A Hybrid Approach to Interactive Machine Translation - Integrating Rule-based, Corpus-based, and Example-basedMethod

Abstract

With rapid development of the Internet, demand is rising high for a personal tool to support writing foreign language document such as e-mail. However, translation result of an automatic MT system is often not satisfactory for this purpose and requires post-editing. In addition, a purely rule-based system does not necessarily provide a satisfactory result for specific expressions because of lack of corresponding rules, nor purely example-based system for expressions not covered by examples. A hybrid approach is worthwhile to pursuit, where automatic and interactive approaches , as well as rule-oriented and data-oriented approaches are integrated. In this article, we propose a hybrid interactive machine translation method that combines rule-based, corpus-based and example-based approach with an interactive man-machine interface. We show that the previously proposed rule-based model can be naturally integrated with different translation paradigms. The interactive operations, previously introduced and shown to be useful for disambiguation in the rule-based transfer, are shown to be also useful to control covering by and selection of the matching examples, two major decisions in the example-based translation method. We also mention an online learning scheme of translation pairs from the user interaction.