Exploring Naturalness in MOOC Video Lecture MT Subtitles: A Case Study

Abstract

This case study investigates the naturalness of subtitle translation using massive open online course (MOOC) video lecture subtitles that were machine translation (MT)-generated by a computer-assisted translation (CAT) tool, Termsoup, after the pre- and post-editing phases. Both the pre- and post-editing phases were conducted by translator trainees while supervised by a professional translation trainer. With the accuracy secured, the naturalness of the MOOC translated subtitles from Chinese to English is further explored. In this study, distributions of lexical bundles (LBs) in terms of structural and functional properties are used to investigate the academic register in the MT output. The result shows that while in previous literature classroom teaching extensively uses a taxonomy of bundles in both written and spoken modes, the investigated corpus tends toward the ‘literate’ side rather than the ‘oral’ side. Nonetheless, the increase in the number of discourse organizers discovered in the corpus not only implies that the MOOC language may share comparable purposes with TED talks but also aids in the formation of spoken elements in the register. The study ends with suggestions for extending the corpus of MOOCs to gain a better grasp of specific formulaic languages in subtitle translation, along with proposals for pre-editing the source text to enhance the academic register before the MT process.

Keywords

• naturalness of subtitles
• massive open online course video lectures
• lexical bundles
• academic register
• machine translation output