Post-editing is decidedly different from carrying out a “traditional” translation. Just as how completing a translation has intersubjective differences in terms of quality and time, similarly, considerable differences can often be noticed between post-editors in successfully completing a post-editing job. But with the same quality of post-edited text, what differentiates a fast post-editor from a relatively slower one?
This is not a futile question; indeed, several studies have focused on exactly this subject, halfway between linguistics and neuroscience. In fact, academic interest in these topics can offer very important insights into what skills, “tricks” and abilities a post-editor should have in order to be productive. While it is entirely reasonable to predict that the vast majority of workflows will pass through machine translation in the near future, it follows that the ability to make post-editing profitable is a crucial aspect for all linguists. In fact, considering the generally lower rates paid for post-editing, the time variable is the most important one to act on in order to produce more, in terms of turnover, in the same unit of time. So what suggestions can scientific research already offer to those who want to become a good (and fast) post-editor?
Scientific research and post-editors
Several methods have been developed in the scientific literature to evaluate the post-editing effort of a text. Some of them essentially consist of metrics that measure the number of changes made and are normalized for the total number of words. If these metrics generally correlate well with the quality of the machine translation output used, it has been highlighted that in some cases the post-editor’s cognitive effort does not correlate well with the number of changes made; there can be small changes which, however, consume substantial time (and therefore impose a greater cognitive effort from the translator), and vice versa. Clarifying the needs and cognitive effort required of post-editors and identifying the strategies that the best and fastest post-editors put in place to complete their work could give more scientifically and pragmatically based suggestions.
Some studies show that faster post-editors make better use of machine output by copying and pasting more often; rather than erasing and rewriting text from scratch). A good post-editor should therefore be able to understand how the machine works, that is, they should be able to rather reliably predict the output quality and the most frequent errors. In addition, the ability to more quickly address the errors that require a greater cognitive effort from the post-editor (and require more time) is crucial. The literature shows that the machine translation errors that require the most cognitive effort are idiomatic expressions, punctuation, errors in word order, incorrectly translated words and missing words.
Instead of paying equal attention to all possible problems, post-editors should instead focus on these types of errors and hone their cognitive abilities to promptly identify and resolve them.
English translation and adaptation by Sarah Schneider