After months spent away from the language-learning app Duolingo, my level-five French skills were in decline. The "food" category was particularly threatened, coded red (for danger) with just one "strength bar" remaining. I clicked it, and was asked to translate: Je mange un repas. No problem. "I eat…" Wait, what was repas? My mind drifted to arepas, the Colombian snack. Defeated, I Google Translated. A meal! I should have intuited this from the English "repast". But, in the moment, I forgot.
Learning is forgetting; or, more accurately, it's virtually forgetting that we know something, but then being able to magically retrieve it when called upon. As Ulrich Boser, author of Learn Better, suggested to me, the human mind is not simply a computer; we will forget things, at a fairly predictable rate. So should I have simply drilled French food vocabulary every morning over my petits déjenuers? No, Boser says. The best thing "is to learn a word right when you're about to forget it". With each instance of effortful relearning, you remember longer.
Boser is referring to the "forgetting curve", pioneered by 19th-century psychologist Hermann Ebbinghaus, and it's precisely why Duolingo had coded my food vocab red: there was a good chance some of those words were going to slip into oblivion.
But as Burr Settles, chief researcher for the Pittsburgh-based company, told me, Duolingo was frustrated by how poorly its curves were working. "Users in the forums were saying, 'I don't feel this is capturing my understanding of the language,'" he said. Then it dawned: Duolingo could create its own curve. "We're tracking the data for every single user at every single point, and we know the statistics about how many times in the past they've gotten the word right or wrong." Rather than use an ad-hoc heuristic, it could create its own model
With its omniscient big-data eye cast upon millions of users toiling away through hours of language acquisition, Duolingo has a uniquely incisive view on learning as it is actually happening. It's learned a lot. It knows which countries tend to learn which languages (typically those of neighbouring countries, except for English, which is universal).
It knows which users are least likely to progress (English speakers learning Turkish and Irish, it turns out). But it is also learning about learning. In the beginning, "We had no idea how to teach a language," says founder Luis von Ahn (a Guatemalan inspired to launch the company by his struggles to learn English in his native country). Now, they do.
One way the company learns is by looking at people's most common mistakes. This often depends, says von Ahn, on which language people are coming from. Speakers of French, for example, seem to struggle with English contractions. "In English, contractions are optional," he says, "whereas in French they're not." Similarly, speakers of Chinese, which lacks articles like "a" and "the", don't need to face those concepts in the first English lessons. "It's demotivating," says Settles.
Motivation is a key, if sometimes under-appreciated, factor in learning. So, rather like video-game designers, who often use devices such as "pity timers" (which grant players some breathing room after they have suffered repeatedly) to keep users from quitting, Duolingo sometimes contravenes its own model of learning in order to boost motivation.
One behavioural enticement is a percentage figure showing fluency in a particular language: the number, drawn from a benchmarking system devised by the EU is not so much a precise diagnostic as a way to keep people engaged - and around. It's the carrot and the stick; or, as they say in German, "mit zuckerbrot und peitsche" - sweet bread and the whip.
But what Duolingo also showed, via its new algorithm, is that a learning system that responds better to users' own abilities is its own reward. The company's new model was better at predicting which words users would forget, which meant people weren't endlessly practising words they already knew. Users were taking on more lessons, but practising less. They were brushing up on language skills, but Duolingo was building its strength bars in a larger sense: learning how we learn.
Post A Comment:
0 comments: