Automatic Acquisition of Translation Knowledge Using Structural Matching Between Parse Trees

Abstract

In this paper we present a rule-based formalism for the representation, acquisition, and application of translation knowledge. The formalism is being used successfully in a Japanese-English machine translation system. The translation knowledge is learnt automatically from a parallel corpus using structural matching between the parse trees of translation examples. We have developed a comfortable user interface, which makes it possible to invoke the translation functionality directly from MS Word. The user can customize the translation knowledge by simply correcting translation results in MS Word. Our system is mainly intended for language students, therefore, we also offer the display of detailed information about linguistic and translation knowledge, in particular token lists, parse trees, translation rules, and a single step trace mode to provide a better understanding of the translation process.