Pronominal anaphora and verbal tenses in machine translation

Abstract

Coherence and cohesion are discursive properties of a text. They hold together pieces of information, making the text comprehensible, and not just a group of sentences put together. In our work, we investigate two different linguistic devices of cohesion and coherence, i.e., pronouns and verb tenses, and their implications for machine translation. We look at two different approaches for pronoun translation: rule-based translation with classic anaphora resolution and cross-lingual pronoun prediction without anaphora resolution. Concerning verb tenses, we assess the usefulness of grammatical tense and boundedness to improve their machine translation. All our experiments concern the translation from English to French, but they can potentially be applied to other language pairs.