A theory of phonological weight By Larry M. Hyman (review)

REVIEWS A theory ofphonological weight. By Larry M. Hyman. (Publications in language sciences, 19.) Dordrecht, Holland, & Cinnaminson, NJ: Foris, 1985. Pp. 136. /40.00. Reviewed by David Odden, Ohio State University TPW is interesting because it presents many provocative ideas, but frustrating because these ideas are not always fully developed and integrated into the theory—and because, in numerous cases, Hyman fails to present any rule which captures the generalization under discussion. Chap. 1 considers two unexplained universals of syllable weight and their treatment in the CV Phonology of Clements & Keyser 1983. One is that a language may treat VC rimes as heavy only if it does the same for VV rimes. A second is that onsets are irrelevant for stress.1 Hyman points out that—in theories with rime, nucleus, and margin—onset irrelevance must be stipulated (usually by projecting only rimes). Further, the notion 'branching rime' cannot be defined simply in terms of a characteristic geometry, since the term technically means branching rime or branching nucleus. Hyman gives the following for ta: and tam: (D R R I G\ N NM N I I W VC I/ I I a am He points out that this problem does not arise with a nucleus node: a heavy syllable (for stress) is one with a branching nucleus. If rime Cs make a syllable heavy, final Cs must be in the nucleus (Vs are always in the nucleus, so vowel length makes a syllable heavy before consonant clustering does). The task which Hyman undertakes is to provide a better explanation ofproblems relating to 'weight'.2 Chap. 2 presents Hyman's theory. His basic concept is the Wfeight] U[nit], represented as X. Each segment contributes one WU, and the universal Ofnset] C[reation] R[ule] fuses a sequence of C and V into one mora: (2) (g).,.-X L--'-'" I [ + cons] [-cons] An underlying sequence ta has two WU's which are reduced by the OCR to 1 But stress in Pirahä is partly conditioned by the presence and voicing ofthe onset C (cf. Everett & Everett 1984). 2 What counts as 'weight-related' is by no means obvious, and Hyman has no clear suggestions. It is not even certain that 'weight' is a uniform property: in Classical Greek, stress placement counts only Vs, but poetic meter treats both long Vs and consonantally closed syllables as heavy. 669 670LANGUAGE, VOLUME 62, NUMBER 3 (1986) one; hence onsets cannot count as weight-bearing. At the same time, the OCR explains the common observation that, in a VCV sequence, the C universally syllabifies with the following V.3 In addition to the OCR, language-specific rules are allowed which fuse one or more WU's; a typical case is a languagespecific M[argin] C[reation] R[ule], which fuses a V and a following obstruent into a single WU. Such a rule accounts for a language where obstruents in the rime are not weight-bearing. Chap. 3 gives Hyman's program for predicting phonetic syllabicity without a C/V distinction, or even syllables. He proposes that the most sonorous element dominated by a WU is the 'bearer' of syllabicity. Thus, in a syllable ba, the syllabicity of a (and the non-syllabicity of b) follow from the fact that the two segments constitute a single beat (via the OCR) and from the fact that a is the most sonorous member of that beat. The problem with this program for predicting syllabicity from sonority is that contrasts like Eng. 'ear' [ir] vs. (reduced) 'your' [yr] cannot be represented. The segmental make-up is the same, but somehow syllabic r must be diacritically 'more sonorous' than nonsyllabic r. Again contrasting WU theory with CV Phonology, Hyman argues that, while his theory cannot contrast weight-bearing non-syllabic vs. syllabic Cs (being syllabic means being weight-bearing), CV theory could contrast V-dominated vs. C-dominated nucleus segments. If syllabicity and phonological weight were interchangeable, as Hyman suggests, then WU theory would score points over CV Phonology in restrictiveness. But the contrast between C and V is well supported in Turkish, as discussed by Clements & Keyser—who show that long Vs may (lexically) be VCdominated or VV...