Detecting Salient Mispronunciations of Oral English for Chinese Students Using Word-dependent Lexicon Extension and Effectiveness Criterion

【摘要】：正This paper attempts to automatically detect salient mispronunciations based on speech recognition with word-dependent lexicon extension for aiding English learning of Chinese students as a second language.A strategy of using data-driven based word-dependent error rules instead of the universal rules is proposed,owing to the unique error forms for different target words.For how to select salient error patterns for each word,an effectiveness criterion is also introduced to balance the tradeoff of error coverage and loss cost.Based on a large corpus with annotations,the experiments show that the performances of the automated detection system are equivalent to the moderate annotator and a little better than the bad annotator. Moreover,the results imply that it is not necessary to make strenuous attempts to cover all the possible mispronunciation patterns,the choice of using a set of salient rules is a compromise solution.