In all human languages, we find words that are similar in some respect or another – usually because they are morphologically related. If speakers wish to determine the degree of similarity between two morphologically related forms, they must decide which elements to compare. Perhaps the most intuitive approach would be to compare the segments, and determine the degree of featural similarity they display. This approach might be thought of as ‘segmental comparison’. A schematic representation for two forms undergoing segmental comparison is given below.

formula here

An equally plausible although less intuitive alternative is also available – prosodic elements could be compared. For example, the most prominent syllable node of each form might be compared, or the main-stressed nucleus of each form might be compared. This approach might be called ‘prosodic comparison’. Importantly, prosodic comparison might result in the comparison of elements that are associated with very different segmental melodies – for example, imagine a pair of words in which a given prosodic role (for example, head of prosodic word) has a different location in each form. Such a situation might occur in a language where stress shifts under affixation; in such a case, the linear order of segments would remain the same in both words, but the location of the prosodic heads would differ. A schematic representation of this sort of situation is given below.

formula here

In this paper, I will consider just such a language – the Saipanese dialect of Chamorro – and demonstrate that the phonology of this language requires both segmental and prosodic comparison. This characteristic of the sound pattern of Chamorro holds great significance for theories which attempt to formalise phonological similarity effects, such as Steriade's (1996) theory of Paradigm Uniformity, Kenstowicz's (1995) theory of Uniform Exponence and McCarthy & Prince's theory of Correspondence (McCarthy & Prince 1993, 1994, 1995, McCarthy 1995). For simplicity of exposition, from this point on my analysis will be cast in terms of the theoretical framework of correspondence (discussed below). It should be remembered, however, that the main result of this analysis – the existence of prosodic phonological comparison – is relevant to any theory which seeks to give a comprehensive treatment of phonological similarity effects.

Developments in phonological theory have led to the recognition that phonological representations have a layered constituent structure. Many, perhaps all, of these constituents contain elements which can be identified as heads. Heads enter into various kinds of relations with their dependents. In this article, we identify a phenomenon which is quite pervasive in every part of phonology which has heads and dependents, namely, the existence of head–dependent asymmetries (henceforth HDAs). While various particular manifestations of these asymmetries are well known and have been much studied, this is the first attempt, to our knowledge, to unite a broad range of seemingly different phenomena under one heading. We identify various types of HDAs, and propose constraints on possible HDAs.

Most importantly, we distinguish between HDAs that involve complexity, and those that involve visibility. These have properties which potentially contradict each other. We propose that they apply in fundamentally different types of cases: unlike complexity HDAs, visibility HDAs are limited to mappings from one phonological plane to another, and so are related to the notion of projection (cf. Vergnaud 1977).

We also wish to show that an understanding of HDAs reveals general structural principles that play a role in diverse phenomena at various levels of the phonological hierarchy. For example, the fact noted in the Optimality Theory literature that certain positions tend to be more ‘faithful’ to underlying specifications (Beckman 1998) is a consequence of the fact that heads allow more complexity. These principles act as constraints on possible constraints, and on possible mappings from one plane to another.

In de Chene & Anderson's (1979) article on compensatory lengthening, the authors make two strong claims as to the universal nature of compensatory lengthening. These claims are: (i) that compensatory lengthening occurs in two stages, involving the weakening of a consonant to a glide and the subsequent merger of the resulting diphthong; and (ii) that compensatory lengthening can only occur when there is a pre-existing vowel-length contrast in the language in question.

Both of these claims have received considerable attention in the literature. The first claim has never gained widespread acceptance, and has been challenged in several studies. Challenges have come from, for example, Hock (1986), Poser (1986), Sezer (1986) and more recently Gildea (1995). Each of these scholars provides a strong case against the view that compensatory lengthening is always decomposable into two distinct stages. The ensemble of their arguments renders this claim simply untenable.

In Chichewa, a Bantu language spoken mainly in Malawi, there is a contrast between high and low tone, as illustrated by such minimal pairs as mtengo ‘price’ vs. mténgo ‘tree’. But there is a strong asymmetry between the two tones in their phonological behaviour: high tone is phonologically active, while low tone is phonologically inert. Tone changes occur in Chichewa only if there is a high tone present in the phrase; a phrase composed only of low-toned morphemes is always realised unchanged with all low tones. The tonal phonology of the language can be described completely without reference to low tone (Kanerva 1989), as is typical for the Bantu languages (Stevick 1969).

I argue in this paper that this asymmetry is due to underspecification. The contrast in Chichewa is a privative one between high tone and no tone. Low tone is phonologically inert because it is simply the absence of tone. In particular, low tone is absent from surface representation. Syllables that are not specified as high-toned are assigned F0 by a non-linear transition function, as proposed for English intonation by Pierre-humbert (1980).

Vowel harmony systems have presented descriptive challenges for virtually every well-articulated theory within the framework of generative phonology. Significantly, no comprehensive and completely satisfactory account in a rule-based theory exists for one of the best studied of these systems, that of Hungarian. The novel approach of Optimality Theory (henceforth OT), as originally developed by Prince & Smolensky (1993) and McCarthy & Prince (1993a, b, 1995), has been shown to offer insightful solutions to vexing problems of prosodic phonology and morphology. This paper seeks to relate the insights of OT to the description of Hungarian vowel harmony: it provides a detailed description of the facts, offers solutions to heretofore unresolved problems, and draws conclusions for general theoretical issues within the OT model.

In §2 we present the facts of backness harmony as the empirical backdrop to the ensuing discussions. In §3 we present an analysis of backness harmony in OT. The ‘spreading’ of the feature [±back] is accounted for by an alignment constraint which is formulated as a constraint prohibiting vowels from intervening between the right edge of a backness feature and the right edge of the word, following proposals of Ellison (1995), Kirchner (1993) and Zoll (1996). We analyse certain roots with floating features, adopting a proposal by Zoll (1996) which ensures that floating features are in fact realised in outputs (unless blocked by satisfaction of higher-ranked constraints). We also assume, following much recent work in OT (Beckman 1995, 1997, 1998, McCarthy & Prince 1995, Steriade 1995, Zoll 1996), that certain prominent positions (e.g. roots) may be subject to more stringent faithfulness constraints than are less prominent positions (e.g. affixes). We further demonstrate that inventory constraints interact with other constraints to determine optimal outputs. In §4 roundness harmony data are presented. We argue that while backness harmony involves alignment constraints, so-called ‘roundness harmony’ does not, and hence that it is a mistake to assume that all cases of vowel harmony involve alignment constraints.

One of the major advances in phonological theory during the past twenty years has been the refinement of a theory regarding the representation and behaviour of tones (Goldsmith 1976, 1990, Fromkin 1978, Clements & Goldsmith 1984, Hyman 1986, 1993, Pulleyblank 1986, 1997), particularly for African tone languages and East Asian tone languages (for recent reviews see Odden 1995 and Yip 1995). A general outline of such a theory, using an autosegmental framework, might be something like the following: (a) Tones or tone melodies are represented in underlying phonological representations (UR); in some cases they are linked to specific ‘tone-bearing units’ (TBUs) such as syllables, moras or vowels on other tiers in UR, and in other cases they are unlinked. (b) Phonological rules will associate tones with correct TBUs according to universal and language-specific principles, including a universal well-formedness condition, such that no tones or TBUs which remain at the end of the derivation are unassociated. Tones may be delinked and omitted or reassociated by phonological rules, which may involve tone spreading, tone sandhi and a number of other phenomena. (c) Tones have a primarily lexical rather than syntagmatic function. (d) Contour tones are represented in UR as a sequence of two or more level tones, and function as tone sequences in tone rules, such that for example an underlying HL sequence may surface as either a falling contour tone on a single TBU or a high–low sequence on two adjacent TBUs. (e) It is assumed that tonal systems in all languages follow the same universal principles, which underlie the theory outlined here.