Nothing but promises, and no real innovations!

Replacing the first translator: revision of the target text without "knowing" the source language At the end of the 80s, hopes concerning automatic translation, i.e., translation carried out totally by a software program, were rather high. And naturally, we are not talking here about CAT (Computer Assisted Translation) of which Trados is the world leader as opposed to Transit, Déjà Vu, Star and other brands. CAT has come a long way and is used on a daily basis by the majority of professional translators. After around fifteen years, which have seen the accomplishment of an IT revolution (from total compatibility to the Internet), what is the real level of use and reliability of automatic translation systems? And, moreover, what are the results compared with the fundamental expectations of the time, i.e., the elimination of the first translator and their assignment to revisions of texts translated by machine? The double conference on automatic translation organized by Teleport, on 9 March 2005 in Brussels, was the occasion to find out the answers to these three questions. The two speakers, Dominique Orban of REVER and Josep Binet of the European Commission dedicated themselves to presenting The progression of the number of pages per year translated automatically within the European Union speaks for itself.

respectively "Les industries de la langue" ("Language industries") and "Machine Translation in a production environment! The example of the European Commission". In summary (see textbox), the responses were as follows. An increase in use of automatic translation but a stagnation in its reliability The application areas of automatic translation have been multiplied in an extraordinary way to the point that a new industry notion is born: that of "Language Industries" ("all activities in the research, development and commercialization of products and services rely on the automatic processing of languages", according to the definition provided by Dominique Orban himself). The studies of the Vandijk office in Brussels, which have been studying the phenomenon of translation for over twenty years—especially on behalf of the EU—were presented in detail. Thus, nine market segments were identified: document management content management production and editing linguistic resources intelligent search engines e-business traanslations e-learning vocal, multimodal interfaces From application to machine and car controls to legal translations for public institutions, the scope of the demand market was apparent. But there was a need to observe the lapse regarding performance levels in automatic translation. Since the 80s,

progress has been almost insignificant. In fact, it is linked to the work of "feedback" from translations obtained automatically and revised by translators. "Correcting" enriches the artificial intelligence of machines, that is to say by introducing semantic, ortho-syntactic and lexical corrections in the software. Private companies are unable to dedicate themselves to research into automatic translation This may seem paradoxical, but "machine translation" learning processes follow the same pedagogic principles of human intelligence and knowledge: quite simply, there is a need to correct and repeat. However, this requires enormous amounts of "feedback" work (see textbox) that is very costly in terms of manpower. What private companies are able to invest to a sufficient degree in this R&D task? During the past twenty years, we have witnessed a more or less naive or dishonest flood of attempts to produce new translation systems. Infant mortality decimated if not liquidated these attempts, which evidently underestimated the task. Just think of the initiatives from all the universities, provinces, regions and businesses that launched themselves into the adventure. Whereas the demand market continues to grow, that of supply (for the language industry) is sinking into financial difficulties resulting from very considerable investments. The hopes that we place on automatic translation remain, even though its application is still very (excessively) limited.

Automatic translation will never replace translators. Yet… Systran: automatic translation thanks to spying
Intelligibility of more than 95% for automatic translations At the beginning of the 90s, the Eurologos-Brussels office was the representative of the Systran system for the private sector in Belgium. The current European Union was (and remains) the owner of Systran for the public sector. Systran is still the best performing automatic translation system in the world for many language pairs. Its performances follow from "feedback" work that the American Department Defense (and, consequently, the Soyuz project for Apollo) put to good use in order to spy on the radio communications of the Russians from as early as the 60s, when the United States bought Systran from its inventor, the Hungarian Peter Toma. Thus the Russian-English pair reached a performance rate of around 96% (even at 90% the translated text is virtually unintelligible), approaching a level of linguistic translation quality allowing for revision of the English (target language) without knowledge of Russian (source language). A Benedictine work of feedback and the sabotage of the Luddites Obviously, the English-Russian symmetric pair performs a lot less well due to the absence of this corrective Benedictine work. It is on this basis that so many hopes vis-à-vis automatic translation, that was to have liberated translators from their tedious (and slow) work of first translation, were founded: thus they were to magically transform themselves into revisers and adaptors	of edited texts and this at a very high quality level and at low prices. The question that many ask themselves is why the EU did not assign 5 to 10% of its army of translators (over 1,200 at the Commission alone) to automatic translation and to "feedback" work nearly 30 years ago. Currently we could have had numerous well performing language pairs. On the other hand, the testimonies of cultural sabotage of Systran translation by translator-workers during the 70s-80s are numerous. This in the manner of the "Luddites", the English obscurants who were opposed to steam locomotives, seeing them as job destroyers…   A diagram of the most used language pairs in the European Union