The State of Machine Translation in Europe

This first half of this general survey covers MT and translation tools in use, including translators workstations, software localisation, and recent commercial and in-house MT systems. The second half covers the research scene, multilingual projects supported by the European Union, networking and evaluation. In comparison with the United States and elsewhere, the distinctive features of activity in Europe in the field of machine translation and machine-aided translation are: (i) the development and popularity of translator workstations, (ii) the strong software localisation industry, (iii) the vigorous activity in the area of lexical resources and terminology, (iv) and the broad based research on language engineering supported primarily by European Union funds. 1. Translator's workstations Recent years have seen great advances in the development and exploitation of support tools for translator. The four most widely used translator's workstations originate from Europe: Trados' Translation Workbench, IBM's TranslationManager, Eurolang's Optimizer, and STAR's Transit. In addition, Europe has been the centre for most of the background and current research on workstations: the TWB project funded by the European Union, and the Commission's own EURAMIS project. For professional translators, the attraction of the workstation is the integration of tools from simple word processing aids (spelling and grammar checkers) to full automatic translation. The translator can choose to make use of whichever tool seems most appropriate for the task in hand. The vendors of these systems always stress that translators do not have to change their work patterns; the systems aim to increase productivity with translator-oriented tools which are easy to use and fully compatible with existing word processing systems. In facilities and functions, each offer similar ranges: multilingual split-screen word processing, terminology recognition, retrieval and management, translation memory (pre-translation based on existing texts), alignment software for users to create their own bilingual text databases, retention of original text formatting, and support for very wide range of European languages, both as source and target languages. Integration to MT systems is now provided by three of the workstations. In the case of Trados access is provided to the Transcend software from Intergraph; IBM Translation Manager links up with Systran; and Eurolang Optimzer with Logos. After its disappointing experience with Eurotra, the European Commission has devoted most of its research support to the development of practical tools for translators and to the creation of essential lexical resources. Most of these projects will be described later. A major project was the TWB (Translator's Work Bench) project -within the ESPRIT framework -which began in 1989 and ended in 1994. The project, led by Triumph-Adler and involving 10 members from companies and universities, investigated the requirements of translators and proposed most of the features which are now commonplace in translator workstations: multilingual editor, document converters, access to lexica and terminology databases (e.g. Eurodicautom), access to MT systems (in this case, METAL), tools for term bank building, pre-translation and translation memory, and in particular a tool kit (System Quirk) for the analysis of texts and the development of lexical resource databases (thesauri, knowledge bases) from corpora and term banks. (A full description of the TWB project has recently been published: Kugler et al. 1995). A second project in this area has been TRANSLEARN, an LRE project for an interactive corpusbased translation drafting tool (a prototype translation memory system) based on EU regulations and directives from the CELEX (European Union law) database. The languages involved have been English, French, Portuguese and Greek. A third project, this time funded by Volkswagen, is investigating the use of a domain knowledge base integrated with a linguistic database as a translation tool; the languages are German and Bulgarian. The Translation Service of the Commission itself is now developing its own workstation: EURAMIS. The aim is to optimize the efficiency of the translation resources already available (e.g. the termbank Eurodicautom), to create a database of translated EU documents (as a 'translation memory'), and to provide easy access to MT systems. It will allow individual translators to develop their own tailor-made resources and facilities, with tools for text corpus management, glossary construction, and text alignment. A particular emphasis will be on the integration of MT and translation tools, including the mutual enrichment of Systran dictionaries and Eurodicautom lexical databases. The European strength in the area of terminology continues. A recent list of termbanks from InfoTerm in Vienna reveals that over two thirds of the 100 terminological databases recorded are based in Europe. The links between terminologists and translators have been a marked feature on the European scene -links which are now extending to the MT community. Terminology and the lexicon was the topic of an EAMT workshop in 1993 (Steffens 1995), and this August another EAMT workshop was held in conjunction with the international terminology conference in Vienna.