Sunday, April 04, 2010

Weaving Trust

Here is an excerpt from a recent interview with Mark Tappling, the CEO of Language Weaver, a venture-backed firm which develops language translation software:

"What we've discovered on the commercial business side, is that there is a massive need for translation of digital content on the Internet. But, because of the large volume of digital content out there, most of it simply is not being considered for translation, because there's just not enough money, people, or time. That said, we made an entry into the commercial market in the digital content space. In order for that to be credible, we found that there was a lack of trust--most people do not think translation software works well enough to be commercially viable. Our strategy was twofold--one, was to go out and get brand name customers, and to translate their content to such a high quality, with no human oversight, that they'd be willing to post that on the internet. We've been able to achieve that (...) The second thing we needed to do, was to engage our users trust. In order to do that, they needed to know when a translation was good or bad. So, we established a patented algorithm--a trust score--which automatically ranks a translation from 1 to 5, five being great, one being poor, so that a customer can determine the trust of that. We calibrated that to our customers own, native speakers, who scored them, so that everyone has trust in the score. That has taken the risk out of the conversation, because we only ask customers to pay when the translations are good."

Is that bit about "no human oversight" giving anyone a chill? If their MT is such high quality, how come Google is struggling? Something is telling me that their average score is around 3/5 and that they sort of provide a tiered system: if it is mediocre, we will get it reviewed by humans, unless of course you are happy with being mediocre in language X (which most of their clients would be, since any language other than English is an enigma and for a pragmatic English-speakers, enigmas are best left at mediocre levels). So close enough is good enough, and if their algorithm produced "3" is 30% better than the "5"s of other MTs, they aren't doing bad.

For all your English to Arabic and vice versa translations that will help you expand your business into the Middle East visit Arabic Language Experts at

No comments: