Automatic prosody evaluation and grammatical mistake detection for English learning by Japanese native speakers

点击翻译按钮获取中文摘要

I have developed two methods that enable speaking exercise in a computer-assisted language learning (CALL) system : a method for evaluating prosody of an English utterance made by a learner, and a method for detecting grammatical mistakes included in the learner's utterance. As for prosody evaluation, I developed an estimation method of word importance factors using a decision tree, and obtained a high correlation to human assessment score, which is comparable to correlation between scores given by human evaluators. As for the grammatical mistake detection, I proposed a method for training an n-gram language model from artificially generated sentences with mistakes, and obtained 89.2% word accuracy.