Home Blog MT Post-Editing (MTPE) at ATA63 and Tips for Evaluating an MTPE Project

MT Post-Editing (MTPE) at ATA63 and Tips for Evaluating an MTPE Project

Machine translation post-editing

By Brooke A. Cochran

In October, I attended the American Translators Association (ATA) 63rd conference, the fourth one I’ve attended since starting my translation career. It didn’t disappoint. With its supportive and engaging atmosphere, the conference renewed my passion for being a translator. The variety of useful presentations was astounding as usual, making it difficult to choose just one to attend per session. While my roommates opted for presentations on interpreting, I mainly learned about productivity, audiovisual translation, and machine translation post-editing (MTPE). The latter is the focus of this blog post.

MT Then and Now

Just 5 years ago, machine translation and the MTPE task were controversial topics. People criticized them for many reasons, which I won’t go into here. However, the tone was different this year. MTPE was accepted rather than shunned and was touted as a reliable source of income to be handled differently than standard translation. There were around 10 sessions on machine translation and MTPE. I attended two but the session descriptions tell me they were all great resources for better understanding how to effectively approach this type of work. After attending Matthew Schlecht’s presentation “Confessions of an MT Post-Editor: A Report from the Trenches of the World’s Newest LSP Profession” and “20+ Linguistic Skills for Machine Translation Projects Beyond Just Post-Editing”  by Adam Wooten, plus the MTPE “Breakfast Club,” I was inspired to share the main messages I heard and to contribute to the conversation by offering my tips for evaluating an MTPE job.

First, the two sessions made it clear that MTPE is here to stay. From there, they diverged with Mr. Schlecht sharing his experience and insights as an MT post-editor and Mr. Wooten discussing skills for the future to make the translator a more valuable part of the MTPE process. I enjoyed how Mr. Schlecht presented MT engine mistakes as hiccups. It was a clever image to describe the various mistakes engines tend to make and to show post-editors what to look for so they can correct them. On the other hand, Mr. Wooten aimed to inspire translators to step outside the boundaries of post-editing and find ways to shift “from basic to better.” He suggested that translators can become the “lead linguist” and help clients prepare the MT engine so that it will provide better output.

MTPE Breakfast Club

I also attended the MTPE “Breakfast Club” on Saturday during the last breakfast and before the longest day of the conference. As conference attendees came in for coffee and bagels and broke into groups of friends and new acquaintances, our group stayed small, less than 10 of us, and we interacted like we all knew each other. For example, when someone dropped their knife on the ground, another member of our group rushed to replace it. That kind of respect made it possible to have both small talk and a mature discussion about this controversial topic.

Two main points were made. One reinforced what I learned in the above mentioned sessions: MTPE is not something to fear and will not mean less work for translators. Instead, it’s an opportunity to add a new skillset and source of income. The second point was that it is best to decide how much you will polish the MT output based on rate and client expectations (i.e., light or full PE, for publication or information only). If a client is paying you for a couple hours of work on a document that the client will be reading for information, then you polish enough to fill those two hours. It was agreed that over-polishing, taking more time than the rate covers or making it publication quality when that’s not necessary, is not desirable, since this results in a much lower hourly rate for the translator.

Tips for evaluating an MTPE project

Negotiating MTPE rates

It is best to charge an hourly rate, as many people do with regular editing. This frequently entails negotiating with the client. If the client works with you regularly and will not negotiate/accept a higher rate, then you just don’t accept the job. But if they leave room for negotiation, please advocate for yourself, and negotiate when appropriate. The worst thing they can do is reject your offer and then you don’t get the job, which you didn’t want at their rate anyway.

In my experience working with large translation agencies, negotiating is done via email or on their platform. On their platform is straightforward – I look at their rate and enter my negotiation, then wait for their counter or acceptance. Via email, I counter their rate and include an explanation, such as “I’m asking for this rate because of the abundance of MT errors, because of the poor quality, which will make it take longer to post-edit to publication quality…”

On the other hand, if this is a new client that you’d like to “win over,” you may want to accept the job at their proposed lower rate, within reason, and send them an email explaining that normally you would ask for a higher rate because of such-and-such reason. This lets them know that you won’t always accept their low rate and that you have a perfectly good, professional reason for that.

Evaluating the project

First of all, you want to look at it through the eyes of an editor more than those of a translator. As I learned at the conference, MTPE is like editing poor quality human translations. You want to check that the source text has been translated accurately and ensure it reads smoothly

Next, as you evaluate a project, you want to consider how much time various aspects will take to edit. I suggest using a numeric scale to rate them from 1-10 with 1 denoting no/little time and 10 indicating lots of time. You may find a different method and that’s fine. This is just a starting point. Rate each of the features below; then, based on the spread of numbers, consider whether the client’s proposed rate is appropriate. If it’s not, ask yourself, will it take enough time that you should ask for more than your usual rate? If the answer is yes, then how much more will you ask for OR will you turn down the job?

Here are some important aspects I suggest evaluating. You only need to scan the MT output in your CAT tool to get a general idea. No need to check every single word. These aspects are a starting point. Feel free to add what works for you and your specialization.

Aspect No. 1:

Accuracy of the MT output: How accurate is it? Are terms translated correctly and consistently? Are numbers reproduced correctly or did the MT engine insert fuzzy matches that changed the number completely? Poor translation quality, incorrect figures, and incorrect terminology will take time to correct.

Aspect No. 2:

Number and length of unedited MT segments and fuzzy matches: Use the filter to view only these segments. Are there just a few of them scattered among 100% matches or do they make up the majority of the document? What kind of information do they contain? Only or mainly numbers, or brief headings and titles? Or are they longer, detailed sentences? The latter take longer to translate than the former.

Aspect No. 3:

Repeats: This is similar to above. Use the filter to view only the repeats. Based on what you see, if you translate one segment, will it complete 100 repeats or just 1? Or think of it as, are there a few segments that repeat dozens of times or are there dozens that repeat once?

Aspect No. 4:

Number of tags: How many are there? Is the output riddled with them or are there only a few, confined to headings and the like? How many tags are there within individual segments? A single segment with 15 tags can be a pain to edit, especially when this occurs over many instances. As you can guess, the more tags there are, the more time you will spend ensuring their correct placement.

Thank you for taking the time to read this post. I hope I have allayed some of your concerns about taking MTPE jobs. I’m just another translator like you who has learned from experience. Therefore, if you have a different approach to evaluating these jobs, great! Feel free to share. The more the better so that readers can experiment and find an approach that works for them.

Author biography:

Brooke A. Cochran began her translation career in Seattle, WA while attending the translation program at Bellevue College. She has now been a full-time translator for 7 years. She specializes in the medical field, in particular clinical trial and pharmaceutical documentation, but also enjoys translating subtitles for cultural TV from France and Canada. To learn more about her, visit her website and you can find her on LinkedIn.

You may also like