LOOSE PHRASE EXTRACTION WITH n-BEST ALIGNMENTS

摘要

Loose phrase extraction method is proposed and applied for phrase-based statistical machine translation. The method extracts phrase pairs that are not strictly consistent with word alignments. Two types of constraints on word positions are investigated for this method. Furthermore,n-best alignments are introduced for phrase extraction instead of the one-best. Experimental results show that the proposed approach outperforms the baseline system, Pharaoh system, for both one-best and n-best alignments.