Human Translation versus Post-Editing of Machine-Translated Figurative Expressions: A Pilot Study

Abstract

This research paper presents a pilot study of temporal effort and translation quality in human translation (HT) and machine translation post-editing (MTPE) by trainee translators. It integrates a process- and product-oriented approach comparing HT with MTPE with respect to efficiency (translation process) and effectiveness (translation product) in translating figurative expressions from Arabic into English. Ten senior translation students were selected to perform the two tasks in different sessions. In the HT task, participants were asked to translate Arabic figurative expressions into English without the assistance of machine translation (MT). In the MTPE task, they were asked to post-edit comparable English figurative expressions from the same source text that had been translated by Google Neural Machine Translation (GNMT). The Arabic novella, Returning to Haifa, by Ghassan Kanafani, served as the source text from which the figurative expressions were extracted. Its published English translation by Harlow and Riley (2000) was used as the reference translation. The study employs a mixed quantitative-qualitative approach. For observing the process, the keylogging software, Translog II, was used to track the time of the tasks. The human-translated and post-edited texts were evaluated by human raters using the Multi-Dimensional Quality Metrics (MQM). The study primarily investigates HT versus MTPE while including the BLEU score of the initial unedited machine translation output. Findings reveal that MTPE improved processing speed, while HT produced better quality. Thus, the study raises certain concerns about MTPE of figurative expressions, which should be considered when designing courses in post-editing of literary discourse.