21 Mar 2006

Sherry Ou

This talk aims at investigating the learning mechanisms which guide L2 word stress acquisition. Two general views are evaluated: (i) UG-based parameter-setting, which claims that L2 stress acquisition is a process of setting an a priori set of metrical parameters, and (ii) input-based statistical learning, which sees the development of L2 stress patterns as a process of statistical extraction based on the distribution of stress patterns in the input data.

In a perceptual preference experiment with English non-words, Chinese participants preferred initial stress for CVCC words when they were presented as nouns, but preferred final stress when they were presented as verbs, paralleling the behaviour of the English subjects. In trisyllabic words, some Chinese subjects preferred penultimate stress when the penult was closed by a consonant (CVC) and antepenultimate stress when the penult only contained a lax vowel (CV). The data show that some Chinese subjects were sensitive to the noun-verb contrast but not to the stress contrast conditioned by the penult of trisyllabic nouns. It is argued that this developmental pattern runs counter to the prediction of the parameter-setting model. If learners had acquired the noun-verb stress contrast via parameter-setting, they would also have acquired the weight implications of CV and CVC, hence the stress contrast in trisyllabic nouns, which is conditioned by the CV/CVC distinction. The observed pattern can however be explained by the input-based statistical learning model, which allows independent learning of the noun-verb contrast via statistical learning. Another type of data that favours the input-based statistical learning account comes from a corpus-based analysis of segmental effects on English stress, which shows that penults tend to be stressed when they contain a certain vowel (i.e., /ε/), regardless of syllable structure. This anomaly was reflected in the Chinese participants' data.

[back to PWorkshop]

<owner-pworkshop@ling.ed.ac.uk>