Document-level translation quality estimation: exploring dicsourse an pseudo-references

Predicting the quality of machine translations is a challenging topic. Quality estimation (QE) of translations is based on features of the source and target texts (without the need for human references), and on supervised machine learning methods to build prediction models. Engineering well-performing features is therefore crucial in QE modelling. Several features have been used so far, but they tend to explore very short contexts within sentence boundaries. In addition, most work has targeted sentence-level quality prediction. In this paper, we focus on documentlevel QE using novel discursive features, as well as exploiting pseudo-reference translations. Experiments with features extracted from pseudo-references led to the best results, but the discursive features also proved promising.