Estrategia multidimensional para la selección de candidatos de traducción automática para posedición

Abstract

An efficient integration of a machine translation (MT) system within a translation flow entails the need to distinguish between sentences that benefit from MT and those that do not before they are presented to the translator. In this work we question the use of Krings' (2001) post-editing effort dimensions separately to classify sentences into suitable for translation or for post-editing when training predictions models and propose a multidimensional strategy instead. We collect measurements of three effort parameters, namely, time, number of post-edited words and perception of effort, as representative of the three dimensions (temporal, technical and cognitive) in a real post-editing task. The results show that, although there are correlations between the measurements, the effort parameters differ in the classification of a considerable number of sentences. We conclude that the multidimensional strategy is necessary to estimate the overall post-editing effort.