Multiple Reference Translations For European Languages
Project leader: Eva Hasler
Desirable skills for participants: experience with Moses pipeline, C++/perl
This project will look at automatically creating multiple reference translations for the news/news commentary data sets. A problem with single-reference data sets is that insignificant changes to the translations can have an impact on translation scores, hence making it difficult to evaluate changes in performance. The idea is to create functional paraphrases of the development and test sets that include variants of frequent, semantically insignificant phrases and punctuation, but not variants of translations that would occur in a different context.