Coherence in Machine Translation Output

Abstract

Coherence is a cognitive process. It plays a key role in argumentation and thematic progression. To be characterised by appropriate coherence relations and structured in a logical manner, coherent discourse/text should have a context and a focus. However, it receives little attention in Machine translation systems that considers the sentence the largest translation unit to deal with, the fact that excludes the context that helps in interpreting the meaning (either by human or automatic translator). In addition to that, Current MT systems suffer from a lack of linguistic information at various stages (modelling, decoding, pruning) causing the lack of coherence in the output. The present research aims at, first, capturing the different aspects of coherence, and second, introducing this notion in texts generated by machine translation based on sentence-by-sentence basis, in order to see and discuss the several phenomena that can lead to incoherent document translations with different language pairs.