Levels in post-editing illustrated with popular science genre
Despite the stellar performance of machine translation engines in recent years which are impressive with high quality output, machines cannot understand text nor identify errors. As the world increasingly demand speed in disseminating information, post-editing has evolved to be a significant part of...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis-Master by Coursework |
Language: | English |
Published: |
Nanyang Technological University
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/148781 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Despite the stellar performance of machine translation engines in recent years which are impressive with high quality output, machines cannot understand text nor identify errors. As the world increasingly demand speed in disseminating information, post-editing has evolved to be a significant part of the translation process. There have been many studies devoted to studying errors made by various translation engines and recommending post-editing methods and strategies for specific domains. Post-editing guidelines were also proposed to overcome the varying factors that could affect the final product. The emergence of neural machine translation, nonetheless, called for a re-look into post-editing accompanying the technological breakthrough. This study expands on the post-editing dichotomy proposed by TAUS and illustrates the application of post-editing at different levels on popular science genre. Two machine translation outputs (Baidu and DeepL) and two human versions (one published translation and a reference translation done by a trained translator) were compared. Our analysis showed that if the goal of machine translation is to rival human translation, five levels of post-editing, namely at word level, sentence level, paragraph level, contextual level and inter-paragraph level need to be dealt with. Progressing from word to inter-paragraph, each level is a subset of the next, indicative of increasing post-editing efforts. Regardless of the purpose of modifying MT outputs, speed in PE could be enhanced when the post-editors are mindful of registers, sensitive to syntactic usage in both source and target languages, cognizant of the background and context as well as highly aware of cultural nuances. |
---|