Semi-supervised Phonetic Category Learning: Does Word-level Information Enhance the Efficacy of Distributional Learning?

Till Poppels,Daniel Swingley

doi:10.5334/jeps.ce

Abstract

To test whether word-level information facilitates the learning of phonetic categories, 40 adult native English speakers were exposed to a bimodal distribution of vowels embedded in non-words. Half of the subjects received phonetic categories aligned with lexical categories, while the other half received no such cue. It was hypothesized that the subjects exposed to lexically-informative training stimuli that were aligned with the target categories would outperform the control subjects on a perceptual categorization task after training. While the results revealed no such group differences, the data indicated that many subjects used the relevant dimension for categorization before having received any training. Implications regarding experimental design and suggestions for future research based on the results are discussed.

Highlights

To test whether word-level information facilitates the learning of phonetic categories, 40 adult native English speakers were exposed to a bimodal distribution of vowels embedded in non-words
One of the first challenges infants face is learning about the speech sounds of the language that surrounds them
Werker and Tees (1984) compared adult speakers of Nthlakampx and adult English speakers on their ability to discriminate two speech sounds that were contrastive in Nthlakampx, but not in English

Summary

Introduction

To test whether word-level information facilitates the learning of phonetic categories, 40 adult native English speakers were exposed to a bimodal distribution of vowels embedded in non-words. Six-month-old infants who were learning English and had never been exposed to Nthlakampx were as good at discriminating the speech sounds as native Nthlakampx speakers. By the time phonetic categories are learned, the infant’s lexicon is still relatively small (Bergelson & Swingley, 2013) and is likely to contain mostly phonetically dissimilar words (Caselli et al, 1995) that do not lend themselves to minimal pair analysis. Another possibility, which is the predominant approach in the literature at present, is that infants discover phonetic categories by implicit analysis of the distributional

Methods

Results

Conclusion