Paper: Hierarchical MT Training using Max-Violation Perceptron

ACL ID P14-2127
Title Hierarchical MT Training using Max-Violation Perceptron
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2014

Large-scale discriminative training has be- come promising for statistical machine translation by leveraging the huge train- ing corpus; for example the recent effort in phrase-based MT (Yu et al., 2013) sig- nificantly outperforms mainstream meth- ods that only train on small tuning sets. However, phrase-based MT suffers from limited reorderings, and thus its training can only utilize a small portion of the bi- text due to the distortion limit. To address this problem, we extend Yu et al. (2013) to syntax-based MT by generalizing their latent variable ?violation-fixing? percep- tron from graphs to hypergraphs. Exper- iments confirm that our method leads to up to +1.2 BLEU improvement over main- stream methods such as MERT and PRO.