Specialisation versus machine translation

green_digitalAt the METM16 conference, I saw a talk that caused me to ponder something I don’t think about very much: machine translation (MT). I had tried it on occasions but had always quickly cast it aside. It just didn’t seem to help. So every time I heard someone talking about the importance of MT (for good or bad), I ended up feeling guilty for not paying attention to something that will, and probably already is, affecting my livelihood. However, this talk sparked a revelation into why I care so little for it.

The talk was actually three talks: one by an MT expert, Diego Bartolome, and two by users of personal MT software, Alan Lounds and Emma Goldsmith. Predictably, the expert sung the praises of MT. He seemed mainly to be talking about using MT as part of a post-editing process in which the translator gets the MT output to edit. So he wasn’t quite coming to us with a translator’s perspective in mind. In any case, he made it clear that MT is improving and becoming more widely used by language service providers.

After Diego, Emma and then Alan explained their experiences with two different MT tools. Because they are both translators, what they had to say aligned with how translators would use machine translation. They talked about the ins and outs of using personal MT products that you either customise yourself or that adapt to how you work. From what they said, I could see how these tools might be useful.

However, I still had doubts about how much I would benefit from such a tool given that it didn’t seem much of a step up from what some CAT tools are already doing, as someone at the talk pointed out. And I’m not talking here about using an external MT engine through the CAT tool but the native CAT functions for assembling segments. In any case, both these technologies might converge in the long run. CAT tools will become freelance MT tools and vice versa.

During the questions at the end of the talk, while I was sitting there trying to reconcile the overwhelmingly positive view of MT coming out of the talk with my overwhelmingly negative experience with it, Tim Barton asked Diego if speed comparisons had been done between translating with voice recognition software versus MT. As far as Diego was aware, no such tests had been done. (Logical, if you think about, given that Diego’s focus seemed geared towards a post-editing model in which the translator doesn’t do the whole job.) And with this question I thought I had my answer to why I’d written off machine translation: it probably saved little or no time compared to translating with voice recognition software, which was also more or less the point of the person asking the question. Of course, you only get this time-saving if you’re doing the whole job yourself.

So, thinking I’d solve the dilemma as to why MT had proven useless to me, I let the matter rest there. However, a few days after the conference, I did a job that made me change my mind. It was a legal/marketing text on real estate that turned out to have quite a few pages full of construction terminology. It was very dense. And so it occurred to me that I could use machine translation to work my way through the thickets of building terms. I thought I’d turned to MT because those positive comments about MT still echoed in my brain, but later I realised something else probably made me think of using MT at that moment.

So I used Google Translate. And it did help me get some technical terms. But after post-editing a couple of paragraphs of the MT output, I started to translate from scratch and only use Google’s attempts as a guide. Then I only used it occasionally as a dictionary for some terms. In the end, I found translating from scratch quicker than fixing up the MT output, even for sentences clogged with technical jargon.

With practice and using a better tool, I imagine I could streamline the process, but the main point is that I later realised why I don’t use machine translation. I surely had turned to it on this occasion because I was being stared down by menacing patches of text outside my fields of specialisation. I rarely translate subjects I don’t know something about, and I realised I’d never consider using MT for any field I’m comfortable with and definitely wouldn’t use it for those I know back to front. I just don’t think it would be quicker — even if I didn’t use voice recognition software. So my conclusion is that specialisation trumps MT.

I should say that often in the fields I know well, I do come across areas I’m not familiar with. In these cases, I sometimes have to do a lot of research, but that doesn’t bother me as I see it time well spent on getting to know my areas of specialisation better. So even in this case, I doubt I’d be tempted by even a near-perfect MT tool.

This is, of course, all from the translator’s perspective — when you’re translating the text from start to finish. An outsourcer can obviously benefit from MT because, following the post-editing model, they do the machine translation, and the translator just does the revising. So, after the initial investment and cost to use the tool, an outsourcer saves by paying less to the translator.

So now I know where my lack of interest in machine translation comes from. I’m sure my apathy will come back to bite me someday, but at the moment it seems warranted, at least for my situation. As I said at the beginning, it is possible MT has already affected my livelihood in some way. But, as one-man show that doesn’t rely on a high volume of jobs, it’s impossible for me to see any real changes.

Written by Rob

Rob Lunn is a freelance legal translator based in Spain. He translates from Spanish and Catalan into English.

5 comments to “Specialisation versus machine translation”
  1. Thank you for this article, Rob. Just beginning my freelance practice, I recently did a volunteer translation En>Es of a short waiver and release form. The non-profit provided a MT version they found for free online and I was a bit nervous to look at it, already wondering if it is a wise decision to focus my work on translation and I would soon be obsolete. I quickly realized I would need to start from scratch to do the client and the document justice, as trying to edit the MT version would have taken longer. I’ll still keep my eye on the progress of MT and adjust my business as necessary, but I do hope to have a rewarding career as a translator and not be replaced but rather assisted by automation.

    • Hi Rebecca! Your example certainly demonstrates the point of the post. Because you are (I imagine, judging by your website) very familiar with the subject matter, you instantly knew you’d be wasting your time with the MT output. MT would probably only have seemed helpful if you had had no experience with the subject matter and maybe not even then depending on your intentions (i.e., to learn something rather than just to get the job done quickly). Would you have gotten the same translation in that case and using MT? I doubt it without your current knowledge.

      I share your hope that there are still a few years left in this job. And yes, may it be a rewarding or at least half interesting job for those years! Thanks for commenting.

  2. Pingback: Legally Yours from Spain | Specialisation versu...

  3. I think a professional translator should run a session in which they show how they use MT. I’d be happy to volunteer!

    When your text contains, say, Universidad de Córdoba, then having a machine translation version containing University of Córdoba (which might be hard to dictate, especially if you want to keep the diacritic) is very useful.

    Even if the rest of the sentence is complete gibberish, I can still make use of the term that is correctly translated.

    My workflow might be as follows.

    -When I move to the row (in MemoQ), it automatically inserts the MT version and places my cursor at the start.
    -I translate my text up to the term, so my segment now looks, say, like this: “Her children studied at the His sons studied at the University of Córdoba, where they…” (so, the part before “His” is my human translation and the part after “His” is the machine translation)
    -Use shift+control+right to select “His sons studied at the” and hit delete.
    -If the term were more complex, at this point I would select the term and say “Google that” in the microphone to look it up, but it wouldn’t be necessary in this case.
    -Use shift-right to place the cursor after “Córdoba” and continue dictating.
    -When I reach the end of my translation and I’ve said “full stop”, I use the voice command “Delete and validate” to delete the remaining machine translation part to the right of the full stop and validate the row.

    If I’m dealing with a sentence where I don’t use the machine translation, I just translate as normal, but I say “Delete and validate” instead of “Confirm row”, so it literally takes no extra time, but the machine translation is there in case there’s anything that might be useful. If you don’t use voice-activated macros, you would just press shift-ctrl-pagedown (or shift-ctrl-end in DVX) then press delete.

    To sum up, while I agree that post-editing is often far slower, I also think that having MT switched on is useful and it doesn’t have to slow you down.

    PS Feel free to credit me as the “someone” who posed the question at MET!

    • A session like that would be good, and you sound like the person to give it! I admit I probably haven’t given MT a good enough go to be able to write it off completely. I’d be in a much better position to do so (or not) if I could see exactly how you someone like you uses it. PS You have been credited!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.