For my mother and father English Sound Structure John Harris BLACKWELL BLACKWELLCopyright © Jon Haris, 1994 “The righ of John Harris to be identified as author ofthis work hasbeen sserted in accordance with the Copyright, Designs and Patents Ad 1988. First published 1994 Reprited 1996 Blackwell Publishers Lad 108 Cowley Road Oxford OX UF, UK Blackwell Publishers Inc. 238 Main Suet Cambridge, Masachusatis 02142, USA. All rights reserved. Exoopt forthe quotation of shot passages forthe purposes (ol eticisn an eview, no pat ofthis publication may be eproduced, red fn rarieval system, or wanamied, in ony form or by any means letranic, iechaneal, photocopying, recording or therwise without the pir permission ‘ofthe publisher. Except inthe United States of America, his ook i sold mbjact to the endtion| that isha act by way of trade or olherwise, be leat, cold hsed out, or caherisecncalted without the publishers prior consent ins Tors of binding “or over ater than ha a wich itis published and without s similar endiion "ncludng this condition being imposed ea the subsequent purchase British Library Cataloguing in Publication Data ‘ACCP oraloguereord for this bok is svulable from the British Library Library of Congress Cataloging in Publication Dutt Harris, Join, 1954 agli sound aructurlTobn Harris Includes bibliographical references and index ISBN 0-631-18261-6— ISBN 0-6B1-18741-3 (pbk) 1 English language ~ Phonology. L Title ME LISRH36S 1994 421,5—de00 95-44174 CIP ‘Typeset in 10 on 12.5 pt Sabon bby Pure Tech Corporation, Pondicherry aia Printed and bound in Great Britain by Hartnolls 14d, Boi, Cornwall ‘This bok is printed on acid fee paper Contents Note on transcription Preface 1 Sounds and words 1.1 Phonology in generative grammar 1.2. Phonological phenomena 12.1 Alternations 12.2 Distribution 1.3. Representation and derivation 1.3.1. Principles vs. rules 1.3.2. Linear rules and representations 1.3.3 Non-linear representations 1.34 Parametric variation 1.4 Words and roots 1.4.1 English morphology 14.2 The phonology of roots and words 1.4.3 Morphological vs. phonological domains Exercises 2. Constituency 2.1 Non-linear phonological representations 22 Timing 2241 Introduction 22.2 Compensatory lengthening 22.3. Diphthongs and afficates 2.3 Phonological constituent structure 23.1 Introduction 23.2. Stress and constituency 23.3 The ‘syllable’ 23.4 Phonotacties 24 Onsets, chymes, nuclei 24.1. Some traditional assumptions 32 32 33 33 34 37 42 42 42 45 46 $3 53vi Contents 24.2 Onsets 24.3 Rhymes 244 ‘Codas’ 24S ‘Appendices’ 24.6. The rhyme: summary 2.5 Syllabification 2.6 Summary Exercises 3. Melody 3.1 Introduction 3.2. The primes of melodic representation 3.2.1 Sounds are componential 3.2.2 Privativeness vs. equipollence 3.2.3 Phonetic interpretation 3.3. Elements for vowels, 33.1 ALU 3.3.2 Autosegmental representations 3.3.3 Dependency relations between elements 3.3.4 Elements and sound patterns 3.3.5. The neutral element 3.3.6 English vowel systems 3.4 Elements for consonants BAL ‘Place’ 34.2. Lenition 3A3 “Manner doing things t0 a 3.5 Melodic geometry 3.6 Laryngeal elements 3.7 Conclusion 3.8 Appendix: specification of elements Exercises 4 Licensing 4.1. Introduction 4.2 The phonological hierarchy 421 Headedness 422 Constituent parameters 423 The prosodic hierarchy 424 Projection 4.3. Phonological licensing 43.1. Prosodic and autosegmental licensing 43.2. Constituent and projection licensing 43:3. Inter-constituent licensing 54 66 81 82 83 84 84 148 148, 149 149 149 151 152 154 154 156 157 4.3.4. Final consonants revisited 4.3.8. Deriving binary branching 43.6 Spreading and autosegmental licensing 4.4 Licensing and melodic complexity 441 Government 442. Complexity relations within onsets 44.3 Complexity relations within nuclei 444. The melodic content of codas 44.5. Deriving sonority 44.6 Complexity relations between nuclei 4.5 Principles and parameters of phonological structure 46 Empty positions 4.6.1. Domain final empty nuclei 4.6.2 Vowel syncope 4.6.3. Resyllabification? 4.64. Preserving prosodic structure 4.6.5. Proper Government 4.7 Lenition 4.7.1 Introduction 4.7.2. Four elenition systems 4.7.3 Coda analyses 4.7A. Licensing analysis 4.7.5. Licensing Inheritance 4.7.6 Cyclic effects 4.7.7 Preceding consonants 4.7.8 Following consonants 4.8 Summary Exercises 5. Floating sounds 5.1 Introduction 5.2. English r-systems 5.2.1. The most imperfect of consonants 3.2.2. Four esystems 5.23 Intrusive ¢ 5.3 A linear analysis of smooth r 5.4 Acoda analysis SAL RDropping S43 Now-Rhoticity 5.5. Floating r 5.6 Historical interlude Contents vii 160 163, 164 167 167 170 172 174 176 17 178 179 179 182 184 187 191 194 194 194 198 203 205 21 217 222 2as 225 230 230 230 230 232 235 237 239 239 243 245 248 252viii Contents 5.7 Vowels before r S.7.1 Rhotie systems 5.7.2 Nuclear $.7.3 Non-rhotic systems 5.7.4 MERRY-MARY-MARRY 5.7.5 Summary 5.8 Conclusion Exercises Epilogue Notes References Subject index Index of vowel word-classes 254 254 237 262 263 264 265 265 269 273 298 310 317 Note on Transcription ‘The alphabetic system of phonetic transcription followed in the book is essentially that used in most generative treatments of English, although somewhat modified in the direction of the International Phonetic Association model. In running text, forms cited in normal spelling are enclosed in angled brackets, eg. bit (bit). Vowels Front Non-front Non-round_| Round | Non-ound | Roma tense i [a a Hit ls 1 e ‘ o Tense e % ° Mid [tax e e | * 2 Low = a > Consonants ‘The phonetic values of the consonant symbols employed in the buok are for the ‘most part self-evident. The following particular conventions should be noted: voiceless palato-alveolar fricative voiced palato-alveolar fricative voiceless palato-alveolar affricate voiced palato-alveolar affricate palatal glide voiceless velar fricative voiceless labial-velar fricative BRN MOMSPreface There are two immediate decisions that face anyone setting out to write a book ‘on modern English phonology. One has to do with the nature of the material to be presented: to what extent should it reflect the dialect diversity that is part and parcel of the present-day language? The other concerns the terms within which the material is to be described and analysed: how much discussion should be devoted to the theoretical issues it raises? Books on the subject tend to fall into one of three categories according to the answers their authors give to these questions. One type is the primarily descriptive textbook which focuses on some standard variety of English, either exclusively or pethaps with some passing acknowledgement of dialect differences. Kenyon’s American Pronunciation and Jones’ An Outline of English Phonetics are bench- ‘mark examples. Authors of such books are typically not felt to be under much obligation to justify the theoretical assumptions that are implicit in the particular descriptive framework adopted. The second type differs from the first in explicitly concerning itself with theoretical issues but is similar in drawing its data from primarily standard sources. The classic of this genre is Chomsky and Halle’s The Sound Pattern of English. The thied type is usually marketed under the heading of sociolinguistics or dialectology. It differs from the others in giving pride of place to varieties which rarely if ever make it into the standard textbooks but generally shares with the first a lack of interest in phonological theory. The example which offers pethaps the most comprehensive coverage is Wells’ Accents of English. ‘Much less common is a type of book that is based on a fourth permutation of answers to the questions posed above ~ a book in which coverage of a wide range ‘of English varieties, standard and non-standard, is combined with an interest in matters theoretical. The present book is designed ro fill that gap. It discusses a range of phonological phenomena drawn from different types of English and, in analysing them, introduces the reader to recent theoretical advances in phono- logy. This orientation is not born out of a desire merely to analyse dialect material for its own sake, still less out of a belief that such an investigation would be rewarding only for whatever phonological curios it might turn up. The main rationale stems from the strategic importance that studies of dialect variation have assumed in the recent development of linguistic theory. Linguistic theory of late has witnessed a move away from an emphasis on rule-based analyses of structural differences among and within languages. What has emerged in its place is a framework in which linguistic variation is seen asxii Preface occurring along a limited number of parameters, in a manner that is determined by universal principles. Where previously a particular phenomenon in a grammar ‘would have been characterized in terms of a language-specific rule, now it is more likely to be treated as one of a fixed number of choices forced on the language system by Universal Granunat. This theoretical development is already well established in syntax and is now beginning to have a significant impact on phonology. With this shift in theoretical priorities has come a renewed interest in dialect variation. The comparative study of closely related language systems provides a controlled research environment in which individual linguistic variables can be investigated while surrounding variables in the grammar are held constant. Reflecting these recent developments, this book sets out to use phonological variation within English asa test-bed for the principle-oriented approach. The choice of variables to be anclysed is necessarily rather selective and has been determined primarily by matters of theoretical interest. It is thus not part ofthe brief of the book to provide an exhaustive description of English dialect phonology. Nor is there any reason to say any more about the regional and social distribution of the variables, than is minimally necessary to establish the credentials of particular data sets. With a blithe indifference to the social evaluation of phonological differences, the researcher is free to ransack the English data-base for whatever material makes promising grist for the theoretical mill. And the data-base, it has to be said, is pretty comprehensive ~ in part a reflection of the language's wide geographical dispersion. In fact, a good case can be made for saying that in English we find a hefty sample of the phono.ogical phenomena that are possible in any natural language. ‘The introductory chapter discusses the place of phonology in generative gram- mar and briefly reviews the types of phonological regularity that have occupied the attention of researchers over the years. We consider some of the reasons for the flight from earlier rule-oriented analyses of phonological processes. The chapter also secks to identify the types of phonological phenomena in English that ‘ate most likely to provide fruitful material against which to assess the more recent principle-based approach. Chapter 2 is about how sequences of sounds are organized into syllabic consti- tuents. This level of structure, we will see, has a significant role to play in such phenomena as the relative duration of sounds and the restrictions that are imposed on their ability to occur next to one another. One conclusion of the chapter i that a binary limit is imposed on the number of positions that syllabic constituents can contain. ‘The next two chapters address a number of fundamental questions about the nature of phonological processes: what is the set of possible processes, where do they occur, and why do they occur where they do? Chapter 3 tackles the first of these issues by considering how the internal make-up of sounds influences their susceptibility to phenological processing. Here we investigate the basic building blocks of sounds, how they are phonetically manifested, and how they are organized within phonological representations. Chapter 4 seeks to answer the questions of where and why processes occur by establishing an intimate link between the make-up of a sound and its ability to Preface xiii occupy particular positions in syllable structure. We consider evidence supporting the view that relations between adjacent positions are asymmetric. That is, within a domain formed by a pair of positions, one acts as the head. Headedness is observable not only within syllabic constituents but also between them, as well as, within wider phonological domains, including the word and the phrase. We will consider arguments in favour of the view that it is these asymmetries that create the conditions for phonological processes to take place. The chapter discusses, how headedness as well as the maximal binarity of constituents can be derived from fundamental principles which regulate phonological representations and indeed linguistic structure in general. Drawing on the theoretical insights reviewed in the first part of the book, chapter 5 presents a detailed analysis of a collection of phenomena related to r in English. These illustrate how differences in the realization of a particular sound are directly relatable to the syllabic position it occupies. Various sections of the book are given over to analyses of data relating to particular phonological variables in English. In each case, the presentation begins with a description of the different manifestations of the variable in a selected number of systems. Discussion of various competing analyses then provides the opportunity to compare rule-oriented and principle-oriented approaches. Each chapter concludes with sets of exercises which can be used in tutorial work. The treatment of this material offers further opportunities to consider the issues discussed in the body of the chapter. It is my pleasure to thank various colleagues who were prevailed upon to take me quietly on one side in an attempt to talk me out of committing to print the indiscretions of earlier drafts of the book. Three people in particular deserve special commendation for taking on this task ~ Michael Kenstowicz for his detailed and perspicacious comments, Neil Smith for his patient and thorough reading of several drafts, and Edmund Gussmann for allowing me to drive him up the wall with an early rough cut. Let me be the first to hold up my hand and accept blame for any follies which, for reasons of oversight or sheer bloody- mindedness, [ have failed to expunge from the text. Others who are in a position to call in hefty debts of gratitude include Monik Charette and Eugeniusz Cyran, who provided helpful criticism of various por- tions of the manuscript. Thanks also to Martine Grice, Phil Harrison and Theo Maniski for saving me from error in a number of places. Pethaps the biggest IU is due to the students at University College London who let me try out the book’s material on them. There is nothing quite like the justified puzzlement of a captive audience for sending a coursebook author scurrying back to the word-processor. I am also in hock to the numerous participants in other courses, where the material was given trial runs, particularly at the Universitat Autonoma de Barcelona, the E6vés Lordnd University in Budapest, the Central Institute for English and Foreign Languages in Hyderabad, and the University of Helsinki John Harris London October 19931 Sounds and Words 1.1 Phonology in generative grammar ‘The capacity to use spoken language boils down to an ability to assign meaning to sequences of speech sounds. It thus involves a harnessing of two different types of knowledge. One interacts with the central conceptual systems and controls how we compute meaning. The other is phonological and regulates the issuing of instructions to the motor-perceptual systems for the purposes of producing and recognizing speech sounds. These two types of knowledge are usually viewed as being located at separate levels of a generative grammar which models linguistic competence: the level of logical form (LF) interfaces with the central conceptual systems, while that of phonetic form (PF) interfaces with articulation and perception. ‘The mapping between LF and PF is mediated by syntax, which, by drawing on the lexicon, generates or defines well-formed sentence structures. The lexicon is the repository of all those properties which are idiosyncratically associated with each word ~ its meaning, its pronunciation and its syntactic and morphological characteristics." ‘As depicted in (1), the function of LF and PF is to assign respectively semantic and phonetic interpretations to sentences. () GRAMMAR Lr --- ~ Semantic interpretation syntax PE ~~~ = Phonetic interpretation Linguistic semantic interpretation is based on a combination of the lexical meanings of words and the relations these contract with one another in sen- tences.? In somewhat similar fashion, the phonetic interpretation of a sentence makes reference not only to the phonological shape of individual words but also to the phonological effects that result from combining these words in2. Sounds and Words sequence. For reasons to be considered presently, the significance of this point is that phonological facts in the grammar are distributed between the lexicon and PF. Some aspects of pronunciation are peculiar to individual words and reflect the essential arbitrariness of the sound-meaning pairing.’ The idiosyncratic nature of such facts is underlined by the observation that the same concept different languages can be signified by arbitrarily different sound sequences. (Compare, for example, the forms dak, Ranar (French) and letata (Sesotho) for the notion ‘duck’.) Not being subject to any general phonological or other grammatical principles, pronunciation facts of this type have to be listed in the lexicon. ‘Other properties of pronunciation, however, display regular phonological pat- terning, in the sense that they are characteristic of particular sound sequences rather than individual words. And not being lexically idiosyncratic, these patterns tend to recur in different languages wherever the relevant sequences occur. ‘One such pattern involves a process whereby a particular type of sound system- atically undergoes a change when it comes to stand next to another particular type of sound. This situation can arise, for example, when two words or mor- phemes are juxtaposed in a sentence, and the final sound of the first form comes into contact with the initial sound of the second. In the prototypical case, such a process will affect any sequence of morphemes containing the relevant string of sounds. Here are some examples that occur naturally in conversational English: (2) (a) tel} te{m] birds ula] ralm] past rufa} clear (b) miss {s]_—_miss you wy faze {2} faze you byl gett] get your yl did{d) did Eunice Gy) (6) senfa) senfd] Anne sen(f] two fin] fin(d] ic fin{d] them holla) hold} on holidl tight The first two examples illustrate processes involving assimilation of place of articulation. In (2a), a word-final alveolar nasal assimilates to the place specifica tion of the consonant at the beginning of the following word. Something similar happens in (2b), where we see how a word-final alveolar obstruent (plosive or fricative) becomes palatalized (to a palato-alveolar) under the influence of a following palatal glide. (2c) is an example of a truncation process; in this case, d in a word-final consonant cluster drops when the next word begins with a consonant. Sounds and Words 3 1.2 Phonological phenomena 1.2.1 Alternations The examples in (2) illustrate the point that a morpheme can appear in one of a number of shapes or alternants. On the basis of the facts given in (2b), for ‘example, we can identify two alternants associated with the morpheme (miss), namely misand mi. To put it somewhat differently, the sin (miss) alternates with J. Two properties of the examples in (2) are worth remarking on. First, the relationship between the alternating sounds can be expressed in phonological terms; f in (miss), for example, can be described as a palatalized relative of s. Second, the context in which the alternations take place can also be specified phonologically; that is, the selection of alternants is determined by the sounds a morpheme comes into contact with in neighbouring morphemes. These properties are characteristic of a large class of the alternations that are encountered in the world’s languages. Not all alternations conform to this pattern, however. In the case of suppletive morphology, for example, no phonological connection exists between the different shapes of a morpheme; examples in English include (go~ went) and (good-better).. The question now is how alternations are to be represented in the grammar. One possibility might be to list the different alternants of a morpheme in the lexicon. Given that the lexicon is the repository of idiosyncratic information, this would certainly seem to be the appropriate solution for suppletions such as BIC__D (b). Structural description: CAD Structural change: CBD ‘The letters here can stand for feature matrices which identify individual sounds or classes of sounds. The arrangement in (8a) specifies that A, the rule's input, is rewritten as B, the outpat, in the context of a preceding C and a following D. C and D, either or both of which may be empty, constitute the rule’s environment. ‘The effect of the rule is to take a string CAD (the rule's structural description in (8b)) and transform it into CBD (the structural change). To take a concrete example, we can formalize the labial portion of the place- assimilation process illustrated in (2a) in the following terms: - silaie] (9) [+ nasal] > f+ rabiaty —[; Iebial That is, a nasal acquires a labial place of articulation when a labial consonant follows. The letters A or B in (8a) can also stand for null; C or D can also stand for a morpheme boundary. Where null occupies the input (f — B), the rule is inter- preted as inserting a sound. Where the output is null (A — 9), a deletion process is represented. The truncation process in (2c), for example, is expressed along the following lines: (10) 4 G/{- syllabic] ___# {- syllabic]10 Sounds and Words In prose, d is deleted when preceded by a consonant and followed by a word boundary (#) which is itelf followed by another consonant. There has never been any real doubt about the rewrite rule's ability to represent any observed phonological process. However, it suffers from a number of major and long-recognized flaws, First, it permits the expression of an enormous set of possible processes, the vast majority of which are unobserved in the languages of the world. Second, the model fails to capture significant asymmetries in the distribution of those processes that are attested. Some processes are evidently ‘more favoured or natural than others, in the sense that they show a greater propensity to recur in different grammars. These two points ate not likely to be uunconnected; so let us consider them in unison. One aim of early generative theory was to provide a direct correlation between the naturalness of a rule and its formal simplicity. According to this notion, the ‘more natural a process is, the simpler should be the corresponding rule, as gauged by the number of units contained in its structural description and change. It was soon acknowledged, however, that the rewrite-rule model patently failed to ‘measure up to the simplicity criterion.” To illustrate this point, consider just a few of the permutations that can be performed on the place assimilation rule in (9). A rule which specified a nasal as, say, [- labial] as opposed to [+labial] in the context of a following [+labial] constant would be formally just as simple as (9). However, the process it ex- presses, one of dissimilation, is much less favoured than assimilation. Examples of other rules minimally different from (9) include one which specifies the input nasal as [+labial] in, say, the context of a following vowel, or another which specifies the nasal as [+lateral] before a [+labial] consonant. The latter two ‘examples express processes which are disfavoured to the point of being complete- ly unattested in any natural language. This rather limited illustration points up a very general problem with the rewrite-rule format: for every rule that corresponds to a naturally occurring process, it is possible to formulate an unfortunately large array of minimally distinct rules which express processes that are either heavily disfavoured or downright impossible. To be fair to proponents of the SPE apprach, this failing was acknowledged from an early date. In response, the sub-theory of markedness was proposed, in ‘which a set of universal conventions designated certain features or feature combina- tions as being unmarked, that is, more favoured than others." Only marked features were then deemed to contribute to a rule’s formal complexity. The most highly valued rule would thus be one which contained nothing but unmarked features. A subset of the conventions defined the conditions under which assimi- latory rules such as thet in (9) were evaluated more highly than disfavoured permutations such as those mentioned in the last paragraph. ‘The markedness solution was but one of a number of approaches to the central problem of the rewrite-rile model, its capacity for over-generation. Each of these proposals in its own way illustrates a quite general development in the emergence, elaboration and demise of scientific paradigms. As dissatisfaction grows with a particular theory’s inability to fit the facts it is intended to explain, the research community is faced with a familiar choice: either it modifies the theory or Sounds and Words 11 abandons it altogether. Naturally the former option is likely to be preferred in the absence of a better competing theory. If the model’s main flaw is one of excessive generative power, some modifications may amount to no more than patch-up jobs, devised to mask a fundamental design fault. Markedness theory could be said to fulfil this role in relation to rewrite rules. Suppose, however, that an alternative model becomes available, one which is not afflicted by the expressive exorbitance of its competitor. As it is developed, the new theory may show signs of under-generation. That is, in eschewing generative promiscuity, it may apparently go too far the other way and fail to account for sets of data which its predecessor was able to handle quite comfort- ably. At this point, there may be some temptation to fall back on the earlier theory and carry on with attempts to reform it. Alternatively, the researcher can per- severe with the new model and earmark those areas where it seems to under- generate as problems for future research. Both orientations are in evidence in current attitudes to linear rewrite rules, sometimes even within the work of the same researcher. Few still assume a full-blooded SPE version of the rewrite-rule model. Some have abandoned it altogether in favour of the more recent principle-based approach briefly reviewed in the last section. Still others implicity work with some kind of hybrid, reserving the right to invoke SPE-type rules whenever they feel the newer model lets them down through apparent under-generation. Inherent in the principle-based approach to phonological processing embraced in this book is a well established minimalist methodology, widely assumed in scientific endeavour and based on the following procedure. Given two competing, models, begin by preferring the more impoverished. Initially set aside for future investigation any data for which no immediate fit with the model can be found. Only abandon the model if, after persistent disconfirmation, the empirical bal- ance tilts decisively in favour of the more enriched competitor. Returning to our consideration of the SPE markedness approach, we may note yet another problem: the universal markedness conventions are themselves no less arbitrary than rewrite rules. One convention, for example, identifies front rounded vowels as being more marked than their non-round congeners. This correctly captures the more favoured status of front non-round vowels. But simply stating the asymmetry in these terms provides no explanation of why things should be this way rather than the other way round. Arbitrariness is in fact one of the inherent weaknesses of the rewrite rule model which contributes to its over-generating capacity. One of the things an account of a given process has to do before it can be considered non-arbitrary is to establish a direct connection between the process and the context in which it occurs.'* That is, it should offer an explanation of why a particular process takes place where it does. Rewrite-rule analyses fail on this count for the reason that they provide no formal link between a process and the context in which it occurs. This weakness stems from the representation of the output and environment as independent entities in a rewrite operation. This property of the model contributes to the equal facility with which disfavoured non-assimilatory processes and favoured assimi- atory processes can be formulated. Intuitively, features acquired by the input to12 Sounds and Words an assimilation process are somehow ‘the same’ as features represented in its environment. It is of course possible to declare that rules in which a copy of a feature from the environment appears in the output are more highly valued than rules which manipulate dissimilar features. But the need for such a stipulation amounts to an admission that the rewrite rule fails to capture the intuition directly. ‘Another arbitrary characteristic of the rewrite-rule format is that it establishes no principled limie on the length of phonological strings that can be referred to in a rule's structural description. With the possibility of including more than one segment or boundary symbol in a rule’s environment (as in (10), for example) ‘comes the prediction that the trigger of a process can occur at any distance from the site at which the process takes place. Again this allows for the formulation of ‘a host of rules which express process-types that are unattested in any language. For example: ‘delete the third segment in a sequence of seven segments’. The same kind of criticism was levelled at the use of transformational rules in early generative syntax. In recent times, the response has been to assume that all syntactic operations ar: subject to the constraining principle of locality.*® That is, a relation of adjacency must exist between two phrase-structure positions before they can participate in a particular syntactic movement. This constraint is now seen as one of the central principles regulating processes in both phonology and syntax. In phonology, :he locality principle requires that the target and trigger of a process be adjacent." Adjacency in phonology, no less than in syntax, cannot be equated with super- ficial contiguity. That is, in many cases, two positions must be considered ad- jacent at some level of structure, even though they are superficially separated by other positions. Classic cases of this type include vowel harmony, in which all vowels within a certain span (usually the word) agree with respect to a particular feature, irrespective of the number of intervening consonants that might be present. By the same token, the fact that two positions are superficially contigu- us is no guarantee that they are structurally adjacent. Exactly what constitutes adjacency in a phonological representation is one of the main themes running through this book. ‘Within the SPE framework, itis not at all clear how the locality principle can be applied. Restricting rules to those that refer only to adjacent segments and boundary symbols will certainly exclude environments containing strings of arbit- rary lengths, while still allowing for the formulation of local assimilation pro- cesses such as that in (5). Unfortunately, however, it will also rule out a significant class of attested processes which do display long-distance effects, such as the vowel-harmony phenomenon just mentioned. This problem stems as much from the linear nature of SPE-type representations as from the rewrite-rule model itself. Intuitively, we would want to say that vowels in a harmonic domain are adjacent at some level that is inaccessible to consonants. But such a level has no formal expression within a representational framework in which sounds are concaten- ated in a single linear string. Sounds and Words 13 1.3.3 Non-linear representations Since the late 1970s, research into phonological processing has seen a concerted effort to redistribute the explanatory burden between derivation and repres- entation. By enriching representations, it has proved possible to reduce reliance on potentially arbiteary derivational machinery. There are two main respects in which this development can be seen to yield a more constrained model of derivation. First, the set of formal operations that can be performed on a repres- entation is now deemed to be considerably smaller than at first thought. On current assumptions, there are in fact only two fundamental operations ~ one which joins together different parts of a representation, and another wi sunders them. A further contraction in the role of derivation is achieved if we adopt the simplest possible hypothesis regarding the applicability of phonological processes, namely that they take place freely wherever their conditions are satis- fied. Under this view, there is no provision for processes to be arbitrarily prevented from occurring in contexts where they otherwise could; nor is there any mechanism whereby one process could be arbitrarily stipulated as applying before another. These developments, it has to be said, have not been universally taken on board by phonologists. But they remain fundamental to the principle-based research programme. ‘The move away from derivational concerns would not have proved particularly fruitful, had it only succeeded in simplifying one aspect of the theory at the ‘expense of complicating another. As the focus shifted, so the need to discover general constraints on the shape of representations became all the more pressing. On the success of this enterprise hinges the possibility of reaching a stage where all processes, or at least a substantial proportion of them, can be viewed as necessary consequences of particular conditions prevailing in phonological repres- entations. ‘This development has indeed enabled some progress to be made on the issues of locality and non-arbitrariness. It is now widely acknowledged that features are arrayed on independent tiers rather than being lumped together in bundles.2® The synchronization of features belonging to a single segment is expressed by means of association lines, as illustrated in the following rough approximation of the representation corresponding to a labial plosi (11) [labial] [stop] [oral] During the course of the book, we will examine in more detail the nature of this, mode of representation and the evidence supporting it. For the moment, however,14 Sounds and Words let us accept that multi-tiered representations are well motivated and note how they allow us to characterize assimilatory processes in a non-arbitrary fashion. Allied to the multi-tiered mode of representation is the hypothesis alluded to above, that most if not all phonological processing can be reduced to two primitive operations ~ the insertion oF deletion of association lines.” All assim lation processes are represented as line insertion, an operation that permits a single feature to be linked simultaneously to more than one segment. It is this possibility that gives rise to the term non-linear, used to describe representations Of this type. The place assimilation in (2a), for example, can be characterized in terms of the following representation of a labial nasal-plosive cluster: (12) (a) {labial} (b) labial] [stop] [stop] {stop} [stop] [nasal] [oral] {nasal] [oral] ‘The homorganicity of the cluster is represented as the insertion of an association line linking the [Labial] feature of the plosive to the preceding nasal. The labial] feature can be said to spread from the plosive to the nasal. The advantage of expressing assimilation in this fashion is that it establishes a non-arbitrary rela- tion between the process and its operating context.2* This it does in the most direct way possible — by identifying the output of the process and the context as ‘one and the same feature, in this case [labial]. The question now is whether, in spite of its non-arbitrariness, the formulation in (12) is any less of a stipulative statement than the rewrite rule in (9). One response has been to assume that spreading is automatic and thus does nor need to be represented as a rule operating in individual grammars. According to this view, the operation is universally defined and occurs wherever it can in individual systems, as long as the necessary conditions prevail. One condition involves a locality requirement: the segments participating in the spreading operation must be adjacent. Another involves one segment being specified for a particular feature while the other is not, as depicted in (12a). Automatic spreading will result in the ‘unspecified segment acquiring the missing feature from its neighbour. For this account to be successful, there needs to be some principled means of determining the directionality of spreading and the contexts within which it does and does not take place. Is spreading always unidirectional, or can it proceed in either direction? What constitutes adjacency for the purposes of spreading? If these questions elicit answers that invoke general principles, rather than lan- ‘guage-specific statements, then we are on the way towards providing a non-stipu- lative account of assimilation and perhaps ultimately of phonological processing in general. Sounds and Words 15 In the succeeding pages, we will consider precise formulations of a number of principles, including those relating to locality and directionality, and consider how they offer answers to the fundamental question of why phonological pro- cesses occur where they do. 1.3.4 Parametric variation ‘A major preoccupation of modern linguistics has been the study of Universal Grammar. This notion can be understood both as a theory of those universal properties that constrain the form of individual grammars and as an account of the human language faculty in general. Studying individual language systems from this perspective involves distinguishing between a core, containing those linguistic traits which reflect systematic properties of Universal Grammar, and a periphery, containing traits which display unsystematic or eccentric behaviour. In the latter category belong such syntactic phenomena as one-off idioms and freakish constructions. (An example of the latter in English would be construc- tions of the type (secretaries general), which fail to conform to the otherwise regular pre-nominal positioning of attributive adjectives.) ‘At the phonological level, the periphery consists of those idiosyncratic pronun- ciation properties of individual lexical items that reflect the arbitrariness of the sound-meaning pairing. Included in this category are accidental lexical gaps — phonologically possible forms that happen not to be utilized as lexical addresses in a particular language, such as *blik in English.}* Some peripheral properties represent historical debris washed up in the lexicon as a result of once-active phonological processes becoming extinct over time. For example, the fact that the dental fricatives of words such as (mother, together, either) are voiced bears ‘witness to a process of inter-vocalic fricative voicing that was once fully product- ive in Old English. The extinction of this process is evidenced in the considerable number of (mostly borrowed) words in the modern language which contain medial voiceless 6, such as (ether, method, mathematics). One source of dissatisfaction with earlier generative theory was an inability to provide a principled distinction between core and peripheral facts of grammar. ‘The transformational rule format is unfortunately as adept at characterizing phenomena which recur across different grammars as characterizing those which are peculiar to specific constructions or phonological sequences in particular languages. At the phonological level, this failing is linked to the essentially open-ended nature of the rewrite-rule model. The set of possible phonological processes that can be expressed within this theory is of vast proportions. And this ‘generative excess is compounded by the possibility of stipulating different order- ing relations between pairs of rules.2” As a result, the choice of derivational routes ‘whereby the different shapes of an alternating morpheme can be mapped from a single lexical representation is enormously wide. This arrangement cannot be considered a sound basis for plausible models of phonological acquisition or speech recognition. It would place excessive processing burdens on learners16 Sounds and Words seeking to construct and listeners seeking to recognize lexical representations on the basis of variant phonetic shapes appearing in linguistic input." ‘The overall effect of the rewrite-rule model is that the set of possible phonolo- sical grammars it generates is to all intents and purposes infinite. This prediction fails to tally with the current perception that core phonology is finite. That is, differences among phorological systems occur within fixed bounds laid down by Universal Grammar. The investigative task is to discover where these limits lie and to determine the precise points at which grammars are free to vary. Variability in core syntax is nowadays characterized in terms of a relatively small number of parameters, each of which defines a choice (usually binary) between particular typological characteristics.” One well-known example has to do with whether or not a subject pronoun position can be left empty: in languages such as Italian and Spanish it can, in others such as French and English it cannot. ‘The core grammar of a language is now viewed as a collection of specific choices, or settings on such parameters. It is increasingly being acknowledged that cross-grammar variability in core phonology should be conceived of in the same parametric terms.” Precisely how much of phonological variation is amenable to this kind of treatment continues to be subject to debate. There seems to be general agreement that certain gross typological differences 10 do with such dimensions as syllable structure and stress assignment can be characterized in this way. One such parameter determines whether stress prominence is located at the left or the right edge of a rhythmic constituent. At the leve! of the phonological word, for example, French takes up the right-hand option, while Hungarian opts for the left-hand setting.” In chap- ters 2 and 4, where we examine how phonological strings are organized into syllabic constituents, we will see that the syllable structure of a given language can be defined in terms of a small number of options which determine the shape of different types of constituent. It is now widely recognized that parameterization can and should be extended to phonological processes. For example, the general parameter that regulates the left versus right placement of stress can also be taken to control leftward versus rightward spreading in long-distance harmony pro- ‘The overall picture that is emerging is one in which Universal Grammar defines rigid limits beyond which it is impossible for individual phonological systems to stray, At the same time, it identifies particular areas where phonological structure is only partially determined; it is at these interstices that differences among core phonological grammars are located. ‘One of the methodological challenges associated with the parametric research programme is familiar from other fields of scientific enquiry. It concerns the necessity of ensuring that the results of investigating a particular set of data are not contaminated by effects emanating from some related set. For example, before attributing a specific structural difference between two languages to @ particular parameter, we need to be sure that the divergence is not simply an incidental reflection of some other entirely independent distinction between the grammars. Sounds and Words 17 To take a concrete example: languages vary in terms of the type of vocalic clusters (vowels and glides) they permit, an effect that is frequently expressed in terms of a parameter regulating the relative sonority of neighbouring sounds. (Very crudely, sonority can be understood as the degree of articulatory opening involved in the production of a sound. We will examine the notion in more detail in the next chapter.) In English, the second member of a vowel-vowel or vowel glide cluster is generally restricted to w, i and (in some dialects at least) a, as in the second portion of the diphthongs in (cow, die, fear). Compare this with the situation in Sesotho, a Bantu language spoken in southern Africa, in which sequences of vocalic pairs are more or less unrestricted ~ hence the occurrence of clusters such as ai, ac, ae, a3, a0, au. The mismatch between the two languages can be described at some level in terms of sonority differences: unlike English, Sesotho apparently imposes no restrictions on the sonority profile of vocalic clusters. However, to ascribe the divergence directly to a difference in sonority restric- tions in this instance would be misguided. The mismatch is in fact a superficial reflection of a more deep-seated difference between the languages which itself involves a parameter that is quite independent of sonority. Each of the English vyocalic clusters just described is diphthongal in so far as it is contained within the same syllable. The Sesotho clusters do not qualify as diphthongs in this sense; instead they constitute sequences of separate syllables. (This conclusion is sup- ported by the observation that each vowel in such strings bears a separate tone.) In fact, Sesotho is like many languages in forbidding more than one segment to occupy the nucleus of syllable (that portion most typically occupied by a vowel). ‘The fundamental distinction between Sesotho and English in this matter thus stems from different settings on a parameter which regulates the number of positions permitted to occur in the nucleus. This example illustrates the pitfalls awaiting the researcher trying to isolate the source of a difference between two languages when more than one parameter is potentially involved. The ideal experimental environment in which to test a particular parametric hypothesis is one where all variables in the data are held constant save the one which is the object of investigation. The varying effects of sonority constraints in vocalic clusters, for example, can be more accurately gauged by comparing languages with like settings on the nuclear parameter just mentioned. The more closely related the two language systems are, the more likely wwe are to approximate laboratory-like conditions. For this reason much of the work on parametric variation in syntax has taken the form of contrastive studies of genetically related languages. It is in the light of this experimental orientation that comparative dialectology takes on a strategic significance it rarely enjoyed in earlier rule-based approaches to linguistic variation.”? The considerable degree of structural relatedness that typically unifies dialects of the same language provides a favourable research environment within which to focus on one linguistic vari- able at a time, while minimizing the danger of contamination from extraneous noise due to other variables. This point provides one of the main motives for the present book's treatment of dialect variation in English.18 Sounds and Words 1.4 Words and roots 1.4.1 English morphology Before proceeding any further with the discussion, we need to come to some decisions about the nature of the data we should be investigating. In particular, it is necessary to establish what kind of phonological phenomena in English are most likely to furnish the relevant material against which to test the validity of the principle-based approach outlined in the last section. This requires us to take account of the manner in which English morphology contributes to the conditions ‘under which phonological phenomena of different types occur. ‘The phonological phenomena found in any given grammar are not necessarily evenly distributed throughout the morphology. In many languages, including English, different sutsets of patterns are observable within different sorts of morphological domains. By way of illustration, compare the phonological beha- viour of the prefixes (in) and (un-). (in-) has a number of alternants, the basic one, in-, being evident before vowel-initial stems (as in (13a)) as well as before certain consonants (13b). Before certain other consonants, however, other altern- ants show up: im- before labial plosives (13c), # before liquids (For r, as in (134)) or nasals (13e). (13) (a) inexcusable, ineligible, inoperative (b) inflammable, intrepid, insouciant (C)_ impossible, implicit, imbued (4) illegal, irregular, irresponsible (©) innocuous, immaterial, immature Note that, in spite cf the spelling, the forms in (13d) and (13e) do not contain geminate (double or long) consonants. That is, a word such as (innocuous) is pronounced {j{nJocuous) rather than *(i{nnJocuous). This makes this prefix quite different from (un-). Ifthe latter alternates at all, it does so only optionally; the place of articulation of the nasal may be subject to the assimilation process in (2) {as in (ufmbpleasant, ufg)kind)). But the most significant difference is that the nasal of (un-), unlike that of (in-), is consistently present. One consequence of this is that the attachmen: of (un-) to a stem can result in the creation of geminate sequences. Hence forms such as (14) ulnnjerved, ufunjecessary, ufunjacural ‘The different behaviour of (in-) and (un}) illustrates a fundamental distinction between two types of morphology in English. (in-) is representative of what is often referred to as root-level affixation, (un-) of word-level affixation. Very Sounds and Words 19 broadly speaking, different morphological operations are assigned to each of these categories along the following lines:» (15) English morphology fa) Root-level morphology: AMfixation: (in, “ity, -ic-al, ory, -ate, ion, ant, -th,...) ‘Strong’ verbs/nouns: (blew, brought, sang, feet, mice, .. > (b) Word-level morphology: Affixation: (un-, ed, (e),-ing, -ness, ly, ful, ship, -hood, -ment, > Compounds, e.g. (cart horse, seagull, blackboard, (Apart from a handful of largely unproductive Germanic suffixes (including (-th), as in (fifth)), root-level affixes are overwhelmingly of Greek or Latinate origin. The ‘strong’ verb and noun forms are nearly all Germanic. So are the bull of word-level affixes, although a few Latinate forms ((-ment), for example)) belong, to this group. Certain pairs of root and word affixes are Latinate cognates: compare root-level (su(C)-) (where C stands for some orthographic consonant, as in (suffix, sustain)) with word-level (sub-) (as in (subfreezing, subtend)). The use of terms such as Latinate and Germanic in this context is purely descriptive and is not meant to imply that etymological considerations play a synchronic role in lexical organization.) The root and word categories are distinguished by a collection of semantic, morphological and phonological characteristics. At the semantic level, the ‘meaning of forms derived by means of word-level morphology is typically com- positional; that is, the overall meaning can by and large be extrapolated from the component morphemes. This is not necessarily the case with root-level morpho- logy. Compare for example the (in-) and (un-) prefixes already mentioned. (an-) attaches to free morphemes — that is, to minimal forms which can stand as words in their own right. Under such circumstances, the reading contributed by (un-) is regularly analysable as negative (as in (unkind)) or reversive (as in (untie)). In contrast, (in-) is frequently prefixed to bound morphemes, forms which do not exist as independent words, such asin (impeach, illusive, imminent). Three basic ‘meanings are usually associated with (in-): negative (as in (inapplicable)), intens- ive or causative (as in (imperil)), and directional (as in (immigrate)). But there are ‘many words in which the presence of (in-) has no readily identifiable autonomous ‘meaning, including (inane, inveigh, imminent). Morphologically, some degree of ordering relation evidently exists between the ‘wo levels, Root-level affixes can be attached inside other root-level affixes (as in (16a)) or inside word-level affixes (as in (166)).** (16) (a) nation-al-ity (b) nation-al-s (c) nation-hood-s (d) *nation-hood-al, *nation-s-ity20 Sounds and Words In contrast, although word-level affixes can be attached inside other word-level affixes (as in (16c)), typically they cannot be attached inside root-level affixe: hence the ungrammaticality of forms such as those in (16d). One way of account- ing for this discrepancy has been to view root-level operations as taking place at an earlier stage of morphological derivation than word-level operations. The sequencing generalization is, however, not absolute; a word such as (un-gram- ‘matical-ity), for example, evidently involves the attachment of the root-level suffix (ity) to the word-level-derived form (un-grammatical).” 1.4.2 The phonology of roots and words ‘The most consistent differences between English root and word morphology are phonological. That is, there exist sets of phonological phenomena that occur at one level but not the other. Let us first examine examples of root-level alterna- tions that are absent from the word level. Root-level phenomena are actually of two types: some represent generalizations that extend to morphologically underived forms, while others are peculiar to root morphology. We have already seen one example of the first type in the failure of ‘geminate consonants to appear in forms containing the root prefix (in-). Accord ing to one analysis, this reflects the operation of a process of degemination, illustrated in the forms in (17a). This process can be assumed to operate vacuously within underived forms, since they too lack geminates in English. (This is why words borrowed from languages with geminates only ever show short consonants in English. Thus a typically English pronunciation of (spaghetti) lacks the geminate tt of Italian.) (17) (a) Degemination ifn}-effectual ifgl-nocuous ifm}-probable ifyh}-mature (b) Closed-syllable shortening Long VVC Short VEC perceive perceptive describe descriptive reduce reduction thieve theft ‘The examples in (17b) illustrate another process that is common to root-level morphology and underived forms — so-called closed-syllable shortening.” In the next chapter, we will examine the syllabic conditioning of this phenomenon in some detail, but for the time being itis sufficient to note that long vowels before a single final consonant alternate with short before a cluster of two consonants. (Also implicated here is a set of vowel-quality changes, on which more presently.) The same pattern is evident in underived forms; hence the non-existence of Sounds and Words 21 sequences such as *-ukt, *-ipt-, *-ayf within the same morpheme. (That is not to say that such sequences never occur in English. They do, but only when they contain a word-level morpheme boundary, as in stipt (steep-ed). We will return to this point in 2.4.4.) Some alternations, on the other hand, are specific to root level morphology. ‘That is, they involve patterns that generalize neither to underived forms nor to word-lével morphology. The examples given in (18) are identified by terms adopted in SPE and still widely used in the literature. (18) (a) Velar Softening electrifk] lectrifs}-ity ccritifk} critifs]-ism smystifk] smysti(s}ism (b) Spirantization pirate pirals}-y president presiden(s}-y permit permiss-ive conclude corrode deride (c) Vowel Shift and Trisyllabic Laxing veyn vain venuti van-ity sori serene sorent sereni divayn divine devant divin-ity In Velar Softening (18a), a stem-final k alternates with s when a root-level suffix beginning with 1 is attached. As shown in (19a), this pattern is not evident in word-level suffixes beginning with the same sound. (19) (a) panick-ing —*panifs}-ing (b) flighele-y ——*figh{s]-y {c)_ might-i-ly *mfytily teeter-ing *ele}ter-ing Neither does Velar Softening extend to &r sequences within underived forms. Hence the lack of Velar Softening in, say, (king) (*— srg). Under Spirantization (18b), stem-final ¢d alternate with s when followed by suffix-initial 4. Compare the alternation-inducing behaviour of the root-level nominalizing suffix (-y) in (18b) with the lack of alternation associated with the word-level adjectival suffix (-) illustrated in (19b). Like Velar Softening, Spiran- tization does not manifest itself morpheme-internally; hence (tea) (*—> si}, (deem) (* sim), for example. ‘The forms in (18c) illustrate two patterns: Trisyllabic Laxing, in which long (including diphthongal) vowels alternate with short in suffixed forms, and Vowel22 Sounds and Words Shift, in which the resulting short-long contrast is accompanied by a difference in vowel quality. As shown in (19c), neither of these alternations is evident in forms derived at the word level. Nor do they correspond to any generalization that can be made about underived forms; hence examples such as (salient) ((sfeyllient) not *(sfeliens)) and (lenient) ((f:}nient) not *4fe}nient)) Tis characteristic of specifically root-level alternations of the type illustrated in (18) that they sustain lexical exceptions; the presence of i: in both (obese) and (obesity), for example, shows an absence of Trisyllabic Laxing and Vowel Shift. In this respect, root-level generalizations which also hold of underived forms are quite different. Patterns such as those illustrated in (17) are exceptionless. Nevertheless, in spite of the differences just detailed, there is one significant respect in which the two types of rootlevel alternations are identical: they faithfully conform to the set of phonological structures associated with underived forms.** For example, they introduce no segments not already found in morpho- logically simple forms. Moreover, as we will see in the next chapter, there exist severe restrictions on the co-occurrence of segment sequences across a root boundary, and these are identical to those operating morpheme-internally. To put it another way, the collection of phonological traits that characterize forms derived by root-level morphology is indistinguishable from that associated with underived forms. The phonological shape of root-derived forms thus provides no clue as to their morhologically complex structure. For example, there is nothing, in the phonological shape of root-derived forms such as (dorsal, polar, tonic) which would in principle exclude them from being morphologically simplex ‘words; compare these examples with, say, (morsel, molar, panic). Indeed in many cases, ‘strong’ root forms are homophonous with unrelated simplex forms; com- pare (blew-blue, feet-feat, taught-taut). ‘All this is in stark contrast to word-level morphology, which introduces ‘novel’ segmental contrasts and segment sequences not found in underived and root-level forms. Most of these patterns word-level morphology shares with phrase and sentence domains. Examples of sound sequences which straddle a word-level morpheme boundary but which are impossible within morphologically simplex or root-derived forms are particularly plentiful in compounds. Here, just as across word boundaries at sentence level, there are no restrictions on the co-occurrence of morpheme-final and following morpheme-initial consonants. Impossible mor pheme-intemal sequences that occur quite freely within compounds include p-m, v-tand 6-6, illustrated in (20a). (20) (a) Compound (b). Phrase/sentence pm hp marker stop me ve cove tail live to ob oth ball path belongs The same freely occurring sequences are evident across words at phrase and sentence level, as the examples in (20b) show. Sounds and Words 23 Moreover it is precisely at word-level morpheme boundaries that we find the gemination effect that is absent from root-level morphology. The gemination associated with (un-), illustrated in (14), isin fact characteristic of the word level in general. Thus, we find geminate nn not only in forms such as (u{n-n}atural) but alco in suffixed forms such as those in (212). Indeed any consonane-initial word-level suffix is free to create a geminate, as the forms in (21b) demonstrate. 21) (a) nm keen-ness, brown-ness (b) I cool-ly, tail-less ff trough-ful (c)__nighe time, sack cloth, tail light (d)_ good day, take care, pass slowly ‘The same phenomenon occurs at compound-internal boundaries, as in (21c), as. ‘well as at the phrasal level, as in (21d). Furthermore, whenever intra-compound and morpheme-internal sequences might be expected to coincide, they usually differ with respect to their phonetic realization. A sequence transcribed as tr, for ‘example, is pronounced quite differently according to whether or not a word-level morpheme boundary intervenes. A typical pattern is to find a released plosive followed by an aspirated r in, say, (nitrate, petrol) but an unreleased stop followed by an unaspirated r in, say, (night rate, hat rack). Although cross-morpheme sequences involving word-level affixes are not to- tally unrestricted in the way that cross-word sequences are, the possibilities of co-occurrence are nevertheless very much greater than within underived and root-level forms. One of the few restrictions we find holding of word-level suffixes involves the phenomenon of voice assimilation which affects the suffixes ed) and (-(e)s). Among the various alternants exhibited by each of these affixes, a voiceless type, t or s, occurs when the preceding morpheme ends in a certain class of voiceless consonant (as in (hopped, missed, tops, laughs)). There are grounds for taking the most widely distributed alternants of these suffixes, d and z (as in (stayed, days)), to be equivalent to their lexical shapes. The voiceless alternants can then be derived by means of an assimilation process. ‘Apart from this one assimilatory phenomenon, the set of sound sequences occurring across word-level affix boundaries is equivalent to that occurring at the boundaries between words. Many of these sequences, exemplified in (22), are unattested either morpheme-internally or across root-level boundaries. (22) (a) Word-affix (b)Cross-word th parent-hood parent who fa stiff-ness ifnone ml harm-less ‘come late ‘On the other hand, there are some sound sequences which can occur word- internally at the end of roots but which are impossible word-finally. This discrep- ancy is due to a set of consonant-zero alternations in which the zero form is24 Sounds and Words peculiar to the right edge of the word level. That is, as the following examples (23) illustrate, the consoaant in question fails to appear in absolute word-final posi- tion (column (ii)) or morpheme-finally before a word-level suffix (i (23) ® i it Root-affix Word-final Word-affix (a) 92 sifgnJature sign signing resilgn}ation resign resigning (b) mn da{mnjation dama damging conde{majation condem# —_—_condemying hy{malal hymna (©) mb bolmbjard bomb bomber crufmbjle crumb crumby A similar case in point isthe cluster ng. Inside simplex forms, 9 only ever occurs before a velar plosive, at least in most dialects.*? We thus find it immediately followed by g, as in (24a), but not by, say, a vowel, as the ungrammaticality of simplex forms such as those in (24b) demonstrates. (24) (a) filngler, atngler (b) ‘filgler, “algler {o)_lofag}-er, strofag er (lofgl-et, *strofg}-er) (4) sila}-er, tato]-er (e) lofolg, strotnlg, sitnlg ‘What is significant is that the distribution of ag is sensitive to the difference between root-and word-level affixes. This point can be illustrated by comparing two independent suffixes both spelt -er), the root-level comparative adjectival marker (as in (quick-er)), and the agentive nominal marker (as in ¢fight-er)). The distributional pattern found in root-derived forms is identical to that found in simplex forms, as a comparison of (24a) and (24c) shows. In the gg clusters of such forms, the velarity of the nasal is entirely dependent on the place of articulation of the following plosive, a pattern that is evidently related to the assimilation process already illustrated in (2). Inside word-derived forms, in contrast, 9 can occur pre-vocalically, including before the agentive suffix, shown in (24d). This pattern is related to the fact that the cluster 9 cannot occur in word-final position, where instead we only find 7, as in (24e). In the latter context, a potentially present g is thus suppressed. In exercise 1 at the end of the next chapter, we will have the opportunity to consider a detailed analysis ofthese and related facts. For the moment, all we need note are the different influences roots and words exert on the distribution of 9. The velar nasal cannot occur independently of a following velar plosive unless a word-level morpheme boundary follows, as in (24d, e). A root-level boundary in this context Sounds and Words 25 cannot support an immediately preceding 9. In this respect, root-derived forms can once again be seen to behave identically to morphologically simplex forms. 14.3 Morphological vs. phonological domains The discrepancy in the phonological behaviour of root-level and word-level ‘morphology raises the issue of whether domains that are morphologically and syntactically relevant necessarily coincide with domains that are phonologically relevant. That is, how much morpho-syntactic structure is visible to phonological processes? It is generally agreed that the word in English, irrespective of its internal morphological structure, constitutes a phonological domain. Beyond the word, it is clear that larger phonological domains are defined by phrase and sentence constituents, which play a role in regulating such phenomena as sentence accent and intonation. What is perhaps not so clear, however, is the extent to which it is possible to recognize phonological domains below the level of the word, Compounding is a relatively straightforward case. Since each component of a compound in English is by definition a word and since the word is well established as a phonological domain, it follows that compounds contain at least as many phonological domains as they contain component words. Thus a form such as Geagull), besides comprising a domain in its own right, also contains two inde- pendent sub-domains, (sea) and (gull). The bracketing that this domain structure implies is’ thus: [{sea] [gull]]. In this respect, the structure of compounds in English can be said to be analytic (using this term in a rather specific sense). That is, from the viewpoint of their phonological make-up, they are analysable into more than one domain. ‘The question is whether affixation also helps define phonologically relevant domains. It is generally agreed that word-level affixes involve analytic structure. This conclusion is consistent with the observations made above about the related ness of phonological phenomena occurring across internal word-level morpheme boundaries and across full word boundaries. Affixes are not generally assumed to constitute independent domains, however. So a form such as (quick-ly) can be taken to comprise two domains, one formed by the word (quick), the other by the ‘word (quickly); thus [{quick} ly]. To say that a particular phonological string displays analytic structure isto say that it contains a number of sub-strings, each defining an independent domain within which phonological processes can occur without needing to make refer- ence to material from some other domain. This means that, in a form such as, [quick] ly}, processes have in principle two opportunities to take place ~ once within the domain defined by (quick], and again within that defined by {quickly). In other words, phonological processing has the potential to proceed cyclically through successive domains. The form (god-li-ness) presents three cycles: [god],26 Sounds and Words [godly], and (godliness). (seagull) provides two: an inner cycle, on which [sea] and {gull] are separately processed, and an outer one constituted by [seagull] In the SPE framework and its closest descendants, rules are individually design- ated as applying at particular points in cyclic derivation. Some rules, for example, are permitted to reapply during the earliest cycles, while others are held ar hay until later cycles. Such stipulations are of course inconsistent with the position that processes take place freely wherever their conditions are met. In a model incorporating the latter principle, all phonological processing proceeds cyclically. That is, each and every domain is individually accessible to each and every process. Less clear-cut than word-level morphology is the case of root-level affixation. Under one view, the English root defines an independent phonological domain no less than the word. One corollary of this position is that every morphological domain in English constitutes a phonological domain. Nevertheless, the demon- strably different phonological behaviour of root- and word-level morphology has to be accounted for somehow. According to one variant of this approach, known 4s Lexical Phonology, all alternations are characterized in terms of phonological rules, which may apply either within the lexicon (where word formation occurs) ‘or outside the lexicon (where words are concatenated into phrases and sen- tences).*5 Within the lexicon itself, roor-level and word-level alternations are represented in terms of separate strata of rule application. Phenomena such as _Nelar Softening, Spicantization and Vowel Shift are designated as applying at Serarum 1, ¢ consonant-zero alternations in (23) are treated in terms of word-final deletion rules assigned to Stratum 2. Within the context of the present discussion, the significant point about the Lexical Phonology account is the assumption that the root level defines a domain for phonological processing. A root-derived form such as (im-mature) is thus deemed to have the same basic morphological shape as a word-derived form such as (un-nerve), viz. {im {mature]] and [un [nerve] ]. What differentiates them is that the former but not the latter is subject to the Stratum-1 phonological rule of degemination. Root-derived [{damn] ation] and word-derived {{dama] ing] are similarly structured, except in this case itis a Stratum-2 rule that distinguishes the two forms, namely consonant-deletion; hence (dam[nJation) but (dam[]ing). According to an alternative account, the root level in English does not constitute an independent phonological domain.** For reasons to be introduced below and to be expanded on ia the following chapters, this is the position adopted in this book. Underlying this view is the assumption that a potential mismatch can exist between domains that are morphologically relevant and those that are phonolo- gically relevant. Thas a root-derived form such as (secrecy) is deemed to be identical to a simplex form such as (lottery) in containing but one phonological domain, In this sense, root-level affixation is considered non-analytic; that is, root-derived forms are not phonologically analysable into their component mor- phemes. ‘One consequence of this position is that alternations which are restricted to the ro0t-level morphology are not treated in terms of phonological processes. That is, Sounds and Words 27 the root-level alternants of a non-suppletive morpheme are not derivable from a common lexical base by means of phonological processes but are listed in the lexicon. The relatedness of such alternants must then be captured by non-phono- logical means, for example through non-derivational lexical rules or by reference to some notion of proximity in lexical storage.#7 There seems little doubt that alternations such as Velar Softening, Spirantization and Vowel Shift were at one time active phonological processes, either at earlier stages in the history of English or in the languages from which Latinate forms were borrowed. However, accord- ing to the non-derivational account of root-level morphology, these processes are extinct in modern English; the alternations they gave rise to are, at least from a phonological viewpoint, no more than historical relics. The idea that root-level alternants, like morphologically simplex forms, are listed in the lexicon is consistent with the observation that the collections of phonological characteristics associated with the two types of form are indistin- guishable. It also squares with the fact that root alternations typically sustain lexical exceptions. That is not to say that we should ignore root-level patterns altogether. Recall that a subset of such phenomena, those illustrated in (17), clearly correspond to distributional regularities that are also evident in underived forms. Irrespective of whether or not we consider it appropriate to represent these patterns in terms of dynamic processes, there remains the question of how we account for the fact that such generalizations apparently do not hold of the word level. In fact, the bulk of the material to be considered in the next chapter consists of distributional regularities of this sort. ‘On the other hand we have root-level alternations, such as those in (18), which have very little if anything to do with distributional zegularities found in underived forms. There are at least two reasons why phenomena of this sort will barely figure in the following chapters. For one thing, they are stable across different dialects. This immediately reduces their interest in the context of this book, given the aim of holding minimal distinctions between phonological grammars up to the mirror of a principle- driven theory. Thus, even if morphologically simple and root-derived forms are represented differently in the grammar, this difference is located at a level that is inaccessible to the sorts of phonological phenomena we are interested in here. Where variability is observable at the root level, it is clearly morphological or lexical rather than phonological in nature. (An example is the reduced paradigm of strong-verb morphology evident in many dialects. Compare, say, standard (do-did-done) with non-standard (do~done-done).) This immunity to variability seems to reflect the special status of root-level morphology: itis typically associ- ated with the more learned vocabulary; it is acquired later than word-level morphology; and establishing connections between alternants at this level appears to be influenced by orthographic factors.** There is in fact a rather more mundane reason for not paying too much attention to specifically root-level alternations. They are among the most fre- ‘quently studied grammatical phenomena in English, and accounts of them are28. Sounds and Words widely available elsewhere in the literature. It would be a pity to devote too much space to them, at the expense of other material which might prove to be of more direct relevance to the concerns of this book. Not that removing root-specific alternations from consideration leaves us scraping the barrel for phonological tivbits. As T hope to show in the succeeding chapters, the word and phrase domains in English furnish us with a rich seam of data against which to assess theories of phonological processing.** Much is still to be learnt about the dialect differences that manifest themselves at these levels. Exercises Vowel contrasts ‘The data Each of the examples below illustrates a particular vowel contrast that ‘occurs in some dialects of English but not in others. In some dialects, for example, the first vowel in (ladder) is different from that in (madder). Each of the contrasts in question may be described as marginal, in the sense that the different vocalic variants are in near-complementary distribution. The only contexts in which the sounds directly contrast turn out to be distinguishable on the basis of morphological structure. ‘The data are arranged in rows, each containing words which illustrate a particular combination of phonological and morphological contexts. The task For the time being, we may set on one side the issue of how the qualitative differences between the vowel sounds in cach example should be represented. Nor need we be concerned with the issue of whether one vowel in each contrast can or should be derived from the other (or indeed whether both should be derived from some third source). “The focus here is rather on the contexts in which the different vocalic variants occur. (a) Identify both the phonological and morphological dimensions of the ‘contexts in question. (bh) What do the contrasts reveal about the visibility of different types of morphological domain to patterns of phonological distribution? I PAUSE-PAWS In vernacular London English, the nucleus of words such as (paw, sauce, thought, hawk, fawn) typically contains one of two main vowel variants: a centring diph- thong 92 and a closing diphthong ow.** (In this dialect, words such as (go, boat, bone) have aw, while those such as (cow, shout, crown) have zw.) The dialect is non-rhotic; that is, unlike in rhotic dialects, ris only ever pronounced before a Sounds and Words 29 vowel. (More on this in chapter 5.) The quality of the nucleus in words such as (pork, court, born), which are r-less in this system but r-ful in ehotic dialects, is identical to that in (though?) ‘ow thought, hawk, broad, board/bawd, pork ‘ow saucelsource, coarse, pause, gauze ‘ow yawn, lawn, call, torn, form 92 saw/sore, paw/pour/poor, bore 29 poorly, sawed, bored, paws/pours, yourn (‘yours, your one’), soreness 92 door knob, saw-tooth, draw bridge IE MOLAR-ROLLER, Another characteristic of vernacular London English is the occurrence of two qualitatively distinct vowels in the nuclei of words such as (go, boat, bone, foal): aw and ow.* aw go, toe, slow aw boat, road, slope, soak, brooch aw loaf, most, both, stove; rose aw bone, loan, home pw goal, hole, bowl, roll, told, gold, shoulder aw molar, Roland, cola ow roller, goalie, bowling ow coal effect, pole axe, goal area pw roll about, hole in, shoal of Il DazE-Days ‘Many varieties of northern Irish English have two vocalic variants in the nuclei of words such as (day, fate, fade, pain): rond e.** fate, staid, tape, lake, cater, baby face, faith, save, daze, lazy, station vain, game, fail, Daley day, stay, pay, ray days, stays, stayed, frayed, daily, greyness, playful ray gun, pay cheque, day time pay them, stay behind, pray for IV LADDER-MADDER The original set of words that contained historically short stressed & in English includes (bat, bad, pass, man). Many dialects of English bear the marks of a change30 Sounds and Words whereby this vowel has split into two main variants or reflexes, one lax, the other tense. Tensing is actually a cover-term for a range of developments, common to all of which is an inczease in duration. This is typically accompanied by one or more qualitative charges, involving amongst other things the appearance of an off-glide, hacking to a, or front-raising to £2, 2, or even higher In some dialects, tensing has led to a full-blown split in the original short-2 category, the tense reflex merging with back vowels from other historical sources (such as that in (calm, palm)). In this case, the result is that a word such as (gas) with « has a different vowel from, say, (pass) with a. In the type of dialect illustrated in the data below, however, the phonological contexts in which the tense and lax reflexes appear remain more or less com- plementary, the distribution being determined largely by the nature of the follow- ing consonant. In its broadest outlines, this pattern is widely represented in a range of dialects spoken in various parts of the eastern United States, southern England, Australia and Ireland, although the precise class of tensing consonants varies from system to system. The particular pattern illustrated here is one that is, ‘common to metropolitan New York and Belfast.** ereflex 1 lax bat, back, tap, hatch 2 lax shang, sang. 3 tense —_bad, ag, dab, badge 4 tense man, ban, damn, lamb S tense —_laugh, graph, path, pass, class, gas 6 7 8 lax ladder, wagon, dagger, adder (‘snake’) lax manner, panel, panic, damage lax placid, passage, saffron 9 lax vanity, sanity, opacity, classical, graphic, pallor 10 tense __madder, dragging, baddie, adder (‘one who adds’) 11 tense manning, damning, lambing 12 tense passer, laughing, classy 13. tense lag effect, gas emission, lamb enclosure 14 tense anit, pass around, bad adjustment ‘V TIDE-TIED ‘One characteristic of Lowland Scots and Scottish English is a phenomenon known, as Aitken’s Law or the Scottish Vowel Length Rule, according to which certain vowels show long and short reflexes in a manner that is determined by the following context.* The following Scottish English data illustrate how this pat- tern affects the three vowel classes represented by words such as (see, feed), (two, food) and (go, road). Scottish English lacks the contrast between o (as in (good)) and ur ((food)) found in most other types of English. Sounds and Words 31 1 shore feet, keep, leak, peach, meter boot, loop, Luke/look, hootch, stupid boat, soap, soak, brooch feed, greed, league, cedar food, rude, good, Cooder road, globe, rogue, ogre peace, teeth, leash, leaf, recent goose, tooth, hoof, lucid loaf, dose, both, grocer keen, mean, seam, penis ‘moon, soon, loom, Souness bone, moan, roam, bonus feel, wheel, steal, feline, Healy tool, foob/ull, bullet foal, hole, goal, molar 6 long see, tea, agree two, brew, blue 20, slow, row 7 long seize, teethe, leave, beaver lose, bruise, smooth, move, music rose, loathe, rove, Ambrosia 8 long _fear, peer, eerie poor, moor, lurid door, more, glory 9 long —_keys, keyed, agreed, freely, gleeful brews, brewed, stewed, truly, blueness rows, rowed, slowly, woeful, slowness 10 long _bee line, stew pot, snow drop 11 long three guitars, two bananas, go behind 2 shore 3 short 4 shore 5 short The effect of Aitken’s Law on the diphthong in words such as (die, fight, line) has additional qualitative consequences: the long reflex is ay, while the short is ay or the like. 12 2y fight, ripe, like tide, bribe, idle, spider tice, lif, rifle line, time, minor file, pilot, Reilly 13 ay tie, sigh tise, lithe, arrive, miser tire, biro 14 ay tied, ties, sighed, dryly, dryness tie pin, fly net fly beneath, tie down2 Constituency 2.1 Non-linear phonological representations ‘This chapter focuses on the question of how sequences of sounds are organized within phonological representations. ‘The standard SPE response to this question, recall, was to view a representation as consisting of a linear string of segments and morpho-syntactic boundary symbols, each segment being characterized as a matrix of binary-valued features (see 1.3.2). In this type of arrangement, features code two very different sets of facts. One set consists of those aspects of a segment’s identity that can be defined in purely phonetic terms. Features of this sort typically have a relatively stable articulatory or acoustic interpretation. They include such categories as (coronal, [continuant], [sonorant] and [high]. Features of the other type are essentially relational or structural in nature, in the sense that they code information about relations between segments within phonological strings; {syllabic], [stress] and [long] are prime examples. There is no such thing as a stable phonetic definition of [long], for instance. The distinction between a [4long] and a [-long] vowel cannot be expressed in terms of absolute temporal values (milliseconds, say); rather it manifests itself as a difference in the relative duration of the two segments when these are compared in identical contexts. ‘As noted in 1.3.3, she SPE view has since given way to a non-linear conception of phonological representations. One aspect of this is a recognition of the need to keep phonetic and relational information formally distinct. Features, it is now widely assumed, are to be reserved for phonetically definable aspects of repres- entation, while relational aspects are to be characterized in terms of hierarchical structure. The justification for this separation is straightforward: the phonetic and structural aspects are observed to behave independently of one another and are subject to different sets of organizing principles. In the next chapter, we will consider how the phonetically definable aspects of segments should be characterized. The focus of the present chapter is on two relational dimensions of phonological representations: quantity and constituency. ‘We will review some of the main arguments which support a hierarchical as opposed to a featural characterization of these aspects of phonological structure. At its lowest levels, the phonological hierarchy can be shown to be composed of syllabic constituents into which segment-strings are organized. Constituency 33 One finding that will emerge from our investigation of the phonological hier- archy is that a binary limit imposes itself on the amount of branching structure each constituent can support. One syllabic constituent is the nucleus, correspond- ing to the vocalic portions of forms such as (say, see, saw) (one nucleus each) and (Betty, city, window) (two nuclei each). Maximal binarity means that the nucleus can contain one or two segments but no more. As illustrated in (1), a one-segment nucleus corresponds to a short vowel (1a), a two-segment nucleus to a diphthong (1b) (or long monophthong). Anything larger than this, such as the ternary- branching structure in (1c), is il-formed. (1) f@) Nucleus (b) Nucleus (e), Nucleus 1 1 \ ry e ey teow In chapter 4, we will sce that the relation between positions within a branching constituent is asymmetric, a property that is reflected in the differing abilities of various segment-types to occupy particular positions. In this chapter, the coverage of constituency in English phonology is not meant to be exhaustive. In some areas, I present views which synthesize conclusions reached from a variety of theoretical perspectives. In other areas, where two or more currently held views are incompatible, I will present arguments which I feel favour one position over others. In some other cases, the present state of our knowledge does not permit the drawing of any firm conclusions on the superiority of any particular approach. In such cases, I will simply draw attention to prob- lems which for the time being remain unresolved. 2.2 Timing 2.2.1 Introduction We begin our examination of the phonological hierarchy by considering the dimension of phonological quantity. Quantity has to do with the amount of “space’ a segment occupies in a phonological string. Informally, the values on this dimension are often referred to in such terms as short versus long or light versus heavy. Like tone or stress, quantity manifests itself phonetically in a relational rather than an absolute manner. A phonologically long segment is realized with relatively greater duration than a short segment appearing under otherwise iden- tical conditions, just as a phonological high tone is realized on a relatively higher pitch than a low tone in the same context. The independence of quantity from other dimensions is exemplified in a number of phenomena, two of which we will concentrate on here. One isthe widespread process of compensatory lengthening. The other is the Jekyll-and-Hyde-like behaviour of a class of sounds whose quantitative and qualitative properties do not match up.34 Constituency 2.2.2 Compensatory lengthening ‘Compensatory lengthening is the name given to processes in which the loss of a particular segment is compensated for by the lengthening of an adjacent segment in the phonological string. By way of illustration, consider the fate of the voiceless velar fricative x in English, specifically as it appears in the historical sequence ixt, reflected in the spelling of such words as (right, night, sight, might). Something like the original pronunciation is retained in Scots: (2) Scos English arxt nayt, nayt, « (night) xt rayt, ray, (right) smuxt mayt,moyt,... (might) ‘A form in Scots such as nxt has the same sort of structure as, say, fist (fist); that is, we find a short reflex of historical i followed by a voiceless fricative. In most types of English, however, the velar fricative has been lost from this context, indeed in most cases from all contexts. If all that were involved here were the simple loss of a consonant, we would expect the general English pronunciation of (night) to be something like mt with a short 1 Instead, what we get is some kind of diphehongal reflex, as in nayt, nayt or the like. This leads us to conclude that, ‘at some stage in the pas: historically ix forms must have had a long i nucleus identical to that in words such as (bite, kite) (which never contained the velar fricative).* The reasoning behind this conclusion is this: the current diphthongal reflexes illustrated in (2) are the result of the historical Vowel-Shift change briefly mentioned in 1.4.2, and this only affected long vowels. The sequence of events in (night), for example, thus went something like this: nixt > nit > nayt. The first stage in this sequence illustrates the phenomenon of compensatory lengthening, In linear terms, we could try to account for the pre-Vowel-Shift part of the change by means of two ordered rules. One would lengthen i before x to produce nixt> nixt. The other would then delete x, producing ni:xt> nit. Note that these rules have to be ordered. Deletion must not be allowed to apply before lengthen- ings otherwise it would obliterate the context in which the latter operates.? There are at least two problems with this analysis. For a start, there is no historical or comparative dialect evidence to support an intermediate stage in which long co-existed with x in*nixe More seriously from a theoretical point of view, the two-rule account fails to capture the unitary nature of the process. The intuition is that there is an intimate connection between loss of x and lengthening of the i. It is as though the vowel somehow expands to fill s vacuum created by the disappearance of the consonant. But this connection is not established by the two rules, which are in principle quite independent of one another.” Under this analysis, we would expect one or other of the rules to occur independently in other systems. But there is no evidence to indicate that velar fricatives particularly favour vowel lengthening. Constituency 35 The compensatory lengthening that is evident in this and many similar cases in the world’s languages illustrates what we might call quantity stability. That is, the quantity dimension of a phonological string is in principle capable of remaining stable while the phonetic-quality dimension is in a state of flux. In order to be able 10 characterize this independent behaviour, we need to represent cach dimension fon a separate level. The way this notion has been implemented in non-linear theory is to separate quantitative information from other aspects of the repre- sentation and deploy it on an independent tier consisting of a sequence of slots or positions. Each position, usually indicated by x, constitutes a bare unit of phono- logical timing which, since it has no featural content, is often referred to as a skeletal point.* The term used to describe the level from which quantiative (and, as we will see, other relational) information is absent is the melodic tier. (The use of the term melodic stems from early non-linear studies of tone languages, in which suprasegmental pitch ‘melodies’ can be shown to exist independently of non-tonal aspects of a representation.‘) The synchronization of melodic units with timing positions is indicated by association lines, as shown in (3). (3) Skeletal tier x x Association lines | | Melodie tier nix et In the normal case, each melodic unit is associated to one timing slot. However, the independence of the skeletal and melodic tiers leaves open the possibility of setting up one-to-many associations. For example, we have the possibility of having a single melodic unit linked to two skeletal points. Since a multiply-linked melodic unit takes up more timing space than a singly-linked one, this arrange- ‘ment provides us with a straightforward way of representing the contrast between short and long vowels:# (4) (a) Short vowel (b) Long vowel 1 Vo Compensatory lengthening in the iat > i case van uow be accounced for in the following terms. The association between the velar fricative and its skeletal point is severed, as shown in (Sa).’ Then, as shown in (Sb), a new association is established between the now vacant slot and i. The former operation is known as delinking, the latter as spreading.” (5) (a) Delinking. (b) Spreading. FETT STE T TTP STIG oe A eo ee36 Constituency Spreading in this instance results in a doubly-linked i which, as we saw in (4b), defines a long vowel. There are two assumptions that have to be made in order for this account to work. First, the process of delinking a melodic unit from its position is not reversibles that is, a delinked melodic unit does not automatically relink to ite original position. Second, for a melodic unit to be realized at the end of a derivation, it must be anchored to a skeletal point.? Since the velar fricative in (5b) is unanchored, it fails to be pronounced. The second of these assumptions is derivable from a more general principle which we will discuss in chapter 4. Like the rule-based linear analysis, the timing-tier account of compensatory engthening involves two operations, in the latter case delinking and spreading. However, now a clearer connection is established between the two events. The spreading of ican only take place as a result of a vacant timing slot having been ‘made available through the delinking of the velar fricative. The notion of quantity stability is captured by the manner in which units on the timing tier are able to ride out changes that aifect units on the melodic tier. In capturing more ditectly the independence of the quantitative and qualitative dimensions, the non-linear account is superior to the linear. Nevertheless, at least in the form given here, the non-linear analysis cannot yet be considered to have established a causal connection between consonant deletion and compensatory ‘vowel lengthening. To approach that goal requires appeal to be made to certain additional theoretical considerations to be discussed in 2.4 and chapter 4. One problem, already alluded to above, is that delinking and spreading are in principle independent operations. Confirmation of this fact comes from numerous exam- ples where the loss of a consonant is not automatically adjusted for by the compen- satory lengthening of a neighbouring segment. A case in point is the truncation of final Cd clusters in English, discussed in the last chapter. The suppression of in a phrase such as (send two) is not accompanied by the lengthening of, say, the preceding 1. Evidently certain specific conditions have to be met before deletion of one segment is counterbalanced by lengthening of another segment. One generaliza tion that can be made about compensatory vowel lengthening, for example, is that it only occurs in systems already possessing long vowels." In a more technical sense to be explained in the succeeding sections, the syllabic nuclei of such systems must be capable of containing two timing slots, something that is by no means true of all languages. This suggests that compensatory vowel lengthening is sensitive to the manner in which skeletal positions are gathered into syllabic constituents. Related to this is the issue of the directionality of compensatory lengthening. We might ask why itis that spreads in (Sb), rather than the following f, or indeed even the word:-initial a, The first of these alternatives in fact corresponds to a real ‘option that is taken up in some languages. For instance, certain Latin two-conson- ant clusters show up in Italian as geminate (long) segments (e.g. nokte > note ‘night’), which can be analysed as a delinking of the first consonant in the cluster accompanied by spreading of the second. Determining whether a particular Constituency 37 language is likely to plump for rightward or leftward spreading in such cases again involves making appeal to aspects of phonological constituent structure. However, the other alternative in which some non-adjacent melodic unit spreads into a vacant skeletal slot (for example, the m in (Sb)) seems on the face of it implausible. In any event, we would want to exclude it on the general theoretical ‘grounds that it contravenes the principle of locality. The question is whether there is anything in the nature of a multi-tiered representation to rule out the crossing of association lines that non-local spreading of this type implies: (0 | | ‘Two properties of non-linear representations express relations that have a direct, bearing on this issue: precedence and temporal overlap. Relations of precedence are expressed in the linear order of units on each tier: in (6), x1 < (precedes) x2 < x3 < x4 on the skeletal tier, while n< i< ¢ on the melody tier. Each association line meanwhile represents a situation in which the duration of a melody unit overlaps at some point with the duration of the skeletal point to which itis linked. ‘The interaction between these two properties rules out line crossing. Its logically impossible for a melody unit to overlap with two discontinuous skeletal positions while simultaneously maintaining its precedence with respect to a melody unit attached to an intervening position. Inherent in the ill-formed representation in (6) is the contradiction that all of m precedes j while part of i precedes n.4 2.2.3 Diphthongs and affricates Let us now take a more detailed look at the bilinear representation of long versus short vowels shown in (4). A long vowel can have either a monophthongal or a diphthongal quality. The proportion of diphthongs to long monophthongs in English varies from dialect to dialect. Some systems, such as Scottish English, have a full range of long monophthongs (for example in (see, say, Shah, saw, 50, t0o))s some have nothing but diphthongal realizations: others fall somewhere in be- tween. From the viewpoint of phonological quantity, however, this variation is insignificant. This is because long vowels in English, irrespective of whether they are monophthongal or diphthongal, belong to a single natural class. The unified nature of the class is revealed in the distributional characteristics of vowels as well as in their participation in particular processes. For example, unlike short vowels, iphthongs and long monophthongs can occur in word-final stressed open sylla- bles. Hence variant forms such as ga/gow (go), sefsey (say)."® Compare these with ungrammatical final-stressed forms with short vowels such as *s1, *se. If we assume that diphthongs (at least of the English type we are concerned with here) ‘occupy two timing slots, it is not difficult to see that the class of long vowels can38. Constituency be distinguished from short on the basis of the skeletal tier. Diphthongs and long monophthongs both occupy two timing slots, short vowels only one: m Long (b) Shore Monophehong Diphehong Vr It ' © ey < The behaviour of long monophthongs illustrates a class of segments that display 1 dual identity when viewed in terms of their quantity and quality properties. They are qualitatively simplex but quantitatively complex. These properties are straightforwardly captured in the bi-linear representation shown in (7a). Long e is simplex from the viewpoint of the melodic tier but complex from the viewpoint of the skeletal tier. The same type of behaviour is exhibited by ‘true” geminate (long) consonants. Such geminates are true in the sense that they contrast lexically with non-gemin- ate (short) consonants, as in Italian grato'T scratch’ versus grato ‘grateful (masc.)’ “Fake” geminates are of the type we encountered in 1.4.2; that is, they are pairs of identical consonants which accidentally occur together as a result of the juxtapo- sition of two morphemes, as in the an of English (un-nerved). For quantitative purposes, true geminates are to two-consonant clusters what long monophthongs are to diphthongs. For example, true geminates count as two consonants for purposes of syllabification. Syllabification has to do with the location of syllable boundaries within polysyl- labic strings of segments. In this book, the term will not be used to refer to the sort of activity in which speakers are asked to make judgements about where syllable boundaries lie, for example by getting them to utter polysyllabic words in separate and allegedly syllable-sized chunks. For reasons to be expanded on below, elicitation methods of this type can be quite unreliable. Rather, as the notion is employed here, syllabification is to be established purely on the basis of empirically verifiable phonological evidence. As the discussion unfolds, we will see that syllabification is subject to quite general principles. This raises the ‘question of whether syllable structure is inherently present in phonological repres- centations or is built up trom scratch by means of some kind of dynamic algorithm. ‘We will take up this matter in 2.5, once we have had a chance to see what the main principles are. For the time being, we may simply assume that syllable structure is in place and forms a backdrop against which other phonological phenomena are to be examined. As far as the syllabification of true geminates is concerned, they behave exactly like VG.GV clusters (where C; and Gare different consonants and “’ indicates a syllable boundary). That is, the first part of a geminate in a VC:V sequence closes the first syllable. Returning to the Italian example, we may note that grat.to ‘I scratch’ is syllabified in the same way as a form such as ven.to ‘wind’. There is Constituency 39 plenty of phonological evidence to support this syllabification. In standard Ita- lian, for example, the vowel of an open syllable (that is, one not closed by a consonant) is lengthened when stressed; hence gra: to ‘grateful’."* No such leng- thening occurs in closed syllables, something which applies as much to geminates a5 to consonant clusters straddling a syllable boundary (*gra:o ‘I scratch’, * vemto). ‘The parallelism between clusters and true geminates is straightforwardly cap- tured in terms of their identity on the skeletal tier. Compare the representations in (8) with those in (7) (8) (a) Long (b) Short Geminate Cluster xx ox x x \r 1 I a nod 1 Having discussed a configuration in which a single melody unit is linked to two timing slots, we can now consider how the bi-linear approach makes available another possible configuration, one in which the opposite pattern of association prevails - two melody units linked to a single timing slot. This possibility can be exploited in order to characterize another class of dual-identity segments, those which are the converse of geminates in being qualitatively complex but quantitat- ively simplex. This class is illustrated by the affricates ¢ and j. As the alternative transcriptions if and ds indicate, an affricate consists of a stop portion followed by a fricative portion and is thus qualitatively similar to any homorganic stop- fricative cluster. (The two types are not necessarily identical, though. Compare the medial consonants of, say, (recharge, catch it) on the one hand with those in {pet shop, cat shit) on the other.) Quantitatively, however, affricates can be shown to line up with simplex consonants rather than with clusters. ‘We can check the quantitative behaviour of affricates by considering how they interact with stress assignment. The placement of word-stress in English and many other languages is conditioned by the manner in which segments pattern into constituents which, for the time being at least, we can informally label syllables. Specifically, stress in such languages is what is known as quantity-sens- itive; that is, its appearance on particular syllables in a word is sensitive to whether they are heavy or light. A heavy syllable contains either a long vowel (monophthong or diphthong, which we can symbolize as VV) or a short vowel and a syllable-closing consonant (VC). A light syllable contains a short vowel (V).. For the purposes of calculating the weight of a word-final syllable in English (and many other languages), final consonant is disregarded, a phenomenon known as ‘extea-metricality (to be considered in more detail in 2.4.4).!* Heavy syllables tend to attract stress, a pattern that is evident in English verbs. ‘The general rule for English verbs is that stress falls on the final syllable if itis heavy; otherwise the penult receives stress. This is illustrated by the following forms:40 Constituency (0) (b) ) tormént —_cajle édit lamént —maintéin—_ast6nish collépse caréuse cancel In a verb such as (tormént) (9a), with final extra-metrical 4 the last syllable men is heavy and thus receives main stress. Similarly, the final syllable of the forms in (9b) is stressed by virtue of being heavy, in these cases as a result of its containing a long vowel. On the other hand, in (édit) (9c), with extca-metrical ¢ stress falls ‘on the penult, since the final syllable (spelt (di) is light. ‘We now have a straightforward way of testing the quantity of ¢ and j. The test involves polysyllabic verbs that end in a short vowel followed by an affricate. If affricates are quantitatively complex, stress should be assigned as in (9a). Only the second (fricative) portion will be extra-metrical, yielding a word-final se- ‘quence of the pattern VaG3)em (where { Jem indicates extra-metricality), The final syllable will be heavy (closed by the stop portion) and should thus receive final stress. If, on the other hand, we are dealing with a quantitatively simplex conson- ant, stress should be assigned on a par with the forms in (9c). That is to say, the whole affricate will be extra-metrical, the final syllable will be light, and stress should be penultimate. We find the answer in (10): : (10) ménage pillage damage forage encourage ‘These forms line up wich the pattern in (9c). In other words, j behaves like a final simplex consonant." In order to express the notion that affricates are qualitatively complex but quantitatively simplex, we can represent them in terms of a contour structure: that is, one in which two melodic units are linked to a single skeletal position, as, depicted in (11). ay x r\ co (Like melody, the term contour is borrowed from non-linear analyses of tone. A contour tone consists of two level tones attached to a single tone-bearing unit (cypically a vowel). A falling pattern, for example, is represented as a High tone followed by a Low.)"” We can now assume that extra-metricality operates on the skeletal level, rather than on the melodic level: (2) x x x x Gem Pretty mezn i d 3 (ménage) Constituency 41 In this way, we capture the fact that, in spite of their phonetically dynamic nature, affricates function exactly like simplex consonants for quantitative purposes. In (12), the whole of the affricate is extra-metrical, with the result that the final syllable is light and thus fails to attract stress. In the SPE framework, it was difficult to capture the dual identity of sounds which display a mismatch between their quantitative and qualitative properties. One linear way of characterizing what we now think of as contour segments was in terms of a feature such as (delayed release] for affricates. This arrangement correctly identified such segments as quantitatively simplex, but at the expense of introducing dynamic feature definitions. These were anomalous in a framework in which quality features otherwise described steady-state phonetic characteristics {coronal}, {labial}, {high}, for instance).'* One way of avoiding this anomaly would be to split the representation of such sounds into two matrices, each containing a steady-state feature which specified the beginning and end points of the dynamic articulation. In the case of affricates, for example, this would mean [-continuant] for the stop-closure matrix followed by [+continuant] for the fricative-release matrix. Unfortunately, the two-matrix treatment suggests that affricates should behave quantitatively like two-consonant clusters, which, as our brief review of the stress facts indicates, is incorrect. By splitting the quantity dimension off from other aspects of the representation, we avoid being caught between two stools in this way. As our discussion of compensatory lengthening revealed, the move also allows us to account for quantity stability in a steaightforward manner. Furthermore, the non-linear treat ment of quantity has an important theoretical consequence for the representation of suprasegmental phenomena in general. In a linear framework, any feature is in principle capable of being implicated in stress assignment. In natural language, however, the only systems in which stress is partly or wholly dependent on aspects of segmental structure are those described as quantity-sensitive. In SPE terms, as ‘we will ee in a moment, this is reflected in the incorporation in stress-assignment rules of features such as [long] or [tense] or those needed to characterize the contrast between C and V. But, according to the internal logic of the framework, wwe should not be surprised to encounter systems in which stress placement is sensitive to any feature or feature combination {low}, [continuant], or [glottal suction}, say. Significantly, robust evidence supporting the existence of such systems is hard to come by. ‘With the abandonment of featural representations of the quantity dimension, we now have the possibility of greatly restricting the predictive power of the stress-handling component of the theory. We can now assume that stress place- ment is completely blind to the featural make-up of segments. Instead only structural aspects of the representation can be scanned by stress processes. So far the only structural dimension I have attempted to formalize in any way is the skeletal tier. Informally, however, I have been referring to the role of syllabic structure in quantity-sensitivity. This indicates that positions on the skeletal tier are not concatenated in linear sequence but are gathered into constituents. It is to the issue of how this notion is formalized that we now turn.42. Constituency 2.3 Phonological constituent structure 2.3.1 Introduction The notion most familiarly applied to the constituent dimension of phonologi- cal representations is the syllable. The concept has a long history in phonological scholarship, although within generative theory it has not always been as firmly established as in other traditions. In SPE, the nearest we get to its recognition is in the feature [syllabic], designed primarily to characterize the contrast between, for example, vowels ([+syllabic}) and glides ({-syllabic}). From the early 1970s ‘onwards, however, the rivotal role of constituency in phonological structure has come to be acknowledged more and more in generative theory.” Today there is near-universal agreement that syllabic structure is an integral part of phonolo- gical representations. However, it would be fair to say that there remains less than general agreement about the precise nature of that structure. In the following sections, we will consider some of the arguments which suggest that certain types of phonological phenomena are more adequately characterized in terms of constituent structure than in terms of linear segmental strings. The cevidence is of three main types, involving the statement of (a) suprasegmental patterns; (b) phonotactic constraints (i.e. constraints on the combinability of segments in sequence); and (c) the domains in which certain segmental processes ‘operate. The focus of this chapter is on the first two types of evidence, but at various points I will make reference to the third. Actually, most of the arguments based on the evidence to be reviewed below turn out to support the existence of phonological constituency rather than the notion syllable per se. 2.3.2 Stress and constituency In our discussion of English verb stress, we have already begun to see the role played by constituent structure in suprasegmental processes. In order to illustrate the superiority of a constituent-based treatment of stress over a linear account, we will now look at the phenomenon of quantity-sensitivity in more detail Consider the following facts relating to primary stress placement in a large class of English nouns: (13) (a) ) © (a) balléon horizon agénda_-—cinema domdin aréma _—verdnda_—_andiiysis ‘muséum amalgam = América aréna uténsil Canada Let us try to characterize the general pattern that is evident in these forms in linear terms. In each column of (13), the stressed vowel (underlined in (14)) is located in the following respective contexts: Constituency 43 (14) (a) WV (C) In: stress the rightmost vowel if it is long, regardless of ‘whether or not the noun ends in a consonant. (b) VV (C) V (C) ]h: stress the penultimate vowel if it is long and is followed by zero or more consonants followed by a short vowel, regardless of whether or not the noun ends in a consonant. () VCC V(C) In: stress the penultimate vowel if it is followed by two consonants followed by a short vowel, regardless of whether or not the noun ends in a consonant. (4) V (©) V (©) V (C) Iw: stress the antepenultimate vowel if this is followed by two short vowels, with or without intervening single consonants, regardless of whether or not the noun ends in a consonant. It is possible to collapse these conditions under a single linear stress-assignment rule, the detailed formalisms of which need not concern us here." ‘The statements in (14) can be simplified if the recutring condition ‘regardless of whether or not the noun ends in a consonant’ is replaced by the declaration, already appealed to in relation to verb stress, that final consonants are extra- metrical. However, the linear account remains inadequate for various reasons. Not the least of these is that it fails to capture the intrinsic connection that exists between the length of a vowel and some of the consonants in the string. For example, it fails to identify the equivalence of conditions (14b) and (14), in which, for the purposes of attracting stress, a long vowel ‘counts the same as’ a short vowel bracketed with a following consonant. ‘The linear account is essentially a collection of circumlocutions referring to conditions that can be much more straightforwardly formulated in terms of constituent quantity. As a preliminary formulation in these terms, all we have to say is that stress in English nouns falls on the rightmost heavy syllable and otherwise on the antepenult. (By default, it will fall on the penult in words consisting of two light syllables, as in city).) We have already seen how we can express the quantitative equivalence of long vowels and certain VC sequences in terms of an identical two-point structure on the skeletal tier. But not all VC sequences count as heavy; the important point is that the C in a heavy sequence must be bracketed with the preceding V. So we need to impose some type of constituency on the skeletal tier. In (verénda) (15a), for example, the bracketing of n in the same syllable asthe preceding vowel means that the stressed syllable contains the same number of timing points as that in, say, (aréma) (15b): G5) @) od & x od ed Prot tr td vo cen do (b) fk} ok x xd fk x) | ! |44 Constitency How many timing slots does a syllable have to have before it qualifies as heavy? ‘And do all slots in a syllable contribute co its weight? The stressed syllables in (15) both contain three timing points. But comparison of these forms with, say, (payéla) reveals that a consonant does not necessarily have to be present in the initial portion of a syllable in order for the syllable to count as heavy: (16) ok x xd od ed trrrt tl pey ow lo To calculate weight, it seems, we only need to take into account the vowel and whatever follows. Light syllables are these in which only one stot is occupied by the vowel and no other material follows. The final syllables in (15) and (16) are of this type and thus fail to attract stress. Allowing for extra-metricality, we find the same final light pattern in a form such as (horizon): 7) fe xd ke x fe Xd Gen rhorirtt he rayzon ‘The fact that weight is measured without regard to whatever precedes the vowel within the syllable is confirmed when we increase the number of syllable-iniial consonants beyond one. The second syllable of, for example, (Katrina) is no heavier than the stressed syllables of (aréma) or (payéla), even though it begins with two consonants. Take a form such as (élgebra), syllabified as in (18a). 1B (18) @) eat d oe non | el (b) fxd x x x) & xd rrrt dd 2 freko “o¥ The final syllable of (algebra) has three skeletal slots, as many as the heavy syllables in (15). However, it counts as light, as is evident from the fact chat it fails to attract stress. Instead, the word is stressed like those in (13d), i.e. on the antepenult, in accordance with condition (14d). For a similar reason, (Africa) (18b) also bears antepenultimate stress in this case, iti the second syllable of the word that contains CCV but still counts as light. “Thus, in establishing the weight of a syllable, we only count material appearing in a span which runs from the left edge of the vowel to the right edge of the syllable. In the data we have looked at so far, if this span contains two timing points, the syllable is heavy; if it contains one, the syllable is light. Quantity facts Constituency 45 such as these have been taken as evidence supporting the existence of two sub-syllabic constituents: the rhyme, containing those positions which determine weight, and the onset, containing those positions which precede the rhyme and are irrelevant to the determination of weight.” When we say that stress assign- ment is more adequately characterized in syllabic than in linear-segmental terms, we are thus using the notion of syllable in a very loose sense. The quantity facts support the onset-rhyme dichotomy but are silent on the question of whether it is necessary to recognize the syllable as an autonomous constituent. When we speak of syllable weight, we are thus really referring to chyme weight. 2.2.3 The ‘syllable’ Throughout this book, we will have many opportunities to examine diverse phonological phenomena in English which are clearly sensitive to constituent structure. None of them, it turns out, needs make reference to an independent node directly corresponding to the syllable, at least as this notion is usually understood in the western tradition of versification. Whether this means that such. 4 constituent has no place in Universal Grammar whatsoever is another matter and depends naturally on whether motivating evidence is to be found in languages other than English. All that need be said at this point is that, for our present Purposes, we can manage quite happily without a syllable node.%* We will have ‘occasion to return to the issue of its theoretical status later in this chapter as well as in chapter 4. The term syllable is in any event potentially misleading in phonological dis- course, since, in spite of what is commonly believed, the notion has no pre-theoret- ical standing. One reason we should be wary of employing the notion is that its status as a metalinguistic term in literate societies derives almost exclusively from conventions governing written language and versification. And these conventions vty, often quite widely, according to different orthographic traditions. The term can be formally taught as a means of labelling some aspect of phonological reality, but it is by no means always obvious exactly what that reality is. In alphabetic traditions, calculating the number of syllables in a word typically involves identifying the number of peaks of perceptual prominence, each typically associ- ated with a vowel. In some other traditions, by contrast, any VC sequence is deemed to straddle a syllable boundary, even when the C in question is not followed by an audible vocalic peak.®* In a CVCV word, the counts by both methods will converge on two: CV.CV (where the full stop indicates a syllable boundary). However, a CVC word will be rated as monosyllabic by the first method (CVC,) but as disyllabic by the second (CV.C.). Moreover, even within one and the same alphabetic tradition, there can be disagreements over the location of syllable boundaries in polysyllabic words (as the notoriously variable conventions regarding the placement of line-breaks in the printing of long words testify). As the discussion in this chapter unfolds, it will become clear that constituent nodes in the phonological hierarchy can be labelled46 Constituency without reference to the syllable. Nevertheless, I will occasionally employ the term in an informal sense where its use is unlikely to give rise to ambiguity. Having established the onset (O) and the rhyme (R) as independent nodes in the phonological hierarchy, our next task is to determine their internal structure. The weight of a rhyme is straightforwardly indicated by the number of positions it contains. Employing a tee-structure representation of constituency, as in syntax, wwe can establish the weight of a rhyme on the basis of whether or not it branches. In the following forms, the medial rhymes are branching and thus heavy (hence stress-attracting), while the initial and final rhymes are non-branching and thus light. 1) oR OR OR R OR OR rrooi ary tt rorya it trorir out rorrdorl vo ren do 2 row ma We have now established the outlines of a rudimentary hierar entation in which skeletal positions form the terminal nodes of onset and chymal constituents. ‘A number of questions arise in connection with this mode of representation, three of which will come up for discussion in this chapter. What constraints are there on the number of skeletal positions that can occur within a constituent? ‘What constraints are there on the type of segments that can occupy these posi- tions? And is there any evidence to suggest that these constituents have any internal hierarchical organization? The last question relates primarily to the hyme, since it is sometimes assumed that this constituent is itself composed of ‘wo sub-constituents: a nucleus (usually but not always containing a vowel) and a coda (containing post-vocalic consonants). Informally speaking, this implies the following structure:”” (20) Rhyme on) Nucleus Coda The question which we will return to below is whether either the nucleus or the coda or both need to be recognized as independent nodes in constituent structure. 2.3.4 Phonotactics So far, the main motivation we have seen for the skeletal tier has to do with its function as a means of coding phonological quantity. If this were its sole function, there would be good grounds for confining it to rhymes, given the non-participa- tion of onsets in the determination of the quantity relations examined up to now. However, we have clear evidence that skeletal positions have at least one other Constituency 47 important role to play in phonological representations: in so far as they form the terminal nodes of syllabic constituents, they help define the level at which phonotactic relations are established. Phonotactic constraints place restrictions ‘on the ability of particular segment-types to stand next to one another. And there is cound evidence that onsets are subject to these restrictions no less than rhymes ‘As briefly illustrated in 1.2.2, it is a fundamental property of phonological systems that systematic gaps exist in the distribution of different segment-types at different points in phonological strings. These phonotactic lacunae can be viewed as reflecting the reduced capacity of particular positions in constituent structure to support a full range of melodic contrasts. What we typically find is that such positions occur next to other positions which enjoy a fuller contrastive potential. The skewed distributional patterns thus involve adjacent segments, where ad- jacency is established on the basis of the constituent contexts occupied by skeletal positions. Phonotactic asymmetries of this sort do not occur randomly across’ phonolo- gical strings; rather they are systematically restricted to certain contexts ~ call them phonotactic domains ~ which are defined in terms of constituent structure. ‘Three domains which are most often cited as displaying phonotactic constraints involve adjacent positions in the following contexts: (21) Phonotactic domains {a)_within an onset (e.g. plin (play))s {b) within the nuclear portion of the thyme (e.g. ow in (go)); and {e) within a cluster composed of a coda followed by an onset (e.g. at in (winter)). We will examine each of these contexts in some detail below. By way of a preliminary illustration of the distributional patterns involved, consider the fol- lowing facts. ‘Within a two-position onset or nucleus, the distribution of segments in each slot is tied up with that of its sister slot. The example of (play) in (21) illustrates the classic branching-onset pattern in which the first position is filled by an obstruent and the second by a liquid or glide; hence grammatical pl, kr tw-and the like (as in (play, ctow, twelve)) but ungrammatical *km., *ps-, *bt, etc. (In fact, as we will see later, this is probably the only type of branching onset permitted univer- sally. This, in spite of appearances to the contrary, such as the initial cluster of a form like (splay).) On the other hand, when two positions are separated by an onset-rhyme boundary, we typically find that they are phonotactically inde- pendent of one another.» A single onset position in English displays a near-max- imal inventory of consonantal contrasts, disregarding to a large extent the nature of the vowel that occurs in the following chyme. This independence provides further justification for the onset-rhyme split ‘Within a branching nucleus (21b), the distributional possibilities in the left- hand position are much greater than in its sister. The right-hand slot can either be filled by an off-glide, specifically y, w or 9(as in ey, aw, ia); or, as in the case of48 Constituency a long monophthong, it must be identical to its sister on the left. By contrast, the vocalic content of the left-hand slot is in principle unrestricted (although the particular set of contrasts varies from dialect to dialect). Some of the phonotactic patterns evident in the cross-constituent sequence of a coda-onset cluster (21c) are the reverse of those found within branching onsets. Thus, while obstruent-liquid makes for a well-formed onset, liquid-obstruent belongs to the set of well-formed coda-onset clusters. Compare onset ki, 7, trin, say, (clay, flay, tree) with coda-onset /k, If tin, say, (falcon, dolphin, quarter). (The 11 sequence is of course only found in rhotic dialects, those which allow constricted r to appear before consonants. More on this in chapter 5.) Other coda-onset clusters are of course possible, including nasal-obstruent (as in (winter), fricative-plosive ((mister)) and dual oral stops ((chapter)). In forms of the shape represented by (winter), itis the onset ¢ that partially determines the identity of the coda n to its left. This particular example illustrates one of the systematic phonotactic constraints that operate intramorphemically in this con- text: a coda nasal must be homorganic with a following onset obstruent. Hence the grammaticality of forms such as (pamper, winter, a[n}chor) and the ungram- maticality of clusters such as *-gp- and *-mk-. Another example involves the impossibility of ¢ appearing in the coda position of such clusters. Compare ill-formed *-tp., *-tk- with well-formed -pe., -kt-(as in (doctor, chapter)).”” ‘Our initial motivation for recognizing the skeletal and constituent levels of representation was based exclusively on quantitative considerations. The fore- going brief preview of English phonotactics introduces an independent source of additional supporting evidence. In tandem, these two representational levels define the conditions of adjacency under which phonotactic constraints operate. That is, restrictions on the ability of different melodic units to occur next to one another are stated in terms ofthe skeletal positions to which the units are attached and the place occupied by these positions in constituent structure. However, unlike quantitative relations, chose involving phonotactics are not specific to the rhyme; rather they are defined both within and between onsets and rhymes. This observation indicates the need for a level of representation that is common to both types of constituent. The skeletal tier, already independently motivated by quantitative facts, fits this bill.”” ‘When we think more about the coda-onset phonotactics just mentioned, appat- ent counterexamples might spring to mind. At first sight, the embargo ou the occurrence of ¢ in codas, for instance, would seem to have been lifted in a form such as (catcall) It is significant, however, that the ¢C sequence in this and similar examples contains an intervening word-level morpheme boundary. The phono- logical shape of such forms testifies to their word-level structure, even if the morphology in question ceases to be synchronically active. This is particularly true of proper names, as in the pronunciation of (Aitken) with -tk-. The variant eykon shows the effects of phonological restructuring, inline with what would be expected of a morphologically underived form (cf. {bacon)). For reasons touched on in 1.4 and to be expanded on presently, a pair of consonants flanking @ Constituency 49 wordslevel morpheme boundary are not truly adjacent and thus cannot be reckoned to reveal systematic phonotactic interactions. The general issue at stake here is this: defining adjacency in terms of the constituent status of skeletal positions presupposes a determination of the relev- ant morpho-syntactic domains within which this constituency is established. In ‘other words, we have to take into account the relation between segment strings in syllabic constituents on the one hand and segment strings in morphemes, words, phrases or sentences on the other. In the case of English and many other lan- guages, itis not usual to take the sentence or phrase as the relevant domain within which the basic set of possible constituents is established. Thus, a word-final consonant would not generally be considered to form a constituent with a following word-initial consonant. This fact is reflected in the lack of any system- atic phonotactic dependency between the two positions. The sequence s-Bin (Tess thinks), for example, does not manifest any systematic phonotactic interaction; rather itis the accidental result of the juxtaposition of two morphemes. Below the phrase level, itis often assumed that the word domain has a privi- leged position in the establishment of phonological constituency. According to this view, the set of possible syllabic constituents in a given language can be determined by simply inspecting word structure. In particular, it is frequently taken for granted that well-formed onsets and rhymes can be identified on the basis of segment sequences at word edges. That is, the set of well-formed onsets supposedly coincides with the set of word-initial consonant sequences, and the set of well-formed rhymes with the set of word-final sequences consisting of the rightmost vowel and any following consonants." From a general theoretical point of view, the claim that syllable structure is equivalent to word structure, if correct, actually constitutes a very good argument for not recognizing syllable structure as an integral part of phonological repres- entations at all. If, for example, phonotactic conditions stated at the syllable level simply duplicated those stated at the word level, it would follow that one of the levels is superfluous. Since there is independent morpho-syntactic motivation for word structure, it is clear which level could be dispensed with. This is essentially the position adopted in SPE, where all phonotactic constraints are formulated in terms of morpheme-structure conditions. But one of the main arguments leading to the rehabilitation of the syllable in generative phonology hhas been precisely that phonotactic constraints cannot be adequately expressed without reference to syllable structure as an independent dimension of phonolo- gical representation. There is a further, this time empirical problem with the position that syllable structure can be directly read off word structure. This concerns the observation that there exist systematic asymmetries between the distribution of segmental sequences at the edges of words and within words. In particular, as we will see in the next couple of sections, the set of word-internal segmental sequences is much smaller than that found at word edges. This discrepancy partly reflects the fact that not all segment sequences which are possible in morphologically complex words are also possible in single morphemes. Because of the central role played50 Constituency by suffixation in English, the mismatch is especially evident word-finally. For example, the word (six-th-s) contains the right-edge sequence -ks6s, with one ‘morpheme boundary intervening between the first s and the 8 and another between @ and the second s. If this sequence is assumed to constitute a coda, we have to explain why codas of this shape never occur morpheme-internally. The question then is whether the cross- morpheme sequence s@ in (sixth) provides any more or less evidence of systematic phonotactic interaction than the same cross morpheme sequence in (Tess thinks). As we will see presently, there is good reason to conclude that we are dealing with arbitrarily juxtaposed segments in the former case no less than it the latter. One consequence of this conclusion is that segments in a word-internal sequence cannot automatically be considered ad- jacent in terms of constituent structure if they are split by a morpheme boundary. Moreover, as we will sec, the strings that constitute the difference between word-edge and word-medial sets of sequences display patterned behaviour which has to be accounted for in some way of another. There can be two responses to the distributional discrepancy between mor- pheme-level and word-level segment sequences. One alternative is to continue with the basic assumptior. that syllabic structure can be directly read off word structure, in which case the over-generation of syllable types in morpheme-internal contexts somehow has to be staunched. The other is to determine the set of possible constituent structures purely on the basis of morpheme-internal sequences and then, if necessary, make special provision for the distributional peculiarities of word-edge strings, In the next two sections, I will demonstrate some of the problems inherent in the first approach and rehearse some of the main arguments that have convinced many phonologists of the correctness of the second. Let us explore the phonotactic discrepancy between intra- and cross-morphemic sequences in more detail. In order to do so, we need to take into account the distinction between root-level and word-level morphology referred to in 1.4. As noted there, from the viewpoint of phonological structure, root-derived forms are non-analytic in the sense that they are indistinguishable from morphologically simplex forms. On the other hand, forms derived at the word level are analytic in the sense that their phonological shape more often than not advertises theit morphological complexity. This distinction is particularly striking in the realm of phonotactics. As illustrated in the examples in (22) (some repeated from the last chapter), word-level forms frequently contain segment sequences which are either ‘marginal or impossible in underived and root-level forms but which coincide with cross-word sequences found at the sentence level. (22) Compound _Phrase/sentence level pm lapmarker stop me vet — dove tail live to @6 — moth ball path belongs Word-suffix _Phrase/sentence level th — parenthood parent who Constituency $1 ben drabness drab neighbourhood vl lively live lynx In the context of the present discussion, the immediate relevance of the distinc- tion between analytic and non-analytic forms is that it helps delimit the morpho- syntactic domain within which the set of well-formed phonological constituents is appropriately identified. We are proceeding on the assumption, remember, that the existence of phonotactic dependencies between neighbouring segments pro- vides evidence of adjacency in constituent structure. The distributional inde- pendence of segments occurring either side of an analytic morpheme boundary suggests that word-level derived forms fall outside the domain within which the basic set of well-formed constituents is established, at least in English.!* The major systematic phonotactic constraints of English operate within non-analytic domains; and it is for this reason that the following discussion of phonological constituents is restricted in the first instance to underived and root-level forms. It will still of course be necessary to say something about how segment sequences occurring at word-level boundaries are integrated into phonological structure. Later, however, I will argue that this integration does not involve constituents which straddle analytic domains. In other words, two segments separated by a word-level boundary are never syllabified within the same constituent. A factor to bear in mind when establishing the relevant data over which phonotactic restrictions are defined is that the analyticity or otherwise of a form does not necessarily remain stable over time. For example, as a result of one well-known type of historical change, a compound can become reanalysed as a morphologically simplex form. This is usually accompanied by a phonological restructuring whereby the form loses its original analyticity, a development that is not always reflected in the spelling. Examples include (bosun) (from (boat- swain)), (gospel) (from Old English (g6d) ‘good’ plus (spell) ‘message’) and forad (forehead). Historically, these words contained the cross-morpheme sequences &-5, d-s and rh (the last still evident in the variant ferrhed) which never occur intra- morphemically. Now, however, they exhibit the same kind of non-analytic struc ture as we find in forms such as (basin), (cascal) and (torrid). (On the other hand, as noted briefly above, some forms continue to bear witness to historically analytic structure, even though they no longer participate in productive morphological alternations. This is particularly true of many proper ‘names. Originally compound names such as (Sopwith, Babcock, Nazeby) contain sequences (p-w, b-k, 2-b) which do not otherwise occur morpheme-internally and are identical to those encountered in productive compounds, cf. (tapeworm, lab, coat, rose border). Proper names themselves are sometimes subject to historical restructuring, as illustrated by the word (Aitken) already mentioned. To take another example, the former compound (Greenwich), when pronounced grenij or grens’, displays the same kind of non-analyticity as (manage). Since forms which continue to put their historical analyticity on phonological view fail to exhibit the systematic phonotactic restrictions encountered in non-derived forms, itis necessary52. Constituency to exclude them from the data set on which the following discussion of intra-con- stituent clusters draws. ‘A pertinent question to ask at this point is whether sound sequences missing from particular morpheme-internal contexts reflect synchronically active phono- logical constraints ur are mexely accidental lacunae in the vocabulary transmitted from earlier stages in the history of English. To put it in terms introduced in 1.3.4, do these gaps constitute core facts of the language that we should attempt to derive from general and possibly universal principles? Or are they unsystematic facts that belong to the periphery of the grammar? The oft-cited pair *bmik and * bik illustrates the point at issue here.” Lexical gaps such as * bik can be shown to fall into the peripheral category. This particular form is grammatical, even if unettested (at least in standard English). Its non-occurrence cannot be put down to systematic restrictions on any of the sound sequences it contains. The initial bi cluster, for example, shows up as a well-formed branching onset in a host of attested lexical items, including (blow, blot, black). Nor is there any bar on ir or bir sequences (cf. (limp, blink)). The same goes for tk (ef. (trick, click)). “The form *bmk, on the other hand, must be judged ungrammatical. The com- plete lack of any words in English beginning with bn suggests that chis cluster is not a possible onset in the language. Moreover, this absence is evidently related to a more general ban on plosive-nasal orset clusters. That this is a systematic property of the grammar, in need of principled explanation, is confirmed by the fact that precisely the same constraint is found in numerous other languages, including many with little or no historical affiliation with English. “There are various pieces of external evidence which bolster the conclusion that lexical gaps of the *bnuk type, unlike those of the *blak type, are not simply accidental residues bequeathed by earlier stages of language history. For example, the phonotactic constraints associated with the former type vigorously assert themselves when new vocabulary items enter the language, as foreign borrowings, neologisms, acronyms, product brand names, or whatever. They also figure prominently in characteristically English second-language errors. The absence of ‘word-initial plosive-nasal sequences in English, for example, means that the German borrowing (gneiss) (pronounced gnays in the donor language) has been adopted as nays. For the same reason, English learners of German frequently render initial gn- and kn- sequences as n-, or else insert an intervening vowel, as in koni: (German kni: knee’). Note that errors of this type cannot be put down to some vague notion of articulatory difficulty. Englishspeakers have no problem getting their tongues round plosive-nasal sequences in, for example, cross-word contexts, asin (big noise, pick no). The source of such errors is phonological and has to do with the deep-seated nature of phonotactic constraints operating in particular constituent contexts. ‘The situation is quite different with accidental gaps such as bik. It is quite possible to imagine blsk as a neologism in English. It would make a perfectly good product brand name (unlike * nuk, for example). And English learners have n0 problem pronouncing this form in German (where it means ‘glance’). Constituency 53 All of the phonotactic patterns briefly surveyed in this section and to be discussed in much more detail in the following sections can be shown to be non- accidental in the sense just described. That is, they fall within the domain of core grammar and as such call for principled explanation. 2.4 Onsets, rhymes, nuclei 2.4.1 Some traditional assumptions A traditional specification of the set of possible syllabic constituents in English usually runs something like this: ine (23) (a) The onset can contain between zero and three positions, illustrated in the word-initial portions of (eye, pie, pry, spry). (b) The coda part of the rhyme contains between zero and four positions, illustrated in the word-final portions of (see (f), sick (-k), six (-ks), sixth (-ks6), sixths (-4309)). (e) The nuclear part of the rhyme contains at least one and at most two postions, illustrated by the short-long vowel contrast in (bid) versus Conclusions (23a) and (23b) are based on the assumption that the set of possible onsets and codas is established on the basis of clusters occurring at word edges. In (23c), the setting of a lower limit of one position on the nucleus reflects the obligatory presence of this constituent within a syllable. Taken together, the statements in (23) define a well-formedness template for the English syllable which we might formalize as follows: (24) [xB]oauer PewcteusXf]coda Of course a template of this sort only establishes the gross limits on the number of positions that can occur in each constituent. It says nothing about additional restrictions that exist on the co-occurrence of different segment-types within each Position. Having established the onset and coda portions of the template (24) on the basis: of word-initial and word-final consonant sequences, we run into the problem already alluded to: we find that the maximal inventory of possible syllable types defined by the template is massively reduced when we turn our attention to word-internal contexts. Moreover, the potential for individual onset and coda Positions to support phonological contrasts decreases spectacularly as we move to the periphery of the onset and the coda. In other words, we have a serious case of. over-generation: the template defines a set of possible syllable-types which is Significantly larger than the set that is actually attested. The immediate response54 Constituency supplementing it with to this problem might be to try patching up the template by supplementing it wit conditions which eliminate a subset of the syllables it allows to be generated. We ‘an illustrate the issues that are at stake here by first examining the onset condition in (23a). 2.4.2 Onsets In this section, we will review the evidence which suggests that English onsets maximally contain not three positions, as claimed in (23a), but two. ‘Single-position onsets in English display virtually the entire gamut of consonant contrasts in the language, while in two-position onsets we find a reduced set of distinctions involving the obstruent-glidefliquid restriction already mentioned. However, itis a well-known fact about English (and many other languages) that, in the first position of a word-initial three-consonant cluster, the distinctive possibilities reduce to one. Only a voiceless coronal fricative can occur in this Context ~ in English, s. If word-initial structure is equated with onset structure, as implied in (23a), we would have to formulate some kind of co-occurrence con- straine limiting the first position in a three-slot onset to s. However, there are at least two sorts of evidence which indicate that this solution is unsound. One involves the syllabification of s in word-medial clusters, the other its phonotactic interactions with following consonants. Both point to the conclusion that s in initial clusters is not integrated into onsets in the way that other consonants are. In order to discuss the syllabification of s in medial clusters, itis necessary to say something about the syllabification of word-internal consonants in general. ‘There is a consensus among phonologists that the universally unmarked pattem in VCV sequences is for the inter-vocalic consonant to be syllabified in the onset Of the second syllable rather than in the coda of the first. So (city), for example, ‘would be syllabified as suti. This principle is sometimes referred to as Onset Maximization: (25) Onset Maximization Syllable-initial segments are maximized to the extent consistent with the syllable structure conditions ofthe language in question. (The universality of this principle is compromised in some frameworks by allow- ing V.CV to undergo resyllabification to VC.V under certain circumstances, 2 point we will return to in chapter 4.) Onset maximization also forces VCCV ‘sequences to be syllabified as V.CCY, if the two-consonant sequence constitutes a well-formed onset, a8 in pe.trol (petro). Otherwise, an inter-vocalic cluster of two consonants will be syllabified as VC.CY, i.e. with the first consonant in the coda of the syllable on the left and the second in the onset of the one on the right, as in (plen.ty). Constituency 55 If s in initial clusters occurs in an onset, we might expect this syllabification to carry over into medial clusters. That is, on the basis of onset maximization, medial VsCV sequences should be consistently syllabified as V.sCV, as in m.stor (mister). There is at least some evidence in English that this expectation is unfounded. There are certain stress facts, for example, which seem ( suggest Vs.CV syllabification. This pattern is evident in, say, (Aléska, Nebréska, aspi- distra), where the penultimate rhymes are heavy, as demonstrated by the fact that they attract stress, This indicates that s in these forms is syllabified not in the onset of the final syllable but rather in the coda of the penult, the same pattern as that found in, for example, {magénta, propaginda, Esmeralda). It has to be acknowledged that not all words containing medial sC clusters display this kind of stress pattern. We find antepenultimate rather than penultimate stress in, for example, (industry, orchestra). Nevertheless, the inconsistency of stress place- ‘ment in words with pre-consonantal s shows that the existence of medial sC onsets can certainly not be taken for granted. In certain other languages with medial sC clusters, the relevant syllabification evidence is quite unambiguous: s in this context clearly occupies a coda. In standard Italian, for example, stressed vowels are long in open syllables but short in closed. Hence forms such as vita ‘life’ (CV.CV) versus vinta ‘defeated (fem.)” (CVC.CV). If sC constituted an onset, a form such as vista ‘sight’ would be expected to contain a long vowel (*vi.sta), on the pattern of vita. However, the fact that vista lines up with viata in having a short vowel confirms that VsCV is syllabified as VCV There are two responses we could have to the English and Italian facts just outlined. The weaker position would be to assume that s in medial clusters is free +o vacillate between coda and onset. Later in the chapter, I will present arguments for favouring the stronger view that medial sC clusters universally conform to the ‘sC pattern. If this position is correct, then the fluctuating stress patterns evident in forms such as (aspidistra) and (industry) must be attributed to the unsystematic lexical selectivity which, quite independently of syllabification, afflicts the English stress system as a whole. If we persist in the belief that s in word-initial clusters is syllabified in onset position, we now have to come up with an explanation for why it should mysteriously migrate to the coda in medial clusters. But, even if we confine out attention to initial clusters, we find additional distributional evidence against sC onsets. If s in word-initial clusters does indeed occur in the same constituent as a following consonant, we might expect the same sort of co-occurrence restrictions to be evident in this context as in other complex onsets. On the face of it, the fact that there is a phonotactic dependency here at all (the constraint on the first of an initial three-consonant cluster being s) does seem to be symptomatic of just this kind of situation. However, the other distributional properties of sin this context suggest that we should be wary of drawing any such conclusion. In order to assess this evidence, we now need to examine in more detail the nature of the rather severe restrictions that are placed on the co-occurrence of segment-types within syllabic constituents.56 Constituency ‘The main dimension that has traditionally been cited as being implicated in both intra- and inter-constituent phonotactic constraints is sonority.”” This is a scalar property attributed to segments which, very broadly speaking, manifests itself perceptually as differences in loudness. (Acoustically, this correlates with dif ferences in the ourput of energy and in the extent to which this is concentrated in particular frequency bands.) In articulatory terms, this corresponds roughly to the degree of aperture in the vocal tract. Segments arrange themselves on a hierarchy ranging from least to most sonorous, the latter being characterized by maximiza- tion of energy output and articulatory aperture. The outlines of this hierarchy are generally assumed to be universal:* (26) Sonority hierarchy Low vowels Glides/high vowels Increasing | Liquids sonority Nasals Fricatives Plosives ‘The role of sonority in restricting the co-occurrence of different segment-types in syllabic structure has two aspects. One involves the principle of sonority sequencing: (27) Sonority sequencing ‘An optimal syllable consists of a sonority peak, corresponding to the nucleus, optionally flanked by segments which decrease in sonority the further they occur from the nucleas. ‘Viewed from left to right, segment-sequences within an onset thus have an upward sonority slope, while those in rhymes have a downward slope. The other aspect involves the observation that languages impose a minimum sonority dis- tance on neighbouring segments, which can be calculated by reference to the ranks occupied by individual segment-types on the hierarchy." Setting aside the issue of sC clusters for a moment, we observe in English a general requirement that segments within an onjet should be separated by at least one sonority ranks hence the illegitimacy of onsets such as "ks-, "ip *ml~ Sonority on its own is insufficient to derive all the characteristic cluster restric- tions that recur across different languages. For example, it cannot account for the bar on plosive-nasal and t/d/ onset clusters evident in English. This raises the issue of whether the whole collection of cluster constraints is derivable from some more fundamental representational property of which sonority is but one reflection. This is a matter we will take up in chapter 4. For the time being, we can simply ‘employ the sonority hierarchy as a notational device for describing phonotactic constraints. Constituency 57 A related question is whether restrictions imposed by sonority or its repres- centational source have universal force. At first sight, they would appear not to, at least if available language descriptions are taken at face value. For example, the literature is replete with examples of supposed onset clusters which flagrantly contravene sonority sequencing constraints. Well-known cases include the ‘multiple word-initial consonant sequences of a language such as Polish. In lgnde ‘adhere’, Isniet ‘shine? and mdlet ‘faint’, for example, we find three-consonant clusters in which the sonority profile falls and then rises again before the follow- ing vowel. One response has been to deny the universality of the sonority hierarchy and seek refuge in the notion that it does no more than express preferences which can be overridden on a language-specific basis."' This of course immediately robs the hierarchy of whatever explanatory force it might have had. ‘At worst, it leaves us with little more than a redescription of the very distribu- mal facts the hierarchy was designed to account for in the first place. The alternative view to be taken here is that the restrictions underlying sonority effects are universally imposed. It then has to be demonstrated that any alleged counterexamples have been inaccurately described. For instance, in the case of supposed onset clusters which clash with the sonority sequencing generalization, it has to be shown that they are not onsets at all but represent other constituent configurations. Onset treatments of this type can indeed be exposed as misana- lyses founded on the misconception that any word- initial consonant sequence automatically qualifies as a well-formed onset.#? As we will now see, this is precisely the point that can be made in respect of initial sC clusters. ‘When we compare the distribution of s in initial rwo-consonant clusters with that of other segments appearing in this position (all of them obstruents), we find some significant differences, some of which can be expressed in terms of sonority relations. The differences are set out below:t* (28) GQ -torw petkmaoa a @po++- ---- - to o- 44 oo a a Dn ne a a othe 2 eee Aoeee 236542 6 - ++ - oe Os t+ - + +e eet ‘Omitted from this array are co-occurrence possibilities which are of marginal status, such as vr- ((vroom)) and sf ((sphere)). I have also omitted y from C2, because I want to devote more attention to it alittle later. sris unattested in all but a few dialects; elsewhere the corresponding cluster is fr (as in (shrimp, shrink).58 Constituency There is one obvious similarity between the distributions displayed in (28a) and (286). In both series, C, can be occupied by a liquid or glide. The clearest difference between the two distributional patterns is that s can co-occur with a larger set of segments than other C, consonants. ‘The two main co-occurrence restrictions operating in (28a) can be considered characteristic of intra-onset phonotactics. One is a bar on homorganic clusters with J and w (hence inadmissible * 1, *dl-, *OF, *pw-, *bw-). The other is a restriction on C; being occupied only by a liquid or glide, a pattern that is expressible in terms of minimum sonoriy distance. Note that, whenever an ‘obstruent and liquid occur in an order which is the reverse of that in (28a), the sonority sequencing principle will debar them from being syllabified within an onset. Thus 1t, for example, is an impossible onset; on the other hand, it con- stitutes a well-formed coda-onset sequence, as in (shel.ter). ‘Neither of these restrictions holds of (28b). The homorganic s/- sequence is admissible (as in (sling)). And after s C: can be not only a liquid or glide, as in (28a), but also a nasal or oral non-continuant. This means that, were initial sC clusters to be treated as onsets, they would constitute the sole exception to the otherwise general principles of sonority sequencing and minimum sonority dis- tance. Note that C represents a downwaré sonority slope, in contradistinction to the uniformly upward patterns evident in (28a). And, again unlike the pattern in (28a), the segments in an s-plus-plosive cluster occupy adjacent positions on the sonority hierarchy. “The clusters in (28b) are not without their own phonotactic restrictions. How- ever, these cannot be considered exclusively diagnostic of onset constituency, since they operate in other environments as well. The single factor which has the most far-reaching effect of constricting the contrastive possibilities in Cin (28b) is voicing assimilation between s and a following plosive. This results in 2 suspension of the p-b, #dand k-g distinctions, each of which is neutralized under 1 voiceless unaspirated stop in this contex:. In so far as this process provides any evidence bearing on the constituency of adjacent segments, it actually favours the interpretation that they are not sisters. One of the other contexts in which it occurs is across constituent boundaries, specifically between a coda consonant and a following onset consonant. This is not only true of s-plus-obstruent clusters, where the obstruent is always voiceless; we get, for example, (aspic,master, basket) but no forms containing *-sh-, *-sd-, *-s0-."* But it is also true of other obstruent clusters; we can havd, for example, (hefty, chapter, factor) but no forms with * -fd-, *-pd-, *-kd-. Voicing assimilation also operates in word-final clusters, both with 5 (as in (clasp, test, ask) (*-8b, *-sd, *-sg)), and with other obstruent sequences (as in (aft, apt, act) (*-f4, *-pd, *-kd)). We have yet to discuss the syllabification of the latter context; but, whatever else it might be, it certainly does not involve sC onsets. So the absence of the b-p, d-t, g-k contrasts from Grin (28b) cannot be attributed to a phonotactic restriction that is exclusive to onsets. ‘We can summarize the phonotactic possibilities in initial sC clusters as follows. ‘They fail to conform to the principles of sonority sequencing and distance which Constituency 59 otherwise constrain onset clusters. Moreover, the set of consonants appearing afer initial sis equal to the ser of consonants appearing in simplex onset minus those that are ruled out by a voicing assimilation process that can be shown to ‘operate independently of onset contexts. In short, the C2 position in sC clusters enjoys a degree of distributional freedom which ie not at all in keeping with the assumption that the two positions are onset sisters on a par with those in (28a). The relative autonomy of s from a following consonant is further underlined when we extend our investigation to three-consonant clusters: 2) a & Q S ipek) thew) The distribution here can be derived by combining the restrictions associated with two-consonant clusters in (28a) and (286). That is, the set of combinatorial possibilities involving C2 and C; in (29) is equal to the set of C;-Cs possibilities in (28a) minus those that are ruled out by the constraints operating in (28b). Generally speaking, the Ci-Cs part of an initial three-term cluster thus looks just like a two-term obstruent-liquid/glide cluster. The only phonotactic dependency involving the initial s concerns the exclusion of sC sequences by the voicing assimilation process which, we have seen, cannot be considered diagnostic of onset constituency. Forms with initial *sdr- or *sbl-, say, are ruled out for the same reason that forms with the same sequences in medial Vs.CCV strings are. Staying with initial three-segment clusters, let us turn to one last piece of distributional evidence which further calls into question the onset syllabification of initial pre-consonantal s. In doing so, we will discover additional support for the onset-rhyme split. The evidence concerns the constituent status of y in clusters, which I excluded from (28). Let us start by considering the significance of using the y transcription to represent a high front non-round vocalic segment. On purely phonetic grounds, this means that the sound is related co and possibly identical to that usually transcribed as for 1, The reason for having distinct symbols here is a phonological rather than a phonetic one. The distinction is designed to reflect the different positions that a high front vocoid (vowel or glide) can occupy in syllabic struc- ture. The symbol / usually indicates sole occupancy of a nucleus, while yis usually reserved for the offglide portion of a nucleus (as in dey (day)) or for onset position (as in yet (yet)). The same relation holds of the symbols u and wz wis to was yis to i. Taking the symbol I to stand for a melodic unit which is high, front and non-round, we can represent the transcriptional difference between onset ¥ and nuclear (rhymal) / purely in terms of constituent structure:*” (30) (a) yb) 7 ° R 1 j 1 f I I60. Constituency Exploiting the representational distinction in (30), we can now examine some additional evidence in support of the onset-rhyme split. So far, we have con- sidered evidence involving syllable quantity and the statement of phonotactic constraints. This time, the evidence relates to the role of syllable structure in conditioning, phonological processes. We will look at two such processes which operate in many (but by no means all) dialects of English, One is responsible for the alternation between (a) and (an) in the indefinite article, the other for the alternation between a and iin (the). The conditions under which the alternations occur are identcal: (31) afan thfovi] (a) anapple th{i] apple anatrival ——_thli] arrival an eel th{i] eel (b) a pear thf] pear a nut tha} nut alord tha] lord (c) ayache tho] yacht awillow tha] willow ‘Asan informal statement of the conditions governing these alternations, we might say that (an) and (th{i}) are selected when a vowel follows but (a) and (ch{a)) when a consonant follows. This is only accurate if we understand the notions vowel and consonant in syllabic terms. The phonetic identity of the segment following the article is irrelevant, as can be seen when we compare (31a) with (31c). In both sets, the noun begins with an vocalic segment (i.e. one produced without consonantal constriction), but they behave differently with respect to the alternation. The crucial determining factor is whether or not the article is fol- lowed by an onset that is filled by a segment of some sort. fit is, the alternants a) and (the{o}) are selected, ierespective of whether the sound occupying the onset is consonantal, as in (31b), of vocalic, as in (31¢). The (an) and (ch{i) alternants are only selected if an immediately following vocalic sound occurs in a rhyme, as in (31a). The difference between the filled and embty onset conditions is illus- trated below: co OR oR Vt x x.. Lx. 4 1 ahs] 1 Ayachey antth{i] 1 {eel ‘The representational difference here suggests an analysis under which the 2 of {a(a)) only appears when it is able to attach itself to a following onset not already occupied by some other segment. We will not pursue this posibility here, but in Constituency 61 chapter 5 we will see how just this sort of analysis can be developed for another consonant-zero alternation. We are now ina position to determine the syllabification of y in CyV sequences. Historically, y did not feature in complex onsets at all. It occurred in nuclei as pact of the diphthong yu: (from carlicr iw, a reflex that still survives in some dialects). In this position it was free to co-occur with virtually any onset conson- ant (including sonorants). This is exactly what we would expect, given the general paucity of phonotactic dependencies between an onset and a following thyme. However, the signs are that the glide has since been reanalysed into onset position. The arguments supporting this conclusion are brought out by an analysis of the various reflexes of historical yur given in exercise 2 at che end of this chapter. For the time being, let me simply mention the piece of supporting evidence that is most relevant to the present discussion: Cy sequences are now subject to the same sort of co-occurrence restrictions as are characteristic of C-plus-sonorant clusters in onsets. In some dialects, y in this environment has either disappeared under certain conditions or coalesced with a preceding coronal to produce palato-alve- olar reflexes of historical alveolars. For example, dialects vary according to whether a y appears in all, some or none of the following forms: (tune, due, lieu, new, sue, enthusiasm)). Hence variants such as tyum-turn-Curn (tune), dyun-ducr- Juun (dune) and ssyu-su (issue). The effect of such changes has been to introduce ‘constraints on the co-occurrence of y with a preceding coronal." The most conservative dialects, those in which post-consonantal y retains a sclatively wide distribution, provide striking confirmation of the conclusion that sin word-initial clusters is not integrated into the onset. In conservative stand- ard dialects most closely associated with the south of England, y potentially ‘occurs after any coronal. (The only proviso is that this is true of r only if an unstressed vowel follows. Hence (virlylulent) but rurd (rude) (*zyucd).) Any sug- ‘gestion that this phonotactic independence might reflect a continuation of the historical nuclear status of yin yu is contradicted by the fact that the occurrence of the sound is dependent on the relative complexity of a preceding cluster. As seen in (33), the glide is free to occur after a lone consonant. This is true whether the consonant is non-coronal (33a) or coronal (T) (33b), including 1 (33c). However, it cannot appear after a lateral when this forms part of a two-consonant cluster (33d). (The missing historical palatal in forms such as (blue) is still to be heard in dialects which retain the carlicr iw reflex.) (33) @) Cyu (b) Tyu (yu (a) Cla (*Ciya) cute tune liew blue pewter dune lewd clue beauty new lucid plumage music suit luminous glue ‘The pattern in (33d) is consistent with the assumption that English onsets can have a maximum of two positions. In (33a, b, c), y appears in a two-slot onset, as depicted in (34a).62 Constituency 34) (a) (b) eH og (b) ‘The failure of y to appear in (33d) can be attributed to the fact that both positions in the onset are already occupied. The blocking of the glide’s historical migration, into the onset is illustrated in (34b). ‘We now have a way of testing the alleged onset status of sin initial sC clusters. If sC is an onset, it should behave just like the CC onsets in (33d) in never being followed by y. The forms in (35) show this prediction to be false. (35) Cyn slyw stew slew skew sluice spew ‘These forms show that Cy is a well-formed onset cluster in the relevant dialects, irtespective of whether it is preceded by s. (Initial sCCy clusters (e.g. *splyu-) are excluded for the same reason as the non-occurring CCy clusters in (33d).) If we are to maintain the explanation that y is excluded from two-position onsets because these are already saturated (as in (33d), we are forced to conclude that the initial sC sequences in (35) do not count as onsets. ‘We have just reviewed various pieces of evidence bearing on the issue of how s is syllabified in initial clusters. All point to the same negative conclusion: as shown in (36), s docs not belong to the onset occupied by the following consonant. (36) 1 \ i C © y (stay) ~\ So what exactly is the syllabic status of sin this context, not only in English but also in other languages with similar initial clusters? One proposal is that s in initial consonant sequences is directly adjoined to some higher node in phonolo- gical structure (the syllable or the word, for example).*° According to another proposal, s in this context occurs in a coda preceded by an unrealized nucleus." It is not easy to come by convinciag evidence from English that supports this analysis. Striking confirmation, however, comes from Romance languages. The difference between, for example, Italian (stadio) and Spanish (estadio) ‘stadium, it has been argued, resides in whether or not the nucleus preceding s is phonetically realized; in Spanish it is, in Italian it is not. Nevertheless, the presence of the initial ‘silent’ nucleus in Italian is indicated by Constituency 63 the fact that words with initial sC clusters behave exactly like vowel-initial words with respect to certain alternations. For example, the general pattern for conson- ant-initial words is to select the (i) form of the masculine plural definite article (eg. (i doni) ‘the gifts’). However, sC-initial words line up with vowel-initial words in selecting the alternant (gli) (c.g. (gli alberi) ‘the wees’, (gli stadi) “the stadiums’). That this selection is conditioned by the special constituent status of s in clusters rather than by its phonetic quality is confirmed by the fact that words with a single onset s take the regular consonant-initial alternant (e.g. (i signori) “the gentlemen’). (Parallel alternations involving the same conditioning factors affect such forms as {il/(o)) (masculine singular definite article) and (dei/degli) (masculine plural indefinite article).) We will not pursue the question of which of these alternatives is the most satisfactory. However, we will return to the issue of how the distributional peculiarities at word edges should be treated, when we move on to discuss the structure of rhymes in the next section. In the meantime, it is worth emphasizing one important implication of the recognition that s is not integrated into initial onset clusters. We now reject the claim made in (23a) that English onsets can contain up to three positions. Instead, we must assume that they can contain no more than two slots. The same conclusion has been reached in regard to other languages. In fact, itis now widely assumed that the imposition of a two-position limit on onsets is universal, although a book devoted to English is not the place to review the relevant evidence. The next question to be asked is whether this limit is peculiar to onsets or is imposed on all syllabic constituents. 2.4.3. Rhymes Up to now, I have been referring informally to two sub-portions of the rhyme, depicted in (20) ~ the nucleus and the coda. In this section, we will consider whether or not these represent independent nodes in constituent structure. Let us start by examining the status of the nucleus. Earlier we saw that, for the purposes of calculating quantity, a rhyme counts as heavy as long as it contains two positions, irrespective of whether these are occupied by a long vowel or by a short vowel followed by a coda consonant. This equivalence seems to suggest a unitary treatment in which both types of sequence ‘occur in a rhymal constituent lacking an independent nuclear node: GB) og R R ry ON trovroaut (The symbols v and ¢ stand for melodic units typically containing vocalic and consonantal material respectively.) However, there are certain distributional con- siderations which indicate that this conclusion is incorrect.64 Constituency Consider the following facts which show that the distributional relations invol- ving VV rhymes are of a different order to those involving VC rhymes. The coda of a VC rhyme is subject to phonotactic restrictions imposed by a following onset consonant. One example is the homorganic dependency that a coda nasal exhibits in relation to a following obseruent, as illustrated in (ca{m)per, wilnlter, alglchor). No such interactions are observed between a V or VV rhyme and a following ‘onset; virtually the whole gamut of vowel qualities is able to manifest itself in the first nucleus of a V(V).CV sequence, regardless of the nature of the onset con- sonant. On the other hand, there are very severe constraints on the co-occurrence of melody units within a vocalic rhyme, whereas there is little or no phonotactic dependency between the positions of a VC rhyme. True, there are restrictions on the length of the vowel before a coda consonant, involving the phenomenon of closed-syllable shortening (on which more in 2.4.4) which excludes long vowels, from this context. But there is no restriction on the type of vocalic melody unit that can occur before a coda consonant. This is in sharp contrast to the situation in a ViV2 rhyme, where the identity of the melody unit in V2 is heavily con- strained. Specifically, it is limited to one of the following off-glides: y, w and 2, ‘Thus, we can have two-position diphthongs such as ey, ay, ow, aw, iobut not, say, ia, “uo, *ue.# This is not to deny that vocalic sequences which might be transcribed in terms such as ia/ya, uo/wo, ue/we do not occur. (Bear in mind the equivalence of iand y and of wand w remarked on earlier.) However, when they do, they can be shown to form part of structures which ate different from the heavy-thyme type we are discussing here. They can represent one of the following, alternatives: an onset-nucleus sequence, or a sequence of two independent nuclei, or a short diphthong (i.e. a contour segment in which two vocalic melody units are associated to a single skeletal point). ‘The defective distributional property of the right-hand position of a vocalic thyme is strongly reminiscent of the situation that obtains in onsets. The vo- calic integrity that is indicated by this asymmetry has been taken as evidence for the existence of a nuclear constituent within the rhyme. Thusit appears necessary to recognize two types of bi-positional rhyme, one containing a branching nu- cleus, the other a single nuclear position: = — 8) @ oR ) oR f I XN ¥ \ it 1 Given the structures in (38), the weight of a chyme is still identifiable on the basis of the number of skeletal points it contains. Now, however, we cannot define @ heavy rhyme as one which directly branches. We have to specify that it dominates Constituency 65 two positions, regardless of whether this involves indirect domination of both positions, as in (38a), or immediate domination of the right-hand position and indirect domination (via N) of the nuclear position, as in (386). We have yet to consider the possibility that rhymes may contain more than wo positions. A number of questions arize in this connection. What is the maximum, number of positions that can occur in a nucleus? Is there a limit to the number of positions that can occur in the post-nuclear portion of the rhyme? If there is, is this limit independent of the number of positions in the nucleus? Is there any evidence that the post-nuclear hymal portion contains an independent coda constituent? The least controversial of these issues concerns the nucleus. There is general agreement among phonologists that nuclei are universally restricted to two posi- tions. Not all languages have branching nuclei, just as not all languages have branching onsets. The difference between branching and non-branching nuclei is the formal expression of a contrast between long and short vowels, and not all languages have such a distinction. Claims have occasionally been made that some systems display a three-way vowel-length contrast, which would imply the pos- sibility of ternary-branching nuclei. However, reported cases of so-called over- long vowels can be shown to involve other structures, for example a sequence of, ‘two or more nuclei, or the occurrence of a long vowel under word stress.5* Given the maximally binary branching structure of onsets and nuclei, we might reasonably ask whether rhymes pattern the same way.* If this is the case, and setting aside for the moment the possibility of there being an autonomous coda constituent, we expect the four logically possible types of rhyme depicted in (39). 39) (@) R o oR © oR @ og 1 \ i i N N N N 1 uN 1 rN So far we have seen evidence for the light structure in (39a) and the heavy structures in (39b) and (39c). The pattern in (39d), for which we have not yet seen any evidence, presents the maximal structure allowed for under binary branching. Note that, although the chyme contains three positions, neither the rhymal nor the nuclear node branches more than twice. In the next section we will see that it is necessary to recognize such ‘super-heavy’ rhymes, although they only occur under very specific conditions. ‘On the face of it, the statement in (23b), which allows for the occurrence of up to four positions in the English coda, suggests not only that the coda exists as an additional independent constituent but also that it is not subject to the same binarity restriction as the other constituents. If this is correct, it greatly increases the number of possible rhyme-types defined by the theory. The following, for ‘example, would have to be considered grammatical in English:66 Constituency R 40) R R R oA A 1 WA N Coda N Coda N Coda N Coda bry IN TN Pry IN TIN However, we will now examine a considerable amount of evidence which indic- ates that the assumption in (23b) is simply incorrect. 2.4.4 ‘Codas’ “This section surveys the main structural and distributional properties of the thyme in English. The primary purpose at this stage is simply to determine what the relevant facts are rather than to suggest explanations for why the rhyme takes the shape does The facts fall no ewe broad etegoris Ox the one hand here are those which help establish the gross typological outlines of the rhyme in English, relating specifically to the number of positions the constituent can “On the other handy there are facts which relate 10 the Cathal case t thyme. Some of the distributional teristics of particular positions within the chyme. § finding are quite robust. We wll se, for example tat he abity of he coda position to support consonantal contrasts is much more tightly constraine tha of eran ote poston, Other dstbuina facts may Seema fist siht somewhat arcane For example, whee a coda focus ina speeayshyme, win in only be occupied by a coronal consonant, Hence grammati Ghoul caldron) ves ungrnmmaal cowfbor* arn orth ke. Some of the apparencly more esoteric distributional patterns to be reviewed here contiane to lade saisfactory explanation In sch css, i hast beconcede, currently available formalizations ofthe relevant facts amount to little more than carefully devised stipulations. Later in this section, we will discus the sllabification of English word-fnal clusters, traditionally ‘considered to exhibit the full range of possible rhyme structures in a language. First, however, let us consider the range of possible Sequences that occur before a morpheme-internal fiset consonant or onset cluster” As illustrated by the firs syllables of the following forms, the four systematic- ally occurring medial rhyme-types in English are as follows: al b) WV. (@ ve. (@) wwe, ey ee faaoe Sulder ladder putrid chapter angel petrol booty kestrel chamber baton capon biscuit council Constituency 67 Forms with internal rhymes show either an open syllable, as in (41a) and (41b), or a syllable closed by a single coda consonant, asin (41c) and (414). Both types occur with a contrast between short and long nuclei: (41a,¢) versus (41b,d). In forms with a super-heavy VVC thyme (41d), the phonotactic latitude accorded the coda position is somewhat more restricted than that observed in (41¢). To gain an idea of the inequality of the distributional possibilities, we now need to take @ more detailed look at the phonotactics of coda-onset clusters briefly ‘touched on in the last section. The distributional profile of internal coda-onset clusters is subject to an exten- sion of the sonority sequencing generalization given in (2: (42) Sonority sequencing II In an optimal coda-onset cluster, the first consonant is no less sonorous than the second, Typical coda-onset clusters show a falling sonority slope (viewed froin left to right). In this respect, this pattern is the reverse of that found between the two Positions of an onset cluster (although the precise distributional details of the {two contexts are by no means mirror-images). Coda-onset clusters of this type include those consisting of a sonorant (liquid or nasal) followed by an obstrucnt, as in (filter, dolphin, winter, fancy). Ina large number of languages with codas, this is in fact ehe only permissible pattern. Fricative-plosive clusters also display the falling shape, as in (mister, biscuit). The formulation in (42) also allows for consonants of equal sonority to occur in the coda-onset context, as in (chapter, factor). Apparent counterexamples to (42) spring readily to mind, such as the obstruent« sonorant sequences in, for example, (atlas, cutlass, athlete, kidney, atmosphere), AAs with the sonority sequencing clause in (27), such evidence has confirmed some Phonologists in the belief thatthe relations expressed in the sonority hierarchy are merely preferences rather than absolute universals. Once again, however, we should be wary of reaching too hasty a conclusion on this matter. Interpreting such eases as genuine counterexamples is based on the assumption that any internal consonant sequence not forming a branching onset automatically qualifies as a coda-onset sequence. This position is rately if ever explicitly defended, and in chapter 4 I will argue that it is unfounded. Such obstruent-s, orant sequences, it can be shown, constitute bogus clusters; that is, the consom. ants in question are not truly adjacent and as such are not subject to the kind of systematic phonotactic restrictions that are evident in authentic coda-onset clus- ters. To anticipate the discussion, the position to be defended in chapter 4 is that the 61 of, say, (athlete) is split by a nuclear position (which is audible in some dialects). That is, the alleged cluster is similar to the 61 of, say, (cath(ollic), where the intervening nucleus is optionally expressed. If this line of reasoning is sound, it follows that whatever representational principle itis that underlies the general ization in (42) can be maintained as a universal.68 Constituency In true coda-onset clusters, the identity of the second consonant partially determines the identity of the first. This results in a number of distributional asymmetries, two of which we have already seen as being characteristic of this context. First, coda nasals cannot support a place contrast; they must be homor- tganic with the onset consonant. Second, the coda cannot be occupied by a coronal plosive (bearing in mind the comment just made on words such as (kidney, cutlass)). A coda plosive of any kind is in fact only possible ater a short vowel, and this is but one of a number of phonotactic restrictions which set VVC rhymes apart from VC rhymes. Thus although we can have, say, (chapter, doctor) with a short vowel, there are no occurrences of underived or root-level forms such as, say, *eyptor ot *owktor. A further constraint is that, in a VC.CV sequence, a coda obstruent lacks a distinctive voice value, taking on instead the zoice specification of the following obstruent; in the overwhelming number of cases such clusters are voiceless.” Subject to these restrictions, the coda position after a short nucleus can be ‘occupied by an oral stop (43a), a fricative (43b), a nasal (43c), or /(43d): (43) (a) chapter (b) blister. (c).- pamper (4) shelter factor whisker winter balcony doctor whisper anchor children after timber dolphin wisdom finger fancy (An additional constraint restricts coda fricatives to s, z and f} Forms containing historical r before an internal onset consonant (as in (quarter, party, farthing)) line up with the VV. pattern in (41b). This is more obvicus ih the case of non-thotic dialects, in which the rin question has been vocalized; the first rhyme of a form such as pati (party) (originally rful) displays the same branching- nucleus structure as that in, say, aso (father) (in which the first chyme is historically r-less). In rhotic dialects, the retained rin this context can be shown to occupy the second position of the branching nucleus. (More on this in chapter 5.) This means that, in these dialects too, forms such as (party, quarter) pattern with (father, capon, booty). ‘Compare the distributional array in (43) with that encountered in VVC thymes. There are two basic patterns here: the coda is filled either by a fricative (as in (oyster)) or by a sonorant (as in (council)). As noted above, there is a bar on superheavy rhymes containing a plosive coda. Many super-heavy sequences result from formerly short vowels having undergone various historical lengthen- ings, which have affected individual dialects to differing exten's. In the following description, it will be necessary to distinguish those super-heavy forms which are firmly established throughout English from those with a more restricted geo- graphical distribution. ‘The general pattern evident in sonorant-coda VVC thymes is illustrated below: Constituency 69 (44) (a) launder (b) shoulder saunter cauldron ancient boulder council poultry danger One observation we can make about forms of this type is thae the onset followin the super-heavy shyme is almost invariably occupied by a coronal. Exceptions to this generalization are few and far between: (chamber, cambric), for example. (To these can be added (example, sample) in those dialects with lengthened a: or 2 in these forms.) Second, the consonant sequences in question are all of the partial-geminate type - homorganic sonorant-obstruent clusters, It is usually assumed that the place value of any kind of geminate consonant is distinctively specified in the onset and spreads into the preceding coda position.** This is illustrated in the following representation of (dainty) (where the italicized N stands for a nasal stop which acquires its place specification from the following, coral stop): (45) (dainty) Only one of two fricatives can occur in the coda of a super-heavy rhyme: s or £VVF rhymes are only found in dialects which show lengthened reflexes of historically short vowels (most usually only 2) in this context. Examples from southern English and derivative dialects include (after, laughter, rafter) with a:, ©; ot some similar long vowel. A further restriction is that the following onset in such cases can only be occupied by 4 something that is also true of coda fafter a short nucleus (as in (hefty)). Forms with a VVs thyme are mote widespread. Besides examples such as those in (46a), which occur in most if not all dialects, we also find cases such as those in (46b), but only in dialects that show the historical vowel-lengthening just described. (46) (a) easter (b) bastard oyster caster pastry basket boisterous exasperate Unlike in the majority of forms with sonorant-coda VVC rhymes, the onset following a VVs thyme is not restricted to a coronal, as forms such as (basket, exasperate) with possible ay: indicate.70 Constituency Let us now turn our attention to the constituent structure of word-final conson- ants and consonant clusters, traditionally assigned to codas. There are several respects in which these sequences differ significantly from internal codas. Let us briefly note two which will be discussed in more detail below. First, unlike internal closed rhymes, a lone final consonant, irrespective of its melodic content, tolerates a short-long contrast in a preceding nucleus: (47) VC]__ lid, run, back, top, step, foot WVC] _ slide, spoon, rake, soap, steep, boot Since the supposed coda in a final (V)VC] sequence is not followed by an onset within the same domain, we must consider it to be distinctively specified for all melodic content, including the place dimension. This immediately sets it apart from an internal coda, which, as we have seen, is subject to a variety of restric- tions on the melody units it can contain. 'A second noteworthy fact about word endings is that they display consonant clusters which cannot occur within an internal rhyme. Some two-consonant examples: (48) (a) fact, apt, list, bend, help, find, old (b) rungs, tipped, summed, likes, dames I is the assumption that such clusters are exhaustively syllabified in the rhyme that underlies the traditional claim in (23c) that the English coda can contain up to four positions. Recall from our earlier discussion that the distributional dis- ‘crepancy between medial and final clusters is bound up with the morphological structure of the latter. Note that some ‘over-weight’ sequences, those illustrated in (48b), involve an analytic morpheme boundary. As we have already estab- lished, the distributional independence of segments occurring tither side of such a boundary indicates that word-level derived forms fall outside the domain within which the major systematic phonotactic constraints of English operate. ‘This point is particularly clear in the case of word-level derived forms where the second of two morphemes begins with a consonant or consonant cluster followed by a nucleus. This configuration obtains in compounds and suffixed forms with (hood, nest, ly), etc. In such cates, it is uncontroversial to assume that the ‘morpheme-initial consonant occupies its own onset, rather than being syllabified in the same constituent as the final segment of the preceding morpheme, es. (pot.light, hat.rack, bad.ly, late.ness). Assigning such trans-morphentic sequen ces to separate constituents is consistent with the phonotactic independence that they display in relation to one another. Rather less straightforward are consonant-only suffixes, in particular the td and s-z alrernants of (-ed, -s) illustrated in (48b). Apart from the voicing interac- tion which is responsible for the alternations, these suffixes exhibit the same sort of phonotactic independence from the preceding stem-final segment as analytic morphemes with vocalic content. The simplest interpretation of this Constituency 71 relative independence is to assume that (-ed) and (-s) ; at -ed) and (-5) are no more qualified to be syllabifed in the same constituent as a preceding stem-final segment than other word-level no are. Rather, a consonant-only suffix occurs in a separate constituent. (Exactly what that constituent is a question we will take up in 4.6. that a final consonant of this type occupies a coda position incorporated in the same syllabi structure asthe sem, This ignores the fact that such consonants fail to conform to otherwise general phonotactic constraints which are establis independently of cross-morphemic contexts. ised As soon as we confine our attention to the di shich : lomain in which systematic phono- tactic constraints operate, i-e. to forms containing no analytic morphological structure, the distributional possiblities in word-final envivonmente. decrease considerably from those allowed for in (23c). The full set of sequence-types appearing in the final stressed portions of forms containing no anal aprearingin Pe ontaining no analytic boundary (49) 2-position sequence (@) Ww day, so, see, pie () ve pat, tip, den, gull 3-position sequence (©) VVC late, ight, town, feat (€@) VCC fist, fact, desk, gulp 4-position sequence (c) VVCC "paste, paint, rind (f) VCCC text (The absence of final stressed light V is due to an independent constraint operat- ing in all Germanic languages, whereby a stressed nucleus in this position is fequied to branch, More on thi in chapter 5.) an immediate consequence of excluding forms containing analytic morpho- logical structute, the upward limit on final consonant lunes sate from the four positions allowed for in (23) to three. Four-term clusters invariably involve word-level suffixation, e.g. (text-s, sixth-s). As a further consequence of this narrowing of focus, we may observe that the number of final consonants is in part dependent on the quantity of the preceding nucleus. For example, three-conson- ant clusters only occur after a short nucleus, which means that a VVCCC. sequence can never form part of a non-analytic domain. A sequence of this shape always indicates the presence of word-level suffixation, as in (point-s, pounce-d). As our investigation proceeds, it will become clear that the interdependence of nnuclear quantity and the length of a following consonant cluster is actually more far-reaching than this. Although two-consonant sequences following. a long nu leus do occur (as in (49e)), their distributional potential is considerably cur- tailed. By far the most favoured maximal final sequences are VVC (49c) and VCC (49d). In other words, the preferred maximal sequence preceding a final72 Constituency consonant is a heavy VX shyme (where X represents either a nuclear or a coda position). ‘We have come across the configuration of a heavy rhyme followed by a final consonant before, but under different circumstances. Recall the extra-metrical status of final consonants in stceas assignment; in verbs, for example, we caw how a heavy VV or VC rhyme preceding a final extra-metrical consonant attracts stress. The parallel between the stress and syllabification facts, which at first blush may seem rather odd, turns out under close inspection to be anything but coincidental. ‘The clusters which on the face of it present the most convincing evidence in favour of an independent coda constituent are the two- and three-consonant sequences in (494, e, f). However, we will now consider evidence which indicates that, of all the clusters shown in (49), only (49a) is exhaustively syllabified in a thyme. The absolute final consonant position illustrated in (49b) through (49f) ccan be shown to exhibit certain distributional peculiarities which are at odds with the notion that it forms part of a coda. We will start by considering how this point relates to the three-position sequences in (49c) and (49d). If the word-final consonant of a V(V)C or VCC cluster were syllabified in coda position, it would be reasonable to expect it to display the same kind of distribu- tional characteristics as morpheme-internal coda consonants. In fact, it does not. If anything, it behaves just like a morpheme-internal onset. One rather obvious respect in which the final consonant differs from an internal coda consonant is that the former is not constrained by the phonotactic dependency that the latter displays in relation to a following onset. For example, a nasal in an internal coda is dependent on the following onset consonant for its place specification. A word-final nasal, on the other hand, has its own distinctive place value (even though in the case of a coronal nasal this is optionally overridden by the place value of a following word-initial consonant, as in {tel — ml past)). Another distributional difference is that t can occur word-finally but not in an internal coda. In terms of distributional potential, the final position of a V(V}C or VOC sequence is in most respects identical to an internal onset; in both positions, we find a mote or less full range of consonantal contrasts. ‘An even more striking parallel between final consonants and internal onsets relates to the manner in which both positions interact with the quantitative charac- teristics of a preceding nucleus. As illustrated in (47), it is a quite general characteristic of single word-final consonant positions that they ate able to support a short-long contrast in a preceding nucleus; compere (49b) with (49). In this respect, consonants in this position are identical to the internal onset consonants illustrated in (41a) and (41b). Moreover, a word-final consonant is able to support a preceding VC cluster in exactly the same way that an internal ‘onset can; compare (49d) with (41c). The failure of consonants in absolute word-final position to demonstrate coda- like behaviour has led many phonologists to the conclusion that they are not integrated into the preceding rhyme. Exactly what the syllabic status of such consonants is a matter of debate. One proposal is that final consonants are ‘extra- Constituency 73 syllabic’. That is, at least until some stage of derivation, they remain unsyllabi- fied. As it stands, the notion of exta-syllabcity contravenes general principle of grammar whereby each unit in a representation is legitimized by virtue of belonging to some higher node in the linguistic hierarchy. (This is the notion of licensing on which much more in chapter 4.) Ie follows that the extca-syllabicity of a position can only be contingent; at some point in a derivation the position ‘must become integrated into the phonological hierarchy in order for it to be phonetically expressed.** This integration, it has been proposed, involves attach- ing the extra-syllabic position to the coda of the preceding syllable. The result is a mismatch between the set of possible coda types defined as well-formed in lexical representations and a larger set legitimized in derived representations. (In chapter 4, we will assess the desirability of permitting a discrepancy of this type.) The notion of extra-syllabicity obviously parallels that of extra-metricaity. Each notion is motivated by an independent array of facts, relating to syllabifica tion in the first instance and to stress assignment in the second. But, at least as far as the facts we have reviewed are concerned, the effect isthe same - the separation of a word-final post-nuclear position from a preceding thyme. This commonality has led to the two notions being subsumed under a more general dimension known as extra-prosodicity.* However, there is at least one reason for rejecting the extra-syllabicity analysis: it makes wrong predictions about the phonotactics of final CC clusters. As depicted in the following VCC] structure, the independence of an extra-syllabic consonant (indicated by (x)q) from a preceding coda implies that the two posi- tions should be phonotactically independent: (50) I »\ ! \ x KOs a a This is incorrect. There are very strict distributional dependencies operating in this context. And significantly they are more or less identical to those holding of internal VC.CV sequences. Compare the main distributional patterns that are evident in the two contexts:74 Constituency (51) Medial -C.C- and final -CC] clusters Medial inal ‘Medial Final Stop-stop Sonorant-stop chapter apt pamper damp vector sect winter ‘Aline filter guilt wrinkle rink scalpel scalp Stop-fricative Sonorant-fricative mister mist cancer ‘manse after raft dolphin golf whisper wisp whisker brisk ‘The only distributional difference between the two contexts involves an inde- pendent development which bars the domain-final occurrence of mband, in some dialects, ag. (See exercise 1 at the end of this chapter.) Thus we find medial mb in, say, (clamber) but not in, say, (climb) where the (b) corresponding to historical b is now silent. ‘We might try to formulate the phonotactic restrictions on the final clusters illustrated in (51) in terms of an interaction between an extra-syllabic position and a preceding coda. This would amount to treating extra-syllabicity as an independent constituent node. But distributional statements couched in these terms would simply duplicate statements relating to internal coda-onset clusters. ‘The simplest solution is to assume that the distributional parallel reflects parallel constituent structures. In other words, the second C of a final -CC} cluster ‘occupies an onset position in the same way as the second C of an internal -C.C- cluster. This means that a form such as (mist) is syllabified as mist, with the final occurring as the onset of a ‘degenerate syllable’. Degeneracy here refers to the absence of an audible nucleus. ‘The immediate advantage of recognizing domain-final onsets is that it allows us ‘0 unify the statement of phonotactic restrictions on medial -C.C- and final -C.C] clusters. Both of these involve coda-onset interactions, as illustrated in the follow- ing representations of (mister) and (mist):"” (82) (@) OR (b) R ; ; eH\ or 98 ¢ ' 1 1 1 1 I 1 1 1 mor sot 2 (r) mit sot Constituency 75 ‘The representation in (52b) contains what might be termed a ‘stray’ onset ~ one that is not followed by a nucleus. However, as the discussion proceeds, it will Become clear that an onset ofthis pe requires suppor fom a following sent Assigning the second position of a final -CC] cluster to an onset raises the question of whether this analysis should be extended to single final consonants, ice. those occurring in -VC] and -VVC] sequences. Answering this question in the affirmative brings with it a number of advantages. First, we are able to explain why a single final consonant position enjoys a degree of distributional freedom that is similar or identical to that of an internal onset but is quite unlike the more restricted pattern associated with internal codas. Second, as we will now see, we are in a better position to account for the quantitative characteristics of final sequences. The notion of degenerate syllable allows us to assign constituent structure to the four consonant-final sequences illustrated in (49b) through (49e) without adding to the set of rhyme-types given in (39). That is, we can dispense with the coda as. an independent node without compromising the position that constituents are maximally binary branching. The set of rhyme-types preceding a final onset consonant is thus identical to the set encountered morpheme-internally. The difference between -VC] and -VVC] is now shown to be identical to that between medial -V.C- and -VV.G, illustrated in (41a,b). As shown in the following repres- ‘entations of (am) and (aim), the difference consists in a contrast between light and heavy rhymes preceding an onset: (3) @ Ro R 0 I ! a N ! aN | ! tt ' 2 om ey om Note that, by replacing the notion of extra-syllabicity by the notion of degener- ate syllable, we may also dispense with extra-metricality as an independent diacritic. As we saw in our brief survey of quantity-sensitivity in stress assignment (2.2.3), a final consonant in English (and indeed other languages) fails to contri- bute to the weight of a preceding rhyme. We now have additional reasons, quite independent of stress, for concluding that this consonant is not part of the preceding rhyme. The extra-metrical behaviour of this consonant is captured if we make the assumption that the degenerate syllable in which it occurs is not projected onto the level at which metrical relations are computed. As illustrated in (54), this means that a final consonant is ignored for the purposes of stress assignment. In (54a), the final £ of (lamént) occurs in the onset of a degenerate syllable; the preceding rhyme en, being heavy, attracts stress.76 Constituency (54) (a) R (b) \ Oo NON ° N ONO foroto ' roid Prati nn 12 memet edict In (dit) (S4b), on the other hand, the rhyme preceding the firal onset ris light (occupied by the short vowel spelt (i)) and thus fails to ateract stress. That is not say that final degenerate syllables are invariably invisible to metrical computation. In languages such as English, they happen to be. But in other languages they play a full role in stress assignment. In Spanish, for example, stress tends to fall on the penultimate syllable of a vowel-final word: (sabéna, pist6la, perdida). However, it falls on the last vowel if the word enés in a consonant {unless the consonant represents a suffix): (civil, mercéd, altér). This can be captured as a unitary penultimate pattern if we assume that, in languages of this, type, final degenerate syllables are projected metrically.® This supports the conclusion that a word-final onset is followed by a silent nucleus, which, in Spanish at least, is counted for stress purposes. (We will explore the notion of silent nuclei in a good deal more detail in chapter 4.) Thus a corsonant-final form such as (ii (witha final silent nucleus) can be said oa penultimate stress in the sa abana) {with a final audible nucleus). “Tnng no oth four poston final sequenes sated 3) and (49) we may note two quite different patterns. There is a very limited number of forms containing a plosives cluster, in apparent contravention of the sonority sequencing principle. Included in this class are (text, coax, traipse), which we will return to below. Inthe vast rtajority of cases, however, a final four-position sequence consists of a long nucleus followed either by a coronal partial geminate (as in (paint)) or a fricative plus plosive (as in (paste)). In other words, the configuration in quest pretty much identical to that encountered in medial VVC.C sequences of the type we find in, say, (council, oyster). The similarities between the medial and final contexts in this regard can be seen by comparing the partial-geminate forms in the two contexts (some of the medial examples repeated from (44) and (46). (55) Final Medial (a) Wat saint, mount, pint fountain {b) VVnd rind, sound flounder (c) Was pounce, ounce council (d) Wns range, scrounge angel (¢) Wit revolt, colt poultry (Wid child, cold, field shoulder (56) Final Medial (a) Wst paste, boast, boost pastry (b) (VIVsC cast, clasp, cask basket Constituency 77 The final partial-geminate forms are entirely restricted to coronals after VV. That is, there are no examples equivalent to the handful of labial partial gemi- nates found medially after VV ((chamber, cambric)). We do not expect to find final VVmb forms in any event, in view of the independently attested bar on final mb clusters already alluded to. But the preference for coronaliry in post-VV partial geminates is rigidly enforced in final position. There are no forms of the shape *-ump or *-eynk. ((Oink) must be considered marginal.) As with the medial cases, coronality is also clearly favoured in final fricative-plosive clusters after VV (VVsC in (56)). VVstforms such as those in (56a) are the best established {although in some dialects final thas been lost in this context). Long-vowel forms such as those in (S6b) are only found in dialects affected by the lengthening of istorically short vowels already described. The effects of the latter development are most widespread in the case of historical «, where we find lengthening not only before st but also before sp and sk, as the examples (clasp, cask) illustrate. But the change has extended to other vowels in various dialects, 0 that the VVsC. class can include forms such as (frost, cost) (historically short o} and (nest, desk) (historically short e). The final VVCC forms submit to exactly the same analysis as that proposed earlier for the medial VVC.C cases. That is, we have a superheavy thyme followed by an onset (in this case, the onset of a final degenerate syllable). The coronal place value of the partial gefhinate in, say, (paint) is distinctively specified in the onset ¢and spreads into the preceding nasal, just as in (dainty) (45). The facts we have discussed so far in this survey of the English rhyme can be ‘summarized as follows. As far asits gross typological outlines are concerned, we may note that all four ofthe structures depicted in (39) are attested. There is no evidence to support any of the branching-coda structures illustrated in (40). In other words, the coda postion of the rhymes contains at most one position. Moreover, these are restricted to contexts where they are followed by an onset. That is, there are good ‘grounds for rejecting the traditional belief in the existence of word:-final codas, a ‘word-final consonant occurs instead in the onset of a degenerate syllable. On the distributional side, the ability of the coda to support consonantal contrasts is severely curtailed, more so in the case of VVC rhymes than in VC. The restrictions on super-heavy rhymes can be summarized as follows:” (57) English super-heavy chymes. (a) The coda position is restricted to a sonorant or fricative, (b) A.coda sonorant is unable to support a distinctive place contrast. (©) Im the case of (b), the favoured place category determined by the following onset consonant is coronal. Each of these stipulations refers to the nature of the melodic units that are Permitted to occur in particular syllabic positions. We cannot say much more about them until we have had a chance to examine the internal make-up of segments in more detail in the next chapter. All we need note at this stage is that78 Constituency the constraints on the qualitative (place and manner) content of the coda reflect the general phonotactic dependency that exists between this position and the following onset in both heavy and super-heavy rhymes. Clause (57a) reflects the general applicability of Sonority Sequencing II (42) to all coda-onset contexts. Clauses (57b) and (57¢) describe the preferred partial coronal geminate pattern found in coda-onset clusters after a branching nucleus. Tt remains true, however, that nothing akin to the qualizative dependency between a coda and a following onset is to be observed between the coda and a preceding nucleus. That is, the quality of a coda consonant is largely independent of that of the preceding nucleus.” As noted in 2.4.3, this is one of the main motivations for recognizing an autonomous nuclear constituent in the first place. There is a quantitative relation between the nucleus and coda - the phenomenon of closed-syllable shortening mentioned in 1.4.2 and to be discussed in more detail immediately below. But the qualitative independence of the two sorts of position is demonstrated by the fact that, irrespective of the aature of the coda consonant, we find full sub-systems of short and long vowels in closed heavy and super-heavy rhymes respectively. The full set of six short stressed vowels (the system associated with southern English and derivative dialects) is illustrated in the closed heavy rhymes of, for example, (filter, shelter, alp, halt, gulp, pulpit) The potential for the complete set of long vowels to occur in super-heavy rhymes, is exemplified by (east, paste, boast, boost, joust, heist, moist) (to which we can add (past) in dialects with lengthened a: or :). In short, the curtailed distributional potential of the coda is evidently due to a combination of two factors - the quality of the consonant in the following onset and the quantity of the preceding nucleus. In chapter 4, we will consider how these dependencies might be derived from general conditions governing the wel- formedness of rhymes. For the present, we may further explore the interacting effects of these two factors by taking a closer look at closed-syllable shortening. This, recall, manifests itself in root-level alternations between long and short vowels in forms such as the following.” (58) “ (i) (a) receive receptive perceive perception describe descriptive reduce reduction scribe scripture (b) retain retentive intervene intervention five fifty wise wisdom Note how the rightmost long vowel of the morphologically simple roots in column {S8i) alternates with a short vowel in the internal closed rhymes ofthe suffixed forms in column (58ii). In accordance with the position argued for in 1.8, we may assume that the root-suffixed forms have non-analytic structure. That is, they are lexically Constituency 79 listed forms which are not phonologically derived from the corresponding unsuffixe forms. Viewed in these terms, closed-llable shortening ss perely ana dinate sional regularity. The term closed-shyme shormessisthus perhaps more appropriate (The alternations exemplify several other root specific phenomena, Vowel Shift and asmment he wo and mana yeti fh 0 ial consonant, ich m ly relevant tothe point at hand. Again on the bass of ents in 14.3, we may ead theseas lexical maters) Ons es ftbeartuments ¢ forms in (58b), unlike those in (58a), contain partial geminates or coda facies these cass he closet hye shores fe carb coir an automatic response to local conditions in the representation, This s bese shortened forms of this shape co-exist with otherwae similarly stecewen tone such as (dainty) (see (44) ) and (easter) (se (46) } which nevertheless display an unshortened super-heavy pattern. In contrast, the shortness effect in colune (i) of (58a) is an automatic consequence of the fact that the forms contain other than partial geminates or coda fricatives. That is the conditions in (57) rale out fore with super-heavy chymes such as "-ayp. "ick -and the like Were we to continue labouring under the misapprehension that a final consonant occupies a coda, we would be ata loss to explain why closed-thyme shortness afferes say, (receptive) but not (eceive) Afterall, the relevant nucleus, according to the cody view, supposedly occurs ina closed rhyme in both forms. On the other hand, by invoking the notion of final degenerate syllable, we are able to account for the immunity to closed-rhyme shortness displayed by (eceive) and similar forms such ag those in column (i) of (58a). In these cates, the final consonant occupies an onset with the result thatthe preceding chyme is open and its nucleus free to branch, (59) R 1 ON 0 R O i 1 1 ay 1 | ! ' Vrot Compare this with (receptive), where the root-final p, occurring word- medially, occupies a coda. Were the first nucleus of receptive) to preserve the wag ‘thas in (ceccive), the result would be a medial super-heavy rhyme ~ something like “risiptiv. But the occurrence of the labial stop in the coda falls foul of the condition. in (57a), with the result that such a form would be ungram- matical”! (60) * na-n-0 «-K-080. Constituency i is shor, In the correct form riseptiv, the vowel in the medial rhyme closed by p : resulting in a heavy rhyme, in which configuration the coda consonant is no longer subject to the conditions in (57): Not (61) 2 N 1 Pp nH -0 ox -0 eK ZS <-%=-0 yuicement that a This outcome follows as an automatic consequence of the req tha i ici ‘an upward limit s-heavy rhyme must satisfy the conditions in ($7). Otherwise an up of two positions is automaticaly imposed on the thyme, thus producing the closed-rhyme shortness effect. Closed:thyme shortness is not confined to medial chymes in English. As the following root-level alternations show, the same pattern also occurs in final -VCC] clusters: (62) (@)_ keep kept deep depth wide width leap leape sleep slept (b) deal dealt heal health wide width clean cleanse leave left thief theft As with the medial cases ilusated in (58), we cannot assume tha hornet automatic in forms such as those in (62b), since they contain partial geminates ot a coda fiatve (ct the forms with super heavy rhymes shown in (33) ) In (62 however, shortness is a direct consequence of condition (57a). Compare (weep) containing a root-final onset consonant preceded by a long nucleus, with (wept), in which the root-final consonant occupies a coda: © @ go R oO ® o oR ° 1 t N m\ a x ox TAT 1 1 1 1 w Si a Constituency 81 Under condition (57a), the coda p renders ill-formed the super-heavy rhyme that would result from the retention of the long nucleus; hence the short vowel in (wept) In accordance with their root-level status, each of the alternants of the type shown in (62) constitutes a single phonological domain. This is reflected in the different syllabic affiliations of the root-final p in (63): it occupies a domain-final onset in (weep) (63a) but a domain-internal coda in (wept) (63b). Compare this with the form (seeped), which contains the regular word-level suffix (-ed).™ Here we are dealing with two independent phonological domains, viz, [sip] i. That is, the p of (seep) remains domain-final irrespective of whether the form is inflected with a word-level suffix or not. This means that the consonant occurs consistently in the onset of a degenerate syllable, with the result that the preceding aucleus is free to branch in both (seep) and (seeped). (The syllabic affiliation of the word-level suffix consonant in such forms is something we take up in chapter 4.) 24.5 ‘Appendices? The last set of word-final sequences ro be discussed are those containing clusters d in (494). The number of forms displaying this pattern after a long nucleus (or nucleus plus 7) is extremely small: (64a) exhausts the set of apparently genuine examples. Rather more frequent are instances of this duster following a short nucleus (64b). The occurrence of the same sequence followed by another consonant is limited to the -VAst string found in (next, text) and derivative forms of the latter (64c). of plosive followed by s, exempli (64) (a) coax, hoax, traipse, corpse (b) lapse, box, tax, eclipse (c)_next, text, context, pretext The sonority profile of these Cs clusters poses a problem for the view that the ‘wo consonants are adjacent in constituent structure. This is true regardless of whether adjacency is understood as co-occupancy of a coda, as under a tradi- tional analysis, or as occurrence in a coda-onset cluster, as per the analysis developed in this chapter. In both instances, we have a contravention of the otherwise general principle of sonority sequencing: the final consonant is more sonorous than the one preceding it, not less as the principle would predict if they were co-constituents or formed a coda-onset cluster. In recognition of the peculiar status of final s in consonant clusters, some Phonologists have assigned it to an extra-syllabic margin or appendix.” It has been proposed that this position is directly integrated into the phonological hierarchy at word level, bypassing, as it were, intermediate levels of constituent Sructure such as the chyme.”* Whatever the precise status of this site might be, we should be careful to distinguish it from the position occupied by other consonants82 Constituency in final clusters. This slot, as we have noted, systematically behaves like an onset and, in English at least, is not subject to any major constraints on the type of melodic unit that can occupy it. This is in stark contrast to the defective distribu. tion of the appendix. The exclusion of all but s from this site is puzzling, but it does af least contain a number of parallels with restrictions operating in certain other contexts. Let us briefly note two of these. It is a matter for future research to determine whether and how these parallels are to be accounted for in a unified way. First, the appearance of s in domain-edge consonant cluste:s, in contravention of the otherwise general principle of sonority sequencing, is of course not re- stricted to word-final contexts. We have already discussed the similarly anoma- lous position it enjoys in word-initial clusters (see 2.4.2). Second, in forms containing analytic morphological structure, over-weight seg- ‘ment strings can result from the juxtaposition of morphemes at word level. Thi means that underived forms terminating in s are phonologically indistinguishable from derived forms to which the s alternant of (-s) is attached. Compare the underived forms in (65a) with the suffixed forms in (65b).” (65) (a) fix (b) sticks box locks lapse laps hoax folks traipse apes ‘This relation between the appendix and word-level morphology provides one of the arguments for the claim that the appendix is integrated into the phonological hierarchy at the level of the word, rather than within a constituent such as the thyme. Indeed many of the words in (64) derive historically from forms in which the appendix was clearly not adjacent to the preceding consoaant, either because the terminal segment was an independent word-level suffix, or because the final ‘wo consonants were separated by a vowel which has since been syncopated. For example, (coax) is historically related to the plural form of an obsolete noun ¢coke) ‘fool’, hoax) to (hocus), (corpse) to (corpus), and (next) to the superlative of a cognate of (near). 2.4.6 The rhyme: summary ‘Summing up the evidence reviewed in the last two sections, we can say that binary limit is imposed on the branching structure of English chympes. As with ‘onsets and nuclei, this conclusion is consistent with findings reported for other languages. We take this ability to branch as a reflection of the autonomous status of onsets, rhymes and nuclei as nodes in the phonological hierarchy. This conclu- sion removes any motivation for recognizing the coda as a coastituent node in its ‘own right. In other words, the term coda is no more than an informal label for @ Constituency 83 post-nuclear rhymal position, and this is the usage I will adopt throughout the rest of this book. In sum, only the four types of rhyme shown in (39) and repeated in (66) are possible. (66) (a) light (b) heavy (e) heavy (a) super-heavy R R R R i 1 i 1 N x N N I nN | rN . x x x x x A light chyme (66a) is composed of a single nuclear position. Of the two heavy types, one contains two nuclear positions (6b), the other a single nuclear position followed by a position directly dominated by the rhymal node (66c). The super-heavy type (66d), whose occurrence is constrained by the conditions in (57), consists of two nuclear slots followed by a single position directly dominated by the rhyme. 2.5. Syllabification One question we have only briefly touched on is how strings of skeletal positions come to be associated with constituent structure. In this connection, we might ask whether constituent structure is already present in lexical representation or is constructed by some sort of syllabification algorithm. The very possibility of postulating general principles which regulate constituency, such as onset maximi aation and sonority sequencing, testifies to the fact that this aspect of repres- entation is predictable in at least some respects. The predictability of constituent structure has led some phonologists to assume that itis absent from underlying representations and is built by algorith."! The latter incorporates universal principles of ylabification supplemented by lan- guage-specific constraints on the co-occurrence of different segment-types in differen syllabic contexts (expressed, for example, in terms of minimum severity distance). The algorithm typically begins by locating nuclei, in recognition of the pivotal status enjoyed by these constituents as obligatory heads of onset-thyme Pairs. In subsequently assigning non-nuclear positions to onsets and rhymes, the Algorithm implements onset maximization by giving precedence to the former. Under an alternative view, lexical representations redundantly contain syllable Structure. Syllabification principles are then interpreted not as derivational pro- ‘esses but rather as well-formedness conditions on the structure of lexical repre- sentations." According to one version of this approach, universal principles and language-particular phonotactic constraints together define for each lan. Buage a set of syllable structure templates. The constituent structure of a given84 Constituency representation is then deemed well formed if it can be parsed in accordance with the templates. Whichever of these views of syllabification is adopted, this much is clear: constituent structure must be established before forms are submitted to deriva- tion. This conclusion is based on the observation that many types of phonological process are conditioned by aspects of constituency. In other words, constituent structure must either be present lexically, as assumed under the template ap- proach, or be constructed ‘first thing’ before phonological processes proper come into operation. This point means that any pair of templatic and algorithmic accounts that incorporate the same general principles and language-specific con- straints are probably no more than notational variants. 2.6 Summary In this chapter, we have examined some of the evidence which indicates that segment strings are organized into constituents. The evidence involves the statement of (a) phonotactic dependencies; (b) suprasegmental patterns; and (c) the condit under which certain types of phonological process occur. We have considered further arguments which lend support to the views that (a) there are only three constituent nodes, namely onset, nucleus and chyme; and (b) a binary limit is uniformly imposed on the branching structure of constituents. Outsize segment clusters, which at first sight seem to contradict these conclusions, can be shown to dissolve into smaller sub-sequences, some of which are peculiar to word edges. Te would be extraordinary if the convergent behaviour of all three constituents with respect to the binarity of branching structure were purely accidental. In chapter 4, we will go on to consider whether this property might be derivable from some more general principle of grammar. Another important issue is whether there is something in the nature of different sound-types that influences their ability to occur in particular constituent posi- tions. To answer this question, we nced to subject melodic units to an internal inspection - the task of the next chapter. Exercises 1 English nasals In 1.4.2, we briefly looked at the distribution of 9 in English. Now we have the ‘opportunity to undertake a more detailed analysis which also takes account of a related set of phenomena involving the other nasals. The data The data below illustrate the realization of nasals and nasal-plus- plosive clusters in Scots and ewo types of English. The three systems ar: the general Constituency 85 southern pattern found in many parts of the English-speaki sof the English-speaking world (A); a pattern ‘ypically associated with the north and Midlands of England (B); and Scots (C). Each ofthe data rows illustrates a particular combination of phonological and morphological environments. Where two or more nasals or nasal clusters appear ina word, the emboldened character inataniates the contest ilistated by een The task Provide a comparative phonological account ofthe three systems, an discuss the general theoretical issues that your analysis raises. yeremsy and Questions relating specifically to this set of data include the following, (a) What is the phonological distribution of d ution of m, n and 9 in each of the systems? (6) What role does constituent structure play in conditioning the distribution? (c) What role does morphological structure play in conditioning the distribu- tion? (a) In regard co (b), what is the significance ofthe vowel-length facts outlined (c) Is the velar nasal lexically distinctive? (f) How should we treat the alternation between g and zero that is evident after ‘Tin one of the systems? (=) If your own system does not tally exactly with any of those represented ere, construct your own data column, and work out precisely in wh respects it differs from the others. Spec in war I CORONAL A BC o onoa need, nail, note nd nd a hand, send, wand ne ont at rant, sent, flint a hang@Pbundle, chunder, cinder non on funnel, money, spanner nd ond on sending, sender nd ond on kinder, blander LABIAL A BC 1mm m more, met, mail 2m m m lamb, thumb, limb, dumb 3m mm sum, tim, ram 4 mp mp mp pump, lamp, limp S mb mb m thimble, tremble, number86 Constituency 6m mm plumber, lambing m dumber summer, pumice TI VELAR ABC rete Foot-initial position (* gs * bopét) 20 wo sing, sang, sung, strong, long, hang, hung, string 3 ok ok ok sink, sank, sunk, thank, monk | 4 9 5 finger, bangle, single, hunger, linge, anger 5 9 England, angry, hungry 6k om x donkey, monkey, rankle, wrinkle 7) woo banger, banging, singer, hanger, songster, stringer, longing 8 m9 0 longer, stronger IV VOWEL LENGTH BEFORE NASALS In all three systems, the contrast between long and short vowels holds before word-final a and a: Systems A, Band C Short Long Short Long fin fine dim time pen vain rum plume Before word-final (9), Systems A and B only ever show short vowels: eg. (sing, hang, lung) but *seyg(g), * /awa(@), * tig. System C, on the other hand, maintains the short-long contrast in this context: System C Short Long sung rey (eign) hang, Jen (‘long’) sing kin (‘king’) 2 Palatal glides in English As briefly noted in 2.4.2, dialects of English vary according to whether or not they show a palatal glide in certain pre-vocalie environments. For example, (new) is Constituency 87 pronounced as nywin some dialects, as muin others, and as niw in still others. In this exercise, we investigate the distribution of y in more detail The data Five different types of dialect are represented in the data below. System Ais the standard pattern most closely associated with the south of Englands B is widely represented in North America and parts of southern England; C includes most Scottish and Irish varieties; D is characteristic of vernacular usage in East ‘Anglia; E is found in rural south Wales." The data illustrate the incidence of the palatal approximant (glide or vowel) in syllable-initial sequences consisting of one or two consonants followed by a long high back vowel. Each set of words illustrates a particular configuration of phonological conditions relating primarily to stress and the preceding conson- antal context. Each of the five realizations given in each data-row records whether or not a palatal appears in that particular context and, where appropriate, identifies the peeceding consonant. In several cases, a consonant-plus-y cluster in some dialects corresponds to a single palatalized reflex in others: 4-<¢, dy-j, hy-g, sf (see data-rows 27, 29, 31, 36-9) The task (a) What is the contrastive potential of the palatal approximant in each of the systems? (b) State the distribution of the palatal approximant in each of the systems, focusing on both segmental and stress conditions. (c) Account for the palatalized consonants that appear in two of the systems. (4) What issues arise in determining whether a pre-vocalic palatal approxim- ant occurs in the onset or the nucleus? (6) Itis quite possible that you are familiar with a y-system that is not exactly like any of those represented here. In that case, using the word list pro- vided, collect as much relevant data as you can, and reconsider the four questions above. System A B C DE Ly ye yu yur yur you, ewe, youth fyu: fy fyu: fu fiw few, futile, future fu: fu fu fur fur fool 1 pye py ye pu piw 20 mye my: mye mu miw music, mule, mew 30 me me ome ome ome ‘moon, moot 4 bye bye by: bu iw beautiful, bureau, abuse Sku bw bebo: boon, booze 6 yy yyw wy vu vw view, revue Tov ove ve ove ow voodoo 8 9 1 Pew, pewter, spew88 u 12 13 14 15 16 7 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 4 42 43 44 45 46 47 48 49 50 Constituency pepe bye Aye keke aye me he he syn se zu ae yu Ou oe On yu te aye du an a tye ye hehe nye aye Iu te ye me se fe ty be aye jue aye pre pre pre pre pin ple plu, pls bie Bi blu blu bre bre tre bre tun mu au dr dre pu ku pu kiw ku: tiw hiw ie diw diw pool, spoon cute, queue, cure cool, coot new, continuity lewd, lew oom, loose Ruth, root assume, pursuit suicide, suit, sue soon, soothe presume, Zeus enthuse thuja Tuesday, perpetsity 100, tool dew, duty, during doom, do huge, Hugh who, hoot continue, annual value, volume erudite, virulent virtue, perpetual residual, incredulous residue prude, prune proof pleurisy plume blew blue bruise, brewed brood’ truant troop, true drew st 2 3 4 55 56 x7 58 9 60 6 dru lu: gnu one flu: flu: Oru: Constituency 89 droop eroon clue, include grew groom fruit, frugal flew flu threw through3 Melody 3.1 Introduction ‘A major goal of phonological theory is to provide a model of the set of possible processes to which sounds may be subject. One aspect of this task is to specify where processes occur; this was a theme which we touched onin the last chapter and which we will return to in the next. Another aspect concerns what can happen t» 4 sound, the copie co be disused inthis chap. Clete anrwer to this question depends to a large extent on what we take to be the set of primitive operations in phonological processing. It also crucially depends on what we consider the internal composition of a sound to be and how this influences its susceptibility to processing, 3.2. The primes of melodic representation 3.2.1 Sounds are componential So far in this book, units on the melody tier have appeared as phonemic alphabetic symbols attached to slots in the skeletal tier. This is not supposed to imply that melody units are indissoluble entities. Rather, the symbols are no more than convenient shorthand for phonological expressions, each of which is decom posable into smaller components. In this chapter, we will consider how these ‘components can be identified and how they are organized within melodic expres- sions. Responses to the more specific questions raised by these issues range from those on which there is near-universal agreement among phonologists to those which continue to be the subject of lively debate. Let us begin by staking out the common ground. First, as just indicated, segments are decomposable into components which constitute the primes of melodic representation. These are most widely known a features, although they are also referred to by such terms as elements, gestures and particles.’ One of the most convincing pieces of evidence in favour of this view is that sounds pattern into non-random sets, known as natural classes, with respect to their participation in phonological processes. The unified behaviour of Melody 91 a given natural class is explicable if itis assumed that the members of that class share a particular property. The number of overlapping natural classes to which a particular sound belongs thus provides an indication of the number of distinet properties of which itis composed. Second, the set of primes is universal and comparatively small. According to the literature, the going rate is somewhere between eight and twenty, although there have been attempts to reduce components to combinations of even more fun- damental atoms.* The set of primes is initially established on the basis of phono- logical evidence, relating primarily to the participation of sounds in processes and their organization into systems. From this evidence emerge patterns which recur with such massive regularity across different languages that it is difficult to avoid the conclusion that melodic substance, no less than other aspects of linguistic organization, is rooted in a genetic endowment that is common to all humans. It is a matter of ongoing research to determine exactly how much of this substance should be attributed to universals of physiology, reflecting constraints imposed by vocal and auditory anatomy, and how much to general cognitive or specifically linguistic universals. What is beyond dispute amongst those committed to this programme is that melodic primes have relatively stable interpretations in articu- lation, the acoustic signal and audition. This remains true even if the mapping between primes and their phonetic exponents is not always direct. Of course, it is always possible to profess a lack of interest in the phonetic and psychological dimensions of melodic content, in which case primes are treated as purely abstract mathematical constructs whose only function is to classify phono- logical contrasts in particular systems.’ However, the impact that Platonist views such as this have had on the recent development of phonological theory (and linguistic theory in general) has been minimal. Third, phonological substance consists of two quite different types of informa- tion, each of which requires its own independent mode of representation. As we saw in the last chapter, relational aspects of a representation, involving such details as quantity, syllabicity and stress, are encoded in terms of skeletal and constituent structure. On the other hand, the strictly componential aspect of a segment’s representation, comprising those attributes that have a comparatively stable phonetic interpretation, is deployed on a separate plane devoted to melodic content. So much for the issues on which there is broad agreement within the phonolo- sical community. Beyond this point, opinions differ to varying extents. Let us ‘now consider some of the more significant moot points. 3.2.2 Privativeness vs, equipollence Each melodic prime defines a bifurcation of segments.‘ One source of disagree- ment is the question of whether the two classes produced by a given bifurcation are of equal status. To make the matter more concrete, take a pair of segments that are minimally distinct; that is, they are identical in all respects save in relation92. Melody to a single prime. In such a situation, does one segment possess the prime in question while the other lacks it? If so, we are dealing with what is known as a privative opposition. Or is it a question of both segments possessing the prime and differing in terms of the value they assign to it? If this is the case, there is a further question regarding the range of values that the prime may assume, According to one view, there are two values (usually expressed in terms of plus versus minus); that is, primes define equipollent oppositions. According to an- other view, the coefficient associated with a prime varies along a scale of three ot more values. The scalar mode, if recognized at all, is usually assumed to be confined to the level of phonetic implementation and to be inappropriate for expressing the bifurcations into natural classes that characterize phonological dis- tinctions. The issue thus boils down to the question of whether phonological oppositions are privative or equipollent. The tradition that at least some oppositions are privative goes back to the Prague School linguists (to whom we owe the terms privative ard equipollent) and has continued in various theoretical guises up to the present day.* By contrast, the position adopted in SPE is that distinctions are uniformly expressed in terms of equipollent features. More recent incarnations of feature thecry, however, have shifted in the direction of privativeness.© ‘The distinction berween the equipollent and privative formats is not a trivial cone. They differ quite clearly in terms of their empirical content. Consider again the case of a minimal distinction, in which a single prime divides sounds into two classes. Equipollence implies that the classes should behave symmetrically. That is, as the term indicates, the two poles of the opposition are predicted to stand an ‘equal chance of being accessed by phonological processes involving that particu- lar prime. Coupled to a rule-based treatment of processes, this arrangement greatly multiplies the number of possible processes defined by the theory.’ Take for instance a situation in which a two-valued feature such as [round] is deemed to be active in vowel harmony. Equipollence generates at least two types of system: one in which [4round] is active and another in which {-round] is active, Indeed it generates a third system, one in which both values are active. Unfortu- nately, this tripartite prediction fails to correspond to the observation that one value of each two-valued feature is universally preferred. (In the case of [round], it isthe plus value.) Early recognition of the asymmetries that are to be observed in the phonological behaviour of features helped spur the development of the theory of markedness, touched on in 1.3.2." In any given opposition, one of the terms is unmarked in the sense that it is universally preferred. This preference manifests itself in various ‘ways: for example, the unmarked term is more widely distributed throughout the phonological systems of the world’s languages; the presence in a given system of the marked term implies the presence of the unmarked one, but not vice versa; and in systems possessing both terms, itis the unmarked one that appears first during the course of child language acquisition. According to the SPE interpretation of this notion, Universal Grammar contains statements to the effect that certain feature specifications are unmarked. The fewer the marked values that a rule incorporates, the more highly valued it is adjudged to be. Melody 93 In a recent descendant of this theory, segments in lexical representation are vuaderspecified, in the sense that they contain only a subset of the feature values they display in phonetic representation.? The default case is for unmarked feature values to be left blank underlyingly and to be filled in by redundancy rules, some or all of which are universal. However, in one version of this theory, individual languages are free to select either the marked or the unmarked value as being lexically distinctive for a particular feature." The potential for overriding univer. sal markedness conventions in this way means that the set of possible processes defined by the theory remains comparatively large. What markedness statements and underspecification tell us is that the possibilities are unequal. The generative capacity of the model further snowballs as a result of allowing redundancy rules to be ordered in various ways in relation to rules which characterize phonological processes. The set of markedness conventions constitutes an independent look-up table against which the values of individual features are gauged. A more radical alternative is to build markedness relations directly into phonological repres- entations. This notion can be implemented quite straightforwardly within a framework in which phonological oppositions are uniformly privative. Privative- ness implies an asymmetry in the phonological behaviour of a pair of segment classes that are distinguished by the presence versus absence of a particular prime. Only the class possessing the prime can participate in a process involving that prime. There can be no complementary process involving the class which lacks the Prime in question. This conclusion is based on the premise that phonological Processes can only access what is present in a representation. This arrangement has the potential for greatly reducing the set of possible Processes defined by the theory." Take for example the case of labial harmony just alluded to, Let us assume for the sake of argument that the equipollent feature [round] corresponds to a single privative prime [round}, which is present in round vowels and absent from non-round vowels. While the equipollent feature predicts the three harmony patterns mentioned above, the privative prime pre- dicts only one system, namely one in which {round] is active. Within a strictly Privative framework, there is no way of expressing a complementary system in which ‘absence-of-{round)’ is harmonically active. Any attempt to introduce a negation operator (such as ~{round], ‘not-{round’) immediately reclassifies the feamework as equipollent.!” All other things being equal (in particular, assuming an equal number of primes), a privative model of phonological oppositions is more constrained than one based on equipollence. A priori, this would lead us to favour the former over the latter — provided of course that the privative model can be shown to be observationally adequate. Potential counterevidence to the more restrictive model comes in the form of any equipollent analysis which incorporates a rule contain- ing a feature value for which there is no direct privative equivalent. And it has to be acknowledged that such accounts are myriad, given the almost unchallenged ascendancy that the equipollent view enjoyed during a period which included the publication of SPE. In some cases, such examples can be straightforwardly94 Melody reanalysed in terms of the opposite feature value. In other words, they turn out ‘not to constitute counterevidence at all but rather reflect one of the recurring problems that is symptomatic of over-generation - the possibility of analysing a single phenomenon in more than one way. Nevertheless, there no doubt exists a corpus of more robust cquipollent accounts that need to be reassessed on a case-by-case basis. 3.2.3 Phonetic interpretation The position adopted with respect to privativeness versus equipollence has an indirect bearing on another important representational question: does each prime enjoy independent phonetic interpretability?" In the case of orthodox features, the answer is no. Although each feature is taken to have its own stable phonetic signature, this cannot actually manifest itself unless it is harnessed to those of other features. For example, a {+high] segment cannot be concretely realized as such unless it also includes a slate of feature specifications (some distinctive, others redundant) such as {-back}, {-round] and [-consonantal], in which case it surfaces as a palatal approximant. In combination with other specifications, [+high] can show up in, say, 1, kx or ¢. This arrangement, when coupled to the ‘underspecification view briefly outlined above, has the consequence that segments are not fully interpretable until all redundant features are filled in. In SPE, all lexically blank specifications were filled in immediately prior to the entry of forms into phonological derivation. One objection to this proposal is that it results in redundant feature values being carried through derivation even though, in most cases, they fail to participate actively in phonological processes and do no more than clutter up the formulation of rules. Take for example @ language in which the feature [round] is non-contrastive; all front vowels are non-round and all non-low back vowels are round. Say the language has a rule which fronts u to i. Fronting is represented as [+tback] —> [-back], which on its own yields *d. So there has to be an additional ‘clean-up’ rule which adjusts [round] from plus to minus (to give *d— #) even though this feature is redun- dant." In response to this objection, more recent versions of feature theory allow for redundant values to be underlyingly unspecified and to remain suppressed until later stages of derivation. The redundancy rules which fill in missing values are interspersed among the phonological rules proper. The result is that segments are not phonetically interpretable until the final stage of derivation, the level some- times known as systematic phonetic representation.'* ‘Three remarks about the implications of the underspecification approach are in order here. First, allowing redundancy rules to be intercalated among phonolo- gical rules, it has been claimed, is justified on the grounds that the same feature value can be shown to be active at different stages of derivation in different grammars. However, as already noted, the multiplicity of possible rule orderings that this arrangement permits grossly inflates the expressive power of the theory. Melody 95 Second, one of the original motives behind the postulation of redundancy-free lexical representations seems to have been the assumption that long-term memory constraints compel speakers to confine storage to idiosyncratic information and to maximize the rule-based computation of predictable information.'* The psy- cholinguistic validity of this idea has uever been seriously defended. In fact, the notion implies a model of lexical access which is not particularly plausible. Economizing on the amount of phonological information contained in lexical representations would incur a significant computational cost to the speaker or hearer: it would greatly add to the amount of work that has to be done when forms are retrieved from the lexicon."” Before the underspecified form of a lexical entry could be submitted to articulation or recognition, it would first have to be unpacked by having its missing feature values filled in. Third, underlying this arrangement is the assumption that the main raison d'étre of the phonological component is to convert abstract underlying representations into ever more concrete phonetic ones. The mismatch between the underlying and systematic phonetic levels has sometimes been justified on the grounds that they allegedly perform quite different roles: the former serves the function of memory and lexical storage, while the latter supposedly serves as input to articulation and Perception. According to one interpretation of this notion, the derivation of phonetic forms occurs ‘on line’ ~ that is, as the speaker is articulating or the listener decoding a particular utterance." This view places the phonological component outside the domain of generative grammar proper, at least if we understand this in the usual sense of a model of internalized linguistic knowledge which exists independently of its performance. It is not simply a matter of the systematic phonetic level constituting a buffer between internalized phonological competence and its articulatory or perceptual externalization. Under this view, the phonological component as a whole is designed to carry out the essentially extra-grammatical task of assembling phonetic forms for production or reception on particular occasions. Compare this performance-centred view with one in which phonology remains wholly rooted in the domain of linguistic competence. To be consistent with the latter notion, phonological derivation should do no more than capture generaliza- tions governing alternations and distributional regularities. This task can be performed quite independently of any provision that needs to be made for articulation and perception. Conceived of in this way, derivation is stricly gener- ative in the sense that it does no more than define a set of well-formed phonologi- cal representations. Processes map phonological objects onto other phonological ‘objects, rather than onto phonetic ones. So where does phonetic interpretation fit into this scheme? If processes map like onto like, it follows that one phonological representation should be no more or less interpretable than another. As a result, an initial representation cannot be considered any less phonetically interpretable or ‘concrete’ than a final repres- entation. Such an arrangement is of course incompatible with the notion that non-final representations are underspecified and thus not directly interpretable. ‘What is evidently called for is a conception of melodic form according to which96 Melody segments are phonetically interpretable at all levels of derivation. This might seem. to imply a return to the SPE notion of full feature specification throughout derivation ~ and a reprise of the problem whereby superfluous feature values make unwelcome appearances in the statement of phonological processes. How- ever, we are only forced into that corner, if we persist in a long-held belief of feature theory ~ that the smallest phonological unit capable of independent phonetic interpretation is the segment. A corollary of this view is the idea that subsegmental status deprives each feature of the ability to be interpretable with- ‘out support from other features occurring within the same segment. This is in fact not a necessary conclusion at all. It is perfectly possible to conceive of a phonological prime as being ‘small’ enough to fit inside a segment and yet ‘big’ enough to enjoy stand-alone phonetic interpretability. For example, fone such prime that is widely assumed in privative approaches is independently manifested as a. Another is interpreted as i. The ability of each of these primes to be made phonetically manifest is not contingent on its being combined with other primes. That is, a and imay be viewed in some sense as ‘primitive’ segments; each is composed of a single prime and is thus the direct phonetic embodiment of that particular prime."* Nevertheless, primes of this type can be combined to form ‘compound or non-primitive segments. Amalgamating a's and ’s primes, as we will see in the next section, yields e. Conceived of in these terms, primes are appropriately labelled elements.2° The analogy with physical matter is apt, in view of the ability of each element both to exist independently and to enter into ‘compounds with other elements. This view of melodic form has the desired consequence that segments are phonetically interpretable at any level of derivation. Within a framework such as this, there is nothing resembling feature underspecification and thus nothing akin to blank-filling operations of the type that can be ordered in relation to phono- logical processes. Since phonological representations uniformly adhere to the principle of full phonetic interpretability, there is no motivation for recognizing an autonomous level of systematic phonetic representation. Any phonological representation at any level of derivation can be directly submitted to articulatory or perceptual interpretation. Derivation is thus not an operation by means of which abstract phonological objects are transformed into increasingly concrete physical objects. Rather it is a strictly generative function which defines the ‘grammaticality of phonological strings. ‘The representational model just sketched is the one to be expznded on in the remainder of this chapter. It takes the following stances on the issues raised in the foregoing discussion. Melodic expressions are decomposable into a relatively, small set of elements, which are uniformly privative and have stable and inde- pendent phonetic exponence. This type of approach is restrictive in the following respects, Relative to an equipollent framework which posits the same number of primes, privativeness narrows the set of segment classes that are potentially active in phonological processing. The principle of full phonetic interpretability ex- cludes the multiple orderings that are possible in a model in which segments are only partially specified and in which redundancy rules are interpolated among phonological processes. Melody 97 A continuing research task is to demonstrate that the observed facts of phon methodological orientation is the minimalist one outlined in 1.3.2: start with an available model which generates a relatively small set of possibilities, and identify as research problems thoze arcas where it apparently under-generates, This isin preference to an orientation which starts with an over-generating model and supplements it with constraints (such as markedness conventions) designed to filter out or disfavour possibilities that fail to be confirmed by observation, 3.3 Elements for vowels 3.3.1 A,1,U In different privative approaches, there is broad agreement on three elements which are considered to play a pivotal role in the representation of vowels. The lependent phonetic exponents of these elements are the three ‘comer vowels a and u2" I will adopt one established practice of symbolizing these elements as A, Tand U respectively: (1) Element Independent interpretation A a 1 i U u Any given vowel is composed either of a single element (a simplex expression) or of a fusion of two or more elements (a compound). For example, fasing A with T results in e; fusing A with U yields o. Informally, we may think of the addition ot 3 ring a more open version a vow defined hy either Lor U. Or & ind U represent primary colours which, when mixed. yield such seconds coloursaseand on A ann The combinatorial possibilities of the elements A, I and U, occurring either ely or in two-element compounds, yield the six melodic expressions shown ia (2) (dis a close front round vowel) 2) Simplex Compound a fal e [AD i 0 o {AU} u © wT ‘The sound defined by the three-element compound [A, I, U] may be thought of as. @ rounded version of g; in other words, it is «It will be necessary to refine the notion of fusion and to recognize further elements, in order to derive other dimensions of vocalic contrast, including nasality and tenseness (the latter in. volved in the representation of distinctions such as is, uu, ele and a9).98 Melody Let us examine some of the main phonological arguments that support the postulation of A, [and U. One important source of evidence used in determining the set of melodic primitives is the behaviour of sounds when they are subject to phonological processing. The reasoning is that, when it moves, asound reveals its stripes. In order to be in a position to interpret evidence of this sort, we need to have a clear idea of the set of possible processes to which a sound is potentially subject. One of the most significant developments in modern phonological theory, discussed in 1.3, is a move away from the open-ended model of phonological processing that is inherent in orthodox rewrite rules. In its place has evolved a more restrictive theory in which all processes are reduced as far as possible to two fundamental operations, which may be characterized as composition and decom- position.®® Composition processes result in the fusion of melodic material. De- composition involves fission, the operation by which the fusion of elements within an expression is undone. Fission may or may not be accompanied by the suppression of all or part of the melodic material that is lexically specified in a particular position. This view of phonological activity can be further constrained by insisting that processes of composition must have a local cause. That is, they involve melodic material lodged in one position being fused with melodic material lodged in a neighbouring position. Limiting all phonological activity to these two basic operationshas the potential for excluding a significant class of non-occurring process-types that are readily expressible in terms of traditional rewrite rules. For example, it forbids in prin- ciple the random substitution of one piece of melodic material by another, something that can easily be accommodated in the input and output portions of a rewrite rule. Rather, the introduction of any melodic material into a position must have some local sources that is, it must propagate from an adjacent position. (In 3.3.5 we will consider a very limited class of cases in which, at least on the basis of our present understanding, it seems necessary to relax this principle.) This view of phonological processing becomes even more restrictive when coupled to a uniformly privative model of phonological distinctions. The two lines of thinking come together to place a strict limit on the number of processes a sound can undergo. In particular, the set of decomposition processes to which a sound is susceptible is logically limited by the number of elements of which itis ‘composed. Moreover, as we observe a sound breaking up under fission, we are privy to aspects of its melodic identity that might otherwise have remained concealed. Conceiving of phonological processes in these terms helps us identify what the individual elements are. Various types of process provide insights into the fusion of elements within vocalic compound expressions, including raising, lowering, diphthongization, monophthongization and coalescence. These constitute a sizeable portion of the recurrent processes that affect vowels and can be straightforwardly characterized in terms of operations involving the elements A, I and U. Many of theexamples of such processes to be reviewed below involve historical changes which in some form oF another have left their mark on Modern English. Not all of these, however, remain. active as processes in the present-day language. That is, although the input and Melody 99 output stages of a change may co-exist in the same lexicon as etymologicall related reflexes, this is no guarantee that they continue to be decivationally linked through some phonological process. For reasons outlined in 1. 4.3, this is particu- larly true of the procession of historical raisings and diphthongizations, collect- ively known as the Great Vowel Shift, which have affected long tuclei* Nevertheless, though now phonologically extinct, the individual stages of this particular series are observed to have active counterparts in other languages. More. over, as we will see, some of the changes are now in the of act of being reprised in English (although affecting a fresh slate of word classes), with successive develop- ments showing up as different phonological variants in different present-day dialects. Coalescence involves the mutual assimilation of pairs of adjacent sounds. A recurrent pattern in vowel coalescence produces ¢ and 0 from the sequences a-i and aurespectively. This can be seen in Zulu, for instance, where the attachment of the proclitic na- (‘and, with’) to a vowel-initial noun yields alternations such as zna-inkosi—> nenkosi (‘and the chiet’) and na-umuntu ~» nomuntu (‘and the person’), A similar effect manifests itself in the type of historical monophthongization by which the Early Modern English diphthongs ay, and aw, developed into e: and >: respectively. I will adope the practice of identifying the class of words containing 4 particular vowel sound by means of a representative head-word, printed in capitals. In this example, as shown in (3), the relevant word classes are BAIT and CAUGHT. The extent of these classes in Early Modern English can be roughly deveined on the basis of their spelling. Words in the original BAIT set are generally spelt with (ai) or (ay), for example (bait, maid, day, stay). The spelli of CAUGHT-class words includes (au, aw, augh, ough, all), as tn eevee caught, bought, call) ° (3) Earlier > later English ay > BAIT. aw CAUGHT The type of process illustrated in (3) is steaightforwardly expressible as the ‘compacting of sequentially ordered elements into a single melodic expression, The first portion of the diphthongs ay and aw consists of a skeletal position occupied by As the second portion consists ofa position occupied by an off-glide represented by I y) oF U (= w). As depicted in (4), the fusion of these elements results in a long ‘mid-monophthong ~ that is, a single vocalic expression attached to two positions: 4 @ aoe (by) aw > N N N i NN \ I \ "i fa} - fal i xx > x \ \ i oi100 Melody Later in its history, the front monophthongal outcome of the process shown in (4a) underwent merger with e: (the latter occurring in the MATE class, including words such as (make, fate, same, tale)). This vowel and its back congener (BOAT, as in (boat, home, go, road)) then became subject toa series of diphthon- gization changes, the results of which show up as different reflexes in different present-day dialects. The principal developments are the following: BAIT = MATE BOAT () @ > ey > ay o> ow > aw ‘The original mid monophthongal reflexes are retained in some dialects spoken in Scotland, Ireland and parts of England (principally the North and the West). The ‘most widespread subsequent development has been diphthongization to ey and ow. In the southeast of England, this process has proceeded as far as ay and aw, the former also showing up in the southern hemisphere.”* Representationally speaking, the diphthongizations in (5), examples of which could be furnished from any number of languages, are simply the converse of the monophthongization processes shown in (4). They can be directly expressed in terms of fission, the linear unpacking of a previously fused structure. As shown in (6a), diphthongization of e:to ay takes the form of a fission of the elements A and 1. The lone association of I to the second position of the nucleus defines the palatal off-glide of the diphthong. (6) (a) @ > ay (b) a > aw N N N N 4 | \d | Hole ow fs {A} [A] ‘As shown in (6b), the development of o: to aw is expressed in parallel fashion. 3.3.3 Autosegmental representations ‘The task of representing the intermediate developments in (5), eto ey and to ow, raises an important question about the organization of elements within melodic expressions. Up to now, I have been depicting compound expressions as matrices of elements with no internal structure. As briefly discussed in 1.3.2, this is essentially the format assumed in SPE-style linear representations. It is now acknowledged, however, that this mode of representation is inappropriate, since it fails to give expression to an important characteristic of primes ~ a propensity to behave independently of one another under phonological processing. In the sce nape Melody 101 stern phonological tradition, the formal implementation Sock atleast to the 194052” Is adoption within generative ghonclon ee the 1970s since when it has become widely accepted tht pines sotto separate ters, rather shan beng amped together in unordered bundles, With is arrangement, each prime is capable of behaving as an autonoun ere pla the last chapter, we examined evidence supporting the dissection of phono- ical representations into skeletal and melodie planes. The autosepienen sigh is thatthe melodic plane itself is spit into a series of substion, seeker nin, i occupied by a particular prime. Par the sine being, we ay eae sssoiation lines, which synchronize material onthe melodic and skeletal love upply direct link between each prime and a position. In this “botte-bra mode of representation, a melodic expression comprises a theese configuration of associations to particular elements: ssesimensional ” —_T_—— u —— up — Skeleton x x x \ ' ! * -1—4_f} 2h Employed in this way, the Funct is way, the function of association lines is to indicat particular clement is co-indexed with a particular skeletal poine The lece nae es in (7) simply reflects the constraints of representing a three dinensoral object on a two-dimensional page. It is not meant to on of precedene ona . tis not meant to imply a relation of Peoweenclements that are co-egistered with the same skeletal slot (The set, ents calls for certain notational refinements; tte introduced later. As discussed in 2.2.3, contour segments such acting oes ive Acs contour segments such as affricates com point.) For space-saving reasons and wh articulated autosegmental represen 2 sist of expressions of opposing quality attached to the same lerever expository considerations permit, fall tations such as those in (7) an be compressed ain (8) (8) Skeleton x x oxw x orot tf ft Ff * 1 I I 1 ~— 1 horton tt A Aa — | i 1 — Uv -u 4 _ io a ouie o a@ 6102 Melody ‘The vertical-line graphies here are not intended to imply a relation of dominance or dependency between elements. In other words, the association of each element 10 a position is direct and is not mediated through some other element. (Neverthe less, as we will see shortly, there is evidence to support the recognition of dependency rclations within melodic expressions. Ie will be necessary to introduce some additional notational conventions to distinguish such relations from the cco-indexing that is indicated in (7).) Some languages, such as French and German, exploit the full range of vocalic contrasts generated by the free co-occurrence of A, I and U in (8). Others fail to do so most dialects of modern English, for example, lack front round vowels. The exclusion of particular melodic combinations can be accounted for by assuming that systems have the option of conflating two or more axtosegmental tiers Conflation of the Land U tiers, for example, prevents the fusion of these elements and results in the following five-way set of contrasts: 9) x ox x x x Pororot og I—1———_u —u root -A— A—A— ie a o w ‘The independent behaviour of elements, which supports the autosegmental mode of representation, can manifest itself in one of two ways: either an clement can delete without affecting the integrity of elements on cther tiers; or it can propagate from one segment to another without dragging other elements along with it. These process-types can be directly expressed in terms of composi- tion and decomposition, effected by means of two formal operations referred to in 2.2.2: line deletion (delinking) and line insertion (spreading). The rele- vant evidence is of the same order as that adduced to motivate the split between the skeletal and melodic levels. Just as a whole melodic expression can spread without setting off changes in the skeleton, so can an individual element spread independently of other elements. By the same token, quantity stability, the phenomenon whereby a skeletal position can ride out the delinking of a complete melodic expression, is matched by element stability. That is, an clement on one tier can be suppressed, while those on other tiers remain un- scathed. The latter operation can be illustrated by the diphthongizations e*> ey and o:> ow exemplified in (5). The process involves the partial breaking-up of the two-cle- ‘ment compounds representing the mid monophthongs. In the case of e > ey, shown in (10a), the elements A and I remain fused in the first position of the nucleus; however, the link between A and the second slot becomes severed, without affecting the double association of I. In the second position, the lone association to I defines the palatal off-glide. Melody 103 ‘As shown in (10b), diphthongization of o: to ow can be treated in like fashion. This example illustrates how the link between an element and a position can be independently removed without implicating links involving other elements in the same expression. In this particular case, the delinking element, A, retains one of its associations and thus remains phonetically interpretable. ‘The same autonomy of elements is demonstrated in cases where delinking results in the total suppres sion of an element while other elements remain intact. Certain vowel-raising processes in English illustrate this point. The relatively recent changes shown in (5) can be viewed as continuations of an ongoing tendency towards vowel shifting in English. Each of the individual stages in this overall development provides evidence in support of A, I and U as the pivotal elements of vocalic structure. One stage saw mid front and back mono- phthongs raising in tandem: (11) Earlier > tater English MEET ((green, feet, feel, ...)) BOOT ((soon, loop, lose, . ..)) The original mid quality of these vowels is retained in root-level alternants with shore nuclei (which remained unaffected by the Great Vowel Shift); compare, for example, (keep-kept, serene-serenity, lose-lost, goose-gosling). Viewed in terms of its elemental effects, mid-vowel raising is a reduction process. That is, it involves the suppression of part of the melodic content of a sound (in this case A): (2) @ >i eee N N N N IN NN N NN Ve AE Vi NF " in vr 1 1 I I u u Logically we should expect another type of reduction process to affect mid vowels, one involving the loss of I or U. In this case, decomposition results in lowering to a, the manifestation of the remaining A element. An example from the104 Melody history of English is the lowering and unrounding of Early Modern short 2 to a (eg. in (pot, top, lock, solid)), the effects of which show up in some present-day dialects.” ‘The cates of diphthongization and easing represented in (10) and (12) exer ify one type of evidence supporting the autosegmental mode of representation — the ability of procewes to target mclodic expeesions in wack a way that they selectively take out particular elements without disturbing others. As mentioned above, the other main source of supporting evidence comes frcm the observation that part of an expression can display mobility under processing while the rest remains inert. We will see numerous examples of this phenomenon in the follow- ing pages. A rather simple example will give some preliminary idea of what is involved. In many types of English, the direct juxtaposition of vowels across a word boundary tends to lead to the ereation of an intervening segment, typically a glide ‘or a glottal stop, as in (two [w] of, two [2] of). The excrescent segment can be viewed as a hiatus-breaker which fills the otherwise vacant onset between the two nuclei. The glide realizations, w and y, only occur after a certain class of vowels, namely i us (see (13a)) and, depending on the dialect, either up-gliding diph- thongs (ie. those ending in y or w) or the monophthongs e a(see (13b)) (13) (a) iV > in three [y] and uN > aw two [w] of (b) eV > eyyV, eV > EW day [y] of ow owwV, oV—>awV go [w] and (c)_*Shah {y] of, *Shah (w] of The backness of the glide is evidently determined by the nature of the preceding vowel - front yafter a front vowel, back w after back. Moreover, the conditios vowel must have some element of palatality or labiality init, which explains why neither glide appears after a low non-round vowel (see (13c)). (Some dialects show so-called ‘intrusive’ r in the latter context, as in (Shah [r] of, but that is another topic for another time (chapter S).) The hiatus glide can be straightforwardly explained as the spreading of an clement from the first nucleus into the vacant onset. Assuming that a skeletal slot is automatically created in the onset to accommodate the incoming element, we can see in (14a) that it is lin a vowel such as i that gives rise t0 y. (14) (a) NV iw (b) uV > uw IN 1 1 IN 1 1 VV I VV | Melody 105 By the same token, as shown in (14b), itis U in a vowel such as u that yields w (V indicates some vocalic expression). In dialects with up-gliding diphthongs, the excrescent glide is simply the continua- tion of the I or U that is the lone occupant of the second position of the nucleus. What is significant from the viewpoint of the present discussion is that exactly the same process is operative in dialects which have long mid monophthongs (e: and er) in place of up-gliding diphthongs (ey and ow). This means that I and U can spread independently of the element A, with which they are fused in the triggering nucleus: (15) (a) eV ey (b) a > ow N ON N ON ee moto xx x x xx ox x A 1 In this independent spreading behaviour we find further evidence in support of the autosegmental mode of representation. 3.3.3 Dependency relations between elements Most approaches to melodic structure which incorporate the primitives A, I and U subscribe to the notion that fusion involves some kind of asymmetric relation between the elements of a compound. This is understood as a means of expressing the notion that the phonetic manifestation of a compound reflects the preponde- ance of one element over another. The most widely established implementation of this idea has been to designate one element in a compound as the head of that expression and any other elements that may be present as dependents.» In a simplex expression, the lone element is the head. ‘Take for example two A-Icompounds, one headed by I, the other by A. Informally, we can think of the Lheaded expression as an open version of an essentially palatal vowel; the A-headed expression meanwhile can be considered a pala- talized version of an essentially open vowel. Let us assume that these asymmetric fusions define the vowels e and & respectively (head element underlined):* (16) (a) (b) Boman o bon x @106 Melody By the same token, the asymmetric fusion of A and U defines the vowels o and respectively: (17) (a) (b) p-le-x © Imoa-x 2 The fusion asymmetries illustrated in (16) and (17) provide a straightforward ‘means of representing widely attested processes of raising and lowering involving low and mid vowels. One example involves the portion of the English Great ‘Vowel Shift which produced the inputs to the changes shown in (5). Early Modern English ex and or are the raised reflexes of Middle English 2: and or respectively: (18) Earlier > later English ez oe MATE = BAIT & >o BOAT In terms of their elementary make-up, the inputs and outputs of each of these raisings are isomers.» That is, the elements of which they ar: composed are identical but are arranged in different ways: A is a head in both #: and but a dependent in both e: and a. In other words, the raising of low vowels involves neither the loss nor the addition of elements; it consists rather in a switch in the headedness of a melodic expression: (9) @ = > ) & > & i 7 i 4 1o4 uo ou io 17 AOA AA Lowering of mid vowels presents the inverse operation, in which A in a com- pound switches from dependent to head status. This can be illustrated by one of the stages in the recent history of originally short 2 briefly mentioned above. (The representational difference between 9 and o is something to be taken up later.) This vowel displays three main dialect reflexes in present-day English, each of ‘which recapitulates a stage in its diachronic development: (20) 9 > 2 > a POT ({top, lock, solid, . ..) ‘The sequence of events involves the following representational changes, one of which (21b) takes the form of an isomeric alteration: Melody 107 (21) fa) 9 (b) 9 (lowering) {c) a (unrounding) x x x I i u u r ' Ss S A In what follows, I will only indicate headedness where it is, the point at hand. ctly relevant to 5.3.4 Elements and sound patterns At this point, itis in order to say something more about the phonetic interpreta- tion of elements and element fusion. One question is whether the phonetic specification of elements should be couched in terms of the articulatory mechan- isms of the speaker, or the acoustic signal, or the auditory perception of the hearer. In view of the assumption that a grammar, as a model of linguistic competence, is neutral as between speakers and hearers, it might reasonably be expected that each element should be definable with reference to all three do- mains. From a purely procedural viewpoint, however, it would seem to make sense to begin with the acoustic dimension, since this is the communicative experience that is shared by speaker and hearer. This stance was explicitly adopted by Roman Jakobson in carly distinctive-feature theory but was largely abandoned in generative phonology by the time of SPE..” Since the 1960s, the primacy of articulatory-based definitions has gone largely unchallenged in feature theory. The assumption made in much of this work, usually implicitly, is that the articulatory specifications of features can be mapped (not necessarily directly) conto acoustic and auditory specifications. * It is certainly possible to provide articulatory definitions of A, I and U. This can be done by referring to the two major resonating cavities, the oral and the pharyngeal, which are involved in the production of a, i and u, the independent manifestations of the elements in question. The lowering and retraction of the tongue body that produce a result in an expansion of the oral tube and a constriction of the pharyngeal tube. In i, the oral cavity is constricted and the pharyngeal cavity expanded. The production of u involves a trade-off between expansion of both the oral and pharyngeal tubes. Nevertheless, we should not lose sight of the fact that phonological primes are not in themselves articulatory events. Rather, like all aspects of phonological representation, they are cognitive categories which serve the grammatical func- tion of helping to code lexical contrasts. From a Jakobsonian viewpoint, the phonetic interpretation of these categories involves in the first instance a mapping onto the acoustic signal. Speech production and perception are then to be con- sidered parasitic on this mapping relation.108 Melody Applied to the model of melodie content being outlined in thischapter, this view implies that elements should be thought of as internally represented objects which map onto sound patterns. Sound is being used here in its literal sense to refer to the acoustic signal, rather than in the loose metaphorical sense frequently en- countered in articulatory definitions of segments. That is, elements constitute pattern templates by reference to which speakers orchestrate and monitor articu- latory output and listeners decode auditory input. In speech production, the speaker marshals whatever articulatory resources are available or necessary to create signal mappings of particular elements. In speech perception, the listener seeks to detect sound patterns which can be matched to particular elements. Viewed in these terms, the phonetic effects of element fusion consist primarily in the amalgamation of sound patterns. The articulatory results of the operation are then specified by identifying whatever vocal-tract gestures are required for achieving particular targets represented by composite sound patterns. , ‘One of the arguments in favour of this view relates to the indirectness of the link between the acoustic identity of a sound and the articulatory mechanisms by which it is produced. For example, it is not always possible to establish a direct correlation between the acoustic characteristics of a vowel and traditional articu- latory labels referring to tongue position, such as [high], [low] ce [back]. Features such as these are usually defined in terms of the point of maximum tongue constriction relative to the palate. But this factor on its own dees not provide an invariant specification of the supralaryngeal vocal-tract configurations that generate the acoustic profiles of different vowels. It is by manipulating the overall shape and size of the supralaryngeal airway that the speaker targets particular sound patterns (as any good ventriloquist will confirm). In vowel production, determining factors other than tongue contour involve adjustments to total vocal- tract length, including the height of the larynx, the size of lip aperture, and the relative protrusion or retraction of the lips. Most vowels can be generated by means of many different articulatory configurations. A given token of a vowel sound classified as {low}, for example, may indeed be produced with a lower tongue position than a vowel classified as {high]; but another token of the same sound may actually be achieved with a higher tongue position than a high] vowel.” This is perhaps not the place to provide detailed specifications of the signal ‘mappings of elements or of the acoustic consequences of fusion. Interested readers can pursue the matter in the appendix to this chapter (3.8). 3.3.5 The neutral element The sound patterns associated with a, i and u are inherently large and distinct. This is a reflection of the fact that these vowels, the universally limiting articula- tions of the vowel triangle, represent extreme departures from 3 neutral position of the vocal tract. The supralaryngeal vocal-tract configuration associated with the neutral position approximates that of a uniform tube and generates a schwa- Melody 108 like auditory effect. The resonating characteristics of this configuration are such that it lacks the bold sound patterns found in a, ‘and u. Most researchers within the AU tradition accord this neutral quality some special status, either by treating it as a segment devoid of any active elementary content or by taking it to be the independent manifestation of a fourth element, which I will symbolize here se.” Building on the colour metaphor, the clement @ can be thought of as a blank canvas to which the bold strokes represented by A, I and U can be applied, From the perspective of production, this metaphor reflects the point just made that A, I and U are realized by means of articulatory manoeuvres that perturb the vocal tract from its neutral state. From an acoustic point of view, the amorphous signal presence of @ constitutes a base-line on which the well-defined sound patterns associated with A, I and U are superimposed. Schwarlike vowels can occur in stressed nuclei; they are to be found, for example, among the various reflexes of the vowel in CUT ((cut, blood, sun), etc.) and, in non-rhotic dialects at least, of that in THIRD ((third, shirt, burn), etc). However, realizations of this sort are particularly prevalent in contexts favouring vowel reduction. This is the phenomenon whereby prosodically recessive nuclear positions display a considerably smaller set of vocalic contrasts than are found in ‘other positions. Positions of this type manifest recessiveness in one of a number of ways: they may bear weak stress, for example; or they may occur within a harmonic span in which their melodic identity is partially or wholly determined by a neighbouring vowel; or their presence in a representation may be solely motivated by syllabification constraints. In such contexts, many languages display only one vocalic reflex, its quality varying from system to system. In Spanish, for instance, it is e, in Japanese 4 and in Telugu (Dravidian, southern India) u*! Reflexes of this type can be thought of as default vowels which show up spontan- cously when no other quality is to hand. Although the elementary composition of such vowels is not lexically distinctive, it cannot be attributed to spreading from 4 neighbouring position. Cases of this type thus appear to call for a relaxation of the requirement that all phonological processes have a local cause. The appear- ace of a default element may be viewed as reflecting a language-specific quality that is latently omnipresent in representations. Of the various vocalic reflexes that appear in recessive nuclei, by far the most common are those of schwa-like quality. Frequently, the curtailnuent of distinctive potential in such contexts involves the neutralization of peripheral vowel qualities under a centralized reflex. In English, a stress-sensitive version of this phenome- non shows up in root-level alternations between full and reduced vowels in forms such as (photograph) fowrsgre#{photégraphy) fotografi. Indeed in some lan- guages, a schwa-like reflex is the only vowel to occur in prosodically weak contexts. Because of this behaviour, schwa is widely acknowledged to be the reduction vowel par excellence. That is, it isthe default reflex that surfaces when other vocalic material is absent or suppressed. ‘As a means of transcribing reduction reflexes, the symbol 9 is used to cover a relatively wide range of qualities, some of which are not even central. (For110 Melody ‘example, employed in the transcription of French, the symbol stands for a vowel of mid front round quality, as in promye ‘first”.) Even amongst non-peripheral reduction reflexes, there can be considerable variation in the quality of vowels symbolized as 2, Rather than being symptomatic of sloppy transcriptional prac- tice, this fluctuation can be put dowa to the fact that the distinctive potential associated with the non-palatal non-low region of the traditional vowel diagram is not as great as its celatively large area might suggest. Variability between systems with respect to the favoured reduction vowel can be taken to reflect differences in the fixing of the base line on which other resonance components are superimposed. From a speech production viewpoint, this variability is sometimes characterized in terms of different articulatory or vocal settings. ‘The range of values attested in the non-peripheral region can be identified by reference to the neutral quality represented by the element @. Other centralised categories that are potentially distinct from this baseline can then be thought of as displaced versions of the neutral quality, expressed as the fusion of @ with some other element. For example, the relatively open w value of the vowel transcribed as o for some types of English and other languages (Catalan, for instance) can be represented as a compound of @ and A. In some dialects of English, this relatively open quality is distinct from a closer non-peripheral vowel, as illustrated by the contrast between the second nucleus of (Rosas) (with 2) and that of (roses) (with . This suggests a distinction between (A, @] and (@]. Let us take a closer look at the English alternations between full and reduced vowels in forms such as (photograph-photography). The full vowels occur in nuclei bearing primary or secondary word stress, while reduced reflexes show up in alternants in which the nucleus is unstressed. These alternations are all of the root-level type, which means that, following the line of reasoning outlined in 1.4.3, vowel reduction in this case takes the form of a static distributional regularity. That is, words such as (photographic) and (photography) are not phonologically derived from (photograph). ‘There are two main types of vowel-reduction system in English, illustrated by the emboldened characters in (22). In one, System A in (22), schwa occurs as the sole weak reflex; in the other, System B, an additional weak reflex, 1, is attested. (22) (a) Reduced vowel =a (Systems A and B) > product-production, photography-photograph ow —_photograph-photography ® photograph-photography {b) Reduced vowel =o (System A), 1 (System B) é demon-demonic ay _horizon-horizontal e telepathic-telepathy Note that the contrast between reduced rand 2 in System B retains one aspect of the distinctions holding of full vowels: the weak reflex is o unless the correspond- Melody 111 ing full vowel is front or ends ina front glide. In element terms, the latter pattern implies that the only element supported by the weak nucleus in question is I, a proper subset of the expressions defining various types of front vowel in the corresponding stressed nuclei. Failure to sustain other than I in a weak context may be seen as a distributional analogue of dynamic processes of melodic decom position such as we have already seen at work in various forms of vowel shifting, Extending the decomposition account to cases involving weakening to schwa would have the desirable consequence of providing us with a unified treatment of vowel reduction. It is not immediately clear, however, whether such an extension is possible. The problem is that the reduction of a full vowel to 2 apparently involves substitution of one set of elements by another. Alternations between, fot example, band 2 or between € and 9 would be characterized as follows: 23) (a) oo ) ee - 2 (4U) fe) an [ey On the face of it, the expression defining 2 is not a proper subset of the expressions defining full vowels. The operation of element substitution that this implies runs counter to the principle that the class of possible phonological Processes is restricted to operations of delinking or spreading. However, viewing reduction to schwa in terms of substitution reflects a miscon- ception about the behaviour of @ under fusion. A solution to the apparent problem in this case lies in the recognition that @ fails to display the sort of active resonance properties that characterize the other elements discussed so far, ‘The notion that @ is without an active resonance component can be captured by assuming that the only circumstances under which its autonomous phonetic signature will be made manifest are when it occurs as the head of an expression. Ie will fail to contribute anything to an expression in which it occurs as a dependent. This silence is consistent with the assumption that @ is latently present as a dependent in all vocalic expressions and has the potential to become audible only when other elements in a compound are suppressed for some reason. In other words, rather than occupying its own melodic tier, @ should be viewed as being resident at any intersection of a tier and a skeletal position that is not already filled by some other element. This means that the expressions corresponding to 4 iand mare most accurately represented as in (24). (24) (a) (b) i (ou @->—% » Ic-@-x I The @-headed isomers of these expressions define Is expres lefine various non-peripheral Qualities, including the relatively open vowel already discussed:112 Melody (25) fa) & 8 v0 t I 1 A @ @ 1 1 T @ I u The representation of lax vowels such as sand u as @-headed compounds is a matter to be considered in more detail in the next section. The idea that @ defines the baseline on which other resonances are superim- posed is thus implemented by assuming that it is omnipresent in vocalic expres- sions but fails to manifest itself wherever it is overridden by the presence of another element. Viewed in these terms, reduction to a centralised vocalic reflex does not involve the random substitution of one set of elements by @. Rather it consists in the promotion of a latently present @ to the status of head of an expression. Under such circumstances, other elements are either totally sup- pressed (through delinking/deletion) or demoted to dependent status. As depicted in (26a), the first operation derives reduction to some schwa-like reflex; the second, illustrated in (26b), derives reduction to one of the other centralized reflexes given in (25). (26) (a) i> 2 ios N N NON H 1 1 1 i 4 an 1 @ 1 I i 1 1 1 e e @ e 1 will henceforth follow the practice of recording the presence of @ in an expression only when it occurs as a head. ‘One of the advantages of viewing centralization in these terms is that it unifies the representation of the process with that of certain processes of raising and lowering which, although not involving reduction to non-peripheral reflexes, nevertheless occur under the same prosodically weak conditions. A widespread phenomenon in the world’s languages is a tendency for mid vowels to be banished from prosodically recessive nuclear positions. In metrical systems, recessiveness refers to positions of weak stress; in harmony systems, it refers to nuclei whose harmonic identity is determined by an adjacent dominant position. Under such conditions, it is common to find neutralization of vocalic contrasts in favour of cither non-peripheral reflexes or the ‘corner’ vowels a, i, u or some mixture of both. In Bulgarian, for example, the stressed five-vowel system (i, ¢ a 0, 4) contracts to three vowels under weak stress: a undergoes centralization to w, while He and u-o are neutralized under i and u respectively. In Catalan, a stressed Melody 113 seven-term system (i, €, 6 & 9, 0, u) gives way to three terms under weak steess: the a-e-e contrast is neutralized under #, 2~o-u under u, and i remains as i: In vowel-height harmony systems, such as occur for example in Central Bantu languages and certain Romance dialects, mid vowels are barred from appearing in harmonically recessive positions unlecs eanctioned by the presence of anid vowel in the harmonically dominant position.” The fact that the processes just mentioned, raising or lowering of mid vowels and centralization, all occur in the same general context indicates that we are dealing with a single phenomenon. Although this commonality has long been recognized, it has not always been clear how it should be captured formally. In terms of element structure, however, the processes in question are uniformly expressible as decomposition (with or without a concomitant switch in the headedness of the expression). As illustrated in (27), all involve the total or partial suppression of melodic material. (27) (a) Raising (b) Lowering (c) Centralization N N N N N N 1 I 1 ' 1 I x o> x xo x xs x 1 I ! I A AoA A i i i uu u u 3.3.6 English vowel systems This is an appropriate point at which to exemplify the combinatorial possibilities of the elements introduced so far by providing a specification of the vocalic contrasts utilized in English. The size and shape of the maximal inventory of distinctions appearing in stressed nuclei varies from dialect to dialect. For illustrat- ive purposes, however, it will be sufficient to focus on those patterns that are Widely reported in the literature. ‘The classic descriptions of English vocalic phonology typically begin by noting 4 fundamental typological property that the language shares with its West Ger- ‘manic cousins such as German and Dutch ~a bifurcation of the inventory into sets of vowels referred to variously as long versus short or tense versus lax.** (The tense—lax dimension, it is now widely recognized, can be subsumed under the distinction between advanced tongue root (ATR) and non-ATR vowels.) The differential phonological behaviour that reflects the long-short dichotomy mani- fests itself in a number of ways. In English, for example, phenomena involving only vowels from the long set include an ability to attract stress and an ability to ecur in word-final stressed open syllables.114 Melody It is usually acknowledged that, at least as far as the West Germanic family of languages is concerned, long ends to correlate with tense ard short with lax, However, there has not always been agreement on the question of which of the two 4 dimensions is the more basic in English. In SPE, it is assumed that length is derivative of tenseness. According to an older tradition, length is the fundamen tal distinctive property, which is not necessarily uniformly correlated with tenseness.% The latter view has re-established itself in more recent generative theory.*" This is partly a reflection of a desire to restrict the conditioning of metrical processes to aspects of prosodic structure. As discussed in 2.3.2, the quantity-sensitivity of stress assignment is expressed by granting metrical pro- cesses access to the arboreal structure internal to rhymes. By representing the contrast between long and short vowels as a difference between branching and non- branching nuclei, we derive the result that heavy nuclei pattern with closed rhymes in terms of their ability to attract stress. To grant metrical processes access to melodic material, including whatever component is involved in the expression of tenseness, is to open the way for the generation of unattested systems in which stress assignment is sensitive to such dimensions as vowel height, backness or roundness. ‘Once length is acknowledged to be the fundamental dichotomizing property of the English vowel system, the question of how tenseness/ATR should be represented takes on a good deal less significance than when considered in relation to many other languages. English shows litte evidence of phonological activity involving the ATR ‘dimension; it certainly has nothing equivalent to the vigorous ATR harmony systems found in many other languages (particularly well represented in West Africa). “There have been two approaches to the characterization of ATR in privative frameworks. One is to recognize an independent ATR element. The more widely adopted solution, is to derive the contrast between ATR and non-ATR vowels by means of different combinations of the established A, I or U elements. According to the specific implementation of this proposal briefly illustrated in the last section, lax/non-ATR vowels are represented as neutral-headed versions of their tense/ATR counterparts. This provides us with contrasts such as those depicted in (28) - long tense i:versus short lax 1 (28a) and tense diphthongal ey versus short lax « (28b).. (28) (a) é: 1 (b) ey e N N N N my ' rN ' xx x xx x Vr | 1 1 | A A 1 1 I I 1 I @ I ~ e ‘The full set of neutral-headed compound vowels generated by the theory is seven, One of these is (A, @], the open e described in 3.3.5. The other six define Melody 115 the lax counterparts of the peripheral non-low vowels that are generated by the remaining possible combinations of A, I and U: 29) (a) 1) ‘ W,@] ov [A,1, @] © 1A,U,@] > () 0,0, @) a (ALU,@) « Lax vowels in English are typically drawn from the subset in (29a), although the precise selection varies from dialect to dialect. Standard German is an example of a language which displays the full range of contrasts in (29). ‘The most familiar subsystem of short stressed vowels in English is the six-term pattern that is historically associated with the south of England and is now widely «established across the English-speaking world. Typical realizations of the six-way contrast, observed in the nuclei of BIT, BET, BAT, POT, CUT and PUT, include a subset of the neutral-headed expressions in (29), supplemented by a selection headed by A: (G0) @ BT + LQ (b) BET e [A,L@) () BAT 2 {1,4} @ PUT ov (U,@) () POT» [U,A} a [Al cur 4 f@ = (Ae Well established realizational variants of two of the vowels are shown in (30e) and (308). (The a reflex in POT occurs in parts of the United States, Ireland and England, the o reflex in Canada, the southern hemisphere, and other parts of England and the USA. The more open e variant of the CUT nucleus is usual in southeastern England and the southern hemisphere.) Consideration of the long sub-system of vowels returns us to a discussion of the Great Vowel Shift and some of the more recent diphthongizations mentioned earlier in the chapter. With the element @, we are now in a position to fill in a ‘umber of representational details involving vocalic developments not yet touched on. The historical raising of originally mid e: and o:in MEET and BOOT shown in (11) might have been expected to threaten merger with originally high rand u: in BITE and SHOUT. That this did not happen was due to the fact that the latter were deflected out of the path of the raising monophthongs through diphthongiza- tion, first to ay and aw (reflexes retained in some present-day dialects), and subsequently to ay and aw. In a recapitulation of the initial stage of this process, some present-day dialects in England show the effects of a more recent diphthongization of the MEET and BOOT vowels to ay and aw. Representationally, this ‘breaking’ of an original monophthong involves a dissociation of either the element I (in MEET) or U (BOOT) from the left-hand position of the branching nucleus. As shown in (31b) and (314), this allows latent @ to manifest itself in the vacated slot:116 Melody G1) (a) é (b) ay (ow (a) ow MEET BOOT N N N a wN\ wN iN W\ a 1 Vi | I u io Pog e A similar proposal accounts for one of the developments to have affected the vowels of the MATE and BOAT classes. Historical 2: and arin these classes were first raised to e: and or, reflexes which survive to this day in some conservative dialects. As shown in (5), these vowels have been subject to subsequent diphthong. ization, first to ey and ow and more recently in some dialects to ay and aw. Another diphthongization, not yet mentioned in connection with these vowel classes, has resulted in the breaking of e: and o: to ia and uo. Dialects showing these reflexes include some spoken in the northeast of England, parts of Ireland and the Caribbean. Representationally, this development has involved the sup- pression of A (producing raising) accompanied by a severing of the link between the remaining element (1in MATE, U in BOAT) from the second nuclear position, ‘As in the reflexes represented in (32d) and (32h), this has resulted in the surfacing of latent @, giving rise in this case to a 9 in-glide. The full range of MATE and BOAT reflexes just discussed can now be summarized as follows: (32) MATE fa) e ) (ay @) ie N N a rN " aS \ Vi AF AT You ap fk i 1 1 p 1 L BOAT () « ) ow (8) aw ee N N Mi iM TN rN iN IN VT 1 | ! A A A es j i u u u u Melody 117 The vowels fa and uo are examples of in-gliding diphthongs (those endin schwa) which have developed spontaneously from original suonovbehonre: chapter 5, we will discuss the representation of another clas of in-gliding vowels those that result from the presence of following historical ras in fi (fens). The remaining upgliding diphthongs in the system (i.e. those ending in y or w) are those found in the word classes BITE, SHOUT and TOY: (33) (a) ay BITE (b) aw SHOUT () sy Toy Finally, there are two long-vowel classes, CAUGHT and CAL lasses, and CALM, the seatus of which differs quite dramatically from dialect to dialect, in terms not only of their realization but also of their distribution across the lexicon, Both elazeee Parisian various patterns of eoatextal or total merger with each other at well as with other classes. The mesger patterns include CAUGHT CALM, CAUGHT POT, CALM-BAT, although there are some dialect in which either e,or both ofthe clases in quesonsemain distinc.” Typical realization age (34) (a) os CAUGHT (bl) @ CALM —(c)_ & CAUGHTT/CALM N N N ry ray Wy x x x x x x vy iw] Vs A t u cn The vowels in (34) clearly fall into the long sub-system, as demonstrated by their ability o occur in final stressed open syllables, as in (paw, law) (CAUGHT) and (ma, bra) (CALM). In some dialects, the classes have fallen together with classes containing on reflexes of historical r, so that the nuclei of CAUGHT. a ’ for example, are identical, as are also the nuclei ang FARM. (More on this in chapter 5.) ° OF CALM and118 Melody 3.4 Elements for consonants 3.4.1 Place ‘The melodic composition of consonants is frequently described in terms of the traditional articulatory dimensions of place, manner and voic:. While this ar- rangement might suffice for presentation purposes, the distinction between place and manner will tum out to be somewhat blurred when viewed from an element- based perspective. I will postpone discussion of the laryngeal elsments until 3.6, Until chen, ic is possible to abstract away from this dimension as we focus on the other two aspects of consonantal representation. {As regards the place dimension, one question that can be posed right away is whether the elements used to represent vowels can and should be extended to expressions attached to non-nuclear positions. In SPE, the cavity characteristics of vowels and consonants are represented by means of separate sets of features, a ‘view that is still held in some current frameworks. However, the majority position nowadays, based primarily on evidence involving assimilatory interactions be- tween adjacent consonants and vowels, is that identical or overlapping sets of ‘components are appropriate for both types of sounds (the so-called “one-mouth’ principle). The widespread palatalization of a consonant in the vicinity of a front vowel, for example, can be straight forwardly expressed as the spreading of a single palatal component from a vowel to a consonant. [As far as the elements introduced up to now are concerned, we have already seen (2.4.2) how the onset glides y and w are simply instances of the elements I and,U occurring outside nuclei. The participation of these sounds in certain assimilatory processes justifies the decision to deploy the same elements in nucleat and non-nuclear contexts. One such process is the palatalization phenomenon just referred to. As it affects coronal plosives, this produces one of two main results ~ an alveolar with a secondary palatal articulation (as in Russian and Irish 1d"), oF a displaced palato-alveolar affricate. As briefly discussed in 1.1, itis the latter development that occurs widely in English. In collocations containing @ ‘word-final alveolar followed by y, recall, it is quite usual in connected speech to find palatal assimilation: G5) te dj sos 243 bet you did you kissyou buzz you hityou would you bless you faze you This assimilatory behaviour appears quite natura if itis assumed that the element representing the palatal glide spreads into the preceding position occupied by the coronal. This indicates that the distinction between a plain and a palatalized consonant consists in the presence of I in the latter but not in the former. In the case of P/d’, palatalization can be assumed to result in a contour segment. AS Melody 119 noted in 2.2.3, these are segments containing two separate melodic expressions attached to a single skeletal point. In this particular instance, the structure is one in which the incoming 1 fails to fuse with the other elements in the position affected by spreading. With the palato-alveolar series, I does fuse with other elements in the segment. In this ease, it is necessary (o account for the additional sibilance that accompanies affrication. We will return to this issue once we have had the opportunity to consider the manner dimension of consonantal contrasts in more detail. As an example of assimilation involving U in an onset, we may note the historical change whereby original in many dialects became rounded and retracted to p following 1, as in (want, wad, swan, quality). The original front vowel is retained in all contexts in some conservative dialects (in Scotland and Ireland, for example) as well as in all dialects in forms containing following velars, which blocked the change (as in (wax, wag, twang)).” The rounding is straightforwardly expressed as the spreading of U from the onset glide into the following nucleus. The conflation of the I and U tiers, responsible for the absence ‘of front rounded vowels in English, means that the I in & is usurped by the incoming U. Spreading of U produces rounding, while suppression of I produces retraction. This is depicted in (36) (where indicates the delinking of I). (36) oN 1 x x we > wo 1 A A ur The spreading of I to create palato-alveolars constitutes one piece of evidence favouring the extension of the resonance elements to sounds which, unlike glides and vowels, are produced with consonantal constriction. It is a natural step to conclude that the element U inheres in labial consonants and A in uvular and pharyngeal consonants (which, like a, are produced by lowering and retracting the tongue body). Since @ independently defines a sound which is articulated as. dorsal (produced with the tongue body), non-palatal, non-lahial and non-open, it can be taken to specify velarity in consonants. By employing the four main resonance elements reviewed so far, we thus identify all the major cavity dimen- sions of consonants save one, coronality, which evidently requires an independent element. 3.4.2 Lenition: doing things to at In seeking to determine the additional elements needed to define coronality as well as the stricture characteristics of consonants, we can turn to the same sort of.120 Melody evidence as that adduced for the resonance elements introduced so far. That is, by ‘observing the behaviour of consonants under phonological processing, we can hope to gain an idea of how they are made up. The consonantal equivalent of the vocalic fission processes that produce diphthongization and vowel reduction generally goes under the name lenition or weakening. The class of processes falling under this rubric includes vocalization (weakening to a glide or liquid, as in p— w), spirantization (the development of a plosive into a fricative, as in ¢—> 4) and debuccalization (loss of supralaryngeal gesture, as in sf) ‘The term lenition makes implicit appeal to the notion of relative segmental strength. Unless interrupted for some reason, consonantal weakening processes typically pass through a series of stages which ultimately culminate in segment deletion.“ The weakest sounds are those that constitute the penultimate stage of this progression; that is, they represent the last vestige of a segment before it disappears altogether. The strongest sounds are those occurring at the other end of the scale.“ In articulatory terms, lenition manifests itself as an opening of ‘consonantal stricture. So for example, close approximation of the articulators marks a fricative gesture out as weaker than the complete closure of a stop and as stronger than the open approximation associated with a glide. As they affect oral consonants, opening processes follow one of three trajec- tories: (37) Opening trajectories (a) Spirantization > ‘aspiration’ > deletion plosive > fricative > h > o (b) Glotzalling > deletion plosive > 2 > § (6) Vocalization > deletion non-continuant > resonant > § ‘The notion of trajectory here refers in the first instance to the historical develop- ‘ment of sound changes. It is not meant to imply that every lenition process inexorably leads towards elision. As noted above, progression through the vari- ‘ous stages on a particular path may be arrested at some point. The synchronic relevance of this observation is that two or more stages on a particular trajectory may be retained within the same grammar as stable alternants or distributional variants. English provides numerous examples which illustrate this state of af fairs. A good place to start a review of this evidence is with t which in recent times has proved to be one of the most unstable consonants in the language. Examining the fate of this sound under lenition teaches us a good deal about the internal composition of consonants in general. ‘At various times and in various dialects, English thas found itself on all three of the weakening trajectories shown in (37). The phonological contexts in which it has been susceptible to lenition can be identified informally as word-final and inter-vocalic within a foot: Melody 121 (38) Word-final: (pet, bit, late, boat, ...) Foot-internal: (pity, photograph, Peter, ...) In particular cases, t has been targeted by lenition in either or both of these coutexts. Au important question concerns why it should be this particular combina~ tion of conditions that favours ¢lenition rather than any other. This is an issue that we will devote a good deal of attention to in the next chapter. What is relevant for our present purposes is the set of processes to which ¢has been subject in these environments, One of the opening processes to have affected fin one or both of the contexts in (38) is glottalling (37b), a phenomenon that is particularly prevalent in Scotland and England. This takes the form of debuccalization, the loss of the coronal gesture, with the residual reflex being realized with glottal stricture, as in (bife]), pilzly).® ‘Another development has resulted in tapping or flapping ~ vocalization to tap 1 (37c), a process that has also affected d (as in (teady)). Tapping is firmly established in most of North America, as well as in Australia, Ireland and parts of England. Tapping (in English at least) generally occurs in pre-vocalic contexts; that is, it takes place both foot-internally and word-finally before a vowel. However, as illustrated in (39), word-finally before a consonant or pause, tapping dialects show an unreleased and sometimes pre-glottalized reflex of t (indicated here by 9. Word-final Before V Before C or pause pilely file] us fife} me (39) Foot-internal ‘Yet another weakening process to have affected © is spirantization, a phenom- enon that is most firmly established in Ireland and the Merseyside area of England. The fricative reflex, illustrated in (40a), can be of the ‘slit’ type, in which case it remains distinct from the grooved manifestation typical of original s. (40) (a) ges (get) —eso—_(letter) () eh Ge, oh (not) deh (that) buh (but) Nevertheless, the potential for merger here is sometimes realized, with the result that, for some speakers at least, (letter) can be identical to (lesser). As the forms in (40b) show, spirantization has given way to debuccalization to h in function words which typically bear weak stress. Summarizing, we may note that lenited tin modern English can show up as one of the following reflexes: glottal stop, tap, unreleased coronal stop, coronal fricative, or h, One of the advantages of a framework in which melodic expres- sions are composed of privative, independently interpretable elements is that the Processes responsible for these reflexes, and indeed lenition processes in general,122 Melody receive a uniform representation. Specifically, every weakening process can be characterized as the suppression of some aspect of the elementary content of ‘a segment. According to this approach, lenition along each of the trajectories in (37) involves a progressive decrease in the melodic complexity of a seg. ment, where complexity is straightforwardly gauged by the number of elements ‘of which a segment is composed. Thus each trajectory corresponds to a scale ‘of complexity, where the strongest segment-types are the most complex and the weakest the least complex. It then makes sense to suppose that the ‘weakest segments, those occupying a stage immediately prior to deletion, contain only one element. From the perspective of a framework in which elements are hheld to be independently interpretable, this observation is highly significant: pre-deletion stages in lenition chains allow us actually to ‘hear’ individual ele- The full set of ‘primitive’ segments which show up as pre-deletion targets in lenition is as follows: (a1) Pre-deletion Element target Glottalling 2 ? ‘Aspiration’ =k h Vocalization R y 1 w u y e (x symbolizes a velar approximant.) Given the line of argumentation being pursued here, each of these sounds should be the autonomous phonetic instantia- tion of a particular element. And indeed, three of them are independently motiv- ated as the manifestation of resonance elements already introduced, namely w = U, y=Land y= @. (y indicates the non-nuclear counterpart of i, in the same way that w, say, corresponds to u.) The lenition facts suggest the need to recognize three additional elements, independently realized as r, ?and hand labelled in (41) as R, ? and h respectively. 3.4.3 Manner’ The element ? may be thought of as a stop or an edge pattern which maps onto the acoustic signal as an abrupt decrease in overall amplitude In articulatory terms, this effect is achieved by a non-continuant gesture of the type that charac- terizes oral and nasal stops and laterals. In isolation, the element is interpreted as a glottal stop, since this is the only articulatory means of achieving an amplitude drop without introducing resonance characteristics into the signal. That is, glottal is not a defining property of ? but is simply an articulatory by-product of the Melody 123 speaker targeting an edge elemental pattern. In compound expressions, the loca- tion of the constriction necessary for producing ? is determined by one of the other constituent elements. For example, the articulatory execution of a com- pound expression containing U and ? calls for a sustained vocal-tract closure at the lips, resulting in a labial non-continuant. Stops with other places of arricula- tion are formed by the fusion of ? with I (palatal), @ (velar), A (uvular), or R (coronal). ‘The signal mapping of the element R is the set of resonance transitions associ- ated with coronal gesture. When not harnessed to any other component, the transition is rapid; in articulatory terms, this means that the independent interpreta- tion of R is a coronal tap. The elemental pattern of h may be defined as noise, which maps onto the speech signal as aperiodic energy. The articulatory targeting of this effect involves a narrowed stricture which produces turbulent airflow. This element contributes a noise component to the class of obstruent consonants (not to sonorants, which are characterized by periodicity). This property, in as far as itis deemed phonologic- ally significant, is usually only associated with fricatives and affricates. Aperi- odic energy, in the form of a noise burst, also characterizes the release phase of genuine plosives (as opposed to unreleased stops). However, this effect has generally not been considered distinctive for this class of segments within ortho- dox feature frameworks. Nevertheless, as we will now sce, the lenition evidence suggests that a noise component is indeed part of the phonological identity of plosives. . The element h in isolation is produced as a glottal fricative. As with 2, the absence of any supralaryngeal gesture here simply reflects the fact that the element lacks its own resonance property. In a compound expression, a place- defining element will indicate the location of the noise-producing gesture. Thus, fricatives with supralaryngeal resonance are formed by combining h with, for example, R (5), with U (f), or with @ (3). Pursuing the notion of lenition as progressive decomplexification, let us con- sider the various stages on the opening trajectory in (37a). If his the least complex segment, the plosive input must be the most complex and oral fricatives of intermediate complexity. It is reasonable to assume that an oral fricative differs from h by one degree of complexity: the former contains a place-defining element hat is absent from the latter. By the same token, the internal structure of a plosive includes whatever material is present in a homorganic fricative but is more ‘complex than the latter by virtue of the presence of an additional element, the stop property represented by ?.*7 This line of reasoning leads us to conclude that hn inheres in all released obstruents, both plosives and fricatives. We are now in a position to show how consonantal lenition can be directly tepresented as the suppression of melodic material. Continuing to abstract away from the laryngeal dimension, we may represent movement along opening traject- ory (37a) as a progressive loss of elementary content, illustrated here for coronal place:124 Melody (42) (@)—) (d) ) womeriK worm Plosive tin (42a) is defined as a coronal (contributed by R) stop (2) accompanied by noise release (h). Loss of closure yields a coronal fricative (42b). Further loss Of the supralaryngeal component contributed by R leaves an expression contain- ing h, which on its own defines a glottal fricative (42c). Loss of the remaining element results in elision of the melodic expression, which may or may not be accompanied by loss of the position to which it was attached (42d). Similar decompositions can be assumed for other place types. ‘The three-way place contrast among plosives found in English and most other Ianguages can now be represented as follows: (43) Plosives (a) Labial (b) Coronal (c),-Velar ' 1 i bh h h t 1 1 U R @ t H 1 ? ? 2 ‘We might ask what type of sound itis that lacks h but inchades ? fused with a place-defining element. Such a compound naturally defines 2 non-continuant without noise release. Three categories of segment fit this bill: unreleased oral stops (‘applosives’), laterals, and nasal stops. To complete the representation of the last of these, we need an additional element which will be introduced presently. ‘As shown in (39), an unreleased oral stop is one of the reflexes of f that shows up in dialects which otherwise have tap r. We now see how the process that gives rise to this reflex, the loss of audible release, is straightforwardly represented as the suppression of h. The treatment of this process as melodic recuction is supported by the fact that it recurs in what are generally regarded as lenition-favouring environments. In many languages, all place-types are affected by loss of noise release in such contexts (for example, in word-final position in Korean and Thai)- ‘The three main reflexes of t that show up in tapping dialects of English are represented in (44a, b, c). Loss of h from plosive ¢ (44a) results in an unreleased stop (44b)." Suppression of both h and ? leaves a lone R element, which inde- pendently manifests the tap reflex (4c). ‘Melody 125 4) @) h) (a) t t © ? It bh 1 R R R 1 1 Suppression of all elements save ? yields a stop devoid of supralaryngeal content and noise release; in other words, it results in ?, the independent realization of ? in (44d). There is every reason to suppose that ? also inheres in another subclass of noiseless non-continuants, namely laterals. One reason for reaching this conclu- sion is that laterals can be observed to pattern with coronal plosives in certain processes. Evidence of a phonological connection happens not to be particularly easy to come by in English. One example is the dialect-specific process mentioned in 2.4.4 whereby historically short 2 (in (man, pass), etc.) is tensed before certain consonants. In a subset of the dialects in question, can be seen to line up with ‘oral coronal stops in inhibiting tensing; 2 is tense before mand s (as in (man, ppass)) but remains lax not only before ¢ and d ({bat, bad)) but also before 7 (pal) In many languages (well represented among the Bantu family of African languages, for example), /altemnates with dor tin certain contexts; in some cases, the two types of sound are in complementary distribution. In Sesotho, for in stance, the pattern shows up in alternations such as the following: bal-a ‘read, count’, bad-ile ‘have read, have counted’; lat-a ‘fetch’, n-tat-a “fetch me It has been proposed that, within an element-based approach, the phonological relation between coronal stops and laterals can be captured by means of isomeric representations.” In particular, both types of coronal non-continuant contain R and 2; the former can be taken as the head of expressions defining stops, the latter as the head of those defining laterals. This asymmetry is consistent with the observation that the extent of coronal stricture is different in the two types of consonant. The preponderance of R in td is reflected in the fact that the coronal closure is tight, including both medial and lateral occlusion. In J, on the other hand, coronality is limited to medial closure. Representation of the third class of noiseless non-continuants, nasal stops, requires recognition of a nasality element, which may be labelled N.” This ‘manifests itself inthe signal as low-frequency broad-band ‘murmur’, an effect that is achieved through lowering of the velum. The same element can be assumed to inhere in nasalized vowels. The autonomy of N is reflected in its participation in assimilatory processes, such as long-distance nasal harmony and the more lo- ‘alized nasalization of vowels in the vicinity of nasal stops.”$ The isomeric relation between expressions corresponding to laterals and oral stops raises the more general issue of the headedness of consonantal repres- entations. Take first the relation between a resonance element and ? occurring126 Melody within the same expression. We have already seen how an Rcheaded expression ‘can be taken to define a coronal stop, while ?-headedness characterizes 1. In the case of labial non-continuants, the phonetic manifestation of a segment can be expected to vary in the following way, according to whether or not U enjoys head status. As a head, U is free to azsert the full elemental pattern it displays in isolation, typically achieved by labial-velar articulation. In conjuction with 2, this configuration thus provides us with a representation of doubly articulated labial. velar stops (kp, gb) such as occur in many languages, perhaps the best known of which are spoken in West Africa. When U occurs as a dependeat, the full effect of its elemental pattern is attenuated; in this case, fusion with a head ? defines a bilabial stop.”* Velar stops, represented in (43c), are necessarily @-headed. This follows from the assumption that this element is unable to contribute to the phonetic interpretation of an expression unless it occurs as a head. The head-dependent relation also provides a means of representing the dif. ference between strident and non-strident fricatives and a‘fricates. h-headed expressions can be expected to display greater stridency or ‘noisiness’ than those in which h occurs as a dependent. The contrast between strident s and non- strident @ can thus be expressed as in (45a) and (45b). (45) (a) (b) ©) () fe) (f) a) fom Ix to toot io hob hob hook Tot Toot toot RoR uu R @ 1 1 ‘The distinction between strident £(45e) and non-strident a (45d) involves the same kind of relation. The voiceless labial-velar fricative occurs in some dialects of English (chiefly spoken in Scotland, Ireland and parts of the United States), where it contrasts with w in pairs such as (which-witch, why-Y, what- watt). As with labial-velar stops, the double articulation of «reflects its U- hheadedness. ‘Other fricatives relevant to English are shown in (45c) and (4Sf). As indicated by the assimilation process illustrated in (35), a palato-alveolar fricative of affricate should be considered a palatalized version of a plain alveolar. In element terms, this means that f includes the h and R elements contained in s, sup- plemented by the palatal clement I. Palatalization, evident in a phrase such as (miss you), can thus be expressed as the spreading of I from the position occupied by the initial glide of (you) into the position occupied by the fticative of (miss). ‘The representation in (454) defines the velar fricative that occurs as a distinctive lexical category in some Scottish and Irish dialects (where, for example, the x at the end of (lough) contrasts with the kof (lock). It also appears as the spirantized reflex of kin those dialects that display forms such as those in (40b). Melody 127 3.5 Melodic geometry ‘As mentioned at the outset of the chapter, one of the motivations for recognizing the dissolubility of segments into smaller primes is the observation that they pattern into natural classes on the basis of shared phonological behaviour and certain phonetic commonalities. A given natural class is captured by identifying the common phonetic property with a particular prime. The unified phonological behaviour of the class is then accounted for by assuming that itis the shared prime that is being targeted by processes. In order to tighten this notion further, we need to exclude the possibility of capturing unnatural (that is, non-occurring) classes. Achievement of this goal clearly depends to a great extent on successfully identifying the appropriate set of cements. But this is only part ofthe story. Even with an appropriate set of elements, there still exists a danger of capturing unnatural classes, if phonological processes are allowed simultaneously to manipulate random conjunctions of two or more elements. To prevent this, we might make appeal to the following principle.” (46) Each phonological process can access only one unit in a representation, However, (46) turns out t0 be too restrictive, at least when allied to the bottle-brush mode of autosegmental representation we have been working with up to now. The view that elements are directly linked to the skeletal tier implies that each element is individually accessible to phonological processing, inde- pendently of all other elements. This much is in accord with the empirical record. ‘The problem is that there exists a set of processes which do indeed display a tendency to access more than one element simultaneously. In other words, ele- ‘ments no less that segments pattern into natural classes. The number of element classes involved is relatively small, and, like segment classes, they are identifiable ‘on the basis of certain phonetic commonalities. ‘One of the most frequently recurring element classes is one that can informally be specified in terms of resonance characteristics or place of articulation.” We have in fact alluded to this grouping at several points over the last couple of chapters. An example of a phonological phenomenon that makes exclusive refer- ‘ence to the place dimension is the set of conditions governing the melodic content of the coda of an English super-heavy rhyme (see 2.4.4). One condition, recall, prevents a sonorant in this position from bearing a distinct specification for place of articulation. In element terms, this means that A, U, I, @ and R form a class that is lexically barred from this context. Related to this is the general homorga- icity of nasal-obstruent clusters, as in (palm|per, wifnlter, alq|chor), expressible as the spreading of a place-defining element from the obstruent on to the nasal. Ie may seem a rather obvious point to make, but adopting a fully autosegmental view of phonological representation should not cause us to lose sight of the original insight that melodic expressions can and often do behave as units. In other words, certain processes can access all elements in a melodic expression128 Melody simultaneously. These are processes which either spread or suppress a segment in its entirety.” Note that the integrity of a melodic expression cannot be defined by reference to the skeletal point to which itis attached. The fundamental motivation for the split between the skeletal and melodic dimensions militates against such a move. There is plenty of evidence to indicate that entite melodie expressions must be allowed to spread or delink quite independently of the skeletal tier. And there are many cases where the total suppression of melodic material evidently occurs without a concomitant loss of a positions witness the phenomenon of compensat- ory lengthening discussed in 2.2.2. The latter process also exemplifies the poten- tial for whole melodic expressions to spread into vacant slots. ‘Another subset of elements that can function as.a class are those that code laryngeal information, to be discussed in the next section, The relevant contrasts associated with this dimension, including tone in vowels and phonation-type in consonants, are frequently accessed as a group and quite independently of other dimensions. On the other hand, there exist many logically possible combinations of elements which simply never show such group behaviour. Take for example the non-occurring three-element subset composed of A, h and R. Together these define a class that includes non-high vowels, coronal, fricatives and plosives. No attested process ‘operates on all of these segment-types simultancously. Other examples of unattested three-element classes, selected more or less at random, include: 2, I, Nz h, N, U, @, A? The challenge then is somehow to be able to pinpoint the small number of recurring element classes without also capturing the large number of non-occurring classes that an abandonment of the principle in (46) would imply. In autosegmental theory, the response has been to modify the bottle-brush model of representations. In its place has evolved a hierarchical model in which the functional unity of pacticular groups of primes is expressed by having them gathered under class nodes, which mediate between primes and the skeleton. The result is a geometric model of the sort schematized in (47), where the non-terminal nodes indicated by upper-case characters represent class nodes, while the terminal nodes in lower- ‘case represent primes (features or elements, depending on the framework).”” a7) peo 7 B DAN As before, elements are arrayed on separate tiers, and association lines indicate co-indexing with the skeletal tier. Thus, unlike the trees of syntactic theory, the geometric model does not imply that terminal elements are ordered in sequence (except in a special case to be discussed presently). The hierarchical arrangement Melody 129 is, however, similar to a syntactic tree in defining relations of dominance between mother and daughter nodes; in (47), a class node dominates terminal elements, cither immediately (as in the case of B and C) or indirectly via an intervening class node (as in the case of A). Applied to the geometric model, the principle in (46) allows each phonological process to target either a single terminal node or a single class node. The functional unity of groups of elements is captured by assuming that a process targeting a class node automatically affects any nodes it dominates. For example, delinking a class node results in the delinking of its daughters, whether these be class nodes themselves or terminal elements. The class nodes for which there is the firmest empirical support involve the recurrent groupings mentioned above. As shown in the geometric fragmen (48), elements defining resonance or place-of-articulation properties are gathered under a PLACE node, while laryngeal clements are dominated by a LARYNGEAL node. These class nodes are grouped under a ROOT node, the matrix which defines the integrity of the melodic expression.*® (48) ' ROOT, / LARYNGEAL « aoe \, PLACE More controversial is the question of whether the phonological behaviour of primes defining degree and manner of stricture warrants representation under an independent class node. Even amongst phonologists who assume there is evidence ‘to support such a view, there is considerable disagreement about where such a node should be located in the geometric model and about whether this dimension is itself subdivided into further class nodes." Here we may follow one well estab- lished view on this matter, namely that the number of class nodes should not be multiplied beyond that which is minimally necessary to account for the uncon- troversial evidence. The implication of this position is that the ‘manner’ elements N, h and ? should be directly and independently attached to the ROOT node.®= cote, Bzgmetsc arrangement of the elements discussed up to now is thus as (49) f ROOT» ____ LARYNGEAL «130 Melody The functioning of the ROOT and PLACE nodes can be illustrated by compar. ing the representation of full and partial geminates. Complete identity between the melodic contents of adjacent positions is represented as the sharing of a single ROOT node. A full geminate consonant, such as the tt shown in (50), consists in the association of ROOT to two positions. Sharing ROOT implies sharing all the class and terminal nodes it dominates. (The LARYNGEAL node can be omitted for the time being, pending further discussion in the next section.) (50) (The directionality of the spreading indicated here is something to be taken up in the next chapter.) A partial geminate such as nt, on the other hand, implies the presence of independent ROOT nodes, one dominating ? and N (in the case of the nasal stop), the other ? and h (in 4. As shown in (51), the homorganicity of the cluster is represented as the sharing of the PLACE node together with the element it dominates, R. (51) —* ROOT ¢ In this case, the shared place element happens to be R; in other cases, it can be, say, U (as in mp) or @ (as in gf). The point is that the general homorganicity of nasal-plosive clusters is captured by referring to PLACE, rather than to individual elements separately. ‘The mechanism of PLACE sharing can be extended to the representation of homorganicity in certain types of contour segment, including affricates. By treating such sounds as occurrences of two melodic expressions attached to a single skeletal point, recall, we are able to capture the fact that they count as, single segments for the purposes of phonological weight but as two segments when viewed in terms of their melodic make-up. Implicit in the notion of con- tour structure is the idea that elements can co-exist in a position without necess- arily being fused. In the case of affricates, this suggests a configuration in which a single position is associated to two ROOT nodes, one dominating ? the Melody 131 other b. The homorganicity of the closure and release phases is represented as the sharing of PLACE. In the case of &/j (in (chew, jaw)), this implies the following representation, in which the fusion of R and I defines palato-alveolar place: (52) e Note that, unlike the geometric representations considered above, the one in (52) contains two nodes on a single tir, that of the ROOT node. This can be taken to indicate sequential ordering, which allows us to capture the two-phase nature of contour segments. The stop-closure phase of the affricate is characterized in the first ROOT node containing ?; the fricative-release phase corresponds to the second ROOT node dominating h. The separateness of the two ROOT nodes indicates a lack of fusion between ? and h. On the other hand, each is inde- pendently fused with the PLACE elements R and I. Besides the lexically distinctive affricates ¢ and j, some dialects of English also display alveolar ts and dz, which result from the affrication of t and d respectively. Hence forms such as (piftsly), (Itslen), ({dzlip).*¥ In some cases, this process is evidently a precursor of spirantization on the opening trajectory in (37a). A typical sequence of events in this case is thus something like 1 > 1s > s, (Parallel developments affecting other places of articulation have produced chains such as P> pf> fand k > kx> x in other languages and at other times in the history of English." The structure of affricates other than € differs from that in (52) only with respect to the elemental content of the PLACE node: for ¢s the televant element is R, for pFU, and for kx @. The process of affrication consists in the fission of a plosive’s single ROOT node into two. In the case of the alveolar series, the resulting contour segment is identical to the palato-alveolar in (52), save for the fact that it lacks the palatal clement I. Viewed in these terms, the process does not involve element loss, unlike other lenition processes. Nevertheless it does qualify as decomposition in the sense that, like other opening processes, it results in a reduction in the degree of fasion contained in a position. ‘The affricate-contour structure also figures in an assimilatory phenomenon which affects nasal-fricative clusters. Nasal stops by definition contain not only the nasal element but also, as indicated in (51), the stop clement 2. The full representation of, say, m thus includes N, ? and U. The behaviour of ? in nasal132. Melody consonants serves to illustrate a class of processes usually referred to as forttion oo strengthening. It is worth emphasizing that fortition is not simply the inverse of lenition. The latter is a much more generalized phenomenon than the former, in that ie affects a wider range of contexts and segment-types. In terms of its melodic effects, fortition involves the transformation of a continuant sound into stop, as in f > por w> p. In other words, it isa ‘closing’ process, taking the form ‘of movement in the opposite direction to that expressed in portions of the opening tzajectories in (37a) and (37c). Representationally, all such processes consist in the addition of ? toa melodic expression. Some historical cases of strengthening are apparently spontaneous, in the sense that the emergence of a stop component cannot be attributed to spreading from a neighbouring position. Compare, for example, Latin yin mayor with its geminate affricate development in Italian majore. Indeed many such spontaneous fortitions {g0 hand in hand with gemination, something that is in all likelihcod related to the propensity of such sounds to display what is known as inalterebility.” That is, they typically resist lenition processes which affect their non-geminate counter- parts, There is evidently some general pressure on geminate consonants to appear in the guise of stops, although precisely what its representational motivation is remains somewhat unclear. “The majority of strengthening processes, particularly those that are synchronically active, can be shown to have a local cause. The usual site is one where the target sound is adjacent to some kind of a non-continuant consonant. A classic example is found in many Bantu languages, in which a stemvinitial fricative, for example, hardens into a homorganic stop when a nasal prefix is attached. Com- ppare the Sesotho root -fa (‘give’) with the derived form m-po (‘gift’). The fact that ‘such processes have a local stop trigger indicates that strengthening ofthis type is simply a sub-case of assimilation. In element terms, the compositional nature of fortition is expressed as the spreading of ? from one position to another. In the Sesotho example, the elements U and h, which together define the root-initial are fused with the ? spreading from the prefix nasal stop; the resulting three-ele- ment compound defines a labial plosive. In English, fortition is evident in the widespread excrescence of an oral stop within nasal-fricative and lateral-fricative clusters. This process manifests itself in some dialects as the appearance of a t between mor J and s in forms such as prince, else). The imerval of voiceless occlusion is shorter than that which characterizes a lexically present t in forms such as (prints, belt)."¥ This may be taken as evidence that the excrescence does not involve epenthesis of a skeletal point. Rather it should be seen as the strengthening of the fricative to a homor- ganic affricate. Viewed in these terms, excrescent ¢ arises through ? spreading from the preceding non-continuant lateral or nasal into the position occupied by the fricative. There ? fuses with R but not with h. This implies a contour structure with one ROOT node containing the fusion of R and ? (defining the excrescent and another containing the fusion of R and h."? The process is illustrated in the following representation of ns > #fs: Melody 133 (53) (a) (b) nis i i A * ROOT, «, AN “| ae I\ xl \N\ PLACE «, h PLACE «. sh NZ NZ In the following pages, I will continue with the practice of using the truncated representational format illustrated in (8) and only spell out geometric details where the exposition so requires. 3.6 Laryngeal elements We may now consider the melodic content of the LARYNGEAL node in (48). A range of different vocal-fold settings is available for executing contrasts in phona- tion-type (traditionally referred to in such terms as voiced versus voiceless) and pitch. The representation of laryngeal contrasts raises a number of issues, some of which continue to be subject to debate. One point is generally agreed on: the labels voiced and voiceless are inadequate for exhaustively classifying the phonetic and phonological aspects of this dimension. Their usefulness is limited to the informal identification of the two-way laryngeal distinction amongst consonants that typifies many phonological systems, However, this description ‘obscures the fact that the phonetic manifestation of such a contrast is by no means constant across languages. ‘Two of the most widespread patterns of realization are exemplified by English and French. In the context of a word-initial CV sequence, the main exponent of laryngeal contrasts amongst plosives is voice onset time, the interval between the release of oral closure and the onset of vocal-fold vibration in the vowel.” In a truly voiced plosive, vibration commences prior to release and continues into the articulation of the following vowel. A voiceless aspirated plosive (sumetimes referred to as fortis) is produced if there exists a time-lag between closure release and the onset of voicing. Where closure release and the onset of voicing coincide, the result is a ‘neutral’ voiceless unaspirated plosive. In French and many other languages, the contrast in this position between, say, the sounds symbolized as b and p is realized by means of the first and last of these types. That is, b is fully voiced, and p is voiceless unaspirated. In most dialects of English, on the other hhand, the corresponding -distinction alphabetized in the same manner is mani- fested as voiceless unaspirated 6 versus voiceless aspirated p2" This mismatch underlines the potentially misleading nature of the gross terms voiced and voice- less, to say nothing of the potential transcriptional confusion inherent in the use134 Melody of symbols such as p and b. In the initial CV context, the allegedly ‘voiced’ baly series in English is in fact not phonetically voiced at all and is to all intents and purposes identical to the neutral pli/k series of French. In most systems of transcription, the selection of symbols such as p and bis thus guided by contrast- ive considerations rather than by any attempt to represent the physical properties of the sounds in question. Some languages, such as Thai, utilize all three of the voice-onset categories just ‘mentioned for consonantal contrasts. Other languages make use of other phona- tion types. Various proposals exist for providing a more accurate charac- terization of the laryngeal dimension than the simple voiced-voiceless bifurcation allows. According to one approach, which has been widely adopted in current phonological theory, the relevant categories are defined in terms of vocal-fold tension and the degree of glottal aperture.*® A detailed exposition of the phonetic correlates of this classification would take us well beyond the bref of this book. Besides, much of the relevant phonological evidence comes from languages other than English. For our present purposes, it will suffice to mention two categories within this system which play a role in the consonantal contrasts discussed in the previous paragraph. These are slack vocal cords and stiff vocal cords, each of which we may take to correspond to a particular element. ‘The slack element inheres in truly voiced obstruents, while the class of forts obstruents (including voiceless aspisated plosives) contains the stiff element, Neutral obstruents can be assumed to possess neither of these elements. The signal characteristics of slack and stiff are respectively lowered and raised fun- damental frequency. This observation has led to the suggestion that these same ‘two categories are also involved in the specification of pitch contrasts in vowels, especially as employed in tone systems. There is some phonological evidence to support such a connection. For example, certain processes indicate a correlation between full voicing in a consonant and low tone in a neighbouring vowel, as well as between aspiration and high tone. It is for this reason that the slack element is sometimes symbolized as L and the stiff element as H. According to this view, a mid-toned vowel in a three-tone system is equivalent to a neutral consonant in lacking an active laryngeal element.” Nevertheless, the unification of tonal and consonant-phonation categories remains controversial. Since lexical tone con- trasts play no role in English, this is not an issue we need dwell on here. ‘What is relevant, however, is to determine which segment-types in English bear which laryngeal elements. In traditional feature-based treatments of laryngeal contrasts in the language, it is usual to specify the distinction in terms of (voice). Aspiration is deemed to be non-contrastive and is assigned to [-voice} plosives by rule.” An element-based approach is quite different. The lexical representation of fortis obstruents contains the element H. Aspiration is the particular interpreta- tion this element receives when it is present in an expression defining a fortis plosive. On the other hand, obstruents in the neutral series, transcribed as &, dd J ¥50, 2 5 lack a laryngeal clement. The two types represent the maximal set of English contrasts found in certain onset contexts, including in the initial CV site. In other contexts, however, the contrast is suspended in favour of the neutral Melody 135 series. This is true, for example, of an onset preceded by a fricative. In the CC clusters of, say, (stay, mister, after), the plosive is unaspirated (hence neutral) in most dialects of English. In systems displaying tapping, the neutralized reflex of ‘td (as in (fiter-bidder)) also lacks a laryngeal element. Fully voiced obstruents of the type that occur in French contain the element L, contrasting in this case with neutral. The manner in which the laryngeal specifica tion of the plosive series in French differs from English is illustrated in (54). (54) Element English French Voiced L = bo ‘beautiful’ Neutral = (bay) po ‘skin’ Voiceless aspirated H pay) = ‘Thai differs from French and English in displaying all three of the categories in (54). Free combination of elements in fact predicts a fourth category, one in which Land H are fused. This can be taken to represent the so-called ‘voiced aspirat or breathy-voice class of consonants. Gujarati is an example of a language whik exploits all four possibilities. The following forms illustrate the Th: plosive systems:”* ind Gujarati (55) Element Thai Gujarati Voiced L baa ‘shoulder’ bar ‘twelve’ Neutral = paa ‘forest’ por ‘last year’ Voiceless aspirated H phaa ‘split? phodz ‘army* Breathy LH = bbhar ‘burden’ The default condition for sonorants (both vowels and consonantal resonants) is to be produced with vocal-cord vibration, a property that gives them their characteristic periodicity inthe signal. This attribute cannot on its own be taken as evidence that sonorants contain the element L. The reason for this is that voicing in sonorants has a very low phonological profile when compared to true voicing in obstruents. The latter property systematically participates in a range of phono- logical processes which rarely if ever affect sonorants. For example, the classic Process of final devoicing, such as occurs in any number of languages including German, Russian, Catalan and Old English, only ever affects obstruents. Voicing assimilation between segments (some cases of which we will consider presently) almost always affects obstruents to the exclusion of sonorants. And in some cases, sonorants are transparent to a laryngeal category spreading between flanking ‘obstruents.” This behaviour has led to the recognition of two types of voicing: the spontaneous type that characterizes sonorants and the non-spontaneous or active type associated with obstruents.'® It is the active type that is predominantly, and in many cases exclusively, involved in phonological processing. A consequence of this view is that the representation of phonation categories in terms of laryngeal136 Melody elements is either restricted to obstruents or only allowed for in sonorants under special circumstances. ‘This conclusion leads us to rethink certain processes which previously have been treated in terms of laryngeal assimilation involving sonorants. A well-known, example from English conecens the inter vocalic tapping of ¢ diceussed in the previous section. This has sometimes been analysed as a voicing process; specific. ally, the voicing of the surrounding vowels is supposedly assimilated by the target consonant." There is litle doube that the tap can be phonetically voiced, a fact ‘we might try to account for by saying that vowels contain a phonation category which they are free to pass on to an adjacent consonant. If correct, such an analysis would contradict the claim that sonorants do not bear an active laryngeal clement. However, an alternative analysis is available which is consistent with the latter view. According to the account of tapping outlined in the preceding section, the process primarily affects the manner rather than the laryngeal aspect of the consonant, taking the form of a suppression of the elements ? aad h. Viewed in this way, the phonetic voicing of the tap reflex is no more than a secondary effect, reflecting the fact that the residual element (R) is independendy realized as a spontaneously voiced sonorant. For a similar reason, it is necessary to rethink the traditional analysis of the durational difference displayed by English vowels before different classes of consonant." Very broadly speaking, a vowel is relatively longer before a neutral obstruent or a sonorant than itis before a fortis obstruent. In the past, this has been treated as a case of lengthening before a ‘voiced’ consonant. The allegedly unified nature of the ‘voiced’ class is difficult to sustain in the face of the evidence just reviewed: neither sonorants nor the relevant sub-class of obstruents can be said to possess a phonologically active laryngeal component. This means that the durttional phenomenon should rather be thought of as a type of shortening, conditioned by a following fortis obstruent. (The opportunity of examining just such a proposal is provided in exercise 1.1 of chapter 4.) In terms of the account being offered here, the class of conditioning sounds in this case is identified by the presence of H. ‘A number of assimilatory phenomena in English illustrate the autonomous nature of laryngeal elements. One, which we will examine in detail in the next chapter, involves a distributional restriction whereby obstruents in coda-onset sequences ‘must agree with respect to their laryngeal identity, as in the medial clusters of forms such as (mister, chapter, factor). Another concerns the dynamic voice’ alternation affecting the suffixes (-s) and (-ed), touched on in 1.4.2, These have three alternants each, conditioned by the nature of the final segment of the stem t0 which they are attached: 2, s, 0 and d, t, ad respectively. The vocalic reflexes, evident in forms such as (dishes) and (fitted), occur whenever the stem ends in a consonant which shares certain place and/or manner characteristics with the consonant of the suffix; specifically, az occurs after a sibilant and od after tor d, We will postpone discussion of this aspect of the alternation until the next chapter. As shown in (56), the appearance of the other alternants of (-s) and (-ed) is. dependent on the laryngeal property of the stem-final segment: Melody 137 (56) ¢s) Ged) (a cs “ cats packed tops stopped safes doffed (by -z -d dogs logged robs robbed pods dodged oz “d days stayed goes bowed sons sunned calls called There are good grounds for taking z and d to be the lexical forms of (-(e)s) and Ged), since these are the shapes they assume when, as in ($6c), they occur next to sonorants." If we are right in assuming that sonorants lack an active voicing ‘component which could be passed on to an adjacent segment, then it follows that the laryngeal value of the suffix consonants in this context must reflect their lexical identity. In terms of theit melodic make-up, z and d in English share with other neutral obstruents a lack ofa laryngeal element. The alternants in (56a) thus result from assimilation to a preceding fortis obstruent. The process consists in the rightward spreading of H from a stem-final position occupied by a fortis ‘obstruent to the position occupied by the suffix consonant. This is illustrated in the following relevant fragment of a representation containing t-2- > ts: (57) ) ‘The fortis-neutral contrast among obstruents appears to be by far the best established pattern in English. Nevertheless, there are some dialects which have a fully voiced series corresponding to the otherwise general neutral set (in (bin, din, un), etc.). As in French, these contrast not with voiceless aspirates but with neutral (in (pin, tin, kin), etc.). (One area where this pattern is attested is the northeast of England.) The indication that the first series is truly voiced, and thus contains the element L, is confirmed by the observation that a subset of the dialects in question displays a voicing assimilation process not found in other mow w—x138 Melody dialects. This is a word-edge anticipatory phenomenon, in which fully voiced obstruents pass on their laryngeal identity to a preceding obstruent:'™ (58) top gun tofb g]un football foofd bjall pitch black: pij bllack backbone baja blone In this instance, it is che element L that spreads leftwards, a process that is ‘widespread in other languages (including Polish and Russian). 3.7 Conclusion In this chapter we have reviewed some of the evidence supporting the existence of sub-segmental primes. The bulk of this evidence derives from the behaviour of segments under phonological processing. ‘The representationsl model that has emerged is one in which the set of processes to which a sound is susceptible is circumscribed, on the one hand, by the number and nature of the elements it contains and, on the other, by the requirement that all processing be expressible in terms of the fundamental operations of composition and decomposition. In order t0 constrain our model of phonological processing, still further, we could require of every process that it have a local cause. Leaving aside the special case of spontancously appearing elements (such as appear in certain examples of fortition and vowel epenthesis), we can see how compositional processes begin to satisfy this criterion. The mechanism of autosegmental spreading, by which an clement from one position becomes fused with an element in a neighbouring position, establishes a intimate bond between a process and its triggering context. ‘Two further questions connected to the local-cause requirement remain to be answered, however. First, what determines whether two positions are adjacent and thus form a domain within which spreading can take place? Second, it may not seem immediately obvious how decompositional processes can have a local cause. What is the nature of the relation between process and context in such cases, and what is it about this relation that triggers delinking? These are issues 10 be taken up in the next chapter. 3.8 Appendix: specification of elements 3.8.0 Elements are cognitive categories which serve the grammatical function of coding lexical contrasts." They are mapped in the first instance onto the acoustic signal- ‘This mapping relation guides the manner in which elements are input to production Melody 139 and perception. That is, elements constitute internalized templates by reference to which listeners decode auditory input and speakers orchestrate and monitor their articulations. ‘The purpose of this appendix is (a) to specify the internalized elemental patterns which, according to one set of proposals, represent the elements introduced in this chapter; and (b) to suggest how these parterns are mapped onto the acoustic signal. 3.8.1 Resonance elements ‘We may think of each pattern associated with a resonance element such as A, 1 and U as being displayed in a frame mimicking a spectral slice, in which the vertical axis corresponds to intensity and the horizontal axis to frequency. The fa) [A] IY NI (b) ao \/ [u} — Figure 1. Elemental patterns: (a) A (Ass), (b) (dp, (c) U(eUmp) Note: The contours are schematic spectal envelopes, plotted in frames which map onto acoustic spectral slices (vetcal axis: amplitude; horizontal axis: frequency). Source: Haris and Lindsey in press,140 Melody latter coincides with what may be termed the sonorant frequency zone, corres. ponding to the frequency band within which the perceptually significant resonant characteristics of vowels are concentrated. Roughly speaking, this region covers the range between 0 kHz and 3 kHz, in which the fist three formants are located, Within this zone, the convergence of two formants produces prominent peaks of energy (spectral peaks) corresponding to various vowel qualities. cis these peaks that match the peak-valley profiles of individual elemental patterns. The elemental patterns of A, I and U, shown in figure 1, are plotted in frames of the type just described. Figure 1a shows what can be tetmed a mAss pattern for A. This shows energy minima at the top and bottom of the frame; ie. there is a mass of relatively higher energy in the middle of the frame. The precise outline of the energy mass is nota crucial part of the definition of this pattern and is thus left blank in the diageam. Figure 1b depicts a dlp pattern for I, with energy lower in the middle of the frame than on either side. The rUmp pattern of U, shown in figure 1c, is characterized by energy of a higher amplitude in the lower part of the frame (its exact outlines not criterial) than in the upper part. The following summaries of each element provide (a) an informal label by ‘which it can be identified; and (b) a specification of its signal mapping. ‘A (a) mAss; (b) a spectral peak (representing the convergence of Formants 1 and 2) located in the middle of the sonorant frequency zone. 1 (a) dips (b) low first formant coupled with a spectral peak (representing the convergence of Formants 2 and 3) at the top of the sonorant frequency zone. U_ (a) rUmp; (b) a spectral peak (representing the convergence of Formants 1 and 2) at the bottom of the sonorant frequency zone. @ (a) neutral; (b) dispersed formant structure, i.e. no salient spectral peak. R (a) ‘coronal’; (b) the set of formant transitions associated with coronals. 3.8.2 Manner’ elements h(a) noise; (b) aperiodic energy. ? (a) stop; (b) abrupt and sustained decrease in overall amplivude N (a) nasal; (b) broad resonant peak at lower end of the frequency range. 3.8.3 Laryngeal elements L (a) ‘slack’; (b) low fundamental frequency. HL (a) ‘sti; (b) high fundamental frequency. 3.8.4 Element fusion The effects of element compounding are derived by overlaying one elemental pattern on another. Figure 2 illustrates two complex profiles which result from the pair-wise combination of patterns displayed in figure 1. The profile of ¢ in Melody 141 (a) e L“N_”\] (b) ly \N_ | Figure 2 Compounded elemental patterns: (a) A and I (e}; (b) A and U (0) Source: Harcs and Lindsey in press figure 2a might be viewed as ‘dlp within a mAss’. That i, itis an amalgam of two patterns: (a) energy higher in the middle of the frame than at top or bottom, indicating the presence of A; and (b) energy lower in the middle than on either side, indicating the presence of I. In the case of 0, depicted in figure 2b, we can again identify two patterns: (a) energy higher in the middle of the frame than at top or bottom, i.e. A, and (b) no energy above the middle, i.e. U. Both of these compound patterns map rather directly onto the spectral envelopes of e and o. Figure 3 shows patterns resulting from various combinations of 2, h and U, displayed in stylized spectrographic frames (in which the horizontal axis mimics time, the vertical axis frequency, and darkness intensity). Fusion of all three elements defines a labial plosive (figure 3a). Different lenition reflexes are defined by the suppression of particular elemental patterns: spirantization (figure 3b), debuccal- ization to h (figure 3c), debuccalization to ? (figure 34), and vocalization (figure 3e).. Exercises 1 Scots vowel systems The data The data below show the subsystems of vowels found in final open stressed syllables in five Scots dialects." Each row of words (listed in English as142 Melody Melody 143 () glowal ficative (a) labial plosive {hy 2] (h] ; iL I~ a (@) glowa stop 2] {u] (b) labial fricative th) iL NY [U} Figure 3. Stylized spectrograms showing various combinations of the elements ? (top) b uw {noise) and U (rUmp): (a) labial plosive, (b) labial fricative, (c) glottal fricative, (d) glottal stop, () labial approximant. Sources Harts and Lindsey in press Figure 3 contd,144 Melody opposed to Scots spelling) gives a very rough guide to the lexical incidence of each vowel. The task For each different vowel, provide a representation which includes details of its nuclear structure and melodic content. FETLAR, SHETLAND it die, eye, give, knee, tree clay, have, no, say, toe fall, gnaw, two, snow, who, wall & low, below, row cow, full, how, through yu: dew, new, spew % do, she, shoe ey pay, stay, weigh, aye ey fry, kye (‘cattle’), ay vy boy, joy, toy ‘2w ewe, grow, hoe, roll, knoll, through KEIss, CAITHNESS e or die, eye, give, key, tree clay, day, do, no, toe, too away, fall, gnaw, two, wall, who below, low, row cow, fll, how, now, pull, shoe new, spew, stew aye, hay, pay, stay, weigh, why ay, die, fry, way, whey joy, toy dew, ewe, grow, knoll, roll, through NEW LUCE, WIGTOWNSHIRE die, eye, tree, give = do, flea, have, no, shoe e day, say % away, fall, gnaw, law, two, who % — below, low u cow, full, how, now, pull, through yu dew, new, spew Melody 145 ey aye, clay, hay, pay, stay, way, whey vy fry, kye, why ay _jey, toy, buoy aw ewe, grow, hoe, knoll, rol BALLYWALTER, CO. DOWN (IRELAND) i die, eye, give, knee, tree 10, Tay clay, day, do, have, tea, toe, show, say = gnaw, fal, law, snow, wall x two, who, away «below, low through, t00, 200 yi dew, new, spew, stew ey aye, clay, hay, pay, weigh, why ay fry, kye vy ay, boy, joy, toy $w cow, ewe, full, gcow, knoll, roll GRETNA GREEN, DUMFRIESSHIRE ey die, eye, key, knee, tree, aye, clay, hay, whey x flea, give, no, pay, she, toe, two, who day, say, stay, way = fall, gnaw, law, snow & below, low 2x cow, full, how, now, through, you iw blue, dew, new, spew % do, shoe, too ey ay, weigh ay fry, kye, why ey buoy, joy, toy wy boy 2w ewe, grow, hoe, knoll, oll 2 Yorkshire assimilation The data The set of laryngeal contrasts in Yorkshire English is the same as that found in most other dialects. However, for many Yorkshire speakers, the forms (bag pack) and (back pack) are homophones."®” The process responsible for this ‘merger is more fully illustrated in the data below.146 Melody The task What is the representation of the process? What element(s) does it target? What does it reveal about the representation of laryngeal contrasts in different classes of sounds? 1 Rahenes Bolh gloes Bob came Bofp klame 2 rubout ru(b] out rub shoulders rulp floulders 3 told Viv tolld vliv told Fred tollt fred 4 dread it dreafd) it dreadful dreafef}ul 5 bad boy bald bloy bad time bale tlime 6 judge ruled jut] ruled judge took jul tJook 7 bridge the brifible bridge party brilé plarty 8 big nose bilg) nose big thumb bifk Jum 9 ragand rafg] and rag trade alk t}rade 10love Davy lofv] Davy love Pat loff plat 11 smooth over smoofd] over smooth top smool® tlop 12 rose garden rol} garden rose petal rols pletal 13 ribbed back rifbd black ribbed pattern rifpt] pattern 14 forged more forljd] more forged cheque forlét Eleque 15 rigged local rilgd] focal rigged sample rifkt sJample 16 loved Zoe lolvd zJoe loved Fiona loffe fiona 17 squeezed Joe squeef2d jJoe squeezed past squeefst plast 3. More Merseyside consonants The data The following data expand on the description of Merseyside English given in 3.4.2. (Symbols: = voiceless bilabial fricative; = neutral bilabial fricative; y= neutral velar fricative.) Melody 147 The task Suggest representations which account for the consonantal realizations shown in the bracketed portions of the words. 1 people people pepper Peller 2 later lalsler matter malsler 3. joker jolxler packet pafx}et 4 Robbie RofBlie rabbit ralBlit ready reafzly Maddy malzi 6 baggy ——_balyly boggle bofyile 4 Nasals in African American English The data In some varieties of African American English, nasal sounds parti in the processes illustrated in the data below." pate The task Provide representations which (i) distinguish the vowels in columns (a, b) from those in column (c); and (ii) capture the alternations evident in columns (a) and (b).. {a) Before pause (b) Before vowel (c) Non-nasal 1 he (him) (h{ism] and) brad (bid) 2 we (win) (w{t5n] a) 3 ks (king) «k(q] of) 4 st (Sam) (s[@5m] and) swad (sad) 5 pes (pan) (lé3n] of 6 nes (hang) (h{é5q] a) 7 tas (time) (t{43m] of) aad (hide) 8 sas (sign) (s{a3n] of)4 Licensing 4.1 Introduction This chapter develops some of the notions of phonological constituent structure introduced in chapter 2. There we reviewed arguments in favour of the view that syllabic constituents (onsets, rhymes and nuclei) are maximally binary branching. Here we will examine a range of facts which support the conclusion that the relation between positions within a constituent is asymmetric. A uniform pattern of left-headedness can be shown to manifest itself in all constituents; that is, the left-hand position is dominant in relation to its recessive sister on the right. We will go on to consider the question of headedness as it relates to other domains, including those formed by coda-onset clusters and those iavolving relations between nuclei, for example at the level of the word. ‘A major concern of the chapter is to establish that the properties of binarity and 3 headedness in phonological constituent structure are derivable from more funda- ‘mental principles of grammar. One such principle, that of licensing, integrates the various parts of a representation by requiring that each unit be bound in some way to some other unit. In order to be phonetically interpretable, a melodic expression, for example, must be associated to a skeletal position; a position itself must belong to a syllabic constituent; and each constituent must also be incorpor- ated into some larger domain, such as the word. Licensing defines the circumstances under which a pair of positions can be considered adjacent. In so doing, it establishes the necessary conditions of locality that must prevail in order for phonological processes involving an interaction between the two positions to occur. In this respect, licensing provides an answer | 10 the question of where phonological processes take place. ; will also attempt to demonstrate how the theory of phonological licensing can begin to answer the more fundamental question of why phonological processes occur where they do. This involves making appeal to a principle which imposes severe constraints on the relation between the melodic content of a skeletal position and the niche occupied by the position in the constituent hierarchy. By ‘way of illustrating this relation, I will devote the second part of the chapter to @ detailed analysis of the contextual conditions that regulate the various processes of ¢lenition in English introduced in 3.4.2. Licensing 49 4.2. The phonological hierarchy 4.2.1 Headedness Terms such as onset and nucleus refer to categories of syllabic structure. Head, on the other hand, is not a categorial term but rather refers to a phonological function or relation, specifically one that is contracted between positions. The headedness of a branching constituent can be determined in the first instance on the basis of the phonotactic possibilities associated with each posi- tion." In a metrically prominent context (for example, in the dominant syllable of a word), a non-branching constituent typically displays a maximal or near-max- imal set of contrasts. In a branching constituent, on the other hand, the distribu- tional possibilities diminish, but in an unequal fashion. A¢ least in the case of ‘onsets and nuclei, itis clear that the position on the left enjoys a greater degree of distributional freedom than its sister. As we saw in 2.4.2, in a branching onset, ‘occupation of the right-hand position is restricted to a liquid or a glide, while the position on the left can be filled by any of a range of plosives or voiceless fricatives. In a branching nucleus, the right-hand slot can only be distinctively specified as a glide, whereas the melodic specification of the position on the left potentially covers a full range of vocalic contrasts (sce 2.4.3). Let us assume that this asymmetric division of phonotactic potential is symptomatic of a headed relation and further that the greater degree of distributional freedom is invested in the dominant member of a constituent. On this basis, we may conclude that onsets and nuclei are left-headed. This distributional argument does not extend directly to branching rhymes. The incidence of melodic contrasts in a coda is indeed constrained, but not by the auclear position on its left; instead, itis dependent on the onset appearing to its right. Under the assumption just outlined, this fact establishes the coda as a recessive position, but it fails to signal any kind of relation between the positions of a branching rhyme. Nevertheless, phenomena such as closed-rhyme shortness and rhymal weight indicate that some kind of link exists between these positions, although it does not immediately suggest a pattern of headedness. However, there is at least one observation we can make which does point to an asymmetry in this ‘context: a nuclear position is obligatorily present in a rhyme, whereas the coda is aly optionally so. This is reflected in an implicational universal according to which no language possesses branching rhymes to the exclusion of non-branching, thymes. On the assumption that obligatoriness indicates dominance and that optionality indicates recessiveness, we may conclude that chymes are left-headed. 42.2 Constituent parameters Research into a wide range of languages of quite diverse genetic affiliation indicates that the properties of left-headedness and maximal binarity are univers-150. Licensing ssible constitu. ally characteristic of phonological constituents. The full set of pos > ent structures is thus as set out in (1). From now on, it will be useful to adopt the convention of representing constituent heads by means of a vertical line. a Non-branching Branching Onsets ° ° 1 Nuclei N N 1 i Rhymes R R 1 Q N N n Nv x(%) x(%) x Languages vary with respect to whether they possess one or more type of branching constituent. English happens to be a language in which all three constituents may branch. Onsets in some languages, however, can only be non- branching (Arabic, for example). Rhymes in some languages, a subset of so-called ‘open-syllable’ (i.e. coda-less) systems, can only be non-branching (Zulu, for instance). Nuclei in some languages can only be non-branching (Yoruba, for example), with the result that they lack vowel-length distinctions. We can charac- terize these cross-linguistic differences in terms of three parameters which deter- sine the structure of each constituent: (2) Constituent structure parameters Branching constituent OFF] ON (a) Onset Arabic English (b) Nucleus Yoruba English () Rhyme Zulu English ‘Whether these parametric settings can be assumed to hold throughout derivation is'a matter we take up later in 4.6.4. | implicit in (2) are two implicational niversls. According to one, the tein zation of branching structure in a particular constituent (corresponding to an O? setting on one of the parameters) implies that the language in question also * possesses the non-branching variant; the reverse implication does not hold. This establishes the OFF setting as the unmarked value (indicated in (2) by the square bracketing), an observation that is confirmed by a range of other facts. For example, exclusively non-branching structures enjoy a much wider distribution throughout the word's languages and are the first to be acquired by native speakers of languages displaying one or more of the ON options.? Second, Licensing 151 some reason that is at present not clearly understood, a positive setting on (2a) depends on (2c) also being positive. That is, no language possesses branching onsets without also possessing branching rhymes. 42.3 The prosodic hierarchy For some time, it has been recognized that the asymmetric nature of intra-constitu- ent relations shows certain parallels with that of relations holding between units in suprasegmental stcucture. These similarities are particularly cleat in the case of the metrical relations involved in stress assignment. Indeed itis now usual to view the skeletal and constituent dimensions as being integrated into a hierarchy of prosodic levels which includes the foot, the phonological word, the phonological phrase, and pethaps even larger domains. Together the various domains make up the prosodic hierarchy: (3) Prosodic hierarchy Phonological phrase Phonological word Foot Syllabic constituent Skeletal position A foot, recall, is a rhythmic unit consisting of one or more syllables — more technically, rhymes. In the event of a foot containing two or more rhymes, one is dominant. English is one of a large class of languages inhabited by that most exotic of beasts, the left-headed foot. This is revealed in the location of primary and secondary word-stress. A word such as (Cinderélla) contains two feet: (Cinde-) and (-rélla). The word (city) contains one. There have been various proposals for capturing the parallelism of relations at different levels of the prosodic hierarchy. One approach has been to represent all relational aspects of phonological structure in terms of metrical tree structures in which each branch is labelled as dominant/strong or recessive/weak.! A related mode of representation is the metrical grid, in one version of which units at different levels of the prosodic hierarchy are gathered into bracketed constituents, with one unit in each constituent being designated as the constituent head.¢ According to related proposals, asymmetries within the prosodic hierarchy are ‘characterized in terms of relations of dependency or government (on which more resently).” The representation of constituent structure in terms of trees or bracketing has a clear precedent in syntax. We can pursue this parallel further by examining some of the structural and relational properties of the prosodic hierarchy. Focusing, initially on the lower reaches of the hierarchy, let us consider a number of generalizations that can be made about constituent heads and the positions that «an follow them.ing 153 152 Licensing Licensing 15: head of X’ and the ultimate head of the overall rhymal constituent X”, Under X’, Y is the complement of X. Thus the xs positions in (4b) and (4c) are respectively nuclear and onset complements (in much the same way that a noun phrase is the complement of a head verb within an X’ verb phrase). Under X” in (5), Y (the ‘cada’) is an adjunct of X. (The nearest ayntactic analogue is a configuration in which an adverbial or prepositional phrase is said to be the adjunct of a verb within an X” verb phrase.) The term chymal adjunct is thus technically more accurate than coda (which, as we saw in 2.4.4, has no categorial status), There is evidence, to be reviewed below, that the prosodic hierarchy extends ‘upwards’ beyond the level of the rhyme.” In English, the head of a bi-rhymal foot, for example, is the rhyme on the left (as in (city)). Feet in their turn enter into headed relations at the level of the phonological word. Of the two feet that make up the example (Cinderélla), mentioned above, it is the one on the right ((-rélla)) that is dominant, as revealed in the fact that its prominent rhyme bears the primary word stress. These prominence relations at different levels of the prosodic hierarchy are illustraced in (6), where we follow the precedent of representing headship by means of a vertical line. 4.2.4 Projection id to reflect certain ‘There is a sense in which a constituent category can be sai properties ofits head position. For the time being, we can think ofthis relation in terms of the notion of sonority. (As mentioned in 2.4.2, this is merely an informal label for a more fundamental property that is directly coded in melodic structure, 4 point to be expanded in 4.4.5.) Typically, the nucleus is the locus of relatively more sonorous or *V-like’ melodic material, while the orset is the locus of relatively less sonorous, more ‘C-like? material. However, within each of these constituents, the burden of maintaining this distinction is nct equally shared. In both instances, it is the head position which optimizes the sonority profile of is constituent. A’ recessive position, in contrast, contains material which com- promises the profile. For example, the right-hand slot of a branching onset can contain a V-like glide. | The novion we would like tobe able to capture here is tht a head postion somehow a better representative of its constituent than is its sister. Suppose we provisionally assign the category V to the head position of a nucleus. Then the category ofthe nuclear node itself should match V. ln phratestrucre erin (6) Word i ology, the nucleus can be referred to as V’ (V-bar).* To make the analogy wit Foot . ee feven more explicit: V (the nuleus) sa projection ofV (its head potion), . i— in much the same way hat a verb phrase is a projestion of ts head ver. By the Rhyme R R R same token, ifan onset head position i , then the onser category is Cs Nucleus ENCORE ROE Generalizing across both the onset and nucleus domaits, we may 1 yr 7 2 F OF category variable X to stand for any constituent head position and Y for an x oes eee optionally present sister position. X and Y are terminal nodes of the prosodic 1 r 1 fF Ff FF ® hierarchy. The onset and nuclear nodes are then immediate projections of X: a ef ee wm x ) ON © 0 Here we see that the foot and the phonological word contain projections of i 1 in nuclear positions. Thus nuclear heads have the potential to be projected through xy x1 xm Successive levels of the prosodic hierarchy. The ultimate head of the whole word in (6) is the nuclear position containing the vowel e. Beyond the word, nuclear heads project first to the phonological phrase and ultimately to the intonational phrase. Prominence relations at these levels mani- fest themselves in phrasal stress and sentence accent. Very broadly speaking, within a two-word phrase in English, the ultimate head is the dominant nucleus of the word on the right, as in (prétty bonnet). As for the rhyme, itis properly considered a projection of the nucleus. That is the term shyme doesnot refer to an independent constituent category tis rather an informal label for the second projection (X” or X-double-ber) of a nuclear he position: “xe R The representation in (6) illustrates one important respect in which the phono- 1 1 logical implementation of X-bar principles differs from that in syntax. The x\ N Prosodic hierarchy lacks the recursive property of syntactic phrase structure Vy 1 whereby a constituent of one category can be embedded within a constituent of a x x % different category. A noun phrase, for example, can be embedded within a verb Phrase; but a nucleus, for example, cannot be embedded within an onset. The absence of recursion lies behind the “spinal” configuration of head projections in the prosodic hierarchy. That is, the ultimate head of a phonological repres- «referring to particular Some additional syntactic terminology will be useful for referring to particu relations that hold between different nodes in the hierarchy. X is the immediate154 Licensing entation is defined by a straight line which runs through all prosodic levels and which contains nothing but projections of a single nuclear position. 4,3, Phonological licensing, 43.1 Prosodic and autosegmental licensing In this section, we address the question of whether the left-headed and maximally binary-branching properties of phonological constituents are derivable from grammatical principles which govern constituent structure ia general. We begin by considering one such principle, that of licensing, as originally developed in syntactic theory. ‘One requirement of sentence well-formedness is that each unit in a repres- entation must be integrated into the phrase-structure hierarchy in order for it to be semantically interpretable. For this integration to be achieved, the presence of each node in sentence structure has to be sanctioned or licensed by the presence of some other node. Licensing thus defines a binary asymmetric relation between units in the grammatical hierarchy. For example, within a verb phrase, a noun phrase complement is licensed by the verb that is its head. The only unit in given structure that is exempt from the licensing requirement is the ultimate head of the ‘whole sentence (in technical terms, the matrix node of a root sentence). I is now usual to assume that licensing extends ‘downwards’ into the phono- logy, where it controls the phonetic interpretability of units in a representation, The phonological application of the principle can be viewed as an attempt to formalize the traditional idea that all sounds in a string have to be gathered into syllable structure. That is, itis not possible to have ‘stray’ sounds which are syllabically unaffiliated.’ Expanding this notion to the entire prosodic hierarchy results in the requirement that each unit within a representation must belong to some higher-order unit ~ a skeletal position to a syllabic constituent, a constituent to a foot, a foot toa word, and so on up the hierarchy. To put it somewhat differently: the presence of each unit within a phonological representation must be sanctionec by some other unit. Licensing is the mechanism by which this authorization is granted. ‘Within the phrase-structure and prosodic hierarchies, it is assumed that a position is licensed by its head. Thus within a branching onset or nucleus, for example, the head position licenses its complement. At the foot and word levels, ‘one nuclear position is licensed by another. We can go further and view licensing as also regulating the relation between melodic and prosodic structure. Specific- ally, a melodic expression is licensed through association with a skeletal point. Licensing, then, can be thought of as the mortar that binds together all compon- ‘ents of the phonological hierarchy, both prosodic and melodic. Purely as a matter of convenience, we can label licensing relations differently according to the levels of the phonological hierarchy at which they hold. Under prosodic licensing, each Licensing 155 unit in the prosodic hierarchy is required to belong to some higher-order struc- ture.!" Autosegmental licensing regulates the attachment of melodic material to skeletal slots.!? These terms simply refer to different facets of the same fundamen- tal mechanism, rather than to two independent mechanisms. ‘Any unit in the phonological hierarchy which faile to be sanctioned by phono- logical licensing is phonetically uninterpretable. This situation arises, for example, in cases where a melodic expression becomes detached from its skeletal slot. Our discussion of compensatory lengthening in 2.2.2 implicitly invoked the ‘autosegmental dimension of licensing; recall that, once the velar fricative of a form such as nixt (night) becomes dissociated from its position, it is no longer phonetically expressed. Some versions of prosodic theory allow for circumstances under which certain constituents or skeletal points fail to get integrated into the phonological hierarchy; under the prosodic aspect of licensing, such stray mater- ial is then assumed to be erased at some point in derivation." Each level of the phonological hierarchy defines a licensing domain. At the autosegmental level, this takes the form of a skeletal point together with its associated melodic material. A prosodic licensing domain corresponds to a par- ticular level of projection in the prosodic hierarchy. Reflecting the asymmetry of licensing relations, the presence of any non-head within a given prosodic domain is sanctioned by the presence of the head of that domain. Within the domain of a branching nucleus, for example, the directionality of the licensing asymmetry is necessarily head-initial, reflecting the fact that constituents are uniformly left- headed." In this configuration, the left-hand position licenses its sister by virtue ‘of being the head of that domain and thus being projected up to the next (nuclear) level of structure. With this notion of domain, we may define phonological licensing as follows: within a domain, all phonological units must be licensed save one, the head of that domain." The unlicensed head of a domain is itself licensed at some higher level of projection. To illustrate this last point, let us train our sights on a fragment of the network of licensing relations holding in the representation of (Cinderella) in (6). A subset, of the relations contracted specifically by the first nucleus are shown in (7), where the arrows indicate the directionality of licensing. In order to focus on the licensing aspect of the representation, I characterize syllabic constituency here by means of labelled bracketing rather than arbureally. — Word rd Rhyme bo UalN a re Ga ad fIn bs10 bel s roa do o9 ff e©€ JT 29 ”156 Licensing The nuclear position x2 in (7) autosegmentally licenses the melodic expression defining 1. By virtue of being the head of its shyme, the same position prosodically licenses the coda position x; on the chyme projection.'* And by virtue of being the head of its foot, it licenses the nuclear position xs on the ‘oot projection. On the word projection, xs is itself licensed by x, the nuclear head of the second foor. There thus exists a chain of licensing extending from the lowest to the highest level of phonological structure. From this it follows that only one unit within the phonological hierarchy may remain unlicensed. This is the ultimate head of a representation ~ the head of the highest-level domain in phonological structure, which in this respect is thus equivalent to the root node of a matrix sentence in syntax.” Consideration of the representation in (7) reveals that directionality is one of the fundamental notions in terms of which licensing is defired. It also demon- strates that the direction of licensing varies according to the level of the prosodic hierarchy at which it operates. The other fundamental property of licensing is that itis subject to the condition of locality.¥* That is, a licensor must be adjacent to its licensee. This observation is obviously true of a branching constituent, in which the relation between the two positions is local on the skeletal tier. But the extension of this principle to all levels of the prosodic hierarchy is perhaps not so immediately obvious, in view of the fact that one or more skeletal positions may intervene between a licensed position and its licensor. Turning again to (7), we may note for example that, although the nuclear positions x2 and xs enter into a licensing relation at the foot level, they are separated by two other positions (one a coda, the other an onset). The key to understanding the uniformly local nature of licensing lies in the insight that adjacency is defined in terms of the particular level at which two positions are projected. Licensing between sister positions within a constituent involves locality at the level of the skeletal tier, a relation sometimes known as strict locality or string-adjacency."” On the other hand the licensing relation between nuclear projections within a foot involves locality at the foot level; this remains true even though the skeletal positions dominated by these nodes are not necessarily string-adjacent. Summarising the discussion to this point, we may define the Phonological Licensing Principle as follows: (8) Phonological Licensing (a) Within a domain, all phonological units must be licensed save one, the head of that domain. {b) Licensing relations are local and directional, 43.2. Constituent and projection licensing ‘As noted briefly above, the directionality of licensing appears to vary from one level of the prosodic hierarchy to another.” In view of the universal left-headedness of Licensing 157 onsets, nuclei and rhymes, we may conclude that licensing at this level proceeds from left to right: (9) Constituent licensing Within constituents, licensing relations are head initial. Licensing between the projections of nuclear heads at higher levels of the prosodic hierarchy ~ projection licensing for short ~ is evidently not fixed in this way. Here directionality, revealed in a range of phenomena including stress, harmony, vowel reduction and syncope (on which more in 4.6.2), vaties from one language to another. As illustrated in (6) and (7), English is one of those languages in which relations on the foot projection are left-headed. That is the strong-weak stress pattern that characterizes this domain involves a configuration in which a dominant nucleus licenses a recessive nucleus to its right. In other languages, the opposite directionality obtains. ‘The same cross-linguistic variability is in evidence at the word level. Word- stress, for example, displays left-dominance in some languages (such as Hunga- rian) but right-dominance in others (French, for example). Indeed it is quite usual for the same language to display opposite directional patterns at different metrical levels. Thus in English, while feet are left-headed, the main pattern at the word level (to generalize grossly) is one of right-headedness, as illustcated in (7). That is to say, where a word contains two or more feet, itis the rightmost one (more particularly its dominant thyme) which is typically prominent. (Of course, this pattern cannot manifest itself in a word containing only one foot. The left-domi nant pattern of, say, (city) is a foot-level matter.) ‘These choices in suprasegmental directionality, itis now generally assumed, are determined by parameter: (10) Projection licensing Between the projections of nuclear heads, licensing relations are parametrically head-initial or head-final. 4.3.3. Inter-constituent licensing We have yet to consider whether licensing relations are also observed to hold between string-adjacent positions occurring in different constituents. The sort of sequence at stake here is illustrated by the mt of (winter). We should start by considering which of the possible two-position inter-constituent sequences invol- ving each syllabic category are actually attested. In view of the left-headedness of constituents, the second position in every case will be a (nuclear or onset) head. In surveying the relevant facts, we should also bear in mind that string-adjacency implies occurrence within the same morpho-syntactic domain. The fact that some of the possibilities turn out to be systematically excluded calls for some kind of principled explanation.158 Licensing (11) lists the set of logically possible sequences in which the first position occurs within an onset. The rightmost column either records an example of an attested type (with the relevant sequence emboldened) or asterisks an ungrammatical type, ay First position Second position fa) Onset Nucleus head head ip) complement head ip) (b) Onset Onset head head ‘ complement head ‘As (11a) shows, an onset position is free to precede a nucleus. Moreover, as we know from the discussion in 2.3.4, no systematic phonotactic dependencies are to be observed in this context, one of the considerations that led us to recognize the split between these two categories in the first place. In the absence if any obvious compelling evidence to the contrary, there seems to be no support for the existence of string-adjacent clusters of positions occurring in independent onsets, the pattern in (1b). ‘As shown in (12), all permutations of sequences in which a nuclear or thymal position is followed by an onset are attested: (12) First position Second position Nucleus/thyme Onset (a) Nuclear head head (lizard) {b) Nuclear complement head leyzi (lazy) (c)_ Rhymal adjunct head (winter) ‘The independence of the onset and nucleus categories is further evidenced by the lack of systematic distributional dependencies in (12a) or (12b). That is, an onset conso- nant is free to follow either a non-branching nucleus (12a) or a kranching one (12b). By contrast, the coda-onset type of sequence in (12c) constitutes a context where phonotactic interactions are very much in evidence (illustrated by the homorganicity of the nasal-plosive cluster in (winter)). It was this observation that prompted us to set up the distinction between the nucleus and rhyme in 2.4.3. In terms of its ability to influence the distributional potential of a preceding position, an onset can evidently penetrate a branching rhyme but not a nucleus. In the next chapter, we will see further independent support for the nucleus- rhyme distinction. ‘As (13a) and (13b) show, two-nucleus sequences are also possible in English. (13) First position Second position Nucleus/thyme Nucleus (a) Nuclear head head (various) Licensing 159 (b) Nuclear complement head (variety) (©) Rhymal adjunct head . ‘The distribution of the two patterns in (13a, b) is to some extent skewed by a tendency for the first nucleus to be lengthened or tensed in this context. That is, the pattern in (13a) gives way to that in (13b), although the strength of this tendency varies from dialect to dialect.%* There is apparently something quite special about string-adjacent nuclei, since there is strong pressure on them to be collapsed. This raises the question of whether positions in this context can be said to enter into any kind of licensing relation with one another. We must postpone consideration of this matter (until 4.7.4), pending discussion of a number of additional points germane to the issue. (13c) defines a sequence in which a coda is immediately followed by a nucleus. ‘The syllabification it implies (VC.V, as in (pit.y)) violates the otherwise quite robust principle of onset maximization. As we saw in 2.4.2, one of the effects of this principle is to ensure that the C of a VCV sequence is universally syllabified in onset position. As we will see in 4.6.3 and 4.7.3, there is little or no inde- pendent support for analyses which allow for deviations from this basic pattern. Of the various types of inter-constituent sequence just described, two are worthy of immediate attention. One, the coda-onset cluster, involves an attested pattern in which systematic phonotactic restrictions are in evidence. The other involves two patterns for which we have no independent evidence and which we would wish to exclude on general theoretical grounds, namely a coda-nucleus sequence and a cluster of two onsets. The existence of a distributional interaction between a coda and a following ‘onset indicates that a headed licensing relation holds in this context. In this case, the direction of the phonotactic dependency is the opposite of that holding within constituents; itis the right-hand position which enjoys a greater degree of dis- tributional freedom than that on the left. ‘The right-to-left directionality that is indicated by this asymmetry, it can be argued, extends to another of the inter-constituent sequences just reviewed. It is, traditionally assumed that every onset must be supported by a following nucleus. This implies that a nucleus licenses a preceding onset, rather than vice versa. Although onsct-nucleus sequences are not subject to anything like the same degree of phonotactic restrictiveness as coda-onset sequences, the directionality implicit in this assumption is further bolstered by the observation that onset heads, unlike their nuclear counterparts, are not projected to higher levels of the prosodic hierarchy. We conclude therefore that licensing relations in the inter-constituent domains depicted in (15) universally proceed from right to left: (14) Inter-constituent licensing Between constituents, licensing relations are head-final.160 Licensing a5) @ 9 N ® R69 1 \ 1 a) el The inter-constituent licensing relation in (15a) is perhaps the nearest we have ‘come to some traditional notions of the syllable. As we have seen, however, there is little or no motivation for recognizing the syllable as an independent node in phonological structure.#* Instead, itis has been proposed that segment strings are “organized into iterated constituent pairs consisting of an onset followed by a nuclear projection.2* Since the question of whether or not constituents may branch is subject to parametric variation (as shown in (2)), constituent licensing will manifest itself to differing degrees in different grammars. Licensing relations within a particular constituent will naturally only be evident in a language if the branching parameter is set at ON for that constituent. In contrast, there are grounds for assuming that the inter-constituent licensing relation shown in (15a) is obligatorily present in all languages, while that in (15b) is obligatorily present in all languages with branch- ing rhymes. In other words, every onset must be licensed by a nucleus, and every coda must be licensed by an onset. Let us formulate these constraints in terms of the following universal principles:?” (16) (a) Onset Licensing ‘An onset head position must be licensed by a nuclear pos (b) Coda Licensing A thymal adjunct position must be licensed by an onset posit Given the head-final directionality of inter-constituent licensing, the licensing position in both cases of (16) will necessarily occur to the right. By means of Onset Licensing we exclude clusters of two onsets. In the absence of an intervening nucleus, the first onset would remain unlicensed. By means of Coda Licensing we derive the onset maximization effect mentioned above, From the perspective of licensing, the exclusion of VC.V syllabification follows from the requirement that a coda be licensed by a following onset. A syllabification such as * prt. (pity) is ill-formed in terms of (16b), since the coda f is unlicensed.** 4.3.4 Final consonants revisited Consider now the bearing that Coda Licensing has on the syllabification of consonants appearing at the end of an analytic domain.” In 2.4.4, we considered a number of arguments supporting the view that, rather than occurring in coda position as assumed insome traditions, such consonants actually occupy the onset Licensing 161 of what we termed a degenerate syllable. Here is a brief recap of the main evidence. (a) In failing to induce closed-rhyme shortness, final consonants behave ‘quantitatively like word-internal onsets rather than like internal codas. (b) The final consonant position enjoys more or less the same degree of distributional {freedom as an internal onset. (c) The systematic phonotactic dependencies observ- able within final two-consonant clusters are more or less identical to those operating in internal coda-onset clusters. In addition to this empirical evidence, we now sce that the syllabification of a final consonant in coda position must in any case be rejected on theory-internal grounds; itis automatically ruled out by the Coda Licensing principle. The following representation of (pit), in which the ‘appears in a coda, is ill formed for the same reason that a syllabification such as itis: in both cases, there is no following onset to license the coda, a7). oR 0 oN ro Kk x w bord poiot ‘The main arguments for syllabifying domain-final consonants in onset position make appeal to quantitative and distributional considerations which are inde- pendent of any particular theory of phonological constituent structure. Having made the decision to recognize the onset status of such consonants, we then reach the following conclusion which may at first sight seem rather surprising but which wwe are driven to by considerations that are internal to the theory of licensing: a domain-final onset consonant is not strictly speaking final at all but is followed by a nucleus of some kind, This conclusion is a direct consequence of the Onset Licensing, principle (16a). In other words, a final ‘degenerate syllable’ is not degenerate in the sense of lacking a nucleus; it does possess a nucleus, albeit one that is inaudible.” According to this view, a form such as (pit) contains two ‘onsets, each of which is licensed by a following nucleus, the second of which is not phonetically expressed: 8 0 N ON torte ko x x x toro poi The notion that ‘empty’ positions exist in phonology suggests a parallel with empty categories in syntax, a matter we will pursue below. Actually, the term empty turns out to somewhat infelicitous. As we will see presently, there are certain specific circumstances under which such positions do receive phonetic interpretation.162 Licensing In fact, there is yet another reason for rejecting the assumption that final consonants are syllabified in coda position: it makes false predictions about the ‘major types of syllabic systems attested in the world’s languages. According to the final-coda view, the typology is exhaustively characterized in terms of a straight. forward bifurcation into open syllable ‘CV" systems versus clased syllable ‘CVC: systems. Inherent in this classification is the notion that internal codas and final consonants have exactly the same status. The facts, however, contradict this taxonomy. The internal-coda and final-consonant dimensions are actually inde- pendent of one another, as a result of which there are four rather than two major types of syllabic system. Two of the attested types do indeed correspond to the predictions of the final-coda view. Some languages, such as Zulu, lack both internal codas and final consonants; that is, word-final polysyllabic sequences can only be of the shape . .. VCV]. Others, such as English, potentially have conson- ants in both contexts (. .. V(C)CV(C)]). However, the two other types are not allowed for under the simple CV-versus-CVC taxonomy. A third type of sys- tem (Italian and Telugu are examples) has internal codas but no final consonants (.... V(C)CV)). In the fourth type, we find final consonants but no internal codas (... VCV(C)}), as in Luo (a Western Nilotic language spoken in northern Kenya) ‘The four-way classification of syllabic systems is derivable from the intersection of two independent parameters. The presence versus absence of (internal) codas, results from a setting on parameter (2c), which determines whether or not rhymes may branch, We can characterize the presence versus absence of final consonants in terms of a parameter which determines whether or not final empty nuclei are sanctioned in the language: (19) Final-empty-nucleus parameter Final empty nucleus licensed? [OFF¥ON In a system which selects a positive setting on this parameter, a final empty nucleus is able to license a preceding onset, which is occupied by the rightmost consonant in the word. In a system in which final empty nuclei are not sanctioned, the rightmost onset consonant of the word will always be followed by a phonetic- ally realized nucleus. (20) summarizes the four language types defined by the intersection of this parameter with the branching-thyme parameter:** 20) Branching chyme [OFF] ON [OFF] VCv] — ViC)CV) Licensing of final empty nucleus Zalu Telugu ON VCC) VICICVIC)) Luo English The unmarked setting on the final-empry-nucleus parameter can be taken as OFF. This is consistent with implicational patterns such as the following: the existence of final VC] sequences in a language implies the existence of V] but not Licensing 163 vice versa; languages which sanction final consonants are outnumbered by those that do not; and in languages permitting both V] and VC] the former is acquired before the latter. 43.5. Deriving binary branching Together, the principles of constituent and inter-constituent licensing derive the set of constituent structures shown in (1). Any other shape of constituent is automatically ruled out on the grounds that it will contravene licensing require- ments. Take for example the following representations of a ternary-branching, anucleus:!? rT ® oN AN LN XX. Xs 2 cee Ett ‘The structure in (21a), in which the licensing position is flanked by two licensees, is ill-formed because, although locality is respected (both licensed positions are adjacent to the licensor), the strict directionality of constituent licensing is con- travened: position x; licenses simultaneously from right to left and from left to right. (21b) is ill-formed because, although strict left-to-right directionality is respected, locality is not: x, is not adjacent to the licensing position x. In this way, we derive one of the quantitative effects that are observable at this level of structure — a binary limit on vowel-length contrasts (short versus long). Under constituent licensing, we expect three-position super-heavy rhymes to be ruled out for the same reason that ternary-branching nuclei are: (21) (a) (22) oN 4 bs In this instance, locality is not respected: the rhymal adjunct position x3, not being adjacent to the constituent head x, remains unlicensed. Nevertheless, as we saw in 2.4.4, super-heavy rhymes do occur (as in (paint, bold, moist)). The well-formedness of a three-position rhyme, however, depends on the additional support of inter-constituent licensing. To see this, compare the snsing properties of the coda in heavy and super-heavy rhymes. In both cases, the position in question must be followed by an onset under Coda Licensing (16b). This is illustrated in the following representations, which correspond to the164 Licensing ‘emboldened portions of forms such as (fact, fist, chapter, winter) (23a) and (paint, wild, mountain, shoulder) (23b):*" 23) @ R 1 N ° ' ' I The chymal adjunct of a heavy rhyme (23a) is doubly licensed: once by the preceding nuclear position (constituent licensing) and again by the following onset position (inter-constituent licensing). The coda of a super-heavy thyme (23b) is, however, only singly licensed, specifically by the following onset. Inthe latter case, itis the complement position of the branching nucleus that is constitu- ent-licensed. A language for which the branching-hyme parameter in (2c) is set at ON but which, unlike English, lacks super-heavy rhymes is subject to the constraint that codas must be doubly licensed. ‘A rhyme containing any more than one coda consonant is ill-formed for the reason that it would feature at least one position that is not sanctioned by either constituent or inter-constituent licensing. In the following fous-term rhyme, for ‘example, position xs remains unlicensed: (24) 4.3.6 Spreading and autosegmental licensing Under the autosegmental aspect of the Phonological Licensing Principle, we have seen, the phonetic interpretation of an element is contingent on ts being attached toa skeletal position. The appearance or non-appearance of a given element & at particular locations within a representation hinges on whether or not ais granted aan autosegmental licence. The overall melodic complexion of a representation will thus be shaped to a large extent by local responses to a fundamental choice facing individual positions ~ to license « or not to license a. In the following sections, we will try to identify the general constraints that influence or force decisions on this matter. The immediate concern of this section, however, isto explore the significance that this conception of melody-to-prosody association Licensing 165 has for our understanding of the operations of spreading/composition and delink- ing/decomposition. Delinking may be thought of as a specific response to conditions under which an element's autosegmental licence is revoked or withheld. Viewed in this way, the dynamic connoration that attaches to the notion of severing the association, between a melodic expression and a position is not particularly fortunate. The operation of withdrawing an element's autosegmental licence can indeed be involved in dynamic alternations. There are plenty of cases where an element occupying a particular position is phonetically interpreted in one set of contexts but suppressed in another. Think, for example, of one of the first processes discussed in this book, one in which a domain-final ¢ or d is realized before a ‘vowel but not before a consonant (as in (send it) versus (send Tom); see 1.1). This, phenomenon is straightforwardly represented as the delinking of an entire melodic expression in the second context. Nevertheless, there are also plenty of cases in which the withholding of autoseg- ‘mental licensing is not manifested in any dynamic way but is restricted instead to purely static distributional regularities. When we say, for example, that a laryn- geal clement cannot be distinctively associated to a coda position, we are not claiming that such an element is lexically present and then actively delinked in this context. What we are assuming rather is that the element is permanently denied lependent access to this position. However, it would be wrong to conclude from this that two autonomous formal ‘operations are required for the sorts of phenomena just described. This is con- firmed by cases in which what is evidently one and the same phonological process displays both dynamic and static effects, a point already made in 1.2.2. For example, the process whereby a domain-final n loses its coronality and takes on the place property of a following consonant (as in (te{a ~> m] past)) can be viewed asa dynamic analogue of the static homorganicity exhibited by morpheme-inter- nal nasal-plus-consonant clusters. As pointed out in 1.2.2, one way of conceptua~ lizing this situation is to think of place-delinking as operating dynamically in one context and vacuously in the other." This all suggests that the fundamental mechanism at work in all such cases is the denial of autosegmental licensing under particular conditions. The active sever- ance of an association line and the failure of a particular element ever to appear ina particular context are thus superficially different implementations of this one basic mechanism, A similar point can be made in relation to spreading. The term has connotations of dynamic activity, but in the case of many phenomena the device must be assumed to work in a static fashion. The place-sharing of partial geminates in the mi:t (see 2.2.2). Here there is a clear sense in which the source of the vocalic material which fills the shoes of the disappearing velar fricative is located in the skeletal slot to the left. The extension of directionality to static cases of melody-sharing is motivated by the headed nature of relations between adjacent positions. The inequality with which distributional possibilities are divided between adjacent positions in certain contexts canbe explained if we assume thet al or part of the pont: iterpreta ion of a licensed position is determined by the melodic content of its licensor. ‘This implies thatthe relevant melodic content i distinctively lodged inthe & Position, even though its phonetic manifestation is smeared over both lice positions. Oblique-line graphics of the type in (25) are simply a way of making the directionality of spreading notationally explicit. This is further illustrated in the following coda-onset representation of the partial geminate -mb-: 2) R ° NO i N oh 4 “Ni u In (26), the element U is distinctively atcached to the licensing onset position, while the licensed coda is unspecified for a place element. Or to put it somewhat Licensing 167 differently, U is autosegmentally licensed by the onset position, whereas the coda fails to license an independent place element. Note that the spreading indicated by the oblique line does not result in a copy of U being inserted into the coda; nor does it result in the same U being moved from one part the representation +0 another, Spreading is not a derivativual device which bestows on the coda an autosegmental licensing ability it lacks in lexical representation. What the oper- ation does is help indicate the distinctive locus of U as well as demarcate the domain over which the element is phonetically interpreted. Understood in these terms, spreading is a purely interpretive matter, as opposed toa representational one. Thatis, it triggers no changes in the structure or content of a representation. It simply specifies how part of a representation is to be phonetically interpreted. A technical definition of the operation might thus run as. follows: (27) Spreading A licensed position is identified with its licensor with respect to the phonetic interpretation of some element that is distinctively lodged in the later. We have yet to consider the precise conditions under which autosegmental licensing can be granted or refused. As we will see in the next section, the conception of spreading in (27) will turn out to have a significant bearing on how ‘we approach this question. 44 Licensing and melodic complexity 44.1 Government In this section, we will see how an apparently disparate range of phonological phenomena can be derived by means of a single principle which defines an intimate connection between licensing and melodic structure. The principal phe- nomena to be treated in this way are the following: consonantal weakening, vowel reduction and syncope, and sonority relations as reflected in the phonotac. tic dependencies that manifest themselves within and between constituents. All of these can be related to a fundamental asymmetry in the ability of different types of melodic expression to occupy particular positions of licensing. Concretely, some sounds make better licensors than others. Let us begin by taking stock of the criteria by which the directionality of licensing within different domains is established. One criterion has to do with whether or not the presence of a position is mandatory within a given domain. It is partly on the basis of the assumption that obligatoriness indicates headship that nucleus is identified as the licensor of an optionally present onset (to its left) or thymal adjunct (to its right). By the same criterion, Coda Licensing (16b) estab-168 Licensing lishes the obligatorily present onset as the head of a coda-onset domain. A second criterion concerns quantitative interactions within the chyme: as observed in the phenomenon of closed-rhyme shortness, the presence of a branching nucleus greatly curtails the availability of a rhymal adjunct position. Third, the head of certain licensing, domains can be established by reference to the greater degree of distributional latitude it enjoys relative to its licensee. It is on the last of these criteria that we focus our attention in this section. ‘The first point to note is that the pivotal role played by the nuclear head in licensing networks appears to insulate it from distributional dependencies with surrounding onset and coda positions. It is as if its obligatory presence as a licensing anchor accords it the privilege of selecting melodic material inde- pendently of whatever material is contained in these other positions. Hence the lack of any systematic phonotactic interaction between a nuclear position and a following rhymal adjunct. In fact, this distributional independence is one of the ‘main motivations for recognizing an autonomous nuclear constituent in the first place. By the same token, onset-nucleus sequences display little or no phonotactic interplay. Given this proviso about the special status of nuclear positions, we may observe various types of distributional asymmetries at all levels of phonological licensing. However, it is in the following three domains, identified in 2.3.4, that quite particular phonotactic restrictions are universally in force: (28) Phonotactic domains: (a) within a branching onset; (b) within a branching nucleus; (c)_ between a coda and a following onset. In each of these contexts, the licensed position displays a seriously depleted set of distributional options. This indicates that tke licensing relations involved are of a quite speci government.» That is, each of the following licensing domains corresponds to a domain of government: (29) Governing domains (®) w () R °. N ° NY, \ 1 xe x Xe xm Ut Lt t In the sub-cases of constituent licensing in (29a) and (29b), a constituent head governs its complement. In the sub-case of inter-constituent licensing in (29¢), an onset head governs a preceding rhymal adjunct. ‘What formally unifies the three governed positions in (29) is that they are what right be termed persistent non-heads; that's, they never enjoy head status at any type. We may refer to this restrictive sub-case of licensing as _ Licensing 169 level of projection. This immediately sets chem apart from all other positions, which, although they may be non-heads at some level of projection, are neverthe- less guaranteed headship at some lower level. This is true of nuclear heads, which have the potential to function as such throughout the prosodic hierarchy. Tt also applies to onset heads, whose headship extends to both constituent and inter-con- stiruent domains. To put it somewhat differently, we can say that (30) Heads are ungoverned. (30) is a necessary but not sufficient condition for factoring out the set of domains in (29). The following configurations also contain non-head positions, but they must be excluded as possible domains of government: (31) fa) Rhymal domain td (31a) constitutes a licensing domain, as demonstrated by such phenomena as closed-thyme shortness and chymal weight. It does not, however, constitute a governing domain, as demonstrated by the phonotactic independence of the nucleus and the coda. Evidently, the notion of immediate projection is crucial to the definition of government. Specifically, within a branching constituent, the governor must be the immediate head of its governee. Thus the complement x; positions in (29a) and (29b) are governed, by virtue of the fact that the onset and nucleus categories are immediate projections of their x; heads. The adjunct x, position in (31a), on the other hand, is not governed within its constituent, since the rhymal node is not an immediate projection of its head xy. Neither (31b) nor (31¢) constitutes a licensing domain; and, since govern- is a sub-case of licensing, this rules them out as governing domains. How- , it is instrictive to” Consider why this should be so, since the notion of immediate projection can again be shown to be implicated. As noted briefly above, for the purposes of imposing phonotactic restrictions on a preceding position, an onset-head position can evidently penetrate a rhyme at the level of the adjunct, as in (2%), but not at the level of the nuclear complement, as in170 Licensing (31D). In syntactic terms, the nucleus beraves as an ‘island’, insulated from potential governing relations emanating from outside the constituent. This effect can be interpreted as a specific response toa general grammatical constraint, the Minimality Condition. According to one clause of the Condition, a position A ig prevented from governing a position B, if the immediate projection of B's head excludes A.?” This means, among other things, that an external position cannot ‘reach inside’ a nucleus to govern a position within the constituent, as in (31b). The same principle blocks a nuclear head from governing inside an onset, as, in (31c). The coda-onset domain (29c) is exempt from this constraint. That is, an onset head is able to govern inside a rhyme, since the latter does not itself constitute a governing domain. The idea to be developed in the following sections is that the asymmetries that are observable in the phonotactic domains listed in (28) reflect inequalities in the ability of different types of melodic expression to occur in particular positions of government. The specific proposal is that the capacity of a sound to occupy a particular position is determined by the complexity of its elemental make-up. Complexity is gauged by the degree of fusion embodied in a melodic expression. For most purposes, this boils down to a straightforward calculation of the number of elements of which the sound is composed. For example, the two-element compound (U, h] (defining f) is more complex than the simplex expression {h] (defining). The basic idea is that, within the domains specified in (29), the melodic expression occupying the governing position must be at least as complex as the expression occupying the governee. This principle may be formulated as follows: (32) Complexity Condition Let a and B be melodic expressions occupying the positions A and B respectively. Then, if A governs B, Bis no more complex than a. 44.2 Complexity relations within onsets The bottom line drawn by (32) is that the complexity differential between a governor and a governee can be zero. Otherwise, a more stringent requirement holds, whereby the governor must be more complex than the governee. We will now see that the more stringent condition controls the relation between a head position and its complement within an onset constituent. Since the directionality of constituent licensing is head-initial, this means that the well-formedness of a branching onset depends on there being a downward complexity slope between the two positions (viewed from left to right). This pattern manifests itself in well-formed branching onsets which, as we saw in 2.4.2, consist of an obstruent followed by a liquid or glide.” As illustrated in (33), the element profile of this type of sequence is uniformly downward. Licensing 171 (33) (a) dr (b) tw (c) pl (a) ° °. ° ° m™ Im IN I x x x ox x x x ox tot tod toot toot R R R U UR UR 1 | ro 1 1 1 h h h h \ 1 1 H H H Compare the well-formedness of these structures with the malformedness of the following, two of which (34a, b) display an upward complexity profile and one (B4c) a level pattern. (34) (a) ed (b) st {c) sm "oO *o I~ I tod 1 1 I RR RR u I ! ' ? 2 ? \ i i h h bh bh N io \ HO H ‘As a comparison of (33) and (34) shows, the complexity of an onset complement is restricted to one element (glides and 1) or maximally two (In some languages, a maximum of one element is tolerated in this position. (Witness, for example, the historical reduction of / in this position to a simplex segment in some Romance languages: r (R) in Portuguese and y (1) in Italian, cf. French (plat), Portuguese {prato), Italian (piatto) ‘dish’.) ‘The Complexity Condition goes a long way towards deriving the sonority effects that are observable within branching onsets. Other phonotactic restric- tions operating in this context are attributable to an independent constraint on the amount of melodic material that the head and complement of an onset can have in common. Complete identity, for example, appears to be universally culed out. That is, there are no known languages possessing onset geminates. The upward limit on the number of elements that can be shared between the ‘wo positions appears to be one. Thus, while tr/drare well-formed onsets (one element ‘common, R), the following are malformed: ¢/d! and in/dn (all with two172 Licensing elements in common, R and 2).#° Note thet st clusters are doubly ruled out as onsets (see (346) ) - once for having an upward complexity gradient, and again for sharing more than one element. 4.43 Complexity relations within nuclei ‘When we turn our attention to branching nuclei, we find at least one similarity with the complexity profile of onsets: the number of elements that can be contained in the complement position is severely restricted, in this case to one. The latter can be I, U or @, each of which characterizes an off glide, as illustrated in diphthongs such as ey, ow, ia. With this configuration, a branching nucleus displays either a level or a downward complexity profil (35) (a) aw (b) ia (ow (@ N N N N IN Im™N m™N I™N ee yx x x x x A e A A f 1 u 1 u 1 The toleration of a level complexity differential is not the only characteristic that distinguishes branching nuclei from onsets. The existence of long mono- phthongs indicates that, unlike in onsets, gemination is possible in nuclei. As we saw in 3.5, true geminates are represented as the simultaneous association of one melodic expression to two skeletal point. I: is widely accepted that this arrange- ‘ment reflects the operation of the following convention:"! (36) Obligatory Contour Principle (OCP) At the melodic level, adjacent identical units are disfavoured. The term principle here may not be entirely felicitous, atleast if we take it to imply universal inviolability. The OCP was initially designed to account for certain tonal regularities but has subsequently been extended to a range of other phenomena, including some in which its effects are evidently not enforced in all languages. Although it is thus more appropriate to think of the OCP as embodying a strong. tendency in natural language rather than a rigid principle, the name has stuck.” Under the OCP, apparently consecutive identical segments are represented as @ single melodic expression attached to two skeletal points. There has been some debate about the precise workings of the OCP and the domain within which itis ‘operative. It is generally agreed that minimally the convention acts as a static condition on the well-formedness of morphemes as they appear in the lexicon. According to this aspect of the OCP, any intra-morphemic geminate cannot Licensing 173 constitute a sequence of independent melodic expressions (asin (372) ) instead it can only be represented as the simultaneous association of a single expression to ‘two skeletal positions (as in (37b)). G7 (a) te x ) x x tot Vs ao a AA further assumption regarding the operation of the OCP is that it remains active during derivation. In this guise, it intervenes in a language-particular manner to block processes which would violate it or to set off processes which repair such violations. In the latter function, the convention triggers the coalescence of identical melodic expressions which accidentally become juxtaposed as a result of morphological concatenation. This affects so-called ‘fake’ (i.e. non-lexical) gemi- nates, such as the nm in (un-nerved). In all instances, true gemination involves a relation of licensing between two positions. In the case of nuclear geminates, the relation is one of constituent licensing. As indicated in 4.3.6, the melodic content of a geminate can be assumed to be lexically specified in the licensing position and spreads automatically into the licensee. Given the left-headedness of constituents, this means that spreading within a nucleus is from left to right. Thus in the representation of a long ‘monophthong such as (38) N IN Ly A the governed position is empty. Its phonetic interpretation depends on its being identified with the melodic material that is lexically specified in its governor. Interpreted in this way, long monophthongs, and indeed all geminates, display a complexity slope between governor and governee. ‘A well-formed branching nucleus then is one which displays either a downward or a level complexity profile, The following upward patterns are ruled out by the Complexity Condition:** (39) (a) ae (b) io (¢) wo “N °N °N IN IN Im to 1 i AOA A A i 1 | I rou uu174 Licensing (Recall from the discussion in 2.4.3 how vocalic sequences transcribed in terms such as ae, io, uo are analysed in terms other than branching nuclei, for example as sequences of independent nuclei or as contour structures attached to a single nuclear point.) 4.44. The melodic content of codas ‘The amount of melodic material that a coda position can sustain is tightly constrained, an effect that is due to the governance of this position by a following ‘onset. That is, the Complexity Condition can also be observed to operate within the inter-constituent domain shown in (29). Here a complexity gradient appears to be universally in force, in this case an upward incline (viewed from left to right) in line with the right-headedness of licensing at this level, ‘The most stringent enforcement of the Complexity Condition in the inter-con- stituent context is observable in so-called ‘Prince’ languages, i.e. those languages in which coda-onset sequences are restricted to full or partial geminates.* As indicated above, the distinctive source of melodic material in a geminate is the licensing position. The melodic identification of the licensee with the licensor consists in the spreading of the entire melodic expression (under the ROOT node). The relevant geometric detail is supplied in the illustration in (40). As in the case of a long monophthong, the governed position of a full geminate consonant has zero complexity: (40) ROOT wom—s—v—+ x -0 In the case of nasal-obstruent partial geminates, illustrated in (26), the coda rust be assumed to license the nasal element (otherwise mb, say, would be indistinguishable from bb). We must also conclude that ? is distinctively specified in the coda rather than having its source in the licensing onset. This conclusion is Licensing 175 supported by the fact that the occlusive property of a nasal stop manifests itself irespective of whether it is followed by a plosive (which also contains ?) or a {ricative (which lacks it). That is, although the stop property of a nasal could conceivably be attributed to spreading of ? from a following plosive, this analysis cannot be extended to nasal-fricative clusters. In fact, as we saw in 3.5, a nasal ‘can pass its occlusion on to a following fricative to create an affricate of the type seen in prints (prince). ‘On the other hand, the place-defining element in a homorganic cluster is specified in the onset and spreads (under the PLACE node) into the coda. This observation, together with the observation that the noise element and, if present, a laryngeal element are also lodged in the onset, confirms the existence of an upward complexity slope in partial geminates. The pattern is illustrated in the following representation of -at- (41): (41) ae R N \ The Complexity Condition can also be observed to operate in non-Prince languages, i.e. those that permit obstruent clusters which are not necessarily homorganic. In such sequences, we must conclude that a place-defining element is distinctively present in the coda, since obstruents in this position contrast with respect to this dimension in languages of this type. In English, we have a contrast between sand before ¢ (for example, (castor) versus (after)), as well as between and k before ¢ ((chapter) versus (factor)).** In fricative-stop clusters, the onset position contains at least one more clement than the coda, namely 2. A charac- teristic of oral stop clusters is that only the second sound is released, an effect that can be interpreted as an optimization of the complexity differential between the two positions. The governing onset position contains an element, h (contributing noise release), which is absent from the governed coda position. Abstracting away from the laryngeal dimension for the moment, we can see from the follow- ing structures how an upward complexity gradient is maintained in obstruent clusters: wom—-r—~--x176 Licensing (42) (a) (b) pe R R 4 i 9 N ~ | \ | \ toot | 3 , t bo \ | + h 1 \ ur uo oR ‘Another phonotactic restriction on coda obstruents can be viewed as a complex: ity effect the absence ofa distinctive laryngeal property. We noted in 2.4.2 that Cada-onset clusters of obstruents uniformly agree with respect to voice. As with fgeminates, this state of affairs reflects the datinctive specification in the governor of melodie material that spreads into the governee. In fact, it has been proposed by some researchers that lexically specified laryngeal elements are uniformly ‘excluded from all governed positions and are universally present in all obstruents ‘occupying governing positions.” 4.4.5 Deriving sonority To sum up the discussion so far: the Complexity Condition accounts for the phonotactic relations that obtain within the contexts identified in (28) and formalized in terms of government in (29). In chapter 2, these distributional patterns were informally described in terms of sonority. In some approaches, cis potion is accorded theoretical status and is embodied in the principle of sonority Sequencing (discussed in 2.4.2), which regulates the co-occurrence of different segment types within the same set of contexts. ‘One respect in which the complexity proposal is similar to the sonority ap- proach i in the presentation of a fixed range of choices which determine phono” tactic differences among languages. Under the sonority account, what varies isthe ‘Minimal sonority distance a language imposes on pairs of segments in each syllabic context. Roughly speaking, this coreesponds to variability inthe steep- nese of the complexity gradient between adjacent positions, where the direction of the slope is determined by the direction of licensing. Tn other respects, however, the sonority and complexity approaches are not at all similar, Rather than being an intrinsic property of segments that can be Sireety read off representations, sonortys calculated by reference roan external look-up table, the sonority hierarchy, which translates the feature make-up of & ‘ement into'a rank on the hierarchy. According to the complexity propose) Sonority has no theoretical status. Instead it is no more than an informal label for Licensing 177 ‘an effect which derives from a more fundamental property, namely melodic fomplexity. And this property, being directly coded in melodic steucture, can be Geeermined without reference to any independent hierarchy. In this respect, the complexity approach is simpler than the sonority account. But not only that ~ the empirical domain it covers is much larger. The complexity dimension can also te shown to be involved in consonantal weakening and vowel reduction, pro- cesses which do not readily submit to interpretation in terms of sonority. Tn spiantization, for example, suppression of ? results in a reduction in the ‘melodic complexity of the lenited segment. Viewed in this way, lenition is simply the dynamic counterpart ofthe static distributional patterns that are traditionally Uescribed in terms of sonority. What is pethaps not so obvious, however, is how the licensing aspect of the complexity approach can be made to account for the Contexts in which consonantal lenition typically applies. One of the classic Ienition sites is word-final position, where, according to the licensing principles iGscussed in this chapter, the target consonant occupies an onset followed by an empty nucleus. Although this context does not directly coincide with any of the Ticensing domains described so far, the reduction in melodic complexity that ‘occurs here can be shown to be indirectly sensitive to licensing. We will return to this issue later in the chapter. 4.4.6 Complexity relations between nuclei ‘The contexts in which phonological interactions between positions are observable are not limited to those listed in (28). A whole range of both distributional and ‘dynamic phenomena can also be witnessed at higher levels of the phonological hierarchy, Processes which reflect close interactions between nuclear projections include stress-conditioned vowel reduction, vowel syncope, vowel harmony and tonal alternations, As we saw in 3.3.5, vowel reduction results in a contraction of the melodic contrasts capable of appearing in positions of weak stress. In the case fof vowel syncope (to be discussed in detail lar in 4.6), the melodic content of a nucleus is completely suppressed under a specific set of conditions, one of which involves the nucleus occurring within the same licensing domain as a neighbour- ing nucleus. In the case of harmony, one nucleus determines one or more of the ‘quality characteristics of other nuclei within the same domain (usually the word). The parametric variability to which the directionality of projection licensing is subject is no less evident in harmony than in stress systems. In some harmony systems, for example, the harmonic category propagates from the left edge of @ harmonic span, whereas in others it propagates from the right. In spite of their superficially disparate nature, all of these phenomena share a pattern in which a nuclear licensor enjoys a greater degree of distributional leeway than its nuclear licensee. Tn terms of their melodic effects, each of these phenomena displays a particular type of complexity interaction between nuclei at some level of projection. Vowel reduction in a metrically weak nucleus consists in the suppression of one or more178 Licensing of the elements it contains. In height harmony systems of the type discussed in 3.3.5, the absence of mid vowels from certain harmonically recessive nuclei consists in the suppression of A. What these processes have in common is a diminution in the melodic complexity of a licensed position in relation to its licensor. Vowel syncope manifests the same phenomenon but ina more extreme guise. resulting in the complexity ofa licensed position being reduced to zero. in keeping with the cross-linguistic variability that characterizes other phenomena triggered by nu- clear licensing, complexity differentials between positions are not strictly en- forced at higher levels of the prosodic hierarchy. Not all languages exhibit vowel reduction, height harmony or syncope, just as not all languages have, say, stress feet. 4.5 Principles and parameters of phonological structure Reviewing the discussion to this point, let us now summarize the principles and parameters which control the prosodic and melodic construction of phonological representations. Together with the Complexity Condition, the Phonological Licensing principle, in its various incarnations at different level of the phonologi- cal hierarchy, defines the most important design properties of the phonological structure of natural language. The main effects of the Phonological Licensing principle ate: (a) to make the phonetic interpretability of representational units dependent on their integration into the phonological hierarchy; (b) to impose a binary limit on the branching structure of constituents; and (c) to define the conditions of locality under which phonological processes operate. The Complexity Condition controls the associ- ation of different types of melodic expression to particular positions in phonolo- sical strings and in this way derives the effects of sonority sequencing. ‘The scope for cross-linguistic variability in the design of phonological structure is circumscribed by a number of parameters, including the following: (43) (a) Branching constituent Onset {OFFVON Nacleus {OFFVON Rhyme [OFEVON (b) Licensing of final empty nucleus (OFFON (c)_ Direction of projection licensing —/e— (4) Complexity gradients within governing domains Settings on the constituent parameters (43a) determine whether of not a particu lar language displays such properties as distinctive vowel length or closed sylla- bles. (43b) controls the sanctioning or otherwise of word-final consonants. The setting on (43c) establishes the difference between, for example, left-dominant and right-dominant stress patterns. Different values on (43d) determine such factors as the sonority distances between sounds in adjacent positions. Licensing 179 For each language, the universal principles of Phonological Licensing and the Complexity Condition, supplemented by particular settings on the parameters in (43), define a set of well-formed prosodic templates and a set of well-formed melodic association patterns. The grammaticality of a phonological repres- entation is dependent on its being parsable in terms that satisfy these prosodic templates and melodic patterns.** Ieis now time to examine in more detail the issues raised by the postulation of the type of position referred to in parameter (43b) ~ the empty nucleus. 4.6 Empty positions 4.6.1 Domain-final empty nuclei So far, the only argument I have advanced in favour of the recognition of empty tuclei is a theory-internal one. As we have seen, there is sizeable body of independent evidence supporting the view that word-final consonants occur in onset position. The principle of Onset Licensing (16a) then leads us to conclude that this position must be sanctioned by the presence of a following nucleus, even if this apparently has no phonetic expression. Among the questions raised by this, analysis are the following. Is there any additional, independent support for empty positions? Is the occurrence of empty nuclei restricted to the right edge of words? If not, what constraints are there on their appearance in other contexts? Are empty nuclei truly empty in the sense of being completely and consistently devoid of all phonetic content? Do empty positions have any analogues elsewhere in phonology or indeed in grammar in general? On the last of these questions, the precedent of recognizing empty positions in phonology was originally set not for nuclei but for onsets. One of the first analyses to incorporate the notion of an empty onset position was proposed to handle the phenomenon of f-aspiré in French." Vowel-initial words in French pattern into two sets according to whether or not they trigger certain phonolo- gical processes, one of which involves the suppression of the vowel of the singular definite article 1 (masculine) or Ja (feminine). In words such as ami (‘fiend’), the yowel of the article fails to appear: lami (‘the friend’). The vowel-zero alternation is usually treated in cerms of a process which deletes the first of two adjacent auclei: 12 ami —> 1 ami, In b-aspiré words, however, the vowel of the article is retained, as in /o ariko (‘the bean’). Forms of the latter kind, it has been proposed, begin with an empty onset position which separates the nucleus of the article from the first nucleus of the noun and thus shields the former from deletion. (As the term -aspiré suggests, the empty position is the present-day reflex of what historically was an onset occupied by f. The /thas long since been lost, although its original presence is still recorded as (h) in French orthography.) A little later, ‘we will consider the question of whether the notion of empty onsets and nuclei in phonology has anything at all in common with empty categories in syntax.180. Licensing Turning to another of the questions raised above, let us now consider a number of arguments in support of the view that so-called ‘empty’ nuclei are not entirely without melodic content. Rather they have latent content which manifests itself phonetically under certain specific ciccumstances. This isa suitable point at which to return to the suffixes (-{c)s) and (-ed), since the manner in which they alternate sheds light on this issue. ‘Of the three alternants displayed by each of these suffixes, we have already seen in 3.6 how two, 2-s and d-tcan be accounted for in terms of voice assimilation, Let us now focus on the third alternant, the one found in forms such as those in (44) containing a vocalic reflex before the suffix consonant. (In (44), the vowel in question is represented in one of its most widespread guises, namely 2, Other attested reflexes include 1) (44) (s) (ed) “on vod kisses voted houses faded bushes catches judges The vocalic reflex occurs whenever the stem ends in a consonant which shares certain place and/or manner characteristics with the consonant of the suffix: oz occurs after a sibilant and ad after t or d. The presence of the vowel is usually interpreted as breaking up a potential sequence of identical or similar consonants. Traditionally, this effect has been treated in terms of a rule of vowel epenthesis ~ in non-linear terms, the insertion of a nuclear position. Underlying this analysis is the now-discredited assumption that the stem-final consonant occupies a coda. Once epenthesis has applied, the stem-final consonant undergoes re-syllabifica- tion; that i, itis moved out of its original coda position and into the onset of the newly created syllable. For (voted), this yields the following derivation: vowt(d} > vowt.ad {epenthesis) —> vow.tad (resyllabification).*" This particular movement operation is representative of a whole class of resyllabification transformations which have been proposed in the literature. We will have cause to question their validity in due course. Suppose, on the other hand, we take the vocalic reflex of (-(e)s) and (-ed) to be part of the stem domain rather than the suffix. We then have no need to invoke resyllabification if, as depicted in the representation of (voted) in (45), we assume that the vowel is the phonetic expression of a latently present reutral element (@) in the stem-final ‘empty’ nucleus.* (Here and elsewhere, I wll only indicate the presence of those elements that are directly relevant to the point at hand. The remaining melodic content of a representation can be abbreviated by means of phonemic symbols in inverted commas.) Licensing 181 (45) ON oN ON rin bout Ie x x x x] x x} (voted) toro | 1 vow e R R 1 1 1 \ h h 1 H The idea that @ is latently present in an ‘empty’ nucleus is in accord with the assumption made in 3.3.5 that it inheres in every intersection between a skeletal position and an autosegmental tier that is not filled by some other element. It is consistent with this view to consider an ‘empty’ nucleus as comprising intersec- tions which contain nothing but @. Ina system in which a nuclear position of this type is parametrically sanctioned domain-finally, this melodic material remains autosegmentally unlicensed and thus phonetically unexpressed. (The absolute final nucleus in (45), the one following the dof the suffix, thus also contains latent @, although in this context it is not made phonetically manifest.) However, should special circumstances arise under which the melodic content of the nucleus is phonetically interpreted, it should have the schwa-like quality that inde- pendently manifests @. In this case, the special circumstances evidently involve a ‘ban on consecutive identical elements of a particular type, those shared by the sibilants or by the coronal plosives in (45). This constraint can be viewed as an instantiation of the OCP (36). ‘The claim that a final ‘empty’ nucleus is latently realizable as the neutral element raises the question of how this is to be distinguished from a final nucleus which is consistently realized as schwa. In what way are the representations of, for example, (ine) and (Dinah) distinct? In the case of some dialects to be discussed in the next chapter, there are grounds for assuming that there is no genuine contrast here at all, since stable schwa is absent from this context. These are the systems that display so-called ‘intrusive’ r; pre-vocalically, a form such as (Dinah) shows final r and is thus indistinguishable from (diner) (so that (Dinahfr] is), for example, is identical to (diner is)). As we will see, such forms can uniformly be assumed to end in a lexical r which fails to materialize pre-consonantally. This means that all apparently schwa-final forms are in fact rfinal in these systems. ‘Nevertheless, those dialects which lack intrusive rand thus contrast (Dinah) and (diner) do have final stable schwa in words of the first type. The manner in which this vowel is distinguished from a final empty nucleus in a form such as (dine), Twill assume, has to do with whether the melodic content of the position in question is headed or not. Specifically, when the melodic content of an ‘empty’ position is not autosegmentally licensed, it has no element as its head. As shown in (46), this is182 Licensing in contradistinetion to a non-empty nucleus corresponding to stable schwa, in ‘which the relevant position autosegmentally licenses a headed melodic expression, (46) ‘Empty’ nucleus Schwa N N 1 D 1 @ @ 4.6.2 Vowel syncope Focusing now on another of the issues raised by empty nuclei, let us consider whether they are restricted to domain-final contexts. A recurring theme in chapter 2 and the present chapter has been the claim that the existence of systematic interactions (phonotactic or quantitative, for example) between two positions can be taken as evidence that they are adjacent in phonological structure. Conversely, the absence of any such interdependence between two superficially abutting segments indicates lack of adjacency. As we saw in 2.3.4, cross-word contexts are one of the sites where no systematic distributional interactions are evident in English, in this case since they involve arbitrary juxtapositions of segments created by lexical insertion. Under the licensing account developed above, the lack of phonological adjacency between two consonants appearing in this con- figuration is reflected in the fact that they are separated by the empty nuclear position which appears on the right edge of the first word. This is illustrated in a string such as (.... kid lay... ) ({axdf] [Jeyl] (where indicates the presence of ‘an empty nucleus), in which no phonotactic dependency exists between the dand 1. The same point can be made with respect to word-internal segment sequences separated by an analytic morpheme boundary, as in (bad-ly) [beef ‘Temight be objected that the lack of distributional interactions in such contexts ‘can be explained purely in terms of morpho-syntactic domairs, without recourse to empty nuclei. Indeed, it might be tempting to assume that all cases of phonological non-adjacency between superficially contiguous consonants are due to the inter- vention of an analytic morpheme boundary. This assumption, however, turns out to be wrong. The world’s languages abound with examples where superficially abutting but phonologically non-adjacent consonants co-occur within the same morpheme. The cases in question involve contexts in which a pair of consonants flank a site displaying an alternation between the presence and absence of a vowel. In linear rule-based frameworks, the decision to treat alterations of this type in terms of vowel insertion (epenthesis) or deletion (syncope) hinges on the predict ability or otherwise of the alternating vowel. The insertion analysis is only available if both the quality and the incidence of the vowel are predictable. (The incidence is predictable as long as there are no forms containing the epenthesis context which systematically fail to show the vowel.) If either of these conditions is not satisfied, the alternating vowel is represented lexically and then deleted by rule. Licensing 183 ‘An example of a language in which the incidence but not the quality of the syncopating vowel is predictable is Tonkawa, an Amerindian language of Texas." The syncope site is the second nucleus in the word (counting from the left). Thus the second vowel of a stem such as picena ‘cut’ is suppressed in picna-no? ‘she is cutting’. When a prefix containing a vowel is artached co the steam, however, the second stem-vowel is no longer second in the word, and it is the first stem-vowel that is suppressed, as in we-pcena-no? ‘she is cutting them’. Various forms of syncope occur in English, although these are typically optional rather than obligatory as in Tonkawa. One type, which is subject to regional and stylistic variation, is reflected in the suppression of the parenthesized vowel in words such as (reffe)rence, choc(o)late). The optionality of the process is reflected in the fact that it is favoured in casual speech styles and is more likely to affect high-frequency words than rarer words.** The syncope site in this case is located in the second of three nuclei in a strong-weak-weak metrical configuration. That is, as shown in the representation of (sep(a)rate) (adjective) in (47), syncope occurs in the post-tonic nucleus of a left-dominant foot which forms the left branch of what is sometimes termed a super-foot. 47) Super-foot 1 Foot Oo NON ONOWN trorririd x x x x x x x x rororoareroar se p @ rat The sensitivity of syncope to the presence of a following weak nucleus is demon- strated by the fact that the process fails if the post-tonic nucleus is followed by a secondary-stressed nucleus occurring in an independent foot. Thus, while syncope is possible in the adjective sep(o)rot (séparate), it is ungrammatical in the differ- ently stressed verb separeyt (séparate) (*sepreyi). In most dialects, the suppressible vowel appearing in the syncope context is some kind of schwa. The process thus presents the converse case of syncope to that found in Tonkawa, one in which the quality of the alternating vowel is Predictable (9) but not its incidence. Compare, for example, the f(a)r string of ‘reffe}rence), in which an optional schwa intervenes, with fr in (Africa), which never contains schwa. This unpredictability renders an epenthesis solution im- possible; a rule inserting schwa in (ref'rence) would also erroneously insert it in Africa), yielding *(Af{olrica). The obvious alternative is to assume that, in lexical representation, a vowel intervenes between the f and r of (reference) (but not of {Africa)) which is then suppressed by syncope. (In some dialects, the identity of the alternating vowel is unpredictable, since there is a contrast in this context between o (as in ¢ref({2})rence)) and 1(as in (def(te)site)).*7)184 Licensing 4.6.3 Resyllabification? The linear treatment of syncope as vowel deletion converts into non-linear format in one of two ways: either (a) delete the nuclear position together with its melodic content; or (b) delete the vocalic melody while leaving the nuclear position intact, ‘The consequences of these two approaches are quite different. The first requires the mechanism of resyllabification to repair the damage inflicted on constituent structure by the obliteration of a nuclear position. When applied to (sep(a)rate), the latter operation casts the preceding onset consonant adrift, as illustrated in (48). (48) @) {b) Syncope ONONONON ONOONON prrrrrte Prrrrre xxx xxx xx xxxxxxx pett tte rrererte separat seprat Two resyllabification strategies are potentially available for reintegrating the stray onset consonant of (48b) into constituent structure: it can join either the preceding hyme (49a) or the following onset (496). (49) (a) ) OR 1 7 ONO NON ox\ oxen Prin ved tt rrr xxx xxx x xx xxxx x trigded tropes seprat se prat In this particular instance, appeal might be made to sonority sequencing and onset maximization; together, these would favour (49a) over (49b) on the grounds that prexists independently of the syncope contexts as a well-formed onset in Eng lishs* However, this account cannot be extended to all cases of this type of syncope, since, as we will now see, many of the consonant strings resulting from the process do not coincide with the set of grammatical onsets. The type of English syncope under discussion is sensitive to the identity of the consonant immediately following the syncope site. In the overwhelming majority of cases, the process affects forms in which this segment is a resonant. Thus we find syncope in words such as (op(e)ra, especi(a)lly, pers(o)nal) but not in the likes of (bracketing, gossiping, menacing). For most forms which display unsyncopated and fully syncopated variants, there exists an intermediate variant containing @ syllabic reflex of the consonant in question: for example, (oplrla, especllly, pers{alal). The triggering of syncope by a following resonant is thus related to the Licensing 185 fact that these are the only consonants in English that can enjoy syllabic status. (This is a characteristic they exhibit in other contexts as well; cf. forms such as (bortle, button) with final syllabic and n respectively and, in thotic dialects, (adder) with syllabic r) In forms in which the consonants flanking the syncope prone nucleus are both coronal resonants, it is frequently difficult to distinguish, the syllabic from the fully syncopated variant (presumably because of the poten tially continuous sonorant gesture involved). Nevertheless, there are dialects in which syncope appears to be disfavoured in this context, to some extent after 2 (as in (finally, general)) but more particularly where the consonant on the left is a liquid (as in (felony, salary, quatrelling, icony)). As with the vocalic ¢e)s, -ed) reflexes discussed above, this pattern evidently constitutes another case where the suppression of vocalic material is blocked by the OCP, if the process would result in sequences of identical or similar melodic expressions. By way of a more extensive illustration of the consonantal conditioning of this, type of syncope, compare the forms in (50a, b), in which the syncopated variant is readily attested in many dialects, with those in (50c), in which it is either marginally possible or downright impossible (indicated by 2/*):*° (50) (a) sepfa)rate (b) miste)ry des(ojlate temp(e)rature evlelry especially elab(o)rate surg(e}ry fin(a)lly factlo)ry cent(u)ry fam(i)ly bound(a)ry urs(e}ry ped(a)lling liclo)rice cam(e)ra jav(ellin choc(o}late treastu)ry gen(e)ral myst(ery pris(o)ner ‘marg(i)nal refle}rence dec(imal comp(a)ny_ quandia)ry deffijnite pers(o)nal awf(u)lly ople)ner nati(o)nal () rock(e}ting ball(o}ting mon(i}tor opac(idty goss(i)ping menta)cing Note that the difference between (50a, b) and (S0c) cannot be explained by saying that syncope only occurs between consonants that can potentially form a branching onset. True, the grammatical forms in (S0a) demonstrate syncopated consonant strings which superficially correspond to onset clusters that are at- tested independently of the syncope site, including pr, br, tr, dr, kr, Kl, ff, Or. Moreover, many of the ungrammatical sequences that would result from syncope in (50c) correspond to independently established coda-onset clusters rather than onsets, (kt It, af, st, sp, ns, for example), This result is, however, fortuitous in view Of the fact that the largest set of forms, those in (S0b), contains syncope-derived strings which fail to correspond either to onset or to coda-onset clusters that are186 Licensing independently identifiable outside the syncope site. These include 21, mr, ér, jr, 3% V4, ai, fh pt, fa, 20, ry fr. low then does a tesyllabificaton account deal withthe forms in (506)? Since the resyllabification of the syncopated strings in such forms is not responsive to pre-existing phonotactic constraints, it will have to be guided dy some other set of factors. One possibility would be to assume that the sonority sequencing principle continues to operate after syncope but with a relaxation in the severity with which the requicement on minimum sonority distance is enforced. In this way, strings such as ml, fn, pn, for example, could be made :o pass muster as onsets. ‘Another tack would be to assume that sonority sequencing switches off at some point in derivation, in which case any syncope-generated string that does not correspond to a pre-existing onset (those shown in (5Ob)) could be parsed as a novel coda-onset sequence. The problem then is how to derive the novel clusters while preventing the creation of the very clusters that correspond to otherwise ‘well-formed coda-onset clusters, those illustrated in (50c). A furcher problem with this alternative is that, in some of the syncopating forms, the rhyme of the tonic syllable is already occupied by a coda consonant (the m in (comp(a)ny) for instance). In such cases, resyllabification of a post-syncope consonant into the preceding rhyme would result in a two-consonant coda, something for which there is absolutely no independent evidence, as we saw in 2.4.4, ‘Whichever of these two alternative solutions is preferred, one result remains the same: two sets of phonotactic conditions are deemed to obtain in the grammar, a more restrictive pre-syncope set, and a less resttictive post-syncope set. ‘At this point, it is reasonable to ask what justification there is for transforma tions which modify constituent structures established in lexical representation. If we follow the lead of much of the phonological literature in taking the validity of resyllabification for granted, then we are wedded to the notion that the con- straints which hold over lexical structures can be overturned during the course of derivation. We should be very wary of taking such a step, since some of the constraints at stake involve extremely robust cross-linguistic generalizations. Any concession to the idea that they might be revoked during derivation represents @ retreat from the strongest possible interpretation of this fact, namely that they are universal. For example, one widely invoked constituent transformatior to be considered below detaches a consonant from its original onset position and moves it into a preceding coda, as for instance in (c.ty) — (cit-y)* This resyllabification can only be achieved at the expense of overriding the Coda Licensing principle, in order to undo the onset maximization effect which lies behind the initial syllabification (city). ‘One empirical result of adopting resyllabification is a significant increase in the set of possible grammars defined by phonological theory. This proliferation stems not just from the prediction that some grammars may take up the resyllabification option while others may not but also from the possibility that the nature of the structural and phonotactic mismatch between underlying and resyllabified Licensing 187 representations varies from system to system. Moreover, acceptance of the notion that such discrepancies may exist between different levels of derivation allows for the possibility that later levels may actually enforce more stringent constraints on constituency than earlier levels. In practice, analyses which invoke the notion almost invariably conform to the pattern in which constraints arc less restrictive at later stages of derivation. This observed asymmetry is only derivable by ‘means of some additional mechanism. One such proposal is the Strong Domain Hypothesis, according to which all rules (including those which formalize syl- lable-structure constraints) are available from the outset of derivation and may be switched off at some later point; however, once switched off, they cannot be subsequently reactivated. Before considering the empirical consequences of resyllabification in any more detail, we should remind ourselves that the device forms part of a research hypothesis that has to be justified not just in relation to the data itis designed to account for but also in relation to any competing analyses that happen to be available. In the first instance, it should be weighed up against the null hypothesis ~ in this case the assumption that nothing happens to constituent structure during the course of a phonological derivation. Under the latter view, constraints on syllable structure remain stable throughout derivation. As with any research hypothesis, we should only accept the resyllabification model once it has been shown consistently to out-perform and not just match a model based on the null hypothesis. ‘At various points in the remainder of this chapter, we will consider whether the position that constituent structure is immutable can be maintained in the face of evidence involving a number of process-types for which resyllabification analyses hhave been proposed. At least as far as the syncope facts are concerned, the resyllabification account can be shown to come out second best. 4.6.4 Preserving prosodic structure Inherent in the resyllabification account of syncope is a distributional paradox that can be summed up as follows. According to phonotactic constraints that can be identified independently of the syncope context, one type of cluster supposedly generated by syncope submits to reanalysis as a branching anset. On the other hand, there is another type which resists resyllabification into a either a branching ‘onset or a coda-onset sequence without forcing a relaxation of the same con- straints. In other words, consonants flanking the syncope site are not subject to the stringent phonotactic constraints that independently can be shown to operate within intra- and inter-constituent domains. All that can be said about the melodic aspect of the syncope conditions is that the right-hand consonant must be a resonant. A conclusion that may reasonably be drawn from these observations is that consonants bordering a syncope site are bogus clusters, since there are no system- atic phonotactic dependencies between them. The lack of true adjacency that is188. Licensing suggested by this independence obviously stems from the fact that the consonants are separated by a vowel in lexical representation. But what reason is there to suspect that this lack of adjacency is not retained even when the vowel is syncopated? The simplest hypothesis that is consistent with this observation is that syncope hae no effect on constituent structure. That is, consonants flanking a vowel-syncope site continue to occupy independent onset positions after the process has occurred. This view is expressible in the second of the non-linear approaches to syncope alluded to briefly above. Under this account, itis only the vozalic melody of a syncopating vowel that is suppressed. The constituent structure is meanwhile preserved intact, as shown in the following representations of (def(i)nite): (sa) Syncope ONONONON ONONONON rrrrrred rrrrrri rriurrdigd braid defonat def nat ‘What remains after syncope is an empty nucleus. The non-adjacency of fand n, reflected in their abiding distributional independence, is accounced for by the fact that they continue to be separated by a nuclear position even after the latter has lost its melodie content. The same type of analysis can be extended to all cases of syncope. ‘To facilitate assessment of the proposal that constituent structare remains stable throughout derivation, let us consider how the view might be formally expressed in terms of phonological licensing. Resyllabification transformations have the effect of altering licensing relations established at the level of lexical repres- entation. These modifications can be viewed in terms of their impact not only on the structure of individual forms but also on the set of licensing conditions that hold of lexical representations in general. For example, from the first perspective, the deletion of the nuclear position shown in (48) results in a destruction of the relation of projection licensing which that position contracts with the preceding nuclear head. The resyllabification of the resulting stray position into a following onset, shown in (49a), creates a new relation of constituent licensing. Resyllabifi- cation into a coda, shown in (49b), creates a new relation of inter-constituent licensing. Viewed in terms of their effect on the set of constituent structures defined by the prosodic licensing conditions operating in English, neither of these transformations results in the creation of novel structures not encountered else where in lexical representation. This happens to be true of these particular cases; in 4.7.3 we will discuss a couple of analyses in which resyllabification does involve overriding pre-existing prosodic constraints. Nevertheless, the transformations shown in (49) do introduce one type of innovation: they generate melodic associations which fail to conform to the autosegmental licensing conditions holding at the lexical level. Many of the Licensing 189 clusters supposedly arising from the resyllabification of the syncopated forms in (50b) fail to respect the Complexity Condition, which we have seen to be strictly enforced in lexical representations. Take for example the alleged coda-onset sequence resulting from the restructuring of (cam(e)ea): (52) (a) (b) . R ' ONONON ON Oo oN rrrrdd rot tot rrrrd tobob tog kemoara ‘ke UR 1 2 I N In (S2b}, the downward complexity slope observable within the inter-constituent domain formed by the coda, occupied by m, and the following onset, occupied by +, constitutes a flagrant contravention of the Complexity Condition. The assump- tion inherent in resyllabification analyses is thus that licensing conditions, includ- ing those slating vo melodic complexity, can be overidden during the couse of rivation. The contrary view that phonological structure remains immutable throughout derivation amounts to the claim that licensing relations at derived levels of representation consistently match those established in lexical representation. In some respects, this parallels the situation in syntax and at LF, where the major properties of any level of representation are independently carried over from the lexicon. According to this requirement (known as the Projection Principle), for example, the lexically specified argument structure of a verb (which determines such properties as whether or not it takes a direct object) is respected identically atall levels of derivation.‘ The effect of this constraint is that syntactic movement ‘operations are structure-preserving; that is, they cannot alter the categorial status of a syntactic position whose presence is required in underlying structure. For example, a position projected as an noun phrase underlyingly cannot be turned into a verb phrase position during derivation. Extending the Structure Preservation Principle into the phonological domain has the following consequences for prosodic and autosegmental structure. If Phonological Licensing requires the presence of a skeletal position in lexical representation, it will also require its presence at derived levels of representation. One implication of this is that certain skeletal positions projected as particular constituent categories in lexical representation cannot be assigned to other ca~ tegories during derivation. For example, under Coda Licensing, a coda position requires the presence of a following licensing onset position. In lexical repres-190 Licensing entation, the syllabification of a VCV sequence is thus universally established as, V.CV. A transformation which reassigned the onset position to the coda of the preceding chyme, as proposed in a number of analyses to be discussed below, ‘would fail to be structure-preserving, since it would create a structure lacking the position required to prosodically license the coda. At the autosegmental level, structure preservation requires that the Complexity Condition be observed in both lexical and derived representation. “The phonological instantiation of the structure preservation principle may be stated as follows: (53) Structure Preservation Principle Licensing conditions holding of lexical representations also hold of derived representations. ‘The principle prevents phonological processes in a language from augmenting the set of prosodic templates and patterns of melodic association defined at the lexical level by universal principles and specific parametric setings. Phonological frameworks which permit resyllabifying translormations require structure preservation to be shut off at some point in dervation.©” That is, creating novel structures through resyllabification implies overriding particular licensing conditions that hold at lexical representation. Conceptually, this view is ‘out of tune with current assumptions in syntax. The alternative position adopted here is that phonological processes preserve prosodic structure throughout deriva- tion. This view, I will try to show, can be maintained in the face of evidence involving processes for which resyllabification analyses have been proposed. We will consider a number of well-known examples of such accounts later in this chapter. ‘As suggested by the analysis of syncope illustrated in (51), one of the keys to maintaining the stability view of prosodic structure lies in the concept of empty phonological categories. One of the points illustrated by this particular account is that empty nuclei are not restricted to the domain-final context where their function is to license a preceding onset. The empty nuclei for which we have just seen evidence occur stem-internally. If an internal empty position really is in some sense the same object as a parametrically licensed final empty position, we should expect the two of them to function identically for at least some phonological purposes. This expectation is in fact borne out. In one case to be discussed in 4.7.8, internal and final empty nuclei constitute a unified context before which tlenition operates in various dialects of English. In the Tonkawa case briefly mentioned abore, we find that the ‘two positions behave identically with respect to syncope. Recall that the second vowel of a word is suppressed in this language, as revealed in the alternation evident in forms such as picna-no? versus we-peena-no?. Now consider verb forms which contain a consonant-final prefix such as nes- (causative). In line with licensing principles, the domain-final consonant in the prefix occupies an onset position followed by an empty nucleus. As illustrated in (54), the latter posit Licensing 191 constitutes the second nucleus of a prefixed verb form and thus coincides with the syncope site. (54) 0 NON ONONON,,, ee ee) Ik x ox x ko x x x x x. trod rororrd 4g nes p i c e nm aw. The prediction is that syncope in this case should operate vacuously in the context of the second nucleus and thus should not affect the first nucleus of the stem itself, This is exactly what we find: nes-picena-no? as opposed to *nes-poena-n0?. 4.6.5. Proper Government Inherent in the acceptance of morpheme-internal empty nuclei is the danger that the concept might be abused through unconstrained deployment in phonological analyses. We might now ask whether there is any principled restriction on the occurrence of empty categories in phonology. The key 0 this question lies in an observation that can be made with respect to both the Tonkawa and English syncope cases: a vowel can only be suppressed if tis adjacent to a vowel which is itself not suppressed. This characteristic in fact extends to the vast majority of examples of the process reported in the literature. It is particularly striking in languages displaying an alternating succession of syncopated and non-syncopated vowels that is iterated across the word domain. __ The adjacency requirement suggests that the target and trigger vowels involved in syncope enter into a relation of licensing at some level of nuclear projection. This has led to the proposal of a constraint on the occurrence of empty nuclei which is formulated as a more restrictive case of projection licensing known as Proper Government. To be empty, a nucleus must either be parametrically sanctioned domain-finally (according to the parameter in (19)) or be properly governed by another nucleus. In the latter instance, Proper Government requires that the licensing nucleus must itself not be empty. In all reported cases, the Proper governor stands to the right of its proper governce. By way of illustration, consider the following representation of (fam(illy) (in which the arcow indicates Proper Government), fen (55) an —z ° —Ze Z moe 0192 Licensing The nuclear position Nz is empty, by virtue of its being properly governed by N, The fact that vowel syncope is obligatory in some languages, not present in others and optional in others (as in English) indicates that Proper Government is not strictly enforced in al systems. Vowel-uero alternations in English ate also evident in cetain 1oo-derived forms, although whether this phenomenon belongs to the general (fam(i)ly) type is questionable. According to one resyllabification account, the lexical repres- entation of roots involved in alternations such as (simple-simplicity, metre~ metric) and (couple-couplet) contains a CC cluster which is broken up by an ‘epenthetic vowel in word-final context. Hence kapl > kapal versus (coufpl]-et).?> Under this analysis, a form such as (couplet) contains the same foot structure as the syncopated variants of the forms in (50). Note, however, that the allegedly epenthetic vowel is obligatorily absent from root-derived forms; hence the im- possibility of *(coupfallet, metfa}ric), etc. Forms of this type are thus phonolo- ically indistinguishable from forms such as (poplar, petrol, patron) which have tho final CVC] alternants. (Poplar trees no more popple than patrons pater.) For the reasons given in 1.4.3, I assume that root-level alternants such as (simplicity, couplet, metric) are simply listed in the lexicon. On the other hand, word-level alternations can display exactly the same op- tional syncope pattern as that illustrated in (50): hence (puzz((a))-ing, batt(e)r- ing, butt(o)n-ing, fidd({o)ler), etc. The lexical representation of the morphological base in such cases, we may assume, contains an internal empty nucleus.” (56) (a) (fiddle) fidat (b) (fiddler) fidtae) roto oc oN OM ON ON ON ON re ee | rorertorer ot tora \ toi rod foud 1 fou d 1 3@ In (56a), Nz fails to be properly governed (indicated by //) by the empty position Ny since the latter is itself licensed (by parameter). Nz must therefore be phone- tically expressed, either as the latently present @ (yielding fidoi) or as a syllabic lateral (idl, through spreading from the following onset). (S6b), on the other hand, displays the same Proper Government configuration as (fam(i)ly) in (5). Nis filled (by the vowel of the suffix (-er)) and is thus able to license Nis the lateet thus remains phonetically unexpressed. Proper Government ensures that a syncopated vowel is always adjacent to an unsyncopated vowel. The constraint thereby derives the alternating . .. VOPCVCH. .. pattern that is characteristic of languages in which syncope iterates across the word. A syncope ‘lapse’ in which adjacent vowels are both suppressed, Licensing 193 for example in the sequence ... OPCUCV ..., is ruled out on the grounds that the second nucleus, being properly governed by the third, is unable to properly govern the first. One of the constraining effects of Proper Government is thus to debar adjacent empty nuclei from phonological representations. In response to one of the questions raised at the beginning of 4.6.1, we may now conclude that an ‘empty’ nuclear position is not really empty at all. Rather it contains latently present melodic material in the shape of @, together with which it forms a potential autosegmental licensing domain. Whether the licensing poten- tial is activated and thereby supplies an association between the position and the element depends on whether certain conditions are satisfied. In the case of syncope, this autosegmental licensing requirement is not met if the position is properly governed (as in an alternant such as (dePnite)). But under different conditions, when the position is not properly governed (as in the alternant (definitive)), ic is empowered to license melodic material. T have already mentioned the recognition of empty onsets as having set a phonological precedent for empty nuclei. It does not take much to think of analogues elsewhere in the grammar. Mention of the Projection Principle above suggests a parallel between empty nuclei in phonology and empty categories in syntax. Indeed, it has been proposed that the two are subject to the same universal constraint, the Empty Category Principle (ECP), one sub-clause of which requires that a category or position must continue to be licensed by its head, even when it hhas been vacated as a result of syntactic movement or the delinking of melodic material.” We might say that an empty nucleus is properly governed by its head nucleus in the same way that, say, the trace of a noun-phrase complement which has been subject to WH-Movement is governed by its verb head. We can develop this parallel further by picking up again on the theme that no systematic interactions will take place between superficially neighbouring posi- tions if an empty position intervenes. A well-known example of this phenomenon in English syntax involves the alternation between (want to) and its contracted form (wanna). The contraction is sensitive to whether or not an empty category intervenes between (want) and {to).” As illustrated in (57a), contraction is an option if the two forms are strictly adjacent. (57) (a) She’s {the woman]; I want to talk to [e]: She's the woman I wanna talk to. {b) She's {the woman]; I want [e]; to be president "She's the woman I wanna be president. Compare (57a) with the sentence in (57b), where (want) is separated from (to) by the trace of its complement (co-indexed with (the woman)); in this case, contrac- tion is blocked. The failure of contraction here, reflecting a lack of adjacency between the categories occupied by (want) and (to), is strongly reminiscent of the failure of phonotactic dependencies to manifest themselves between onsets that are separated by an empty nucleus.194 Licensing 4.7 Lenition 4.7.1 Introduction ‘The remainder of this chapter is devoted to the analysis of consonantal lenition, In 3.4.2, we considered the melodic effects of this type of process. Now we turn. our attention to the issues of where and why it occurs. This may seem an inordinate amount of space to devote to a single phenomenon. However, there is good reason to go into the matter in considerable detail. Our search for a non-arbitrary account of phonological processing has focused on establishing an intimate bond between the effects of a process and the context in which these make themselves felt. The first class of process to submit to analysis in these terms was that involving assimilation. The mechanism of spreading sets up a connection between the target and trigger of such processes in the most direct way possible ~ by having them associated to one and the same piece of melodic material. (Of the remaining process-types, a substantial proportion, pechaps the majority, fall under the rubric of lenition or weakening. Until recently, it had proved difficult to subject this class of phenomena to the same sort of non-acbitrary treatment as that successfully applied to assimilation. The particular challenge presented by Ienition processes is that they lack an immediately obvious local cause. Of course it has long been recognized that consonantal lenition, like vocalic reduction, characteristically takes place in ‘weak’ contexts. Nevertheless, the goal of formalizing this insight has turned out to be rather elusive. Informally, vowel reduction can be said to occur in metrically recessive nuclei, typically positions that are assigned weak stress or undergo harmony. The most-cited locations for consonantal lenition are codas, word-final contexts and footinternal inter- vocalic contexts. It would be desirable to be able to subsume this apparently disparate range of contexts under a single dimension of prosodic weakness. In the rest of this chapter, we will consider a number of analyses of tlenition in English which illustrate two fundamentally different ways of approaching this challenge. ‘We begin in the next section by reviewing the main contextual details of +lenition in different types of English. 4.7.2 Four lenition systems The data on which the discussion is based are drawn from four types of English, illustrating patterns of f realization which are distributed over relatively large geographical areas. As we will see, the conditions under whick tlenition mani- fests itself in these systems overlap to a significant extent. Nevertheless, they vary in interesting ways in a manner which, I hope to show, sheds light on the role played by autosegmental and prosodic licensing in conditioning phonological processes. Licensing 195 ‘The four illustrative systems are as follows: (58) Lenition systems glottalling (restricted distribution) B: glottalling (wide distribution) Cz tapping (unreleased stop in certain contexts) D: spirantization Systems A and B exemplify glottalling as it operates in many areas of England and Scotland. The phonological distribution of the glottalled reflex in System A is more restricted than that in B. For example, while both systems show lenited ¢ utterance-finally, as in (pif?)), only B shows it foot-internally, as in (pif?ly). A unidirectional implicational pattern reportedly exists in the geographical distribu- tion of these two systems: in some areas, both systems exist side by side (with various patterns of social distribution); but, while the occurrence of B invariably implies that A is also present, the converse is not necessarily true. Viewed historically, the occurrence in certain areas of System A to the exclusion of B suggests that the latter is a more recent development of the former.”* Cis a typical tapping system, showing a tapped reflex of ¢ foot-internally (e.g. Gilely)) and an unreleased, often pre-glottalized stop reflex in, for example, utterance final position (e.g. (pift")). The particular system from which the data given in this section are derived is one encountered in metropolitan New York. However, tapping is firmly established all over the United States, as well as in Canada, Australia, and some areas of England and Ireland. System D displays the effects of spirantization to s, a process most charac- teristically associated with parts of southern Ireland and Merseyside in England. In the Irish English data supplied below, the spirantized reflex occurs both foot-internally and word-finally. Lenition of ris by no means an isolated phenomenon in English. In terms of both realization and distribution, it is related to several other processes which con- siderations of space preclude us from discussing in detail here. Foot-internal tapping, for example, extends to d, with the result that the #d conteast is potentially neutralized in this context in many dialects (so that (ladder) = (latter), for instance).”* The distribution of pre-glottalized stop reflexes holds not just of ¢ but of all voiceless plosives.” In some glottalling dialects, glottalling affects not only but also pand k, albeit under a more restricted set of conditions.”* And, as. we saw in 3.4.2, spirantization can extend to other plosives and, in the case of f, can proceed as far as debuccalization to h. In some areas where ¢-weakening is less firmly entrenched than in others, lenited reflexes typically alternate with unlenited reflexes under different social condi- tions. This variability will not concern us in the following discussion. Rather the focus will be on the phonological conditions under which the appearance of a lenited reflex is grammatical (even if optional) in each of the four illustrative systems.196 Licensing have selected the four systems described here on the basis of the way they point up differences in the conditioning of tlenition by two main factors - the presence of a following licensed nucleus, and the immediate segmental coatext. The first of these relates to the occurrence of ¢ inter-vocalically (as in (pity)) and domain- finally (as in (pity)), the second to the effect of surrounding consonants (a3 in (mister) and (petrol). A further conditioning factor, which we will not discuss, involves an inter-vocalic context in which the two nuclei occur in separate feet within the same word.” In some dialects, ¢ can be lenited here if the second nucleus bears secondary stress, as in (détal, lét2x). This also applies to bipedal forms whose primary stress shifts according to the rhythmic context within the phrase. Forms of this shape which contain a lenition-prone ¢ include (thirteen, fourteen, eighteen). (To see the effects of stress shift, compare (éightéen women) with (ightéen bandnas).") The manner in which this last dimension influences Henition cuts across the four-way classification of the systems to be discussed here. ‘As in previous chapters, the data are presented in word sets, each of which illustrates a particular configuration of phonological and morphological condi- tions. To start with, we will confine our attention to segmental contexts in which tis either inter-vocalic (as in (pity)) or post-vocalic and word-fnal (as in (pit). We will consider the conditioning effects of neighbouring conscnants later. ‘We begin our comparison by noting one context in which ¢ is resistant to lenition in all four systems. As shown in (59), an unlenited plosive reflex occurs in a foot-initial onset, both word-initially (59a) and word:nternally (59). (Where a word contains more than one orthographic (1), the emboldened letter, corresponds to the t which appears in the illustrated context.) (59) ABCD (a) tot tt time,tin, trade ib) t tt € pretend, boutique, politician, imitation Inter-vocalically within a morpheme-internal foot, System B shows a glottal reflex, Ca tap and Da spirant, while A retains an unlenited reflex. This pattern holds regardless of whether the dominant nucleus of the foot has primary (60a) or secondary word-stress (60). (60) ABCD @t 2 6 s )t 2 © s pretty, Peter, water, automatic photographic, automatic As shown by the forms in (61), the same pattern manifests itself in an onset flanked by the two weak nuclei of a super-foot: (1) A BOC OD to? es sanity, parity, political Licensing 197 As discussed in 4.6.2, this configuration contains two metrical domains: the dominant nucleus is bracketed with the following nucleus on the foot projection and with the third nucleus on the super-foot projection. In a form such as (competitor), both are susceptible to lenition in Systems B, C and D, the first by yistuc of being foorincennaly the second by virtue of being intemal to a supe Word-finally before a consonant or a pause (indicated in (62) by IA joins B and D in showing a weakened reflex. It isin this context that the cherwiserap ping System C has an unreleased stop. 7 (2) ABCD (a)? 2? © s — fetl, putl, tight! (b) ? 2% © 5s letme, put by, light rain ‘Word-finally before a vowel, however, C again displays the tap: (a) let a, put it, light again, (b) Pacer (3) A t t let 6ff, put 6ver, light ép BCD Pes Res Here, B and D also have lenited reflexes. Note that weakening occurs in this context irrespective of whether the following vowel is unstressed (63a) or stressed (63b). This means that, in System C, tapping before a stressed vowel may take place across a word boundary but not word-internally; compare the C data in _ (63b) and (596). Based on the data presented so far, the main diffe ha : > the main differences among the four systems aa respect 10 the distribution of lenited ¢ can be summarized informally as (64) (b) Word-final before V A (a) Word-final before Corl? (c) Foor-internal t wae naga eueg ‘There are three main aspects of the data that any account has to get to grips with: (a) the unified behaviour ofthe domain-final and footnternal conve (0) thc relation between the tap and unreleased-stop alternants in System C; and (c) the apparent sensitivity to morpho-syntactc structure of the latter alternation, as wel as of the alternation between fand ?in System A. Note that the word-final context in (64a) contains the conjunction consonant or Pause. In linear approaches, a conjunction such as this was diectly incorporated into a rule's environment by means of brace (‘curly-bracket’) notation: (Cy l}!2 Couched in these terms, the gloalling rule fr System A looks something ike this:19 te can fq Brace notation was originally introduced into the SPE-type framework to allow rules with overlapping structural descriptions to be collapsed. In (65), there are two sub-rules, one applying before C, the other before |. The notation is be interpreted in such a way that both subparts apply (a convention known as conjunctive rule ordering). The device was an early target for criticism of the rewrite-rule format, on the grounds that it vastly increases the expressive power of the model, to the extent that any grammar the model generates is in all likelihood unlearnable.* In principle, any two or more feature matrices or bound- ary symbols can be unified by means of the curly-bracket device. With this comes the prediction that, even if we restrict ourselves to combinations of two units, as in rule (65), each combination has an equal chance of being instantiated in rules characterizing processes in the world’s languages. This is simply not correct. The vast majority of the expressible conjunctions are not attested in any natural language. And expanding the set of conjoined environments beyond two produces an exponential increase in the number of expressible permutations, again well nigh all of which correspond to nothing in the empirical record. In contrast, the specific combination illustrated in (64a) is one that recurs with disturbing regularity in linear analyses of different languages and language families, although this fact remains no more than accidental within a theory which incorporates brace notation. This reflects a more general failure to evaluate a very small class of attested processes any differently from a vast array of unattested types. For some time now, it has been recognized that the (C, !} conjunction is an inadequate circumlocution referring to a single context that can only be satisfac- torily characterized in terms of constituent structure. Indeed it was partly in respons¢ to this observation that linear representations were abandoned in favour of syllabically structured representations in the first place. 4.7.3 Coda analyses ‘We will now compare three constituent-based analyses of tlenition, the first two of which were initially formulated exclusively to account for the process as it ‘occurs in System-C dialects. Both of these illustrate a more gereral approach to consonantal weakening, one in which the coda is identified as the prime weak context. The main focus here will be on the formal machinery these analyses call on, specifically resyllabification, rule ordering and, in one case, ambisyllabicity. ‘An ambisyllabic consonant is one that belongs simultaneously to a coda and a following onset. According to this type of approach, domain-final rin a form such as (pit) is syllabified in a coda. A further assumption is that, in line with onset maximization, a foot-internal fin a form such as (pity) is attacked to an onset at the level of basic syllabification; subsequently, however, the consonant undergoes Licensing 199 resyllabification into the preceding rhyme, an operation we may refer to as Coda Capture. Under one such analysis, the rule by which a foot-internal consonant is captured into the coda ofthe dominant syllable on the left canbe informally paraphrased as follows: (66) Coda Capture Within a foot, attach the consonant associated with the onset of the unstressed syllable to the coda of the stressed syllable. Note that the way this operation is formulated leaves the initially established onset attachment undisturbed. The result is a foot-internal ambisyllabic conso- rant, as in the £ of (pit (6) OR OR Pine gt pou i A further resyllabification rule that has a bearing on élenition affects a word- final consonant in context (64b), as in (get a). Again paraphrasing, we may formulate this operation as follows: ¢ (68) Onset Capture Attach a word-final coda consonant to an unoccupied onset at the beginning of a following word. As with (66), this rule creates ambisyllabic consonants, such as in ft} al (ater nicates the smetaneous acupevon of cole and on one Mol that Coda Capture in (66) is stress-sensitive: ¢ is ambisyllabic in ({piley}}, where it is foot-internal, but not in ({pre}{ténd]), where it is foot-initial. Onset Capture in (68), on the other hand, is insensitive to stress: ¢ is ambisyllabic in both (gé(t} a)) and (gett) tf) ‘Together, the two resyllabification rules set the scene for lenition, itself formulated ‘as two sub-rules, one of which generates a tap, the other an unreleased stop:"” (69) Lenition @ cor bb) toe if ambisyllabie if not syllable-initial ‘The derivation of the three forms in (70) illustrates the workings of this analysis (skeletal tier suppressed):** (70) (ity) Get) (get a) OROR OR OR R read ray rind prti ger get oe200. Licensing CodaCapture OR OR ting t = _ pai Onset Capture OR OR - bins t ger 2 Lenition (69) OR R OR OR oR PIV A mN Pivet pai oer ger io ‘The following devices play a crucial role in this analysis: resyllabification, ambisyllabicity and rule ordering. In the case of Coda Capture and Lenition (69a), the ordering is intrinsic; that is, the latter cannot logically become operat- ive until after the former has applied. However, in the case of Onset Capture and Lenition (69b), the ordering must be extrinsically imposed. These two rules could logically apply in either order. But were the operation of Lenition (69) to precede that of Onset Capture, the fof a form such as (get off) could only be realized as, an unreleased stop rather than as the attested tap. Thus, a System-C grammar has to contain a stipulation to the effect that Onset Capture must be ordered before Lenition (69b). Technically, the former rule bleeds the latter. ‘The device of ambisyllabicity has not been widely accepted in phonological theory.” The reason has to do with its anomalous status with respect to the bracketing of units in constituent structure. An otherwise uniform property of constituent structure in both phonology and syntax is that of proper bracketing, according to which no unit in a string may be simultaneously bracketed within ‘two (or more) independent constituents. No word can simultaneously belong to, say, a verb phrase and an independent noun phrase (i.e. one that is not properly included in the former). To countenance the improper bracketing that is inherent in ambisyllabicity is to lose sight of this uniformity. ‘A second type of coda analysis dispenses with the need for ambisyllabicity However, it retains the notion that the process operates in codas and thus also the need for resyllabification. The latter is reduced to a single operation in which the attachment of a foot-internal consonant to a preceding coda is accompanied by its ‘detachment from its original onset position.” This may be informally stated as (71). (71) Coda Capture IL Within a foot, detach the consonant from the onset of the unstressed syllable and attach it to the coda of the stressed syllable. ‘As with the first analysis, resyllabification prepares the ground for tlenition. Jn order to characterize the relation between the tap and unreleased-stop reflexes of in a type-C system, the following two-stage analysis of lenition, incorporating the binary feature [release], has been proposed. First, a rule assigns values for the feature to t on the basis of the immediately following context. Specifically, if tis followed by a consonant or pause, it acquires a minus values otherwise itis plus:"? Licensing 201 (72) Release A syllable-final oral stop is (a) [-telease] before a consonant or a pause; and (b) [erelease] elsewhere. A syllable-final ¢ will thus be [-release] in forms such as (get by) and (get §) (under (72a)) but [+release] in (get a) (under (72b)). A [-release] rsubsequently yields 4. A [trelease] f occurring in an onset is realized as a plosive (with aspiration if certain other conditions are met). In a coda, however, [+release] 1 is subject to tapping. The two lenited reflexes are derived by means of the following rules:?? (73) Lenition tt @ of [+release] + r in coda ot [-release] = The following derivations illustrate how these rules operate ([+ release] abbre- viated to [+ r)): (74) (pity) (get f) (get a) OROR OR OR R rrr’ tay riot prti get ger OR R riyit 5 - Coda Capture pot ti OR R OR OR R PIV rota tro Release(72) pr ti get get [er] EA fer) OR R OR R rivod rivit Lenition (73a) Pot or i ger OR IN Lenition II (73b) ger puti get ero202. Licensing ‘The main advantage this analysis enjoys over the first is its ability to account for the lenition facts without recourse to ambisyllabicity. It is also on the face of it simpler in that it dispenses with Onset Capture (68) (although, es we will see ina moment, this apparent saving may be illusory). Moreover, it dispenses with the need for extrinsic rule ordering. The tapping cule (73b) is intrinsically or- dered after Coda Capture II (71); both lenition rules in (73) are intrinsically ordered after Release (72). However, these savings do not come without costs being incurred elsewhere in the analysis. One price to be paid is the introduction of the ad hoc feature [release], for which there is litele or no phonological motivation beyond this particular account, at least within an orthodox feature framework. Moreover, the sub-rule in (72a), which assigns (-release] to t, retains the problematical disjunc- tion (C, I} that marred carlier linear analyses. ‘Taking stock of the formal devices just reviewed, we may note that each of the coda-based accounts invokes two or more of the following: extrinsic rule ordering, an ad hoc feature, ambisyllabicity and resyllabification. Each of these devices in its own way loosens the restrictiveness of the theory within which the analysis is framed. Resyllabification is incompatible with the most restrictive interpretation of the Structure Preservation Principle (53), according to which well-formedness conditions on prosodic structure remain in force throughout derivation. Coda Capture Il (71), for example, creates a VC.V structure in which a coda position is unlicensed (in violation of Coda Licensing (16b) ). Underlying particular coda analyses of lenition is a more general assumption that melodic restrictions on domain-final consonants closely match or duplicate ‘those operating in domain-internal codas. If this were true, it would provide some support for the view that both contexts are codas. It certainly is the case that there can be distributional overlap between the two positions, which co-occur in many classic examples of lenition and defective distribution.” However, this evidence cannot be considered sufficient to clinch the case for the coda assignment of final consonants. Even if we set aside the theoretical reasons we now have for rejecting this analysis, itis flatly contradicted by the substantial body of other empirical evidence reviewed in 2.4.4. In any case, the distributional relationship between internal codas and final consonants is by no means as close as is often supposed. The evidence discussed in 2.4.4 shows that, in this respect, the two contexts are in fact quite different in English. Moreover, some of the best-known examples from other languages which supposedly demonstrate the relationship turn out, under close inspection, to be rather less than convincing.”* ‘There is in fact a more fundamental flaw in the coda account of lenition: while manages to shoe-horn the set of weak consonantal sites into a single constituent ‘context, it provides no reason for why it should be the coda rather than any other syllabic context that particularly favours such processes. According to the theory within which the coda approach is formulated, lenition rules such as those in (69) Licensing 203 and (73) would be no more highly valued than rules that weakened a consonant in, say, an onset or in the onset of a stressed syllable. Neither of the latter types coincides with the empirical record, at least not as an exhaustive characterization of a weakening context. 4.7.4 Licensing analysis The question to be taken up in this and the following sections is whether an approach which has no recourse to resyllabification or indeed to any of the other machinery mentioned in the last section is up to the task of characterizing the same set of facts. I hope to show that a valid analysis of this type is available within a framework which incorporates independent principles of pho- nological licensing. Moreover, these same principles provide an answer not only to the question of why weakening occurs where it does but also to the even more fundamental question of why there exist distributional asymmetries atall. ‘We have already seen one respect in which an element-based approach is better equipped than one based on features (let alone ad hoc features) to represent weakening processes such as ¢lenition in a constrained fashion (see 3.4.2). With binary features, lenition has to be expressed as the arbitrary substitution of one feature value by another. With univalent elements, on the other hand, all weakening processes are directly characterized as the suppression of melodic material that is present in lexical representation. The question now is whether the licensing framework outlined in this chapter is capable of capturing the condi- tions under which melodic decomposition takes place. I will try to show that an account of this type is indeed possible, not just for the System-C facts we have been concentrating on up to this point but for the other three illustrative systems. as well. Let us start by considering how to account for the selective conditioning of tlenition by context (64a) (as in (get by)) versus (64b) ((get a)). According to the first of the coda analyses above, this difference is attributable to the availability or otherwise of an onset position following the target segment. Specifically, a vacant onset is made available if, as in (64b), a vowel-initial word follows but not, as in (64a), if the target segment is succeeded by a consonant-initial word {in which the onset is already occupied) or a pause. The former case illustrates a recurring configuration in languages in which a word-final consonant is syllabified with following word-initial vowel. Under a coda-based account, this effect is only achievable through resyllabification; specifically, an under lying word-final coda consonant is captured into a following onset if one becomes available during derivation. This aspect of the analysis is simply untrans- latable into the licensing framework presented in this chapter. For one thing, the resyllabification operation is ruled out if Structure Preservation is assumed to persist throughout derivation. Moreover, we have seen sufficient evidence to204 Licensing support the view that a word-final consonant already occupies an onset position in any case. If we assume that t in forms such as (get by) and (get a) in both instances ‘occupies an onset slot, we must conclude that the distinction between contexts (64a) and (64b) cesides in the nature of the following nucleus, In (get by), the position in question (a word-final parametrically licensed enpty nucleus) is succeeded by an onset position: 75) ON ON ON rrr rt rx x x x & x x trot irl ger bay (get a), on the other hand, lacks a following onset position. Te notion that the word-final consonant is syllabified with the following word-initial vowel can be implemented by assuming that an empty nucleus deletes next to a filled nucleus ‘when no onset position intervenes: (76) (a) O, Ny, O; Nr Ny rrr (b) O; Ni Q Ns vot t4 ro get ge The operation in (76) might be judged inadmissible under an extreme reading of the Structure Preservation Principle, since it results in the replacement of one specific licensing relation (that holding between Nz and Os in |76a)) by another (between Ny and O> in (76b)). It is, however, consistent with an interpretation under which general conditions on licensing remain in force throughout derivation. The relevant condition in (76) is the requirement (Onset Licensing (16a)) that Or be licensed by some nucleus, whatever the particular identity of that nucleus might be. To say that the onset is licensed ty one nucleus in one domain and by a different nucleus in another implies no amendment to the set of prosodic templates defined over lexical representation by universal prin- ciples and parametric settings. The suppression of an empty position in (76) certainly does not entail the type of resyllabification whereby a position is reassigned from one constituent category to another, for example from a coda to ‘The operation whereby 2 nucleus deletes next to another nucleus is not just ‘motivated to achieve the cross-word V.C] [V syllabification effect shown in (76)- Ik can be viewed as part of a wider phenomenon, alluded to in 4.3.3, in which sequences of adjacent nuclei are disfavoured. Cross-linguisticaly, this manifests itself as a condition under which intra-morphemic sequences of filled nuclei end Licensing 205 to be absent from lexical representation. It also manifests itself in processes which serve to disengage potential nuclear sequences resulting from morphological concatenation. A process of this type demonstrates one of two strategies. One involves breaking up the sequence by supplying the intervening onset with melodie content. This frequently takes the form of a hiatus glide consisting of an element spread from one of the flanking nuclei. In some dialects of English, for example, a w glide appears in such contexts when the first vowel is round, as in {too (w]early). In this instance, the spreading element is U.” The other strategy for avoiding cross-word nuclear sequences is to suppress one of the nuclei. This is not widely employed in English (although it is evident in a handful of lexicalized contractions, such as (I'm, you're)). However, it is firmly entrenched in many other languages. Witness, for example, the French apocope case briefly discussed in 4.6.1, with forms such as (('ami) (from (le/la ami) ‘the friend’), or the similar Yoruba pattern illustrated by forms such as gba ers — gber5 (‘receive meat’).”* The nucleus-suppressing strategy is the one that is formally related to the oper- ation shown in (76). ‘The avoidance of consecutive nuclei is strongly reminiscent of the avoidance of consecutive identical melodic expressions that is embodied in the OCP (36). Indeed, it has been proposed that both phenomena are reflections of a more general version of the convention which applies not only to melodic expressions but also to constituents together with their positions.”” 4.7.5. Licensing Inheritance ‘The next question is how to unify the word-final pre-vocalic context (64b) with the foot-internal context (64c). Both sites correspond to onsets, but on the face of it the licensing conditions in these lenition-favouring contexts are no different from those obtaining in other onset contexts in which lenition is disfavoured. For example, a foot-internal onset (occupied by ¢ in, say, (letter)) is inter- constituent-licensed by a following nucleus in the same way that a foot-initial onset is (as in (time)). And the latter is precisely one of the contexts that is typically resistant to lenition. What turns out to be important, however, is that this identity only holds of licensing conditions obtaining in the immediate vicinity of the positions in question. Panning out to higher levels of projection, we discover that indirect licensing relations which contribute to the integra- tion of foot-initial and footinternal onsets into the prosodic hierarchy are significantly different. In particular, the licensing nucleus is itself unlicensed on the foot projection in the first instance (the e in (letter)) but licensed in the second (the 2{r) in (letter)). This distinction, I will now try to show, has a crucial jes of different onset positions to support melodic ‘Consonantal lenition, viewed as melodic decomposition, can be taken as evid- cence of a position’s diminished ability to support melodic content, Asymmetries206 Licensing in the capacity of different onset positions to sustain particular levels of melodic complexity are strongly reminiscent of the distributional inequalities obtaining within the governing domains in (29) (branching onsets, branching nuclei and coda-onset clusters). The Complexity Condition (32) is essentially a statement abour che aucosegmental licensing power of different positions within this type of domain, where greater power implies a greater toleration of melodic complexity. ‘What the condition says is that a governed position can never have a greater capacity to license melodic material than its governor. By means of this con- straint, we correctly derive the complexity gradients that are ebservable in these contexts. As it stands, however, the Condition has nothing to say about positions which, although not standing in a relation of direct licensing, nevertheless also display systematic discrepancies in melodic complexity. This is true of the two types of onsets that figure in the differential conditioning of lenition. The two sites in question, foot-internal and word-final, involve positions where the tend- ency towards low melodic complexity cannot be directly attributed to their status as licensees. Rather, as pointed out in the last paragraph, theit weakness celative 10 foot-initial onsets seems to stem indirectly from relations involving the follow- ing nuclei that license them, ‘These observations suggest an intimate relationship between the prosodic and autosegmental aspects of licensing and further justify subsuming them under the single principle of Phonological Licensing in (8). A question worth pursuing is whether this relation is responsible for all asymmetries in melodic complexity, not just those involving direct licensing. This would mean thet the Complexity Condition is reducible to some more fundamental principle which regulates the connection between the two mechanisms. The fundamental insight thar initiates this line of enquiry is that the capacity of a position to suppor: melodic contrasts is determined in large part by its place in the prosodic hierarchy. To develop this idea, it will be convenient to be able to speak of a position’s autosegmental licensing potential. This may be understood as the ability of a position either to directly license melodic content or to confer autosegmental licensing potential on another position. Viewed in these te:ms, distributional asymmetries reflect the fact that a prosodically unlicensed position has a greater degree of autosegmental licensing potential at its disposal than its licensee. ‘Two assumptions will help give this notion formal substance. First, as stated in (77), licensed position acquires its ability ro license melodic material from its licensors!" (77) Licensing Inheritance Principle ‘A. prosodically licensed position inherits its autosegmental licensing potential from its licensor. Second, the stock of autosegmental licensing potential invested in an unlicensed position is finite and is dissipated through transmission to licensed positions. This is illustrated in the branching onset structure shown in (78) (where & stands for some melodic expression defining a consonant). Licensing 207 (78) (a) Direct autosegmental (b) Indirect autosegmental licensing licensing 1 tm x10 fa x10 i cr w The immediate source of the autosegmental licensing potential available within the onsets the constituent head x. (As we will se presently, x itself inherits this potential from outside the onset, specifically from the following nucleat head by Which itis licensed.) Within the onset, as shown in (78a), the autosegmental lzesing of the melodic expresono by xs diet On the other han ae shown in (78b), the licensing of cs by xis indirect, since the position’ licensing potential in 78) he esing of by 5nd since the positon’ s ens pon A licensed position can be thought of as a sort of resistor which attenuates the autosegmental licensing charge delivered by its licensor. This notion, together with the Licensing Inheritance Principle, allows us to derive the reduced distribu tional leeway of a licensed position vis-d-vis that of its licensor. The defective distributional characteristic of the right-hand slot of a branching onset is thus a reflection of the fat that its autosegmental licensing potential is diluted asa result of being inherited from another position, namely it licensor on the let Al ofthe distributional inequalities derived by means of the Complexity Con- dition may now be viewed as instantiations of the more general principle of Licensing Inheritance. The downward complexity gradient enforced within branching onsets and favoured within branching nuclei can now be understood as reflecting the fact thatthe right-hand position in both eases acquires its autoseg. mental licensing potential from its constituent-governor on the left. Similarly the upward complexity slope observable in coda-onset clusters results from the directionality of inter-constituent government: the coda inherits its autosegmental licensing potential from the following onset governor. In fact, the autosegmental licensing power transmitted to the coda by the following onset is doubly diminished. As shown in (79), the onset x2 which governs coda x; is itself licensed by the following nuclear head 10, (79) Codac RR \ ox Yor Xo % tit This means that the autosegmental licensing potential of the coda is inherited at two removes fom it source; that is its depleted at wo stages ona licensing path.208 Licensing Before extending this insight to the other lenition sites, it is worth making brief excursus at this juncture to consider the relevance of the point just made to the distributional peculiarities of codas discussed in chapter 2. We are now in a better position to explain the fact that the coda of a heavy (two-position) chyme hhas rather more distributional leeway than that in a super-heavy (dce-position) rhyme. For example, as noted in 2.4.4, the former, unlike the latter, can support aan oral stop; hence single-domain forms such as (factor, apt) but none on the lines of * féktor or * eypt. Or another example: a sonorant occupying the coda of a super-heavy rhyme must be place-linked to the following onset, as in (mountain, shoulder). In the case of a heavy chyme, on the other hand, a coda sonorant can bear an independent place element. Thus we find forms such as (dolphin, bulk) but none of the shape * duifn or * bowik. Distributional discrepancies of this sort, we may surmise, are attributable to the relational distinction between the two types of coda pointed out in 4.3.5. As shown in (23a), the coda of a heavy rhyme is simultaneously licensed by the preceding nuclear head and by its governing onset. From the viewpoint of Licens- ing Inheritance, this means that it receives two injections of autosegmental licensing power. Compare this with the single transfusion rectived by the coda of a super-heavy rhyme, which is licensed only by the following onset (see (23b) }. ‘The second-hand quality of the licensing potential handed down by the governing ‘onset in (79) means that neither type of coda makes a particulaely strong autoseg- mental licensor. But the additional source of power available to the doubly licensed coda leaves it rather more distributional room to manoeuvre. “The two-stage nature of the licensing configuration in (79) provides the key to unifying the foot-internal and word-final sites where consorantal lenition takes place. The distributional disparity between foot-initial and fcot-internal onsets is related to the different licensing potentials of the following nuclei. On the foot projection, the dominant nucleus is identified as a powerful autosegmental licen sor by virtue of the fact that itis unlicensed within its domain. The licensed status of a recessive nucleus, on the other hand, marks it out as possessing correspond- ingly less autosegmental licensing potential. Hence the recurrent pattern whereby a maximal inventory of vocalic contrasts is to be found in dominant nucle, whereas recessive nuclei frequently display reduced inventories. The phenomenon of vowel reduction, typical of recessive nuclei, is expressible as the suppression of tlements when the autosegmental licensing power of the position to which they would otherwise be attached, inherited from the dominant nacleus, is insufficient to ensure their interpretation." ‘The distributional asymmetry between the nuclei of a foct is potentially mir- rored in the onsets they license at the inter-constituent level. Just as the distribu" tional leeway of the dominant nucleus is less'tightly restricted than that of its recessive sister, so the onset licensed by the former enjoys a greater distinctive potential than the onset licensed by the latter. Hence the wid:spread tendency for contrasts holding in foot-initial onsets to be neutralized foot-internally. This imbalance can be seen to follow from the Licensing Inheritance Principle, if we ‘compare the different licensing paths involved in the two contexts, illustrated by’ Licensing 209 the ¢in (cunny) (80a) and that in (pity) (80b). given on the right, the relevant licensing detai In the fragments of representations are highlighted. (80) (a) Foot-initial C ONON tri — xx x x JO [XIN [x]O. [xJN rid 1 (b) Foot-internal C ONON trad c fl i i ii fx]O [XN Ee f1N piti t From (80a) it can be seen that the source of a foot-initial onset’s autosegmental licensing potential occurs at one remove from its prosodic licensor, the dominant nucleus of the foot. The diminished autosegmental licensing power of a foot-in- ternal onset, by contrast, stems from the fact that it is inherited at two removes, as shown in (80b). ‘The word-final context can be identified as a weak autosegmental licensing site for similar reasons to those holding foot-internally. The onset position occupied by a final consonant is licensed by a following empty nucleus which is itself licensed, this time parametrically. This is illustrated in (81) for the rin (get) (again with the relevant licensing relations featured on the right). (81) Domain-final C Final-empty nucleus parameter ON oN 0 gN + trou t 1 RoR oR x IO IN (0 TIN oe t t By invoking Licensing Inheritance, we have succeeded in unifying the three classic consonantal weakening contets ~ coda, foorintenal, and domait-fnal, ‘As depicted in (79), (80b) and (81), each involves a position whose autosegmental licensing potential is depleted at two stages on a licensing path. In short, the diminished ability of all three sites to support melodic contrasts stems from th relatively lowly status within the prosodic hierarchy. Under such circumstances, [enition results from the inability of a position to issue an autosegmental licence to particular elements. Of course not all three contexts necessarily figure simultaneously as conditions on particular weakening processes. The differences between the foot-internal and210 Licensing domain-final sites, for instance, can be captured by referring to the status of the following licensed nucleus. A process that only occurs foot-internally is condi- tioned by the presence of a following nucleus that is licensed on the foot projec. tion, as in (80a). A process that only occurs word-finally is triggered by the presence of a parametrically liveused empty nucleus, as in (81). For a process occurring in both contexts, we need only specify that the following nucleus is licensed. To illustrate these different conditions, let us briefly consider three ‘weakening processes other than ¢lenition. ‘An example of a lenition process that is confined to fcot-internal contexts is, provided by the suppression of 8 that optionally occurs in some dialects (for ‘example, those spoken in the north of Ireland). As a result of this weakening, forms such as (mother, father, other, together, weather) may lack the medial 6 ‘which appears in most other dialects. Hence a pronunciation such as faar (father), {In Dutch, a similar process has affected cognate d in the same position; cf, English (brother) and Dutch (broer).) Formally, this process can be expressed as the failure of a nucleus that is licensed on the foot projecticn to transmit sufficient autosegmental licensing power to enable the onset to support the relevant ele- ments (head R, dependent hi ‘The widespread process of obstruent devoicing provides an example of weaken- ing that occurs word-finally. Examples can be cited from a host of languages, including some of English's close West Germanic relatives such as Dutch and German. The version of the process that was once partially active in Old English is now more or less extinct in the modern language. Final devoicing has, however, developed independently in a few present-day dialects. It is found, for example, in certain types of African American English, where forms such as (bit = bid), (bifk] = big), (lilft} = lived) are attested.'®* (While the distinction between the two series of obstruents is neutralized in this case, the original lexical contrast is preserved in the length of the preceding vowel, as in brat (bid) versus bu (bit), for example.) Devoicing consists in the suppression of the slack-vocal-cords element, an event that reflects the diminished autosegmental licensing power of a domain-final ‘empty nucleus." The distributional properties of hin most dialects of Engish illustrate a weaken- ing phenomenon that occurs both foot-internally and wored-finally. The sound can appear in foot-initial onsets, both word-initially (as in (822) ) and word-internally (a3 in (82b) ). (82) (a) {hlit, fhJum, (hot (b) _befhjalf, be(hJind, apprefhend (¢) vethjicular —vétiicle profhjibie profflibition ‘As shown in (82c), however, f is excluded from foot-internal onsets, a patter 3 which manifests itself in root-level alternations such as (ve(h}icular) (foot-initial) versus (vélicle) (foot-internal). Furthermore, the sound is never found domait- finally in such dialects. (There are thus no words such as * reeh.) According to on€ Licensing 211 approach, the unification of the two weakening contexts in this case is achieved by identifying both with coda position. This relies on the now-discredited as- sumptions that word-final consonants occupy a coda and that foor-internal onset consonants become resyllabified into a preceding coda." The Licensing Inherit- ance proposal enables us to dispense with the latter device. All chat has to be eaid in this case is that a licensed nucleus fails to bestow on a preceding onset the necessary capacity to license a lone h element. 4.7.6 Cyclic effects ‘We are now in a position to assess how the licensing proposal can be applied to the analysis of ¢lenition contexts in our four illustrative dialects of English." In tandem, the licensing configurations in (80b) and (81) define the conditions under which the various Flenition processes operate: (3) NON rr where x, is licensed either x 7 bad (a) by x1, or Fe (b) parametrically Condition (83a) defines the foot-internal site represented by (pity) in (84a)s, condition (83b) defines the word-final site in (pit) (84b). (84) (a) {b) Parameter (19) ON or J ONON ONON rood Horo k x x x k x x x} frog Had poe pit Informally stated, the two glottalling systems, A and B, differ minimally with respect to whether lenition occurs pre-vocalically or not (see (64) ). I have included these two systems in the comparative exercise for one express purpose — to show how the difference between them sheds light on the nature of the ‘operation whereby the melodic content of a position is suppressed. ‘Although suppression of a melodic expression implies a withdrawal of its ‘autosegmental licence, note that this does not necessarily imply that itis deleted. In principle, we can conceive of the delinking and deletion of melodic material as two independent operations. The simplest view would of course be that either but not both of these operations is a necessary part of the theory, or perhaps that deletion is no more than an automatic consequence of delinking. However, the conclusion that many phonologists have reached on the basis of the available evidence is that delinking and deletion are independent operations with different212. Licensing consequences. More accurately, delinking does not necessarily entail immediate deletion, although it is usually assumed that any melodic material not associated toa skeletal point at the end of a derivation is erased at that stage (one aspect of an operation known as Stray Erasure).!”” ‘One type of evidence supporting this view has to do with the manner in which phonological processes interact with morpho-syntactic structure. Let us remind ‘ourselves of the assumption, outlined in 1.4.3, that processes occur cyclically: thar is, their conditions are repeatedly checked through a succession of morpho-syn- tactic domains which are visible to the phonology. A given process which results in the suppression of melodic material is going to have quite different cyclic effects according to whether it is cast in terms of deletion of delinking without deletion. Any material that is deleted on a particular cycle will not be recoverable ‘on a later cycle (one of the consequences of a principle known as the Strict Cycle Condition).‘% On the other hand, if melodic material is delinked but not deleted ona given cycle, it is potentially recoverable, should differert licensing conditions become available on a subsequent cycle. That at least is the prediction which follows from the independence view of delinking and deletion; and it appears to be by and large correct, as we will now see.! Let us now try to specify the precise conditions under which lenition occurs in each of our illustrative systems. As summarized in (64), the distributional dif ference between the two glottalling systems, A and B, resides in the fact chat lenition in A only occurs domain-finally and then only if a consonant or pause follows. Under the ambisyllabicity approach, the latter conditions are implicitly identified as the context in which ¢ fails to undergo resyllabification into a following onset (by rule (68)). Under the licensing account, since ¢ already occupies an onset, the conditioning can only be expressed in terms of the nature of the following nucleus. In System B, glottalling occurs irrespective of whether the following nucleus is filled or empty. In System A, on the other hand, glottall- ing only occurs if the following nucleus is empty. “The initial representation of fin the lenition site contains the elements 2, R and bh. Glottalling consists in the suppression of the elements R and hh in the repres- entation of ¢ (see 3.4.2). In System B, these particular elements are unlicensed in context (83). In A, the same elements are unlicensed in context (83), but only if xis empty. The manner in which this withdrawal of autosegmental licensing is implemented differs in the two systems. As will become clear, it is necessary to specify that the unlicensed elements are delinked in A but ¢eleted in B: (85) tlenition A/B In context (83): (a) (System A) delink R and h, iff xs is empty; (b) (System B) delete R and h. ‘The System-A alternation between word-final t and 7 can now be shown 10 result from the selective manner in which the delinking conditions in (85a) ate Licensing 213 satisfied in different morphological domains. In a phrase such as (get by), word- final ¢is followed by an empty nucleus, both within the inner domain enclosing, ((get]) and within the outer domain enclosing {{get by]). According to (85a) then, R and h remain unlicensed on both cycles, with the result that r manifests the glorralled reflex. This is shown in (86), where delinked elements are paren thesized. (86) (get by) gezbay (System A) First cycle Oo NON ON 1 toto ' I (ik x x ox & x a t tot 1 tot “% ev? boa y I @®) Henition A (85a) ! ) Second cycle O NON ON Hororit 1 1 ik x x «x xx borer I 1 i! % ef? "ay" : y ®) HHenition A (85a) i a) On the other hand, in a form such as (get a) ris followed by an empty nucleus ‘only on the frst cycle ([get). Under the extended OCP, this nucleus is suppressed ‘on the second cycle where it would otherwise stand next to the nucleus of the form (a). The results that, within the phrasal domain ge a) 1s fllowed by @ (87) Get 2) get 0 (System A) First cycle Se On ee rrr fe x x xd bg 1 ! aoe? Ea | ® tlenition A (85a)214 Licensing Second cycle oO NON tot i oce ik x tot aed ° 1 \ 1 R lenition A (85a) 1 conditions not met h (On the second cycle, the presence ofa filled nucleus following the ‘means that the conditions for Flenition (85a) ate no longer satisfied. The filled nucleus supplies, the preceding onset position with the necessary capacity to license the elements h and R, with the result that an unweakened plosive reflex manifests itself. This account is in line with the view that delinking does not imply immediate deletion fof melodic material. In (87), h and R are unlicensed in ({get]) and thus cannot receive phonetic interpretation on that cycle. They nevertheless remain present in the representation where they are potentially available for interpretation, should the required licensing conditions be met on a later cycle, as they are in ({get a}. In the case of System B, the situation is quite straightforward. The deletion of R and h on an inner cycle correctly implies that these elements are not recoverable ‘on subsequent cycles. Hence the retention of a glottalled reflex in both (get by) land (get a). The same state of affairs holds in spirantizing System D, with the difference that it is the element ? that fails to be licensed in context (83) and is deleted: tlenition D In context (83) delete 2 (88) ‘We can now subject this analysis to a direct comparison with the coda accounts, by considering how it extends to System C. One aspect of the earlier accounts which can profitably be retained is the insight that the alternation between the tap and unreleased stop should be characterized in terms of two stages. The justifica- tion for this decision rests in the observation that the two stages are guided by different (though overlapping) conditions. In terms of their melodic composition, what the two reflexes have in common is an absence of noise release. Under an clement-based account, this commonality is expressed as the deletion of a lexic- ally present b. This constitutes one stage of lenition in System C and occurs in context (83), irrespective of whether x; is filled or empry. Ar this point, the weakened tis interpretable as an unreleased stop, the reflexthat appears in a form such as (get by). The second stage of lenition takes this already weakened tand reduces it further by deleting 2, yielding a tap, but only if a phonetically inter~ preted nucleus follows. In summary: Licensing 215 Henition C (a) In context (83) delete hy co) oN delete 2, iff x2 tot hhas melodic content. ' R Note that the two stages are intrinsically ordered, which is full : red, which is fully in accord with the principle that phonological processes occur freely whenever their conditions are met. The tapping part of tlenition (89b) is only free to apply once h has been suppressed by (89a). In other words, only a t without noise release is tapped. The derivations in (90 lustre she two stage aspect ofthis analysis (Here and in subsequent derivations, deletion is represented by a slash through the relevan element().) ” * " (90) (System C) (pity) (pit |) ONON Oo NON rordd raid radi iii * » eR mp oR 1 1 ? ? i I tlenition C (89a) uv u Oo NON prdid radia oR i ; tlenition C (89b) i conditions not met pic pit. Under this analysis, the alternation between the tap in (get a) and the unreleased stop in (get by) is attributable to the different ways in which the two lenition Processes in (89) operate within different morpho-syntactic domains. In accord- ance with (89a), h fails to be licensed on the inner cycle ({get]). On the next cycle,216 Licensing nce wit fails to be ? continues to be licensed in ({get byl) but, in accordance with (89b), licensed in ({get al). The following derivation demonstrates the resulting differen. ces between the two forms: (01) (gee by) get-bay (get a) gera (System ©) ‘Word cycle: t-lenition C (89a) ON ON ON ON ON N tot tt mois i 1 1 fi a x x x x} ke x XM x x ox x) bd i 1 I Jt tt to 1 ohn | & R "bo a y” "9 eo” R 1 1 i i ¥ ¥ Phrase cycle: tlenition C (89b) ON ON ON oN ON tot ot bin io I 1 [x x x x) x x xj ik x x xl tt I Hae I ' 1 ay "| e R “boa y “ge a "2" 1 e i low does this analysis account for the observation that tapping in System Ccan < oe fice before a stressed vowel word-finally (as in (get ff) but not word- medially (as in (cetdin))? The difference resides in the fact that, as the dominant position within the foot and word domains, the stressed nucleus following the tin (cetdin) is prosodically unlicensed, with the result that the conditions for h-loss defined in (89a), are never met. In (get Off), on the other hand, the being domain-fina, is followed by a paramettically licensed empty nucleus on the fist cycle (Igel), where the conditions for suppression of h (89) are met. Within the phrasal domain ({get off)), the presence of a following filled nucleus satisfies conditions for the deletion of ?(89b), resulting inatapped.eflex. © The analysis presented in this and the previous sections draws on principles o licensing which are motivated by a diverse range of phenomena apparetty unrelated to English ¢lenition. I have tried to show that, within this framewor the facts of tlenition can be treated without resorting to any of the following devices relied on by previous analyses: resyllabification, ambisyllabicity, extrinsic rule ordering, and ad hoc features. Moreover, the question of why lenition occu where it does is answered by identifying particular positions in the pr hierarchy as weak autosegmental licensors. Licensing 217 47.7 Preceding consonants So far, our examination of the tlenition facts has been confined t0 vocalic contexts. We now turn our attention tthe influence that neighbouring conson- ants can exert on the process. Here too, licensing relations between adjacent positions can be shown to play a decisive conditioning role. ‘The consonantal sites in question involve the constituent configurations in (92). (As before, « indicates some melodic expression corresponding to a consonant.) (2) @ R (b) 1 w\ oN NO N I rod PooIN D x ox x x x ox x «x 14 14 ae toe Iewill prove significant that these constitute a subset of the governing domains in (29). In (92a), tis preceded by a coda consonant (as in (mist, mister)); in (92b), t forms part of a branching onset (as in (petrol)). In each of these contexts, tis at least potentially susceptible to lenition, provided the second nucleus is licensed. It turns out, however, that the process is liable to be blocked in both instances, an observation that has led to such contexts sometimes being referred to as ‘pro- tected environments’.!" Let us first examine condition (92a). In none of our illustrative systems is lenition grammatical if the preceding consonant is an obstruent. This goes for both foot-internal (93a) and domain- final (93b) po: (93) A BCD @roetere after, mister, pistol, chapter, doctor et tice left, fist, bust, at, apt, adopt (Domain-finally in this context t (and, after sonorants, d) is subject to the independent process of total suppression discussed in 1.1. In some types of English, this is restricted to casual speech styles and is triggered when a conso- aant-initial word follows, as in (bes¢ do, ming! the). In other dialects, the lexical representations of such words have apparently been cestructured to exclude the final coronal altogether.™) AA preceding resonant does not have the same categorical blocking effect. Forms which historically contained r in this context show lenition, irrespective of ‘whether the modern reflex has consonantal constriction (chotic pronunciation) or has been vocalized (non-rhotic). Each of the four lenition systems is attested in both rhotic and non-rhotic versions, and in every case the lenition facts are218 Licensing identical to those associated with historically vocalic contexts. Compare the following results with those in (93): (94) ABCD @) t 2 ee b) 2? 2 © s party, quarter, forty ure, court, part | ‘The presence of a preceding lateral, on the other hand, exerts an inhibiting influence on lenition. As shown in (95), tapping and spirantization are blocked in this context. (95) ABCD fa) t 2? te filter, shelter b) 2 2 we faul, belt It is possible to get a stop reflex with no noise release here, but in this case the homorganicity of the coronal Itcluster results in a merger of ?and ¢~. That is, while an unreleased stop component realizes the f, coronality manifests itself over the cluster as a whole. In Systems A and B, however, historically dark /in this context is susceptible to vocalization to w; in this event, the debuccalization that results from glottalling removes the coronal gesture altogether, producing realizations such as bew? (belt). The effect of a preceding nasal also varies from system to system, as demon strated in (96). (96) ABCD @? @ re hint, sent, rant bc 2 Pe winter, twenty, plenty In parallel with the post-/ context, both spirantization and tapping are blocked here. In tapping dialects, ris either retained after m or lost altogether; inthe event of the latter, a form such as (plenty) rhymes with + + + after you, before one Note that the informal term pre-consonantal here refers to the syllabic position of a following sound rather than to its phonetic quality. (4a) and (4b) illustrate the same syllabic context, namely one in which the word following historical r begins with an onset. As the data here confirm, it makes no difference whether the sound ‘occupying the onset is characterized by some kind of consonantal constriction, as in (4a), or is a (non-consonantal) glide, as in (4b). ‘The pattetn illustrated in (4) is also found whenever historical r occurs word- finally before a pause (): () ABCD etre bear I, star, poor I, clear I, alter! Turning now to morpheme-final historical r when followed by a vowel, we find constricted reflexes appearing in Systems A, B and C but not in D. This pattern is, evident both word-internally (6a) and across words (6b). It B and C, the ful variant is obligatory in the former context and optional (though preferred) in the latter, (6) ABCD (a) + + + * bearing, starry, altering (b) + + + * — bearup, star of, poor Eva, clearup, alter a, afterall, before eight In non-thotic B and C, we now see an alternation between r and zero in words such as (bear, star, poor); linking r shows up pre-vocalically, as in (6), while the zero alternant appears pre-consonantally or pre-pausally, as in (4) and (5). On the face of it, the distribution in (6) appears to be identical to thet of pre-vocalic rin a morpheme-internal context (cf. the forms in (3) ). However, as we will see late, there can be quite striking discrepancies in the range of vocalic distinctions that hold before rin VAJV as opposed to Vi¥V sequences. ‘Taking stock of the distributional facts reviewed so far, we can say that rhotic System A displays consonantal rin the full range of contexts examined so far. The non-rhotic systems, on the other hand, only show constricted reflexes in some oF all pre-vocalic positions. In Systems B and C, we find rzero alternations in morpheme-final position, with the consonantal alternant appearing pre-vocalic~ ally. System D, in contrast, shows no such alternation; constricted rfails to appear ‘morpheme-finally and is confined to foot-initial onsets. Floating Sounds 235 5.2.3 Intrusive ¢ ‘One question that arises at this point is whether loss of post-vocalic constricted r in the non-rhotie systems has resuleed in mergers with nuclei from other historical sources. In some instances, this is indeed what has happened. Consider first the morpheme-internal pre-consonantal context illustrated in (1). The nucleus of words which historically contained ar (bar, cart, lard), etc.) is in most non- shotic dialects identical to the long low vowel that comes from various non-rhotic sources, such as in (calm, palm) (containing a vocalized development of historical ‘al and (father, Rajah). On the other hand, in many non-thotic systems the in-gliding diphthong in the BEARD word class is the exclusive reflex of historical Vr in this context; it remains distinct from diphthongs and long vowels which have no thotic source, such as those in BADE or BEAD. The synchronic relevance of the merger issue should become clear when we consider whether there is any evidence to support the presence of a lexical r in forms which are superficially less. On the face of it, there seems nothing to suggest that a child learning a categorically non-thotic system should have any ‘more reason to reconstruct a post-vocalic r in a form such as fam (farm) (historically rful) than in a form such as kasm (calm) (historically less). The question becomes less straightforward when we consider morpheme-final posi- tion where the zero alternation comes into play in Systems B and C. Here there would scem to be a prima facie case for contemplating the existence of an underlying r which gets deleted under certain circumstances."? In order to be able to get to grips with this issue, we need to extend our comparison of the four illustrative systems by examining whether the historical distinction between Vr} and V] sequences is maintained. First consider what happens with schwa in word-final position. In particular, compare forms such as (after, fear) (etymologically rful), with those such as (sofa, idea) (etymologically ress) ” ABCD @ +r after ten, better not, fear them torte sofa by, Sheena Kelly, idea to In the pre-consonantal context illustrated in (7), the rless pattern of the three znon-rhotic systems results in a merger of the Vil versus V] contrast. In chotic A, the contrast is retained. However, the non-rhotic systems diverge when we consider the occurrence of the same forms in pre-vocalic position: D, @ . (by after all, better apple, fear of @) A + . . sofa and, Sheena Easton, idea of ete c +236 Floating Sounds In (8a), B and C optionally display the linking r already illustrated in (6). D is not expected to show linking rin this position, given the total absence of morpheme. final r from this system. Note that, in the pre-vocalic context, non-rhotic B joins thotic A in retaining the contrast between Vij and V]. What is striking about non-thotic C here is that it actually exhibits a wider incidence of constricted r than rhotic A. That is, rshows up not only in the etymologically ful forms in (8a) but also in the etymologically less forms in (8b), as in (Sheenal7] Easton). In the latter case, the ris, from a diachronic point of view, unetymological or ‘intrusive’. ‘The domain within which optional rsandhi operates in Systems B and C is apparently not syntactically circumscribed. All that is required is that no pause intervene after the word-final context. The following examples illustrate how intrusive rin C can occur phrase-internally (9a), across a phrase boundary (96), across a clause boundary within a root sentence (9c), and even across sentences (9a). (9) (a) All that’s left is a vast area (7 of debris. (b) Hondalr] own this circuit. (c) This could be a problem for Villa[d as he floats one in for Klinsmann. (4). You can see Sennaff). I ean’t hear him. ‘A pattern similar to that evident in (7) and (8) emerges when we extend our comparison of historical Vil versus V] sequences to other vowels. Pre-consonan- tally, only A contrasts PAW with PORE and PA with PAR: (10) ABCD os lore to, sore knee far too, star turn law takes, saw them ma said, Shah can Pre-vocalically across a word boundary, B and C show linking r (11a), while C also shows intrusive r(11b): (11) ABCD fa) + + + * lore of, sore eye far off, star of aw of, saw it ma and, Shah of beta In Cand D, (lore of) and (law of) are homophones. In D, neither form contains as a result of (lore of) lacking linking r. In C, both phrases contain 1 the ‘etymological variety in (lore of) and the intrusive variety in (law of). Pre-vocalic intrusive ris also possible word-internally in PAW forms in C, as in (saw{r]ing wood), (draw(rling a picture) (even if this particular pronunciation touches a raw nerve with ~ to borrow again from John Walker ~ ‘the learned and the polite’). Floating Sounds 237 Superficially at least, there is a restriction on the sandhi pattems illustrated in (6), (8) and (11): they only hold if the vowel involved in the contrast is non-high. The sound in question may be schwa, in which case it occurs cither alone in an unstressed nucleus (as in the second syllable of ba: (bar). ‘The question now arises as to how we ensure that the r alternant in forms such as (bat) is preserved in pre-vocalic contexts. A solution that springs to mind is to suggest that such rs are not syllabified in rhymal position at the time R-Dropping applies. Since they are immediately followed by a nucleus, we can take it that they occur in an onset, where they fail to meet the structural description of the R-Dropping rule. This is illustrated in (18), where we can compare the fate of pre-consonantal rin (barred) with that of pre-vocalic rin (barring). (18) Dropping OR OR PIN rims band ba d bad OR OR not applicable bt 7 IN bad in to One requirement of this analysis is that an unoccupied onset be made available for pre-vocalic rto be syllabified into. This involves resyllabification, a version of the Onset Capture rule in 4.7.3.2 Initially, stem-final r is attached to a chymal position on the inner cycle where the form (bar) is syllabified. Then the consonant is re-attached to a following onset when this becomes available on the next cycle in the form (barring). This is more fully illustrated in (19). (19) Inner eyele Outer eycle ‘Onset Capture R-Dropping (a) (barred) OR not applicable OR ran rin (ib ae dl [b a “d) (b) (barring) OR R OR OR not applicable EIN IN tr 7 IN (ba Him bear iW242. Floating Sounds In (barring), the vacant onset happens to be supplied by a word-level suffix. But the position can also be provided by a following word in a phrasal ot larger domain, as the following derivation of (bar of) illusteates: (20) Word cycle Phrasal eycle Onset Capture R-Dropping OR R OR OR PIN IN rrof IN ‘bad b&b W.1 far pb Wy not applicable Implicit in the derivations in (19) and (20) is the assumption that phonological strings contain syllable structure on all cycles. In order to dispense with resyllabi- fication under the coda analysis, it would be necessary to take the alternative view that strings are syllabified once only and ‘last thing’. That is, syllable structure would not be present on inner cycles but would be constructed post-cyclically after all morpho-syntactic operations had been completed. This would mean that rwould only ever appear in onset position in (barring) and in coda position in (barred). However, this alternative is not open to us. As we saw in 2.3.4, syllable structure has to be made available from the outset of derivation, since it provides the backdrop against which phonotactic relations in nor-analytic domains are defined. A further requirement of the coda analysis is that the deletion of morpheme- final r must somehow be held at bay until the consonant has had a chance to become syllabified into a following onset, if one is available. Deletion must not be allowed to apply on the inner cycle; otherwise, r would be erroneously erased in (barring) at a stage when the consonant is still in the rhyme: (21) Inner cycle Outer cycle R-Dropping Onset Capture OR R OR R net applicable tt 1 ri t {lb ™ ial (ba Ji a * bam (barring) This undesired result could of course be prevented by the brute-force expedient of extrinsically ordering the two rules." However, a rather more attractive alterna tive, which dispenses with rule ordering, is to assume (a) hat the delinking of r occurs freely wherever its structural description is met; and (b) that it does not entail immediate deletion. This would allow r to be delinked on the inner cycle (bar) in (18)) and then subsequently relinked should a vacant onset become available on a later cycle, as in (barring). Any delinked r which failed to find sanctuary in an onset on any cycle would then be erased atthe end of derivation, 4s in the case of (barred). Floating Sounds 243 One striking consequence of the coda analysis, as formulated to this point, is that the phonological distribution of r in underlying representations in thotic System A and non-rhotic B is very similar. In both cases, underlying rcan appear in chymes as well as onsets. Moreover, the lexical incidence of the consonant is, also identical in onsets and in morpheme-final position and differe only in that r ‘occurs before a morpheme-internal consonant in A but not in B. The main distinction between the dialects is relatively superficial and has to do with whether or not R-Dropping is present in the grammar. 5.4.2. R-Epenthesis Now we can consider where System C fits into this classification. The C group of dialects, recall, displays the so-called intrusive-r pattern, one symptom of which is the absence of a contrast between alternating Vj-V] and non-alternating V] words. All words ending in a particular class of vowels in pre-consonantal or pre-pausal position show a variant containing constricted r in pre-vocalic posi- tion. The immediate question is whether or not an ris lexically present in these forms. If there is one, we need some kind of deletion rule by which to derive the zero variants. If there is not, we need some kind of insertion rule. Either way, the lexical incidence of ris going to differ from the patterns found in Systems A and B. If there is an underlying rin these cases, it will appear not only in those morphemes which have it in A and B (the etymologically ‘correct’ ones such as ar, lore), etc.) but also in many forms which lack it in the other systems (the etymologically “incorrect” ones such as (pa, law, idea), etc.). Ifthere is no under- lying r, it will be missing from both historically less and rful forms. ‘The particular class of vowels which precedes alternating r in System C is in principle identical to that preceding historical rin both Systems A and B. (D has its own special developments, on which more presently.) In most dialects, whether rhotic or non-thotic, the sub-system of vowels in this context is quite unlike that occurring in other contexts. For one thing, the set of contrasts holding before historical ris much reduced from the maximal inventory, as found say before This is illustrated in the following comparison of the two stressed sub-systems in typical southern English non-rhotic system (B or C): (22) Maximal inventory Inventory before historical r BIT ' FEAR 1B BEAT i BET e BEAR eo BAIT ey BITE ay WIRE a BAT ® CART = BAR } a0 SHOUT — aw HOUR cor >244 Floating Sounds CAUGHT — = cur 4 BOAT ow FOUR } 28 PUT o POOR BooT The collapse of distinctions before historical ris evidently due to a number of qualitative and quantitative developments, some of which we will examine later in the chapter. The fact that similar developments show up in most chotic dialects confirms that, a least historically speaking, these various effects can be attributed directly to r itself. For the moment, it will be sufficient to speak of a general lowering influence that historical r exerts on a preceding vowel. ‘A standard rule-based approach to the r-2ero alternation in C-type dialects isto treat it in terms of segment insertion.** The relevant rule turns out to be more or less an inverted form of R-Dropping. The main justification for this analysis is that the surface distribution of morpheme-final ris allegedly fully predictable in a type-C system. Given this predictability, so the argument goes, the segment is not distinctive in the alternating site and thus does not need to be represented underlyingly.”” This account is founded on the assumption that the appearance of rrin this context is dependent on the quality of a preceding vowel rather than vice versa — an inversion of the historical pattern just outlined. In other words, the special set of vowels before historical r now supposedly forms an independent sub-system which is represented in the lexicon and constitutes the right-hand portion of the environment into which ris inserted. ‘An orthodox linear formulation of the insertion rule looks something like thi (23) v § > s/{-high) ‘The initial portion of the rule’s environment refers to the generalization that the rralternant in System C only appears after non-high vowels; compare the data in (6), (8) and (11) with those in (12). It is not necessary for the insertion site to be specified as morpheme-final position; the rule can be assumed to apply vacuously in morpheme-internal contexts, where + is already present underlyingly (as in (dairy)? Reinterpreting (23) in syllabic terms, we can state that ris inserted into an unoccupied onset, provided that it is preceded by a non-high vowel: (24) R-Epenthesis NO Pov tot [high] — Floating Sounds 245 (The horizontal bar under the onset indicates the insertion site.) The operation of the rule is illustrated in the following derivations of (saw = soar) in System C (in those dialects which have a mid round nucleus in this class of words}: 5) R-Epenthesis OROR 9% tu (sawisore to) trad not applicable {ls 9} fe ull ORR OR OR soir m (sawlsore in) rt it 1\ Us 9] fi al} ans OR sod (sawed/soared) ri not applicable Us 2} 4] ORR OR OR soir {sawinglsoaring) NN rE yon (soli soe ig 5.4.3 Non-Rhoticity Let us take it that the coda analyses just outlined are correct at least in so far as. they identify constituent structure as the conditioning factor in non-rhoticity. To be in a position to assess the validity of the remaining aspects of the analyses, it will be useful for us to have some theory-neutral bench mark which couches the statement of non-rhoticity solely in terms of distributions that are observable in the data. Any statement formulated in terms of R-Dropping fails to fit this bill, since it incorporates certain assumptions about the nature of the rule's lexical input which may or may not be justified. Specifically, it presupposes a mismatch between the incidence of r at the initial and final stages of derivation. ‘What is required is a statement formulated as an output condition on derivation — ‘one which specifies the distributional goal that a successful derivation must target bbut which is neutral on the question of how that goal is achieved. Thatis, it makes no claims about the lexical status or derivational history of r in non-rhotic systems. R-Dropping focuses on the conditions under which lexical r fails to surface in non-rhotic systems. Suppose we now switch the perspective and reformulate our bench-mark statement in terms of where r does appear. Based on the observed distribution of rin constituent contexts, an output condition for both Systems B and C might be expressed as follows: (26) Non-Rhoticity appears exclusively in onsets,246 Floating Sounds Any analysis of non-rhoticity must capture this generalization in some way. The simplest analysis is one which requites only one statement :0 do so, A deletion analysis of System B can in principle meet this criterion by specifying the comple- ment set of contexts, Whether the particular deletion account offered by the R Dropping rule in (16) can be judged successful in this respect depends on whether the coda environment ints structural description correctly characterizes this complement set. As we will now see, there are grounds for concluding that it does not One aspect of the deletion account of non-rhoticity that is common to both the SPE and coda formulations ofthe process concerns the point made earlier about vacuous application. Under the coda view, there are two contexts where deletion operates vacuously ~in morpheme-internal hymes (asin the ist sla. Bes of ther, party) and in final hymes closed by some other consonant (ain farm, harp)). As a dynamic process which produces an active r-zero alternation, theme has relatively restricted domain of application. Although the formula- tion of the rule in (24) might give che impression thr deletion affects any syllable rhyme, all of the observable alternation action actually only takes place in what, unde this analysis, constitutes a word-final rhyme and even then only when this is not closed by some other consonant. This is of itself not necessarily a bad thing, since the rule succeeds in subsuming the alternating and non-alternat- ing contexts under a single generalization about the surface distribution of non-rhoticity. What is peculiar, however, is that the dynamic aspect of the coda analysis boils down to an account of what happens in absolute domain-final position ~ the very position whose supposed rhymal status we now have good reason to reject. In short, the coda context referred to in R-Dropping (16) does not accurately identify the set of contexts that complements the onset site referred 10 in (26). 2 significant ‘What then of the epenthesis account of System C? In expressing a signi generalization about the surface distribution of 1 the Non-Rhoticity condition (26) is as much true of the System-C grammar as it is of B. R-Epenthesis (24) on it own isnot sufficient ro capture he full distributional facts of rin this ope of dialect. In other words, the insertion analysis requires two statements regulating, the surface distribution of r— R-Epenthesis and Non-Rhoticity. A simpler and therefore preferable account would be one in which the distribution of the nt in intrusive-r dialects is reduced to a single statement. - “any cee there ae atonal reson for questioning the valid of ERE: thesis. The rule is potentially arbitrary in two respects. First, the process itsel ‘ust be considered arbitrary, unless grounds can be provided for assuming that it smust be rtha is insereed eather than any other randomly selectable sound. Thee is no obvious local source in the surrounding vowels. The potential damage of this cicism might be mitigated by invoking the universally unmarked status of coronals. That is, it might be argued that the coronality of r is specified by universal default rule which, by viwe of being supplied by Universal Grammar, comes cost-free to the System-C grammar. There are certainly other examples of languages which make use of what looks like a spontaneously appearing rin Floating Sounds 247 certain contexts. The question remains, however, as to why it is r that epen- thesizes rather than some other coronal such as or d, particularly since ob- struents such as the latter are considered less marked than sonorants. There is another respect in which R-Epenthesis is in need of justification. Why should it take place in the context it does, between vowels as long as the first is non-high? Would we have been surprised if it had applied in any other environ. ment? The inter-voealic aspect of the context may seem to be motivated by a universal preference for syllables with filled onsets. That is, the epenthetic seg- ment supplies a realized onset which breaks up a potential sequence of nuclei." Bur if the rule were an instantiation of this preference, we would expect it to split up any potential V-V sequence. The puzzle is why it should only fulfil this function when the preceding vowel is non-high. Epenthesis, recall, fails to apply after high vowels in sequences such as (fee of, two and, day of, bow and) (see the forms in (12)). To counter this criticism, we might note that this is the very context where we find hiatus-breaking y and w glides; hence (fee [ylof, two {wland). As mentioned in 4.7.4, this phenomenon is straightforwardly treated as the spreading of I or U from a nucleus into a following vacant onset. What unifies the set of vowels preceding the intrusive-r context, it might be argued, is that they lack either of these elements; r then is the default hiatus-breaker which appears in the absence of a locally available high glide.2® ‘These arguments in defence of R-Epenthesis are based on the assumption that the rule accurately captures a surface-true generalization about intrusive-r sys- tems. The force of this claim is, however, undermined by the observation made above that rexerts a lowering influence on a preceding vowel. This tendency, which is attested cross-linguistically, is sometimes invoked as an explanation for the occurrence of {-high] in the left-hand environment of the rule. But, if any- thing, the observation actually constitutes an argument against R-Epenthesis, At stake is the question of which segment is influencing which. Is the appearance of intrusive dependent on the quality of the vowel, or is the quality of the vowel dependent on the #2 The R-Epenthesis analysis implies the former answer, But the latter answer is implicit in the claim that rhas a lowering effect on a preceding vowel. This effect has to be accounted for in any event. That is, there has to be some independent provision in the grammars of all our illustrative systems which accounts for the reduced set of vowel contrasts found in this context. The epenthesis account of intrusive r fails in this respect. A preferable analysis of non-rhoticity would be one which meets the follow- ing criteria. First, is should be consistent with the view that domain-final r ‘no more occupies a final coda than does any other consonant. Second, it should if possible derive each of the observed patterns of rdistribution by means of a single statement. Third, it should account for the qualitative influence that r exerts on a preceding vowel. In the following sections, we will consider an alternative account which comes closer to fulfilling these goals than those dis- ‘cussed so far.248 Floating Sounds 5.5 Floating r Expressed in terms of autosegmental licensing, (26) might provisionally be refor- mulated as follows: (27) Now-Rhoticity’ Only an onset position licenses 1. ‘The condition in (27) correctly allows for the appearance of r in word-initial and internal onsets, as in (red, carry). What of rin other positions? Let us accept ‘the argument that the only other context in which r is lexically present in non-rhotic System B is domain-finally in forms which showthe r-zero alternation, as in (bar). The question now is what syllabic position such occurrences of domain-final roccupy, if it is denied that they are syllabified in the ehyme. In line with the conclusions reached in chapters 2 and 4, a reasonable first assumption would be that lexical r in this context occupies an onset licensed by an empty nucleus, as in: (28) ON IN i wf This analysis seems attractive, since it would permit us ¢0 unify r loss with the various sorts of weakening discussed in the last chapter. Non-Rhoticity could thus be understood as reflecting the weak licensing capacity of the final empty nucleus. More specifically, the parametrically licensed nuclear position could be viewed as failing to supply the preceding onset with sufficient autosegmental licensing power to support the melodic expression defining r. However, as we will see below, there are grounds for considering the representation in (28) to be appro- priate only for one quite specific type of system but not for others. Enquiring into the syllabic position occupied by rbegs another question, namely whether it is necessary to assume that a consonant in this context occupies any Kind of position at all. In fact, phonological theory makes available another possibility ~ that domain-final ris lexically a ‘floating’ segment with no syllabic position of its own. The sort of configuration envisaged here is illustrated by the following lexical representation of (bat):?? nx -0 ook 2 oN tod 1 1A Bae Floating Sounds 249 In (29), we have an rwhich isnot atached to any syllabic postion. As long as the melodic expression remains in this decached state, itis autosegmentally unlicensed and thus unrealizable. For it to be phonetically expressed, it must acquire some attachment to an available postion. In (29), the only available position is the nearest poinc dominated by the nucleus. Flowever, according to condition (26), rean only be licensed by an onset in non-rhotic systems. A candidate position does become availble, we contin with the assumption, already made inthe coda account of non-rhoticity, that an unoccupied onset is supplied by a following vowel initial 1morpheme. The association ofa floating segment toan available onset we may asune automatically triggers the creation of a skeletal point. In a form such as (bar a) (30a), the floating rcan thus ‘dock’ on to a position which is able to license it2 (30) (a) (bar a) ON ° Ph oF PR Pr Yah IY SAA CL, (b) (bar the) PR PT PR et bi di ti bi The difference between a docked and an unlicensed ris the difference between ful and eles aleemants in the non-rhotc B system. In (bar the) (30, the onset following the floating r is already occupied. Since there is nowhere for the segment to dock, it remains unlicensed and hence unrealized, Under the floating-segment analysis, the rule of R-Dropping in System B is replaced by a condition requiring r to be licensed by an onset position. Where does this leave the rule of R-Epenthesis which characterizes the pattern of intru- sive r in System C? In B, the lexical representations of alternating Vj-V] mor- Dhemes sch as bat) ate detngusbed from thous of now sheave ¥] morphemes such as (Shah) by the presence of a domain-final floating r: (31). System B ON ON Hin ray xr x xxx bale iat (bar) (Shah)250 Floating Sounds ‘The Non-Rhoticity condition in (27) correctly characterizes the surface distribu- tion of r in System C no less than in B. But what consequences does this observation have for the lexical representation of morphemes which show the zero alternation in C? Given that the historical distinction between final and rless forms lias been obliterated in this system, there are in principle ewo possible ways of representing forms such as (bar) and (Shah) under the floating. Segment account: either both contain domain-final floating r (32a), or neither does (326). (32) System © @ oN ON ray 1X Tis Tis bate dae 6) ON ON ry Vi\ rif rif ba” med (bar) (Shah) With representations such as those in (32a), the zero alternation in System C can be derived in exactly the same way as in B, ie. in terms of the Non-Rhoticity’ condition (27). No additional rules or conditions are required. The difference between B and Cis thus purely a matter of lexical incidences floating ris present in the lexical representation of a greater number of forms in C than in B (and than in A, for that matter). I might be surmised that the lexical incidence of floating r in System C is restricted to those forms in which the consonant is preceded by a non-high vowel. If this were true, the analysis would then be open to criticism on the grounds that it failed to capture this clear generalization. However, this objection presupposes one of the assumptions underpinning the epenthetic-r analysis alluded to above, namely that it is the quality of a preceding vowel which determines the appear ance of r, rather than vice versa. This analysis, as we saw, is itself vulnerable to criticism on the grounds that it fails to account for the general lowering influence of On the other hand, if the alternating r is lexically present, as in (32a), available as a lowering trigger. (Precisely what it is about rethat has this effect is a question to be addressed in 5.7.2.) Under this analysis, the absence from the data of sequences of a high vowel followed by intrusive ris not the reflection of some mysterious lexical lacuna. In principle, floating ris free to occur after any vowel in lexical representation. The distributional gap is a reflection of a process it is Floating Sounds 251 which lowers vowels in this context. Thus there are lexical contrasts such as fi ee) versus fir (—» fiar) (fear). Let us summarize the floating-segment analysis of non-thotic Systems B and C. In both systems, as expressed in the Non-Rhoticity’ condition (27), the surface distribution of constricted + is limited to onset position ~ see (33a) (where a indicates some vocalic expression). (33) @ 0 (b) RK Zz RK -Z 1 t © © In lexical representation, r occurs either in onset position (33a) (as in (ced, carry) cor domain-finally as a floating consonant (33b) (as in (bar)). The phenomenon of rsandhi is formally expressed as follows: the licensing of a lexically floating ris contingent on an unoccupied onset being made made available for it to dock on to; otherwise it fails to surface. Amongst the non-rhotic systems, D is distinguished from B and C both in terms of the lexical incidence of rand in terms of the manner in which ris licensed by ‘onsets. In the absence of alternations motivating the lexical construction of morpheme-final r, it must be assumed that D lacks floating r. Thus D possesses rrless lexical representations such as those in (32b). As it stands, the Non-Rhoticity’ condition (27) is inadequate for System D. While any onset is capable of licensing r in B and C, only foot-initial onsets can do so in D (see (3)). This in fact renders the distribution of rin D identical to that of hin most dialects of English, as discussed in 4.7.5: (34) General English h System D Foot-initial {head [eled Foot-internal —_velficle veltly Domain-final *kah caldl ‘The same kind of licensing conditions can be assumed to apply in each case. That is, in line with the Licensing Inheritance Principle, the ability of an onset to autosegmentally license r varies according to whether the following nucleus from which that ability is acquired is itself licensed. The licensing charge delivered by the dominant nucleus of 2 foot, unlicensed within that domain, is passed relatively directly on to the preceding onset. The autosegmental licensing power of a foot-internal onset, on the other hand, is attenuated as a result of being acquired vvia the licensed nucleus on the foot projection. In D at least, this diminished potential manifests itself as an inability to support not only A but also r. In Systems B and C, any nucleus can empower its onset to license 7 in D, by contrast, only an unlicensed nucleus has the capacity to do so.252. Floating Sounds 5.6 Historical interlude By way of a brief detour, let us at this point consider some historical evidence which seems to favour the floating-r analysis over one incorporating rewrite rules such as R-Dropping and R-Epenthesis. The sequence of events leading up to the present pattern of dialect divergence among rsystems looks quite different when viewed in terms of the two types of analysis. Before examining what the main } { 0 } FOR = oo (Pre-r neutralizations are also evident in contexts where the consonant appears in domain-internal onsets; hus there are potential mergers in the MERRY- MARRY-MARY series as well as in SPIRIT-EERIE, on which more below.) As illuserated in (40), the qualitative adjustments that accompany the shrinking of the distinctive space inhabited by pre-r vowels include a widespread lowering, the disfavouring of tenseness and the presence of an off-glide. For the sake of brevity, we may focus only on the last of these effects, one which results from a process sometimes known as breaking.** Dialects which have up-gliding diph- thongs such as ow, ey in contexts which display the maximal system of contrasts tend to lack these before r. A typical instantiation of this pattern is to have the up-glide replaced by a schwa in-glide, a development which is usually accompa- nied by a lowering of the first portion of the diphthong; compare, say, ein (care) with ey in (fade). In another development, we find the up-glide being retained and serving simultaneously as the onset of a separate syllable containing 27, ¢8- ekey.yar (care), fay. yar (fire). ‘A lot of ink has been spilt in debates about how the restricted set of vowels before historical r should be assigned to phonemic categories set up on the basis Floating Sounds 257 of the maximal inventory found in other contexts." The phonemic assignment problem is essentially a transcriptional one; it involves taking each of the pre-r vowels and relating it to one of the vowels in the maximal inventory. This problem simply does not come up for discussion within a non-phonemic phono- logy, particularly if we are not wedded to the assumption that the same system of phonological distinctions necessarily holds in all contexts. What does need to be explained, however, is the observation that the set of vocalic contrasts is severely reduced before r in other than Scottish systems, Why should we find such a defective pattern in this context and not before other consonants? (Actually, something similar is to be found before the vocalized reflex of Jin some dialects. See exercise 3 at the end of this chapter.) 5.7.2 Nuclear ‘The idea we want to be able to capture formally is that post-vocalic rin other than Scottish systems somehow constricts the potential for vocalic distinctions to ‘manifest themselves. As noted above, the relative freedom of vowel contrasts to hold independently of following domain-final consonants other than r can be taken to reflect the syllabification of such VC sequences into separate constituents (sce (38b) ). Suppose we take the vocalie restrictions that obtain before domain- final rto reflect a closer relationship between segments in a Vrsequence. Specific- ally, let us make the assumption that r in these contexts is syllabified in the nucleus itself. The presence of r within a nucleus would then be expected to result in a compression of the distinctive space available to vocalic elements. By way of a first draft of this proposal, the representation in (41) illustrates the occurrence of nuclear rin the rhotic form of (bar). The branching nucleus here takes the form ofa diphthong with a rhotic off (1) ON ON rit bar In rhotic systems other than Scottish, the defective distribution of pre r vowels is not restricted to contexts where the consonant occurs domain-finally. Similar merget patterns to those illustrated in (40) are to be found domain-internally before a consonant. Compare, for example, the vocalic sub-systems found before -ntand -r1 clusters, shown in (42). Again we can use the unmerged Scottish English pattern as a reference point. (42) Scottish Other rhotic FLINT 1 SKIRT ' nn BLUNT HURT a258 Floating Sounds This suggests that the type of syllabification shown in (41) should be extended to domain-internal contexts, as in the following representation of (part). (8) ON ON rive bird part (On the representation of », more presently.) In this way, the defective distribu- tion of vowels before rC clusters can be attributed to the fact that some of the contrastive space available within the nucleus is taken up by r ‘The nuclear status of non-onset r is further supported by the behaviour of pre-consonantal rin tlenition, discussed in 4.7.7. Recall that glottalling, for example, can occur if £ is directly preceded by a nucleus, as in a. The latter development, which (as we saw in 4.4.3) consists in the rightward spreading of melodic material from the first nuclear position into the second, preferentially affects relatively lower vowels. The most extreme outcome of this development occurs in dialects in which 19 in BEER remains as the only in-gliding diphthong; otherwise we find monophthongal a: in BAR, a: in FOR = FOUR = POOR and e: in CARE. 5.7.4. MERRY-MARY-MARRY Another site where breaking effects are to be observed in some dialects is before onset r within morphemes, as in (carry, spirit, sorry). Once again the result is varying degrees of merger, which suggests that the nuclear incorporation account ‘can be extended to this position. Since medial onset ris lexically present in both thotic System A and non-rhotic B/C, the distribution of the merger patterns cuts ight across the classification of the three systems. It is worth comparing such dialects with those in which the generalization of breaking has failed to take place, because the latter provide further evidence of a systematic phonological lifference between lexically floating and anchored r, As in domain-final position, breaking in medial position has had the effect of neutralizing the historical distinction between long and short vowels in favour of long. In cases where breaking has affected the domain-final but not the domain- internal context, we get contrasts such as hari (hurry) (short vowel before medial 1) versus fou} {furr-y) (long before domain-final 1. The extension of breaking to medial environments can lead to large-scale merger of VAV versus Vir(}JV con- trasts, so that (hurry) and (furry), for example, become perfect rhymes. (52) shows264 Floating Sounds wth some iin historical ditnctions that can be affected inthis way, together Tapeesenative vocaic elzains The mergers lfc diferent dal vo ifr: ing extents (52) spirit 1 } e spearling ia Mary >) ee ere le marry 2 8 uy 8} furly sorry 2 | os story 0 i at the nuclear incor- ‘The operation of breaking in medial contexts indicates that the poration of @ can affect rnot only when itis floating, as shown in (51), but also when iis lexically anchored to an onset. An originally short nucleus affected by this change undergoes historical restructuring, resulting in potential merger wit an originally long vowel. Compare the following unbroken (53a) and broken (53b) variants of (merry): (83) (a) meri (merry) (b) meari (merry (= Mary)) oN ON o NON tort ee ee i pit i 1 i mw | oR i mt \ns a \I te i 1 1 5.7.5 Summary in 5.7, we have examined the special length and quality effects that appear in vowels peeling historical r in other than Scottish systems. Under the deletion and epenthesis accounts of non-rhoticity discussed in 5.4, the nature of these patterns remains mysterious. The effects can, however, be explained if we assume that r, ot at least some part of its melodic make-up, has invaded the nuclear space inhabited by the vowels in question. Breaking, for example, can be attributed to the presence of the vocalic component of approximant rin a nucleus. The main distinction between non-Scortish rhotic systems on the one hand and non-rhotic Floating Sounds 265 systems on the other is that the element R has been absorbed into the nucleus in the former case but not in the latter. 5.8 Conclusion In this chapter, we have seen how the distributional peculiarities of r in Eng- lish reflect the eather special status it has in various dialects. Its nearest relat- ives in the system are y and w. Like them, it can occur in nuclei as well as in onsets. Its appearance in nuclei accounts for the squeezing of the space that is available to vocalic distinctions in Vrcontexts. In non-rhotic systems, rin sandhi environments is represented as a lexically floating segment which attaches itself to an unoccupied onset when this is made available during the course of a derivation. ‘The glide-like status of ris reflected not only in its distributional but also its phonetic properties. In terms of its coronal constriction and ability to occur in ‘onsets, it lines up with the true consonants. In terms of its secondary vocalic characteristic and its ability to occur in nuclei, it lines up with the true vowels. In non-thotic dialects where it is the sole participant in sandhi, its floating repres- entation makes it unlike any other segment in the system. John Walker was surely not wide of the mark when he described r as the most imperfect of the consonants. Exercises 1 Other rsystems The data Get your own, There are various systems which do not correspond exactly to any of the four described in this chapter. If you have access to speakers of such a system (yourself, for example), collect data that you can draw on in considering the following questions. The task (a)_ Precisely in what respects does the system differ from those discussed in this chapter? (b) To what extent is the divergence attributable to differences in the lexical incidence of {c) If the system exhibits an rzero alternation, can the floating-r analysis account for it? If not, how might the analysis be modified so that it does? (4) Determine the set of vowel contrasts before historical . IFit is not identical to those found in other contexts, account for the266 Floating Sounds 2 Breaking and the southern drawl ‘The data ‘The data below illustrate the front vowel series before historical rin a particular non-rhotic southern dialect of United States English.% The range of vowel qualities and their lexical incidence are not directly equivalent to anything described in the standard textbooks, but this point can be set on one side for the purposes of this exercise. The system exhibits a version of the in-gliding pattern discussed in this chapter. Note, however, that breaking only occurs in a specific suaset of historical-r contexts. This results in superficial contrasts between broken and unbroken reflexes; compare, say, the €9 of (caring) with the eof (Mary). The task (a) Provide a full analysis both of the r-zero alternation an¢ of breaking in this system. (b) How are the breaking and non-breaking contexts distinguished under the floating-r analysis? (c) Determine the morpho-syntactic domains within which the different vo- calie variants occur. (a) What is the elemental make-up the various vowels represented in the data, and how is this affected by breaking? 1 beer, fear, jeer 2 wre beery, fearing, jeering. 30 ir... Erie, hero, series 4 ef ‘care, hear, scare 5 cor. caring, hearing, scaring 6 er. dairy, hilarious, Mary 7 er... berry, heron, merry 8 wot air, bear, dare 9 wor... airy, bearer, daring. 10 ar... carrot, sparrow, marry 3 London laterals The data In many dialects of English, lis vocalized to w in certain phonological contexts. In some cases, this results in Fw alternations rather similar to those involving rand zero; for example, (pee(lling-pee[w]). Moreover, like the rpattern, vocalization can have quite significant effects on the quality of a preceding vowel. Perhaps not surprisingly then, the lateral pattern raises issues very similar to those discussed in this chapter. Floating Sounds 267 ‘The particular vocalizing pattern illustrated here is one that is firmly established in London and is apparently gaining ground throughout the southeast of England. ‘The data in I provide information about the conditions under which the vocalized reflex occurs. As usual, the data are arranged in word rows, each of which represents a particular configuration of phonological and morphological condi- tions. The data in II show the influence vocalized / exerts on the quality and length of a preceding vowel. Note that the system is non-rhotic, a point that is of relevance in rows I1.4(b) and Il.5. For comparative purposes, the system's max- imal vocalic inventory, as found for example before 4 is given in IIl For ease of reference, the row numbers in II and Ill indicate rough correspondences between the two inventories. The task (a) Determine the conditions under which the vocalized reflex appears. (b) Account for the /-w alternation. What is the lexical representation of forms that show the alternation? (c) Compare the reduced vowel inventory in II, which occurs before the vocalized reflex of J, with the maximal inventory given in III. Account for the effect that Fvocalization has on preceding vowels. I / VERSUS w live, loud play, glow trolley, tally, yellow belt, fold, help, bulge tell, pull | feel |, goal | tell Joe, pull through, feel bad, goal down felled, bowled, sealed telling, pulling, feeler, goalie tell us, pull over, feel it, goal up weakens mmeeee rn II VOWELS BEFORE VOCALIZED / 1 ow (a) fil, still, build, guilt (b) feel, steal, field 2 ew tell, bel, else, weld 3 ew (a) pal, Val (b) tail, pale, mail (c) owl, fowl, growl 4 aw (a) tle, pile, smile (b) snarl, Charles268. Floating Sounds 5 ow cwirl, girl, hurl 6 ow (a) doll, solve (b) goal, pole, bold (6) dull, gull, bulb 7 ow (a) full, pull, bull (b) fool, pool, school (c) fall, crawl, fault TI MAXIMAL VOWEL INVENTORY 101 BIT sy BEAT 2 © BET 30 = BAT ey BAIT BOUT 4 oy BITE a CART So SHIRT 6 > POT aw BOAT a BUT 7 0 Pur ew LOOT co: CAUGHT Epilogue eee ‘What makes English sound English? The question can be approached from two angles. Looking at it from the outside in, as it were, we might ask whether anything distinguishes English phonology from the phonologies of other lan- ‘guages. That is, can we put our finger on a collection of phonological properties that will uniquely identify English to the exclusion of other languages, especially its close Germanic relatives? Or, bearing in mind the considerable phonological diversity that accompanies the wide geographical dispersal of the language, we might consider the issue from the inside out: what if anything unites different dialects of English at the phonological level? Put concretely, is there something, specifically ‘English’ about the phonologies of, say, Sydney, Detroit or Glasgow English that immediately sets them apart from dialects of, say, Frisian, Dutch or German? ‘Viewed from a classical SPE perspective, the relation between the phonological systems of cognate dialects involves the two areas of the grammar where pronun- ciation facts are accommodated in this model - the lexicon and the phonological component, Although either or both of these might be expected to be implicated in dialect diffecentiation, the emphasis in the relevant literature has been placed firmly on the phonological component. This leads to an apparently straightfor- ward interpretation of the view that related dialects share some kind of under- lying structural identity. Panlectal unity is supposedly manifested in a shared set of phonological representations in the lexicon and, possibly, a shared core of phonological rules. Dialect divergence is then deemed to involve discrepancies." the organization of phonological rules. These arise historically from the select’. manner in which such rules undergo addition, loss, reordering or restructuring: individual grammars. Even taken on its own terms, an extreme interpretation of this view ~ that ay dialect variation is expressible in terms of rule differences - does not stand up tf scrutiny. At stake here is the question of whether the phonological non-uniform ity of related dialects is in principle any different from cross-language diversity. Most research which has specifically addressed this issue has come up with a consistent answer ~ a resounding no. As is by now well known, the distinction between language and dialect has no basis in linguistic reality but is defined in purely social, political or cultural terms. This may seem a rather disappointing conclusion, because it would appear to undermine the intuitively appealing notion that purely linguistic criteria can be used to group together particular270 Epilogue systems as “dialects of the same language’. The abiding attraction of this assump- tion is reflected in the way it has continued to figure with varying degrees of explicitness in the phonological literature. However, all the evidence suggests that ‘no neat cutoff point exists at which ‘dialects’, distinguished purely by the contents of their phonological rule components, give way to ‘languages’, distinguished by differing lexical representations. In any event, the whole question of dialect and language variation takes on ‘quite different complexion once we begin to move away from a rule-oriented approach to phonological phenomena. Indeed, it is now reasonable to ask whether it is even appropriate to think of phonology as identifying an inde- pendent module in the grammar. “The response to the equivalent question in relation to syntax is now no. Syntax, on current understanding, no longer refers to an autonomous grammatical com- ponent, as it did in earlier generative theory. Ic is instead an informal term applied to the mapping between LF and PF, a relation that is governed by quite general constraints rather than by grammar-specific rules. In the light of the theoretical developments outlined in the preceding chapters, itis reasonable to suggest that phonology is on the way to achieving something of the same status. The term can still be applied to the derivational function whereby ‘one set of phonological representations is mapped onto ancther. But the scope of activity attributed to this function has shrivelled quite dramatically from what was countenanced in the heyday of SPE. The notion that there exists an inde- pendent phonological module, chock-full of language-specific rules, is now obso- lescent. The explanatory burden previously borne by the phonological rule component has gradually been taken over by the enriched model of repres- entations that has emerged during the last twenty years or so. Thus when we nowadays speak of the ‘phonology of language X’, we are referring primarily to the phonological structure of its lexical entries. So is the phonological structure of English in any way peculiarly English? We cean ask the question again, this time from the viewpoint of a theory which dispenses with rewrite rules. Taking the ‘external’ perspective first, we might set to identify a combination of settings on particular phonological parameters _at occurs in English but in no other language. This is a highly unlikely resule, in Sw of the small number of such parameters and the comparatively large number * patural languages. Particularly when we narrow the focis ta exchide all but its est Germanic sisters, it becomes clear that English is anything but special. Take, ror example, the English settings on five major parameters controlling prosodic structure: branching onsets (ON), branching rhymes (ON), branching nuclei (ON), foot-level licensing (HEAD-INITIAL), and licensing of final empty nuclei (ON). In terms of these settings, English looks just like any other West Germanic language, not to mention a considerable number of its more distant Indo-Euro- pean cousins. Switching to the ‘internal’ perspective, by asking what unifies different phono- logical grammars of English, begs the question of what qualifies as a dialect of English. It is probably true that the parametric settings just mentioned hold for Epilogue 271 the vast majority of systems to which the label English might uncontroversially be attached. Rather less certain is the status of systems which have predominantly English lexical stock but whose genetic affiliation to the language is at best ‘obscure. These are precisely the cases with divergent settings on the parameters in ‘question. Many anglophone creoles and pidgins, for example, display an absence of one or more of the following characteristics: complex onsets, contrastive vowel length, closed rhymes, and word-final consonants. Each of these reflects an OFF setting on the relevant parameter. The main role of phonological representations is to provide addresses for lexical storage and retrieval. Much of the phonological content of individual lexical entries is perforce idiosyncratic, affirming the arbitrariness of the sound-meaning pairing. However, the overall design of the representations in terms of which entries are structured reflects quite general properties of the human language faculty. In other words, these properties define the set of possible lexical ad- dresses. Each language exploits a subset of these addresses, the selection being circumscribed by the particular combination of settings the system opts for on a small number of phonological parameters. As to the matter of derivation, the null hypothesis is that itis constrained by the same principles as shape the design of lexical representations. Within an authen- tically generative model of grammar, phonological derivation does no more than define the distributional and alternation regularities that hold over phonological representations, both in the lexicon and in larger syntactic frames. It does not, for example, serve the extra-grammatical function of preparing representations for submission to articulation and perception. Within a full-blooded principle-based approach, phonological derivation is in large part automatic. An automatic derivational effect is one that, in response to universal constraints, is the inevitable consequence of some combination of conditions obtaining within a representation. The conditions are primarily pros- dic, the effects primarily melodic. Each of the latter takes the form of a decision about whether the prosodic licensing status of a particular position enables it to support the phonetic interpretation of a given element ct. In short, all phonolo- sical processing can be distilled into the universal schema license a. The massive ‘over-generation that flows from this generalization is stemmed by the interven- tion of general constraints which, among other things, ensure that there is a host of prosodic conditions under which cannot be licensed. That said, it needs to be stressed that the emerging model stops short of being fully deterministic. It still possesses a certain degree of ‘slack’, permitting situ- ations in which more than one derivational outcome is well-formed. This is as it should be. For example, any analysis of tlenition in English has to make allow- ances for the fact that different dialects opt for different weakened reflexes. Nevertheless, the number of melodic options that are attested in such cases amounts to no more than a handful. The challenge is to ensure that the set of possible derivations defined by our model matches this observation as closely as. possible.272 Epilogue In sum, phonological differences between grammars come in three forms. Some are systematic and reflect divergent permutations of parametric settings. Others may take the form of differing choices from a limited set of derivational outcomes, allowed for by particular combinations of parametric settings and universal constraints. The remainder are a matter of idiosyncratic distinctions in the plo- nological content of individual lexical entries. None of these differences is the preserve of specifically ‘dialect’ as opposed to ‘language’ variation. The main value of focusing on the former lies in the opportunity it affords us of investigat- ing the scope of cross-grammat differences in laboratory-like conditions. Notes Chapter 1 Sounds and Words 2 10 n 2 13 4 as 16 For an overview of developments inthis model, in the eration of Chomsky (1365, 1981), se van Riemadie and Willams (1986: 1709 The evel mance imerpreation of 2 sentence, when werd in context, of course depends on more than the purely grammar-internal factors mentioned here. Exra- trammatcal factors eating tothe real-world sting in which the sentence is urtered atualy have « significant role to play in this regard. For one approach co the ‘Fealsemantis issue, see Sperber and Wilton (1986) and Kempson (1988), This nigh perhaps mone commonly associated with de Sausure (1916) In the phonological lteratare ofthis Century, the “henvand-procss' mods! described here hus been associated with generative phonology since it earliest days (8. Halle 1959), Earlier peecedens are to be found inthe writing of, amongst others, Bloomfield (0933). ‘Analtenative analysis under which the lexical presentation of end) is sen, ith the den being ase pre-vocaicaly i ued out forthe reason tha it woud insert Seer al wfnal morphemes. This would result in very large number of erroneous derivations, uch as "sand fr (an) For more derailed look a dae differences relating to Cy clusters, see exercise 2at the end of chapter 2. See van de Huls and Smith (1982) for discussion ofthis development. Foran excellent summary of the ordered-rle approach, see Kenstowict a (19739). See van Rizmsdik and Wiliams (1986) fora summary of early rle-orented workin fencratve sya and for adiscusion of more recent reactions aginst it. Textbook ‘presentations of the classic SPE mode! of phonology include those of Schane (1973), Hyman (1975), Kenstowicz and Kisseberth (1979), Katamba (1988), Durand (1990) tnd Care (1993). ‘Asin so van Riemsd and Wiliams (1986), Sec for example Chomsky (1992) and Brody (1992). Se especialy Bromberger and Halle (1988). Se in particular Kay, Lowenstamm and Vergnaud (1985, 1990) One ofthe earies exp declaration of this view i that of Kaye, Lowenstamm and Vergnaud (1985), See now also, for example, Kaye, Lowenseamm and Vergaaud (1990), Pine and Smeoensky (1993) and the papers in Goldsmith (1993) See Anderson (1982), Levin (1985) and the discussion i chapter 4. Full explications of the formal devices of the rewrite-rule model in phonology are to be found for example in Schane (1973) and Hyman (1975) Kisseberth274 Notes 17 For discussion of this point, se for example Kaye (1989: 588. 1B. SPE (ch. 9), Kean (1975). 19. Kaye, Lowenstamm and Vergnaud (1990) 20. Haegeman (1991: 16740), Manzini (1992) 21 On the notion of locality in phonology, see for example It (1986) and Archangeli and Palleyblank (1992: ch. 1). 22 Thisis the aucosegmental view of phonological representation, pioneered by Goldsmith (1976) and to be discussed in some detail in chapter 3. 23 See, for example, Kaye, Lowenstamm and Vergnaud (19851, Mascaré (1987), Yip (1988a), Piggott (1988), Avery and Rice (1989) and Archangeli and Pulleyblank (1992), 24 Kaye (1989: 894). 25_ See for example Chomsky (1981: ch. 1; 1986a). 26 Chomsky (1964: 64). 27. For an extensive critique of the over-generating properties of rule ordering in phono- logy, sce Pullum (1976). 28 On this point, see for example Kaye (1989: ch. 5}. 29. Chomsky (1981), 30. Parametric variation in phonology is one of the main themes of chapter 4. For explicit discussions ofthe topic in the phonological literature, see for example Hayes (1980), Ito (1986), Kaye, Lowenstamm and Vergnaud (1990) and Archangeli and Pulleyblank (1992: ch. 4). 31. Halle and Vergnaud (1987). 32. See especially the collection of articles in Beninca (1989). 33 The root and word terminology is traditional. See Selkirk (1982a) for detailed discus- sion and analysis, Roughly equivalene classifications use such terms as primary and secondary affixation (Bloomfield 1933) and Class-1 and Class morphology (Siegel 1974), 34 Allen (1978). 35. See for example Siegel (1974) 36. Siegel (1974). 37. Sproat (1985), Fabb (1988). 38 Borowsky (1989). 39° Myers (1987). 40 That is, such alternations are structure-preserving in the sease of Kiparsky (1985). ‘More on this notion in chapter 4. 41. See, for example, Halle and Mohanan (1985) and Borowsky (1986: 2320. 42. The facts in (24) are true of the majority of dialects. Other patterns of 9 distribution are attested, some of which we will consider in exercise 1 at the end of chapter 2. 43. The use of the term analytic in this specific sense is attributable to Kaye and Vergnaud (1990). 44 On the non-independence of domains created by affixation, see for example Kiparsky (19822). 45 For example, Kiparsky (1982a, 1985), Halle and Mohanan (1985), Mohanan (1986), Pulleyblank (1986), Booij and Rubach (1987). For a summary and references, see Kaisse and Shaw (1985). 46 For example, Kaye and Vergnaud (1990). 47 For ponderivational approaches to rootleve alternations, st for example Lieber (1982), Marantz (1982) and Kaye and Vergnaud (1990). Notes 275 48. For discussion of external evidence bearing on the representation of root-level alterna Son in Engh ce for cample Jagr 1986), Mecaoy 500 aa a Derwing (1986). O86) and Wang and 49 On this point, sce Kahn (1976: 8 50 Well (1982: 310-19) ‘51 Wells (1982: 31260). 52. Haris (19904) 53 As well as summarizing the main developments, Harris (1989) provides ‘ewe sunmaing pment, Haris (1989) provides references to 54 For detailed descriptions ofthese particular systems, (1972) (New York City) and Harris (1985) (Bellas), $5 Forth dea of Ate Lay se Aitken 1981) 2nd Lass (1974. Lowland Scot, a Germanic relative of English, has 4 long history of independent development in codand Scotsh English eles io those dacs tha bane ene oe a tesuleof she more ecent importation of English For adscuston of he enc between the two languages, see Aitken (1984a).. ° S6 One morpologsl envy of ike’ Low fr example Hac McMahon (1991), pie Hasse (19708) and see Labov, Yaeger and Steiner Chapter 2 Constituency 1 Jones (1989: 15 provides a discussion of the historical deal aswell ate the classic literature on the topic. ss references to 2 Technically, deletion would be said bleed lengthening if ed lengthening iit applied st. To derive the correct result, the opposite order has tobe impose, in which ose thecal ane een stand ina counter-leding relation. The best discussion of ralcondenne enue. probably that of Kenstowiz and Kisseberh (1979: ch, 8) 3 The erature contains various proposals for esablhing puicowise elation between linear els, such a those described here, which eden conspne none ten eutcome. According to one suggestion, x particular principe of ie cadena Tule ordering favours counter-bleedng elation as outlined in note 2) over bleeing reanon eee (1982b: ch. 4) for discussion and references. , an ad 4 Inits orga incarnation, the seca or ning : Or ining ie was represented asa sequence of Gand los (McCarthy 1979, Clemens and Key 1983) Under tha rope ak slot contained a value forthe feature [syllabic]: minus for C plus for V. Belore lone was acknowledged hat hi mode of tersetaton onetoed Tones ore fsdundany, sine the sylabi tatu of postion was coded ace once tee nt for illabic), and again in constvent stecure.Siees af we wil se nent are independent grounds fr retaining constituency, the conchaion oat ec featral characterction of sylabicity could be dapensed with siopether Thay 5 Th scar aang memprony scale nn an ee he dase stsegmena weamen of tone onignaly Slog by ae Goldsmith (1976). “my aaa 6 Mecarty (1379, lle nd Vergaud (1980), Clements nd Keyer 983), 7 For seal teainens of compensorylenhenng eae Laon, ee for examle Strade (1982), Clemens and Keer (aan Sm nels for276 Notes 2 13 4 1s 16 7 18 19 20 241 2 2B ‘Werzels and Sezer (1985). More recent treatments of the phenomenon, based on the mora rather than the skeletal point as the fundamental unt of timing, include that of Hayes (1989). ‘The terms and theic associated concepts ate originally due to Goldsmith (1976). This isthe Linkage Condition discussed by, among others, McCarthy (1979), James Harris (1980) and Goldsmith (1990: 52-3). De Chene and Anderson (1987). ‘The no-crossng cffect was originally obtained by means of a phonological constrain, the Well-Formedness Condition of Goldsmith (1976: 48). Sagey (1988), however, shows that it can be derived from the more general logical principles referred to here (see ako Bird and Klein 1990). “These examples illustrate primary final stress. The same patera is evident in final open syllables with secondary stress, eg. meralmerow (marsow), bele/beley (bale) Kaye, Loweastamm and Vergnaud (1990). Thanks to Nel Smith forthe (cach i-(ca shit) minimal pai. The term is due to Hayes (1982). For discussion and further references, see Hogg and McCally (1987: 106i). ‘There are apparent counterexamples ia, for instance, (detach, attéch, allege). However, these forms have final stress for the same reason as, say, (deter, abit, permit, conc, ‘compel: the intial syllable coincides with an unstressable prefix: On this point, see SPE (0. 9). For an catly non-linear analysis of contour tones, see Leben (1973). Other dynamic segment-rypes which can be treated in the same contour terms include short diphthongs (ecg. ya, a, as in (a)) and pre-nasalized stops (nd, mb, ec, asin (b)) (x (b) x WN 1N ya ad ‘Type. (a) occurs, for example, in French (eg. irwa (trois) three’), type (b) in, for ‘example, Venda (southern Bantu, South Africa, e.g. mbudeigcat’). Neither structure is arrested in English For discussion, sce van der Hulst and Smith (19823). Early attempts to integrate syllable struccure into generative phonology include those of, Vennemann (1972a), Hooper (1972) and Kahn (1976). ‘These facts have been widely discussed in the literature on word stress. See, for example, SPE (pp. 71 ff), Liberman and Prince (1977) and Kaye (1989: 79 () See SPE (ch. 3). (On the evidence for the onset-rhyme split, see for example Piks and Pike (1947), Kiury- fowice (1948), Pike (1967), Fudge (1969, 1987), Selkirk (1982b) and James Harris (1983). Ik is sometimes claimed that a certain type of reduplication process targets (‘circum- scribes’) a syllable-sized constituent. A typical example is provided by Tagalog, where the recent perfective prefix is a CV template in which the segmental material is a copy of the initial CV portion of the base to which itis artached. The inflected form of the base palit ‘get mach, for instance, & ka-ga-galit ‘ust got mad? (che reduplicative prefix emboldened). McCarthy and Prince (1986) treat this as the prefixation of a syllable (1986). However, see now Kaye (1991) for an alternative analysis which dispenses with all reference to the syllable node and makes use of the notion of licensing (to be discussed in chapter 4). 4 2s 26 27 28 29 30 3 32 3 35 36 37 38 39 40 a 2 4B Notes 277 Reasons for dispensing with an independent syllable node are discussed by Aoun (1979) and Kaye, Lowenstamm and Vergnaud (1990), ‘This tradition is evident in a number of writing systems, These include the various Beahamt-derived scripts of India, used for writing a wide range of Indo-European, Dravidian and other languages. One example ic the Devanagari script, the system woed for Sanskrit and various modern languages such as Hindi, Nepali and Marwari. The ‘same basic principle underlies the Ethiopic script, the original Ge'ez system which has since been adapted for Amharic, Oromo, Tigr, Tirgina, and other modern languages of Ethiopia. (For a useful summary, see Coulmas 1989.) ‘An example from a widely used typists” manual: ‘words should be divided between syllables, e.g. (win-dow, pic-ture). Many words can be divided in a number of ways. For example, the word (qualifying) could be divided: (qualifying, qualifying)’ (Mackay 1982: 33), See the references in note 22. (On the phonotactic evidence for the onset-chyme split, see Fudge (1969, 1987), Selkirk (19826) and Kaye (1985). For reasons be discussed presently, forms such as (catkin) cannot be considered counterexamples to this generalization. Such a uniform characterization is impossible if positions are assumed not to be syllabic skeletal slots but rather morae assigned solely on the basis of chyme weight (Hyman. 1985). In moraic theory, onset segments are adjoined either to the first mora of a syllable (c.g. Zec 1988) or directly to the syllable node (e.g. Hayes 1989). The exclusive ‘concentration on rhyme-internal quantity in this approach precludes a unified account ‘of phonotactic dependencies, since these also involve extea-thymal relations, namely those illustrated in (21a) and (2c). There is thus no single dimension with reference t0 ‘which phonotactic restrictions can be stated. The relations involved can be (a) between ‘non-moraic melody units (within the onset); or (b) between morae (within the nucleus); for (c) between moraic and non-moraic material (coda-onset clusters) These assumptions are rarely made explicit. On the reasons for doubting their validity, see Kaye, Lowenstamm and Vergnaud (1990), On this point, see especialy Selkirk (1982b}, Chomsky (1964). For example O'Connor and Trim (1953), Abercrombie (1967: ch. 3), Kreidler (1989: 77 ff.) and Giegerich (1992: 137 Ef.) This formulation adapted from Clements and Keyser (1983: 37). See also Selkirk (1982b: 359). ‘The relevance of Italian vowel length to the syllabification of medial sC clusters is discussed by Kaye, Lowenstamm and Vergnaud (1990), ‘The notion of sonority, under various guises, has a long history in phonological scholarship. See for example Jespersen (1913), Saussure (1916), Jakobson and Halle (1956), Vennemann (19722), Steriade (1982), Selkirk (1984), However, various proposals have been made to the effect that language-specific sub- hierarchies exist within each of the classes defined in (26), e.g. Stetiade (1982), Levin (1985), Jakobson and Halle (1956), Selkirk (1984: 116), Vennemann (1988). Selkirk (1984), Levin (1985), Steriade (1982), Levin (1985), Vennemann (1988). For a reanalysis ofthe Polish clusters just mentioned, see Gussmann and Kaye (1993). Cf. Clements and Keyser (1983: 41)278 Notes clue rappers in some rypeno Azan America El Wels 1962: 558) and ean on ural ate of England oe the enres under he) (ten IV. 5.2 inthe Sarvey °F eg Dacre (Orton and Barty 190, Orton and Ting 190) 15 Ce hcruat ste smiation ely operates within he ot For pest pa co aed nena eins roup compos ether of) + oe seed shies or) seed sable followed by a estesed sab (sonra chap 4) he al ois one ui oa Seeenins task epe The genratcaon soot ve snsiilaton beeen we ds hl he ey ocr separ ee Ths ty he ed an de afr composed of or eso ant dove There ata fw b'ai/i ‘devotion’, dab.ba > da.bba ‘tin can’. In other words, after resyllabification there are essentially no restrictions on what constitutes an ‘onset in the language. A principle-based approach to the same set of facts would call for a radically different analysis. Kiparsky (1985), Borowsky (1986: 12 H), Myers (1991b) ‘On this point, see Kaye (1990a) and Harris (1992b). See for example van Riemsdijk and Williams (1986: 251 ff. The application of the Structure Preservation Principle to prosodic licensing is origi ally due to Selkirk (1982b). Kaye, Lowenstamm and Vergnaud (1990) express the notion in terms of the phonological instantiation of the Projection Principle, In the phonological literature, the term structure preservation has also been used in the rather different sense of imposing restrictions on the combinability of feature values within a single melodic expression, without regard to sequential position (eg. Kiparsky 1985). ‘We touched on this notion in 1.4.2. ‘This view is explicitly espoused by, for example, Ib (1986), Other examples include Moroccan Arabic (Kaye 1990b), French (Charette 1991) and. Polish (Gussmann and Kaye 1993) Kaye (19906), Charette (1991), ‘Mohanan (1985). This is essentially the analysis proposed by Brockhaus (1992) for the same general pattern in standard German. For a useful summary and discussion of the ECP, see Haegeman (1991: ch. 8). On the phonological implementation of the principle, see especially Kaye (1990b) and Charette (1991), More accurately, itis a particular type of empty category that blocks contraction, namely PRO. For discussion and analysis of this phenomenon, see van Riemsdijk and Williams (1986: 149 £9. Data on Systems A and B are drawn from Leslie (1983) and Harris and Kaye (1990). Material relating to System C is widely available in the literarure (e.g. Kahn 1976, Kiparsky 1979, Selkirk 1982b, Hacris and Kaye 1990). System-D data are from Harris (19909). See Leslie (1989) on this point. ‘The merger is by no means complete, since in some cases a durational difference in the preceding vowel provides evidence of the original distinction. This is illustrated by the famous (rider-writer) example; in most tapping dialects, the former word retains a relatively longer stressed vowel than does the latter. See for example Wells (1982: 260 ff Thisis true of London Vernacular English (Sivertsen 1960). For discussions of how Flenition operates differently within foot-level and word-level prosodic domains, see Leslie (1983) and Harcis and Kaye (1990). On this pattern of stress shift, see for example Liberman and Prince (1977). Some System-A dialects show the effects of a special development whereby word-final ‘alternates with approximant r (identical to that in (berry)) after a short vowel, e-g- gel] off). See Wells (1982: 370).292 Ey 3 84 86 87 88 9 90 1 2 93 4 95 96 97 98 99 100 Notes “The use of this conjunction is typical of a large number of linear analyses of various phenomena. For examples sce the references in James Harris (1983: ch. 3). For the arguments, se for example Kahn (1976) and James Harris (1983). ibid. Coda Capture isin all essential details equivalent ro Kahn's Rule Ill (1976: 48), which forms part of his analysis of tapping in a system of type C. Adaptations of his approach include Leslie's (1983) account of glocalling systems. Gussenhoven (1986) {teats English oral stop allophony in similar terms. ‘This is equivalent to Kahn's Rule V (1976: 53). Kahn’s versions of these rules are respectively Flap and Glottalization (1976: 99-100). Although Kahn's (1976) characterization of syllable structure is not identical to that sven here, the differences do not materially affect the presentation of his analysis. For ‘example, his representations contain syllable nodes lacking internal constituency. For arguments against ambisyllabiity, se for example Selkirk (1982b). ‘The spirit of this analysis is already present in Stampe (1973). The execution of the analysis, as well as its extension to other aspects of English plosive allophony, is essentially the work of Selkirk (1982b), subsequently adopted by others including Borowsky (1986). In its essentials, this isthe version formulated by Borowsky (1986: 265). ‘This cule is not actualy formalized by Selkirk (1982b) but can be extrapolated from her discussion. Selkirk (1982b: 373). ‘See Harris (1992b) for a summary and discussion of some ofthe classic examples. ‘A case in point is Lardil (Australi), frequently cited as a language in which non-coro- ‘nal consonants fail to appear in either internal codas or final position. This, ithas been ‘proposed, can be accounted for by a single condition which bars the feature (-coronal] from associating to a coda position (Itd 1986: 84 ff). In fac, the sets of consonants permitted in each context are not identical. Internal codas are restricted to (coronal) liquids (as in pinen ‘woman’, relka *head’) or nasals which, if not coronal, must be place-linked to a following onset (e.g, kupka ‘groin’, Gempe ‘mother’s father’) (dara from Kenstowicz and Kisseberth 1979: 109ff). Final position also only permits coro- nals; however, this set includes not only liquids and nasals (eg. yalul ‘flame’, yaraman ‘horse but also, unlike internal codas, plosives (e.g. yarput‘suake, bird’). This pattern certainly demonstrates the special status of coronals as wel as the special autosegmen- tal licensing properties of the two contexts. However, it fails to indicate that the contexts necessarily have the same constituent status. See Vergnaud (1982) and Gussmann and Kaye (1993) for ciscussion. Formally, this “operation is somewhat akin to a notion made available in other frameworks, in which ‘a contonant is extra-metrical whenever it occurs domain-finally (as in (get)) but loses its extra-metrical status once i¢ appears domain-internally (in this case within the phrase containing (get a). See for example Ito (1986). ‘Another widespread hiatus-filler isthe glottal stop, asin (coo (Pearly. This example belongs to a class of phenomena, referred to in 3.3.5, which present a challenge to the view that all processes have a local cause. Specifically, they involve the apparently spontaneous appearance of melodic material which has no obvious local source. Why it should be ? rather than some other element that emerges in the onset context is something of a mystery ‘Thanks to Nikee Ola for the Yoruba data. See for example Charette (1991: 90). ‘This discussion of Licensing Inheritance is based on Harris (19928). Notes 293 101. Fora somewhat similar use ofthe notion licensing path, see Goldsmith (1990: 113 ff. 102. It seems reasonable to consider vowel syncope an extreme form of reduction, This suggests that Proper Government, which defines the conditions under which syncope ‘occurs, must be viewed as independent of licensing between nuclei on the foot and word projections. Proper Government may be thought of as a mechanism which completely guts a position of is ability to license melodic material. 103. Luelsdorff (1975: 67-8). 104 This analysis is proposed by Brockhaus (1992), with exempliicaton from German, Polish and Catalan. 105. Thisis the proposal advanced, for example, by Borowsky (1986) 106 The following analysis develops the accounts proposed in Harris (1990) and Harris and Kaye (1990). 107 On Stray Erasure and the question of where it operates in derivation, see for example Steriade (1982) and I6 (1986) 4108. For discussions of this principle, se for example Mascaré (1976), Kiparsky (1982a) and Halle and Mohanan (1985), 109 The mechanism by which Ib (1986) derives the independence of failure to link and Stray Erasure is eather different to what is presented here. For her, the question of Whether a segments erased or retained on a given cycle depends on whether o not it is licensed by extra-prosodicty within that domain, 110 Lass and Anderson (1975: ch. 5) 111 Fasold (1972), Neu (1980). 112. Kahn (1976: 99 ff), Selkirk (1982b: 373), Leslie (1983). 113 The choice between optional and obligatory syncope is not peculiar to the context tnder discussion here, ic. one where the consonant on the lft ofthe syncope site is 1d, while that on the right isa resonant. The same patter i also evident in a form such as (every), in which syncope is obligatory for ome speakers and optional for others. 114 These are four of seven phenomena which Wells (1990) analyses in coda terms. 115 Data from Alleyne (1980) Chapter 5 Floating Sounds 1 4.1990 draft ofthis chapter circulated under the ttle “The most imperfect of consonants" 2. Walker (1791: 50), 3 Ibid (p. 50). 4 Accounts of A (my own native system) have been around in various forms for as long a5 there has been a scholarly tradition of English phonetics, 5 Bis the standard southern British system recommended in the classic pronunciation guides such as Jones (1956) and Gimson (1962). However, as these authors acknow- Tedge, most if not all native speakers of southern British standard actually use some version of System C. 6 Thave collected the System-C data myself, but they can easily be verified in any number cof works, including Wells (1982: 222 ff). For reports on intrusive rin the United States, see Kurath and McDavid (1961: ch. 2) and McCarthy (1991) 7 Non-rhotie System D lies at the ‘deep vernacular’ end of the Southern United States dialect continuum. The data are based primarily on my own observations, supported by reports by, among others, Bailey (1969) and Wells (1982),294 Notes 8 Detailed information regarding the social distribution of competing rhotic and non- thotic variants is available for several areas. See especially La3ov’s (1966) study of New York City. For summaries of the regional distribution of this variable in the eastern United States, see Kurath and McDavid (1961); for England, see Wells 1982 (vol. 2) 9 Wels (1982: 367 ff) 10 Sledd (1966). II Insome dialects, rails to show up in onsets occupied by @. Hence forms such as (thfec, tivow. This is true, for instance, of some otherwise System-D dialects in the southern United States (Bailey 1969: 189) as well of some (System-A) dialect in Ireland. 12. This is by no means true of all non-rhotic dialects. In some Caribbean varieties, the in-gliding io vowel of BEARD can merge with that in BADE. This isthe pattern in, for example, non-thotic Jamaican English (Cassidy and LePage 1967). In some north: eastern United States dialects, there is a potential merger beween the BEARD nucleus and the in-gliding reflex of historical short # in BAD (Labov, Yaeger and Steiner 1972). 13 For early discussion of this issue, see Sledd (1966). 14. The type of case illustrated in (9d) is apparently only possible when a close pragmatic or stylistic relationship holds between the two sentences. For discussion of the prosodic domains within which rsandhi operates, see Nespor and Vogel (1986: 234 ff), 15. According to Wells, the appearance of intrusive rafter the PAW vowel is more likely to raise hackles than it does in other contexts (1982: 225). It may be necessary to recognize 4 variant of System C in which intrusive r occurs after x Kmafr] and pa)) or schwa (Gdeafe] of) bue not after the PAW vowel (*(saw{¢] it). On this point, see Gimson (1962: 204), 16 In its essential details, this isthe rule proposed by, among others, Kahn (1976: 23), Wells (1982: 218 ff) and Gussmann (1980: 38). 17 Kahn (1976: 23). 18 For example, Sledd (1966). 19, Wells (1982: 557). 20 This is essentially the rule proposed for System-B dialects by, among others, Kahn (1976: 109) and Mohanan (1986: 36). 21 Kenstowicz and Kisseberth (1979: ch. 3). 22 Mobanan adopts a more restricted form of Onset Capture which applies only to r (1986: 36). 23. The wrong result here is due to the counter-bleeding order of R-Dropping and Onset Capture. The desired ordering relation would be one of bleeding; in applying first, ‘Onset Capture would remove forms such as (barring) from the purview of R-Dropping. 24 Cis, however, like B in displaying the contrast at the end of certain roots; compare Asphefe-sphe(rical) with (formula-formula{ lic). More on this anon. 25 Nespor and Vogel scem to ignore this point when they claim that rsandhi in both System B and C can be tceated as rinsertion (1986: 228). Such an arrangement derives intrusive r correctly in C but erroneously in B. 26 Some have argued that deletion and insertion co-exist within an intrusive-r grammar. ‘According to one version of this view, these take the form of independent rules ~ R- Dropping, which derives the zeto variant of etymological linking r, and rinsertion, which derives unetymological intrusive r(e.g. Gussmann 1980: 34 ff, Lodge 1984: 12 f). This proposal seems to be founded on the erroneous assumption that linking ris ‘obligatory, while intrusive r is optional. In fact, speakers who variably suppress unetymological rare also observed to suppress the etymological variant. This indicates, that che two reflexes are non-distinct for such speakers. 7 28 29 30 31 2 3B 4 35 37 38 Notes 295 According to an alternative proposal (McCarthy 1991), deletion and epenthesis covoceur in intrusive-r dialects not as autonomous rules but as different effects of a single symmetrical rule (‘alternates with ). The selection of alternants is guided by separate output constraints. One of these filters out coda r another, which amounts t0 ‘bar on hiatus, is satisfied by the appearance of sandhi r. The etymological provenance of the latter is irrelevant under this analysis. The formulation of the antihiatus ‘constraint is, however, somewhat problematical (se note 41). ‘This argument is widely peated in the literature; ee, for example, Wells (1982: 222 ff) and Tradgill (1986: 71). ‘This is essentially the formulation given by, for example, Wells(1982: 226) and Nespor and Vogel (1986: 228-9). In intrusive-rdialects, the contrast between historically rfl and rless forms survives in ‘a number of cootlevel alternations. Thus r occurs in one alternant of pairs such as Aephese-spheftlical, explate-explofrlation, cleaf—clealrance) but notin, say, (Formula formulaic, algebra~algebraic). The insertion analyst runs into trouble, if it assumed that root evel alternations are derived by phonological process. As formulated in (23) ‘oF (24), the rinsertion rule would be unable o deal with alternations such as (sate satirical, admife-admifejation,aspite-aspfrJaion). In these cases, discussed by Guse mann (1980: 34 ff, the vowel preceding the supposedly epenthetic + i high, not non-high as required by the rule. ‘Aversion ofthis rule is provided by Kahn (1976: 111) Broadbent (1991), McCarthy (1991). A rather more appealing alternative isto assume that sandhi rhas its source in the preceding non-high vowel itself. According to one such proposal, ris the onset manifes: {ation of an expression consisting of head @ and dependent A, the latter spreading from 2 preceding nucleus (Broadbent 1991). The floating teatment of English rsandhi isin the spirit of related analyses proposed for other languages. The idea that a melodic uni can lack association to a skeletal point ‘was originally applied to che phenomenon of floating tones (se, for example, Gold: smith 1976). Closee precedents for the analysis to be presented here ae the treatments of French liaison consonants developed by Vergnaud (1982), Piggott and Singh (1985), Prunet (1986), Encrevé (1988) and Charette (1991). In Charete's analysis, for example, the form kafe (chachet ‘seal (soun} contains a final floating ¢ which is audible when i attaches toa following onset, asin kafte (cachet) ‘o seal (1991: 188 i). ‘As proposed by Vergnaud (1982) ‘The autosegmental docking metaphor is originally due to Goldsmith (1976) ‘The notion that, in non-rhotic systems, etymologically less forms have assimilated to the lexically rful patter (rather than vice versa) isto be found, for example, in Gimson (1362: 92), Lindsey (1990) and Scobie (1992) ‘Vennemann (19728), Wells (1982: 222 ff, Trudgill (1986: 71 f), MeMahon (1992). For a critique of historical rule inversion, sce McCarthy (1991) and Scobbie (1992). “There is certainly a widespread impression among the general public that intrusive ris a recent (and mostly unwelcome) innovation. An example: there is one recent develop ment which needs hiting on the head. I isthe revanchist expansion of our old enemy, the intrusive *r",... now pitching for end-vowels. This gives us most hideously the perversion of “draw” into “droc” and the subsequent scream-inducing “droring” * (Gdward Pearce, “Unspoken shifts in the spoken word’, The Guardian (Manchester and London), 3 October 1990). The notion that System Cis a recent offshoot of Bis taken for granted in much ofthe phonological literature; ee for example the references in the Previous note.296 Notes 39 Sheridan (1762: 34). 40 The phenomenon is sometimes teferced to as hyper-thoticty (Wells 1982: 221-2, ‘Trudgill 1986: 74 f), In such systems, the appearance of rin, say, (commale) is not restricted to sandhi contexts, which indicates that i as a lexcal source (as opposed to oti dialects are similar to non-rhotc intrusive-r 41. The proposal that System C is subject to some kind of word-steucture constraint is also made by McCarthy (1991). His solution takes the form of a filter which places aan embargo on final ‘vowels’ in intrusive-r dialects. The term vowel in his formula tion (*VIwe) is an informal reference to a nuclear position occupied by other than a high vocalic segment. In other words, it is a re-statement ofthe left-hand environment ofthe Epes 24),Wocd ending ior ara i atin dtm 0). The constraint triggers rinsertion in words ending in nor-high vowels such as 2, 2 fr 2 This analysis is only valid for dialects with upgliding diphthongs in DAY (e. ey) and GO (eg. ow). It cannot account for intrusive-r dialects in which the cor- responding nuclei are occupied by the mid monophthongs & and a (such as the northern English varieties described in 3.3.6). The second portion of these mono- phehongs, being non-high, qualifies as a ‘vowel’ in McCarthy's sens. It should thus be expected to give rise to intrusive r. This is not the case. As noted in 3.3.2, e: and 0: behave like ey and ow in yielding respectively y and w glides in hiatus; hence de: yov day of, so: w2 (60 a). 42 The contrastive possibilities diminish to one oF two vocal contrasts in unstressed contexts, as in the final vowels of words such as (rapid, abbor, ribbon). But this is a function of the prosodic weakness of the nucleus within the foot and has nothing to do with the presence or absence ofa following consonant. 43. For the details of this regularity, which figures in exercise V of chapter 1, see Aitken (1981) and Lass (1974). 44, Sarveys of pre-r vocalic systems, including those illustrated here, include Kucath and MeDavid (1961: ch. 4), Wells (1982: 153 ff) and Kreidler (1989: 52 ff). Some very approximate examples of the geographical spread of the diferent sub-systems in (40): (404) (i) Ireland, (ii) England, North America; (40b) (i) easteen New England, metro- politan New York, the Upper South, rural Ireland, (ii) some central areas of the eastern United States, (ii) western Pennsylvania and points west; (Cc) (i) Ireland, (i) parts of the southern United States, (ii) most of the northeastern United States apart from New England. 4S. Shedd (1966), Wells (1982: 213 ff). 46. Wells's (1982) survey of different vocal systems before historical rincludes a particu- larly clear discussion of this issu. 47 Fromkin (1973: 251-2), 48 Ibid. 49° More recent proposals for nuclear rinclude those of Selkisk (19826) and Mohanan (1986: 37 50 Ladefoged (1982: 78). St Sledd (1966). ‘52 The neutral element must be assumed to be the head of the expression defining dark approximant r. This follows from the clement’ inability to contribute to the phonetic interpeetation of any expression in which it occurs as a depeadent (see 3.3.5). 53 This is a formalization of an observation made, for example, by Kenyon and Knott (1953: xix-xx). For a related analysis, see Clements and Keyser (1983: 32). In non- 54 ss 56 Notes 297 thotic dialects, the cognate vowel in non-alternating contexts (as in bam (buen), where historical ris unrecoverable) should presumably be represented as follows: N iN ! WA ‘The idea that the migimal word can be prosodicaly defined is due to McCarthy and Prince (1986). According tothe interpretation ofthe notion given here, both CVV and CVC forms satisfy the requirement that words in English minimally contain wo nuclear postions. In a CVV word such at (se), the two postions occur within the same nucleus. A CVC word such as (pit is well formed for the saroe reason that disyllabc (pity) is. Both these words contain wo nuclear postions, the frst of whichis occupied bythe shore vowel «In the cas of (pity), the second nuclear positon i the one oceupi by the unstressed vowel. Inthe case of (pit) the relevant postion isthe parametteally licensed inal empty nucleus ‘Again se the surveys in Kurath and McDavid (1961) and Well (1982). ‘The dialec in question is the Atlanta system described by Sledd (1958), to whom I also ‘owe the title ofthis exercise (1966).References ‘Abercrombie, David (1967). Elements of General Phonetics. Edinburgh: Edinburgh Univer- sity Press. ‘Aitken, A.J. (1981). The Scottish vowel-length ule. In M. Benskinand M.L. Samuels (eds), ‘So Mony People Longages and Tonges: philological essays in Scots and mediaeval English presented to Angus Mcintosh, 131-57. Edinburgh: Middle English Dialect Project. ‘Aitken, A. J (1984a). Scots and English in Scotland. In Trudgill 1984, 517-32. Aitken, A.J. (1984b). Scottish dialects and accents. In Trudgill 1984, 94-114 Alen, Margaret (1978). Morphological investigations. Ph.D. dissertation, University of Connecticut. Alleyne, Mervyn C. (1980). Comparative Afro-American: an historical-comparative study of English-based Afro-American dialects of the New World. Ann Arbor: Karoma. Anderson, John M. (1986) Suprasegmental dependencies. In Durand 1986, 55-133. ‘Anderson, John M. and Jacques Durand (eds) (1987). Explorations in Dependency Phono- logy. Foris: Dordrecht, Anderson, John M. and Colin J. Ewen (1980). Introduction: « sketch of Dependency Phonology. In John M. Anderson and Colin J. Ewen (eds), Studies in Dependency Phonology. Ludwigsburg Studies in Language and Linguistics 4. Ludwigsburg: Univer- sity of Ludwigsburg. ‘Anderson, John M. and Colin J. Ewen (1987). Principles of Dependency Phonology. ‘Cambridge: Cambridge University Press. ‘Anderson, John M. and Charles Jones (1974). Thee theses concerring phonological repres- entations. Journal of Linguistics 10. 1-26. ‘Anderson, John M. and Charles Jones (1977). Phonological Structure and the History off English, Amsterdam: North-Holland. ‘Anderson, Stephen R. (1978). Tone features. In Victoria Fromkin (ed.), Tone: a linguistic ‘anthology, 133-75. New York: Academic Press. ‘Anderson, Stephen R. (1982). The analysis of French schwa; or, hew to get something from nothing. Language 58, 534-73. ‘Anderson, Stephen R. (1985). Phonology in the Twentieth Century. Chicago: University of Chicago Press. ‘Aoun, Joseph (1973). Is the syllable or the supersyllable a constituent? MIT Working Papers in Linguistics 1. 140-8. Archangeli, Diana (1984). Underspecification in Yawelmani Phonology and Morphology. PAD dissertation, MIT. Published 1988, New York: Garland. ‘Archangeli, Diana (1988). Aspects of Underspecification Theory. Phonology 5. 183-205. “Archangel, Diana and Douglas Pulleyblank (1992). Grounded phonology. MS. University of Arizona and University of British Columbia. References 299 ‘Aronoff, Mark and Richard T. Oerhle (eds) (1984). Language and Sound Structure: studies in phonology presented to Morris Halle by bis teacher and students. Cambridge, MA: MIT Press, Avery, Peter and Keren Rice (1989). Segment structure and coronal underspecification, Phonology 6. 178-200. Backley, Pillip (1993). Coronal: the undesirable element. UCL Working Papers in Lingus ties $. 301-23, Bailey, Charles-James N. (1969). Introduction to Southern States phonetics: chapters 5 and 6. University of Hawaii Working Papers in Linguistics 6. 105-203. Barry, William and Martine Grice (1991). Auditory and visual factors in speech database analysis. Speech, Hearing and Language: Work in Progress $. 9-32. London: Depart ment of Phonetics and Linguistics, University College London, Bell, A. and J. B. Hooper (eds) (1978). Syllables and Segments. Amsterdam: North-Holland. Bendor-Samuel, J. T. (1960). Some problems of segmentation in the phonological analysis of Terena. Word 16. 348-55, Reprinted in Palmer 1970, 214-21. Beninca, Paola (ed.) (1989). Dialect Variation and the Theory of Grammar. Dordrecht: Foris. Bird, Steven and Ewan Klein (1990). Phonological events. Journal of Linguistics 26. 33-56. Bloomfield, Leonaed (1933). Language. New York: Holt, Booij, Geert and Jerzy Rubach (1987). Postcyclic versus postlexical rules in lexical phono- logy. Linguistic Inquiry 18. 1-44, Borowsky, Toni J. (1984). On resyllabification in English. In M. Wescoat (ed.), Proceedings of the West Coast Conference on Formal Linguistics 3, 1-15. Stanford, CA: Stanford Linguistics Association. Borowsky, Toni J. (1986). Topics in the lexical phonology of English. PhD dissertation, University of Massachusers. Borowsky, Toni J. (1989). Structure Preservation and the syllable coda in English. Natural Language and Linguistic Theory 7. 145-66. Beaine, Martin DS. (1974). On what might constitute learnable phonology. Language 50. 270-99, Broadbent, Judith (1991). Linking and intrusive r in English. UCL Working Papers in Linguistics 3. 281—302. Brockhaus, Wiebke G, (1992). Final devoicing: principles and parameters. PhD dissertation, University College London. Brockhaus, Wiebke G. (in press). Skeleton and suprasegmental structure within Govern- ‘ment Phonology. To appear in Durand and Katamba, in press. Brody, Michael (1992). A note on the organization of grammar. UCL Working Papers in Linguistics 4. 3-9. Bromberger, Sylvain and Morris Halle (1988). Why phonology is different, Linguistic Inquiry 20. 51-70. Carr, Philip (1993). Phonology. London: Macmillan, Cassidy, F. G. and R. B. LePage (1967). Dictionary of Jamaican English. Cambridge: ‘Cambridge University Press. ‘Charette, Monik (1989). The minimality condition in phonology. Journal of Linguistics 25. 159-87. Charette, Monik (1991). Conditions on Phonological Government. Cambridge: Cambridge University Press. (Charette, Monik (1992). Mongolian and Polish meet Government Licensing. SOAS Work- ing Papers in Linguistics and Phonetics 2. 275-92.300 References Chaterte, Monik and Jonathan Kaye (1993) The death of ATR. Ms, School of Oriental and ‘African Languages, University of London. Chene, E. B. de and §, R. Anderson (1987). Compensatory lengthening. Language 5S. 05-35. Chomsky, Noam (1964). Current Issues Linguistic Theory. The Hague: Mouton. Chomsky, Noam (1965). Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Fors Chomsky, Noam (19863). Knowledge of Language: its nature, origin and use. New York: Praeger. Chomsky, Noam (1986b). Barriers. Cambridge, MA: MIT Press. Chomsky, Noam (1992). A minimalist program for linguistic theory, MIT Occasional Papers in Linguistics 1 Chomsky, Noam and Morris Halle (1968), The Sound Pattern of English. New York: Harper and Row. Clements, George N. (1981). Akan vowel harmony: a nonlinear analysis. In George N. ‘Clements (ed), Harvard Studies in Phonology 2, 108-77. Clements, George N. (1985), The geometry of phonological features. Phonology Yearbook 2. 223-50. Clements, George N. (1987). Phonological feature representation and the description of intrusive stops. CLS 23, Part 2: parasesion on antosegmental and metrical phonology, 23-50. Clements, George N. (19918). Place of articulation in consonants and vowels: a unified theory. Working Papers ofthe Cornell Phonetics Laboratory 5. 77-123. Clements, George N. (19918). Vowel height assimilation in Bantu languages. Working Papers of the Cornell Poneties Laboratory 5. 37-76 Clements, George N. and Susan R. Hert: (1991). Nonlinear phonology and acoustic interpretation. Procedings of the XIItb International Congress of Phonetic Sciences, vol 1/5, 364-73, Provence: Universit de Provence. Clements, George N. and Elizabeth V. Hume (in pres). Segment structure. To appear in Goldsmith in press. Clements, George N. and Samuel Jay Keyser (1983). CV Phonology: a generative theory of the syllable. Cambridge, MA: MIT Press. Cocnya, Wiliam S. (1964), Outline of Birmese grammar. Language 20, Supplement Language Dissercations, vol. 38. Coulmas, Florian (1989). The Writing Systems of the World. Oxford: Blackwell. den Dikken, Marcel and Harcy van der Hols (1988), Segmental ierarchitecture, In van der Holst and Smith 1988, Pare, 1-78. Dobson, E. | (1968). English Promunciation 1500-1700. vol. 2, 2nd ed, Oxford Claren- don Pres. : Durand, Jacques (ed) (1986). Dependency and Non-Linear Phonology. London: Croom Heim, Durand, Jacques (1990). Generative and Nos-Linear Phonology. Harlow, Essex: Longman. Durand, Jacques and Francis Katamba (eds) (in press). Frontiers of Phonology Atoms, stracturesand derivations. Harlow, Extx: Longman Einarsson, S. (1945). Icelandic Grammar, Texts, Glossary. Baltimore: John Hopkins Press. Ekwall, Eilert (1975). A History of Modern English Sounds and Morphology. Translated and edited by Alan Ward. Oxford: Blackwell Encrevé, P. (1988). La Liaison avec et sans enchainement: phonologie tridimensionnelle et usages du francais. Paris: Editions du Seuil. References 301 Ewen, Colin J. (in press). Dependency relations in phonology. To appear in Goldsmith, in ress Ewen, Colin J. and Harry van der Hulst (1988). [high], [low] and {back] or [1], (A) and [U}. Im P. Coopmans and A. Hulk (eds), Linguistics in the Netherlands 1988, 51-60. Dordrecht: Fors. Fabb, Nigel (1988). English suffixation is constrained only by selectional restrictions. Natural Language and Linguistic Theory 6. 527-38. Farid, M. Onn (1980). Aspects of Malay Phonology and Morphology: a generative ap proach, Bangi: Universiti Kebangsaan Malaysia. Fasofd, Ralph (1972). Tense Marking in Black English, Aslington, VA: Center for Applied Linguistics. Ferguson, Charles (1972). ‘Shore ain Philadelphia English. In M. E. Smith (ed), Studies in Linguistics in Honor of George L. Trager, 259-74. The Hague: Mouton. Fischer-Jorgensen, Eli (1975). Trends in Phonological Theory. Copenhagen: Academisk Forlag. Foley, James. (1977). Foundations of Theoretical Phonology. Cambridge: Cambridge University Pres. Fourakis, M. and R. Port (1986). Stop epenthesis in English. Research in Phonetics and Computational Linguistics $, 37-71. Bloomington, IN: Indiana Univers Fromkin, Victoria A. (1973). A sample of speech errors, In Victoria A. Fromkin (ed.), Speech Errors as Linguistic Evidence, 243-69. The Hague: Mouton. Fudge, Eric C. (1967). The nature of phonological primes. Journal of Linguistics 3. 1-36 Fudge, Eric C. (1968). Syllables. Journal of Linguistics 5. 253-86. Fudge, Eric C. (1984). English Word-Stress. London: Allen and Unwin. Fudge, Erie C. (1987). Branching structure within the syllable. Journal of Linguistics 23. 359-77. Fujimura, Osamu and Julie Lovins (1978). Syllables as concatenative phonetic units In Bell and Hooper 1978, 107-20. Giegerich, Heinz J. (1985). Metrical Phonology and Phonological Structure: German and English. Cambridge: Cambridge University Pres. Giegerich, Heinz J. (1992). English Phonology: an introduction. Cambridge: Cambridge University Pres. Gimson, A.C. (1962). An Introduction to the Pronunciation of English. London: Arnold, Goldsmith, John A. (1976). Autosegmental Phonology. PhD dissertation, MIT. Published 1979, New York: Garland. Golfsmith, John A. (1981). Subsegmentals in Spanish phonology: an autosegmental ap- ‘roach, In W. W. Cressey and D. J. Napoli (eds), Linguistic Symposium on Romance Languages 9, 1-16. Washington, DC: Georgetown University Press. Goldsmith, John’A. (1985). Vowel harmony in Khalkha Mongolian, Yaka, Finnish and Hungarian. Phonology 2. 251-74. Goldsmith, John A. (1989) Licensing, inalterability and harmonic rule application. CLS25. 145-56. Goldsmith, John A. (1990). Autosegmental and Metrical Phonology. Oxford: Blackwell. Goldsmith, John A. (ed.) (1993). The Last Phonological Rule. Chicago: University of Chicago Press Goldsmith, John (ed) (in press. A Handbook of Phonology. Oxford: Blackwell. Guerssel, Mohand (1990). On the syllabification pattern of Berber. Ms. Université du Québec & Montréal.302. References Gussenhoven, Carlos (1986). English plosive sllophones and ambisyllabicity. Gramma 10. 119-41. Gussmann, Edmund (1980). Introduction to Phonological Analysis. Warsaw: Pafstwowe ‘Wydawnicewo Naukowe Gussmana, Edmund and Jonathan Kaye (1993), Polish notes from a Dubrovnik café, Ms. ‘School of Oriental and African Studies, University of London. Haegeman, Liliane (1991). Introduction to Government and Binding Theory. Oxford: Blackwell Halle, Morcis (1959). The Sound Patter of Russian. The Hague: Mouton. Halle, Morris and G.N. Clements (1983). Problem Book in Phonology. Cambridge, MA: MIT Press. Hille, Morris and K. P. Mohanan (1985). The segmental phonology of modern English Linguistic Inquiry 16. S7-116. Halle, Morris and Kenneth N. Stevens (1971). A note on laryngeal features. Quarterly Progress Report of the Research Laboratory of Electronics (MIT) 101. 198-213. Halle, Mortis and Jean-Roger Vergnaud (1980). Thrce-dimensional phonology. Journal of Linguistic Research 1. 83-108. Halle, Morris and Jean-Roger Vergnaud (1987). An Essay on Stress. Cambridge, MA: MIT Press, Harris, James (1980). Nonconcatenative morphology and Spanish plurals. Journal of Linguistic Research 1. 15-31 Harris, ames (1983). Syllable Structure and Siress in Spanish. Cambridge, MA: MIT Press. Harris, John (1985). Phonological Variation ond Change. Cambridge: Cambridge Univer- sity Press. Harris, John (1989). Towards a lexical analysis of sound change in progress. Journal of Linguistics 25. 35-56. Harris, John (19902). Derived phonological contrasts. In Ramsaran 1990, 87-105. Harris, John (19908). Reduction harmony. Paper presented at the 13th GLOW Phonology ‘Workshop, London. Hares, John (1990c). Segmental complexity and phonological government. Phonology 7. 255.300. Harris, John (19922). English goes west: on the tail of short « Ms. University College London. Harris, John (1992b). Licensing Inheritance. UCL Working Papers in Linguistics 4. 359- 406. Harris, John and Jonathan Kaye (1990). A tale of ewo cities: London glotalling and New ‘York City tapping. The Linguistic Review 7. 251-74. Harris, John and Geoff Lindsey (1992). Segmental decomposition and the signal. Paper ‘presented atthe 7th International Phonology Meeting, Krems, July 1992. Harris, John and Geoff Lindsey in press). The elements of phonological representation. To ‘appear in Durand and Katamba, in press. Harr, John (1569). An Orthographic, Conteyning the Due Order and Reason Howe to Paint Thimage of Mannes Voice, Most Like to the Life or Nature. Reprinted 1969, English Linguistics 1500-1800, Menston: Scholar Press. Haudricourt, A-G. (1961). Biparttion et tipatition des systémes de tons dans quelques langues d'Extréme-Orient. Bulletin de la Société de Linguistique de Paris S6. 163-80. Hayes, Bruce (1980). A Metrical Theory of Stress Rules. PhD dissertation, MIT. Published 1981, New York: Garland. Hayes, Bruce (1982). Extrametrcalty and English stress. Linguistic Inquiry 13. 27-76. References 303 Hayes, Bruce (1984). The phonetics and phonology of Russian voicing assimilation. In ‘Aronoff and Ochrle 1984, 318-28. Hayes, Bruce (1986). Inalterability in CV phonology. Language 62. 321-51 Hayes, Bruce (1989). Compensatory lengthening. in moraic phonology. Linguistic Inquiry 20. 253-306 Hayes, Bruce (1990). Diphthongisation and coindexing. Phonology 7. 31-72. Hayward, Dick (1984). The Arbore Language: a first investigation. Hamburg: Helmut Buske. Hogg, Richard and C.B. McCully (1987). Metrical Phonology: a coursebook. Cambridge: ‘Cambridge University Press. Hier, H. (1946). Tonkawa. In Linguistic Structures of Native America. New York: Viking ‘Fund Publications in Anthropology. Hooper, Joan B. (1972). The syllable in phonological theory. Language 48. 525-40. Hooper, Joan B. (1976). Word frequency in lexical diffusion and the source of morpho- phonological change. In W. Christie Jr (ed), Current Progress in Historical Linguistics, 95-108. Amsterdam: North-Holland. Hooper, Joan B. (1978). Constraints on schwa-deletion in American English. In Jacek Fisiak ed), Recent Developments in Historical Phonology, 183-207. The Hague: Mouton, Hulst, Harry van der (1989). Atoms of segmental structure: components, gestures and dependency. Phonology 6. 253-84 Holst, Harry van der (in press). Radical CV phonology. To appear in Durand and Katamba, in press. Hulst, Harry van der and Aone van Engelenhoven (1993). Metachesis effects in Tutukeian- Letinese. Paper presented at the Fist Holland Instirure of Linguistics Conference in Phonology, January 1993. Hulse, Harry van der and Norval Smith (19822). An overview of aurosegmental and ‘metrical phonology. In van der Hulst and Smith 1982b, Pare I, 1-46. Hulse, Harry van der and Norval Smith (eds) (1982). The Structure of Phonological Representations. Two parts. Dordrecht: Fors. Hulst, Harry van der and Norval Smith (1985). Vowel feacures and umlaut in Djingil, ‘Nyangumarda and Warlpti. Phonology 2. 275-302. Hulse, Harry van der and Norval Smith (eds) (1988). Features, Segmental Structure and Harmony Processes. Two pats. Dordrecht: Foris. Hyman, Larry M. (1975). Phonology: theory and analysis. New York: Hol, Rinchart and Winston. Hyman, Larry M, (1982). The representation of nasality in Gokana. In van der Hiulst and Smith 1982b, Pare I, 111-30 Hyman, Larry M. (1985). A Theory of Phonological Weight. Dordrecht: Foris Its, Junko (1986). Syllabic Theory in Prosodic Phonology. PhD dissertation, University of Massachussets. Published 1988, New York: Garland. Ito, Junko and R. Armin Mester (1986). The phonology of voicing in Japanese. Linguistic Inquiry 17. 49-73, Jaeger, Jeri J. (1986). On the acquisition of abstract representations for English vowel. ‘Phonology 3. 71-98. Jakobson, Roman (1939). Observations sur Je classement phonologique des consonnes. Proceedings of the Third International Congress of Phonetic Sciences. 31-41. Jakobson, Roman and Morris Halle (1956). Fundamentals of Language. The Hague: Mouton.304. References Jakobson, Roman, Gunnar M. Fant and Morris Halle (1962). Preliminaries to Speech ‘Analysis. Cambridge, MA: MIT Press. Jespersen, Otto (1909). A Modern English Grammar on Historical Principles, Part 1 sounds and spellings. London: Allen and Unwin. Tespersen, Orto (1913). Lehrbuch der Phonetit. Leipzie: Teubner. Jones, Charles (1989). A History of English Phonology. Harlow, Essex: Longman ‘Jones, Daniel (1956). An Outline of English Phonetic. 8th edn. Cambridge: Heffer. Kahn, Daniel (1976). Syllable-Based Generalizations in English Phonology. PhD disserta- tion, MIT. Published 1980, New York: Garland, Kaisse, Ellen M. and Patricia’ A. Shaw (1985). On the theory of Lexical Phonology. Phonology Yearbook 2. 1-30. Katamba, Francis (1988). An Introduction to Phonology. Harlow, Essex: Longman. Kaye, Jonathan (1985). On the syllable structure of certain West African languages. In Didier Goyvaerts (ed), African Linguistics: essays in memory of M.W.K. Semikenke, 285-308. Amsterdam: John Benjamins. Kaye, Jonathan (1989). Phonology: a cognitive view. Hillsdale, Nj: Lawrence Erlbaum, Kaye, Jonathan (1990a). ‘Coda’ Licensing. Phonology 7. 301-30, Kaye, Jonathan (19906). Government in phonology: the case of Moroccan Arabic. The Linguistic Review 7. 131-59 Kaye, Jonathan (1991), Head projection and indexation: a theory of reduplication. Paper resented at the 14th GLOW Colloquium, Leiden. Kaye, Jonathan (1992). Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics . 293-314. ~Kaye, Jonathan and Jean Lowenstamm (1981). Syllable structure and markedness theory. In A. Belleti, L. Brandi and L. Rizzi (eds), Theory of Markedness in Generative Grammar, 287-316, Pisa: Scuola Normale Superiore. Kaye, Jonathan and Jean Lowenstamm (1984), De la syllabicieé. In Frangois Dell, Daniel Hirst and Jean-Roger Vergnaud (eds), Forme sonore du langage, 123-59. Paris: Her- Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud (1985). The internal struc- ture of phonological elements: a theory cf charm and government. Phonology Year- book 2. 305-28. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud (1990). Constituent structure and government in phonology. Phonology 7. 193-232. Originally published 1989 as Konstituentenstruktur und Rektion in der Phonologie, Linguistische Berichte 114. 31-75. Kaye, Jonathan and Jean-Roger Vergnaud (1950). Phonology, morphology and the lexicon. Paper presented at the 13th GLOW Colloquium, Cambridge, England. Kean, Marie Louise (1975). The theory of matkedness in generative grammar. PhD disser- tation, MIT. Kempson, Ruth (1988 Logical form: the grammar-cognition interface. Journal of Linguis- tics 24, 393-431. Kenstowicz, Michael and Charles Kisseberth (1979). Generative Phonology: description ‘and theory. New York: Academie Press. Kenyon, J.S. (1958). American Pronunciation. 10th edn. Ann Arbor, MI: George Waht Kenyon, J. &. and T. A. Knott (1953). A Pronouncing Dictionary of American English Springfield, MA: Merriam. Keyser, Samuel J. and Wayne O'Neill (1985). Rule Generalization and Optionality in Language Change. Dordrecht: Foris. References 305 Kerswill, Paul E. (1987). Levels of linguistic variation in Dusham. Journal of Linguistics 23. 25-48. Kiparsky, Paul (1979). Metrical structuce is cyclic. Linguistic Inquiry 8. 421—A2. Kiparsky, Paul (1981). Remarks on the metrical stcture ofthe syllable, In W. U. Dressler, (©. E, Peiffer and J. R. Rennison (eds), Phonologica 1980: Akten aus den Vierten Interationalen Phonologie-Tagung, Wien, 245-56. Innsbruck: Inasbrucker Beitrige zur Sprachwissenschaft. Kiparsky, Paul (1982) From Cyclic Phonology to Lexical Phonology. In van der Hulst and ‘Smith 1982, Pare L 131-75. iparsky, Paul (19828). Explanation in Phonology. Dordrec Kiparsky, Paul (1985). Some consequences of Lexical Phonology. Phonology 2. 85-138. Kiparsky, Paul (1992). Catalexs. Ms. Stanford Universi Kenig, Werner (1978). dev-Adas ur deutschen Sprache. Municha: Deutscher Taschenbuch Verlag. Kreidler, Charles W. (1989). The Promunciation of English. Oxford: Blackwell Kurath, Hans and Raven I. MeDavid Je (1961). The Pronunciation of English in the Atlantic ‘States. Ann Arbor: University of Michigan Pres. Kaurylowice, Jerzy (1948), Contributions a la théorie de la syllabe. Bulletin dela société ‘polonaise de inguistque 8. 80-114. Labor, William (1966). The Social Stratification of English in New York City. Washington, DC: Center for Applied Linguistics. Labor, William, Malcah Yaeger and R. Steiner (1972). A Quantitative Study of Sound ‘Change in Progress. Philadelphia, PA: US Regional Survey. Ladefoged, Peter (1971). Preliminaries to Linguistic Phonetics. Chicago: University of Chicago Press. Ladefoged, Pete (1982). A Course in Phonetics. 2nd edn. New York: Harcourt, Brace and Jovanovich Ladefoged, Peter, J. DeClerk, M. Lindau and G. Papcun (1972). An auditory motor theory of speech production. UCLA Phonetics Laboratory Working Papers in Phonetics 22 48.76. Lass, Roger (1974). Linguistic orthogenesis? Scots vowel quantity and the English length conspiracy. In John Anderson and Charles Jones (eds), Historical Linguistics II: theory and description in phonology, Proceedings of the First International Conference on Historical Linguistics, Edinburgh, September 1973, 311-52. Amsterdam: North-Hol- land. Lass, Roger (1976). English Phonology and Phonological Theory. Cambrid ‘University Pres Lass, Roger (1984). Phonology: an introduction to asic concepts. Cambridge: Cambridge ‘University Press. Lass, Roger and John M. Anderson (1975). Old English Phonology. Cambrid Universicy Press Laver, John (1980). The Phonetic Description of Voice Quality. Cambridge: Cambridge University Press. Leben, William (1973). Suprasegmental phonology. PRD dissertation, MIT. Lehiste, Ilse (1978). The syllable as a structural unit in Estonian In Bell and Hooper 1978, 73-83. Leslie, David (1983). Left capture and British voiceless stop allophony. Paper presented at the 19th Congress of the Intemational Association of Logopaedics and Phoniatrcs, University of Edinburgh, August 1983. 1 Cambridge Cambridge306 References Leslie, David (1989). Lenition systems. Paper presented at the London Phonology Seminar, University of London. Levin, Juliette (1985). A metrical theory of sllabicity. PhD dissertation, MIT. Liberman, Mark and Alan Prince (1977). On stress and linguistic rhythm. Linguistic Inquiry 8. 249-336, Lieber, Rochelle (1982). Allomocphy. Linguistic Analysis 10. 27-52. Lieberman, Philip and Sheila E. Blumstein (1988). Speech Physiology, Speech Perception, ‘and Acoustic Phonetics. Cambridge: Cambridge University Press. Lindsey, Geoff (1990). Quantity and qualcy in British and American vowel systems. In ‘Ramsaran 1990, 106-18. Lindsey, Geoff and John Harris (1990). Phonetic interpretation in generative grammar. UCL Working Papers in Linguistic 2. 355-69. Lodge, K. R. (1984). Studies in the Phonology of Colloguial English. London: Croom Helm. Lowenstamm, Jean (1979). Topics in Prosodic Phonology. PhD dissertation, University of Massachussetts Lowenstamm, Jean (1988). Prosodic government. Langues orientales anciennes: philologie et linguistique 2. 221-3 Lowenstamm, Jean and Jonathan Kaye (1985). Compensatory lengthening in Tiberian Hebrew. In Wetzels and Sezer 1985, 97-132. Luelsdorff, Philip (1975). A Segmental Phonology of Black English. The Hague: Mouton. Luick, Karl (1921). Historische Grammatik der englischen Sprache. Leipzig: Tauchnitz McCarthy, John J. (1979). Formal problems in Semitic morphology and phonology. PhD dissertation, MIT. Distributed by Indiana University Linguistics Club. McCarthy, John J. (1986). OCP effects: gemination and antigemination. Linguistic Inquiry 17, 207-63. McCarthy, John J. (1988). Feature geometry and dependens 84-108. McCarthy, John J. (1991). Synchronic rule inversion. Proceedings of the Seventeenth ‘Annual Meeting of the Berkeley Linguistics Society. 192-207. McCarthy, John J. and Alan Prince (1985). Prosodic morphology. Ms. University of Massachusetts, MeCawley, James D. (1986). Today the word, tomorrow phonology. Phonology 3. 27-44. Mackay, Edith (1982). Universal Typing: advanced level. London: Pitman. McMahon, April (1991). Lexical phonology and sound change: the case of the Scottish ‘Vowel Length Rule. Journal of Linguistics 27. 29-53. McMahon, April (1992). Rule inversion: a lexicalist reanalysis. Ms. University of Cam- bridge. Manuini, M. Rita (1992). Locality: « theory and some uf its empirical consequences Cambridge, MA: MIT Press. Marantz, Alec (1982). Re reduplication. Linguistic Inquiry 13. 435-82. ‘Mascaré, Joan (1976). Catalan phonology and the phonological cycle. PhD dissertation, MIT. Distributed by Indiana University Linguistics Club. Mascaré, Joan (1983). Phonological levels and assimilatory processes. Ms. Universitat Autonoma de Barcelona. Mascaré, Joan (1987). A reduction theory of voicing and other sound effects. Ms. Univer- sitat Autonoma de Barcelona. Mather, J. Y and HLH Speitel (1986). The Linguistic Atlas of Scotland, Scots Section, Volume Ill: phonology. London: Croom Helm. review. Phonetica 45. References 307 Matisoff, J. A. (1973). Tonogenesis in southeast Asia. In Larry M. Hyman (ed.), Consonant Types and Tone, 71-95. Los Angeles: University of Southern California Mohanan, K. P. (1983). The structure of the melody. Ms. MIT. Mohanan, K. P. (1985). Syllable structure and lexical strata in English. Phonology Year. book 2. 137-54. Mohanan, K. P. (1986). The Theory of Lexical Phonology. Dordrecht: Reidel Mohanan, K.P. (1991). On the bases of Radical Underspecfication Theory. Natural Language and Linguistic Theory 9. 285-325. Mohanan, Tara (1988). Syllable steucture in Malayalam. Linguistic Inquiry 20. 589625. Myers, Scott (1987). Vowel shortening in English. Natural Language and Linguistic Theory 5. 485-518. Myers, Scott (19913) Persistent rules. Linguistic Inquiry 22. 315-44. Myers, Scott (1991b). Structure Preservation and the Strong Domain Hypothesis. Linguistic Inquiry 22. 379-85. Nespor, Marina and Irene Vogel (1986). Prosodic Phonology. Dordrecht: Fotis. Neu, H. (1980). Ranking of constraints on Ayd/ deletion in American English: a statistical ‘analysis. In William Labov (ed.), Locating Language in Time and Space, 37-54. New York: Academic Press. °Connor, J. D and J. L. M. Trim (1953). Vowel, consonant and syllable: a phonological definition. Word 9. 103-22. Odden, David (1988). Ant-antigemination and the OCP. Linguistic Inquiry 19. 451-74. Orton, Harold and Michael V. Barry (1969). Survey of English Dialects, Basic Material: the West Midland counties. Leeds: Arnold. Orton, Harold and Philip M. Tillng (1969). Survey of English Dialects, Basic Material: the East Midland counties and East Anglia. Leeds: Arnold Palmada Félez, Blanca. 1991. La fonologia del catala i els principis actius. Doctoral dissereation, Universitat Autonoma de Barcelona. Palmer, Frank (ed.) (1970). Prosodic Analysis. London: Oxford University Press. Paradis, Carole and Jean-Frangois Prunet (eds) (1991). The Special Status of Coronal: internal and external evidence. Phonetics and Phonology 2. San Diego, CA: Academic Press Petterson, Thore and Sidney Wood (1987). Vowel reduction in Bulgarian and its implica tions for theories of vowel reduction: a review of the problem. Folia Linguistica 21. 261-79. Piggott, Glynne (1988). The parameters of nasalisation. Ms. McGill University. Piggott, Glynne and R. Singh (1985). The phonology of epenthetic segments. Revwe Cara- denne de Linguistique 30. 415-53. Pike, Kenncth (1967). Language in Relation to « Unified Theory of Human Behaviour. The Hague: Mouton. Pike, Kenneth and Eunice Victoria Pike (1947). Immediate constituents of Mazateco sylla- bles. International Journal of American Linguistics 13. 78-91. Prince, Alan S. (1984). Phonology with tiers. In Aronoff and Ochrle 1984, 234-44. Prince, Alan S. and Paul Smolensky (1993). Optimality theory: constraint interaction in generative grammar. Technical Report 2 ofthe Rutgers Center for Cognitive Science, Rutgers University. Prunet, Jean-Frangois (1986). Spreading and locality domains in phonology. PhD disserta- sion, McGill University. Palleyblank, Douglas (1986). Tone in Lexical Phonology. Dordrecht: Reidel,308 References Pulleyblank, Douglas (in press). Feature geometry and underspecfication. To appear in Durand and Katamba, in press. Pallum, Geoff K. (1976). Rule Interaction and the Organization of a Grammar. PhD dissertation, University College London. Fublished 1978, New York: Garland. Ramsaran, Susan (ed.) (1990). Studies in the Pronunciation of English: a commemorative ‘volume in honour of A.C. Gimson. Londcn: Routledge. Rennison, John (1984). On tridirectional featace systems for vowels. Wiener linguistische Gazette. 33-4, 69-93. Reprinted in Durand 1986, 281-303. Rennison, John (1987). Vowel harmony and tidirectional vowel features. Folia Linguistica 21. 337-54. Rennison, John (1990). On the elements of phonological representations: the evidence from vowel systems and vowel processes. Folia Linguistica 24. 175-244. Rice, Keren D. (1992). On deriving sonority: a structural account of sonority relationships. Phonology 9. 61-100. van Riemsdijk, Henk and Edwin Williams (1986). Theory of Grammar. Cambridge, MA: MIT Press. Robins, R. H. (1957). Aspects of prosodic analysis. Proceedings of the University of Durham Philosophical Society 1. 1-12. Reprinted in Palmer 1970, 157-73. Sagey, Elizabeth (1986). The representation of features and relations in non-linear phono- logy. PRD diseertation, MIT. Sagey, Elizabeth (1988). On the illformedness of crossing association lines. Linguistic Inquiry 19. 109-18, Saussure, Ferdinand de (1916). Cours de linguistique générale. Paris: Payor. Schane, Sanford S. (1973). Generative Phonology. Englewood Clifs, NJ: Prentice-Hall, Schane, Sanford S. (1984a). The fundamentals of Particle Phonology. Phonology 1. 129-56. Schane, Sanford S. (1984b). Two English vowel movements: a particle analysis. In Aronoff ‘and Ochele 1984, 32-51 Scobbie, James M. (1992). Against rule inversion: the development of English {c]-sandhi. Poster paper, 7th International Phonology Meeting, Krems. Selkirk, Elizabeth ©, (1980). The role of prosodic categories in English word stress. Linguistic Inquiry 11. 563-605, Selkirk, Elizabeth ©. (1981). Epenthesis and degenerate syllables in Cairene Arabic. In agit Borer and Joseph Aoun (eds), MIT Working Papers in Linguistics 3. 209-32. Selkitk, Flizabeth O. (1982a). The Syntax of Words. Cambridge, MA: MIT Press. Selkirk, Elizabeth ©. (1982b). The syllable. In van der Hulst and Smith 1982b, Part Il, 337-84, Selkick, Elizabeth ©. (1984). On the major class features and syllable theory. In Aronoff and Ocrhle 1984, 107-36. Selkirk, Elizabeth O. and Jean-Roger Vergnaui (1973). How abstract is French phonology? Foundations of Language 10. 249-54. Sheridan, Thomas (1762). A Course of Lectuces on Elocution. Reprinted 1968, Menston: Scholar Press. Siegel, Dorothy (1974). Topics in English morphology. PhD dissertation, MIT. Sivertsen, E. (1960). Cockney Phonology. Oslo: Oslo University Press. Sledd, James H. (1958). Some questions of English phonology. Language 34. 252-60. Sledd, James H. (1966). Breaking, umlaut, and the Southern drawl. Language 42. 18-41 ‘Smith, Norval. (1988). Consonant place fearures. In van der Hulst and Smith 1988, Pare I, 209-36. References 309 Sperber, Dan and Deirdre Wilson (1986). Relevance: communication and cognition. Ox- ford: Blackwell, Sproat, R. (1985). On deriving the lexicon. PhD dissertation, MIT. Stampe, David (1973). A dissertation on Natural Phonology. PAD dissertation, University of Chicago. Steriade, Donea (1982). Greek prosodies and the nature of syllabification. PhD dissertation, MIT. ‘Trager, G. Land H. L. Smith (1951). An Outline of English Structure. Studies in Lingwis- ties, Occasional Papers 3, Norman, OK: Battenburg Press. Trubetzkoy, N. (1939). Grundziige der Phonologie. Travaux du cercle linguistique de Prague 7. ‘Trudgil, Peter (ed.) (1984). Language in the British Isles. Cambridge: Cambridge University Press. ‘Trudgil, Peter (1986). Dialects in Contact. Oxford: Blackwell Uriagercka, Juan (1986). Government functions and the Classical Arabic verb paradigm. Paper presented at the 9th GLOW Colloquium, Barcelona/Girona Vennemann, Theo (19728). On the theory of syllabic phonology. Linguistische Berichte 18. 148. Vennemann, Theo (1972b). Rule inversion. Lingua 29. 209-42. ‘Vennemann, Theo (1988). Preference Laws for Syllable Structure and the Explanation of ‘Sound Change. Berlin: Mouton de Gruyter. Vergnaud, Jean-Roger (1982). On the theoretical bases of phonology. Paper presented atthe ‘Sth GLOW Colloguium, Paris. Walker, John (1791). A Critical Pronouncing Dictionary and Expositor of the English Language. London: Robinson and Cadell. Reprinted 1968, English Linguistics 1500- 1800 117. Menston: Scholar Press. Wang, H. Samuel and Bruce L. Derwing (1986). More on English vowel shift: the back vowel question. Phonology 3. 99-116. Walls, J.C. (1982). Accents of English. 3 vols. Cambridge: Cambridge University Press. Wells, J.C. (1990). Syllabification and allophony. In Ramsaran 1990, 76-86. Werzels, Leo and Engin Sezer (eds) (1985). Studies in Compensatory Lengthening. Dor- deecht:Foris. Wolfe, Patricia M. (1972). Linguistic Change and the Great Vowel Shift in English. Berkeley and Los Angeles: University of California Press. Yip, Moira (1988a). Template morphology and the direction of association. Natural Language and Linguistic Theory 6. 551-77. Yip, Moira (1988b). The Obligatory Contour Principle and phonological rules: a loss of identity. Linguistic Inquiry 19. 65-100. Yip, Moira (1991). Coronals, consonant clusters, and the coda condition. in Paradis and Prunet 1991, 61-78. Zec, Draga (1988). Sonority constraints on prosodic structure. PhD dissertation, Stanford ‘Universi. Zwwicky, Arnold M. (1975), Setling on an underlying form: the English inflectional ending. In D. Cohen and J. R. Wirth (eds), Testing Linguistic Hypotheses, 129-85, New York: Wiley.Subject index (alan) 60 adjacency 12, 14, 48-51, 156, 163, 182, 187-8, 191-3 string 156-9 adjunct position 153, 158, 160, 163-4, 168-9 advanced tongue root see tense vs. lax affricate 39-41, 131 ‘African American English final devoicing in 210 nasals in 147 sr clusters in 278 0.44 Aitken’s Law 30-1, 255 alternation 3, 60-1, 63, 125, 165-6, 180, 182, 192-3, 197, 240-1, 266-7 of rand zero 231, 234-5, 238, 240-52, 266 20 also root-level alternation ambisyllabicity 198-200, 202 analytic morphology 25, 50-1, 70-1, 82, 182 appendix 81-2 Arabic ‘onsets in 150 Arbore lottalling in. 284 n.63 assimilation ‘of manner 132 of place 2,9, 14, 18, 72, 118-19, 165 ‘of voice 23, 58-9, 136-8, 145-6, 278 nas association line 13-14, 35, 37, 101, 128, 165-6 no crossing constraint on 37 se-tensing/lengthening 29-30, 69, 77-8, 125 ATR see tense vs. lax Aust tapping in 121, 195 autosegmental licensing see licensing autosegmental representation 101-2, 105, 127-8 Belfast ‘e-tensing in 30 Berber igeminates in 288 n.39 binary branching 63, 65, 82-3, 163, 178, 279 0.56 bleeding order see rule bogus consonant cluster 67, 182, 187, 222-3, 288 n.39 bound morpheme 19, 290 n.53 brace notation 197-8, 202, 238-9 breaking see vowel breathy voice 135 broad ¢ 282-3 n.32 see also a-tensing/lengthening. Bulgarian vowel reduction in 112 Canada ‘hoticity in 232 tapping in 115, 195 ibbean breaking in 116, 278 0.53, 294 0.12 shoticity in 232 Caralan ‘devoicing in 135, ‘vowel reduction in 110, 112-13 charm 289 n.43, 289 n.47 class node 128-9) closed-rhyme shortness 20, 55, 78-80, 149, 161, 168-9, 226 closed-syllable shortening see closed-rhyme shortness Coda Capeure 198-202, 225 Coda Licensing 160-1, 163, 165, 186, 189, 202 compensatory lengthening 34, 37, 128, 155, 166 ‘complement position 153-4, 164, 168-72, 222 Complexity Condition 170-8, 189-90, 206-7, 219, 221 ‘complexity of melodic unies 122-2 170-8, 205-6, 218-19, 221-2, 289 2.46 ‘composition 98, 132, 164-5 ‘constituent licensing see licensing contour segment 40, 64, 118-19, 130-3, 174, 276 n.17, 278 n.54, 284 n.68, 288 0.40 ‘core grammar 15, 52-3 creole 227,271 ‘eurly-bracket notation see brace notation ceycle 25-6, 212-16, 241-2, 262 debuccalization 120-1, 124, 141, 143, 195, 218, 224 decomposition 98, 103-4, 111-13, 131, 164-5, 203, 205 degemination 20,26 degenerate syllable 74-7, 79, 81, 160-1 see also empry nucleus deletion of consonant 2~ 124, 165, 217, 227 ing, 35-6, 102-3, 112, 115-16, 129, 155, 164-5, 211-14, 240-2 derivation 4, 6, 13, 94-6, 186-90, 241-2, 245, 270-1 devoicing 210 dip elemental pattern 139-41 diphthong. short 276 n.17, 278 n.54 diphthongization’ 100, 102-3, 115-17 ditectionality of licensing 155-6, 159-60, 163 of spreading 166-7 docking 249, 251 Dutch and English vowel systems 113 and Saramaccan 227 ‘consonantal lenition in 210 devoicing in 210 9, 36, 120, del Early Modern English lowering in 103-4 Subject index 311 monophthongization in 99 raising in 106 East Anglia palatal glides in 87-9 ECP see Empty Category Principle Ged) 23, 70-1, 136-7, 180-1 Empty Category Principle 193 see also Proper Government ‘empty nucleus 62, 74, 76, 161, 179-82, 188, 190-3, 209-13, 216, 222-4, 253-5 ‘empty onset 179 England Midlands 85-6 north 85-6, 100, 232, 280 9.72-3, 284 n.54, 296 n. 41 northeast 116, 137 POT in 115 rin 231-2 south 69, 78, 95-9, 115, 232, 243-4, 267, 282 n.26 southern standard English 61, 226-7, 232, 293.5 sr clusters in 278 0.44 tapping in 121, 195 West Country 100, 232 cepenthetic stop 132-3, 227, 279 n.60 ‘equipollence 92-4 Ger) 24 Estonian vowel length in 278-9 n.55 Excrescent stop see epenthetic stop extra-metricality 39-40, 43-4, 72-3, 75, 292.96 cextra-prosodicity 73, 293 1.109 extra syllabicity 72-5, 81 feature 8-10, 90 1s condition on repenthesis 244, 247 ‘8 condition on ¢lenition 200-2, 218 cequipollence of 924 non-linear representation of 13-14 phonetic interpretation of 32, 94, 107-8 representation of quantity 32, 41 representation of voice contrasts 134 underspecification of 93-5, 281 a.10, 289 0.46 fission 98, 100, 120, 131 see also decomposition flapping see tapping312. Subject index Flemish historical development from 279 n.60 foot 120-1, 151, 153, 156-7, 183, 192, 195-224 passim, 233-4, 251, 278 nas see also super-toot fortis 133-7, 227 fortition 132-3 free morpheme 19 : French Clonsets in 171 final branching onsets in 222, 279-80 2.67 liaison in 295 0.33 schwa in 110 short diphthongs in. 276 0.17 stress in 16, 157 voice contrast in 133, 135, 137 vowel deletion in 179, 205, see also h-aspiré fusion 97, 105-8, 110-11, 119, 130-2, 135, 140-1, 170 see also composition geminate 18, 20, 23, 36, 38, 130, 132, 171-4, 289 046 fake 38,173 inalterability of 132, 286 n.87 partial 69, 76-80, 130, 165-6, 174-5 German and English vowel systems 113, devoicing in 135, 210 Fen clusers in 52 lax vowels in 115 Germanic 269-70 ‘devoicing in 210 final branching nuclei in 261 short vs. long vowels in. 113-14 slide see hiatus glide, in-glide, off-glide, palatal glide, up-glide eloralling 120-1, 125, 143, 195-8, 211-14, 218-20, 224, 258, 284 n.63, ‘government 168-78, 206-8, 217, 219-20, 222-3, 289 n43 Great Vowel Shift sce Vowel Shift Gujarati voice contrasts in 135 Guyana shoticity in 232 Hart, John 258 hraspiré 179,290 0.50 head immediate 152-3, 156, 169 of prosodic domain 149-59, 166, 169-70, 193 vlaumate 152-4 see also persistent non-head hhead-dependency (within segment) 105-7, 111-14, 125-6, 181-2, 253, 296 0.82. heavy rhyme 33, 39-40, 43-4, 46, 55, 6425, 72, 75, 78, 83, 114, 163-4, 208 hiatus glide 104-5, 205, 247, 260, 296 nal homophone 5, 22, 236, 239 Hungarian stress in 16, 157 hhyper-rhoticity 296 n.40 Icelandic diphthongs in 278 n.54 implicational universal 150-1, 162-3, Gn) 18-20, 26 inalterabilty see geminace Indo-European 270 in-glide 116-17, 233, 235, 254, 256, 259, 263, 266, 294 n.A2 inter-constituent licensing see licensing, Irish English breaking in 116 Cy dusters in 87-9 DAZE vs. DAYS in 29 deletion of medial d in 210 long monophthongs in 100 POT in 115) rhoticity in 231-2 Scots vowel system in 145 SHOUT in 280 0.72 spirantization in 121 tapping in 121, 195 thr?) in 294 n.11 wa: in 119 (wh) in. 126 Irish Gaelic palatalization in 118 isomer 106-7, 111-12, 125 Italian codas in 162 Cy onsets in 171 ‘geminates in 20, 36, 38-9, 132 sC clusters in 62-3, vowel length in. 55 Jakobson, Roman 107 Jamaica in-gliding diphhongs in 294 n.12 thoticity in 232 Japanese default vowel in 109 Korean unreleased stops in 124 Lancashire thoticity in 232 Lardil ‘codas in 292.n.95 Latin historical development from 132, 279 1.60, 281 n.79 Latinate vocabulary 19 lenition 120-3, 131, 141-3, 17, 194-225, 239, 248, 258, 271 rules of 199-203 Lexical Phonology 26 lexical representation 4-5, 73, 83, 95, 186, 188-90, 192, 204, 238-9, 252, 2 LF see Logical Form licensing autosegmental 155-6, 164-7, 181-2, 188, 193, 206-12, 219, 248-9, 251, 2m constituent 157, 163-4, 170, 173, 188 projection 157, 177-8, 188, 191 prosodic 154-6, 206, 271 see also licensing domain, Licensing Tnheritance Principle, licensing path, Phonological Licensing Principle licensing domain 155, 168-70, 177 Licensing Inheritance Principle 206-11, 251 licensing path 207-9 light rhyme 33, 39, 40-1, 43-4, 46, 65, 75-6, 83 locality 12, 14, 37, 156, 163, 178 Logical Form 1, 270 London affrication in 286 n.85 slotalization in 284 0.63, Subject index 313 intrusive rin 253 Levocalization in 266-8 rnon-rhoticity in 231 PAUSE vs. PAWS in 28-9 Lowland Scots see Scots Luo final consonants in 162 Malay gloctaling in 284 1.63 Malayalam ccodas in 290-1 n.62 markedness 10-11, 92-3 mAss elemental pattern 139-41 merger 117, 235, 238-9, 256, 262-4, 2910.76 Merseyside debuccalization (to b) in 121, 195 spirantization in 121, 146-7, 195 metathesis 285 n.78 metrical grid/tree 151 Midale English final schwa in 253, raising in 106 minimal word 261, 297 1.54 Minimality Condition 170, 288 1.37 minimum sonority distance 56, 58, 83, 176, 178, 186 monophthongization 99 mora 2772.30 ‘morpheme-structure condition 49 nasal clement 125, 131-2 nasalization see vowel natural class 90-1, 127 neutral obstruent 133-7, 146 neutralization 58, 112, 135, 195, 208, 210, 218, 224-5, 256, 263, New York ‘e-tensing in 30 rin 2940.8 tapping in 195 noise elemental pattern 123, 140, 142-3 non-analytic morphology 26, 50-1, 71, 78, 165, 242 non-thotic 28-9, 68, 217-18, 220-1, 226-7, 230-65 passim ‘Non-Rhoticty condition 245-6, 248-52, 263 North America314 Subject index Cy clusters in 87-9 tapping in 121 Obligatory Contour Principle (OCP) 172-3, 181, 185, 205, 213, 224-5, 290 n.53 off-glide 30, $9, 64, 99, 100-3, 172, 237, 256-7, 260, (Old English final devoicing in 135, 210 fricative voicing in 15 historical development from 51, 253 Old Saxon historical development from 279 n.60 Onset Capture 199-200, 202, 241-2 Onset Licensing 160-2, 179, 204 onset maximization 5475, 83, 159-60, 184, 186, 198, 2871.28 opening. 120-1, 123-4, 131-2 copen-yllable lengthening 55 output constraint 245, 294-5 n.26, 296 nl palatal glide 5, 61-2, 86-9, 104-5, 118 palatalization 2-3, 5-6, 87, 118, 126, 222 parameter 16-17, 270-1 ‘constituent-strucrure 150, 160, 162, 164, 178-9, 228, 270 directionality ‘157, 178-9, 181 final-empty nucleus 162, 178-9, 190-2, 209-11, 216, 248, 253, 270, 297 n.54 partial geminate see gerninate peripheral grammar 15, 52 persistent non-head 168-9 Phonetic Form (PF) 1-2, 5-7 Phonological Licensing Principle 156, 178, 189, 206 phonological phease 151, 153-4 phonological word 151, 153-4, 156-7 phonotactic domain 47, 168-70 place node 129-31, 133, 175 Polish consonant clusters in 57, 288 0.39 voice assimilation in. 138 Portuguese and Saramaccan 227 Cr clusters in 171 Prague School 92 pre-fortis clipping 226 pre-nasalized stop 228, 276 n.17 pre-vocalic tensing 287 n.24 Prince language 174-5 privativeness 92-4, 121 projection 152-3, 155-7, 168-70 licensing see licensing Projection Principle 189, 193 proper bracketing 200 Proper Government 191-3 102 prosodic hierarchy 152-7, 209 prosodic licensing see licensing protected environment 217, 219 , 222-3, 293, quantity 33,39, 41, 43-4, 46, 71-1 113-14, 163, s0e aleo heavy rhyme, light rhyme, superheavy rhyme ‘quantity-senstivity 39, 41-2, 75 quantity-sability 35-6 78, approximant 232, 238, 259-60, 262 intrusive 104, 181, 231-2, 236-7, 243, 1246-7, 249-50, 252-4, 294 n.15, 294-5 n.26, 295 0.29, 295 0.38 linking 231-2, 234, 236, 253, 262, 294-5 n.26 rough vs. smooth 230-1, 237 tap 226-7, 232, 259, 261 uvular 232 r-colouring 232, 259, 261 deletion rules 237-43, 245-6, 252, 294-5 n.26 r-epenthesis rules 243-7, 252, 294-5 1.26, 295 n.29, 296 a4 r-sandhi 231, 236-7, 251, 254, 294-5 1.26, 295 32-3, vocalization 117, 217-18, 220, 233, 235,259 redundancy rule 93-4 [release] rule 201-2 resyllabification 54, 180, 184, 186-90, 192, 198-200, 202-4, 211, 241-4, 290-1 n.62 thotic 48, 68, 217-18, 220-1, 230-65 assim root-level alternation 18, 20-1, 23-4, 26-8, 78-81, 103, 109-10, 192, 210, 294 1.24, 295 0.29 ‘morphology 19-20 phonology of 22, 24, 50-1 root node 129-33, 174 rule 7-12, 15, 26, 34, 93-4, 98, 182, 197, 203, 218 change 252, 269 inversion 24, 252, 254 ordering 34, 94, 200: bleeding 200, 275 n.2-3, 294 n.23; conjunctive 198; extrinsic 200, 202, 242; intrinsic 200, 202 structural change of 9-10 structural description of 9-10, 12, 198, 241-2, 246 Ump elemental pattern 139-41 Russian devoicing in 135 palatalization in 118 voice assimilation in 138 (3) 23, 70-1, 136-7, 180-1 sandhi seer Saramaccan syllable structure of 227-9 Scotland thoticity in 232 (wh) in 126 Scots 275 0.55 SHOUT in 280 n.72 velar fricative in 34-6 velar nasal in 84-6 vowel systems of 141, 144-5, wein 119 Scottish English 275 n.55 Cy clusters in 87-9 slotalling in 121, 195 long monophthongs in 37, 100 pre-r vowels in 255-6, 261 TIDE vs. TIED in 30-1 Scottish Vowel Length Rule sce Aitken’s Law Sheridan, Thomas 253 Sesotho fortition in 132 lateral~ stop alternation in 125 vocalic sequénces in 17 simplicity criterion 10 skeletal slot as syllabic position 46-8, 275 n.4 as unit of timing 35, 37-41, 43-4 slack vocal cords 134, 140, 210 Subject index 315 sonorant frequency zone 139-40 sonority 17, 56-7, 152,171, 176 hierarchy ‘56-8, 67, 176 sequencing 56-8, 67, 76-7, 81-3, 176, 178, 184, 186 see also minimura sonority distance Sound Pattern of English sce SPE southern hemisphere ddiphthongization in 100 snon-thoticity in 232 POT in 115 southern standard English see England Spanish default vowel in 109 esC clusters in 62 stcess in 76 SPE xi, 7-12 features in 41-2, 92-4, 96, 107, 114, 118 linear seructure in 32, 100 rules in 21, 26, 198, 237-8, 246, 269-70 see also morpheme-structure condition spirantization 21, 26-7, 120-1, 124, 126, 131, 141-2, 177, 195-7, 214, 218, 224 spontaneous vs. non-spontaneous voicing 135-6 spreading 14, 35-7, 102, 118-19, 132, 164-7, 173-6, 194, 224 stiff vocal cords 134, 140 stop elemental pattern 122-3, 140, 142-3 Stray Erasure 212, 293 1.109 stray segment 74, 154-5, 184, 188 strengthening see forttion stress. 16, 39, 41, 75-6, 87, 151, 153, 157, 177-8, 183, 196-7, 199-200, 216, 237-8 in English nouns 42-4, 55 in English verbs. 39 shift 196 Strict Cycle Condition 212 strident 126 string adjacency see adjacency Strong Domain Hypothesis 187 structural change see rule structural description see rule Structure Presezvation Principle 189-90, 202, 204, 291 n.66 super-foot 183, 196-7, 222, 224316 Subject index superheavy chyme 65-9, 77-81, 83, 127, 163-4, 208, 223-4 suppletion 3 surface predictability 240-1 Surinam see Saramaccan, syllabic consonant 184-5, 224-5, 258-9 syllabification 38, 83-4, 242 syllable 42-6, 62, 160, 276 n.23 syllable-structure template $3, 83-4 systematic phonetic representation 94-6 Tagalog reduplication in 276 0.23 tapping 121, 124-5, 135-6, 143, 195-7, 199-202, 214-16, 218-19, 224 Telugu default vowel in 109 syllable structure of 162 tense vs. ax 112-15, 284 0.53 Thai unreleased stops in 124 voice contrasts in 134-5 Athe) 60 tone 35,40, 134 Tonkawa vowel syncopein 183, 190-1 Trisyllabic Laxing 7, 21-2, 226 truncation of consonant see deletion of (un-) 18-19, 23, 26 underspecification see feature United States of America 293 n.6 eastern 232-3, inteusive rin 294 0.8 northeastern 294 n.12 POTin 115 thoticity in 232 South 232-3, 266, 293 n.7, 294 n-11 tapping in 195 (wh) in 126 Universal Grammar (UG) 15-16, 45, 92, 246 upglide 104-5, 117, 233, 237, 256, 260, 296 n41 vacuous operation of processes 6, 20, 165, 191, 238, 244, 246 velar fricative 34-6, 123, 126, 155, 166 velar nasal 24, 84-6 Velar Softening 7,21, 26-7 Venda pre-nasalized stops in 276 n.17 vocalization 120-2, 141, 143, 218, 220, 266-7 see also r-vocalization voice onset time 133-4 vowel breaking 115-16, 256, 259, 263-4, 266 coalescence 99 cepenthesis 180, 182-3, 192 harmony 12, 92-3, 113, 177-8 lowering 106-7, 112-13, 280 n.7: lowering before r 244, 247, 250-1, 256 nasalization 147, 21-2 raising 103, 106, 112-13, 115-16 eduction 109-13, 177-8, 208, 253 rounding 119 shortening 136 syncope 82, 177-8, 182-93, 222-4, 290 n.60, 293 n.113 word-class 99 see also Index of vowel word-clastes Vowel Shift 7,21-2, 26-7, 34, 99, 103, 106, 115, Walker, John 230-1, 236, 265 (wanna}-contraction 193 weakening see lenition weight 39, 43-6, 64, 75, see also quantity ‘Welsh English Cy clusters in 87-9 word-class see Index of vowel word-classes word-level morphology 18-28, 48-51, 70-1, 81-2 X-bar structure 152-3 Yorkshice pre-aspiration in 284 0.63 voice assimilation in 145-6 Yoruba nuclei in 150 vowel deletion in 205 Zulu ‘coalescence in 99 syllable structure of 150, 162 Index of vowel word-classes Following the practice adopted in the text, vowel contrasts are identified here according to the classes of words in which they occur, indicated by means of capitalized head-words. Rather than being listed alphabetically, the indexed head-words are grouped according to the following rough historical-phonological criteria: (a) short versus (b) long/diphthongal versus (c) pre-r. Shore BIT 115, 243, 255, 268 BET 115, 243, 255, 268 BAD/BAT 115, 117, 243, 255, 268, 294 nid COTPOT 106, 115, 117, 243, 255, 268 BUT/CUT 109, 115, 243, 255, 268 PUT 115,243,255, 268 Long/diphhongal BEAD/BEATMEET 103, 115, 235, 243, 255, 268 BADE/BAIT/DAY/MATE 99, 100, 106, 116, 235, 243, 255, 268, 294 0.12, 296 n.41 BOATIGO 100, 106, 116, 244, 255, 268, 296 n.41 BOOTILOOT 103, 115, 244, 255, 268 BITE 115-16, 243, 255, 268 BOUTSHOUT 115-16, 243, 255, 268, 280 n.12 Toy 117 CAUGHTIPAW 99, 117, 236-7, 244, 254-5, 268, 294 n.15 CALMIPA 117, 236-7 Before historical r BEARD/BEER/FEAR 235, 243, 254-5, 263, 294 n.12 EERIE 256 SPIRIT 256 BEAR/CARE/FAIR 243, 255,263 MARY 256, 263-4 MERRY 256, 263-4 MARRY 256, 263-4 BAR/CARTIFAR/FARM/PAR 117, 236-7, 243, 255-6, 263, 268 COURT/FOUR/PORE 117, 236-7, 244, 254-6, 263 FOR 255-6, 263 POOR 244, 255-6, 263 FIRSHIRTISKIRT/ THIRD 109, 25 268 FUR/HURT 255-7 PER/PERT 255-7 WIRE 243, 255, 256 HOUR 243, 255-6