24 Feb 2009

Hierarchical Phrase-based Translation

February 25, 2009

Location: TASC1 9408

Presenter: Baskaran Sankaran

Time: 12:30pm  -1:30 pm

I will be presenting the Hierarchical Machine Translation system (Chiang, 07 and Chiang 05). The original motivation is to have a fusion of both syntax-based and phrase-based systems. The translation model is synchronous context-free grammar, learnt automatically from bitext without syntactic annotations of a treebank. Instead of the normal phrases of the phrase-based systems, here the phrases are hierarchical, (i.e) sub-phrases within large phrases.  Apart from covering the system in detail, I will be focussing on two aspects: i) phrase-extraction and ii) language model integration.