A Maximum Entropy/Minimum Divergence Translation Model

I describe two methods for incorporating information about the relative positions of bilingual word pairs into a Maximum Entropy/Minimum Divergence translation model. The better of the two achieves over 40% lower test corpus perplexity than an equivalent combination of a trigram language model and the classical IBM translation model 2.

• Type: Conference
• Year: 2000
• Where: ACL
• Authors: George F. Foster
• Keywords: ACL 2000, ACL 2007, Divergence Translation Model, Translation Model, Trigram Language Model