Publikation
DFKI System Combination using Syntactic Information at ML4HMT-2011
Christian Federmann; Yu Chen; Sabine Hunsicker; Rui Wang
In: Proceedings of the International Workshop on Using Linguistic Information for Hybrid Machine Translation (LIHMT 2011) and of the Shared Task on Applying Machine Learning Techniques to Optimise the Division of Labour in Hybrid Machine Translation (ML4. Shared Task on Applying Machine Learning Techniques to Optimise the Division of Labour in Hybrid Machine Translation (ML4HMT-11), November 19, Barcelona, Spain, META-NET, 11/2011.
Zusammenfassung
We present a substitution approach for the combination of machine translation outputs. Using a translation template derived from the output obtained from a rule-based translation engine, we identify parts of the template that could possibly be improved by adding in segments from other MT output. Substitution candidates are determined based on their part-of-speech. Alternative translations from the additional engines are retrieved by using word alignment. Substitution is based on several decision factors, such as part-of-speech, local left-/right-context, and language model probabilities. Our approach differs from other methods as it puts its main focus on preserving the syntactic structure inherited from the rule-based translation template. For the language pair Spanish-English an improvement in BLEU score can be observed.