This article orginally appeared in the May-June 1994 issue of Language Industry Monitor The Russian prototype was developed by a group of eight computational linguists and computer scientists from the former East German Academy of Sciences. The group, led by Gerda Klimonow, took the plunge into the market economy as the Gesell, schaft fur Multilingual Systems (GMS), a company set up in Germany for this purpose by a Munich, based software house. Working under the auspices of Sietec, the team was able to exploit Sietec’s development environment and the existing Metal modules. The AFA had selected Sietec to develop the system after a thorough, two’year evaluation of existing MT systems, with particular attention paid to their extensibility. In turn, Sietec discovered the Academy team at the Coling ‘90 conference in Helsinki demonstrating its PC,based German,Rus, sian verb translation system. The prototype has a good coverage of the grammatical complexity of the language pair. With some 7,000 entries, it boasted a set of lexicons consider’ ably larger than is usual for a prototype. It was demonstrated with texts from two domains, nuclear power and aeronautics. Naturally, the prototype was fine-tuned to these texts, and the system produced ready-to-use texts of a quality unknown to ordinary Metal users, but in no way did Klimonow pretend the system was perfect. She listed no less than eleven major grammatical obstacles that were as yet only partially resolved. Some of these are specific to Russian-German transfer, such as the interpretation of the Russian aspect category, which does not exist in German, and the generation of definite and indefinite articles in German, which are likewise unknown in Russian. The achievement shown in Berlin is due to a combination of expertise and thoroughness not often found in the MT world. Gerda Klimonow’s team has a long,standing record of work both in German and Russian grammar as well as in compu, tational linguistics. The team has maintained fruitful cooperation with well,respected sites in Russia and the Czech Republic, where — for want of hardware — they have for many years been forced to think, design, and plan before embarking upon implementation. In their strategy, they also stick to the often neglected distinction of linguistics, computational linguistics, and implementation, which in my experience is extremely expedient to all work in language technology. The speakers in the Berlin event had much to say about future plans. Peer van Driesten, head of GMS, argued in favor of reversing the language pair, tackling other Slavic languages, and taking other steps toward remote cultures. Gregor Thurmair of Sietec mentioned language pairs like Russian, Engish and Russian,french along with Polish and Ukraninan and dwelt for quite some time on the possibility of using Metal modules for non-MT purposes, such as information skimming and routing or index term generation in another languages. Stephan Boden (AFA) advised Sietec to pay more attention to the integration of the system with the customers’ work’processes, emphasizing the impor’ tance of the workbench concept. Despite the noble sentiments espoused about the future, the Russian-German Metal project is at the end of its funding, and it is uncertain whether the prototype will be developed into a marketable product. By acquiring a complete team from the Academy, Sietec has acquired expertise far above the value of the three years or so now spent in this project It would be a shame not to pursue the project. Sietec may not reach the break,even point, however, unless it develops the German,Russian system too. Only the latter truly addresses the needs of German industry rather than the information, gathering priorities of authorities who ordered this prototype. Technological know, how and developmental impetus cannot be put into the deep-freeze. Sietec should not drop the ball — or its competitive edge will be lost. |