Research methods and system designs in machine translation: a ten-year review, 1984-1994

Abstract: 1 The five eras of MT Research on MT has passed through five eras to the present day. The first period began with the memorandum from Warren Weaver in 1949 which effectively launched MT research. The second began with the 1954 demonstration of a simple system for translation from Russian to English, which encouraged government agencies in the US and elsewhere to support large-scale projects. This period was brought to an end by the notorious ALPAC report in 1966, which highlighted the `failure' of MT research to meet its promises. The third, `quiet', era, when MT was virtually ignored, lasted until about 1975, with a revival of interest in Canada, Europe and Japan. Whereas the systems of the first two eras were generally based on thedirect' approach, the dominant framework after ALPAC was the various transfer and interlingual approaches based on linguistic rules. This fourth period lasted until the end of the 1980s with the emergence of corpus-based approaches (the use of bilingual text corpora, statistical methods, example-based approaches), and also the development of new rule-based methods using unification and constraint-based grammars. (For general descriptions of the historical development of MT see Hutchins 1986, 1988, 1993). 2 The situation in 1984 The revival of MT, which had begun some ten years after the ALPAC report of 1966, was firmly established by the mid 1980s. The SYSTRAN system had already been in use by the US Air Force for 15 years, for translating from Russian into English. The use of SYSTRAN at the Commission of the European Communities had started in 1977 with the initial English-French version, and since then a number of other language pairs had already been added. Since 1978 also SYSTRAN had been successfully operating at Xerox using a controlled English as input. Elsewhere, in Canada, the sub language system Meteo for translating weather reports had been in operation since 1976; Fujitsu introduced its ATLAS systems in the early 1980s; at the same time, the first commercial MT systems appeared: ALPS and Weidner; and they were to be followed in the mid 1980s by a number of other, mainly Japanese commercial systems. There were a few well-established research groups: the GETA team at Grenoble in France, the SUSY project at Saarbrucken in Germany, and the METAL project at the University of Texas (by now funded by the German Siemens Company). More significantly, the early and mid 1980s saw the establishment of a large number of MT research teams in Japan, primarily by the computer companies. There had been research in Japan in earlier years (notably at Kyoto University) but it came to international notice only in the 1980s. For Europe the most important development in MT research at this time was the establishment of the Eurotra project.