Abu-MaTran: automatic building of machine translation

Incorporating Paraphrasing in Translation Memory Matching and Retrieval Rohit Gupta and Constantin Or˘asan RGCL, Research Institute of Information and Language Processing, University of Wolverhampton, Stafford Street, Wolverhampton WV11LY, UK {R.Gupta, C.Orasan}@wlv.ac.uk Abstract Current Translation Memory (TM) sys- tems work at the surface level and lack se- mantic knowledge while matching. This paper presents an approach to incorpo- rating semantic knowledge in the form of paraphrasing in matching and retrieval. Most of the TMs use Levenshtein edit- distance or some variation of it. Generat- ing additional segments based on the para- phrases available in a segment results in exponential time complexity while match- ing. The reason is that a particular phrase can be paraphrased in several ways and there can be several possible phrases in a segment which can be paraphrased. We propose an efficient approach to incor- porating paraphrasing with edit-distance. The approach is based on greedy approx- imation and dynamic programming. We have obtained significant improvement in both retrieval and translation of retrieved segments for TM thresholds of 100%, 95% and 90%. 1 Introduction Translation Memories (TMs) are tools commonly used by professional translators to speed up the translation process. The concept of TM can be traced back to 1978 when Peter J. Arthern pro- posed the use of a translation archive (Arthern, 1978). A TM system helps translators by retriev- ing previously translated segments to extract the relevant match for reuse. TMs also help them in c⃝ 2014 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. maintaining the consistency with previous work and use of appropriate terminology. Lagoudaki (2006) surveyed the use of TMs by professional translators in 2006, and 721 out of 874 (82.5%) replies confirmed the use of a TM. Although, extensive research has been done in Natural Language Processing (NLP) with empha- sis on improving the performance of automatic Machine Translation (MT), there is not much re- search on improving the TM systems by using NLP techniques. So far, most of the research in TM has been carried out mostly in industry with more focus on improving user interface and user experience in general rather than employing lan- guage technology to improve matching and re- trieval. Recent research (Koehn and Senellart, 2010; Zhechev and Genabith, 2010) on TM fo- cus more on improving machine translation using TMs. The TMs currently used by translators find matches for a given segment on the basis of sur- face form. This means that even if a paraphrased segment is available in the TM, the TM systems have no way to retrieve such segments. In this pa- per we try to mitigate this problem by using ex- isting paraphrase databases. To achieve this, we have incorporated paraphrasing in the TM match- ing process. A trivial approach to incorporating paraphrasing would be to generate all the possible segments based on paraphrases available. How- ever in this approach the number of segments in- creases exponentially and hence can not be applied in …

Content truncated to 500 words