An acoustics-based assessment of error patterns in English sounds produced by Cantonese ESL speakers

Abstract

Pronunciation problem has always been the major challenge for Cantonese ESL speakers due to discrepancy between the Cantonese and English sound inventories. Traditionally, accent identification was done by listening to the accented English produced by Cantonese speakers. This is a subjective and unreliable method for learning a second language (L2). The present study made use of automatic speech recognition (ASR) technologies in identifying the error pattern associated with accented English produced by Cantonese ESL speakers. Cantonese ESL speakers are instructed to produce a list of isolated English words and a short passage. Pattern and frequency of errors in L2 English pronunciation were objectively identified from the speech samples using ASR software. Results of error analysis were compared with what linguists predicted. Results on L2 English error pattern were discussed in relation to L1 Cantonese.