SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University


© Webmaster

Contributions: Abstract

Local parsing strategies of low-proficiency language learners

Kazuo Amma
Tamagawa University
Japan

This study was intended to identify a response behaviour in grammaticality judgement as a function of general language proficiency. The main research question, what is the common parsing strategy taken by low-proficiency learners., was sought through exploratory and confirmatory multivariate analyses.

The participants were 1152 Japanese university learners of English as a foreign language, who also took a C-test for measuring their general language proficiency. The test data, in a combined multiple-choice and dichotomous format, was first analysed by logistic regression analysis in order to obtain the probability of the response which reflected a local, short-range parsing of the stimulus sentence and the probability of the other responses. By combining these two types of probabilities for two test items as the magnitude indices of a contingency table, the association measure of the two items was calculated. This process was repeated for all combinations of the 38 items, for 9 proficiency levels. Using the association measure as the similarity variable, a multidimensional scaling analysis identified a cluster of items which became stronger as the proficiency level went lower.

What, then, was observed from the items in the cluster was a common parsing strategy with which one closes the interpretation of a quasi-sentential unit; for example, The girl who ate with Cathy loved ice creamwas interpreted as Cathy loved ice cream. A classification of the test items by whether or not they comprised this local parsing indicated that the average magnitude of inter-item association was statistically stronger among the items which allowed local parsing than among the items which did not only when the proficiency was low. This phenomenon suggests that low-proficiency test takers are more susceptible to a common cognitive constraint, leading to a limited span of text to be processed.