PIMS-SFU CSC Seminar: Andrew R. Plummer

The concept of equivalence has played a central role in mainstream linguistic theory over the last century, e.g., in the definition of linguistic "form" in Bloomfield's (1926, 1933) attempts to axiomatize linguistic theory, in debates on the nature of mutual entailment of propositions in formal semantic theories (see Lewis, 1923, 1943; Carnap, 1947; inter alia), etc. Within Bloomfield's axiomatic approach the definition of a phonological form, or "phoneme," in a given language involves equivalence classes of speech sounds that are judged to be similar by native speakers of that language, and each language has a set of phonemes that act as the minimal building blocks used to construct its words and phrases. Yet, it was argued by Chao (1934) that "there are usually more than one possible way of reducing [the sounds of a language] to a system of phonemes," who concludes that "these different systems or solutions are not simply correct or incorrect, but may be regarded only as being good or bad for various purposes" (p. 363). Attempting to address this perceived difficulty, using the source-filter model of speech production (Chiba and Kajiyama, 1941; Fant, 1960) as a basis, Jakobson, Fant, and Halle (1951) formulated a theory of speech analysis that postulated the existence of "distinctive features" that underlie the speech sounds of all languages.

In line with this theoretical development, it is often assumed in current speech research that speech sounds possess underlying invariant properties that render them separable into equivalence classes despite the massive variation in their observable properties (as revealed, say, by acoustic analysis) both within and across speakers. In this presentation, we situate the assumption of underlying invariant properties of speech sounds within the context of phonological acquisition and present evidence that undermines its validity. We then present the basis of an alternative view wherein speech sound equivalence is constructed by infants relative to their social interactions with caretakers over the course of ontogeny. We conclude with presentation of a modeling framework that formalizes the construction process as a "manifold alignment" computation that provides the means for the infant to view their vocalizations as commensurate with those of their caretakers.