Monday, 26 May 2014

Two Sad Things about MT (+ Analysis of MT Copy)

Here are the two really sad things about MT:

  1. Companies will pay for MT — and the development of it, further research, staff to operate it etc. — in order to avoid having to pay translators.
  2. MT is now — ironically! — combatting its image of an easy thing to do, just like translation should. Not translation but MT is being presented as the respectable art or science which requires skill.

Look at Asia Online's banner close to the bottom of Pangeanic's blog post.

Here's also what they claim at AsiaOnline:

On the other hand, the complexity of high quality MT can be daunting. To succeed requires a significant amount of skill, a deep understanding of the different approaches to MT and MT technology, a deep understanding of the data used to customize MT and a range of tools, skills and knowledge that permit optimizing, managing, manufacturing and fine tuning of the data to deliver the highest possible quality. These skills are generally in short supply and are not available to most in-house. (Emphasis added.)

The text is a fragment of content-driven copy, i.e. text which still sells or achieves a PR goal (e.g. awareness, loyalty, incentive to switch), but does so through creating informative content rather than advertising.

Let me show you a couple of characteristic PR/marketing moments in the text. To avoid any misunderstanding: PR/marketing is not pejorative. It's not (necessarily) a dark sales ploy. It's persuasion and advocacy, and its techniques are often used without conscious deliberation by people who want to convince others of something. I'm not 'unmasking' here, just showing you how persuasion works here.

It 'leverages' scarcity, i.e. presents MT as rare goods, which sell better because demand outgrows supply.

It 'leverages' the techy side of MT by justifying the client in not being able to grasp it fully — it's too complex for you, don't worry, it's not your fault and you aren't any less a smart person if you can't fully grasp it. NOT: You can't hope to understand it, forget doing it on your own even if you're a full bilingual, just hire a professional and pay up — which is often the way we, translators, talk.

So, the MT guys keep the tech aspect well short of losing them the client's attention and interest. The result is respect and admiration rather than headache, loss of interest and dismissal of the idea as uncomprehensible and of doubtful viability. Which is different from what we, translators, achieve or even try. Rather, we tell the client we're necessary whether he likes it or not — which is not quite endearing — or we try to make ourselves useful, often in ways conflicting with self-respect, which clients and agencies do pick up on.

By contrast, look how MT dodges both!

The copy also still reassures the client that the highest possible quality can be achieved, even: 'delivered'. It would be unfair to make too much of 'deliver', since just about everybody else talks like that, and 'deliver quality' is pretty much an established collocation, but 'deliver' is still a bit of a marketing buzzword which 'leverages' mental associations with physical delivery (right in your mailbox, on time) or delivering on a promise.

In other words, the paragraph presents MT as something too complex for the average person or business to 'deliver' but still something a reasonable businessperson would want to use.

Like a good ad, the copy keeps the editorial style while unloading a whole slew of positive factors. It conveys just enough complexity to earn the reader's respect and concession of status, but not enough to make the reader lose faith in his own ability to make an informed judgement regarding its viability. Sort of like engine parameters in a sophisticated car ad or encyclopaedic details in a high-class food ad. See Ogilvy classics. I particularly have in mind the Rolls-Royce ad and the Guinness ad as good illustrations of the editorial/tidbit technique.

Other than a potentially transpiring notion that MT is capable of reliably achieving the highest possible translation quality — one can't disagree with the facts in that copy.

In short, they're doing their PR and marketing right. They're getting it right as far as copy goes.

The problem, from a translator's point of view, is that:

  1. A machine's skill to translate is valued. A human's is not.
  2. MT skills are clearly being the subject of a PR initiative to raise their profile, which means both awareness and status, plus positive emotions and associations such as respect, admiration, reliability and robustness.
  3. MT is an investment, human translation is a cost. Guess which one wins? Okay: Investment always wins over cost.
  4. R+D time and money, and marketing and PR budgets, go to MT and not human translation. There will be or already are clients who prefer — other things being equal — to put money in machine translation rather than human translation as a matter of preference.

Re: #2 but also in connection with #4, don't count on the MT-driven part of the 'translation industry' to raise the profile of human translation, make it respected, admired and well-rewarded. It's in their interest to keep the profile of human translation low — and lower — and raise the profile of MT instead.

Individual translations can't do that, either, because we don't have the 'leverage' and we didn't become translators to be half-time PR and marketing paraprofessionals.

Rather, it's up to translator associations to give human translators some serious PR. Can't do much of that when they rely on financing from MT-related companies, though.


  1. But Łukasz, you forget the important role that Asia Online assigns to humans in the HAMPsTr workflow. It is your privilege to serve the machine and help to refine its dictionaries, correct its syntactic stumbles. What higher purpose could one serve? Human translators are a renewable resource which could be mass-produced in Central European production camps if there were truly enough demand for them. But they don't say "to err is human" for no reason. Instead, with well-chosen engineering one can receive the linguistic wisdom of algorithmic beauty and wonder, more mysterious and latent with promise than they utterances of ancient oracles.

  2. Hello Łukasz. This is Manuel, from Pangeanic I must thank you for the time you reading our post and finding yourself on our side of MT overblown claims.
    We develop engines and tools which we have used for production internally here at Pangeanic. So, in one way, you could consider us another machine translation "demon", but our approach is radically different to the approach taken by AO.

    Pangeanic is a translation company employing full-time translators and a good bunch of freelancers, mostly dealing with work into the electronics fields. This means that we have historically dealt with pretty controlled language. Due to client pressures, machine translation was the only path we could find many years ago (with rule-based systems). Eventually, we met some researchers here in Valencia and we opted for developing our own "technology" to help us do our work - what you call "customized engines", fit for our work and our clients - a productivity enhancer.

    AO claims are often exaggerated and very overblown. They attack anyone who might look like a competitor. They claim a level of sophistication and technology knowledge which looks somehow funny if you follow their ex-employees profiles in LinkedIn. Our approach to machine translation is not radical. It is useful for certain things, it does help and we do not plan to squeeze translators out of a job. It would be unfair to charge a client 1000 new words if 999 had a 99% match from his/her TM. The same goes with machine translation. Some things simply have a short shelf life and clients are not in a position to pay standard rates. Different apples have different prices, they are still apples. It does not mean some apples are really bad and some are really good. It does not mean the farmer is out to cheat, I'm sure the farmer tries his best to grow the best apples and uses different techniques to maximize his crop.

    Translation companies using MT as a "cost-down" strategy a sharks. I've been shock to find out MT+PE rates in certain Central American countries where translation is done to serve the US Hispanic market. Pretty much the situation that Kevin describes. But let's face it, there were sharks in the translation industry before the advent of machine translation. MT has just sharpened their teeth.

    There are many other uses where MT is useful and does not threat translators' jobs: for gisting, for real-life understanding, for information and online chat. At least, as an ex-translator myself, I do not see it as a risk.

    For me, the danger always lays with the people who think they hold the absolute truth.


Annoyed They're Asking for a Free Sample? You Should Be Happy Instead! And let Me Tell You Why.

Before we move on, let's keep in mind a sample is not the same as a test . There is a difference between being sampled and being more o...