Charlie Bavington

French to English Freelance Translator - I.T. specialist

Bringing a pragmatic eye to meeting your needs

PEMT haze

June 26th, 2015 | Categories: agencies, business, translation issues

This year thus far appears to be the year, in some quarters, of getting mighty indignant about the evils of post-editing machine translation (PEMT).

Might as well stick my oar in. It’s been a while.

There seem to be two basic angles of attack employed by those getting their knickers in a twist about PEMT.

First, the financial aspect. PEMT is evil because it’s poorly paid, both absolutely and relatively. Absolutely, because the per-word (or hourly, as the case may be) rate offered is rarely generous. Relatively, because as we know, review/editing is always a fraction (say a quarter or thereabouts) of the overall cost of a translation, and with the first draft of that translation now done free of charge by a computer, you’ll find yourself being offered a fraction of the money (say a quarter or thereabouts) to produce any given translation job.

But that doesn’t make PEMT evil per se. It just means the adopters of the PEMT model are the kind of operators who compete more or less solely on price and endeavour to exploit their suppliers to screw them for all they can get out of them. There does seem to be a correlation between those offering PEMT work and those offering somewhat less than generous rates for conventional translation, but if you’re looking for a fair day’s pay for a fair day’s work, these outfits are surely best avoided, full stop.

Consider this. If you were offered €1 per word for PEMT, would you do it? If you would (and I would), then the argument merely becomes one of rate negotiation or market segmentation, or possibly both, and not one of PEMT being evil in itself because the remuneration offered is rotten.

Second, at least a couple of anti-PEMT pieces I’ve seen have attempted to compare the usual human translation process against PEMT in order to reach the conclusion that the use of MT followed by PEMT is a more time-consuming way to reach an acceptable target text. There may be some truth in that, but it is not a phenomenon restricted to PEMT alone. If we take such procedures or workflows and insert “assign translation to incompetent arse” in the place of “run text through MT”, we get exactly the same long-winded process. Any process where the first translation is barely comprehensible crud will have the same result. This is not an exclusively PEMT problem.

I have seen a third angle, usually presented more calmly for some reason although it is arguably the most bonkers, namely the idea that PEMT somehow harms the translator’s faculties and adversely affects their subsequent translation output. I’m not sure what to make of such claims, other than to suggest that if you find this happening to you, then it’s probably a good reason not to forge a career in PEMT. On no scientific grounds whatsoever, I find it hard to believe such an effect would be produced in everyone (see, I can talk bollocks as well as the next man), therefore I would not seek to deny everyone the chance to forge a career in PEMT, not for this reason.
(If you do find your own output seriously affected by your environment, then I hope you are careful about the company you keep, where you live, the books you read, and so on.)

So it pays poorly, takes ages and turns your brain to rice pudding. Nonetheless, it is almost certainly here to stay. MT is a permanent feature on the landscape now. Unless its output improves markedly, it will always need editing, when used for purposes other than gisting.

I would therefore suggest that the job of MT post-editor is likely to grow, more PEMT work is likely to be on offer, eventually perhaps by some otherwise respectable operators; it would therefore be more sensible to accept it and strive to ensure it is properly remunerated for the time and skill it takes. Rather, than, say, post metaphorically spit-speckled rants twice a quarter about how evil it is. Wouldn’t it?

  1. June 27th, 2015 at 07:06
    Quote | #1

    Hi Charlie,

    I’m going to be brave, although there are others far more knowledgeable on the subject who will hopefully come forth. In no particular order:

    1) PEMT will never be remunerated for the time it takes because this flies in the face of the whole business model. I am as certain of that as night follows day. If anything, you should be getting paid well over the odds for populating their machine with some half-decent quality.
    2) You are contributing to the false promises that these snakeoil salesmen make to the end clients.
    3) If language contagion didn’t exist we wouldn’t now have the bastardised version of English that is used across European institutions.

    • Charlie
      June 27th, 2015 at 13:30
      Quote | #2


      1) My objection is only one of unsound logic, or weak arguments that prove nothing. In this case, I’ve seen too many people saying, in essence, “PEMT is undesirable because it is poorly paid”. By that logic, nursing and street cleaning are also undesirable. I want to see “PEMT is undesirable because [a, b and c where a, b, and c are intrinsic to the activity] AND it happens to be poorly paid”, and then I’ll see if I agree or not.

      2) See above. I would in fact accept a criticism that my contribution adds little. Criticising those who are against a thing is not the same as supporting the thing in question, though. We could draw a parallel with the famous Proz red “P”. Probably a good idea, but extremely poorly implemented. Without implying that PEMT is a good idea, the current commercial implementation PEMT certainly seems to be poorly implemented from the translators’ standpoint, and the operators best avoided for a raft of reasons, including remuneration. Or as with the term “best rate”, it’s not the words themselves that are intrinsically undesirable, but more that the people using them are not usually the kind with whom a fruitful business relationship is very likely.

      3) I was careful with my words (or those ones, anyway – see also comment to Kevin!). The point of the post was to counter the stance (as I perceive it) of those saying “PEMT is always and only ever an undesirable thing” and supporting that stance with inadequate arguments. I start to sound like an extra from Last Of The Summer Wine as soon as I’m north of St Albans, so I know how easy it is to adopt features of style from one’s environment. We’ve all seen people whose native language is “polluted” by living in a source-language country. But it doesn’t happen to everyone or consistently, the point is we all need to guard against factors that lower the standard of our output, we all encounter them all the time, and the issue is not unique to PEMT and therefore, I would contend, not a convincingly strong argument against it intrinsically.

      Hope all is well with you, and thanks for responding :)

  2. June 27th, 2015 at 08:35
    Quote | #3

    Your assessment of the “third angle”, Charlie, is interesting as it ignores published research from the field which I have referenced on occasion. The mental effects of prolonged exposure to machine-generated content first came to my attention after Jost Zetzsche began working in this area and noted some difficulties for a time in his subsequent translation work. These effects were also acknowledged from the experience of an agency owner who was part of a webinar I did not long ago with Asia Online’s Dion Wiggins and Kirti Vashee, and this was confirmed again in the discussion with Dion and Scott at memoQ Fest this year.

    This is not a question of whether to engage in post-editing of machine pseudo-treanslation (PEMpT) or not. It is a basic issue of occupational health for which certain measures are necessary, just like you should get up and exercise at regular periods when sitting at your desk on a long work day.

    Of course, other environmental factors can also affect one’s work; try dating a kindergarten teachers and that will quickly become apparent in most cases, and many colleagues who spend excessive time correcting non-native work or the work of incompetents eventually lose their edge unless they expose themselves to a lot of better material.

    Here you appear to be too susceptible to the disingenuous propaganda of interested apologists like Ms. Massardo and others whose facts and motivations do not hold up under closer examination. Machine pseudo-translation processes will certainly be part of the language “landscape” in many roles, and I have never argued that they have no place. However, the current processes are simply not sustainable for all those involved, and the wishes of some and lies of others will not change that. Ergonomically and psychologically the current methods are deeply flawed. I don’t get involved in the economic arguments, because I frankly don’t give a damn. There are a number of better options available to those who finf PEMpT compensation unsatisfactory, including prostitution and drug sales. New methods, such as the non-interfering dictation workflows developed by David Hardisty at Universidade Nova show great promise for improving the usefulness of machine pseudo-translation, translation memory resources and similar reference material in translation work and doing so in a way in which the output exceeds what can be achieved with the sort of sausage-factory, damaging processes typically inflicted on the cannon fodder in our sector.

    • Charlie
      June 27th, 2015 at 13:32
      Quote | #4

      Hi Kevin,

      I should no doubt clarify that the post was mainly a response to postings on some of the more populist blogs and forums, where intellectual rigour is sometimes lacking, in my less than humble view, in that regard at least. If I added contemplation and discussion of the likes of TAUS into the mix, I’d never get any work done!

      Despite unwisely referring to the idea as “bonkers” (I’ll leave it, though, because to err is human and it would be unethical to remove it just because I’ve reconsidered), the point was that contamination is not, in my view, universally and predictably a consequence of indulging in PEMT for everyone (and therefore does not of itself make PEMT undesirable). By implication, therefore, it can be practised by some, sometimes. Which is (my interpretation of) what you said, provided appropriate safeguards are in place. Theoretically at least – it may well be that the currently implementation of PEMT processes requires foolhardy workloads, but as I mentioned in replying to Lisa (and as you also said), that’s a problem with implementation, not PEMT itself.

      It is no doubt to be hoped, to extrapolate a little from your final paragraph, and to address the “PEMT takes longer” argument, that advances in speech recognition could force a re-think on PEMT implementation, although other issues would arise.

      And please rest assured I’m not (intentionally) supporting apologists or anyone else. My beef was with the weak arguments against PEMT, not an attempt to support how it currently seems to be applied.

      Thanks for your comprehensive comments, of course :-)

      • June 27th, 2015 at 14:21
        Quote | #5

        Charlie, we both know that for most people, the implementation will comprise their full understanding of the technology, and I think it’s fair to say that much of the implementation is, in one way or another, pure professional crap. But you’ll hear the same from many of the MpT and PEMpT advocates who put the technologies in their proper context and presumably keep their own work in that scope. But when you look at the naked lies and idiocy of SDL’s marketing of MpT, etc. you quickly realize that the examples of what I might consider acceptable use merely serve the marketing carnival barkers out to pick the pockets of those suckers born every minute. Not our problem, really, as fools will find one way or another to be parted from their money and it might as well be for the technology which has been five years away from perfection for nearly 70 years.

        There aren’t many of the MpT crowd who are willing to engage real questions of productivity and examine the workflows properly, acknowledge their rather pitiful output compared to dictation alternatives or consider the rather interesting possibilities of combining dictation in a non-interfering PEMpT workflow as Prof. Hardisty and others have done. John Moran and some others get it, but too many would rather just grind linguistic sausage and ignore the human cost and the simple waste of time and money.

        • Charlie
          June 28th, 2015 at 15:11
          Quote | #6

          Is not the dismissal of acceptable use (whether in theory or practice and no doubt something of a subjective assessment) throwing a potentially useful baby out with some admittedly extremely murky bathwater? You seem to be saying that because of the generally despicable nature of the current purveyors of this technology, the whole shebang is beyond redemption.

  3. June 27th, 2015 at 10:45
    Quote | #7

    Hi Charlie:

    I agree with you.

    PEMPT is not evil per se.

    But what it does is evil. It’s purpose is to completely redefine our profession. I don’t know if you really believe what you are saying, or whether you just want to have some fun by ruffling a few feathers. I suspect it is the letter. Once when I felt the need to provoke some people, I wrote a post titled “Why Are All Sign Language Interpreters Women?” It created a veritable shit storm, but it was fun. Maybe that is what you are doing now, and if so, more power to you.

    But here is my argument:

    Nothing is evil per se. Hitler was not evil per se. He liked dogs, painting, art, built highways that Germans are still using, and gave a lot of people work.

    But what he did was evil.

    1. The whole purpose of PEMT is to save money that would have to be paid for real (human) translation. The compensation for post-processing will be always miserable, so your first argument is just silly.

    2. PEMT may work quite well for some translations from French to English. But believe me, from other languages, more complicated than French (different from English),such as Japanese or Czech, to English and vice versa, the result is total garbage and the whole exercise is counterproductive. Nevertheless, agencies will try to apply the same method to all languages and all fields as long as they think that there is a buck in it for them the way they did it with CAT software.

    3. Have you been doing some PEMTing already on the sly? If so, it may be starting to affect your brain already.

    I wish you peace and good health, brother.

    • Charlie
      June 27th, 2015 at 13:33
      Quote | #8

      Hi Steve,

      It would be fair to say I anticipated ruffled feathers. It would also be fair to say I’ve ruffled some feathers I didn’t expect to ruffle, and have not ruffled some I did (the ruffling and non-ruffling is not confined to these comments). But it’s not gratuitous WUMmery and trolling. I’m just asking people to consider the idea of PEMT separate from its implementation (hence the question “would you do it for €1 per word?”), and not everyone wants to, or perhaps doesn’t see the point of it. That’s their prerogative.

      That seems to address some of your point 1. See also my comment to Lisa about whether a thing is intinsically undesirable (downgraded from “evil”!) merely because it is poorly paid in the current environment.

      Your second point seems perfectly valid. It’s not useful all the time. I didn’t say it was. My argument is that is doesn’t always have to always regarded as always undesirable, because that was the standpoint that my post was seeking to counter. In the same way that criticism of criticism of X is not equal to support of X, saying that X is not always bad is not the same as saying X is always good.

      To answer point 3 – no. While trimming my post on grounds of length, I cut out a section about how much I dislike editing, reviewing, revising, correcting or proof-reading, and when I am reluctantly obliged to do so to keep a direct customer sweet (on English text produced by French speakers), they pay handsomely for it. Perhaps the answer should be “no, not knowingly”. Sometimes, I suspect it’s not inconceivable they have sought assistance from Mountain View.

      Keep on keeping on :)

  4. Charlie
    June 28th, 2015 at 15:14
    Quote | #9


    Prompted by a tweet by Shai Navé and other comments, it seems worth allowing some scope creep. The suggestion was that PEMT is undesirable because one’s efforts are added to the MT database for future use. It is, after all, a feature of MT systems that they need training and more data in order to improve (and thus perpetuate the cycle of exploitation seemingly unavoidable with current operators – hence its undesirability).

    So yes, it’s certainly an inherent feature of PEMT. Is it undesirable? I can only speak for myself, but I am certainly in the habit of providing unclean files to those who ask for them, and I’m not alone. They are produced anyway in my workflow. I assume they are used to populate TMs, and thus potentially provide fuzzy matches and therefore potentially lower incomes for future translators. Even if I didn’t provide the unclean version, using the translation to populate a TM is entirely feasible. Is providing the data to update agency TMs undesirable? I believe some people think it is; that is fair enough but I’m not one of them. So I’m prepared to shrug off that objection to PEMT when expressed in that way.

    However comma, as my history teacher used to say, it’s possible some underlying assumptions are different. When we hand over unclean files to agencies, we naturally keep the translation in a TM too, so we can re-use stuff. As I ploughed through quite a number of videos yesterday, I noticed that one firm underlined how translators could download TMXs if they wanted. Is it then standard practice for MT post-editors NOT to get to keep a copy of what they’ve done? Hmmm. If so, while clearly not an inherent undesirable attribute of PEMT (since it’s clearly not the only way to do things), it’s certainly more fuel for the “but the current implementation of PEMT is truly appalling” fire.

  5. June 28th, 2015 at 18:11

    Sorry not to offer anything of more weight or import, but”“assign translation to incompetent arse” did make me properly laugh out loud :)

  6. Paula Rennie
    July 9th, 2015 at 10:19

    What Rob said. And good to find your blog, Charlie – I miss you over at Strido.

    • Charlie
      July 9th, 2015 at 16:10

      Hi Paula – Well, I do sometimes miss the general quality of discussion on Strido (especially of late, as it happens), but sometimes other considerations have to take priority. That was one such time, I’m afraid. Still, onwards and upwards, eh? :-)
      pip pip,

Comments are closed.