Learning Phrase Boundaries for Hierarchical Phrase-based Translation

摘要

Hierarchical phrase-based models pro-vide a powerful mechanism to capture non-local phrase reorderings for statis-tical machine translation (SMT). How-ever, many phrase reorderings are arbi-trary because the models are weak on de-termining phrase boundaries for pattern-matching. This paper presents a novel approach to learn phrase boundaries di-rectly from word-aligned corpus without using any syntactical information. We use phrase boundaries, which indicate the be-ginning/ending of phrase reordering, as soft constraints for decoding. Experi-mental results and analysis show that the approach yields significant improvements over the baseline on large-scale Chinese-to- English translation.