Human strategies in translation and interpreting : what MT can learn from translators

Abstract

Translation - which we think of as a broader concept above written translation as well as interpreting - is basically a complex decision process. The decisions are based on available information. Translation problems arise when the translator does not have necessary information available at the moment of the translation. This is where translation strategies come into effect, which translators use consciously or subconsciously. We think that both forms of translation use basically the same type of strategies, which are, however, not easy to detect or to measure. Furthermore, we think that the model of translation as a decision process also applies to machine translation. In our paper, we try to prove this using the example of reduction as a translation strategy. Reduction is used both in written translation and in interpreting, but is more prominent in the latter. In our work, we focus upon dialogue interpreting, a non-simultaneous type used in face-to-face interactions. We try to outline how reduction strategies could be modelled in a machine interpreting system (such as VERBMOBIL), using the concept of the target of translation.