Automatic disfluency removal for improving spoken language translation

Statistical machine translation (SMT) systems for spoken languages suffer from conversational speech phenomena, in particular, the presence of speech dis uencies. We examine the impact of dis uencies from broadcast conversation data on our hierarchical phrasebased SMT system and implement automatic dis uency removal approaches for cleansing the MT input. We evaluate the ef cacy of proposed approaches and investigate the impact of dis uency removal on SMT performance across different dis uency types. We show that for translating Mandarin broadcast conversational transcripts into English, our automatic dis uency removal approaches could produce signi cant improvement in BLEU and TER.

• Type: Conference
• Year: 2010
• Where: ICASSP
• Authors: Wen Wang, Gökhan Tür, Jing Zheng, Necip Fazil Ayan
• Keywords: Automatic Dis, Dis Uencies, Dis Uency Removal, ICASSP 2010, Signal Processing