Extrinsic Evaluation of Sentence Alignment Systems

Parallel corpora are usually a collection of documents which are translations of each other. To be useful in NLP applications such as word alignment or machine translation, they first have to be aligned at the sentence level. This paper is a user study briefly reviewing several sentence aligners and evaluating them based on the performance achieved by the SMT systems trained on their output. We conducted experiments on two language pairs and showed that using a more advanced sentence alignment algorithm may yield gains of 0.5 to 1 BLEU points. Posted at the Zurich Open Repository and Archive, University of Zurich ZORA URL: https://doi.org/10.5167/uzh-62565 Accepted Version Originally published at: Abdul-Rauf, Sadaf; Fishel, Mark; Lambert, Patrik; Noubours, Sandra; Sennrich, Rico (2012). Extrinsic evaluation of sentence alignment systems. In: Workshop on Creating Cross-language Resources for Disconnected Languages and Styles, Istanbul, 27 May 2012 27 May 2012, 6-10. Extrinsic Evaluation of Sentence Alignment Systems Sadaf Abdul-Rauf, Mark Fishel, Patrik Lambert, Sandra Noubours, Rico Sennrich LIUM, University of Le Mans, France sadaf.abdul-rauf,[email protected] Institute of Computational Linguistics, University of Zurich, Switzerland sennrich,[email protected] Fraunhofer FKIE, Wachtberg, Germany [email protected]