Treebanks in Machine Translation

We present an approach using treebanks in machine translation. Our experiment in Czech-English machine translation is an attempt to develop a full machine translation system based on dependency trees (Dependency Based Machine Translation, DBMT). We use the following resources: Prague Dependency Treebank, a newly created Czech-English parallel corpus of Penn Treebank, English monolingual corpus, and translation lexicons. The fully automatic process includes analysis of the Czech input into tectogrammatical (semantic) representation, lexical and structural transfer, a simple rule-based system for generation to English surface realization, and an -gram language model for scoring and choosing from translation hypotheses. The results are evaluated quantitatively with BLEU score.