A multi-method inquiry into the cognitive effort in human translation and neural machine translation post-editing processes / Wang Yu

• Main Author: Wang, Yu
• Format: Thesis
• Published: 2023
• Subjects: P Philology. Linguistics
• Online Access: http://studentsrepo.um.edu.my/15408/2/Wang_Yu.pdf; http://studentsrepo.um.edu.my/15408/1/Wang_Yu.pdf; http://studentsrepo.um.edu.my/15408/
• Institution: Universiti Malaya

Due to the inaccuracies of Machine Translation (MT), Post-editing (PE) by human translators is inevitable. While most studies have reviewed HT and PE as end products, questions concerning whether the effort made by the human to polish an MT is worth it and which texts would be more efficient to be post-edited remain unanswered in the area of the cognitive translation process, especially from an empirical perspective. This study, therefore, adopts a multi-method approach, incorporating quantitative data derived from keystroke-logging and eye-tracking, with qualitative data from retrospective interviews to examine a) the differences in cognitive effort between HT and PE in terms of the whole process and the sub-phases (orientation, drafting, and revision); b) the differences in the cognitive effort between HT and PE according to text types; c) the underlying reasons for the different cognitive effort in HT and PE as well as in different text types in Chinese to English translation. A total of 25 participants from two different groups, a non-experienced group (15 students) and an experienced group (10 professional translators) were recruited to manually translate and to post-edit six texts of three text types (2 informative texts, 2 expressive texts, and 2 operative texts). Total task time, total fixation duration, fixation count, pause, and the number of keyboard activities were used as the indicators of cognitive effort. R language software was performed for statistical analysis. T-test and Liner Mixed-effect Model were used to test the significant difference and interaction between the variables. Thematic analysis was used for the qualitative data gathered from the retrospective interview. Through the comparison of the effort between HT and PE, this study concluded that PE emancipates significantly reduces human effort temporally, cognitively, and technically. The comparison of the three phases shows that translators tend to spend a long time reading the source texts before they carry out the PE works while spending a shorter time on drafting. Translators are also aware of the importance of revision after the HT and PE drafting work. The investigation of the influence of the text types on cognitive effort reveals that texts with different types bring an obvious change to translators’ time investment, mental process, and technical input in the translation process, and this change is also affected by the task type (HT and PE). Generally, in translating operative texts, translators benefit most from the “MT plus PE” mode from the perspective of effort saving. The interview data further confirmed that in translating expressive texts, the transfer of the mode is challenging, which requires translators to adopt a free translation, while in translating informative text type, the difficulty lies in the transfer of field, a literal translation method is thus adopted. Translators with different translation experiences also demonstrate different behavioural tendencies in HT and PE. Overall, student translators are more careful in carrying out the demanding task of HT while professional translators are more careful in carrying out the less demanding PE task. The study has practical implications for both translators and translator training.