Empirical Machine Translation and its Evaluation

Our goal was twofold. On the one side, we studied the problem of Automatic MT evaluation. We analyzed the main deficiencies of the current evaluation methodology and suggested several complementary improvements. On the other side, we built an empirical MT system and have analyzed several of its limitations. We incorporated linguistic knowledge into the system with the aim to improve overall translation quality. In particular, we addressed the problem of lexical selection in Statistical Machine Translation. As a side question, we also studied one of the main criticisms against Empirical MT systems, i.e., their strong domain dependence, and how its negative effects may be mitigated by properly combining outer knowledge sources when porting a system into a new domain.