Identifying confusable contexts for automatic generation of activities in second language pronunciation training

This paper presents a new technique in the development of tools for Computer-Aided Language Learning (CALL) in Second Language (L2) pronunciation training. Instead of the usual pinpointing and feedback strategies, which address all production errors in the L2 speaker equally, this technique aims at honing in on where a pronunciation error can lead to a confusion in the transmission of the message and providing these contexts for practice. To provide this new correction strategy, it is necessary to first identify the words and contexts which can lead to a confusion in a message when there is a mispronunciation in them. This paper describes a Basic Identification of Confusable Contexts (BICC) procedure for detecting these cases providing a measure of their relevance. It also includes a methodological validation of BICC with humans assessing the quality and comprehensibility of a dataset of automatically generated sentences which shows that the prediction of these contexts is possible with good precision.