Machine Translation: Tips for the translator evaluating a post-MT editing job before accepting it

After attending the ATA 63 conference, specifically several sessions* on post-machine translation editing (PMTE), I want to contribute to the conversation of how to make the best of PMTE jobs.

(*Note: I will say that the overall message I got was that more translators are accepting this technology. Per a discussion over breakfast, it’s here to stay and if you resist it, you’re really doing yourself a disservice. I'll also add that I’m aware this is a generalization, so there will of course be exceptions, i.e. translators who can get away with not taking PMTE jobs.)

How might a translator evaluate a proposed PMTE job and decide if it is worth their time to take?
After all, translators must examine any job before taking it. With PMTE jobs, the general idea is identical to pure translation jobs - make sure you’re comfortable with the content and can translate it by the deadline. Nevertheless, there are some concepts specific to PMTE that you should evaluate, too.

Below is a checklist of steps specific to PMTE, which I apply regularly. Feel free to use or adapt as needed for your own needs.

 1. Check the content and overall accuracy:

• Of course, by all means, look at the source when available. Also, to save time, look at the initial MT segments for clues as to the topic and document type. For example, headings, letterheads, table of contents, etc. Personally, as a medical translator, I avoid things related to chemistry and pathology, as these are subjects I’m not comfortable with. When I see section headings about these things, I will turn down the job.
• Also, scan the machine-translated segments for overall accuracy. I will check for terminology. How well the MT engine did its job is a major factor for me. If multiple segments show that it incorrectly translated basic terminology or arranged words in a confusing order, I know this will be extra work for me and will factor into my acceptance of the job.

2. Check how much translation will be required: I like to use the filter ”all except 100% and context match segments,” then scroll quickly through those. This gives me an idea of the extent of fuzzy match, unedited MT, and repeat segments. Are there 20 long segments? Or 100 short, numbers-only segments? The former takes a different amount of time to translate than the latter, and depending on the balance and rate proposed, I will decide if it’s worth my time.

3. Evaluate the repeats: The job description will typically provide a breakdown of the number of each type of segment. However, it is two different perspectives to read that there are 100 repeats versus seeing those 100 repeats. I’m checking to see how many segments in general (I don’t need an exact number) will be populated once I translate and commit the first one in the group. If half of the job is repeats, it will go quickly. But, if those 100 segments only make up 1/4 of the job, they will have less of an impact.

4. Evaluate the tags: How many are there, especially how many are in a single segment? (Disclaimer: I don’t completely understand tags, but I know they are used to maintain formatting.) Too many tags in a segment can be time-consuming, tedious, and down-right frustrating! This happens when the client does the formatting and leaves all kinds of “extras,” then doesn’t review the job in a CAT tool where those tags become clear. Many of them can be removed, so the translator can focus on the translation and not trying to put the tags in the correct place. Personally, if one or two segments are overrun with tags, that's fine. However, if a dozen or half of the project is, I will refuse the job.

In summary, the goal is to make sure you’re comfortable with the subject and to evaluate how much time it will take you to do the work based on a few parameters that can slow down or speed up the process.