Fuzzy-match repair using black-box machine translation systems: what can be expected?

Fuzzy-match repair using black-box machine translation systems: what can be expected? John E. Ortega jeo10@dlsi.ua.es Felipe S´anchez-Mart´ınez fsanchez@dlsi.ua.es Mikel L. Forcada mlf@dlsi.ua.es Dept. de Llenguatges i Sistemes Informatics, Universitat d’Alacant, E-03071, Alacant, Spain Abstract Computer-aided translation (CAT) tools often use a translation memory (TM) as the key re- source to assist translators. A TM contains translation units (TU) which are made up of source and target language segments; translators use the target segments in the TU suggested by the CAT tool by converting them into the desired translation. Proposals from TMs could be made more useful by using techniques such as fuzzy-match repair (FMR) which modify words in the target segment corresponding to mismatches identified in the source segment. Modifications in the target segment are done by translating the mismatched source sub-segments using an ex- ternal source of bilingual information (SBI) and applying the translations to the corresponding positions in the target segment. Several combinations of translated sub-segments can be ap- plied to the target segment which can produce multiple repair candidates. We provide a formal algorithmic description of a method that is capable of using any SBI to generate all possible fuzzy-match repairs and perform an oracle evaluation on three different language pairs to ascer- tain the potential of the method to improve translation productivity. Using DGT-TM translation memories and the machine system Apertium as the single source to build repair operators in three different language pairs, we show that the best repaired fuzzy matches are consistently closer to reference translations than either machine-translated segments or unrepaired fuzzy matches. 1 Introduction Computer-aided translation (CAT) tools often use a translation memory (TM), containing trans- lation units (TU), as the key resource to assist translators. TU are made up of source and target language segments and translators use the target segments in the TU suggested by the CAT tool by converting them into the desired translation. When an exact match (100%, s = s′) is not available one can use a fuzzy-match repair method to repair a translation proposal t. The aim of these methods is to replace the sub-segments in t that are the translation of the sub-segments in s that do not appear in s′ by the translation of the corresponding sub-segment in s′. Fuzzy-match repair is gaining more traction in modern tools such as DejaVu1 and MemoQ2 as a reliable method of replacing words in a proposed target-language (TL) segment t by using a source of bilingual information (SBI) such as the very same TM being used, a dictionary, or an on-line translation tool. 1http://www.atril.com/content/10-deepminer-fuzzy-matches-repair-458 2https://www.memoq.com/whats-new-in-memoq-2015 ������������������������������������������������������ ���������������������������������������� Ortega et al. (2014) describe a method that is capable of using any SBI for fuzzy-match repair. This method first aligns the words in the SL segment s of the TU being repaired (s, t) with the words in the segment to be translated s′ and identifies the mismatched words in s and s′, i.e. the sub-segments they do not have in common. It then uses the SBIs available to …

Content truncated to 500 words