Learning through crowdsourcing is deaf to the language challenge

Opinion

In Germany, folk tradition tells of the Nuremberg funnel, a device used to pour language skills into the learner's head with little or no effort on the part of teacher or learner.

In The Hitchhiker's Guide to the Galaxy, the satirical science fiction novel by Douglas Adams, reluctant space traveller Arthur Dent was able to understand any alien language by inserting a Babel Fish into his ear.

Combine these two notions and you come close to the aspirations of a recently launched internet project called Duolingo. Branded by its creators as "the next chapter in human computation", the goal is to translate vast amounts of web content into a variety of languages while at the same time providing free language lessons for millions of users. The logic runs like this: Over 1.2 billion people are currently learning a foreign language whereas large sections of the web exist only in a single language, usually English. So why not combine these two activities and create an online language course in which the learners translate web texts as part of the learning experience?

Let's say you want to learn Spanish. The program gives you a sentence from a website and asks you to translate it. If you can't, the system helps you to understand difficult items by providing "educational examples" or translations of individual words. You then submit your translation and receive further feedback from your peer group.

According to the inventors, a pool of just 100,000 people could translate the whole of Wikipedia into Spanish in just five weeks and the ultimate goal is to attract over 100 million users.

If this all sounds too good to be true, that's because it is.

The developers believe they have addressed the crucial question of how learners can produce publication-ready translations when working into a language of which they have little or no knowledge. Inexperienced translators can see how other learners have rendered a given text and evaluate each other's work, thus creating an optimised translation through sheer force of numbers.

This is a technique known as "crowdsourcing", a term recently defined by web strategist Henk van Ess as "channelling the experts' desire to solve a problem and then freely sharing the answer with everyone".

Actually, this is not a new idea. In pre-internet days the Oxford English Dictionary made an open call for volunteers to index all words in the English language and received some six million submissions over a 70-year period. Internet crowdsourcing also has a role to play in language teaching, since it kicks in every time a teacher or student shares a problem with other learners or educators via blogs or social networking sites.

In the translation sector, such projects already exist, too. Over the 18 months or so since its international launch, Linguee, a bilingual dictionary and translation search engine has amassed over 100m sentences in various language pairs. Unfortunately, sheer quantity does not guarantee quality and although Linguee can serve as a point of reference, its user-rated entries include a vast number of dubious items as inexperienced translators unwittingly duplicate each other's mistakes. To coin another term, it has been crowdcrunched. And I suspect Duolingo will suffer the same fate.

The weakness in the Duolingo model lies not in the desire to harness the crowd but in the composition of the crowd itself. I am a long-standing member of an online translation community called ProZ.com, a moderated forum and database run by professionals for professionals. It is a community of translators who are prepared to pool some of their expertise, collaborate on projects and provide mutual assistance beyond what the web can offer. Putting a complex task such as translating the often highly specific content of an encyclopaedia into the hands of learners is, however, not the best idea.

But what about Duolingo's language teaching component? The software is currently undergoing "private beta testing" and apparently has a waiting list of 100,000 users, and although I have twice signed up (for Spanish and German lessons), no further information has so far been forthcoming. From the sparse details available, it is difficult to test the creators' claim that the free lessons are "as effective as professional e-learning packages such as Rosetta Stone with a price tag of $500".

Since Duolingo is a text-based system, it is a fair assumption that it will offer mainly vocabulary practice, and indeed the website's promotional video indicates that devices such as flashcards will feature strongly in the learning process. This is not a bad thing – in fact, there are plenty of free vocabulary trainers out there already, including those operated by crowdsourced multilingual dictionaries such as leo.org and dict.cc – but it is simply not enough.

Translation can play an important part in language learning, but a methodology constructed specifically around the translation by learners of decontextualised text snippets reveals a view of language learning which is simplistic in the extreme.

The developers appear blissfully unaware of the fundamental tenets of communicative language teaching, second language acquisition and the hazards of adopting fellow students rather than a teacher as the prime role model.

This project is the brain child of Guatemalan-born computer scientist Luis von Ahn, better known for his work in developing those internet security routines that ask you to identify a pair of distorted words to prove that you are a human being. To his credit, his previous projects have successfully harnessed the power of the crowd to solve discrete digitisation tasks.

Unfortunately for von Ahn, translation and language teaching are tasks of a much higher order than word recognition. This time he has bitten off more then he can crunch.

Maurice Claypole is a professional translator, a member of the Charted Institute of Linguists and author of Controversies in ELT