Changing Expectations on Translation Quality
Against the backdrop of the development of artificial intelligence (AI), machine translation has advanced to a stage where it can change expectations on translation quality.
Translation is a difficult problem for machines, and hence, machine translations are imperfect and require human post-editing. Post-editing is the process whereby a machine-generated translation is human-edited to produce an acceptable final translation.
Speed is one of the key advantages of machine translation. Too much time spent on post-editing by humans chips away from machine translation’s advantage.
How much post-editing is necessary? Answer: enough to meet the client’s expectations of quality. Importantly, these expectations need to be managed and negotiated.
In the world of post-editing, people speak of “light” post-editing where the focus is mainly on accuracy and “full” post-editing where style and fluency also come into play. So do we give clients a choice of “light”, “medium” or “full” post-editing? Alternatively, it might be easier to define expectations of quality in terms of the types of errors (and number of each type of error) that may be tolerated in the final product: for example, a maximum of two grammar mistakes.
On the side of human post-editors, there is a need to retrain today’s translators into tomorrow’s efficient post-editors and monitor post-editing to prevent over-editing (and under-editing).
At DeepTranslate, our vision is to transform professional communications, intelligently. We hope to transform expectations on translation quality — for both clients and post-editing professionals — into a new normal so as to obtain the most benefit from state-of-the-art machine translation.