Article,
Has There Been a Revolution in Machine Translation?
Machine Translation, 16 (1): 1--19 (2001)
Abstract
When we compare the contributions on MT in the proceedings of Coling 1988 and Coling-ACL 1998, it seems obvious that in the period between them a revolution has taken place. Often this intuition is formulated as the replacement of linguistic approaches by statistical approaches. On closer inspection, however, this position cannot be defended. An analysis of Rosetta, concentrating on the different levels of discussion and of underlying assumptions, shows that the choice of knowledge from linguistic theories or information theory and corpora is by itself not a decisive issue. More important is the question of how the problem to be solved by an MT system is defined. An analysis of the decisions underlying Verbmobil, resulting in a list corresponding point by point to the one for Rosetta, shows how far-reaching the new approach to defining the problem of MT is. As it is shown that these systems are representative of the work in MT as it was done ten years ago and today, it can reasonably be argued that a revolution in MT has taken place, though not in exactly the way it is often believed.
