Learning machine translation strategies using commercial systems: discovering word reordering rules

Commercial machine translation (MT) systems seldom publicize their MT strategies; however, unlike well-documented experimental systems, they are easily available, some even as a free Internet service. Some of these systems seem to follow very closely a basic model (a sort of advanced morphological transfer strategy), described in detail in this paper, which focuses on the translation from English to Spanish, and, in particular, on the mechanisms used by the systems to obtain the correct word order for the target language. The basic model is so simple that a laboratory assignment based on it allows students to discover interesting details about the operation of a number of real MT systems, such as the reordering rules used.