Pages

Mar 8, 2014

The carnival is over. The MpT emperor still has no clothes.

The late Miguel Llorens once commented about David Grunwald, machine pseudo-translation (MpT) developer and advocate and owner of GTS Global Translations:
"... I disagree with Mr. Grunwald about most things. His ideas about translation as a commodity are depressing and I wouldn’t work for him unless something with a bit more dignity—such as “circus freak”—weren’t a viable career option (for whatever reason)."
However, Miguel also respected the man's efforts to expose the sleazy scam of a Canadian translation technology company called Ortsbo a few years ago. I also find many of Mr. Grunwald's views troubling, particularly statements that "one translator is easily replaceable with another" based upon a long string of unsupportable suppositions. His company's blog often contains interesting and useful insights into events, actors and issues of interest to translators but his obsession with machine pseudo-translation (MpT) and fanatical devotion to the ideology of commoditization over the years wore me out, and I prefer not to spend my energies contemplating the campaigns of one who seems to be on a personal mission as a mental battering ram directed against individuals who are professional language service providers.

So I was surprised when I found his recent guest post on the TAUS blog, which is too often a semi-coherent organ for the hucksters in the MpT carnival. It's more or less what I've expected to hear for a while and many of his points can be clearly picked out of arguments that Kirti Vashee and others make (and which are often overlooked or contradicted by their sources on other occasions). But Mr. Grunwald's piece strikes me as the clearest, most comprehensive and honest statement of current art that I've heard from the MpT camp so far.

I'm not quite the enemy of automation and machine pseudo-translation that some take me for. I am simply against lies, liars and (un)professional abuse in forms such as the human-assisted machine pseudo-translation (HAMPsTr) processes that so many piratical organizations and their enablers push. There are clear cases where automated translation processes offer value, but damned few of these have anything to do with my fields and level of work, and the attempts of SDL and other organizations to pretend otherwise are dishonest and/or deluded at best.

Read the TAUS post and think about it. You might wonder why an MpT advocate would make such unambiguous admissions. Well, unlike some in that camp, Mr. Grunwald never struck me as dishonest, merely as one who inhabited a stratum of the barrel where translators are perhaps indeed interchangeable. He clearly has a good mind, a sense of ethics which seems sound enough in most respects and perhaps a little taste for shaking things up. But as he points out, the money has gone elsewhere now.
"The VCs have rendered their decision: MT is out, human translation is in. In the last 2-3 year a number of venture capital companies have poured millions into companies that develop human translation automation platforms."
And
"Post-edited MT is not as good as from-scratch. Everyone has heard the ‘you get 2 out of 3’ saying. When you deliver post-edited translations, it will be cheap and fast, but will not be (as) good."
The whole MT carnival for years has reminded me of The Great Y2K Scam aka The Last Hurrah of Cobol Programmers. Grab the cash as fast as you can as long as the suckers leave it on the table. Things did not change much a few years ago in a technical sense when MpT became all the rage in the bottom tiers. What did change was the perception that there was money to be made, a fix of VC heroin to dream of language automation at least, and those in the pay of special interests began to brand skeptics like Mr. Llorens as "haters and naysayers" and worse.

Now the money has gone away; it's time to wake up and face reality - or the latest deceptions.

31 comments:

  1. The TAUS article is highly significant - thank you for bringing this to the attention of a wider audience, Kevin. I'm astonished that the translation world isn't abuzz with the news.

    ReplyDelete
  2. "I'm astonished that the translation world isn't abuzz with the news."

    There is a new trend: discredit those professionals who are worried about the many threats our profession is facing. It's easier to picture a bright Disneyland where Minnie and Mickey dance happily. We are living the worst crisis ever, and should be more informed than ever: to know what is going on, and to know who is who in our profession (as some emphasize on the need to avoid problems just because they are nearsighted, but others --danger!-- do so because they have other interests, totally opposite to ours.) Luckily Kevin, Steve, Valerij, Rose, and other bloggers are here to speak out only on behalf of language professionals.

    ReplyDelete
  3. Consider this sentence:

    "We use an entirely human translation process."

    Where do you suppose that sentence can be found?

    It's in the very first paragraph of website of David Grunwald's company: GTS Translation Services. All the other usual suspects are there on the home page of the website, too -- "high-quality," "professional," and, yes, "translators."

    So I'm not in the least surprised when somebody who makes all their money from an "entirely human translation process" -- and, incidentally, feels the need to point that out in the first paragraph -- comes clean on the brutal realities of MT.

    And it's not just about the money.

    I've written about this on my blog before ("The MT Debate is All Wrong" http://bit.ly/1ioodvZ). The MT market fundamentally misunderstands what LSPs are selling. They're not selling translations.

    They're selling TRUST. As long as there is doubt about the accuracy of a product the client cannot assess, and in the case of MT errors are randomly distributed throughout the client's OWN COMMUNICATIONS TO THEIR CUSTOMERS, the game stops being about translation and becomes all about trust and risk.

    When I owned ASET, I gave more client pitches to Fortune 500 corporations than Don Draper ever did in the entire decade of the 1960s. In both Don's and my case, clients made decisions based on who in that room they trusted. You can't prove quality in translation or in advertising.

    Customers buy from people they trust and like.

    MT intrinsically undermines trust. It poisons it. You are never going to win a battle between humans you can trust, and MT you cannot.

    End of story.

    ReplyDelete
    Replies
    1. I wonder if that sentence was there five years ago, Kevin. That's it's there now is no great wonder given Mr. Grunwald's current stated position. I'll admit to not having spent time on his company's main pages, because the attitude conveyed toward human translators as disposable commodities in the blog over the years did not encourage me to hang out for the party. I'm curious how he pulls off the "high quality" with interchangeable cogs in the wheel.

      You've probably heard the same story I have about a recent high-profile debacle with Lionbridge where every factor of form promised by technology was delivered flawlessly - perfect formatting, consistent terminology and style, fast turnaround - with unusable garbage for words, because they bought cheap translation "commodities" on the digital slave market. Good translation is anything but a commodity.

      You're certainly right about the toxicity of MT as it is often inflicted on processes and people. The videos of discussions by MT advocates in the HAMPsTr article make some of the issues very clear. The trust issues too often extend to the issues of compensation for the post-editors caught in the gears. I'm always amused by the voodoo people come up with to answer a basically simple question of how to pay for work.

      If the money is indeed drying up and attention is moving to other areas perhaps we can expect some more useful graduate research theses in the years ahead as well.

      Grunwald's TAUS piece is one step back toward sanity, a necessary one. Much of what TAUS has promoted in recently years is of course shamefully irresponsible with respect to machine translation and its realistic applications. And of course, the distortions will continue for a long time with those hawking MpT "solutions". One of the most amusing of these I've been hearing lately is the clam that a calculated match percentage for a machine translation is "just like" a translation memory match. Give me a break. In the first case - MpT - you have synthesized garbage patched together algorithmically without a true understanding of language or context, where the only real hope of making sense probably lies in the text being pulled from a TM anyway, while in the case of a properly created TM asset you have a text created according to human understanding of language (provided it was not simian-generated, of course). The MpT "match" may involve language patterns which can in fact adversely affect the linguists command of language without proper mental rehabilitation, while the TM assets should reflect normal language at least.

      There was another interesting link I "collected" today which in my mind relates to Grunwald's article. It talked about the "maturing" of YouTube and how recently the great quantity of amateur video has led to a stronger market for more polished work. If this is the case, will we see a similar shift in public attitudes with respect to translation some day? Interesting speculation perhaps.

      Delete
    2. The YouTube aspect is certainly interesting, since it reflects where mass media needs are going and how they are best fulfilled. I recently bought a (not-inexpensive) consumer article made by a well-known Dutch company.
      Fill out the guarantee form with dealer's stamp? Huh? Register online and get a free extra year's warranty!
      Read the manual? O.K. It contains no words; simply a weird list of drawings by someone who was
      probably a non-communicative cartoonist on speed. Right! Go online to look at the real manual in *.pdf form! Same shit as the *.pdf. ;-(
      Go to Youtube, and we have a cheery chap who says "Today, We are going to talk about how to...." Relief!!
      You already recognised this, Kevin, with your outstanding brief step-by-step training "memoQuckies" on YouTube at http://www.youtube.com/user/kslossner. So will we have: "...How to check the oil level of a Rolls Royce without RTFM" in 4 painless minutes"?
      Will be see product-based L10N working more towards subtitling? With all that information already stashed on the Net, are people getting too lazy to even read and just want to select a friendly voice from the menu to answer their question – no more and no goddamned less? Well: that friendly voice will HAVE TO BE talking sense, rather than post-edited codswallop that still baffles!
      Food for thought that I fancy some are already chewing.

      Delete
    3. @Chris: I hadn't considered the YouTube matter in quite the same way, because I was busy extrapolating it to other productions, but I do believe you're right, as you usually are in matters of marketing. I resisted anything to do with videos for years as some know, because I'm unimpressed by most of what I see (and because it wasn't trivial to rethink some approaches to presentation that might work on paper but fail utterly in a screencast).

      I think subtitling still has a future, even same-language subtitling. I've done a little and intend to do more. But when I think of the high-value B2B voice recording services that some are launching right now, it's clear that anything with even a whiff off MT has no place there, regardless what the marketing slides of your friends across the parking lot in Maidenhead might say.

      That's not to say that I am against people using MT to produce video language content. The combination of voice recognition and machine translation found on YouTube can provide hours of drunken amusement. And that's just the sort of production I hope to see from those who should be buried to fertilize the growth of better businesses.

      Delete
  4. While I agree with much of what Dave is saying, and agree that there has been too much hype and over-promising by many in the MT vendor community, I think you may be glossing over the point that in-spite of this, there is a "huge" amount of MT successfully in use in the world at large and a growing amount of MT use amongst informed practitioners even in the professional translation business. Just Google and Microsoft alone perhaps service ~500 million individual translation requests a day. We have already reached a point where 99%+ of the translation in the world is being done by computers and your dismissal of the technology is not noticed by those who use it.

    Dave also made the point that customization is difficult " Training MT engines is a game for big boys" and this is why many many "professional" efforts at using MT fail -- complicated tools being used by people who don't understand them.

    As far as investment interest is concerned I think you may be surprised to see how much Google and Microsoft invest in developing this technology and will continue to do so. hey have simply decided that they can do it better than anybody in the translation industry.

    Since we live in an industry where a company like SDL is considered a "technology leader" it is clear that what is called the translation industry is NOT a place that will attract any serious VC interest. When you contrast that Instagram and WhatsApp are applications that are considered worth billions, one perhaps gets a sense that ANY professional translation business pales in comparison since LIOX and SDL are usually worth half of their sales volumes or around $200M on any given day. From an investor perspective there is very little to get excited by in the professional translation business and the money raised by Smartling and others is really miniscule in the overall universe of VC investments (who have a terrible record of identifying what investments might appreciate anyway.) I think MT will continue to make steady if sometimes slow progress in the professional translation business, and we see that already most of the TAUS corporate members are translating a higher volume of words with MT than with humans mostly because their customers demand it.

    The basic language translation issue is still a very big deal to global commerce and communication and thus I expect MT technology will continue to attract people who will try to solve this problem. It is unlikely that they will ever replace humans but it is also quite likely IMO that the technology will get better, and it's use get more informed and appropriate, and MT will offer definite competitive advantage to those enterprises in the professional translation business who learn to use it with real skill.

    ReplyDelete
  5. Kirti, I think perhaps you are misunderstanding the nature of the "dismissal."

    Let's see if I can explain this by means of analogy while fully recognizing that you are one of the few people who is reasonable and intelligent about the proper and useful applications of MT.

    A million gazillion bazillion people use cell phones to take digital pictures every single day. Market valuations of Instagram, Facebook, Twitter lie in the $$$billion range; VCs sprinting after startups fighting over clever 20-year-olds with the next new great idea. Almost all of them -- well into the 90% range -- fail.

    While digital photography on cell phones -- which is "good enough" quality that has gotten a LOT better in recent years -- has so massively penetrated daily life that people think to photograph and upload images of their meals before consuming them, x 400 million x 12 dozen platforms, no corporation, media conglomerate, law firm, investment bank, pharmaceutical company or any of a hundred other major corporate entities anywhere are going to produce an ad campaign or photograph cover stories or produce annual reports or take any kind of risk on their image by using amateurs toting around "good enough" digital cell phones.

    This is not going to happen because people who work at these companies like their jobs and would like to remain employed.

    What professional photography and editing and creative "vision" bring, aside from far better production values, is that the customers' product benefits incomparably from the leveraging of human skills -- judgment, assessment, insight, creativity.

    So as risk rises for these companies, the need to have humans to trust in those creative transactions skyrockets.

    Keeping those jobs, you know.

    Where translators object so vehemently is when the happy cell phone amateurs start nattering on about how they are going to be shooting and producing the next global Samsung TV campaign using the cool pocket technology they just bought since a million bazillion gazillion people are using this wonderful technology right now. Isn't is just AWESOME!

    This is, of course, delusional.

    This does not mean that you can't incorporate cell phone digital photography into stunningly creative content -- it's done all the time. But where the real money lies right now -- why Dave feels the need to actually point out that his company uses an "ENTIRELY human translation process" ("Entirely!") is to build a protective moat and 50-feet brick walls and place armed guards around the idea that you damn well can trust us to use HUMANS ("Entirely!") on your project, not the MT that you may see I am selling out of the back door of my van on weekends.

    The reason the MT industry has the reputation of Amway in the translation industry is that they keep behaving like Amway -- instead of just admitting the limitations on the technology and focusing on proper applications, we are forever subjected to the next idiotic claim about how we all damn well better get on the bandwagon before it runs us all over.

    By people who are not even translators.

    At least the technology gurus in IT have the good taste to actually be coders.

    ReplyDelete
  6. But I think the point that many naysayers and yaysayers miss is that this technology actually has a very bright future as tool for humans to use to speed up repetitive/similar translations where the MT generates pseudo-translations that are actually helpful and welcomed by human editors.

    For high quality translation it will always need to be machine + human and this is just a new kind of TM capability not a replacement for human translation except for material that can be left in raw MT form and which would would just never get translated otherwise.

    Much of the noise today is from vendors with outrageous claims and instant solutions that promise to take you to nirvana tonight. Which I find as tiresome as any translator, or even more since I am often lumped together with reckless proponents of MT. I think this is still many years away from being a real do it yourself technology even though you may see occasional flukes.

    ReplyDelete
    Replies
    1. Kirti, "high quality" translations have nothing to do with the machines if we are talking about linguistic quality. Just like a rescue job on a crappy human translation will not produce a result as good as if the job had simply been done by a competent translator in the first place, you will not take MpT spew and polish it to equivalent linguistic quality at any reasonable time or cost. The output figures you've quoted a number of times on your blog for the HAMPsTr process to produce "usable" results with post-editing also do not usually measure up to what I have witnessed myself with good translators producing better quality with dictation systems.

      Even some of the "repetitive/similar" work out there can be handled quite nicely without MT by configuring pattern-based rules with regular expressions. I had that recently with an automotive parts catalog data set with a lot of part numbers, manufacturing dates and repetitive brand names. After about half an hour of rules configuration, I was translating slowly, and carefully using the memoQ auto-translation rules and carefully configured fragment assembly rules... at a rate of about 2500 words/hour. I would have gone faster, but I had to take some time to document source text errors which were revealed by the regex rules. No 6-figure investment in MT there, just my €400 license for a translator's copy of memoQ. (I actually suspect that well-tuned MT would have done better in this case and told the others on the test team as much. But for a one-off job it's not worth it.)

      Most of the so-called MT success cases that get cited are in very limited domains which have nothing to do with the application claims of SDL carnival barkers and other charlatans. To your credit, you usually make clear distinctions and point out the many areas where MpT technology really has little to offer (like almost every domain I work in). But then your Asia Online CEO Dion Wiggins steps in with his scare tactics and nonsense, proclaiming that translation agencies must "get on the MT boat or drown" (that infamous quote from memoQfest a few years ago). To call that daft and dishonest is being too generous. And it undermines a lot of the common sense that you try to add to discussions. And the GALA discussion in Berlin as well as Mr. Grunwald's post point out that the cost and engineering required for usable MT systems is beyond the means of most agencies and other SMEs. Will they "drown" as a result? I think not. The translation service providers working in marketing, finance and legal domains will probably notice nothing of the irrelevant, MpT nonsense as it regurgitates text for OK buttons in software interfaces and churns out flawless inventory lists for toilet assembly kits.

      Saying that MpT technology will get better is wishful thinking at best. Better minds than yours and mine have described very clearly why we've gone about as far as we are likely to with SMpT, and the corpse of rules-based MpT isn't likely to reanimate in any useful way. The very limited domains where MpT is useful for more than gisting are such a small part of the market and not one where good translators are usefully occupied anyway.

      Delete

    2. There was an interesting discussion on a private translators' forum a few months ago where one of my colleagues shared a fascinating experience with one of her direct customers which decided that they were better off using MpT to get the gist of gigaliters of that content tsumani that you and others are so fond of talking about. I think she actually lost some business for a while. But soon the client found that it was just too mentally exhausting to skim all the MpT crap, which was apparently accurate enough to understand to the extent required. The came back to her and hired her instead to skim the original content an summarize it. And from the way she described this work, it paid very well (better than the previous translations) and was a nice alternative activity. An interesting case and not one I've heard repeated yet, but I wonder if there aren't others like this and more to come. If corporate translation consumers find that all the straight MpT spew and HAMPsTrized content dulls their mental edges with prolonged contact as these people did, maybe they'll simply think of better ways to surf that tsunami that to swallow gallons of salt water - like using human resources in more intelligent ways to extract essential knowledge.

      Delete
  7. Where the MT is mining high-quality human-translated content and producing pseudo-translations that translators can leverage, then sure. MT as a tool is quite similar to TM technology in that respect.

    But it also shares some of the same (serious) downsides. It encourages a lot of grudging acceptance of "ok" translations in the name of speed and efficiency, and discourages the kind of creative thinking required to actually make the translations better by reworking them from scratch.

    You see a lot of PEMT guidelines in presentations that scream "Don't edit for style!" and "Correct blatant errors only!" that would never, ever be accepted as best practices in human translation. When the bar falls down that far, it hurts both the immediate output and the next generation leveraging of the "accepted" content.

    You also might see a whole lot less naysaying if the translators who were goaded or dragged or begged into mopping up MT's incontinence were asked to do so in the name of producing a better product and not (perpetually) in the name of reducing costs and driving up efficiency to the detriment of everything else.

    In other words, pay them better -- a whole lot better. There is no way to work at your best on what feels like a machine-controlled production line when the whole reason for the machine being there in the first place is to drive down your cost and value in the equation as much as possible. That ends up feeling a lot like a "master-slave" automation process, even though the translator is told that nominally "you are in charge" (as long as the efficiencies hold up).

    It would make a whole lot more sense to treat and pay translators as the masters of that process rather than the slaves to it.

    And just for the record -- there are still whole swaths of the translation industry valued in the multi-billion-dollar range -- the "premium" market -- where MT is clearly generations away from even getting anywhere near the front door. Let's keep repeating that so it doesn't get swept under the rug in what often feels like conspiratorial silence.

    Quick and fast IT user interface strings for a mobile app or a real-time help forum do not come anywhere close to constituting a major percentage of the translation market, or to what humans do and get paid handsomely for every day.

    ReplyDelete
  8. Hi Kevin,

    First of all I am flattered that my guest post on TAUS would spin-off a post of your own. Thanks for your kind words. I would like to say first of all that nobody that I know in the MT world is dishonest. I find that a poor word choice. Disillusioned? Maybe. Ignorant? Perhaps. But not dishonest.

    As to myself, I still maintain that translators are interchangeable. Especially when you have a good TM and glossary. And in my business where the customers make insane delivery demands, how can we function unless we have a range of translators that can be selected as required? I'll prove my point by bringing up Miguel Llorens, a man that I respected. Who is doing his work now? Certainly there is no Internet where he is now. But the work continues.

    I make my living by working with professional translators. They are our second most valuable asset (customers are the first as they are the ones funding the enterprise). And we work with some brilliant translators. I still say that translators and MT should not be enemies-they should be friends.

    ReplyDelete
    Replies
    1. David, I think you fail to understand the meaning of the word "interchangeable". And I would have thought that some of your own client experiences would have taught you that one cannot simply plug in any monkey to a process.

      The open, dishonest representations of MpT by SDL and others is quite apparent and deliberate. The FUD tactics of quite a few others are also perfectly clear. There are quite a few abusive opportunists in the MpT world, and those involved at the highest levels of the translation sector often speak openly and name names over a beer. The charlatans and carnival barkers do a lot to discredit those honest souls who dedicate their loves and loves to better automation of software menu, catalog and parts list translation.

      Delete
    2. An interesting discussion that underscores how very segmented the market is -- all those apples and oranges (and kiwi fruit and horned melons and...). Tthose working in bulk appear to be unaware of existence the high end of the market. Of how big it is, how lucrative it can be and, oh, how *demanding* it is.
      Sure, there's no disputing the need for and demand for fast & cheap & "good enough" in some areas, but only where the cost of failure is low. Which would include cases where buyers and sellers can agree that it is "better than nothing," for example. Or super-fast and super-cheap in triage mode. Fair enough.
      But a statement like "I still maintain that translators are interchangeable. Especially when you have a good TM and glossary" disqualifies David's operation from the type of work that Kevin L and Kevin H are talking about. With all due respect, demanding clients (many of whom have been burned by bulk providers) would laugh him out of the room.
      So surely the takeaway is that each provider profile should stick with what they know and can do; should not oversell (however tempting that might be); and should avoid generalizations about "what clients require". Since they are clearly talking about what their own client profile requires.

      Delete
    3. Well, Chris, I never assumed David was in the same market. For all his enthusiasm as a salesman, he simply lacks the background and education to understand the issues the markets the you and Kevin H and most others I deal with routinely are involved with. To his credit, at least he is aware that glossaries serve a useful function. Most of those businesses I encounter in the bulk markets rely on translation memories alone, which contain more dubious contributions from a questionable community than the bed linen in a cut-rate Bankok brothel after a long, steamy summer. I admire him as a man of faith; I lost mine long ago after seeing how even well-maintained translation memories, terminology lists and style guides are often not enough to guide even experienced translators to an acceptable result if they do not possess the requisite subject knowledge and linguistic judgment. But when one lives in the kinder, gentler world of revolving door service like OneHourTranslation, there are few such unreasonable expectations of understanding and surely any warm body at a keyboard will do :-)

      Delete
  9. See if this feels right: "I still maintain humans who claim surgical skills are interchangeable. Especially when you have a sharp scalpel and good lighting."

    If you are the patient in this scenario -- you know, the client -- jump off that table and RUN.

    I really miss the days I had competitors who thought this way. It's how we grew my company 20% a year in revenue, every year, by taking market share away from people who thought like this.

    And when the Fortune 500 company that that looked at over 300 privately-owned translation companies in search of a single one to buy in 2008, they chose to buy ours.

    Oh, who am I kidding, they bought us because we were making a TON of money, year after year. That's because we were owned by translators, run by translators and we valued translators above all so we went out and recruited the very best ones we could find anywhere, treated them like surgeon-rock stars and paid them accordingly.

    The very last thing I would ever have done is insult them by calling them "interchangeable" and putting their importance in the queue behind a TM and a glossary! When you put the cart before the horse like that, you are going to be fighting for fractions of a penny in the bulk market with all the other translation companies trying frantically to keep their heads above water and trying desperately to figure out why they can't make a ton of money in this business.

    When the industry loses a towering talent like Miguel Llorens, his poor clients do what they do in the NFL and they go out and scour the world for a replacement. They don't just grab a fan out of the stands since everybody is "interchangeable." My guess is that Miguel Llorens work is now being done by somebody (badly) until his poor clients, who no doubt feel this loss immensely, can find somebody in his same class of talent.

    That's what we used to do for a living, and man did we have a ton of fun doing it.

    ReplyDelete
  10. Hey Kevin, like yourself I am also a multimillionaire. I don't work in the translation business for the money, it is just for fun. PS I never heard of you or your big bucks company before but hey, that doesn't mean anything.

    ReplyDelete
    Replies
    1. Seems easy enough to track ASET down, David. Press releases and other public records all over the place. Looks like the have a lot to do with some of the areas the CSA always seems to forget about.

      Delete
  11. Kevin,

    After a hiatus of several months in my blog-writing efforts, the piece I did for TAUS (and your post) has wet my appetite. I wrote a new post on my blog which addresses some of the points raised by yourself and by the other Kevin. You can read it here:
    http://www.gts-translation.com/gtsblog/2014/03/12/when-is-a-chevy-good-enough/

    Best, David

    ReplyDelete
  12. Well, I read David's blog post. It was greatly entertaining.

    I guess it doesn't surprise me much that he admits his company produces "good enough" Chevys.

    It's a bit odd that the "good enough" that he admits in that blog post doesn't appear anywhere on the homepage of website he shows to his customers, though. There, the world "quality" is splashed all over the place.

    The central thesis of the blog post reminds me of how astrology (and later, religion) got started in a world where people who thought their tiny little mote of dust in space was the center of the universe, and then extrapolated their almost painfully limited view to argue that the whole universe was just like them.

    This is the central fallacy that's driving the false conclusion that translators are "interchangeable."

    I've posted a response on David's blog. If David doesn't approve it or materially revises it, Kevin you can, if you so choose, publish my response here.

    ReplyDelete
    Replies
    1. Kevin, you know that sentence you pointed out from David's corporate web site - "We use an entirely human translation process."? Well, it seems to have been left off of the machine-translated version of the site in French. As I understand it, language quality isn't quite as important in France, where they prefer their Chevys without hubcaps. Or axles. Or a working engine.

      Delete
  13. [Here's the text -- if he continues to censor it, please feel free to publish at your will]

    What's so amusing about this post is that it screams "My tiny view of my tiny corner of the market constitutes the reality of the entire market."

    It's also mildly entertaining to see somebody whose claim to expertise is simply the number of years spent occupying the same tiny little island, claiming that the duration on that tiny island gives him license to extrapolate into what reality is in areas all over a world he's actually never even seen.

    He's obviously never seen cases where a translator's unique choice of how to render a phrase saved a life, or whose special knowledge of culture prevented a political disaster, or whose skills are so critical to national security -- and have in fact played a crucial role in history -- that the translators don't ever talk about the content of those jobs, ever (my job on the Presidential Hotline when I was 23 is just one of those -- there are thousands.)

    If we take this example to its extreme -- another way of framing the argument in this blog post is that "the graveyards are full of irreplaceable people" -- it means all human expertise and brilliance and gifts are just simply meaningless. Abraham Lincoln was replaceable by Andrew Johnson. They are interchangeable because one died and another took over.

    If this is your view of the world and of life, I think you might want to seriously reconsider it.

    But leaving aside the logical fallacies, what's important to point out here is that the translation market has huge sectors where the entire per-word argument on which this post is based is completely false.

    Yesterday I had a very enjoyable lunch with colleagues about how robust and lucrative and intellectually stimulating their work is right now in the premium segment of the market -- where you are paid spectacularly well, you engage CEOs of multinationals directly and you are treated as an absolute peer by the best minds in their industry.

    Dollars per word? Think higher. Profit margins in the 30% range? Think higher.

    The premium sector of the industry is expanding and can't even find talented translators fast enough.

    It's still unclear where the top of the income limit is in the premium market.

    There is also a multibillion-dollar market in national security and the classified markets where translators are paid hourly rates higher than most working attorneys and many physicians. Their pay is bid UPWARDS by companies competing FOR their talent. They are competing for translators and interpreters who they deem quite irreplaceable.

    There are conference interpreters in Europe and Asia who negotiate with nobody because they are also fought over for their expertise. They make about $150,00 per year on average.

    These are just a few examples I can think of off the top of my head -- the market is, of course, incomparably more rich and varied than I've even described here.

    So this completes your tour of a bit of the rest of the translation world. We now return you to your regular programming featuring arguing over fractional per-word rates, driving around in Chevys, and being quite certain that your tiny little island is all the reality there is.

    ReplyDelete
    Replies
    1. I think you mean $150,000 Kevin. Not much at current exchange rates, but since I know translators who manage that in euros, I'm sure the good conference interpreters aren't doing too badly either. There are all sorts of interesting opportunities with a wide range of rewards - financial or otherwise - for those who aren't inclined to accept the role of interchangeable hamster on a rusty commoditized wheel and have put in their 10,000 hours or so :-)

      Delete
  14. @Kevin H ...

    I agree that the translator market is much more varied than many in the LSP business seem to understand but I would be wary of promoting some of the "well paid" US Govt market opportunities.

    I was quite involved with the US Govt market for awhile (with Language Weaver) and was quite aware of the dearth of translators/ interpreters in languages like Arabic, Farsi, Dari and Pashto.

    I recall many Afghani cab drivers complaining about "intelligence community" recruiters hounding them to come and work for the euphemistically named "reconstruction effort" (many told me they would rather starve than do that kind of work and be ostracized by friends and family, not to mention earning the wrath of "enemy combatants".) In fact many middle eastern restaurants in the DC area had notices offering $125K starting salaries to interpreters for "being part of the Iraqi reconstruction effort". What they forgot to tell you was that interpreters were the favorite target of hostile snipers and there was a pretty high probability that you got yourself shot and wounded at least, if not killed. In fact it was a common practice for interpreters to wear hoods when walking around with US troops so that they could later go and see their friends (sans the hood) and not get recognized, captured and tortured. In spite of these precautions there was a very high death rate for these guys which you can see described here http://www.nytimes.com/2013/04/15/world/asia/american-visa-delays-put-safety-out-of-afghan-interpreters-reach.html or this http://warnewsradio.org/read-more/iraqi-interpreters/

    Excerpt from HuffPost article:
    “We are abandoning Afghans who worked with U.S. troops to a certain death,” says Ronald Payne, a U.S. Army nurse who served in Afghanistan and founded the Allied Freedom Project, a Texas-based nonprofit that assists Afghans who are applying for an SIV."

    So yeah you might get some money for awhile but then you pay with your life.

    Much of the interpreting work also involved being the intermediary for US interrogators of "suspected hostiles", hardly appetizing work -- in fact I know many who would gladly choose post-editing over that kind of work where often inducements are offered (involving water perhaps) to get that information to help the reconstruction effort and preserve US national security.

    We also discovered (using MT) that sometimes interpreters were telling "suspected hostiles" (= often random brown guys in the wrong place at the wrong time) what to say and not to say when the transcripts were run through machine translation. Language Weaver pretty much got all their revenue from this list of languages and as far as I know it was the most anybody has ever spent on MT ($12 Million at the time of the SDL buyout) if you don't count the really stupid $45M that SDL paid for Language Weaver.

    So while I am sure there are very lucrative jobs for the elite few who are involved with high level diplomacy and finance/trade negotiations, as you rightfully point out, I would be wary of promoting the joys of working for "national security" to support wars that make little sense to anyone with common sense, as ultimately I think the money they were paid was very much and literally a double-edged sword without a handle. I am also willing to bet that many of the people involved as interpreters for these core languages listed above are far from the best, they were often desperate, as I have seen many sample transcripts of the work they have done.

    Translations involving humans can be very messy, especially when you bring war, politics and big money to the table and think that these elements have little or no impact on linguistic outcomes.

    ReplyDelete
  15. Kirti, yes, I know about the in-country interpreters and all the related safety issues. I was quite active with ATA PR back during the period we were trying to get the Iraqi interpreters out of the country and safely resettled in the U.S., a situation that was repeated with eerie similarity a few years later in Afghanistan. There were tragedies unfolding everywhere, and one could only look on a bit in horror.

    I was actually referring in my posts principally to the translators in those languages employed safely behind secure walls and inside secure facilities all over the world. The U.S. federal government spends well over $1 billion every year just on contracted language services, not even including those who are full-time employed. These contract linguists are not recruited from the cab driver cohort, but rather tend to be exceedingly well-educated intellectuals. Their salaries are bid up not only because of their relative scarcity, but also because of the value of their security clearance and their excellent work ethic. Part of their job is to remain silent about where they work and what they do, and where they come from, so they fly well under the radar, as it were.

    There is also a lot of activity on the military side that is non-combat-related. The US military is active in 143 countries around the world, and believe it or not, much of that activity is non-combat-related and involves building or sustaining infrastructure, providing aid, etc. Sure, there are political reasons for all these activities, but the amount of goodwill building by the military would shock even the most cynical among us (there are U.S. military planners deeply concerned that the services have turned increasingly into hybrid aid organizations and have evolved away from combat, leaving that role increasingly to special forces or drone fleets controlled from afar).

    A great deal of that language work in the field is also contracted out to private companies, largely because of the absence of solid language skills in our military services, a situation they have recognized and are trying to change. And of course private contracting (to major companies, most of whose names are unrecognizable in the traditional translation industry) gives far greater flexibility to the government in all areas -- intelligence, military, diplomacy, aid, and disaster relief -- to rapidly spin up major language capabilities in times of urgent crisis or need. It's incomparably more efficient than attempting to train and sustain people by trying to forecast your language needs a decade or more in advance. The demands change far too quickly for that, often nearly in real time.



    ReplyDelete
  16. Could I throw my 2 cents in here? Just to make the point, which echoes that of Chris about the variety of fruit on the stall, or as Jeanette Winterson's novel has it "Oranges are not the Only Fruit", that I have been translating full-time for five years in France, with a small but successful company I choose to call a translation studio, and with a range of great direct clients in the arts and culture sectors, and never once, in all that time, have I been asked to post-edit, read or even glance at a Machine Translation. Not once. Am I missing something?

    ReplyDelete
  17. Not at all Andrew, MpTPE (post-editing) is a very small niche. It is well under 5% of turnover amongst LSPs that respond to CSA questionnaires and that is probably a small fraction of the world's LSPs. It would amaze me if more than 1% of all words human translated over the next 24 hours across the globe were officially MpTPE.

    In his article Dave Grunwald stated “MT is as mature as it will ever be”. For the time being I agree with that assertion, but only on an algorithmic level. Phrase-Based SMpT seems to be hard to beat except for highly inflected languages like Finnish. I hear no rumors of game changers on the horizon from the academic world where most of the man-years are being invested in MT research. What remains to be seen is how the industry and tools adapt to the technology. As Kevin remarked, it is clearly still struggling on the payment front.

    Another thing that surprised me until I had a little think about it is that it seems that SMpT may have a positive impact on terminological consistency. I have yet to see that verified in published PE research but it has been said to me by quality managers in large LSPs. I’ll explain why I think that might make sense…

    For the kind of work that Chris D. does a company books a single translator to translate a single text (e.g. an annual report). That translator has a deep understanding of the domain (finance, chemical engineering and physics to take examples from some participants above). Here, terminological consistency is hardly a major concern. The wetware does that. However, for documentation for machine manuals or software where time to market must be fast due to a high rate of innovation it is not realistic to have a client wait around while a single translator translates 2.5k-ish words per day. In theory this is where terminology databases should help but few clients have the foresight to invest in good termbases and few LSPs have the margin to do it on their behalf. Let’s face it - most are perfunctory.

    In this case MT can be useful and this is true even if the source and target language are mixed in the target segment- as is the case with small training corpora. The terms you want to make sure are consistent are precisely those that are frequent so SMpT can work well there.

    My feeling is that if we can get young translators trained into using ASR (Dragon Dictate or better still something integrated into the CAT tool), a combination of MT for terminological consistency and ASR may work very well on this kind of team translation work. I would even go as far as to say to preserve style translators should be prevented from being able to post-edit MT, e.g. by presenting proposals as text in an image.

    Not many people are as privy to as much industrial scale PE productivity data as I am. Based on the numbers I have seen, with a few years of experience riding the Dragon, ASR beats PEMT on quality and words per hour. I worked with a guy last week that did 8k words per day of medical device software translation using Dragon on a large project. Ok, it took him 30 years of translation experience and 12 years with Dragon to get to that. While that is just north of turkey on Dion’s asinine scale http://www.asiaonline.net/images/WordsPerDay.png) it is pretty fucking rockstar in my world-view. I snail along at about 2.5k w/day - and I am not alone there I think.

    ReplyDelete
    Replies
    1. John, far less effort than would go into SMpT for the terms can yield suitable company-specific termbases, and the related QA procedures involving these to check the consistency of language in translated manuals, annual reports, etc. are also not that difficult. The main barrier here is that the people involved in the processes are not trained, and even given the right tools, many seem reluctant to learn how to use them. I've seen automated term extraction tools like those built by Lou Cremers do some pretty impressive things, and even a well-planned extraction session with Kilgray or SDL tools can identify and prepare a few hundred key terms in a handful of working hours.

      At the LSP level the main barrier I have seen here is the installation and use of server-based translation environment tools as wishware. The purchasers wish that an investment of €10,000 to €30,000 would solve all their technical process problems, but they are not willing to invest sufficiently in the brains and bodies to operate the processes. Thus, for example, I've seen cases where a year after investment and "training" not a single simple project has been managed successfully. Or where projects are initiated, this happens in the most primitive and unsophisticated way. In recent work with a client who runs a memoQ server, I sent a bilingual file to the project manager, and she had no idea what that was. To talk of six-figure investments with major engineering (*MpT) to solve problems already manageable with investments an order of magnitude lower is hilarious. By investing in the people involved, one could actually manage these processes at an even lower cost level. Our faith in software and hardware is great news for those who sell and support it, but sometimes a little faith in people coupled with higher expectations for basic learning (= better instruction, current practices largely suck) would get us much farther.

      Delete
  18. Kevin, I agree terminology extraction with manual pruning of candidate lists serves a similar purpose (though nothing beats manually adding terms during translation). I guess my point is that ASR and MT are not either/or technologies. It is easy for me to imagine use cases where they can be usefully combined.

    ReplyDelete
    Replies
    1. Although one always refines the terminology in the course of the work, I think an up-front term survey for translation projects of any size is definitely the trump card. It's indispensable for a well-run collaborative project, and it's a great way of avoiding many issues after delivery.

      I think speech recognition largely renders MpT pointless - it would simply slow the work down and confound the thinking of the translator, unless it involves no more than slight adaptive correction of very high fuzzy TM matches as DVX has done for ages. You're right that it's not an either/or proposition, because the only really sensible choice if you want good quality work is to get a competent specialist trained to dictate and probably keep MpT out of the equation. I've seen top-notch translators frequently put out over 10,000 words of finished, proofread work per day - non-trivial legal and scientific texts. Nothing in the HAMPsTr universe can touch that volume with that quality.

      My own workflow usually involves preliminary term surveys and selection of the particular main terms and transferral of them to a primary termbase which then feeds the tool tips which appear on highlighted source terms as I pass the cursor over them in memoQ while I dictate. It's a beautiful little "dynamic dictionary" that lets me stay focused on the text and speak the translation calmly and systematically. Badly constructed syntax and grammar from MpT is an unwanted, if occasionally humorous distraction in such a workflow. It might be enough for a dealer of used Chevrolets, but I'm afraid my clients would not approve of me fetching my parts from a scrapyard full of rust and rats :-)

      Delete

Notice to spammers: your locations are being traced and fed to the recreational target list for my new line of chemical weapon drones :-)