Pós-edição e cognição: uma investigação empírica entre as traduções humana e automática

Abstract

The aim of this research was to analyze, in the reception of sentences translated from English into Portuguese, the preference of translators and non-translating English teachers among the types of translation: machine translation, human translation and post-editing. From the perspective of Natural Language Processing (NLP), the utilities of Machine Translation (MT) and its systematic evolutions were pointed out in the congruence relationship between human and machine translations. In parallel, it was considered the interface between Translation Studies and Psycholinguistics regarding the relevance of the translator's role in translation and post-editing tasks, reflecting on the mental processes involved in this context. Thus, the theoretical axis was based on the following themes: (1) translation studies; (2) natural language processing; (3) post-editions and cognition; and, (4) psycholinguistics and translation. Starting from an empirical and experimental type of research (COZBY; BATES, 2018), the systematics of effect observations, of this research, followed in two experimental groups, according to the professional occupation of the participants: (1) translators, which was subdivided into (1.1) translators and (1.2) translators and English Language (IL) teachers; and, (2) non-translators English teachers. The application of the experiment corresponded to a translation task with multiple-choice format, designed in the PsyToolKit software (STOET, 2010; 2017), involving sentences from machine translations, human translations and post-editions, in two types of textual genre: journalistic and literary. The research instruments used were: the consent form; the biographical questionnaire; the English language vocabulary test; and, the translation task. The statistical analyses of the data collection followed in descriptive and inferential form (GAZZANIGA; HEATHERTON; HALPERN, 2018), with application of the Chi-Square test (PEARSON, 1900) and the Mixed Linear Model (BATES et al., 2014), to verify the investigative theme.e results obtained considered the contrast between the translation most chosen by the participant, and the response time spent during the task. In this way, a ranking order was considered for these translations, in which the machine translations In this way, a ranking order was considered for these translations, in which machine translations were indicated as being the most preferred in the literary genre, while in the journalistic genre, the post-edited and the machine translations had an almost equal percentage, with the post-edited ones standing out slightly more. Furthermore, it was observed that the average response time was higher for the choice of post-edited sentences, suggesting a higher processing cost. On the other hand, machine translations had a shorter processing time compared to the other types of translation. Thus, as already mentioned in previous studies on post-editing, machine translation, and cognition (ALVES et al., 2016; KOGLIN; MOURA; SILVEIRA, 2020), it can be concluded that both machine translations and post-editions are relevant for translation practice in the technological setting, considering the time-benefit. Therefore, it is encouraged the academic community of the aŕea to continue empirical investigations in the axis of Translation Studies, psycholinguistics and translation technologies, in view of the cost-benefit and productivity of the translation professional in their practices in the labor market