Flaps and other variants of /t/ in American English: Allophonic distribution without constraints, rules, or abstractions

Abstract

The distribution of the flap allophone [ɾ] of American English, along with the other allophones of /t/,[t^h,t⁼, ʔ, t] has been accounted for in various formal frameworks by assuming a number of different abstract mechanisms and entities. The desirability or usefulness of these formalisms is not at issue in the present paper. Instead, a computationally explicit model of categorization is used (Skousen 1989, 1992) in order to account for the distribution of the allophones of /t/ without recourse to such formalisms. The simulations that were carried out suggest that they are not needed because analogy to surface apparent variables such as phones and word boundaries is sufficient to predict allophony.

In analogy, the particular allophone of /t/ (i.e. [ɾ, t^h, t⁼, ʔ, t]) that appears in a given context is determined on the basis of similarity to stored exemplars in the mental lexicon. From an acquisitional standpoint, categorization by analogy to stored exemplars dispenses with the need for rule induction although it does suggest that speakers group functionally related sounds into mental categories, a process that is influenced to a great deal by orthography.

Analogy also explains the stochastic nature of linguistic performance. In the present study, 3,719 tokens of the allophones of the phoneme /t/ were extracted from the TIMIT corpus and constitute the database from which analogs were chosen. The variables used included the three phones or boundaries on either side of /t/, and the stress of the syllables preceding and following /t/. The model proves quite successful in predicting the correct allophone, and the errors made are generally possible alternative pronunciations (e.g. moun[ʔ]ain, moun[t^h]ain). The success rate changes little when only small sub-samples of the database are incorporated. In addition, exemplar-modeling is found to be quite robust because even when a feature such as stress is eliminated, (a feature which is critical in most rule approaches), allophony is still highly predictable.