The Ins and Outs of Machine Translation Post Editing
by Andy Benzo
In the words of Benjamin Franklin: “An investment in knowledge pays the best interest.” Never has that phrase been more accurate than now. With all the recent technological advances, we need to be at the top of our game. Therefore, continuing education is fundamental.
On March 18, I attended an NCTA workshop on Machine Translation Post Editing (MTPE) presented by Carola Berger in Oakland. Carola is not only well-known for her presentations but is also well-versed in this topic; it was a brilliant investment!
Carola is the NCTA webmaster as well as Board Member. She holds a master’s degree in engineering physics and a Ph.D. in Theoretical Physics. She is ATA-certified in English <> German and has been working as a patent and technical translator and post editor. In 2022, she co-authored a presentation on qualitative metrics for the MT Summit of the American Machine Translation Association.
The workshop started with an overview of the history of machine translation (MT). We talked about rule-based MT in the 50s, the transformation to statistical MT in the 90s, and the arrival of neural MT almost ten years ago. Carola explained all these approaches to MT and the differences between them. We are now mainly dealing with neural machine translation. Furthermore, she pointed out ISO 18587:2017, which distinguishes between full post editing and light post editing. The process of full post editing is intended to obtain a product comparable to a human translation while light post editing yields a text that is merely comprehensible. We discussed how it would be impossible to handle the present-day volume of text and data that is translated through human translation alone.
To be a good post editor, Carola emphasized that we need to have source and target language skills, subject matter expertise, and excellent knowledge of terminology management and CAT tools. She also recommended using external grammar and spellcheck as well as QA tools. She offered a list of the most common MT errors, and a list of which documents are suitable for MT and which are not. She repeatedly stressed the importance of automating the translation process workflow as much as possible to save time: having a glossary and turning on terminology QA in your CAT tool, and third-party proofing tools, in particular PerfectIt, if you work into English.
One of the key questions many of us had was how to price our post-editing work; we examined various scenarios for calculating how to bill on an hourly basis to ensure profitability keeping in mind the state minimum wage and typical salary components such as healthcare, vacation time, and self-employment tax. We then talked about the importance of quality metrics. We examined the differences between BLEU, Translation Edit Distances and MQM, which is one of the best metrics. We looked at a sample scoring card from MQM and also its decision tree to evaluate error instances.
Carola concluded this portion of the workshop by discussing how to track errors found in order to provide feedback to the MT engineers refining MT systems.
And that was only half of the workshop! In the second half, we practiced post editing many English-to-English segments (since the workshop was intended to be language-neutral). A major takeaway from this workshop is that there is a lot we need to learn about how to post edit machine translation output and how it differs significantly from editing a human translation. The workshop was very interactive, and we all learned many tips and tricks on how to handle MTPE. This workshop explains in detail the tools and skills we must add to our toolbox.
It was great attending this workshop in person, as the exchange of ideas and being able to reconnect “live” always brings positive energy. We are witnessing massive changes in our profession because of MT and AI. We won’t stop the advances in technology, but we can adapt, adopt, and evolve.
Andy Benzo holds Law and Translation degrees from her native Argentina. She has been a freelance Legal Translator for over 30 years and specializes in civil lawsuits, labor union agreements, and wills and trusts. She joined the ATA in 1999, is ATA-certified, and is now an ATA Board Member. Andy has taught Legal and Business Translation for the T&I program at UCSD and has given many presentations at ATA conferences and other national and international forums.