Joining translation and statistics?
Abstract
We tackle the foundations of statistical machine translation by means of an exercise presented almost fifteen years ago by the senior researcher Kevin Knight in the prestigious Artificial Intelligence Magazine. Specifically, the exercise consists of a challenge to the reader: the goal is to get good quality translations between Arcturan and Centauri. The fact that this two languages are unknown for the reader allows to cope with this problem from a statistical point of view. In this context, the basic assumptions typically made within statistical translation framework as well as the strong and weak points of the method become apparent. Finally, on account of the exercise developed with extraterrestrial languages, a brief reflection around the challenges presented by current machine translation is carried out.
