Modeling second language acquisition processes with logistic regression

Robert Berdan

Sun. 11-1:05 B

This paper is a continuation of previous work looking at the data published by Schumann (1978) on adult second language acquisition. Schumann's informant Alberto is widely cited as the paradigm example of pidginized or fossilized second language acquisition. Using logistic regression it has been possible to show that Alberto's use of negation moves substantially toward conventional English during the 10 months of observations. Now it is also possible to show that that acquisition process is not strictly linear. Using logistic regression, the more generalized procedure underlying VARBRUL, to disentangle variation associated with change over time from variation associated with significantly differentiated syntactic and discourse contexts, this paper traces the acquisition process through time, within and across the particular contexts in which acquisition is most notable. The dichotomized predicted values typically used in logistic regression models for evaluation of goodness of fit classification tables are readily interpretable in the framework of second language acquisition. Comparing these predicted values with the observed variation it is possible to trace the predicted patterns of variation through successive transcripts, and to compare the predicted forms with the observed realizations.