Machine Translation Post-Editing: Enhancing Efficiency and Automating Repetitive Tasks

Technology and productivity play an increasingly fundamental role in the localisation industry, which has responded to clients’ requests for ever-larger volumes to be translated at a rapid pace by integrating artificial intelligence into the translation workflow. This talk explores the under-researched field of productivity and workflow optimisation in the translation industry, focusing on analyses of machine translation post-editing tasks from Toppan Digital Language, a leading language service provider.

After a brief overview of the current landscape, it presents findings from the first large-scale quantitative analysis of speed in human translation and post-editing, based on real-world data for 90 million words translated by 879 linguists across 11 language pairs. It then introduces RECAP (Repetition, Error, Change, Action, Post-editing), the most detailed typology available to date for classifying the types of edits that linguists make to correct machine translation output.

Using this categorisation, a mixed-methods analysis of edit types was conducted on a small corpus of English-to-Italian post-editing tasks requiring different levels of effort from translators. Among a wide range of findings, the analysis revealed that linguists frequently had to make the same corrections repeatedly. This talk concludes by presenting a possible solution to this issue, i.e. AREA (Automating Repetitive Editing Actions), an algorithm that could automate up to 46% of repetitive corrections in post-editing in the English-to-Italian language pair.