Duolingo, a new free language-learning site, says it can help you learn a language for free while simultaneously using your learning exercises to translate the web.
A pretty big claim, but at the heart of Duolingo is an interesting idea – that people can use computers to do something while computers, in the background, achieve something else.
It’s much the same idea behind the founder of Duolingo, Luis von Ahn’s other venture – reCAPTCHA. This program uses the simple tests that prove you’re a human (and not an internet bot) to also help identify words scanned from old books.
Released to the general public in June 2012, Duolingo already boasts over 300,000 active learners and has raised $US15 million of venture capital. But despite a great deal of attention, all the information we have on how it works is from one condensed TED talk. And as yet, no formal independent review of the program has been published.
Earlier this year, I took the program for a test-drive (as a reviewer, not as a genuine learner), to check whether what Duolingo offers matches von Ahn’s promises on language learning and translating.
The new lingo
Von Ahn claims that with Duolingo people can learn a language about as well as with the leading language learning software through translation.
For centuries, translation was the core of formal instruction for language learning but then, in the sixties, the communicative approach replaced it. Duolingo is basically bringing the translation model of language learning back.
So does it work for language learning?
It does, surprisingly well. Translation gives the learner control over the process and a sense of achievement – translating words is often easier than dealing with lots of interaction, especially if you’re a beginner.
Duolingo also uses gamification to great effect and always gives immediate, automated feedback and points.
The program also allows learners to review and look at how other learners have translated text, which can help with understanding their own translations.
But the program also makes mistakes, and the more you advance, the more you notice them, which can be frustrating.
Is it as good as the leading language learning software as promised? It is good, but lacks many of the bells and whistles available in others: audiovisual context (BBC Languages), speech recognition and synthesis (Rosetta Stone), use of virtual words (Avatar English), native-to-learner interaction, not just peer-to-peer (Livemocha).
Lost in translation
The other half of the Duolingo project is translation.
For translating the web, machine translation is not good enough and relying only on professional translators is far too expensive. Duolingo offers a third way, with translation as a by-product of language learning – making it notionally almost as cheap as if done by machines and almost as good as if by professionals.
Once learners make some progress, they are asked to participate with real texts in real translation projects. It is by clients paying for these learner’s translations that Duolingo will make money.
It’s a clever idea but, does it work?
Below are two examples, one that featured in von Ahn’s Ted Talk and the second from my own own test-drive.
The first indeed shows a translation equivalent to a professional. But my example shows a translation equivalent to one done by a machine.
My first attempt (in Figure 2) also got disappointing feedback, but when I entered it into the Google Translate version, it got “94% agreement with correct solutions from others”.
So what’s happening here? My guess is that the system gets machine translation as the reference on which to base its automatic feedback. Then they hope learners, by voting, will end up getting the final version right.
That’s what they did in the German example above (in Figure 1) – which must have had the professional version entered as a learner’s input.
If I’m right, this makes a professional quality translation uncertain, unless there was at least one professional quality version for each sentence in the database – and that’s unlikely.
The final verdict
Duolingo will only work for texts for which accurate and elegant translation is not critical, and only if it’s capable of retaining massive numbers of advanced learners. It needs to capture the “whole” language learning market and be recognised as the undisputed leader – the Facebook or Google of language learning.
But there is fierce competition out there and expectations are very high. Major tweaking in its feedback algorithms is still needed.
Technology-wise, taking the source from the web, translating it in Duolingo and inserting the target back in the web seems less complicated than what reCAPTCHA does with words – but language is not about words, it’s about meaning. And ultimately, this is where Duolingo falls short.