CCG Syntactic Reordering Models for Phrase-based Machine Translation

Statistical phrase-based machine translation requires no linguistic information beyond word-aligned parallel corpora (Zens et al., 2002; Koehn et al., 2003). Unfortunately, this linguistic agnosticism often produces ungrammatical translations. Syntax, or sentence structure, could provide guidance to phrasebased systems, but the “non-constituent” word strings that phrase-based decoders manipulate complicate the use of most recursive syntactic tools. We address these issues by using Combinatory Categorial Grammar, or CCG, (Steedman, 2000), which has a much more flexible notion of constituency, thereby providing more labels for putative nonconstituent multiword translation phrases. Using CCG parse charts, we train a syntactic analogue of a lexicalized reordering model by labelling phrase table entries with multiword labels and demonstrate significant improvements in translating between Urdu and English, two language pairs with divergent sentence structure.