Machine Translation Post-Editing
Joss Moorkens, Andy Way, Séamus Lankford · 2024
While there are many use-cases where the raw machine translation (MT) output suffices, in most cases, it needs to be post-edited by a human expert. This chapter examines the process of post-editing, from when it first came to be recognised as an integral step in the translation pipeline to the environments in place today to facilitate the task, and how they might be improved still further. One way of improving MT performance is to pre-edit the text, but this is rarely done nowadays. We distinguish ‘light’ post-editing from ‘full’ post-editing, where essentially two different levels of quality are envisaged. Despite post-editing being an essential part of most translation workflows today, and many studies existing which demonstrate its effectiveness, we examine the extent to which translators like doing it, and where this is not the case, what factors might be playing a role in that. We describe other methods developed to help translators in this task, such as interactive translation prediction, and summarise some of the negative impacts on translation quality which arise from post-editing, such as post-editese. Finally, with MT quality continuing to improve, we describe efforts to post-edit MT output for literature.