Use of Linguistic Resources like Translation Memories in Machine Translation Systems

EAMT Workshop, TKE’96, Vienna, Austria, 29 - 30 August 1996. Proceedings 101 Use of Linguistic Resources like Translation Memories in Machine Translation Systems Lee Humphreys Abstract The context and general technical strategy for MAHT and MT at ERLI is presented. After presenting the commercial context of translation, we outline the architecture of AlethTR, ERLI’s assisted-translation platform. Special attention is paid to the question of linguistic resources : general lexicon, terminology, translation memory. We consider briefly the integration of translation memory with MT and its extension to the sub-sentence level. Lee Humphreys Lee Humphreys currently heads the Grammar group in the Research and Development Department of ERLI. He is also involved in the design of translation systems. He was previously responsible for French-English MT development at SITE-EUROLANG (Paris) (1992-1994). Prior to this he was part of the CL/MT Group in the Department of Language and Linguistics at the University of Essex (UK), working on MT Evaluation and linguistic specifications in the EUROTRA project. GSI-Erli Created in 1977, ERLI is Europe’s leading language and document engineering company. With a turnover of MFF 31 in 94, the company has a team of 60 computational linguists and software engineers. The principal areas of activity are natural-language based indexation and document retrieval (AlethIR), text generation (AlethGen), controlled-language checking (AlethCL), translation (AlethTR), linguistic knowledge manipulation (AlethKES), terminology management (AlethGT) and general lexical management (AlethGD). Lee Humphreys ERLI 1 place des Marseillais F-94227 Charenton-le-Pont CEDEX France Tel: +33 (01) 48 93 81 21, Fax: +33 (01) 43 75 79 79 E-mail: info@erli.fr WWW: http://www.erli.fr 102 EAMT Workshop, TKE’96, Vienna, Austria, 29 - 30 August 1996. Proceedings Introduction AlethTRTM is a translation support tool from ERLI that integrates • bilingual terminology management • identification and default translation of candidate new terms • translation memory (TM) • light machine translation (MT) together with text handling, project costing software, translation utilities and user interfaces. As has become standard in translation engineering, the MT engine is used to translate - if required - text which is not found in translation memory. The grammar of the current engine can be customised by ERLI to suit the particular type of text routinely handled by a customer. Translation Context ERLI’s translation tool engineering effort has historically concentrated on supporting high-volume professional technical translation. It is worthwhile for the translator to invest a considerable amount of time and effort in a project preparation phase since this is recouped by improved quality and speed during the translation phase. A project typically involves from one hundred to several thousand pages, where all the texts in the project belong to the same domain and text type. The French texts we have looked at have syntactic and lexical restrictions characteristic of sub-languages: restrictions on verb form Verbs restricted to 3rd person, mostly present tense, limited use of past and future, perhaps complete exclusion of the French passé simple. Interrogatives are restricted to very specific document parts such as fault- finding e.g. Is the main circuit stop valve closed? anaphora …

Content truncated to 500 words