The application of explicit semantic analysis in translation memory systems

Abstract

Although translation memory systems have become one of the most important computer-assisted translation tools, the development of systems able to retrieve Translation Memory (TM) files on the basis of semantic similarity has hitherto been limited. In this study, we investigate the use of Explicit Semantic Analysis (ESA), a semantic similarity measure that represents meanings in natural language texts by using knowledge bases such as Wikipedia, as a possible solution to this problem. While ESA may be used to improve TM systems, at present the evaluation of semantic processing techniques in the context of TM is not fully developed because the use of semantic similarity measures in TM systems has been limited. The study hence aims to evaluate ESA for its specific application in TM systems. The evaluation is performed within a knowledge management framework as this provides a suitable technical context. A software platform called the ESA Information Retrieval platform was designed to test the performance of ESA in TM system tasks using three different text genres: technical reports, popular scientific articles and financial texts. The aim of the evaluation was not only to improve our understanding of how ESA can be applied to TM systems, but also to examine certain textual factors that may have an impact on their performance. It was found that the use of ESA was able to create different ways of utilising translation suggestions. On the basis of the results obtained, both the existing problems of using ESA in TM systems and the future perspectives of TM systems are discussed. This study not only contributes to our understanding of employing semantic processing techniques in TM systems but also presents a new knowledge management perspective for the development of translation technology.