Duolingo’s AI learns what you need to learn
spectrum.ieee.org
spectrum.ieee.org
IIRC Duolingo only claims to get to even B1 level for a couple of their most robust language courses. Not sure what you view as sufficient for real world but B1 is about where I'd personally draw the line.
All that said, it was actually a really great introductory program that made language learning more approchable. As an early intro and vocabulary supplement it is great. After that you will want to seek greater depths of learning and get immersed in content of the target language.
Defined as verbal communication only? It has got me far enough so far that I can read certain newspapers with good comprehension. (I am at Intermediate 2 on Duo, so I suspect this will improve further)
For what purpose? You either subscribe or you don’t, and once you subscribe you’d think they would want to show you real results so you continue to subscribe. If it was just shallow learning then they would lose subscribers over time once people wised up to this, but that doesn’t seem to be happening. It seems incentives are aligned to actually learn! How do you monetize what you are saying?
So I guess I agree that while Duolingo alone isn't enough to get good at a language I have been frequently amazed by how it "gets" where I'm at on my learning journey.
Especially recently I've started taking the occasional tests it provides and that seems to change my learning journey in a convincing way.
I disagree that this is (only) about keeping you using the app. When I studied piano, it was critical to work through pieces slowly and in small snippets that you could play. If that meant a quarter the speed marked, so be it (if slower than that, you'd bitten off more than you could chew). Otherwise, you weren't practicing the music, you were practicing your own mistakes. So a 95+ success is really about practicing reading and producing the language correctly as it is about playing games (the games may help ease the conscious boredom of the same thing over and over while the reflexes build).
Also, something like 'the owl' and 'the mouse' may seem trivial and logo-ed, but the language angle is 'le hibou' vs 'la souris' (masculine vs feminine gender) and these don't follow across different languages ('die Eule, die Maus' - both feminine) so repeating this along with exercises that flesh this out has value. I do think that the exercises don't often enough provide the opportunity to get it wrong, but that may be the 95% thing.
When I speak with an actual person through italki, I never talk about weather. I find out I need to learn how to recognize Haber, Poder, Ir, and Tener in preterite and future tenses at conversational speed. Otherwise I can't hold the context for whatever is going on (we tend to talk about what I did the last week, which I love. Throughout the week I lookup words and build a vocabulary in anticipation for our conversations)
I'd really like Duolingo to focus more on audio lessons, and answering questions about stories. I pay way more attention for those questions and really feel like I am getting better at hearing the language. However, I spend more time, and I do better at those questions. I fly through the other questions and tend to make typos which are bad enough where I get the question wrong. I don't know if this trains the AI to keep giving me easier things that I don't pay much attention to.
I have learned WAY more with Duolingo in 6 months (10 to 15 mins a day, but I started slower) than I learned in 4 years through High School (US). Just one data point, but I've heard similar from friends. TBH I am more motivated today than I was in school, so I'm sure that plays a part.
Some of the stories, maybe only at higher levels, prompt you to respond to or summarize the story in 20-40 words. Then it applies some basic grammar and spelling corrections. Nice to get some feedback, but I don’t have any real way to know how correct or comprehensive it is.
I feel the same (approx 205 days of French). While the multi-modal aspect(text&audio) was helpful especially at the beginning of the course, I feel like text now is a crutch at times. I wish some of the tasks were audio only and required either answering from among text responses, or single word speech responses.
Because of this crutch, I feel well equipped to approach written French in my day-to-day life, but freeze up at even simple questions like "Parles-tu français ?" I feel like I would freeze less if I could practice these types of exchanges in app, but part of me also knows that the best solution to this problem is simply getting over it with a language partner or teacher.
I finished the French course through level 1, and was about 60% done through level 5. That’s advanced enough to know a lot of what you don’t know, which makes filling in the gaps a lot easier. I think my word list in the app was 2k.
For Spanish I only got about 30% to level 1. The gaps there are obviously much wider but since it’s got a lot of similarities to French, and the fact that there are tons of learning material, it’s not too hard to make steady progress.
I tried Russian on Duolingo for a while, and it was actually great for learning the Cyrillic alphabet, but the explanations of cases, declensions, word order, etc were really lacking and just grinding through memorizing individual sentences was obviously not going to be effective.
But I never felt that the ai was really doing anything helpful… I got tons of the same easy questions, and even “hard” lessons didn’t seem much different except for being slightly longer sentences.
What’s better than duo? Other people have talked about other apps being better, I haven’t tried anything but Anki.
I don’t have any structured conversation time set up yet; italki or something similar seems like a good idea. Actually producing the language on demand is definitely a weak spot. But I do a fair bit of reading, listening to podcasts/Netflix/youtube, and have built up good Anki decks.
Your observation about irregular verbs and conjugation in various tenses is spot on. That’s one of the harder things about keeping track of the flow of a podcast for me too. Video is easier with more visual cues, and especially if I have cc on ;)
Studying the decks daily is not only far more time efficient than duo, I’m finding that the learning sticks much better. (Part of this I think is because I have had to look up the rules and exceptions to be able to make up cards for some focused study topics. I think it also helps when new vocab or sentences come from a book I read.)
There is absolutely something special about building decks too. I also get pretty much the same motivation from the timeline statistics in Anki as the streaks in Duolingo.
I think there should be a "never show/ask me this again" button on every slide/question in every learning app. This alone can make their "AI"s a lot smarter.
Through italki (spanish) I make tu/usted (informal/formal) mistakes, I make plural mistakes, and I make gender mistakes. My italki instructor still understands me and we can focus on other things.
I tried Duolingo Spanish pretty much every day for 1-2 years at the highest rate (5 lessons/day IIRC) and honestly... I don't remember any of it.
The vocab is extremely selective (and limited). It's too easy to guess answers without knowing them. Simple things like capitalization giving away the first word of the answer and trailing puncutation giving away the last. It seems like your recorded "fluency" just faded at a simplistic rate rather than there being any true attempt at spaced repetition.
And then there was the gamification. The main goal of this was, of course, to get you to pay. At some point (years ago now) they added this hearts system where guessing wrong would hurt you. This is completely counterproductive (IMHO). The streaks, the badges, etc. It just seemed like it was shallow engagement bait.
But an AI that learns what you need to learn? No way. Not even close.
This might be a newer feature and perhaps only on the browser version, as I don't use the app. But you can configure Duolingo to make you type in all the words freeform instead of suggesting words for you. This helps get rid of the little cheat you mention above.
Also the cheat mentioned above only sort of works because the app does remove them when it wants to make a question more difficult I've found.
Ugh. I want so badly to believe that this kind of technology of manipulation would be used for nothing but good.
Unfortunately I know better. Duolingo, and education in general, is a great use-case for many of these technologies. The reality of how they'll be employed has been, and will continue to be, less rosy.
I switched to Babbel last week. It seems to do a better job at teaching you the basics.
It's possible this is just the language I'm learning, but the "duolingo learning what you need to learn" seems iffy in my experience. It frequently prioritizes three words for me: "spring", "autumn" and "bag". It's unclear why, but every time I do an "review of old material" practice there are those three lessons.
There's a lot of material which I make actual mistakes on that it could prioritize, but it never does. It used to be you could get a the whole list of words (and duo's rating of your progress) but they killed the endpoint with the redesign. (Probably reasonable, the main use case for it was to dump the course into something like Anki... certainly that's what I was going to do with it).
I definitely made a fair amount of progress using Duo, but never got beyond the "struggle to communicate in any meaningful way" stage. Fluent Forever (a competing app) got me a lot farther (but is expensive and requires a fair amount of invested effort), and some basic flashcards got me much closer to reading comprehension than grinding Duo lessons.
All in all I can't tell if this is a marketing speak that is only loosely connected to reality or if duolingo just has inconsistently rolled out features.
How far did you go in Duolingo? I have been working away at it for 10-30 min/day for around 18 months and am around halfway through Intermediate 2. I find I can read and write quite well but my speaking has a lot of room for improvement. I feel like if I fully immersed myself in the language, the base I’ve built would allow me to get to oral fluency quite quickly.
The most recent changes to UI reduced user choice on what to learn and practice significantly. Some of it is being repaired, but if you are continuing to use Duolingo you are deferring to its algorithm significantly on how you should learn.
This is probably a good thing in some ways as they are potentially better at identifying weaknesses I'm not aware of. However, I know for a fact that I'm weak on some things and I want to practice those things with their exercises. It'd be really helpful if it gave me the freedom to make my own decisions more easily. For instance, scrolling back in lesson history was lousy previously but at least possible. With the path UI it's much worse - still technically possible but so much scrolling.
Solving for this would be pretty easy. "Jump scrolling" as I call it, not sure if that's the correct term, where the scrolling shows indexes based on letter or number that you can scroll through more quickly would let me go back to lessons I want to revisit more easily. I'm also sure that people who work on algorithms like this are capable of coming up with other ways to take into account how I might know or understand my own weaknesses that their observations can't just detect. I see no curiosity about that in this article though.
It's pretty clear that Duolingo is more invested in improving their algorithms and all of what is discussed is entirely based on observing the learner rather asking the directly or providing them easy ability to make their own learning path when you feel Duo isn't getting it right.
I still use Duolingo and think it's been a great way for me to learn a language that I probably wouldn't learn otherwise. However I'll always be slightly disappointed in how restricted it feels. It's a tough balance for sure and I can probably solve my concerns with outside solutions like supplemental reading and flash card apps. But I would love to have Duolingo adapt to my preferences in some ways and take into account how I know my own deficiencies in learning. I just don't think they're culturally or financially motivated to make those changes, and I live with that comfortably for the most part. At least until I read articles like this that informs me that I'm paying for some amazing learning technology that has no interest in directly listening to me.
I hope they consider changing, I won't be surprised or upset if they don't.
Thanks for posting this article!
Complaining directly is useless in the face of a rigid data driven approach (assuming the correct data is even being gathered), so I've decided to vote with my feet and hit em in the OKRs by switching to another app. I've been using Lingodeer
Sounds like it's not going to be changed back. It's a shame, really.
Thanks for the recommendation!
Learning languages is awesome, and I fucking hated it and was awful at it. So I’m curious if new approaches will be more successful overall, but after some initial head scratching I’m curious to learn more about language pedagogy and hope my kids do better than I did!
What if what humanity needs to realize is that language learning is a waste of time, unless you are studying the history of language. There's no reason we shouldn't have only one planetary language, besides the whims of governments and nationalists.
I dream of a advanced designed language that incorporates the best features of existing written and verbal systems. Now that's a goal worth spending time on.
Learning languages is fun, and the people you can meet (which you wouldn't otherwise) and things you learn about your own native languages is worth it to me.
I think anything that allows other people to get outside their box and learn more about (and from!) people who don't share the same cultural background is a positive for the world.
What if what humanity needs is to realize that you don't have to scale to the limits of the planet for something to be worth while.
Maybe the "time better spent elsewhere" should be spent learning that we are all humans with amazing cultural backgrounds, and what could be a better way to engage with others than through language?
I've just been reading way to much history of South America recently to accept this idea, but let's be honest, the entire world is based on this concept. Every time somebody tries to setup a "Planetary Language/Religion/Culture" it comes at the expense of "the others". Fuck this idea.
All this is lost if we all switched to Esperanto tomorrow.