Contextual BitextDerived Paraphrases in Automatic MT Evaluation.pdfVIP

Contextual Bitext-Derived Paraphrases in Automatic MT Evaluation Karolina Owczarzak Declan Groves Josef Van Genabith Andy Way National Centre for Language Technology School of Computing Dublin City University Dublin 9, Ireland {owczarzak,dgroves,josef,away}@computing.dcu.ie the string similarity of the candidate translation to Abstract one or more reference strings, and will penalize any divergence from them. In effect, a candidate In this paper we present a novel method translation expressing the source meaning accu- for deriving paraphrases during automatic rately and fluently will be given a low score if the MT evaluation using only the source and lexical choices and syntactic structure it contains, reference texts, which are necessary for even though perfectly legitimate, are not present in the evaluation, and word and phrase at least one of the references. Necessarily, this alignment software. Using target language score would not reflect a much more favourable paraphrases produced through word and human judgment that such a translation would re- phrase alignment a number of alternative ceive. reference sentences are constructed auto- The