epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (2024)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (3)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (4)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (5)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (6)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (7)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (8)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (9)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (10)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (11)

epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (12)

Prévia do material em texto

<p>Complexity Scales and Licensing in Phonology</p><p>Studies in Generative Grammar 105</p><p>Editors</p><p>Harry van der Hulst</p><p>Jan Koster</p><p>Henk van Riemsdijk</p><p>De Gruyter Mouton</p><p>Complexity Scales</p><p>and Licensing in Phonology</p><p>by</p><p>Eugeniusz Cyran</p><p>De Gruyter Mouton</p><p>The series Studies in Generative Grammar was formerly published by</p><p>Foris Publications Holland.</p><p>ISBN 978-3-11-022149-7</p><p>e-ISBN 978-3-11-022150-3</p><p>ISSN 0167-4331</p><p>Library of Congress Cataloging-in-Publication Data</p><p>Cyran, Eugeniusz.</p><p>Complexity scales and licensing in phonology / by Eugeniusz</p><p>Cyran.</p><p>p. cm. � (Studies in generative grammar ; 105)</p><p>ISBN 978-3-11-022149-7 (hardcover : alk. paper)</p><p>1. Grammar, Comparative and general � Phonology. 2. Dis-</p><p>tinctive features (Linguistics) I. Title.</p><p>P217.3.C958 2010</p><p>414�dc22</p><p>2010003183</p><p>Bibliographic information published by the Deutsche Nationalbibliothek</p><p>The Deutsche Nationalbibliothek lists this publication in the Deutsche</p><p>Nationalbibliografie; detailed bibliographic data are available in the Internet</p><p>at http://dnb.d-nb.de.</p><p>” 2010 Walter de Gruyter GmbH & Co. KG, Berlin/New York</p><p>Printing: Hubert & Co. GmbH & Co. KG, Göttingen</p><p>� Printed on acid-free paper</p><p>Printed in Germany</p><p>www.degruyter.com</p><p>Preface</p><p>This book is an attempt to demonstrate that the basic principles of phono-</p><p>logical organisation boil down to the interaction between the strength of</p><p>nuclei as licensers of phonological structure and various non-rerankable</p><p>scales of complexity occurring at different levels of phonological represen-</p><p>tation. The licensing relation between nuclei and the preceding onsets on</p><p>the one hand, and governing relations between consonants, which are to a</p><p>great extent determined by their internal melodic structure, allow us to view</p><p>the phonological representation as a self-organizing system.</p><p>As a starting point of our discussion, we take the theory of Principles</p><p>and Parameters in phonology, also referred to as (standard) Government</p><p>Phonology (Charette 1991, Harris 1990, 1994, 1997, Kaye 1990, 1995, Kaye,</p><p>Lowenstamm and Vergnaud 1985, 1990). The central underlying principle</p><p>of the self-organization in phonology due to the interaction between com-</p><p>plexity scales and licensing strength leads to a number of dramatic modifi-</p><p>cations of the standard model. Firstly, a lot of most cherished principles</p><p>and parameters are eliminated or redefined as part of non-rerankable scales.</p><p>Secondly, a change of philosophy is proposed concerning the employment</p><p>of empty nuclei in representation: from striving to develop mechanisms of</p><p>their licensing – muting mechanisms which allow empty nuclei to remain</p><p>silent – to determining their own licensing properties. Their formal function</p><p>is viewed as generally the same as that of other nuclei, while their special</p><p>status stems from the fact that they are substantively empty. And thirdly,</p><p>the phonological representation is viewed as a consecution of CVs (Lowen-</p><p>stamm 1996, Polgárdi 1998, Rowicka 1999, Scheer 2004), which is not just</p><p>an assumption. Some arguments for the CVCV structure are also adduced.</p><p>Complexity itself is not a new concept in Government Phonology, but it</p><p>has mostly been discussed in the context of the melodic make-up of seg-</p><p>ments (Harris 1990, 1994). In Chapter 1, various melodic complexity ef-</p><p>fects are discussed in Irish, Polish and Welsh. It is shown that such aspects</p><p>of segmental phonology as sonority effects, relative markedness, segmental</p><p>inventories and their susceptibility to phonological processes, as well as</p><p>the interaction between consonants in syllabification may to a great extent</p><p>be derived from the substantive complexity of segments defined as the</p><p>number of elements they contain. Additionally, an extension to the Element</p><p>Theory is proposed in the form of parameterizing the occurrence of some</p><p>Preface vi</p><p>elements. Chapter 2 deals with formally defined complexity – at the syl-</p><p>labic level – and its interaction with the melodic level. The proposal trans-</p><p>forms the original idea of Government Licensing (Charette 1990, 1992)</p><p>into a non-rerankable scale of progressively more complex structures which</p><p>demand progressively stronger licensers. The resulting model may account</p><p>for both fairly basic and also quite complex issues connected with syllabi-</p><p>fication and word structure, such as phonotactics and clustering, syllabi-</p><p>cally driven phonological processes, syllable typology, markedness, and</p><p>acquisition. This chapter contains a new analysis of Polish initial conso-</p><p>nant clusters. Chapter 3, considers issues connected with phonologically</p><p>conditioned aspects of word structure. Its first part deals with the interac-</p><p>tion between foot structure and syllabic organisation in the context of the</p><p>historical development in Slavic languages called liquid metathesis. It is</p><p>shown that the model is fully compatible with the predictions made by the</p><p>Licensing Inheritance theory (Harris 1997). The interaction between licens-</p><p>ing and complexity may now be treated as an organising agent present at</p><p>all levels of phonological representation which enables us to reinterpret the</p><p>familiar notion of structural analogy found in Dependency Phonology (e.g.</p><p>Anderson and Ewen 1987). Finally, the problem of word edges is returned</p><p>to with a view to demonstrating that the new model predicts such anoma-</p><p>lies of word structure as complex clusters at word edges in Polish, or Super</p><p>Heavy Rhymes in English and Dutch. This allows us to adopt a different</p><p>view on extra-syllabicity, that is, one in which such notions need no longer</p><p>be necessary.</p><p>I wish to express my gratitude to the following friends and colleagues</p><p>for their generous assistance and comments at various stages of writing the</p><p>book. First and foremost, many thanks are due to Edmund Gussmann, who</p><p>taught me phonology, and whose constant support and interest in my work</p><p>greatly contributed to the feeling that my efforts may be worth pursuing.</p><p>His numerous comments on the earlier version of this book were more than</p><p>helpful. They also saved me form a number of blunders. Of course, I take full</p><p>responsibility for the remaining ones. I did not take all of Ed’s criticisms in</p><p>to account, but I know I will be forgiven, as always. I am also extremely</p><p>grateful to Jonathan Kaye and Tobias Scheer who were always more than</p><p>willing to discuss my proposals and phonology in general, and who have</p><p>greatly influenced my thinking. Thanks are also due to the friendly, vi-</p><p>brant, and ever-growing group of people working within the broadly under-</p><p>stood model of Government Phonology. They provided a lot of inspiration</p><p>for my research. In particular, I would like to thank Monik Charette, John</p><p>Preface vii</p><p>Harris, Harry van der Hulst, Jean Lowenstamm, Krisztina Polgárdi, John</p><p>Rennison, Nancy Ritter, Grażyna Rowicka and Péter Szigetvári. A lot of</p><p>the initial research for this work was carried out during my stay at the Lin-</p><p>guistics Department of the University of California, Los Angeles in the</p><p>years 1998-1999. I am for ever grateful to Vicky Fromkin for her hospital-</p><p>ity and help. While in Los Angeles, I benefited considerably from the ex-</p><p>changes of ideas with Henning Andersen, Heriberto Avelino, Morris Halle,</p><p>Bruce Hayes, Pat Keating, Ian Maddieson, Tomás Ó Cathasaigh and Donca</p><p>Steriade. I am also extremely grateful to Aidan Doyle for his useful sugges-</p><p>tions and expertise as regards both Irish and English, and to Mark Ó Fion-</p><p>náin for proofreading the text.</p><p>Last but not least, I would like to thank Marta Cyran for her patience</p><p>and support. This book is dedicated to her.</p><p>Contents</p><p>Preface.......................................................................................................... v</p><p>Chapter 1</p><p>Substantive complexity</p><p>1. Introduction ..............................................................................................1</p><p>2. The Element Theory in Government Phonology .....................................2</p><p>2.1. Representing vowels ........................................................................4</p><p>[±BACK] idealizes the facts slightly even in the case of Connemara Irish –</p><p>as shifts of the type [a~i], [a~e] and [o~i] do occur in this dialect (15c-e).</p><p>The system with [I] and [E], coupled with spreading backness, predicts</p><p>symmetry in behaviour, and it is completely unable to subsume the Mun-</p><p>ster facts which involve height distinctions on a regular basis. These would</p><p>have to be dealt with by means of additional patch-up rules.</p><p>What seems to be required is an analysis which would be able to cover</p><p>all the facts and the dialectal variation in a simple and elegant fashion. It</p><p>should also be able to explain how and why vowels of any height tend to</p><p>alternate with [i] in Munster, and occasionally in Connemara. In other</p><p>words, there seems to be an asymmetry in the effects of backness spreading</p><p>which are difficult to express in an equipollent feature system in which</p><p>[+BACK] should be no different from [−BACK].37 One would also like to</p><p>have some explanation of the interesting correlation between the height of</p><p>the target vowels and the corresponding complexity of the facts. Note that</p><p>the [u~i] alternation is almost exceptionless in both dialects. The [o~e]</p><p>alternations are almost regular in Connemara, and almost non-existent in</p><p>Munster, in which [o~i] is the norm. The latter type is only marginal in</p><p>Connemara, though, much better established than [o~e] in Munster. This</p><p>asymmetry in the behaviour of high and mid targets of spreading has no</p><p>37 There is no denying that one may always resort to feature co-occurrence restric-</p><p>tions to derive these effects. The point is, however, that there is nothing inherent in</p><p>the two values of the feature [BACK] that would directly express such asymmetries.</p><p>24 Substantive complexity</p><p>expression in the I/E distinction in (16a). Thus, the lower we get in terms</p><p>of the height of the targets, the more complicated the picture gets, and the</p><p>targets are less and less susceptible to spreading, with the low vowel /A/</p><p>being fully resistant, like long vowels. This also brings up the question of</p><p>the representation of the short opaque vowels.38</p><p>Let us briefly compare the story of backness spreading with an element-</p><p>based analysis. In Cyran (1997), palatalized consonants are defined by the</p><p>presence of the element (I) which spreads to the preceding nucleus as the</p><p>head (I). The velarized series of consonants contains the (U) element,</p><p>which spreads as an operator. The asymmetry in the status of the spread</p><p>element is responsible for the fact that (I)-ness dominates the vocalic sys-</p><p>tem of Munster Irish, as we saw above. The so called opaque nuclei are</p><p>headed, e.g. (A), while the targets of spreading are headless. The short</p><p>vowel system of Irish may be viewed as a vertical system, very much in the</p><p>spirit of Ní Chiosáin (1994), but with some crucial differences.</p><p>(17) Short vowels in Irish</p><p>(_)</p><p>(A._)</p><p>(A)</p><p>Two comments are in order here. Firstly, we should immediately say that</p><p>the headed (A) is realized as back after velarized consonants (CA), and as</p><p>front after palatalized ones (C´a). In the latter context palatalization spread-</p><p>ing from the right hand context may affect such nuclei leading to [a~i] and</p><p>[a~e] alternations under the specific contextual conditions discussed in</p><p>Cyran (1997: 56). Secondly, the reason why (I) and (U) are banished from</p><p>the lexical representations of alterable short vowels is because these prop-</p><p>erties are always available from the consonants. However, the headless</p><p>targets should not be viewed as underspecified vowels which will receive</p><p>phonetic interpretation only once filled with (I) and (U). In a system in which</p><p>38 It is conceivable that they may be lexically specified with the feature [±BACK],</p><p>similarly to long vowels, which do not participate in the alternations.</p><p>Substantive complexity effects in Irish 25</p><p>there would be no element spreading to such nuclei, they would still be</p><p>interpretable in some way.39</p><p>Finally, a word of comment is in order concerning the concept of spread-</p><p>ing, which is fundamentally a derivational notion. In non-derivational par-</p><p>lance, we may say that the domain of phonetic interpretation of the secondary</p><p>articulation of consonants in Irish is wider than one skeletal position. In this</p><p>respect, the phonetic interpretation of alterable short vowels in Irish may</p><p>be said to involve two overlapping domains, that is, the melody lodged in</p><p>the nucleus and the superimposed melody of the secondary articulation of</p><p>the following consonant. In what follows the term spreading should be</p><p>understood in the non-derivational sense. We will also use the term super-</p><p>imposition of properties in overlapping domains.40</p><p>The alternation [u~i] is derived by means of the superimposition of (I)</p><p>or (U) from the following consonant on a nucleus which otherwise has no</p><p>melodic content. Since consonants in Irish are always either palatalized or</p><p>velarized, this structure will always be interpreted as either [i] or [u]. Note</p><p>that the height of these vowels needs no further specification because this</p><p>property is inherent in the two elements. In addition to that, because in this</p><p>instance (I) and (U) meet no other element in the nucleus, this type of al-</p><p>ternation is the most regular, as it does not involve any interaction between</p><p>the lexically present elements and the spread ones.</p><p>The situation is different in the case of lexical (A._), responsible for the</p><p>[o~e] and [o~i] alternations. Here the dialectal differences between Con-</p><p>nemara and Munster are most clear, but only when we talk about palatali-</p><p>zation contexts. Note that the spreading of (U) into the nucleus represented</p><p>as (A._) yields (A.U._), that is [o] in both dialects. The problem lies in the</p><p>way (A) and (I) combine in the two dialects. It appears that in Connemara</p><p>the incoming (I) may assume the head position to produce (A.I), hence the</p><p>regular alternation [o~e].41 In Munster Irish, [e] is an extremely restricted</p><p>39 Languages with vertical vowel systems of this type do exist. For example, Ka-</p><p>bardian has only two short vowels [È, E], but it has a full set of five long ones [i:, u:,</p><p>e:, o:, a:] (Maddieson 1984: 417).</p><p>40 The scope of the superimposition of secondary articulation in Irish is a complex</p><p>issue. There are generally two blockers restricting overlapping domains: a) the so</p><p>called ‘opaque vowels’, b) the specification of the preceding non-adjacent conso-</p><p>nant, which marks the beginning of a new domain (Cyran 1997: 50).</p><p>41 It is also possible that in Connemara the element (I) spreads as an operator just as</p><p>(U) does. This could explain why the effects of spreading into a nucleus specified</p><p>as (A._) are symmetrical, that is, (A.U._) and (A.I._).</p><p>26 Substantive complexity</p><p>vowel. On the basis of other phenomena involving the interaction between</p><p>(A) and (I), Cyran (1997: 101) proposes that the well-formed (A-I) com-</p><p>pound in this dialect is (A)-headed, that is, (A.I).42 Thus, in the case of</p><p>vowel-consonant interaction, the incoming (I) element cannot form a</p><p>grammatically licit compound with (A._) and it suppresses the (A) element,</p><p>or, to put it differently, the element (A) cannot be licensed in the nucleus</p><p>headed by (I). The same type of (A)-suppression is observed in the [a~i]</p><p>alternation, for example fear / fir [f´ar ~ f´ir´] ‘man / gen.sg.’, while in</p><p>[a~e] the phonetic mid vowel survives, but it must receive additional sup-</p><p>port from the following nucleus, for example, deas / deise [d´as ~ d´eS´]</p><p>‘nice / gen.sg.’.</p><p>Thus, the complexity of the effects concerning the mid vowels in Irish</p><p>follows directly from the melodic design of such vowels. Here, in contrast</p><p>to high vowels, the nucleus already has one property which is lexically</p><p>present in the nucleus. The effects of I/U-spreading are therefore depend-</p><p>ent on the combinatorial</p><p>possibilities between (A) and the incoming (I) and</p><p>(U). It appears then, that we in fact predict exactly where dialectal varia-</p><p>tion is more likely to occur: it is when the targets already have a lexical</p><p>property which will interact with the incoming elements.</p><p>The low vowels behave in the most irregular fashion because they are</p><p>represented by a headed element (A) which interacts with (I)-ness only</p><p>under strict conditions, if at all. This is parallel to the long vowel system.</p><p>Thus, it seems that the vertical system presented above is able to capture</p><p>not only the reflexes of spreading and their different outcomes by referring</p><p>to dialect-specific constraints on element combinability, but also straight-</p><p>forwardly captures the correlation between the height of the targets and the</p><p>relative regularity of the observed alternations. This analysis fully incorpo-</p><p>rates the properties of backness and height which eluded a uniform analysis</p><p>in a feature-based model. All the necessary asymmetries can be derived</p><p>from the general nature and workings of representations. Finally, it is</p><p>worth noting that the short vowel system in Irish resemble the scales of (U)</p><p>and (I) presence in the place of articulation of consonants illustrated in (8)</p><p>and (9) respectively. Here, we are dealing with a similar ternary scale of</p><p>(A) presence, in that, it may be a) absent (regular [u~i] alternation across</p><p>dialects), b) present as operator (leading to dialectal distinction: [o~e] in</p><p>Connemara vs. [o~i] in Munster), and c) present as head (general immunity</p><p>to element spreading). The scale (_)-(A._)-(A), with all the accompanying</p><p>42 For a recent exposition of the so called Licensing Constraints defining element</p><p>combinability in GP see e.g. Kaye (2001).</p><p>Substantive complexity effects in Irish 27</p><p>phonological contrasts and behaviour, also suggests a solution for the opac-</p><p>ity of some mid vowels. For example, the [o] in cois [koS] ‘leg’ and scoil</p><p>[skol´] ‘school’ is opaque because it is lexically represented as (U.A), that</p><p>is, a headed vowel.</p><p>We will now return to consonants and discuss some aspects of Irish</p><p>phonotactics in which the melodic complexity seems to play a role.</p><p>3.2. Substantive conditions on Irish epenthesis</p><p>The idea that complexity may replace sonority or strength scales is attrac-</p><p>tive for the simple reason that it is now a derivative of the internal repre-</p><p>sentation of consonants rather than a separately proposed look-up scale</p><p>whose role in the phonological system is arbitrary and unclear.43 Ideally,</p><p>once the representations of consonants are established for a given system</p><p>they should display consistent behaviour for all possible phonological phe-</p><p>nomena where complexity, sonority, or strength are assumed to play a role.</p><p>We should expect a convergence of unrelated aspects of a given phono-</p><p>logical system in the internal structure of its segments.</p><p>In this section, we look at two such aspects of the phonological system</p><p>of Irish. The first is phonotactics, or more specifically, the interaction be-</p><p>tween consonants in clustering. The second aspect concerns the segmental</p><p>inventory of Irish and some effects connected with it. The third area which</p><p>calls for an analysis in terms of elements are the initial consonant muta-</p><p>tions, which is attested in other Celtic languages as well. This phenomenon</p><p>will be only mentioned in passing here. It will be dealt with more fully in</p><p>the following section in relation to Welsh which will be assumed, quite un-</p><p>controversially, to share some properties of its consonantal system with Irish.</p><p>Let us first concentrate on the phenomenon concerning the syllable struc-</p><p>ture of Irish, which is common to all dialects, and which appears to be</p><p>conditioned by the segmental structure of consonants.</p><p>Irish displays a consistent phonotactic pattern in which certain coda-</p><p>onset contacts are disallowed. To be more precise, only a subset of poten-</p><p>tially possible clusters of falling sonority is found in this language. The</p><p>instances of sequences which are not grammatical are normally assumed to</p><p>be broken up by epenthesis, which has received a lot of attention in the</p><p>literature. Descriptions of the facts can be found in, for example, Ó Cuív</p><p>43 To be fair, phonological theory has witnessed quite a few attempts to encode</p><p>sonority effects in the internal representation of segments (e.g. Steriade 1982, Cle-</p><p>ments 1990, Dogil and Luschützky 1990, Rice 1992, Zec 1995)</p><p>28 Substantive complexity</p><p>(1975), de Bhaldraithe (1945), Ó Dochartaigh (1987), Ó Sé (2000), Ó Si-</p><p>adhail (1989), Sjoestedt (1931), Sjoestedt-Jonval (1938), Wagner (1959).</p><p>Formal accounts include, among others, Cyran (1996a), de Búrca (1981),</p><p>Green (1997, 2003), Ní Chiosáin (1991, 1999), Ó Baoill (1980). Let us</p><p>consider some examples below, in which the epenthetic vowel is given in</p><p>superscript.</p><p>(18)</p><p>a. ["f´er´´g´´] feirge ‘anger, gen.sg.’</p><p>["g´er´´b´´] geirbe ‘scab, gen.sg.’</p><p>["l´er´´g´´] leirge ‘slope, gen.sg.’</p><p>["bol´g´m] bolgam ‘mouthful’</p><p>["Ser´´v´i:S] seirbhís ‘sevice’</p><p>["ar´´g´´d] airgead ‘money’</p><p>b. ["d´ar´f´] dearfa ‘proved’</p><p>["kon´f ́ ] confadh ‘anger’</p><p>["fur´x´] forcha ‘beetle’</p><p>["dor´x´] dorcha ‘darkness’</p><p>["for´m´du:l´] formadúil ‘envious’</p><p>[f´er´´"m´o:r´] feirmeoir ‘farmer’</p><p>["an´´m´] ainm ‘name’</p><p>For the sake of the argument, the data include only those forms in which the</p><p>cluster is followed by a vowel rather than word-final, so that we can uncon-</p><p>troversially speak of the impossibility of establishing coda-onset contacts.</p><p>The same pattern, however, is attested for these clusters also in the word-</p><p>final context, for example, fearg [f´ar´g] ‘anger’, gearb [g´ar´b] ‘scab’,</p><p>learg [l´ar´g] ‘slope’, bolg [bol´g] ‘belly’, balbh [bAl´v] ‘dumb’. In the</p><p>following chapter, an attempt will be made to unify these two seemingly</p><p>disjoint contexts.</p><p>There is some agreement among linguists concerning the synchronic</p><p>status of this type of epenthesis. The main argument for this view is based</p><p>on the way secondary articulation affects clusters in Irish and Scots Gaelic</p><p>(Clements 1986, Cyran 1996a, Ní Chiosáin 1999).44 Specifically, lexical</p><p>clusters always agree in terms of palatalization or velarization. For exam-</p><p>ple, in Irish cearc [k´ark] ‘hen’, the cluster is velarized, while in circe</p><p>44 In fact, the discussion in Clements (1986) concerns similar instances of epenthe-</p><p>sis in Barra Gaelic, e.g. those described in Borgstrøm (1937).</p><p>Substantive complexity effects in Irish 29</p><p>[k´ir´k´´] ‘hen, gen.sg.’ it is palatalized. The same condition holds for the</p><p>assumed epenthetic sequences, for example, fearg ["f´ar´g] ‘anger’ vs. feir-</p><p>ge ["f´er´´g´´] ‘anger, gen.sg.’, but not those sequences in which the inter-</p><p>vening schwa is unambiguously lexical rather than epenthetic, for example,</p><p>capall [kAp´l] ‘horse’ vs. capaill [kAp´l´] ‘horse, pl.’, and not *[kAp´´l´].</p><p>On this basis, we may claim that the epenthesized sequences of consonants</p><p>are lexically adjacent.</p><p>It seems that one of the reasons for the impossibility of establishing</p><p>coda-onset contacts is strictly connected with the sub-segmental structure</p><p>of the second consonant. The structural description of the process of epen-</p><p>thesis varies depending on which data are assumed to be part of the phe-</p><p>nomenon. For example, de Búrca (1981) defines it as vowel insertion “be-</p><p>tween a sonorant and a non-homorganic voiced obstruent when the cluster</p><p>is preceded by a short vowel”. However, as the examples in (18b) clearly</p><p>demonstrate, the second consonant need not be voiced, e.g. dearfa ["d´ar´f´]</p><p>‘proved’, and it need not be an obstruent, e.g. ainm [an´´m´] ‘name’. A</p><p>more precise definition of the context for epenthesis is given in Ó Siadhail</p><p>(1989), where it is described as occurring within a coda-onset cluster if the</p><p>vowel preceding the cluster is short, and the sonorant is followed by a non-</p><p>homorganic consonant other than a voiceless stop. This formulation is in</p><p>fact an accurate description of the contacts which are possible in Irish. In</p><p>other words, the surviving contacts are those which involve homorganicity,</p><p>or those in which the second consonant has a particular value for sonority</p><p>(Ní Chiosáin 1999). The question is, however, if there is any way of cap-</p><p>turing where and why the phenomenon of epenthesis actually occurs.45</p><p>Let us look at some data illustrating the good contacts in Irish, which</p><p>involve a sonorant followed by an obstruent.46 Again, we limit ourselves to</p><p>the word-medial context, which differs slightly from the final one. How-</p><p>ever, these clusters also occur finally, for example, beirt [b´er´t´] ‘two peo-</p><p>ple’, olc [olk] ‘bad’, corp [korp] ‘body’.</p><p>45 There are three crucial aspects of the context: a) the preceding short vowel, b) the</p><p>sonorant, c) the non-homorganic consonant other than a voiceless stop (Cyran</p><p>1996a). Ní Chiosáin (1999) adds data in which the prosodic structure also seems to</p><p>play a role.</p><p>46 Irish also has clusters of falling sonority with [s, S, x] followed by an obstruent.</p><p>Here too, certain restrictions hold. For example, [x] can be followed by [t], but not</p><p>[p] or [k].</p><p>30 Substantive complexity</p><p>(19)</p><p>a. ["Sk´ulp´] sciolpa ‘splinter’</p><p>[kIr "́p´ax] coirpeach ‘criminal’</p><p>[il´k´´s] oilceas ‘evil’</p><p>["k´ir´k´´] circe ‘hen, gen.sg.’</p><p>[p´r"tAx] portach ‘bog’</p><p>["bAl´t´´] bailte ‘home, pl.’</p><p>b. ["t´i:m´p´´l] timpeall ‘round’</p><p>["mi:n´t´Ir´] muintir ‘people’</p><p>["ri:N´k´´] rince ‘dance’</p><p>["baund´] banda ‘band’</p><p>["frauNk´x] Francach ‘French’</p><p>["m´i:l´S´] milse ‘sweet, pl.’</p><p>["tIrS´] tuirse ‘tiredness’</p><p>The above data do not include all the possible types of sonorant – obstruent</p><p>word-internal clusters, but they illustrate the main tendencies. (19a) shows</p><p>that clusters in which a sonorant is followed by a voiceless stop are not</p><p>broken up by epenthesis. On the other hand, (19b) illustrates an additional</p><p>interesting phenomenon which is optional in Connemara but fairly regular</p><p>in Munster: the homorganic clusters are often preceded by a long nucleus.</p><p>Some of these are cases of lengthening before the homorganic sonorant −</p><p>obstruent clusters, for example, milse ["m´i:l´S´] ‘sweet, pl.’, others are</p><p>much more complicated.47 However, the main point is that for cluster in-</p><p>tegrity to be maintained, the second member should be either homorganic</p><p>(milse, banda), or a voiceless stop (circe, sciolpa). We will not pursue the</p><p>question of how homorganicity contributes to cluster integrity, though this</p><p>is not an uncommon situation in languages in general (e.g. Hayes 1986).</p><p>We are interested in seeing how sub-segmental representation in terms of</p><p>elements can enable us to understand the fact that, apart from cases of ho-</p><p>morganicity, only the voiceless stops make good contacts for the preceding</p><p>sonorants.</p><p>In models operating with sonority the obvious solution to the above</p><p>question would be that the voiceless stops are the least sonorous, and there-</p><p>fore, together with the preceding sonorant, they provide the biggest sonor-</p><p>ity slope. Likewise, in models using consonantal strength in the definition</p><p>47 For a formal analysis of the difference between milse [m´i:l´S´] ‘sweet, pl.’ and</p><p>tuirse [tIrS´] ‘tiredness’ see e.g. Cyran (1996a).</p><p>Substantive complexity effects in Irish 31</p><p>of preferred contacts, the voiceless stops are considered to be the strongest.</p><p>The Element Theory deals with sonority sequencing in a similar way.</p><p>Namely, the best syllable contacts, as well as the most preferred branching</p><p>onsets, are those displaying the greatest differential in complexity, which</p><p>corresponds to sonority slope. Note however, that the complexity-based</p><p>description has already been shown to be more advantageous than sonority</p><p>and strength scales. As we saw in the case of branching onsets in English</p><p>and Polish in (14), sometimes the preferred sequence involves a voiced</p><p>rather than a voiceless obstruent. Thus, the preference for /vr/ over /fr/ in</p><p>Polish is inexpressible in terms of sonority distance, because then, the</p><p>preference should be the reverse of what it is.</p><p>When sonority scales are fine-tuned to include distinctions among a</p><p>group of stops or fricatives, the voiceless congeners are at the bottom of</p><p>sonority, not the voiced ones. On the other hand, in strength systems, we</p><p>may put voiced obstruents at the top of the strength scale, but only as a</p><p>result of an arbitrary decision. The Element Theory, on the other hand, may</p><p>deem voiced stops or fricatives stronger (more complex) than their voice-</p><p>less congeners by representing the former as containing the low tone ele-</p><p>ment and representing the voiceless ones as neutral. This decision, how-</p><p>ever, must follow from a thorough analysis of the behaviour of the</p><p>obstruents in a given system and should never be taken contingently, just to</p><p>account for one effect.</p><p>Given that Irish voiceless obstruents are marked and contain the high</p><p>tone element as argued in e.g. Cyran (1997), the Element Theory correctly</p><p>identifies the voiceless stops in Irish as the class with the highest complex-</p><p>ity, in which case the interpretation of the preferred contacts would go</p><p>along the same lines as in sonority and strength models.48</p><p>Below, in (20), a first approximation of the representation of Irish ob-</p><p>struents is attempted. The representations clearly demonstrate that stops</p><p>are the most complex of obstruents as they have the additional element (/).</p><p>Among the stops, however, the voiceless ones are still more complex than</p><p>the voiced ones because they possess the high tone element. Thus, it seems</p><p>that we are able to point to a precise place on the complexity scale pertain-</p><p>ing to the Irish obstruents where voiceless stops begin to pattern on their</p><p>own, in contradistinction to the remaining obstruents.</p><p>48 In the following chapter complexity is integrated into a model of consonant inter-</p><p>action, based on governing relations in which the concept of complexity slopes will</p><p>become more meaningful.</p><p>32 Substantive complexity</p><p>(20) Irish obstruents49</p><p>[p] [b] [t] [d] [k] [g] [f] [v] [s] [S] [x] [V]</p><p>U</p><p>h</p><p>/</p><p>H</p><p>U</p><p>h</p><p>/</p><p>A</p><p>h</p><p>/</p><p>H</p><p>A</p><p>h</p><p>/</p><p>_</p><p>h</p><p>/</p><p>H</p><p>_</p><p>h</p><p>/</p><p>U</p><p>h</p><p>H</p><p>U</p><p>h</p><p>A</p><p>h</p><p>H</p><p>I</p><p>h</p><p>H</p><p>_</p><p>h</p><p>H</p><p>_</p><p>h</p><p>Note, however, that voiced stops have exactly the same complexity as the</p><p>voiceless fricatives, that is, three elements. Thus, we may predict that if the</p><p>cut-off point was made one element lower on the complexity scale, then</p><p>voiceless stops would pattern with voiced stops and voiceless fricatives.</p><p>However, in Irish, the cut-off point sets the voiceless stops aside from the</p><p>remaining obstruents.</p><p>At this point, the question arises as to how we should treat the place-</p><p>defining elements with respect to complexity slopes. If they count, then</p><p>Irish [k] should pattern with voiced stops, and we should expect it to fea-</p><p>ture in the data illustrating epenthesis, which is not the case. This leaves us</p><p>with two alternatives. Either velarity must be represented by a real element,</p><p>for example, by means of the neutral element, or resonance elements must</p><p>be claimed not to contribute to the inherent complexity of segments. Be-</p><p>low, we assume that the latter option is correct, that is, generally resonance</p><p>elements do not contribute to the complexity slopes.50 However, they play</p><p>an important role in the aforementioned homorganicity condition.</p><p>Let us look at the representation of a) good (e.g. coirpeach [kIr´"p´ax]</p><p>‘criminal’), b) acceptable (e.g. banda ["baund´] ‘band’) and c) illegal sylla-</p><p>ble contacts (e.g. geirbe ["g´er´´b´´] ‘scab, gen.sg.’) with respect</p><p>to their</p><p>complexity slopes.</p><p>The structures in (21) represent a scale of preferences which can be in-</p><p>terpreted in the following way. Irish clusters with a complexity differential</p><p>of three elements display integrity (21a) and are never broken up by an</p><p>epenthetic vowel. The same complexity slope is found in the case of other</p><p>49 The list ignores the contrast between palatalized and velarized consonants. It also</p><p>does not include the glottal fricative /h/ whose status is unclear.</p><p>50 For the opposite conclusion see e.g. Scheer (2004: 59), who restricts complexity</p><p>to counting only the resonance elements, with sonority being derived by three pa-</p><p>rameters: a) the status of the aperture element (A), b) syllabic position (nucleus or</p><p>onset), and c) presence or absence of manner elements (h) and (/).</p><p>Substantive complexity effects in Irish 33</p><p>voiceless stops, e.g. oilceas [il´k´´s] ‘evil’, portach [p´r"tAx] ‘bog’, because</p><p>they all contain noise, stopness and the high tone. On the other hand, clus-</p><p>ters in which the differential is two elements or less must be broken up by</p><p>epenthesis (21c). This ‘class’ of segments includes voiced stops and voice-</p><p>less fricatives, as in e.g. feirge ["f´er´´g´´] ‘anger, gen.sg.’, dearfa ["d´ar´f´]</p><p>‘proved’, forcha ["fur´x´] ‘beetle’. However, excluded from epenthesis are</p><p>those obstruents which share the same place elements (21b), as in partial</p><p>geminates, e.g. banda ["baund´] ‘band’.</p><p>(21) a. /kI r´ p´ ax/ b. /bau n d ´/ c. /g´e r´ b´ ´/</p><p>A U A=A A U</p><p>h N h h</p><p>/ / /</p><p>H</p><p>The representations in (21) show that complexity, coupled with additional</p><p>conditions such as the one on homorganicity, can account for Irish phono-</p><p>tactics. It should be stressed that the distinction between voiceless and</p><p>voiced stops which in this model follows directly from the specification of</p><p>the laryngeal dimension is not always obvious in models based on strength</p><p>or sonority.51</p><p>One question which remains unanswered is why such stringent condi-</p><p>tions on clustering should hold in Irish.52 Harris (1990), in his first discus-</p><p>sion of element complexity and syllable contacts claims that sometimes the</p><p>accepted coda-onset clusters may have even equal complexity. Why is an</p><p>obstruent which is more complex from the preceding sonorant by two ele-</p><p>ments not a good contact? The answer to this question is two-fold. Substan-</p><p>tive complexity merely provides a non-arbitrary scale with cut-off points.</p><p>However, where exactly the grammar of a particular language chooses to</p><p>place the divisions is an arbitrary property of that grammatical system.53</p><p>Thus, so far, the Element Theory may be said to cover similar empirical</p><p>ground as sonority and strength-based models. It also has the potential of</p><p>51 For example, Clements (1990) proposes a universal sonority scale in which no</p><p>distinction is made between stops and fricatives, not to mention one between voi-</p><p>ced and voiceless stops, a distinction which seems to be called for in Irish.</p><p>52 In the following chapter, we will see that in Dutch the conditions on cluster integ-</p><p>rity are slightly different.</p><p>53 In chapters 2 and 3 we will look at other conditions on cluster integrity, e.g. the</p><p>shape of the following vowel.</p><p>34 Substantive complexity</p><p>doing a little better when it comes to phonotactics without losing any of its</p><p>restrictiveness. This can be achieved if two assumptions are made. Firstly,</p><p>there is a small universal set of privative primes called elements (A, I, U, h,</p><p>/, L, H). Secondly, there is no such thing as a universal uniformity of seg-</p><p>mental representations. The latter assumption is necessary to avoid such</p><p>theoretical pitfalls as providing melodic representations for all the contrasts</p><p>found in the IPA chart – a tendency which besets most distinctive feature</p><p>theories and some element-based ones. Rather, the small set of elements must</p><p>be utilized in representations on the basis of an in-depth system analysis.</p><p>Despite the paucity of the model, there are a few variables which allow</p><p>for a fairly accurate description of any system. One of these parameters</p><p>involves the status of the resonance elements (headedness), which was</p><p>most visible in the discussion of the vowel system of Irish. In consonants,</p><p>the status of resonance elements is inherently connected with primary and</p><p>secondary place. Another variable is connected with the language-specific</p><p>utilization of the source elements, which leads to particular subtle distinc-</p><p>tions between the marked and the neutral obstruents.</p><p>Ideally, the proposed representations should find support in other areas</p><p>of a given phonological system. Convergence of disparate effects is one of</p><p>the symptoms of correctly established representations. Below, we consider</p><p>yet another variable, this time involving the utilization of the noise element</p><p>(h) across languages. This will allow us to see better how various aspects</p><p>of the system of Irish consonants converge.</p><p>3.3. The h-parameter</p><p>It seems that we may further refine our understanding of the Irish conso-</p><p>nantal phonology by making a particular systemic claim.</p><p>The claim concerns the systematic absence of the noise element (h) in</p><p>the Irish language, and possibly also Welsh, as will be shown in the follow-</p><p>ing section. This proposal was first made in Cyran (1996b) and argued for</p><p>on the basis of a number of phonological phenomena in Irish.54 Let us first</p><p>see how the scale of preferred clusters looks after this innovation, before we</p><p>54 There is a general tendency in GP to eliminate the manner elements (h, /) com-</p><p>pletely (e.g. Jensen 1994, Kaye 2001, Ploch 1999, Pöchtrager 2006, Ritter 1997).</p><p>In this work, it is assumed that both of them are necessary, while noise can be ab-</p><p>sent in systems as a parametric choice, just as some languages may choose not to</p><p>utilize any of the tone elements, and have just one series of obstruents, e.g. Malak-</p><p>malak (12).</p><p>Substantive complexity effects in Irish 35</p><p>offer some additional support for the claim, which is based on the analysis of</p><p>the segmental inventory of Irish.</p><p>(22) a. /kI r´ p´ ax/ b. /bau n d ´/ c. /g´e r´ b´ ´/</p><p>A U A=A A U</p><p>/ N / /</p><p>H</p><p>Notice that the systematic absence of the noise element makes all the ob-</p><p>struents less complex. However, the relevant distinctions which were dis-</p><p>cussed under (21) still hold. The voiceless stops continue to be the most</p><p>complex. What needs to be said in the case of the new representations in</p><p>(22) is that the cut-off point between good contacts and bad ones is below</p><p>two manner elements. If the obstruent has two such elements, it may form a</p><p>good contact. If the obstruent is less complex, the cluster will have to be</p><p>broken-up by epenthesis. As mentioned earlier, contacts such as those in</p><p>(22b) survive due to homorganicity, not complexity.</p><p>It will be recalled that substantive complexity merely provides a non-</p><p>arbitrary scale with cut-off points, whose choice is then an arbitrary property</p><p>of particular grammars. However, given the general principle that good</p><p>contacts are those with steep complexity slope, we may venture a compari-</p><p>son between Irish and other languages, in which the consonantal system is</p><p>not deprived of the noise element and the complexity slopes are thereby</p><p>steeper. It may be the case that the internal representation of Irish obstru-</p><p>ents has some influence on the fact that the clustering in this system is so</p><p>stringently conditioned, and why seemingly the same cluster, say [rb], is</p><p>better off in a system like English or Polish.55</p><p>(23) a. Irish b. English c. Polish</p><p>r b r p r b r p r b r p</p><p>A U A U A U A U A U A U</p><p>/ / / / / /</p><p>H h h h h</p><p>H L</p><p>55 Let us assume that we are dealing with a rhotic variety of English, that is one in</p><p>which [r] is pronounced in the coda.</p><p>36 Substantive complexity</p><p>In fact, the best [rb] contact is found in Polish in which voiced obstruents</p><p>have an additional element (L). Note that even in contexts for devoicing,</p><p>that is, (L)-delinking, the Polish cluster [rp] will still exhibit the same</p><p>complexity slope as English [rb] (23b) or Irish [rp] (23a). Thus, Irish [rp]</p><p>patterns with Polish and English in terms of its complexity differential as</p><p>well as in showing no epenthesis. Note that the Irish [rb] sequence exhibits</p><p>some degree of complexity steepness but is still disallowed, in contradis-</p><p>tinction to Harris’ position that coda-onset sequences may be even equal in</p><p>terms of complexity. This question will be taken up in the following chap-</p><p>ters. It seems that some additional factor is at play, which makes such con-</p><p>tacts possible in, for example, English, but totally illicit in Irish.</p><p>In what follows we will take a closer look at elemental representations</p><p>of Irish consonants in the light of this new assumption concerning the noise</p><p>element, and see how the segmental inventory and certain phonological proc-</p><p>esses can now be understood better.</p><p>3.4. Segmental inventories and complexity</p><p>There are two aspects of the consonant system of Irish which attract one’s</p><p>attention immediately. The first one concerns the quality distinction be-</p><p>tween the palatalized and velarized series, which was briefly mentioned</p><p>earlier in the discussion of vocalic alternations. The second characteristic</p><p>feature is the presence of word-initial consonant mutations which occur in</p><p>particular morpho-syntactic contexts.56 In the list of consonants below,</p><p>palatalization is marked by the diacritic ‘´’, while velarization is not repre-</p><p>sented by any diacritic.</p><p>(24) Irish consonants</p><p>Labial p, p´, b, b´, f, f´, v, v´, m, m´</p><p>Coronal t, t´, d, d´, s, S, n, n´, l, l´, r, r´57</p><p>Velar k, k´, g, g´, x, x´, V, V´, N, N´</p><p>Glottal h</p><p>A few comments are in order here concerning the status of the consonants</p><p>listed above. In word-initial position [x, x´, v, v´, V, V´] occur only in leni-</p><p>56 An analysis of initial mutations in a related language, that is, Welsh is offered in</p><p>the following section.</p><p>57 The palatalized version of [s], that is, [S] is in fact palatal.</p><p>Substantive complexity effects in Irish 37</p><p>tion contexts, that is, they are derived, as it were, from [k, k´, S, b, m, f, b´,</p><p>m´, f´, g, d, g´, d´]. Of these restricted fricatives, [x] has the widest distri-</p><p>bution as it also occurs intervocalically, in clusters, and word-finally. [v]</p><p>tends to freely alternate with [w] word-initially. Both [v] and [v´] are found</p><p>finally, but they tend to be elided in intervocalic position, as will be shown</p><p>later. On the other hand, [f] is restricted in word-final position to two items</p><p>which may be native and a handful of borrowings (see Doyle and Guss-</p><p>mann 1996: 135). The pair [V, V´] is not found outside the initial mutation</p><p>context, while [h] not only does not have a palatalized congener (Ó Cu-</p><p>ív1975: 11), but it is restricted to initial position of lexical items which are</p><p>mostly borrowings. For this reason, [h] will be kept out of the discussion of</p><p>the Irish consonant system below.58</p><p>In general, what is striking about the system of consonants in Irish is the</p><p>restricted distribution of fricatives, of which [f, s, S, x] seem to fair best,</p><p>and a very low profile which is kept by the voiced fricatives, a point which</p><p>calls for a principled account.</p><p>In order to see the peculiarities of the Irish system better it will be com-</p><p>pared with that of Polish, and Malakmalak.59 The palatalized / velarized</p><p>distinction in Irish is disregarded in what follows.</p><p>(25) plosives affricates fricatives</p><p>Polish p t k t °s t°Ç t°S f s Ç S x</p><p>b d g d °z d °Û d °Z v z Û Z −</p><p>Irish60 p t k − − − f s − S x/h</p><p>b d g − − − v? − − − V?</p><p>Malakmalak p t tj k − − − − − − − −</p><p>− − − − − − − − − − − −</p><p>58 Some examples are: haircín ‘hurricane’, hata ‘hat’, hidrigin ‘hydrogen’, histéire</p><p>‘hysyeria’, héileacaptar ‘helicopter’.</p><p>59 Malakmalak is an Australian language whose consonantal system involves the</p><p>following objects (Maddieson 1984: 327): Stops: p, t, tj, k; Nasals: m, n, nj, N; Liq-</p><p>uids: r, ®, l, lj, j, w. The superscripted ‘j’ denotes palato-alveolars, contrasting with</p><p>alveolars.</p><p>60 Marginally, one comes across instances of [d °Z] and [z] in Irish. However, they</p><p>can hardly be treated as part of the phonological system.</p><p>38 Substantive complexity</p><p>Let us first identify the similarities and differences between Polish and</p><p>Irish obstruents. In broad phonemic terms, the two languages seem to have</p><p>analogous systems of stops. Practically, this is where the surface similari-</p><p>ties end. Polish has a group of affricates while Irish has none.61 And fi-</p><p>nally, while Polish has a fairly symmetrical system among the fricatives in</p><p>terms of voicing which is also reflected in its affricates and stops, Irish has</p><p>a defective system in which the voiced fricatives are highly restricted, or</p><p>virtually non-existent. One might ask a number of questions concerning the</p><p>defective Irish system, for example, why there are voice contrasts among</p><p>the stops but not among the fricatives? Why the systems of stops appear to</p><p>be similar in Irish and Polish while everything else is so different? Could</p><p>we expect the reverse situation? What is the nature of the gap concerning</p><p>the affricates? Could we predict a system in which there are no fricatives,</p><p>but there are voiceless affricates? Is there any formal connection between</p><p>the absence / presence of voice contrasts among fricatives and the absence</p><p>/ presence of affricates? How Malakmalak fits the picture?</p><p>It turns out that most of the above questions can be answered by refer-</p><p>ring to a single representational aspect which makes all the difference be-</p><p>tween Polish and Irish. The answer involves the aforementioned hypothesis</p><p>that Irish does not make use of the noise element (h). Metaphorically, we</p><p>may say that Polish is a ‘noisy’ language, while Irish is ‘noiseless’. Let us</p><p>compare the representations of obstruents in Polish and Irish beginning</p><p>with the stops.</p><p>(26) Polish stops Irish stops</p><p>[b] [p] [d] [t] [g] [k] [p] [b] [t] [d] [k] [g]</p><p>U</p><p>h</p><p>/</p><p>L</p><p>U</p><p>h</p><p>/</p><p>A</p><p>h</p><p>/</p><p>L</p><p>A</p><p>h</p><p>/</p><p>_</p><p>h</p><p>/</p><p>L</p><p>_</p><p>h</p><p>/</p><p>U</p><p>/</p><p>H</p><p>U</p><p>/</p><p>A</p><p>/</p><p>H</p><p>A</p><p>/</p><p>_</p><p>/</p><p>H</p><p>_</p><p>/</p><p>The above representations show how identically looking segmental inven-</p><p>tories of stops are dramatically different phonologically. They show that</p><p>making lists of segments or phonemic inventories is both futile and mis-</p><p>leading if a thorough analysis of the internal structure does not go with it.</p><p>61 In Donegal Irish palatalized dentals are pronounced with affrication (Ó Dochar-</p><p>taigh 1987). However, this need not be viewed as the presence of affricates in the</p><p>system.</p><p>Substantive complexity effects in Irish 39</p><p>Firstly, the voice specification is different in that Polish uses (L) versus</p><p>nothing, whereas Irish has (H) versus nothing. Secondly, the noise element</p><p>(h) features with a vengeance in Polish, but is completely missing in Irish.</p><p>It is only to be expected that the two systems will also behave differently.</p><p>In Polish we have devoicing of obstruents, while in Irish the more complex</p><p>character of the voiceless stops plays an important role in clustering, as we</p><p>saw in the previous sub-section. What is worth noting is the representation</p><p>of [g] in Irish which is as different from its Polish counterpart as can be. In</p><p>Polish, this consonant is made up of three elements while in Irish it con-</p><p>tains only one, the stopness itself. How is this single element interpreted</p><p>as</p><p>a voiced velar plosive? Firstly, it is a plosive because it has the stopness</p><p>element. It is velar because it has no place specification, and it is voiced</p><p>because the system interprets unmarked obstruents as voiced. If Polish was</p><p>a ‘noiseless’ language like Irish, but continued to have the same specifica-</p><p>tion of voicing (L vs. nothing), then the representation of the stopness ele-</p><p>ment alone would give a phonetic [k].</p><p>As we will see shortly, the simplex representation of [g] in Irish allows</p><p>us to understand why this is the only stop which is deleted intervocalically.</p><p>To conclude our discussion of Polish and Irish stops, we must emphasize</p><p>the fact that, despite the impoverished representations, all the existing con-</p><p>trasts in Irish are captured in our system.</p><p>The problem of the absence of affricates in Irish may follow from the</p><p>absence of (h) as well. Harris (1990) proposes that the representation of</p><p>affricates involves a contour structure whereby the relation between stop-</p><p>ness (/) and noise (h) is broken up. Below we give tentative representations</p><p>of Polish affricates, with headedness deliberately unspecified.</p><p>(27) Polish affricates</p><p>[t°s] [d °z] [t°Ç] [d °Û] [t°S] [d °Z]</p><p>A A A.I A.I I I</p><p>/ h / h / h / h / h / h</p><p>L L L</p><p>Whether affricates are indeed contour structures, or the mere presence of</p><p>‘noise’ brings out the effect of affrication, one thing is clear. A system</p><p>40 Substantive complexity</p><p>without ‘noise’ should not have affricates, and Irish is such a system.62 On</p><p>the other hand, a language with ‘noise’ may have affricates and Polish is an</p><p>example.</p><p>Let us now turn to the difference between Polish and Irish fricatives.</p><p>This class of segments is the most interesting with respect to the proposal</p><p>that ‘noise’ is missing in Irish, because this category is normally responsi-</p><p>ble for aperiodic energy in the acoustic signal, that is, friction. As shown in</p><p>the representations of Polish fricatives in (28), the noise element is the</p><p>inherent property of this class of obstruents just as stopness was the defin-</p><p>ing property of plosives. Irish has to do without the noise category, and it</p><p>does. The question is, how?</p><p>It has been proposed within Government Phonology that the headedness</p><p>of resonance elements, which produces tenseness in vowels, may also bring</p><p>about the stronger articulation in non-vocalic positions, whereby, the rep-</p><p>resentation of fricatives without noise becomes possible (Cyran 1996b,</p><p>Ritter 1997). Thus, it is possible to say that a labial fricative is a phonetic</p><p>interpretation of a headed (U) element, which, when headless, represents</p><p>the labial-velar glide [w]. Then, the fluctuations of the type [v~w] and in-</p><p>deed [r~|] in Irish can be viewed as head switches (U)~(U) and (A)~(A)</p><p>respectively.63</p><p>(28) Polish fricatives Irish fricatives</p><p>[v] [f] [z] [s] [Û] [Ç] [Z] [S] [x] [f] [v] [s] [S] [x]</p><p>U</p><p>h</p><p>L</p><p>U</p><p>h</p><p>A</p><p>h</p><p>L</p><p>A</p><p>h</p><p>A.I</p><p>h</p><p>L</p><p>A.I</p><p>h</p><p>I</p><p>h</p><p>L</p><p>I</p><p>h</p><p>_</p><p>h</p><p>U</p><p>H</p><p>U A</p><p>H</p><p>I</p><p>H</p><p>_</p><p>H</p><p>Concentrating now on the remaining Irish fricatives, it will be noted that</p><p>this way of representing the Irish fricatives suggests that the voiceless se-</p><p>ries contrasts directly with glides and liquids, phonologically speaking, and</p><p>62 See Rubach (1994) for a proposal that Polish affricates are strident stops from</p><p>the point of view of phonology, and also Rennison (1998) who considers other</p><p>formal devices to replace contour structure. Rennison’s proposal still relies on a</p><p>physical presence of particular primes, in this case the ‘noise’ element.</p><p>63 See Cyran and Nilsson (1998) for a discussion of the Slavic shift [w] > [v] which</p><p>involves two different alternations: [w~v], that is (U)~(U), and [v~f], that is</p><p>(U,h,L)~(U,h). See also Golston and van der Hulst (2000) who derive stricture</p><p>from structure rather than from a separate melodic prime.</p><p>Substantive complexity effects in Irish 41</p><p>not with voiced fricatives. A true phonological voiced fricative in Irish is</p><p>possible only if ‘noise’ is part of the system. Thus, Irish [f] contrasts di-</p><p>rectly with an object which is sonorant-like in character, as it is represented</p><p>only by one resonance element.64 Note that [s] directly contrasts with (A),</p><p>which stands for the phonetic trill [r]. This ‘noiseless’ system has therefore</p><p>a direct influence on the absence of [z] in Irish.65 Furthermore, Irish [S]</p><p>directly contrasts with [j], hence there is no [Z]. In fact, the palatal glide in</p><p>Irish also has a near-fricative realization, that is, [J], which is the result of</p><p>the headed status of the element (I). It seems that we can also account for</p><p>the most restricted Irish fricative, that is [V], which occurs only as a result</p><p>of lenition word-initially. Depending on the way we represent velarity – it</p><p>will be recalled that velarity may also be represented by means of the so</p><p>called neutral element – this segment may be the realization of a neutral ele-</p><p>ment, or, as suggested by the above representations, it is the phonetic inter-</p><p>pretation of an empty onset.66</p><p>Finally, a word of comment is in order concerning the fricative [x]. It is</p><p>a headless object because it has no place element. On the other hand, it</p><p>contains the high tone element which is also responsible for voicelessness</p><p>and aspiration. Quite possibly we are dealing here with a third source of</p><p>phonetic friction in the Element Theory. Namely, next to the noise ele-</p><p>ment, which is absent in Irish, and headedness of the resonant element,</p><p>which appears to be utilized to the full in this system, also the high tone</p><p>64 The voiced labial fricative is notorious for displaying sonorant-like characteris-</p><p>tics cross-linguistically. This is true of, for example, Russian (Andersen 1969),</p><p>Polish (Gussmann 1981, 2002), Slovak (Rubach 1993), Hungarian (Siptár 1996,</p><p>Szigetvári 1998).</p><p>65 Irish seems to be the type of system that, provided it had a process of H-deletion,</p><p>would exhibit the rhotacism of [s] > [r] instead of the voicing of [s] > [z] in a proc-</p><p>ess like Verner’s Law which generally voices fricatives (Cyran 1997: 192).</p><p>66 Empty onsets in Irish may license velarization or palatalization (Cyran 1997),</p><p>hence, [V] and [V´] could indeed be treated as empty onsets with secondary articu-</p><p>lation, a situation comparable to the [u~i] alternations discussed in 4.1., where</p><p>nuclei had no specification and were interpreted as [u] or [i] depending on what</p><p>secondary specification was lodged on the following consonant. Note that [V´] is in</p><p>fact [J], that is, a palatal fricative, and as we remember, the (I) element of palatali-</p><p>zation affects objects as the head, hence the friction is expected in each such case.</p><p>For this reason Irish does not exhibit the alternation [J]~[j], that is a (I)~(I) fluctua-</p><p>tion (Ó Cuív 1975: 42).</p><p>42 Substantive complexity</p><p>may be responsible for the phonetic effect of friction.67 This richness of the</p><p>potential sources of one phonetic effect shows manifestly that any ad hoc</p><p>representation of phonetic facts by means of elements is likely to be wrong,</p><p>as phonetic effects of different theoretical categories may overlap. This,</p><p>however, is not a drawback of the element system but rather its inbuilt</p><p>positive potential to cover, for example, such phenomena as phonologiza-</p><p>tion, reanalysis, and other phenomena leading to language change.68</p><p>In general, the existing phonetic voiced fricatives in Irish are not fully-</p><p>fledged phonological objects, which is a result of the missing noise ele-</p><p>ment and the fact that voiced friction is derived by other, less stable means,</p><p>that is headedness. Thus, it seems that the systematic absence of (h) almost</p><p>single-handedly accounts for the main distinctions between Polish and Irish</p><p>obstruent systems. What initially appeared to be a defective and asymmet-</p><p>rical system, turns out to be</p><p>perfectly symmetrical, given the resources it</p><p>has at its disposal. Its defective nature follows from the fact that not all</p><p>universally recognized primes are utilized. Specifically, Irish makes no use</p><p>of the ‘noise’ category. On the other hand, we see that Polish must have</p><p>this category to be able to express the additional contrasts, that is, affrica-</p><p>tion and voice among the fricatives. Hence, we propose that the utilization</p><p>of the noise element is subject to parameterization.69</p><p>It should be borne in mind that almost every obstruent in Irish differs</p><p>markedly from the phonetically corresponding object in Polish. This also</p><p>refers to the seemingly identical systems of plosives (25). What these re-</p><p>sults demonstrate is that it is impossible and hence erroneous to determine</p><p>the phonological composition of segments on phonetic grounds, or on the</p><p>basis of a superficial analysis. Phonetic contrasts in voicing do not directly</p><p>correspond to particular laryngeal elements. Likewise, some phonetic con-</p><p>trasts in manner of articulation need not directly correspond to particular</p><p>elements. This is evident in the case of friction.70</p><p>67 In fact, the effects formerly ascribed to the noise element (h) are now often at-</p><p>tributed to the dual behaviour of the high tone (H). See e.g. Kaye (2001).</p><p>68 For an example of phonologization and language change in the Element Theory</p><p>see Cyran and Nilsson (1998), which concerns the Slavic shift from [w] to [v].</p><p>69 It seems now, that any attempt to eliminate the noise element in the Element</p><p>Theory should be able to offer a new interpretation of noisy and noiseless lan-</p><p>guages like Polish and Irish, respectively.</p><p>70 We make no particular claims concerning stopness (/) and nasality (N). See how-</p><p>ever Jensen (1994), Nasukawa (2005), Pöchtrager (2006), Ploch (1999).</p><p>Substantive complexity effects in Irish 43</p><p>Let us briefly return to the comparison in (25) and in particular to the</p><p>system of Malakmalak consonants. For fear of contradicting the above</p><p>conclusions we may tentatively suggest what parameters should be checked</p><p>to account for what this language has. It seems that it can be characterized</p><p>first of all as ‘toneless’, that is, neither (H) nor (L) is used. For this reason,</p><p>there is only one series of stops, that is the voiceless unaspirated one. Ma-</p><p>lakmalak also appears to be ‘noiseless’. The absence of (h) excludes affri-</p><p>cates and fully-fledged fricatives. Potentially, such a system could have</p><p>voiced fricatives if its resonance elements could be headed. Since there are</p><p>only glides and liquids, we may suspect that this language does not utilize</p><p>headedness either. Recall, that fricatives have three representational sour-</p><p>ces: a) the noise element (h), b) headedness of resonance elements, and c)</p><p>high tone, as in the case of Irish [x]. Since no analysis of Malalkmalak is</p><p>provided here, the above definition of its consonantal system must, of</p><p>course, be viewed as a sheer speculation.</p><p>Returning, again, to the Irish consonants we must emphasize that the</p><p>impoverished (h-less) system allows us to understand better quite a number</p><p>of seemingly unrelated issues. Firstly, we saw that the less complex nature</p><p>of Irish obstruents sets it clearly apart from ‘noisy’ languages such as Eng-</p><p>lish and Polish in terms of phonotactic restrictions, or more specifically,</p><p>clustering (23).71 The uniform absence of ‘noise’ also accounts for some</p><p>crucial aspects of the segmental inventory of Irish consonants, in particu-</p><p>lar, the absence of voice oppositions among the fricatives and the complete</p><p>absence of affricates.72 It seems that we also gain an insight into a few o-</p><p>ther phenomena concerning Irish consonants. For example, in lenition con-</p><p>texts, [m] and [b] lenite to [v]. If the fricative were a typical obstruent con-</p><p>taining (h), we would be able to understand [b] > [v], but not [m] > [v], in</p><p>which case the phenomenon would have to involve the loss of nasality and</p><p>the addition of noise.73 Under the ‘no-noise’ assumption this problem is</p><p>non-existent, as the shift [m] > [v] is simply (U,N) > (U).</p><p>Earlier, in our discussion of Irish stops, we mentioned that [g] is now a</p><p>simplex object (/._). On the other hand, in models operating with sonority</p><p>or strength scales the position of this segment in Irish should be relatively</p><p>71 Note that the representations in (23) show distinctions which are hardly calcula-</p><p>ble in sonority-based systems.</p><p>72 One should bear in mind that the presence of (h) does not guarantee the presence</p><p>of affricates. What is meant here is that the absence of (h) in a system means that it</p><p>will not have such objects.</p><p>73 The following section deals with Celtic mutations in much more detail.</p><p>44 Substantive complexity</p><p>analogous to that of [g] in English or Polish. For these models, deletion of</p><p>[g] in intervocalic position, or in Welsh mutations (see section 4), is a com-</p><p>plete accident. In a model operating with sub-segmental complexity in terms</p><p>of elements, this is fully predicted, given the correctness of the ‘no-noise’</p><p>hypothesis.</p><p>To conclude this section, let us consider the following phenomenon.</p><p>Two segments, [g] and [v], tend to be deleted in intervocalic position in</p><p>Munster Irish. As a result, a long vowel is created. The velar plosive is lost</p><p>in the verbal system when the first person ending −im is added to a stem</p><p>ending in this consonant (29a). That the personal ending contains a lexical</p><p>vowel is shown by such forms as las / lasaim [lAs / lAsIm´] ‘light / I light’.</p><p>The labial fricative, on the other hand, is lost in the nominal system when a</p><p>vocalic ending is added (29b). In this case, we are dealing with the same</p><p>kind of genitive formation as in cearc / circe [k´ark / k´ir´k´´] ‘hen /</p><p>gen.sg.’, that is, by addition of the ending −e, which palatalizes the preced-</p><p>ing consonant.</p><p>(29) Imperative Ist person sg. ([-im´])</p><p>a. [n´ig´] nigh [n´i:m´] ním ‘wash’</p><p>[sig´] suigh [si:m´] suím ‘sit’</p><p>b. [uv] ~ [i:]74 ubh / uibhe ‘egg/gen.sg.’</p><p>[n´iv´] ~ [n´i:] nimh / nimhe ‘poison/gen.sg.’</p><p>74 This form is pronounced as [iv´´] in Connemara and Donegal Irish.</p><p>The question is what [g] and [v] have in common to be deleted intervocali-</p><p>cally, or what makes them different from other consonants? A quick look at</p><p>the representation of these objects in terms of elements tells us that the two</p><p>consonants are mono-elemental. Note that in the lenition trajectories dis-</p><p>cussed in, for example, Lass (1984), or Harris (1990, 1996) the pre-deletion</p><p>stages usually involve simplex objects like glottal stops [/], or glottal frica-</p><p>tives [h], which in the Element Theory are simplex objects. What the pre-</p><p>deletion stages of Irish [g] and [v] have in common with other known types</p><p>of deletable objects is precisely the same complexity, that is, being repre-</p><p>sented by only one element. This connection does not follow from any</p><p>scale of sonority or strength unless the scales are seriously manipulated, or</p><p>arbitrarily set on the basis of observation. In our model, such facts follow</p><p>directly from the internal representation of consonants in a given system,</p><p>Initial consonant mutations in Welsh 45</p><p>which must be first arrived at through analysis. Note that the representation</p><p>of Irish obstruents has been shown to converge on a few disconnected as-</p><p>pects of the phonological system. These areas are phonotactics, segmental</p><p>inventories, and phonological processes.</p><p>In the following section the complexity-based model is tested against</p><p>the well-known phenomenon of consonant mutations. This will be done on</p><p>the basis of data from Welsh, which, like Irish, is a Celtic language, and is</p><p>also h-less.75</p><p>4. Initial consonant mutations in Welsh</p><p>4.1. Introduction</p><p>Alternations of initial consonants, called mutations, are among the most</p><p>distinctive traits of Celtic</p><p>languages.76 As a result of these mutations, a few</p><p>different surface forms of a given lexical item can be observed depending</p><p>on the grammatical context. For example, the Welsh word cath [ka:T] ‘cat’</p><p>begins with a voiceless velar plosive in the phrase eu cath [i ka:T] ‘their</p><p>cat’, but there is a corresponding voiced plosive in the phrase ei gath [i</p><p>ga:T] ‘his cat’, a voiceless velar fricative in ei chath [i xa:T] ‘her cat’, and a</p><p>voiceless velar nasal as in fy nghath [v´ N(a:T] ‘my cat’. Similar effects are</p><p>observed in Irish as well. For example, the same word cat [kAt] ‘cat’ ap-</p><p>pears as [´ xAt] ‘his cat’, [´ gAt] ‘their cat’, and [´ kAt] ‘her cat’.</p><p>The above examples serve to illustrate two points. Firstly, they show the</p><p>phonetic correspondences between the various reflexes of the initial con-</p><p>sonant [k] in Welsh and Irish, which are clearly phonologically related and</p><p>do not form arbitrary sets of forms, even though, analogous grammatical</p><p>75 For a thorough analysis of the Irish initial mutations within the Element Theory</p><p>can be found in Jaskuła (2006). In what follows some comparison will be made</p><p>between Irish and Welsh to show that the switch of language in our presentation is</p><p>warranted.</p><p>76 The study of initial consonant mutations in Celtic languages is a well-ploughed</p><p>area. There are numerous accounts of the phenomenon within a number of theoreti-</p><p>cal frameworks. The list of authors which follows is only partial and includes works</p><p>on Welsh, Irish and Scottish Gaelic: Awbery (1973), Ball and Müller (1992), Ewen</p><p>(1982), Green (2003), Grijzenhout (1995), Gussmann (1983, 1986), Hamp (1951),</p><p>Morgan (1952), Ní Chiosáin (1991), Oftedal (1962), Ó Cuív (1986), Ó Siadhail</p><p>(1989), Pilch (1975), Pyatt (1997), Stewart (2004), Thurneysen (1949).</p><p>46 Substantive complexity</p><p>contexts lead to different effects in the two languages. Thus, some phono-</p><p>logical description of the mutations is necessary and warranted. Secondly,</p><p>the above data illustrate the fact that phonetically identical contexts (post-</p><p>vocalic) trigger disparate effects on the initial consonant. In other words,</p><p>the different types of mutations, which are complementary in the respective</p><p>grammatical contexts, are the only exponents of the different meanings.</p><p>The relationship between the phonological context and the effect is broken.</p><p>Thus, at the outset of this discussion we must realize that we are not go-</p><p>ing to discuss a real phonological process that is taking place synchronic-</p><p>ally (Gussmann 1983, Green 2003, Jaskuła 2006). Rather, we will concen-</p><p>trate on the representational relationship between the alternating forms</p><p>with a view to demonstrating that the Element Theory and the concept of</p><p>melodic complexity offer a good tool to understanding these relationships.</p><p>The following analysis will not be that of a live phonological process, but</p><p>it will nevertheless attempt to provide some answers concerning the scope</p><p>of different mutations, the nature of exceptions, as well as shed some light</p><p>on the problem of the learnability of such complex phenomena in language</p><p>acquisition. Phenomena such as reradicalization consisting in mistaken</p><p>uses of mutation patterns to derive wrong basic forms (Ball and Müller</p><p>1992, Chudak, in prep.) suggest that learners do attempt to construct the</p><p>phonological regularity of mutations despite the fact that they are no longer</p><p>truly phonological.</p><p>It is generally assumed among historians of the Celtic languages that o-</p><p>riginally (4th – 5th century AD.) the mutations were purely phonological</p><p>phenomena. We may refer to them as sandhi effects, triggered by clearly</p><p>defined phonological contexts which arose in close syntactic configurations</p><p>such as preverb + verb, article + noun, or noun + adjective. What is inter-</p><p>esting is that the external sandhi phenomena mirrored similar effects within</p><p>the word, which can be illustrated by such Welsh borrowings from Latin as</p><p>apostlus > abostol [abostol], peccātum > pechod [pexod], or Adam >Ad-</p><p>daf [a:Dav]. The effect of lenition illustrated by the above forms may be</p><p>given a structural description as occurring in intervocalic position within</p><p>the word. There is some historical evidence which allows us to assume that</p><p>the sandhi contexts created similar environments to those in which conso-</p><p>nants were lenited word-medially. The identification of the two contexts</p><p>can be schematically illustrated in the following fashion.</p><p>(30) ...VCV... = ...V#CV...</p><p>Initial consonant mutations in Welsh 47</p><p>The sound shifts within the word were lexicalized, giving pechod [pexod]</p><p>and Addaf [a:Dav] in modern Welsh, while the sandhi alternations later</p><p>became grammaticalized (6th century) due to the fact that, for the most part,</p><p>the phonological triggers disappeared together with the loss of final sylla-</p><p>bles. Thus, the mutations themselves continued to act as the exponents of</p><p>gender, number, or case, and de facto became part of morpho-syntax rather</p><p>than phonology proper.</p><p>Before we look at possible triggers for the various mutations let us first</p><p>get a more general view of what can happen to consonants when they are</p><p>mutated. In order to facilitate the comparison between Irish and Welsh a</p><p>somewhat simplified picture is presented where the secondary articulation</p><p>distinctions on Irish consonants are ignored.</p><p>(31) Effects of mutations on a consonant in Irish and Welsh</p><p>Irish</p><p>Basic k g t d p b s f m</p><p>Lenition x V h V f v h P v/w</p><p>Eclipsis g N d n b m - v -</p><p>Welsh</p><p>Basic k g t d p b s f m Ò r 9</p><p>Soft M. g P d D b v - - v l r</p><p>Aspirate M. x - T - f - - - - - -</p><p>Nasal M. N( N n9 n m9 m - - - - -</p><p>It has been noted that, if we disregard other minor changes and adjustments</p><p>in individual cases, the mutations of initial consonants can be captured in</p><p>terms of manipulating only three features (e.g. Awbery 1973, Ball and</p><p>Müller 1992, Fife 1993). This idea is represented below graphically.</p><p>(32) C</p><p>Eclipsis [+voiced] Soft</p><p>Irish [+continuant] Aspirate Welsh</p><p>Lenition [+nasal] Nasal</p><p>The consonants affected by initial mutations can either become voiced,</p><p>spirantized, nasalized, or be deleted. Individual modern Celtic languages,</p><p>however, differ with respect to the actual implementation of the shifts.</p><p>Thus, in Irish, the process of eclipsis, which historical descriptions refer to</p><p>as nasalization (e.g. Thurneysen 1946), results in the voicing of some ob-</p><p>struents and the turning of others into nasal consonants. Lenition in this</p><p>48 Substantive complexity</p><p>language turns stops into fricatives, weakens [s] to [h], deletes [f], and also</p><p>turns [m] into [v/w], as the list in (31) demonstrates.</p><p>In Welsh, on the other hand, lenition, which we will call soft mutation,</p><p>either voices or spirantizes stops, but it also deletes [g], turns [m] into [v],</p><p>and voices [Ò, r 9] to [l, r]. With respect to the spirantization of stops, soft</p><p>mutation coincides in its effects with aspirate mutation in that both proc-</p><p>esses involve turning stops into spirants. However, aspirate mutation only</p><p>affects the voiceless stops, which in soft mutation are voiced rather than</p><p>spirantized. Finally, unlike eclipsis in Irish, the nasal mutation in Welsh</p><p>turns all its targets into nasal sounds, whether voiced or voiceless.</p><p>Before we consider some data illustrating the mutations in Welsh, let us</p><p>briefly look at the putative phonological triggers of the mutations, which</p><p>we can reconstruct to some extent. In general the three main contexts for</p><p>the consonant changes may be represented in the following fashion.</p><p>(33) ...V#CV... > lenition</p><p>...s#CV... > spirantization or no effect</p><p>...n#CV... > nasalization</p><p>Lenition required that the initial consonant found itself in an intervocalic</p><p>environment. If the first element of a close syntactic unit ended with an [s],</p><p>this resulted either in spirantization of the following initial, or in no</p><p>change</p><p>at all. On the other hand, a nasal consonant in that position brought about</p><p>nasalization of the following initial. Some examples concerning Welsh and</p><p>Irish follow below. The capital letter following the forms denotes the type</p><p>of mutation they caused on the initial consonant of the following attribu-</p><p>tive adjective. S = Spirantization / Aspirate mutation, L = Lenition / Soft</p><p>Mutation, N = Nasalization / Nasal Mutation / Eclipsis.</p><p>(34) A reconstructed Brittonic o-stem declension (Russell 1995: 123)</p><p>Singular</p><p>Nom. *mapos > **mab (+S in Breton)</p><p>Acc. *mapon > **mabN</p><p>Gen. *mapi @ > **meibL</p><p>Dat. *mapu@ > **meibL</p><p>Plural</p><p>Nom. *mapi @ > **meibL</p><p>Acc. *mapu@s > **meibS</p><p>Gen. *mapon > **mabN</p><p>Dat. *mapobi/os > **mabofL/S</p><p>Initial consonant mutations in Welsh 49</p><p>Although in modern Welsh most of the case endings are lost, the former,</p><p>reconstructed forms for mab [ma:b] ‘son’ allow us to identify the connec-</p><p>tion between the shape of its final syllable and the type of mutation the</p><p>noun used to impose on the following syntactically close material. A vo-</p><p>calic context led to lenition, a nasal one to nasalization, and the presence of</p><p>[s] either led to spirantization or to nothing. It must be emphasized, how-</p><p>ever, that the absence of mutation refers to the basic form, sometimes</p><p>called the ‘radical’ form, which continued to play a role in the alternations</p><p>and later in the exposition of the particular meanings, just as much as mu-</p><p>tated forms did.</p><p>The following development in the history of Irish illustrates similar ef-</p><p>fects, with the additional comparison between masculine and feminine</p><p>declensions showing the origin of the intricate complementarity of muta-</p><p>tion effects which pervades the grammatical systems of Celtic languages</p><p>even today.</p><p>(35) Development of the Old Irish declension (Russell 1995:41)</p><p>io-stem (masc.) ā-stem (fem.)</p><p>Proto- Proto- Old Proto- Proto- Old</p><p>Celtic Irish Irish Celtic Irish Irish</p><p>Nom. donii8os dunei8ah duine cf. to @ta @ to @ta túathL</p><p>Voc. donii8e donii8e duiniL</p><p>Acc. donii8on donii8e duineN</p><p>Gen. donii 8i @ dunii8i duiniL cf. to @tii8a @s to @tei8ah túaithe</p><p>Dat. donii 8u@ dunii8u duiniuL</p><p>What is interesting in the data above is that the historical mutation effects</p><p>are retained in modern Irish regardless of the fact that the contexts were</p><p>opaque as early as the Old Irish period. Note that the nominative form of the</p><p>masculine noun duine ‘man’ did not mutate the following attributive adjec-</p><p>tive even though the context was vocalic. Likewise, the feminine noun túath</p><p>‘tribe, people’, did cause lenition although it ended with a consonant. How-</p><p>ever, if we go back far enough in the reconstruction of these forms (to Proto-</p><p>Celtic), we may identify the right contexts and gain some inkling as to</p><p>when the mutations were full blooded phonological phenomena, and when</p><p>they became mere exponents of grammatical information such as gender,</p><p>case, etc.</p><p>After this brief and rather general introduction let us examine the ef-</p><p>fects in more detail.</p><p>50 Substantive complexity</p><p>4.2. Soft Mutation (SM)</p><p>Soft mutation, or lenition, is the most pervasive of the initial consonant</p><p>alternations in that it involves the greatest number of targets and triggers. It</p><p>is also the most complex mutation in terms of the number of processes</p><p>involved (Awbery 1973, 1986, Thomas 1992, Ball and Müller 1992, Buc-</p><p>zek 1995). To some extent, SM may also be claimed to be a fairly produc-</p><p>tive phenomenon. Watkins (1993: 306) gives an example of the English</p><p>borrowing chips which begins with a consonant which is not even part of</p><p>the phonological inventory of Welsh but still gets regularly lenited in col-</p><p>loquial speech producing [dZIps] as in a bag of chips [bag o dZIps]. All of</p><p>the targets, with a sample group of triggers, are represented below.77</p><p>(36) Soft Mutation</p><p>Target Example and Trigger</p><p>p > b pen [pen] ‘top, head’ > ar ben [ar ben] ‘on top’</p><p>(nouns after prepositions)</p><p>t > d tad [ta:d] ‘father’ > Duw Dad [diu da:d] ‘God the Father’</p><p>(a noun in apposition)</p><p>k > g dw$r [du:r] ‘water’, ci [ki:] ‘dog’ > dwrgi ["durgi] ‘otter’</p><p>(compound)</p><p>b > v bach [ba:x] ‘little’, merch [merx] ‘girl’ > merch fach [merx va:x]</p><p>‘little girl’ (adjective following a feminine noun)</p><p>d > D daeth [da:iT] ‘he came’ > yr un a ddaeth [´r i:n a Da:iT]</p><p>‘the one who came’ (after the relative particle a)</p><p>g > P gardd [garD] ‘garden’ > yr ardd [´r arD] ‘the garden’</p><p>(feminine singular noun after definite article)</p><p>Ò > l llanw [Òanu] ‘filling, VN.’ > wrth lanw [urT lanu] ‘by filling’</p><p>(verbal noun after preposition)</p><p>r 9 > r rhaff [r 9a:f] ‘rope’ > ei raff [i ra:f] ‘his rope’</p><p>(noun after masculine singular possessive pronoun)</p><p>m > v merch [merx] ‘girl’ > un ferch [i:n verx] ‘one girl’</p><p>(a feminine noun following un ‘one’)</p><p>Although, as mentioned earlier, all the initial mutations in Celtic languages</p><p>can be captured by using the three major features mentioned earlier, that is</p><p>77 For more exhaustive surveys of the triggers of all types of mutations in Welsh</p><p>see, for example, Williams (1980), Ball and Müller (1992).</p><p>Initial consonant mutations in Welsh 51</p><p>[+voice], [+continuant] and [+nasal], a quick glance at only one type of</p><p>mutation in Welsh shows that the situation is rather more complex. In pre-</p><p>theoretical terms we may describe soft mutation as involving five proc-</p><p>esses. The voiceless stops [p, t, k] become voiced [b, d, g], the voiced ante-</p><p>rior stops [b, d] become the corresponding voiced fricatives [v, D], the voiced</p><p>velar plosive [g] is deleted, the voiceless liquids [Ò] and [r9] are voiced, that is,</p><p>they become modal liquids, and the bilabial nasal [m] becomes the labio-</p><p>dental [v].</p><p>Ball and Müller (1992), among others, working within a feature-based</p><p>derivational model, show that these five processes can be collapsed into</p><p>three if certain assumptions are made. Thus, we may talk about one major</p><p>voicing process which turns [p, t, k, Ò, r 9] into their voiced congeners [b, d,</p><p>g, l, r]. The second process would involve the spirantizing of voiced stops</p><p>including [g]. For this description to be upheld, we must assume that at some</p><p>stage of the derivation a voiced velar fricative *[V] is produced which is later</p><p>deleted by other rules. And finally, the process turning [m] into [v] may be</p><p>included in the second group, that is spirantization, under the proviso that,</p><p>here too, the derivation is allowed to involve an abstract stage with a non-</p><p>existent nasalized voiced labio-dental fricative *[v)], that is m > *v ) > v.</p><p>Here, parallel to the derivation of g > *V > P, the simplification is effected</p><p>at the cost of introducing abstract stages and having to posit further ad-</p><p>justment rules to arrive at the correct result.</p><p>As for the voicing process, one has to express some reservations con-</p><p>cerning the theoretical validity of this generalization. The problem con-</p><p>cerns not so much the formulation of the rule as the representation of the</p><p>targets and products of the rules. It is normally assumed, at least in priva-</p><p>tive feature models, that voicing of sonorants is a default property which</p><p>need not be specified phonologically, while it must be specified in obstru-</p><p>ents (e.g. Lombardi 1995). The treatment of plosives and liquids as a natural</p><p>class for the purpose of unifying the process of SM, in total disregard of the</p><p>asymmetry in their universal voice specification, seems to be rather dubious.</p><p>The spirantizing rule, on the other hand, raises the question of the ab-</p><p>stractness of phonological derivations, while leaving us with no way of</p><p>getting round the problem of the need for a number of phonetic adjustment</p><p>rules, for example, one turning bilabial place into labio-dental in m > v.</p><p>Though we can posit a late rule of *V-deletion or *v )-denasalization, this</p><p>does not explain why it</p><p>is the former that is deleted altogether and not the</p><p>latter. The m > v change also involves a change of major class features.</p><p>The above reservations are not real objections as they only describe the</p><p>necessary additional mechanisms which allow for a full derivation of soft</p><p>52 Substantive complexity</p><p>mutation within a particular model. The problem, rather, is that the reduc-</p><p>tion of SM rules to three major ones and a host of minor ones does not</p><p>convince us that we have simplified anything or indeed understood more</p><p>than if we reverted to an analysis with five processes. The criterion for a</p><p>successful analysis of the mutations should probably be one of a better</p><p>understanding of the phenomenon rather than one of a more general de-</p><p>scription. As Ball and Müller (1992: 92) themselves rightly note, the com-</p><p>plexity of SM may be only apparent and may simply stem from the compli-</p><p>cated theoretical mechanisms, invoked to explain it.</p><p>The aim of this section is to demonstrate that the key to understanding</p><p>the mutations better lies not in the reduction of the types of rules in the</p><p>derivation but in the internal structure of the targets. Thus, we will first try</p><p>to arrive at some approximation of what the targets of SM are made up of,</p><p>hoping that the internal structure will not only be reflected in SM – we can</p><p>always make assumptions concerning the representation which will yield the</p><p>right results – but will also connect with other areas of Welsh phonology,</p><p>such as consonant inventory, distribution, phonotactics, or indeed other,</p><p>seemingly unrelated phonological phenomena. Once we have arrived at the</p><p>correct representations within the Element Theory, the formulation of rules</p><p>or constraints will hopefully be dramatically simplified. First, however, let</p><p>us look at the remaining types of mutation in Welsh.</p><p>4.3. Aspirate Mutation (AM)</p><p>The term ‘aspirate’ is used here in line with other major works on Welsh</p><p>phonology, and there is no linguistic reason why the more apt term ‘spi-</p><p>rant’ should not be used in this context, other than to avoid terminological</p><p>confusion with SM discussed earlier.</p><p>The process of spirantization affects only three radicals, that is, [p, t, k].</p><p>It has more triggering environments than nasal mutation, discussed below,</p><p>but fewer than soft mutation. All the triggers are of a lexical type, for ex-</p><p>ample, ei ‘her’, gyda ‘along with’, chwe ‘six’, or tua ‘towards’. Some ex-</p><p>amples are given below.</p><p>(37) Aspirate Mutation</p><p>p > f pen [pen] ‘head’ > ei phen [i fen] ‘her head’</p><p>t > T ty  [ti:] ‘house’ > ei thy$ [i Ti:] ‘her house’</p><p>k > x ci [ki:] ‘dog’ > ei chi [i xi:] ‘her dog’</p><p>Initial consonant mutations in Welsh 53</p><p>This process can be given a straightforward description in terms of the</p><p>feature [+continuant], as all the voiceless plosives more or less become the</p><p>corresponding voiceless spirants.78 The question, however, which one</p><p>would like to answer is why only this set of consonants is targeted by AM.</p><p>Does this fact follow from something, or is it totally accidental? In our</p><p>analysis we will try to show that there may be a phonological reason why</p><p>voiceless plosives participate in mutations on a larger scale than any other</p><p>types of consonants. This reason has something to do with the internal</p><p>complexity of segments. Let us now look at the last major mutation type in</p><p>Welsh.</p><p>4.4. Nasal Mutation (NM)</p><p>This process affects all six stops by nasalizing them regardless of whether</p><p>they are voiced or voiceless. This results in a series of three corresponding</p><p>nasal consonants for the voiced stops [b, d, g], which become [m, n, N], and</p><p>in a series of voiceless nasals corresponding in place to the voiceless tar-</p><p>gets [p, t, k], which become [m9, n9, N(].79 NM has the smallest number of</p><p>triggering environments, and it is the only initial mutation type which</p><p>seems to have retained its original phonological context in the triggers.</p><p>That is, with some exceptions, almost all triggers of NM contain a nasal</p><p>themselves.80 Below we use examples adapted from Buczek (1995: 203).</p><p>The nasalization is in some sense reciprocal as the nasal of the preposition</p><p>assimilates to the place of articulation of the following consonant.</p><p>78 We ignore such details as the adjustment of place in the lenition of [p] to [f] from</p><p>labial to labio-dental, parallel to the earlier discussed shifts [b]>[v] or [m]>[v], or</p><p>the fact that [x] in ei chi [i xi:] ‘her dog’ tends to be pronounced as uvular [X].</p><p>These phonetic adjustments have no consequences on the representations and do</p><p>not bear crucially on the analysis.</p><p>79 The nature of the voiceless nasals is debatable. Some realizations are clearly</p><p>voiceless nasals, while others sound like aspirated voiceless nasals.</p><p>80 An interesting case in point is the possessive pronoun fy ‘my’ which causes nasal-</p><p>ization in the following noun, for example, fy mhen m Bangor > yn Mangor [´m maNgor] ‘in Bangor’</p><p>d > n Dyfed > yn Nyfed [´n n´ved] ‘in Dyfed’</p><p>g > N Goginan > yn Ngoginan [´N "Noginan] ‘in Goginan’</p><p>p > m9 Powys > yn Mhowys [´m m9owis] ‘in Powys’</p><p>t > n9 Tresaith > yn Nhresaith [´n n9re"saiT] ‘in Tresaith’</p><p>k > N( Caerdydd > yn Nghaerdydd [´N N(air"di:D] ‘in Caerdydd’</p><p>One should also note that NM once again raises the question of changes in</p><p>major class features. Although intuitively this is a simple process of [–nasal]</p><p>> [+nasal] in the relevant context, the change of [obstruent] to [sonorant] is</p><p>not as negligible an adjustment as that of bilabial to labio-dental place in</p><p>for example b > v in SM. The point is that this would not normally be</p><p>regarded as a minor adjustment, or at least, the criteria for minor and major</p><p>adjustments are not entirely clear.</p><p>4.5. Hard Mutation (HM)</p><p>Finally, we must mention one more type of consonantal change which re-</p><p>sembles the mutations discussed so far, except that it occurs less regularly</p><p>and in a different context. Hard mutation, sometimes referred to as</p><p>‘provection’, represents effects which seem to be the reverse of soft muta-</p><p>tion (Ball and Müller 1992: 286) in that the voiced stops [b, d, g] and, less</p><p>regularly, the voiced fricatives [v, D] are turned into their voiceless conge-</p><p>ners [p, t, k, f, T]. The changes occur in morpheme-final context and are</p><p>triggered mainly by the phonological environment, that is, by the following</p><p>element of a compound or certain derivational endings, for example, verb</p><p>formation suffixes, which may be characterized as beginning with either</p><p>[h] or a voiceless obstruent.</p><p>(39) Hard Mutation</p><p>b > p bwyd+ha > bwyta ["buIta] ‘to eat’</p><p>d > t abad+ty > abaty [a"bati:] ‘abbey’</p><p>g > k gwag+ha+u > gwacau [gwa"kai] ‘to empty’</p><p>In the following sections we will try to understand the various consonantal</p><p>mutations by first proposing a representation for the relevant consonants.</p><p>This will lead to a more precise formulation of the changes in phonological</p><p>terms. It must be remembered that this phonological characterization mere-</p><p>Initial consonant mutations in Welsh 55</p><p>ly aims at understanding the phenomenon of initial mutations in terms of</p><p>the necessary modifications that the segments undergo. Since most of the</p><p>time the changes are divorced from any traceable phonological triggers,</p><p>except for nasal and hard mutations, we must content ourselves with sim-</p><p>ply accounting for the effects.</p><p>4.6. Representing Welsh consonants</p><p>In existing accounts of the mutations in Welsh we note intuitively that the-</p><p>re seems to be a distinction between statements which</p><p>2.2. Representing consonants..................................................................8</p><p>2.2.1. Place ......................................................................................9</p><p>2.2.2. Manner ................................................................................12</p><p>2.2.3. Source..................................................................................15</p><p>2.3. Complexity and syllabification ......................................................18</p><p>3. Substantive complexity effects in Irish ..................................................20</p><p>3.1. Features vs. elements in vocalic alternations.................................20</p><p>3.2. Substantive conditions on Irish epenthesis ....................................27</p><p>3.3. The h-parameter .............................................................................34</p><p>3.4. Segmental inventories and complexity ..........................................36</p><p>4. Initial consonant mutations in Welsh.....................................................45</p><p>4.1. Introduction....................................................................................45</p><p>4.2. Soft Mutation (SM)........................................................................50</p><p>4.3. Aspirate Mutation (AM) ................................................................52</p><p>4.4. Nasal Mutation (NM).....................................................................53</p><p>4.5. Hard Mutation (HM)......................................................................54</p><p>4.6. Representing Welsh consonants ....................................................55</p><p>4.7. A new analysis of Welsh mutations...............................................61</p><p>4.8. The morpho-phonology of mutations.............................................68</p><p>5. Summary and conclusions......................................................................71</p><p>Chapter 2</p><p>Formal complexity</p><p>1. Introduction ............................................................................................75</p><p>2. Syllabification ........................................................................................76</p><p>2.1. Basic facts ......................................................................................76</p><p>Contents x</p><p>2.2. Government....................................................................................79</p><p>2.3. Licensing........................................................................................82</p><p>3. Syllable markedness as a scale of formal complexity............................84</p><p>3.1. Syllable markedness.......................................................................85</p><p>3.2. Problems with parameters..............................................................87</p><p>3.3. Syllabic complexity is scalar .........................................................91</p><p>4. The licensing properties of different nuclear types ...............................96</p><p>4.1. The schwa vowel in Dutch.............................................................96</p><p>4.2. Light and heavy clusters ..............................................................102</p><p>4.3. The word-final context and the scale of licensers........................105</p><p>4.4. The syllabic space ........................................................................107</p><p>4.5. Licensing scale and linguistic variation.......................................110</p><p>4.6. Complexity Scales and Licensing model</p><p>– a first approximation..............................................................114</p><p>5. Sources of empty nuclei and licensing mechanisms in standard GP ...116</p><p>5.1. Introduction..................................................................................116</p><p>5.2. Governing relations and empty nuclei .........................................118</p><p>5.3. Other sources of empty nuclei in phonological representation....122</p><p>5.4. Interonset Government.................................................................124</p><p>5.5. Lured by mgła? ............................................................................125</p><p>5.6. True or False? English and Polish initial clusters........................129</p><p>5.7. Principles and parameters in conflict – towards a solution .........135</p><p>5.8. Licensing of clusters without licensing of empty positions.........139</p><p>5.9. Conclusion ...................................................................................141</p><p>6. Polish as a CV language?.....................................................................143</p><p>6.1. Introduction..................................................................................143</p><p>6.2. Branching onsets in Polish? .........................................................145</p><p>6.2.1. Proper Government across branching onsets ....................146</p><p>6.2.2. Government Licensing ......................................................147</p><p>6.2.3. BrO vs. ONO and verbal prefixation in Polish .................148</p><p>6.2.4. Three types of nuclei in Polish..........................................153</p><p>6.2.5. RIO in word-final context .................................................158</p><p>6.2.6. Substantive restrictions on final RIO ................................160</p><p>6.2.7. Conclusion.........................................................................166</p><p>6.3. Branching rhymes lost .................................................................167</p><p>6.3.1. The ‘missing’ structure ........................................................168</p><p>6.3.2. The distribution of floating melodies in Polish ...................174</p><p>7. CSL – summary and conclusions .........................................................180</p><p>Contents xi</p><p>Chapter 3</p><p>The phonological structure of words</p><p>1. Introduction ..........................................................................................187</p><p>1.1. Theoretical assumptions...............................................................189</p><p>2. Lexicon optimization ...........................................................................192</p><p>2.1. Elimination of liquid diphthongs in Slavic ..................................192</p><p>2.2. Distribution of licensing in the phonological word .....................199</p><p>2.3. Late Common Slavic prosody......................................................201</p><p>2.4. Elimination of rt clusters – a foot-based analysis ........................206</p><p>2.5. Irish metathesis as a stress related phenomenon..........................214</p><p>2.6. Modern Bulgarian ´r / r´ shifts – a case for phonologically</p><p>grounded optimality? ................................................................227</p><p>2.7. A typology of expected liquid metathesis....................................235</p><p>2.8. The non-exceptionality of turt in Slavic ......................................239</p><p>2.9. Jers and clusters in the history of Slavic......................................248</p><p>2.10. The phonological conditions on liquid metathesis</p><p>− conclusions ............................................................................255</p><p>3. Clustering at word edges......................................................................259</p><p>3.1. Introduction..................................................................................259</p><p>3.2. Branching rhymes revisited .........................................................260</p><p>3.3. Branching rhymes CV-lized.........................................................263</p><p>3.3.1. Epenthesis and compensatory lengthening........................263</p><p>3.3.2. The double licensing of LIO?............................................267</p><p>3.3.3. More on Dutch schwa .......................................................273</p><p>3.4. The magic of the left edge............................................................277</p><p>3.5. ‘Magic’, left, right and centre ......................................................280</p><p>3.6. Conclusions..................................................................................285</p><p>4. Chapter summary .................................................................................286</p><p>achieve the status of</p><p>true generalizations and those which are mere descriptions. The former</p><p>kind includes statements which manipulate the features [continuant], [voi-</p><p>ce] and [nasal], while the latter type supply redundant properties, such as</p><p>place adjustment, or even a change of major class, as in the shifts m > v, or</p><p>b > m. Thus, there seems to be a set of primes which are pertinent to a par-</p><p>ticular phenomenon – they tend to coincide with the features surviving in</p><p>the underlying representations in underspecification frameworks – and</p><p>properties which consistently turn out to be redundant. The latter form an</p><p>arbitrary set, and the battery of redundancy rules which is needed to deal</p><p>with them provides no insight into the phenomenon in question.</p><p>Let us briefly remind ourselves of the basic tenets of the melodic repre-</p><p>sentation in Government Phonology, and apply the model to the Welsh</p><p>system. The Element Theory in Government Phonology attempts to rid</p><p>representations of redundancy altogether. It replaces features with primes</p><p>that enjoy stand-alone interpretability. Each such prime individually or in</p><p>combination with others is directly mappable onto articulation and auditory</p><p>perception. Phonological representation in this model is therefore simulta-</p><p>neously redundancy-free and fully interpretable, that is, ready for phonetic</p><p>implementation with no further specification required (Harris 1996).</p><p>The advantage of this model with respect to the problem of redundancy</p><p>lies in the fact that phonological statements, be they rules or output con-</p><p>straints, refer only to lexically pertinent primes and there is no need to</p><p>distinguish between true generalizations and other, less important state-</p><p>ments. All statements are, in a sense, true generalizations. This model also</p><p>restricts the nature of rules or constraints. For example, they may refer to</p><p>combinability of primes (Charette and Göksel 1998, Cobb 1997, Kaye 2001),</p><p>or their ability to occur in particular prosodic positions (Harris 1997).</p><p>Returning to the representation of Welsh consonants, we will tentatively</p><p>assume the following representations of place. The labels are not meant to</p><p>be precise phonetic definitions of place but rather act as rough functional</p><p>56 Substantive complexity</p><p>distinctions. Since place of articulation does not play any crucial role in</p><p>mutations the proposed representations will not be argued for. 81</p><p>(40)</p><p>U labials b, p, m, v, w</p><p>A-I dentals t, d, T, D, n</p><p>A alveolars r, l, r 9, Ò</p><p>I palatals S</p><p>_ velars k, g, N, x</p><p>The manner dimension is more complex and a few comments concerning</p><p>this aspect of the representation of consonants are needed before we con-</p><p>sider the Welsh consonants. Let us again consider the list of manner ele-</p><p>ments below, based on the work of Harris (1990), Harris and Lindsey (1995).</p><p>(41)</p><p>/ occlusion b, p, t, d, k, g, l?, Ò?, nasals?</p><p>N nasality nasals</p><p>H high tone aspirated voiceless obstruents</p><p>L low tone fully voiced obstruents (not in Welsh)</p><p>h noise fricatives, affricates, released stops</p><p>The presence of the occlusion element in plosives is rather uncontroversial.</p><p>On the other hand, it is not clear why it should be part of the representation</p><p>of Welsh laterals and nasals, hence the question mark. It is true, that the</p><p>acoustic definition of (/) is ‘a drop in overall amplitude of the signal’,</p><p>which may mean complete silence as in stops, or attenuation of energy, as</p><p>in nasals and laterals. Thus, in theory, there is no reason why occlusion</p><p>should not be allowed in these consonants. However, there is also no rea-</p><p>son why it must be present. Other sonorants, for example, glides, also ex-</p><p>hibit attenuation in the signal as compared to vowels, but it would be inap-</p><p>propriate to claim that this is the result of the presence of occlusion. In this</p><p>81 The compound A-I is a shorthand notation which avoids resolving the question as</p><p>to which of the two elements is the head and which one is the operator. It must be</p><p>added that the representation of dentality in Welsh does not constitute a universal</p><p>assumption concerning this place, although the same compound is used to define</p><p>dentality in Irish and Polish in Cyran (1997: 222).</p><p>Initial consonant mutations in Welsh 57</p><p>work we will retain this element in laterals. As for nasals, the situation is</p><p>not straightforward.82</p><p>It is an inherent property of the Element Theory that to obtain a nasal</p><p>consonant it should be sufficient to have a nasal element and some place</p><p>specification, underlying or acquired through assimilation. The role of</p><p>occlusion in this class of consonants requires further research, especially in</p><p>the light of common assimilations to stops. We will assume that it is absent</p><p>in nasals unless there are reasons for its presence, in which case we will</p><p>still get a nasal consonant, albeit a more complex one. In other words, ei-</p><p>ther (U,N) or (U,N,/) will yield [m], and it is phonology that will tell us</p><p>which construct we are dealing with. This ambiguity, which is inbuilt in</p><p>the Element Theory, has its advantages as we will shortly discover.83</p><p>Returning to nasality for a moment, the general introduction to Celtic</p><p>mutations at the beginning of this section showed that there seems to be</p><p>some affinity between nasalization and voicing. For example, the historical</p><p>process of nasalization in Old Irish diverged into two phenomena, that is,</p><p>nasalization and voicing. There has been a lot of research into the nature of</p><p>post-nasal voicing in phonological theory (e.g. Herbert 1986, Itô and Mester</p><p>1986, Pater 1995). There are also proposals within the Element Theory</p><p>which merge the elements (N) and (L) into one, for example, Nasukawa</p><p>(1998, 2005), and Ploch (1999). In this work we would like to at least bear</p><p>in mind the possibility that voicing and nasality might be two sides of the</p><p>same coin. We will represent this option by supplying both elements in</p><p>relevant contexts, that is (N/L).</p><p>The use of the tone elements in the definition of voicing has been de-</p><p>scribed in detail at the beginning of this chapter. Let us only repeat that in</p><p>systems which have a two-way laryngeal distinction only one of the ele-</p><p>ments is employed – in the marked series, while the neutral series remains</p><p>undefined. In Welsh, just as in Irish and English, the marked series is the</p><p>voiceless or aspirated one, containing the high tone element (H). In the</p><p>previous section we saw how this asymmetry plays out in the phonotactics</p><p>of modern Irish. Similar effects can be traced in Welsh. For example, clus-</p><p>ters of the type rg, lg are absent word-finally and morpheme internally, and</p><p>82 There are facts from other languages which point to the presence of stopness in</p><p>laterals and nasals. See the analysis of Icelandic pre-aspiration in Gussmann (2000).</p><p>83 Ambiguous representations yielding identical phonetic effects but at the same</p><p>time displaying disparate phonological behaviour have been called ‘double agents’</p><p>in the Government Phonology literature (Gussmann 2001, 2002).</p><p>58 Substantive complexity</p><p>the best nasal+stop clusters word finally are those with the voiceless stop</p><p>and only one voiced, that is, nd.84</p><p>However, the most significant innovation that we would like to propose</p><p>with respect to Welsh concerns the noise element (h), or rather its absence</p><p>from the Welsh consonantal system. We put forward this proposal on the</p><p>basis of our findings in Irish, discussed in the previous sections. The Welsh</p><p>system shows all the relevant diagnostic aspects for such a move. First of</p><p>all, as mentioned above, lenis obstruents in cluster formation are generally</p><p>weaker. Secondly, systemically acceptable affricates are missing from the</p><p>system of consonants. And finally, there are no real voice distinctions a-</p><p>mong the fricatives. The voiceless fricatives, specified as possessing (H),</p><p>contrast directly with sonorants or sonorant-like</p><p>fricatives. These diagnos-</p><p>tic criteria seem to point heavily to the fact that Welsh, like Irish, is an h−less</p><p>language. The tables below compare two approaches to Welsh consonants</p><p>and their consequences for the analysis of mutations. One is Buczek (1995)</p><p>in which the noise element (h) is very much part of the system and has a</p><p>very important function to play, and the other is that advocated in this</p><p>work, in which (h) is absent altogether from the Welsh system. First we</p><p>look at Buczek’s proposal.85</p><p>(42) Welsh consonants (adapted from Buczek 1995)</p><p>[p] [b] [t] [d] [k] [g] [l] [r] [m] [n] [N]</p><p>U</p><p>/</p><p>h</p><p>H</p><p>U</p><p>/</p><p>h</p><p>A-I</p><p>/</p><p>h</p><p>H</p><p>A-I</p><p>/</p><p>h</p><p>_</p><p>/</p><p>h</p><p>H</p><p>_</p><p>/</p><p>h</p><p>A</p><p>/</p><p>A</p><p>U</p><p>N</p><p>A-I</p><p>/</p><p>N</p><p>_</p><p>/</p><p>N</p><p>[f] [v] [T] [D] [x] [Ò] [r 9] [m9] [n9] [N(]</p><p>U</p><p>h</p><p>H</p><p>U</p><p>A-I</p><p>h</p><p>H</p><p>A-I</p><p>h</p><p>_</p><p>h</p><p>H</p><p>A</p><p>/</p><p>h</p><p>A</p><p>h</p><p>U</p><p>/</p><p>N</p><p>h</p><p>A-I</p><p>/</p><p>N</p><p>h</p><p>_</p><p>/</p><p>N</p><p>h</p><p>84 For a recent discussion of similar restrictions on word-final nasal+voiced stop</p><p>clusters in the dialects of modern English see Gussmann (1998), and Tóth (2002).</p><p>85 In Buczek (1995), coronality is represented by the element (R) which has since</p><p>been eliminated from GP. We replace this element with (A) and (A-I).</p><p>Initial consonant mutations in Welsh 59</p><p>Since there is no crucial difference in the use of place elements and the</p><p>tone specification of voiceless obstruents between the above table and our</p><p>proposal below, these aspects will be ignored in the discussion.</p><p>The noise element (h) is used in all stops in Buczek (1995), in accor-</p><p>dance with the assumption of the early Element Theory that this element is</p><p>present in released stops (e.g. Harris 1990). By definition, this element is</p><p>also present in fricatives, but not in all of them. It is missing in the voiced</p><p>labio-dental fricative, which is represented only by means of (U). This</p><p>move is dictated by the fact that in soft mutation [m] becomes [v]. A de-</p><p>scription of this process would have to involve not only the loss of nasality,</p><p>but also the addition of noise to obtain a fricative, a rather complicated</p><p>procedure. It is also interesting to note that (/) is missing in [m] as opposed</p><p>to other nasals, clearly to avoid delinking two elements at a time in the</p><p>shift m > v. However, given the representations of [v] and [b], Buczek</p><p>cannot avoid delinking two elements in the SM shift b > v (U,/,h) > (U),</p><p>unless we assume that there are two types of [v] – one with noise and an-</p><p>other without it. These small inconsistencies aside, Buczek also makes an</p><p>important claim that the noise element is used to represent voicelessness</p><p>and aspiration in [Ò, r 9, m9, n9, N(].</p><p>The table below illustrates the option in which the relevant Welsh con-</p><p>sonants are devoid of the noise element.</p><p>(43) An h-less representation of Welsh consonants</p><p>[p] [b] [t] [d] [k] [g] [Ò] [r 9] [m] [n] [N]</p><p>U</p><p>/</p><p>H</p><p>U</p><p>/</p><p>A-I</p><p>/</p><p>H</p><p>A-I</p><p>/</p><p>_</p><p>/</p><p>H</p><p>_</p><p>/</p><p>A</p><p>/</p><p>H</p><p>A</p><p>H</p><p>U</p><p>N/L</p><p>A-I</p><p>N/L</p><p>_</p><p>N/L</p><p>[f] [v] [T] [D] [x] [l] [r] [m9] [n9] [N(]</p><p>U</p><p>H</p><p>U</p><p>A-I</p><p>H</p><p>A-I</p><p>_</p><p>H</p><p>A</p><p>/</p><p>A</p><p>U</p><p>N/L</p><p>H</p><p>A-I</p><p>N/L</p><p>H</p><p>_</p><p>N/L</p><p>H</p><p>Note that the stops are still stops because they possess the occlusion ele-</p><p>ment (/), which together with place defining elements and the laryngeal</p><p>distinction: high tone vs. nothing, fully suffices to define the system. What</p><p>is interesting is that Welsh stops are generally less complex than in other</p><p>languages, for example Polish, and that the voiced series is still weaker,</p><p>60 Substantive complexity</p><p>with the voiced velar plosive having only one element, the occlusion itself.</p><p>The same situation was earlier shown to hold also in Irish.86</p><p>It will be recalled that the phonetic realization of (/) as [g] cannot be</p><p>treated as universal. In Welsh, just as in Irish, it follows from two condi-</p><p>tions; the systemic absence of (h), and the fact that obstruents use (H) in</p><p>their laryngeal specification. This means that in another h–less system in</p><p>which (L) is used in obstruents, the sole presence of (/) may produce [k]</p><p>because [g] in that system would have the low tone. The phonetic imple-</p><p>mentation of a given element in Government Phonology depends not only</p><p>on its overall acoustic signature but also its place in a particular phonologi-</p><p>cal system. This is one of the reasons why it is erroneous to try and provide</p><p>element based representations for all the sounds represented in, for exam-</p><p>ple, the IPA chart. The representations should always be system based.</p><p>To return to the discussion of the above table, the simplex representation</p><p>of [v] is now a systemic effect rather than an ad hoc solution. Note that we</p><p>bypass the problem of how many elements will be lost in the shift m > v, or</p><p>b > v. In each case, only one element needs to be lost: either nasality or</p><p>occlusion respectively.</p><p>The absence of (h) in the system requires that a different proposal be</p><p>made concerning fricatives, sounds in which this element was traditionally</p><p>indispensable. We assume, following Cyran (1996b, 1997) and Ritter (1997),</p><p>that the headedness of the resonance element may be responsible for the</p><p>effect of friction.87 This is true of Welsh [v] = (U), and [D] = (A-I). Such</p><p>fricatives are sonorant-like in terms of their behaviour.88 It is interesting to</p><p>note that the two fricatives tend to be lost in word-final context in Welsh</p><p>(Thomas 1992). For example, gof ‘blacksmith’ is pronounced as [go:] in</p><p>the North and [go:v] in the South. Similar effects are found in words like</p><p>barf [bar(v)] ‘beard’ and gardd [gar(D)] ‘garden’. The absence of the noise</p><p>86 Although the representation of [g] in Irish and Welsh is identical, the effects of</p><p>lenition of that object are different. In Irish, [g] becomes [V], while in Welsh its [g]</p><p>is simply deleted. Historically, Welsh was like Irish. The difference may lie in the</p><p>way the two modern languages interpret the empty onset that remains after the</p><p>deletion of the stopness element.</p><p>87 Recall that in the case of the velar fricative [x] friction results from (H), as there</p><p>is no place element to be headed.</p><p>88 Recall that [v] is notorious for acting as a ‘double agent’ not only in Welsh but</p><p>also in Polish (Gussmann 1981), Irish (Cyran 1997), Russian (Andersen 1969),</p><p>Hungarian (Szigetvári 1998), Slovak (Rubach 1993) and in the history of a number</p><p>of Slavic languages (Cyran and Nilsson 1998).</p><p>Initial consonant mutations in Welsh 61</p><p>element (h) also enforces a different representation for voiceless liquids</p><p>and nasals, that is, one in which the high tone (H) is responsible for the</p><p>voiceless effect. In this way, this group of consonants can be more easily</p><p>related to the fortis plosives and fricatives.</p><p>Let us now look at how the two different approaches to the representa-</p><p>tion of Welsh consonants affect the analysis of the consonant mutations.</p><p>4.7. A new analysis of Welsh mutations</p><p>The table below is a comparison of the two different approaches to repre-</p><p>sentations within the Element Theory, which were described above. One of</p><p>them attempts to ascribe particular representations to Welsh consonants</p><p>rather automatically introducing some modifications only contingently.89</p><p>This is compared with the ‘h-less’ approach, which attempts to understand</p><p>the system in terms of general parameters on element availability.</p><p>The Element Theory allows for only two types of processes: decompo-</p><p>sition, that is, the deduction of elements, and composition, which is the</p><p>addition of elements.90 These correspond to lenition and fortition phenom-</p><p>ena and are represented in the table as (–) and (+) respectively. The dotted</p><p>line separates those classes of sounds which require a different analysis in</p><p>terms of element deduction or addition. Ideally, the groups of consonants</p><p>which are intuitively perceived as natural classes should receive identical</p><p>interpretation.</p><p>One immediate</p><p>observation that can be made on the basis of the table</p><p>below is that there are certain aspects which the two analyses share. Some</p><p>of the effects in Soft Mutation (SM), the entire Aspirate Mutation (AM)</p><p>and Hard Mutation (HM) receive an identical interpretation, but this is</p><p>where the similarities between the two approaches end.</p><p>The first set of targets in SM, that is [p,t,k] are uniformly turned into</p><p>voiced stops by means of suppressing the high tone element. Since the high</p><p>tone may to some extent be equated with the feature [–voice] in deriva-</p><p>tional frameworks, this fragment of the analysis is not only identical to</p><p>Buczek (1995) but also to Awbery (1973) and Ball and Müller (1992).</p><p>89 Recall the discussion of the representations in (42).</p><p>90 One of the conditions on composition in GP is that the element that is added must</p><p>have a local source, that is, be present in the representation of the trigger, if this</p><p>process is to be viewed as phonological.</p><p>62 Substantive complexity</p><p>(44) A comparison between a h-full and h-less analyses</p><p>Soft M. Buczek 1995 h-less analysis</p><p>p > b</p><p>t > d</p><p>k > g</p><p>b > v</p><p>d > D</p><p>g > P</p><p>Ò > l</p><p>r9 > r</p><p>m > v</p><p>–H</p><p>–/, –h?</p><p>–/</p><p>–/, –h</p><p>–h</p><p>–N</p><p>–H</p><p>–/</p><p>–H</p><p>–N/L</p><p>Aspirate m.</p><p>p > f</p><p>t > T</p><p>k > X</p><p>–/</p><p>–/</p><p>Nasal. M.</p><p>p > m9h</p><p>t > n9h</p><p>k > N(h</p><p>b > m</p><p>d > n</p><p>g > N</p><p>+N, –H</p><p>+N, –h</p><p>+N/L</p><p>Hard M.</p><p>b > p</p><p>d > t</p><p>g > k</p><p>v > f</p><p>D > T</p><p>+H</p><p>+H</p><p>As for the voiced targets, that is [b,d,g], the h–less analysis needs only to</p><p>deduct the occlusion element. Recall that the generative accounts of SM</p><p>also strove to see the shift [b,d,g] > [v,D,P] as one process. However, the</p><p>unification resulted in a special treatment of the shift [g] > [P] which in-</p><p>volved a postulation of abstract intermediate stages such as g>*V>P. This</p><p>representation-based analysis bypasses this problem and shows that there is</p><p>no need for a derivational approach to the seemingly disparate behaviour of</p><p>[g]. Once we arrive at the correct representations, the formulation of the</p><p>process is quite simple. Since Welsh [g] is represented only by one element</p><p>Initial consonant mutations in Welsh 63</p><p>(/), its deletion results in nothing.91 Note that in Buczek (1995) the situa-</p><p>tion gets a little complicated here. The shift [b] > [v] must involve either</p><p>deletion of two elements (U,/,h) > (U), or alternatively, we must assume</p><p>that there are two representations of [v] in Welsh, (U) and (U,h). Similarly,</p><p>two elements must be lost in the deletion of [g]. While, theoretically speak-</p><p>ing, there is perhaps nothing wrong with losing two elements in some</p><p>processes, it is not clear how we decide which segments will lose two and</p><p>which ones only one element in Buczek’s approach. In the h-less analysis</p><p>this dilemma is non-existent: it is always only one element which is lost.</p><p>The interpretation of the voicing of the voiceless liquids naturally de-</p><p>pends on what prime is made responsible for the effect of voicelessness. In</p><p>this respect, the two analyses appear to be equally appealing in that only</p><p>one prime is lost. The advantage of one over the other may only be as-</p><p>sessed on the basis of the overall system. One argument which we can</p><p>wheel out in favour of the h–less analysis is that the voicing of liquids may</p><p>indeed be grouped with the voicing of fortis plosives, an observation which</p><p>was also made in generative accounts. Here we are dealing with the uni-</p><p>form deletion of the high tone element.92 No such unification is possible</p><p>under Buczek (1995) in which the prime which is lost in liquids is (h),</p><p>while the plosives lose the tone element (H).</p><p>The final shift in SM, that is (m > v), clearly involves deletion of nasal-</p><p>ity. It must be recalled however, that in Buczek (1995) this elegant inter-</p><p>pretation is achieved at the cost of two assumptions. Firstly, [m] is the only</p><p>nasal sound with no occlusion element, and secondly, [v] is the only frica-</p><p>tive with no noise element. Both facts may however fall out directly from</p><p>91 An empty position in Government phonology, be it onset or nucleus, may also</p><p>have a phonetic interpretation (Kaye, Lowenstamm and Vergnaud 1990, Charette</p><p>1991). We may tentatively assume that what is referred to as the historical reflex</p><p>*[V] may have been the interpretation of the empty intervocalic onset which re-</p><p>sulted from the loss of occlusion, just as it is in modern Irish. Likewise, word-initial</p><p>empty onsets in vowel initial words may be realized as glottal stops. Such effects</p><p>need not be treated as the realization of separate elements, but as mere language-</p><p>specific realizations of empty positions which for one reason or another cannot</p><p>remain totally empty, for example, due to some language specific constraints on</p><p>such structures.</p><p>92 Note that we also bypass the problem of specifying liquids as [+voice] as in the</p><p>generative formulations. The marked case is represented by the voiceless liquids,</p><p>which have an additional element, and the voicing of the liquids has no particular</p><p>exponent in the form of a feature or element. They simply become modal liquids,</p><p>which are voiced.</p><p>64 Substantive complexity</p><p>more general assumptions concerning the way elements are harnessed in the</p><p>definition of sound systems. It should be mentioned that in this analysis the</p><p>problem of major class features does not arise. The loss of (N/L) from (N/L,</p><p>U) leaves us with only the labiality element, which has the interpretation of a</p><p>voiced labio-dental fricative in Welsh. In this respect our account seems to be</p><p>more adequate than the generative accounts also, which have to adjust major</p><p>class features as well as the place of articulation in the m > v shift.</p><p>To conclude the discussion of soft mutation let us attempt to formulate</p><p>the process in terms of what transpires as a result of the shifts discussed</p><p>above. First of all, in the Element Theory we may provide a formulation of</p><p>SM which will only involve the manipulation of one prime at a time. Given</p><p>the nature of this model, no further adjustments are necessary – the dele-</p><p>tion of an element creates a structure which is also independently interpret-</p><p>able, unless what remains is nothing, when we do get phonetic nothing, as</p><p>in g > P. Secondly, we lose none of the generalizations made in generative</p><p>studies which tried to reduce SM to three processes. It seems, in fact, that</p><p>we can achieve the same generalizations in a more compact fashion be-</p><p>cause we avoid a lot of the additional adjustments that those frameworks</p><p>required. What is more, given the possibility of conflation of nasality (N)</p><p>and Low tone (L) elements, we are able to reduce SM to only two proc-</p><p>esses, that is, tone deletion, and if the tone is missing in the target conso-</p><p>nant – as in the neutral series [b, d, g] – then occlusion is deleted. Note that</p><p>given the representations of Welsh consonants, there is nothing else left to be</p><p>deleted except for the place defining elements. But these do not seem to be</p><p>targets of lenition in Welsh. Thus, our representations also restrict possible</p><p>types of processes, in that what we observe is everything that could occur.93</p><p>It goes without saying that our analysis simplifies the most complex of</p><p>the initial mutations in a considerable way. The overall principle of SM</p><p>may be formulated as first targeting tone elements, and if they are missing,</p><p>then other manner elements. Place elements seem to remain unscathed.</p><p>(45) A scale of mutation targets in Welsh</p><p>Laryngeal > Manner > Place</p><p>H, L/N L/N, /</p><p>93 Note that in an ‘h-full’ analysis this additional element creates an additional op-</p><p>eration that could occur. It is difficult to imagine what deletion of (h) could bring</p><p>about in Welsh.</p><p>Initial consonant mutations in Welsh 65</p><p>The above scheme does not resolve the issue of conflation of nasality</p><p>with</p><p>the low tone (L/N), and how this new construct should be categorized. If it</p><p>is part of the laryngeal node then the mutations may be more readily de-</p><p>scribed as predominantly tonal phenomena, an option to which we will</p><p>return below.94</p><p>As mentioned earlier, SM is the most common type of mutation and one</p><p>with the greatest number of targets and triggers.95 It seems to be the domi-</p><p>nant type of mutation, especially when compared with aspirate mutation. In</p><p>our earlier discussion of the latter we asked the question why AM affects</p><p>only three targets, while the formulation of AM as delinking of occlusion</p><p>allows us to cover all stops. We can understand this problem better under</p><p>two conditions: if the dominant character of soft mutation is recognized,</p><p>and if the formulation of SM given above is correct. Soft mutation, in a</p><p>sense, uses up some of the options that aspirate mutation could have utilized.</p><p>It is not clear how the dominance should be formalized. It is not impos-</p><p>sible that it may be best understood in terms of historical precedence, in</p><p>which case no theory of the dominance of some types of mutations over</p><p>others needs to be postulated. On the other hand, some formalization of the</p><p>precedence of laryngeal elements as opposed to manner elements as targets</p><p>of SM seems in place, if only to answer the question why the voiceless</p><p>stops in SM are voiced, rather than spirantized on a par with the other plo-</p><p>sives.96</p><p>Whether there is a universal hierarchy determining the propensity of</p><p>particular elements to undergo deletion, and how this should be expressed</p><p>in the Element Theory, is an issue for further research. At this stage we just</p><p>state the fact that in Welsh laryngeal elements tend to be lost more readily</p><p>than manner elements. Of course, the voiced stops in SM lose occlusion</p><p>because there is no laryngeal element that could be lost. As for aspirate</p><p>mutation, we must note that the voiceless plosives in Welsh are the only</p><p>obstruents that have one remaining ‘deletable’ prime that has not been</p><p>affected by SM. Hence, the occlusion in voiceless plosives is lost in AM.</p><p>The voiced series of stops cannot be included in this type of mutation be-</p><p>94 The use of the term ‘laryngeal node’ may suggest that we are dealing with some</p><p>melodic geometry within the Element Theory (Harris 1994: 127).</p><p>95 Ball and Müller (1992) refer to it as the ‘default mutation’.</p><p>96 The problem discussed here hinges on what Lass (1984) distinguishes as ‘sonori-</p><p>zation’ versus ‘opening’ in lenition processes. Apparently, the former is less dra-</p><p>matic a change than the latter. However, Lass does not provide a rigorous linguistic</p><p>argument in support of this impressionistic view.</p><p>66 Substantive complexity</p><p>cause this would lead to grammatical ambiguity – they have already lost</p><p>their only deletable prime in SM. Thus, the special status of voiceless stops</p><p>in Welsh is due to the fact that they are the most complex consonants (43).</p><p>Recall that voiceless stops are treated as the strongest and are placed at</p><p>the beginning of the lenition trajectories discussed in Lass (1984: 178).</p><p>The Element Theory, in which voiceless stops are the most complex only if</p><p>specified by (H), also predicts that in systems in which it is the voiced se-</p><p>ries that is specified – with (L) – the voiceless stops are not the strongest.</p><p>This allows the Element Theory to avoid the pitfalls of the universal mark-</p><p>ing of certain segment types as possessing particular characteristics (Foley</p><p>1977), and of treating such processes as devoicing in German or Polish as</p><p>fortition (Lass 1984: 179), rather than the more intuitively correct lenition,</p><p>or element deduction in prosodically weak positions as the Element Theory</p><p>has it.</p><p>Let us return now to the remaining two mutation types and their analy-</p><p>sis within the Element Theory. Nasal Mutation is the only initial consonant</p><p>mutation which consists in element addition. Recall that most of the trig-</p><p>gers contain the nasal property.97 Here, the h-less analysis is simple again.</p><p>The stops, whether voiceless or voiced, are affected by nasality. As a re-</p><p>sult, nasal or nasalized consonants appear as predicted by the Element The-</p><p>ory. Given the specification of the targets, where the voiceless series con-</p><p>tains (H), the effects of NM seem to be straightforward, though not entirely</p><p>unusual.98 Nevertheless, assuming that nasality may combine with (H) in</p><p>some way, we predict that the aspirated / voiceless stops will yield voice-</p><p>less / aspirated nasals, and voiced stops will simply produce nasals with the</p><p>same place of articulation.</p><p>Admittedly, the nasals which result from NM are representationally dif-</p><p>ferent from the nasals provided in table (43). The stops contain the occlu-</p><p>sion element and therefore, this time, we must assume its presence in the</p><p>nasalized forms. As mentioned earlier, the presence or absence of (/) in</p><p>nasal consonants changes nothing in terms of phonetic realization. Both</p><p>(N/L,U) and (N/L,U,/) will be realized as [m]. However, while in the case</p><p>of the basic [m] we have evidence of the absence of occlusion, which is</p><p>97 On this basis it could be claimed that perhaps Nasal Mutation is still a phono-</p><p>logical phenomenon.</p><p>98 Voiceless or aspirated nasals are certainly a marked phenomenon, just as the</p><p>voiceless liquids discussed earlier. Some dialects of Welsh, notably the South Gla-</p><p>morgan dialect (Thomas 1984) eschew such structures and of all the voiceless so-</p><p>norants of Standard Welsh retain only [Ò].</p><p>Initial consonant mutations in Welsh 67</p><p>reflected in the effects of SM (m > v), in NM shifts b > m and p > m9 there</p><p>is positive evidence of its presence, and hence we must accept this presence</p><p>as a fact.</p><p>For Buczek (1995), nasal mutation is unduly complex, which is a direct</p><p>consequence of the assumption that the noise element is responsible for the</p><p>voicelessness of sonorants. Thus, to get the voiceless nasals, she needs to</p><p>add nasality and delete the high tone, while for the voiced stops to become</p><p>nasals, she must add nasality and delete the noise which is present in all</p><p>stops and which would otherwise produce the voicelessness effect. Addi-</p><p>tionally, in Buczek’s analysis a process which clearly consists in element</p><p>addition must be also assumed to delete elements, where the choice of</p><p>which element is to be deleted is quite arbitrary.</p><p>Finally, hard mutation, which is not an initial consonant mutation but</p><p>rather a morpheme-final one, is also a composition process. It involves addi-</p><p>tion of the high tone element to the toneless series of stops and, less com-</p><p>monly, the toneless voiced fricatives. Here too, the phonological context is</p><p>still retained and there are good reasons to assume that it may still be a</p><p>phonological phenomenon.</p><p>To conclude the analysis of Welsh mutations we note that the h-less ap-</p><p>proach clearly shows that all the consonant mutations in Welsh can be</p><p>fairly well understood as predominantly tonal phenomena. The picture is of</p><p>course more convincing if nasality can indeed be conflated with the low</p><p>tone into some sort of Janus-faced laryngeal element. Each process targets</p><p>only one prime, which allows for a neat formulation of the mutations under</p><p>any framework which decides to use element-based representations, be it a</p><p>rule- or constraint-based model. This analysis works under certain assump-</p><p>tions concerning the internal structure of Welsh consonants, for example, the</p><p>fact that the noise element is totally missing, and that radical nasal conso-</p><p>nants do not contain stopness. These assumptions find some support within</p><p>the Welsh system itself, as well as in other Celtic languages such as Irish.</p><p>In Welsh, we witness a convergence of effects in the mutations with a</p><p>number of diagnostic criteria for the absence of (h). These include the ab-</p><p>sence of affricates, and no voice distinctions among the fricatives, whereby</p><p>the existing</p><p>voiced fricatives behave as sonorant-like and the voiceless</p><p>fricatives contrast directly with liquids and glides by means of the (H) ele-</p><p>ment. Just as in Irish, the weakest stop is the voiced velar one, and like in</p><p>Irish, it is deleted in the relevant contexts. In Welsh, this happens in SM,</p><p>while in Irish, [g] is deleted in the verbal system when intervocalic, for</p><p>68 Substantive complexity</p><p>example, suigh ‘sit’ [sig´] loses the stop in suím ‘I sit’ [si:m´], when the first</p><p>person singular ending –im is added.99</p><p>It must also be emphasized that this analysis throws new light on the</p><p>apparent complexity of initial mutations in Welsh from the point of view of</p><p>learnability and the interaction between morphology and phonology. As for</p><p>the former issue, given that mutations are predominantly tonal phenomena,</p><p>the difficulty they present for learners is roughly of the same kind as voice</p><p>neutralization is in Polish or German. On the other hand, more needs to be</p><p>said about the morpho-phonological nature of mutations. We turn to a brief</p><p>discussion of this problem below.</p><p>4.8. The morpho-phonology of mutations</p><p>At the outset of our discussion of the initial consonant mutations in Welsh,</p><p>we deemed these effects to be morpho-phonological due to the fact that</p><p>they occur in morpho-syntactic rather than phonologically defined con-</p><p>texts. This is no place to deal exhaustively with the concept of morpho-</p><p>phonology itself.100 Our aim in this section is merely to set out some crite-</p><p>ria for deciding whether we are dealing with pure phonological or morpho-</p><p>phonological regularities. We will not attempt to define the nature of the</p><p>interaction between phonology and morphology, though the element-based</p><p>analysis opens some options which will be mentioned below.</p><p>The historical perspective on the initial mutations in Celtic languages</p><p>and the analysis of the modern Welsh phenomena point to the fact that</p><p>phonological and morpho-phonological regularities may look very similar.</p><p>The reason for this state of affairs is simple and follows from the fact that,</p><p>at least in the case of the mutations, we are dealing with grammaticaliza-</p><p>tion of phonological regularities. In this respect, the morpho-phonological</p><p>effects may be easily confused with phonology proper, because they reflect</p><p>former rather than current phonological patterns, and, as we saw above,</p><p>they may look very regular. We will assume that morpho-phonology is a</p><p>term which covers cases of petrification of phonological regularities when</p><p>the phonological system itself develops in such a way that the pattern can</p><p>no longer be phonological. Thus, morpho-phonology is a case of systemic</p><p>99 Interestingly, also the fricative [v] is lost in the two languages, for example,</p><p>Welsh gof [go:] ‘blacksmith’, and Irish nimh / nimhe [n´iv´] / [n´i:] ‘poison/gs.’.</p><p>100 For some discussion of the place of morphophonology in linguistic theory see,</p><p>for example, Dressler (1977, 1985), Gussmann (1985), Kowalik (1997), Laskowski</p><p>(1975), Trubetzkoy (1931).</p><p>Initial consonant mutations in Welsh 69</p><p>conservatism in that it petrifies former states of the phonological system.101</p><p>How exactly is the relationship between the alternating consonants captured</p><p>is another story.</p><p>At this stage we must mention some criteria on the basis of which some-</p><p>thing is deemed phonological or is relegated to morphophonology. The</p><p>Element Theory provides us with some useful tools in this respect, for ex-</p><p>ample, by restricting phonological operations at the melodic level to com-</p><p>position and decomposition. As for composition, the model requires that</p><p>the added property be locally present. This, to some extent, obtains for the</p><p>Welsh nasal and hard mutations. However, there is a purely grammatical</p><p>reason why mutations in Celtic languages should not be viewed as auto-</p><p>matic phonological effects. The mutations have taken on the grammatical</p><p>functions of distinguishing gender, case or even tense, and are restricted to</p><p>particular morpho-syntactic contexts. In other words, not every context</p><p>where nasality is present will produce nasal assimilation. This last point</p><p>significantly undermines the phonological status of the nasal mutation.</p><p>As for decomposition, that is, element loss, we may speak of the re-</p><p>quirement of a phonologically determinable context triggering the loss. For</p><p>example, we saw earlier that [g] and [v] are deleted intervocalically in</p><p>modern Irish. However, the contexts in which the initial mutations in both</p><p>Irish and Welsh originally arose are no longer present due to the historical</p><p>loss of final syllables. It will be recalled that in present-day Welsh and</p><p>Irish, the identical phonetic context triggers disparate types of mutations.</p><p>The facts are repeated below for convenience.</p><p>(46) Irish cat [kAt] ‘cat’</p><p>a chat [´ xAt] ‘his cat’</p><p>a cat [´ kAt] ‘her cat’</p><p>a gat [´ gAt] ‘their cat’</p><p>Welsh cath [ka:T] ‘cat’</p><p>ei gath [i ga: T] ‘his cat’</p><p>ei chath [i xa: T] ‘her cat’</p><p>eu cath [i ka: T] ‘their cat’</p><p>Thus, to determine whether a regularity is phonological or morphopho-</p><p>nological, we may use phonological criteria such as the requirement that</p><p>101 See for example Árnason (1985: 22) for similar views.</p><p>70 Substantive complexity</p><p>added primes be locally present and that this phenomenon is obligatory</p><p>every time the appropriate context is created, or the requirement that the</p><p>suppression of particular primes occurs in specific phonologically defin-</p><p>able contexts. On the basis of such criteria we may quite definitely claim</p><p>that, in Welsh, soft and aspirate mutations belong to morpho-phonology,</p><p>while hard mutation can be viewed as phonological according to the crite-</p><p>ria we have laid down. In fact, this phenomenon resembles regressive voice</p><p>assimilation in, for example, the Polish phrase nasz dom ‘our home’, which</p><p>is pronounced as [naZ dom] rather than *[naS dom]. As for nasal mutation,</p><p>some criteria, like the local presence of nasality, point to one solution,</p><p>while others, such as the lexicalization of the contexts, point to the other.</p><p>We will not take a firm position concerning NM and simply accept that</p><p>morpho-phonological effects may overlap with phonological ones. The full</p><p>picture of the mutations in Welsh that transpires from our analysis grades</p><p>the individual effects on the basis of the criteria we have suggested.</p><p>(47)</p><p>morphophonology ??? phonology</p><p>Soft Mutation Nasal Mutation Hard Mutation</p><p>Aspirate Mutation</p><p>It is interesting to note that the uncertain status of nasal mutation finds</p><p>some reflection in the way it is intuitively perceived by Welsh speakers.</p><p>Recall that there is one particle in standard Welsh, that is, fy [v´] ‘my’,</p><p>which causes NM while it does not possess nasality itself. In spoken collo-</p><p>quial Welsh, however, this particle is pronounced as [´n], thus perhaps</p><p>reflecting the intuitions of the speakers that nasal mutation has its source in</p><p>the preceding nasal sound. In the following chapter, similar overlapping</p><p>effects leading to ambiguous interpretations by speakers will be shown to</p><p>take place also at the syllabic level.</p><p>We will refrain from making particular theoretical claims as to the</p><p>status of morpho-phonological phenomena like initial mutations in gram-</p><p>mar.102 Proposals in this respect may oscillate between full lexicalization of</p><p>all mutated forms, where the choice of a particular exponent resembles the</p><p>selection of affixes (Green 2003), through segment replacement rules, whe-</p><p>reby a separate module called morpho-phonology would implement rules</p><p>102 To put it bluntly, the term ‘morpho-phonological’ that transpires from this dis-</p><p>cussion is very much synonymous with ‘non-phonological’</p><p>Summary and conclusions 71</p><p>of segment exchange, to element manipulation, an option that this analysis</p><p>of Welsh mutations</p><p>reopens.103</p><p>Even if all the mutated forms are indeed lexicalized, it seems that the</p><p>regularities may sometimes take a form of rules. This can be observed in</p><p>the phenomenon called reradicalization (Ball and Müller 1992, Chudak, in</p><p>prep.) whereby learners and speakers of Welsh or Irish mistake the mutated</p><p>form for the radical and apply regular mutations to the new radical forming</p><p>a new pattern for a given lexical item. This phenomenon, however, does</p><p>not provide a conclusive argument as to the nature of morpho-phonology.</p><p>5. Summary and conclusions</p><p>In this chapter, an attempt was made to show that sub-segmental complex-</p><p>ity, which may also be called substantive complexity, is an important factor</p><p>in the organization of phonological representation. The basic tenets of the</p><p>Element Theory were introduced and applied to a number of phonological</p><p>and morpho-phonological phenomena. The main characteristics of the me-</p><p>lodic primes in this model, that is their autonomous interpretability and</p><p>privativity, allow us to view the phonological representation as redun-</p><p>dancy-free and fully interpretable at any stage of derivation.</p><p>It has been shown that a number of phonological phenomena depend on</p><p>the internal complexity of segments. The relative substantive complexity</p><p>contributes to an understanding of both the static and dynamic aspects of</p><p>phonological systems. By static aspects we understand typological and mark-</p><p>edness effects as well as such system-specific areas as the segmental inven-</p><p>tory and phonotactics. To the dynamic aspect belong historical as well as</p><p>synchronic phonological processes such as the vocalic alternations of Irish,</p><p>voice neutralization in Polish, or consonant mutations in Irish and Welsh.</p><p>While the vocalic alternations of Irish are interpretations of morphologically</p><p>manipulated phonological representations (composition), the devoicing in</p><p>Polish consists in the interpretation of segments in which the laryngeal ele-</p><p>ment is not licensed in prosodically weak positions (decomposition).</p><p>The melodic representations of speech sounds should always be deter-</p><p>mined on the basis of a particular system rather than on the basis of pho-</p><p>netics. The fact that representations in the Element Theory are always sys-</p><p>tem dependent and are not uniform across languages, does not preclude</p><p>103 Recall that analyses based on distinctive features are unable to reduce the muta-</p><p>tions to simple and non-derivational operations.</p><p>72 Substantive complexity</p><p>similarities such as those found between the Irish and Welsh consonantal</p><p>systems, but the model eschews the automatic assignment of representa-</p><p>tions on phonetic basis alone.</p><p>Relatively complex objects behave like strong ones in cases where there</p><p>is a syntagmatic relation with other segments, for example, in phonotactics.</p><p>In this sense, complexity is able to replace both sonority and strength in pho-</p><p>nological theory, while also being able to account for markedness effects.</p><p>Complexity defined in terms of the number of privative elements in a</p><p>segment predicts the existence of asymmetries in phonological systems.</p><p>The phenomena discussed in this chapter demonstrate that this is a wel-</p><p>come prediction. For example, the relative complexity of the target vowels</p><p>in Irish vocalic alternations tallies with the relative regularity of the alter-</p><p>nations. The asymmetrical representation of laryngeal contrasts, on the</p><p>other hand, leads to asymmetries in neutralizations, and to particular pho-</p><p>notactic patterns concerning both complex onsets (English vs. Polish fr/vr</p><p>in section 2.3) and coda-onset contacts (Irish ‘epenthesis’ in 3.2).</p><p>One of the more important features of the Element Theory is the possi-</p><p>bility that not all universally recognized elements may be used in individ-</p><p>ual systems. Traditionally, this claim has been made about the use of the</p><p>two laryngeal elements (H/L), in that systems with two-way laryngeal dis-</p><p>tinctions have been assumed to employ only one of the elements. In this</p><p>chapter, we considered yet another case of the systemic absence of a prime,</p><p>which concerned the ‘noise’ element (h). This single parameter accounts</p><p>for the absence of affricates and voiced fricatives in Irish and Welsh, as</p><p>well as for the phonological behaviour of consonants in the two languages.</p><p>If systems can be differentiated by systemic choices of not using par-</p><p>ticular primes of a universal set, we predict that there may be systems</p><p>which develop additional primes which the Element Theory has not yet</p><p>established. A whole area of research opens up here touching on such is-</p><p>sues as typology, markedness and acquisition of categories. It should be the</p><p>task for further research to determine the status of the currently used</p><p>primes that can be dispensed with in individual systems, as well as the</p><p>potential directions in which the model could expand, should the existing</p><p>set of primes turn out to be insufficient.</p><p>Finally, the discussion of the consonant mutations in Welsh aimed at</p><p>demonstrating how the Element Theory is able to simplify the analysis of</p><p>this phenomenon if the consonantal system itself is first properly defined.</p><p>Even though the mutations are not truly phonological processes, the de-</p><p>scription of the effects in terms of elements allows us to understand the</p><p>logics of the phenomenon and gives us some inkling as to its nature when it</p><p>Summary and conclusions 73</p><p>was phonological. Given certain necessary representational assumptions,</p><p>the mutations can be defined as primarily tonal effects, very similar in kind</p><p>to obstruent devoicing and voicing. The simplification of the analysis is</p><p>particularly beneficial in that it offers a simpler view of the interaction be-</p><p>tween phonology and morphology, if one assumes that morphophonological</p><p>operations may manipulate the same categories as phonology proper.</p><p>As for the criteria for determining whether a given phenomenon is pho-</p><p>nological or morpho-phonological, it is clear that they must primarily fol-</p><p>low from the phonological theory. The theory must first be able answer the</p><p>question of what is a possible phonological process. In the Element Theory,</p><p>there are two main types of processes: composition and decomposition, both</p><p>of which must occur in a phonologically definable context.</p><p>In the following chapter, the prosodic level of phonological representa-</p><p>tion is scrutinized. We will first develop a phonological model to deal with</p><p>syllabification, which will be shown to be based on the interaction between</p><p>substantive and formal complexity on the one hand, and licensing on the</p><p>other.</p><p>Chapter 2</p><p>Formal complexity</p><p>1. Introduction</p><p>In the previous chapter it was shown that sub-segmental representation in</p><p>terms of the elements of Government Phonology may directly account for</p><p>quite a range of melodic phenomena, such as vowel reduction, vowel qual-</p><p>ity alternations, neutralization of voice, consonant mutations, phonotactics,</p><p>or even the segmental inventories of given languages. In most of these</p><p>phenomena the complexity of the representation, measured in terms of the</p><p>number of elements, appears to be as important a factor as any other, such</p><p>as, for example, the actual elements involved, homorganicity, etc. We also</p><p>saw that it is possible to derive some markedness effects from the melodic</p><p>(substantive) complexity. More importantly, substantive complexity has</p><p>been shown to play a key role in how consonants interact in syllabification.</p><p>In this respect, complexity profiles may replace sonority and strength scales</p><p>in determining the syllabification of consonants.</p><p>The aim of this chapter is to demonstrate how exactly substantive com-</p><p>plexity is incorporated into the higher level of phonological organization.</p><p>We will be mainly concerned with syllabic organization in its formal as-</p><p>pect, that is the structure of syllabic units. It will be shown that complexity</p><p>effects are also observed at this level. Since</p><p>they concern structural con-</p><p>figurations, the term formal complexity will be used. The relative complex-</p><p>ity of syllabic structure, if captured correctly in a formal model, provides</p><p>direct access not only to the definition and understanding of individual</p><p>systems and syllabically driven phonological processes, but also to syllabic</p><p>typology and markedness.</p><p>The attraction of deriving syllabic markedness from a formally defined</p><p>complexity scale is evident. However, to achieve this goal, a few serious</p><p>modifications of the model of Government Phonology will have to be in-</p><p>troduced consisting mainly in simplifying its apparatus to the bare mini-</p><p>mum. In general, this minimum involves the presence of governing relations</p><p>between consonants and licensing relations between nuclei and onsets. On</p><p>the other hand, some principles and parameters which defined phonologi-</p><p>76 Formal complexity</p><p>cal structure and constituted the core of standard Government Phonology</p><p>in the past will be eliminated from the model and replaced with formally</p><p>defined scales. This modified model will be shown to be fully compatible</p><p>with the hypothesis that syllable structure is in fact a sequence of simplex</p><p>onsets and nuclei, that is CVs (Lowenstamm 1996).</p><p>Descriptively, we will look at phenomena which are crucial for the un-</p><p>derstanding of syllabification, such as the distribution of clusters, vowel</p><p>epenthesis, and vowel – zero alternations in Dutch, French and Polish. It</p><p>will be shown that the modified model of Government Phonology is fully</p><p>falsifiable, and that, apart from being able to capture all the phenomena</p><p>connected with syllabification, it provides tools which may enable us to</p><p>understand dialectal and register variation, historical shifts, the acquisition</p><p>of syllable structure, the interaction between phonology and morphology,</p><p>and the role of phonology in determining word structure, an issue which</p><p>will be addressed in more detail in chapter 3. We begin the discussion by</p><p>reviewing some basic facts concerning syllabification.</p><p>2. Syllabification</p><p>2.1. Basic facts</p><p>The structure we assume for the syllable is fairly well-established. We</p><p>begin by providing some examples of fairly uncontroversial syllabification</p><p>using this template, in order to be able later to introduce the alternative</p><p>model of syllabification which is advocated in Government Phonology.</p><p>The syllable is often equated with the presence of a vowel which assumes</p><p>the position of the nucleus. The consonant, or consonants preceding the</p><p>nucleus belong to the onset, while those which follow the nucleus belong</p><p>to the coda of the syllable.</p><p>(1) σ</p><p>σ = syllable</p><p>O R O = onset</p><p>R = rhyme</p><p>N C N = nucleus</p><p>C = coda</p><p>x x x...</p><p>| | |</p><p>k œ n d I</p><p>Syllabification 77</p><p>Let us now observe how syllabic divisions are made in the following three</p><p>words: baby, vulgar, and cobra.</p><p>(2)</p><p>a. ba.by b. vul.gar c. co.bra</p><p>While most linguists will probably agree with the syllabification of the</p><p>words above, the means of arriving at such divisions may differ across</p><p>models. Also, views on the correctness of particular divisions may differ</p><p>once more complex, or less obvious, clusters are taken into account.</p><p>As mentioned above, nuclei are said to be the most important ingredient</p><p>of the syllable, therefore, they will be projected onto the prosodic level</p><p>first, as heads of syllables. What we can ascertain at this stage is that all</p><p>three words in (2) are bisyllabic. However, we must now prove that the</p><p>consonants are adjoined to the syllable heads in the way illustrated in (2).</p><p>There are two basic questions. First, what makes a single intervocalic con-</p><p>sonant end up in the onset of the second syllable in ba.by rather than as the</p><p>coda of the first (*bab.y)? And second, on what basis are the consonant</p><p>clusters in (2b) and (2c) separated by a syllable boundary (2b), or syllabi-</p><p>fied together as a branching onset in (2c)? We expect that a model which</p><p>produces the intuitively correct divisions in (2) will also rule out the incor-</p><p>rect forms, e.g. *bab.y, *vu.lgar, *cob.ra.</p><p>The answer of standard generative models to the questions posed above</p><p>consists in establishing syllable building procedures, or rules intertwined</p><p>with general cross-linguistic principles and language specific constraints.</p><p>One such principle, which interacts with language specific constraints,</p><p>pertains to the maximization of onsets.</p><p>(3) Maximal Syllable Onset Principle (Selkirk 1982)</p><p>In the syllable structure of an utterance, the onsets of syllables are maxi-</p><p>mized, in conformance with the principles of basic syllable composition of</p><p>the language.</p><p>This principle ensures that the intervocalic consonant in baby is assigned to</p><p>the onset of the second syllable in (2a). It also tells us why *cob.ra is not</p><p>correctly syllabified. Given that br is a well-formed branching onset in</p><p>78 Formal complexity</p><p>English, which it is, it must be syllabified as such.1 On the other hand, *lg</p><p>is not a possible branching onset and this sequence must be separated by a</p><p>syllable boundary, hence vul.gar. The choice between a well-formed branch-</p><p>ing onset and a coda-onset sequence is determined by a principle relating to</p><p>the inherent sonority / resonance of segments, or an inherent scale of seg-</p><p>mental strength (Murray 1988).2</p><p>(4) Sonority Sequencing Generalization (Harris 1994)</p><p>An optimal syllable consists of a sonority peak, corresponding to the nu-</p><p>cleus, optionally flanked by segments which decrease in sonority the further</p><p>they occur from the nucleus.</p><p>Thus, we may say that the division in *vu.lgar is incorrect because the</p><p>sonority slope of the cluster decreases towards the syllable nucleus, while</p><p>it should increase. In the previous chapter, it was demonstrated that in the</p><p>Element Theory this aspect is dealt with by means of complexity slopes.</p><p>To summarize briefly, there are three aspects of syllabification which</p><p>seem to be important: the supremacy of nuclei, the precedence of onsets in</p><p>the syllabification of consonants, and principles of phonotactics. The latter</p><p>term covers a wide area as it first of all involves language specific deci-</p><p>sions concerning the types of formal structures to be allowed, for example,</p><p>whether branching onsets or coda-onset clusters are present. These major</p><p>parameters or constraints are further supplemented by conditions on what</p><p>good branching onsets and good coda-onset contacts are. This can be con-</p><p>trolled in terms of sonority, strength, or complexity distance between con-</p><p>sonants, as we saw in the previous chapter. However, these conditions are</p><p>dependent on major syllable structure decisions, and are immaterial in sys-</p><p>tems which have no clusters at all.</p><p>In more recent models such as Optimality Theory (Prince and Smolen-</p><p>sky 1993), syllabification procedures are replaced with an interaction of</p><p>markedness constraints pertaining to the syllable structure in the output.</p><p>1 The data from Polish to be discussed below will demonstrate how naive this sta-</p><p>tement in fact is. The fact that br is a well-formed branching onset does not guaran-</p><p>tee that this structure should be imposed on any such surface string.</p><p>2 The syllable contact law (e.g. Murray and Vennemann 1983) redefines the sonor-</p><p>ity hierarchy as one of consonant strength, where the values are the converse of</p><p>those in the sonority model. In such a model a preferred syllable is defined as one</p><p>in which the strength of consonants consistently decreases from the outer margins</p><p>to the nucleus.</p><p>Syllabification 79</p><p>For example, the respective phonological constraints ONSET, NUCLEUS,</p><p>*CODA, *COMPLEX CODA, and *COMPLEX ONSET express the observation</p><p>that preferred syllables have onsets and nuclei and avoid having simplex or</p><p>complex codas, and complex onsets. The unmarked</p><p>syllable structure, that</p><p>is CV, does not involve the violation of either of the above listed constraints.</p><p>However, a violation of any of these constraints is possible, thus producing</p><p>more marked syllable types, relatively speaking.3</p><p>After this simple introduction to syllabification, let us proceed to a dis-</p><p>cussion of the views of Government Phonology (GP) on the subject.</p><p>2.2. Government</p><p>Government Phonology translates the syllable contact laws into depend-</p><p>ency or governing relations between consonants. Syllabification, therefore,</p><p>proceeds from governing relations contracted between consonants.</p><p>Whether a consonant is a governor, which we will symbolically represent</p><p>by the capital letter (T), or a governee-(R) in such relations is determined by</p><p>their segmental complexity differential. It will be recalled from the discus-</p><p>sion in the previous chapter that, to some extent, complexity reflects sonor-</p><p>ity to the effect that the more complex the segment the less sonorous it is.4</p><p>Note that complexity profiles are comparable with sonority or strength</p><p>slopes, and the theory of government finds a role for these slopes to play.</p><p>Thus, a more complex segment always governs a less complex one, regard-</p><p>less of their linear order in a string, as illustrated below.</p><p>(5)</p><p>g l l g</p><p>| | | |</p><p>T R R T</p><p>( ) = direction of government, T = governor, R = governee</p><p>3 This approach echoes earlier generative work on syllable markedness and evalua-</p><p>tion of structural markedness (e.g. Cairns and Feinstein (1982).</p><p>4 In fact, the complexity of consonants which is defined in terms of the number of</p><p>phonological elements present in their melodic make-up corresponds to a large</p><p>extent to the strength scale proposed in Vennemann (1972). Since the complex</p><p>consonants are governors, applying the term ‘strong’ to them is also very apt.</p><p>80 Formal complexity</p><p>Let us disregard the exact substantive complexities of g and l at this stage</p><p>and assume that in a sequence of two consonants T, the governor, is more</p><p>complex than R, that is, the governee. Though it is not impossible to assign</p><p>a fixed function to some segments as typical governees, for example,</p><p>glides, or typical governors, for example stops, we will assume that these</p><p>functions are always worked out for any given sequence.5 For example, f is</p><p>likely to be a governor when adjacent to a liquid, as in fling or alpha, or a</p><p>governee when followed by a stop, as in hefty.</p><p>In general, as we saw in the previous chapter, obstruents have more</p><p>complex representations than sonorants, therefore, when g and l stand next</p><p>to each other in a string, g will always be the governor because it is more</p><p>complex than l.6 Note that this fact leads to two types of situations: one in</p><p>which the governing relation goes from left to right, and another one in</p><p>which the direction of government is reversed.</p><p>In terms of the actual syllabic configurations, the rightward governing</p><p>relation defines what we traditionally understand as branching onsets, and</p><p>the leftward direction specifies the relation between the onset and the pre-</p><p>ceding non-vocalic complement of a branching rhyme, that is, the coda. We</p><p>illustrate this by providing the relevant fragments of the syllable structures</p><p>of the now familiar words.</p><p>(6) a. co.bra b. vul.gar</p><p>O N R O N</p><p>k ´ U b r ´ v ø l g ´</p><p>| | | |</p><p>T R R T</p><p>Government, however, should not be viewed as a mere theoretical rephras-</p><p>ing of contact laws and sonority sequencing. One advantage of the model is</p><p>that the nature of government restricts possible syllabic types, because in</p><p>any given direction only two positions – the adjacent ones – may contract a</p><p>governing relation. This, effectively, allows only for maximally binary</p><p>5 Such labelling of consonants with a fixed function was attempted in the early</p><p>Element Theory (Kaye, Lowenstamm and Vergnaud 1985). It is known as the The-</p><p>ory of Charm and was promptly abandoned in GP (see e.g. Harris 1990).</p><p>6 Unless we are dealing with a system like Irish or Welsh, discussed in the previous</p><p>chapter, in which a mono-elemental g can hardly be considered a good governor.</p><p>Syllabification 81</p><p>branching constituents: branching onsets, nuclei, and rhymes. This makes</p><p>the model highly constrained. On the other hand, there is nothing in the</p><p>standard generative models or Optimality Theory to constrain the size of</p><p>syllables other than observation turned into language specific constraints.</p><p>All the possible syllabic constituents which are recognized in standard GP</p><p>are listed below.</p><p>(7) a. onset b. nucleus c. rhyme</p><p>O O N N R R O</p><p>|</p><p>N N</p><p>| |</p><p>x x x x x x x x x x</p><p>As for the simplex structures, it must be mentioned that a non-branching</p><p>rhyme is in fact identical to a non-branching nucleus and refers simply to a</p><p>short vowel. Logically, since the nucleus is subsumed under the rhyme, the</p><p>latter may contain a branching nucleus as well, which is not shown in the</p><p>above structures. Branching constituents, on the other hand, may be de-</p><p>fined as involving governing relations which are from left to right. The</p><p>only governing relation which goes in the opposite direction is that between</p><p>an onset and the preceding rhymal complement as in (7c). Note that ternary</p><p>structures would either violate adjacency between governor and governee,</p><p>or the directionality of governing relations. Therefore, there are no ternary</p><p>branching constituents.7 The model allows for a simple definition of the</p><p>syllable structure of a given system in that what is required is a statement</p><p>concerning the ability of particular constituents to branch, a statement</p><p>which may be couched in terms of parameters, for example.</p><p>It is interesting to note that in standard GP a branching rhyme involves</p><p>a very complex structure in which not only is the rhymal complement gov-</p><p>erned by the head from the left, but is also governed by the following onset</p><p>(7c). It will be shown later that this rhymal complement is in fact the only</p><p>structural instance in which we may speak of a coda. What should be borne</p><p>7 Some cases of complex onsets and rhymes where binarity seems to be breached</p><p>will be returned to. This problem concerns, for example, Polish initial consonant</p><p>clusters and English super-heavy rhymes, in which the branching rhyme contains a</p><p>branching nucleus.</p><p>82 Formal complexity</p><p>in mind, however, is the fact that the structure of the branching rhyme</p><p>would not be possible without the two relations.8</p><p>2.3. Licensing</p><p>Having seen how consonantal clusters are syllabified in GP we must return</p><p>to the question of the role of nuclei and also to the precedence of onsets in</p><p>the syllabification of consonants. Like other approaches, GP assumes that</p><p>vowels / nuclei constitute an indispensable part of the syllable. One reason</p><p>for this assumption is the simple fact that while we can have monosyllabic</p><p>words without an onset consonant, a monosyllable cannot be deprived of a</p><p>melodically filled nucleus. Another reason for treating nuclei as special is</p><p>their participation in higher prosodic organization, that is, foot and word</p><p>structure. In this respect, nuclei are assumed to be the carriers of prosodic</p><p>information in the phonological representation. It is through nuclei that the</p><p>prosodic licensing is distributed within the phonological word. Before we</p><p>examine an example of prosodic phenomena connected with this type of</p><p>licensing, let us look at the lowest level of licensing relations, the one hold-</p><p>ing between the nucleus and its onset.</p><p>(8)</p><p>O N</p><p>| |</p><p>C V</p><p>licensing relation</p><p>It is assumed that each nucleus must license its onset, a relation</p><p>which</p><p>encapsulates the two aspects of syllabification which we discussed above.</p><p>Firstly, it directly reflects the supremacy of the nucleus within the syllable.</p><p>It is indispensable because it is the licenser. It is the organizing agent in the</p><p>utterance, without which the syllable would not exist. Secondly, the exis-</p><p>tence of the relation with the preceding onset, and not with the following</p><p>one, accounts for the fact that single intervocalic consonants are syllabified</p><p>as onsets in words such as ba.by. In other words, by recognizing the exis-</p><p>tence of the licensing relation between the nucleus and its onset, we are</p><p>8 In the ensuing discussion it will be shown that a branching Rhyme is not an inde-</p><p>pendent constituent. More detailed discussion and a concrete proposal concerning</p><p>this problem can be found in chapter 3. See also Lowenstamm (1996), Takahashi</p><p>(1993) and Scheer (1996) for similar conclusions.</p><p>Syllabification 83</p><p>able to account for basic syllabification without resorting to additional</p><p>principles such as Onset Maximization, which in reality, merely state the</p><p>facts and do not provide the theoretical means of deriving them.</p><p>Let us briefly return to the forms vul.gar and co.bra whose syllabifica-</p><p>tion was explained above in (6). Note that, in these cases, the nuclei which</p><p>directly follow the clusters should also remain in a licensing relation with</p><p>their onsets.9 What is more, we may now view the governing relations be-</p><p>tween the consonants as an extension of the licensing coming from the</p><p>nucleus. This way, each position within the word appears to be licensed</p><p>one way or another. Assuming that the stressed vowel is the head of the</p><p>prosodic domain called the word, the distribution of prosodic licensing</p><p>down to the level of interconsonantal relations can be illustrated in the</p><p>following way.10 For clarity of presentation the projection of the nucleus at</p><p>the level of the foot is represented as R = rhyme.</p><p>(9) a. b. c.</p><p>R R R R R R</p><p>| | | | |</p><p>O N O N O N O N O N O N</p><p>| | | | | | | | |</p><p>b e I b I v ø l g ´ k ´ U b r ´</p><p>governing relation licensing relation</p><p>In each case, the licensing goes from the nucleus to its onset which, on the</p><p>other hand, may stand alone, for example, the second b in baby (9a), or</p><p>find itself in a governing relation with a neighbouring consonant of lower</p><p>complexity (higher sonority). Rightward government defines the constitu-</p><p>ent called a branching nucleus (9a), branching rhyme (9b), or branching</p><p>onset (9c), whereas a leftward relation obtains between consonants belong-</p><p>ing to two separate constituents (9b).</p><p>A careful reader will have noticed that now we may claim that indeed</p><p>all syllabification is somehow connected with nuclei licensing their onsets</p><p>which, in turn, find themselves in different prosodic arrangements. We will</p><p>9 It was Charette (1990) who first proposed that governing relations between con-</p><p>sonants must be licensed by nuclei.</p><p>10 See Harris (1997) for a fully articulated theory of prosodic licensing and its role</p><p>in such phonological processes as lenition and fortition.</p><p>84 Formal complexity</p><p>come back to this observation in the following section. Having noted the</p><p>licensing relation between a nucleus and the preceding onset, the next logi-</p><p>cal question should be whether languages may differ with respect to the</p><p>licensing properties of their nuclei, leading to a cross-linguistic variation in</p><p>types and sizes of the onsets. To answer this question, we must establish</p><p>how the licensing properties of nuclei may differ. As well as this, we need</p><p>to find out whether there is a phonologically definable property of onsets</p><p>which would allow us to gauge the licensing abilities of nuclei. This is</p><p>what we will turn to now.</p><p>3. Syllable markedness as a scale of formal complexity</p><p>In the previous chapter, we considered only one way to deal with marked-</p><p>ness effects in Government Phonology, which involves making reference to</p><p>the relative complexity of segments, that is, substantive complexity. The</p><p>purpose of this section is to demonstrate that the same basic mechanism,</p><p>that is, the interaction between licensing and the relative complexity of the</p><p>structures, may capture markedness and typological tendencies concerning</p><p>syllabification. To refer to these structural effects, the term Formal Com-</p><p>plexity will be used.</p><p>Recall that since segments in this model are composed of privative ele-</p><p>ments, the actual cost of licensing particular objects is calculated straight-</p><p>forwardly from the number of elements involved. It is to be expected that</p><p>in prosodically weak contexts, the less complex segments should have a</p><p>better chance of survival than compounds. This prediction is borne out by</p><p>phonological phenomena such as the lowering or raising of mid vowels in</p><p>unstressed nuclei as in, for example, Bulgarian and Catalan (Harris 1994).</p><p>Obstruent devoicing, as in Polish or German, is captured in exactly the</p><p>same way as vowel reduction. Simply, the element defining the laryngeal</p><p>activity is unlicensed in prosodically weak positions. Thus, the general</p><p>principle responsible for markedness phenomena in segmental structure in</p><p>GP is viewed as the distributing of various complexities within a word in</p><p>such a way that the amount of phonological material tends to be greater in</p><p>strong positions and reduced in weak ones. Harris (1997), for example,</p><p>proposes a coherent theory of neutralization, which unifies the intimate rela-</p><p>tionship between the distribution of prosodic licensing within a word and the</p><p>allocation of melodic contrasts.</p><p>Later in this chapter, we will see how the complexity of consonantal seg-</p><p>ments may account for cross-linguistic patterns concerning their occurrence</p><p>Syllable markedness as a scale of formal complexity 85</p><p>in word-final position. First, however, we illustrate how syllable typology</p><p>and markedness can be captured in GP by referring to the same concepts as</p><p>in the case of segmental markedness, that is, complexity and licensing. Let us</p><p>begin by reviewing some facts concerning syllable markedness.</p><p>3.1. Syllable markedness</p><p>Kaye and Lowenstamm (1981) observe an implicational relationship that</p><p>seems to hold cross-linguistically between branching rhymes and branch-</p><p>ing onsets, that is between forms such as vul.gar and co.bra. The observa-</p><p>tion stipulates that a language which has branching onsets must also pos-</p><p>sess in its syllabic inventory the structure of a branching rhyme. Since the</p><p>implication cannot be reversed, the following scale of progressively</p><p>marked syllabic structures is derived.11</p><p>(10)</p><p>a. O b. R O c. O</p><p>N</p><p>|</p><p>b e I b I v ø l g ´ k ´ U b r ´</p><p>The implications illustrated above are traditionally understood in the fol-</p><p>lowing way. The least marked syllable structure is that with a simplex on-</p><p>set and a short nucleus (CV). The second step on the scale of markedness</p><p>is represented by a syllable which has a coda (10b), that is CVC, and the</p><p>presence of this structure obviously implies the unmarked structure in</p><p>(10a). Finally, the most marked structure is that with a branching onset</p><p>(10c), the presence of which necessarily implies the previous less marked</p><p>structures.</p><p>Thus, Kaye and Lowenstamm divide the syllabic complexities into three</p><p>major levels corresponding to the choices which languages make concern-</p><p>ing their syllable structure.</p><p>11 For a discussion of typology and markedness in syllable structure see, for exam-</p><p>ple, Blevins (1995), Cairns and Feinstein (1982), van der Hulst and Ritter (1999),</p><p>Kaye, Lowenstamm and Vergnaud (1990), McCarthy and Prince (1994), Prince</p><p>and Smolensky (1993).</p><p>⊂ ⊂</p><p>86 Formal complexity</p><p>(11) Three levels of syllable</p><p>markedness</p><p>I CV Zulu, Desano = (10a)</p><p>II CV, CVC Hungarian, Japanese = (10a,b)</p><p>III CV, CVC, CCV Polish, English = (10a,b,c)</p><p>The question that must be answered concerns the theoretical relationship</p><p>between all three structures, which must be established for the purpose of</p><p>accounting for the markedness scale in a non-arbitrary fashion. Especially</p><p>troublesome is the distinction between the branching onset and the branch-</p><p>ing rhyme, because there seems to be no formal connection between the</p><p>two structures. On the other hand, the unmarked nature of CV appears to</p><p>be rather uncontroversial.</p><p>In order to account formally for the implications shown in (10), Kaye</p><p>and Lowenstamm propose to index the markedness scale in the following</p><p>way (1981: 292).</p><p>(12)</p><p>The markedness values are established separately for the onset and for the</p><p>rhyme. For this reason branching onsets end up having the same marked-</p><p>ness value as the branching rhyme, contrary to the classification in (11)</p><p>which suggests that the two structures must constitute separate levels. To</p><p>amend this situation, Kaye and Lowenstamm postulate that the implication</p><p>CCV ⊃ CVC may be handled by a separate condition stipulating that the</p><p>maximum markedness value for the onsets m may be equal but should not</p><p>exceed that for the rhyme n (m ≤ n). Despite this little glitch, one cannot</p><p>but admire the ingenuity of the observation, given that no obvious formal</p><p>connection between branching onsets and branching rhymes can be readily</p><p>supplied in any phonological model to this day.</p><p>The following section demonstrates that the basic insight of Kaye and</p><p>Lowenstamm (1981), summarized in (10) and (11) above, may receive a</p><p>fairly non-arbitrary description within a slightly modified model of Gov-</p><p>ernment Phonology, and that there is no need for a separate condition dif-</p><p>Onset Rhyme Markedness</p><p>C V 0</p><p>P P 1</p><p>CC VC 2</p><p>CCC VCC 3</p><p>C1...Cn VC1...Cn-1 n</p><p>Syllable markedness as a scale of formal complexity 87</p><p>ferentiating branching onsets and rhymes, because they are not of the same</p><p>markedness value.</p><p>3.2. Problems with parameters</p><p>From our earlier discussion of syllabification in standard GP it transpired</p><p>that governing relations between consonants are not only indicative of the</p><p>presence of branching constituents. They in fact restrict the maximal struc-</p><p>ture of branching constituents to binary relations. Thus, given that govern-</p><p>ment is able to define all syllable types, that is, cover the syllable typology</p><p>while retaining binarity, it may be possible to define syllabic systems by</p><p>means of simple parameters on branchingness.12 This is, in fact, the stan-</p><p>dard way of capturing syllable typology in GP, which is illustrated below.</p><p>(13) Branching</p><p>Onset ON/OFF</p><p>Rhyme ON/OFF</p><p>Nucleus ON/OFF</p><p>If the parameter for branching onsets is set in the OFF position, the system</p><p>only has simplex onsets. If the parameter is switched ON, the system pos-</p><p>sesses both branching and simplex onsets. The same concerns nuclei. On</p><p>the other hand, the parameter for branching rhymes in fact determines the</p><p>existence of internal codas, and, in a system which has long vowels, the</p><p>possibility of having super-heavy rhymes, e.g. bold, find, etc. A discussion</p><p>of the latter problem is delayed till chapter 3.</p><p>According to this set of parameters, the syllabic systems of Polish and</p><p>English differ in terms of one parameter: in Polish the parameter for</p><p>branching nuclei is switched OFF. This effectively deprives Polish of long</p><p>vowels and super-heavy rhymes. Otherwise the systems may be said to be</p><p>similar; however, the complex initial and final clusters in Polish require an</p><p>additional explanation.13</p><p>12 In section 5.4 we deal with cases of structures seemingly exceeding binarity.</p><p>13 Some discussion of Polish clusters will be offered later in this chapter and in chap-</p><p>ter 3. For more extensive analyses, the reader is referred to, for example, Bargiełówna</p><p>(1950), Cyran and Gussmann (1999), Gussmann and Kaye (1993), Kuryłowicz</p><p>(1952), Rubach (1977), Rubach and Booij (1990a, 1990b), Rowicka (1999).</p><p>88 Formal complexity</p><p>Before we show how the above mentioned parameters fail to account</p><p>for the basic markedness tendencies discovered in Kaye and Lowenstamm</p><p>(1981), let us look briefly at the other ingredient of syllabification which</p><p>was mentioned earlier, namely, licensing.</p><p>Charette (1990, 1992) proposes that both types of governing relations</p><p>between consonants, that is, rightward (b→r) and leftward (l←g), must be</p><p>licensed by the following nucleus. She distinguishes between indirect and</p><p>direct Government Licensing (GL), respectively, as separate licensing prop-</p><p>erties of nuclei. The distinction direct vs. indirect takes into account the</p><p>adjacency of the licenser and the licensee at the skeletal level, because at</p><p>the constituent level no such distinction really exists, as can be seen in</p><p>(14).14 The following symbols are used below: T = governor, R = governee,</p><p>a = any vowel.</p><p>(14) a. indirect GL b. direct GL</p><p>O R R O R</p><p>| |</p><p>N N N</p><p>| | |</p><p>... T R a ... a R T a</p><p>government, government licensing</p><p>Because the distinction between direct and indirect GL in Charette (1990,</p><p>1992) is not used for broad typological purposes, but rather for concrete</p><p>analyses of the interaction of licensing with Proper Government, that is,</p><p>relations between nuclei, it is not clear if the distinction corresponds to that</p><p>between two independent parameters.15</p><p>The positive setting of the two parameters may be assumed to condition</p><p>the presence of governing relations of the T→R and R←T type in a given</p><p>language, and, in effect, of branching onsets and branching rhymes. It must</p><p>be noted, however, that the relationship between the individual government</p><p>licensing parameters and the corresponding branching constituents is not</p><p>identical. The parameter for indirect government licensing refers directly to</p><p>the two consonants that form the branching onset. In a sense, then, the ef-</p><p>14 This is one of the reasons why the constituent-based views of standard GP will be</p><p>modified later in this chapter.</p><p>15 See however the table in Charette (1992:289) where direct and indirect GL are</p><p>treated as separate parameters.</p><p>Syllable markedness as a scale of formal complexity 89</p><p>fect of this parameter overlaps with the parameter on branching onsets. The</p><p>same cannot be said about the relationship between direct government li-</p><p>censing and the existence of branching rhymes. Direct government licens-</p><p>ing is responsible only for the governing relation between the onset and the</p><p>coda consonant in the preceding syllable. It is not clear how this could</p><p>evoke a branching rhyme structure, which itself is defined in standard GP,</p><p>like any other branching constituent, by a left-headed relation between the</p><p>nucleus and the rhymal complement. This mismatch is probably the reason</p><p>why the government licensing parameters have never fully replaced the pa-</p><p>rameters for branching constituents. It must be noted, though, that the two</p><p>types of parameters, that is, those referring directly to branching constitu-</p><p>ents and those which define the licensing properties of nuclei, are overlap-</p><p>ping and potentially conflicting.</p><p>Let us now see how these two types of parameters, that is, parameters</p><p>for branching constituents and parameters for the presence of government</p><p>licensing, fare separately and in conjunction with respect to the observation</p><p>made in Kaye and Lowenstamm (1981) that there is an implicational rela-</p><p>tionship between branching onsets and branching rhymes. We begin with</p><p>the parameters for branching constituents.</p><p>Though the parameters for branching constituents can describe typo-</p><p>logical variation, they are unable</p><p>Conclusion................................................................................................289</p><p>References ................................................................................................291</p><p>Subject index............................................................................................309</p><p>Language index ........................................................................................311</p><p>Chapter 1</p><p>Substantive complexity</p><p>1. Introduction</p><p>The aim of this chapter is to demonstrate that one of the crucial organizing</p><p>properties of phonological representation at the melodic level is subseg-</p><p>mental complexity, which is of a scalar character. Substantive complexity,</p><p>as we will call it, will be shown to play a pivotal role in phonological sys-</p><p>tems, contributing to the understanding of certain static aspects of these</p><p>systems, for example, segmental inventories, phonotactics, typology, mark-</p><p>edness effects etc., as well as a number of dynamic characteristics such as</p><p>phonological processing, in both its synchronic and historical dimension.</p><p>The chapter is organized in the following way. First, in section 2, the</p><p>Element Theory is introduced and illustrated by focusing on both simple</p><p>and more complex aspects of sub-segmental representation, and by show-</p><p>ing that complexity may successfully replace such concepts as sonority,</p><p>and strength in all the areas of phonological theory where they were used</p><p>to account for phonological systems, including the syllabification of con-</p><p>sonants (section 3). Then, in section 4, we look more deeply at the system</p><p>of modern Irish with a view to illustrating how the model can be practically</p><p>applied to a range of phenomena within one phonological system. First, we</p><p>deal with vowel quality alternations and show the advantages of a privative</p><p>model employing elements over an equipollent feature system in capturing</p><p>the existing alternations, as well as capturing the peculiar pattern whereby</p><p>the relative regularity of the phenomenon is strictly dependent on the</p><p>height distinctions of the target vowels. The second aspect of the phono-</p><p>logical system of Irish which is dealt with concerns the role of substantive</p><p>complexity in determining grammatical coda-onset contacts. Here, a modi-</p><p>fication of the model will be proposed, which consists in allowing the utili-</p><p>zation of a particular melodic prime to be subject to parameterization. This</p><p>move will be shown to facilitate a better understanding of Irish phonotactics</p><p>and to have additional, far-reaching consequences for the types of segments</p><p>that this phonological system may theoretically employ. Some systemic dis-</p><p>tinctions leading to typological variation between consonantal systems will</p><p>2 Substantive complexity</p><p>be proposed, of which the distinction involving internal complexity seems</p><p>to be the most important. Finally, we focus our discussion on the phe-</p><p>nomenon of initial consonant mutations in Welsh – another Celtic language</p><p>– with a view to showing how this seemingly complex phenomenon can</p><p>receive a fairly simple analysis within the Element Theory.</p><p>2. The Element Theory in Government Phonology</p><p>The smallest units of phonological representation in Government Phonol-</p><p>ogy are called elements.1 The term has been chosen not only to oppose this</p><p>construct to the traditional features, but also to convey the similarity of</p><p>their behaviour to physical elements, in that they can occur in isolation –</p><p>simplex structures, or in compounds – complex structures. In a nutshell,</p><p>the elements can be characterized as privative, cognitive units which enjoy</p><p>a stand-alone phonetic interpretability. Privativeness, as opposed to equi-</p><p>pollence, means that each relevant property of melodic representation is</p><p>defined by the physical presence of a given prime, and phonological proc-</p><p>esses may refer only to actively present elements, rather than to their ab-</p><p>sence, or to a negative value for them. The term ‘cognitive unit’ is used to</p><p>convey the fact that elements which encode lexical contrasts are neither</p><p>articulatory nor auditory in nature.2</p><p>...continuing the essentially Jakobsonian line of thinking, we consider their</p><p>phonetic implementation as involving in the first instance a mapping onto</p><p>sound patterns in the acoustic signal. Viewed in these terms, articulation and</p><p>perception are parasitic on this mapping relation. That is, elements are inter-</p><p>nally represented templates by reference to which listeners decode auditory</p><p>input and speakers orchestrate and monitor their articulations.</p><p>Harris and Lindsey (1995: 50)</p><p>1 This section draws heavily on Harris (1990, 1996) and Harris and Lindsey (1993,</p><p>1995). Early GP proposals on elements also include Kaye (1989), Kaye, Low-</p><p>estamm and Vergnaud (1985, 1990), Rennison (1987, 1990). Other contributions</p><p>are Backley (1993, 1995), Backley and Takahashi (1998), Brockhaus (1995), Cha-</p><p>rette and Göksel (1996, 1998), Cobb (1993, 1997), Cyran (1996b, 1997), Denwood</p><p>(1993), Harris (1997), Jensen (1994), Kaye (2001), Nasukawa (1998, 2005), Ploch</p><p>(1999), Pöchtrager (2006), Ritter (1997), Rennison (1998), Rennison and Neubarth</p><p>(2003), Scheer (1996, 2004), Szigetvári (1994).</p><p>2 See, for example, Coleman (1998) for a review of various arguments concerning</p><p>the nature of linguistic primes, in which he arrives at similar conclusions.</p><p>The Element Theory in Government Phonology 3</p><p>As far as autonomous interpretability is concerned, it is assumed that each</p><p>element that is linked to a skeletal position can be directly realized as a</p><p>speech sound, either alone, or in combination with other elements. The</p><p>phonological representations remain privative and redundancy-free throug-</p><p>hout the derivation. There is no place for any default fill-in procedures. For</p><p>example, sonorants are non-specified for voice lexically, and they remain</p><p>so at every stage of the derivation.3 Thus, there is no need for a level of</p><p>systematic phonetic representation (Harris and Lindsey 1993, 1995: 46).</p><p>The details of the Element Theory will transpire as we proceed. It will</p><p>also become obvious that some assumptions which are fit for an introduc-</p><p>tion to the Element Theory must be verified and confronted with particular</p><p>phonological systems. Let us first look at an exhaustive list of what we</p><p>assume to be a standard set of elements in GP. The following table defines</p><p>the elements in terms of their acoustic patterns and the necessary articulatory</p><p>execution required in their production (adapted from Harris 1996: 314).</p><p>(1)</p><p>Acoustic pattern Articulatory execution</p><p>A</p><p>Mass: central spectral energy mass</p><p>(convergence of F1 and F2)</p><p>Maximal expansion of oral tube; ma-</p><p>ximal constriction of pharyngeal tube</p><p>I</p><p>Dip: low F1 coupled with high spec-</p><p>tral peak (convergence of F2 and F3)</p><p>Maximal constriction of oral tube;</p><p>maximal expansion of pharyngeal tube</p><p>U</p><p>Rump: low spectral peak (conver-</p><p>gence of F1 and F2)</p><p>Trade-off between expansion of oral</p><p>and pharyngeal tubes</p><p>/</p><p>Edge: abrupt and sustained drop in</p><p>overall amplitude</p><p>Occlusion in oral cavity</p><p>h</p><p>Noise: aperiodic energy Narrowed stricture producing turbu-</p><p>lent airflow</p><p>N</p><p>Nasal: low frequency of first reso-</p><p>nance</p><p>Lowered velum; air flow through the</p><p>nasal passage</p><p>H</p><p>High tone: raised pitch on vowels;</p><p>VOT lag (aspiration) in obstruents</p><p>Stiff vocal cords</p><p>L</p><p>Low tone: lowered pitch on vowels;</p><p>VOT lead (full voicing) in obstruents</p><p>Slack vocal cords</p><p>3 The modal voicing of sonorants in the Element Theory may be said to follow from</p><p>the fact that they are typically represented by the same primes as vowels, that is,</p><p>resonance elements, to be introduced below. Most sonorants exhibit spectral pat-</p><p>terns similar to vowels.</p><p>4 Substantive complexity</p><p>Before we continue the discussion, it must be emphasized that the rough</p><p>universal cues inherent in the elements listed</p><p>to account for the syllable markedness</p><p>observation made by Kaye and Lowenstamm. To see this clearly, let us con-</p><p>sider all the possible configurations concerning the parameters for branching</p><p>onsets and rhymes which are predicted by the model.</p><p>(15)</p><p>parameters a. b. c. d.</p><p>TR branch onset ON OFF OFF ON</p><p>RT branch rhyme ON OFF ON OFF</p><p>English Zulu Japanese ???</p><p>Note that the system in (15d), that is, one which has branching onsets but</p><p>no branching rhymes, is fully predicted by the model, even though it is</p><p>precisely what Kaye and Lowenstamm (1981) found to be impossible.16</p><p>16 Kaye and Lowenstamm (1981) discuss some apparent cases corresponding to the</p><p>settings in (15d) and dismiss them. The following chapter offers an extensive ana-</p><p>lysis of similar systems in the history of Slavic. See also Cyran (2001) for a discus-</p><p>sion of a similar problem in Malayalam.</p><p>90 Formal complexity</p><p>The problem lies in the nature of the parameters in general, or rather in</p><p>their independent status. Since each parameter is set separately, the only</p><p>way to preclude (15d) above is to resort to the arbitrary designation of such</p><p>settings as marked or downright impossible. This would be a highly unsat-</p><p>isfactory move, because there would be nothing in theory to prevent us</p><p>from imposing similar constraints on the correct settings in (15a-c).</p><p>A similar problem of arbitrariness besets models of phonological de-</p><p>scription which employ ranked constraints to derive the typology of sylla-</p><p>ble structure. In Optimality Theory, the relevant constraints responsible for</p><p>the relation between branching onsets and rhyme-onset sequences, that is,</p><p>internal codas, are *COMPLEX ONSET and *CODA respectively. While it is</p><p>difficult to see how the two constraints could interact with each other, the</p><p>tendency to avoid complex onsets in the absence of codas would require</p><p>that *COMPLEX ONSET be inherently ranked higher than *CODA with re-</p><p>spect to Faithfulness constraints, or that *COMPLEX ONSET be undomi-</p><p>nated whenever *CODA is too. However, the reverse ranking, or the reverse</p><p>implication must be somehow precluded. That is, if *COMPLEX ONSET is</p><p>undominated, *CODA must be too. In this respect, constraint ranking faces</p><p>the same problem as the parameter system of GP, because there is nothing</p><p>inherent in the model that would express this implicational relationship.</p><p>A more serious problem for standard GP is that as long as parameters</p><p>for the government licensing properties of nuclei and parameters for</p><p>branching constituents are allowed to coexist in the model, we cannot ex-</p><p>clude conflicts between these two types of parameters. For example, we</p><p>must assume that the presence of branching onsets is due to two theoreti-</p><p>cally unconnected parameters – one which allows onsets to branch, and</p><p>refers to the structure of the constituent, and another, which defines the</p><p>licensing properties of the nuclei in a given language.</p><p>(16)</p><p>parameters a. b. c. d.</p><p>branch onset ON OFF OFF ON</p><p>indir. gov. lic. OFF ON OFF ON</p><p>??? ??? Zulu Polish</p><p>What (16) illustrates is that it is not clear what the possible conflicting</p><p>settings of the two parameters would yield. They must be assumed, there-</p><p>fore, to be switched ON or OFF in conjunction to account for the observable</p><p>facts, which suggests that either the two parameters require additional jus-</p><p>Syllable markedness as a scale of formal complexity 91</p><p>tification to be maintained in the grammar, or some external mechanism</p><p>must be evoked to link them. The same applies to the interaction between</p><p>the parameter for branching rhymes and direct government licensing. Since</p><p>the two disparate types of parameters must be switched in conjunction, the</p><p>problem signalled in (15d) remains unsolved. Below, we will pursue yet</p><p>another option which consists in modifying the approach to parameters in a</p><p>dramatic way, though the model of Government Phonology will be changed</p><p>only slightly.</p><p>3.3. Syllabic complexity is scalar</p><p>Since syllabification in GP is indeed a reflection of governing and licens-</p><p>ing relations, let us assume that we can do without parameters for branch-</p><p>ing constituents and derive the syllable typology only by reference to the</p><p>licensing properties of nuclei. The latter will not be defined in terms of</p><p>separate parameters but rather as a scale on which the cut-off points are</p><p>defined by the complexity of the syllabic configuration to be licensed.17</p><p>As mentioned earlier, the primary function of nuclei in phonological</p><p>strings is to license their onsets. These onsets, however, may find them-</p><p>selves in different configurations and each configuration requires different</p><p>degrees of licensing strength from the following nucleus. Given the two</p><p>types of governing relations between consonants discussed in an earlier</p><p>section, we appear to have three possible structural configurations, or, to</p><p>put it differently, there are three levels of formal complexity, each of which</p><p>puts different demands on the licenser, that is, the nucleus. These struc-</p><p>tures are repeated below.</p><p>(17)</p><p>a. Simple b. Direct Government c. Indirect Government</p><p>Licensing Licensing Licensing</p><p>C N C C N C C N</p><p>| | | | |</p><p>(C) R T T R</p><p>17 Note that the elimination of parameters on branching constituents from the model</p><p>does not affect such fundamental notions as, for example, the binary theorem. The</p><p>maximally binary nature of constituents is guaranteed by the way governing rela-</p><p>tions are contracted and need not be doubly secured.</p><p>92 Formal complexity</p><p>In (17), we illustrate the formal differences between particular configura-</p><p>tions of onset licensing. Thus, (17a) represents the simplest arrangement,</p><p>where a nucleus licenses a simplex onset of any substantive make-up what-</p><p>soever. It may be any consonant which is present in a given linguistic sys-</p><p>tem, be it a sonorant or an obstruent.18 It may also be an empty onset, if the</p><p>language-particular settings allow for it. The structures (17b) and (17c) are</p><p>formally more complex because the onset, which receives licensing from</p><p>its nucleus, is itself in a relation with another consonant.</p><p>It is clear that the latter two structures are more demanding in terms of</p><p>licensing than (17a), which explains the unmarked status of CV syllables.</p><p>On the other hand, to distinguish between the licensing demands imposed</p><p>by (17b) and (17c) on the nucleus, we will assume after Charette (1990)</p><p>that the relevant distinction derives from the fact that in (17b) the nucleus</p><p>is directly adjacent to the governor and therefore this structure is formally</p><p>easier to license than (17c), in which the onset head is separated from the</p><p>nucleus by the complement of the governing relation. Since syllabification</p><p>is now viewed as the interaction between formal complexity and the licens-</p><p>ing strength of the nuclei which sanction such structures, the relative dis-</p><p>tance between the licenser and the licensee should rightly play a role in the</p><p>relation. This model, therefore, predicts that the opposite placement of the</p><p>relevant structures, that is, one in which branching onsets would be simpler</p><p>structures than coda-onset clusters, should be theoretically impossible.</p><p>This formal difference should alone suffice to establish the relative mark-</p><p>edness of the structures in (17). Note that the syllabic complexity scale,</p><p>which is derived from government and licensing, directly corresponds to the</p><p>levels of markedness proposed by Kaye and Lowenstamm (1981) and dis-</p><p>cussed under (11). An extended version of the picture in (17) is given below.</p><p>(18) Syllabic complexity scale</p><p>I II III</p><p>O N</p><p>x x</p><p>T/R a</p><p>R O N</p><p>N</p><p>|</p><p>x x x x</p><p>a R T a</p><p>O N</p><p>x x x</p><p>T R a</p><p>CV CV, CVC CV, CVC, CCV</p><p>18 Hence, we use the symbol C rather than R or T.</p><p>Syllable markedness as a scale of formal complexity 93</p><p>The common formal denominator in establishing the complexity scale is</p><p>the fact that in each instance there is a licensing relation between a nucleus</p><p>and the preceding onset. The growing licensing demand at particular levels</p><p>depends strictly on the function of the onset, that is, whether it is simplex</p><p>or whether it is a governor. In the latter case it is the direction of govern-</p><p>ment that determines the formal difference in the complexity of levels II</p><p>and III.19 Thus the markedness levels above appear to act like stable regions</p><p>in syllable complexity, where the increasing complexity of consonantal con-</p><p>figurations directly corresponds to the growing demand on the nuclei which</p><p>are called on to license these formal structures. We assume, then, that the</p><p>crucial factor in systemic decisions as to how much syllabic structure is to be</p><p>allowed can be reduced to one theoretical aspect of phonological organiza-</p><p>tion: the licensing properties of nuclei, or better, their licensing strength.</p><p>Linguistic variation in this model consists in languages choosing arbi-</p><p>trarily how much complexity their nuclei will license along the nonarbi-</p><p>trary complexity scale, as illustrated in (19) below.20</p><p>(19) Licensing strength of nuclei</p><p>structure example example language</p><p>I Ca baby Desano</p><p>II R.Ta winter Japanese</p><p>III TRa trap English</p><p>C = any consonant, T = governor, R = governee, a = any full vowel</p><p>Either of the three choices (I–II–III) is available, but the scale itself is by no</p><p>means arbitrary. The three steps, or ‘quantal regions’, to borrow a term from</p><p>19 Though the relative complexity of these structures is implicit in the terminology</p><p>proposed by Charette, that is, direct vs. indirect GL, one may think of quite a few</p><p>arguments supporting the ranking in (19) and very few reasons to contradict it. For</p><p>example, it is characteristic of (true) branching onsets that they are much more</p><p>constrained melodically than coda-onset sequences, which could be taken to be a</p><p>reflection of their more costly nature in terms of licensing. Thus, sufficient sonority</p><p>distance in branching onsets is nothing else than making the governing relation</p><p>‘easier’ for the indirect licenser, where ease is defined as relative to the steepness of</p><p>the complexity slope.</p><p>20 At this stage, the term nucleus is tantamount to an unreduced vowel. In the fol-</p><p>lowing sections a finer distinction is made between different types of nuclei.</p><p>94 Formal complexity</p><p>phonetic theory, along the scale of syllabic complexity are non-reversible</p><p>or re-rankable.</p><p>The above table recapitulates the hierarchy proposed in Kaye and</p><p>Lowenstamm (1981) and solves the problem of the formal expression of</p><p>the markedness values for branching onsets and branching rhymes. They</p><p>are not equally marked. The branching onset is formally more costly.</p><p>The fixed nature of the complexity scale – allowing for easy falsifica-</p><p>tion – is not its only advantage. The simplex onset in CV syllables is the</p><p>least marked because this is where the scale begins and thus it plays the</p><p>role of a crucial reference point. The scale also offers a fresh look at the</p><p>concept of markedness itself. More complex structures need not be viewed</p><p>as violations of any universal conditions or constraints, but rather, as the</p><p>utilization of all logically possible structural configurations, some of which</p><p>happen to be more costly to license than others. In this respect, complexity</p><p>and markedness are synonymous terms.21 Additionally, the model of Gov-</p><p>ernment Phonology imposes limits on the structural possibilities them-</p><p>selves. These follow from the nature of government. Since governing rela-</p><p>tions are contracted between two agents, the constituents formed in this</p><p>way may be maximally binary, that is, may occupy maximally two posi-</p><p>tions, e.g. a branching onset.22</p><p>One should mention a few consequences of the model presented above,</p><p>which will be taken up in chapter 3. One of them concerns the fixed nature</p><p>of the complexity scale. It is very easy to falsify the proposal, in that the</p><p>existence of languages which possess branching onsets (level III), but lack</p><p>branching rhymes, that is, codas (level II) should be ruled out. This is be-</p><p>cause, nuclei that can license the most complex structures are predicted to</p><p>license the less complex / marked ones.23</p><p>Another problem concerns the status of branching rhymes. In standard</p><p>GP, this structure involves a governing relation between the head, that is,</p><p>the nucleus, and the rhymal complement which is at the same time gov-</p><p>erned by the following onset. In the model presented here, the crucial aspect</p><p>of what has hitherto been considered to be a branching rhyme is shifted to the</p><p>governing relation between the consonants. The consequences of this move</p><p>are far-reaching. First of all, the status of the branching rhyme is now un-</p><p>21 Recall that a similar relationship between complexity and markedness is ob-</p><p>served at the sub-segmental level of representation (see chapter 1).</p><p>22 Cases where government is not contracted between two consonants, as well as</p><p>consonant sequences exceeding binarity will be discussed shortly.</p><p>23 This problem is referred to as ‘skipped steps’ in Cairns and Feinstein (1982).</p><p>Syllable markedness as a scale of formal complexity 95</p><p>dermined, and the phonological phenomena typically ascribed to this struc-</p><p>ture, for example, closed syllable shortening and stress attraction in Eng-</p><p>lish, will have to be captured in a different way. More importantly, we</p><p>predict that whether a given system has internal codas depends on the li-</p><p>censing strength of the nucleus in the following syllable, thus undermining</p><p>the status of the syllable itself as a linguistically valid unit.24</p><p>We may illustrate the shift in focus by the following diagram. The dot-</p><p>ted area illustrates the traditional way of looking at syllabic constituents.</p><p>This perspective required reference to parameters on branching constitu-</p><p>ents. The solid-line area marks the domains of interaction that transpire</p><p>from our discussion, which allow for a scalar understanding of syllable</p><p>markedness.</p><p>(20) O R O N</p><p>|</p><p>N</p><p>|</p><p>t r E n d I</p><p>T R V R T V</p><p>The consequences of this move will be discussed in the remainder of this</p><p>book.25 The syllable typology given in (19) above deals with variation in</p><p>the licensing strength of nuclei across languages. As such, it must be treated</p><p>as an observation-based proposal which requires further substantiation. For</p><p>this purpose, we will now consider the question whether nuclei may have</p><p>differing licensing strengths within a single phonological system.</p><p>24 Government Phonology has always claimed that there is no such prosodic unit as</p><p>the syllable (Kaye, Lowenstamm and Vergnaud 1990). Takahashi (1993) and</p><p>Scheer (2004) claim that governing relations can effectively replace any notion of</p><p>syllabic constituency, a position which is supported by this discussion. For other pro-</p><p>posals denying the existence of the syllable see e.g. Dziubalska-Kołaczyk (1995,</p><p>2002) and the references therein.</p><p>25 A similar shift from arboreal structure to lateral relations in phonological repre-</p><p>sentation, although in a slightly different model, can be found in e.g. Scheer (2004).</p><p>96 Formal complexity</p><p>4. The licensing properties of different nuclear types</p><p>4.1. The schwa vowel in Dutch</p><p>So far we have seen that nuclei containing a full vowel exhibit different li-</p><p>censing properties across languages. These properties were gauged against</p><p>the complexity of the syllabic configurations that demanded the licensing.</p><p>Syllabification, therefore, appears to result from a tug of war between</p><p>the</p><p>relative structural complexity of onset configurations and the licensing</p><p>strength of nuclei. In this section, we will further extend the model by</p><p>looking at different types of nuclei in order to see if within a single lan-</p><p>guage they may also exhibit differing licensing properties.</p><p>We know that vowels may differ in quality and quantity, and it would</p><p>be prudent to see if these distinctions have any bearing on their licensing</p><p>properties. If licensing strength is taken seriously, it predicts that weaker</p><p>vowels can license less, not more. The question, of course, is what is a</p><p>weak or weaker vowel. We will first concentrate on the difference between</p><p>full vowels (unreduced) and reduced ones, and then go one step further.</p><p>English possesses the relevant distinction, as most of its unstressed</p><p>vowels are reduced to the so called schwa [´]. However, as the words vul-</p><p>gar [vøl.g´] and cobra [k´U.br´] suggest, there is no difference in the li-</p><p>censing abilities of full and reduced vowels in English. For our purposes, it</p><p>would be worrying if schwa licensed more than a full vowel. We therefore</p><p>provisionally assume that the two kinds of vowels may have similar licens-</p><p>ing properties in this language.26</p><p>In order to neatly illustrate the differing licensing abilities of nuclei we</p><p>will look at restrictions on consonantal clusters and following vowels in</p><p>Dutch.27 Among the many characteristics of the schwa vowel in Dutch, the</p><p>one which is most interesting for us is its constrained distribution with</p><p>respect to preceding clusters. Kager (1989: 212) notes that pre-schwa clus-</p><p>ters in Dutch behave as if they were word-final. In other words, schwa</p><p>behaves as if it was a word boundary rather than a nucleus which is able to</p><p>26 There are facts in English phonology, described, e.g. in Gussmann (1998), which</p><p>seem to suggest that in some contexts schwa is banned and only a full vowel will</p><p>do. The phenomenon concerns the absence of sequences such as *...´mp, *...´Nk,</p><p>etc. in this language. Although, these facts are closely connected with our proposal,</p><p>they will not be discussed until chapter 3.</p><p>27 The discussion is based on Booij (1995), Kager (1989), Kager and Zonneveld</p><p>(1986), Trommelen (1984), and van Oostendorp (1995, 2000).</p><p>The licensing properties of different nuclear types 97</p><p>construct its own syllable. We will look at both rising and falling sonority</p><p>clusters in pre-schwa position as they seem to behave in a way which sug-</p><p>gests that the effects are not at all accidental. First, we take clusters of in-</p><p>creasing sonority, that is, branching onsets (TR). Such clusters are said to</p><p>occur only before full vowels, as the data taken from Kager (1989: 213)</p><p>illustrate.</p><p>(21) a. b. c.</p><p>*[katr] *[ka.tr´l] [ka.trOl] katrol ‘pulley’</p><p>*[dypl] *[dy.pl´] [dy.plo] duplo ‘duplicate’</p><p>There are no word-final clusters of rising sonority (21a), or before a schwa</p><p>vowel (21b). Branching onsets in Dutch require a full vowel to follow as</p><p>shown in (21c). In terms of the model of licensing we have introduced so</p><p>far the difference between (21b) and (21c) may be captured by referring to</p><p>the weaker status of schwa as a licenser. To put it differently, the govern-</p><p>ing relation from left to right which is present in branching onsets can only</p><p>be licensed by a full vowel. As yet, we have little to say about the illicit</p><p>forms in (21a) and why they are excluded.</p><p>Two more comments must be made about the data in (21). Firstly, al-</p><p>though so far we have not discussed the phonological nature of word-final</p><p>clusters such as those in (21a), it appears to be quite unusual for a word-</p><p>final cluster to be compared to pre-schwa clusters which, as most linguists</p><p>will agree, in most languages constitute an onset. Secondly, it is not true</p><p>that (21a) and (21b) are equally unacceptable. While there are indeed no</p><p>word-final clusters with rising sonority in Dutch, one can find a few inter-</p><p>esting exceptions to the pre-schwa context. First of all, there is a well-</p><p>defined group of words, mostly of Greek origin, where clusters of rising</p><p>sonority do occur before a schwa, although, admittedly, these clusters do</p><p>not look like well-formed branching onsets, e.g. Dafne [dafn´] (Kager</p><p>1989: 213). Secondly, well-formed branching onsets are found in pretonic</p><p>position in words like: fregat [fr´gat] ‘frigate’, brevet [br´vEt] ‘patent’. So,</p><p>in fact we are dealing here with a sort of gradation of acceptability of clus-</p><p>ters in the three contexts in (21); from absolutely excluded, through re-</p><p>stricted, to fully acceptable. This scale is presented below in a symbolic</p><p>way. This will facilitate the comparison of the restrictions holding in clusters</p><p>of rising sonority with those of falling sonority to be presented below.28</p><p>28 ‘a’ stands for a full vowel</p><p>98 Formal complexity</p><p>(22) *TR#</p><p>below, that the word-final context (_#)</p><p>is not exceptional in the treatment of consonantal restrictions in Dutch in</p><p>that it forms an integral part of the gradation of restrictions. As it stands,</p><p>the Sonority Sequencing Generalization provides no platform for compari-</p><p>sons between the word-final, pre-schwa, and pre-full vowel contexts, and</p><p>the hierarchy (_a>_´>_#) makes very little theoretical sense.33 The re-</p><p>ranked scales *(_a>_#>_´) or *(_#>_a>_´) can only be excluded on obser-</p><p>vational and not on theoretical grounds. To see this better, let us look at RT</p><p>clusters in Dutch where neither homorganicity nor dentality of the obstru-</p><p>ent is involved.</p><p>Clusters of a liquid and a non-dental consonant are subject to schwa ep-</p><p>enthesis in two contexts: at the end of the word (syllable) and before a</p><p>schwa (Kager 1989: 214). Thus, once again the pre-schwa situation is iden-</p><p>tified with the end of the word. However, the status of the epenthesis in the</p><p>two contexts is not identical. While epenthesis is almost obligatory in word-</p><p>final context (25a), it is only optional in pre-schwa position (25b), and it is</p><p>excluded in contexts preceding a full vowel (25c).</p><p>31 See section 3.2. in chapter 1.</p><p>32 It must be said that there are problems with the description of exceptional struc-</p><p>tures in syllabic analyses, and in effect, resorting to such contingencies as extrasyl-</p><p>labicity or appendices, is a direct consequence of operating with syllabic constituents</p><p>to establish syllable templates for a given system. In such approaches, exceptions</p><p>ruin the otherwise clear-cut picture.</p><p>33 The hierarchy (_a>_´>_#) simply says that the pre-full vowel context is better for</p><p>clusters than pre-schwa, and pre-schwa is better than word-final.</p><p>100 Formal complexity</p><p>(25) a. ´-epenthesis obligatory (RT# R´T#)</p><p>[har´p] harp ‘harp’</p><p>[kEr´k] kerk ‘church’</p><p>[bal´k] blak ‘beam’</p><p>[hEl´m] helm ‘helmet’</p><p>b. ´-epenthesis optional (RT´ R(´).T´)</p><p>[kar(´).p´r] karper ‘carp’</p><p>[kEr(´).k´r] kerker ‘dungeon’</p><p>[stal(´).k´r] Stalker ‘Stalker’</p><p>[hEl(´).m´r] Helmer ‘first name’</p><p>c. ´-epenthesis excluded (RTa R.Ta)</p><p>[har.pun] harpoen ‘harpoon’</p><p>[kar.kas] karkas ‘carcass’</p><p>[bal.kan] Balkan ‘Balkan’</p><p>[hEl.ma] Helma ‘first name’</p><p>What we observe in (25) is a gradation of RT integrity depending on what</p><p>follows the cluster, which is reminiscent of the restrictions on TR clusters</p><p>depicted in (22). Compare the two scales of contextual strength below.</p><p>(26) a. b. c.</p><p>okTRa > */okTR´ > *TR#</p><p>okR.Ta > ok/´-epenR.T´ > ´-epen/okRT#</p><p>The similarity in the distribution of TRs and RTs lies in the fact that in both</p><p>cases we are dealing with the same scale of contexts (_a>_´>_#). The cru-</p><p>cial difference is that in each respective context RT fares better than TR,</p><p>which we mark by shifting the RT scale of integrity slightly to the left.34</p><p>These effects are fully predicted in our model, as TR is formally more de-</p><p>manding. Thus, to use our terminology, a full vowel can license both RT</p><p>and TR clusters (26a). Both direct and indirect government licensing ob-</p><p>tains in the presence of this strong licenser, hence, Dutch is said to have</p><p>both branching rhymes and branching onsets, or, to put it in constituent-</p><p>34 Although before an unreduced vowel (_a) the full set of TR and RT clusters can be</p><p>present, the slanted line is used to express the fact that the two types of clusters will</p><p>still display different degrees of melodic freedom.</p><p>The licensing properties of different nuclear types 101</p><p>neutral terms, Dutch full vowels license both leftward (R←T) and right-</p><p>ward (T→R) governing relations. The schwa is much weaker as a licenser.</p><p>It can only license TR in pre-tonic position, while its RT clusters often un-</p><p>dergo optional epenthesis. What is interesting is that while epenthesis is</p><p>excluded in RT followed by a full vowel, it is also excluded before a schwa</p><p>if the cluster is a nasal+homorganic obstruent or liquid+dental, e.g. culte</p><p>[koel.t´] ‘cult’. Recall that these clusters are also found in word-final con-</p><p>texts.</p><p>The scales in (26) provide a general picture of the gradation of contexts</p><p>with respect to the licensing of the two types of consonant clusters. Now,</p><p>each of these individual situations merits a discussion with respect to the</p><p>observed effects. Here we focus only on the optional epenthesis in the pre-</p><p>schwa RT clusters. The analysis of this phenomenon within our model</p><p>hinges on two aspects of Dutch phonology. Firstly, we must determine</p><p>what the licensing strength of the schwa vowel is, and propose some ac-</p><p>count for the optionality of the epenthesis. Secondly, to account for the</p><p>clusters which do not get broken up by epenthesis, we must propose some</p><p>way to deal with exceptional strings.</p><p>The mechanism of epenthesis itself receives a fairly straightforward ac-</p><p>count within the licensing model. All we need to say is that in Dutch, the</p><p>licensing strength of schwa is such that it can barely license level II of</p><p>structural complexity. We use the word barely because schwa is able to</p><p>license partial geminates, which we will assume to be the easiest RT clus-</p><p>ters to license at level II. However, in words such as karper [kar(´)p´r]</p><p>‘carp’ (25b), where the cluster is of the ‘heavier’ type, optionality of the</p><p>effects are predicted. Either the licensing potential of the nucleus is suffi-</p><p>cient to license the governing relation, or it is not. For this reason, the clus-</p><p>ter may be broken up by an epenthetic vowel, or not.35 Note that the failure</p><p>of schwa to license rp de facto leads to a situation where a branching</p><p>rhyme in the preceding syllable is impossible, which is exactly what the</p><p>model predicts. Thus, we seem to have found some empirical support for</p><p>35 We bypass the question of resyllabification as a result of epenthesis. Since all</p><p>governing and licensing relations are contracted in the phonological representation,</p><p>we may assume that the difference between epenthesized forms and those which</p><p>retain the cluster ([karp´r]) lies in the different representations. Another option</p><p>which can be pursued is to assume a CVCV model of phonological representation</p><p>where all that happens phonologically is that a nucleus is filled with a melody,</p><p>while the syllabic structure remains the same. See section 6 for more discussion.</p><p>102 Formal complexity</p><p>the assumption that a branching rhyme is determined by the nucleus in the</p><p>following syllable, as it were.</p><p>(27) a. b.</p><p>O N O N O N</p><p>| | | | | |</p><p>k a r p ´ r k a r ´ p ´ r</p><p>In (27a), we see that the nucleus is unable to directly government license</p><p>its onset. Hence, the governing relation, and thereby, a coda-onset contact</p><p>is impossible. Epenthesis is a strategy providing a licenser for the liquid</p><p>(27b).36 As for the optionality of epenthesis we are forced to say that,</p><p>within a particular level of structural complexity the licensing abilities of</p><p>nuclei will vary, from speaker to speaker and also between registers.37 Reg-</p><p>ister difference in this model is viewed as manipulation, conscious or not,</p><p>of the licensing strength of the nuclei. The schwa in Dutch is able to li-</p><p>cense some leftward governing relations, e.g. culte [koelt´] ‘cult’, thus its</p><p>licensing strength reaches level II of structural complexity and hence, fluc-</p><p>tuations within this level are rather unsurprising. Note that such manipula-</p><p>tion of the licensing properties of nuclei must be viewed as an abstract</p><p>phenomenon, because we are not dealing with a stronger articulatory effort</p><p>− a schwa is a reduced vowel and remains so. It is its licensing properties</p><p>that are up- or down-graded depending on the register. Let us now expand</p><p>the idea of easy and difficult governing relations.</p><p>4.2. Light and</p><p>heavy clusters</p><p>We will not go into much detail concerning the distinction between light</p><p>and heavy clusters here, terms which we find quite appropriate for a model</p><p>operating with the licensing strength of nuclei. It is clear that such distinc-</p><p>tions must exist to account for the exceptional pre-schwa and word-final RT</p><p>clusters in Dutch. At this stage we may offer the following criterion, which</p><p>36 Recently, a different strategy is also used in Dutch, namely, instead of epenthesis,</p><p>the coda liquid is vocalised (van Oostendorp p.c.). In other words, instead of provi-</p><p>ding a stronger licenser for the coda consonant, the consonant itself is weakened.</p><p>This situation is expected as one of strategies of resolving coda-onset clusters.</p><p>37 A connection between epenthesis and register has been noted for languages other</p><p>than Dutch. See e.g. van Oostendorp (1995), Mohanan (1986).</p><p>The licensing properties of different nuclear types 103</p><p>was briefly mentioned in the previous chapter when we discussed the dif-</p><p>ferences between English and Polish fr/vr, and the complexity-based condi-</p><p>tions on Irish epenthesis. Since two consonants contract a governing rela-</p><p>tion on the basis of their complexity differential, it may be assumed that in</p><p>clusters where the differential is greater, government will obviously be</p><p>easier to contract. Such clusters will also be easier to license than clusters</p><p>with near equal complexity. In this respect, geminates (28a) and partial</p><p>geminates (28b) are the easiest RT clusters to license because the comple-</p><p>ment of the governing relation has little melodic content or none.</p><p>(28) ease of licensing</p><p>light heavy</p><p>a. geminate b. partial geminate c. ordinary RT cluster</p><p>R T N R T N R T N</p><p>.</p><p>TR completely, it does allow for a restricted set of RT clusters. In</p><p>other words, the context (_#) behaves very much like other licensers except</p><p>that it is consistently the weakest in the hierarchy.</p><p>Given the gradation system shown in (26) above, we may reverse the</p><p>initial observation of Kager (1989) and claim that it is not that schwa be-</p><p>haves like a word boundary, but that the word boundary (#) behaves very</p><p>much like a nucleus. Let us then assume that # is in fact a nucleus, except</p><p>38 Similar criticisms may be applied to sonority sequencing, and such concepts as</p><p>sonority distance or degrees of sonority steepness.</p><p>39 In this respect the recent development within Natural Phonology called the Beats</p><p>and Binding Theory (Dziubalska-Kołaczyk 1995, 2002) is a very similar proposal.</p><p>40 Although the difference between a full vowel and a schwa is clearly melodic, we</p><p>must not forget that to a great degree this difference is connected with the prosodic</p><p>position of vowels. Schwas are reduced, unstressed vowels, thus referring to mel-</p><p>ody as the distinguishing factor may be insufficient and we should rather refer to</p><p>the weak licensing characteristics of prosodically weak positions. See chapter 3.</p><p>106 Formal complexity</p><p>that it is melodically empty.41 We are not introducing anything new within</p><p>the model of Government Phonology, in which it has always been claimed</p><p>that final consonants are not codas but onsets followed by an empty nu-</p><p>cleus. What is new here is that the model which draws on syllabic com-</p><p>plexity and licensing strength provides additional support for a view which</p><p>has been put forward by other authors. Only by assuming that words end-</p><p>ing with consonants on the surface structurally end with an empty nucleus,</p><p>are we able to compare the word-final context with the pre-schwa and pre-</p><p>full vowel situations in a coherent and meaningful way. Thus the observa-</p><p>tion that schwa in Dutch sometimes behaves like a word boundary was not</p><p>entirely incorrect. Only now, we can define better what the alleged bound-</p><p>ary is.42 The gradation of contexts in (26) is in fact a hierarchy of licensers.</p><p>(29) Scale of licensers</p><p>N N N</p><p>| | |</p><p>a > ´ > P</p><p>This assumption, which will be further illustrated in the following section,</p><p>solves two problems. Firstly, TR and RT clusters before a final empty nu-</p><p>cleus are now formally identical to the same clusters in pre-schwa and pre-</p><p>full vowel contexts. TRs are always branching Onsets, as it were, and RTs</p><p>are Coda-Onset sequences in all three contexts. Secondly, the scale of con-</p><p>textual strength (_a > _´ > _ø) receives a non-arbitrary explanation now, in</p><p>that a full vowel licenses better than a prosodically weaker schwa. Both</p><p>schwa and the final empty nucleus are weak licensers, but schwa has mel-</p><p>ody and is therefore a better licenser than the empty nucleus. The theoreti-</p><p>cal difference between the contexts _# and _ø cannot be underestimated.</p><p>This is illustrated below.</p><p>41 See Kaye (1990) for more discussion concerning this proposal, as well as Guss-</p><p>mann and Kaye (1993), and Harris and Gussmann (1998) for a survey of convinc-</p><p>ing arguments against final codas and in favour of final empty nuclei.</p><p>42 In the following discussion we refer only to the licensing properties of empty</p><p>nuclei in word-final position.</p><p>The licensing properties of different nuclear types 107</p><p>(30) a. context</p><p>(traditional)</p><p>effect b. licensing scale</p><p>_a unmarked, no restrictions _a</p><p>_´ more marked, some restrictions _´</p><p>_# most marked, severe restrictions _ø</p><p>Clearly, the contexts understood in the traditional way as in (30a) do not</p><p>constitute a uniform set, and the placement of _# at the bottom of the mark-</p><p>edness hierarchy is arbitrary and based only on observation. On the other</p><p>hand, the scale of licensers in (30b) leaves no space for re-ranking. An</p><p>empty nucleus cannot license more than a schwa, and a schwa cannot li-</p><p>cense more than a full vowel. Thus, this model is easily falsifiable. As</p><p>regards the licensers, a discrepancy to the effect that more melodic material</p><p>(substantive complexity) or more syllabic structure (formal complexity)</p><p>can be found before weaker licensers than before stronger ones would be</p><p>potentially detrimental to this model. Likewise, given that branching onsets</p><p>are more marked than coda-onset contacts, a system with TR but no RT clus-</p><p>ters would also be problematic.43</p><p>4.4. The syllabic space</p><p>Given the three levels of formal complexity (I−II−III) and the three-way</p><p>scale of licensers (a–´–P), the following syllabic space and syllable mark-</p><p>edness can be proposed. C stands for any consonant, [a] is any full vowel.</p><p>RT is a coda-onset contact, which means that it is not a word-initial cluster.</p><p>TR is a branching onset in any position in a word.</p><p>(31) Syllabic space</p><p>[a] [´] [ø]</p><p>I C_ Ca ⊂ C´ ⊂ CP</p><p>∩ ∩ ∩</p><p>II RT_ RTa ⊂ RT´ ⊂ RTP</p><p>∩ ∩ ∩</p><p>III TR_ TRa ⊂ TR´ ⊂ TRP</p><p>43 See chapter 3 for a discussion of an apparent example of such a system, that is,</p><p>Common Slavic. The other example, Malayalam, is discussed in Cyran (2001).</p><p>108 Formal complexity</p><p>The syllabic space in this model is defined by the interaction between the</p><p>vertical vector of the structural complexity scale (I–II–III), where govern-</p><p>ment, its presence and type, plays the key role, and the horizontal vector of</p><p>the scale of licensers (a–´–P). Neither the levels of complexity nor the</p><p>types of licensers can be re-ranked, and the syllabic space as defined by</p><p>complexity and licensing is finite.44</p><p>From this scheme it follows that for any given licenser the same full ty-</p><p>pology of possible syllabic structures and the same markedness relation-</p><p>ships are available. That is, potentially, each type of nucleus can license a</p><p>single onset (CV), an onset governing a preceding coda (RT), and a branch-</p><p>ing onset (TR). The difference, of course, is that the melodic and structural</p><p>options will decrease as we move away from Ca, that is, a CV syllable</p><p>containing a simplex onset licensed by a full vowel.</p><p>Ca is the least marked syllable type because here the easiest structure is</p><p>licensed by the strongest licenser. Thus we do not need any separate con-</p><p>straints or principles to derive this fact. The ‘unmarked’ syllable type</p><p>emerges from the basic theoretical assumptions on syllabification and not</p><p>from a set of extraneous principles or constraints. TRP, the word-final</p><p>branching onset, is on the other hand the most marked structure. Marked-</p><p>ness increases with the extension of one or both vectors, that is, (I–II–III)</p><p>and (a–´–P).</p><p>The vectors allow us to establish the implicational relationships be-</p><p>tween structures in a straightforward fashion. For example, the presence of</p><p>RT´ in a given system ensures the existence of C´, RTa and Ca by direct</p><p>implication or transitivity. On the other hand, the presence of TRP suggests</p><p>that all possible configurations shown in the syllabic space scheme in (31)</p><p>should be also present.</p><p>The integration of the empty nucleus in the licensing scale unifies struc-</p><p>tural licensing in that the typology and markedness of the right edge of</p><p>words may be given the same account as word-medial simplex onsets and</p><p>clusters.45 This includes the fact that word-medially the maximal number of</p><p>consonants in a cluster is typically three. Note that so far, there was no</p><p>mention of three-consonant clusters or bigger. However, the syllabic space,</p><p>as defined in (31) does in fact predict the existence of three-consonant</p><p>44 Note that the syllabic space does not include empty onsets, long vowels and</p><p>clusters of consonants that are not in a governing relations. Some of these will be</p><p>discussed in the following sections.</p><p>45 See Harris and Gussmann (1998)</p><p>who point to the similarity between intervocalic</p><p>and word-final phonotactics in English.</p><p>The licensing properties of different nuclear types 109</p><p>clusters. Recall that the structural complexity is defined by the presence of</p><p>government. The conditions on government in GP allow a relation between</p><p>only two consonants in a given direction. Government is bidirectional.</p><p>Therefore, if a governor T governs one complement to the right and one to</p><p>the left (R←T→R), then what we obtain is a licit ternary cluster, which,</p><p>however, should not be confused with a ternary syllabic constituent. An</p><p>example of such a configuration word-medially can be easily found in a</p><p>language like English. In fact, the very name of the language contains a</p><p>ternary cluster (/IN←g→lISP/).</p><p>A ternary cluster is also possible word-finally, although it is very rare.</p><p>In Polish, which has word-final branching onsets as in wiatr [vjatr]</p><p>(36) Licensing properties of nuclei in Polish</p><p>[a] [ø]</p><p>I C_</p><p>II RT_</p><p>III TR_</p><p>The obvious question that can be raised at this stage concerns the status of</p><p>the scale of licensers. Polish seems to exemplify a system which does not</p><p>47 One immediate advantage of this proposal is that inflection does not require re-</p><p>syllabification of any sort, but only provides a melody for the existing nucleus.</p><p>48 These facts are discussed in more detail in section 6 below.</p><p>The licensing properties of different nuclear types 113</p><p>use the full scale. Recall, that the scale of formal complexity precludes</p><p>skipped steps, in that the presence of branching onsets necessarily implies</p><p>the presence of coda-onset contacts. Why should the scale of licensers be</p><p>different? The answer is simple. The role of nuclei is to license structure,</p><p>be it substantive or formal. It is immaterial what phonological shape the</p><p>nuclei assume, as long as they are granted the necessary licensing power.</p><p>One may point to two criteria determining the shape of the scale of li-</p><p>censers. With respect to schwas, it is the presence of vowel reduction, or</p><p>the presence of lexical schwas. As far as empty nuclei are concerned, the</p><p>condition on their occurrence word-finally has been identified earlier and</p><p>boils down to granting a melodically empty nuclear position the ability to</p><p>license. At this stage, we predict the following linguistic systems in terms</p><p>of the types of licensers they employ.</p><p>(37) Licenser types</p><p>a. [a]</p><p>b. [a] – [´]</p><p>c. [a] – [P]</p><p>d. [a] – [´] – [P]</p><p>The above typology stems directly from the phonological representation. A</p><p>nucleus may either have melody, or not. Just like onsets. Nuclei, may addi-</p><p>tionally be melodically reduced to schwa.49 System (37a) does not have any</p><p>distinctions among licenser types. It has no word-final consonants and no</p><p>vowel reduction. (37b) illustrates a system which has schwas but no word-</p><p>final consonants. Polish represents system (37c). It has no reduced vowels,</p><p>but allows word-final empty nuclei to license consonants and clusters. The</p><p>full scale in (37d) has been discussed at length above. It is a matter of fur-</p><p>ther research to establish if it may be further expanded, for example, to</p><p>include distinctions in licensing strength between different melodies of full</p><p>vowels.</p><p>To summarize, linguistic variation concerning syllable structure stems</p><p>from the choice of the types of licensers in (37) and the settings defining</p><p>the strength of these licensers. The strength is in a sense gauged against the</p><p>49 At this stage we refer to the object called schwa without making its definition</p><p>very precise. An attempt to provide a clear functional definition will be made in the</p><p>following chapter.</p><p>114 Formal complexity</p><p>formal complexity scale, which itself is defined by the presence and type of</p><p>a governing relation between consonants.</p><p>The model of complexity scales and licensing also points to impossible</p><p>systems. Two restrictions must be mentioned that seem to hold in connec-</p><p>tion with the settings of licensing strength between two types of licensers</p><p>in a given system, for example, between full vowels and empty nuclei.</p><p>(38) Impossible systems</p><p>*a. [a] [P] *b. [a] [P]</p><p>I C_ ? I C_</p><p>II RT_ ? II RT_ ?</p><p>III TR_ III TR_</p><p>First, what we do not expect in this model are skipped steps, that is, a dis-</p><p>continuity of licensing potential of a given licenser, as illustrated in</p><p>(38a).50 A second restriction concerns the possibility that a language may</p><p>select higher licensing potential for its empty nuclei than for its full vowels</p><p>(38b). This excludes a number of impossible systems, for example, one in</p><p>which consonant clusters are found word-finally but not word medially or</p><p>initially. This restriction also excludes languages in which full vowels do</p><p>not license anything, i.e. systems with only an arbitrary repetition of onsets</p><p>and empty nuclei.</p><p>4.6. Complexity Scales and Licensing model – a first approximation</p><p>We have seen in the above sections how parameters known from standard</p><p>GP, such as those on branching constituents or government licensing, can</p><p>successfully be replaced by scales, which, by their very nature, account for</p><p>gradient phenomena such as markedness in a superior fashion. A coherent</p><p>model of Complexity Scales and Licensing (CSL) based on the interaction</p><p>between substantive and formal complexity scales and licensing strength of</p><p>nuclei can be achieved only if certain assumptions are made about the na-</p><p>ture of phonological representations. Crucial in this model is the structure</p><p>50 The typology of syllabic structures presented in e.g. Blevins (1995) generally</p><p>supports the tendency which we wish to capture here, but she does quote a couple</p><p>of languages which seem to have complex onsets but no codas, for example, Maza-</p><p>teco or Arabella. Such languages must be looked into. In chapter 3, we discuss a</p><p>similar problem concerning Common Slavic.</p><p>The licensing properties of different nuclear types 115</p><p>of segments, which are defined in terms of privative elements. Their num-</p><p>ber in a given segment provides the necessary complexity slopes required</p><p>for any two consonants to contract governing relations. The two types of</p><p>relations, that is, R←T (right-to left) and T→R (left-to-right), which must be</p><p>licensed by the following nucleus, display an asymmetry as regards the</p><p>licensing demand. Hence the formal complexity scale (C–RT–TR). Intersect-</p><p>ing the complexity regions is another scale of nuclear types ([a–´–P]), re-</p><p>flecting the gradation of relative licensing potential. The empty nucleus</p><p>plays a pivotal role in the hierarchy of licensers, but more importantly, its</p><p>presence in the model affords a fresh view on word-final consonants,</p><p>which may be regarded as onsets and integrated into the system of prefer-</p><p>ence scales in a straightforward fashion.</p><p>So far, in our discussion of the three types of licensers (a–´–P), we were</p><p>mostly concerned with the right edge of words. This was the only context</p><p>in which we saw the empty nucleus in action. Until further evidence is</p><p>found, we assume that full vowels will have identical licensing properties</p><p>in a given system, regardless of their position in the word. On the other</p><p>hand, in the case of the schwa vowel, we noted an interesting variation</p><p>concerning its licensing power in Dutch. Namely, the word-final schwa</p><p>could not license branching onsets (TRs), while a pretonic one could, e.g.</p><p>brevet [br´vEt] ‘patent’. This is not an entirely surprising fact.</p><p>In accordance with the Licensing Inheritance principle (Harris 1997),</p><p>the same types of nuclei may exhibit slightly different licensing properties</p><p>depending on their position in the licensing network within the word. In</p><p>this respect the licensing scales discussed in this book and Licensing In-</p><p>heritance are complementary aspects of the phonological representation.</p><p>We return to Licensing Inheritance in chapter 3 when we discuss the law of</p><p>open syllables in Slavic. In what follows, we look at the consequences of</p><p>utilizing empty nuclei in other contexts than the word-final one. We review</p><p>the conditions on the distribution of empty nuclei in standard GP and pro-</p><p>pose to shift the focus of the phonological apparatus from licensing of</p><p>empty nuclei to their own licensing properties, which would be more com-</p><p>patible with the tents of the CSL model.</p><p>116 Formal complexity</p><p>5. Sources of empty nuclei and licensing mechanisms in standard GP</p><p>5.1. Introduction</p><p>In the previous sections we introduced a new entity into the inventory of</p><p>phonological units, that is, an empty nucleus. The reasons for postulating</p><p>this object were based solely on the discussion of the licensing strength</p><p>scale. It followed from the system that such a structure must</p><p>exist if only to</p><p>be able to account for the uniformity and relationships between the con-</p><p>texts: pre-vocalic, pre-schwa, and word-final. This way, the relative mark-</p><p>edness gradation can be understood better than in other approaches which</p><p>identify the word-final context with the coda. Thus, the word-final context</p><p>is now fully incorporated into a coherent system of syllabification in which</p><p>the right edge is the most restricted because it is licensed by the weakest</p><p>possible licenser.</p><p>Empty positions play an important role in the theory of phonological</p><p>government (Kaye 1990, Charette 1991, Gussmann and Kaye 1993, Harris</p><p>and Gussmann 1998). Their presence is not only justified, but in fact, ex-</p><p>pected given the nature of phonological representation advocated not only</p><p>in Government Phonology, but also in any other framework which adopts</p><p>the three-dimensional model. It is true, however, that only GP treats empty</p><p>nuclei as an indispensable aspect of representation. One objection which is</p><p>typically levelled against empty nuclei, is that such a construct is too ab-</p><p>stract. This overlooks the fact that anything beyond the melody level in the</p><p>phonological representation is abstract. The skeleton is abstract, and so is</p><p>the syllable with its constituents. These separate levels have been proposed</p><p>and independently argued for as autonomous (Harris 1994). In this respect</p><p>three-dimensional phonology predicts the existence of melodically empty</p><p>onsets and nuclei, and if they are sufficiently argued for, they should be</p><p>accepted, just like any other abstract units of phonological analysis. We</p><p>will assume that both the filled and empty positions illustrated in (39) are</p><p>theoretically predicted.</p><p>(39) O O N N</p><p>| | | |</p><p>x x x x</p><p>| |</p><p>α α</p><p>β β</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 117</p><p>Another justification for using empty positions can be drawn from proc-</p><p>esses of melodic depletion such as the lenition of consonants (e.g. Lass</p><p>1984), and the historical shift from high vowels, through jers, to zero in</p><p>Slavic (e.g. Stieber 1973), which we can also treat as depletion of melody,</p><p>as shown in the previous chapter.</p><p>(40) a. opening lenition b. Slavic rise and fall of jers</p><p>[p] > [f] > [w] > [P] [u] > [ъ]</p><p>α α α [P]</p><p>β β [i] > [ь]</p><p>γ</p><p>Since the two phenomena have been discussed at length in chapter 1, they</p><p>will not be given an airing here.51 It is worth mentioning however, that while</p><p>empty consonantal positions are widely accepted in phonological theory,</p><p>abstract vowels which can be to some extent equated with empty nuclei, also</p><p>have their own history in the literature. Abstract vowels in place of lost jers</p><p>have been proposed to account for vowel – zero alternations in the phonol-</p><p>ogy of Polish in, for example, Gussmann (1980), Rubach (1984), and Szpyra</p><p>(1992), to name but a few proposals.52</p><p>In an attempt to keep the model as constrained as possible, it is gener-</p><p>ally assumed in standard Government Phonology that the distribution of</p><p>empty positions, once we accept them, must be subject to certain restric-</p><p>tions. Thus, not only does the very occurrence of empty positions derive</p><p>from the nature of the phonological representation involving government</p><p>and licensing, but in the phonological representation itself one may seek to</p><p>discover the mechanisms which would license or justify such positions.</p><p>The interaction between the source of the occurrence, and the source of the</p><p>licensing of empty positions appears to underline their distribution, that is,</p><p>where they occur, and whether they remain empty or must surface melodi-</p><p>cally. Below, a summary is given in a tabular form of the contexts in which</p><p>empty nuclei may occur in phonological representation as well as of the</p><p>mechanisms which license them. Although most of the examples will come</p><p>from Polish this is not meant to be a full analysis of Polish phonotactics.</p><p>51 The problem of historical jers is also discussed in chapter 3.</p><p>52 See Scheer (2004, 2006) for an elaborate discussion of the connection between</p><p>these proposals and the status of empty nuclei in GP.</p><p>118 Formal complexity</p><p>In general, one may speak of a sort of assumed equilibrium between the</p><p>sources of the presence of empty nuclei and the licensing mechanisms</p><p>which make sure that such empty positions remain silent. In the absence of</p><p>licensing, the nucleus must be phonetically realized.</p><p>(41)</p><p>context source (due to) licensed by</p><p>word-initial</p><p># Ps+C...</p><p>governing relations</p><p>(s + C = interconstituent</p><p>government</p><p>parameter (‘magic’)</p><p>word-medial</p><p>...CPC...</p><p>governing relations (kPto)</p><p>grammar (parameter on</p><p>Branching: OFF)</p><p>lexicon (v ~ P)</p><p>Proper Government</p><p>Interonset Government</p><p>word-final</p><p>...CP #</p><p>governing relations</p><p>(coda licensing)</p><p>domains (...P]...P])</p><p>parameter (domain-final)</p><p>The contexts provided above suggest that empty nuclei may in fact occur in</p><p>all possible positions within the word. However the licensing mechanisms</p><p>for dealing with these instances differ depending on the context. Let us</p><p>begin the discussion by defining the way governing relations introduce</p><p>empty nuclei into representations. This source of their existence appears in</p><p>all three contexts.</p><p>5.2. Governing relations and empty nuclei</p><p>To see how government enforces the presence of empty nuclei let us first</p><p>recall the basic conditions underlying this relation.</p><p>(42) Conditions on government</p><p>a. melodic complexity profiles (in which the governor, symbolized as (T), is</p><p>melodically more complex than the governee (R).</p><p>b. adjacency (the two consonants must be adjacent in the relevant sense).</p><p>c. licensing (governing relations, just as simplex segments, require licens-</p><p>ing from the nucleus following such a segment or relation).</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 119</p><p>Condition (42a) refers to the necessary complexity differential between the</p><p>governor and the governee, and expresses more or less the same principle</p><p>as the Sonority Sequencing Generalization. Hence, government is respon-</p><p>sible for phonotactics. Adjacency, (42b), is defined in standard GP at the</p><p>level of skeletal positions. This effectively means that two consonants</p><p>separated by an empty nucleus are not adjacent and therefore they cannot</p><p>contract a governing relation. This is because an empty nucleus has a skeletal</p><p>point. It just lacks a melody linked to it. The third condition, (42c), is self</p><p>evident.</p><p>If all the above conditions are fulfilled, government between two con-</p><p>sonants must be contracted. On the other hand, a failure of one of these</p><p>conditions entails a failure of government and the two consonants must be</p><p>separated by an empty nucleus. The resulting structure is a bogus or false</p><p>cluster (43b).</p><p>(43) a. Government failure b. False cluster</p><p>C C V → C P C V</p><p>The false clusters are also subject to conditions. Note that the onset fol-</p><p>lowed by the empty nucleus must be licensed by that nucleus. We predict</p><p>that false clusters may occur only in those systems which grant licensing</p><p>properties to such empty nuclei. Let us look at concrete examples from</p><p>Polish.</p><p>(44) a. b. c.</p><p>O N O N O N O N O N O N O N</p><p>| | | | | | | | ↑ |</p><p>t k a t°Ç l n u l e n</p><p>tkać [tkat°Ç] ‘weave’ lnu [lnu] ‘flax, gen.sg.’ len [len] ‘flax, nom.sg.’</p><p>Since two obstruents and two sonorants do not form sufficient complexity /</p><p>sonority slope, they must be separated by an empty nucleus. Now, if this</p><p>empty nucleus is granted licensing power, the false cluster may be gram-</p><p>matical, provided that the empty nucleus is itself licensed by the following</p><p>full vowel. The relation responsible for this licensing in standard GP is</p><p>known as Proper Government and is marked by a solid arrow in (44).</p><p>Thus,</p><p>there are two conditions on false clusters. Firstly, the empty nucleus inside</p><p>120 Formal complexity</p><p>that cluster must be a licenser. And, secondly, the empty nucleus must be</p><p>properly governed by the following melodically expressed nucleus. If one</p><p>of these conditions is not fulfilled then, either the given cluster is ungram-</p><p>matical and therefore impossible, or the empty nucleus must be vocalized</p><p>as in (44c).</p><p>Thus, there is a precisely predicted typology of effects concerning two</p><p>adjacent consonants of particular melodies. If the three conditions on gov-</p><p>ernment listed in (42) are fulfilled then the two consonants contract a gov-</p><p>erning relation. We may call such surface consonant sequences true clus-</p><p>ters (45a), as opposed to those, in which no governing relation can be</p><p>contracted. The false clusters (45b) have their own conditioning: the empty</p><p>nucleus must be a licenser for its onset, and it must be itself licensed by the</p><p>following vowel through a relation of Proper Government. If one of these</p><p>conditions fails, then no surface cluster is possible (45c).</p><p>(45)</p><p>C1 C2 a. R←T, T→R true clusters (governing relations)</p><p>| | b. CPC false clusters</p><p>α β c. no cluster</p><p>Thus, clusterless languages are those which cannot have governing rela-</p><p>tions between consonants, or do not allow empty nuclei to license their</p><p>onsets word-internally. The situation in (45c) in fact subsumes a number of</p><p>possible effects. For example, let us imagine a situation that a given lan-</p><p>guage may have true clusters but not false ones. If for some reason two</p><p>consonants cannot contract a governing relation in that system, then we</p><p>predict a number of possible outcomes. Firstly, the two consonants will be</p><p>separated by an epenthetic vowel as in Dutch harp [har´p] ‘harp’. A con-</p><p>sonant simplification or deletion may also be expected. We will return to</p><p>the distinction between true and bogus clusters below. Let us briefly look</p><p>at two other examples where the nature of government enforces the pres-</p><p>ence of empty nuclei and point to the mechanism of their licensing.</p><p>The first one concerns the word-initial context and the problem of ‘s+C’</p><p>clusters. It is claimed in standard GP that such clusters always form a left-</p><p>ward governing relation (Kaye 1992) in which ‘s’ is in the coda and not in</p><p>a branching onset with the following consonant. Recall that governing</p><p>relations depend on the complexities of the participants. Word-medially,</p><p>this presents no problem, as can be seen in (46a) below. On the other hand,</p><p>in word-initial context this has far-reaching consequences. If in a sequence</p><p>[str], the fricative is governed to the left, then it is automatically assigned</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 121</p><p>to the coda of the preceding rhyme. This word-initial rhyme must contain</p><p>the head, the nucleus, which is empty (46b). Note, that the syllabic con-</p><p>figurations for the string [str] are identical, that is, coda+branching onset,</p><p>and recall that the same structure is also given to this string word-finally,</p><p>e.g. in sióstr [Çustr]</p><p>above become fully meaning-</p><p>ful only when they are viewed as part of a particular sound system. As we</p><p>will see presently, it may be the case that a given phonological representa-</p><p>tion will not correspond to identical phonetic interpretations across lan-</p><p>guages. Here we differ markedly from Kaye, Lowenstamm and Vergnaud</p><p>(1990: 194) who assume that “the same physical object will receive uni-</p><p>form interpretation across phonological systems”. Since they made their</p><p>proposal it has been found that the same representation will not always</p><p>yield identical phonetic effects or vice versa. That is, identical phonetic</p><p>objects may have disparate phonological representations across systems.</p><p>2.1. Representing vowels</p><p>The first three elements (A), (I), and (U) in (1) define vocalic expressions</p><p>and place of articulation in consonants. The discussion of vowel systems</p><p>within the Element Theory will serve the purpose of a rather sketchy illus-</p><p>tration of some of the points made above. However, in general, more em-</p><p>phasis will be placed on consonantal systems in this work.4</p><p>A basic three-vowel system, for example [a,i,u], reflects simplex repre-</p><p>sentations involving only one element in each case (2a). These are the least</p><p>marked vowels which utilize the phonetic vowel space most efficiently.</p><p>We may define this space either in terms of articulation, using familiar</p><p>properties like HIGH, LOW, BACK, FRONT, or in terms of acoustic dimen-</p><p>sions.5 At any rate, the simplex character of the three corner vowels re-</p><p>flects their universally unmarked status (Crothers 1978, Maddieson 1984).</p><p>The schwa vowel represents the neutral state of articulators and, typically,</p><p>evenly spaced-out formants. In Government Phonology this vowel may be</p><p>viewed as a realization of a neutral element or nothing, a point which will</p><p>be returned to when we discuss headedness.</p><p>4 For more extensive studies of how resonance elements function in phonological</p><p>systems the reader is referred to Backley and Takahashi (1998), Bloch-Rozmej</p><p>(1998), Charette and Göksel (1998), Cobb (1997), Cyran (1997), Polgárdi (1998),</p><p>Rennison (1998), Scheer (1996).</p><p>5 See, for example, Ladefoged (2001: 39ff) for a discussion of how, with a certain</p><p>amount of theoretical gymnastics, the same phonetic space can be defined in terms</p><p>of F1 and F2 values.</p><p>The Element Theory in Government Phonology 5</p><p>(2) a. b.</p><p>I U I U</p><p>i u i ü u</p><p>´ E O</p><p>A A</p><p>A A</p><p>Other vowels are combinations of the elements (I), (A), (U), for example,</p><p>(A-I) = [E], (A-U) = [O], (I-U) = [ü] (2b). It follows from the illustrations in</p><p>(2) that the more complex and marked vowel systems have more complex</p><p>representations in terms of combinations of elements. Thus, the relation be-</p><p>tween markedness and representational complexity is inherent to the model.</p><p>The relative markedness of mid vowels is reflected in the fact that they</p><p>are the first vowels to be eliminated in prosodically weak positions. Let us</p><p>look at some typically quoted instances of vowel reduction in unstressed</p><p>positions (Harris and Lindsey 1995).</p><p>(3) Bulgarian vowels Catalan vowels</p><p>under stress i e a o u i e E a O o u</p><p>↓ ↓</p><p>unstressed i ´ u i ´ u</p><p>Note that in both languages the surviving melodies in unstressed positions</p><p>are simplex. We do not wish to make any particular claims concerning the</p><p>representation of schwa vowels in the two systems, that is, whether they</p><p>still contain the element (A). However, one thing is clear, compound struc-</p><p>tures cannot be maintained in prosodically weak positions in some languages.</p><p>We must note two immediate advantages of the Element Theory in the</p><p>description of vowel reduction. Firstly, the relative markedness is directly</p><p>read-off from the representations rather than extrinsically encoded on the</p><p>basis of observation. Here, mid vowels are marked because they are com-</p><p>plex objects. Secondly, there is a direct and logical connection between</p><p>vowel reduction and the context where it occurs. Prosodically weak posi-</p><p>tions simply eschew complex vocalic structures, therefore, the latter must</p><p>be reduced in complexity.</p><p>So far, we have seen how to represent vowel systems possessing between</p><p>three and six objects, and the obvious question is what happens in systems</p><p>6 Substantive complexity</p><p>with more than six vowels, or in those in which there are two types of mid</p><p>front and mid back vowels as shown in Catalan in (3). At this point, one</p><p>more aspect of representations in the Element Theory must be introduced.</p><p>This additional mechanism is called headedness.</p><p>When two elements combine to form a compound, for example, (A-I), it</p><p>is assumed that the elements may enter into an asymmetrical relation in</p><p>which one of the elements may dominate the other, thus yielding a differ-</p><p>ent object than if the situation was reversed.6 Roughly speaking, a com-</p><p>pound structure (A-I) which is I-headed, that is (A.I), may correspond to</p><p>phonetic [e], while (A.I) should give [œ]. In other words, due to the reversed</p><p>head-operator relations, we are dealing with an essentially high front vowel</p><p>which is lowered, and an essentially low vowel which is fronted and raised,</p><p>respectively.7</p><p>The use of headedness has been extended to two other situations. One</p><p>of them concerns simplex structures. Here we find two different represen-</p><p>tations, that is, a simplex structure which is headed, and a headless one.</p><p>Thus, the contrast between a lax [I] and a tense [i] may be expressed by</p><p>referring to a headless (I._) vs. headed (I), respectively. Similarly, a com-</p><p>pound as a whole may also be headless, for example, (A.I._). This structure</p><p>may correspond to the open front mid vowel [E].</p><p>Thus, the introduction of headedness is meant to account, among other</p><p>things, for tense/lax contrasts, introducing greater generative potential into</p><p>the simple theoretical system which uses only three basic categories. Note</p><p>that now we are able to define much richer systems, including such con-</p><p>trasts as the one between [e] and [E], which we saw earlier in the system of</p><p>Catalan. In fact, the introduction of headedness allows the model to define</p><p>at most twenty independent vocalic objects, and attempts have been made</p><p>to propose mechanisms or parameters which would restrict the generative</p><p>power of the Element Theory with respect to individual systems (e.g. Cha-</p><p>rette and Göksel 1998, Backley 1995, 1998, Cobb 1993, 1997, Kaye 2001).8</p><p>6 This idea is familiar from such models as Dependency Phonology (e.g. Anderson</p><p>and Ewen 1987). Headedness will be represented by underlining the relevant element.</p><p>7 Throughout this work the elements will be used in parentheses and underlined</p><p>when headed, unless headedness is irrelevant for the discussion. Compounds in</p><p>which head specifications are deliberately omitted will be represented as e.g. (A-I).</p><p>8 One must add that apart from the three resonance elements, (L), (H), and (N) may</p><p>also be used in vowels. They represent tonal patterns – low and high pitch – and</p><p>nasalization respectively.</p><p>The Element Theory in Government Phonology 7</p><p>As for schwa vowels, there are various options to consider. It is not im-</p><p>possible that some schwas do have an active resonance element in operator</p><p>position, for example, (A._), (U._), (I._). In other words, the nuclei still</p><p>contain elements, though they are headless. This would account for the</p><p>various qualities of schwa vowels, not only across languages but also</p><p>within one system, for example, English. Within the Element Theory, it has</p><p>also been proposed that there is an additional, neutral element (@) which is</p><p>present in all representations but only shows up, as it were, if the full-</p><p>blooded elements are absent (Harris and Lindsey 1995). Other proposals</p><p>boil down to the assumption</p><p>is applied iteratively from right to left.</p><p>Thus, while the final nucleus (P3) is licensed by parameter, the second nu-</p><p>cleus (P2) from the right would have to be realized, thus providing the li-</p><p>censing for the first nucleus (P1). In order to account for this problem, it is</p><p>proposed that the diminutive suffix is analytic, that is, it constitutes a do-</p><p>main of its own /[[p´P1sP2]P1kP2]/. This gives two sequences of empty nu-</p><p>clei, which are independent of each other because they occur in different</p><p>cycles. The two domain-final empty nuclei are licensed by parameter.</p><p>Therefore, in each sequence only the first nucleus (P1) is realized, yielding</p><p>the correct phonetic form [p´esek]. The genitive form of this diminutive is</p><p>pieska [p´eska]</p><p>this condition on government needs to be redefined to hold</p><p>either at some level of the projection of onsets, or at the level of melodies.</p><p>These are the only two levels where the two consonants might see each other.</p><p>Finally, the analysis in (48b) appears to involve two competing licens-</p><p>ing mechanisms which are potentially conflicting. We must somehow make</p><p>sure that the interonset relation is contracted prior to the application of</p><p>Proper Government to obtain the correct results. This would suggest some</p><p>sort of ordering or ranking of the licensing principles, a consequence which</p><p>is at odds with the non-derivational stand of GP.</p><p>Even if we accept the position that Interonset Government takes prece-</p><p>dence over Proper Government, this would have some grave consequences</p><p>for the latter, in that it would be reduced to nothing more than a kind of</p><p>‘sweep-up’ mechanism with very limited application in Polish phonology.</p><p>To see this better, let us return to the standard analysis of the three-</p><p>consonantal clusters involving two empty nuclei (Gussmann and Kaye</p><p>1993, Cyran and Gussmann 1999). For the purpose of illustration we</p><p>choose a form which exhibits a vowel – zero alternation, which is typically</p><p>dealt with by means of Proper Government (44b-c).</p><p>(49)</p><p>a. O N1 O N2 O N3 b. O N1 O N2 O N3</p><p>| | | | | | ↑ |</p><p>m g w a m g e w</p><p>| | | | | |</p><p>C T R C T R</p><p>[mgwa] mgła ‘mist’ [mg´ew] mgieł ‘mist, gen.pl.’</p><p>The analysis is as follows. In (49a), the two onsets, which constitute a ris-</p><p>ing sonority pattern, contract an interonset governing relation, thus locking</p><p>or licensing the intervening empty nucleus. The final vowel N3 is, there-</p><p>fore, able to properly govern the first empty nucleus N1 and the form is</p><p>rendered grammatical. In (49b), on the other hand, the final nucleus N3 is</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 127</p><p>empty and disallows an interonset relation. Therefore, the preceding empty</p><p>nucleus N2, must be realized phonetically because it is not properly gov-</p><p>erned. However, having received melody it is able to properly govern N1.</p><p>Forms like mgła / mgieł cannot be overestimated as they prove inde-</p><p>pendently, through the vowel – zero alternation, that there is indeed an</p><p>empty nucleus P2 in the pattern CP1TP2RV.61 However, the consequences of</p><p>the analysis involving Interonset Government in such forms are quite det-</p><p>rimental to the model of standard GP. One upshot of this analysis, which</p><p>was already mentioned above, is that Interonset Government takes prece-</p><p>dence over Proper Government not only in those rising sonority clusters</p><p>like [kn], where standard GP inserts an empty nucleus because they are not</p><p>acceptable as branching onsets, but also in clusters which could form licit</p><p>branching onsets, but they cannot due to the lexical presence of an empty</p><p>nucleus. This lexical presence of the empty nucleus in mgła / mgieł follows</p><p>from the presence of the vowel – zero alternation. This in turn means, that</p><p>a fair number of regular cases of vowel – zero alternation, occurring in the</p><p>strings of the pattern /TPRV ~ TeRP/, which were traditionally viewed as</p><p>instances of the application of Proper Government, must now be reana-</p><p>lysed as involving Interonset Government. This concerns both word-initial</p><p>and word-final strings, for example, gra / gier [gra ~ g´er] ‘game, nom.sg.</p><p>/gen.pl.’, cukier / cukru [t °suk´er ~ t °sukru] ‘sugar, nom.sg. /gen.sg.’, etc.</p><p>Thus the correct analysis of gra should be that in (50a) and not (50b).</p><p>(50) a. *b. c.</p><p>O N O N O N O N O N O N</p><p>| | | | | | | ↑ |</p><p>g r a g r a g´ e r</p><p>In the analysis of such alternations, we must assume that the final vowel</p><p>does not properly govern the preceding empty nucleus because it will be</p><p>superseded by Interonset Government (50a). The vowel may, at best, pro-</p><p>vide licensing for the interonset relation, and it is the latter mechanism that</p><p>licenses the intervening empty nucleus. On the other hand, in gier (50c) we</p><p>must say that the first empty nucleus is realized because there is no proper</p><p>governor or interonset relation to license it, therefore, it must surface. In</p><p>other words, the mechanism of Proper Government is needed in the ac-</p><p>61 There are about three examples of this pattern: mgła / mgieł ‘mist, nom.sg.</p><p>/gen.pl.’, źdźbło / źdźbeł ‘blade of grass, nom.sg. /gen.pl.’, and pchła / pcheł ‘flea,</p><p>nom.sg. /gen.pl.’.</p><p>128 Formal complexity</p><p>count of these forms only to say that its absence causes vocalization of the</p><p>empty nucleus.</p><p>It is more than obvious that some simplification of the model is in order</p><p>and Proper Government appears to be a good candidate for elimination.</p><p>This is possible if some proposal can be made concerning the forms in</p><p>which Proper Government seems to be the only licensing mechanism avail-</p><p>able, for example, with initial clusters which do not exhibit a rising sonor-</p><p>ity profile /TPRV/, e.g. kto [kto]</p><p>relations behave identically, and both can be called</p><p>true clusters, if only because both involve a governing relation. On the</p><p>other hand, we need a third structure which behaves as if it contained a</p><p>visible empty nucleus even if the surrounding consonants could contract an</p><p>interonset relation, a false cluster. This structure is excluded in (50b) on</p><p>the basis of the forms like mgła / mgieł (49), but it is also shown in (51a) to</p><p>be necessary. It appears, then, that mgła / mgieł could have lured us into</p><p>making wrong proposals, namely, that onsets flanking an alternating vowel</p><p>could contract a governing relation. We will return to the behaviour of</p><p>branching onsets and Interonset Government, as opposed to false clusters</p><p>in section 6, and argue that next to false clusters only one structure of true</p><p>clusters is necessary.</p><p>In the following subsection, we look in more detail at the distinction be-</p><p>tween true and false clusters in relation to the problem of initial consonant</p><p>clusters in English and Polish.</p><p>5.6. True or False? English and Polish initial clusters</p><p>There are a few theoretical points concerning the distinction between true</p><p>and false clusters we introduced earlier, which require clarification. One</p><p>question is whether the model predicts any implicational relationship be-</p><p>tween the two types of consonant sequences. Another question concerns</p><p>diagnostic contexts and effects which tell us which structure we are dealing</p><p>with. Finally, the obvious question is if this theoretical distinction corre-</p><p>sponds to real empirical aspects of phonological systems.</p><p>130 Formal complexity</p><p>Judging by the conditions on government in (42), which underlie true</p><p>clusters (R←T and T→R), as well as those on false clusters (CPC), it must be</p><p>said that theoretically speaking there can be no implicational relationship</p><p>between the two structures. Melodic and adjacency issues aside, the crucial</p><p>conditions on true and false clusters boil down to two different aspects of</p><p>one mechanism: licensing. While, in the case of true clusters it is the abil-</p><p>ity of nuclei to government license (52b-c), false clusters require that the</p><p>intervening empty nucleus can license its onset (52a).</p><p>(52) a. false b. true RT c. true TR</p><p>C v C V C C V C C V</p><p>| | | | | | | | |</p><p>C C α R T α/P T R α/P</p><p>Note that in a false cluster both onsets have their own licensers. It is clear</p><p>that for a false cluster to be viewed as grammatical, an empty word-medial</p><p>nucleus must be able to license its onset (P is a Licenser). Additionally, this</p><p>empty nucleus cannot be followed by another empty one.62 On the other</p><p>hand, true clusters require government licensing (Ns License to Govern)</p><p>and this property can be possessed by both filled and empty nuclei, as we</p><p>saw in earlier sections, depending on language specific choices.63 The two</p><p>parameters provide us with the following typology of possible systems</p><p>with respect to the occurrence of consonant sequences, of which only one</p><p>combination yields structural ambiguity between true and false clusters.</p><p>(53) A B C D</p><p>P is a Licenser – – + +</p><p>Ns License to Govern – + – +</p><p>The above typology shows clearly that the two parameters, which are inde-</p><p>pendently manipulated, allow for no implicational relationship between</p><p>true and false clusters. System A is one in which no surface consonant</p><p>62 In other words, sequences of two empty nuclei are ungrammatical (*P–P).</p><p>63 Recall from section 4 that decisions as to which types of nuclei government li-</p><p>cense, and at which level of syllabic complexity, are language specific.</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 131</p><p>sequences can be found, as they are eliminated by the negative settings of</p><p>both parameters. System B contains only true clusters, that is governing</p><p>relations. Whether both RT and TR clusters will be used depends on the</p><p>language particular setting of the licensing strength. By contrast, system C</p><p>will only have false clusters. It is a system in which empty nuclei may li-</p><p>cense simplex onsets, while no government licensing is possible for any</p><p>type of nuclei. Finally, system D is the most complex. It has both true and</p><p>false clusters. It is in this type of system that we need to ask about the di-</p><p>agnostic contexts which would identify the type of structure at hand. We</p><p>will briefly comment on two.</p><p>The right edge of words appears to be one of the most reliable diagnos-</p><p>tic contexts in this model. Recall that an empty nucleus inside a false clus-</p><p>ter cannot be followed by another empty nucleus (52a). Thus, only true</p><p>clusters, that is, governing relations can survive at the right edge of mor-</p><p>phologically simplex words.</p><p>Word-initial and medial contexts are more ambiguous in that here a</p><p>given sequence is pre-vocalic and it is not immediately obvious whether</p><p>that vowel licenses a governing relation or simply does not disallow a false</p><p>cluster. In these contexts the lack of melodic restrictions on the consonant</p><p>sequences can be diagnostic. This brings us to the distinction between Pol-</p><p>ish and English initial clusters.</p><p>In English, word-initial clusters are highly restricted. Two consonant</p><p>clusters are limited to branching onsets, that is rightward governing rela-</p><p>tions (T→R), e.g. try, and the ‘magic’ sC, e.g. stop. The latter sequence is</p><p>in fact a coda-onset relation (R←T), but because of its special status we</p><p>will continue to refer to it as sC. There are stringent melodic restrictions on</p><p>the structure of the branching onsets in this language, which have tradi-</p><p>tionally been captured in terms of sonority distance and constraints on</p><p>homorganicity. These disallow initial clusters like *[pn, kn, tf] and *[pw,</p><p>tl, dl] respectively. On the other hand, sC sequences are virtually unre-</p><p>stricted except for the sequence *[sr]. Clusters of three consonants in Eng-</p><p>lish are a combination sC and TR, in that they all must take the pattern sTR,</p><p>e.g. string.</p><p>Given that both sC and TR are instances of true clusters, that is, govern-</p><p>ing relations, we may rightly conclude that English does not allow false</p><p>clusters on the left edge of words. Technically, this decision may be ex-</p><p>pressed by a parameter on the licensing potential of empty nuclei in this</p><p>context – since they are not granted licensing abilities, false clusters are</p><p>out. It must be stressed that the ban on false clusters in English concerns</p><p>only the left edge. English does have post-vocalic, that is word-medial false</p><p>132 Formal complexity</p><p>clusters, for example, bottling [bOtlIN]</p><p>the empty nucleus is realized phonetically, provide additional sup-</p><p>port for claiming that [ln, kt, lv] cannot be branching onsets. They must be</p><p>false clusters in which the empty nucleus is granted licensing potential, that</p><p>is, they license their onset. Thus, vowel – zero alternation is another diag-</p><p>nostic phenomenon for the presence of the empty nucleus inside a cluster.</p><p>It appears that this empty nucleus in Polish is able to license more than</p><p>just a simplex onset, which is not surprising. Theoretically speaking, any</p><p>64 The sequence [tl] must be viewed as a false cluster for two reasons. Firstly, it</p><p>cannot be a branching onset, for homorganicity reasons, and it cannot be a coda-</p><p>onset governing relation, for sonority / complexity reasons.</p><p>65 See Scheer (2004) for a completely different interpretation of the absence of</p><p>false clusters on the left edge, which refers to the presence of an empty CV site at</p><p>the beginning of English words.</p><p>66 As we saw in the previous subsection, false clusters in Polish may also take the</p><p>rising sonority pattern obstruent + sonorant. This problem is taken up again in the</p><p>following section.</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 133</p><p>nucleus should be able to license any level of formal complexity (CV – RTV</p><p>– TRV) in any position within the word. Consider the following structures</p><p>proposed for three-consonant clusters in Polish.</p><p>(55) # C1C2C3</p><p>a. s←T→R strawa [strava] ‘food, nom.sg.’</p><p>b. CPT→R tkliwy [tklivÈ] ‘tender’</p><p>c. T→RPC krwi [krf´i] ‘blood, gen.sg.’ cf. krew [kref] ‘ibid., nom.sg.’</p><p>d. CPs←C bzdura [bzdura] ‘nonsense’67</p><p>e. s←CPC szkło [Skwo] ‘glass, nom.sg.’ cf. szkieł [Sk´ew] ‘ibid., gen.sg.’</p><p>The structures of ternary initial clusters in Polish appear to utilize the same</p><p>configurations as English, that is, sTR, which is the only possible structure</p><p>of a ternary true cluster, as well as the combinations of sC and TR with a</p><p>single consonant separated by an empty nucleus. This way, we get five</p><p>predicted structural patterns, all of which find instantiations in real data.</p><p>(55b-c) are combinations of a branching onset and a single onset. Note that</p><p>in krwi [krf´i]</p><p>as licensers in a given system. In Polish, for example, the issue does</p><p>not arise because empty nuclei may license both types of clusters, that is,</p><p>T→R and R←T relations, as we may observe in such forms as krwi [krf´i]</p><p>in Polish which were discussed earlier, we must mention one more mecha-</p><p>nism responsible for the vocalization of empty nuclei. This is connected</p><p>with the ban on sequences of such objects, that is *P−P. Recall that the</p><p>vocalization of the first nucleus in such a sequence was thought to be the</p><p>result of the absence of Proper Government. This option is now unavail-</p><p>able. For the time being let us assume that there is a universal constraint on</p><p>this structure. Rowicka (1999: 54), for example, refers to this constraint as</p><p>NO LAPSE, thus attempting to ground it in the universal rhythmic organiza-</p><p>tion of speech, whereby sequences of unstressed syllables are avoided.</p><p>This structure is always resolved as a strong – weak sequence, reminding</p><p>us of the trochaic foot organization.76 This second mechanism is crucial to</p><p>account for such alternations in Polish as cukier / cukru [t °suk´er ~ t °sukru]</p><p>‘sugar/gen.sg.’. Recall that the nominative form has a sequence of two</p><p>empty nuclei, that is, /t °sukPrP/. Referring solely to the licensing properties</p><p>of empty nuclei would not be sufficient, as they can license not only simplex</p><p>75 The ‘magic licensing’ parameter should also be eliminated for consistency’s</p><p>sake. This point will be discussed in chapter 3.</p><p>76 In fact, Rowicka (1999) retains Proper Government in her model but she reverses</p><p>the direction and views it as a trochaic relation. Thus, the first nucleus is realized</p><p>and forms the stronger part of the foot, which means it can properly govern the</p><p>second empty nucleus. It seems, however, that this intuitively correct approach can</p><p>be maintained without recourse to Proper Government.</p><p>Sources of empty nuclei and licensing mechanisms in standard GP 139</p><p>but also complex onsets, for example, wiatr [v´atr]</p><p>and synchronic justifica-</p><p>tion, as well as a purely theoretical one. The existence of empty nuclei is</p><p>predicted, or at least not excluded, by the very model of three-dimensional</p><p>phonology in which the prosodic and melodic levels of representation are</p><p>relatively independent of each other.</p><p>78 The underlined empty nucleus means that the surrounding onsets are in a gover-</p><p>ning relation.</p><p>142 Formal complexity</p><p>The main source of empty nuclei in representation is connected with</p><p>governing relations. They may appear also due to grammatical settings of</p><p>parameters and as a result of purely lexical distribution – like any other</p><p>vowel. We looked at three contexts in which empty nuclei occur and con-</p><p>sidered the licensing mechanisms which are used in standard GP to silence</p><p>them.</p><p>Since in the model of Complexity Scales and Licensing (CSL), which</p><p>we are trying to develop in this book, the primary job of nuclei is to license</p><p>the preceding onset and the formal configurations in which this onset might</p><p>be involved, we assumed that an empty nucleus can be used in a given</p><p>system only if it is afforded some licensing properties. Thus, we shift the</p><p>focus from licensing of empty nuclei to licensing properties of the nuclei in</p><p>question. This allows us to simplify the model by eliminating a number of</p><p>mechanisms known from standard GP, which were responsible for licens-</p><p>ing empty nuclei, namely, the magic licensing parameter, the domain final</p><p>parameter and Proper Government. Recall that earlier we eliminated pa-</p><p>rameters on branching constituents, which were replaced by the complexity</p><p>scale of formal configurations interacting with the licensing properties of</p><p>nuclei. The only mechanism which is left from standard GP and may be</p><p>viewed as a licensing instrument is Interonset Government (IO) which,</p><p>however, does not license empty nuclei in the traditional sense. First of all,</p><p>IO is a governing relation that is not motivated by a need to license the</p><p>intervening empty nucleus. It just is. It is an automatic relation that must be</p><p>contracted if all the conditions are fulfilled. The fact that, IO ‘locks’ the</p><p>intervening empty nucleus and makes it invisible to other nuclei and to the</p><p>constraint on sequences of empty nuclei (*P–P) should rather be viewed as</p><p>a side-effect.</p><p>The CSL model predicts that some onsets may be licensed by empty</p><p>nuclei, not only word-finally but also word-medially and initially. Depend-</p><p>ing on the licensing properties of such empty nuclei, we naturally predict</p><p>the existence of complex clusters such as those in Polish and their absence</p><p>in languages like English. The two systems differ not only in terms of what</p><p>their empty nuclei can license, leading to the distinctions in the word-final</p><p>context, but also with respect to the particular positions within the word.</p><p>The absence of empty nuclei at the left edge in English effectively elimi-</p><p>nates strings like *#kt…, *#tkl…, or *#krf…, which are found in Polish</p><p>kto, tkliwy and krwi and leaves only those structures which are simplex</p><p>onsets or true clusters, that is, C (tap), sC (stop), TR (trap), and sTR (strap).</p><p>In the following section, we will take things a step further and show that</p><p>the model of complexity scales and licensing strength should be redefined</p><p>Polish as a CV language? 143</p><p>as a model in which the syllable structure is assumed to be a sequence of</p><p>Cs and Vs (Lowenstamm 1996, Scheer 1996, 1998b, 2004). One of the</p><p>immediate advantages of this move will be elimination of branching onsets,</p><p>that is, the structure that behaves identically to interonset relations. Conse-</p><p>quently, the status of branching rhymes and nuclei will also have to be</p><p>reconsidered.</p><p>6. Polish as a CV language?</p><p>6.1. Introduction</p><p>In this section an attempt is made to demonstrate that the model of com-</p><p>plexity scales and licensing strength (CSL) is fully compatible with the</p><p>radical hypothesis that syllable structure is in fact a sequence of consonan-</p><p>tal and vocalic positions, that is, simplex onsets and nuclei (Larsen 1994,</p><p>Lowenstamm 1996, Scheer 1996, 1998b, 2004, Rowicka 1999).79 It is also</p><p>better suited for handling a number of questions concerning Polish phono-</p><p>tactics. The purpose of this exercise, however, is not limited to the mere</p><p>redefinition of a model with maximally binary syllabic constituents, which</p><p>is already fairly constrained, as a more abstract model in which every con-</p><p>sonant is structurally followed by a nucleus. There are a few reasons why</p><p>this step seems to be necessary.</p><p>One of the reasons why the strict CV assumption appears to be more at-</p><p>tractive than branching constituents is connected with the internal logic of</p><p>the complexity scales and licensing model. Note that in the modified view</p><p>of phonological organization in which word structure is an effect of a tug</p><p>of war between formal complexity and the licensing strength of nuclei, the</p><p>entire syllable typology as well as language specific settings are now dealt</p><p>with by referring to the formal configurations of the onsets, and their li-</p><p>censing relation with the following nucleus. Thus, in effect, we have al-</p><p>ready been dealing with a pattern of onsets and nuclei. Since the onset</p><p>configurations beyond level I of syllabic complexity, that is CV, involve</p><p>79 Rowicka (1999) is the first study employing the strict CV assumption in the ana-</p><p>lysis of Polish clustering. It is also an attempt to eliminate parameters from stan-</p><p>dard GP. However, her model replaces parameters with violable universal con-</p><p>straints, thus attempting to connect the GP way of viewing phonological repre-</p><p>sentation with the Optimality Theory of constraint interaction. A similar attempt</p><p>within GP is found in Polgárdi (1998).</p><p>144 Formal complexity</p><p>governing relations between two consonants, whether they are strictly ad-</p><p>jacent or adjacent in the sense that no vocalic melody separates them does</p><p>not make much difference and may be a matter of general assumption, or a</p><p>question of which of the two options is more suited to explain particular</p><p>phonological phenomena.80</p><p>More importantly, the CSL model relies heavily on formal distinctions</p><p>between structures. Recall that the distinction between true R←T and T→R</p><p>clusters lies in the distance between the licenser and the head of the gov-</p><p>erning relation (Direct vs. Indirect Government Licensing). It appears then,</p><p>that the introduction of an empty nucleus inside similar, though, interonset</p><p>relations should have consequences on the understanding of the syllabic</p><p>space introduced in section 4.4. In each case, whether it is leftward or</p><p>rightward interonset, we would be dealing with greater distances. This</p><p>should lead to a clear difference between skeletally adjacent governing</p><p>relations and those of interonset type, which would be reflected in empiri-</p><p>cal facts. Thus, from the point of view of CSL it would be best to be deal-</p><p>ing with one type of government, either interonset or one involving skeletal</p><p>adjacency.</p><p>The second reason for considering the CV assumption is that most of</p><p>the problematic cases in Polish phonotactics already receive a CV analysis.</p><p>This concerns not only the sequences of two consonants, as in mchu</p><p>/mPxu/ ‘moss, gen.sg’, and kto /kPto/ ‘who’, which have been shown to</p><p>contain an empty nucleus, but also three-consonant sequences, for exam-</p><p>ple, tknąć /tPkPnõt °ÇP/ ‘touch’, etc., in which two empty nuclei must be</p><p>postulated on theory internal grounds, and the entire word is formed of</p><p>sequences of simplex onsets and nuclei.</p><p>Additionally, the forms which already reflect a CV pattern occur along-</p><p>side ones with assumed branching constituents, thus producing a variety of</p><p>formal configurations which seem to cover similar if not the same empiri-</p><p>cal ground. For example, the purpose of introducing the structural overlap</p><p>between branching onsets and interonset relations in the analysis of</p><p>tknąć</p><p>[tkno≠t °Ç]</p><p>cism of this mechanism.</p><p>148 Formal complexity</p><p>the same interpretation under the CV assumption, for example, krtań</p><p>[krta≠]</p><p>onsets. How-</p><p>ever, there seems to be one problem with the replacement of BrO by RIO –</p><p>the nature of government. If RIO is contracted in bryzgać, there is no rea-</p><p>son why it should not be present also in brać.</p><p>Recall that government must be contracted if all conditions are fulfilled.</p><p>Namely, if two consonants are adjacent at a relevant level, they form a</p><p>sonority / complexity slope, and they are licensed by the following nucleus.</p><p>It seems that the last two conditions must be viewed as fulfilled in brać.</p><p>Specifically, the sequence [br] is melodically identical in bryzgać and brać,</p><p>thus, the complexity slope should equally favour government in both in-</p><p>stances. Also, in both cases the sequence [br] is followed by a full vowel</p><p>which is a perfect licenser.84 The only condition which may distinguish</p><p>between brać and bryzgać is that of adjacency.</p><p>In fact, adjacency is an equally pressing problem for standard GP analy-</p><p>sis and for the model we are trying to develop here. It will be recalled, that</p><p>once interonset is admitted in standard GP – this was argued for on the</p><p>basis of forms like tknąć (48b) – adjacency at the level of skeleton is no</p><p>longer valid, and the intervening empty nucleus is no longer a blocker to</p><p>government.85 Thus, before a systemic elimination of BrO and replacing it</p><p>with RIO we need to be able to distinguish between RIO and ONO in</p><p>forms like bryzgać and brać, respectively. Since in both cases the onsets</p><p>are separated by an empty nucleus, the nucleus cannot be a blocker to gov-</p><p>ernment in one string and not in the other, unless the empty nuclei are not</p><p>of the same kind. The question is, then, what blocks RIO in (64c)? An at-</p><p>tempt to answer this question will be made in the following section. In</p><p>84 Note that the governing relation in [br] can be licensed also by an empty nucleus,</p><p>as in, e.g. brnąć [brno≠t°Ç] ‘wade’, regardless of whether the governing relation is</p><p>viewed as the branching onset (/b→rPnõt°ÇP/), or interonset type /bPrPnõt°ÇP/.</p><p>85 The problem was discussed in section 5.5 and illustrated in (50).</p><p>152 Formal complexity</p><p>what follows, however, we will consider an alternative way of approaching</p><p>the distinction between the stems in (63).</p><p>It has been proposed in the literature that the difference between the</p><p>data sets in (63a) and (63b) may lie in the way prefixed verbs are bracketed</p><p>(Booij and Rubach 1984, Szpyra 1989, Rowicka 1999). Without going into</p><p>too much detail, it is assumed, based on evidence from other phenomena</p><p>involved in prefixation such as palatalization spreading, that only prefixes</p><p>attached to a selected number of stems form with them one phonological</p><p>word (synthetic affixation), a domain within which phenomena like jer</p><p>vocalization may occur. The stems which require such affixation exhibit</p><p>the morphologically conditioned vowel-zero alternations of the type brać /</p><p>bierze ‘take / (s)he takes’, prać / pierze ‘wash / (s)he washes’, and so on</p><p>(63a). On the other hand, prefixes attached to other stems, that is, to those</p><p>lacking a jer, are said to form a separate (analytic) domain. Thus, according</p><p>to this proposal, zbryzgać must be bracketed as /[zP][brÈzgat °ÇP]/,86 while</p><p>zebrać has a one domain structure /[zPbPrat °ÇP]/.</p><p>From the point of view of standard GP, this leads to a peculiar situation</p><p>in Polish in that zbryzgać, which could be easily derived in the same way</p><p>as the independently motivated case of tkliwy (48a), that is, as a single</p><p>phonological domain (65a), is offered an additional mechanism securing</p><p>the absence of prefix vocalization by means of analytic bracketing (65b).87</p><p>(65) a. O b. O</p><p>[z P1 b→r È z g a t°Ç P] [z P] [b→r È z g a t°Ç P]</p><p>cf. [t P1 k→l i v È] BrO (48a)</p><p>[t P1 k P n õ t°Ç P] RIO (48b)</p><p>Both approaches to the structure of the initial cluster [br], that is, the stan-</p><p>dard GP analysis with a branching onset (BrO) and the one proposing an</p><p>interonset relation (RIO), are perfectly capable of handling the zbryzgać as</p><p>a synthetic domain. The nucleus P1 is not required to vocalize, and does not</p><p>need to be separated by a domain as in (65b).</p><p>86 The exact bracketing is irrelevant. See Booij and Rubach (1984), Szpyra (1989)</p><p>and Rowicka (1999) for proposals in this respect. The distinction can be broadly</p><p>made by referring to analytic versus non-analytic (synthetic) morphology.</p><p>87 The same argument holds even if [br] were viewed as a RIO /bPr/ locking the</p><p>intervening empty nucleus, as shown in the analysis of tknąć /tPkPnõt°ÇP/ (48b).</p><p>Polish as a CV language? 153</p><p>On the other hand, ironically, zebrać, which is assumed to form one</p><p>phonological domain, defies the established interpretation of three-onset</p><p>sequences shown in (65a). The first nucleus is vocalized. In this respect,</p><p>zebrać is as surprising as *megła and *teknąć would be. The analysis of</p><p>zebrać as opposed to tknąć involves one crucial difference, that is, an in-</p><p>teronset governing relation is absent in the former case, and present in the</p><p>latter.</p><p>(66) a. b.</p><p>*N N RIO</p><p>[z P b P r a t°Ç P] [t P1 k P n õ t°Ç P]</p><p>↑</p><p>e</p><p>The absence of RIO in (66a) creates a sequence of two unlocked empty</p><p>nuclei which must be resolved by vocalization due to the constraint *P−P,</p><p>whereas in (66b) there is only one visible empty nucleus P1. Thus we return</p><p>to our initial question of what conditions the fact that RIO is contracted or</p><p>not, which in fact is a question pertaining to the difference between what</p><p>we can call a true cluster and a false one.</p><p>6.2.4. Three types of nuclei in Polish</p><p>So far, we have seen that in all the diagnostic contexts which allow us to</p><p>detect the structure of the branching onset in Polish, the competing struc-</p><p>ture of rightward Interonset Government (RIO) is able to replace it, be-</p><p>cause it is predicted to behave in exactly the same way. If there is any func-</p><p>tional difference between phonetically identical strings of rising sonority in</p><p>Polish, it is always the case that BrO and RIO pattern together in opposi-</p><p>tion to the so called false clusters ONO, in which no governing relation is</p><p>found. The ultimate elimination of BrO from the phonology of Polish re-</p><p>quires, however, that a solution be found to the question why some se-</p><p>quences of the /TPR/ type, do not contract a governing relation, e.g. brać</p><p>/bPrat °ÇP/, even though all the necessary conditions seem to be fulfilled.</p><p>The answer must be sought in the representation. More precisely, there</p><p>must be something in the representation of brać that blocks RIO. Since</p><p>government is obligatory, it appears that its absence in brać is due to the</p><p>fact that one of the conditions on government is contravened. Recall that</p><p>melodically speaking, the sequence [br] in brać and bryzgać is identical.</p><p>154 Formal complexity</p><p>Therefore it is not the complexity / sonority slope requirement that prevents</p><p>government in the former. Additionally, in both forms [br] is followed by a</p><p>full vowel, hence, government licensing is also above suspicion. The only</p><p>condition on government that remains is adjacency. Normally, empty nu-</p><p>clei should not block government. It appears, however, that some of them</p><p>do, and they are typically the nuclei which sometimes appear as vowels.</p><p>Following Scheer (2004) we assume that there are two types of empty</p><p>nuclei. Representationally they differ in one respect. A truly empty nucleus</p><p>(P) is just a nuclear position in phonological representation (67c), while the</p><p>empty nucleus which alternates with vowels contains unassociated / float-</p><p>ing melody (Pe).</p><p>88 Let us compare these representations with regular vowels.</p><p>(67) a. full vowel b. alternating vowel c. empty nucleus</p><p>N N N</p><p>|</p><p>α α</p><p>The three structures of nuclei in (67) in fact represent all the logically pos-</p><p>sible configurations which follow from the three-dimensional</p><p>that schwa may have no representation in</p><p>terms of elements, that is, phonologically speaking it is a phonetically in-</p><p>terpreted nuclear position which has no melodic content. Under this pro-</p><p>posal, the difference between schwa and an empty nucleus proper lies only</p><p>in the fact that the former is interpreted phonetically and the latter remains</p><p>silent.9 Let us see how these options may be applied to the well-known</p><p>phenomenon of the rise and fall of jers in Slavic.</p><p>(4)</p><p>[u] > [ъ]</p><p>[P]</p><p>[i] > [ь]</p><p>Generally speaking the short high back and front vowels [u] and [i] were</p><p>weakened to the so called jers [ъ] and [ь], which were later lost in particu-</p><p>lar positions.10 Given the current assumptions of Element Theory, we may</p><p>provide three descriptions of the events depending on our view on the</p><p>structure of schwa and the status of the neutral element.</p><p>(5) [u/i] [ъ/ь] [P]</p><p>a. (U/I) > (@) > (_)</p><p>b. (U/I) > (U/I._) > (_)</p><p>c. (U/I) > (_) > (_)</p><p>9 More on empty nuclei can be found in the following chapters.</p><p>10 The development of jers will be discussed at length in chapter 3.</p><p>8 Substantive complexity</p><p>All three options agree in their interpretation of the last stage in which</p><p>there is no melody left in the nucleus. In (5a), the rise of jers is accompa-</p><p>nied by the complete loss of the melodies (U) and (I). What remains in the</p><p>representation is the neutral element. This analysis assumes that the oppo-</p><p>sition between back and front jers has been shifted onto the preceding con-</p><p>sonant, in that now front jers occur after palatalized consonants, while back</p><p>jers follow non-palatalized consonants. The interpretation in (5b) assumes</p><p>that the jers are schwa-like but they still contain the resonance elements as</p><p>operators, and only when these elements are lost is a phonetic zero possi-</p><p>ble. Under this view, only after the loss of jers should palatalization be</p><p>represented on consonants. The last view, represented in (5c), is similar to</p><p>(5a) in assuming that jers have no active resonance elements and that the</p><p>opposition between palatalized and velarized or neutral should be repre-</p><p>sented on consonants. However, it assumes that schwas and schwa-like</p><p>vowels may be representationally identical to empty nuclei. The difference</p><p>lies in the context-based interpretation of such constructs.</p><p>In this work, we will follow the assumption that there is no such thing</p><p>as a neutral element, which narrows down the options in (5) to two. How-</p><p>ever, the problem of the phonological structure of schwa, or of the jers,</p><p>cannot be dismissed with one sweeping statement. More detailed discus-</p><p>sion of these objects will be provided in the relevant contexts in the follow-</p><p>ing chapters. An example of an element-based analysis of a vocalic system</p><p>will be provided in section 3.1. Let us now turn to the representation of</p><p>consonants in the Element Theory.</p><p>2.2. Representing consonants</p><p>In the previous section we saw how vowels are represented in the Element</p><p>Theory and how a phonological representation may be affected in phono-</p><p>logical processing. Vowel reduction, for example, is a phenomenon in</p><p>which the internal structure of a vowel is decomplexified by means of de-</p><p>ducing primes, e.g. (A-I) > (I), or reducing their status from head to opera-</p><p>tor, e.g. (A) > (A._). Both cases are instances of weakening and their direct</p><p>contextual connection with weak prosodic positions is a welcome effect.</p><p>Besides decomposition, the Element Theory also predicts composition as</p><p>another possible type of phonological event. This process involves element</p><p>addition, as in vowel harmony or the strengthening of consonants. In both</p><p>instances a condition must be satisfied whereby the added element is lo-</p><p>The Element Theory in Government Phonology 9</p><p>cally present.11 Let us now look in more detail at the representation of con-</p><p>sonants in the Element Theory.</p><p>2.2.1. Place</p><p>The resonance elements discussed above define primary and secondary</p><p>places of articulation in consonants.12</p><p>(6) (I) = palatal, e.g. [j, ç, c]</p><p>palatalized, e.g. [pj, kj]</p><p>(U) = labial, e.g. [p, b, v, f, w]</p><p>labialized, e.g. [kw, gw]</p><p>(A) = coronal, e.g. [r, t, s]</p><p>retracted (uvular, pharyngeal), e.g. [R, q, G, ?]</p><p>(_) = velar, e.g. [k, g, x]</p><p>velarized, e.g. dark [:] in English</p><p>The categories given in (6) must be taken as rough indications rather than</p><p>exact representations. It will transpire presently that the best way to talk</p><p>about the Element Theory is within the context of a particular system. The</p><p>parsimony of the model must be striking for anyone familiar with the IPA</p><p>chart. However, it is also true that no language uses all the place, or indeed</p><p>manner distinctions found in the world’s languages. Thus, it must be borne</p><p>11 This is probably too general a statement. Some historical processes of consonant</p><p>strengthening, for example, [w] > [v] in the history of Slavic languages, require a</p><p>more complicated, and less idealized analysis (Cyran and Nilsson 1998). In a nut-</p><p>shell, since the weakening processes involve either element deduction or demotion,</p><p>it is logical that strengthening may involve element addition or promotion to hea-</p><p>ded status. Cyran and Nilsson claim that in Slavic strengthening in which there is</p><p>no source for the added elements, two stages are necessary: first element promo-</p><p>tion, e.g. (U)>(U), yielding [w~v] alternations, and then phonological reanalysis of</p><p>(U) as (U,h,L), yielding systems with [v~f] alternations. Mixtures of the two sys-</p><p>tems are also possible, e.g. in Slovak (Rubach 1993: 244).</p><p>12 There is no agreement as to the use of resonance elements in defining place of</p><p>articulation. For example, the old dilemma whether coronal or velar consonants</p><p>should be unmarked for place remains unsolved. See e.g. Backley (1993) and</p><p>Scheer (1996, 2004). In this work, we assume that velarity has no place element,</p><p>while coronality is represented by the element (A), or its combination with (I), that</p><p>is (A-I), as will soon become apparent.</p><p>10 Substantive complexity</p><p>in mind that the actual representations of consonants in a given system</p><p>must follow an in-depth analysis and should not be assumed a priori.</p><p>Before we consider the manner and source elements, let us briefly look</p><p>at an illustration of how primary and secondary articulations as defined by</p><p>resonance elements may interact in the description of certain historical</p><p>shifts in consonantal place of articulation.</p><p>In Celtic languages there was regular labialization of Indo-European</p><p>*gw to [b] as in, for example, IE *gwou-, ‘cow, ox’ > Old Irish bó, Welsh</p><p>bu, or IE *gwena@, ‘woman’ > Old Irish ben, Welsh benyw. A similar phe-</p><p>nomenon affected the proto-Celtic voiceless labialized velar *kw, but only</p><p>in the Brittonic subgroup, thus leading to the linguistic division into the so</p><p>called P– and Q–Celtic groups.13</p><p>(7) *kwetu8ores *kweis *makwkwo-</p><p>p (Brittonic) pedwar pwy mab</p><p>*kw</p><p>k (Goidelic) cethar cía macc</p><p>‘four’ ‘who’ ‘son’</p><p>Given that the representation of velars has no active element, the secondary</p><p>labialization is best represented as the presence of the (U) element in op-</p><p>erator position. The shift from [gw] to [b] in Celtic in general, or [kw] to</p><p>[p/b] in Brittonic, is thus directly captured as a switch in the status of the</p><p>resonance element from operator to head.14 For the moment we ignore the</p><p>other elements making up the velar plosive and concentrate on place only.</p><p>(8) velar labialized velar labial</p><p>[g] vs. [gw] vs. [b]</p><p>(_) (U._) (U)</p><p>The distinction between the three types of segments can be described as a</p><p>scale of (U) presence. While it is completely absent in plain velars, it af-</p><p>fects the labialized consonants as an operator – adds the labial colouring as</p><p>it were, or, in the case of</p><p>model of</p><p>representation.89 Polish appears to utilize all three structures. Full vowels</p><p>(67a) are complete representations with melody associated to the nuclear</p><p>position. These vowels do not alternate with zero. They also block interac-</p><p>tion between the surrounding onsets. Alternating vowels (67b), which</p><p>structurally constitute a halfway house between full vowels and empty</p><p>nuclei, contain unassociated melody, which may or may not be linked to</p><p>the nucleus depending on the shape of the following nucleus. The empty</p><p>nucleus (67c), on the other hand, is deprived of any melody, and does not</p><p>block Interonset Government.</p><p>The dual function of the alternating vowel is such that, as a licenser, it</p><p>patterns with the empty nucleus. If the melody is linked, then, quite logi-</p><p>cally, it behaves like a full vowel. On the other hand, even if the melody</p><p>remains unassociated, the alternating vowel behaves like a full vowel in</p><p>that it blocks government between the flanking onsets. This assumption</p><p>88 The floating melody in Polish is typically [e], hence the symbol Pe. There are also</p><p>alternations with [o], e.g. kozioł / kozła [koÛow ~ kozwa] ‘male goat, nom.sg.</p><p>/gen.sg.’ in which case we are dealing with Po.</p><p>89 The level of skeletal positions is conflated with the level of Ns for simplicity. All</p><p>three structures may be represented with an x-slot.</p><p>Polish as a CV language? 155</p><p>clarifies the dilemma at which level onsets see each other in interonset</p><p>government. Recall that once interonset is introduced into phonological</p><p>theory, adjacency defined at skeletal level is no longer valid. The two op-</p><p>tions we mentioned in the previous sections were either the level of onset</p><p>projection, or the melodic level. The effect of blocking RIO by the floating</p><p>melody unequivocally points to the latter level. The presence of vocalic</p><p>melody, whether associated or not, blocks this interaction. Given that gov-</p><p>erning relations between consonants are strictly related with their melodic</p><p>make-up it stands to reason that the interaction must take place at the me-</p><p>lodic level.</p><p>We are now ready to eliminated BrO completely from Polish phonology</p><p>and illustrate the representational difference between brać and bryzgać as</p><p>that between RIO and ONO, that is, a true and a false cluster.</p><p>(68) a. RIO b. ONO</p><p>O N O N O N O N</p><p>| | | | | |</p><p>b r È z g a t°Ç b e r a t°Ç</p><p>bryzgać ‘splash’ brać ‘take’</p><p>A true cluster is one which involves government between the consonants</p><p>(68a). Government may to some extent be viewed as a binding mechanism</p><p>which extends the domain of licensing. In other words, government, though</p><p>ontologically different from licensing, is de facto forming structures bigger</p><p>than one segment, whose individual players exist due to a single source of</p><p>licensing – the nucleus that directly follows the second consonant. Thus</p><p>true clusters may be compared to compounds in morphology.</p><p>A few words are in order concerning the ‘locked’ empty nucleus. At</p><p>this stage we assume that it is invisible to phonological processes, in that it</p><p>may not vocalize if followed by another empty nucleus, and may not cause</p><p>vocalization of the preceding empty nucleus. Additionally, as transpires</p><p>from the representation in (68a), it does not license its onset. All these</p><p>functions become available to the empty nucleus only once it is, or be-</p><p>comes unlocked.</p><p>The false cluster in (68b) contains an unlocked empty nucleus. Conse-</p><p>quently, it must be a licenser to its onset, and it is visible to all phonologi-</p><p>cal phenomena connected with nuclei. For example, it causes vocalization</p><p>of the jer in prefixes, e.g. zebrać [zebrat °Ç] ‘collect’ (69a), and is itself sub-</p><p>156 Formal complexity</p><p>ject to vocalization if followed by another visible empty nucleus, as in gra</p><p>/ gier [gra ~ g´er] ‘game, nom.sg. /gen.pl.’ (69c).</p><p>(69) a. b. c.</p><p>* *</p><p>O N1 - O N2 O N3 O N4 O N1 O N2 ~ O N1 O N2</p><p>| ↑ | | | | | | | | ↑ |</p><p>z e b e r a t°Ç g e r a g e r</p><p>[zebrat°Ç] [gra] [g´er]</p><p>Note that N1 in (69a) is also proposed to possess a floating melody now.</p><p>The melody is linked due to the universal (unviolable) constraint *P−P.</p><p>The melody under the nucleus N2 is proposed on the basis of the morpho-</p><p>logical alternations, e.g. bierze ‘(s)he takes’, rozbierać ‘undress’. N3 is the</p><p>only lexical full vowel in that form. On the other hand, N4 is a regular</p><p>empty nucleus.90 Such empty nuclei may remain unlocked not only word-</p><p>finally. Note that in words like kto ‘who’</p><p>the labial, it assumes the head position. Thus, one</p><p>13 This shift also occurred in other IE languages, e.g. Italic (Oscan and Umbrian *kw</p><p>> p), and to some extent in Greek.</p><p>14 The [p/b] variation in Welsh is due to lenition which is discussed in some detail</p><p>in section 4 below.</p><p>The Element Theory in Government Phonology 11</p><p>way to distinguish between primary and secondary articulation of conso-</p><p>nants is by referring to the status of the resonance element.15</p><p>A similar description can be offered for parallel shifts in Slavic. This</p><p>time the property that affects a velar consonant is the element (I), responsi-</p><p>ble for palatalization. Typically, three different types of velar palatalization</p><p>are mentioned in the literature on Polish. These are: surface velar palatali-</p><p>zation in which, the velar plosives [k, g] and the fricative [x] are palatal-</p><p>ized to [kj, gj, xj] before front vowels, as in bok – boki ‘side, nom.sg.</p><p>/nom.pl.’, noga – nogi ‘leg, nom.sg. /nom.pl.’, historia ‘history’; the so</p><p>called 1st velar palatalization (e.g. Gussmann 1978, 1980, Rubach 1981) in</p><p>which [k, g, x] alternate with palatal [t °S, Z, S], as in bok – boczek ‘side</p><p>nom.sg. /dim.’, noga – nóżka ‘leg, nom.sg. /dim.’, ucho – uszko ‘ear,</p><p>nom.sg. /dim.’; and the 2nd velar palatalization, occurring in the dative and</p><p>locative singular and producing alternations between [k, g, x] and [t °s, d °z, S]</p><p>respectively, as in rzeka – rzece ‘river, nom.sg. /loc.sg.’, noga – nodze ‘leg,</p><p>nom.sg. /loc.sg.’, mucha – musze ‘fly, nom.sg. /loc.sg.’.16 Ignoring the 2nd</p><p>velar palatalization in which the corresponding sounds have very little in</p><p>common, let us look closer at a possible representational contrasts between</p><p>ordinary velars, and those affected by surface and 1st velar palatalization</p><p>respectively. These contrasts may be given a similar interpretation to the</p><p>one involving the different degrees of labialization of velars in Celtic.</p><p>(9) velar vs. palatalized velar vs. palato-alveolar</p><p>[k] [kj] [tS]</p><p>(_) (I._) (I)</p><p>lok ‘hair lock’ loki ‘pl.’ loczek ‘dim.’</p><p>[lok] [lokji] [lot°Sek]</p><p>The plain velar is devoid of any secondary articulation. The palatalized</p><p>velar – through surface palatalization – contains the element (I) in operator</p><p>15 Another possibility that may be considered for the purpose of capturing secon-</p><p>dary articulation is connected with structural distinctions, for example, the use of</p><p>contour structures.</p><p>16 See Gussmann (1978) for arguments that the so called 2nd velar palatalization has</p><p>no synchronic reality as a phonological regularity, and Gussmann (1997b) for say-</p><p>ing the same about the 1st velar palatalization.</p><p>12 Substantive complexity</p><p>position. On the other hand the element (I) as the head produces a palato-</p><p>alveolar consonant which concomitantly undergoes affrication.17</p><p>Let us now turn to the remaining elements defining other dimensions in</p><p>the representations of consonants.</p><p>2.2.2. Manner</p><p>The manner dimension in consonants is defined by five elements of which</p><p>only two (/, h) can be called truly consonantal, in that they are not used in</p><p>vowels. This has been one of the reasons why the status of these elements</p><p>is shaky.18 As mentioned above, nasality, as well as high and low tones are</p><p>also used in vowel systems. The latter two will be discussed in more detail</p><p>in the following sub-section.</p><p>(10)</p><p>(/) = occluded, e.g. [p, t, k]</p><p>(h) = ‘noisy’, e.g. [s, S, x]</p><p>(N) = nasal, e.g. [n, m, N]</p><p>(H) = voiceless aspirated [ph, th]</p><p>(L) = fully voiced [b, d, g]</p><p>Each of the elements above deserves comment. The occlusion element is</p><p>assumed to be present in plosives but some researchers also place it in na-</p><p>sal consonants and laterals (Kaye, Lowenstamm and Vergnaud 1985, Har-</p><p>ris 1990). The noise element is assumed to be present in all released</p><p>stops.19 The status of nasality as an independent prime has been challenged</p><p>in the work of Nasukawa (1998, 2005) and Ploch (1999). Both researchers</p><p>attempt to merge nasality with low tone (L) in some way.</p><p>Leaving aside the laryngeal elements for the moment, let us observe how</p><p>some basic consonants may be represented by means of the manner elements</p><p>17 Some phonological reasons for this affrication, couched in terms of the Element</p><p>Theory, are provided in Cyran (1997: 214), Harris (1990: 270), Rennison (1998).</p><p>18 For discussion related to ‘stopness’ and ‘noise’ see e.g. Cyran (1996b), Golston</p><p>and van der Hulst (2000), Jensen (1994), Pöchtrager (2006), Ritter (1997).</p><p>19 This view is challenged in Cyran (1996b) who proposes that the noise element may</p><p>in some systems be completely missing even in released stops. We will return to this</p><p>idea shortly in the discussion of Irish clustering and Welsh consonant mutations.</p><p>The Element Theory in Government Phonology 13</p><p>just mentioned. The representations below only serve the purpose of illustrat-</p><p>ing how the Element Theory captures such phenomena as lenition.20</p><p>(11) lenition trajectory of the opening type</p><p>[p] > [f] > [w] > [P]</p><p>U U U</p><p>h h</p><p>/</p><p>Since each element on its own and each possible combination of elements</p><p>can be independently interpreted in production and perception, each of the</p><p>stages along the trajectory can be described as the effect of losing one pho-</p><p>nological prime, that is, decomposition. Thus there is a logical connection</p><p>between the fact that lenition is a weakening process and the idea that de-</p><p>composition leads to progressively less complex structures. Recall that</p><p>vowel reduction in unstressed position consists in precisely the same pro-</p><p>cedures though, admittedly, the contexts for consonantal lenition are dif-</p><p>ferent from those for vowel reduction. Nevertheless, we can describe both</p><p>contexts uniformly as prosodically weak (Harris 1997).</p><p>It is obvious now that sonority in Element Theory is the inverse of sub-</p><p>segmental complexity.21 The question is if complexity can successfully</p><p>replace sonority in all those aspects of phonology where the latter played a</p><p>central role. For one thing, it seems that the complexity scale captures the</p><p>lenition trajectory better than sonority. As noted by Harris (1996), if the</p><p>sonority hierarchy is anything to go by then we should expect nasals to</p><p>appear along the lenition trajectories of obstruents as they are more sono-</p><p>rous than, say, [p] or [f]. Secondly, it seems that complexity is able to solve</p><p>two apparent paradoxes connected with the weakening of consonants and</p><p>vowels. The first one concerns the fact that in terms of sonority the weak-</p><p>20 This discussion of lenition draws heavily on the work of Harris (1990, 1996,</p><p>1997) and Harris and Lindsey (1993, 1995). Note that so far we limit ourselves to a</p><p>discussion of the effects produced on a given segment, and little reference is made</p><p>to the link between lenition phenomena and the contexts in which they occur. The</p><p>typical sites for lenition or neutralization can be roughly defined as the intervocalic</p><p>and coda positions. The latter context is understood in a dramatically different way</p><p>in Government Phonology than in other current frameworks (see e.g. Kaye 1990,</p><p>Harris and Gussmann 1998).</p><p>21 See e.g. Rice (1992) for the reversed relationship between sonority and complex-</p><p>ity of structure.</p><p>14 Substantive complexity</p><p>ening of vowels, such as the rise and fall of jers in Slavic discussed above</p><p>([u/i] > [ъ/ь] > [P]), results in less and less sonorous objects, in contradis-</p><p>tinction to the weakening of consonants which results in more and more</p><p>sonorous ones. It is interesting that the sonorization of consonants ends</p><p>with a stage where the object is the least sonorous one, that is silence ([p] ></p><p>[f] > [w] > [P]).22 In terms of complexity, both phenomena receive a uni-</p><p>form interpretation. Simply, all stages of vowel weakening and consonant</p><p>lenition are of the same nature: depletion of melodic complexity.</p><p>The element-based analysis of lenition also bypasses the pertinent prob-</p><p>lem of major class feature changes.23 In this model, what remains as the</p><p>outcome of any decomposition process is as interpretable as the previous</p><p>stage, as shown in (11) above.</p><p>There are two more points to be made here. Firstly, in the model of rep-</p><p>resentations introduced in this section the range of possible processes that</p><p>a given segment may undergo is logically limited by its phonological struc-</p><p>ture. For example, a stop may either lose its release (h), be spirantized by</p><p>losing (/), debuccalized by losing the resonance element defining place,</p><p>voiced or devoiced. All these will be exemplified in section 5, when we</p><p>discuss consonant mutations in Welsh. Secondly, the pre-deletion stages</p><p>typically involve a simplex segment, for example, [h]=(h), [/]=(/), as well</p><p>as [w]=[U], [j]=(I), and [|]=(A), while their sonority values differ markedly</p><p>(Harris 1994: 122). Thus, elemental complexity offers a uniform account</p><p>of such phenomena in contradistinction to sonority scales.</p><p>In general, it appears that complexity can quite successfully replace so-</p><p>nority in lenition. On the other hand, complexity may replace another term</p><p>used with relation to lenition, and indeed syllabification, namely, strength.24</p><p>In Element Theory, the plosive seems to be the most complex and at the</p><p>same time the strongest consonant. This direct relation between complexity</p><p>22 I was made aware by Péter Szigetvári (p.c.) that the last point may be erroneous,</p><p>in that that net result of the last stage in the lenition trajectory is the most sonorous</p><p>stage, because what is left is the vocalic context flanking the consonantal position.</p><p>Though essentially true, this point does not diminish the merits of the complexity-</p><p>based treatment of lenition in any way.</p><p>23 For a critical evaluation of various proposals to deal with this issue see Harris</p><p>(1990, 1996).</p><p>24 The concept of strength has a long history in phonological theory. It typically</p><p>refers to inherent properties of segments which determine their behaviour in lenition</p><p>processes as well as phonotactics (e.g. Sievers 1901, Vennemann 1972, Hooper 1976,</p><p>Foley 1977, Murray 1988).</p><p>The Element Theory in Government Phonology 15</p><p>and strength follows from the internal representation rather than being</p><p>assumed in an arbitrary fashion on the basis of observation. In the follow-</p><p>ing sections and chapters it will be shown how strength defined as com-</p><p>plexity is exploited in syllabification. In the meantime, let us deal with the</p><p>last two elements, which define the laryngeal distinctions.</p><p>2.2.3. Source</p><p>The Element Theory uses only two elements to express all the possible</p><p>phonation types: (L) which is found in fully voiced obstruents, and (H)</p><p>which is found in voiceless fortis obstruents.25 It is assumed that laryngeal</p><p>specification is typically asymmetrical. For example, in a system like Eng-</p><p>lish, which exhibits voiceless aspirated stops as opposed to weakly voiced</p><p>ones, the opposition is expressed by marking the fortis series with the high</p><p>tone element (H), while the so called lenis series bears no laryngeal ele-</p><p>ment. In other words, the lenis obstruents are neutral. On the other hand,</p><p>languages like Polish in which the opposition among the obstruents is that</p><p>of fully voiced as opposed to voiceless, it is assumed that the voiced series</p><p>is the marked one and contains the low tone element (L), while the voice-</p><p>less series is unspecified.26 It follows then that from the phonological point</p><p>of view, the same phonological representation of, for example, neutral</p><p>stops, yields quite different phonetic results in Polish and in English. How-</p><p>ever, we must remember that the respective interpretations belong to two</p><p>distinct systems in which the neutral stop is perceived and produced with</p><p>sufficient phonetic difference from the series to which it is opposed in the</p><p>system. If the marked series is fully voiced, as in Polish, then the neutral</p><p>series tends towards the voiceless reflex, and conversely, if the opposite</p><p>series is voiceless then the neutral series tilts towards the voiced one.27 A</p><p>simple acoustic analysis of English and Polish plosives reveals that the</p><p>25 This description is deliberately simplified. The system of H/L tone elements is</p><p>also able to express more rare laryngeal articulations, e.g. Sahakyan (2006) demon-</p><p>strates that it is the ejectives in South-East Armenian and not the aspirated voice-</p><p>less stops that contain the high tone element.</p><p>26 For a discussion of the relationship between tone and voice see Matisoff (1973).</p><p>27 The term phonetic polarization may be used to describe this effect. This is remi-</p><p>niscent of the Dispersion Theory (Liljencrants and Lindblom 1972), which has re-</p><p>cently been harnessed into Optimality Theory in the form of SPACE constraints (e.g.</p><p>Flemming 1995, Ní Chiosáin and Padgett 2001).</p><p>16 Substantive complexity</p><p>supposedly distinct neutral series are very similar, thus supporting our views</p><p>on how the opposition should be represented.</p><p>One of the ways to define laryngeal distinctions in phonetics is by</p><p>means of Voice Onset Time, that is, VOT (Lisker and Abramson 1964).</p><p>This is the interval between the release of a stop and the start of a follow-</p><p>ing vowel. In general, the neutral obstruents in English have a short VOT</p><p>and a little voicing occurring before the release, to which we may refer as</p><p>VOT lead. The fortis series has a long VOT, also called VOT lag (e.g.</p><p>Harris 1994, Ladefoged 2001). On the other hand, the neutral series in</p><p>Polish and Spanish have a short VOT in the voiceless series, as opposed to</p><p>distinct voicing during closure, that is, a long VOT lead in the voiced se-</p><p>ries. Generally, the Element Theory assigns elemental representations to</p><p>the long VOT lead (L), and the long VOT lag (H), but no element defines</p><p>the short VOT type. The typology of phonation types in obstruents sup-</p><p>ports the view that the short VOT class is the unmarked one. For example,</p><p>if a system has only one series of stops it is typically voiceless unaspirated,</p><p>that is, having short VOT, or, in terms of elements, no laryngeal specifica-</p><p>tion.28 The majority of languages exhibit the two-way distinction of the two</p><p>main types: fully voiced vs. plain voiceless, and voiceless aspirated vs.</p><p>voiced. Let us look at a simple typology of laryngeal distinctions and see</p><p>how the Element Theory can capture the VOT distinctions. The typology is</p><p>based on Harris (1994), Ladefoged (2001) and Maddieson (1984). The</p><p>unmarked series of stops, with short VOT, and their elemental representa-</p><p>tion is represented as ‘_’, that is nothing.</p><p>(12) VOT opposition representation examples</p><p>Malakmalak _ (_) p</p><p>Spanish, Polish lead _ (L) , (_) b, p</p><p>English, Irish _ lag (_) , (H) b, ph</p><p>Thai lead _ lag (L), (_), (H) b, p, ph</p><p>Hindi lead _ lag, lead/lag (L), (_), (H), (LH) b, p, ph, bH</p><p>It seems that both the VOT and the element system share the ability to</p><p>capture one important aspect of the above typology, namely, that with the</p><p>increase of the number of contrasts, the number of VOT combinations and</p><p>28 In fact 98% of such systems in the UPSID data base show this tendency (Mad-</p><p>dieson 1984: 28).</p><p>The Element Theory in Government Phonology 17</p><p>the complexity of representations in terms of elements also increase.29</p><p>Thus, once again the relative markedness of particular systems goes hand</p><p>in hand with the relative complexity of representations. Both the acoustic</p><p>and elemental models of description have a neutral series in each system of</p><p>oppositions, and they seem to be able to directly express laryngeal neu-</p><p>tralizations</p><p>in a straightforward fashion: as the simplification of laryngeal</p><p>activity, giving rise to the unmarked variant. This advantage of privative</p><p>models over equipollent ones is well-established in phonological theory</p><p>(e.g. Lombardi 1995, Brockhaus 1995).</p><p>Let us look at how the phenomenon of obstruent devoicing is captured</p><p>in this model. As mentioned above, in Polish the voiced series of obstru-</p><p>ents is marked and bears the element (L), while the voiceless obstruents</p><p>have no specification.</p><p>(13) voice contrasts in Polish stops devoicing</p><p>[b] [p] [d] [t] [g] [k] [b] > [p]</p><p>U U A A _ _ U U</p><p>h h h h h h h h</p><p>/ / / / / / / /</p><p>L L L L</p><p>In an asymmetrical system of privative specification of voice, devoicing is</p><p>understood as delinking of the property responsible for voice due to licens-</p><p>ing failure in prosodically weak positions. Again, there is a direct relation</p><p>between the structural description of the phenomenon and the fact that we</p><p>are dealing with neutralization, or weakening. We do not attempt a full</p><p>analysis of devoicing in Polish here, suffice it to say that predominantly it</p><p>is due to the weak licensing that the obstruent receives in a particular con-</p><p>text, for example, word-finally.30</p><p>It appears then that Polish and English have quite different complexity</p><p>asymmetries in the representation of their obstruents. In the following sec-</p><p>tion we will look at one possible indication in the phonotactics of the two</p><p>languages which might directly fall out from the different laryngeal speci-</p><p>29 For a more advanced discussion of the relation between the Element Theory and</p><p>VOT types see Harris (1994: 133).</p><p>30 An exhaustive and satisfactory analysis of all the voice phenomena in Polish</p><p>within the Element Theory has not been proposed yet. For surveys of all the rele-</p><p>vant issues and recent feature-based analyses see Bethin (1992), Gussmann (1992)</p><p>and Rubach (1996).</p><p>18 Substantive complexity</p><p>fications employed in the two systems. More intricate complexity effects</p><p>will be described in the ensuing sections.</p><p>2.3. Complexity and syllabification</p><p>In the above discussion we saw how the concept of complexity is able to</p><p>capture a number of segmental phenomena, successfully replacing such</p><p>notions as sonority or strength. The advantage of complexity over the other</p><p>two concepts is that the scales of relative complexity fall out from the in-</p><p>ternal composition of segments and, therefore, are directly incorporated</p><p>into phonological processing, rather than being arbitrarily postulated as</p><p>look-up scales. Syllabification and phonotactic restrictions is another area</p><p>of phonology in which sonority and strength play an important role. The</p><p>aim of this and the following section is to demonstrate that complexity may</p><p>replace these constructs also here, and also provide some new insights into</p><p>the nature of syllabification.</p><p>In definitions of well-formed branching onsets or good syllable con-</p><p>tacts, that is, coda-onset clusters, the sonority profile plays an important</p><p>role (e.g. Selkirk 1982, 1984, Yip 1991, Itô 1986). A good coda-onset con-</p><p>tact is one in which the coda is more, or at least no less sonorous than the</p><p>following onset (e.g. Harris 1994). In models operating with strength of</p><p>segments (e.g. Vennemann 1972, 1988, Murray 1988), the preferred con-</p><p>tacts are similarly defined as those in which the strength differential be-</p><p>tween the coda and the following onset is greater, in favour of the latter.</p><p>The strength scale, however, is the inverse of sonority, therefore, the onset</p><p>will be stronger, or higher on the scale of strength, and the preceding coda</p><p>will be weaker.31 This is no place to introduce the syllabification principles</p><p>of Government Phonology. Suffice it to say that in terms of phonotactics it</p><p>is no different from sonority- or strength-based models, in that the best</p><p>contacts are those with the greatest complexity differential. For ease of</p><p>comparison with the other models, the most complex segments in the Ele-</p><p>ment Theory are obstruents, that is, they are the least sonorous in the for-</p><p>mer theory and the strongest in the latter.</p><p>Much stricter conditions constrain well-formed branching onsets. Here,</p><p>the condition of sufficient sonority distance is usually referred to in order</p><p>to account for the fact that onsets of the type [pl, kl], [pj, kj], or [tr, kr] are</p><p>better than [ks, pf, kn]. In fact, most of the latter group are normally viewed</p><p>31 This understanding of strength will be returned to in more detail in the following</p><p>chapter where we take up the problem of syllabification in Government Phonology.</p><p>The Element Theory in Government Phonology 19</p><p>as impossible onsets, at least in English. Thus, the best branching onsets</p><p>are those which involve an obstruent as the first element and a glide or</p><p>liquid as the second. What is required then is sufficient distance in terms of</p><p>sonority, strength, or complexity between the two consonants.</p><p>Below, we compare a fragment of the phonotactics in English and Pol-</p><p>ish, in which the preferences seem to be contradictory. While in the sonor-</p><p>ity and strength systems this problem cannot be solved without arbitrary</p><p>reshuffling of the scales, in the complexity-based model the facts fall out</p><p>directly from what we know about the representation of obstruents in the</p><p>two languages. Specifically, the differences will depend on the way the</p><p>laryngeal contrasts are specified.</p><p>Both English and Polish have branching onsets of the type [pr, br].</p><p>However, once we move down the scale of complexity of the other labial</p><p>obstruents in the two languages, we encounter restrictions to the effect that</p><p>while [vr] is a well-formed onset in Polish, for example, wrota ‘gate’, wróg</p><p>‘enemy’, wrona ‘crow’, in English this option is not utilized in native vo-</p><p>cabulary, except for the onomatopoeic vroom, or some obsolete forms and</p><p>French borrowings. On the other hand, while [fr] is a perfect branching</p><p>onset in English, for example, free, front, freak, etc., in Polish, words be-</p><p>ginning with this cluster are mostly borrowings, for example, fryzura ‘hair-</p><p>style, frytki ‘fries’, frykatywa ‘fricative’, frustracja ‘frustration’. Admit-</p><p>tedly, [fr] in Polish fares much better than [vr] in English, as most of the</p><p>borrowings are fully integrated into the language and one might even find</p><p>some forms which sound native, for example, fruwać ‘to fly’, which ap-</p><p>pears to be of onomatopoeic origin, like the English vroom.32</p><p>It seems that complexity as understood in the Element Theory may pro-</p><p>vide some rationale for these asymmetries between English and Polish. The</p><p>representations below are limited to the relevant labial obstruents and [r],</p><p>which is the second element of the branching onset.</p><p>(14) some English consonants some Polish consonants</p><p>[p] [b] [f] [v] [r] [b] [p] [v] [f] [r]</p><p>U U U U A U U U U A</p><p>h h h h h h h h</p><p>H H L L</p><p>32 The gap in native Polish vocabulary may be due to the fact that most of the mod-</p><p>ern instances of [f] are either borrowings or due to the devoicing of [v].</p><p>20 Substantive complexity</p><p>Recall, that the specification of the laryngeal contrasts in English involves</p><p>the presence of high tone in the voiceless obstruents, while in Polish the</p><p>voiceless series is unmarked. It transpires from the representations above</p><p>that [fr] in English is parallel to [vr] in Polish in terms of complexity dif-</p><p>ferential, an effect which in sonority-based accounts must result from arbi-</p><p>trary manipulation of the scale. In both languages preference is given to the</p><p>clusters with the greater complexity differential. Theoretically, neither</p><p>English [vr], nor Polish [fr] are completely illegal because there is some</p><p>complexity slope, but their ‘toned’ counterparts</p><p>are understandably pre-</p><p>ferred.33 In the following chapter the role of complexity in syllabification</p><p>will be defined in more detail. It is hoped that we will be able to provide an</p><p>answer to the question why clusters with identical complexity slopes (Eng-</p><p>lish [vr] and Polish [fr]) still show a different degree of acceptability. This</p><p>will be connected with conditions on syllable structure which are of more</p><p>importance than substantive constraints on well-formed onsets.</p><p>The following section discusses some complexity effects in modern</p><p>Irish in which we try to demonstrate the connection between phonotactics,</p><p>syllable structure, and phonological processes on the one hand, and sub-</p><p>segmental representations on the other.</p><p>3. Substantive complexity effects in Irish</p><p>33 One might wish to extend this analysis to another asymmetry in English, namely,</p><p>[Tr] vs. *[Dr], or [Sr] vs. *[Zr].</p><p>In this section, we bring together a few aspects of the phonological system</p><p>of Irish in order to demonstrate how the element-based model is employed</p><p>in concrete analyses of linguistic facts, and how various aspects of one</p><p>phonological system converge on the internal representation of its conso-</p><p>nants and vowels. Since the discussion is limited to substantive complexity</p><p>effects, some aspects of the data reviewed in this section will receive a</p><p>fuller interpretation once other principles of phonological organization are</p><p>introduced in the following chapters.</p><p>3.1. Features vs. elements in vocalic alternations</p><p>From the presentation of the Element Theory it follows that an element may</p><p>be equal to a segment, for example, (I) defines the vowel [i] on its own, while</p><p>some segments contain combinations of elements. In this respect, elements</p><p>Substantive complexity effects in Irish 21</p><p>are bigger units than features of the SPE type (Chomsky and Halle 1968).</p><p>Note that in order to get the same vowel in any feature-based model, we</p><p>need at least two features, for example [+HIGH] and [–BACK], neither of</p><p>which means anything in isolation, because the former defines all high,</p><p>while the latter refers to all non-back vowels. On the face of it, it seems</p><p>that feature systems are able to provide more precise and subtle descrip-</p><p>tions of phonological objects. The question however is if analyses in terms</p><p>of elements fail to cover the empirical facts, and, more importantly, if they</p><p>can account for the same phenomena better or worse than feature-based</p><p>systems. Let us briefly look at a comparison of two analyses of vowel qual-</p><p>ity alternations in Irish, one couched in the equipollent version of feature</p><p>specification (Ní Chiosáin 1994), and the other within the Element Theory.</p><p>In all dialects of Modern Irish consonants are grouped into two quality</p><p>series: palatalized and velarized.34 These consonants affect the preceding</p><p>phonologically short vowels by spreading their secondary articulation prop-</p><p>erty.35 In the data below (C) refers to Connemara and (M) to Munster Irish.</p><p>(15)</p><p>a. (u ~ i) [muk] ~ [mik´] muc / muic ‘pig / dat.’ (C,M)</p><p>b. (o ~ e) [sop] ~ [sep´] sop / soip ‘wisp / gen.sg.’ (C)</p><p>c. (o ~ i) [sop] ~ [sip´] sop / soip ‘wisp / gen.sg.’ (M)</p><p>[kod´] ~ [kid´] coda / cuid ‘portion, gen.sg./nom.’ (C,M)</p><p>d. (a ~ i) [f´ar] ~ [f´ir´] fear / fir ‘man / gen.sg.’ (M)</p><p>[f´œ:r] ~ [f´ir´] fear / fir ‘man / gen.sg.’ (C)</p><p>e. (a ~ e) [d´as] ~ [d´eS´] deas / deise ‘nice / gen.sg.’ (M)</p><p>[d´œ:s] ~ [d´eS´] deas / deise ‘nice / gen.sg.’ (C)</p><p>Although the preceding onset is not unimportant, for the sake of simplicity</p><p>we will limit the discussion to the context VC, in which the quality of the</p><p>consonant affects the nucleus to its left.</p><p>34 The distinction palatalized vs. velarized is typically represented as C´ vs. C. The</p><p>consonant inventory of Irish, with a degree of simplification, is as follows: Labial</p><p>(p, p´, b, b´, f, f´, v, v´, m, m´), Coronal (t, t´, d, d´, s, S, n, n´, l, l´, r, r´), Velar (k,</p><p>k´, g, g´, x, x´, V, V´, N, N´), Glottal (h, h´).</p><p>35 Consonants also affect the following vowels although on a smaller scale. This</p><p>effect may to some extent be called phonetic. See Ní Chiosáin (1991) and Bloch-</p><p>Rozmej (1998) for thorough analyses of these effects in Connemara Irish, and</p><p>Cyran (1995, 1997) for the Munster dialect.</p><p>22 Substantive complexity</p><p>The alternation [u~i] is the most regular across all dialects, and the ef-</p><p>fects are identical in Munster and Connemara. The alternation [o~e] is</p><p>rather limited to Connemara Irish, and the corresponding alternation in</p><p>Munster is that of [o~i].36 Nevertheless the [o~i] alternation is also found</p><p>in the descriptions of western dialects, for example, in Connemara (de</p><p>Bhaldraithe 1945). As for (15d), the alternation may be said to be identical</p><p>in both dialects, despite the difference in the pronunciation of the stressed</p><p>[a] which comes out as [œ:] in Connemara. Similarly, the alternation [a~e]</p><p>seems to be analogous in the two dialects. However, this alternation is</p><p>highly conditioned. To obtain [e], the nucleus must be flanked by palatal-</p><p>ized consonants on both sides, and, additionally be followed by a schwa</p><p>vowel in the following nucleus.</p><p>In Ní Chiosáin (1994), the alternations illustrated above are due to</p><p>spreading of the feature [±BACK] from the consonants into the nucleus.</p><p>[−BACK] and [+BACK] define the palatalized and the velarized consonants</p><p>respectively. Short nuclei, which are the targets of the spreading, are un-</p><p>derspecified for backness. Ní Chiosáin proposes that the inventory of short</p><p>vowels involves only three objects: two underspecified ones, that is, [I] and</p><p>[E] which correspond to high and mid vowels, and a low [A] which has a</p><p>phonetic variant [a/œ:] after a palatalized onset (C´a/œ:). Thus, Ní Chi-</p><p>osáin predicts that only high and mid vowels are targets of backness</p><p>spreading, while the alternations in (15d) and (15e), which involve manipu-</p><p>lation of height, are not part of the rule Spread [BACK]. This move is, of</p><p>course, logical. It is difficult to expect that spreading of [±BACK] should</p><p>cause changes in height. However, the effects involving the low vowels</p><p>occur in exactly the same phonological contexts as the high and mid vowel</p><p>alternations, that is, when the quality of the following consonant changes</p><p>from [+BACK] to [−BACK]. Thus, to capture this fact one would need to</p><p>refer to backness to account for height shifts, despite the lack of formal</p><p>connection between the two phonological dimensions. Another prediction</p><p>that the feature-based analysis makes is that the vocalic alternations take</p><p>place in a symmetrical fashion along the same height. This idea is repre-</p><p>36 It must be added that both dialects exhibit opaque vowels which are not affected</p><p>by the property of the following consonant. For example, scoil [skol´] ‘school’, cois</p><p>[koS] ‘leg’, rather than the expected *[skel´ / skil´] or *[keS/ kiS] (e.g. Ó Cuív 1975,</p><p>de Bhaldraithe 1945). Similar behaviour concerns the back low vowel [A]. For</p><p>example, bainne ‘milk’ is pronounced as [bAn´´] in Munster and [bA:N´´] in Con-</p><p>nemara ([N] stands for a tense coronal nasal). The lengthening in Connemara is</p><p>phonetic.</p><p>Substantive complexity effects in Irish 23</p><p>sented graphically below in (16a). On the other hand, (16b) shows the di-</p><p>rections of changes which transpire from the data in (15) above. Admit-</p><p>tedly, they concern Munster Irish to a greater extent, but they occur in both</p><p>dialects. This suggests that, some small differences notwithstanding, the</p><p>two dialects should be offered a uniform analysis, in which we would be</p><p>able to incorporate height as well as backness.</p><p>(16) a. Spread [±BACK] b. Munster Irish</p><p>i I u i u</p><p>e E o e o</p><p>a A a A</p><p>The illustrations above clearly suggest that the analysis based on spreading</p>
  • (a) Interdiscursividade é o termo que explica o uso que um texto faz de discursos, a ele anteriores, ou contemporâneos Há duas maneiras de uso de outros discursos
  • (44) 9 9 16 2-8 92 8 Apresente brevemente o tema da Reforma Protestante e destaque a importância de se compreender este evento histórico (1)
  • Apresente brevemente o tema da Reforma Protestante e destaque a importância de se compreender este evento histórico (2)
  • Assim diz o SENHOR dos Exércitos, o Deus de Israel Ajuntai os vossos holocaustos aos vossos sacrifícios e comei carne (1)
  • Assim diz o SENHOR dos Exércitos, o Deus de Israel Ajuntai os vossos holocaustos aos vossos sacrifícios e comei carne (2)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - HISTÓRIA DA IGREJA I - 54_2024 (1)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - HISTÓRIA DA IGREJA I - 54_2024 (2)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - HISTÓRIA DA IGREJA I - 54_2024 (3)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - HISTÓRIA DA IGREJA I - 54_2024 (4)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - HISTÓRIA DA IGREJA I - 54_2024 (5)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - MÉTODOS DE ESTUDOS BÍBLICOS NO ANTIGO TESTAMENTO - 54_2024 (1)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - MÉTODOS DE ESTUDOS BÍBLICOS NO ANTIGO TESTAMENTO - 54_2024 (2)
  • (44) 9 9 16 2-8 92 8 ATIVIDADE 1 - TEOL - MÉTODOS DE ESTUDOS BÍBLICOS NO ANTIGO TESTAMENTO - 54_2024 (3)
  • Para a Sociolinguística, toda língua natural poderia ser conceituada como o “conjunto de realizações possíveis dos recursos expressivos que estão à...
  • São exemplos de gramaticalização no Português do Brasil (PB), EXCETO:Grupo de escolhas da perguntaA constituição do sufixo formador de advérbios...
  • Jakobson (1987) revolucionou o mundo da linguística ao apresentar seis funções da linguagem, baseadas nos seis fatores envolvidos na comunicação: O...
  • Em qual das situações abaixo observamos que está acontecendo a escuta ativa? Assinale a alternativa correta. Questão 3 Escolha uma opção: Em uma re...
  • Observe a seguinte situação: “Em uma reunião pedagógica, os professores têm a oportunidade de expressar suas ideias, sugestões e preocupações. Ele...
  • Da formação de ídolo + latria (= adoração), chegamos ao vocábulo “idolatria” pela supressão de segmento semelhante, uma haplologia.
  • O coach tem conhecimento técnico e de software que O corte tem conhecimento técnico e de software que
  • Alfabetização língua escrita e normal culta vivemos numa sociedade tradicional pouco letra nesse sentido bastaria lembrar que temos ainda por volta...
  • É um instrumento através do qual a comunicação se realiza em uma ação de comunicação apresentamos nossos conceitos ideias pensamentos enfim nossa v...
  • Segundo o texto da Unidade I, por que a Bioética deve ser estudada de forma interdisciplinar?
  • Assinale a alternativa que completa o parágrafo seguinte: Falar que os gêneros são entidades comunicativas que refletem as necessidades de comunica...
  • Faça a relação entre a definição de estudo de texto e as fases para se chegar à disciplina LT de nossos dias. A- 1° momento – Análise Transfrástica...
  • Que tipo de conexão foi aplicada na frase: Como o sol não costuma dar trégua, as praias são sempre uma ótima opção. (Anúncio de uma Agência de Viag...
  • refucc81gio sp as histocc81rias de refugiados e imigrantes em sacc83o paulo joacc83o vieira 31326625
  • Apostila Português
epdf pub complexity-scales-and-licensing-in-phonology-studi - Linguística (2024)

References

Top Articles
Latest Posts
Recommended Articles
Article information

Author: Lilliana Bartoletti

Last Updated:

Views: 6451

Rating: 4.2 / 5 (53 voted)

Reviews: 84% of readers found this page helpful

Author information

Name: Lilliana Bartoletti

Birthday: 1999-11-18

Address: 58866 Tricia Spurs, North Melvinberg, HI 91346-3774

Phone: +50616620367928

Job: Real-Estate Liaison

Hobby: Graffiti, Astronomy, Handball, Magic, Origami, Fashion, Foreign language learning

Introduction: My name is Lilliana Bartoletti, I am a adventurous, pleasant, shiny, beautiful, handsome, zealous, tasty person who loves writing and wants to share my knowledge and understanding with you.