowel. Vowel deletion, on the other hand, could result from transfer of a phonological rule of the substrate production grammar, allowing contractions of the type CVLV > CLV. Such contractions are the source of surface clusters in several of the substrate languages (cf. section 8.3).

302 | 8 Explaining creole phonotactic restructuring one attested in Haitian Creole. As is the case in Haitian, French rhotics surface as glide [w] before rounded vowels also in the Trinidadian creole. However, Trinidad FC has [w] for etymological rhotics also in clusters with labial C1 s, irrespective of the quality of the following vowel, so we might be dealing with only a partial parallel here. Russell Webb (2010) also suggests that certain consonant deletions in creoles may be the result of nonnative perception. He ascribes the deletion of coda rhotics in Haitian Creole to the weakness of their auditory cues, assuming that substrate speakers did not associate such weak cues with a consonantal segment. Other coda consonants with more robust cues are usually preserved in Haitian Creole. Also Shinohara (2006) argues that segments with low acoustic and auditory salience may be simplified in loanwords, whereas more salient segments may surface intact. To give an example, word-final nasal-plosive sequences (NP#) are less salient than nasal-fricative sequences (NF#). The former are taken to be more likely to undergo consonant deletion than the latter essentially because the contrast between NP# and N# is smaller than the contrast between NF# and N#. Unlike supporters of a perception-only approach to loanword adaptations, however, Shinohara (2006) proposes that deleted consonants are perceived by the speakers, but not parsed in production. She argues that covert knowledge about the perceptual salience of particular segments or sequences in different contexts is incorporated into speakers’ grammars (P-map hypothesis, cf. Steriade 2001). This knowledge is assumed to take the form of universal perceptibility scales which can interact with other grammatical constraints. Such an interaction may lead to the preservation of all pertinent elements or it may cause the deletion of elements at the less salient end of the scale while at the same time allowing more salient elements to surface intact. The same perceptibility scales are assumed to be active in language acquisition, where more salient elements are expected to emerge before relatively less salient elements on the same scale. Under Shinohara’s approach, the deletion of perceptually weak segments is thus seen as resulting from the interaction of linguistic universals and L1 grammatical constraints. Whether a perception- or a production-oriented explanation is more appropriate may have to be determined individually for different deletion processes. It is not inconceivable that some of the deletions took place in perception. For others, production-based simplification seems more likely. Cases in my data in which perceptually weak segments or sequences are targeted include coda rhotic deletion in Guiana FC, Trinidad FC and Negerhollands. Also the deletion of coronal plosives in final clusters in St Kitts falls into this category. Finally, Negerhollands offers a direct parallel to the process observed by Shinohara, systematically retaining word-final nasal-fricative clusters, but variably restructuring nasal-plosive clus-

8.4 Testing SLA approaches to creolisation

| 303

ters. Especially for the latter cases of cluster simplification, adaptation in production appears more plausible than adaptation in perception. The same clusters which lose their final plosive in Negerhollands are repaired by vowel paragoge in Berbice Dutch. The formerly final plosive thus surfaces intact in a medial nasalplosive sequence in Berbice Dutch, which means that the plosive must have been perceived by substrate speakers. The fact that access to the superstrate was actually more limited for speakers of Berbice Dutch than for speakers of Negerhollands renders a perceptual origin of the deletions in Negerhollands highly unlikely. As the above examples show, vowel epenthesis is clearly not the only process compatible with the assumption that creole phonotactic restructuring is to a large extent motivated by an initial transfer of L1 grammatical restrictions and processing patterns and subsequent incomplete L2 acquisition. Also interesting for an SLA connection is the observation that processes which are likely to originate in perception, such as vowel epenthesis or metathesis, are particularly pervasive in creoles with limited superstrate contact. Those creoles which had better access to the superstrate do not show the same predominance of potentially perception-based repairs. The case of word-final clusters in the Dutchbased creole related above is particularly interesting here. What we might have expected to find is that creoles with more intensive contact to the superstrate show basically the same restructuring choices as creoles with more limited superstrate contact, albeit with relatively lower application rates. In the case of final clusters, however, we find that Negerhollands not only shows less restructuring overall, but it also differs from Berbice Dutch in the preferred choice of restructuring mechanism. Berbice Dutch employs vowel paragoge, whereas Negerhollands employs consonant deletion for variable simplifications. Given that the substrate grammars were reasonably similar in both cases, such a difference in repair cannot be explained as a transfer effect. It can, however, plausibly be accounted for as a difference between adaptation in perception and adaptation in production. As perception is assumed to precede production at least in untutored SLA (cf. Major 2008), we would expect intermediate stages to occur at which perceptual procedures have already been reorganised, but at which constraints at work in production may still impose non-targetlike restrictions on the output. The situation in Negerhollands might be just such a case. Although the above discussion has been far from exhaustive, it should suffice here as a demonstration that many of the restructuring processes attested in the creole data can receive a plausible explanation under second language acquisition approaches to creolisation such as the ones suggested by Plag (2009) and Uffmann (2009). Since the two creolisation hypotheses contained no predictions regarding the choice between different adaptation processes, it could only

304 | 8 Explaining creole phonotactic restructuring be shown here that the attested processes are in line with the general assumptions concerning mechanisms in creolisation.

8.4.10 Summary In the above sections it was tested in how far the predictions made by Plag and Uffmann are born out by the empirical findings for the six creoles investigated in this study. We found support for the two creolisation hypotheses at the level of restriction constellations, where the vast majority of creole outcomes proved to be consistent with either transfer of substrate grammatical restrictions or (partial or complete) acquisition of superstrate structures. Also cross-creole differences in the degree of restructuring were in accordance with the predictions. Creoles which had only very limited contact with their respective superstrates showed a comparatively smaller L2 acquisition component than creoles with better access to the superstrate. At a less abstract level, it was shown that the general cross-creole tendencies described in chapter 7 are in line with the assumption that L1 transfer and L2 acquisition are the main mechanisms in creolisation. Cases in which the creoles show similar preferences as the substrates were interpreted as signs of L1 transfer. Regarding an L2 acquisition component, we saw that creole syllable structure shows evidence of both markedness and salience effects in much the same way as L2 phonological systems do. Finally, numerous parallels between restructuring mechanisms in creoles and loanword adaptations were pointed out. These parallels support the hypothesis that phonotactic restructuring takes place in creoles because an L2 input is (at least initially) subjected to L1 grammatical restrictions. As outlined above, there is reason to believe that at least some of the observed adaptations originate in perception rather than in production. Many of the specifics are still unclear, however, and much research remains to be done on the interaction of perception and production. In the next section I will briefly discuss the consequences of the above findings for theoretical phonological models of the emergence of creole syllable structure. I will outline the requirements that such a theoretical model would have to fulfil and comment on previous approaches to modelling creole syllables in phonological theory.

8.5 Creole syllable structure in phonological theory |

305

8.5 Creole syllable structure in phonological theory: previous approaches and future challenges Based on the above conclusions concerning mechanisms in creolisation and the motivation for repair choices, a theoretical account of the development of creole syllable structure would have to make use of a fairly complex grammatical model. It was argued in the previous section that restructuring of some lexifier input in creoles may result from a variety of factors including the interpretation of an L2 auditory form based on L1 perceptual processing patterns as well as the application of L1 grammatical restrictions. A theoretical analysis of the emergence of creole syllable structure should therefore take into account not just speech production, but also perception. In addition, interactions between restrictions at the syllabic level and constraints on other types of structures may need to be incorporated. Finally, the model needs to allow for changes in the grammar that reflect partial L2 acquisition and result in a gradual approximation to the superstrate system. Not many theoretical accounts exist that address the development of creole syllables. Most formal analyses focus on one particular stage of a given creole language and do not discuss the relation between the creole grammar and the grammars of substrate and superstrate, respectively. One study which proposes such relations is Brousseau and Nikiema (2006). The authors take a theoretical approach to consonant sequences in Haitian Creole and its main contributing languages, French and Gbe, using the framework of Government Phonology to account for the observed structures in the three languages. They then propose a scenario that is based on the Full Transfer/Full Access Mode Hypothesis (Schwartz and Sprouse 1996) and relates the different grammars via resetting of parameters of Universal Grammar. Brousseau and Nikiema take all adaptations to occur in production. A potential perceptual origin of phonotactic restructuring is not discussed, nor are interactions between restrictions on, for instance, the syllabic and the word level. It thus remains unclear how (or whether) these aspects could be formalised under a government phonological approach. The problem that perceptual adaptations are not considered also applies to analyses in a strictly production-oriented optimality theoretic framework (e.g. Alber and Plag 2001, Uffmann 2008). The interaction between restrictions on different types of structures, on the other hand, can easily be modelled in this framework. In OT, all grammatical restrictions are represented in terms of a set of violable ranked constraints. Not the mere presence of a constraint in the grammar, but its ranking relative to other (faithfulness and markedness) constraints, determines the output of the production grammar. In OT accounts, differences between

306 | 8 Explaining creole phonotactic restructuring substrate, creole and superstrate syllable structure result from differences in this relative ranking. Second language acquisition and partial approximation to the superstrate can be formalised in terms of constraint reranking. In recent years, some researchers have attempted to model not only production, but also perception in OT (see Boersma and Hamann 2009 and Russell Webb 2010 for proposals regarding adaptations in loanwords and creoles, respectively). In these proposals, perception is assumed to involve an interaction between structural constraints and constraints that evaluate the correspondence between an auditory input and a phonological structure to which that input is mapped. The relative ranking of constraints determines which mappings are possible and which ones are not. Like reorganisation of the production grammar in traditional OT, also an adaptation of perceptual processing patterns can be modelled as a reranking of individual constraints. Approaches of the latter type, allowing for restructuring to take place in both perception and production, seem the most promising for an account of creole syllable structure. On the downside, a theoretical analysis along these lines cannot be applied to early creole data. In order to posit cue constraints of the type used by Boersma and Hamann (2009), we would need detailed phonetic evidence for the superstrate input as well as for substrate and creole forms. Obviously, such evidence is not available for non-contemporary data. As far as the emergence of creole syllables is concerned, we might thus have to rely on comparisons with loanword adaptation or restructuring in interlanguages, for which also the influence of perception can be investigated in detail.

8.6 Summary In this chapter, I have discussed the question of how we can explain the syllable structures and patterns of phonotactic restructuring found in the six creoles under investigation. To this end, two current hypotheses about the emergence of creole phonology were evaluated in the light of the empirical findings presented in chapters 4 through 6. It was shown that the predictions made by Plag (2009) and Uffmann (2009) with respect to the choice between retention and repair of some lexifier structure in the creoles are largely correct. Repair mechanisms typically only apply if the substrates impose tighter constraints on some structure than the superstrate. This is in line with the assumption of initial full transfer of L1 (i.e. substrate) grammatical restrictions. Variable restructuring or systematic retention of superstrate structures, on the other hand, can be interpreted as reflecting partial or successful L2 (i.e. superstrate) acquisition. Only a small number of structural categories showed

8.6 Summary |

307

creole outcomes that were not compatible with either a transfer or an L2 acquisition interpretation. Some of these could be accounted for under the assumption that grammatical restructuring takes place in language transmission from one creole generation to the next. Others were truly unexpected. A closer look at the pertinent data, however, indicated that most of the unpredicted repairs were unlikely to be motivated primarily in constraints on syllable structure. Several potentially confounding factors could be identified. Since these factors were not systematically investigated, no definite conclusion as to the source of the unexpected patterns in the creoles could be reached. Given this situation, their occurrence was not counted as evidence against the two creolisation hypotheses. Also general tendencies holding across creoles were discussed briefly in relation to the hypotheses and interpretations in terms of either L1 transfer or L2 acquisition were suggested for the different patterns. At the level of concrete restructuring mechanisms, it was shown that numerous parallels exist between creole phonotactic restructuring and loanword adaptations, which further supports the hypothesis that creoles start out with a (levelled) substrate sound system. Based on comparable suggestions for loanwords and findings from studies of nonnative perception, I have argued that at least some processes of phonotactic restructuring in the creoles are likely to originate in perception rather than production. As far as cross-creole differences are concerned, it was demonstrated that creoles which developed in closer contact with their respective superstrate show a greater approximation to superstrate structures and allow a greater range of marked syllable structures than creoles which only had limited access to the lexifier. These findings lend additional support to the assumption of an L2 acquisition component in creolisation. They also provide evidence for the claim that the relative impact of different factors on the creole outcome is determined by the social conditions under which the creoles developed. I have shown in particular that the relative influence of L1 transfer and L2 acquisition may vary across different contact settings. Some aspects relating to the hypotheses could not be discussed exhaustively. In particular, I only briefly touched upon the question of whether cross-creole differences in restrictions on particular subsets of a given structure can be accounted for under the tested hypotheses. However, it was argued that at least the individual restrictions themselves are likely to receive an explanation in terms of SLA. More specifically, the observation that perceptual salience and/or markedness appear to play a role in many cases is in line with the L2 acquisition hypothesis. With regard to perceptual salience, we have seen that the creoles show the same type of effects that would also be expected to occur in language acquisition, with low perceptual salience of some structure making deletion in the creole more likely.

308 | 8 Explaining creole phonotactic restructuring Also markedness shows similar effects in creoles and L2 acquisition. In particular, singleton coda consonants are more likely to be retained in the creoles than coda clusters, a relation which finds a parallel in the acquisition of shorter syllable constituents before longer ones in SLA. Only some of the attested language-internal differences were discussed in detail. Others went undiscussed. While I trust that the majority of such differences can be accounted for by the mechanisms discussed above, it is entirely possible that individual cases occur for which a straightforward explanation is not so readily available. Not only markedness and perceptual salience would have to be taken into account. As indicated by the discussion of vowel-initial words in section 8.4.7 above, it may not be sufficient in all cases to look at syllable structure alone. Co-occurrence restrictions on particular consonants and vowels, allophonic variation in the substrates, word-level constraints on prosodic structure as well as morphological reanalysis of lexifier forms on the basis of substrate regularities are all factors which might potentially interact with restrictions at the syllabic level. All of the above factors would have to be incorporated in a theoretical phonological account of the emergence of creole syllable structure. As far as a perceptual component in phonotactic restructuring is concerned, such an account may have to draw on findings from contemporary language varieties.

9 Creole syllable structure: A final assessment In this book, I have presented a detailed empirical study of syllable structure and phonotactic restructuring in six Caribbean creoles with Dutch, English and French as main lexifiers. A comparison of the results for the individual creoles was conducted to determine the extent of cross-creole variation. It was then tested whether current claims about the emergence of creole phonology can account for the observed patterns. It was shown in chapters 4 through 7 that creole syllable structure is much more variable than has often been assumed in the past. Although some general tendencies can be observed, we find considerable variation even across creoles with similar contributing languages. Cross-creole differences in general may concern – – – –

the degree of syllable complexity, the probability of restructuring for a given structural category, restrictions on the segmental content in different syllable positions and the preferred choice among different restructuring mechanisms.

As far as syllable complexity is concerned, cross-creole differences are relatively limited, but not to the effect that only CV syllables are allowed. Instead, the results of the empirical investigation make it clear that a description of creole syllable structure cannot be reduced to a comment on its unmarked status. None of the six creoles allows less than six different syllable types, including CV, but also V, VC, CVC, CCV and CCVC. Negerhollands and Early St Kitts also permit complex codas, thus adding VCC, CVCC and CCVCC syllables to their inventories. Even trisegmental onset are possible in some creoles, at least in word-initial position. However, the mere attestedness of a given syllable type does not imply that structures of this type are never targeted by repairs in the respective creole. To the contrary, we have seen that many restrictions on syllable structure are not categorical. Instead, the probability of restructuring may vary significantly across different categories, reflecting differences in the relative permissibility of those structures. Two general statements can be made in this respect. First, retention rates are overall higher for syllable onsets than for syllable codas and second, complex codas are the most dispreferred type of syllable structure in all six creoles. If we take not only syllable complexity, but also constraints on segmental content into account, then stating parallels between the creoles becomes more difficult. Generalisations can only be expressed in terms of tendencies, i.e. frequent patterns which hold for several, though not necessarily all, creoles. The most commonly permitted structures are obstruent-approximant onsets and simple nasal

310 | 9 Creole syllable structure: A final assessment codas. The most frequent targets of restructuring include initial [s]-plosive clusters, onsetless structures in words of three or more syllables, coda rhotics and complex codas. Apart from these general tendencies, however, we also find a number of distinctions between subsets that apply only in individual creoles without any obvious parallels in the remaining languages. Creole syllables are thus neither maximally unmarked nor particularly uniform. Neither are they identical to the syllable structure of the substrate or superstrate languages. Each of the creoles investigated here allows also structures which are not common to their respective substrates. On the other hand, none of the creoles retain the whole range of lexifier structures, i.e. we find evidence of phonotactic restructuring even in the most permissive of creoles. Consequently, in terms of syllable structure, the investigated creoles are typically more permissive than their substrate languages, but more restrictive than their superstrates. Exceptions from this rule exist, however. In a small number of cases, a structure which is permitted in the substrates is nonetheless restructured in the creole. It was shown in the final part of this book (chapter 8) that many of the observed patterns can receive a plausible explanation if we assume that initial full L1 transfer, substrate levelling and processes of L2 acquisition feature prominently among the mechanisms in creolisation (cf. Plag 2009, Uffmann 2009). We found support for such an assumption also in the fact that the degree of restructuring varies across creoles depending on the intensity of contact between substrate and superstrate speakers in the respective contact setting. In particular, creoles which had only limited access to the superstrate impose tighter restrictions on syllable structure. Under the above approach, these differences in the degree of restructuring can be interpreted as differences in the degree of superstrate acquisition in the sense that more intensive or prolonged contact leads to the acquisition of more (and more complex) superstrate structures. Limited access to the superstrate, on the other hand, also limits acquisition and thus causes creole speakers to retain more L1 grammatical restrictions. The adaptation of structures that are permitted in the substrates warrants a different explanation. If the targeted structures did not occur in superstrate lexical items, such cases were interpreted as resulting from grammatical restructuring in language transmission (cf. Uffmann 2009). Repairs of structures which are possible in both substrate and superstrate are the only cases which present a potential problem for the tested hypotheses. It was shown, however, that even such unpredicted patterns may find an explanation if aspects other than syllable structure are taken into account. On a fairly general level, the findings on creole syllable structure were thus shown to be consistent with the predictions made by Plag (2009) and Uffmann (2009). The same was done for concrete cross-creole tendencies regarding the

9 Creole syllable structure: A final assessment

| 311

relative acceptability of different types of structures. What could not be investigated systematically in this study is whether all creole-particular distinctions between subsets of a given category can be accounted for equally well. Only some aspects relating to this issue were discussed. Markedness was argued to be a relevant factor here, whose effects can be observed, for instance, in the relatively greater acceptability, i.e. earlier acquisition, of simple codas in contrast to complex codas. Similarly, perceptual salience was shown to influence the probability of retention of some structures in the creole. All else being equal, perceptually salient elements were more likely to be retained, i.e. more easily acquired, than perceptually weak elements. Linguistic universals were thus taken to affect not so much the starting point of the creole phonological system, but rather the order of acquisition of superstrate structures. As far as variation in terms of restructuring mechanisms is concerned, we found a similarly diverse situation as for the degree of restructuring in general. Vowel epenthesis and consonant deletion are the two most common restructuring options overall, but they are not evenly distributed across either creoles or structural categories. Cross-creole differences exist in particular with respect to the preferred choice of restructuring mechanism. Early Saramaccan and Berbice Dutch show a predominant use of vowel epenthesis, particularly paragoge. In the other creoles, consonant deletion is more common. It is not the case, however, that a given creole employs only a single default strategy to repair all illicit structures. Instead, restructuring choices may vary across different types of structures. Based on their similarity to loanword adaptations, I have argued that also processes of phonotactic restructuring in creoles can be understood as minimal transformations of some nonnative input. It was assumed that such transformations can originate in both perception and production. Under this assumption, the choice of repair strategy depends on and can vary according to – the auditory characteristics of the input, – L1 perceptual processing patterns and – L1 grammatical restrictions. Following relevant suggestions in the literature, I have proposed that many cases of vowel epenthesis, but also processes such as liquid-vowel metathesis and certain segmental substitutions are likely to occur during speech perception. For cases of consonant deletion, on the other hand, a perceptual explanation is not always convincing. Many of the observed deletions thus may be more likely to originate in speech production. Clear evidence for the origin of individual processes is not always available, however, and opinions on this issue are divided in the loanword adaptation literature. Future research in this area is clearly needed and will hopefully also further our understanding of repair choices in creole phono-

312 | 9 Creole syllable structure: A final assessment tactic restructuring. Studies involving the same language constellations as in the creole contexts would be of particular interest here. For instance, an investigation of the production and perception of English structures by native speakers of Gbe, Kikongo and Twi could serve as a basis for comparison with the findings for Saramaccan. Such studies could help determine more clearly whether individual restructuring mechanisms originate in perception or production. They would be interesting also because they might reveal differences between groups of speakers with different L1s, which would be relevant to the discussion of levelling effects in creoles. Although not the primary focus of the present study, my findings have implications also for the typology of creole syllables. In particular, the interpretation of creole syllable structure as being shaped mainly by L1 transfer and L2 acquisition makes important typological predictions. For one thing, it predicts that creoles should not restrict their inventory to only unmarked syllables, unless either their substrates or their superstrates impose such restrictions. Even if the substrates allow only CV syllables, partial acquisition of the superstrate would most likely lead to more complex structures in the creole. These expectations are confirmed by the results of Klein’s (2011) typological study of creole phonology. Not a single creole in his set of 32 languages showed only CV syllables. At the other end of the markedness scale, in contrast, there may be no absolute cut-off point. We have seen in the case of Negerhollands that even highly complex syllable structures can occur in creoles. Even though not all structures are categorically retained, in terms of syllable templates, Negerhollands is comparable to its superstrate Dutch, both of them allowing (s)(C)(C)V(C)(C) structures.¹ In principle, then, creoles can exhibit the same degree of complexity as their lexifier languages. Given the SLA interpretation, however, we would not expect this to be typical in cases in which the substrate is considerably more restrictive than the superstrate. Rather, we would expect most creoles to occupy some middle ground between substrate and superstrate in such cases. For creoles whose contributing languages are more similar in terms of syllable structure, a comparable degree of complexity in creole and superstrate would be more likely. Overall, however, syllable structures at the upper end of the complexity scale should be retained only in a subset of those creoles whose input contains them. Consequently, unless languages with highly complex syllable structures are massively overrepresented among the superstrates of existing creoles, such structures should be less frequent in creoles than in non-creole languages. This latter point is consistent with Com-

1 Recall, however, that the theoretically possible sCCVCC does not occur in the set of etyma for Negerhollands.

9 Creole syllable structure: A final assessment

| 313

rie’s (2011) preliminary typological findings on complex syllable structures in creole and non-creole languages.² Also Klein’s (2011: 183) conclusion that “creoles cluster in the higher end of the typological middle” is well in accordance with the above predictions concerning structures at both the upper and the lower end of the markedness scale. In order to thoroughly test these predictions, we would have to investigate not only the typological features of creoles, but also those of their contributing languages. However, the typological findings available so far and the results of the present study support the following three generalisations about creole syllable structure. First, creoles fall in a range from relatively unmarked to highly marked systems, although the latter are comparatively infrequent. Second, the majority of creoles allow syllables with at least a medium degree of complexity. Finally, and crucially, although unmarked syllable types are common in creoles, a creole system that allows only maximally unmarked syllables is highly unexpected. Thus, somewhat ironically, strict CV syllable structure, which was long believed to be typical of creole languages, may turn out to be one of the least likely outcomes of creolisation.

2 His conclusion that simplification may be the driving force in the creation of creole phonology, on the other hand, is contradicted by the non-attestedness of pure CV systems.

Bibliography Aceto, Michael (1996). Early Saramaccan syllable structure: An analysis of complex onsets from Schumann’s 1778 manuscript. In: Journal of Pidgin and Creole Languages 11, 23–44. Alber, Birgit and Ingo Plag (2001). Epenthesis, deletion and the emergence of the optimal syllable in creole. In: Lingua 111, 811–840. Alleyne, Mervyn C. (1980). Comparative Afro-American: An historical-comparative study of English-based Afro-American dialects of the New World. Ann Arbor: Karoma. Andersen, Janet I. (1987). The markedness differential hypothesis and syllable structure difficulty. In: Ioup, Georgette and Steven Weinberger (eds.), 279–291. Anderson, Roger W. (1978). An implicational model for second language research. In: Language Learning 28, 221–281. — (1983a). Pidginization and creolization as language acquisition. Rowley, MA: Newbury House. — (1983b). Transfer to somewhere. In: Gass, Susan M. and Larry Selinker (eds.), Language transfer in language learning. Rowley, MA: Newbury House, 177–201. Andresen, Jutta, Jette Mez and Eckhard Bodenstein (eds.) (1995). Langenscheidts Taschenwörterbuch Dänisch. Berlin: Langenscheidt. Arends, Jacques (1995). Introduction. In: Arends, Jacques and Matthias Perl (eds.), Early Suriname Creole texts: A collection of early Sranan and Saramaccan documents. Frankfurt/Madrid: Vervuert, 11–71. — (2002). The history of the Surinamese creoles I. In: Carlin, Eithne B. and Jacques Arends (eds.), Atlas of the languages of Suriname. Leiden: KITLV Press, 115–130. Aub-Büscher, Gertrud (1969). Introduction to The theory and practice of creole grammar by John J. Thomas. London/Port of Spain: New Beacon Books. Baker, Philip (1999). Investigating the origin and diffusion of shared features among the Atlantic English creoles. In: Baker, Philip and Adrienne Bruyn (eds.), 315–364. Baker, Philip and Adrienne Bruyn (eds.) (1999). St Kitts and the Atlantic Creoles. The late 18th century texts of Samuel Augustus Mathews in perspective. London: University of Westminster Press. Baker, Philip, Adrienne Bruyn, Neville Shrimpton and Lise Winer (1999). The texts of Samuel Augustus Mathews arranged and annotated by Philip Baker, Adrienne Bruyn, Neville Shrimpton and Lise Winer with a grammatical commentary by Alain Kihm. In: Baker, Philip and Adrienne Bruyn (eds.), 5–47. Bakker, Peter (2009). The Saramaccan lexicon: Verbs. In: Selbach, Rachel, Hugo C. Cardoso and Margot van den Berg (eds.), Gradual creolisation: Studies celebrating Jacques Arends. Amsterdam/Philadelphia: John Benjamins, 156–172. Beckman, Mary E. (1996). When is a syllable not a syllable? In: Otake, Takashi and Anne Cutler (eds.), Phonological structure and language processing. Berlin/New York: Mouton de Gruyter, 95–123. Bentley, William Holman (1887). Dictionary and grammar of the Kongo language as spoken at San Salvador, the ancient capital of the old Kongo empire, West Africa. London: Baptist Missionary Society. Republished in 1967 by Gregg, London. Bhatt, Parth (2007). If creoles are simple, why is Haitian Creole so complex? The case of syllable structure. Paper presented at the University of Siegen, Germany. Bickerton, Derek (1981). Roots of language. Ann Arbor: Karoma.

316 | Bibliography — (1984). The Language Bioprogram Hypothesis. In: The Behavioral and Brain Sciences 7, 173–221. Blevins, Juliette (1995). The syllable in phonological theory. In: Goldsmith, John A. (ed.), Handbook of phonological theory. Oxford: Blackwell, 206–244. Blevins, Juliette and Andrew Garrett (1998). The origins of consonant-vowel metathesis. In: Language 74, 508–556. Boersma, Paul and Silke Hamann (2009). Loanword adaptation as first-language phonological perception. In: Calabrese, Andrea and W. Leo Wetzels (eds.), Loan phonology. Amsterdam/Philadelphia: John Benjamins, 11–58. Booij, Geert (1995). The phonology of Dutch. Oxford: Oxford University Press. Boretzky, Norbert (1983). Kreolsprachen, Substrate und Sprachwandel. Wiesbaden: Otto Harrassowitz. Bosch, Anna R. K. (2011). Syllable-internal structure. vol. 1, In: van Oostendorp, Marc, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.), The Blackwell companion to phonology. Oxford: Wiley-Blackwell, 781–798. Braun, Bettina, Kristin Lemhöfer and Nivedita Mani (2011). Perceiving unstressed vowels in foreign-accented English. In: Journal of the Acoustical Society of America 129(1), 376–387. Braun, Maria (2009). Word-formation and creolisation: The case of Early Sranan. Tübingen: Niemeyer. Brousseau, Anne-Marie (2005). The sociolect of 17th-18th century French settlers: Phonological clues from French Creoles. In: Selected Papers from New Ways of Analyzing Variation (NWAVE) 32. Philadephia: Penn Linguistics Club, 45–60. — (2007). Le maintien du h aspiré au 18e siècle: L’éclairage des créoles français. In: Combettes, Bernard and Christiane Marchello-Nizia (eds.), Études sur le changement linguistique en français. Nancy: Presses Universitaires de Nancy, 65–85. — (2011). One substrate, two creoles: The development of segmental inventories in St. Lucian and Haitian. In: Lefebvre, Claire (ed.), Creoles, their substrates, and language typology. Amsterdam/Philadelphia: John Benjamins, 95–125. Manuscript. Brousseau, Anne-Marie and Emmanuel Nikiema (2001). Phonologie et morphologie du français. Saint-Laurent, Québec: Fides. — (2006). From Gbe to Haitian: The multi-stage evolution of syllable structure. In: Lefebvre, Claire, Lydia White and Christine Jourdan (eds.), 295–330. Cancino, Herlinda, Ellen J. Rosansky and John H. Schumann (1975). The acquisition of the English auxiliary by native Spanish speakers. In: TESOL Quarterly 4, 421–430. Capo, Hounkpati B. C. (1991). A comparative phonology of Gbe. Berlin/New York: Foris Publications. Carlisle, Robert S. (2001). Syllable structure universals and second language acquisition. In: International Journal of English Studies 1(1), 1–19. Christaller, Johann Gottlieb (1875). A grammar of the Asante and Fante languages called Tshi. Basel: Evangelical Missionary Society. Republished 1964 by Gregg, Ridgewood. Clements, George N. (1990). The role of the sonority cycle in core syllabification. In: Kingston, John and Mary E. Beckman (eds.), Papers in laboratory phonology 1: Between the grammar and physics of speech. New York: Cambridge University Press, 283–333. Clements, G. N. and Samuel J. Keyser (1983). CV phonology: A generative theory of the syllable. Cambridge, MA: MIT Press. Comrie, Bernard (2011). Creoles and language typology. In: Lefebvre, Claire (ed.), 599–611.

Bibliography

|

317

Cooper, Vincent O. (1999). St Kitts: The launching pad for Leeward Island Creole. In: Baker, Philip and Adrienne Bruyn (eds.), 379–386. de Josselin de Jong, Jan Petrus Benjamin (1924). Het Negerhollands van St. Thomas en St. Jan. In: Mededeelingen der koninklijke Akademie van Wetenschapen, Afdeeling Letterkunde 57, 55–71. — (1925). Het huidige Negerhollandsch. Teksten en woordenlijst. In: Verhandelingen der koninklijke Akademie van Wetenschapen te Amsterdam, Afdeeling Letterkunde 26, 1–107. Delafosse, Maurice (1929). La langue mandingue et ses dialectes. Paris: Librairie Orientaliste Paul Geuthner. Delattre, Pierre (1969). An acoustic and articulatory study of vowel reduction in four languages. In: International Review of Applied Linguistics and Language Teaching (IRAL) 7, 294–325. den Besten, Hans (1987). Review article: Einführung in die Geschichte und Struktur des Afrikaans by Edith Raith. In: Journal of Pidgin and Creole Languages 2(1), 67–92. Deslandes, H. (1781). A new introduction to the pronunciation of the French tongue. Farmington Hills: Thomson Gale. Dobson, Eric John (1968). English pronunciation 1500-1700. Oxford: Oxford University Press. Duanmu, San (2009). Syllable structure: The limits of variation. Oxford: Oxford University Press. Dupoux, Emmanuel, Yuki Hirose, Kazuhiko Kakehi, Christophe Pallier and Jacques Mehler (1999). Epenthetic vowels in Japanese: A perceptual illusion? In: Journal of Experimental Psychology: Human Perception and Performance 25(6), 1568–1578. Dupoux, Emmanuel, Erika Parlato, Sonia Frota, Yuki Hirose and Sharon Peperkamp (2011). Where do illusory vowels come from? In: Journal of Memory and Language 64, 199–210. Eckman, Fred R. (1991). The structural conformity hypothesis and the acquisition of consonant clusters in the interlanguage of ESL learners. In: Studies in Second Language Acquisition 13, 23–41. Eckman, Fred R. and Gregory K. Iverson (1993). Sonority and markedness among onset clusters in the interlanguage of ESL learners. In: Second Language Research 9(3), 234–252. Ellis, Alfred Burdon (1894). The Yoruba-speaking peoples of the slave coast of West Africa. Their religion, manners, customs, laws, language, etc. with an appendix containing a comparison of the Tshi, Gã, Ewe and Yoruba languages. London: Curzon Press, 2nd ed. Reprinted in 1974. Fleischhacker, Heidi (2001). Cluster-dependent epenthesis asymmetries. In: UCLA Working Papers in Linguistics 7, 71–116. — (2005). Similarity in phonology: Evidence from reduplication and loan adaptation. Ph.D. thesis, University of California Los Angeles. Foley, James (1972). Rule precursors and phonological change by meta-rule. In: Stockwell, Robert P. and Ronald K. S. Macaulay (eds.), Linguistic change and generative theory. Bloomington: Indiana University Press, 96–100. Fox, John and Robin Wood (1971). A concise history of the French language: Phonology and morphology. Oxford: Blackwell, 3rd ed. Franken, Johan Lambertus Machiel (1953). Taalhistoriese bydraes. Cape Town: A. A. Balkema. Fudge, Erik C. (1969). Syllables. In: Journal of Linguistics 5, 253–286. Gelderen, Icarus van (1993). Wolters Handwörterbuch Deutsch-Niederländisch / Niederländisch-Deutsch. Berlin: Langenscheidt. Revised by Wilhelmus H. Wallis. Gimson, Alfred Charles and Alan Cruttenden (1994). Gimson’s pronunciation of English. London: Arnold, 5 ed. Revised by Alan Cruttenden.

318 | Bibliography Good, Jeff (2009a). A twice-mixed creole? Tracing the history of a prosodic split in the Saramaccan lexicon. In: Studies in Language 33(2), 459–498. — (2009b). Loanwords in Saramaccan, an English-based Atlantic creole of Suriname. In: Haspelmath, Martin and Uri Tadmoor (eds.), Loanwords in the world’s languages: A comparative handbook. Berlin: Mouton de Gruyter, 918–943. Goossens, Jan (1974). Historische Phonologie des Niederländischen. Tübingen: Niemeyer. Hald, Katja and Marieluise Schmitz (eds.) (2008). PONS Großwörterbuch Spanisch-Deutsch / Deutsch-Spanisch. Stuttgart: Ernst Klett Sprachen. Hallé, Pierre A. (2008). Language-specific aspects of the perception of illegal clusters. Paper presented at the workshop ’Consonant clusters and structural complexity’, 31 July - 2 August 2008, Munich, Germany. Hammond, Michael (1999). The phonology of English: A prosodic optimality-theoretic approach. Oxford: Oxford University Press. Hancin-Bhatt, Barbara (2008). Second language phonology in optimality theory. In: Hansen Edwards, Jette G. and Mary L. Zampini (eds.), 117–146. Hansen Edwards, Jette G. and Mary L. Zampini (eds.) (2008). Phonology and second language acquisition. Amsterdam/Philadelphia: John Benjamins. Harry, Otelemate (2004). Aspects of the tonal system of Kalabari-Ijo. Stanford, CA: Center for ˙ ˙ ˙ ˙ the Study of Language and Information. Haspelmath, Martin (2010). Comparative concepts and descriptive categories. In: Language 86(3), 663–687. Hockett, Charles F. (1955). A manual of phonology. Baltimore: Waverly Press. Holm, John (1988). Pidgins and creoles. Vol 1: Theory and structure. Cambridge: Cambridge University Press. — (2000). An introduction to pidgins and creoles. Cambridge: Cambridge University Press. Ioup, Georgette and Steven Weinberger (eds.) (1987). Interlanguage phonology: The acquisition of a second language sound system. Rowley, MA: Newbury House Publishers. Irmen, Friedrich and Ana Maria Cortés Kollert (eds.) (1995). Langenscheidts Taschenwörterbuch Portugiesisch. Berlin: Langenscheidt. Jennings, William (1995). The first generations of a creole society: Cayenne 1660-1700. In: Baker, Philip (ed.), From contact to creole and beyond. London: University of Westminster Press, 21–40. — (2009). Demographic factors in the formation of French Guianese Creole. In: Selbach, Rachel, Hugo C. Cardoso and Margot van den Berg (eds.), Gradual creolisation: Studies celebrating Jacques Arends. Amsterdam/Philadelphia: John Benjamins, 373. Jones, Daniel (1950). The pronunciation of English. Cambridge: Cambridge University Press, 3 ed. Kabak, Baris (2003). Perceptual processing of second language consonant clusters. Ph.D. thesis, University of Delaware. Kahn, Daniel (1980). Syllable-based generaligeneral in English phonology. New York: Garland. Kang, Yoonjung (2003). Perceptual similarity in loanword adaptation: English postvocalic word-final stops in Korean. In: Phonology 20, 219–273. — (2011). Loanword phonology. In: van Oostendorp, Marc, Colin J. Ewen, Elizabeth Hume and Keren Rice (eds.), The Blackwell companion to phonology. Oxford & Malden, MA: WileyBlackwell. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud (1990). Constituent structure and government in phonology. In: Phonology 7(2), 193–231.

Bibliography

|

319

Kenstowicz, Michael and Atiwong Suchato (2006). Issues in loanword adaptation: A case study from Thai. In: Lingua 116(7), 921–949. Klein, Thomas B. (2011). Typology of creole phonology: Phoneme inventories and syllable templates. In: Journal of Pidgin and Creole Languages 26(1), 155–193. Kouwenberg, Silvia (1994). A grammar of Berbice Dutch Creole. Berlin/New York: Mouton de Gruyter. — (2012). The Ijo-derived lexicon of Berbice Dutch Creole: An a-typical case of African lexical influence. In: Bartens, Angela and Philip Baker (eds.), Black through white? OR African words and calques in creoles and transplanted European languages. London: Battlebridge. Kouwenberg, Silvia and John Victor Singler (2008a). Introduction. In: Kouwenberg, Silvia and John Victor Singler (eds.), 1–16. Kouwenberg, Silvia and John Victor Singler (eds.) (2008b). The handbook of pidgin and creole studies. Malden: Wiley-Blackwell. Laman, Karl Edvard (1912). Grammar of the Kongo Language (Kikongo). New York: Christian Alliance Publ. Co. Laman, Karl Edvard and Carl Meinhof (1928). An essay in Kongo phonology. In: Zeitschrift für Eingeborenen-Sprachen 19, 12–40. Laver, John (1994). Principles of phonetics. Cambridge: Cambridge University Press. Lefebvre, Claire (1998). Creole genesis and the acquisition of grammar: The case of Haitian Creole. Cambridge: Cambridge University Press. Lefebvre, Claire (ed.) (2011). Creoles, their substrates, and language typology. Amsterdam/Philadelphia: John Benjamins. Lefebvre, Claire and Anne-Marie Brousseau (2002). A grammar of Fongbe. Berlin/New York: Mouton de Gruyter. Lefebvre, Claire, Lydia White and Christine Jourdan (eds.) (2006). L2 acquisition and creole genesis: Dialogues. Amsterdam/Philadelphia: John Benjamins. Levin, Juliette (1985). A metrical theory of syllabicity. Ph.D. thesis, Massachusetts Institute of Technology. Lewis, M. Paul (ed.) (2009). Ethnologue: Languages of the World. Dallas, TX: SIL International, 16 ed. Online version: http://www.ethnologue.com/. Lipski, John (2000). Epenthesis vs. elision in Afro-Iberian language: A constraint-based approach to creole phonology. In: Papia 10, 23–39. Lipski, John M. (2008). Spanish-based creoles in the Caribbean. In: Kouwenberg, Silvia and John Victor Singler (eds.), 543–564. Major, Roy C. (2008). Transfer in second language phonology: A review. In: Hansen Edwards, Jette G. and Mary L. Zampini (eds.), 63–94. McCarthy, John J. (1979). On stress and syllabification. In: Linguistic Inquiry 10, 443–465. McCarthy, John J. and Alan S. Prince (1994). The emergence of the unmarked: Optimality in prosodic morphology. In: Papers from the Annual Meeting of the North East Linguistic Society 24, 333–379. — (1996). Prosodic morphology 1986. Technical Report #32, Rutgers University Center for Cognitive Science. McWhorter, John (2004). Saramaccan and Haitian as young grammars: The pitfalls of syntactocentrism in creole genesis research. In: Journal of Pidgin and Creole Languages 19.1, 77–138.

320 | Bibliography Meade, Rocky R. (1995). An analysis of Jamaican /s/-stop cluster reduction within Optimality Theory. In: UWI working papers in linguistics 1. Mona: University of the West Indies, 31– 39. Migge, Bettina (1998). Substrate influence in the formation of the Surinamese plantation creole: A consideration of sociohistorical data and linguistic data from Ndyuka and Gbe. Ph.D. thesis, Ohio State University. — (2003). Creole formation as language contact: The case of the Surinamese creoles. Amsterdam: John Benjamins. Mühlhäusler, Peter (1986). Pidgin and creole linguistics. Oxford: Basil Blackwell. Müller, Friedrich (1876). Grundriss der Sprachwissenschaft. Wien: Alfred Hölder. Munro, Martin (2009). The French Creoles of Trinidad and the limits of the Francophone. In: French Studies 63(2), 174–188. Odden, David (2005). Introducing phonology. Cambridge: Cambridge University Press. OED Online, The Oxford English Dictionary (2012). Oxford University Press. www.oed.com. Paradis, Carole and Darlene Lacharité (1997). Preservation and minimality in loan word adaptation. In: Journal of Linguistics 33, 379–430. Parkvall, Mikael (1995). A dual approach to French Creole genesis. Master’s thesis, Stockholm University, Institutionen för Lingvistik. — (1999a). A short note on the peopling of English St Kitts. In: Baker, Philip and Adrienne Bruyn (eds.), 63–74. — (1999b). Feature selection and genetic relationships among Atlantic creoles. In: Huber, Magnus and Mikael Parkvall (eds.), Spreading the word: The issue of diffusion among the Atlantic creoles. London: University of Westminster Press, 29–66. — (2000). Out of Africa. African influences in Atlantic Creoles. London: Battlebridge Publications. Peperkamp, Sharon (2005). A psycholinguistic theory of loanword adaptation. In: Ettlinger, Marc, Nicholas Fleisher and Mischa Park-Doob (eds.), Proceedings of the Berkeley Linguistic Society 30. Berkeley, CA: Berkeley Linguistics Society, 341–352. Peperkamp, Sharon and Emmanuel Dupoux (2003). Reinterpreting loanword adaptations: The role of perception. In: International Congress of Phonetic Sciences 15, 367–370. Peperkamp, Sharon, Inga Vendelin and Kimihiro Nakamura (2008). On the perceptual origin of loanword adaptations: Experimental evidence from Japanese. In: Phonology 25, 129–164. Peterson, Gordon E. and Ilse Lehiste (1960). Duration of syllable nuclei in English. In: Journal of the Acoustical Society of America 32(6), 693–703. Pierret, Jean-Marie (1981). Phonétique du français: Notions de phonétique générale et phonétique historique du français. Louvain-la-Neuve: Cabay. Pike, Kenneth L. and Eunice V. Pike (1947). Immediate constituents of Mazateco syllables. In: International Journal of American Linguistics 13, 78–91. Plag, Ingo (1993). Sentential complementation in Sranan. Tübingen: Niemeyer. — (1999). Where did Sranan’s English base come from? Some phonological evidence from 18th-century St Kitts and Barbados texts. In: Baker, Philip and Adrienne Bruyn (eds.), 173–191. — (2009). Creoles as interlanguages: Phonology. In: Journal of Pidgin and Creole Languages 24(1), 119–138. Plag, Ingo, Gero Kunter and Mareile Schramm (2011). Acoustic correlates of primary and secondary stress in North American English. In: Journal of Phonetics 39, 362–374.

Bibliography

|

321

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: Bhatt, Parth and Ingo Plag (eds.), The Structure of creole words: Segmental, syllabic and morphological aspects. Tübingen: Niemeyer, 131–150. Plag, Ingo and Christian Uffmann (2000). Phonological restructuring in creole. The development of paragoge in Sranan. In: Neumann-Holzschuh and Edgar Schneider (eds.), Degrees of restructuring in creole languages. Amsterdam/Philadelphia: John Benjamins, 309–336. Price, Richard (1976). The Guiana maroons: A historical and bibliographical introduction. Baltimore: Johns Hopkins University Press. R Development Core Team (2007). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.org. Rey-Debove, Josette and Alain Rey (eds.) (2008). Le nouveau Petit Robert. Paris: Dictionnaires Le Robert, 41st ed. Rickard, Peter (1989). A history of the French language. London: Routledge, 2nd ed. Rietveld, Toni C. M. and Florina J. Koopmans-van Beinum (1987). Vowel reduction and stress. In: Speech Communication 6, 217–230. Robertson, Ian (1989). Berbice and Skepi Dutch: A lexical comparison. In: Tijdschrift voor Nederlandsche Taal- en Letterkunde 105, 3–21. — (1990). Some patterns of language use in the period of transition from Dutch to British rule in the Guiana colonies. In: International Journal of the Sociology of Language 85, 51–60. — (1994). Berbiciaansche woorde. In: Amsterdam Creole Studies 11, 67–74. Romaine, Suzanne (1988). Pidgin and creole languages. London: Longman. Rose, Yvan and Katherine Demuth (2006). Vowel epenthesis in loanword adaptation: Representational and phonetic considerations. In: Lingua 116(7), 1112–1139. Rotter, Jessica (2002). Die Silbe im Saramakkan. Master’s thesis, University of Siegen. Rougé, Jean-Louis and Emmanuel Schang (2006). The origin of the liquid consonant in Saotomense Creole. In: Bhatt, Parth and Ingo Plag (eds.), The structure of creole words: Segmental, syllabic, and morphological aspects. Tübingen: Niemeyer, 23–37. Rountree, S. Catherine (1972). The phonological structure of stems in Saramaccan. In: Grimes, Joseph E. (ed.), Languages of the Guianas. Norman, OK: Summer Institute of Linguistics, 22–27. Russell Webb, Eric (2010). Creole phonological restructuring: The role of perception in contactinduced change. In: Journal of Pidgin and Creole Languages 25(2), 263–288. Russell Webb, Eric and Travis G. Bradley (2009). Rhotic metathesis asymmetries in Romance: Formalizing the effects of articulation and perception on sound change. In: Masullo, Pascual J., Erin O’Rourke and Chia-Hui Huang (eds.), Romance Linguistics 2007: Selected papers from the 37th Linguistic Symposium on Romance Languages (LSRL). Amsterdam: John Benjamins, 321–337. Sabino, Robin (1990). Towards a phonology of Negerhollands: An analysis of phonological variation. Ph.D. thesis, University of Pennsylvania. — (1993). On onsets: Explaining Negerhollands initial clusters. In: Byrne, Francis and John Holm (eds.), Atlantic meets Pacific: A global view of pidginization and creolization. Amsterdam/Philadelphia: John Benjamins, 37–44. Saint Jacques Fauquenoy, Marie (1972). Analyse structural du créole guyanais. Paris: Éditions Klincksieck. Saint-Quentin, Auguste de (1872). Etude sur la grammaire créole. In: Saint-Quentin, Alfred de, 127-205.

322 | Bibliography Scheer, Tobias (2004). A lateral theory of phonology. Vol 1: What is CVCV and why should it be? Berlin: Mouton deGruyter. Schlegel, J. Bernhard (1856). Schlüssel zur Ewe-Sprache. Bremen: Valett and Co. Schuchardt, Hugo (1914). Die Sprache der Saramakkaneger in Surinam. Amsterdam: Johannes Müller. Schumann, Christian L. (1778). Saramakkanisch-Deutsches Wörterbuch. In: Schuchardt, Hugo (ed.), Die Sprache der Saramakkaneger in Surinam. Amsterdam: Johannes Müller, 44–116. Schüttenhelm, Sonja (2002). Die Silbe im St. Kitts Creole. Master’s thesis, University of Siegen. Schwartz, Bonnie and Rex Sprouse (1996). L2 cognitive states and the Full Transfer/Full Access model. In: Second Language Research 12(1), 40–72. Seidel, August and Ivon Struyf (1910). La langue congolaise. Grammaire, vocabulaire systématique, phrases graduées et lectures. Heidelberg: Groos. Selkirk, Elisabeth (1982). The syllable. In: Hulst, Harry van der and Norval Smith (eds.), The structure of phonological representations. Dordrecht: Foris, 337–383. Shinohara, Shigeko (2006). Perceptual effects in final cluster reduction patterns. In: Lingua 116(7), 1046–1078. Siegel, Jeff (1999). Transfer constraints and substrate influence in Melanesian Pidgin. In: Journal of Pidgin and Creole Languages 14, 1–44. — (2003). Substrate influence in creoles and the role of transfer in second language acquisition. In: Studies in Second Language Acquisition 25(2), 185–209. Sierman, Harry (ed.) (1987). Kramers Pocketwoordenboeken Nederlands-Engels. Amsterdam/Brussels: Elsevier Boeken B.V., 34th ed. Silverman, Daniel (1992). Multiple scansions in loanword phonology: Evidence from Cantonese. In: Phonology 9, 289–328. Singler, John Victor (1993a). African influence upon Afro-American language varieties: A consideration of sociohistorical factors. In: Mufwene, Salikoko S. (ed.), Africanisms in AfroAmerican language varieties. Athens/London: University of Georgia Press, 235–253. — (1993b). The African presence in Caribbean French colonies in the seventeenth century: documentary evidence. In: Travaux de recherche sur le creole haitien 16/17, 1–236. (Université du Quebec à Montreal). — (1996). An OT account of pidgin phonology: Coda consonants in vernacular Liberian English. In: Johnson, Ian, Matthew L. Juge and Jeri L. Maxley (eds.), Proceedings of the Berkeley Linguistics Society 22. Berkeley, CA: Berkeley Linguistics Society, 375–386. — (2000). Optimality Theory, the minimal-word constraint, and the historical sequencing of substrate influence in pidgin/creole genesis. In: McWhorter, John (ed.), Language change and language contact in pidgins and creoles. Amsterdam/Philadelphia: John Benjamins, 335–351. Smith, Norval (1987). The genesis of the creole languages of Surinam. Ph.D. thesis, University of Amsterdam. — (1999). The vowel system of 18th-century St Kitts Creole: Evidence for the history of the English creoles? In: Baker, Philip and Adrienne Bruyn (eds.), 145–172. — (2001). Voodoo Chile: Differential substrate effects in Saramaccan and Haitian. In: Smith, Norval and Tonjes Veenstra (eds.), Creolization and Contact. Amsterdam/Philadelphia: John Benjamins, 43–80. — (2003). Evidence for recursive syllable structures in Aluku and Sranan. In: Adone, Dany (ed.), Recent development in creole studies. Tübingen: Niemeyer, 31–52.

Bibliography

|

323

Smith, Norval and Hugo Cardoso (2004). A new look at the Portuguese element in Saramaccan. In: Journal of Portuguese Linguistics 3, 115–147. Smith, Norval, Ian Robertson and Kay Williamson (1987). The Ijo element in Berbice Dutch. In: Language in Society 16, 49–90. Smith, Norval and Marleen van de Vate (2006). Population movements, colonial control and vowel systems. In: Bhatt, Parth and Ingo Plag (eds.), The structure of creole words: Segmental, syllabic and morphological aspects. Tübingen: Niemeyer, 59–82. Steele, Jeffrey and Anne-Marie Brousseau (2006). Parallels in process: Comparing Haitian Creole and French learner phonologies. In: Lefebvre, Claire, Lydia White and Christine Jourdan (eds.), 331–352. Stein, Peter (1984). Kreolisch und Französisch. Tübingen: Niemeyer. Steinthal, Heymann (1867). Die Mande-Neger-Sprachen: Psychologisch und phonetisch betrachtet. Berlin: F. Dümmler. Steriade, Donca (2001). Directional asymmetries in place assimilation: A perceptual account. In: Hume, Elizabeth and Keith Johnson (eds.), The role of speech perception in phonology. San Diego: Academic Press, 219–250. Stetson, Raymond Herbert (1951). Motor phonetics. A study of speech movements in action. Amsterdam: North Holland, 2 ed. Stolz, Thomas (1984). Two chapters in Negro-Dutch (ND) etymology. In: Amsterdam Creole Studies 7, 35–52. — (1986). Gibt es das kreolische Sprachwandelmodell? Vergleichende Grammatik des Negerholländischen. Frankfurt am Main: Peter Lang. Stolz, Thomas and Peter Stein (1986). Language and history in the former Danish Antilles: Non-linguistic evidence for a diachronic description of the Negro-Dutch language. In: Amsterdam Creole Studies 9, 103–122. Strange, Winifred and Valerie L. Shafer (2008). Speech perception in second language learners: The re-education of selective perception. In: Hansen Edwards, Jette G. and Mary L. Zampini (eds.), 153–191. Tarone, Elaine E. (1987a). Some influences on the syllable structure of interlanguage phonology. In: Ioup, Georgette and Steven H. Weinberger (eds.), 232–247. — (1987b). The phonology of interlanguage. In: Ioup, Georgette and Steven H. Weinberger (eds.), 70–85. Thomas, John Jacob (1869). The theory and practice of creole grammar. Port of Spain: Chronicle Publishing Office. Reissued in 1969 by New Beacon Books, London. Thurot, Charles (1881-83). De la prononciation française depuis le commencement du XVIe siècle d’après les témoignages des grammairiens. Geneva: Slatkine reprints [1966]. Tinelli, Henri (1979). Afro-European syllabic patterns and structural stability in Romance and English creoles. In: Orbis 28, 267–276. Uffmann, Christian (2001). Patterns of vowel epenthesis (not only) in Shona loanwords. In: Féry, Caroline, Antony Dubach Green and Ruben van de Vijver (eds.), Proceedings of HILP5. Potsdam: University of Potsdam, 193–211. — (2003). Markedness, faithfulness and creolisation: The retention of the unmarked. In: Plag, Ingo (ed.), Phonology and morphology of creole languages. Tübingen: Niemeyer, 3–23. — (2006). Simplicity vs. unmarkedness in creole phonology. Paper presented at the 7th Creolistics Workshop, Giessen, Germany. — (2007). Vowel epenthesis in loanword adaptation. Tübingen: Niemeyer.

324 | Bibliography — (2008). Vowel epenthesis and creole syllable structure. In: Michaelis, Susanne (ed.), Roots of creole structures: Weighing the contribution of substrates and superstrates. Amsterdam/Philadelphia: John Benjamins, 123–152. — (2009). Creole consonant inventories: How simple? In: Faraclas, Nicholas and Thomas B. Klein (eds.), Simplicity and complexity in creoles and pidgins. London: Battlebridge, 81– 106. van Bergem, Dick R. (1993). Acoustic vowel reduction as a function of sentence accent, word stress, and word class on the quality of vowels. In: Speech Communication 12, 1–23. Van Name, Addison (1869). Contributions to creole grammar. In: Transactions of the American Philological Association 1, 123–167. van Rossem, Cefas and Hein van der Voort (eds.) (1996). Die creol taal: 250 years of Negerhollands texts. Amsterdam: Amsterdam University Press. Vekeman, Herman and Andreas Ecke (1992). Geschichte der niederländischen Sprache. Frankfurt am Main: Peter Lang. Walter, Henriette (1993). L’évolution phonologique du français depuis le XIIIe siècle. In: Schmidt-Radefeldt, Jürgen and Andreas Harder (eds.), Sprachwandel und Sprachgeschichte: Festschrift für Helmut Lüdtke zum 65. Geburtstag. Tübingen: Narr, 211– 220. — (1994). French inside out: The world-wide development of the French language in the past, present and the future. London/New York: Routledge. Translated by Peter Fawcett. Weijnen, Antonius A. (1991). Vergelijkende klankleer van de Nederlandse dialecten. The Hague: SDU uitgeverij. Westermann, Diedrich (1907). Grammatik der Ewe-Sprache. Berlin: Dietrich Reimer. — (1930). A study of the Ewe language. London: Oxford University Press. (translated by A.L. Bickford-Smith). Williamson, Kay (1965). A grammar of the Kolokuma dialect of Ijo. Cambridge: Cambridge University Press. Winford, Donald (1975). Creole culture and language in Trinidad: A socio-historical sketch. In: Caribbean Studies 15(3), 31–56. Yip, Moira (1993). Cantonese loanword phonology and Optimality Theory. In: Journal of East Asian Linguistics 2, 261–291. Zampini, Mary L. (2008). L2 speech production research: Findings, issues, and advances. In: Hansen Edwards, Jette G. and Mary L. Zampini (eds.), 219–249. Zimmermann, Johannes A. (1858). Grammatical sketch of the Akra- or Gã-language. Stuttgart: Basel Missionary Society. 2 volumes. Reprinted in 1972 by Gregg.