We investigate the impact of parse quality on a syntactically-informed statistical machine translation system applied to technical text. We vary parse quality by varying the amount of data used to train the parser. As the amount of data increases, parse quality improves, leading to improvements in machine translation output and results that significantly outperform a state-of-the-art phrasal baseline.

PDF file

In: Proceedings of EMNLP 2006

Publisher: ACL/SIGPARSE
Publisher does not hold copyright.

Details