Posedición y paridad humano-máquina en traducción automática neuronal: Un estudio empírico desde la traducción profesional

Abstract Albeit controversial, human-machine parity in neural machine translation has been claimed, but there is a lack of empirical studies providing concrete evidence. This article aims to present part of the results of an experiment conducted under the NEUROTRAD project where human-machine parity in neural machine translation is approached from the perspective of post-editing effort. Combining descriptive statistics, Friedman test for hypothesis testing, and Spearman correlation coefficients, our study analyses the answers to two questionnaires by 36 Spanish freelance translators in order to test 1) whether the type of translation —human, machine translation, and, as a novelty, post-edited machine translation— impacts on the perceived post-editing effort, time, and quality and 2) whether the participant’s professional experience correlates with the perceived post-editing effort, time, and quality. Results revealed that, on the one hand, the type of translation has no influence on the perceived post-editing effort (p<0.953), time (p<0.439) and quality (p<0.390) and, on the other hand, having more post-editing experience does not correlate with 1) perceived post-editing effort for post-edited machine translation (p<0.094); 2) perceived post-editing time for human translation (p<0.259) and post-edited machine translation (p<0.187) and 3) perceived translation quality for post-edited machine translation (p<0.10).