Computer-Aided Translation: Where Are the Problems
The paper addresses the problems involved in setting up a computer-aided expert system for human translators. It postulates four modules which function as an interactive reference guide in the framework of a translator's workstation: (1) a language component containing the syntactic and lexical correspondences between the SL- and TL-systems, (2) the >textual differential between the various types of SL- and TL-discourse, (3) the encyclopedic or knowledge base(s) underlying bi- or multilingual information transfer, and (4) a repertoire enabling the translator to word-process texts written in different languages.Emphasis is laid on the need for empirical research into the data culled from existing translations, parallel and background texts, diverse linguistic and encyclopedic as well as specialist reference works, and from practical experience of highly competent translators.
For several decades researchers and practitioners, R & D establishments and commercial entrepreneurs have time and again been offering promises and retractions with regard to machine translation programmes. Through increasingly concerted efforts, they have been trying, more or less successfully, to come up with highly specialized solutions for ever more specific translation tasks. The results have been a number of competing attempts to cope with extremely limited text samples at a very high cost. I am not going to comment on the ultimate success of MT. Clearly, it will forever entail the utmost specialization. You won't get quality, quantity, automation and speed in one bargain. The alternative still amounts to a tailor-made solution, and this will continue to be very costly indeed.