Machine-Aided Human Translation and the Paradigm Shift
Pierre Isabelle · 1993
1. The speed at which MT researchers are abandoning or supplementing their traditional rule-based paradigms in favor of corpus-based approaches provides a striking proof of their general lack of enthusiasm for the results of the last three decades of MT research. No doubt, this feeling is fueled in part by the fact these results have not allowed the MT business to penetrate any more than a marginal segment of the global translation market. 2. Martin Kay [3] was already contemplating this failure when, in 1980, he forcefully urged the MT community to shift its near-term goal away from classical MT towards machine-aided human trans-lation (MAHT). While many MT researchers probably felt that Kay's proposed program was a more sensible way to go, most of them declined to sign up for the journey and preferred to stick it out for another decade on their familiar path towards an improbable place. 3. This reluctance to give any serious consideration to MAHT clearly had something to do with the then-prevalent rule-based paradigms of MT. It is in fact very hard to single out, within three de-cades of rule-based MT research, a single result that has any obvious and immediate potential from the point of view of MAHT. There was just no straightforward way of applying the core techniques of rule-based MT systems to the development of translation support tools.