Utilization of Coreferences for the Translation of Utterances Containing Anaphoric Expressions

Abstract

One of the major problems of current machine translation systems is the lack of contextual knowledge which might result in misinterpretations of the translation. Especially, short sentences frequently do not contain enough information to achieve an adequate translation in the context of previous utterances. Contextual information can be achieved through the analysis of coreferential relationships of anaphoric constituents. The utilization of coreferences, however, is not restricted to the word selection of the anaphor itself. We propose a word selection method which analyzes coreferences and uses the semantic features of the antecedent for the translation of the sentence predicate.