Geppetto Swarming

by Translation Guy on December 10, 2009
0 comments

I got an email the other day from my friend Jim Lehner, bilingual, very smart programmer, came out of Sun Microsystems, did the software for ZipCar, etc., a real genius who totally gets workflow. IBM’s announcement on n.Fluent got him thinking about swarm translation. Here’s his model for getting translation done cheap and fast…

You probably know that there are some websites where you can bid out jobs that you want done and pay bidders shockingly small amounts for their time. I have looked at these requests for programming and never been interested because of the miniscule remuneration. I have a feeling that they are appealing more to Indian programmers, let’s say. Anyway, this got me to thinking about a system that might just be interesting for you to consider. Imagine that you could farm out a translation job in replicate form to a number of translators, none of whom you knew. You give them a document and a deadline and a per-word translation fee. You get most or all of them returned by the deadline. Then, with some combination of automation and a human [bilingual] editor, you can take the best of the translations and make a final document that hopefully you feel is worthy of putting your name on for perhaps a fraction of the cost of a US-based translator. Using multiple translations helps to ensure that one bogus or low quality one doesn’t ruin the end product. You could even consider breaking up the project into paragraphs or pages in order to reduce the risk that any one translator is a bad apple or intimidating anyone with the overall size of the project.

The industry is a buzzing bee’s nest over these swarm translations. Facebook and others have taken the plunge. But in our shop the word is still out on swarm. Even among the project tools we use internally, we are not using these strategies much. I think this is because all our customers are very serious about translation.

Here are the problems we run into:

1. Good translation requires context. No word is an island. Context is key for the translator. The more info we can give them the better job they will do before they start. Glossary, style guide, translation memory, those are the golden troika of translation quality. The more info they have the closer they will get on the first pass. One paragraph or a line here must stand with its fellows to be truly understood.

2. Too many cooks spoil the soup. Whenver possible we want one translator per project. One voice, one glossary makes it just easier. Integrating several linguists is a headache. And translators like to get their teeth into a big project. More money, and it goes faster as their familiarity with subject increases.

3. The best translators charge the most money because they are the busiest. So by running a service that hires the cheapest translators you are skewed to receive the worst translation.

4. Translators can translate 300 words an hour, and editors can edit 1000 words per hour, proofreaders at about 2000 words per hour, unless something goes wrong. Then all bets are off. If a translation is less than 97% accurate, throw it back. Give the translator a chance to fix it, or start over. Because the cost to clean it up will exceed the cost of correcting it. The expense of translation is not from translation, although that’s expensive. It’s the quality that gets expensive. There are ways in which all can work more productively, but that’s another story.

5. Lots of translation project management tools make it easy to pool. You can ship a job to hundreds of pre-qualified translators. But our project managers don’t do it. We can’t maintain our ceiling-bumping level of client satisfaction if we don’t pick just the right person for the assignment.

So that means that our project manager has to understand what the client really needs, so they can put together the right team. And once they find the right translator for the job, someone who knows the subject, understands the client’s business, may even have discussed the project with our client, these are the people we want to use. Not the second person in the line, nor the translator offering the cheapest rate, or the one who responded the most quickly.

Not that we don’t need a better model than good ‘ol TEP (translate, edit and proof). Our clients want to bust the iron triangle of cost, quality and speed to market. The patient practice of our Geppetto-like wordsmiths, cobbling together sentences over their Trados workbenches is, sooo pre- 21st century, dude.

But the quality is in the hand-work. The quiet tap…tap…taping of those old mustaches over their their keyboards, bringing words to life. Breakthrough technology promises grow with Pinocchio’s nose. The iron triangle still applies. Will our artisanal cream float to the top of the translation quality pan, or be washed away in a swill of amateurish babel? That is not a rhetorical question. I really want to know.

0 Comments

  1. Ziggy M. says:

    You say tomato, we say too many cars on the road…zipcar is a great concept

  2. Soxxe says:

    You’ll have a lot of unintentionally miscommunication with n.fluent. It would actually make more sense, to teach everyone some man-made language…at least there would be less mistakes, even if I admit, you’d still be subject to the usual problems — i.e. IBM management doesn’t even know what they are saying in their own language…pretty much no way for a machine to correct that.

  3. I agree. But the emphasis in this post is not about translating WORDS, when word-for-word translations simply do not work. A proper translator must intimately understand both languages and convert the intent of what is said in one language to phrases in another language which express the same intent. I like the multiple ‘cheap’ translator approach – way to think outside the box Ken!

  4. Frizzlsnits says:

    You say tomato, we say too many cars on the road…zipcar is a great concept

  5. Zizelevak says:

    Nice concept but AutoShare is cheaper 😉

  6. Cyndrus says:

    It’s not that easy. Good luck, there is room for improvement here, but you can’t just hope throwing a crowd at it, is enough. It’s not enough, not to make it ‘smart.’

    • Ken says:

      Next step… educate the crowd. FaceBook is now giving away badges to their translators. I wonder how our translators would feel about that….

  7. ISmokeDope says:

    n.fluent = nice try…translations programs do not simply tackle mere words, they, and for many years now, tackle colloquialisms, phrases, and sentence structures.

  8. Rodney S says:

    While I applaud IBM’s attemps I have to agree that with best will in the world it’ll never be better than professional human translation

LiveZilla Live Chat Software