Publications
Authors:
  • Sara Stymne
Citation:
Workshop on Extracting and Using Constructions in NLP, Odense, Denmark, 14 May 2009. <BR> The Full text is available <a href=http://www.sics.se/~mange/papers/constructions_workshop.pdf> online </a>
Abstract:
There are two ways to express definiteness in Danish, which makes it problematic for statistical machine translation (SMT) from English, since the wrong realisation can be chosen. We present a method for identifying and transforming English definite NPs which would likely be expressed in a
different way in Danish. The transformed English is used for training a phrase-based SMT system. We show significant improvements of translation quality, of up to 21.1% relative on Bleu, by performing identification and transformation of definiteness, compared to a baseline trained on original English, in two different domains.
Year:
2009
Report number:
2009/024