Using Machine Translation Post-Editing for Urgent Tenders
Feb - 2020
When one of our long-standing clients needed a large Swedish document quickly translated into English, we had the solution!
The document was a detailed tender document, representing a significant opportunity for our client. As the translation needed to be read and digested by a group of people tasked with creating a response on a very short deadline, we proposed using post-edited machine translation (MTPE) to deliver the target English text.
We offered the client two levels of MTPE: ‘basic’, which would involve a human linguist or “post-editor”, who would work with the “raw” machine translation (MT) output to ensure solely that all the inputted content had been translated; and ‘light’, in which the post-editor would work on the raw MT output, reviewing and rewriting where necessary to provide basic accuracy and comprehension, as well as ensuring that all the inputted content had been translated. As the client only required an overall understanding of the document, they chose basic MTPE to be carried out.
In order to ensure that the best possible machine translation was produced, it was essential that the original Swedish document was refined before running it through the MT engine. We therefore engaged one of our native Swedish linguists to thoroughly review and prepare the source document for translation. Part of this preparation involved checking the document for any images or diagrams containing Swedish text. The MT engine would not have been able to pick up this text, thus it was necessary to transcribe it.
Once prepared, translated and post-edited, the result was an English text that was limited in style and accuracy, but that achieved the purpose of providing our client with a basic understanding of the original document. The final English document was then formatted to reflect the appearance of the original Swedish document.
In all, there were 22,000 words for translation. Ordinarily, this would have taken a human linguist 10-12 days to translate. However, through the use of MT, we were able to meet the client’s requirement of a quick turnaround and returned the final English document to them in 3 days.
Although this project was successful, it is worth pointing out that machine translation will never be as accurate as “pure” human translation. Despite this, in situations where time and budget are limited, it is an appropriate solution, as was evident in this project.