Pós-edição humana apoiada por anotações de erro: um estudo de caso para textos científicos de português brasileiro para inglês

Abstract

The internet and globalization have boosted the demand for translation services. In this scenario, the automated translation of texts, by means of computer programs, aims to meet part of this demand and improve the productivity of translators. However, problems with the quality of automated translation are accentuated in specific contexts, due to the characteristics of the domain, such as the translation of scientific texts, an area in which more than 80% of all world publications are in English. The improvement in the fluency of automated translations makes it difficult to identify such errors, demanding additional attention from proofreaders. A case study was conducted to investigate whether error annotations, generated by quality estimation models of automated translations, can support translators in the post-editing of scientific texts, translated from Brazilian Portuguese into English, with a view to increasing the productivity and quality of translations. No evidence of increased productivity was found, associated exclusively with the presence of highlights. The quality assessments are still under analysis and their results will be communicated in due course. Although translators in the highlighted modalities corrected 1.5 times more critical errors than those in the No highlighted modality, no significant differences were found in quality averages between the modalities. In terms of usability, participants acknowledged the influence of highlights on post-editing, but their perception of its usefulness was mixed. Keywords: machine translation; scholarly communication; quality estimation;