Emergent phonological representations: no need for autosegmental architecture

Abstract

This paper examines implications for autosegmental representations of a model that minimizes the role of an innate linguistic endowment in grammar formation. If the innate linguistic endowment is minimized, language learning is from the bottom up and cannot rely on universal structures. Bottom-up grammars share common goals with top-down grammars, among them to identify and characterize phonological patterns. In this paper, we examine vowel distribution in Tiv, a Niger-Congo language of Nigeria. The six Tiv vowels occur in restricted positions in verbs: only 10 of the 36 possible V1(C)V2 sequences occur with any frequency. Tiv vowel distribution has been explained in terms of feature geometry, association rules, and spreading rules. We show that while the vowel distribution can be expressed using such an architecture, it can also be expressed in simpler terms, relying only on nonlinguistic capabilities such as the ability to evaluate input based on similarity and frequency, and the ability to construct symbolic representations of such data. In addition to conceptual arguments in favor of a bottom-up, emergent phonology, the paper provides an example of the analysis of a phonological system under the Emergence hypothesis.