What Language Acquisition Theory(s) Is Used in Duolingo?
Hey Duolingo. I was interested in knowing which language acquisition theories / methodology you used (or inspired you) when creating this program. I have a background in language teaching and I think Duolingo could form a strong backbone of any language course and I always am trying encourage other teachers / students to use it. However, many language teachers are stuck in their ways and often have to be "sold" on a new language learning tool before they implement it in their own classrooms. I wanted to hear what language acquisition theory / point of view you guys use with Duolingo so I can help justify this to my teacher colleagues as a viable tool for language learning classrooms. Muchas gracias por su tiempo y dedicación. :-)
There are several theories behind our design of Duolingo, based on cognitive science and educational psychology research. We're also constantly experimenting with how we implement these theories in the software to improve on them. Some of the core ideas are (with links for further reading):
1. Active recall (the "testing effect"). Instead of flashcards and multiple-choice, most of our challenges use complete sentences that you have to generate from memory (mostly translations). This kind of active (not passive) recall tests your knowledge in a more realistic way, which has been shown more effective for memory retention. Of course, we use a few other challenge types for fun and variety too! :) [https://en.wikipedia.org/wiki/Active_recall, https://en.wikipedia.org/wiki/Testing_effect]
2. Spaced repetition (the "forgetting curve"). Over time, your ability to remember something decays. But the decay rate changes depending on how you practice. Duolingo is smart and schedules practice for words and concepts right when you are on the verge of forgetting them. Many other learning systems (Pimsleur, Anki) also use "spaced repetition" methods like this, but ours is (1) adaptive and personalized to how you use the app, based on the mathematical models we've built and tested, and (2) gets you to practice these words and concepts together in complete sentences (instead of just scheduling flashcards, like most apps). [https://en.wikipedia.org/wiki/Spaced_repetition, https://en.wikipedia.org/wiki/Forgetting_curve]
3. Core vocabulary (word frequency). We design our "skill trees" to teach more common words and concepts first, gradually introducing more rare and complex ideas. For example, after completing half of our French tree, we estimate that you can read/translate more than 90% of the French articles in our translation system! [https://en.wikipedia.org/wiki/Zipf%27s_law]
4. Multi-modal learning (or "representational systems"). People use visual, auditory, and tactile (bodily) information with exercising new concepts. Duolingo uses a combination of these senses in every lesson — listening, reading, interpreting pictures, typing, speaking, etc. — to help internalize the ideas both consciously and subconsciously, in multiple ways. We haven't figured out how to use smell and taste in a computer program yet, though. :) [https://en.wikipedia.org/wiki/Representational_systems_(NLP)]
I hope you find this useful! When I have some more time (and more detailed things to say about these topics), we hope to do a series of blog posts on the cool learning science research here at Duolingo...
For the fourth part , I think duo might provide more pictures linked with nouns.For example,the rudimentary lessons often have pictures for words like knife or girl beer ect... However the pictures are getting less and less along with the lesson.I dont know if this because of the lack of time for implementing or something.But when I level higher,(French)The pictures seem to be completely absent from the courses.And that discourages me a bit,you can't have a vivid input .
I know Luis had once created some picture labeling games ,you may use that to insert more pictures Thanks
I agree about the pictures. We have been adding more to the advanced skills... and plan to keep doing so. (Things are better now than they even were last week!) It gets tougher, though, with abstract nouns and nuanced verbs, which is the case deeper in the tree. :)
Points 1-3 are great, but on point 4 do you realise that the idea of individual learning styles is comprehensively discredited - see eg http://www.psychologicalscience.org/journals/pspi/PSPI_9_3.pdf . What there is however is evidence of different teaching approaches working with different types of content, and also some (though not a huge amount) about multi-sensory integration. It is probably these that you are tapping into, not mythical 'learning styles'.
Thanks for the pointer! To be clear, my point isn't to perpetuate a "learning styles" agenda, but to emphasize that we strive to integrate multiple modes/senses at the same time. I edited the post above to try and reflect this.
My background is in artificial intelligence and modeling behavioral/cognitive processes, and there is a ton of both theoretical and empirical evidence supporting multi-modal learning in computational learning systems (for example, https://en.wikipedia.org/wiki/Co-training and http://courses.csail.mit.edu/6.803/pdf/coen-short.pdf). Intuitively, the ideas can carry over to human learning, too. The idea is that we "index" bits of information by their "features." If we have a variety of features for each bit of information in different "modes" or "views" (sounds, images, words, how our muscles move as we say or write them, etc.) they help reinforce each other with less training (assuming they are conditionally independent, blablabla...).
I'm still getting up to speed on the related work that's been done in more straight-up education research, so any other pointers you have are truly welcome!!
Quoting from the summary of that paper:
Our review of the learning-styles literature led us to define a particular type of evidence that we see as a minimum precondition for validating the use of a learning-style assessment in an instructional setting. As described earlier, we have been unable to find any evidence that clearly meets this standard. Moreover, several studies that used the appropriate type of research design found results that contradict the most widely held version of the learning-styles hypothesis, namely, what we have referred to as the meshing hypothesis (Constantinidou Baker, 2002; Massa Mayer, 2006). The contrast between the enormous popularity of the learning-styles approach within education and the lack of credible evidence for its utility is, in our opinion, striking and disturbing. If classification of students’ learning styles has practical utility, it remains to be demonstrated.
Good point! I was irked by the same thing when I was reading the post. As you say, making use of different senses is a good thing, but perpetuating the learning styles myth in spite of all the counter-evidence isn't good for anyone. Most of all it hinders people who DO have learning difficulties from getting actually useful help instead of some mumbo-jumbo "specially suited" teaching. This is a fairytale built on a nugget of truth that's as harmful as or possibly worse than the similar and equally pseudo-scentific "right hemisphere person"-"left hemisphere person" classification.
The hardest part for me is to hear the spoken sentence so Duolingo is perfect for me. It forces me to go through my vocabulary to find possible candidates for words or phrases I can't identify, then I can use the "slow" playback to, hopefully, find the correct rendition. Then I can repeat at high speed as many times as I need until I can hear it naturally. I am always amazed at how my ability to hear a word is improved by knowing the word.
More pictures would be good, maybe even the ability to save and share mnemonics. I think there's a possibility to use input theory (Krashen) with this system, if you use the readings/translations also. It would be nice if the readings/translations were more integrated into the overall system, somehow. I think you're doing a great job of improving the system constantly though.