Back in the January 31 issue of Time magazine, an article by Bryan Walsh caught my attention. Entitled “Pennies for Your Thoughts,” the article looks at a new trend for crowdsourcing on the Web.
As Walsh describes, Amazon is offering an online service call mTurk. The name draws upon the Mechanical Turk story from 18th century Europe, in which a chess-playing automaton tours Europe defeating luminaries in chess matches. Of course it was a hoax – a chess master was inside the machine and guiding its movements.
The idea for Amazon’s service is to take tasks that we would like computers to be able to handle on their own, but can’t quite do…yet. Via the mTurk service, individuals or organizations can hire cadres of individuals all over the world to carry out repetitive tasks that require human intelligences. So, now the tables have turned, it is human intelligence that is augmenting machine “intelligence”. The service is being used often for academic research, in which there may be a need to tackle high-volume tasks that can best be carried out by humans.
The production model is intriguing, since it combines distributed computing with “distributed human intelligence”. My concern is that there are individuals who believe document translation can be done reliably using this method. Admittedly, there may be some types of content that lend themselves to the randomness of the process. For example, using individuals fluent in two languages to confirm that a sentence in one language is equivalent in meaning to a sentence in another language. If the sentences being compared don’t relate to each other (e.g. they are not from a single document, but are used in isolation) then, possibly, mTurk might be useful. However, in the commercial translation space, I can’t think of one project we’ve completed in the last 17 years that would fit this scenario. I also must point out that comparing two sentences between two languages is a far cry from actually generating high quality translation!
Crowdsourcing of translation is gaining traction within the translation/localization industry. Large corporations are starting to use it to address customer support needs. But I, as a customer, would not want to translate content for myself or fellow customers. That’s the manufacturers responsibility…if they want me to buy their product.
I understand the impulse to look for ways to reduce the cost of translation by throwing software and headcount at the problem of exponentially exploding volumes of content filling corporate content management systems. This trend will only continue given the rise of many new technologies. The problem arises when highly technical, complex content is forced through an automated process – be it crowdsourcing or machine translation. The process itself changes how translators perceive the content, understand the context of the project as a whole, and undermines the cohesiveness of well-authored materials. Invariably the results of machine translated or crowdsourced translation will be less than what professional, specialized human translators can create.
The negative impact that highly automated methods have is how they skew cost and timeline expectations of buyers of translation. Many individuals assume already that all translation is done by computers. This makes the task of specialized technical translators that much more difficult, since many managers in both large and small enterprises don’t have direct experience with translation of content and often assume because of Google Translate and BabelFish that the service is cheap and fast – which it isn’t.
The ability of individuals to collaborate over the Internet in order to solve problems and tackle difficult projects is a cornerstone of the company I have created. But, like any tool, the Internet and computer technology have to be applied to problems in the most appropriate way. To paraphrase the old saying – ”if all you have is a hammer, everything is a nail” – if all you have is software and lots a heads, all content becomes cheap and less valuable.