Hello, my name is Julie and I am a Gengo Language Specialist for the IT>EN and IT>EN-GB language pairs. I am a native British English speaker and I have been a translator and language teacher for many years.
In this article we will look at Machine Translation Post-Editing (MTPE) and offer some useful suggestions on the process to ensure quality in our post-edited machine translation.
Modern technology is developing at a breathtaking pace and affects almost all areas of our lives. The translation industry is no exception, and we have seen how the advancements of machine translation (MT) have influenced how we work.
Nowadays, it is increasingly common to see MT integrated into our CAT tools and, undeniably, it can speed up our work, but relying heavily on MT is not something we encourage at Gengo. However, there may be times when we are required to edit raw MT output, called machine translation post-editing, and often abbreviated to MTPE or sometimes PEMT.
What is MTPE?
Machine translation post-editing is the revision of raw machine translation by human linguists and can be divided into two categories: light post-editing (LPE) and full post-editing (FPE). The decision on which one to use will depend on the purpose of the text and client requirements and expectations of quality.
What is the difference between LPE and FPE?
The Translation Automation User Society (TAUS), a think tank and language data network, developed a set of guidelines to aid post-editing. LPE achieves a ‘good enough quality’, corrects major errors but may not be perfect. FPE aims for ‘almost human quality’.
How do we know how much to edit?
When performing LPE, we should aim to produce a ‘good enough quality’ whereby the MT output is accurate and without omissions, additions or mistranslations. Offensive or inappropriate content is edited. LPE ensures there are no incorrect spellings but is not overly concerned with punctuation, capitalisation or grammar rules such as adjectival agreement, etc. It uses as much of the raw MT output as possible.
FPE is applied when the requirement is as close to human translation as possible and the text is needed for publication or certification. We should ensure correct punctuation and capitalisation and accurate terminology. Adherence to target language syntax may lead to some sentences being restructured.
For FPE we need to make sure proper names, trademarks and other DoNotTranslate elements have not been translated by mistake. Measurements, date formats and locale-specific punctuation need to be checked. This is especially true when translating into both British and American variants of English. Always check the client style guide for the required target language spelling.
Mistakes are also made when abbreviations are not handled correctly: UE from Italian not being rendered as EU is an obvious one. But recently I saw ‘cmq’, which is a texting abbreviation of ‘comunque’, copied over into the target text. This made no sense in English and was obviously an MT error not picked up by the translator.
Recommended considerations
Check with the client or project manager on the required quality of the MT output. Do they require LPE or FPE?
Follow the TAUS Post-editing Guidelines which can be found here: https://www.taus.net/resources/reports/mt-post-editing-guidelines,
Finally, always proofread your work!
0 comments