Introduction

The speech literature abounds in evidence that language-specific phonotactic patterns affect perception. Phonotactics affect placement of phoneme category boundaries (Massaro and Cohen 1983), segmentation of nonce forms (Suomi, McQueen and Cutler 1997), and speed and accuracy of phoneme monitoring (Otake, Yoneyama, Cutler and van der Lugt 1996). Papers in previous volumes in this series (Pierrehumbert 1994; Treiman, Kessler, Knewasser and Tincoff 2000) have provided evidence that perceived well-formedness of phoneme combinations is related to their frequency in the language. Coleman (1996) also found that speakers rated neologisms with attested clusters higher than those containing unattested clusters.

These results indicate that speakers generalise over the entries in their lexicons, and respond differently to patterns which are exemplified versus ones which are not. However, patterns may be exemplified to different degrees. This raises the question of whether knowledge of phonotactics is categorical, distinguishing only possible from impossible forms (as predicted by classical generative models), or whether it is gradient, tracking lexical statistics more finely. Some evidence is available from studies which report different outcomes for high- and low-frequency configurations.

Jusczyk, Luce and Charles-Luce (1994) found that 9-month-old infants prefer frequent phonotactic patterns in their language to infrequent ones. Saffran, Aslin and Newport (1996a) showed that 8-month-old infants are sensitive to transitional probabilities in nonsense speech streams. Saffran, Newport and Aslin (1996b) show similar sensitivity in adults.

The metaphors invoked

The phrase ‘phonetic interpretation’ (included in both special themes of the 6th Conference on Laboratory Phonology) invokes a particular view of language sound structure in which phonetics is defined as the semantics of phonology, which implies, conversely, that phonology is the syntax of phonetics. (See Pierrehumbert 1990, inter alia, for earlier explicit articulations of this view.) The paired metaphors are characteristic of a class of phonological theories built on symbolic formalism, just as the theory of ‘ordinary’ syntax is. They also delineate a class of psycholinguistic theories about how our memory of past experiences constrains our own phonological behaviour and our understanding of others' phonological behaviour. The utterances of other speakers are not apprehended and remembered as a collection of unanalysed audio-visual patterns. Nor are our own utterances produced by dredging up from memory the unanalysed articulatory records of some collection of previously produced utterances. Rather, utterances are decomposed into (or composed from) smaller parts that can be recognised as instances of particular syntactic categories standing in particular syntactic relationships to other categories in the context. Thus, phonological behaviour (like ‘ordinary’ syntactic behaviour) can be seen as a particularly elaborate type of categorisation, which allows for very complex and flexible stimulus generalisation. In this paper, we explore the cognitive function of categorisation, to try to understand its ecological advantage.

Introduction

In his paper, Kubozono nicely summarizes the literature on the mora as a salient unit in speech production in Japanese, and then describes new results which he takes as evidence for the mora as a comparably salient unit in speech perception. His experiments presented pairs of English CVC monosyllables, such as hill and top, to which subjects responded with a third word blending the two stimulus words. Responses differed for different subject groups; English speakers combined the onset of the first word with the rhyme of the second (hill + top = hop), whereas Japanese speakers usually took both the onset and the nucleus from the first word and only the coda consonant from the second (hill + top = hip). Pilot results for two Italian and nine Spanish speakers also showed the Japanese pattern.

Kubozono interprets these results as indicating a difference in syllable-internal prosodic structure among the languages, illustrated in (la) for English and (lb) for the other three, (lc) is Derwing, Yoon & Cho's (1993) alternative interpretation of the Japanese pattern of responses, which their Korean subjects also showed in a similar word-blend experiment. The difference between hypotheses (lb) and (lc) should not obscure the premise in common: Kubozono, like Derwing et al. (1993), takes the pattern of blends as evidence that speakers parse syllables exhaustively into some set of syllable-internal prosodic constituents which can differ from language to language.

Introduction

Speech is structured both by the grammatical functions of contrast and organization and by the physical medium for those functions, namely sound as produced by the human vocal tract and perceived by the human ear. In investigating sound structure, neither the grammar nor the medium can be ignored. Each is an external source of constraint and explanation for facets of the other. A striking example of the dangers of ignoring either is the difficulty we have had in characterizing stress. It is only within the last ten years or so that phonological accounts of stress patterns have approached the level of descriptive insight afforded to segmental patterns by the phonetic observations of Panini. Conversely, investigations of physical and perceptual correlates of stress have produced many seemingly conflicting results, at least in part because of confusion concerning the phonological categories and structures involved.

In this paper, we will describe our understanding of what a unified theory of (surface) stress patterns must be like in general in order to be able to account for both the phonology and the phonetics of stress. We will then make a few proposals about how to fill in the specifics of the theory for a description of some parts of the English stress system, and show how this description might help account for the phonetics of stress shift. Finally, we will present some articulatory data in partial support for our specific proposals.

A theory of stress

One of the reasons why stress has been so difficult to characterize phonologically is that its salient function is not the familiar one of contrasting lexical items.

Introduction

There are two durational effects often cited as evidence for very different models of prosodic constituency in English. The first is known variously as “final lengthening,” “pre-boundary lengthening,” or “pre-pausal lengthening” (e.g. Oiler 1973; Klatt 1975; Cooper and Paccia-Cooper 1980). As the last name suggests, this effect is usually interpreted as a durational correlate of the sort of disjuncture that can cause a momentary cessation of speech. The second durational effect has no similar unified set of labels, but it might be called “stresstimed shortening” since it belongs to a class of effects that have been interpreted as indications of a tendency toward isochronous spacing of prosodically strong syllables; a stressed syllable in a polysyllabic word or stress foot is compressed in order to make the overall duration of its word or stress foot closer to that of a contrasting monosyllable (e.g. Huggins 1975; Fowler 1977).

The two effects are similar in that both involve an apparent adjustment of syllable durations which is dependent on some notion of constituency. In the first case, a syllable is lengthened because of its position near to the edge of some constituent, and in the second it is shortened because of the length of a constituent defined by adjacent peaks at some prosodic level. The two effects differ radically, however, in the type of constituency that is implicit in their interpretations. Final lengthening implies a constituent that has well-defined edges; the lengthening occurs before a boundary that could be followed by a pause. But this constituent need not have a phonological head; its internal prosodic structure could be completely flat.

While each of the papers in this volume has its specific individual topic, collectively they address a more general issue, that of the relationship between the phonological component and the phonetic component. This issue encompasses at least three large questions. First, how, in the twin processes of producing and perceiving speech, do the discrete symbolic or cognitive units of the phonological representation of an utterance map into the continuous psychoacoustic and motoric functions of its phonetic representation? Second, how should the task of explaining speech patterns be divided between the models of grammatical function that are encoded in phonological representations and the models of physical or sensory function that are encoded in phonetic representations? And third, what sorts of research methods are most likely to provide good models for the two components and for the mapping between them?

Previous answers to these questions have been largely unsatisfactory, we think, because they have been assumed a priori, on the basis of prejudices arising in the social history of modern linguistics. In this history, phonology and phonetics were not at first distinguished. For example, in the entries for the two terms in the Oxford English Dictionary each is listed as a synonym for the other; phonology is defined as “The science of vocal sounds (= PHONETICS)” and phonetics as “The department of linguistic science which treats of the sounds of speech; phonology.”