Translation memory retrieval methods

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 202–210, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Translation Memory Retrieval Methods Michael Bloodgood Center for Advanced Study of Language University of Maryland College Park, MD 20742 USA meb@umd.edu Benjamin Strauss Center for Advanced Study of Language University of Maryland College Park, MD 20742 USA bstrauss@umd.edu Abstract Translation Memory (TM) systems are one of the most widely used translation technologies. An important part of TM systems is the matching algorithm that de- termines what translations get retrieved from the bank of available translations to assist the human translator. Although detailed accounts of the matching algo- rithms used in commercial systems can’t be found in the literature, it is widely believed that edit distance algorithms are used. This paper investigates and eval- uates the use of several matching algo- rithms, including the edit distance algo- rithm that is believed to be at the heart of most modern commercial TM systems. This paper presents results showing how well various matching algorithms corre- late with human judgments of helpfulness (collected via crowdsourcing with Ama- zon’s Mechanical Turk). A new algorithm based on weighted n-gram precision that can be adjusted for translator length pref- erences consistently returns translations judged to be most helpful by translators for multiple domains and language pairs. 1 Introduction The most widely used computer-assisted transla- tion (CAT) tool for professional translation of spe- cialized text is translation memory (TM) technol- ogy (Christensen and Schjoldager, 2010). TM consists of a database of previously translated ma- terial, referred to as the TM vault or the TM bank (TMB in the rest of this paper). When a trans- lator is translating a new sentence, the TMB is consulted to see if a similar sentence has already been translated and if so, the most similar pre- vious translation is retrieved from the bank to help the translator. The main conceptions of TM technology occurred in the late 1970s and early 1980s (Arthern, 1978; Kay, 1980; Melby and oth- ers, 1981). TM has been widely used since the late 1990s and continues to be widely used to- day (Bowker and Barlow, 2008; Christensen and Schjoldager, 2010; Garcia, 2007; Somers, 2003). There are a lot of factors that determine how helpful TM technology will be in practice. Some of these include: quality of the interface, speed of the back-end database lookups, speed of network connectivity for distributed setups, and the com- fort of the translator with using the technology. A fundamentally important factor that determines how helpful TM technology will be in practice is how well the TM bank of previously translated materials matches up with the workload materials to be translated. It is necessary that there be a high level of match for the TM technology to be most helpful. However, having a high level of match is not sufficient. One also needs a successful method for retrieving the useful translations from the (po- tentially large) …

Content truncated to 500 words