Acquisition of phonological variables of a Flemish dialect by children raised in Standard Dutch: some considerations on the learning mechanisms

This study investigates the learning mechanisms underlying the acquisition of a dialect as a second language. We focus on the acquisition of phonological features of a Flemish dialect by children with Standard Dutch or a regional variety of Dutch as their first language. Data were gathered by means of picture naming and sentence completion tasks. Inspired by Chambers (1992), who found that the data of second dialect learners displayed S-curve patterns which he interpreted as evidence of rule-based learning, we examine whether similar S-curves can be observed in the learner data of our subjects. Contrary to Chambers, our subjects’ data do not display S-curves but bear evidence of word-by-word learning across the board. These data are consistent with analogical memory-based models of language acquisition. In order to further investigate the applicability of memory-based reasoning to our data, we perform a computational classification task in TiMBL (Daelemans & Van den Bosch, 2005), in which the dialect forms of Standard Dutch words have to be predicted on the basis of various amounts of training data. Not only do we compare the accuracy scores of the model with the acquisition scores of our subjects, the classification task also gives us insight into which words constitute the nearest neighbours of a given word. On the basis of this output, we investigate the effect of the number of enemy neighbours on the degree to which the subjects realize the correct dialect variants of words and on the degree to which they make overgeneralization errors. The major finding of this paper is that dialect forms are more often realized incorrectly and that more overgeneralization errors occur in words with a large(r) number of enemy neighbours.