First, some background: Many moons ago, I ran across a
tweet by a translation company called Smartling proudly announcing the launch
of its website in Spanish. I visited said Spanish website and was mildly surprised to discover that it was actually in English (“English” is technical jargon we translation
geeks use for “not in Spanish”). This little vignette came to mind this past week
when I detailed my hilarious encounter with a Pinterest employee who insisted
that a mangled blog post in Spanish announcing its crowdsourced translation
effort was written by “professional translators.” When given evidence to the
contrary, she finally confessed that she had used her mother to translate the
text. This rollicking anecdote closed with a two-sentence comparison between
this incident and the aforementioned “Case of the Smartling Website that was in
Spanish Except that it was in English.” There the incident would have ended,
except that Smartling’s CEO—a Yeti with the whitest, most translucent mane I
have ever spied on an earthly creature a pale, blond chap by the name of
Jack Welde—decided to correct me in a jargony comment. His main arguments were
that: 1) his company does not do crowdsourced post-editing and 2) that
crowdsourcing gives him gigantic brain boners. To which I replied that 1) his
company most certainly does do crowdsourced post-editing and 2) crowdsourced
translation makes me a sad panda because of the abundant evidence that it
provides subpar results.
This prompted another onslaught by the now
very frenzied and irate woodland creature. I now reproduce it with snide interstitial
remarks written by yours truly (because it’s my blog and I do what I want):
Nope, incorrect again. I'll try to keep my response less "jargon-filled", so you can follow.
Douchy and passive-aggressive, but I’ll let
it slide. Go on, Johannes.
1) I've never said we do "post-editing". As a professional translator, you certainly know that the term "post-editing" generally means human editing over machine translation, which is not what we do. You've chosen "jargon" that you hope will be provocative with your readers, even if 100% incorrect.
The following quotes are taken from Smartling’s
website, Jack-O’-Lantern. You add caveats that machine translation is not as
good as human translation (to which I must parenthetically add: “DUH!”) but
then proceed to gush to your clients that “MT
is a great way to see the power of your new language site, and might jump-start your professional or crowdsourced translation
effort. You can choose specific parts of your site / app to be machine translated…
MT can be a valid choice for some organizations.” In the previous paragraph,
you state that “our platform integrates
with several popular MT services, so you can create a fully SEO compatible site
in minutes.” I don’t know, Jackie-Chan, but that sounds a lot like you’re
enabling crowdsourced post-editing to me.
2) We didn't fail to translate our own website.
Beg to differ, Jackeroo. A company that
does website translations and is not capable of translating its own website can
be accurately described as a “web-translation company that failed to translate
its own website.” I think this is irrefutable. However, I suspect you have
taken too many Tony Robbins courses and now you think you can play mind tricks
on inferior minds. Well, you have forgotten that to play Jedi mind tricks, YOU
HAVE TO BE A FLIPPING JEDI!
If you announce to the world that you have
published the Spanish-language version of your website and it turns out to be in
English, you have failed. The Big “F.” FUBAR. Fracasado. Finito. Something in
German that is bad and starts with “f”.
You were clever enough to snap a screenshot almost a year ago that showed some English on our Spanish home page. We had made some last minute changes to our English copy, and the Spanish translation was not yet complete. So we had a choice of 1) delaying the launch, 2) using poor quality MT, temporarily, or 3) leaving it in English for the short period of time before it was fully translated by the professional translators. We chose to launch, and I would make the same decision today. It wasn't a big deal, and the translation was completed quickly and professionally, and was deployed via or software immediately upon completion. Most importantly, this "incident" certainly has not hurt our growth as a company.
Beg to differ again. Your analysis of your
own brilliant decision making is deceptive, Jack-in-the-Box. My recollection of
the incident is slightly different: I followed a link to your homepage
announcing a Spanish-language version, I saw what a piece of crap it was, and
then I tweeted a snarky tweet about it, as is my wont. Since all you tech start-ups
spend more time monitoring Twitter than actually working on your core
competencies, one of your employees asked what the problem was and then immediately
fixed it on the run.
That is different from your version,
Jack-a-rino. You didn’t make a
conscious decision to publish a crappy website. You pushed out a crappy website translation because that is what your
company basically does. Because translation for you is an afterthought. It
is the excuse for vacuuming up all that yummy venture capitalist cash and
buying your little toys. Your company’s mission could just as easily be
copywriting or raising pet rocks or teaching math to Austrian midget horses. People
and companies like you work in reverse to inventors. Innovators see a problem
and engineer a solution. Edison saw darkness and dreamed a light bulb. You see
a fad for crowdsourcing and say: “How can I get the Jack-Dog some of that
action? Woof!” As long as the business plan has “social media” and “website”
and “crowdsourcing” somewhere on page one, you get a foot in the door. The core
mission of the matter is what you solve (or make up) after you have the funding in your bank account.
Of course, the “incident” as you describe
it (why use scare quotes?) is not the end of the world. However, allow me to
break down your decision flowchart as follows: a) I run a company that translates
websites; b) I launch my own company’s
website in another language; c) the version of my website in Spanish is actually
in English.
Faced with this daunting challenge, your options,
as you describe them, were as follows: “1)
delaying the launch, 2) using poor quality MT, temporarily, or 3) leaving it in
English for the short period of time before it was fully translated by the
professional translators.”
Seriously, what would Henry Ford do? Let’s
imagine that the prototype of the Model-T lacked wheels. Imagine Ford’s
decision tree looked like this: 1) delay the launch of the Model-T; 2) replace
it temporarily with a horse; or 3) leave it without wheels for a short period
of time hoping the customers won’t notice. And then imagine that Ford decided
to choose option 3 and tell potential customers that if the car had wheels,
they would be driving through the countryside. Finally, imagine that some
schmuck on the street walked by and said: “Ha! Old Man Ford’s mechanical
carriages don’t have wheels!” And then imagine that Ford berated the
slack-jawed yokel who had the gall to
point out something that obvious.
What would you do if you were Ford’s investor
and he described his options the way you just described yours. Would you: 1)
give him a Chuck Norris roundhouse kick to the nuts?; 2) knock his teeth out
with a baseball bat?; or 3) suspend all future injections of cash into a failed
business?
I am sure that 1 and 2 would be tempting,
but on the whole— given the polite customs of the early 20th century—the
investor would choose 3 and swiftly fly away.
Dude, you run a company that translates
websites and yet your own Spanish website isn’t in bleeping Spanish! And then
you claim that unprofessional crowds can do it just as well! By what possible
measurement? By your own? By the criterion of a company that translates
websites and isn’t capable of translating its
own website?
My melanin-deprived homey, do you really
fail to grasp the beautiful, tender irony of the entire anecdote? Do you really
want to engage in a flame war with a random blogger when the evidence of
incompetence would make most responsible businessmen run into a corner to cry
like a 12-year-old girl?
3) As I said in my prior comment, many of our customers use professional translators to perform translation -- translators like yourself (although you seem pretty angry, and not much fun to work with...)
Does that mean you’re not going to hire me, Jackie-O?
My dream was always to work for a fly-by-night tech start-up that probably
won’t be around six months from now… (Sob!)
Seriously, Jumping-Jack-Flash, I’m a
barrelful of laughs. And you, my Polar-bear-colored friend, are hilarious too.
If we could only hook up, we would create a rocking comedy duo: The Translator
Who Stared at Websites and The Crowdsourcing Snowman (did I mention
that Jack is disturbingly, almost supernaturally,
white? I swear to God that if I didn’t believe in goblins I would have trouble
sleeping after incurring the anger of this elfin woodland creature).
My chromatically challenged friend is, after
all, a garden variety sociopath. A sane man would have realized that the “Spanish website that
was actually in English” is just an embarrassing episode and would have let
sleeping dogs lie, suppressing the memory with alcohol. A sociopath, in
contrast, decides to engage in an angry polemic with the passerby who pointed
out that English and Spanish are, when all is said and done, not the same
language. But there is not even a hint of embarrassment in Jack’s discussion of
his company’s goof. The L10N Web 2.0 companies are so divorced from reality
that a CEO seizes upon overwhelming evidence of his own incompetence as an
opportunity to teach the world the beauties of crowdsourcing.
And then he nimbly shifts from defense and
boldly goes on the offense.
Are you saying that you are a better translator than every other professional translator? I guess the citizens of Web 2.0 only deserve the quality you personally can provide?
This is what is known as a non sequitur,
Hit-the-Road-Jack. Look it up. It is also a tried-and-true rhetorical trick
lifted straight from the playbook of a five-year-old child. When someone lands
a verbal zinger, you scrunch up your nose like a snot-head and go: “I know what
you are, but what am I?” It is a classic, though.
Your argument is tired, Miguel.
It is not an argument, Action Jack-Son. It
is a piece of empirical evidence. Empirical evidence is the basis for an
argument, but it is not an argument in itself. An argument is something akin to
“you are an albino cretin because of A, B and C.” The merits of the argument
would depend upon the way in which A, B and C prove the proposition that you
are, indeed, an albino cretin. Empirical evidence, on the contrary, can only be
refuted by denying that the evidence is real or that it actually happened,
which you have not done. You have just fabricated a counter-fairy tale, cast a
few aspersions, mumbled some conspiracy theories about the UN and Microsoft,
and called it a refutation of an argument.
You are the equivalent of the Microsoft software engineer who claimed that open-source software wouldn't work because only professional software developers working at Microsoft could produce high quality software.
And you, Jack-o’-nine-tails, are the
translation world’s equivalent of the Bush Administration. It’s like we’re
still living in 2007. Why is the “Mission Accomplished” sign so outrageous?
Because the mission wasn’t accomplished and thousands of people were still going to
die! It was, in fact, the opposite of
accomplished. It was like… not
accomplished! Just like your Spanish homepage wasn’t in Spanish, but in
English, which is a whole other language from Spanish.
Why is the “heckuva a job, Brownie” so
outrageous? Because Brownie wasn’t doing a heck of a job and
thousands of people were going to suffer!
The fact is there is plenty of work for professional translators, especially the good ones. And Smartling is delighted to work with some of the best translators in the business;
Really? Because the quality of your
website’s translation indicates otherwise (but that will be the topic of another
blog post I’m writing). Your website is a literal translation that does not
sound very much like Spanish, but rather like a bad transcription of corporate
jargon dictated through a bad cell phone connection.
we respect their craft and the high quality work they do.
Yes, every one of your comments drips
respect. This takes us to the next paragraph, your Nessun Dorma of dill-holiness:
PS: Since you love to point out errors in other people's work, your headline on this blog is inaccurate. From your own narrative above, it sounds like only one employee's mother may have been asked to assist with translation. And yet your headline says "Pinterest Uses Employees' Moms" -- in English, the use of the word "Moms", as well as the the apostrophe after the "s", means that more than one employee's mother was used for translation. But that seems to be inaccurate, from your own story. Were you just trying to be provocative with your headline? Or do you lack the basic understanding of plurals in English (which would make me question your ability as a professional translator)? Should I take a screen shot?
Tsk, tsk, tsk, Hugh Jack-Man. (If you had
edited out this paragraph, you would have saved yourself this public response.
I even gave you a chance to rewrite the comment, remember? But you insisted. So
here you go.) Sticks and stones, my man. Sticks and stones… Passive
aggressiveness is not an attractive trait, especially in a man.
This really is the non plus ultra of
entitlement. Faced with undeniable evidence of your own incompetence, your
decide to go on the attack and question another professional’s competence. But
no defense is better than a good offense.
Miguel, anytime you want to have a real, honest, non-sensational discussion about the merits of professional translation vs. crowd translation (and even MT in limited cases) -- and the best ways to manage the translation process -- I'd be happy to have that discussion. In the meantime, try to be cool.
If this is a morsel of this serious dialog,
you can store it, Jack-meister (OK, I admit it, I ran out of “Jacks”). I can
get more stimulating debate from the homeless dude panhandling on my corner who
constantly warns me that the Queen of England has bad “joo joo.”
Miguel Llorens is a freelance financial translator based in Madrid who works from Spanish into English. He is specialized in equity research, economics, accounting, and investment strategy. He has worked as a translator for Goldman Sachs, the US Government's Open Source Center, and H.B.O. International, as well as many small-and-medium-sized brokerages and asset management companies operating in Spain. To contact him, visit his website and write to the address listed there. Feel free to join his LinkedIn network or to follow him on Twitter.
Miguel Llorens is a freelance financial translator based in Madrid who works from Spanish into English. He is specialized in equity research, economics, accounting, and investment strategy. He has worked as a translator for Goldman Sachs, the US Government's Open Source Center, and H.B.O. International, as well as many small-and-medium-sized brokerages and asset management companies operating in Spain. To contact him, visit his website and write to the address listed there. Feel free to join his LinkedIn network or to follow him on Twitter.
26 comments:
Hahaaaa! I could tell this was coming.
"El pez por la boca muere" - there's a nice phrase to throw into the crowdslation blender and see what comes out.
Thanks for the entertaining read! I wonder if Mr. Welde failed to note the nearly-year-old dateline of your original post on this topic. Seems funny to engage in a flame war over something that's such old news.
My guess is he let it slide the first time, but the little urge to reply lay dormant and was reawakened last week when I made a mention of it.
Dear Blogger,
this is a shame that you can claim to be human or have a sane mind. Translators as yourself are making people want MT so badly. You suck!
Dear Anonymous,
What can I say about people who are so gutless they insult people behind a wall of anonymity? Are you really that scared of bloggers? Do you feel brave insulting people without showing your face? Is that indicative of humanity? I'm peeved at Welde, but at least he has the courage to show his face.
Miguel:
MT and crowdsoourcing is the gold rush future of the translation industry and only luddites such as yourself would try to resist what clearly is unavoidable, as well as conducive to high output and high translation quality.
YOU SIMPLY HAVE TO LEARN HOW TO INTEGRATE MT AND CROWDSOURCING INTO YOUR OWN TRANSLATION PROCESS TOO!
ALL RESISTANCE IS FUTILE!
See e-mail sent to me today by the American Translators Association.
Presenter: Ruben de la Fuente
Date: June 20, 2012
Time: 12 Noon US Eastern Time
Duration: 60 minutes
CE Point(s): 1
Translators frequently overlook the potential for machine translation to
improve their productivity, increase revenue, and maintain quality. It
can be done, but the trick is in knowing how to do it.
This webinar will look at ways to successfully integrate machine
translation into a translator's current workflow. The presenter will also
discuss best practices in using machine translation for efficient post
editing.
Attendees will learn:
* how to select a machine translation tool
* how to leverage existing linguistic assets for machine translation
* how to post edit efficiently
* how to integrate machine translation with CAT tools
ATA Member $35 | Non-Member $50
Previous ATA webinars have sold out in advance. Register now while
space is still available!
Click to REGISTER:
http://www.atanet.org/webinars/ataWebinar111_machine.php
I would love to see one of these seminars stating that some technologies are not useful for all translation scenarios. But unfortunately propaganda doesn't work that way. The thing that is surprising is that some translators are not aware of niches in which some tech tools end up not being that useful.
Hi Miguel
Actually in webinar above I explain some of the features that make MT suitable for some texts but not others. As a rule of thumb, if using TMs make sense, so will MT.
I don't think any technology is useful in all possible use cases, but when it comes to MT many translators don't stop to think if and why it could be useful to them, but rather dismiss it without test driving.
Hi Miguel
Actually in webinar above I explain some of the features that make MT suitable for some texts but not others. As a rule of thumb, if using TMs make sense, so will MT.
I don't think any technology is useful in all possible use cases, but when it comes to MT many translators don't stop to think if and why it could be useful to them, but rather dismiss it without test driving.
Hi, Rubén,
Fair enough. I would like to highlight the fact that I have test driven the technology and I can safely (and sincerely) say it is not useful for me. Secondly, I feel that this will be the case for people who want to get out from the more commoditized areas of the market. I realize that this second statement is more controversial, but still provable.
Hi Miguel
Out of curiousity, what did you test drive? Any tool that can be customized (Systran, ProMT, Moses) to specific style and terminology or just Google Translate?
While Google Translate can deliver great results in some cases, I don't think it is as representative of MT potential as customizable tools.
Google Translate and ProMT and Systran. My conclusion was that ProMT and Systran were quite simply garbage. Google Translate was clearly better. That is why when I hear the spokespeople of paid MT criticizing Google Translate, I make skeptical noises. This alleged superiority of non-free MT has not been p¡roven in my view. And Moses is simply not for the average jobbing translator.
For me, the only reason to use SMT is to take advantage of the breadth of Google's algorithms and corpora. And don't kid yourself: if there is any interest at all outside of the industry in machine translation, it is due to Google Translate. The myopia of MT "experts" in this regard is really quite remarkable. The other options don't give you that twin access to huge corpora and better algorithms. And if you use ProMT or Systan, my feeling is that it is just because you have an MT fetish, not because of any vaunted productivity boost from recent technological advances. Because, let's face it, the technology in those tools dates from the Cold War.
Did you do any customization work for Systran and ProMT, like importing your glossaries, for instance, or adjusting translation rules (formal or informal translation for "you", preferred translation for recurring structures like "by+ing"? Or did you use it "as is".
Google Translate performs quite well in generic use cases, but you cannot customize it for any specific client, neither improve it over time.
Using customizable desktop solutions is no fetish, it just gives you the flexibility to adjust MT output to specific requirements. About the Cold War bit... I don't think so.
I didn't get into the nitty gritty of customization. I only used the Systran plug-in a little with Trados and found it wanting. Also, I compared ProMT's output with Google Translate and I was shocked by how bad it was. When I went into the Spanish dictionaries used by ProMT, I was further shocked at how mediocre they were (even the "specialized" financial "dictionaries"--which are really glossaries). They were just unspeakable crap. It looked like something done by Russian hackers who didn't know any other languages. For me, it was garbage in, garbage out. I am not going to spend six months customizing a program that supposedly helps translation productivity but whose creators were so shoddy. When you factor in the fact that Systran and ProMT are Cold War technology, well...
If you will allow me to say so, the impetus behind MT at the moment is 95% due to Google Translate and advances in the application of statistical algorithms to large Web-based corpora. To tell a translator that "well, what you really should be looking into is rule-based machine translation" for me smacks of a shopper investigating automobiles in 1915 and being told that what he really needs is the top of the line in horse-drawn carriages.
What I would like to know is why there isn't a desktop SMT application. I'm pretty sure the answer would be quite interesting.
Anyone who has ever used a CAT tool (or TEnT, lest Jost the Great slaps my wrist) would find it hard to buy into this "customizable MT" line.
1) If a commercial MT tool isn't better than Google Translate in its pre-customized state, you can bet your [donkey] I won't be paying for it.
2) Whatever it may do better than Google Translate after I've invested my own money, time and effort in customizing it, my CAT tool is already doing.
Please don't take this personally Rubén, but to quote your own words: as a rule of thumb, if using TMs make sense..., what's the point in using MT?
Not to mention Miguel's point on algorythm and corpora. Trying to compete with Google Translate commercially is like trying to sell people the Super Awesome Customizable Search Engine 3000 that will beat Google Search into the ground... once you've fed it all you need to find.
It just doesn't make a lot of sense.
Miguel,
It doesn't take 6 months to customize an MT tool. You can import existing glossaries in minutes and play around with translation rules to match your requirements. Basic customization shouldn't take more than a few hours. And customization is key for MT to produce workable output.
You insist on the Cold War technology bit... RbMT has evolved a lot since then (to the point that some incorporate SMT to refine output further, I'm sure you heard the term "hybrid", very fashionable these days). And it's more user-friendly than SMT for the average translator.
I wouldn't say the impetus behind MT is due to Google. Maybe to SMT, but not Google. Companies investing in MT are looking into ProMT, Systran or Moses, but not Google.
If you are interested in desktop SMT, there's do-moses-yourself (sort of Moses for dummies). Only problem is that it runs in Linux, but you can get it to work without having a computer science degree.
Gueibor,
TMs generally work at a sentence level (with some improvements lately), they only retrieve whole sentences. MT works at phrase level (store recurring phrases in your dictionary and they will be inserted in relevant inflected/conjugated form)
And a final word on test driving: if you want to test properly you should edit the output and time yourself to see if the old saying "takes me less time to translate from scratch" is actually true.
Well, what you call a "bit" is a fact. RbMT has been around since the Cold War and, with all due respect, it is not exactly setting the world on fire. When economists talk abot the Great Stagnation, they mean exactly that: we are in a period of no major innovation but mediocre tweaks to technology that do't increase productivity or bring greater social wealth. I heartily recommend that you read Tyler Cowen. When I hear MT people talking about innovation, it reminds me of some of his quite funny empirical examples. In my experience, SDL's vaunted improvements to Trados are just that: minor tweaks disguised as a tech revolution.
While I have obviously failed miserably in getting you to concede that ProMT is a really ugly piece of software for a profession that allegedly deals with language, I would recommend that you not mistake the tweak-tree for the tech-forest. Also, my recommendation to MT evangelists is that when people look into MT, they want to look into SMT. When you move them along to see a model from the 1950s, they get disappointed. And that may be a reason why the whole MT thing isn't getting much traction outside of the commoditized pennies-per-word segment.
Oh, and another point that is tangentially related to what Gueibor said about TMs doing what a commercial MT program does. I know that isn't exactly accurate, but what about phrase-level matching that some TM programs offer now? Isn't that much simpler for the individual user to implement than to dive into an expensive and not-user-friendly program like ProMT?
Not sure what you mean by tweak-tree and tech-forest. I think one of the reasons why MT has moved forward is that developers have realized it's not possible to build a piece of software that works in all cases, and instead provide something the users can customize so that it works for their particular use case. RbMT might have been around since Cold War. That does not mean it has not improved since or cannot still be useful today.
SMT can look more attractive, but bear in mind that 1) for some language combinations RbMT works better than SMT, 2) users have no control over SMT. If SMT is using wrong terminology, the process to fix the engine is more cumbersome and slow than with RbMT (re-training the whole engine as opposed to modifying a dictionary entry).
Re. phrase-level matching from TM tools, it'd certainly be more convenient, but I'm not familiar enough with them to compare them with MT.
About ProMT's expensiveness and userfriendliness... Cost is roughly the same as a CAT tool (1000 USD; I think I paid 900 EUR for Trados Studio) and I don't find it particularly user-unfriendly, although I prefer Systran.
Tech revolution? Setting the world on fire? Maybe the problem is using such fancy words. All I know is I've been using MT for a while and it takes me less to edit an MT draft than to translate from scratch: no less, no more. And I'm happy with that.
And why did you give up on Google, out of curiousity?
I stopped using Google Translate because the productivity gain was ultimately negligible. I translated just as fast with and without MT because the real added value in my work is in providing a stylish and accurate version of the original. When that is you business, a lower quality first draft (which is what MT provides) isn't too useful.
Did you time yourself? Sometimes productivity increase perception can be deceitful.
Yes, I timed myself. There was barely any difference once way or another. But, mostly, it was noticeable that the quality of the writing was better without the MT output. As I have stated, MT is a crappy first draft and for higher-end "transcreation" a poor first draft is of very little use. However, if one targets the lower-end, lower-rate spectrum of the market, I can understand how MT might be appealing.
Wouldn't you say there's a whole range of shades between "transcreation" (is that what you do? thought you were specialized in finance) and lower end you mention?
And just for the record, I'm not suggesting translators should charge less because they use MT. On the contrary, they should keep their rates and make more profit.
There are infinite gradations between yellow and green, that doesn't mean the difference between the two doesn't exist. Most prophets of automation who talk about the industry consciously ignore any distinction between market strata because for them it is all just one greyish shade of pale.
Incidentally, might I also point out that while it is laudable for a translator to increase productivity while charging the same rates, a translator who expressed expertise in machine translation is probably signaling something quite different. And that signal is assimilated by the market.
A translator who expressed expertise in MT... why would s/he want to express such expertise? If I was a freelancer working with my own home-made MT, I'd make sure to keep that piece of information (and associated revenue) to myself. Can we agree at least on that one and close the thread? ;)
OK, I'm glad we can agree on something. Thread closed.
Post a Comment