hyman_2006_prosodic_typology

Word-prosodic typology* Larry M. Hyman University of California, Berkeley Numerous proposals have been advanced as to how prosodic systems should be typologised. In this paper I distinguish two prototype systems, tone and stress accent, which, unlike systems analysed as ‘pitch accent’, have two inviolable, definitional properties: (i) obligatoriness (every word has at least one stress ac- cent) ; (ii) syllable-dependency (the stress-bearing unit is necessarily the syllable). In contrast, the oft-cited criterion of culminativity (every word has at most one tone/accent) not only includes tone/accent systems that are neither obligatory nor syllable-dependent, but also culminative non-prosodic features that are clearly not accentual. I argue that there is no one pitch-accent prototype. Instead, since tone and stress accent may co-occur, and since languages may ‘pick and choose’ between the non-definitional properties that tend to cluster within the tone vs. stress-accent prototypes, there is a range of intermediate (and possibly inde- terminate) word-prosodic systems which may or may not be best seen as ‘types’. If stress is monarchic, and length oligarchic, we may say that tone is democratic. (Greenberg & Kaschube 1976: 9) 1 Introduction Over the past several decades, many linguists have made pronouncements on how prosodic systems should be typologised. While no other area of phonological typology has attracted as much attention, it is not clear how much consensus there is on where things currently stand. Linguists sometimes speak or write on the subject with great conviction, but the views that are expressed often conflict on both general and specific issues, e.g. whether all human languages must have ‘stress’, whether the prosodic system of Tokyo Japanese is ‘tonal ’ vs. ‘accentual’, and so forth. In this paper I attempt to sort out some of the complexities of word-prosodic typology. I begin in §2 with a brief discussion of problems inherent in doing typology, particularly when applied to phonology. In the next two sections I give brief definitions of tone (§3) and stress accent (§4), followed * This paper was first presented as an invited talk at the conferenceBetween Stress and Tone (BeST). I would like to thank Bert Remijsen and Vincent van Heuven for the invitation and those in attendance for their helpful comments. I have particularly profited from extended discussions with Carlos Gussenhoven and Jose´ Hualde on the tone vs. accent question and good Bantu exchanges with David Odden and Ge´rard Philippson. I also would like to thank the editors and anonymous reviewers for their very thoughtful comments on the original manuscript. Phonology 23 (2006) 225–257. f 2006 Cambridge University Press doi:10.1017/S0952675706000893 Printed in the United Kingdom 225 by a longer section (§5) which addresses the more problematic notions of ‘accent’ and ‘pitch accent’. I argue that the highest-level typological cut consists of identifying two prototypes, one with stress accent, the other with tone. While it is possible to define these prototypes and establish a set of properties that typically co-occur in each, establishing a third pitch- accent prototype is more elusive. This is, in part, because languages which have been identified as pitch accent freely ‘pick and choose’ between the prototypical properties of stress-accent systems vs. tone systems. In other words, there are many intermediate word-prosodic systems which are not best seen as discrete types. While most of the discussion centres around typological issues resulting from structural or systemic difference, §6 ad- dresses the role of phonetics in typologising prosodic systems. §7 provides a brief conclusion. 2 Typology Since the concern of this study is to address the typology of word-prosodic systems, it seems appropriate to begin by asking the questions: What is typology? Phonological typology? Word-prosodic typology? Concerning the first question, Hage`ge (1992: 7) provides a traditional definition of typology as ‘a principled way of classifying the languages of the world by the most significant properties which distinguish one from another’. Vajda (2001) answers the second question in a similar fashion: ‘ it is possible to classify languages according to the phonemes they containº typology is the study of structural features across languages. Phonological typology involves comparing languages according to the number or type of sounds they contain’. It is significant that the above definitions speak of classifying languages rather than subsystems of languages. In any case, only certain parameters impress linguists enough to establish language types: within phonology, there is a class of ‘click languages’, but not ‘ implosive languages’, ‘open syllable languages’, but not ‘onset languages’ (meaning that every syllable has to have an onset). The question is whether there is any reality in classifying languages rather than analysing the properties of the relevant subsystems. This issue is particularly pronounced in the area of prosody, where the practice of distinguishing ‘stress languages’ from ‘tone languages’ is well established and has encouraged some to propose additional types: ‘pitch-accent language’, ‘restrictive tone language’ and so forth. However, as Greenberg (1974: 14) notes, ‘ the same data can be utilized either for a typology of linguistic properties or a typology of individual languages’. A major aim of this study is to show that there is considerably more diversity in prosodic systems than such labels have thus far distinguished. Most phonologists would probably agree that there is little, if any, difference between doing phonological typology and doing phonological theory. At the very least, it is unlikely that one can do insightful typology without addressing the same analytical issues that confront phonological 226 Larry M. Hyman theory. In order to appreciate the complexity involved in doing proper typology, consider the question of what phonologists might seek to typo- logise. There are at least three dimensions to this issue: (i) What are the properties to be typologised? Segments? Systems? Rules? (ii) At what level should the properties be typologised? Underlying (morphophonemic)? Lexical (phonemic)? Surface (phonetic)? (iii) Within what domain should the properties be typologised? The morpheme? Word? Phrase? For many phonological properties, one will obtain quite different re- sults depending on the answers to the above questions. In some languages, word-level prosody is predictable by rule (e.g. demarcative stress, mor- phological tone) and hence is not present in underlying representations (URs). In other languages, the prosodic properties are introduced only at the phrase level. In this study I take the following approach to the above three questions: (i) The properties to be typologised are linguistic systems. Since there is considerable room for interpretation, care must be taken to distinguish the descriptive ‘facts’ of linguistic systems from the analyses that attempt to account for them. On the other hand, a simple inventory is not a linguistic system. (ii) Linguistic systems necessarily capture the relationship between underlying, lexical and phonetic representations. However, this study is primarily concerned with properties which are present at the lexical level, whether from URs or introduced by rule. (iii) As seen in the title, this study is limited to prosodic properties which hold within the word domain vs. properties which make reference to the phrase or utterance levels. Among the frequent assumptions underlying past attempts at phono- logical typology are: (i) EXHAUSTIVITY: all systems can be assigned to a type; (ii) UNIQUENESS: no system can be assigned to two types; (iii) DISCRETENESS: types do not overlap. By these criteria we arrive at the classic bi- and trifurcation of prosodic types in (1a, b). (1) a. Classic bi- and trifurcation of prosodic types pitch accent stress accent b. tonestress accent pitch accent c. tone* tonestress accent Word-prosodic typology 227 In (1a), prosodic systems are identified as either stress accent (SA) or tone (T), while (1b) allows a third category, pitch accent (PA). As indicated by the asterisk, the above assumptions do not condone (1c), where PA belongs simultaneously to two different types, the first with SA, the second with T. As attractive as they are, all three assumptions are problematic. First, despite the discreteness of the putative types (SA, PA, T), prosodic sys- tems may be ambiguous, i.e. analytically indeterminate. In languages such as Kinga (Schadeberg 1973) and Nubi (Gussenhoven 2006), every word must have one and only one high tone. Should such systems be categorised as ‘restricted’ T, as PA or as SA, where stress would be realised simply as H pitch? All three interpretations seem reasonable, perhaps not only to linguists but also to the native speakers. Given this indeterminacy, it may be that such languages provide the transitional ‘pivot’ for change from one type of system to another (cf. Salmons 1992, Gussenhoven 2006). Besides ambiguity, a second issue is that prosodic systems may be mixed. In Saramaccan (Good 2004), in one class of words a single syllable is marked /H/, the remaining syllables being underlyingly toneless. On the other hand, in a second class of words, all syllables must be specified for tone (all H, all L or combinations of H and L). While Saramaccan has a stratified lexicon, which Good analyses as PA vs. T, the same words may have both properties in Cherokee. As analysed by Johnson (2005), a Cherokee word may or may not have a single L*H pitch accent. Whether or not it does, a non-accented syllable may also have a contrastive H or L tone. While Saramaccan and Cherokee combine PA and T, other languages combine SA and T (see §5). Whatever its usefulness, the three- way distinction between SA, PA and T cannot therefore be a discrete, non-overlapping one.1 A final word before addressing SA, PA and T in more detail. The above discussion assumes that word-prosodic typology is limited to this tri- chotomy, or the properties it encodes. We have however not explicitly addressed the question of what counts as word prosody. Words are not only characterised by the classic three suprasegmentals of tone, length and stress (Lehiste 1970, Greenberg & Kaschube 1976), but also by various kinds of harmonies and quantitative/qualitative distributional constraints concerning minimum/maximum mora/syllable count, initial vs. final position restrictions, C/V combinatorics and so forth. If word- prosodic typology were to take all of these issues into account as well, the result would be an explosion of disparate prosodic word types, as indicated in (2). 1 Another kind of mixed system has been occasionally proposed, based on different levels of representation. Although subject to other interpretations, McCawley (1978: 126) suggests that Luganda ‘can be described as having a pitch-accent sys- tem in its deep phonology and a tonal system in its surface phonology’. 228 Larry M. Hyman (2) a. The demarcative word A property marks the beginning or end of the word. b. The culminative word A property occurs only once per word. c. The harmonic word A property is realised throughout the word. d. The metrical word A word consists of hierarchically arrayed moras or syllables. e. The minimum word A word must consist of a minimum number of moras or syllables. f. The maximum word A word can consist of a maximum number of moras or syllables. g. The phonotactic word A word permits only certain output segments/sequences. h. The morphophonotactic word A word permits only certain input segments/sequences. Di‰erent kinds of prosodic word? Beyond practicality there are legitimate reasons to group tone, length and stress together. First, there is the intuitive recognition of all three as overlaid suprasegmentals (Lehiste 1970), whose phonetic variations (F0, duration, intensity) can be realised not only in spoken language, but also on a musical instrument. Second, these three phonetic properties fre- quently interact, e.g. in the realisation of stress and intonation. Finally, tone, length and stress also share functional similarities, e.g. culminativity and rhythmic alternations. For the above reasons, the present study on word-prosodic systems will be limited to T, SA and PA, and their manipulation of F0, duration and intensity. In the following sections, tone, stress and (pitch) accent are taken up in turn. 3 Tone In order to establish prosodic types in a reasonably objective way, it is necessary to provide definitions for tone, stress and accent. I will assume the following as a working definition of tone (Hyman 2001a: 1367, slightly modified from Welmers 1959, 1973): (3) A language with tone is one in which an indication of pitch enters into the lexical realisation of at least some morphemes. As indicated, the domain of tone is the morpheme. Depending on the language, morphemes with tone may be all, many or few. They can be Word-prosodic typology 229 lexical or grammatical morphemes, including tonal morphemes which lack a segmental content. According to (3), the presence of tone is estab- lished on the basis of lexical realisation, i.e. the output of the lexical pho- nology. Tone may thus either be present in URs or be introduced by rule. If this latter occurs at the lexical level, rather than postlexically, this would count as tone by the above definition. Tone is featural and paradigmatic: in a prototypical tone system such as Cantonese or Yoruba, one IDENTIFIES the tone of each tone-bearing unit (TBU). In other words, tone has a ‘distinctive function’ (Martinet 1960). The tone system may consist of a simple binary opposition, e.g. /H, L/, or involve more levels, contour tones and specific tones accompanied by phonation types (glottalisation, breathiness, etc.). The definition in (3) refers to ‘an indication of pitch’ rather than pitch itself. This is intentional, and is designed to extract away from analytic preferences. For example, it was mentioned in §2 that Kinga and Nubi require exactly one H tone per lexical word. The question was also raised as to whether these systems should be thought of as SA, PA or T. They could thus be analysed with a /H/ tonal feature or an abstract accent, which could be said to be realised as [H] by a postlexical pitch-assignment rule. However, since only pitch is involved, it is clear that the abstract accent is nothing but ‘an indication of pitch’. By (3), Kinga and Nubi are tonal. The question that is relevant in this context is whether we want to ty- pologise according to the properties of prosodic systems or according to the analyses given to them by diverse linguists. This issue has arisen in a number of cases, the most prominent of which concerns Tokyo Japanese. As seen in (4), Tokyo Japanese has been analysed both accentually and tonally (Haraguchi 1977, McCawley 1978, Poser 1984, Pierrehumbert & Beckman 1988, etc.) : (4) accentual tonal output output with ª ‘pillow+nom’ ma0kura ga ‘heart+nom’ ‘head+nom’ ‘fish+nom’ koko0ro ga atama0 ga sakana ga makura ga kokoro ga atama ga sakana ga mákùrà gà mákùrà gà H kókórò gà kòkórò gà átámá gà àtámá gà sákáná gá sàkáná gá HH In the accentual account in the first row, the mark 0 indicates where a drop occurs between H and L. (Others have used a tick mark or asterisk.) The tonal analysis in the second row starts with a /H/ prelinked to the mora that precedes the pitch drop. The third and fourth rows show that moras following the accent or /H/ are L, while the remaining moras are H, except where an initialx boundary tone affects the first mora. A final note concerning the notion of ‘word tone’ (Donohue 1997, Mazaudon 2005). In the definition in (3) tone is identified with the 230 Larry M. Hyman morpheme, and nothing is said about the TBU, which can be the syllable or the mora. In Tamang, Kukuya and a number of other languages, there is a fixed number of word- (or stem-) tone patterns independent of the number of syllables or moras (cf. the discussion of Skou in §5). For such systems it is sometimes suggested that the word takes precedence over the morpheme, or is even the TBU, rather than the domain determining the tone (cf. Garde’s 1968: 12–13 distinction between ‘unite´ accentuable ’ and ‘unite´ accentuelle ’ applied to stress systems). Like Tokyo Japanese, there is a sense in which such languages diverge from what I have called proto- typical tone systems. As in the case of vowel-harmony systems, such tone systems have a syntagmatic dimension which is lacking in Cantonese, Yoruba, etc., where tone contrasts paradigmatically on syllables. 4 Stress Whereas tone has to do with pitch features, the general approach to stress has been that it has to do with metrical PROMINENCE. The definition I would propose for stress accent is given in (5). (5) A language with stress accent is one in which there is an indication of word-level metrical structure meeting the following two central criteria: a. obligatoriness: every lexical word has at least one syllable marked for the highest degree of metrical prominence (primary stress); b. culminativity: every lexical word has at most one syllable marked for the highest degree of metrical prominence. In many characterisations of SA, the phrase ‘one and only one’ primary stress is used: ‘every prosodic word contains one and only one head foot, which is the locus of main stress. The existence and uniqueness of the head foot are usually taken to be axiomatic – universal properties of GEN rather than violable constraints’ (McCarthy 2003: 110). As indicated in (5), it is however fruitful to separate the phrase into two properties : obligatoriness and culminativity. Obligatoriness states that every lexical word has to have a primary stress. Although there is variation in the use of the term in the literature, culminativity is intended here to mean that a lexical word cannot have more than one primary stress. In contrast to tone, which is featural and paradigmatic, stress is struc- tural and syntagmatic. Prototypical tone has a distinctive function, while stress accent has a CONTRASTIVE function (Martinet 1960), even where minimal pairs are possible. Whereas one IDENTIFIES the T of each TBU, one LOCATES stress within the lexical word domain. Of the two criteria in (5), obligatoriness is the more important. It is an absolute universal – DEFINITIONAL – of a SA system, which requires that an obligatorily headed metrical constituent be built at the word level. Rather than HEAD(PWd), ‘which says that each phonological word must have a unique head and therefore exactly one accent’ (McCarthy Word-prosodic typology 231 2002: 78), and which thereby conflates obligatoriness and culminativity, I shall borrow from optimality syntax and refer to obligatoriness as OBLIGATORYHEAD (OBLHEAD). Because of obligatoriness, no language fails to assign metrical stress to a word for lack of a specific phonological property. Thus OBLHEAD rules out stress systems such as the following:2 (6) a. Stress the penultimate syllable; monosyllabic words are stressless. b. Stress the last heavy syllable; words that lack a heavy syllable are stressless.3 c. Stress the initial syllable if it has an onset; words beginning with a vowel are stressless. d. Stress the leftmost syllable with /a/, otherwise stress the leftmost syllable containing /e/ or /o/; words with only /i/ or /u/ are stressless. e. Stress the last H tone; words lacking a H tone are stressless What’s wrong with the above systems is that they do not exercise default stress assignment in case stress-attracting phonological properties are lacking. Except for this property, (6a–e) would seem to be a priori reasonable, if not also grounded: stress is often avoided on light syllables, so why not on