Ternary rhythmic systems differ from binary systems in stressing every third syllable in a word, rather than every second. Consider the following examples from Cayuvava (Key 1961), where stress is on every third syllable counting from the end of the word:

(1) a. à.ri.hi.hí.be.e ‘I have already put the top on’

b. ma.rà.ha.ha.é.i.ki ‘their blankets’

c. i.ki.tà.pa.re.ré.pe.ha ‘the water is clean’

Ternary rhythm is well-established for only a small group of languages, including Chugach Alutiiq, Cayuvava and Estonian, and possibly Winnebago. Nevertheless the stress patterns of these languages are sufficiently complex to warrant an ongoing debate about the implications for metrical theory (see Prince 1980, Levin 1985, 1988, Halle & Vergnaud 1987, Halle 1990, Hammond 1990, Dresher & Lahiri 1991, Rice 1992, Hewitt 1992, Kager 1993, 1994, Halle & Idsardi 1995, Hayes 1995, Ishii 1996, Elenbaas 1999, among others).

The reason for a fresh look at ternarity is the rise of Optimality Theory (henceforth OT; Prince & Smolensky 1993, McCarthy & Prince 1993a), a theory abandoning most devices on which rule-based accounts of ternarity were based. It abandons serial derivations and together with it directional foot assignment, a core device in parametric theories of word stress, as well as special parsing modes for ternary rhythm (Weak Local Parsing; Hayes 1995). Derivational mechanisms and parameters are replaced by universal and violable constraints, stating well-formedness on output forms, and ranked in language-particular hierarchies.

The issue then arises whether OT is able to predict the ternary patterns in a descriptively adequate fashion. The first goal of this paper is to argue that adequate and insightful analyses are indeed possible in OT for two ternary stress languages: Cayuvava and Chugach Alutiiq. We argue that these analyses require no ternarity-inducing mechanisms, such as ternary feet or special parsing modes. Instead ternarity emerges by LICENSING, involving interactions of the anti-lapse constraint *LAPSE (banning long sequences of unstressed syllables; Selkirk 1984) with standard foot- alignment constraints (ALL-FT-X, ALIGN-Y; McCarthy & Prince 1993b). Our analysis incorporates Ishii's (1996) insight that ternarity is a kind of underparsing, which is licensed by an anti-lapse constraint, and induced by standard foot alignment.

This paper offers a novel account of a familiar typological observation, namely the tendency of phonological inventories to consist of segments that are dispersed through the available auditory space. In contrast to previous approaches, which have treated dispersion as a goal explicitly encoded in the grammar, this paper shows that the cross-linguistic pattern follows automatically from the interaction of two independently motivated factors: phonological representations in which only contrastive features are specified, and the enhancement of these features in phonetic implementation. The merits of this approach are illustrated by examples involving both vocalic and consonantal inventories.

This article is concerned with the internal structure of coronals. Although coronals are often considered the least marked consonants on the basis of frequency and acquisition, current models of feature geometry assign coronality no special status among places of articulation.1 In this paper, we argue that the structure of coronals differs from that of other consanats in that coronals lack a Place node altogether.

Vowel duration is not contrastive in standard and regional varieties of Italian. However, vowels in stressed open syllables are longer than unstressed vowels or vowels in closed syllables. The increased duration is not equal in all positions. Most notably, the increased duration of a stressed open penultimate syllable is much greater than the duration of a stressed open antepenultimate syllable or a stressed final syllable, which has no noticeable duration increase. Nonetheless, phonological analyses of Italian have characterised length by a single rule (see for example Vogel 1986, Nespor & Vogel 1986) that lengthens non-final main stress vowels regardless of position. Phonetic studies, particularly Farnetani & Kori (1983, 1990) and Marotta (1985), pay closer attention to the duration of stressed vowels in different positions. Although their explanations of stressed vowel duration differ, the common theme is that duration differences are due to shortening vowels as a consequence of word compression or position (antepenultimate or penultimate syllable) in the word.

While the phonetic approaches account for differences in duration due to shortening, the phonological approaches propose lengthening with no regard for actual duration differences. The phonetic and phonological approaches to stressed vowel duration in Italian appear to be diametrically opposed. This paper proposes that lengthening a stressed vowel is the correct characterisation of duration differences in Italian, but there is no single rule that lengthens stressed vowels.

The intonation of Greek wh-questions consists of a rise-fall followed by a low plateau and a final rise. Using acoustic data, we show (i) that the exact contour shape depends on the length of the question, and (ii) that the position of the first peak and the low plateau depends on the position of the stressed syllables, and shows predictable adjustments in alignment, depending on the proximity of adjacent tonal targets. Models that specify the F0 of all syllables, or models that specify F0 by superposing contour shapes for shorter and longer domains, cannot account for such fine-grained lawful variation except by using ad hoc tonal specifications, which, in turn, do not allow for phonological generalisations about contours applying to utterances of greatly different lengths. In contrast, our findings follow easily from an autosegmental-metrical approach to intonational phonology, according to which melodies may contain long F0 stretches derived by interpolation between specified targets associated with metrically strong syllables and prosodic boundaries.

Ambivalent segments are speech sounds whose cross-linguistic patterning is especially variable, creating contradictions for theories of universal distinctive features. This paper examines lateral liquids, whose [continuant] specification has been the subject of controversy because of their ability to pattern both with continuants and with non-continuants, and because phonetically they are situated in the contested ground between two different articulatory definitions for the feature [continuant]. Evidence from a survey of sound patterns in 561 languages shows that lateral liquids, like nasals, pattern with continuants about as often as with non-continuants. Ambivalent phonological behaviour is argued to be natural and expected for phonetically ambiguous segments in a theory of emergent distinctive features where features are the result of sound patterns, rather than the other way around.

This article discusses the syntax–prosody interface, drawing on evidence from Stockholm Swedish. It is shown that a Swedish main clause containing an embedded clause has three prosodic correlates, two of which are non-isomorphic to the syntactic bracketing. However, two coordinated clauses have only one – isomorphic – prosodic correlate. Optimality-theoretic constraints (Prince & Smolensky 1993) are used to derive this variation. A new markedness constraint, EqualSisters, is argued to be responsible for a preference for flat prosodic structures. This constraint requires that sister nodes in prosodic structure belong to the same prosodic category, and therefore sometimes conflicts with match constraints, which call for syntax–prosody correspondence (Selkirk 2009, 2011). When high-ranked, EqualSisters forces syntax–prosody non-isomorphism if the input syntactic structure contains embedding, whereas full isomorphism is predicted in coordinated structures. The previously suggested markedness constraints Non-recursivity and Exhaustivity (Selkirk 1996) cannot replace EqualSisters, and in the present account are rendered redundant.

The Northern and Western dialects of the Mon-Khmer language Kammu use fundamental frequency to distinguish words, while the Eastern dialect relies on the contrastive voicing of initial consonants to distinguish words which display an F0 contrast in the other dialects. The production and perception of tones in the three dialects is investigated, using recordings and perception tests. As expected, Northern and Western Kammu speakers do use F0 to distinguish words, as seen both in their production and in perception tests, which show that they could distinguish small F0 differences. In contrast, Eastern Kammu speakers neither produce nor perceive F0 differences at the lexical level. Furthermore, the voiceless and voiced stops in Eastern Kammu cause smaller F0 perturbations in the following vowel than might be expected for a language undergoing tonogenesis. The significance of these results for tonogenesis is discussed, as well as their relevance for the notion of ‘tone language’.

This paper builds on growing evidence that aspirated or fortis obstruents in languages like English and German are laryngeally marked, but that phonetic voicing in the (unmarked) unaspirated or lenis series is contextually determined. Employing the laryngeal feature set proposed by Halle & Stevens (1971), as incorporated into the ‘dimensional theory’ of laryngeal representation (Avery & Idsardi 2001, forthcoming), we develop an explicit account of this phonetic enhancement of phonological contrasts, which is widely known as ‘passive voicing’. We find that both passive voicing and inherent aspiration have been phonetic and phonological characteristics of the Germanic languages since the break-up of Indo-European, with laryngeally unmarked stops repeatedly enhanced by the gesture of [spread glottis]. A key implication of this view is that Verner's Law was not an innovation specifically of early Germanic, but rather is an automatic (ultimately phonologised) reflex of passive voicing, itself a ‘persistent change’ rising out of the enduring ‘base of articulation’ that came to characterise Germanic.

Focusing on weight-insensitive binary stress systems, the article presents an approach to metrical stress that is more restrictive than standard approaches and more accurate in its predictions. The proposal's restrictiveness derives from a set of constraints and structural assumptions that run counter to prevailing theories' fundamental principles. For example, the proposed account assumes strict succession between prosodic categories, ensuring that syllables are exhaustively parsed into feet. It tolerates improper bracketing of prosodic categories, allowing feet to overlap and to share entries on the metrical grid. Finally, it makes the foot–stress relationship violable, allowing feet to remain stressless under appropriate rankings. The article examines each of these assumptions and demonstrates how they combine to more accurately predict attested typologies.

Blends, also called portmanteau words, are formed by fusing two words into one new word, where internal portions of the base words are often subtracted (one segmental string from the right part of the first word and another from the left part of the second word). For example, the English blend nixonomics has been formed by combining nixon and economics and subtracting the string neco. (For clarity of exposition, blends will be usually represented as nixo <n⋅eco> nomics, where the subtracted material is enclosed in angled brackets and the boundary between the base elements is indicated by ⋅).

Harmonic Grammar is a model of linguistic constraint interaction in which well-formedness is calculated in terms of the sum of weighted constraint violations. We show how linear programming algorithms can be used to determine whether there is a weighting for a set of constraints that fits a set of linguistic data. The associated software package OT-Help provides a practical tool for studying large and complex linguistic systems in the Harmonic Grammar framework and comparing the results with those of OT. We first describe the translation from harmonic grammars to systems solvable by linear programming algorithms. We then develop a Harmonic Grammar analysis of ATR harmony in Lango that is, we argue, superior to the existing OT and rule-based treatments. We further highlight the usefulness of OT-Help, and the analytic power of Harmonic Grammar, with a set of studies of the predictions Harmonic Grammar makes for phonological typology.

This paper argues that rather than just select the best candidate, EVAL imposes a harmonic rank-ordering on the full candidate set. Language users have access to this enriched information, and it shapes their performance. This paper applies this idea to variation. The claim is that language users can access the full candidate set via the rank-ordering imposed by EVAL. In variation, more than one candidate is well-formed enough to count as grammatical. Consequently, language users will access more than just the best candidate from the rank-ordering. However, the accessibility of a candidate depends on its position on the rank-ordering. The higher the position a candidate occupies, the more likely it is to be selected. In a variable process, variants that appear higher on the rank-ordering (i.e. are more well-formed) will therefore also be the more frequent variants. This model is applied to variation in the phonology of Faialense Portuguese and Ilokano.

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.

A fundamental observation of phonological theory is that, out of the rich sound signal of speech, a small subset of phonetic properties is contrastive in any given language. Standardly, this observation is captured by excluding the non-distinctive phonetic properties from some levels of representation, yielding an abstract categorical representation of the contrasts among the speech sounds. Universally non-contrastive features have generally been assumed to be unspecified throughout the phonological component; thus, for example, Keating (1984) and Lombardi (1991) have argued against inclusion of particular laryngeal features in phonological representation on the grounds that they are never contrastive. Language-particular non-contrastive features have generally been assumed to be unspecified only in underlying representation and early stages of the derivation (e.g. Kiparsky 1982, Archangeli 1984, 1988, Archangeli & Pulleyblank 1986, 1994, Steriade 1987, Clements 1988). Moreover, since the discovery of extensive language-particular gradient alternations, it has standardly been assumed that a subsequent phonetic component is needed, to fill in gradient and universally non-contrastive properties (e.g. Pierrehumbert 1980, Keating 1990; but see Pierrehumbert 1994 for an attack on this view of the phonetics–phonology interface). This model is schematised in Fig. 1.

The past decade or so has seen increasing interest in prosodic research, and remarkable progress has been made in the study of accent and intonation, both empirical and theoretical. Research in Japanese intonation is no exception in this respect, as evidenced by such works as Poser (1984) and Beckman & Pierrehumbert (1986), among others, which have developed theoretical discussions on many interesting phonological issues on the basis of their own experimental evidence.

This paper reconciles the standpoint that language users do not aim at improving their sound systems with the observation that languages seem to improve their sound systems. If learners optimise their perception by gradually ranking their cue constraints, and reuse the resulting ranking in production, they automatically introduce a prototype effect, which can be counteracted by an articulatory effect. If the two effects are of unequal size, the learner will end up with a sound system auditorily different from that of her language environment. Computer simulations of sibilant inventories show that, independently of the initial auditory sound system, a stable equilibrium is reached within a small number of generations. In this stable state, the dispersion of the sibilants of the language strikes an optimal balance between articulatory ease and auditory contrast. Crucially, these results are derived within a model without any goal-oriented elements such as dispersion constraints.

This paper is about opaque interactions between phonological processes in the two senses defined by Kiparsky (1971, 1973) and discussed in much recent work on the topic, most notably McCarthy (1999) : underapplication opacity, whereby a process appears to have failed to apply in expected contexts on the surface, and overapplication opacity, whereby a process appears to have applied in unexpected contexts on the surface. Specifically, I demonstrate that there are three distinct types of overapplication opacity in addition to the only case discussed and properly categorised as such in the literature, counterbleeding. The analysis of each type of opacity in terms of rule-based serialism and in terms of Optimality Theory is discussed, emphasising the strengths and weaknesses of the two frameworks in each case.

Varieties of Malay, including Indonesian, have been variously described as having word stress on the penultimate syllable, as having variable word stress and as having a phrase-final pitch accent without word stress. In Ambonese Malay, the alignment of sentence-final pitch peaks fails to support the existence of either word stress or phrase-final pitch accents. Also, the shape of its pitch peaks fails to vary systematically with the information status of the phrase-final word. The two intonation melodies of the language include phrase-final boundary-tone complexes which do not associate with any syllables. The declarative rise-fall would appear to be timed so as to occur within the last word of the sentence. Minimal stress pairs presented in earlier descriptions show a contrast between /a/ and a segmentally distinct weak /ă/, a contrast that also appears in positions that have not been claimed to have stress. A preliminary phonological analysis concludes the account.

We argue that the common phonological assumptions that (i) plain voiceless consonants are less marked than voiceless aspirates and (ii) the unmarked two-way stop system opposes unaspirated voiced and voiceless members are incorrect. A wide range of phonetic and internal and external phonological evidence suggests instead that (i) the maximally unmarked single-series stop is unspecified for laryngeal features and (ii) the unmarked two-way stop system opposes aspirated and unaspirated stops, and the aspirated series may be the unmarked member of this set.